Building Skills in Python

Building Skills in Python
Release 2.6.5
Steven F. Lott
April 20, 2010
CONTENTS
Front Matter
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
5 5 6 7 8 9 9 10
1 Preface 1.1 Why Read This Book? . . . . . 1.2 Audience . . . . . . . . . . . . . 1.3 Organization of This Book . . . 1.4 Limitations . . . . . . . . . . . . 1.5 Programming Style . . . . . . . 1.6 Conventions Used in This Book 1.7 Acknowledgements . . . . . . . .
II
Language Basics
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11
15 15 15 16 19 21 21 24 25 28 31 31 34 36 40 41 42 44 45
2 Background and History 2.1 History . . . . . . . . 2.2 Features of Python . 2.3 Comparisons . . . . . 2.4 Some Jargon . . . . .
3 Python Installation 3.1 Windows Installation . . . . . . . 3.2 Macintosh Installation . . . . . . . 3.3 GNU/Linux and UNIX Overview 3.4 Build from Scratch Installation
4 Getting Started 4.1 Command-Line Interaction . . . . . . . 4.2 The IDLE Development Environment . 4.3 Script Mode . . . . . . . . . . . . . . . 4.4 Getting Help . . . . . . . . . . . . . . . 4.5 Syntax Formalities . . . . . . . . . . . . 4.6 Exercises . . . . . . . . . . . . . . . . . 4.7 Other Tools . . . . . . . . . . . . . . . . 4.8 Style Notes: Wise Choice of File Names
5 Simple Numeric Expressions and Output 47 5.1 Seeing Output with the print() Function (or print Statement) . . . . . . . . . . . . . . . . 47 5.2 Numeric Types and Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 5.3 Numeric Conversion (or Factory) Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.4 5.5 5.6
Built-In Math Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Expression Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Expression Style Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
54 56 60 61 61 61 63 64 66 68 71 71 73 75 78 78 79 80 81 83 83 85 88 90 91 92 93 94
6 Advanced Expressions 6.1 Using Modules . . . . . . . . . 6.2 The math Module . . . . . . . 6.3 The random Module . . . . . . 6.4 Advanced Expression Exercises 6.5 Bit Manipulation Operators . 6.6 Division Operators . . . . . . .
7 Variables, Assignment and Input 7.1 Variables . . . . . . . . . . . . . . . . . . . . . . . . 7.2 The Assignment Statement . . . . . . . . . . . . . 7.3 Input Functions . . . . . . . . . . . . . . . . . . . . 7.4 Multiple Assignment Statement . . . . . . . . . . . 7.5 The del Statement . . . . . . . . . . . . . . . . . . . 7.6 Interactive Mode Revisited . . . . . . . . . . . . . . 7.7 Variables, Assignment and Input Function Exercises 7.8 Variables and Assignment Style Notes . . . . . . . . 8 Truth, Comparison and Conditional Processing 8.1 Truth and Logic . . . . . . . . . . . . . . . . . . 8.2 Comparisons . . . . . . . . . . . . . . . . . . . . 8.3 Conditional Processing: the if Statement . . . . 8.4 The pass Statement . . . . . . . . . . . . . . . . 8.5 The assert Statement . . . . . . . . . . . . . . . 8.6 The if-else Operator . . . . . . . . . . . . . . . 8.7 Condition Exercises . . . . . . . . . . . . . . . . 8.8 Condition Style Notes . . . . . . . . . . . . . . . 9 Loops and Iterative Processing 9.1 Iterative Processing: For All and There Exists 9.2 Iterative Processing: The for Statement . . . . 9.3 Iterative Processing: The while Statement . . 9.4 More Iteration Control: break and continue 9.5 Iteration Exercises . . . . . . . . . . . . . . . . 9.6 Condition and Loops Style Notes . . . . . . . . 9.7 A Digression . . . . . . . . . . . . . . . . . . . 10 Functions 10.1 Semantics . . . . . . . . . . . . . . . . . . 10.2 Function Denition: The def and return 10.3 Function Use . . . . . . . . . . . . . . . . 10.4 Function Varieties . . . . . . . . . . . . . 10.5 Some Examples . . . . . . . . . . . . . . 10.6 Hacking Mode . . . . . . . . . . . . . . . 10.7 More Function Denition Features . . . . 10.8 Function Exercises . . . . . . . . . . . . . 10.9 Object Method Functions . . . . . . . . . 10.10 Functions Style Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
95 . 95 . 96 . 97 . 98 . 100 . 103 . 104 . . . . . . . . . . 107 107 109 110 111 112 113 115 118 121 122
. . . . . . . Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11 Additional Notes On Functions 125 11.1 Functions and Namespaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 ii
11.2 The global Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 11.3 Call By Value and Call By Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 11.4 Function Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
III
Data Structures
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
131
135 135 136 139 139 141 141 141 143 146 146 147 148 151 152 153 155 155 155 156 157 157 158 160 161 163 163 163 164 164 165 166 167 169 170 175 175 176 176 178 178 179 180 181 iii
12 Sequences: Strings, Tuples and Lists 12.1 Sequence Semantics . . . . . . . . . 12.2 Overview of Sequences . . . . . . . 12.3 Exercises . . . . . . . . . . . . . . . 12.4 Style Notes . . . . . . . . . . . . . .
13 Strings 13.1 String Semantics . . . . . . . . . . . . 13.2 String Literal Values . . . . . . . . . . 13.3 String Operations . . . . . . . . . . . 13.4 String Comparison Operations . . . . 13.5 String Statements . . . . . . . . . . . 13.6 String Built-in Functions . . . . . . . 13.7 String Methods . . . . . . . . . . . . . 13.8 String Modules . . . . . . . . . . . . . 13.9 String Exercises . . . . . . . . . . . . 13.10 Digression on Immutability of Strings 14 Tuples 14.1 Tuple Semantics . . . . . . . . . . 14.2 Tuple Literal Values . . . . . . . . 14.3 Tuple Operations . . . . . . . . . 14.4 Tuple Comparison Operations . . 14.5 Tuple Statements . . . . . . . . . 14.6 Tuple Built-in Functions . . . . . 14.7 Tuple Exercises . . . . . . . . . . 14.8 Digression on The Sigma Operator 15 Lists 15.1 List Semantics . . . . . . . . . . . 15.2 List Literal Values . . . . . . . . . 15.3 List Operations . . . . . . . . . . 15.4 List Comparison Operations . . . 15.5 List Statements . . . . . . . . . . 15.6 List Built-in Functions . . . . . . 15.7 List Methods . . . . . . . . . . . . 15.8 Using Lists as Function Parameter 15.9 List Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Defaults . . . . .
16 Mappings and Dictionaries 16.1 Dictionary Semantics . . . . . . . . . . . 16.2 Dictionary Literal Values . . . . . . . . . 16.3 Dictionary Operations . . . . . . . . . . . 16.4 Dictionary Comparison Operations . . . . 16.5 Dictionary Statements . . . . . . . . . . . 16.6 Dictionary Built-in Functions . . . . . . . 16.7 Dictionary Methods . . . . . . . . . . . . 16.8 Using Dictionaries as Function Parameter
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Defaults
16.9 Dictionary Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182 16.10 Advanced Parameter Handling For Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 17 Sets 17.1 17.2 17.3 17.4 17.5 17.6 17.7 17.8 17.9 Set Semantics . . . . . . . . . . . . . . . . Set Literal Values . . . . . . . . . . . . . . Set Operations . . . . . . . . . . . . . . . . Set Comparison Operators . . . . . . . . . Set Statements . . . . . . . . . . . . . . . . Set Built-in Functions . . . . . . . . . . . . Set Methods . . . . . . . . . . . . . . . . . Using Sets as Function Parameter Defaults Set Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 187 187 188 190 191 191 192 194 195 199 199 200 203 204 206 206 207 208 210 211 212 213 213 214 215 216 217 217 218 219 221 221 222 223 224 226 226 228 232 235 235 236 239 242 244 246 248
18 Exceptions 18.1 Exception Semantics . . . . . . . . 18.2 Basic Exception Handling . . . . . 18.3 Raising Exceptions . . . . . . . . . 18.4 An Exceptional Example . . . . . 18.5 Complete Exception Handling and 18.6 Exception Functions . . . . . . . . 18.7 Exception Attributes . . . . . . . 18.8 Built-in Exceptions . . . . . . . . 18.9 Exception Exercises . . . . . . . . 18.10 Style Notes . . . . . . . . . . . . . 18.11 A Digression . . . . . . . . . . . . 19 Iterators and Generators 19.1 Iterator Semantics . . . . . . . 19.2 Generator Function Semantics 19.3 Dening a Generator Function 19.4 Generator Functions . . . . . . 19.5 Generator Statements . . . . . 19.6 Iterators Everywhere . . . . . 19.7 Generator Function Example . 19.8 Generator Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . The nally . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . Clause . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
20 Files 20.1 File Semantics . . . . . . . . . . 20.2 File Organization and Structure 20.3 Additional Background . . . . . 20.4 Built-in Functions . . . . . . . . 20.5 File Statements . . . . . . . . . 20.6 File Methods . . . . . . . . . . . 20.7 Several Examples . . . . . . . . 20.8 File Exercises . . . . . . . . . . .
21 Functional Programming with Collections 21.1 Lists of Tuples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.2 List Comprehensions . . . . . . . . . . . . . . . . . . . . . . . . 21.3 Sequence Processing Functions: map(), filter() and reduce() 21.4 Advanced List Sorting . . . . . . . . . . . . . . . . . . . . . . . . 21.5 The Lambda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.6 Multi-Dimensional Arrays or Matrices . . . . . . . . . . . . . . . 21.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
22 Advanced Mapping Techniques 251 22.1 Default Dictionaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251 22.2 Inverting a Dictionary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252 22.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
IV
Data + Processing = Objects

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
255
259 259 262 263 264 266 269 271 287 287 292 294 296 296 297 299 299 301 303 307 307 310 313 315 319 320 321 322 322 327 329 330 331 332 333 334 336 336 343 343 344 346 348 v
23 Classes 23.1 Semantics . . . . . . . . . . . . . . . . 23.2 Class Denition: the class Statement 23.3 Creating and Using Objects . . . . . . 23.4 Special Method Names . . . . . . . . 23.5 Some Examples . . . . . . . . . . . . 23.6 Object Collaboration . . . . . . . . . 23.7 Class Denition Exercises . . . . . . .
24 Advanced Class Denition 24.1 Inheritance . . . . . . . . . . . . . . . . . . . 24.2 Polymorphism . . . . . . . . . . . . . . . . . 24.3 Built-in Functions . . . . . . . . . . . . . . . 24.4 Collaborating with max(), min() and sort() 24.5 Initializer Techniques . . . . . . . . . . . . . 24.6 Class Variables . . . . . . . . . . . . . . . . . 24.7 Static Methods and Class Method . . . . . . 24.8 Design Approaches . . . . . . . . . . . . . . . 24.9 Advanced Class Denition Exercises . . . . . 24.10 Style Notes . . . . . . . . . . . . . . . . . . . 25 Some Design Patterns 25.1 Factory . . . . . . . . . . 25.2 State . . . . . . . . . . . 25.3 Strategy . . . . . . . . . 25.4 Design Pattern Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26 Creating or Extending Data Types 26.1 Semantics of Special Methods . . . . . . . . . . . . . . . . 26.2 Basic Special Methods . . . . . . . . . . . . . . . . . . . . . 26.3 Special Attribute Names . . . . . . . . . . . . . . . . . . . 26.4 Numeric Type Special Methods . . . . . . . . . . . . . . . 26.5 Collection Special Method Names . . . . . . . . . . . . . . 26.6 Collection Special Method Names for Iterators and Iterable 26.7 Collection Special Method Names for Sequences . . . . . . 26.8 Collection Special Method Names for Sets . . . . . . . . . . 26.9 Collection Special Method Names for Mappings . . . . . . 26.10 Mapping Example . . . . . . . . . . . . . . . . . . . . . . . 26.11 Iterator Examples . . . . . . . . . . . . . . . . . . . . . . . 26.12 Extending Built-In Classes . . . . . . . . . . . . . . . . . . 26.13 Special Method Name Exercises . . . . . . . . . . . . . . . 27 Attributes, Properties and Descriptors 27.1 Semantics of Attributes . . . . . . . . . . . 27.2 Properties . . . . . . . . . . . . . . . . . . 27.3 Descriptors . . . . . . . . . . . . . . . . . . 27.4 Attribute Handling Special Method Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27.5 Attribute Access Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349 28 Decorators 28.1 Semantics of Decorators . . . 28.2 Built-in Decorators . . . . . 28.3 Dening Decorators . . . . . 28.4 Dening Complex Decorators 28.5 Decorator Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351 351 352 354 355 356 357 357 358 358 360 361
29 Managing Contexts: the with Statement 29.1 Semantics of a Context . . . . . . . . . . 29.2 Using a Context . . . . . . . . . . . . . . 29.3 Dening a Context Manager Function . . 29.4 Dening a Context Manager Class . . . . 29.5 Context Manager Exercises . . . . . . . .
Components, Modules and Packages

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
363
367 367 368 370 372 373 375 375 377 379 379 380 381 381 381
30 Modules 30.1 Module Semantics . . . . . . . . . . . 30.2 Module Denition . . . . . . . . . . . 30.3 Module Use: The import Statement 30.4 Finding Modules: The Path . . . . . 30.5 Variations on An import Theme . . 30.6 The exec Statement . . . . . . . . . . 30.7 Module Exercises . . . . . . . . . . . 30.8 Style Notes . . . . . . . . . . . . . . . 31 Packages 31.1 Package Semantics 31.2 Package Denition 31.3 Package Use . . . 31.4 Package Exercises 31.5 Style Notes . . . . 32 The 32.1 32.2 32.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Python Library 383 Overview of the Python Library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383 Most Useful Library Sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385 Library Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395 395 396 397 399 401 401 403 404 405 407 408 409
33 Complex Strings: the re Module 33.1 Semantics . . . . . . . . . . . . . 33.2 Creating a Regular Expression . 33.3 Using a Regular Expression . . . 33.4 Regular Expression Exercises . .
34 Dates and Times: the time and datetime Modules 34.1 Semantics: What is Time? . . . . . . . . . . . . . . 34.2 Some Class Denitions . . . . . . . . . . . . . . . . 34.3 Creating a Date-Time . . . . . . . . . . . . . . . . . 34.4 Date-Time Calculations and Manipulations . . . . . 34.5 Presenting a Date-Time . . . . . . . . . . . . . . . . 34.6 Formatting Symbols . . . . . . . . . . . . . . . . . . 34.7 Time Exercises . . . . . . . . . . . . . . . . . . . . .
vi
34.8 Additional time Module Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 410 35 File Handling Modules 35.1 The os.path Module . . . . . . . . . . . . . . . . 35.2 The os Module . . . . . . . . . . . . . . . . . . . . 35.3 The fileinput Module . . . . . . . . . . . . . . . 35.4 The glob and fnmatch Modules . . . . . . . . . . 35.5 The tempfile Module . . . . . . . . . . . . . . . . 35.6 The shutil Module . . . . . . . . . . . . . . . . . 35.7 The File Archive Modules: tarfile and zipfile 35.8 The sys Module . . . . . . . . . . . . . . . . . . . 35.9 Additional File-Processing Modules . . . . . . . . 35.10 File Module Exercises . . . . . . . . . . . . . . . . 36 File 36.1 36.2 36.3 36.4 36.5 36.6 36.7 36.8 36.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411 413 414 416 417 418 419 419 423 424 425 427 427 428 431 432 434 436 441 446 446 451 451 453 455 458 459 461 461 465 465 466 467 469 477 478 480 485 491
Formats: CSV, Tab, XML, Logs and Others Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comma-Separated Values: The csv Module . . . . . . . . . . . . . . . Tab Files: Nothing Special . . . . . . . . . . . . . . . . . . . . . . . . Property Files and Conguration (or .INI ) Files: The ConfigParser Fixed Format Files, A COBOL Legacy: The codecs Module . . . . . XML Files: The xml.etree and xml.sax Modules . . . . . . . . . . . Log Files: The logging Module . . . . . . . . . . . . . . . . . . . . . File Format Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . The DOM Class Hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . and Batch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
37 Programs: Standing Alone 37.1 Kinds of Programs . . . . . . . . . 37.2 Command-Line Programs: Servers 37.3 The optparse Module . . . . . . . 37.4 Command-Line Examples . . . . . 37.5 Other Command-Line Features . . 37.6 Command-Line Exercises . . . . . 37.7 The getopt Module . . . . . . . .
38 Architecture: Clients, Servers, the Internet and 38.1 About TCP/IP . . . . . . . . . . . . . . . . . . . 38.2 The World Wide Web and the HTTP protocol . 38.3 Writing Web Clients: The urllib2 Module . . . 38.4 Writing Web Applications . . . . . . . . . . . . . 38.5 Sessions and State . . . . . . . . . . . . . . . . . 38.6 Handling Form Inputs . . . . . . . . . . . . . . . 38.7 Web Services . . . . . . . . . . . . . . . . . . . . 38.8 Client-Server Exercises . . . . . . . . . . . . . . 38.9 Socket Programming . . . . . . . . . . . . . . .
the World . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Wide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Web . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
VI
Projects
499
39 Areas of the Flag 503 39.1 Basic Red, White and Blue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503 39.2 The Stars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504 40 Bowling Scores 41 Musical Pitches 507 509
vii
41.1 41.2 41.3 41.4 41.5
Equal Temperament Overtones . . . . . . Circle of Fifths . . . Pythagorean Tuning Five-Tone Tuning .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
510 511 511 512 513 515 515 517 518 519 521 521 522 523 524 525 525 526 528 529 529 531 532 534 535 537 539 542 545 545 549 552 552 553
42 What Can be Computed? 42.1 Background . . . . . . . . . . . . 42.2 The Turing Machine . . . . . . . 42.3 Example Machine . . . . . . . . 42.4 Turing Machine Implementation 42.5 Exercise 1 . . . . . . . . . . . . . 42.6 Test Machines . . . . . . . . . . 42.7 Exercise 2 . . . . . . . . . . . . . 42.8 Better Implementations . . . . . 42.9 Exercise 3 . . . . . . . . . . . . . 42.10 Consequences . . . . . . . . . . . 42.11 Other Applications . . . . . . . 42.12 Alternative Specications . . . . 42.13 Exercise 4 . . . . . . . . . . . . . 43 Mah 43.1 43.2 43.3 43.4 43.5 43.6 43.7 43.8 Jongg Hands Tile Class Hierarchy . . Wall Class . . . . . . . TileSet Class Hierarchy Hand Class . . . . . . . Some Test Cases . . . . Hand Scoring - Points . Hand Scoring - Doubles Limit Hands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
44 Chess Game Notation 44.1 Algebraic Notation . . . . . . . . 44.2 Algorithms for Resolving Moves 44.3 Descriptive Notation . . . . . . . 44.4 Game State . . . . . . . . . . . . 44.5 PGN Processing Specications .
VII
Back Matter
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
555
557 557 557 557 557 557 559 561 563
45 Bibliography 45.1 Use Cases . . . . . 45.2 Computer Science 45.3 Design Patterns . 45.4 Languages . . . . 45.5 Problem Domains 46 Indices and Tables 47 Production Notes Bibliography
viii
Building Skills in Python, Release 2.6.5
A Programmers Introduction to Python
Legal Notice This work is licensed under a Creative Commons License. You are free to copy, distribute, display, and perform the work under the following conditions: Attribution. You must give the original author, Steven F. Lott, credit. Noncommercial. You may not use this work for commercial purposes. No Derivative Works. You may not alter, transform, or build upon this work. For any reuse or distribution, you must make clear to others the license terms of this work.
CONTENTS
CONTENTS
Part I
Front Matter
CHAPTER
ONE
PREFACE
The Zen Of Python Tim Peters Beautiful is better than ugly. Explicit is better than implicit. Simple is better than complex. Complex is better than complicated. Flat is better than nested. Sparse is better than dense. Readability counts. Special cases arent special enough to break the rules. Although practicality beats purity. Errors should never pass silently. Unless explicitly silenced. In the face of ambiguity, refuse the temptation to guess. There should be one and preferably only one obvious way to do it. Although that way may not be obvious at rst unless youre Dutch. Now is better than never. Although never is often better than right now. If the implementation is hard to explain, its a bad idea. If the implementation is easy to explain, it may be a good idea. Namespaces are one honking great idea lets do more of those!
1.1 Why Read This Book?

You need this book because you need to learn Python. Here are a few reasons why you might need to learn Python You need a programming language which is easy to read and has a vast library of modules focused on solving the problems youre faced with. You saw an article about Python specically, or dynamic languages in general, and want to learn more. Youre starting a project where Python will be used or is in use. A colleague has suggested that you look into Python. Youve run across a Python code sample on the web and need to learn more. Python reects a number of growing trends in software development, putting it at or near the leading edge of good programming languages. It is a very simple language surrounded by a vast library of add-on modules. It is an open source project, supported by many individuals. It is an object-oriented language, binding data and processing into class denitions. It is a platform-independent, scripted language, with complete access 5
to operating system APIs. It supports integration of complex solutions from pre-built components. It is a dynamic language, which avoids many of the complexities and overheads of compiled languages. This book is a close-to-complete presentation of the Python language. It is oriented toward learning, which involves accumulating many closely intertwined concepts. In our experience teaching, coaching and doing programming, there is an upper limit on the clue absorption rate. In order to keep within this limit, weve found that it helps to present a language as ever-expanding layers. Well lead you from a very tiny, easy to understand subset of statements to the entire Python language and all of the built-in data structures. Weve also found that doing a number of exercises helps internalize each language concept. Three Faces of a Language. There are three facets to a programming language: how you write it, what it means, and the additional practical considerations that make a program useful. While many books cover the syntax and semantics of Python, in this book well also cover the pragmatic considerations. Our core objective is to build enough language skills that good object-oriented design will be an easy next step. The syntax of a language is covered in the language reference manual available online. In the case of relatively simple languages, like Python, the syntax is simple. Well provide additional examples of language syntax. The semantics of the language can be a bit more slippery than the syntax. Some languages involve obscure or unique concepts that make it diicult to see what a statement really means. In the case of languages like Python, which have extensive additional libraries, the burden is doubled. First, one has to learn the language, then one has to learn the libraries. The number of open source packages made available by the Python community can increase the eort required to understand an entire architecture. The reward, however, is high-quality software based on high-quality components, with a minimum of development and integration eort. Many languages oer a number of tools that can accomplish the same basic task. Python is no exception. It is often diicult to know which of many alternatives performs better or is easier to adapt. Well try to focus on showing the most helpful approach, emphasizing techniques that apply for larger development eorts. Well try to avoid quick and dirty solutions that are only appropriate when learning the language.
1.2 Audience
Professional programmers who need to learn Python are our primary audience. We provide specic help for you in a number of ways. Since Python is simple, we can address newbie programmers who dont have deep experience in a number of other languages. We will call out some details in specic newbie sections. Experienced programmers can skip these sections. Since Python has a large number of sophisticated built-in data structures, we address these separately and fully. An understanding of these structures can simplify complex programs. The object-orientation of Python provides tremendous exibility and power. This is a deep subject, and we will provide an introduction to object-oriented programming in this book. More advanced design techniques are addressed in Building Skills in Object-Oriented Design, [Lott05]. The accompanying libraries make it inexpensive to develop complex and complete solutions with minimal eort. This, however, requires some time to understand the packaged components that are available, and how they can be integrated to create useful software. We cover some of the most important modules to specically prevent programmers from reinventing the wheel with each project. Instructors are a secondary audience. If you are looking for classroom projects that are engaging, comprehensible, and focus on perfecting language skills, this book can help. Each chapter in this book contains exercises that help students master the concepts presented in the chapter. This book assumes an basic level of skill with any of the commonly-available computer systems. The following skills will be required. 6 Chapter 1. Preface
Download and install open-source application software. Principally, this is the Python distribution kit from http://www.python.org. However, we will provide references to additional software components. Create text les. We will address doing this in IDLE, the Python Integrated Development Environment (IDE). We will also talk about doing this with a garden-variety text editor like Komodo, VIM, EMACS, TEXTPAD and BBEDIT. Run programs from the command-line. This includes the DOS command shell in Microsoft Windows, or the Terminal tool in Linux or Apples Macintosh OS X. Be familiar with high-school algebra and some trigonometry. Some of the exercises make heavy use of basic algebra and trigonometry. When youve nished with this book you should be able to do the following. Use of the core procedural programming constructs: variables, statements, exceptions, functions. We will not, for example, spend any time on design of loops that terminate properly. Create class denitions and subclasses. This includes managing the basic features of inheritance, as well as overloaded method names. Use the Python collection classes appropriately, this includes the various kinds of sequences, and the dictionary.
1.3 Organization of This Book

This book falls into ve distinct parts. To manage the clue absorption rate, the rst three parts are organized in a way that builds up the language in layers from central concepts to more advanced features. Each layer introduces a few new concepts, and is presented in some depth. Programming exercises are provided to encourage further exploration of each layer. The last two parts cover the extension modules and provide specications for some complex exercises that will help solidify programming skills. Some of the chapters include digressions on more advanced topics. These can be skipped, as they cover topics related to programming in general, or notes about the implementation of the Python language. These are reference material to help advanced students build skills above and beyond the basic language. The rst part, Language Basics introduces the basic feartures of the Python language, covering most of the statements but sticking with basic numeric data types. Background and History provides some history and background on Python. Getting Started covers installation of Python, using the interpreter interactively and creating simple program les. Simple Numeric Expressions and Output covers the basic expressions and core numeric types. Variables, Assignment and Input introduces variables, assignment and some simple input constructs. Truth, Comparison and Conditional Processing adds truth and conditions to the language. Loops and Iterative Processing. In Functions well add basic function denition and function call constructs; Additional Notes On Functions introduces some advanced function call features. The second part, Data Structures adds a number of data structures to enhance the expressive power of the language. In this part we will use a number of dierent kinds of objects, prior to designing our own objects. Sequences: Strings, Tuples and Lists extends the data types to include various kinds of sequences. These include Strings , Tuples and Lists. Mappings and Dictionaries describes mappings and dictionaries. Exceptions covers exception objects, and exception creation and handling. Files covers les and several closely related operating system (OS) services. Functional Programming with Collections describes more advanced sequence techniques, including multi-dimensional matrix processing. This part attempts to describe a reasonably complete set of built-in data types. 1.3. Organization of This Book 7
The third part, Data + Processing = Objects, unies data and processing to dene the object-oriented programming features of Python. Classes introduces basics of class denitions and introduces simple inheritance. Advanced Class Denition adds some features to basic class denitions. Some Design Patterns extend this discussion further to include several common design patterns that use polymorphism. Creating or Extending Data Types describes the mechanism for adding types to Python that behave like the built-in types. Part four, Components, Modules and Packages, describes modules, which provide a higher-level grouping of class and function denitions. It also summarizes selected extension modules provided with the Python environment. Modules provides basic semantics and syntax for creating modules. We cover the organization of packages of modules in Packages. An overview of the Python library is the subject of The Python Library. Complex Strings: the re Module covers string pattern matching and processing with the re module. Dates and Times: the time and datetime Modules covers the time and datetime module. Programs: Standing Alone covers the creation of main programs. We touch just the tip of the client-server iceberg in Architecture: Clients, Servers, the Internet and the World Wide Web. Some of the commonly-used modules are covered during earlier chapters. In particular the math and random modules are covered in The math Module and the string module is covered in Strings. Files touches on fileinput, os, os.path, glob, and fnmatch. Finally, part ve, Projects, presents several larger and more complex programming problems. These are ranked from relatively simple to quite complex. Areas of the Flag covers computing the area of the symbols on the American ag. Bowling Scores covers scoring in a game of bowling. Musical Pitches has several algorithms for the exact frequencies of musical pitches. What Can be Computed? has several exercises related to computability and the basics of nite state machines. Mah Jongg Hands describes algorithms for evaluating hands in the game of Maj Jongg. Chess Game Notation deals with interpreting the log from a game of chess.
1.4 Limitations
This book cant cover everything Python. There are a number of things which we will not cover in depth, and some things which we cant even touch on lightly. This list will provide you directions for further study. The rest of the Python library. The library is a large, sophisticated, rapidly-evolving collection of software components. We selected a few modules that are widely-used. There are many books which cover the library in general, and books which cover specic modules in depth. The subject of Object-Oriented (OO) design is the logical next step in learning Python. That topic is covered in Building Skills in Object-Oriented Design [Lott05]. Database design and programming requires a knowledge of Python and a grip on OO design. It requires a digression into the relational model and the SQL language. Graphical User Interface (GUI) development requires a knowledge of Python, OO design and database design. There are two commonly-used toolkits: Tkinter and pyGTK. Web application development, likewise, requires a knowledge of Python, OO design and database design. This topic requires digressions into internetworking protocols, specically HTTP and SOAP, plus HTML, XML and CSS languages. There are numerous web development frameworks for Python.
Chapter 1. Preface
1.5 Programming Style

We have to adopt a style for presenting Python. We wont present a complete set of coding standards, instead well present examples. This section has some justication of the style we use for the examples in this book. Just to continune this rant, we nd that actual examples speak louder than any of the gratuitously detailed coding standards which are so popular in IT shops. We nd that many IT organizations waste considerable time trying to write descriptions of a preferred style. A good example, however, trumps any description. As consultants, we are often asked to provide standards to an inexperienced team of programmers. The programmers only look at the examples (often cutting and pasting them). Why spend money on empty verbiage that is peripheral to the useful example? One important note: we specically reject using complex prexes for variable names. Prexes are little more than visual clutter. In many places, for example, an integer parameter with the amount of a bet might be called pi_amount where the prex indicates the scope (p for a parameter) and type (i for an integer). We reject the pi_ as potentially misleading and therefore uninformative. This style of name is only appropriate for primitive types, and doesnt address complex data structures well at all. How does one name a parameter that is a list of dictionaries of class instances? pldc_? In some cases, prexes are used to denote the scope of an instance variables. Variable names might include a cryptic one-letter prex like f to denote an instance variable; sometimes programmers will use my or the as an English-like prex. We prefer to reduce clutter. In Python, instance variables are always qualied by self., making the scope crystal clear. All of the code samples were tested on Python 2.6 for MacOS, using an iMac running MacOS 10.5. Additional testing of all code was done with Windows 2000 on a Dell Latitude laptop as well as a VMWare implementation of Fedora 11.
1.6 Conventions Used in This Book

Here is a typical Code sample. Typical Python Example
combo = { } for i in range(1,7): for j in range(1,7): roll= i+j combo.setdefault( roll, 0 ) combo[roll] += 1 for n in range(2,13): print "%d %.2f%% " % ( n, combo[n]/36.0 )
1. This creates a Python dictionary, a map from key to value. If we initialize it with something like the following: combo = dict( [ (n,0) for n in range(2,13) ] ) , we dont need the setdefault() function call below. 2. This assures that the rolled number exists in the dictionary with a default frequency count of 0. 3. Print each member of the resulting dictionary. Something more obscure like [ (n,combo[n]/36.0) for n in range(2,13)] is certainly possible. The output from the above program will be shown as follows:
1.5. Programming Style
2 0.03% 3 0.06% 4 0.08% 5 0.11% 6 0.14% 7 0.17% 8 0.14% 9 0.11% 10 0.08% 11 0.06% 12 0.03% Tool completed successfully
We will use the following type styles for references to a specic Class, method(), attribute, which includes both class variables or instance variables. Sidebars When we do have a signicant digression, it will appear in a sidebar, like this. Tip: tip There will be design tips, and warnings, in the material for each exercise. These reect considerations and lessons learned that arent typically clear to starting OO designers.
1.7 Acknowledgements
Id like to thank Carl Frederick for asking me if I was using Python to develop complex applications. At the time, I said Id have to look into it. This is the result of that investigation. I am indebted to Thomas Pautler, Jim Bullock, Michal Van Dorpe, Matthew Curry, Igor Sakovich, Drew, John Larsen, Robert Lucente, Lex Hider, John Nowlan and Tom Elliott for supplying much-needed corrections to errors in previous editions. John Hayes provided particular complete and meticulous copy-editing.
10
Chapter 1. Preface
Part II
Language Basics
11
The Processing View

A programming language involves two closely interleaved topics. On one hand, there are the procedural constructs that process information inside the computer, with visible eects on the various external devices. On the other hand are the various types of data structures and relationships for organizing the information manipulated by the program. This part describes the most commonly-used Python statements, sticking with basic numeric data types. Data Structures will present a reasonably complete set of built-in data types and features for Python. While the two are tightly interwoven, we pick the statements as more fundamental because we can (and will) add new data types. Indeed, the essential thrust of object-oriented programming (covered in Data + Processing = Objects) is the creation of new data types. Some of the examples in this part refer to the rules of various common casino games. Knowledge of casino gambling is not essential to understanding the language or this part of the book. We dont endorse casino gambling. Indeed, many of the exercises reveal the magnitude of the house edge in most casino games. However, casino games have just the right level of algorithmic complexity to make for excellent programming exercises. Well provide a little background on Python in Background and History . From there, well move on to installing Python in Python Installation. In Simple Numeric Expressions and Output well introduce the print statement (and print() function); well use this to see the results of arithmetic expressions including the numeric data types, operators, conversions, and some built-in functions. Well expand on this in Advanced Expressions. Well introduce variables, the assignment statement, and input in Variables, Assignment and Input , allowing us to create simple input-process-output programs. When we add truth, comparisons, conditional processing in Truth, Comparison and Conditional Processing, and iteration in Loops and Iterative Processing, well have all the tools necessary for programming. In Functions and Additional Notes On Functions, well show how to dene and use functions, the rst of many tools for organizing programs to make them understandable.
13
14
CHAPTER
TWO
BACKGROUND AND HISTORY

History of Python and Comparison with Other Languages
This chapter describes the history of Python in History. The Features of Python is an overview of the features of Python. After that, Comparisons is a subjective comparison between Python and a few other other languages, using some quality criteria harvested from two sources: the Java Language Environment White Paper and On the Design of Programming Languages. This material can be skipped by newbies: it doesnt help explain Python, it puts it into a context among other programming languages.
2.1 History
Python is a relatively simple programming language that includes a rich set of supporting libraries. This approach keeps the language simple and reliable, while providing specialized feature sets as separate extensions. Python has an easy-to-use syntax, focused on the programmer who must type in the program, read what was typed, and provide formal documentation for the program. Many languages have syntax focused on developing a simple, fast compiler; but those languages may sacrice readability and writability. Python strikes a good balance between fast compilation, readability and writability. Python is implemented in C, and relies on the extensive, well understood, portable C libraries. It ts seamlessly with Unix, Linux and POSIX environments. Since these standard C libraries are widely available for the various MS-Windows variants, and other non-POSIX operating systems, Python runs similarly in all environments. The Python programming language was created in 1991 by Guido van Rossum based on lessons learned doing language and operating system support. Python is built from concepts in the ABC language and Modula-3. For information ABC, see The ABC Programmers Handbook [Geurts91], as well as http://www.cwi.nl/~steven/abc/. For information on Modula-3, see Modula-3 [Harbison92], as well as http://www.research.compaq.com/SRC/modula-3/html/home.html. The current Python development is centralized at http://www.python.org.
2.2 Features of Python

Python reects a number of growing trends in software development. It is a very simple language surrounded by a vast library of add-on modules. It is an open source project, supported by dozens of individuals. It is an object-oriented language. It is a platform-independent, scripted language, with complete access to operating
15
system API s. It supports integration of complex solutions from pre-built components. It is a dynamic language, allowing more run-time exibility than statically compiled languages. Additionally, Python is a scripting language with full access to Operating System (OS) services. Consequently, Python can create high level solutions built up from other complete programs. This allows someone to integrate applications seamlessly, creating high-powered, highly-focused meta-applications. This kind of very-high-level programming (programming in the large) is often attempted with shell scripting tools. However, the programming power in most shell script languages is severely limited. Python is a complete programming language in its own right, allowing a powerful mixture of existing application programs and unique processing to be combined. Python includes the basic text manipulation facilities of Awk or Perl. It extends these with extensive OS services and other useful packages. It also includes some additional data types and an easier-to-read syntax than either of these languages. Python has several layers of program organization. The Python package is the broadest organizational unit; it is collection of modules. The Python module, analogous to the Java package, is the next level of grouping. A module may have one or more classes and free functions. A class has a number of static (class-level) variables, instance variables and methods. Well lookl at these layers in detail in appropriate sections. Some languages (like COBOL) have features that are folded into the language itself, leading to a complicated mixture of core features, optional extensions, operating-system features and special-purpose data structures or algorithms. These poorly designed languages may have problems with portability. This complexity makes these languages hard to learn. One hint that a language has too many features is that a language subset is available. Python suers from none of these defects: the language has only about 24 statements (of which ve are declaratory in nature), the compiler is simple and portable. This makes the the language is easy to learn, with no need to create a simplied language subset.
2.3 Comparisons
Well measure Python with two yardsticks. First, well look at a yardstick originally used for Java. Then well look at yardstick based on experience designing Modula-2.
2.3.1 The Java Yardstick

The Java Language Environment White Paper [Gosling96] lists a number of desirable features of a programming language: Simple and Familiar Object-Oriented Secure Interpreted Dynamic Architecture Neutral Portable Robust Multithreaded Garbage Collection Exceptions 16 Chapter 2. Background and History
High Performance Python meets and exceeds most of these expectations. Well look closely at each of these twelve desireable attributes. Simple and Familiar. By simple, we mean that there is no GOTO statement, we dont need to explicitly manage memory and pointers, there is no confusing preprocessor, we dont have the aliasing problems associated with unions. We note that this list summarizes the most confusing and bug-inducing features of the C programming language. Python is simple. It relies on a few core data structures and statements. The rich set of features is introduced by explicit import of extension modules. Python lacks the problem-plagued GOTO statement, and includes the more reliable break, continue and exception raise statements. Python conceals the mechanics of object references from the programmer, making it impossible to corrupt a pointer. There is no language preprocessor to obscure the syntax of the language. There is no C-style union (or COBOL-style REDEFINES) to create problematic aliases for data in memory. Python uses an English-like syntax, making it reasonably familiar to people who read and write English or related languages. There are few syntax rules, and ordinary, obvious indentation is used to make the structure of the software very clear. Object-Oriented. Python is object oriented. Almost all language features are rst class objects, and can be used in a variety of contexts. This is distinct from Java and C++ which create confusion by having objects as well as primitive data types that are not objects. The built-in type() function can interrogate the types of all objects. The language permits creation of new object classes. It supports single and multiple inheritance. Polymorphism is supported via run-time interpretation, leading to some additional implementation freedoms not permitted in Java or C++. Secure. The Python language environment is reasonably secure from tampering. Pre-compiled python modules can be distributed to prevent altering the source code. Additional security checks can be added by supplementing the built-in __import__() function. Many security aws are problems with operating systems or framework software (for example, database servers or web servers). There is, however, one prominent language-related security problem: the buer overow problem, where an input buer, of nite size, is overwritten by input data which is larger than the available buer. Python doesnt suer from this problem. Python is a dynamic language, and abuse of features like the exec statement or the eval() function can introduce security problems. These mechanisms are easy to identify and audit in a large program. Interpreted. An interpreted language, like Python allows for rapid, exible, exploratory software development. Compiled languages require a sometimes lengthy edit-compile-link-execute cycle. Interpreted languages permit a simpler edit-execute cycle. Interpreted languages can support a complete debugging and diagnostic environment. The Python interpreter can be run interactively; which can help with program development and testing. The Python interpreter can be extended with additional high-performance modules. Also, the Python interpreter can be embedded into another application to provide a handy scripting extension to that application. Dynamic. Python executes dynamically. Python modules can be distributed as source; they are compiled (if necessary) at import time. Object messages are interpreted, and problems are reported at run time, allowing for exible development of applications. In C++, any change to centrally used class headers will lead to lengthy recompilation of dependent modules. In Java, a change to the public interface of a class can invalidate a number of other modules, leading to recompilation in the best case, or runtime errors in the worst case. Portable. Since Python rests squarely on a portable C source, Python programs behave the same on a variety of platforms. Subtle issues like memory management are completely hidden. Operating system
2.3. Comparisons
17
inconsistency makes it impossible to provide perfect portability of every feature. Portable GUIs are built using the widely-ported Tk GUI tools Tkinter, or the GTK+ tools and the the pyGTK bindings. Robust. Programmers do not directly manipulate memory or pointers, making the language run-time environment very robust. Errors are raised as exceptions, allowing programs to catch and handle a variety of conditions. All Python language mistakes lead to simple, easy-to-interpret error messages from exceptions. Multithreaded. The Python threading module is a Posix-compliant threading library. This is not completely supported on all platforms, but does provide the necessary interfaces. Beyond thread management, OS process management is also available, as are execution of shell scripts and other programs from within a Python program. Additionally, many of the web frameworks include thread management. In products like TurboGears, individual web requests implicitly spawn new threads. Garbage Collection. Memory-management can be done with explicit deletes or automated garbage collection. Since Python uses garbage collection, the programmer doesnt have to worry about memory leaks (failure to delete) or dangling references (deleting too early). The Python run-time environment handles garbage collection of all Python objects. Reference counters are used to assure that no live objects are removed. When objects go out of scope, they are eligible for garbage collection. Exceptions. Python has exceptions, and a sophisticated try statement that handles exceptions. Unlike the standard C library where status codes are returned from some functions, invalid pointers returned from others and a global error number variable used for determining error conditions, Python signals almost all errors with an exception. Even common, generic OS services are wrapped so that exceptions are raised in a uniform way. High Performance. The Python interpreter is quite fast. However, where necessary, a class or module that is a bottleneck can be rewritten in C or C++, creating an extension to the runtime environment that improves performance.
2.3.2 The Modula-2 Yardstick

One of the languages which strongly inuenced the design of Python was Modula-2. In 1974, N. Wirth (creator of Pascal and its successor, Modula-2) wrote an article On the Design of Programming Languages [Wirth74], which dened some timeless considerations in designing a programming language. He suggests the following: a language be easy to learn and easy to use; safe from misinterpretation; extensible without changing existing features; machine [platform] independent; the compiler [interpreter ] must be fast and compact; there must be ready access to system services, libraries and extensions written in other languages; the whole package must be portable. Python syntax is designed for readability; the language is quite simple, making it easy to learn and use. The Python community is always alert to ways to simplify Python. The Python 3.0 project is actively working to remove a few poorly-concieved features of Python. This will mean that Python 3.0 will be simpler and easier to use, but incompatible with Python 2.x in a few areas. Most Python features are brought in via modules, assuring that extensions do not change or break existing features. This allows tremendous exibility and permits rapid growth in the language libraries.
18
Chapter 2. Background and History
The Python interpreter is very small. Typically, it is smaller than the Java Virtual Machine. Since Python is (ultimately) written in C, it has the same kind of broad access to external libraries and extensions. Also, this makes Python completely portable.
2.4 Some Jargon

For folks new to developing software, it might help to understand a few distinctions made above. Interperted Not Interpreted (i.e., Compiled) Python is a byte-code interpreter. A Python code object is a sequence of bytes that represent various operations and values. The Python interpreter steps through the bytes, performing the operations. A compiled language (e.g., C, C++, etc.) is translated from source form to executable binary specic to operating system and hardware platform. Java is similar to Python: its compiled and the Java Virtual Machine is a byte-code interpreter. Dynamic Not Dynamic (i.e., Static) Python is a dynamic language. Variables and functions do not have dened data types. Instead, a variable is simply a label attached to an object. A function is a callable object with parameters, but no declared result type. Each object has a strongly-dened permanent class. There is no sophisticated compile-time type checking. Instead, any type mismatches will be detected at run-time. Since many types are nearly interchangeable, there isnt a need for a lot of type checking. For examples of interchangeable (polymorphic) types, see Simple Numeric Expressions and Output . Languages like C, C++ and Java have statically-declared variables and functions. Scripting Non-Scripting The scripting distinction is an operational feature of POSIX-compliant operating systems. Files which begin with the #!/path/to/interpreter will be used as scripts by the OS. They can be executed from the command-line because the interpreter is named in the rst line of the le. Languages like Java, C and C++ do not have this feature; these les must be compiled before they can be executed.
2.4. Some Jargon
19
20
Chapter 2. Background and History
CHAPTER
THREE
PYTHON INSTALLATION
Downloading, Installing and Upgrading Python
This chapter is becoming less and less relevant as Python comes pre-installed with most Linux-based operating systems. Consequently, the most interesting part of this chapter is the Windows Installation, where we describe downloading and installing Python on Windows. Python runs on a wide, wide variety of platforms. If your particular operating system isnt described here, refer to http://www.python.org/community/ to locate an implementation. Mac OS developers will nd it simplest to upgrade to Leopard (Max OS 10.5) or Snow Leopard (Mac OS 10.6), since it has Python included. The Mac OS installation includes the complete suite of tools. Well look at upgrading in Macintosh Installation. For other GNU/Linux developers, youll nd that Python is generally included in most distributions. Further, many Linux distributions automatically upgrade their Python installation. For example, Fedora Core 11 includes Python 2.6 and installs upgrades as they become available. You can nd installation guidelines in GNU/Linux and UNIX Overview . The Goal. The goal of installation is to get the Python interpreter and associated libraries. Windows users will get a program called python.exe. Linux and MacOS users will get the Python interpreter, a program named python. In addition to the libraries and the interpreter, your Python installation comes with a tutorial document (also available at http://docs.python.org/tutorial/) on Python that will step you through a number of quick examples. For newbies, this provides an additional point of view that you may nd helpful. You may also want to refer to the Beginners Guide Wiki at http://wiki.python.org/moin/BeginnersGuide.
3.1 Windows Installation

In some circumstances, your Windows environment may require administrator privilege. The details are beyond the scope of this book. If you can install software on your PC, then you have administrator privileges. In a corporate or academic environment, someone else may be the administrator for your PC. The Windows installation of Python has three broad steps. 1. Pre-installation: make backups and download the installation kit. 2. Installation: install Python. 3. Post-installation: check to be sure everything worked. Well go through each of these in detail.
21
3.1.1 Windows Pre-Installation

Backup. Before installing software, back up your computer. I strongly recommend that you get a tool like Nortons Ghost (http://www.symantec.com/norton/ghost) or clonezilla (http://clonezilla.org/). Products like these will create a CD that you can use to reconstruct the operating system on your PC in case something goes wrong. It is diicult to undo an installation in Windows, and get your computer back the way it was before you started. Ive never had a single problem installing Python. Ive worked with a number of people, however, who either have bad luck or dont read carefully and have managed to corrupt their Windows installation by downloading and installing software. While Python is safe, stable, reliable, virus-free, and well-respected, you may be someone with bad luck who has a problem. Often the problem already existed on your PC and installing Python was the straw that broke the camels back. A backup is cheap insurance. You should also have a folder for saving your downloads. You can create a folder in My Documents called downloads. I suggest that you keep all of your various downloaded tools and utilities in this folder for two reasons. If you need to reinstall your software, you know exactly what you downloaded. When you get a new computer (or an additional computer), you know what needs to be installed on that computer. Download. After making a backup, go to the http://www.python.org web site and look for the Download area. In here, youre looking for the pre-built Windows installer. This book will emphasize Python 2.6. In that case, the kit will have a lename like python-2.6.x.msi. When you click on the lename, your browser should start downloading the le. Save it in your downloads folder. Backup. Now is a good time to make a second backup. Seriously. This backup will have your untouched Windows system, plus the Python installation kit. It is still cheap insurance. If you have anti-virus software [you do, dont you? ] you may need to disable this until you are done installing Python. At this point, you have everything you need to install Python: A backup The Python installer
3.1.2 Windows Installation

Youll need two things to install Python. If you dont have both, see the previous section on pre-installation. A backup The Python installer Double-click the Python installer (python-2.6.x.msi). The rst step is to select a destination directory. The default destination should be C:\Python26 . Note that Python does not expect to live in the C:\My Programs folder. Because the My Programs folder has a space in the middle of the name something that is atypical for all operating systems other than Windows subtle problems can arise. Consequently, Python folks prefer to put Python into C:\Python26 on Windows machines. Click Next to continue. If you have a previous installation, then the next step is to conrm that you want to backup replaced les. The option to make backups is already selected and the folder is usually C:\Python26\BACKUP. This is the way it should be. Click Next to continue. The next step is the list of components to install. You have a list of ve components. Python interpreter and libraries. You want this.
22
Chapter 3. Python Installation
Tcl/Tk (Tkinter, IDLE, pydoc). You want this, so that you can use IDLE to build programs. Python HTML Help le. This is some reference material that youll probably want to have. Python utility scripts (Tools/). We wont be making any use of this in this book. In the long run, youll want it. Python test suite (Lib/test/). We wont make any use of this, either. It wont hurt anything if you install it. There is an Advanced Options... button that is necessary if you are using a company-supplied computer for which you are not the administrator. If you are not the administrator, and you have permission to install additional software, you can click on this button to get the Advanced Options panel. Theres a button labeled Non-Admin install that youll need to click in order to install Python on a PC where you dont have administrator privileges. Click Next to continue. You can pick a Start Menu Group for the Python program, IDLE and the help les. Usually, it is placed in a menu named Python 2.6. I cant see any reason for changing this, since it only seems to make things harder to nd. Click Next to continue. The installer puts les in the selected places. This takes less than a minute. Click Finish ; you have just installed Python on your computer. Tip: Debugging Windows Installation The only problem you are likely to encounter doing a Windows installation is a lack of administrative privileges on your computer. In this case, you will need help from your support department to either do the installation for you, or give you administrative privileges.
3.1.3 Windows Post-Installation

In your Start... menu, under All Programs , you will now have a Python 2.6 group that lists ve things: IDLE (Python GUI) Module Docs Python (command line) Python Manuals Uninstall Python Important: Testing If you select the Python (command line) menu item, youll see the Python (command line) window. This will contain something like the following.
Python 2.6.2 (r262:71605, Apr 14 2009, 22:40:02) [MSC v.1500 32 bit (Intel)] on win32 Type "help", "copyright", "credits" or "license" for more information. >>> ^Z
If you hit Ctrl-Z and then Enter , Python will exit. The basic Python program works. You can skip to Getting Started to start using Python. If you select the Python Manuals menu item, this will open a Microsoft Help reader that will show the complete Python documentation library.
3.1. Windows Installation
23
3.2 Macintosh Installation

Python is part of the MacOS environment. Tiger (Mac OS 10.4) includes Python 2.3.5 and IDLE. Leopard (Mac OS 10.5) includes Python 2.5.1. Snow Leopard (Mac OS 10.6) includes Python 2.6. Generally, you dont need to do much to get started. Youll just need to locate the various Python les. Look in /System/Library/Frameworks/Python.Framework/Versions for the relevant les. In order to upgrade software in the Macintosh OS, you must know the administrator, or owner password. If you are the person who installed or initially setup the computer, you had to pick an owner password during the installation. If someone else did the installation, youll need to get the password from them. A Mac OS upgrade of Python has three broad steps. 1. Pre-upgrade: make backups and download the installation kit. 2. Installation: upgrade Python. 3. Post-installation: check to be sure everything worked. Well go through each of these in detail.
3.2.1 Macintosh Pre-Installation

Before installing software, back up your computer. While you cant easily burn a DVD of everything on your computer, you can usually burn a DVD of everything in your personal Mac OS X Home directory. Ive never had a single problem installing Python. Ive worked with a number of people, however, who either have bad luck or dont read carefully and have managed to corrupt their Mac OS installation by downloading and installing software. While Python is safe, stable, reliable, virus-free, and well-respected, you may be someone with bad luck who has a problem. A backup is cheap insurance. Download. After making a backup, go to the http://www.python.org web site and look for the Download area. In here, youre looking for the pre-built Mac OS X installer. This book will emphasize Python 2.6. In that case, the kit lename will start with python-2.6.2.macosx. Generally, the lename will have a date embedded in it and look like python-2.6.2.macosx2009-04-16.dmg When you click on the lename, your browser should start downloading the le. Save it in your Downloads folder. Backup. Now is a good time to make a second backup. Seriously. It is still cheap insurance. At this point, you have everything you need to install Python: A backup The Python installer
3.2.2 Macintosh Installation

When you double-click the python-2.6.2-macosx2009-04-16.dmg, it will create a disk image named Universal MacPython 2.6.x . This disk image has your license, a ReadMe le, and the MacPython.mpkg. When you double-click the MacPython.mpkg e, it will take all the necessary steps to install Python on your computer. The installer will take you through seven steps. Generally, youll read the messages and click Continue. Introduction. Read the message and click Continue. Read Me. This is the contents of the ReadMe le on the installer disk image. Read the message and click Continue.
24
License. You can read the history of Python, and the terms and conditions for using it. To install Python, you must agree with the license. When you click Continue , you will get a pop-up window that asks if you agree. Click Agree to install Python. Select Destination. Generally, your primary disk drive, usually named Macintosh HD will be highlighted with a green arrow. Click Continue. Installation Type. If youve done this before, youll see that this will be an upgrade. If this is the rst time, youll be doing an install. Click the Install or Upgrade button. Youll be asked for your password. If, for some reason, you arent the administrator for this computer, you wont be able to install software. Otherwise, provide your password so that you can install software. Finish Up. The message is usually The software was successfully installed. Click Close to nish.
3.2.3 Macintosh Post-Installation

In your Applications folder, youll nd a MacPython 2.6 folder, which contains a number of applications. BuildApplet Extras IDLE PythonLauncher Update Shell Prole.command Look in /System/Library/Frameworks/Python.Framework/Versions for the relevant les. In the bin , Extras and Resources directories youll nd the various applications. The bin/idle le will launch IDLE for us. Once youve nished installation, you should check to be sure that everything is working correctly. Important: Testing From the terminal you can enter the python command. You should see the following
MacBook-5:~ slott$ python Python 2.6.3 (r263:75184, Oct 2 2009, 07:56:03) [GCC 4.0.1 (Apple Inc. build 5493)] on darwin Type "help", "copyright", "credits" or "license" for more information. >>>
Enter end-of-le ctrl-D to exit from Python.
3.3 GNU/Linux and UNIX Overview

In Checking for Python well provide a procedure for examining your current conguration to see if you have Python in the rst place. If you have Python, and its version 2.6, youre all done. Otherwise, youll have to determine what tools you have for doing an installation or upgrade. If you have Yellowdog Updater Modied (YUM) see YUM Installation. If you have one of the GNU/Linux variants that uses the Red Hat Package Manager (RPM), see RPM Installation.
3.3. GNU/Linux and UNIX Overview
25
The alternative to use the source installation procedure in Build from Scratch Installation. Root Access. In order to install software in GNU/Linux, you must know the administrator, or root password. If you are the person who installed the GNU/Linux, you had to pick an administrator password during the installation. If someone else did the installation, youll need to get the password from them. Normally, we never log in to GNU/Linux as root except when we are installing software. In this case, because we are going to be installing software, we need to log in as root, using the administrative password. If you are a GNU/Linux newbie and are in the habit of logging in as root, youre going to have to get a good GNU/Linux book, create another username for yourself, and start using a proper username, not root. When you work as root, you run a terrible risk of damaging or corrupting something. When you are logged on as anyone other than root, you will nd that you cant delete or alter important les. Unix is not Linux. For non-Linux commercial Unix installations (Solaris, AIX, HP/UX, etc.), check with your vendor (Oracle/Sun, IBM, HP, etc.) It is very likely that they have an extensive collection of open source projects like Python pre-built for your UNIX variant. Getting a pre-built kit from your operating system vendor is an easy way to install Python.
3.3.1 Checking for Python

Many GNU/Linux and Unix systems have Python installed. On some older Linuxes [Linuxi? Lini? Linen?] there may be an older version of Python that needs to be upgraded. Heres what you do to nd out whether or not you already have Python. We cant easily cover all variations. Well use Fedora as a typical Linux distribution. Run the Terminal tool. Youll get a window which prompts you by showing something like [slott@linux01 slott]$ . In response to this prompt, enter env python, and see what happens. Heres what happens when Python is not installed.
[slott@linux01 slott]$ env python tcsh: python: not found
Heres what you see when there is a properly installed, but out-of-date Python on your GNU/Linux box.
[slott@linux01 slott]$ env python Python 2.3.5 (#1, Mar 20 2005, 20:38:20) [GCC 3.3 20030304 (Apple Computer, Inc. build 1809)] on darwin Type "help", "copyright", "credits" or "license" for more information. >>> ^D
We used an ordinary end-of-le (Control-D) to exit from Python. In this case, the version number is 2.3.5, which is good, but we need to install an upgrade.
3.3.2 YUM Installation

If you are a Red Hat or Fedora user, you likely have a program named Yum. If you dont have Yum, you should upgrade to Fedora Core 11. Note that Yum repositories do not cover every combination of operating system and Python distribution. In these cases, you should consider an operating system upgrade in order to introduce a new Python distribution. If you have an out-of-date Python, youll have to enter two commands in the Terminal window.
26
yum upgrade python yum install tkinter
The rst command will upgrade the Python 2.6 distribution. You can use the command install instead of upgrade in the unlikely event that you somehow have Yum, but dont have Python. The second command will assure that the extension package named tkinter is part of your Fedora installation. It is not, typically, provided automatically. Youll need this to make use of the IDLE program used extensively in later chapters. In some cases, you will also want a packaged called the Python Development Tools. This includes some parts that are used by Python add-on packages.
3.3.3 RPM Installation

Many variants of GNU/Linux use the Red Hat Package Manager (RPM). The rpm tool automates the installation of software and the important dependencies among software components. If you dont know whether on not your GNU/Linux uses the Red Hat Package manager, youll have to nd a GNU/Linux expert to help you make that determination. Red Hat Linux (and the related Fedora Core distributions) have a version of Python pre-installed. Sometimes, the pre-installed Python is an older release and needs an upgrade. This book will focus on Fedora Core GNU/Linux because thats what I have running. Specically, Fedora Core 8. You may have a dierent GNU/Linux, in which case, this procedure is close, but may not be precisely what youll have to do. The Red Hat and Fedora GNU/Linux installation of Python has three broad steps. 1. Pre-installation: make backups. 2. Installation: install Python. Well focus on the simplest kind of installation. 3. Post-installation: check to be sure everything worked. Well go through each of these in detail.
3.3.4 RPM Pre-Installation

Before installing software, back up your computer. You should also have a directory for saving your downloads. I recommend that you create a /opt directory for these kinds of options which are above and beyond the basic Linx installation. You can keep all of your various downloaded tools and utilities in this directory for two reasons. If you need to reinstall your software, you know exactly what you downloaded. When you get a new computer (or an additional computer), you know what needs to be installed on that computer.
3.3.5 RPM Installation

A typical scenario for installing Python is a command like the following. This has specic le names for Fedora Core 9. Youll need to locate appropriate RPMs for your distribution of Linux.
rpm -i http://download.fedora.redhat.com/pub/fedora/linux/development\ /i386/os/Packages/python-2.5.1-18.fc9.i386.rpm
3.3. GNU/Linux and UNIX Overview
27
Often, thats all there is to it. In some cases, youll get warnings about the DSA signature. These are expected, since we didnt tell RPM the public key that was used to sign the packages.
3.3.6 RPM Post-Installation

Important: Testing Run the Terminal tool. At the command line prompt, enter env python, and see what happens.
[slott@localhost trunk]$ env python Python 2.6 (r26:66714, Jun 8 2009, 16:07:26) [GCC 4.4.0 20090506 (Red Hat 4.4.0-4)] on linux2 Type "help", "copyright", "credits" or "license" for more information. >>>
If you hit Ctrl-D (the GNU/Linux end-of-le character), Python will exit. The basic Python program works.
3.4 Build from Scratch Installation

There are many GNU/Linux variants, and we cant even begin to cover each variant. You can use a similar installation on Windows or the Mac OS, if you have the GCC compiler installed. Heres an overview of how to install using a largely manual sequence of steps. 1. Pre-Installation. Make backups and download the source kit. Youre looking for the a le named python-2.5.x.tgz. 2. Installation. The installation involves a fairly common set of commands. If you are an experienced system administrator, but a novice programmer, you may recognize these. Change to the /opt/python directory with the following command.
cd /opt/python
Unpack the archive le with the following command.

tar -zxvf Python-2.6.x.tgz
Do the following four commands to congure the installation scripts and make the Python package. and then install Python on your computer.
cd Python-2.6 ./configure make
As root, youll need to do the following command. Either use sudo or su to temporarily elevate your privileges.
make install
3. Post-installation. Check to be sure everything worked. Important: Testing Run the Terminal tool. At the command line prompt, enter env python, and see what happens.
28
If you hit Ctrl-D (the GNU/Linux end-of-le character), Python will exit. The basic Python program works. Tip: Debugging Other Unix Installation The most likely problem youll encounter in doing a generic installation is not having the appropriate GNU GCC compiler. In this case, you will see error messages from congure which identies the list of missing packages. Installing the GNU GCC can become a complex undertaking.
3.4. Build from Scratch Installation
29
30
CHAPTER
FOUR
GETTING STARTED
Interacting with Python
Python is an interpreted, dynamic language. The Python interpreter can be used in two modes: interactive and scripted. In interactive mode, Python responds to each statement while we type. In script mode, we give Python a le of statements and turn it loose to interpret all of the statements in that script. Both modes produce identical results. When were producing a nished application program, we set it up to run as a script. When were experimenting or exploring, however, we may use Python interactively. Well describe the interactive command-line mode for entering simple Python statements in Command-Line Interaction. In The IDLE Development Environment well cover the basics of interactive Python in the IDLE environment. Well describes the script mode for running Python program les in Script Mode. Well look at the help unction in Getting Help. Once weve started interacting with Python, we can address some syntax issues in Syntax Formalities. Well mention some other development tools in Other Tools. Well also address some style issues in Style Notes: Wise Choice of File Names.
4.1 Command-Line Interaction

Well look at interaction on the command line rst, because it is the simplest way to interact with Python. It parallels scripted execution, and helps us visualize how Python application programs work. This is the heart of IDLE as well as the foundation for any application programs we build. This is not the only way or even the most popular way to run Python. It is, however, the simplest and serves as a good place to start.
4.1.1 Starting and Stopping Command-Line Python

Starting and stopping Python varies with your operating system. Generally, all of the variations are nearly identical, and dier only in minor details. Windows. There are two ways to start interactive Python under Windows. 1. You can run the command tool (cmd.exe) and enter the python command. 2. You can run the Python (Command Line) program under the Python2.6 menu item on the Start menu.
31
To exit from Python, enter the end-of-le character sequence, Control-Z and Return. Mac OS, GNU/Linux and Unix. You will run the Terminal tool. You can enter the command python to start interactive Python. To exit from Python, enter the end-of-le character, Control-D.
4.1.2 Entering Python Statements

When we run the Python interpreter (called python , or Python.exe in Windows), we see a greeting like the following:
When we get the >>> prompt, the Python interpreter is looking for input. We can type any Python statements we want. Each complete statement is executed when it is entered. In this section only, well emphasize the prompts from Python. This can help newbies see the complete cycle of interaction between themselves and the Python interpreter. In the long run well be writing scripts and wont emphasize this level of interaction. Well only cover a few key rules. The rest of the rules are in Syntax Formalities. Rule 1. The essential rule of Python syntax is that a statement must be complete on a single line. There are some exceptions, which well get to below.
>>> 2 + 3 5
This shows Python doing simple integer arithmetic. When you entered 2 + 3 and then hit Return, the Python interpreter evaluated this statement. Since the statement was only an expression, Python printed the results. Well dig into to the various kinds of numbers in Simple Numeric Expressions and Output . For now, its enough to know that you have integers and oating-point numbers that look much like other programming languages. As a side note, integers have two slightly dierent avors fast (but small) and long (but slow). Python prefers to use the fast integers (called int) until your numbers get so huge that it has to switch to long. Arithmetic operators include the usual culprits: + , - , *, / , % and ** standing for addition, subtraction, multiplication, division, modulo (remainder after division) and raising to a power. The usual mathematical rules of operator precedence (multiplys and divides done before adds and subtracts) are in full force, and ( and ) are used to group terms against precedence rules. For example, converting 65 Fahrenheit to Celsius is done as follows:
>>> (65 - 32) * 5 / 9 18 >>> (65.-32)*5/9 18.333333333333332 >>>
32
Chapter 4. Getting Started
Note that the rst example used all integer values, and the result was an integer result. In the second example, the presence of a oat caused all the values to be coerced to oat. Also note that Python has the standard binary-to-decimal precision issue. The actual value computed does not have a precise binary representation, and the default display of the decimal equivalent looks strange. Well return to this in Numeric Types and Operators. Incomplete Statements. What happens when an expression statement is obviously incomplete?
>>> ( 65 - 32 ) * 5 / File "<stdin>", line 1 ( 65 - 32 ) * 5 / ^ SyntaxError: invalid syntax
Parenthensis. There is an escape clause in the basic rule of one statement one line. When the parenthesis are incomplete, Python will allow the statement to run on to multiple lines. Python will change the prompt to ... to show that the statement is incomplete, and more is expected.
>>> ( 65 - 32 ... )*5 / 9 18
Rule 5. It is also possible to continue a long statement using a \ escape at the end of the line.
>>> 5 + 6 *\ ... 7 47
This escape allows you to break up an extremely long statement for easy reading. Indentation. Python relies heavily on indendentation to make a program readable. When interacting with Python, we are often typing simple expression statements, which are not indented. Later, when we start typing compound statements, indentation will begin to matter. Heres what happens if we try to indent a simple expression statement.
>>> 5+6 SyntaxError: invalid syntax
Note that some statements are called compound statements they contain an indented suite of statements. Python will change the prompt to ... and wait until the entire compound statement is entered before it does does the evaluation. Well return to these when its appropriate in Truth, Comparison and Conditional Processing. History. When we type an expression statement, Python evaluates it and displays the result. When we type all other kinds of statements, Python executes it silently. Well see more of this, starting in Variables, Assignment and Input . Small mistakes can be frustrating when typing a long or complex statement. Python has a reasonable command history capability, so you can use the up-arrow key to recover a previous statement. Generally, youll prefer to create script les and run the scripts. When debugging a problem, however, interactive mode can be handy for experimenting. One of the desirable features of well-written Python is that most things can be tested and demonstrated in small code fragments. Often a single line of easy-to-enter code is the desired style for interesting programming
4.1. Command-Line Interaction
33
features. Many examples in reference manuals and unit test scripts are simply captures of interactive Python sessions.
4.2 The IDLE Development Environment

There are a number of possible integrated development environments (IDE) for Python. Python includes the IDLE tool, which well emphasize. Additionally, you can download or purchase a number of IDEs that support Python. In Other Tools well look at other development tools. Starting and stopping IDLE varies with your operating system. Generally, all of the variations are nearly identical, and dier only in minor details.
4.2.1 IDLE On Windows

There are several ways to start IDLE in Windows. 1. You can use IDLE (Python GUI) from the Python2.6 menu on the Start menu. 2. You can also run IDLE from the command prompt. This requires two conguration settings in Windows. Assure that C:Python26\Lib\idlelib on your system PATH. This directory contains IDLE.BAT . Assure that .pyw les are associated with C:\Python26\pythonw.exe. In order to suppress creation of a console window for a GUI application, Windows oers pythonw.exe. You can quit IDLE by using the Quit menu item under the File menu.
4.2.2 IDLE On Mac OS X

In the Mac OS, if youve done an upgrade, you may nd the IDLE program in the Python 2.6 folder in your Applications folder. You can double-click this icon to run IDLE. If you have the baseline application, youll have to nd IDLE in the directory /System/Library/Frameworks/Python.framework/Versions/Current/bin. Generally, this is directory part of your PATH setting, and you can type the command idle & in a Terminal window to start IDLE. When you run IDLE by double-clicking the idle icon, youll notice that two windows are opened: a Python Shell window and a Console window. The Console window isnt used for much. When you run IDLE from the Terminal window, no console window is opened. The Terminal window is the Python console. You can quit IDLE by using the Quit menu item under the File menu. You can also quit by using the Quit Idle menu item under the Idle menu. Since the Macintosh keyboard has a command key, , as well as a control key, ctrl, there are two keyboard mappings for IDLE. You can use the Congure IDLE... item under the Options menu to select any of the built-in Key Sets. Selecting the IDLE Classic Mac settings may be more comfortable for Mac OS users.
34
4.2.3 IDLE on GNU/Linux

Well avoid the GNOME and KDE subtleties. Instead, well focus on running IDLE from the Terminal tool. Since the le path is rather long, youll want to edit your .profile (or .bash_profile ) to include the following alias denition.
alias idle='env python /usr/lib/python2.5/idlelib/idle.py &'
This allows you to run IDLE by entering the command idle in a Terminal window. You can quit IDLE by using the Exit menu item under the File menu.
4.2.4 Basic IDLE Operations

Initially, youll see the following greeting from IDLE.
Python 2.6.3 (r263:75184, Oct 2 2009, 07:56:03) [GCC 4.0.1 (Apple Inc. build 5493)] on darwin Type "copyright", "credits" or "license()" for more information. **************************************************************** Personal firewall software may warn about the connection IDLE makes to its subprocess using this computer's internal loopback interface. This connection is not visible on any external interface and no data is sent to or received from the Internet. **************************************************************** IDLE 2.6.3 >>>
You may notice a Help menu. This has the Python Docs menu item, which you can access through the menu or by hitting F1. This will launch Safari to show you the Python documents available on the Internet. The personal rewall notication is a reminder that IDLE uses Internetworking Protocols (IP) as part of its debugger. If you have a software rewall on you development computer, and the rewall software complains, you can allow the connection. IDLE has a simple and relatively standard text editor, which does Python syntax highlighting. It also has a Python Shell window which manages an interactive Python session. You will see that the Python Shell window has a Shell and a Debug menu. When you use the New menu item in the File menu, youll see a le window, which has a slightly dierent menu bar. A le window has name which is a le name (or untitled ), and two unique menus, a Run and a Format menu. Generally, youll use IDLE in two ways: Youll enter Python statements in the Python Shell window. Youll create les, and run those module les using the Run Module item in the Run menu. This option is usually F5.
4.2.5 The Shell Window

The Python Shell window in IDLE presents a >>> prompt. At this prompt, you can enter Python expressions or statements for evaluation. This window has a complete command history, so you can use the up arrow to select a previous statement and make changes. 4.2. The IDLE Development Environment 35
You can refer back to Command-Line Interaction ; those interactions will look and behave the same in IDLE as they do on the command line. The Shell Window is essentially the command-line interface wrapped in a scrolling window. The IDLE interface, however, provides a consistent working environment, which is independent of each operating systems command-line interface. The Shell and Debug menus provides functions youll use when developing larger programs. For our rst steps with Python, we wont need either of these menus. Well talk briey about the functions, but cant really make use of them until weve learned more of the language. The Shell Menu. The Shell menu is used to restart the Python interpreter, or scroll back through the shells log to locate the most recent restart. This is important when you are developing a module that is used as a library. When you change that module, you need to reset the shell so that the previous version is forgotten and the new version can be imported into a fresh, empty interpreter. Generally, being able to work interactively is the best way to develop working programs. It encourages you to create tidy, simple-looking components which you can exercise directly. The Debug Menu. The Debug menu provides some handy tools for watching how Python executes a program. The Go to File/Line item is used to locate the source le where an exception was raised. You click on the exception message which contains the le name and select the Go to File/Line menu item, and IDLE will open the le and highlight the selected line. The Debugger item opens an interactive debugger window that allows you to step through the executing Python program. The Stack Viewer item opens a window that displays the current Python stack. This shows the arguments and working variables in the Python interpereter. The stack is organized into local and global namespaces, a conceot we need to delve into in Variables, Assignment and Input . The Auto-open Stack Viewer option will open the Stack Viewer automatically when a program raises an unhandled exception. How exceptions are raised and handled is a concept well delve into in Exceptions.
4.2.6 The File Windows

Each le window in IDLE is a simple text editor with two additional menus. The Format menu has a series of items for fairly common source text manipulations. The formatting operations include indenting, commenting, handling tabs and formatting text paragraphs. The Run menu makes it easy to execute the le you are editing. The Python Shell menu item brings up the Python Shell window. The Check Module item checks the syntax for your le. If there are any errors, IDLE will highlight the oending line so you can make changes. Additionally, this option will check for inconsistent use of tabs and spaces for indentation. The Run Module , F5 , runs the entire le. Youll see the output in the Python Shell window.
4.3 Script Mode

In interactive mode, Python displays the results of expressions. In script mode, however, Python doesnt automatically display results.
36
In order to see output from a Python script, well introduce the print statement and the print() function. The print statement is the Python 2.6 legacy construct. The print() function is a new Python 3 construct that will replace the print statement. Well visit this topic in depth in Seeing Output with the print() Function (or print Statement). For now, you can use either one. Well show both. In the future, the print statement will be removed from the language.
4.3.1 The print Statement

The print statement takes a list of values and prints their string representation on the standard output le. The standard output is typically directed to the Terminal window.
print "PI = ", 355.0/113.0
We can have the Python interpreter execute our script les. Application program scripts can be of any size or complexity. For the following examples, well create a simple, two-line script, called example1.py.
example1.py
print 65, "F" print ( 65 - 32 ) * 5 / 9, "C"
4.3.2 The print() function

The print() functions takes a list of values and prints their string representation on the standard output le. The standard output is typically directed to the Terminal window. Until Python 3, we have to request the print() function with a special introductory statement: from __future__ import print_function.
from __future__ import print_function print( "PI = ", 355.0/113.0 )
We can have the Python interpreter execute our script les. Application program scripts can be of any size or complexity. For the following examples, well create a simple, two-line script, called example1.py.
example1.py
from __future__ import print_function print( 65, "F" ) print( ( 65 - 32 ) * 5 / 9, "C" )
4.3.3 Running a Script

There are several ways we can start the Python interpreter and have it evaluate our script le.
4.3. Script Mode
37
Explicitly from the command line. In this case well be running Python and providing the name of the script as an argument. Well look at this in detail below. Implicitly from the command line. In this case, well either use the GNU/Linux shell comment (sharpbang marker) or well depend on the le association in Windows. This is slightly more complex, and well look at this in detail below. Manually from within IDLE . Its important for newbies to remember that IDLE shouldnt be part of the nal delivery of a working application. However, this is a great way to start development of an application program. We wont look at this in detail because its so easy. Hit F5. MacBook users may have to hit fn and F5. Running Python scripts from the command-line applies to all operating systems. It is the core of delivering nal applications. We may add an icon for launching the application, but under the hood, an application program is essentially a command-line start of the Python interpreter.
4.3.4 Explicit Command Line Execution

The simplest way to execute a script is to provide the script le name as a parameter to the python interpreter. In this style, we explicitly name both the interpreter and the input script. Heres an example.
python example1.py
This will provide the example1.py le to the Python interpreter for execution.
4.3.5 Implicit Command-Line Execution

We can streamline the command that starts our application. For POSIX-standard operating systems (GNU/Linux, UNIX and MacOS), we make the script le itself executable and directing the shell to locate the Python interpreter for us. For Windows users, we associate our script le with the python.exe interpreter. There are one or two steps to this. 1. Associate your le with the Python interpreter. Except for Windows, you make sure the rst line is the following: #!/usr/bin/env python . For Windows, you must assure that .py les are associated with python.exe and .pyw les are associated with pythonw.exe. The whole le will look like this:
#!/usr/bin/env python print 65, "F" print ( 65 - 32 ) * 5 / 9, "C"
2. For POSIX-standard operating systems, do a chmod +x example1.py to make the le example1.py executable. You only do this once, typically the rst time you try to run the le. For Windows, you dont need to do this. Now you can run a script in most GNU/Linux environments by saying:
./example1.py
38
4.3.6 Windows Conguration

Windows users will need to be sure that python.exe is on their PATH. This is done with the System control panel. Click on the Advanced tab. Click on the Environment Variables... button. Click on the System variables Path line, and click the Edit... button. This will often have a long list of items, sometimes starting with %SystemRoot%. At the end of this list, add ";" and the direction location of Python.exe. On my machine, I put it in C:\Python26. For Windows programmers, the windows command interpreter uses the last letters of the le name to associate a le with an interpreter. You can have Windows run the python.exe program whenever you double-click a .py le. This is done with the Folder Options control panel. The File Types tab allows you to pair a le type with a program that processes the le.
4.3.7 GNU/Linux Conguration

We have to be sure that the Python interpreter is in value of the PATH that our shell uses. We cant delve into the details of each of the available UNIX Shells. However, the general rule is that the person who administers your POSIX computer should have installed Python and updated the /etc/profile to make Python available to all users. If, for some reason that didnt get done, you can update your own .profile to add Python to your PATH variable. The Sharp-Bang (shebang) Comment The #! technique depends on the way all of the POSIX shells handle scripting languages. When you enter a command that is the name of a le, the shell must rst check the le for the x (execute) mode; this was the mode you added with chmod +x. When execute mode is true, the shell must then check the rst few bytes to see what kind of le it is. The rst few bytes are termed the magic number ; deep in the bowels of GNU/Linux there is a database that shows what the magic number means, and how to work with the various kinds of les. Some les are binary executables, and the operating system handles these directly. When an executable les content begins with #!, it is a script le. The rest of the rst line names the program that will interpret the script. In this case, we asked the env program to nd the python interpreter. The shell nds the named program and runs it automatically, passing the name of script le as the last argument to the interpreter it found. The very cool part of this trick is that #! is a comment to Python. This rst line is, in eect, directed at the shell, and ignored by Python. The shell glances at it to see what the language is, and Python studiously ignores it, since it was intended for the shell.
4.3.8 Another Script Example

Throughout the rest of this book, were going to use this script processing mode as the standard way to run Python programs. Many of the examples will be shown as though a le was sent to the interpreter. For debugging and testing, it is sometimes useful to import the program denitions, and do some manipulations interactively. Well touch on this in Hacking Mode. Heres a second example. Well create a new le and write another small Python program. Well call it example2.py.
4.3. Script Mode
39
example2.py
#!/usr/bin/env python """Compute the odds of spinning red (or black) six times in a row on an American roulette wheel. """ print (18.0/38.0)**6
This is a one-line Python program with a two line module document string. Thats a good ratio to strive for. After we nish editing, we mark this as executable using chmod +x example2.py. Since this is a property of the le, this remains true no matter how many times we edit, copy or rename the le. When we run this, we see the following.
$ ./example2.py 0.0112962280375
Which says that spinning six reds in a row is about a one in eighty-nine probability.
4.4 Getting Help

Python has two closely-related help modes. One is the general help utility, the other is a help function that provides the documentation on a specic object, module, function or class.
4.4.1 The help() Utility

Help is available through the help() function. If you enter just help() you will enter the online help utility. This help utility allows you to explore the Python documentation. The interaction looks like this:
>>> help Type help() for interactive help, or help(object) for help about object. >>> help() Welcome to Python 2.5! This is the online help utility.
If this is your first time using Python, you should definitely check out the tutorial on the Internet at http://www.python.org/doc/tut/. Enter the name of any module, keyword, or topic to get help on writing Python programs and using Python modules. To quit this help utility and return to the interpreter, just type "quit". To get a list of available modules, keywords, or topics, type "modules", "keywords", or "topics". Each module also comes with a one-line summary of what it does; to list the modules whose summaries contain a given word such as "spam", type "modules spam". help>
40
Note that the prompt changes from Pythons standard >>> to a special help-mode prompt of help>. When you enter quit, you exit the help system and go back to Pythons ordinary prompt. To start, enter :samp:modules, :samp:keywords or :samp:topics to see the variety of information available.
4.4.2 Help on a specic topic

If you enter help( object ) for some object, you will be given help on that specic object. This help is displayed using a help viewer. Youll enter something like this:
>>> help("EXPRESSIONS")
Youll get a page of output, ending with a special prompt from the program thats helping to display the help messages. The prompt varies: Mac OS and GNU/Linux will show one prompt, Windows will show another. Mac OS and GNU/Linux. In standard OSs, youre interacting with a program named less; it will prompt you with : for all but the last page of your document. For the last page it will prompt you with (END). This program is very sophisticated. The four most important commands you need to know are the following. q Quit the less help viewer. h Get help on all the commands which are available. Enter a space to see the next page. b Go back one page. Windows. In Windows, youre interacting with a program named more; it will prompt you with -- More --. The four important commands youll need to know are the following. q Quit the more help viewer. h Get help on all the commands which are available. Enter a space to see the next page.
4.5 Syntax Formalities

What is a Statement?
Informally, weve seen that simple Python statements must be complete on a single line. As we will see in following chapters, compound statements are built from simple and compound statements. Fundamentally, Python has a simple equivalence between the lexical line structure and the statements in a Python program. This forces us to write readable programs with one statement per line. There are nine formal rules for the lexical structure of Python. 1. Simple statements must be complete on a single Logical Line. Starting in Truth, Comparison and Conditional Processing well look at compound statements, which have indented suites of statements, and which span multiple Logical Lines. The rest of these rules will dene how Logical Lines are built from Physical Lines through a few Line Joining rules.
4.5. Syntax Formalities
41
2. Physical Lines are dened by the platform; theyll end in standard n or the Windows ASCII CR LF sequence ( \r\n ). 3. Comments start with the # character outside a quoted string; comments end at the end of the physical line. These are not part of a statement; they may occur on a line by themselves or at the end of a statement. 4. Coding-Scheme Comments. Special comments that are by VIM or EMACS can be included in the rst or second line of a Python le. For example, # -*- coding: latin1 -*- 5. Explicit Line Joining. A \ at the end of a physical line joins it to the next physical line to make a logical line. This escapes the usual meaning of the line end sequence. The two or three-character sequences ( \n or \r\n ) are treated as a single space. 6. Implicit Line Joining. Expressions with ()s, []s or {}s can be split into multiple physical lines. 7. Blank Lines. When entering statements interactively, an extra blank line is treated as the end of an indented block in a compound statement. Otherwise, blank lines have no signcance. 8. Indentation. The embedded suite of statements in a compound statement must be indented by a consistent number of spaces or tabs. When entering statements interactively or in an editor that knows Python syntax (like IDLE), the indentation will happen automatically; you will outdent by typing a single backspace. When using another text editor, you will be most successful if you congure your editor to use four spaces in place of a tab. This gives your programs a consisent look and makes them portable among a wide variety of editors. 9. Whitespace at the beginning of a line is part of indentation, and is signicant. Whitespace elsewhere within a line is not signicant. Feel free to space things out so that they read more like English and less like computer-ese.
4.6 Exercises
4.6.1 Command-Line Exercises
1. Simple Commands. Enter the following one-line commands to Python: copyright license credits help 2. Simple Expressions. Enter one-line commands to Python to compute the following: 12345 + 23456 98765 - 12345 128 * 256 22 / 7 355 / 113 (18-32)*5/9 -10*9/5+32
42
4.6.2 IDLE Exercises

1. Create an Exercises Directory. Create a directory (or folder) for keeping your various exercise scripts. Be sure it is not in the same directory in which you installed Python. 2. Use IDLEs Shell Window. Start IDLE . Refer back to the exercises in Command-Line Interaction . Run these exercises using IDLE . 3. Use the IDLE File Window. Start IDLE . Note the version number. Use New Window under the File menu to create a simple le. The le should have the following content.
""" My First File """ print __doc__
Save this le in your exercises directory; be sure the name ends with .py . Run your le with the Run Module menu item in the Run menu, usually F5 .
4.6.3 Script Exercises

1. Print Script. Create and run Python le with commands like the following examples:
print print print print 12345 + 23456 98765 - 12345 128 * 256 22 / 7
Or, use the print function as follows.

from __future__ import print_function print(12345 + 23456) print(98765 - 12345) print(128 * 256) print(22 / 7)
2. Another Simple Print Script. Create and run a Python le with commands like the following examples:
print "one red", 18.0/38.0 print "two reds in a row", (18.0/38.0)**2
Or, use the print function as follows.

from __future__ import print_function print("one red", 18.0/38.0) print("two reds in a row", (18.0/38.0)**2)
3. Interactive Dierences. First, run IDLE (or Python) interactively and enter the following commands: copyright, license, credits. These are special global objects that print interesting things on the interactive Python console. Create a Python le with the three commands, each one on a separate line: copyright, license, credits. When you run this, it doesnt produce any output, nor does it produce an error. Now create a Python le with three commands, each on a separate line: print copyright, print license, print credits.
4.6. Exercises
43
Interestingly, these three global variables have dierent behavior when used in a script. This is rare. By default, there are just three more variables with this kind of behavior: quit, exit and help. 4. Numeric Types. Compare the results of 22/7 and 22.0/7. Explain the dierences in the output.
4.7 Other Tools

This section lists some additional tools which are popular ways to create, maintain and execute Python programs. While IDLE is suitable for many purposes, you may prefer an IDE with a dierent level of sophistication.
4.7.1 Any Platform

The Komodo Edit is an IDE that is considerably more sophisticated than IDLE. It is - in a way - too sophisticated for this book. Our focus is on the language, not high-powered IDEs. As with IDLE, this is a tool that runs everywhere, so you can move seamlessly from GNU/Linux to Wiundows to the Mac OS with a single, powerful tool. See http://www.komodo.com for more information on ordering and downloading.
4.7.2 Windows
Windows programmers might want to use a tool like Textpad. See http://www.textpad.com for information on ordering and downloading. Be sure to also download the python.syn le from http://www.textpad.com/add-ons which has a number of Python syntax coloring congurations. To use Textpad, you have two setup steps. First, youll need to add the Python document class. Second youll need to tell Textpad about the Python tool. The Python Document Class. You need to tell Textpad about the Python document class. Use the Congure menu; the New Document Class... menu item lets you add Python documents to Textpad. Name your new document class Python and click Next. Give your class members named *.py and click Next. Locate your python.syn le and click Next. Check the new Python document class, and click Next if everything looks right to create a new Textpad document class. The Python Tool. Youll want to add the Python interpreter as a Textpad tool. Use the Congure menu again, this time selecting the Preferences? item. Scroll down the list of preferences on the left and click on Tools. On the right, youll get a panel with the current set of tools and a prominent Add button on the top right-hand side. Click Add, and select Program... from the menu that appears. Youll get a dialog for locating a le; nd the Python.exe le. Click Okay to save this program as a Textpad tool. You can check this by using Congure menu and Preferences... item again. Scroll down the list to nd Tools . Click the + sign and open the list of tools. Click the Python tool and check the following: The Command is the exact path to your copy of Python.exe The Parameters contains $File The Initial Folder contains $FileDir The capture output option should be checked You might also want to turn o the Sound Alert option; this will beep when a program nishes running. I nd this makes things a little too noisy for most programs.
44
4.7.3 Macintosh
Macintosh programmers might want to use a tool like BBEdit. BBEdit can also run the programs, saving the output for you. See http://www.barebones.com for more information on BBEdit. To use BBEdit, you have two considerations when writing Python programs. You must be sure to decorate each Python le with the following line: #!/usr/bin/env python. This tells BBEdit that the le should be interpreted by Python. Well mention this again, when we get to script-writing exericses. The second thing is to be sure you set the chdir to Scripts Folder option when you use the the run... item in the #! (shebang) menu. Without this, scripts are run in the root directory, not in the directory that contains your script le.
4.8 Style Notes: Wise Choice of File Names

There is considerable exibility in the language; two people can arrive at dierent presentations of Python source. Throughout this book we will present the guidelines for formatting, taken from the Python Enhancement Proposal (PEP) 8, posted on http://python.org/dev/peps/pep-0008/. Well include guidelines that will make your programming consistent with the Python modules that are already part of your Python environment. These guidelines should also also make your programming look like other third-party programs available from vendors and posted on the Internet. Python programs are meant to be readable. The language borrows a lot from common mathematical notation and from other programming languages. Many languages (C++ and Java) for instance, dont require any particular formatting; line breaks and indendentation become merely conventions; bad-looking, hard-to-read programs are common. On the other hand, Python makes the line breaks and indentations part of the language, forcing you to create programs that are easier on the eyes. General Notes. Well touch on many aspects of good Python style as we introduce each piece of Python programming. We havent seen much Python yet, but we do need some guidance to prevent a few tiny problems that could crop up. First, Python (like all of Linux) is case sensitive. Some languages that are either all uppercase, or insensitive to case. We have worked with programmers who actually nd it helpful to use the Caps Lock key on their keyboard to expedite working in an all-upper-case world. Please dont do this. Python should look like English, where lower-case letters predominate. Second, Python makes use of indentation. Most programmers indent very nicely, and the compiler or interpreter ignores this. Python doesnt ignore it. Indentation is useful for write clear, meaning documents and programs are no dierent. Finally, your operating system allows a fairly large number of characters to appear in a le name. Until we start writing modules and packages, we can call our les anything that the operating system will tolerate. Starting in Components, Modules and Packages, however, well have to limit ourselves to lenames that use only letters, digits and _s. There can be just one ending for the lename: .py . A le name like exercise_1.py is better than the name execise-1.py. We can run both programs equally well from the command line, but the name with the hyphen limits our ability to write larger and more sophisticated programs.
4.8. Style Notes: Wise Choice of File Names
45
46
CHAPTER
FIVE
SIMPLE NUMERIC EXPRESSIONS AND OUTPUT

The print Statement and Numeric Operations
Basic expressions are the most central and useful feature of modern programming languages. To see the results of expressions, well use the print statement. This chapter starts out with Seeing Output with the print() Function (or print Statement), which covers the print statement. Numeric Types and Operators covers the basic numeric data types and operators that are integral to writing expressions Python. Numeric Conversion (or Factory) Functions covers conversions between the various numeric types. Built-In Math Functions covers some of the built-in functions that Python provides.
5.1 Seeing Output with the print() Function (or print Statement)
Before delving into expressions and numbers, well look at the print statement. Well cover just the essential syntax of the print statement; it has some odd syntax quirks that are painful to explain. Note: Python 3.0 Python 3.0 will replace the irregular print statement with a built-in print() function that is perfectly regular, making it simpler to explain and use. In order to use the print() function instead of the print statement, your script (or IDLE session) must start o with the following.
from __future__ import print_function
This replaces the print statement, with its irregular syntax with the print() function.
5.1.1 print Statement Syntax Overview

The print statement takes a list of values and, well, prints them. Speaking strictly, it does two things: 1. it converts the objects to strings and 2. puts the characters of those strings on standard output.
47
Generally, standard output is the console window where Python was started, although there are ways to change this that are beyond the scope of this book. Heres a quick summary of the more important features of print statement syntax. In short, the keyword, print, is followed by a comma-separated list of expressions.
print expression , ...
Note: Syntax Summary This syntax summary isnt completely correct because it implies that the list of expressions is terminated with a comma. Rather than fuss around with complex syntax diagrams (thats what the Python reference manual is for) weve shown an approximation that is close enough. The , in a print statement is used to separate the various expressions. A , can also be used at the end of the print statement to change the formatting; this is an odd-but-true feature that is unique to print statement syntax. Its hard to capture this sublety in a single syntax diagram. Further, this is completely solved by using the print() function. One of the simplest kind of expressions is a quoted string. You can use either apostrophes (') or quotes (") to surround strings. This gives you some exibility in your strings. You can put an apostrophe into a quoted string, and you can put quotes into an apostrophed string without the special escapes that some other languages require. The full set of quoting rules and alternatives, however, will have to wait for Strings. For example, the following trivial program prints three strings and two numbers.
print "Hi, Mom", "Isn't it lovely?", 'I said, "Hi".', 42, 91056
Multi-Line Output. Ordinarily, each print statement produces one line of output. You can end the print statement with a trailing , to combine the results of multiple print statements into a single line. Here are two examples.
print print print print "335/113=", 335.0/113.0 "Hi, Mom", "Isn't it lovely?", 'I said, "Hi".', 42, 91056
Since the rst print statement ends with a , it does not produce a complete line of output. The second print statement nishes the line of output. Redirecting Output. The print statements output goes to the operating systems standard output le. How do we send output to the systems standard error le? This involves some more advanced concepts, so well introduce it with a two-part recipe that we need to look at in more depth. Well revisit these topics in Components, Modules and Packages . First, youll need access to the standard error object; you get this via the following statement.
import sys
Second, there is an unusual piece of syntax called a chevron print which can be used to redirect output to standard error. >>
print file , expression , ...
48
Chapter 5. Simple Numeric Expressions and Output
Two common les are sys.stdout and sys.stderr. Well return to les in Files. Here is an example of a small script which produces messages on both stderr and stdout.
mixedout.py
#!/usr/bin/env python """Mixed output in stdout and stderr.""" import sys print >>sys.stderr, "This is an error message" print "This is stdout" print >>sys.stdout, "This is also stdout"
When you run this inside IDLE, youll notice that the stderr is colored red, where the stdout is colored black. Youll also notice that the order of the output in IDLE doesnt match the order in our program. Most POSIX operating systems buer stdout, but do not buer stderr. Consequently, stdout messages dont appear until the buer is full, or the program exits.
5.1.2 The print() Function

Python 3 replaces the relatively complex and irregular print statement with a simple and regular print() function. In Python 2.6 we can use this new function by doing the following:
from __future__ import print_function
This statement must be one of the rst executable statements in your script le. It makes a small but profuound change to Python syntax. The Python processor must be notied of this intended change up front. This provides us with the following: print([object, ...], [sep= ], [end=n], [le=sys.stdout] ) This will convert each object to a string, and then write the characters on the given le. The separator between objects is by default a single space. Setting a value for sep will set a dierent separator. The end-of-line character is by default a single newline. Setting a value for end will set a dierent end-of-line character. To change output les, provide a value for le. Multiline Output. To create multiline output, do the following:
from __future__ import print_function print( print( print( print( "335/113=", end="" ) 335.0/113.0 ) "Hi, Mom", "Isn't it lovely?", end="" ) 'I said, "Hi".', 42, 91056 )
Redirecting Output. The print statements output goes to the operating systems standard output le. How do we send output to the systems standard error le? This involves some more advanced concepts, so
5.1. Seeing Output with the print() Function (or print Statement)
49
well introduce it with a two-part recipe that we need to look at in more depth. Well revisit these topics in Components, Modules and Packages. First, youll need access to the standard error object. Second, youll provide the le option to the print() function.
from __future__ import sys print( "This is print( "This is print( "This is import print_function an error message", file=sys.stderr ) stdout" ) also stdout", file=sys.stdout )
Adding Features. You can with some care add features to the print() function. When we look at function denitions, well look at how we can override the built-in print() function to add our own unique features.
5.1.3 print Notes and Hints

A program produces a number of kinds of output. The print() function (or print statement) is a handy jumping-o point. Generally, well replace this with more advanced techiques. Final Reports. Our desktop applications may produce text-based report les. These are often done with print statements. PDF or other format output les. A desktop application which produces PDF or other format les will need to use additional libraries to produce PDF les. For example, ReportLab oers PDF-production libraries. These applications wont make extensive use of print statements. Error messages and processing logs. Logs and errors are often directed to the standard error le. You wont often use the print statement for this, but use the logging library. Debugging messages. Debugging messages are often handled by the logging library. The print statement (or print() function) is a very basic tool for debugging a complex Python program. Feel free to use print statements heavily to create a clear picture of what a program is actually doing. Ultimately, you are likely to replace print statements with other, more sophisticated methods.
5.2 Numeric Types and Operators

Python provides four built-in types of numbers: plain integers, long integers, oating point numbers and complex numbers. Numbers all have several things in common. Principally, the standard arithmetic operators of +, -, *, /, % and ** are all available for all of these numeric types. Additionally, numbers can be compared, using comparison operators that well look at in Comparisons. Also, numbers can be coerced from one type to another. More sophisticated math is separated into the math module, which we will cover later. However, a few advanced math functions are an integral part of Python, including abs() and pow().
5.2.1 Integers
Plain integers are at least 32 bits long. The range is at least -2,147,483,648 to 2,147,483,647 (approximately 2 billion). 50 Chapter 5. Simple Numeric Expressions and Output
Python represents integers as strings of decimal digits. A number does not include any punctuation, and cannot begin with a leading zero (0). Leading zeros are used for base 8 and base 16 numbers. Well look at this below.
>>> 355 >>> 355 >>> 355 >>> 3 255+100 397-42 71*5 355/113
While most features of Python correspond with common expectations from mathematics and other programming languages, the division operator, /, poses certain problems. Specically, the distinction between the algorithm and the data representation need to be made explicit. Division can mean either exact oatingpoint results or integer results. Mathematicians have evolved a number of ways of describing precisely what they mean when discussing division. We need similar expressive power in Python.Well look at more details of division operators in Division Operators. Binary, Octal and Hexadecimal. For historical reasons, Python supports programming in octal and hexadecimal. I like to think that the early days of computing were dominated by people with 8 or 16 ngers. A number with a leading 0 (zero) is octal, base 8, and uses the digits 0 to 7. 0123 is octal and equal to 83 decimal. A number with a leading 0x or 0X is hexadecimal, base 16, and uses the digits 0 through 9, plus a, A, b, B, c, C, d, D, e, E, f, and F. 0x2BC8 is hexadecimal and equal to 11208. A number with a leading 0b or 0B is binary, base 2, and uses digits 0 and 1. Important: Leading Zeroes When using Python 2.6, watch for leading zeros in numbers. If you simply transcribe programs from other languages, they may use leading zeros on decimal numbers. Important: Python 3 In Python 3, the octal syntax will change. Octal constants will begin with 0o to match hexadecimal constants which begin with 0x. 0o123 will be octal and equal to 83 decimal.
5.2.2 Long Integers

One of the useful data types that Python oers are long integers. Unlike ordinary integers with a limited range, long integers have arbitrary length; they can have as many digits as necessary to represent an exact answer. However, these will operate more slowly than plain integers. Long integers end in L or l. Upper case L is preferred, since the lower-case l looks too much like the digit 1. Python is graceful about converting to long integers when it is necessary. Important: Python 3 Python 3 will not require the trailing L. It will silently deduce if you need an integer or a long integer. How many dierent combinations of 32 bits are there? The answer is there are 232 ; 2**32 in Python. The answer is too large for ordinary integers, and we get the result as a long integer.
5.2. Numeric Types and Operators
51
>>> 2**32 4294967296L >>> 2**64 18446744073709551616L
There are about 4 billion ways to arrange 32 bits. How many bits in 1K of memory? 1024 8 bits. How many combinations of bits are possible in 1K of memory? 210248 .
print 2L**(1024*8)
I wont attempt to reproduce the output from Python. It has 2,467 digits. There are a lot of dierent combinations of bits in only 1K of memory. The computer Im using has 512 1024K bytes of memory; there are a lot of combinations of bits available in that memory. Python will silently convert between ultra-fast integers and slow-but-large long integers. You can force a conversion using the int() or long() factory functions.
5.2.3 Floating-Point Numbers

Python oers oating-point numbers, often implemented as double-precision numbers, typically using 64 bits. Floating-point numbers are written in two forms: a simple string of digits that includes a decimal point, and a more complex form that includes an explicit exponent.
.0625 0.0625 6.25E-2 625E-4
The last two examples are based on scientic notation, where numbers are written as a mantissa and an exponent. The E (or code:e) , powers of 10 are used with the exponent, giving us numbers that look like this: 6.25 102 and 625 104 . The last example isnt properly normalized, since the mantissa isnt between 0 and 10. Generally, a number, n, is some mantissa, g, and an exponent of c. For human consumption, we use a base of 10. Internally, most computers use a base of 2, not 10. n = g 10c n = h 2d This dierece in the mantissa leads to slight errors in converting certain values, which are exact in base 10, to approximations in base 2. For example, 1/5th doesnt have a precise representation. This isnt generally a problem because we have string formatting operations which can make this tiny representation error invisible to users.
>>> 1./5. 0.20000000000000001 >>> .2 0.20000000000000001
52
5.2.4 Complex Numbers

Besides plain integers, long integers and oating point numbers, Python also provides for imaginary and complex numbers. These use the European convention of ending with J or j. People who dont use complex numbers should skip this section. 3.14J is an imaginary number = 3.14 1. A complex number is created by adding a real and an imaginary number: 2 + 14j. Note that Python always prints these in ()s; for example (2+14j). The usual rules of complex math work perfectly with these numbers.
>>> (2+3j)*(4+5j) (-7+22j)
Python even includes the complex conjugate operation on a complex number. This operation follows the complex number separated by a dot (.). This notation is used because the conjugate is treated like a method function of a complex number object (well return to this method and object terminology in Classes). For example:
>>> 3+2j.conjugate() (3-2j)
5.3 Numeric Conversion (or Factory) Functions

We can convert a number from one type to another. A conversion may involve a loss of precision because weve reduced the number of bits available. A conversion may also add a false sense of precision by adding bits which dont have any real meaning. Well call these factory functions because they are a factory for creating new objects from other objects. The idea of factory function is a very general one, and these are just the rst of many examples of this pattern.
5.3.1 Numeric Factory Function Denitions

There are a number of conversions from one numeric type to another. int(x ) Generates an integer from the object x. If x is a oating point number, digits to the right of the decimal point are truncated as part of creating an integer. If the oating point number is more than about 10 digits, a long integer object is created to retain the precision. If x is a long integer that is too large to be represented as an integer, theres no conversion. Complex values cant be turned into integers directly. If x is a string, the string is parsed to create an integer value. It must be a string of digits with an optional sign (+ or -).
>>> int("1243") 1243 >>> int(3.14159) 3
5.3. Numeric Conversion (or Factory) Functions
53
float(x ) Generates a oat from object x. If x is an integer or long integer, a oating point number is created. Note that long integers can have a large number of digits, but oating point numbers only have approximately 16 digits; there can be some loss of precision. Complex values cant be turned into oating point numbers directly. If x is a string, the string is parsed to create an oat value. It must be a string of digits with an optional sign (+ or -). The digits can have a single decimal point (.). Also, a string can be in scientic notation and include e or E followed by the exponent as a simple signed integer value.
>>> float(23) 23.0 >>> float("6.02E24") 6.0200000000000004e+24 >>> float(22)/7 3.14285714286
long(x ) Generates a long integer from x. If x is a oating point number, digits to the right of the decimal point are truncated as part of creating a long integer.
>>> long(2) 2L >>> long(6.02E23) 601999999999999995805696L >>> long(2)**64 18446744073709551616L
complex(real, [imag] ) Generates a complex number from real and imag. If the imaginary part is omitted, it is 0.0. Complex is not as simple as the others. A complex number has two parts, real and imaginary. Conversion to complex typically involves two parameters.
>>> complex(3,2) (3+2j) >>> complex(4) (4+0j) >>> complex("3+4j") (3+4j)
Note that the second parameter, with the imaginary part of the number, is optional. This leads to a number of dierent ways to call this function. In the example above, we used three variations: two numeric parameters, one numeric parameter and one string parameter.
5.4 Built-In Math Functions

Python has a number of built-in functions, which are an integral part of the Python interpreter. We cant look at all of them because many are related to features of the language that we havent addressed yet. One of the built-in mathematical functions will have to wait for complete coverage until weve introduced the more complex data types, specically tuples, in Tuples. The divmod() function returns a tuple object with the quotient and remainder in division.
54
5.4.1 Built-In Math Functions

The bulk of the math functions are in a separate module, called math, which we will cover in The math Module . The formal denitions of mathematical built-in functions are provided below. abs(number ) Return the absolute value of the argument, |x|. pow(x, y, [z] ) Raise x to the y power, xy . If z is present, this is done modulo z, xy mod z . round(number, [digits] ) Round number to ndigits beyond the decimal point. If the ndigits parameter is given, this is the number of decimal places to round to. If ndigits is positive, this is decimal places to the right of the decimal point. If ndigits is negative, this is the number of places to the left of the decimal point. Examples:
>>> print round(678.456,2) 678.46 >>> print round(678.456,-1) 680.0
5.4.2 String Conversion Functions

The string conversion functions provide alternate representations for numeric values. This list expands on the function denitions in Numeric Conversion (or Factory) Functions. hex(number ) Create a hexadecimal string representation of number. A leading 0x is placed on the string as a reminder that this is hexadecimal.
>>> hex(684) '0x2ac'
oct(number ) Create a octal string representation of number. A leading 0 is placed on the string as a reminder that this is octal not decimal.
>>> oct(509) '0775'
bin(number ) Create a binary representation of number. A leading 0b is placed on the string as a reminder that this is binary and not decimal.
>>> bin(509) '0b111111101'
int(string, [base] ) Generates an integer from the string x. If base is supplied, x must be a string in the given base. If base is omitted, the string x must be decimal.
5.4. Built-In Math Functions
55
>>> int( '0775', 8 ) 509 >>> int( '0x2ac', 16 ) 684 >>> int( '101101101101', 2 ) 2925
The int() function has two forms. The int(x) form converts a decimal string, x, to an integer. For example, int('25') is 25. The int(x,b) form converts a string, x, in base b to an integer. For example, int('25',8) is 21. str(object ) Generate a string representation of the given object. This is the a readable version of the value. repr(object ) Generate a string representation of the given object. Generally, this is the a Python expression that can reconstruct the value; it may be rather long and complex. For the numeric examples weve seen so far, the value of repr() is generally the same as the value of str(). The str() and repr() functions convert any Python object to a string. The str() version is typically more readable, where the repr() version is an internalized representation. For most garden-variety numeric values, there is no dierence. For the more complex data types, however, the resultsof repr() and str() can be very dierent. For classes you write (see Classes), your class denition must provide these string representation functions.
5.4.3 Collection Functions

These are several built-in functions which operate on simple collections of data elements. max(value, ...) Return the largest value.
>>> max(1,2,3) 3
min(value, ...) Return the smallest value.

>>> min(1,2,3) 1
Additionally, there are several other collection-handling functions, including any(), all() and sum(). These will have to wait until we can look at collection objects in Data Structures.
5.5 Expression Exercises

There are two sets of exercises. The rst section, Basic Output and Functions, covers simpler exercises to reinforce Python basics. The second section, Numeric Types and Expressions, covers more complex numeric expressions.
56
5.5.1 Basic Output and Functions

1. Print Expression Results. In Command-Line Exercises, we entered some simple expressions into the Python interpreter. Change these simple expressions into print statements. Be sure to print a label or identier with each answer. Heres a sample.
print "9-1's * 9-1's = ", 111111111*111111111
Heres an example using the print() function.

from __future__ import print_function print( "9-1's * 9-1's = ", 111111111*111111111 )
2. Evaluate and Print Expressions. Write short scripts to print the results of the following expressions. In most places, changing integers to oating point produces a notably dierent result. For example (296/167)**2 and (296.0/167.0)**2 . Use long as well as complex types to see the dierences. 355/113 * ( 1 - 0.0003/3522 ) 22/17 + 37/47 + 88/83 (553/312)**2
5 3. Numeric Conversion. Write a print statement to print the mixed fraction 3 8 as a oating point number and as an integer.
4. Numeric Truncation. Write a print statement to compute (22.0/7.0)-int(22.0/7.0). What is this value? Compare it with 22.0/7.0. What general principal does this illustrate? 5. Illegal Conversions. Try illegal conversions like int('A') or int( 3+4j ). Why are exceptions raised? Why cant a simple default value like zero or None be used instead? 6. Evaluate and Print Built-in Math Functions. Write short scripts to print the results of the following expressions. pow( 2143/22, 0.25 ) pow(553/312,2) pow( long(3), 64 ) long( pow(float(3), 64) ) Why do the last two produce dierent results? What does the dierence between the two results tell us about the number of digits of precision in oating-point numbers? 7. Evaluate and Print Built-in Conversion Functions. Here are some more expressions for which you can print the results. hex( 1234 ) int( hex(1234), 16 ) long( 0xab ) int( 0xab ) int( 0xab, 16 ) int( ab, 16 ) cmp( 2, 3 )
5.5. Expression Exercises
57
5.5.2 Numeric Types and Expressions

1. Stock Value. Compute value from number of shares purchase price for a stock. Once upon a time, stock prices were quoted in fractions of a dollar, instead of dollars and cents. Create a simple print statement for 125 shares purchased at 3 3 8 . Create a second simple print statement for 1 7 150 shares purchased at 2 4 plus an additional 75 shares purchased at 1 8 . Dont manually convert 1 4 to 0.25. Use a complete expression of the form 2+1/4.0, just to get more practice writing expressions. 2. Convert Between |deg| C and |deg| F. Convert temperatures from one system to another. Conversion Constants: 32 F = 0 C, 212 F = 100 C. The following two formulae converts between C (Celsius) and F (Fahrenheit). 212 32 C 100 100 C = (F 32) 212 32 F = 32 + Create a print statement to convert 18 C to F. Create a print statement to convert -4 F to C. 3. Periodic Payment on a Loan. How much does a loan really cost? Here are three versions of the standard mortgage payment calculation, with m = payment, p = principal due, r = interest rate, n = number of payments. Dont be surprised by the sign of the results; theyre opposite the sign of the principle. With a positive principle, you get negative numbers; you are paying down a principle. ( ) r m=p 1 (1 + r)n Mortgage with payments due at the end of each period: m= rp(r + 1)n (r + 1)n 1
Mortgage woth payments due at the beginning of each period: m= rp(r + 1)n [(r + 1)n 1](r + 1)
Use any of these forms to compute the mortgage payment, m, due with a principal, p, of $110,000, an interest rate, r, of 7.25% annually, and payments, n, of 30 years. Note that banks actually process things monthly. So youll have to divide the interest rate by 12 and multiply the number of payments by 12. 4. Surface Air Consumption Rate. SACR is used by SCUBA divers to predict air used at a particular depth. For each dive, we convert our air consumption at that dives depth to a normalized air consumption at the surface. Given depth (in feet), d , starting tank pressure (psi), s, nal tank pressure (psi), f, and time (in minutes) of t, the SACR, c, is given by the following formula. c= 33(s f ) t(d + 33)
Typical values for pressure are a starting pressure of 3000, nal pressure of 500.
58
A medium dive might have a depth of 60 feet, time of 60 minutes. A deeper dive might be to 100 feet for 15 minutes. A shallower dive might be 30 feet for 60 minutes, but the ending pressure might be 1500. A typical c (consumption) value might be 12 to 18 for most people. Write print statements for each of the three dive proles given above: medium, deep and shallow. Given the SACR, c , and a tank starting pressure, s, and nal pressure, f, we can plan a dive to depth (in feet), d, for time (in minutes), t, using the following formula. Usually the 33(s f )/c is a constant, based on your SACR and tanks. 33(s f ) = t(d + 33) c For example, tanks you own might have a starting pressure of 2500 and and ending pressure of 500, you might have a c (SACR) of 15.2. You can then nd possible combinations of time and depth which you can comfortably dive. Write two print statements that shows how long one can dive at 60 feet and 70 feet. 5. Force on a Sail. How much force is on a sail? A sail moves a boat by transferring force to its mountings. The sail in the front (the jib) of a typical fore-and-aft rigged sailboat hangs from a stay. The sail in the back (the main) hangs from the mast. The forces on the stay (or mast) and sheets move the boat. The sheets are attached to the clew of the sail. The force on a sail, f, is based on sail area, a (in square feet) and wind speed, vw (in miles per hour). f = w2 0.004 a For a small racing dinghy, the smaller sail in the front might have 61 square feet of surface. The larger, mail sail, might have 114 square feet. Write a print statement to gure the force generated by a 61 square foot sail in 15 miles an hour of wind. 6. Craps Odds. What are the odds of winning on the rst throw of the dice? There are 36 possible rolls on 2 dice that add up to values from 2 to 12. There is just 1 way to roll a 2, 6 ways to roll a 7, and 1 way to roll a 12. Well take this as given until a later exercise where we have enough Python to generate this information. Without spending a lot of time on probability theory, there are two basic rules well use time and again. If any one of multiple alternate conditions needs to be true, usually expressed as or, we add the probabilities. When there are several conditions that must all be true, usually expressed as and, we multiply the probabilities. Rolling a 3, for instance, is rolling a 1-2 or rolling a 2-1. We add the probabilities: 1/36 + 1/36 = 2/36 = 1/18. On a come out roll, we win immediately if 7 or 11 is rolled. There are two ways to roll 11 (2/36) or 6 ways to roll 7 (6/36). Write a print statement to print the odds of winning on the come out roll. This means rolling 7 or rolling 11. Express this as a fraction, not as a decimal number; that means adding up the numerator of each number and leaving the denominator as 36. 7. Roulette Odds. How close are payouts and the odds? An American (double zero) roulette wheel has numbers 1-36, 0 and 00. 18 of the 36 numbers are red, 18 are black and the zeroes are green. The odds of spinning red, then are 18/38. The odds of zero or double zero are 2/36. 5.5. Expression Exercises 59
Red pays 2 to 1, the real odds are 38/18. Write a print statement that shows the dierence between the pay out and the real odds. You can place a bet on 0, 00, 1, 2 and 3. This bet pays 6 to 1. The real odds are 5/36. Write a print statement that shows the dierence between the pay out and the real odds.
5.6 Expression Style Notes

Spaces are used sparingly in expressions. Spaces are never used between a function name and the ()s that surround the arguments. It is considered poor form to write:
int (22.0/7)
The preferred form is the following:

int(22.0/7)
A long expression may be broken up with spaces to enhance readability. For example, the following separates the multiplication part of the expression from the addition part with a few wisely-chosen spaces.
b**2 - 4*a*c
60
CHAPTER
SIX
ADVANCED EXPRESSIONS
The math and random Modules, Bit-Level Operations, Division
This chapter covers some more advanced topics. The math Module cover the math module. The The random Module covers elements of the random module. Division Operators covers the important distinction between the division operators. We also provide some supplemental information that is more specialized. Bit Manipulation Operators covers some additional bitddling operators that work on the basic numeric types. Expression Style Notes has some notes on style.
6.1 Using Modules

A Python module extends the Python execution environment by adding new classes, functions and helpful constants. We tell the Python interpreter to fetch a module with a variation on the import statement. There are several variations on import, which well cover in depth in Components, Modules and Packages. For now, well use the simple import:
import m
This will import module m. Only the modules name, m is made available. Every name inside the module m must be qualied by prepending the module name and a .. So if module m had a function called spam(), wed refer to it as m.spam(). There are dozens of standard Python modules. Well get to the most important ones in Components, Modules and Packages. For now, well focus on extending the math capabilities of the basic expressions weve looked so far.
6.2 The math Module

The math module is made available to your programs with:
import math
The math module contains a number of common trigonometric functions. acos(x ) Arc cosine of x ; result in radians.
61
asin(x ) arc sine of x ; result in radians. atan(x ) arc tangent of x ; result in radians. atan2(y, x ) y arc tangent of y x: arctan( x ); result in radians. cos(x ) cosine of x in radians. cosh(x ) hyperbolic cosine of x in radians. exp(x ) ex , inverse of log (x). hypot(x, y ) Euclidean distance, x2 + y 2 ; the length of the hypotenuse of a right triangle with height of :replaceable:y and length of x. log(x ) Natural logarithm (base e) of x. Inverse of exp(). n = eln n . log10(x ) natural logarithm (base 10) of x , inverse of 10** x. n = 10log n . pow(x, y ) xy . sin(x ) sine of x in radians. sinh(x ) hyperbolic sine of x in radians. sqrt(x ) square root of x. This version returns an error if you ask for sqrt(-1), even though Python understands complex and imaginary numbers. A second module, cmath, includes a version of sqrt() which correctly creates imaginary numbers. tan(x ) tangent of x in radians. tanh(x ) hyperbolic tangent of x in radians. Additionally, the following constants are also provided. math.pi the value of , 3.1415926535897931 math.e the value of e, 2.7182818284590451, used for the exp() and log() functions. Conversion between radians, r, and degrees, d, is based on the following denition: 360 degrees = 2 radians From that, we get the following relationships: d = r 180 d r 180 ,r = d= 180 The math module contains the following other functions for dealing with oating point numbers. 62 Chapter 6. Advanced Expressions
ceil(x ) Next larger whole number.

>>> import math >>> math.ceil(5.1) 6.0 >>> math.ceil(-5.1) -5.0
fabs(x ) Absolute value of the real x. floor(x ) Next smaller whole number.
>>> import math >>> math.floor(5.9) 5.0 >>> math.floor(-5.9) -6.0
fmod(x, y ) Floating point remainder after division of x y . This depends on the platform C library and may handle the signs dierently than the Python x % y.
>>> math.fmod( -22, 7 ) -1.0 >>> -22 % 7 6
modf(x ) Creates a tuple with the fractional and integer parts of x. Both results carry the sign of x so that x can be reconstructed by adding them. Well return to tuples in Tuples.
>>> math.modf( 123.456 ) (0.45600000000000307, 123.0)
frexp(x ) This function unwinds the usual base-2 oating point representation. A oating point number is m 2e , where m is always a fraction 1 2 m 1, and e is an integer. This function returns a tuple with m and e. The inverse is ldexp(m,e). ldexp(m, e) Calculat m 2e , the inverse of frexp(x).
6.3 The random Module

The random module contains a large number of functions for working with distributions of random numbers. There are numerous functions available, but the later exercises will only use these functions. The random module is made available to your program with:
import random
Here are the denitions of some commonly-used functions.
6.3. The random Module
63
choice(sequence) Chooses a random value from the sequence sequence.

>>> import random >>> random.choice( ['red', 'black', 'green'] ) 'red'
random() A random oating point number, r, such that 0 r < 1.0. randrange([start], stop, [step] ) Choose a random element from range( start, stop, step ). randrange(6) returns a number, r, such that 0 r < 6. There are 6 values between 0 and 5. randrange(1,7) returns a number, r, such that 1 r < 7. There are 6 values between 1 and 6. randrange(10,100,5) returns a number, such that 10 5k < 100. for some integer value of k. These are values 10, 15, 20, ..., 95. randint(a, b) Choose a random number, r, such that a r b. Unlike randrange(), this function includes both end-point values. uniform(a, b) Returns a random oating point number, r, such that a r < b. The randrange() has two optional values, making it particularly exible. Heres an example of some of the alternatives.
demorandom.py
#!/usr/bin/env python import random # Simple Range 0 <= r < 6 print random.randrange(6), random.randrange(6) # More complex range 1 <= r < 7 print random.randrange(1,7), random.randrange(1,7) # Really complex range of even numbers between 2 and 36 print random.randrange(2,37,2) # Odd numbers from 1 to 35 print random.randrange(1,36,2)
This demonstrates a number of ways of generating random numbers. It uses the basic random.randrange() with a variety of dierent kinds of arguments.
6.4 Advanced Expression Exercises

1. Evaluate These Expressions. The following expressions are somewhat more complex, and use functions from the math module. math.sqrt( 40.0/3.0 - math.sqrt(12.0) ) 6.0/5.0*( (math.sqrt(5)+1) / 2 )**2 math.log( 2198 ) / math.sqrt( 6 )
64
Chapter 6. Advanced Expressions
2. Run demorandom.py. Run the demorandom.py script several times and save the results. Then add the following statement to the script and run it again several times. What happens when we set an explicit seed?
#!/usr/bin/env python import random random.seed(1) ...everything else the same
Try the following variation, and see what it does.

#!/usr/bin/env python import random, time random.seed(time.clock()) ...everything else the same
3. Wind Chill. Wind chill is used by meteorologists to describe the eect of cold and wind combined. Given the wind speed in miles per hour, V, and the temperature in F, T, the Wind Chill, w, is given by the formula below. Wind Chill, new model 35.74 + 0.6215 T 35.75 (V 0.16 ) + 0.4275 T (V 0.16 ) Wind Chill, old model 0.081 (3.71 V + 5.81 0.25 V ) (T 91.4) + 91.4
Wind speeds are for 0 to 40 mph, above 40, the dierence in wind speed doesnt have much practical impact on how cold you feel. Write a print statement to compute the wind chill felt when it is -2 F and the wind is blowing 15 miles per hour. 4. How Much Does The Atmosphere Weigh? Part 1 From Slicing Pizzas, Racing Turtles, and Further Adventures in Applied Mathematics, [Banks02]. Pressure is measured in Newtons, N, kg m/sec2 . Air Pressure is is measured in Newtons of force per square meter, N/m2 . Air Pressure (at sea level) P0 . This is the long-term average. P0 = 1.01325 105 Acceleration is measured in m/sec2 . Gravity acceleration (at sea level) g. g = 9.82 We can use g to get the kg of mass from the force of air pressure P0 . Apply the acceleration of gravity (in m/sec2 ) to the air pressure (in kg m/sec2 ). This result is mass of the atmosphere in kilograms per square meter (kg/m2 ). Mm2 = P0 g Given the mass of air per square meter, we need to know how many square meters of surface to apply this mass to. Radius of Earth R in meters, m. This is an average radius; our planet isnt a perfect sphere. R = 6.37 106
6.4. Advanced Expression Exercises
65
The area of a Sphere. A = 4r2 Mass of atmosphere (in Kg) is the weight per square meter, times the number of square meters. Ma = P0 g A Check: somewhere around 1018 kg. 5. How Much Does The Atmosphere Weigh? Part 2. From Slicing Pizzas, Racing Turtles, and Further Adventures in Applied Mathematics, [Banks02]. The exercise How Much Does the Atmosphere Weigh, Part 1 assumes the earth to be an entirely at sphere. The averge height of the land is actually 840m. We can use the ideal gas law to compute the pressure at this elevation and rene the number a little further. Pressure at a given elevation P = P0 e RT z Molecular weight of air m = 28.96 103 kg/mol. m = 28.96 103 Gas constant R, in joule/(K mol). R = 8.314 Gravity g, in m/sec2 . g = 9.82 Temperature T, in K based on temperature C, in C. Well just assume that C is 15 C. T = 273 + C Elevation z, in meters, m. z = 840 This pressure can be used for the air over land, and the pressure computed in How Much Does the Atmosphere Weigh, Part 1 can be used for the air over the oceans. How much land has this reduced pressure? Reference material gives the following areas in m2 , square meters. ocean area: Ao = 3.61 1014 land area: Al = 1.49 1014 Weight of Atmosphere, adjusted for land elevation Ml = P0 g A0 + P g Al
mg
6.5 Bit Manipulation Operators

Weve already seen the usual math operators: +, -, *, /, %, **; as well as the abs() and pow() functions. There are several other operators available to us. Principally, these are for manipulating the individual bits of an integer value. Well look at ~, &, ^, |, << and >>. The unary ~ operator ops all the bits in a plain or long integer. 1s become 0s and 0s become 1s. Since most hardware uses a technique called 2s complement, this is mathematically equivalent to adding 1 and switching the numbers sign. 66 Chapter 6. Advanced Expressions
>>> print ~0x12345678 -305419897
There are binary bit manipulation operators, also. These perform simple Boolean operations on all bits of the integer at once. The binary & operator returns a 1-bit if the two input bits are both 1.
>>> print 0&0, 1&0, 1&1, 0&1 0 0 1 0
Heres the same kind of example, combining sequences of bits. This takes a bit of conversion to base 2 to understand whats going on.
>>> print 3&5 1
The number 3, in base 2, is 0011. The number 5 is 0101. Lets match up the bits from left to right:
0 0 1 1 & 0 1 0 1 ------0 0 0 1
The binary ^ operator returns a 1-bit if one of the two inputs are 1 but not both. This is sometimes called the exclusive or.
>>> print 3^5 6
Lets look at the individual bits

0 0 1 1 ^ 0 1 0 1 ------0 1 1 0
Which is the binary representation of the number 6. The binary | operator returns a 1-bit if either of the two inputs is 1. This is sometimes called the inclusive or. Sometimes this is written and/or.
>>> print 3|5 7
Lets look at the individual bits.

0 0 1 1 | 0 1 0 1 ------0 1 1 1
Which is the binary representation of the number 7. There are also bit shifting operations. These are mathematically equivalent to multiplying and dividing by powers of two. Often, machine hardware can execute these operations faster than the equivalent multiply or divide.
6.5. Bit Manipulation Operators
67
The << is the left-shift operator. The left argument is the bit pattern to be shifted, the right argument is the number of bits.
>>> print 0xA << 2 40
0xA is hexadecimal; the bits are 1-0-1-0. This is 10 in decimal. When we shift this two bits to the left, its like multiplying by 4. We get bits of 1-0-1-0-0-0. This is 40 in decimal. The >> is the right-shift operator. The left argument is the bit pattern to be shifted, the right argument is the number of bits. Python always behaves as though it is running on a 2s complement computer. The left-most bit is always the sign bit, so sign bits are shifted in.
>>> print 80 >> 3 10
The number 80, with bits of 1-0-1-0-0-0-0, shifted right 3 bits, yields bits of 1-0-1-0, which is 10 in decimal. There are some other operators available, but, strictly speaking, theyre not arithmetic operators, theyre logic operations. Well return to them in Truth, Comparison and Conditional Processing.
6.6 Division Operators

In general, the data type of an expresion depends on the types of the arguments. This rule meets our expectations for most operators: when we add two integers, the result should be an integer. However, this doesnt work out well for division because there are two dierent expectations. Sometimes we expect division to create precise answers, usually the oating-point equivalents of fractions. Other times, we want a rounded-down integer result. The classical Python denition of / followed the pattern for other operators: the results depend entirely on the arguments. 685/252 was 2 because both arguments where integers. However, 685./252. was 2.7182539682539684 because the arguments were oating point. This denition often caused problems for applications where data types were used that the author hadnt expected. For example, a simple program doing Celsius to Fahrenheit conversions will produce dierent answers depending on the input. If one user provides 18 and another provides 18.0, the answers were dierent, even though all of the inputs all had the equal numeric values.
>>> 18*9/5+32 64 >>> 18.0*9/5+32 64.400000000000006 >>> 18 == 18.0 True
This unexpected inaccuracy was generally due to the casual use of integers where oating-point numbers were more appropriate. (This can also occur using integers where complex numbers were implictly expected.) An explicit conversion function (like float()) can help prevent this. The idea, however, is for Python be a simple and sparse language, without a dense clutter of conversions to cover the rare case of an unexpected data type. Starting with Python 2.2, a new division operator was added to clarify what is required. There are two division operators: / and //. The / operator should return oating-point results; the // operator will always return rounded-down results.
68
In Python 2.5 and 2.6, the / operator can either use classical or old rules (results depend on the values) or it can use the new rule (result is oating-point.) In Python 3.x, this transitional meaning of / goes away and it always produces a oating-point result. Important: Python 3 In Python 3, the / operator will always produces a oating-point result. The // operator will continue to produce an integer result. To help with the transition, two tools were made available. This gives programmers a way to keep older applications running; it also gives them a way to explicitly declare that their program uses the newer operator denition. There are two parts to this: a program statememt that can be placed in a program, as well as command-line options that can be used when starting the Python interpreter. Program Statements. To ease the transition from older to newer language features, there is a __future__ module available. This module includes a division denition that changes the denition of the / operator from classical to future. You can include the following import statement to state that your program depends on the future denition of division. Well look at the import statement in depth in Components, Modules and Packages.
from __future__ import division print 18*9/5+32 print 18*9//5+32
This produces the following output. The rst line shows the new use of the / operator to produce oating point results, even if both arguments are integers. The second line shows the // operator, which produces rounded-down results.
64.4 64
The from __future__ statement will set the expectation that your script uses the new-style oatingpoint division operator. This allows you to start writing programs with version 2.6 that will work correctly with all future versions. By version 3.0, this import statement will no longer be necessary, and these will have to be removed from the few modules that used them. Command Line Options. Another tool to ease the transition are command-line options used when running the Python interpreter. This can force old-style interpretation of the / operator or to warn about old-style use of the / operator between integers. It can also force new-style use of the / operator and report on all potentially incorrect uses of the / operator. The Python interpreter command-line option of -Q will force the / operator to be treated classically (old), or with the future (new) semantics. If you run Python with -Qold , the / operators result depends on the arguments. If you run Python with -Qnew, the / operators result will be oating point. In either case, the // operator returns a rounded-down integer result. You can use -Qold to force old modules and programs to work with version 2.2 and higher. When Python 3.0 is released, however, this transition will no longer be supported; by that time you should have xed your programs and modules. To make xing easier, the -Q command-line option can take two other values: warn and warnall . If you use -Qwarn, then the / operator applied to integer arguments will generate a run-time warning. This will allow you to nd and x situations where the // operator might be more appropriate. If you use -Qwarnall, then all instances of the / operator generate a warning; this will give you a close look at your programs. You can include the command line option when you run the Python interpreter. For Linux and MacOS users, you can also put this on the #! line at the beginning of your script le.
6.6. Division Operators
69
#!/usr/local/bin/python -Qnew
70
CHAPTER
SEVEN
VARIABLES, ASSIGNMENT AND INPUT

The = , augmented = and del Statements
Variables hold the state of our program. In Variables well introduce variables, then in The Assignment Statement well cover the basic assignment statement for changing the value of a variable. This is followed by an exercise section that refers back to exercises from Simple Numeric Expressions and Output . In Input Functions we introduce some primitive interactive input functions that are built-in. This is followed by some simple exercises that build on those from section The Assignment Statement . Well cover the multiple assignment statement in Multiple Assignment Statement . Well round on this section with the del statement, for removing variables in The del Statement .
7.1 Variables
As a procedural program makes progress through the steps from launch to completion, it does so by undergoing changes of state. The state of our program as a whole is the state of all of the programs variables. When one variable changes, the overall state has changed. Variables are the names your program assigns to the results of an expression. Every variable is created with an initial value. Variables will change to identify new objects and the objects identied by a variable can change their internal state. These three kinds of state changes (variable creation, object assignment, object change) happen as inputs are accepted and our program evaluates expressions. Eventually the state of the variables indicates that we are done, and our program can exit. A Python variable name must be at least one letter, and can have a string of numbers, letters and _s to any length. Names that start with _ or __ have special signicance. Names that begin with _ are typically private to a module or class. Well return to this notion of privacy in Classes and Modules. Names that begin with __ are part of the way the Python interpreter is built. Example variable names:
a pi aVeryLongName a_name __str__ _hidden
Tip: Tracing Execution
71
We can trace the execution of a program by simply following the changes of value of all the variables in the program. For programming newbies, it helps to create a list of variables and write down their changes when studying a program. Well show an example in the next section. Python creates new objects as the result of evaluating an expression. Python assigns these objects to new variables with an assignment statement. Python removes variables with a del statement. The underlying object is later garbage-collected when there are no more variables referring to the object. Some Consequences. A Python variable is little more than a name which refers to an object. The central issue is to recognize that the underlying object is the essential part of our program; a variable name is just a meaningful label. This has a number of important consequences. One consequence of a variable being simply a label is that any number of variables can refer to the same object. In other languages (C, C++, Java) there are two kinds of values: primitive and objects, and there are distinct rules for handling the two kinds of values. In Python, every variable is a simple reference to an underlying object. When talking about simple immutable objects, like the number 3, multiple variables referring to a common object is functionally equivalent to having a distinct copy of a primitive value. When talking about mutable objects, like lists, mappings, or complex objects, distinct variable references can change the state of the common object. Another consequences is that the Python object fully denes its own type. The objects type denes the representation, the range of values and the allowed operations on the object. The type is established when the object is created. For example, oating point addition and long integer objects have dierent representations, operations of adding these kinds of numbers are dierent, the objects created by addition are of distinct types. Python uses the type information to choose which addition operation to perform on two values. In the case of an expression with mixed types Python uses the type information to coerce one or both values to a common type. This also means the casting an object to match the declared type of a variable isnt meaningful in Python. You dont use C++ or Java-style casting. Weve already worked with the four numeric types: plain integers, long integers, oating point numbers and complex numbers. Weve touched on the string type, also. There are several other built-in types that we will look at in detail in Data Structures. Plus, we can use class denitions to dene new types to Python, something well look at in Data + Processing = Objects. We commonly say that a static language associates the type information with the variable. Only values of a certain type can be assigned to a given variable. Python, in contrast, is a dynamic language; a variable is just a label or tag attached to the object. Any variable can be associated with an object of any type. The nal consequence of variables referring to objects is that a variables scope can be independent of the object itself. This means that variables which are in distinct namespaces can refer to the same object. When a function completes execution and the namespace is deleted, the variables are deleted, and the number of variables referring to an object is reduced. Additional variables may still refer to an object, meaning that the object will continue to exist. When only one variable refers to an object, then removing the last variable removes the last reference to the object, and the object can be removed from memory. Also note that expressions generally create new objects; if an object is not saved in a variable, it silently vanishes. We can safely ignore the results of a function. Scope and Namespaces. A Python variable is a name which refers to an object. To be useful, each variable must have a scope of visibility. The scope is dened as the set of statements that can make use of this variable. A variable with global scope can be referenced anywhere. On the other hand, a variable with local scope can only be referenced in a limited suite of statements. This notion of scope is essential to being able to keep a intellectual grip on a program. Programs of even moderate complexity need to keep pools of variables with separate scopes. This allows you to reuse variable names without risk of confusion from inadvertantly changing the value of a variable used elsewhere in a program.
72
Chapter 7. Variables, Assignment and Input
Python collects variables into pools called namespaces . A new namespace is created as part of evaluating the body of a function or module, or creating a new object. Additionally, there is one global namespace. This means that each variable (and the state that it implies) is isolated to the execution of a single function or module. By separating all locally scoped variables into separate namespaces, we dont have an endless clutter of global variables. In the rare case that you need a global variable, the global statement is available to assign a variable to the global namespace. When we introduce functions in Functions, classes in Classes and modules in Components, Modules and Packages, well revisit this namespace technique for managing scope. In particular, see Functions and Namespaces for a digression on this.
7.2 The Assignment Statement

Assignment is fundamental to Python; it is how the objects created by an expression are preserved. Well look at the basic assignment statement, plus the augmented assignment statement. Later, in Multiple Assignment Statement , well look at multiple assignment.
7.2.1 Basic Assignment

We create and change variables primarily with the assignment statement. This statement provides an expression and a variable name which will be used to label the value of the expression.
variable = expression
Heres a short script that contains some examples of assignment statements.
example3.py
#!/usr/bin/env python # Computer the value of a block of stock shares= 150 price= 3 + 5.0/8.0 value= shares * price print value
1. We have an object, the number 150, which we assign to the variable shares. 2. We have an expression 3+5.0/8.0, which creates a oating-point number, which we save in the variable price. 3. We have another expression, shares * price, which creates a oating-point number; we save this in value so that we can print it. This script created three new variables. Since this le is new, well need to do the chmod +x example3.py once, after we create this le. Then, when we run this progam, we see the following.
$ ./example3.py 543.75
7.2. The Assignment Statement
73
7.2.2 Augmented Assignment

Any of the usual arithmetic operations can be combined with assignment to create an augmented assignment statement. For example, look at this augmented assignment statement:
a += v
This statement is a shorthand that means the same thing as the following:
a = a + v
Heres a larger example
portfolio.py
#!/usr/bin/env python # Total value of a portfolio made up of two blocks of stock portfolio = 0 portfolio += 150 * 2 + 1/4.0 portfolio += 75 * 1 + 7/8.0 print portfolio
First, well do the chmod +x portfolio.py on this le. Then, when we run this progam, we see the following.
$ ./portfolio.py 376.125
The other basic math operations can be used similarly, although the purpose gets obscure for some operations. These include -=, *=, /=, %=, &=, ^=, |=, <<= and >>=. Heres a lengthy example. This is an extension of Craps Odds in Numeric Types and Expressions. In craps, the rst roll of the dice is called the come out roll. This roll can be won immediately if the number is 7 or 11. It can be lost immediately if the number is 2, 3 or 12. All of the remaining numbers will establish a point and the game continues.
craps.py
#!/usr/bin/env python # Compute the odds of winning on the first roll win = 0 win += 6/36.0 # ways to roll a 7 win += 2/36.0 # ways to roll an 11 print "first roll win", win # Compute the odds of losing on the first roll lose = 0 lose += 1/36.0 # ways to roll 2 lose += 2/36.0 # ways to roll 3 lose += 1/36.0 # ways to roll 12 print "first roll lose", lose # Compute the odds of rolling a point number (4, 5, 6, 8, 9 or 10)
74
point point point print
= 1 # odds must total to 1 -= win # remove odds of winning -= lose # remove odds of losting "first roll establishes a point", point
Theres a 22.2% chance of winning, and a 11.1% chance of losing. Whats the chance of establishing a point? One way is to gure that its whats left after winning or loosing. The total of all probabilities always add to 1. Subtract the odds of winning and the odds of losing and whats left is the odds of setting a point. Heres another way to gure the odds of rolling 4, 5, 6, 8, 9 or 10.
point point point point print = 0 += 2*3/36.0 # ways to roll 4 or 10 += 2*4/36.0 # ways to roll 5 or 9 += 2*5/36.0 # ways to roll 6 or 8 point
By the way, you can add the statement print win + lose + point to conrm that these odds all add to 1. This means that we have dened all possible outcomes for the come out roll in craps. Tip: Tracing Execution We can trace the execution of a program by simply following the changes of value of all the variables in the program. We can step through the planned execution of our Python source statements, writing down the variables and their values on a sheet of paper. From this, we can see the state of our calculation evolve. When we encounter an assignment statement, we look on our paper for the variable. If we nd the variable, we put a line through the old value and write down the new value. If we dont nd the variable, we add it to our page with the initial value. Heres our example from craps.py script through the rst part of the script. The win variable was created and set to 0, then the value was replaced with 0.16, and then replaced with 0.22. The lose variable was then created and set to 0. This is what our trace looks like so far. win: lose: 0.0 0 0.16 0.22
Heres our example when craps.py script is nished. We changed the variable lose several times. We also added and changed the variable point. win: lose: point: 0.0 0.0 1.0 0.16 0.027 0.77 0.22 0.083 0.66 0.111
We can use this trace technique to understand what a program means and how it proceeds from its initial state to its nal state. As with many things Python, there is some additional subtlety to assignment, but well cover those topics later. For example, multiple-assignment statement is something well look into in more deeply in Tuples.
7.3 Input Functions

Python provides two simplistic built-in functions to accept input and set the value of variables. These are not really suitable for a complete application, but will do for our initial explorations.
7.3. Input Functions
75
Typically, interactive programs which run on a desktop use a complete graphic user interface (GUI), often written with the Tkinter module or the pyGTK module. Interactive programs which run over the Internet use HTML forms. The primitive interactions were showing with input() and raw_input() are only suitable for very simple programs. Important: Python 3.x In Python 3, the raw_input() function will be renamed to input(). The Python 2 input() function will be removed. Its that useless. Note that some IDEs buer the programs output, making these functions appear to misbehave. For example, if you use Komodo, youll need to use the Run in a New Console option. If you use BBEdit, youll have to use the Run in Terminal option. You can enhance these functions somewhat by including the statement import readline. This module silently and automatically enhances these input functions to give the user the ability to scroll backwards and reuse previous inputs. You can also import rlcompleter. This module allows you to dene sophisticated keyword auto-completion for these functions.
7.3.1 The raw_input() Function

The rst way to get interactive input is the raw_input() function. This function accepts a string parameter, which is the users prompt, written to standard output. The next line available on standard input is returned as the value of the function. raw_input([prompt] ) If a prompt is present, it is written to sys.stdout. Input is read from sys.stdin and returned as a string. The raw_input() function reads from a le often called sys.stdin. When running from the command-line, this will be the keyboard, and what you type will be echoed in the command window or Terminal window. If you try, however, to run these examples from Textpad, youll see that Textpad doesnt have any place for you to type any input. In BBEdit, youll need to use the Run In Terminal item in the #! menu. Heres an example script that uses raw_input().
rawdemo.py
#!/usr/bin/env python # show how raw_input works a= raw_input( "yes?" ) print "you said", a
When we run this script from the shell prompt, it looks like the following.
MacBook-3:Examples slott$ python rawdemo.py yes?why not? you said why not?
1. This program begins by evaluating the raw_input() function. When raw_input() is applied to the parameter of "yes?", it writes the prompt on standard output, and waits for a line of input.
76
(a) We entered why not?. (b) Once that line was complete, the input string is returned as the value of the function. (c) The raw_input() functions value was assigned to the variable a. 2. The second statement printed that variable along with some text. If we want numeric input, we must convert the resulting string to a number.
stock.py
#!/usr/bin/env python # Compute the value of a block of stock shares = int( raw_input("shares: ") ) price = float( raw_input("dollars: ") ) price += float( raw_input("eights: ") )/8.0 print "value", shares * price
Well chmod +x stock.py this program; then we can run it as many times as we like to get results.
MacBook-3:Examples slott$ ./stock.py shares: 150 dollars: 24 eights: 3 value 3656.25
The raw_input() mechanism is very limited. If the string returned by raw_input() is not suitable for use by int(), an exception is raised and the program stops running. Well cover exception handling in detail in Exceptions. As a teaser, heres what it looks like.
MacBook-5:Examples slott$ python stock.py shares: a bunch Traceback (most recent call last): File "stock.py", line 3, in <module> shares = int( raw_input("shares: ") ) ValueError: invalid literal for int() with base 10: 'a bunch'
7.3.2 The input() Function

In addition to the raw_input() function, which returns the exact string of characters, there is the input() function. This applies the eval() function to the input, which will typically convert numeric input to the appropriate objects. Important: Python 3 This function will be removed. Its best not to make use of it. The value of the input() function is eval( raw_input( prompt ) ).
7.3. Input Functions
77
7.4 Multiple Assignment Statement

The basic assignment statement can do more than assign the result of a single expression to a single variable. The assignment satement can also assign multiple variables at one time. The essential rule is that the left and right side must have the same number of elements. For example, the following script has several examples of multiple assignment.
line.py
#!/usr/bin/env python # Compute line between two points. x1,y1 = 2,3 # point one x2,y2 = 6,8 # point two m,b = float(y1-y2)/(x1-x2), y1-float(y1-y2)/(x1-x2)*x1 print "y=",m,"*x+",b
When we run this program, we get the following output

MacBook-3:Examples slott$ ./line.py y = 1.25 *x+ 0.5
We set variables x1, y1, x2 and y2. Then we computed m and b from those four variables. Then we printed the m and b. The basic rule is that Python evaluates the entire right-hand side of the = statement. Then it matches values with destinations on the left-hand side. If the lists are dierent lengths, an exception is raised and the program stops. Because of the complete evaluation of the right-hand side, the following construct works nicely to swap to variables. This is often quite a bit more complicated in other languages.
a,b = 1,4 b,a = a,b print a,b
Well return to this in Tuples, where well see additional uses for this feature.
7.5 The del Statement

An assignment statement creates or locates a variable and then assigns a new object to the variable. This change in state is how our program advances from beginning to termination. Python also provides a mechanism for removing variables, the del statement. The del statement looks like this:
del object , ...
Each object is any kind of Python object. Usually these are variables, but they can be functions, modules or classes.
78
The del statement works by unbinding the name, removing it from the set of names known to the Python interpreter. If this variable was the last remaining reference to an object, the object will be removed from memory. If, on the other hand, other variables still refer to this object, the object wont be deleted. C++ Comparison Programmers familiar with C++ will be pleased to note that memory management is silent and automatic, making programs much more reliable with much less eort. This removal of objects is called garbage collection, something that can be rather diicult to manage in larger applications. When garbage collection is done incorrectly, it can lead to dangling references: a variable that refers to an object that was deleted prematurely. Poorly designed garbage collection can also lead to memory leaks, where unreferenced objects are not properly removed from memory. Because of the automated garbage collection in Python, it suers from none of these memory management problems. The del statement is typically used only in rare, specialized cases. Ordinary namespace management and garbage collection are generally suicient for most purposes.
7.6 Interactive Mode Revisited

When we rst looked at interactive Python in Command-Line Interaction we noted that Python executes assignment statements silently, but prints the results of an expression statement. Consider the following example.
>>> pi=355/113.0 >>> area=pi*2.2**2 >>> area 15.205309734513278
The rst two inputs are complete statements, so there is no response. The third input is just an expression, so there is a response. It isnt obvious, but the value assigned to pi isnt correct. Because we didnt see anything displayed, we didnt get any feedback from our computation of pi. Python, however, has a handy way to help us. When we type a simple expression in interactive Python, it secretly assigns the result to a temporary variable named _. This isnt a part of scripting, but is a handy feature of an interactive session. This comes in handy when exploring something rather complex. Consider this interactive session. We evaluate a couple of expressions, each of which is implicitly assigned to _. We can then save the value of _ in a second variable with an easier-to-remember name, like pi or area.
>>> 335/113.0 2.9646017699115044 >>> 355/113.0 3.1415929203539825 >>> pi=_ >>> pi*2.2**2 15.205309734513278 >>> area=_ >>> area 15.205309734513278
7.6. Interactive Mode Revisited
79
Note that we created a oating point object (2.964...), and Python secretly assigned this object to _. Then, we computed a new oating point object (3.141...), which Python assigned to _. What happened to the rst oat, 2.964...? Python garbage-collected this object, removing it from memory. The second oat that we created (3.141) was assigned to _. We then assigned it to pi, also, giving us two references to the object. When we computed another oating-point value (15.205...), this was assigned to _. Does this mean our second oat, 3.141... was garbage collected? No, it wasnt garbage collected; it was still referenced by the variable pi.
7.7 Variables, Assignment and Input Function Exercises

7.7.1 Variables and Assignment
1. Extend Previous Exercises. Rework the exercises in Numeric Types and Expressions. Each of the previous exercises can be rewritten to use variables instead of expressions using only constants. For example, if you want to tackle the Fahrenheit to Celsius problem, you might write something like this:
#!/usr/bib/env python # Convert 8 C to F C=8 F=32+C*(9./5.) print "celsius",C,"fahrenheit",F
Youll want to rewrite these exercises using variables to get ready to add input functions. 2. State Change. Is it true that all programs simply establish a state? It can argued that a controller for a device (like a toaster or a cruise control) simply maintains a steady state. The notion of state change as a program moves toward completion doesnt apply because the software is always on. Is this the case, or does the software controlling a device have internal state changes? For example, consider a toaster with a thermostat, a browness sensor and a single heating element. What are the inputs? What are the outputs? Are there internal states while the toaster is making toast?
7.7.2 Input Functions

Refer back to the exercises in Numeric Types and Expressions for formulas and other details. Each of these can be rewritten to use variables and an input conversion. For example, if you want to tackle the Fahrenheit to Celsius problem, you might write something like this:
C = raw_input('Celsius: ') F = 32+C*(9./5.) print "celsius",C,"fahrenheit",F
1. Stock Value. Input the number of shares, dollar price and number of 8ths. From these three inputs, compute the total dollar value of the block of stock. 2. Convert from |deg| C to |deg| F. Write a short program that will input C and output F. A second program will input F and output C.
80
3. Periodic Payment. Input the principal, annual percentage rate and number of payments. Compute the monthly payment. Be sure to divide rate by 12 and multiple payments by 12. 4. Surface Air Consumption Rate. Write a short program will input the starting pressure, nal pressure, time and maximum depth. Compute and print the SACR. A second program will input a SACR, starting pressure, nal pressure and depth. It will print the time at that depth, and the time at 10 feet more depth. 5. Wind Chill. Input a temperature and a wind speed. Output the wind chill. 6. Force from a Sail. Input the height of the sail and the length. The surface area is 1/2 h l. For a wind speed of 25 MPH, compute the force on the sail. Small boat sails are 25-35 feet high and 6-10 feet long.
7.8 Variables and Assignment Style Notes

Spaces are used sparingly in Python. It is common to put spaces around the assignment operator. The recommended style is
c = (f-32)*5/9
Do not take great pains to line up assignment operators vertically. The following has too much space, and is hard to read, even though it is fussily aligned.
a b aVeryLongVariable d = = = = 12 a*math.log(a) 26 13
This is considered poor form because Python takes a lot of its look from natural languages and mathematics. This kind of horizontal whitespace is hard to follow: it can get diicult to be sure which expression lines up with which variable. Python programs are meant to be reasonably compact, more like reading a short narrative paragraph or short mathematical formula than reading a page-sized UML diagram. Variable names are often given as mixedCase; variable names typically begin with lower-case letters. The lower_case_with_underscores style is also used, but is less popular. In addition, the following special forms using leading or trailing underscores are recognized: single_trailing_underscore_: used to avoid conicts with Python keywords. For example: print_ = 42 __double_leading_and_trailing_underscore__: used for special objects or attributes, e.g. __init__, __dict__ or __file__. These names are reserved; do not use names like these in your programs unless you specically mean a particular built-in feature of Python. _single_underscore: means that the variable is private.
7.8. Variables and Assignment Style Notes
81
82
CHAPTER
EIGHT
TRUTH, COMPARISON AND CONDITIONAL PROCESSING

Truth, Comparison and the if Statement, pass and assert Statements.
This section leading up to the for and while statements, as well as the break and continue statements. The elements of Python weve seen so far give us some powerful capabilities. We can write programs that implement a wide variety of requirements. State change is not always as simple as the examples weve seen in Variables, Assignment and Input . When we run a script, all of the statements are executed unconditionally. Our programs cant handle alternatives or conditions. Python provides decision-making mechanisms similar to other programming languages. In Truth and Logic well look at truth, logic and the logic operators. The exercises that follow examine some subtleties of Pythons evaluation rules. In Comparisons well look at the comparison operators. Then, Conditional Processing: the if Statement describes the if statement. In The assert Statement well introduce a handy diagnostic tool, the assert statement. In the next chapter, Loops and Iterative Processing, well look at looping constructs.
8.1 Truth and Logic

Many times the exact change in state that our program needs to make depends on a condition. A condition is a Boolean expression; an expression that is either True or False. Generally conditions are on comparisons among variables using the comparison operations. Well look at the essential denitions of truth, the logic operations and the comparison operations. This will allow us to build conditions.
8.1.1 Truth
Python represents truth and falsity in a variety of ways. False. Also 0, the special value None, zero-length strings "", zero-length lists [], zero-length tuples (), empty mappings {} are all treated as False. True. Anything else that is not equivalent to False. We try to avoid depending on relatively obscure rules for determining True vs. False. We prefer to use the two explicit keywords, True and False. Note that a previous version of Python didnt have the boolean literals, and some older open-source programs will dene these values. 83
Python provides a factory function to collapse these various forms of truth into one of the two explicit boolean objects. bool(object ) Returns True when the argument object is one the values equivalent to truth, False otherwise.
8.1.2 Logic
Python provides three basic logic operators that work on this Boolean domain. Note that this Boolean domain, with just two values, True and False, and these three operators form a complete algebraic system, sometimes called Boolean algebra, after the mathemetician George Boole. The operators supported by Python are not, and and or . We can fully dene these operators with rule statements or truth tables. This truth table shows the evaluation of not x.
print "x", "not x" print True, not True print False, not False
x True False
not x False True
This table shows the evaluation of x and y for all combination of True and False.
print print print print print "x", "y", "x and y" True, True, True and True True, False, True and False False, True, False and True False, False, False and False
x True True False False
y True False True False
x and y True False False False
An important feature of and is that it does not evaluate all of its parameters before it is applied. If the left-hand side is False or one of the equivalent values, the right-hand side is not evaluated, and the left-hand value is returned. Well look at some examples of this later. For now, you can try things like the following.
print False and 0 print 0 and False
This will show you that the rst false value is what Python returns for and. This table shows the evaluation of x or y for all combination of True and False. x True True False False y True False True False x or y True True True False
Parallel with the and operator, or does not evaluate the right-hand parameter if the left-hand side is True or one of the equivalent values.
84
Chapter 8. Truth, Comparison and Conditional Processing
As a nal note, and is a high priority operator (analogous to multiplication) and or is lower priority (analogous to addition). When evaluating expressions like a or b and c, the and operation is evaluated rst, followed by the or operation.
8.1.3 Exercises
1. Logic Short-Cuts. We have several versions of false: False, 0, None, '', (), [] and {}. Well cover all of the more advanced versions of false in Data Structures. For each of the following, work out the value according to the truth tables and the evaluation rules. Since each truth or false value is unique, we can see which part of the expression was evaluated. False and None 0 and None or () and [] True and None or () and [] 0 or None and () or [] True or None and () or [] 1 or None and 'a' or 'b'
8.2 Comparisons
Well look at the basic comparison operators. Well also look at the partial evaluation rules of the logic operators to show how we can build more useful expressions. Finally, well look at oating-point equality tests, which are sometimes done incorrectly.
8.2.1 Basic Comparisons

We compare values with the comparison operators. These correspond to the mathematical functions of <, , >, , = and =. Conditional expressions are often built using the Python comparison operators: <, <=, >, >=, == and != for less than, less than or equal to, greater than, greater than or equal to, equal to and not equal to.
>>> p1 = 22./7. >>> p2 = 355/113. >>> p1 3.1428571428571428 >>> p2 3.1415929203539825 >>> p1 < p2 False >>> p2 >= p2 True
When applying a comparison operator, we see a number of steps. 1. Evaluate both argument values. 2. Apply the comparison to create a boolean result. (a) Convert both parameters to the same type. Numbers are converted to progressively longer types: plain integer to long integer to oat to complex.
8.2. Comparisons
85
(b) Do the comparison. (c) Return True or False. We call out these three steps explicitly because there are some subtleties in comparison among unlike types of data; well come to this later when we cover sequences, mappings and classes in Data Structures. Generally, it doesnt make sense to compare unlike types of data. After all, you cant ask Which is larger, the Empire State Building or the color green? Comparisons can be combined in Python, unlike most other programming languages. We can ask: 0 <= a < 6 which has the usual mathematical meaning. Were not forced to use the longer form: 0 <= a and a < 6. This is useful when a is actually some complex expression that wed rather not repeat. Here is an example.
>>> 3 < p1 < 3.2 True >>> 3 < p1 and p1 < 3.2 True
Note that the preceding example had a mixture of integers and oating-point numbers. The integers were coerced to oating-point in order to evaluate the expressions.
8.2.2 Partial Evaluation

We can combine the logic operators, comparisons and math. This allows us to use comparisons and logic to prevent common mathematical blunders like attempting to divide by zero, or attempting to take the square root of a negative number. For example, lets start with this program that will gure the average of 95, 125 and 132.
sum = 95 + 125 + 132 count = 3 average = float(sum)/count print average
Initially, we set the variables sum and count . Then we compute the average using sum and count. Assume that the statement that computes the average (average=...) is part of a long and complex program. Sometimes that long program will try to compute the average of no numbers at all. This has the same eect as the following short example.
sum, count = 0, 0 average = float(sum)/count print average
In the rare case that we have no numbers to average we dont want to crash when we foolishly attempt to divide by zero. Wed prefer to have some more graceful behavior. Recall from Truth and Logic that the and operator doesnt evaluate the right-hand side unless the left-hand side is True. Stated the other way, the and operator only evaluates the right side if the left side is True. We can guard the division like this:
average = count != 0 and sum/count print average
86
This is an example that can simplify certain kinds of complex processing. If the count is non-zero, the left side is true and the right side must be checked. If the count is zero, the left side is False, the result of the complete and operation is False. This is a consequence of the meaning of the word and. The expression a and b means that a is true as well as b is true. If a is false, the value of b doesnt really matter, since the whole expression is clearly false. A similar analysis holds for the word or. The expression a or b means that one of the two is true; it also means that neither of the two is false. If a is true, then the value of b doesnt change the truth of the whole expression. The statement Its cold and rainy is completely false when it is warm; rain doesnt matter to falsifying the whole statement. Similarly, Im stopping for coee or a newspaper is true if Ive stopped for coee, irrespective of whether or not I got a newspaper.
8.2.3 Floating-Point Comparisons

Exact equality between oating-point numbers is a dangerous concept. After a lengthy computation, roundo errors in oating point numbers may have innitesimally small dierences. The answers are close enough to equal for all practical purposes, but every single one of the 64 bits may not be identical. The following technique is the appropriate way to do oating point comparisons.
abs(a-b)<0.0001
Rather than ask if the two oating point values are the same, we ask if theyre close enough to be considered the same. For example, run the following tiny program.
oatequal.py
#!/usr/bin/env python # Are two floating point values really completely equal? a,b = 1/3.0, .1/.3 print a,b,a==b print abs(a-b)<0.00001
When we run this program, we get the following output

$ python floatequal.py 0.333333333333 0.333333333333 False True
The two values appear the same when printed. Yet, on most platforms, the == test returns False. They are not precisely the same. This is a consequence of representing real numbers with only a nite amount of binary precision. Certain repeating decimals get truncated, and these truncation errors accumulate in our calculations. There are ways to avoid this problem; one part of this avoidance is to do the algebra necessary to postpone doing division operations. Division introduces the largest number erroneous bits onto the trailing edge of our numbers. The other part of avoiding the problem is never to compare oating point numbers for exact equality.
8.2. Comparisons
87
8.3 Conditional Processing: the if Statement

Many times the programs exact change in state depends on a condition. Conditional processing is done by setting statements apart in suites with conditions attached to the suites. The Python syntax for this is an if statement.
8.3.1 The if Statement

The basic form of an if statement provides a condition and a suite of statements that are executed when the condition is true. It looks like this:
if expression : suite
The suite is an indented block of statements. Any statement is allowed in the block, including indented if statements. You can use either tabs or spaces for indentation. The usual style is four spaces. This is our rst compound statement. See Python Syntax Rules for some additional guidance on syntax for compound statements. The if statement evaluates the condition expression rst. When the result is True, the suite of statements is executed. Otherwise the suite is skipped. For example, if two dice show a total of 7 or 11, the throw is a winner. In the following snippet, d1 and d2 are two dice values that range from 1 to 6.
if d1+d2 == 7 or d1+d2 == 11: print "winner", d1+d2
Here we have a typically complex expression. The or operator evaluates the left side rst. Python evaluates and applies the high-precendence arithmetic operator before the lower-precendence comparison operator. If the left side is true (d1 + d2 is 7), the or expression is true, and the suite is executed. If the left side is false, then the right side is evaluated. If it is true (d1 + d2 is 11), the or expression is true, and the suite is executed. Otherwise, the suite is skipped.
88
Python Syntax Rules Python syntax is very simple. Weve already seen how basic expressions and some simple statements are formatted. Here are some syntax rules and examples. Look at Syntax Formalities for an overview of the lexical rules. Compound statements, including if, while, for, have an indented suite of statements. You have a number of choices for indentation; you can use tab characters or spaces. While there is a lot of exibility, the most important thing is to be consistent. Further, the recommendation is to use spaces. Thats what well show. The generally accepted way to format Python code is to set your editor to replace tabs with 4 spaces. Well show an example with spaces, shown via . a=0 ifa==0: print"aiszero" else: print"aisnotzero" ifa%2==0: print"aiseven" else: print"aisodd" IDLE uses four spaces for indentation automatically. If youre using another editor, you can set it to use four spaces, also.
8.3.2 The elif Clause

Often there are several conditions that need to be handled. This is done by adding elif clauses. This is short for else-if. We can add an unlimited number of elif clauses. The elif clause has almost the same syntax as the if clause.
elif expression : suite
Here is a somewhat more complete rule for the come out roll in a game of craps:
result= None if d1+d2 == 7 or d1+d2 == 11: result= "winner" elif d1+d2 == 2 or d1+d2 == 3 or d1+d2 == 12: result= "loser" print result
First, we checked the condition for winning; if the rst condition is true, the rst suite is executed and the entire if statement is complete. If the rst condition is false, then the second condition is tested. If that condition is true, the second suite is executed, and the entire if statement is complete. If neither condition is true, the if statement has no eect.
8.3.3 The else Clause

Python also gives us the capability to put a catch-all suite at the end for all other conditions. This is done by adding an else clause. The else clause has the following syntax.
8.3. Conditional Processing: the if Statement
89
else: suite
Heres the complete come-out roll rule, assuming two values d1 and d2.
point= None if d1+d2 == 7 or d1+d2 == 11: print "winner" elif d1+d2 == 2 or d1+d2 == 3 or d1+d2 == 12: print "loser" else: point= d1+d2 print "point is", point
Here, we use the else: suite to handle all of the other possible rolls. There are six dierent values (4, 5, 6, 8, 9, or 10), a tedious typing exercise if done using or. We summarize this with the else: clause. While handy in one respect, this else: clause is also dangerous. By not explicitly stating the condition, it is possible to overlook simple logic errors. Consider the following complete if statement that checks for a winner on a eld bet. A eld bet wins on 2, 3, 4, 9, 10, 11 or 12. The payout odds are dierent on 2 and 12.
outcome= 0 if d1+d2 == 2 or outcome= 2 print "field elif d1+d2==4 or outcome= 1 print "field else: outcome= -1 print "field
d1+d2 == 12: pays 2:1" d1+d2==9 or d1+d2==10 or d1+d2==11: pays even money"
loses"
Here we test for 2 and 12 in the rst clause; we test for 4, 9, 10 and 11 in the second. Its not obvious that a roll of 3 is missing from the even money pay out. This fragment incorrectly treats 3, 5, 6, 7 and 8 alike in the else:. While the else: clause is used commonly as a catch-all, a more proper use for else: is to raise an exception because a condition was not matched by any of the if or elif clauses.
8.4 The pass Statement

The pass statement does nothing. Sometimes we need a placeholder to ll the syntactic requirements of a compound statement. We use the pass statement to ll in the required suite of statements. The syntax is trivial.
pass
Heres an example of using the pass statement.

if n%2 == 0: pass # Ignore even values else: count += 1 # Count the odd values
90
Yes, technically, we can invert the logic in the if-clause. However, sometimes it is more clear to provide the explicit do nothing than to determine the inverse of the condition in the if statement. As programs grow and evolve, having a pass statement can be a handy reminder of places where a program can be expanded. Also, when we come to class declarations in Data + Processing = Objects, well see one other use for the pass statement.
8.5 The assert Statement

An assertion is a condition that were claiming should be true at this point in the program. Typically, it summarizes the state of the programs variables. Assertions can help explain the relationships among variables, review what has happened so far in the program, and show that if statements and for or while loops have the desired eect. When a program is correct, all of the assertions are true no matter what inputs are provided. When a program has an error, at least one assertion winds up false for some combination of inputs. Python directly supports assertions through an assert statement. There are two forms:
assert condition assert condition , expression
If the condition is False, the program is in error; this statement raises an AssertionError exception. If the condition is True, the program is correct, this statement does nothing more. If the second form of the statement is used, and an expression is given, an exception is raised using the value of the expression. Well cover exceptions in detail in Exceptions. If the expression is a string, it becomes an the value associated with the AssertionError exception. Note: Additional Features There is an even more advanced feature of the assert statement. If the expression evaluates to a class, that class is used instead of AssertionError. This is not widely used, and depends on elements of the language we havent covered yet. Heres a typical example:
max= 0 if a < b: max= b if b < a: max= a assert (max == a or max == b) and max >= a and max >= b
If the assertion condition is true, the program continues. If the assertion condition is false, the program raises an AssertionError exception and stops, showing the line where the problem was found. Run this program with a equal to b and not equal to zero; it will raise the AssertionError exception. Clearly, the if statements dont set max to the largest of a and b when a = b . There is a problem in the if statements, and the presence of the problem is revealed by the assertion.
8.5. The assert Statement
91
8.6 The if-else Operator

There are situations where an expression involves a simple condition and a full-sized if statement is distracting syntatic overkill. Python has a handy logic operator that evalutes a condition, then returns either of two values depending on that condition. Ternary Operator Most arithmetic and logic operators have either one or two values. An operation that applies to a single value is called unary. For example -a and abs(b) are examples of unary operations: unary negation and unary absolute value. An operation that applies to two values is called binary. For example, a*b shows the binary multiplication operator. The if-else operator trinary (or ternary) It involves a conditional expression and two alternative expressions. Consequently, it doesnt use a single special character, but uses two keywords: if and else. Some folks will mistakenly call it the ternary operator as if this is the only possible ternary operator. The basic form of the operator is
expression if condition else expression
Python evaluates the condition in the middle rst. If the condition is True, then the left-hand expression is evaluated, and thats the value of the operation. If the condition is False, then the right-hand expression is evaluated, and thats the value of the operation. Note that the condition is always evaluated. Only one of the other two expressions is evaluated, making this a kind of short-cut operator like and and or. Here are a couple of examples.
average = sum/count if count != 0 else None oddSum = oddSum + ( n if n % 2 == 1 else 0 )
The intent is to have an English-like reading of the statement. The average is the sum divided by the count if the count is non-zero; else the average is None. The wordy alterative to the rst example is the following.
if count != 0: average= sum/count else: average= None
This seems like three extra lines of code to prevent an error in the rare situation of there being no values to average. Similarly, the wordy version of the second example is the following:
if n % 2 == 0: pass else: oddSum = oddSum + n
92
For this second example, the original statement registered our intent very clearly: we were summing the odd values. The long-winded if-statement tends to obscure our goal by making it just one branch of the if-statement.
8.7 Condition Exercises

1. Develop an or-guard. In the example above we showed the and-guard pattern:
average = count != 0 and float(sum)/count
Develop a similar technique using or. Compare this with the if-else operator. 2. Come Out Win. Assume d1 and d2 have the numbers on two dice. Assume this is the come out roll in Craps. Write the expression for winning (7 or 11). Write the expression for losing (2, 3 or 12). Write the expression for a point (4, 5, 6, 8, 9 or 10). 3. Field Win. Assume d1 and d2 have the numbers on 2 dice. The eld pays on 2, 3, 4, 9, 10, 11 or 12. Actually there are two conditions: 2 and 12 pay at one set of odds (2:1) and the other 5 numbers pay at even money. Write two two conditions under which the eld pays. 4. Hardways. Assume d1 and d2 have the numbers on 2 dice. A hardways proposition is 4, 6, 8, or 10 with both dice having the same value. Its the hard way to get the number. A hard 4, for instance is d1+d2 == 4 and d1 == d2. An easy 4 is d1+d2 == 4 and d1 != d2. You win a hardways bet if you get the number the hard way. You lose if you get the number the easy way or you get a seven. Write the winning and losing condition for one of the four hard ways bets. 5. Sort Three Numbers. This is an exercise in constructing if-statements. Using only simple variables and if statements, you should be able to get this to work; a loop is not needed. Given 3 numbers (X, Y, Z ), assign variables x, y, z so that x y z and x , y, and z are from X, Y, and Z. Use only a series of if-statements and assignment statements. Hint. You must dene the conditions under which you choose between x X, x Y or x Z. You will do a similar analysis for assigning values to y and z. Note that your analysis for setting y will depend on the value set for x; similarly, your analysis for setting z will depend on values set for x and y. 6. Come Out Roll. Accept d1 and d2 as input. First, check to see that they are in the proper range for dice. If not, print a message. Otherwise, determine the outcome if this is the come out roll. If the sum is 7 or 11, print winner. If the sum is 2, 3 or 12, print loser. Otherwise print the point. 7. Field Roll. Accept d1 and d2 as input. First, check to see that they are in the proper range for dice. If not, print a message. Otherwise, check for any eld bet pay out. A roll of 2 or 12 pays 2:1, print pays 2; 3, 4, 9, 10 and 11 pays 1:1, print pays even; everything else loses, print loses 8. Hardways Roll. Accept d1 and d2 as input. First, check to see that they are in the proper range for dice. If not, print a message. Otherwise, check for a hard ways bet pay out. Hard 4 and 10 pays 7:1; Hard 6 and 8 pay 9:1, easy 4, 6, 8 or 10, or any 7 loses. Everything else, the bet still stands.
8.7. Condition Exercises
93
9. Partial Evaluation. This partial evaluation of the and and or operators appears to violate the evaluate-apply principle espoused in The Evaluate-Apply Cycle. Instead of evaluating all parameters, these operators seem to evaluate only the left-hand parameter before they are applied. Is this special case a problem? Can these operators be removed from the language, and replaced with the simple if -statement? What are the consequences of removing the short-circuit logic operators?
8.8 Condition Style Notes

Now that we have introduced compound statements, you may need to make an adjustment to your editor. Set your editor to use spaces instead of tabs. Most Python is typed using four spaces instead of the ASCII tab character (^I). Most editors can be set so that when you hit the Tab key on your keyboard, the editor inserts four spaces. IDLE is set up this way by default. A good editor will follow the indents so that once you indent, the next line is automatically indented. Well show the spaces explicitly as in the following fragment. ifa>=b: m=a ifb>=a: m=b This is has typical spacing for a piece of Python programming. Note that the colon (:) immediately follows the condition. This is the usual format, and is consistent with the way natural languages (like English) are formatted. These if statements can be collapsed to one-liners, in which case they would look like this: ifa>=b:m=a ifb>=a:m=b It helps to limit your lines to 80 positions or less. You may need to break long statements with a \\ at the end of a line. Also, parenthesized expressions can be continued onto the next line without a \\. Some programmers will put in extra ()s just to make line breaks neat. While spaces are used sparingly, they are always used to set o comparison operators and boolean operators. Other mathematical operators may or may not be set o with spaces. This makes the comparisons stand out in an if statement or while statement. ifb**2-4*a*c<0: print"noroot" This shows the space around the comparison, but not the other arithmetic operators.
94
CHAPTER
NINE
LOOPS AND ITERATIVE PROCESSING

The for, while, break, continue Statements
The elements of Python weve seen so far give us some powerful capabilities. We can write programs that implement a wide variety of requirements. State change is not always as simple as the examples weve seen in Variables, Assignment and Input . In Truth, Comparison and Conditional Processing we saw how to make our programs handle handle alternatives or conditions. In this section, well see how to write programs which do their processing for all pieces of data. For example, when we compute an average, we compute a sum for all of the values. Python provides iteration (sometimes called looping) similar to other programming languages. In Iterative Processing: For All and There Exists well describe the semantics of iterative statements in general. In Iterative Processing: The for Statement well describe the for statement. Well cover the while statements in Iterative Processing: The while Statement . This is followed by some of the most interesting and challenging short exercises in this book. Well add some iteration control in More Iteration Control: break and continue, describing the break and continue statements. Well conclude this chapter with a digression on the correct ways to develop iterative and conditional statements in A Digression.
9.1 Iterative Processing: For All and There Exists

There are two common qualiers used for logical conditions. These are sometimes called the universal and existential qualiers. We can call the for all and there exists. We can also call them the all and any qualiers. A program may involve a state that is best described as a for all state, where a number of repetitions of some task are required. For example, if we were to write a program to simulate 100 rolls of two dice, the terminating condition for our program would be that we had done the simulation for all 100 rolls. Similary, we may have a condition that looks for existence of a single example. We might want to know if a le contains a line with ERROR in it. In this case, we want to write a program with a terminating condition would be that there exists an error line in the log le. It turns out that All and Any are logical inverses. We can always rework a for any condition to be a for all condition. A program that determines if there exists an error line is the same as a program that determines that all lines are not error lines.
95
Any time we have a for all or for any condition, we have an iteration: we will be iterating through the set of values, evaluating the condition. We have a choice of two Python statements for expressing this iteration. One is the for statement and the other is the while statement.
9.2 Iterative Processing: The for Statement

The simplest for statement looks like this:
for variable in iterable : suite
The suite is an indented block of statements. Any statement is allowed in the block, including indented for statements. The variable is a variable name. The suite will be executed iteratively with the variable set to each of the values in the given iterable. Typically, the suite will use the variable, expecting it to have a distinct value on each pass. There are a number of ways of creating the necessary iterable collection of values. The most common is to use the range() function to generate a suitable list. We can also create the list manually, using a sequence display; well show some examples here. Well return to the details of sequences in Sequences: Strings, Tuples and Lists. The range() function has 3 forms: range(x) generates x distinct values, from 0 to x -1, incrementing by 1. Mathematicians describe this as a half-open interval and write it [0, x). range(x, y) generates y x distinct values from x to y-1, incrementing by 1. [x, y ). range(x, y, z) generates values from x to y-1, incrementing by z : [x, x + z, x + 2z, ..., x + kz < y ], for some integer k. A sequence display looks like this: []
expression , ...
Its a list of expressions, usually simply numbers, separated by commas. The square brackets are essential for marking a sequence. Here are some examples.
for i in range(6): print i+1
This rst example uses range() to create a sequence of 6 values from 0 to just before 6. The for statement iterates through the sequence, assigning each value to the local variable i. The print statement has an expression that adds one to i and prints the resulting value.
for j in range(1,7): print j
This second example uses the range() to create a sequence of 6 values from 1 to just before 7. The for statement iterates through the sequence, assigning each value to the local variable j . The print statement prints the value.
96
Chapter 9. Loops and Iterative Processing
for o in range(1,36,2): print o
This example uses range() to create a sequence of 36/2=18 values from 1 to just before 36 stepping by 2. This will be a list of odd values from 1 to 35. The for statement iterates through the sequence, assigning each value to the local variable o. The print statement prints all 18 values.
for r in [1,3,5,7,9,12,14,16,18,19,21,23,25,27,30,32,34,36]: print r, "red"
This example uses an explicit sequence of values. These are all of the red numbers on a standard roulette wheel. It then iterates through the sequence, assigning each value to the local variable r. The print statement prints all 18 values followed by the word red. Heres a more complex example, showing nested for statements. This enumerates all the 36 outcomes of rolling two dice.
for d1 in range(6): for d2 in range(6): print d1+1,d2+1,'=',d1+d2+2
1. The outer for statement uses range() to create a sequence of 6 values, and iterates through the sequence, assigning each value to the local variable d1. 2. For each value of d1, the inner loop creates a sequence of 6 values, and iterates through that sequence, assigning each value to d2. 3. The print statement will be executed 36 times. Heres the example alluded to earlier, which does 100 simulations of rolling two dice.
import random for i in range(100): d1= random.randrange(6)+1 d2= random.randrange(6)+1 print d1+d2
1. The for statement uses range() to create a sequence of 100 values, assigns each value to the local variable i. Note that the suite of statements never actually uses the value of i. The value of i marks the state changes until the loop is complete, but isnt used for anything else. 2. For each value of i, two values are created, d1 and d2. 3. The sum of d1 and d2 is printed. There are a number of more advanced forms of the for statement, which well cover in the section on sequences in Sequences: Strings, Tuples and Lists.
9.3 Iterative Processing: The while Statement

The while statement looks like this:
while expression : suite
9.3. Iterative Processing: The while Statement
97
The suite is an indented block of statements. Any statement is allowed in the block, including indented while statements. As long as the expression is true, the suite is executed. This allows us to construct a suite that steps through all of the necessary tasks to reach a terminating condition. It is important to note that the suite of statements must include a change to at least one of the variables in the while expression. When it is possible to execute the suite of statements without changing any of the variables in the while expression, the loop will not terminate. Lets look at some examples.
t, s = 1, 1 while t != 9: t, s = t + 2, s + t
1. The loop is initialized with t and s each set to 1. 2. We specify that the loop continues while t = 9. 3. In the body of the loop, we increment t by 2, so that it will be an odd value; we increment s by t, summing a sequence of odd values. When this loop is done, t is 9, and s is the sum of odd numbers less than 9: 1+3+5+7. Also note that the while condition depends on t, so changing t is absolutely critical in the body of the loop. Heres a more complex example. This sums 100 dice rolls to compute an average.
s, r = 0, 0 while r != 100: d1,d2=random.randrange(6)+1,random.randrange(6)+1 s,r = s + d1+d2, r + 1 print s/r
1. We initialize the loop with s and r both set to zero. 2. The while statement species that during the loop r will not be 100; when the loop is done, r will be 100. 3. The body of the loop sets d1 and d2 to random numbers; it increments s by the sum of those dice, and it increments r by 1. When the loop is over, s will be the sum of 100 rolls of two dice. When we print, s/r we print the average rolled on two dice. The loop condition depends on r, so each trip through the loop must update r.
9.4 More Iteration Control: break and continue

Python oers several statements for more subtle loop control. The point of these statements is to permit two common simplications of a loop. In each case, these statements can be replaced with if statements; however, those if statement versions might be considered rather complex for expressing some fairly common situations. The break statement terminates a loop prematurely. The syntax is trivial:
break
98
A break statement is always found within an if statement within the body of a for or while loop. A break statement is typically used when the terminating condition is too complex to write as an expression in the while clause of a loop. A break statement is also used when a for loop must be abandoned before the end of the sequence has been reached. The coninue statement skips the rest of a loops indented suite. The syntax is trivial:
continue
A continue statements is always found within an if statement within a for or while loop. The continue statement is used instead of deeply nested else clauses. Heres an example that has a complex break condition. We are going to see if we get six odd numbers in a row, or spin the roulette wheel 100 times. Well look at this in some depth because it pulls a number of features together in one program. This program shows both break and continue constructs. Most programs can actually be simplied by eliminating the break and continue statements. In this case, we didnt simplify, just to show how the statements are used. Note that we have a two part terminating condition: 100 spins or six odd numbers in a row. The hundred spins is relatively easy to dene using the range() function. The six odd numbers in a row requires testing and counting and then, possibly, ending the loop. The overall ending condition for the loop, then, is the number of spins is 100 or the count of odd numbers in a row is six.
sixodd.py
from __future__ import print_function import random oddCount= 0 for s in range(100): spinCount= s n= random.randrange(38) # Zero if n == 0 or n == 37: # treat 37 as 00 oddCount = 0 continue # Odd if n%2 == 1: oddCount += 1 if oddCount == 6: break continue # Even assert n%2 == 0 and 0 < n <= 36 oddCount = 0 print( oddCount, spinCount )
1. We import the print_function module to allow use of the print() function intead of the print statement. 2. We import the random module, so that we can generate a random sequence of spins of a roulette wheel. 3. We initialize oddCount, our count of odd numbers seen in a row. It starts at zero, because we havent seen any add numbers yet. 4. The for statement will assign 100 dierent values to s, such that 0 s < 100. This will control our experiment to do 100 spins of the wheel.
9.4. More Iteration Control: break and continue
99
5. We save the current value of s in a variable called spinCount, setting up part of our post condition for this loop. We need to know how many spins were done, since one of the exit conditions is that we did 100 spins and never saw six odd values in a row. This never saw six in a row exit condition is handled by the for statement itself. 6. Well treat 37 as if it were 00, which is like zero. In Roulette, these two numbers are neither even nor odd. The oddCount is set to zero, and the loop is continued. This continue statement resumes loop with the next value of s. It restarts processing at the top of the for statement suite. 7. We check the value of oddCount to see if it has reached six. If it has, one of the exit conditions is satised, and we can break out of the loop entirely. We use the break statement will stop executing statements in the suite of the for statement. If oddCount is not six, we dont break out of the loop, we use the continue statement to restart the for statement statement suite from the top with a new value for s. 8. We threw in an assert statement (see the next section, The assert Statement for more information) to claim that the spin, n, is even and not 0 or 37. This is kind of a safety net. If either of the preceding if statements were incorrect, or a continue statement was omitted, this statement would uncover that fact. We could do this with another if statement, but we wanted to introduce the assert statement. At the end of the loop, spinCount is the number of spins and oddCount is the most recent count of odd numbers in a row. Either oddCount is six or spinCount is 99. When spinCount is 99, that means that spins 0 through 99 were examined; there are 100 dierent numbers between 0 and 99.
9.5 Iteration Exercises

1. Greatest Common Divisor. The greatest common divisor is the largest number which will evenly divide two other numbers. Examples: GCD( 5, 10 ) = 5, the largest number that evenly divides 5 and 10. GCD( 21, 28 ) = 7, the largest number that divides 21 and 28. GCDs are used to reduce fractions. Once you have the GCD of the numerator and denominator, they can both be divided by the GCD to reduce the fraction to simplest form. 21/28 reduces to 3/4.
Greatest Common Divisor of two integers, p and q

Loop. Loop until p = q . Swap. If p < q then swap p and q, p q. Subtract. If p > q then subtract q from p, p p q . Result. Print p 2. Extracting the Square Root. This is a procedure for approximating the square root. It works by dividing the interval which contains the square root in half. Initially, we know the square root of the number is somewhere between 0 and the number. We locate a value in the middle of this interval and determine of the square root is more or less than this midpoint. We continually divide the intervals in half until we arrive at an interval which is small enough and contains the square root. If the interval is only 0.001 in width, then we have the square root accurate to 0.001
Square Root of a number, n

Two Initial Guesses.
100
g1 0 g2 n At this point, g1 g1 n 0 g2 g2 n. Loop. Loop until |g1 g1 n| n < 0.001. Midpoint. mid (g1 + g2 ) 2 Midpoint Squared vs. Number. cmp mid mid n Which Interval? if cmp 0 then g1 mid. if cmp 0 then g2 mid. if cmp = 0, mid is the exact answer! Result. Print g1 3. Sort Four Numbers. This is a challenging exercise in if-statement construction. For some additional insight, see [Dijkstra76], page 61. Given 4 numbers (W, X, Y, Z ) Assign variables w, x, y, z so that w x y z and w, x, y, z are from W, X, Y, and Z. Do not use an array. One way to guarantee the second part of the above is to initialize w, x, y, z to W, X, Y, Z, and then use swapping to rearrange the variables. Hint: There are only a limited combination of out-of-order conditions among four variables. You can design a sequence of if statements, each of which xes one of the out-of-order conditions. This sequence of if statements can be put into a loop. Once all of the out-of-order conditions are xed, the numbers are in order, the loop can end. 4. Highest Power of 2. This can be used to determine how many bits are required to represent a number. We want the highest power of 2 which is less than or equal to our target number. For example 64 100 < 128. The highest power of 25 100 < 26 . Given a number n, nd a number p such that 2p n < 2p+1 . This can be done with only addition and multiplication by 2. Multiplication by 2, but the way, can be done with the << shift operator. Do not use the pow() function, or even the ** operator, as these are too slow for our purposes. Consider using a variable c, which you keep equal to 2p . An initialization might be p = 1, c = 2. When you increment p by 1, you also double c. Develop your own loop. This is actually quite challenging, even though the resulting program is tiny. For additional insight, see [Gries81], page 147. 5. How Much Eort to Produce Software? The following equations are the basic COCOMO estimating model, described in [Boehm81]. The input, K, is the number of 1000s of lines of source; that is total source lines divided by 1000. Development Eort, where K is the number of 1000s of lines of source. E is eort in sta-months. E = 2.4 K 1.05 Development Cost, where E is eort in sta-months, R is the billing rate. C is the cost in dollars (assuming 152 working hours per sta-month) C = E R 152
9.5. Iteration Exercises
101
Project Duration, where E is eort in sta-months. D is duration in calendar months. D = 2.5 E 0.38 Staing, where E is eort in sta-months, D is duration in calendar months. S is the average sta size. S= E D
Evaluate these functions for projects which range in size from 8,000 lines (K = 8) to 64,000 lines (K = 64) in steps of 8. Produce a table with lines of source, Eort, Duration, Cost and Sta size. 6. Wind Chill Table. Wind chill is used by meteorologists to describe the eect of cold and wind combined. Given the wind speed in miles per hour, V, and the temperature in F, T, the Wind Chill, w, is given by the formula below. See Wind Chill in Numeric Types and Expressions for more information. 35.74 + 0.6215 T 35.75 (V 0.16 ) + 0.4275 T (V 0.16 ) Wind speeds are for 0 to 40 mph, above 40, the dierence in wind speed doesnt have much practical impact on how cold you feel. Evaluate this for all values of V (wind speed) from 0 to 40 mph in steps of 5, and all values of T (temperature) from -10 to 40 in steps of 5. 7. Celsius to Fahrenheit Conversion Tables. Well make two slightly dierent conversion tables. For values of Celsius from -20 to +30 in steps of 5, produce the equivalent Fahrenheit temperature. The following formula converts C (Celsius) to F (Fahrenheit). F = 32 + 212 32 C 100
For values of Fahrenheit from -10 to 100 in steps of 5, produce the equivalent Celsius temperatures. The following formula converts F (Fahrenheit) to C (Celsius). C = (F 32) 100 212 32
8. Dive Planning Table. Given a surface air consumption rate, c, and the starting, s, and nal, f, pressure in the air tank, a diver can determine maximum depths and times for a dive. For more information, see Surface Air Consumption Rate in Numeric Types and Expressions. Accept c, s and f from input, then evaluate the following for d from 30 to 120 in steps of 10. Print a table of t and d. For each diver, c is pretty constant, and can be anywhere from 10 to 20, use 15 for this example. Also, s and f depend on the tank used, typical values are s =2500 and f =500. t= 33(s f ) c(d + 33)
9. Computing . Each of the following series compute increasingly accurate values of (3.1415926...) 1 1 1 1 1 =1 + + + 4 3 5 7 9 11
2 1 1 1 = 1 + 2 + 2 + 2 + 6 2 3 4 ) ( 1 )k ( 4 2 1 1 = 16 8k + 1 8k + 4 8k + 5 8k + 6
0k<
102
Building Skills in Python, Release 2.6.5 1 12 123 + + + 3 35 357 10. Computing e. A logarithm is a power of some base. When we use logarithms, we can eectively multiply numbers using addition, and raise to powers using multiplication. Two Python built-in functions are related to this: math.log() and math.exp() . Both of these compute what are called natural logarithms, that is, logarithms where the base is e . This constant, e, is available in the math module, and it has the following formal denition: Denition of e. =1+ e=
0k<
1 k!
For more information on the () operator, see Digression on The Sigma Operator . The n! operator is factorial. Interestingly, its a post-x operator, it comes after the value it applies to. n! = n (n 1) (n 2) 1. For example, 4! = 4 3 2 1 = 24. By denition, 0! = 1.
1 1 1 1 If we add up the values 0! + 1! + 2! + 3! + we get the value of e. Clearly, when we get to about 1/10!, the fraction is so small it doesnt contribute much to the total.
We can do this with two loops, an outer loop to sum up the the k !.
1 k!
terms, and an inner loop to compute
However, if we have a temporary value of k !, then each time through the loop we can multiply this temporary by k, and then add 1/temp to the sum. You can test by comparing your results against math.e, e 2.71828 or math.exp(1.0). 11. Hailstone Numbers. For additional information, see [Banks02]. Start with a small number, n, 1 n < 30. There are two transformation rules that we will use: If n is odd, multiple by 3 and add 1 to create a new value for n. If n is even, divide by 2 to create a new value for n. Perform a loop with these two transformation rules until you get to n = 1. Youll note that when n = 1, you get a repeating sequence of 1, 4, 2, 1, 4, 2, ... You can test for oddness using the % (remainder) operation. If n % 2 == 1 , the number is odd, otherwise it is even. The two interesting facts are the path length, the number of steps until you get to 1, and the maximum value found during the process. Tabulate the path lengths and maximum values for numbers 1..30. Youll need an outer loop that ranges from 1 to 30. Youll need an inner loop to perform the two steps for computing a new n until n == 1; this inner loop will also count the number of steps and accumulate the maximum value seen during the process. Check: for 27, the path length is 111, and the maximum value is 9232.
9.6 Condition and Loops Style Notes

As additional syntax, the for and while statements permits an else clause. This is a suite of statements that are executed when the loop terminates normally. This suite is skipped if the loop is terminated by a 9.6. Condition and Loops Style Notes 103
break statement. The else clause on a loop might be used for some post-loop cleanup. This is so unlike other programming languages, that it is hard to justify using it. An else clause always raises a small problem when it is used. Its never perfectly clear what conditions lead to execution of an else clause. The condition that applies has to be worked out from context. For instance, in if statements, one explicitly states the exact condition for all of the if and elif clauses. The logical inverse of this condition is assumed as the else condition. It is, unfortunately, left to the person reading the program to work out what this condition actually is. Similarly, the else clause of a while statement is the basic loop termination condition, with all of the conditions on any break statements removed. The following kind of analysis can be used to work out the condition under which the else clause is executed.
while not BB: if C1: break if C2: break else: assert BB and not C1 and not C2 assert BB or C1 or C2
Because this analysis can be diicult, it is best to avoid the use of else clauses in loop constructs.
9.7 A Digression
For those new to programming, heres a short digression, adapted from chapter 8 of Edsger Dijkstras book, A Discipline of Programming [Dijkstra76]. Lets say we need to set a variable, m, to the larger of two input values, a and b. We start with a state we could call m undened. Then we want to execute a statement after which we are in a state of (m = a or m = b) and m a and m b. Clearly, we need to choose correctly between two dierent assignment statements. We need to do either m=a or m=b. How do we make this choice? With a little logic, we can derive the condition by taking each of these statements eects out of the desired end-state. For the statement m=a to be the right statement to use, we show the eect of the statement by replacing m with the value a, and examining the end state: (a = a or a = b) and a a and a b. Removing the parts that are obviously true, were left with a b. Therefore, the assignment m=a is only useful when a <= b. For the statement m=b to be the right statement to establish the necessary condition, we do a similar replacement of b for m and examine the end state: (b = a or b = b) and b a and b b. Again, we remove the parts that are obviously true and were left with b a. Therefore, the assignment m=b is only useful when b <= a. Each assignment statement can be guarded by an appropriate condition.
if a>=b: m=a elif b>=a: m=b
Is the correct statement to set m to the larger of a or b. Note that the hard part is establishing the post condition. Once we have that stated correctly, its relatively easy to gure the basic kind of statement that might make some or all of the post condition true. Then we do a little algebra to ll in any guards or loop conditions to make sure that only the correct statement is executed. Successful Loop Design. There are several considerations when using the while statement. This list is taken from David Gries, The Science of Programming [Gries81]. 104 Chapter 9. Loops and Iterative Processing
1. The body condition must be initialized properly. 2. At the end of the suite, the body condition is just as true as it was after initialization. This is called the invariant , because it is always true during the loop. 3. When this body condition is true and the while condition is false, the loop will have completed properly. 4. When the while condition is true, there are more iterations left to do. If we wanted to, we could dene a mathematical function based on the current state that computes how many iterations are left to do; this function must have a value greater than zero when the while condition is true. 5. Each time through the loop we change the state of our variables so that we are getting closer to making the while condition false; we reduce the number of iterations left to do. While these conditions seem overly complex for something so simple as a loop, many programming problems arise from missing one of them. Gries recommends putting comments around a loop showing the conditions before and after the loop. Since Python provides the assert statement; this formalizes these comments into actual tests to be sure the program is correct. Designing a Loop. Lets put a particular loop under the microscope. This is a small example, but shows all of the steps to loop construction. We want to nd the least power of 2 greater than or equal to some number greater than 1, call it x. This power of 2 will tell us how many bits are required to represent x, for example. We can state this mathematically as looking for some number, n, such that 2n1 < x 2n . If x is a power of 2, for example 64, wed nd 26 . If x is another number, for example 66, wed nd 26 < 66 27 , which is 64 < 66 128. We can start to sketch our loop already.
assert x > 1 ... initialize ... ... some loop ... assert 2**(n-1) < x <= 2**n
We work out the initialization to make sure that the invariant condition of the loop is initially true. Since x must be greater than or equal to 1, we can set n to 1. 211 = 20 = 1 < x. This will set things up to satisfy rule 1 and 2.
assert x > 1 n= 1 ... some loop ... assert 2**(n-1) < x <= 2**n
In loops, there must be a condition on the body that is invariant, and a terminating condition that changes. The terminating condition is written in the while clause. In this case, it is invariant (always true) that 2n1 < x. That means that the other part of our nal condition is the part that changes.
assert x > 1 n= 1 while not ( x <= 2**n ): n= n + 1
9.7. A Digression
105
assert 2**(n-1) < x assert 2**(n-1) < x <= 2**n
The next to last step is to show that when the while condition is true, there are more than zero trips through the loop possible. We know that x is nite and some power of 2 will satisfy this condition. Theres some n such that n 1 < log2 x n, which limits the trips through the loop. The nal step is to show that each cycle through the loop reduces the trip count. We can argue that increasing n gets us closer to the upper bound of log2 x. We should add this information on successful termination as comments in our loop.
106
CHAPTER
TEN
FUNCTIONS
The heart of programming is the evaluate-apply cycle, where function arguments are evaluated and then the function is applied to those argument values. Well review this in Semantics. In Function Denition: The def and return Statements we introduce the syntax for dening a function. In Function Use, well describe using a function weve dened. Some languages make distinctions between varieties of functions, separating them into functions and subroutines. Well visit this from a Python perspective in Function Varieties. Well look at some examples in Some Examples. Well look at ways to use IDLE in Hacking Mode. We introduce some of the alternate argument forms available for handling optional and keyword parameters in More Function Denition Features. Further sophistication in how Python handles parameters has to be deferred to Advanced Parameter Handling For Functions, as it depends on a knowledge of dictionaries, introduced in Mappings and Dictionaries. In Object Method Functions we will describe how to use method functions as a prelude to Data Structures; real details on method functions are deferred until Classes. Well also defer examination of the yield statement until Iterators and Generators. The yield statement creates a special kind of function, one that is most useful when processing complex data structures, something well look at in Data Structures.
10.1 Semantics
A function, in a mathematical sense, is often described as a mapping from domain values to range values. Given a domain value, the function returns the matching range value. If we think of the square root function, it maps a positive number, n, to another number, s, such that s2 = n. If we think of multplication as a function, it maps a pair of values, a and b, to a new value, c, such that c = a b. When we memorize multiplication tables, we are memorizing these mappings. In Python, this narrow denition is somewhat relaxed. Python lets us create functions which do not need a domain value, but create new objects. It also allows us to have functions that dont return values, but instead have some other eect, like reading user input, or creating a directory, or removing a le. What We Provide. In Python, we create a new function by providing three pieces of information: the name of the function, a list of zero or more variables, called parameters, with the domain of input values, and a suite of statements that creates the output values. This denition is saved for later use. Well show this rst in Function Denition: The def and return Statements.
107
Typically, we create function denitions in script les because we dont want to type them more than once. Almost universally, we import a le with our function denitions so we can use them. We use a function in an expression by following the functions name with (). The Python interpreter evaluates the argument values in the (), then applies the function. Well show this second in Function Use. Applying a function means that the interpreter rst evaluates all of the argument values, then assigns the argument values to the function parameter variables, and nally evaluates the suite of statements that are the functions body. In this body, any return statements dene the resulting range value for the function. For more information on this evaluate-apply cycle, see The Evaluate-Apply Cycle. Namespaces and Privacy. Note that the parameter variables used in the function denition, as well as any variables in a function are private to that functions suite of statements. This is a consequence of the way Python puts all variables in a namespace. When a function is being evaluated, Python creates a temporary namespace. This namespace is deleted when the functions processing is complete. The namespace associated with application of a function is dierent from the global namespace, and dierent from all other function-body namespaces. While you can change the standard namespace policy (see The global Statement ) it generally will do you more harm than good. A functions interface is easiest to understand if it is only the parameters and return values and nothing more. If all other variables are local, they can be safely ignored. Terminology: argument and parameter. We have to make a rm distinction between an argument value, an object that is created or updated during execution, and the dened parameter variable of a function. The argument is the object used in particular application of a function; it may be referenced by other variables or objects. The parameter is a variable name that is part of the function, and is a local variable within the function body.
108
Chapter 10. Functions
The Evaluate-Apply Cycle The evaluate-apply cycle shows how any programming language computes the value of an expression. Consider the following expression:
math.sqrt( abs( b*b-4*a*c ) )
What does Python do? For the purposes of analysis, we can restructure this from the various mathematical notation styles to a single, uniform notation. We call this prex notation, because all of the operations prex their operands. While useful for analysis, this is cumbersome to write for real programs.
math.sqrt( abs( sub( mul(b,b), mul(mul(4,a),c) ) ) )
Weve replaced x*y with mul(x,y) , and replaced x-y with sub(x,y) . This allows us to more clearly see how evaluate-apply works. Each part of the expression is now written as a function with one or two arguments. First the arguments are evaluated, then the function is applied to those arguments. In order for Python to evaluate this math.sqrt(...) expression, it evaluates the argument, abs(...), and then applies math.sqrt() to it. This leads Python to a nested evaluate-apply process for the abs(...) expression. Well show the whole process, with indentation to make it clearer. Were going to show this as a list of steps, with > to show how the various operations nest inside each other.
Evaluate the arg to math.sqrt: > Evaluate the args to sub: > > Evaluate the args to mul: > > > Get the value of b > > Apply mul to b and b, creating r3=mul( b, b ). > > Evaluate the args to mul: > > > Evaluate the args to mul: > > > > Get the value of a > > > Apply mul to 4 and a, creating r5=mul( 4, a ). > > > Get the value of c > > Apply mul to r5 and c, creating r4=mul( mul( 4, a ), c ). > Apply sub to r3 and r4, creating r2=sub( mul( b, b ), mul( mul( 4, a ), c ) ). Apply math.sqrt to r2, creating r1=math.sqrt( sub( mul( b, b ), mul( mul( 4, a ), c ) ) ).
Notice that a number of intermediate results were created as part of this evaluation. If we were doing this by hand, wed write these down as steps toward the nal result. The apply part of the evalute-apply cycle is sometimes termed a function call. The idea is that the main procedure calls the body of a function; the function does its work and returns to the main procedure. This is also called a function invocation.
10.2 Function Denition: The def and return Statements

We create a function with a def statement. This provides the name, parameters and the suite of statements that creates the functions result.
def name ( parameter , ... ): suite
The name is the name by which the function is known. The parameters is a list of variable names; these names are the local variables to which actual argument values will be assigned when the function is applied. The suite (which must be indented) is a block of statements that computes the value for the function.
10.2. Function Denition: The def and return Statements
109
The rst line of a functions suite is expected to be a document string (generally a triple-quoted string) that provides basic documentation for the function. This is traditionally divided in two sections, a summary section of exactly one line and the detail section. Well return to this style guide in Functions Style Notes. The return statement species the result value of the function. This value will become the result of applying the function to the argument values. This value is sometimes called the eect of the function.
return expression
The yield statement species one of the result values of an iterable function. Well return to this in Iterators and Generators. Lets look at a complete example.
def odd( spin ): """Return true if this spin is odd.""" if spin % 2 == 1: return True return False
1. We name this function odd(), and dene it to accept a single parameter, named spin. 2. We provide a docstring with a short description of the function. 3. In the body of the function, we test to see if the remainder of spin /2 is 1; if so, we return True. 4. Otherwise, we return False.
10.3 Function Use

When Python evaluates odd(n), the following things will happen. 1. It evaluates n. For a simple variable, the value is the object to which the variable refers. For an expression, the expression is evaluated to result in an object. 2. It assigns this argument value to the local parameter of odd() (named spin ). 3. It applies odd(): the suite of statements is executed, ending with a return statement. 4. This value on the return statement is returned to the calling statement so that it can nish its execution. We would use this odd() function like this.
s = random.randrange(37) # 0 <= s <= 36, single-0 roulette if s == 0: print "zero" elif odd(s): print s, "odd" else: print s, "even"
1. We evaluate a function named random.randrange to create a random number, s. 2. The if clause handles the case where s is zero.
110
3. The rst elif clause evaluates our odd() function. To do this evaluation, Python must set spin to the value of s and execute the suite of statements that are the body of odd(). The suite of statements will return either True or False. 4. Since the if and elif clauses handle zero and odd cases, all that is left is for s to be even.
10.4 Function Varieties

Some programming languages make a distinction between various types of functions or subprograms. There can be functions or subroutines or procedure functions. Python (like Java and C++) doesnt enforce this kind of distinction. Instead, Python imposes some distinction based on whether the function uses parameters and returns a value or yields a collection of values. Ordinary Functions. Functions which follow the classic mathematical denitions will map input argument values to a resulting value. These are, perhaps, a common kind of function. They include a return statement to express the resulting value. Procedure Functions. One common kind of function is one that doesnt return a result, but instead carries out some procedure. This function would omit any return statement. Or, if a return statement is used to exit from the function, the statement would have no value to return. Carrying out an action is sometimes termed a side-eect of the function. The primary eect is always the value returned. Heres an example of a function that doesnt return a value, but carries out a procedure.
from __future__ import print_function def report( spin ): """Report the current spin.""" if spin == 0: print( "zero" ) return if odd(spin): print( spin, "odd" ) return print( spin, "even" )
This function, report(), has a parameter named spin, but doesnt return a value. Here, the return statements exit the function but dont return values. This kind of function would be used as if it was a new Python language statement, for example:
for i in range(10): report( random.randrange(37) )
Here we execute the report() function as if it was a new kind of statement. We dont evaluate it as part of an expression. Theres actually no real subtlety to this distinction. Any expression can be used as a Python statement. A function call is an expression, and an expression is a statement. This greatly simplies Python syntax. The docstring for a function will explain what kind of value the function returns, or if the function doesnt return anything useful. The simple return statement, by the way, returns the special value None. This default value means that you can dene your function like report(), above, use it in an expression, and everything works nicely because the function does return a value. 10.4. Function Varieties 111
for i in range(10): t= report( random.randrange(37) ) print t
Youll see that t is None . Factory Functions. Another common form is a function that doesnt take a parameter. This function is a factory which generates a value. Some factory functions work by accessing some object encapsulated in a module. In the following example, well access the random number generator encapsulated in the random module.
def spinWheel(): """Return a string result from a roulette wheel spin.""" t= random.randrange(38) if t == 37: return "00" return str(t)
This functions evaluate-apply cycle is simplied to just the apply phase. To make 0 (zero) distinct from 00 (double zero), it returns a string instead of a number. Generators. A generator function contains the yield statement. These functions look like conventional functions, but they have a dierent purpose in Python. We will examine this in detail in Iterators and Generators. These functions have a persistent internal processing state; ordinary functions cant keep data around from any previous calls without resorting to global variables. Further, these functions interact with the for statement. Finally, these functions dont make a lot of sense until weve worked with sequences in Sequences: Strings, Tuples and Lists.
10.5 Some Examples

Heres a big example of using the odd() , spinWheel() and report() functions.
functions.py
#!/usr/bin/env python import random def odd( spin ): """odd(number) -> boolean.""" return spin%2 == 1 def report( spin ): """Reports the current spin on standard output. if int(spin) == 0: print "zero" return if odd(int(spin)): print spin, "odd" return print spin, "even"
Spin is a String"""
112
def spinWheel(): """Returns a string result from a roulette wheel spin.""" t= random.randrange(38) if t == 37: return "00" return str(t) for i in range(12): n= spinWheel() report( n )
1. Weve dened a function named odd(). This function evaluates a simple expression; it returns True if the value of its parameter, spin, is odd. 2. The function called report() uses the odd() function to print a line that describes the value of the parameter, spin. Note that the parameter is private to the function, so this use of the variable name spin is technically distinct from the use in the odd() function. However, since the report() function provides the value of spin to the odd() function, their two variables often happen to have the same value. 3. The spinWheel() function creates a random number and returns the value as a string. 4. The main part of this program is the for loop at the bottom, which calls spinWheel(), and then report(). The spinWheel() function uses random.randrange(); the report() function uses the odd() function. This generates and reports on a dozen spins of the wheel. For most of our exercises, this free-oating main script is acceptable. When we cover modules, in Components, Modules and Packages, well need to change our approach slightly to something like the following.
def main(): for i in range(12): n= spinWheel() report( n ) main()
This makes the main operation of the script clear by packaging it as a function. Then the only free-oating statement in the script is the call to main().
10.6 Hacking Mode

On one hand we have interactive use of the Python interpreter: we type something and the interpreter responds immediately. We can do simple things, but when our statements get too long, this interaction can become a niusance. We introduced this rst, in Command-Line Interaction. On the other hand, we have scripted use of the interpreter: we present a le as a nished program to execute. While handy for getting useful results, this isnt the easiest way to get a program to work in the rst place. We described this in Script Mode. In between the interactive mode and scripted mode, we have a third operating mode, that we might call hacking mode. The idea is to write most of our script and then exercise portions of our script interactively. In this mode, well develop script les, but well exercise them in an interactive environment. This is handy for developing and debugging function denitions. The basic procedure is as follows.
10.6. Hacking Mode
113
1. In our favorite editor, write a script with our function denitions. We often leave this editor window open. IDLE, for example, leaves this window open for us to look at. 2. Open a Python shell. IDLE, for example, always does this for us. 3. In the Python Shell, import the script le. In IDLE, this is eectively what happens when we run the module with F5. This will execute the various def statements, creating our functions in our interactive shell. 4. In the Python Shell, test the function interactively. If it works, were done. 5. If the functions in our module didnt work, we return to our editor window, make any changes and save the le. 6. In the Python Shell, clear out the old denition by restarting the shell. In IDLE, we can force this with F6. This happens automatically when we run the module using F5 7. Go back to step 3, to import and test our denitions. The interactive test results can be copied and pasted into the docstring for the le with our function denitions. We usually copy the contents of the Python Shell window and paste it into our modules or functions docstring. This record of the testing can be validated using the doctest module. Example. Heres the sample function were developing. If you look carefully, you might see a serious problem. If you dont see the problem, dont worry, well nd it by doing some debugging. In IDLE, we created the following le.
function1.py Initial Version

def odd( number ): """odd(number) -> boolean Returns True if the given number is odd. """ return number % 2 == "1"
We have two windows open: function1.py and Python Shell. Heres our interactive testing session. In our function1.py window, we hit F5 to run the module. Note the line that shows that the Python interpreter was restarted; forgetting any previous denitions. Then we exercised our function with two examples.
Python 2.5.1 (r251:54863, Oct 5 2007, 21:08:09) [GCC 4.0.1 (Apple Inc. build 5465)] on darwin Type "help", "copyright", "credits" or "license" for more information. ************************************************************ Personal firewall software may warn about the connection IDLE makes to its subprocess using this computer's internal loopback interface. This connection is not visible on any external interface and no data is sent to or received from the Internet. ************************************************************ IDLE 1.1.4 >>> ================================ RESTART ================================ >>> >>> odd(2) False
114
>>> odd(3) False
Clearly, it doesnt work, since 3 is odd. When we look at the original function, we can see the problem. The expression number % 2 == "1" should be number % 2 == 1. We need to x function1.py. Once the le is xed, we need to remove the old stu from Python, re-import our function and rerun our test. IDLE does this for us when we hit F5 to rerun the module. It shows this with the prominent restart message. If you are not using IDLE, you will need to restart Python to clear out the old denitions. Python optimizes import operations; if its seen the module once, it doesnt import it a second time. To remove this memory of which modules have been imported, you will need to restart Python.
10.7 More Function Denition Features

Python provides a mechanism for optional parameters. This allows us to create a single function which has several alternative forms. In other languages, like C++ or Java, these are called overloaded functions; they are actually separate function denitions with the same name but dierent parameter forms. In Python, we can write a single function that accepts several parameter forms. Python has three mechanisms for dealing with optional parameters and a variable number of parameters. Well cover the basics of optional parameters in this section. The other mechanisms for dealing with variable numbers of parameters will be deferred until Advanced Parameter Handling For Functions because these mechanisms use some more advanced data structures. Python functions can return multiple values. Well look at this, also.
10.7.1 Default Values for Parameters

The most common way to implement optional parameters is by providing a default value for the optional parameters. If no argument is supplied for the parameter, the default value is used.
def report( spin, count=1 ): print spin, count, "times in
a row"
This silly function can be used in two ways:

report( n ) report( n, 2 )
The rst form provides a default argument of 1 for the count parameter. The second form has an explicit argument value of 2 for the count parameter. If a parameter has no default value, it is not optional. If a parameter has a default value, it is optional. In order to disambiguate the assignment of arguments to parameters, Python uses a simple rule: all required parameters must be rst, all optional parameters must come after the required parameters. The int() function does this. We can say int("23") to do decimal conversion and int("23",16) to do hexadecimal conversion. Clearly, the second argument to int() has a default value of 10. Important: Red Alert Its very, very important to note that default values must be immutable objects. Well return to this concept of mutability in Data Structures. 10.7. More Function Denition Features 115
For now, be aware that numbers, strings, None, and tuple objects are immutable. As we look at various data type, well nd that lists, sets and dictionaries are mutable, and cannot be used as default values for function parameters. Fancy Defaults. When we look at the Python range() function, we see a more sophisticated version of this. range(x) is the same as range(0,x,1). range(x,y) is the same as range(x,y,1). It appears from these examples that the rst parameter is optional. The authors of Python use a pretty slick trick for this that you can use also. The range() function behaves as though the following function is dened.
def range(x, y=None, z=None): if y==None: start, stop, step = 0, x, 1 elif z==None: start, stop, step = x, y, 1 else: start, stop, step = x, y, z Real work is done with start, stop and step
By providing a default value of None, the function can determine whether a value was supplied or not supplied. This allows for complex default handling within the body of the function. Conclusion. Python must nd a value for all parameters. The basic rule is that the values of parameters are set in the order in which they are declared. Any missing parameters will have their default values assigned. These are called positional parameters, since the position is the rule used for assigning argument values when the function is applied. If a mandatory parameter (a parameter without a default value) is missing, this is a basic TypeError. For example:
badcall.py
#!/usr/bin/env python def hack(a,b): print a+b hack(3)
When we run this example, we see the following.

MacBook-5:Examples slott$ python badcall.py Traceback (most recent call last): File "badcall.py", line 4, in <module> hack(3) TypeError: hack() takes exactly 2 arguments (1 given)
10.7.2 Providing Argument Values by Keyword

In addition to supplying argument values by position, Python also permits argument values to be specied by name. Using explicit keywords can make programs much easier to read.
116
First, well dene a function with a simple parameter list:

import random def averageDice( samples=100 ): """Return the average of a number of throws of 2 dice.""" s = 0 for i in range(samples): d1,d2 = random.randrange(6)+1,random.randrange(6)+1 s += d1+d2 return float(s)/float(samples)
Next, well show three dierent kinds of arguments: keyword, positional, and default.
test1 = averageDice( samples=200 ) test2 = averageDice( 300 ) test3 = averageDice()
When the averageDice() function is evaluated to set test1, the keyword form is used. The second call of the averageDice() function uses the positional form. The nal example relies on a default for the parameter. Conclusion. This gives us a number of variations including positional parameters and keyword parameters, both with and without defaults. Positional parameters work well when there are few parameters and their meaning is obvious. Keyword parameters work best when there are a lot of parameters, especially when there are optional parameters. Good use of keyword parameters mandates good selection of keywords. Single-letter parameter names or obscure abbreviations do not make keyword parameters helpfully informative. Here are the rules weve seen so far: 1. Supply values for all parameters given by name, irrespective of position. 2. Supply values for all remaining parameters by position; in the event of duplicates, raise a TypeError. 3. Supply defaults for any parameters that have defaults dened; if any parameters still lack values, raise a TypeError. There are still more options available for handling variable numbers of parameters. Its possible for additional positional parameters to be collected into a sequence object. Further, additional keyword parameters can be collected into a dictionary object. Well get to them when we cover dictionaries in Advanced Parameter Handling For Functions.
10.7.3 Returning Multiple Values

One common desire among programmers is a feature that allows a function to return multiple values. Python has some built-in functions that have this property. For example, divmod() returns the divisor and remainder in division. We could imagine a function, rollDice() that would return two values showing the faces of two dice. In Python, it is done by returning a tuple. Well wait for Tuples for complete information on tuples. The following is a quick example of how multiple assignment works with functions that return multiple values.
rolldice.py
10.7. More Function Denition Features
117
import random def rollDice(): return ( 1 + random.randrange(6), 1 + random.randrange(6) ) d1,d2=rollDice() print d1,d2
This shows a function that creates a two-valued tuple. Youll recall from Multiple Assignment Statement that Python is perfectly happy with multiple expressions on the right side of =, and multiple destination variables on the left side. This is one reason why multiple assignment is so handy.
10.8 Function Exercises

1. Fast exponentiation. This is a fast way to raise a number to an integer power. It requires the fewest multiplies, and does not use logarithms.
Fast Exponentiation of integers, raises n to the p power

(a) Base Case. If p = 0: return 1.0. (b) Odd. If p is odd: return n fastexp(n, p 1). (c) Even. If p is even: p compute t fastexp(n, ); 2 return t t. 2. Greatest Common Divisor. The greatest common divisor is the largest number which will evenly divide two other numbers. You use this when you reduce fractions. See Greatest Common Divisor for an alternate example of this exercises algorithm. This version can be slightly faster than the loop we looked at earlier.
Greatest Common Divisor of two integers, p and q

(a) Base Case. If p = q : return p. (b) p < q. If p < q : return GCD(q, p). (c) p > q. If p > q : return GCD(p, p q ). 3. Factorial Function. Factorial of a number n is the number of possible arrangements of 0 through n things. It is computed as the product of the numbers 1 through n. That is, 1 2 3 n. The formal denition is n! = n (n 1) (n 2) 1 0! = 1 We touched on this in Computing e. This function denition can simplify the program we wrote for that exercise.
118
Factorial of an integer, n
(a) Base Case. If n = 0, return 1. (b) Multiply. If n > 0: return n factorial(n 1). 4. Fibonacci Series. Fibonacci numbers have a number of interesting mathematical properties. The ratio of adjacent Fibonacci numbers approximates the golden ratio ((1 + 5)/2, about 1.618), used widely in art and architecture.
The n th Fibonacci Number, Fn .

(a) F(0) Case. If n = 0: return 0. (b) F(1) Case. If n = 1: return 1. (c) F(n) Case. If n > 1: return F(n 1) + F(n 2). 5. Ackermanns Function. An especially complex algorithm that computes some really big results. This is a function which is specically designed to be complex. It cannot easily be rewritten as a simple loop. Further, it produces extremely large results because it describes extremely large exponents.
Ackermanns Function of two numbers, m and n

(a) Base Case. If m = 0: return n + 1. (b) N Zero Case. If m = 0 and n = 0: return ackermann(m 1, 1). (c) N Non-Zero Case. If m = 0 and n = 0: return ackermann(m 1, ackermann(m, n 1)). Yes, this requires you to compute ackermann(m, n 1) before you can compute ackermann(m 1, ackermann(m, n 1)). 6. Maximum Value of a Function. Given some integer-valued function f(), we want to know what value of x has the largest value for f() in some interval of values. For additional insight, see [Dijkstra76]. Imagine we have an integer function of an integer, call it f(). Here are some examples of this kind of function. def f1(x): return x def f2(x): return -5/3*x-3 def f3(x): return -5*x*x+2*x-3 The question we want to answer is what value of x in some xed interval returns the largest value for the given function? In the case of the rst example, def f1(x): return x, the largest value of f1() in the interval 0 x < 10 occurs when x is 9. What about f3() in the range 10 x < 10?
Max of a Function, F, in the interval low to high

(a) Initialize. x low; max x;
10.8. Function Exercises
119
maxF F(max). (b) Loop. While low x < high. i. New Max? If F(x) > maxF : max x; maxF F(max). ii. Next X. Increment x by 1. (c) Return. Return max as the value at which F(x) had the largest value. 7. Integration. This is a simple rectangular rule for nding the area under a curve which is continuous on some closed interval. We will dene some function which we will integrate, call it f(x)(). Here are some examples. def f1(x): return x*x def f2(x): return 0.5 * x * x def f3(x): return exp( x ) def f4(x): return 5 * sin( x ) When we specify y = f (x), we are specifying two dimensions. The y is given by the functions values. The x dimension is given by some interval. If you draw the functions curve, you put two limits on the x axis, this is one set of boundaries. The space between the curve and the y axis is the other boundary.
a The x axis limits are a and b. We subdivide this interval into s rectangles, the width of each is h = b s . We take the functions value at the corner as the average height of the curve over that interval. If the interval is small enough, this is reasonably accurate.
Integrate a Function, F, in the interval a to b in s steps

(a) Initialize. xa ba h s sum 0.0 (b) Loop. While a x < b. i. Update Sum. Increment sum by F (x) h. ii. Next X. Increment x by h. (c) Return. Return sum as the area under the curve F() for a x < b. 8. Field Bet Results. In the dice game of Craps, the Field bet in craps is a winner when any of the numbers 2, 3, 4, 9, 10, 11 or 12 are rolled. On 2 and 12 it pays 2:1, on any of the other numbers, it pays 1:1. Dene a function win( dice, num, pays). If the value of dice equals num, then the value of pays is returned, otherwise 0 is returned. Make the default for pays a 1, so we dont have to repeat this value over and over again. Dene a function field( dice ). This will call win() 7 times: once with each of the values for which the eld pays. If the value of dice is a 7, it returns -1 because the bet is a loss. Otherwise it returns 0 because the bet is unresolved.
120
It would start with

def field( dice ): win( dice, 2, pays=2 ) win( dice, 3, pays=1 ) ...
Create a function roll() that creates two dice values from 1 to 6 and returns their sum. The sum of two dice will be a value from 2 to 12. Create a main program that calls roll() to get a dice value, then calls field() with the value that is rolled to get the payout amount. Compute the average of several hundred experiments. 9. range() Function Keywords. Does the range function permit keywords for supplying argument values? What are the keywords? 10. Optional First Argument. Optional parameters must come last, yet range fakes this out by appearing to have an optional parameter that comes rst. The most common situation is range(5) , and having to type range(0,5) seems rather silly. In this case, convenience trumps strict adherence to the rules. Is this a good thing? Is strict adherence to the rules more or less important than convenience?
10.9 Object Method Functions

Weve seen how we can create functions and use those functions in programs and other functions. Python has a related technique called methods or method functions. The functions weve used so far are globally available. A method function, on the other hand, belongs to an object. The objects class denes what methods and what properties the object has. Well cover method functions in detail, starting in Classes. For now, however, some of the Python data types were going to introduce in Data Structures will use method functions. Rather than cover too many details, well focus on general principles of how you use method functions in this section. The syntax for calling a method function looks like this:
someObject.aMethod( argument list )
A single . separates the owning object (someObject) from the method name (aMethod()). We glanced at a simple example when we rst looked at complex numbers. The complex conjugate function is actually a method function of the complex number object. The example is in Complex Numbers. In the next chapter, well look at various kinds of sequences. Python denes some generic method functions that apply to any of the various classes of sequences. The string and list classes, both of which are special kinds of sequences, have several methods functions that are unique to strings or lists. For example:
>>> "Hi Mom".lower() 'hi mom'
Here, we call the lower() method function, which belongs to the string object "Hi Mom". When we describe modules in Components, Modules and Packages, well cover module functions. These are functions that are imported with the module. The array module, for example, has an array() function that creates array objects. An array object has several method functions. Additionally, an array object is a kind of sequence, so it has all of the methods common to sequences, also.
10.9. Object Method Functions
121
file objects have an interesting life cycle, also. A file object is created with a built-in function, file(). A le object has numerous method functions, many of which have side-eects of reading from and writing to external les or devices. Well cover les in Files, listing most of the methods unique to le objects.
10.10 Functions Style Notes

The suite within a compound statement is typically indented four spaces. It is often best to set your text editor with tab stops every four spaces. This will usually yield the right kind of layout. Well show the spaces explicitly as in the following fragment. defmax(a,b): ifa>=b: m=a ifb>=a: m=b returnm This is has typical spacing for a piece of Python programming. Also, limit your lines to 80 positions or less. You may need to break long statements with a \ at the end of a line. Also, parenthesized expressions can be continued onto the next line without a \. Some programmers will put in extra () just to make line breaks neat. Names. Function names are typically mixedCase(). However, a few important functions were done in CapWords() style with a leading upper case letter. This can cause confusion with class names, and the recommended style is a leading lowercase letter for function names. In some languages, many related functions will all be given a common prex. Functions may be called inet_addr(), inet_network(), inet_makeaddr(), inet_lnaof(), inet_netof(), inet_ntoa(), etc. Because Python has classes (covered in Data + Processing = Objects) and modules (covered in Components, Modules and Packages), this kind of function-name prex is not used in Python programs. The class or module name is the prex. Look at the example of math and random for guidance on this. Parameter names are also typically mixedCase. In the event that a parameter or variable name conicts with a Python keyword, the name is extended with an _. In the following example, we want our parameter to be named range, but this conicts with the builtin function range(). We use a trailing _ to sort this out.
def integrate( aFunction, range_ ): """Integrate a function over a range.""" ...
Blank lines are used sparingly in a Python le, generally to separate unrelated material. Typicaly, function denitions are separated by single blank lines. A long or complex function might have blank lines within the body. When this is the case, it might be worth considering breaking the function into separate pieces. Docstrings. The rst line of the body of a function is called a docstring. The recommended forms for docstrings are described in Python Extension Proposal (PEP) 257. Typically, the rst line of the docstring is a pithy summary of the function. This may be followed by a blank line and more detailed information. The one-line summary should be a complete sentence.
def fact( n ): """fact( number ) -> number Returns the number of permutations of n things."""
122
if n == 0: return 1L return n*fact(n-1L) def bico( n, r ): """bico( number, number ) -> number Returns the number of combinations of n things taken in subsets of size r. Arguments: n -- size of domain r -- size of subset """ return fact(n)/(fact(r)*fact(n-r))
The docsting can be retrieved with the help() function. help(object ) Provides help about the given object. Heres an example, based on our fact() shown above.
>>> help(fact) Help on function fact in module __main__: fact(n) fact( number ) -> number Returns the number of permutations of n things.
Note that you will be in the help reader, with a prompt of (END). Hit q to quit the help reader. For more information, see Getting Help.
10.10. Functions Style Notes
123
124
CHAPTER
ELEVEN
ADDITIONAL NOTES ON FUNCTIONS

The global Statement
In Functions and Namespaces well describe some of the internal mechanisms Python uses for storing variables. Well introduce the global statement in The global Statement . Well include a digression on the two common argument binding mechanisms: call by value and call by reference in Call By Value and Call By Reference. Note that this is a distinction that doesnt apply to Python, but if you have experience in languages like C or C++, you may wander where and how this is implemented. Finally, well cover some aspects of functions as rst-class objects in Function Objects.
11.1 Functions and Namespaces

This is an overview of how Python determines the meaning of a name. Well omit some details to hit the more important points. For more information, see section 4.1 of the Python Language Reference. The important issue is that we want variables created in the body of a function to be private to that function. If all variables are global, then each function runs a risk of accidentally disturbing the value of a global variable. In the COBOL programming language (without using separate compilation or any of the modern extensions) all variables are globally declared in the data division, and great care is required to prevent accidental or unintended use of a variable. To achieve privacy and separation, Python maintains several dictionaries of variables. These dictionaries dene the context in which a variable name is understood. Because these dictionaries are used for resolution of variables, which name objects, they are called namespaces. A global namespace is available to all modules that are part of the currently executing Python script. Each module, class, function, lambda, or anonymous block of code given to the exec command has its own private namespace. Names are resolved using the nested collection of namespaces that dene an execution environment. Python always checks the most-local dictionary rst, ending with the global dictionary. Consider the following script.
def deep( a, b ): print "a=", a print "b=", b
125
def shallow( hows, things ): deep( hows, 1 ) deep( things, coffee ) hows= 1 coffee= 2 shallow( "word", 3.1415926 ) shallow( hows, coffee )
1. The deep() function has a local namespace, where two variables are dened: a and b. When deep() is called from shallow(), there are three nested scopes that dene the environment: the local namespace for deep(): the local namespace for shallow(), and the global namespace for the main script. 2. The shallow() function has a local namespace, where two variables are dened: hows and things. When shallow() is called from the main script, the local hows is resolved in the local namespace. It hides the global variable with the same name. The reference to coffee is not resolved in the local namespace, but is resolved in the global namespace. This is called a free variable, and is sometimes a symptom of poor software design. 3. The main script by denition executes in the global namespace, where two variables (hows and coffee) are dened, along with two functions, deep() and shallow(). Built-in Functions. If you evaluate the function globals(), youll see the mapping that contains all of the global variables Python knows about. For these early programs, all of our variables are global. If you simply evaluate locals(), youll see the same thing. However, if you call locals() from within the body of a function, youll be able to see the dierence between local and global variables. The following example shows the creation of a gobal variable a, and a global function, q. It shows the local namespace in eect while the function is executing. In this local namespace we also have a variable named a.
>>> a=22.0 >>> globals() {'__builtins__': <module '__builtin__' (built-in)>, '__doc__': None, 'a': 22.0} >>> def q( x, y ): ... a = x / y ... print locals() ... >>> locals() {'__builtins__': <module '__builtin__' (built-in)>, 'q': <function q at 0x76830>, '__doc__': None, 'a': >>> globals() {'__builtins__': <module '__builtin__' (built-in)>, 'q': <function q at 0x76830>, '__doc__': None, 'a': >>> q(22.0,7.0) {'a': 3.1428571428571428, 'y': 7.0, 'x': 22.0}
'__name__': '__main__',
'__name__': '__main__', 22.0} '__name__': '__main__', 22.0}
The function vars() accepts a parameter which is the name of a specic local context: a module, class, or object. It returns the local variables for that specic context. The local variables are kept in a local variable named __dict__. The vars() function retrieves this. The dir() function examines the __dict__ of a specic object to locate all local variables as well as other features of the object. Assignment statements, as well as def and class statements, create names in the local dictionary. The del statement removes a name from the local dictionary. 126 Chapter 11. Additional Notes On Functions
Some Consequences. Since each imported module exists in its own namespace, all functions and classes within that module must have their names qualied by the module name. We saw this when we imported math and random. To use the sqrt() function, we must say math.sqrt, providing the module name that is used to resolve the name sqrt(). This module namespace assures that everything in a module is kept separate from other modules. It makes our programs clear by qualifying the name with the module that dened the name. The module namespace also allow a module to have relatively global variables. A module, for example, can have variables that are created when the module is imported. In a sense these are global to all the functions and classes in the module. However, because they are only known within the modules namespace, they wont conict with variables in our program or other modules. Having to qualify names within a module can become annoying when we are making heavy use of a module. Python has ways to put elements of a module into the global namespace. Well look at these in Components, Modules and Packages.
11.2 The global Statement

The suite of statements in a function denition executes with a local namespace that is dierent from the global namespace. This means that all variables created within a function are local to that function. When the suite nishes, these working variables are discarded. The overall Python session works in the global namespace. Every other context (e.g. within a functions suite) is a distinct local namespace. Python oers us the global statement to change the namespace search rule.
global name
The global statement tells Python that the following names are part of the global namespace, not the local namespace. The following example shows two functions that share a global variable.
ratePerHour= 45.50 def cost( hours ): global ratePerHour return hours * ratePerHour def laborMaterials( hours, materials ): return cost(hours) + materials
Warning: Global Warning The global statement has a consequence of tightly coupling pieces of software. This can lead to diiculty in maintenance and enhancement of the program. Classes and modules provide better ways to assemble complex programs. As a general policy, we discourage use of the global statement.
11.3 Call By Value and Call By Reference

Beginning programmers can skip this section. This is a digression for experienced C and C++ programmers. Most programming languages have a formal mechanism for determining if a parameter receives a copy of the argument (call by value) or a reference to the argument object (call by name or call by reference.) 11.2. The global Statement 127
The distinction is important in languages with primitive types: data which is not a formal object. These primitive types can be eiciently passed by value, where ordinary objects are more eiciently passed by reference. Additionally, this allows a languge like C or C++ to use a reference to a variable as input to a function and have the function update the variable without an obvious assignment statement. Bad News. The following scenario is entirely hypothetical for Python programmers, but a very real problem for C and C++ programmers. Imagine we have a function to2() , with this kind of denition in C.
int to2( int *a ) { /* set parameter a's value to 2 */ *a= 2; return 0; }
This function changes the value of the variable a to 2. This would be termed a side-eect because it is in addition to any value the function might return normally. When we do the following in C
int x= 27; int z= to2( &x ); printf( "x=%i, z=%i", x, z );
We get the unpleasant side-eect that our function to2() has changed the argument variable, x, and the variable wasnt in an assignment statement! We merely called a function, using x as an argument. In C, the & operator is a hint that a variable might be changed. Further, the function denition should contain the keyword const when the reference is properly read-only. However, these are burdens placed on the programmer to assure that the program compiles correctly. Python Rules. In Python, the arguments to a function are always objects, never references to variables. Consider this Python version of the to2() function:
def to2( a ) a = 2 return 0 x = 27 z = to2( x ) print "x=%d, z=%d" % ( x, z )
The variable x is a reference to an integer object with a value of 27. The parameter variable (a) in the to2() function is a reference to the same object, and a is local to the functions scope. The original variable, x, cannot be changed by the function, and the original argument object, the integer 27, is immutable, and cant be changed either. If an argument value is a mutable object, the parameter is a reference to that object, and the function has access to methods of that object. The methods of the object can be called, but the original object cannot be replaced with a new object. Well look at mutable objects in Data Structures. For now, all the objects weve used (strings and numbers) are immutable and cannot be changed. The Python rules also mean that, in general, all variable updates must be done explicitly via an assignment statement. This makes variable changes perfectly clear.
128
Chapter 11. Additional Notes On Functions
11.4 Function Objects

One interesting consequence of the Python world-view is that a function is an object of the class function, a subclass of callable. The common feature that all callable objects share is that they have a very simple interface: they can be called. Other callable objects include the built-in functions, generator functions (which have the yield statement instead of the return statement) and things called lambdas. Sometimes we dont want to call and evaluate a function. Sometimes we want to do other things to or with a function. For example, the various factory functions (int(), long(), float(), complex()) can be used with the isinstance() function instead of being called to create a new object. For example, isinstance(2,int) has a value of True. It uses the int() function, but doesnt apply the int() function. A function object is created with the def statement. Primarily, we want to evaluate the function objects we create. However, because a function is an object, it has attributes, and it can be manipulated to a limited extent. From a syntax point of view, a name followed by () is a function call. You can think of the () as the call operator: they require evaluation of the arguments, then they apply the function.
name ( arguments )
There are a number of manipulations that you might want to do with a function object. Call The Function. By far, the most common use for a function object is to call it. When we follow a function name with (), we are calling the function: evaluating the arguments, and applying the function. Calling the function is the most common manipulation. Alias The Function. This is dangerous, because it can make a program obscure. However, it can also simplify the evoluation and enhancement of software. Heres a scenario. Imagine that the rst version of our program had two functions named rollDie() and rollDice(). The denitions might look like the following.
def rollDie(): return random.randrange(1,7) def rollDice(): return random.randrange(1,7) + random.randrange(1,7)
When we wanted to expand our program to handle ve-dice games, we realized we could generalize the rollDice() function to cover both cases.
def rollNDice( n=2 ): t= 0 for d in range(n): t += random.randrange( 1, 7 ) return t
It is important to remove the duplicated algorithm in all three versions of our dice rolling function. Since rollDie() and rollDice() are just special cases of rollNDice(). We can replace our original two functions with something like the following.
def rollDie(): return rollNDice( 1 ) def rollDice(): return rollNDice()
11.4. Function Objects
129
However, we have an alternative. This revised denition of rollDice() is really just an another name for the rollNDice(). Because a function is an object assigned to a variable, we can have multiple variables assigned to the function. Heres how we create an alias to a function.
def rollDie(): return rollNDice( 1 ) rollDice = rollNDice
Warning: Function Alias Confusion Function alias denitions helps maintaining compatibility between old and new releases of software. It is not something that should be done as a general practice; we need to be careful providing multiple names for a given function. This can be a simplication. It can also be a big performance improvement for certain types of functions that are heavily used deep within nested loops. Function Attributes. A function object has a number of attributes. We can interrogate those attributes, and to a limited extend, we can change some of these attributes. For more information, see section 3.2 of the Python Language Reference and section 2.3.9.3 of the Python Library Reference. func_doc __doc__ Docstring from the rst line of the functions body. func_name __name__ Function name from the def statement. __module__ Name of the module in which the function name was dened. func_defaults Tuple with default values to be assigned to each argument that has a default value. This is a subset of the parameters, starting with the rst parameter that has a default value. func_code The actual code object that is the suite of statements in the body of this function. func_globals The dictionary that denes the global namespace for the module that denes this function. This is m.__dict__ of the module which dened this function. func_dict __dict__ The dictionary that denes the local namespace for the attributes of this function. You can set and get your own function attributes, also. Heres an example
def rollDie(): return random.randrange(1,7) rollDie.version= "1.0" rollDie.authoor= "sfl"
130
Chapter 11. Additional Notes On Functions
Part III
Data Structures
131
The Data View

Computer programs are built with two essential features: data and processing. We started with processing elements of Python. Were about to start looking at data structures. In Language Basics, we introduced almost all of the procedural elements of the Python language. We started with expressions, looking at the various operators and data types available. We described fourteen of the approximately 24 statements that make up the Python language. Expression Statement. For example, a function evaluation where there is no return value. Examples include the print() function. import. Used to include a module into another module or program. print. Used to provide visible output. This is being replaced by the print() function. assignment. This includes the simple and augmented assignment statements. This is how you create variables. del. Used (rarely) to remove a variable, function, module or other object. if. Used to conditionally perform suites of statements. This includes elif and else statements. pass. This does nothing, but is a necessary syntactic placeholder for an if or while suite that is empty. assert. Used to conrm the program is in the expected state. for and while. Perform suites of statements using a sequence of values or while a condition is held true. break and continue. Helpful statements for short-cutting loop execution. def. Used to dene a new function. return. Used to exit a function. Provides the return value from the function. global. Used adjust the scoping rules, allowing local access to global names. We discourage its use in The global Statement . The Other Side of the Coin. The next chapters focus on adding various data types to the basic Python language. The subject of data representation and data structures is possibly the most profound part of computer programming. Most of the killer applications email, the world wide web, relational databases are basically programs to create, read and transmit complex data structures. We will make extensive use of the object classes that are built-in to Python. This experience will help us design our own object classes in Data + Processing = Objects. Well work our way through the following data structures. Sequences. In Sequences: Strings, Tuples and Lists well extend our knowledge of data types to include an overview various kinds of sequences: strings, tuples and lists. Sequences are collections of objects accessed by their numeric position within the collection. In Strings we describe the string subclass of sequence. The exercises include some challenging string manipulations. We describe xed-length sequences, called tuple s in Tuples. In Lists we describe the variable-length sequence, called a list. This list sequence is one of the powerful features that sets Python apart from other programming languages. The exercises at the end of the list section include both simple and relatively sophisticated problems. Mappings. In Mappings and Dictionaries we describe mappings and dictionary objects, called dict. Well show how dictionaries are part of some advanced techniques for handling arguments to functions. Mappings are collections of value objects that are accessed by key objects. 133
Sets. Well cover set objects in Sets. Sets are simple collections of unique objects with no additional kind of access. Exceptions. Well cover exception objects in Exceptions. Well also show the exception handling statements, including try, except, nally and raise statements. Exceptions are both simple data objects and events that control the execution of our programs. Iterables. The yield statement is a variation on return that simplies certain kinds of generator algorithms that process or create create iterable data structures. We can iterate through almost any kind of data collection. We can also dene our own unique or specialized iterations. Well cover this in Iterators and Generators. Files. The subject of les is so vast, that well introduce le objects in Files. The with statement is particularly helpful when working with les. Files are so centrally important that well return les in Components, Modules and Packages. Well look at several of the le-related modules in File Handling Modules as well as File Formats: CSV, Tab, XML, Logs and Others.. In Functional Programming with Collections we describe more advanced sequence techniques, including multi-dimensional processing, additional sequence-processing functions, and sorting. Deferred Topics. There are a few topics that need to be deferred until later. try. Well look at exceptions in Exceptions. This will include the except, nally and raise statements, also. yield. Well look at Generator Functions in Iterators and Generators. class. Well cover this in its own part, Classes. with. Well look at Context Managers in Managing Contexts: the with Statement . import. Well revisit import in detail in Components, Modules and Packages. exec. Additionally, well cover the exec statement in The exec Statement .
134
CHAPTER
TWELVE
SEQUENCES: STRINGS, TUPLES AND LISTS

The Common Features of Sequences
Before digging into the details, well introduce the common features of three of the data types that are containers for sequences of values. In Sequence Semantics we will provide an overview of the semantics of sequences. We describes the common features of the sequences in Overview of Sequences. The sequence is central to programming and central to Python. A number of statements and functions we have covered have sequence-related features that we have glossed over, danced around, and generally avoided. Well revisit a number of functions and statements we covered in previous sections, and add the power of sequences to them. In particular, the for statement is something we glossed over in Iterative Processing: For All and There Exists. In the chapters that follow well look at Strings, Tuples and Lists in detail. In Mappings and Dictionaries , well introduce another structured data type for manipulating mappings between keys and values.
12.1 Sequence Semantics

A sequence is a container of objects which are kept in a specic order. We can identify the individual objects in a sequence by their position or index. Positions are numbered from zero in Python; the element at index zero is the rst element. We call these containers because they are a single object which contains (or collects) any number of other objects. The any number clause means that they can contain zero other objects, meaning that an empty container is just as valid as a container with one or thousands of objects. Important: Other Languages In some programming languages, they use words like vector or array to refer to sequential containers. For example, in C or Java, the primitive array has a statically allocated number of positions. In Java, a reference outside that specic number of positions raises an exception. In C, however, a reference outside the dened positions of an array is an error that may never be detected. Really. There are four commonly-used subspecies of sequence containers. String, called str. A container of single-byte ASCII characters. Unicode String, unicode. A container of multi-byte Unicode (or Universal Character Set) characters.
135
tuple. A container of anything with a xed number of elements. list. A container of anything with a dynamic number of elements. Important: Python 3 This mix of types will change slightly. The String and Unicode types will merge into the str type. This will represent text. A new container, the byte array will be introduced, named bytes. This will represent binary data. tuple and list wont change. When we create a tuple or string , weve created an immutable, or static object. We can examine the object, looking at specic characters or items. We cant change the object. This means that we cant put additional data on the end of a string. What we can do, however, is create a new string that is the concatenation of the two original string objects. When we create a list, on the other hand, weve created a mutable object. A list can have additional objects appended to it or inserted in it. Objects can be removed from a list, also. A list can grow and shrink; the order of the objects in the list can be changed without creating a new list object. One other note on string. While string are sequences of characters, there is no separate character data type. A character is simply a string of length one. This relieves programmers from the C or Java burden of remembering which quotes to use for single characters as distinct from multi-character string. It also eliminates any problems when dealing with Unicode multi-byte characters.
12.2 Overview of Sequences

All the varieties of sequences (string, tuple and list) have some common characteristics. Well identify the common features rst, and then move on to cover these in detail for each individual type of sequence. This section is a road-map for the following three sections that cover string, tuple and list in detail. Literal Values. Each sequence type has a literal representation. The details will be covered in separate sections, but the basics are these: string uses quotes: "string". tuple uses (): (1,'b',3.1). list uses []: [1,'b',3.1]. Operations. Sequences have three common operations: + will concatenate sequences to make longer sequences. * is used with a number and a sequence to repeat the sequence several times. Finally, the [ ] operator is used to select elements from a sequence. The [ ] operator can extract a single item, or a subset of items by slicing. There are two forms of []. The single item format is sequence [ index ]. Items are numbered from 0. The slice format is sequence [ start : end ]. Items from start to end -1 are chosen to create a new sequence; it will be a slice of the original sequence. There will be end start items in the resulting sequence. Positions can be numbered from the end of the string as well as the beginning. Position -1 is the last item of the sequence, -2 is the next-to-last item. Heres how it works: each item has a positive number position that identies the item in the sequence. Well also show the negative position numbers for each item in the sequence. For this example, were looking at a four-element sequence like the tuple (3.14159,"two words",2048,(1+2j)) .
136
Chapter 12. Sequences: Strings, Tuples and Lists
forward position reverse position item
0 -4 3.14159
1 -3 two words
2 -2 2048
3 -1 (1+2j)
Why do we have two dierent ways of identifying each position in the sequence? If you want, you can think of it as a handy short-hand. The last item in any sequence, S can be identied by the formula S[ len(S)-1 ] . For example, if we have a sequence with 4 elements, the last item is in position 3. Rather than write S[ len(S)-1 ], Python lets us simplify this to S[-1] . You can see how this works with the following example.
>>> a=(3.14159,"two words",2048,(1+2j)) >>> a[0] 3.1415899999999999 >>> a[-3] 'two words' >>> a[2] 2048 >>> a[-1] (1+2j)
Built-in Functions. len(), max() and min() apply to all varieties of sequences. Well provide the denitions here and refer to them in various class denitions. len(sized_collection) Return the number of items of the collection. This can be any kind of sized collection. All sequences and mappings are subclasses of collections.Sized and provide a length. Here are some examples.
>>> len("Wednesday") 9 >>> len( (1,1,2,3) ) 4
max(iterable_collection) Returns the largest value in the iterable collection. All sequences and mappings are subclasses of collections.Iterable; the max() function can iterate over elements and locate the largest.
>>> max( (1,2,3) ) 3 >>> max('abstractly') 'y'
Note that max() can also work with a number of individual arguments instead of a single iterable collection argument value. We looked a this in Collection Functions. min(iterable_collection) Returns the smallest value in the iterable collection. All sequences and mappings are subclasses of collections.Iterable; the max() function can iterate over elements and locate the smallest.
>>> min( (10,11,2) ) 2 >>> min( ('10','11','2') ) '10'
Note that strings are compared alphabetically. The min() (and max() function cant determine that these are supposed to be evaluated as numbers.)
12.2. Overview of Sequences
137
sum(iterable_collection, [start=0] ) Return the sum of the items in the iterable collection. All sequences and mappings are subclasses of collections.Iterable. If start is provided, this is the initial value for the sum, otherwise 0 is used. If the values being summed are not all numeric values, this will raise a TypeError exception.
>>> sum( (1,1,2,3,5,8) ) 20 >>> sum( (), 3 ) 3 >>> sum( (1,2,'not good') ) Traceback (most recent call last): File "<stdin>", line 1, in <module> TypeError: unsupported operand type(s) for +: 'int' and 'str'
any(iterable_collection) Return True if there exists an item in the iterable collection which is True. All sequences and mappings are subclasses of collections.Iterable. all(iterable_collection) Return True if all items in the iterable collection are True. All sequences and mappings are subclasses of collections.Iterable. enumerate(iterable_collection) Iterates through the iterable collection returning 2-tuples of ( index, item ).
>>> for position, item in enumerate( ('word',3.1415629,(2+3j) ) ): ... print position, item ... 0 word 1 3.1415629 2 (2+3j)
sorted(sequence, [key=None], [reverse=False] ) This returns an iterator that steps through the elements of the iterable container in ascending order. If the reverse keyword parameter is provided and set to True, the container is iterated in descending order. The key parameter is used when the items in the container arent simply sorted using the default comparison operators. The key function must return the elds to be compared selected from the underlying objects in the tuple. Well look at this in detail in Functional Programming with Collections. reversed(sequence) This returns an iterator that steps through the elements in the iterable container in reverse order.
>>> tuple( reversed( (9,1,8,2,7,3) ) ) (3, 7, 2, 8, 1, 9)
Comparisons. The standard comparisons (<, <=, >, <=, ==, !=) apply to sequences. These all work by doing item-by-item comparison within the two sequences. The item-by-item rule results in strings being sorted alphabetically, and tuples and lists sorted in a way that is similar to strings. There are two additional comparisons: in and not in. These check to see if a single value occurs in the sequence. The in operator returns a True if the item is found, False if the item is not found. The not in operator returns True if the item is not found in the sequence. 138 Chapter 12. Sequences: Strings, Tuples and Lists
Methods. The string and list classes have method functions that operate on the objects value. For instance "abc".upper() executes the upper() method belonging to the string literal "abc". The result is a new string, 'ABC'. The exact dictionary of methods is unique to each class of sequences. Statements. The tuple and list classes are central to certain Python statements, like the assignment statement and the for statement. These were details that we skipped over in The Assignment Statement and Iterative Processing: For All and There Exists. Modules. There is a string module with several string specic functions. Most of these functions are now member functions of the string type. Additionally, this module has a number of constants to dene various subsets of the ASCII character set, including digits, printable characters, whitespace characters and others. Factory Functions. There are also built-in factory (or conversion) functions for the sequence objects. Weve looked at some of these already, when we looked at str() and repr().
12.3 Exercises
1. Tuples and Lists. What is the value in having both immutable sequences (tuple) and mutable sequences (list)? What are the circumstances under which you would want to change a string? What are the problems associated with a string that grows in length? How can storage for variable length string be managed? 2. Unicode Strings. What is the value in making a distinction between Unicode strings and ASCII strings? Does it improve performance to restrict a string to single-byte characters? Should all strings simply be Unicode strings to make programs simpler? How should le reading and writing be handled? 3. Statements and Data Structures. In order to introduce the for statement in Iterative Processing: For All and There Exists, we had to dance around the sequence issue. Would it make more sense to introduce the various sequence data structures rst, and then describe statements that process the data structure later? Something has to be covered rst, and is therefore more fundamental. Is the processing statement more fundamental to programming, or is the data structure?
12.4 Style Notes

Try to avoid extraneous spaces in list and tuple displays. Python programs should be relatively compact. Prose writing typically keeps ()s close to their contents, and puts spaces after commas, never before them. This should hold true for Python, also. The preferred formatting for a list or tuple, then, is [1,2,3] or (1, 2, 3). Spaces are not put after the initial [ or (. Spaces are not put before ,.
12.3. Exercises
139
140
Chapter 12. Sequences: Strings, Tuples and Lists
CHAPTER
THIRTEEN
STRINGS
Well look at the two string classes from a number of viewpoints: semantics, literal values, operations, comparison operators, built-in functions, methods and modules. Additionally, we have a digression on the immutability of string objects.
13.1 String Semantics

A String (the formal class name is str) is an immutable sequence of ASCII characters. A Unicode String (unicode) is an immutable sequence of Unicode characters. Since a string (either str or unicode) is a sequence, all of the common operations on sequences apply. We can concatenate string objects together and select characters from a string. When we select a slice from a string, weve extracted a substring. An individual character is simply a string of length one. Important: Python 3.0 The Python 2 str class, which is limited to single-byte ASCII characters does two separate things: it represents text as well as a collection of bytes. The text features of str gain the features from the Unicode String class, unicode. The new str class will represent strings of text, irrespective of the underlying encoding. It can be ASCII, UTF-8, UTF-16 or any other encoding. The array of bytes features of the Python 2 str class will be moved into a new class, bytes. This new class will implement simple sequences of bytes and will support conversion between bytes and strings using encoding and decoding functions.
13.2 String Literal Values

A str is a sequence of ASCII characters. The literal value for a str is written by surrounding the value with quotes or apostrophes. There are several variations to provide some additional features. Basic String Strings are enclosed in matching quotes (") or apostrophes ('). A string enclosed in quotes (") can contain apostrophes ('); similarly, a string enclosed in apostrophes (') can contains quotes ("). A basic str must be completed on a single line, or continued with a \ as the very last character of a line. Examples:
141
"consultive" 'syncopated' "don't do that" '"Okay," he said.'
Multi-Line String Also called Triple-Quoted String. A multi-line str is enclosed in triple quotes (""") or triple apostrophes ('''). It continues on across line boundaries until the concluding triple-quote or triple-apostrophe. Examples:
"""A very long string""" '''SELECT * FROM THIS, THAT WHERE THIS.KEY = THAT.FK AND THIS.CODE = 'Active' '''
Unicode String A Unicode String uses the above quoting rules, but prefaces the quote with (u"), (u'), (u""") or (u'''). Unicode is the Universal Character Set; each character requires from 1 to 4 bytes of storage. ASCII is a single-byte character set; each of the 256 ASCII characters requires a single byte of storage. Unicode permits any character in any of the languages in common use around the world. A special \uxxxx escape sequence is used for Unicode characters that dont happen to occur on your ASCII keyboard. Examples:
u'\u65e5\u672c' u"All ASCII"
Raw String A Raw String uses the above quoting rules, but prefaces the quote with (r"), (r'), (r""") or (r'''). The backslash characters (\) are not interpreted as escapes by Python, but are left as is. This is handy for Windows les names that contain \. It is also handy for regular expressions that make extensive use of backslashes. Examples:
newline_literal= r'\n' filename= "C:\mumbo\jumbo" pattern= "(\*\S+\*)"
The newline_literal is a two character string, not the newline character. Outside of raw strings, non-printing characters and Unicode characters that arent found on your keyboard are created using escapes. A table of escapes is provided below. These are Python representations for unprintable ASCII characters. Theyre called escapes because the \ is an escape from the usual meaning of the following character.
142
Chapter 13. Strings
Escape \\ \' \" \a \b \f \n \r \t \ooo \xhh
Meaning Backslash (\) Apostrophe (') Quote (") Audible Signal; the ASCII code called BEL. Some OSs translate this to a screen ash or ignore it completely. Backspace (ASCII BS) Formfeed (ASCII FF). On a paper-based printer, this would move to the top of the next page. Linefeed (ASCII LF), also known as newline. This would move the paper up one line. Carriage Return (ASCII CR). On a paper based printer, this returned the print carriage to the start of the line. Horizontal Tab (ASCII TAB) An ASCII character with the given octal value. The ooo is any octal number. An ASCII character with the given hexadecimal value. The x is required. The hh is any hex number.
Adjacent Strings. Note that adjacent string objects are automatically concatenated to make a single string. "ab" "cd" "ef" is the same as "abcdef". The most common use for this is the following:
msg = "A very long" \ "message, which didn't fit on" \ "one line."
Unicode Characters. For Unicode, a special \uxxxx escape is provided. This requires the four digit Unicode character identication. For example, is made up of Unicode characters U+65e5 and U+672c. In Python, we write this string as u'\u65e5\u672c'. There are a variety of Unicode encoding schemes, for example, UTF-8, UTF-16 and LATIN-1. The codecs module provides mechanisms for encoding and decoding Unicode Strings.
13.3 String Operations

There are a number of operations on str objects, operations which create strs and operations which create other objects from strs. There are three operations (+ , * , [ ]) that work with all sequences (including strs) and a unique operation, %, that can be performed only with str objects. The + operator creates a new string as the concatenation of the arguments.
>>> "hi " + 'mom' 'hi mom'
The * operator between str and numbers (number * str or str * number) creates a new str that is a number of repetitions of the input str.
>>> print 3*"cool!" cool!cool!cool!
13.3. String Operations
143
The [ ] operator can extract a single character or a slice from the string. There are two forms: the single-item form and the slice form. The single item format is string [ index ]. Characters are numbered from 0 to len(string). Characters are also numbered in reverse from -len(string) to -1. The slice format is string [ start : end ]. Characters from start to end -1 are chosen to create a new str as a slice of the original str; there will be end start characters in the resulting str. If start is omitted it is the beginning of the string (position 0). If end is omitted it is the end of the string (position -1). Yes, you can omit both (someString[:]) to make a copy of a string.
>>> s="adenosine" >>> s[2] 'e' >>> s[:5] 'adeno' >>> s[5:] 'sine' >>> s[-5:] 'osine' >>> s[:-5] 'aden'
The String Formatting Operation, %. The % operator is sometimes called string interpolation, since it interpolates literal text and converted values. We prefer to call it string formatting, since that is a more apt description. Much of the formatting is taken straight from the C librarys printf() function. This operator has three forms. You can use % with a str and value, str and a tuple as well as str and classname:dict. Well cover tuple and dict in detail later. The string on the left-hand side of % contains a mixture of literal text plus conversion specications. A conversion specication begins with %. For example, integers are converted with %i. Each conversion specication will use a corresponding value from the tuple. The rst conversion uses the rst value of the tuple, the second conversion uses the second value from the tuple. For example:
import random d1, d2 = random.randrange(1,6), random.randrange(1,6) r= "die 1 shows %i, and die 2 shows %i" % ( d1, d2 )
The rst %i will convert the value for d1 to a string and insert the value, the second %i will convert the value for d2 to a string. The % operator returns the new string based on the format, with each conversion specication replaced with the appropriate values. Conversion Specications. Each conversion specication has from one to four elements, following this pattern: %.
[ flags ][ width [ precision ]] code
The % and the nal code in each conversion specication are required. The other elements are optional. The optional ags element can have any combination of the following values: - Left adjust the converted value in a eld that has a length given by the width element. The default is right adjustment.
144
Chapter 13. Strings
+ Show positive signs (sign will be + or -). The default is to show negative signs only. (a space) Show positive signs with a space (sign will be or -). The default is negative signs only. # Use the Python literal rules (0 for octal, 0x for hexadecimal, etc.) The default is decoration-free notation. 0 Zero-ll the the eld that has a length given by the width element. The default is to space-ll the eld. This doesnt make a lot of sense with the - (left-adjust) ag. The optional width element is a number that species the total number of characters for the eld, including signs and decimal points. If omitted, the width is just big enough to hold the output number. If a * is used instead of a number, an item from the tuple of values is used as the width of the eld. For example, "%*i" % ( 3, d1 ) uses the value 3 from the tuple as the eld width and d1 as the value to convert to a string. The optional precision element (which must be preceded by a dot, . if it is present) has a few dierent purposes. For numeric conversions, this is the number of digits to the right of the decimal point. For string conversions, this is the maximum number of characters to be printed, longer string s will be truncated. If a * is used instead of a number, an item from the tuple of values is used as the precision of the conversion. For example, "%*.*f" % ( 6, 2, avg ) uses the value 6 from the tuple as the eld width, the value 2 from the tuple as the precision and avg as the value. The standard conversion rules also permit a long or short indicator: l or h. These are tolerated by Python so that these formats will be compatible with C, but they have no eect. They reect internal representation considerations for C programming, not external formatting of the data. The required one-letter code element species the conversion to perform. The codes are listed below. % Not a conversion, this creates a % in the resulting str. Use %% to put a % in the output str. c Convert a single-character str. This will also convert an integer value to the corresponding ASCII character. For example, "%c" % ( 65, ) results in "A". s Convert a str. This will convert non- str objects by implicitly calling the str() function. r Call the repr() function, and insert that value. i d Convert a numeric value, showing ordinary decimal output. The code i stands for integer, d stands for decimal. They mean the same thing; but its hard to reach a consensus on which is correct. u Convert an unsigned number. While relevant to C programming, this is the same as the i or d format conversion. o Convert a numeric value, showing the octal representation. %#0 gets the Python-style value with a leading zero. This is similar to the oct() function. x X Convert a numeric value, showing the hexadecimal representation. %#X gets the Python-style value with a leading 0X; %#x gets the Python-style value with a leading 0x. This is similar to the hex() function. e E Convert a numeric value, showing scientic notation. %e produces d.ddd e xx, %E produces d.ddd E xx. f F Convert a numeric value, using ordinary decimal notation. In case the number is gigantic, this will switch to %g or %G notation. g G Generic oating-point conversion. For values with an exponent larger than -4, and smaller than the precision element, the %f format will be used. For values with an exponent smaller than -4, or values larger than the precision element, the %e or %E format will be used. Here are some examples.
"%i: %i win, %i loss, %6.3f " % (count,win,loss,float(win)/loss)
13.3. String Operations
145
This example does four conversions: three simple integer and one oating point that provides a width of 6 and 3 digits of precision. -0.000 is the expected format. The rest of the string is literally included in the output.
"Spin %3i: %2i, %s" % (spin,number,color)
This example does three conversions: one number is converted into a eld with a width of 3, another converted with a width of 2, and a string is converted, using as much space as the string requires.
>>> a=6.02E23 >>> "%e" % a '6.020000e+23' >>> "%E " % a '6.020000E+23' >>>
This example shows simple conversion of a oating-point number to the default scientic notation which has a witdth of 12 and a precision of 6.
13.4 String Comparison Operations

The standard comparisons (<, <=, >, >=, ==, !=) apply to str objects. These comparisons use the standard character-by-character comparison rules for ASCII or Unicode. There are two additional comparisons: in and not in. These check to see if a substring occurs in a longer string. The in operator returns a True when the substring is found, False if the substring is not found. The not in operator returns True if the substring is not found.
>>> 'a' in 'xyzzyabcxyzzy' True >>> 'abc' in 'xyzzyabc' True
Dont be fooled by the fact that string representations of integers dont seem to sort properly. String comparison does not magically recornize that the strings are representations of numbers. Its simple alphabetical order rules applied to digits.
>>> '100' < '25' True
This is true because '1' < '2'.
13.5 String Statements

The for statement will step though all elements of a sequence. In the case of a string, it will step through each character of the string. For example:
for letter in "forestland": print letter
This will print each letter of the given string.
146
Chapter 13. Strings
13.6 String Built-in Functions

The following built-in functions are relevant to str manipulation chr(i ) Return a str of one character with ordinal i. Note that 0 i < 256 to be a proper ASCII character. unichr(u) Return a Unicode String (unicode) of one character with ordinal u. 0 u < 65536. ord(c) Return the integer ordinal of a one character str. This works for any character, including Unicode characters. unicode(string, [encoding], [errors] ) Creates a new Unicode object from the given encoded string. encoding defaults to the current default string encoding. errors denes the error handling, defaults to strict. The unicode() function converts the string to a specic Unicode external representation. The default encoding is UTF-8 with strict error handling. Choices for errors are strict, replace and ignore. Strict raises an exception for unrecognized characters, replace substitutes the Unicode replacement character ( \uFFFD ) and ignore skips over invalid characters. The codecs and unicodedata modules provide more functions for working with Unicode.
>>> unicode("hi mom","UTF-16") u'\u6968\u6d20\u6d6f' >>> unicode("hi mom","UTF-8") u'hi mom'
Important: Python 3 The ord(), chr(), unichr() and unicode() functions will be simplied in Python 3. Python 3 no longer separates ASCII from Unicode strings. These functions will all implicitly work with Unicode strings. Note that the UTF-8 encoding of Unicode overlaps with ASCII, so this simplication to use Unicode will not signicantly disrupt programs that work ASCII les. Several important functions were dened earlier in String Conversion Functions. repr(). Returns a canonical string representation of the object. eval(repr(object)) == object. For most object types,
For simple numeric types, the result of repr() isnt very interesting. For more complex, types, however, it often reveals details of their structure.
>>> ... ... ... >>> "'a a="""a very long string in multiple lines """ repr(a) very \\nlong string \\nin multiple lines\\n'"
This representation shows the newline characters ( \n ) embedded within the triple-quoted string. Important: Python 3 The reverse quotes (`a`) work like repr(a). The reverse quote syntax is rarely used, and will be dropped in Python 3.
13.6. String Built-in Functions
147
str(). Return a nice string representation of the object. If the argument is a string, the return value is the same object.
>>> a= str(355.0/113.0) >>> a '3.14159292035' >>> len(a) 13
Some other functions which apply to strings as well as other sequence objects. len(). For strings, this function returns the number of characters.
>>> len("abcdefg") 7 >>> len(r"\n") 2 >>> len("\n") 1
max(). For strings, this function returns the maximum character. min(). For strings, this function returns the minimum character. sorted(). Iterate through the strings characters in sorted order. This expands the string into an explicit list of individual characters.
>>> sorted( "malapertly" ) ['a', 'a', 'e', 'l', 'l', 'm', 'p', 'r', 't', 'y'] >>> "".join( sorted( "malapertly" ) ) 'aaellmprty'
reversed(). Iterate through the strings characters in reverse order. This creates an iterator. The iterator can be used with a variety of functions or statements.
>>> reversed( "malapertly" ) <reversed object at 0x600230> >>> "".join( reversed( "malapertly" ) 'yltrepalam'
13.7 String Methods

A string object has a number of method functions. These can be grouped arbitrarily into transformations, which create new string s from old, and information, which returns a fact about a string. The following string transformation functions create a new string object from an existing string. capitalize() Create a copy of the string with only its rst character capitalized. center(width) Create a copy of the string centered in a string of length width. Padding is done using spaces. encode(encoding, [errors] ) Return an encoded version of string. Default encoding is the current default string encoding. errors may be given to set a dierent error handling scheme. Default is strict meaning that encoding errors raise a ValueError. Other possible values are ignore and replace. 148 Chapter 13. Strings
expandtabs([tabsize] ) Return a copy of string where all tab characters are expanded using spaces. If tabsize is not given, a tab size of 8 characters is assumed. join(sequence) Return a string which is the concatenation of the strings in the sequence. Each separator between elements is a copy of the given string object. ljust(width) Return a copy of string left justied in a string of length width. Padding is done using spaces. lower() Return a copy of string converted to lowercase. lstrip() Return a copy of string with leading whitespace removed. replace(old, new, [maxsplit] ) Return a copy of string with all occurrences of substring old replaced by new. If the optional argument maxsplit is given, only the rst maxsplit occurrences are replaced. rjust(width) Return a copy of string right justied in a string of length width. Padding is done using spaces. rstrip() Return a copy of string with trailing whitespace removed. strip() Return a copy of string with leading and trailing whitespace removed. swapcase() Return a copy of string with uppercase characters converted to lowercase and vice versa. title() Return a copy of string with words starting with uppercase characters, all remaining characters in lowercase. translate(table, [deletechars] ) Return a copy of the string, where all characters occurring in the optional argument deletechars are removed, and the remaining characters have been mapped through the given translation table. The table must be a string of length 256, providing a translation for each 1-byte ASCII character. The translation tables are built using the string.maketrans() function in the string module. upper() Return a copy of string converted to uppercase. The following accessor methods provide information about a string. count(sub, [start], [end] ) Return the number of occurrences of substring sub in string. If start or end are present, these have the same meanings as a slice string[start:end]. endswith(suix, [start], [end] ) Return True if string ends with the specied suffix, otherwise return False. The suix can be a single string or a sequence of individual strings. If start or end are present, these have the same meanings as a slice string[start:end]. find(sub, [start], [end] ) Return the lowest index in string where substring sub is found. Return -1 if the substring is not found. If start or end are present, these have the same meanings as a slice string[start:end].
13.7. String Methods
149
index(sub, [start], [end] ) Return the lowest index in string where substring sub is found. Raise ValueError if the substring is not found. If start or end are present, these have the same meanings as a slice string[start:end]. isalnum() Return True if all characters in string are alphanumeric and there is at least one character in string; False otherwise. isalpha() Return True if all characters in string are alphabetic and there is at least one character in string; False otherwise. isdigit() Return True if all characters in string are digits and there is at least one character in string; False otherwise. islower() Return True if all characters in string are lowercase and there is at least one cased character in string; False otherwise. isspace() Return True if all characters in string are whitespace and there is at least one character in string, False otherwise. istitle() Return True if string is titlecased. Uppercase characters may only follow uncased characters (whitespace, punctuation, etc.) and lowercase characters only cased ones, False otherwise. isupper() Return True if all characters in string are uppercase and there is at least one cased character in string; False otherwise. rfind(sub, [start], [end] ) Return the highest index in string where substring sub is found. Return -1 if the substring is not found. If start or end are present, these have the same meanings as a slice string[start:end]. rindex(sub, [start], [end] ) Return the highest index in string where substring sub is found. Raise ValueError if the substring is not found.. If start or end are present, these have the same meanings as a slice string[start:end]. startswith(sub, [start], [end] ) Return True if string starts with the specied prefix, otherwise return False. The prex can be a single string or a sequence of individual strings. If start or end are present, these have the same meanings as a slice string[start:end]. The following generators create another kind of object, usually a sequence, from a string. partition(separator ) Return three values: the text prior to the rst occurance of separator in string, the sep as the delimiter, and the text after the rst occurance of the separator. If the separator doesnt occur, all of the input string is in the rst element of the 3-tuple; the other two elements are empty strings. split(separator, [maxsplit] ) Return a list of the words in the string, using separator as the delimiter. If maxsplit is given, at most maxsplit splits are done. If separator is not specied, any whitespace characater is a separator. splitlines(keepends) Return a list of the lines in string, breaking at line boundaries. Line breaks are not included in the resulting list unless keepends is given and set to True.
150
Chapter 13. Strings
13.8 String Modules

There is an older module named string. Almost all of the functions in this module are directly available as methods of the string type. The one remaining function of value is the maketrans() function, which creates a translation table to be used by the translate() method of a string. maketrans(from, to) Return a translation table (a string 256 characters long) suitable for use in str.translate(). The from and to parameters must be strings of the same length. The table will assure that each character in from is mapped to the character in the same position in to. The following example shows how to make and then apply a translation table.
>>> import string >>> t= string.maketrans("aeiou","xxxxx") >>> phrase= "now is the time for all good men to come to the aid of their party" >>> phrase.translate( t ) 'nxw xs thx txmx fxr xll gxxd mxn tx cxmx tx thx xxd xf thxxr pxrty'
The codecs module takes a dierent approach and has a number of built-in translations. More importantly, this module contains a number of denitions of the characters in the ASCII character set. These denitions serve as a central, formal repository for facts about the character set. Note that there are general denitions, applicable to Unicode character setts, dierent from the ASCII denitions. ascii_letters The set of all letters, ascii_uppercase. ascii_lowercase The lowercase 'abcdefghijklmnopqrstuvwxyz' ascii_uppercase The uppercase 'ABCDEFGHIJKLMNOPQRSTUVWXYZ' essentially a union of in in the the ascii_lowercase and character character set: set:
letters letters
ASCII ASCII
digits The digits used to make decimal numbers: '0123456789' hexdigits The digits used to make hexadecimal numbers: '0123456789abcdefABCDEF' letters This is the set of all letters, a union of lowercase and uppercase, which depends on the setting of the locale on your system. lowercase This is the set of lowercase letters, and depends on the setting of the locale on your system. octdigits The digits used to make octal numbers: '01234567' printable All printable characters in the character set. This is a union of digits, letters, punctuation and whitespace. punctuation All punctuation in the !"#$%&'()*+,-./:;<=>?@[\]^_`{|}~ ASCII character set, this is
uppercase This is the set of uppercase letters, and depends on the setting of the locale on your system. whitespace A collection of characters that cause spacing to happen. '\t\n\x0b\x0c\r' For ASCII this is
13.8. String Modules
151
13.9 String Exercises

1. Check Amount Writing. Translate a number into the English phrase. This example algorithm fragment is only to get you started. This shows how to pick o the digits from the right end of a number and assemble a resulting string from the left end of the string. Note that the right-most two digits have special names, requiring some additional cases above and beyond the simplistic loop shown below. For example, 291 is two hundred ninety one, where 29 is twenty nine. The word for 2 changes, depending on the context. As a practical matter, you should analyze the number by taking o three digits at a time, the expression (number % 1000) does this. You would then format the three digit number with words like million, thousand, etc.
English Words For An Amount, n

(a) Initialization. Set result Set tc 0. This is the tens counter that shows what position were examining. (b) Loop. While n > 0. i. Get Right Digit. Set digit n%10, the remainder when divided by 10. ii. Make Phrase. Translate digit to a string from zero to nine. Translate tc to a string from to thousand. This is tricky because the teens are special, where the hundreds and thousands are pretty simple. iii. Assemble Result. Prepend digit string and tc string to the left end of the result string. iv. Next Digit. n n 10. Be sure to use the // integer division operator, or youll get oating-point results. Increment tc by 1. (c) Result. Return result as the English translation of n. 2. Roman Numerals. This is similar to translating numbers to English. Instead we will translate them to Roman Numerals. The Algorithm is similar to Check Amount Writing (above). You will pick o successive digits, using %10 and /10 to gather the digits from right to left. The rules for Roman Numerals involve using four pairs of symbols for ones and ve, tens and fties, hundreds and ve hundreds. An additional symbol for thousands covers all the relevant bases. When a number is followed by the same or smaller number, it means addition. II is two 1s = 2. VI is 5 + 1 = 6. When one number is followed by a larger number, it means subtraction. IX is 1 before 10 = 9. IIX isnt allowed, this would be VIII. For numbers from 1 to 9, the symbols are I and V, and the coding works like this. (a) I (b) II
152
Chapter 13. Strings
(c) III (d) IV (e) V (f) VI (g) VII (h) VIII (i) IX The same rules work for numbers from 10 to 90, using X and L. For numbers from 100 to 900, using the symbols C and D. For numbers between 1000 and 4000, using M. Here are some examples. 1994 = MCMXCIV, 1956 = MCMLVI, 3888= MMMDCCCLXXXVIII 3. Word Lengths. Analyze the following block of text. Youll want to break into into words on whitespace boundaries. Then youll need to discard all punctuation from before, after or within a word. Whats left will be a sequence of words composed of ASCII letters. Compute the length of each word, and produce the sequence of digits. (no word is 10 or more letters long.) Compare the sequence of word lenghts with the value of math.pi.
Poe, E. Near a Raven Midnights so dreary, tired and weary, Silently pondering volumes extolling all by-now obsolete lore. During my rather long nap - the weirdest tap! An ominous vibrating sound disturbing my chamber's antedoor. "This", I whispered quietly, "I ignore".
This is based on http://www.cadaeic.net/cadenza.htm.
13.10 Digression on Immutability of Strings

In Strings and Tuples we noted that string and tuple objects are immutable. They cannot be changed once they are created. Programmers experienced in other languages sometimes nd this to be an odd restriction. Two common questions that arise are how to expand a string and how to remove characters from a string. Generally, we dont expand or contract a string, we create a new string that is the concatenation of the original string objects. For example:
>>> a="abc" >>> a=a+"def" >>> a 'abcdef'
In eect, Python gives us string objects of arbitrary size. It does this by dynamically creating a new string instead of modifying an existing string. Some programmers who have extensive experience in other languages will ask if creating a new string from the original string is the most eicient way to accomplish this. Or they suggest that it would be
13.10. Digression on Immutability of Strings
153
simpler to allow a mutable string for this kind of concatenation. The short answer is that Pythons storage management makes this use of immutable string the simplest and most eicient. Responses to the immutability of tuple and mutability of list vary, including some of the following frequently asked questions. Since a list does everything a tuple does and is mutable, why bother with tuple? Immutable tuple objects are more eicient than variable-length list objects for some operations. Once the tuple is created, it can only be examined. When it is no longer referenced, the normal Python garbage collection will release the storage for the tuple. Most importantly, a tuple can be reliably hashed to a single value. This makes it a usable key for a mapping. Many applications rely on xed-length tuples. A program that works with coordinate geometry in two dimensions may use 2-tuples to represent (x, y) coordinate pairs. Another example might be a program that works with colors as 3-tuples, (r, g, b), of red, green and blue levels. A variable-length list is not appropriate for these kinds of xed-length tuple. Wouldnt it be more eicient to allow mutable string s? There are a number of axes for eiciency: the two most common are time and memory use. A mutable string could use less memory. However, this is only true in the benign special case where we are only replacing or shrinking the string within a xed-size buer. If the string expands beyond the size of the buer the program must either crash with an exception, or it must switch to dynamic memory allocation. Python simply uses dynamic memory allocation from the start. C programs often have serious security problems created by attempting to access memory outside of a string buer. Python avoids this problem by using dynamic allocation of immutable string objects. Processing a mutable string could use less time. In the cases of changing a string in place or removing characters from a string, a xed-length buer would require somewhat less memory management overhead. Rather than indict Python for oering immutable string, this leads to some productive thinking about string processing in general. In text-intensive applications we may want to avoid creating separate string objects. Instead, we may want to create a single string object the input buer and work with slices of that buer. Rather than create string, we can create slice objects that describe starting and ending osets within the one-and-only input buer. If we then need to manipulate these slices of the input buer, we can create new string objects only as needed. In this case, our application program is designed for eiciency. We use the Python string objects when we want exibility and simplicity.
154
Chapter 13. Strings
CHAPTER
FOURTEEN
TUPLES
Well look at tuple from a number of viewpoints: semantics, literal values, operations, comparison operators, statements, built-in functions and methods. Additionally, we have a digression on the operator in Digression on The Sigma Operator .
14.1 Tuple Semantics

A tuple is a container for a xed sequence of data objects. The name comes from the Latin suix for multiples: double, triple, quadruple, quintuple. Mathematicians commonly consider ordered pairs; for instance, most analytical geometry is done with Cartesian coordinates (x, y ). An ordered pair can be generalized as a 2-tuple. An essential ingredient here is that a tuple has a xed and known number of elements. A 3-dimensional point is a 3-tuple. An CMYK color code is a 4-tuple. The size of the tuple cant change without fundamentally redening the problem were solving. A tuple is an immutable sequence of Python objects. Since it is a sequence, all of the common operations to sequences apply. Since it is immutable, it cannot be changed. Two common questions that arise are how to expand a tuple and how to remove objects from a tuple. When someone asks about changing an element inside a tuple, either adding, removing or updating, we have to remind them that the list, covered in Lists, is for dynamic sequences of elements. A tuple is generally applied when the number of elements is xed by the nature of the problem. This tuple processing even pervades the way functions are dened. We can have positional parameters collected into a tuple, something well cover in Advanced Parameter Handling For Functions.
14.2 Tuple Literal Values

A tuple literal is created by surrounding objects with () and separating the items with commas (,). An empty tuple is simple (). An interesting question is how Python tells an expression from a 1-tuple. A 1-element tuple has a single comma; for example, (1,). An expression lacks the comma: (1). A pleasant consequence of this is that an extra comma at the end of every tuple is legal; for example, (9, 10, 56, ). Examples:
155
xy= (2, 3) personal= ('Hannah',14,5*12+6) singleton= ("hello",)
xy A 2-tuple with integers. personal A 3-tuple with a string and two integers singleton A 1-tuple with a string. The trailing , assures that his is a tuple, not an expression. The elements of a tuple do not have to be the same type. A tuple can be a mixture of any Python data types, including other tuples.
14.3 Tuple Operations

There are three standard sequence operations (+, *, []) that can be performed with tuples as well as other sequences. The + operator creates a new tuple as the concatenation of the arguments. Heres an example.
>>> ("chapter",8) + ("strings","tuples","lists") ('chapter', 8, 'strings', 'tuples', 'lists')
The * operator between tuple and number (number * tuple or tuple * number) creates a new tuple that is a number of repetitions of the input tuple.
>>> 2*(3,"blind","mice") (3, 'blind', 'mice', 3, 'blind', 'mice')
The [] operator selects an item or a slice from the tuple. There are two forms: the single-item form and the slice form. The single item format is tuple [ index ]. Items are numbered from 0 to len(tuple). Items are also numbered in reverse from -len(tuple) to -1. The slice format is tuple [ start : end ]. Items from start to end -1 are chosen to create a new tuple as a slice of the original tuple; there will be end start items in the resulting tuple. If start is omitted it is the beginning of the tuple (position 0). If end is omitted it is the end of the tuple (position -1). Yes, you can omit both (someTuple[:]) to make a copy of a tuple. This is a shallow copy: the original objects are now members of two distinct tuples.
>>> t=( (2,3), (2,"hi"), (3,"mom"), 2+3j, >>> t[2] (3, 'mom') >>> t[:3] ((2, 3), (2, 'hi'), (3, 'mom')) >>> t[3:] ((2+3j), 6.02e+23) >>> t[-1] 6.02e+23 >>> t[-3:] ((3, 'mom'), (2+3j), 6.02e+23) 6.02E23 )
156
Chapter 14. Tuples
14.4 Tuple Comparison Operations

The standard comparisons (<, <=, >, >=, ==, !=, in, not in) work exactly the same among tuple objects as they do among string and other sequences. The tuple pbjects are compared element by element. If the corresponding elements are the same type, ordinary comparison rules are used. If the corresponding elements are dierent types, the type names are compared, since there is almost no other rational basis for comparison.
>>> a >>> b >>> a True >>> 3 True >>> 3 False = (1, 2, 3, 4, 5) = (9, 8, 7, 6, 5) < b in a in b
Heres a longer example.
redblack.py
#!/usr/bin/env python import random n= random.randrange(38) if n == 0: print '0', 'green' elif n == 37: print '00', 'green' elif n in ( 1,3,5,7,9, 12,14,16,18, 19,21,23,25,27, 30,32,34,36 ): print n, 'red' else: print n, 'black'
1. We import random. 2. We create a random number, n in the range 0 to 37. 3. We check for 0 and 37 as special cases of single and double zero. 4. If the number is in the tuple of red spaces on the roulette layout, this is printed. 5. If none of the other rules are true, the number is in one of the black spaces.
14.5 Tuple Statements

There are a number of statements that have specic features related to tuple objects. The Assignment Statement. There is a variation on the assignment statement called a multipleassignment statement that works nicely with tuples. We looked at this in Multiple Assignment Statement . Multiple variables are set by decomposing the items in the tuple. For example:
14.4. Tuple Comparison Operations
157
>>> x, y = (1, 2) >>> x 1 >>> y 2
An essential ingredient here is that a tuple has a xed and known number of elements. For example a 2-dimensional geometric point might have a tuple with x and y. A 3-dimensional point might be a tuple with x, y, and z. This works well because the right side of the assignment statement is fully evaluated before the assignments are performed. This allows things like swapping the values in two variables with x,y=y,x. The for Statement. The for statement will step though all elements of a sequence. For example:
s= 0 for i in ( 1,3,5,7,9, 12,14,16,18, 19,21,23,25,27, 30,32,34,36 ): s += i print "total",s
This will step through each number in the given tuple. There are three built-in functions that will transform a tuple into another sequence. The enumerate(), sorted() and reversed() functions will provide the items of the tuple with their index, in sorted order or in reverse order.
14.6 Tuple Built-in Functions

The tuple() function creates a tuple out of another sequence object. tuple(sequence) Create a tuple from another sequence. This will convert list or str to a tuple. Functions which apply to tuples, but are dened elsewhere. len(). For tuples, this function returns the number of items.
>>> len( (1,1,2,3) ) 4 >>> len( () ) 0
max(). For tuples, this function returns the maximum item.

>>> max( (1,9973,2) ) 9973
min(). For tuples, this function returns the minimum item. sum(). For tuples, this function sums the individual items.
>>> sum( (1,9973,2) ) 9976
any(). For tuples, Return True if there exists any item which is True.
158
Chapter 14. Tuples
>>> any( (0,None,False) ) False >>> any( (0,None,False,42) ) True >>> any( (1,True) ) True
all(). For tuples, Return True if all items are True.

>>> all( (0,None,False,42) ) False >>> all( (1,True) ) True
enumerate(). Iterate through the tuple returning 2-tuples of ( index, item ). In eect, this function enumerates all the items in a sequence: it provides a number and each element of the original sequence in a 2-tuple.
for i, x in someTuple: print "position", i, " has value ", x
Consider the following.

>>> a = ( 3.1415926, "Words", (2+3j) ) >>> tuple( enumerate( a ) ) ((0, 3.1415926000000001), (1, 'Words'), (2, (2+3j)))
We created a tuple from the enumeration. This shows that each item of the enumeration is a 2-tuple with the index number and an item from the original tuple. sorted(). Iterate through the tuple in sorted order.
>>> (1, >>> (9, tuple( sorted( (9,1,8,2,7,3) )) 2, 3, 7, 8, 9) tuple( sorted( (9,1,8,2,7,3), reverse=True )) 8, 7, 3, 2, 1)
reversed(). Iterate through the tuple in reverse order.

>>> tuple( reversed( (9,1,8,2,7,3) ) ) (3, 7, 2, 8, 1, 9)
The following function returns a tuple. divmod(x, y ) -> ( div, mod ) Return a 2-tuple with ((x-x%y)/y, x%y). The return values have the invariant: div y + mod = x. This is the quotient and the remainder in division. The divmod() functions is often combined with multiple assignment. For example:
>>> q,r = divmod(355,113) >>> q 3 >>> r 16
14.6. Tuple Built-in Functions
159
>>> q*113+r 355
14.7 Tuple Exercises

These exercises implement some basic statistical algorithms. For some background on the Sigma operator, , see Digression on The Sigma Operator . 1. Blocks of Stock. A block of stock as a number of attributes, including a purchase date, a purchase price, a number of shares, and a ticker symbol. We can record these pieces of information in a tuple for each block of stock and do a number of simple operations on the blocks. Lets dream that we have the following portfolio. Purchase Date 25 25 25 25 Jan Jan Jan Jan 2001 2001 2001 2001 Purchase Price 43.50 42.80 42.10 37.58 Shares 25 50 75 100 Symbol CAT DD EK GM :Current Price 92.45 51.19 34.87 37.58
We can represent each block of stock as a 5-tuple with purchase date, purchase price, shares, ticker symbol and current price.
portfolio= [ ( "25-Jan-2001", 43.50, 25, 'CAT', 92.45 ), ( "25-Jan-2001", 42.80, 50, 'DD', 51.19 ), ( "25-Jan-2001", 42.10, 75, 'EK', 34.87 ), ( "25-Jan-2001", 37.58, 100, 'GM', 37.58 ) ]
Develop a function that examines each block, multiplies shares by purchase price and determines the total purchase price of the portfolio. Develop a second function that examines each block, multiplies shares by purchase price and shares by current price to determine the total amount gained or lost. 2. Mean. Computing the mean of a list of values is relatively simple. The mean is the sum of the values divided by the number of values in the list . Since the statistical formula is so closely related to the actual loop, well provide the formula, followed by an overview of the code. xi x =
0i<n
[The cryptic-looking x is a short-hand for mean of variable x.] The denition of the mathematical operator leads us to the following method for computing the mean:
Computing Mean
(a) Initialize. Set sum, s, to zero (b) Reduce. For each value, i, in the range 0 to the number of values in the list, n: Set s s + xi .
160
Chapter 14. Tuples
(c) Result. Return s n. 3. Standard Deviation. The standard deviation can be done a few ways, but well use the formula shown below. This computes a deviation measurement as the square of the dierence between each sample and the mean. The sum of these measurements is then divided by the number of values times the number of degrees of freedom to get a standardized deviation measurement. Again, the formula summarizes the loop, so well show the formula followed by an overview of the code. x =
0i<n
(xi x )2
n1
[The cryptic-looking x is short-hand for standard deviation of variable x.] The denition of the mathematical operator leads us to the following method for computing the standard deviation:
Computing Standard Deviation

(a) Initialize. Compute the mean, m. Initialize sum, s, to zero. (b) Reduce. For each value, xi in the list: Compute the dierence from the mean, d xi x . Set s s + d2 .
s (c) Variance. Compute the variance as n 1 . The n 1. factor reects the statistical notion of degrees of freedom, which is beyond the scope of this book.
(d) Standard Deviation. Return the square root of the variance. The math module contains the math.sqrt() funtion. For some additional information, see The math Module.
14.8 Digression on The Sigma Operator

For those programmers new to statistics, this section provides background on the Sigma operator, . The usual presentation of the summation operator looks like this.
n i=1
f (i)
The operator has three parts to it. Below it is a bound variable, i and the starting value for the range, written as i = 1. Above it is the ending value for the range, usually something like n. To the right is some function to execute for each value of the bound variable. In this case, a generic function, f (i). This is read as sum f ( i ) for i in the range 1 to n. This common denition of uses a closed range; one that includes the end values of 1 and n. This, however, is not a helpful denition for software. It is slightly simpler to dene to start with zero and use a half-open interval. It still exactly n elements, including 0 and n-1; mathematically, 0 i < n. For software design purposes, we prefer the following notation, but it is not often used. Since most statistical and mathematical texts use 1-based indexing, some care is required when translating formulae to
14.8. Digression on The Sigma Operator
161
programming languages that use 0-based indexing.

0i<n
f (i)
This shows the bound variable (i ) and the range below the operator. It shows the function to execute on the right of the operator. Statistical Algorithms. Our two statistical algorithms have a form more like the following. In this we are applying some function, f, to each value, xi of an array. f (xi )
0i<n
When computing the mean, there the function applied to each value does nothing. When computing standard deviation, the function applied involves subtracting and multiplying. We can transform this denition directly into a for loop that sets the bound variable to all of the values in the range, and does some processing on each value of the sequence of values. This is the Python implemention of . This computes two values, the sum, sum, and the number of elements, n.
Python Sigma Iteration

sum= 0 for x_i in aSequence: fx_i = some processing of x_i sum += fx_i n= len(aSequence)
1. Execute the body of the loop for all values of x_i in the sequence aSequence. The sequence can be a tuple, list or other sequential container. 2. For simple mean calculation, the fx_i statement does nothing. For standard deviation, however, this statement computes the measure of deviation from the average. 3. We sum the x_i values for a mean calculation. We sum fx_i values for a standard deviation calculation.
162
Chapter 14. Tuples
CHAPTER
FIFTEEN
LISTS
Well look at list from a number of viewpoints: semantics, literal values, operations, comparison operators, statements, built-in functions and methods.
15.1 List Semantics

A list is a container for variable length sequence of Python objects. A list is mutable, which means that items within the list can be changed. Also, items can be added to the list or removed from the list. Since a list is a sequence, all of the common operations to sequences apply. Sometimes well see a list with a xed number of elements, like a two-dimensional point with two elements, x and y. A xed-length list may not be the right choice; a tuple, covered in Tuples is usually better for static sequences of elements. A great deal of Pythons internals are list -based. The for statement, in particular, expects a sequence, and we often create a list by using the range() function. When we split a string using the split() method, we get a list of substrings.
15.2 List Literal Values

A list literal is created by surrounding objects with [] and separating the items with commas (,). A list can be created, expanded and reduced. An empty list is simply []. As with tuple, an extra comma at the end of the list is gracefully ignored. Examples:
myList = [ 2, 3, 4, 9, 10, 11, 12 ] history = [ ]
The elements of a list do not have to be the same type. A list can be a mixture of any Python data types, including lists, tuples, strings and numeric types. A list permits a sophisticated kind of display called a comprehension. Well revisit this in some depth in List Comprehensions. As a teaser, consider the following:
>>> [ 2*i+1 for i in range(6) ] [1, 3, 5, 7, 9, 11]
163
This statement creates a list using a list comprehension. A comprehension starts with a candidate list ( range(6), in this example) and derives the list values from the candidate list (using 2*i+1 in this example). A great deal of power is available in comprehensions, but well save the details for a later section.
15.3 List Operations

The three standard sequence operations (+, *, []) can be performed with list, as well as other sequences like tuple and string. The + operator creates a new list as the concatenation of the arguments.
>>> ["field"] + [2, 3, 4] + [9, 10, 11, 12] ['field', 2, 3, 4, 9, 10, 11, 12]
The * operator between list and numbers (number * list or list * number) creates a new list that is a number of repetitions of the input list.
>>> 2*["pass","don't","pass"] ['pass', "don't", 'pass', 'pass', "don't", 'pass']
The [] operator selects an character or a slice from the list. There are two forms: the single-item form and the slice form. The single item format is list [ index ]. Items are numbered from 0 to len(list). Items are also numbered in reverse from -len(list) to -1. The slice format is list [ start : end ]. Items from start to end -1 are chosen to create a new list as a slice of the original list; there will be end start items in the resulting list. If start is omitted it is the beginning of the list (position 0). If end is omitted it is the end of the list (position -1). Yes, you can omit both (someList[:]) to make a copy of a list. This is a shallow copy: the original objects are now members of two distinct lists. In the following example, weve constructed a list where each element is a tuple. Each tuple could be a pair of dice.
>>> l=[(6, 2), (5, 4), (2, 2), (1, 3), (6, 5), (1, 4)] >>> l[2] (2, 2) >>> l[:3] [(6, 2), (5, 4), (2, 2)] >>> l[3:] [(1, 3), (6, 5), (1, 4)] >>> l[-1] (1, 4) >>> l[-3:] [(1, 3), (6, 5), (1, 4)]
15.4 List Comparison Operations

The standard comparisons (<, <=, >, >=, ==, !=, in, not in) work exactly the same among list, tuple and string sequences. The list items are compared element by element. If the corresponding 164 Chapter 15. Lists
elements are the same type, ordinary comparison rules are used. If the corresponding elements are dierent types, the type names are compared, since there is no other rational basis for comparison.
d1= random.randrange(6)+1 d2= random.randrange(6)+1 if d1+d2 in [2, 12] + [3, 4, 9, 10, 11]: print "field bet wins on ", d1+d2 else: print "field bet loses on ", d1+d2
This will create two random numbers, simulating a roll of dice. If the number is in the list of eld bets, this is printed. Note that we assemble the nal list of eld bets from two other list objects. In a larger application program, we might separate the dierent winner list instances based on dierent payout odds.
15.5 List Statements

There are a number of statements that have specic features related to list objects. The Assignment Statement. The variation on the assignment statement called multiple-assignment statement also works with lists. We looked at this in Multiple Assignment Statement . Multiple variables are set by decomposing the items in the list.
>>> x, y = [ 1, "hi" ] >>> x 1 >>> y 'hi'
This will only work of the list has a xed and known number of elements. This is more typical when working with a tuple, which is immutable, rather than a list, which can vary in length. The for Statement. The for statement will step though all elements of a sequence.
s= 0 for i in [2,3,5,7,11,13,17,19]: s += i print "total",s
When we introduced the for statement in Iterative Processing: The for Statement , we showed the range() function; this function creates a list. We can also create a list with a literal or comprehension. Weve looked at simple literals above. Well look at comprehensions below. The del Statement. The del statement removes items from a list. For example
>>> >>> >>> [1, i = range(10) del i[0], i[2], i[4], i[6] i 2, 4, 5, 7, 8]
This example reveals how the del statement works. The i variable starts as the list [0, 1, 2, 3, 4, 5, 6, 7, 8, 9 ]. 1. Remove i[0] and the variable is [1, 2, 3, 4, 5, 6, 7, 8, 9]. 2. Remove i[2] (the value 3) from this new list , and get [1, 2, 4, 5, 6, 7, 8, 9].
15.5. List Statements
165
3. Remove i[4] (the value 6) from this new list and get [1, 2, 4, 5, 7, 8, 9]. 4. Finally, remove i[6] and get [1, 2, 4, 5, 7, 8].
15.6 List Built-in Functions

The list() function creates a list out of another sequence object. list(sequence) Create a list from another sequence. This will convert tuple or str to a list. Functions which apply to tuples, but are dened elsewhere. len(). For lists, this function returns the number of items.
>>> len( [1,1,2,3] ) 4 >>> len( [] ) 0
max(). For lists, this function returns the maximum item.

>>> max( [1,9973,2] ) 9973
min(). For lists, this function returns the minimum item. sum(). For lists, this function sums the individual items.
>>> sum( [1,9973,2] ) 9976
any(). For lists, Return True if there exists any item which is True.
>>> any( [0,None,False] ) False >>> any( [0,None,False,42] ) True >>> any( [1,True] ) True
all(). For lists, Return True if all items are True.

>>> all( [0,None,False,42] ) False >>> all( [1,True] ) True
enumerate(). Iterate through the list returning 2-tuples of ( index, item ). In eect, this function enumerates all the items in a sequence: it provides a number and each element of the original sequence in a 2-tuple.
for i, x in someList: print "position", i, " has value ", x
Consider the following list of tuples.
166
Chapter 15. Lists
>>> a = [ ("pi",3.1415946),("e",2.718281828),("mol",6.02E23) ] >>> list( enumerate( a ) ) [(0, ('pi', 3.1415945999999999)), (1, ('e', 2.7182818279999998) 02e+23))] >>> for i, t in enumerate( a ): ... print "item",i,"is",t ... item 0 is ('pi', 3.1415945999999999) item 1 is ('e', 2.7182818279999998) item 2 is ('mol', 6.02e+23)
sorted(). Iterate through the list in sorted order.

>>> [1, >>> [9, list( sorted( [9,1,8,2,7,3] )) 2, 3, 7, 8, 9] tuple( sorted( [9,1,8,2,7,3], reverse=True )) 8, 7, 3, 2, 1]
reversed(). Iterate through the list in reverse order.

>>> tuple( reversed( [9,1,8,2,7,3] ) ) [3, 7, 2, 8, 1, 9]
The following function returns a list. range([start], stop, [step] ) The arguments must be plain integers. If the step argument is omitted, it defaults to 1. If the start argument is omitted, it defaults to 0. step must not be zero (or else ValueError is raised). The full form returns a list of plain integers [ start, start + step, start + 2 step, ...]. If step is positive, the last element is the largest start + i step < stop; ; if step is negative, the last element is the largest start + i step > stop.
15.7 List Methods

A list object has a number of member methods. These can be grouped arbitrarily into mutators, which change the list, transformers which create something new from the list, and and accessors, which returns a fact about a list. The following list mutators update a list object. Generally, these do not return a value. In the case of the pop() method, it both returns information as well as mutates the list. append(object ) Update list by appending object to end of the list. extend(list ) Extend list by appending list elements. Note the dierence from append() , which treats the argument as a single list object. insert(index, object ) Update list l by inserting object before position index. If index is greater than len(list), the object is simply appended. If index is less than zero, the object is prepended.
15.7. List Methods
167
pop([index=-1] ) Remove and return item at index (default last, -1) in list. An exception is raised if the list is already empty. remove(value) Remove rst occurrence of value from list. An exception is raised if the value is not in the list. reverse() Reverse the items of the list. This is done in place, it does not create a new list. There is no return value. sort([key], [reverse=False] ) Sort the items of the list. This is done in place, it does not create a new list. If the reverse keyword parameter is provided and set to True, the tuple is sorted into descending order. The key parameter is used when the items in the tuple arent simply sorted using the default comparison operators. The key function must return the elds to be compared selected from the underlying items in the tuple. Well look at this in detail in Functional Programming with Collections. The following accessor methods provide information about a list. count(value) Return number of occurrences of value in list. index(value) Return index of rst occurrence of value in list. Stacks and Queues. The list.append() and list.pop() functions can be used to create a standard push-down stack, or last-in-rst-out (LIFO) list. The append() method places an item at the end of the list (or top of the stack), where the pop() method can remove it and return it.
>>> stack = [] >>> stack.append(1) >>> stack.append( "word" ) >>> stack.append( ("a","2-tuple") ) >>> stack.pop() ('a', '2-tuple') >>> stack.pop() 'word' >>> stack.pop() 1 >>> len(stack) 0 >>> stack.pop() Traceback (most recent call last): File "<stdin>", line 1, in <module> IndexError: pop from empty list
The list.append() and list.pop() functions can be used to create a standard queue, or rst-in-rst-out (FIFO) list. The append() method places an item at the end of the queue. A call to pop(0) removes the rst item from the queue and returns it.
>>> >>> >>> >>> queue = [] queue.append( 1 ) queue.append( "word" ) queue.append( ("a","2-tuple") )
168
Chapter 15. Lists
>>> queue.pop(0) 1 >>> queue.pop(0) 'word' >>> queue.pop(0) ('a', '2-tuple') >>> len(queue) 0 >>> queue.pop(0) Traceback (most recent call last): File "<stdin>", line 1, in <module> IndexError: pop from empty list
15.8 Using Lists as Function Parameter Defaults

Its very, very important to note that default values must be immutable objects. Recall that numbers, strings, None, and tuple objects are immutable. We note that lists as well as sets and dictionaries are mutable, and cannot be used as default values for function parameters. Consider the following example of what not to do.
>>> ... ... ... >>> >>> [2] >>> [2, >>> [2, >>> >>> >>> >>> [2] >>> >>> [2, >>> [2, def append2( someList=[] ): someList.append(2) return someList looks_good= [] append2(looks_good) append2(looks_good) 2] looks_good 2]
not_good= append2() not_good worse= append2() worse 2] not_good 2]
1. We dened a function which has a default value thats a mutable object. This is simple a bad programming practice in Python. 2. We used this function with a list object, looks_good. The function updated the list object as expected. 3. We used the functions default value to create not_good. The function appended to an empty list and returned this new list object. It turns out that the function updated the mutable default value, also. 4. When we use the functions default value again, with worse, the function uses the updated default value and updates it again. 15.8. Using Lists as Function Parameter Defaults 169
Both not_good and worse are references to the same mutable object that is being updated. To avoid this, do not use mutable values as defaults. Do this instead.
def append2( someList=None ): if someList is None: someList= [] someList.append(2) return someList
This creates a fresh new mutable object as needed.
15.9 List Exercises

1. Accumulating Distinct Values. This uses the Bounded Linear Search algorithm to locate duplicate values in a sequence. This is a powerful technique to eliminate sorting from a wide variety of summarytype reports. Failure to use this algorithm leads to excessive processing in many types of applications.
Distinct Values of a Sequence, seq

(a) Initialize Distinct Values. Set dv list(). (b) Loop. For each value, v, in seq. Well use the Bounded Linear Search to see if v occurs in dv. i. Initialize. Set i 0. Append v to the list dv. ii. Search. while dv [i] = v : increment i. At this point dv [i] = v . The question is whether i = len(dv ) or not. iii. New Value?. if i = len(dv ): v is distinct. iv. Existing Value?. if i = len(dv ): v is a duplicate of dv [i]. Delete dv [1], the value we added. (c) Result. Return array dv, which has distinct values from seq. You may also notice that this fancy Bounded Linear Search is suspiciously similar to the index() method function of a list. Rewrite this using uniq.index instead of the Bounded Linear Search in step 2. When we look the set collection, youll see another way to tackle this problem. 2. Binary Search. This is not as universally useful as the Bounded Linear Search (above) because it requires the data be sorted.
Binary Search a sorted Sequence, seq, for a target value, tgt

(a) Initialize. l, h 0, len(seq ). m (l + h) 2. This is the midpoint of the sorted sequence.
170
Chapter 15. Lists
(b) Divide and Conquer. While l + 1 < h and seq [m] = tgt. If tgt < seq [m]: h m. Move h to the midpoint. If tgt > seq [m]: l m + 1. Move l to the midpoint. m (l + h) 2. Compute a midpoint of the new, smaller sequence. (c) Result. If tgt = seq [m]: return m If tgt = seq [m]: return -1 as a code for not found. 3. Quicksort. The super-fast sort routine As a series of loops it is rather complex. As a recursion it is quite short. This is the same basic algorithm in the C libraries. Quicksort proceeds by partitioning the list into two regions: one has all of the high values, the other has all the low values. Each of these regions is then individually sorted into order using the quicksort algorithm. This means the each region will be subdivided and sorted. For now, well sort an array of simple numbers. Later, we can generalize this to sort generic objects.
Quicksort a List, a between elements lo and hi

(a) Partition i. Initialize. ls, hs lo, hi. Setup for partitioning between ls and hs. middle (ls + hs) 2. ii. Swap To Partition. while ls < hs: If a[ls].key a[middle].key : increment ls by 1. Move the low boundary of the partitioning. If a[ls].key > a[middle].key : swap the values a[ls] a[middle]. If a[hs].key a[middle].key : decrement hs by 1. Move the high boundary of the partitioning. If a[hs].key < a[middle].key :, swap the values a[hs] (b) Quicksort Each Partition. QuickSort( a , lo, middle ) QuickSort( a , middle+1, hi ) 4. Recursive Search. This is also a binary search: it works using a design called divide and conquer. Rather than search the whole list, we divide it in half and search just half the list. This version, however is dened with a recusive function instead of a loop. This can often be faster than the looping version shown above. a[middle].
Recursive Search a List, seq for a target, tgt, in the region between elements lo and hi.
(a) Empty Region? If lo + 1 hi: return -1 as a code for not found. (b) Middle Element. m (lo + hi) 2. (c) Found? If seq [m] = tgt: return m.
15.9. List Exercises
171
(d) Lower Half? If seq [m] < tgt: return recursiveSearch ( seq, tgt, lo, m ) (e) Upper Half? If seq [m] > tgt: return recursiveSearch( seq, tgt, m+1, hi ) 5. Sieve of Eratosthenes. This is an algorithm which locates prime numbers. A prime number can only be divided evenly by 1 and itself. We locate primes by making a table of all numbers, and then crossing out the numbers which are multiples of other numbers. What is left must be prime.
Sieve of Eratosthenes
(a) Initialize. Create a list, prime of 5000 booleans, all True, initially. p 2. (b) Iterate. While 2 p < 5000. i. Find Next Prime. While not prime[p] and 2 p < 5000: Increment p by 1. ii. Remove Multiples. At this point, p is prime. Set k p + p. while k < 5000. prime[k ] F alse. Increment k by p. iii. Next p. Increment p by 1. (c) Report. At this point, for all p, if prime [ p ] is true, p is prime. while 2 p < 5000: if prime[p]: print p The reporting step is a lter operation. Were creating a list from a source range and a lter rule. This is ideal for a list comprehension. Well look at these in List Comprehensions. Formally, we can say that the primes are the set of values dened by primes = {p|0p<5000 if primep }. This formalism looks a little bit like a list comprehension. 6. Polynomial Arithmetic. We can represent numbers as polynomials. We can represent polynomials as arrays of their coeicients. This is covered in detail in [Knuth73], section 2.2.4 algorithms A and M. Example: 4x3 + 3x + 1 has the following coeicients: ( 4, 0, 3, 1 ). The polynomial 2x2 3x 4 is represented as ( 2, -3, -4 ). The sum of these is 4x3 + 2x2 3; ( 4, 2, 0, -3 ). The product these is 8x5 12x4 10x3 7x2 15x 4; ( 8, -12, -10, -7, -15, -4 ). You can apply this to large decimal numbers. In this case, x is 10, and the coeicients must all be between 0 and x -1. For example, 1987 = 1x3 + 9x2 + 8x + 7, when x = 10.
Add Polynomials, p, q
(a) Result Size. rsz the larger of len(p) and len(q ). (b) Pad P? If len(p) < rsz :
172
Chapter 15. Lists
Set p1 to a tuple of rsz len(p) zeros + p. Else: Set p1 to p. (c) Pad Q? If len(q ) < rsz : Set q1 t a tuple of rsz len(q ) zeroes + q. Else, Set q1 to q. (d) Add. Add matching elements from p1 and q1 to create result, r. (e) Result. Return r as the sum of p and q.
Multiply Polynomials, x, y
(a) Result Size. rsz len(x) + len(y ). Initialize the result list, r, to all zeros, with a size of rsz . (b) For all elements of x. while 0 i < len(x): For all elements of y. while 0 j < len(y ): Set r[i + j ] = r[i + j ] + x[i] y [j ]. (c) Result. Return a tuple made from r as the product of x and y. 7. Random Number Evaluation. Before using a new random number generator, it is wise to evaluate the degree of randomness in the numbers produced. A variety of clever algorithms look for certain types of expected distributions of numbers, pairs, triples, etc. This is one of many random number tests. Use random.random() to generate an array of random samples. These numbers will be uniform over the interval 0..1
Distribution test of a sequence of random samples, U

(a) Initialize. Initialize count to a list of 10 zeros. (b) Examine Samples. For each sample value, v, in the original set of 1000 random samples, U. i. Coerce Into Range. Set x v 10. Multiply by 10 and truncate and integer to get a a new value in the range 0 to 9. ii. Count. Increment count [x ] by 1. (c) Report. We expect each count to be 1/10th of our available samples. We need to display the actual count and the % of the total. We also need to calculate the dierence between the actual count and the expected count, and display this. 8. Dutch National Flag. A challenging problem, one of the hardest in this set. This is from Edsger Dijkstras book, A Discipline of Programming [Dijkstra76]. Imagine a board with a row of holes lled with red, white, and blue pegs. Develop an algorithm which will swap pegs to make three bands of red, white, and blue (like the Dutch ag). You must also satisfy this additional constraint: each peg must be examined exactly once. Without the additional constraint, this is a relatively simple sorting problem. The additional constraint requires that instead of a simple sort which passes over the data several times, we need a more clever sort.
15.9. List Exercises
173
Hint: You will need four partitions in the array. Initially, every peg is in the Unknown partition. The other three partitions (Red, White and Blue) are empty. As the algorithm proceeds, pegs are swapped into the Red, White or Blue partition from the Unknown partition. When you are done, the unknown partition is reduced to zero elements, and the other three partitions have known numbers of elements.
174
Chapter 15. Lists
CHAPTER
SIXTEEN
MAPPINGS AND DICTIONARIES

Many algorithms need to map a key to a data value. This kind of mapping is supported by the Python dictionary, dict. Well look at dictionaries from a number of viewpoints: semantics, literal values, operations, comparison operators, statements, built-in functions and methods. We are then in a position to look at two applications of the dictionary. Well look at how Python uses dictionaries along with sequences to handle arbitrary connections of parameters to functions in Advanced Parameter Handling For Functions. This is a very sophisticated set of tools that let us dene functions that are very exible and easy to use.
16.1 Dictionary Semantics

A dictionary, called a dict, maps a key to a value. The key can be any type of Python object that computes a consistent hash value. The value referenced by the key can be any type of Python object. There is a subtle terminology issue here. Python has provisions for creating a variety of dierent types of mappings. Only one type of mapping comes built-in; that type is the dict. The terms are almost interchangeable. However, you may develop or download other types of mappings, so well be careful to focus on the dict class. Working with a dict is similar to working with a sequence. Items are inserted into the dict, found in the dict and removed from the dict. A dict object has member methods that return a sequence of keys, or values, or ( key , value ) tuples suitable for use in a for statement. Unlike a sequence, a dict does not preserve order. Instead of order, a dict uses a hashing algorithm to identify each items place in the dict with a rapid calculation of the keys hash value. The built-in function, hash() is used to do this calculation. Items in the dict are inserted in an position related to their keys apparently random hash values. Some Alternate Terminology. A dict can be thought of as a container of ( key : value ) pairs. This can be a helpful way to imagine the information in a mapping. Each pair in the list is the mapping from a key to an associated value. A dict can be called an associative array. Ordinary arrays, typied by sequences, use a numeric index, but a dicts index is made up of the key objects. Each key is associated with (or mapped to) the appropriate value. Some Consequences. Each key object has a hash value, which is used to place the value in the dict. Consequently, the keys must have consistent hash values; they must be immutable objects. You cant use list, dict or set objects as keys. You can use tuple, string and frozenset objects, since they are immutable. Additionally, when we get to class denitions (in Classes), we can make arrangements for our objects to return an immutable hash value.
175
A common programming need is a heterogeneous container of data. Database records are an example. A record in a database might have a boats name (as a string), the length overall (as a number) and an inventory of sails (a list of strings). Often a record like this will have each element (known as a eld ) identied by name. A C or C++ struct accomplishes this. This kind of named collection of data elements may be better handled with a class (someting well cover in Classes) or a named tuple (see collections). However, a mapping is also useful for managing this kind of heterogeneous data with named elds. Note that many languages make record denitions a statically-dened container of named elds. A Python dict is dynamic, allowing us to add eld names at run-time. A common alternative to hashing is using some kind of ordered structure to maintain the keys. This might be a tree or list, which would lead to other kinds of mappings. For example, there is an ordered dictionary, in the Python collections module.
16.2 Dictionary Literal Values

A dict literal is created by surrounding a key-value list with {}; a key is separated from its value with :, and the key : value pairs are separated with commas (,). An empty dict is simply {}. As with list and tuple, an extra , inside the {} is tolerated. Examples:
diceRoll = { (1,1): "snake eyes", (6,6): "box cars" } myBoat = { "NAME":"KaDiMa", "LOA":18, "SAILS":["main","jib","spinnaker"] } theBets = { }
diceRoll This is a dict with two elements. One element has a key of a tuple (1,1) and a value of a string, "snake eyes". The other element has a key of a tuple (6,6) and a value of a string "box cars". myBoat This variable is a dict with three elements. One element has a key of the string "NAME" and a value of the string "KaDiMa". Another element has a key of the string "LOA" and a value of the integer 18. The third element has a key of the string "SAILS" and the value of a list ["main", "jib", "spinnaker"]. theBets An empty dict. The values and keys in a dict do not have to be the same type. Keys must be a type that can produce a hash value. Since list s and dict objects are mutable, they are not permitted as keys. All other non-mutable types (especially string, frozenset and tuple) are legal keys.
16.3 Dictionary Operations

A dict only permits a single operation: []. This is used to add, change or retrieve items from the dict. The slicing operations that apply to sequences dont apply to a dict. Examples of dict operations.
>>> >>> >>> >>> {2: d= {} d[2] = [ (1,1) ] d[3] = [ (1,2), (2,1) ] d [(1, 1)], 3: [(1, 2), (2, 1)]}
176
Chapter 16. Mappings and Dictionaries
>>> d[2] [(1, 1)] >>> d["2 or 3"] = d[2] + d[3] >>> d {'2 or 3': [(1, 1), (1, 2), (2, 1)], 2: [(1, 1)], 3: [(1, 2), (2, 1)]}
1. This example starts by creating an empty dict, d. 2. Into d[2] we insert a list with a single tuple. 3. Into d[3] we insert a list with two tuples. 4. When the entire dict is printed it shows the two key:value pairs, one with a key of 2 and another with a key of 3. 5. The entry with a key of the string "2 or 3" has a value which is computed from the values of d[2] + d[3]. Since these two entries are lists, the lists can be combined with the + operator. The resulting expression is stored into the dict. 6. When we print d, we see that there are three key:value pairs: one with a key of 3, one with a key of 2 and one with a key of "2 or 3" . This ability to use any object as a key is a powerful feature, and can eliminate some needlessly complex programming that might be done in other languages. Here are some other examples of picking elements out of a dict.
>>> myBoat = { "NAME":"KaDiMa", "LOA":18, ... "SAILS":["main","jib","spinnaker"] } >>> myBoat["NAME"] 'KaDiMa' >>> myBoat["SAILS"].remove("spinnaker") >>> myBoat {'LOA': 18, 'NAME': 'KaDiMa', 'SAILS': ['main', 'jib']}
String Formatting with Dictionaries. The string formatting operator, %, can be applied between str and dict as well as str and sequence. When this operator was introduced in Strings, the format specications were applied to a tuple or other sequence. When used with a dict, each format specication is given an additional option that species which dict element to use. The general format for each conversion specication is:
%( element ) [ flags ][ width [ . precision ]] code
The ags, width, precision and code elements are dened in Strings. The element eld must be enclosed in ()s; this is the key to be selected from the dict. For example:
print "%(NAME)s, %(LOA)d" % myBoat
This will nd myBoat[NAME] and use %s formatting; it will nd myBoat[LOA] and use %d number formatting.
16.3. Dictionary Operations
177
16.4 Dictionary Comparison Operations

Some of the standard comparisons ( < , <= , > , >=, == , != ) dont have a lot of meaning between two dictionaries. Since there may be no common keys, nor even a common data type for keys, dictionaries are simply compared by length. The dict with fewer elements is considered less than a dict with more elements. The membership comparisons (in, not in) apply to the keys of the dictionary.
>>> colors = { "blue": (0x30,0x30,0xff), "green": (0x30,0xff,0x97), ... "red": (0xff,0x30,0x97), "yellow": (0xff,0xff,0x30) } >>> "blue" in colors True >>> (0x30,0x30,0xff) in colors False >>> "orange" not in colors True
16.5 Dictionary Statements

There are a number of statements that have specic features related to dict objects. The for Statement. The for statement iterates through the keys of the dictionary.
>>> colors = { "blue": (0x30,0x30,0xff), "green": (0x30,0xff,0x97), ... "red": (0xff,0x30,0x97), "yellow": (0xff,0xff,0x30) } >>> for c in colors: ... print c, colors[c]
Its common to use some slightly dierent techniques for iterating through the elements of a dict. The key:value pairs. We can use the items() method to iterate through the sequence of 2-tuples that contain each key and the associated value.
for key, value in someDictionary.items(): # process key and value
The values. We can use the values() method to iterate through the sequence of values in the dictionary.
for value in someDictionary.values(): # process the value
Note that we cant easily determine the associated key. A dictionary only goes one way: from key to value. The keys. We can use the keys() method to iterate through the sequence of keys. This is what happens when we simply use the dictionary object in the for statement. Heres an example of using the key:value pairs.
>>> myBoat = { "NAME":"KaDiMa", "LOA":18, ... "SAILS":["main","jib","spinnaker"] } >>> for key, value in myBoat.items(): ... print key, "=", value ...
178
LOA = 18 NAME = KaDiMa SAILS = ['main', 'jib', 'spinnaker']
The del Statement. The del statement removes items from a dict . For example
>>> i = { "two":2, "three":3, "quatro":4 } >>> del i["quatro"] >>> i {'two': 2, 'three': 3}
In this example, we use the key to remove the item from the dict. The member function, pop(), does this also.
>>> i = { "two":2, "three":3, "quatro":4 } >>> i.pop("quatro") 4 >>> i {'two': 2, 'three': 3}
16.6 Dictionary Built-in Functions

Here are the built-in functions that deal with dictionaries. dict([values], [key=value...] ) Creates a new dictionary. If a positional parameter, values is provided, each element must be a 2tuple. The values pairs are used to populate the dictionary; the rst element of each pair is the key and the second element is the value. Note that the zip() function produces a list of 2-tuples from two parallel lists. If any keyword parameters are provided, each keyword becomes a key in the dictionary and the keyword argument becomes the value for that key.
>>> dict( [('first',0), ('second',1),('third',2)] ) {'second': 1, 'third': 2, 'first': 0} >>> dict( zip(['fourth','fifth','sixth'],[3,4,5]) ) {'sixth': 5, 'fifth': 4, 'fourth': 3} >>> dict( seventh=7, eighth=8, ninth=9 ) {'seventh': 7, 'eighth': 8, 'ninth': 9}
Functions which apply to dicts, but are dened elsewhere. len(). For dicts, this function returns the number of items.
>>> len( {1:'first',2:'second',3:'third'} ) 3 >>> len( {} ) 0
max(). For dicts, this function returns the maximum key.

>>> max( {1:'first',2:'second',3:'third'} ) 3
16.6. Dictionary Built-in Functions
179
min(). For dicts, this function returns the minimum key. sum(). For dicts, this function sums the keys.
>>> sum( {1:'first',2:'second',3:'third'} ) 6
any(). Equivalent to any( dictionary.keys() ). Return True if any key in the dictionary are True, or equivalent to True. This is almost always true except for empty dictionaries or a peculiar dictionary with keys of 0, False, None, etc. all(). Equivalent to all( dictionary.keys() ). Return True if all keys in the dictionary are True, or equivalent to True.
>>> all( {1:'first',2:'second',3:'third'} ) True >>> all( {1:'first',2:'second',None:'error'} ) False
enumerate(). Iterate through the dictionary returning 2-tuples of ( index, key ). This iterates through the key values. Since dictionaries have no explicit ordering to their keys, this enumeration is in an arbitrary order. sorted(). Iterate through the dictionary keys in sorted order. The keys are actually a list, and this returns a list of the sorted keys.
>>> sorted( { "two":2, "three":3, "quatro":4 } ) ['quatro', 'three', 'two']
16.7 Dictionary Methods

A dict object has a number of member methods. Many of these maintain the values in a dict . Others retrieve parts of the dict as a sequence, for use in a for statement. The following mutator functions update a dict object. Most of these do not return a value. The dict.pop() and dict.setdefault() methods both update the dictionary and return values. clear() Remove all items from the dict. pop(key, [default] ) Remove the given key from the dict, returning the associated value. If the key does not exist, return the default value provided. If the key does not exist and no default value exists, raise a KeyError exception. setdefault(key, [default] ) If the key is in the dictionary, return the associated value. If the key is not in the dictionary, set the given default as the value and return this value. If default is not given, it defaults to None. update(new, [key=value...] ) Merge values from the new new into the original dict, adding or replacing as needed. It is equivalent to the following Python statement. for k in new.keys(): d[k]= new[k] If any keyword parameters are provided, each keyword becomes a key in the dictionary and the keyword argument becomes the value for that key.
180
>>> x= dict( seventh=7, eighth=8, ninth=9 ) >>> x {'seventh': 7, 'eighth': 8, 'ninth': 9} >>> x.update( first=1 ) >>> x {'seventh': 7, 'eighth': 8, 'ninth': 9, 'first': 1}
The following transformer function transforms a dictionary into another object. copy() Copy the dict to make a new dict. This is a shallow copy. All objects in the new dict are references to the same objects as the original dict. The following accessor methods provide information about a dict. get(key, [default] ) Get the item with the given key, similar to dict[key]. If the key is not present and default is given, supply default instead. If the key is not present and no default is given, raise the KeyError exception. items() Return all of the items in the dict as a sequence of (key,value) 2-tuples. Note that these are returned in no particular order. keys() Return all of the keys in the dict as a sequence of keys. Note that these are returned in no particular order. values() Return all the values from the dict as a sequence. Note that these are returned in no particular order.
16.8 Using Dictionaries as Function Parameter Defaults

Its very, very important to note that default values must be immutable objects. Recall that numbers, strings, None, and tuple objects are immutable. We note that dictionaries as well as sets and lists are mutable, and cannot be used as default values for function parameters. Consider the following example of what not to do.
>>> def default2( someDict={} ): ... someDict['default']= 2 ... return someDict ... >>> looks_good= {} >>> default2(looks_good) {'default': 2} >>> default2(looks_good) {'default': 2} >>> looks_good {'default': 2} >>> >>> >>> not_good= default2() >>> not_good {'default': 2} >>> worse= default2()
16.8. Using Dictionaries as Function Parameter Defaults
181
>>> worse {'default': 2} >>> not_good {'default': 2} >>> >>> not_good['surprise']= 'what?' >>> not_good {'default': 2, 'surprise': 'what?'} >>> worse {'default': 2, 'surprise': 'what?'}
1. We dened a function which has a default value thats a mutable object. This is simple a bad programming practice in Python. 2. We used this function with a dictionary object, looks_good. The function updated the dictionary object as expected. 3. We used the functions default value to create not_good. The function inserted a value into an empty dictionary and returned this new dictionary object. It turns out that the function updated the mutable default value, also. 4. When we use the functions default value again, with worse, the function uses the updated default value and updates it again. Both not_good and worse are references to the same mutable object that is being updated. To avoid this, do not use mutable values as defaults. Do this instead.
def default2( someDict=None ): if someDict is None: someDict= {} someDict['default']= 2 return someDict
16.9 Dictionary Exercises

1. Word Frequencies. Update the exercise in Accumulating Unique Values to count each occurance of the values in aSequence. Change the result from a simple sequence to a dict. The dict key is the value from aSequence. The dict value is the count of the number of occurances. If this is done correctly, the input sequence can be words, numbers or any other immutable Python object, suitable for a dict key. For example, the program could accept a line of input, discarding punctuation and breaking them into words in space boundaries. The basic string operations should make it possible to create a simple sequence of words. Iterate through this sequence, placing the words into a dict. The rst time a word is seen, the frequency is 1. Each time the word is seen again, increment the frequency. Produce a frequency table. To alphabetize the frequency table, extract just the keys. A sequence can be sorted (see section 6.2). This sorted sequence of keys can be used to extract the counts from the dict. 2. Stock Reports. A block of publicly traded stock has a variety of attributes, well look at a few of them. A stock has a ticker symbol and a company name. Create a simple dict with ticker symbols and company names. 182 Chapter 16. Mappings and Dictionaries
For example:
stockDict = { 'GM': 'General Motors', 'CAT':'Caterpillar', 'EK':"Eastman Kodak" }
Create a simple list of blocks of stock. These could be tuple s with ticker symbols, prices, dates and number of shares. For example:
purchases = [ ( 'GE', 100, '10-sep-2001', 48 ), ( 'CAT', 100, '1-apr-1999', 24 ), ( 'GE', 200, '1-jul-1999', 56 ) ]
Create a purchase history report that computes the full purchase price (shares times dollars) for each block of stock and uses the stockDict to look up the full company name. This is the basic relational database join algorithm between two tables. Create a second purchase summary that which accumulates total investment by ticker symbol. In the above sample data, there are two blocks of GE. These can easily be combined by creating a dict where the key is the ticker and the value is the list of blocks purchased. The program makes one pass through the data to create the dict. A pass through the dict can then create a report showing each ticker symbol and all blocks of stock. 3. Date Decoder. A date of the form 8-MAR-85 includes the name of the month, which must be translated to a number. Create a dict suitable for decoding month names to numbers. Create a function which uses string operations to split the date into 3 items using the - character. Translate the month, correct the year to include all of the digits. The function will accept a date in the dd-MMM-yy format and respond with a tuple of ( y , m, d ). 4. Dice Odds. There are 36 possible combinations of two dice. A simple pair of loops over range(6)+1 will enumerate all combinations. The sum of the two dice is more interesting than the actual combination. Create a dict of all combinations, using the sum of the two dice as the key. Each value in the dict should be a list of tuple s; each tuple has the value of two dice. The general outline is something like the following:
Enumerate Dice Combinations

Initialize. combos dict() For all d1. Iterate with 1 d1 < 7. For all d2. Iterate with 1 d2 < 7. Create a Tuple. t (d1 , d2 ). In the Dictionary. Is d1 + d2 a key in combos? Append. Append the tuple, t to the value for item d1 + d2 in combos. Not In the Dictionary. If d1 + d2 is not a key in combos, then Insert. Add a new element d1 + d2 to the combos; the value is a 1-element list of the tuple, t. Report. Display the resulting dictionary.
16.9. Dictionary Exercises
183
16.10 Advanced Parameter Handling For Functions

In More Function Denition Features we hinted that Python functions can handle a variable number of argument values in addition to supporting optional argument values. When we dene a function, we can have optional parameters. We dene a xed number of parameters, but some (or all) can be omitted because we provided default values for them. This allows us to provide too few positional argument values. If we provide too many positional argument values to a function, however, Python raises an exception. It turns out that we can also handle this. Consider the following example. We dened a function of three positional parameters, and then evaluated it with more than three argument values.
>>> def avg(a,b,c): return (a+b+c)/3.0 ... >>> avg(10,11,12) 11.0 >>> avg(10,11,12,13) Traceback (most recent call last): File "<stdin>", line 1, in <module> TypeError: avg() takes exactly 3 arguments (4 given)
First, well look at handling an unlimited number of positional values. Then well look at handling an unlimited number of keyword values.
16.10.1 Unlimited Number of Positional Argument Values

Python lets us dene a function that handles an unknown and unlimited number of argument values. Examples of built-in functions with a unlimited number of argument values are max() and min(). Rather than have Python raise an exception for extra argument values, we can request the additional positional argument values be collected into a tuple. To do this, we provide a nal parameter denition of the form * extras. The * indicates that this parameter variable is the place to capture the extra argument values. The variable, here called extras, will receive a sequence with all of the extra positional argument values. You can only provide one such variable (if you provided two, how could Python decide which of these two got the extra argument values?) You must provide this variable after the ordinary positional parameters in the function denition. The following function accepts an unlimited number of positional arguments; it collects these in a single tuple parameter, args.
def myMax( *args ): max= args[0] for a in args[1:]: if a > max: max= a return max
Heres another example. In this case we have a xed parameter in the rst position and all the extra parameters collected into a tuple called vals.
def printf( format, *vals ): print format % vals
184
This should look familiar to C programmers. Now we can write the following, which may help ease the transition from C to Python.
printf( "%s = %d", "some string", 2 ) printf( "%s, %s, %d %d", "thing1", "thing2", 3, 22 )
16.10.2 Unlimited Number of Keyword Argument Values

In addition to collecting extra positional argument values into a single parameter, Python can also collect extra keyword argument values into a dict. If you want a container of keyword arguments, you provide a parameter of the form ** extras. Your variable, here called extras, will receive a dict with all of the keyword parameters. The following function accepts any number of keyword arguments; they are collected into a single parameter.
def rtd( **args ): if "rate" in args and "time" in args: args['distance'] = args['rate']*args['time'] elif "rate" in args and "distance" in args: args['time']= args['distance']/args['rate'] elif "time" in args and "distance" in args: args['rate']= args['distance']/args['time'] else: raise Exception( "%r does not compute" % ( args, ) ) return args
Heres two examples of using this rtd() function.

>>> rtd( rate=60.0, time= .75 ) {'distance': 45.0, 'rate': 60.0, 'time': 0.75} >>> rtd( distance=173, time=2+50/60.0 ) {'distance': 173, 'rate': 61.058823529411761, 'time': 2.8333333333333335}
The keyword arguments are collected into a dict, named args. We check for combinations of rate, time and distance in the args dictionary. For each combination, we can solve for the remaining value and update the dict by insert the additional key and value into the dict.
16.10.3 Evaluation with a Container Instead of Individual Argument Values

When evaluating a function, we can provide a sequence instead of providing individual positional parameters. We do this with a special version of the * operator when evaluating a function. Heres an example of forcing a 3- tuple to be assigned to three positional parameters.
>>> def avg3( a, b, c ): ... return (a+b+c)/3.0 ... >>> data= ( 4, 3, 2 ) >>> avg3( *data ) 3.0
In this example, we told Python to assign each element of our 3-tuple named data, to a separate parameter variables of the function avg3().
16.10. Advanced Parameter Handling For Functions
185
As with the * operator, we can use ** to make a dict become a series of keyword parameters to a function.
>>> d={ 'a':5, 'b':6, 'c':9 } >>> avg3( **d ) 6.666666666666667
In this example, we told Python to assign each element of the dict, d , to specic keyword parameters of our function. We can mix and match this with ordinary parameter assignment, also. Heres an example.
>>> avg3( 2, b=3, **{'c':4} ) 3.0
Here weve called our function with three argument values. The parameter a will get its value from a simple positional parameter. The parameter b will get its value from a keyword argument. The parameter c will get its value from having the dict {'c':4} turned into keyword parameter assignment. Well make more use of this in Inheritance .
186
CHAPTER
SEVENTEEN
SETS
Many algorithms need to work with simple containers of data values, irrespective of order or any key. This is a simple set of objects, which is supported by the Python set container. Well look at Sets from a number of viewpoints: semantics, literal values, operations, comparison operators, statements, built-in functions and methods.
17.1 Set Semantics

A set is, perhaps the simplest possible container, since it contains objects in no particular order with no particular identication. Objects stand for themselves. With a sequence, objects are identied by position. With a mapping, objects are identied by some key. With a set, objects stand for themselves. Since each object stands for itself; elements of a set cannot be duplicated. A list or tuple, for example, can have any number of duplicate objects. For example, the tuple ( 1, 1, 2, 3 ) has four elements, which includes two copies of the integer 1; if we create a set from this tuple, the set will only have three elements. A set has large number of operations for unions, intersections, and dierences. A common need is to examine a set to see if a particular object is a member of that set, or if one set is contained within another set. A set is mutable, which means that it cannot be used as a key for a dict (see Mappings and Dictionaries for more information.) In order to use a set as a dict key, we can create a frozenset, which is an immutable copy of a set. This allows us to accumulate a set of values to create a dict key.
17.2 Set Literal Values

There are no literal values for set objects. A set value is created by using the set() or frozenset() factory functions. These can be applied to any iterable container, which includes any sequence, the keys of a dict, or even a le. Well return to the general notion of iterable when we look at the yield statement in Iterators and Generators. set(iterable) Transforms the given iterable (sequence, le, frozenset or set) into a set.
>>> set( ("hello", "world", "of", "words", "of", "world") ) set(['world', 'hello', 'words', 'of'])
187
Note that we provided a six-tuple sequence to the set() function, and we got a set with the four unique objects. The set is shown as a list literal, to remind us that a set is mutable. You cannot provide a list of argument values to the set() function. You must provide an iterable object (usually a tuple). Trying set( "one", "two", "three" ) will result in an TypeError because you provided three arguments. You must provide a single argument which is iterable. All sequences are iterable, so a sequence literal is the easiest to provide. set(iterable) Transforms the given iterable (sequence, le or set) into an immutable frozenset.
17.3 Set Operations

There are a large number of set operations, including union (|), intersection (&), dierence ( -), and symmetric dierence (^). These are unusual operations, so well look at them in some detail. In addition to the operator notation, there are also method functions which do the same things. Well look at the method function versions below. Well use the following two set objects to show these operators.
>>> fib=set( (1,1,2,3,5,8,13) ) >>> prime=set( (2,3,5,7,11,13) )
Union. |. The resulting set has elements from both source sets. An element is in the result if it is one set or the other.
>>> fib | prime set([1, 2, 3, 5, 7, 8, 11, 13])
188
Chapter 17. Sets
S 1 S 2 = {e|e S 1 or e S 2} Intersection. &. The resulting set has elements that are common to both source sets. An element is in the result if it is in one set and the other.
>>> fib & prime set([2, 3, 5, 13])
S 1 S 2 = {e|e S 1 and e S 2} Dierence. -. The resulting set has elements of the left-hand set with all elements from the right-hand set removed. An element will be in the result if it is in the left-hand set and not in the right-hand set.
>>> fib - prime set([8, 1]) >>> prime - fib set([11, 7])
S 1 S 2 = {e|e S 1 and e / S 2} S 2 S 1 = {e|e / S 1 and e S 2} Symmetric Dierence. ^. The resulting set has elements which are unique to each set. An element will be in the result set if either it is in the left-hand set and not in the right-hand set or it is in the right-hand set and not in the left-hand set. Whew!
17.3. Set Operations
189
>>> fib ^ prime set([1, 7, 8, 11])
S 1 S 2 = {e|e S 1 xor e S 2}
17.4 Set Comparison Operators

Therer are a number of set comparisons. All of the standard comparisons (<, <=, >, >=, ==, !=, in, not in) work with sets, but the interpretation of the operators is based on set theory. The various operations from set theory are the subset and proper subset relationships. The various comparison mathematical operations of , , , are implemented by <, <=, >, >=. In the following example, the set craps is all of the ways we can roll craps on a come out roll. Also, weve dened three to hold both of the dice rolls that total 3. When we compare three with craps, we see the expected relationships: three is a subset craps as well as a proper subset of craps.
>>> craps= set( [ (1,1), (2,1), (1,2), (6,6) ] ) >>> three = set( [ (1,2), (2,1) ] ) >>> three < craps True >>> three <= craps True
The in and not in operators implement that and / relationships. In the following example, the set craps is all of the ways we can roll craps on a come out roll. Weve modeled a throw of the dice as a 2-tuple. We can now test a specic throw to see if it is craps.
190
Chapter 17. Sets
>>> craps= set( [ (1,1), (2,1), (1,2), (6,6) ] ) >>> (1,2) in craps True >>> (3,4) in craps False
17.5 Set Statements

The for statement works directly with set objects, because they are iterable. A set is not a sequence, but it is like a sequence because we can iterate through the elements using a for statement. Here we create three set objects: even, odd, and zero to reect some standard outcomes in Roulette. The union of all three sets is the complete set of possible spins. We can iterate through this resulting set.
>>> >>> >>> >>> even= set( range(2,38,2) ) odd= set( range(1,37,2) ) zero= set( (0,'00') ) for n in even|odd|zero: print n
17.6 Set Built-in Functions

A number of built-in functions create or process set objects. The set() and frozenset() were described above, under Set Literal Values. Functions which apply to sets, but are dened elsewhere.
17.5. Set Statements
191
len(). For sets, this function returns the number of items.

>>> len( set( [1,1,2,3] ) ) 3 >>> len( set() ) 0
Note that sets do not include duplicates, thats why the length in the rst example is not 4. max(). For sets, this function returns the maximum item.
>>> max( set( [1,1,2,3,5,8] ) ) 8
min(). For sets, this function returns the minimum item. sum(). For sets, this function sums the items.
>>> sum( set( [1,1,2,3,5,8] ) ) 19
Note that sets do not include duplicates, thats why the sum is not 20. any(). For sets, Return True if there exists any item which is True.
>>> set( [0, None, False] ) set([0, None]) >>> any( _ ) False >>> any( set( [0,None,False,42] ) ) True
Note that False and 0 have the same value when constructing a set, and are duplicates. all(). For sets, Return True if all items are True.
>>> all( set( [0,None,False,42] ) ) False >>> all( set( [1,True] ) ) True
enumerate(). Iterate through the set returning 2-tuples of ( index, item ). Since sets have no explicit ordering to their items, this enumeration is in an arbitrary order. sorted(). Iterate through the set elements in sorted order. This returns a set of elements.
>>> sorted( set( [1,1,2,3,5,8] ) ) [1, 2, 3, 5, 8]
17.7 Set Methods

A set object has a number of member methods. The following mutators update a set object. Note that most of these methods dont return a value. The exception is pop.
192
Chapter 17. Sets
clear() Remove all items from the set. pop() Remove an arbitrary object from the set, returning the object. If the set was already empty, this will raise a KeyError exception. add(new ) Adds element new to the set. If the object is already in the set, nothing happens. remove(old ) Removes element old from the set . If the object old is not in the set , this will raise a KeyError exception. discard() Same a set.remove(). update(new ) Merge values from the new set into the original set, adding elements as needed. It is equivalent to the following Python statement. s |= new. intersection_update(new ) Update set to have the intersection of set and new. In eect, this discards elements from set, keeping only elements which are common to new and set It is equivalent to the following Python statement. s &= new. difference_update(new ) Update set to have the dierence between set and new. In eect, this discards elements from set which are also in new. It is equivalent to the following Python statement. s -= new. symmetric_difference_update(new ) Update set to have the symmetric dierence between set and new. In eect, this both discards elements from s which are common with new and also inserts elements into s which are unique to new. It is equivalent to the following Python statement. s ^= new. The following transformers built a new object from one or more sets. copy() Copy the set to make a new set. This is a shallow copy. All objects in the new set are references to the same objects as the original set. union(new ) If new is a proper set, return set | new. If new is a sequence or other iterable, make a new set from the value of new, then return the union, set | new. This does not update the original set.
>>> prime.union( (1, 2, 3, 4, 5) ) set([1, 2, 3, 4, 5, 7, 11, 13])
intersection(new ) If new is a proper set, return set & new. If new is a sequence or other iterable, make a new set from the value of new, then return the intersection, set & new. This does not update set. difference(new ) If new is a proper set, return set - new. If new is a sequence or other iterable, make a new set from the value of new, then return the dierence, set - new. This does not update set.
17.7. Set Methods
193
symmetric_difference(new ) If new is a proper set, return s ^ new. If new is a sequence or other iterable, make a new set from the value of new, then return the symmetric dierence, sset ^ new. This does not update s . The following accessor methods provide information about a set. issubset(other ) If set is a subset of other, return True, otherwise return False. Essentially, this is set <= other . issuperset(other ) If set is a superset of other, return True , otherwise return False. Essentially, this is set >= other.
17.8 Using Sets as Function Parameter Defaults

Its very, very important to note that default values must be immutable objects. Recall that numbers, strings, None, and tuple objects are immutable. We note that sets as well as dictionaries and lists are mutable, and cannot be used as default values for function parameters. Consider the following example of what not to do.
>>> def default2( someSet=set() ): ... someSet.add(2) ... return someSet ... >>> looks_good= set() >>> default2( looks_good ) set([2]) >>> looks_good set([2]) >>> >>> >>> not_good= default2() >>> not_good set([2]) >>> worse= default2() >>> worse set([2]) >>> >>> not_good.add(3) >>> not_good set([2, 3]) >>> worse set([2, 3])
1. We dened a function which has a default value thats a mutable object. This is simple a bad programming practice in Python. 2. We used this function with a set object, looks_good. The function updated the set object as expected. 3. We used the functions default value to create not_good. The function inserted a value into an empty set and returned this new set object. It turns out that the function updated the mutable default value, also. 4. When we use the functions default value again, with worse, the function uses the updated default value and updates it again.
194
Chapter 17. Sets
Both not_good and worse are references to the same mutable object that is being updated. To avoid this, do not use mutable values as defaults. Do this instead.
def default2( someSet=None ): if someSet is None: someSet= {} someSet.add( 2 ) return someSet
17.9 Set Exercises

1. Dice Rolls. In Craps, each roll of the dice belongs to one of several set s of rolls that are used to resolve bets. There are only 36 possible dice rolls, but its annoying to dene the various set s manually. Heres a multi-step procedure that produces the various set s of dice rolls around which you can dene the game of craps. First, create a sequence with 13 empty set s, call it dice. Something like [ set() ]*13 doesnt work because it makes 13 copies of a single set object. Youll need to use a for statement to evaluate the set constructor function 13 dierent times. What is the rst index of this sequence? What is the last entry in this sequence? Second, write two, nested, for-loops to iterate through all 36 combinations of dice, creating 2- tuple s. The 36 2-tuple s will begin with (1,1) and end with (6,6). The sum of the two elements is an index into dice. We want to add each 2- tuple to the appropriate set in the dice sequence. When youre done, you should see results like the following:
>>> dice[7] set([(5, 2), (6, 1), (1, 6), (4, 3), (2, 5), (3, 4)])
Now you can dene the various rules as sets built from other sets. lose On the rst roll, you lose if you roll 2, 3 or 12. This is the set dice[2] | dice[3] | dice[12]. The game is over. win On the rst roll, you win if you roll 7 or 11. The game is over. This is dice[7] | dice[11]. point On the rst roll, any other result (4, 5, 6, 8, 9, or 10) establishes a point. The game runs until you roll the point or a seven. craps Once a point is established, you win if you roll the points number. You lose if you roll a 7. Once you have these three sets dened, you can simulate the rst roll of a craps game with a relatively elegant-looking program. You can generate two random numbers to create a 2-tuple. You can then check to see if the 2-tuple is in the lose or win sets. If the come-out roll is in the point set, then the sum of the 2-tuple will let you pick a set from the dice sequence. For example, if the come-out roll is (2,2), the sum is 4, and youd assign dice[4] to the variable point; this is the set of winners for the rest of the game. The set of losers for the rest of the game is always the craps set.
17.9. Set Exercises
195
The rest of the game is a simple loop, like the come-out roll loop, which uses two random numbers to create a 2- tuple. If the number is in the point set, the game is a winner. If the number is in the craps set, the game is a loser, otherwise it continues. 2. Roulette Results. In Roulette, each spin of the wheel has a number of attributes like even-ness, low-ness, red-ness, etc. You can bet on any of these attributes. If the attribte on which you placed bet is in the set of attributes for the number, you win. Well look at a few simple attributes: red-black, even-odd, and high-low. The even-odd and high-low attributes are easy to compute. The red-black attribute is based on a xed set of values.
redNumbers= set( [1,3,5,7,9,12,14,16,18,19,21,23,25,27,30,32,34,36] )
We have to distinguish between 0 and 00, which makes some of this decision-making rather complex. We can, for example, use ordinary integers for the numbers 0 to 36, and append the string 00 to this set of numbers. For example, set( range(37) ) | set( ['00'] ). This set is the entire Roulette wheel, we can call it wheel. We can dene a number of set s that stand for bets: red, black, even, odd, high and low. We can iterate though the values of wheel, and decide which set s that value belongs to. If the spin is non-zero and spin % 2 == 0, add the spin to the even set. If the spin is non-zero and spin % 2 != 0, add the spin to the odd set. If the spin is non-zero and its in the redNumbers set, add the spin to the red set. If the spin is non-zero and its not in the redNumbers set, add the value to the black set. If the spin is non-zero and spin <= 18, add the value to the low set. If the spin is non-zero and spin > 18, add the value to the high set. Once you have these six sets dened, you can use them to simulate Roulette. Each round involves picking a random spin with something like random.choice( list(wheel) ). You can then see which set the spin belongs to. If the spin belongs to a set on which youve bet, the spin is a winner, otherwise its a loser. These six sets all pay 2:1. There are a some set s which pay 3:1, including the 1-12, 13-24, 25 to 36 ranges, as well as the three columns, spin % 3 == 0, spin % 3 == 1 and spin % 3 == 2. There are still more bets on the Roulette table, but the set s of spins for those bets are rather complex to dene. 3. Sieve of Eratosthenes. Look at Sieve of Eratosthenes. We created a list of candidate prime numbers, using a sequence with 5000 boolean ags. We can, without too much work, simplify this to use a set instead of a list.
Sieve of Eratosthenes - Set Version

(a) Initialize Create a set, prime which has integers between 2 and 5000. Set p 2 (b) Iterate. While 2 p < 5000: Find Next Prime. while not primep and 2 p < 5000: Increment p by 1. Remove Multiples. At this point, p is prime.
196
Chapter 17. Sets
Set k p + p while k < 5000: Remove k from the set prime Set k k + p Next p. Increment p by 1. (c) Report. At this point, the set prime has the prime numbers. We can return the set. In the Find Next Prime step, youre really looking for the minimum in the prime set which is greater than or equal to p. In the Remove Multiples step, you can create the set of multiples, and use difference_update() to remove the multiples from prime. You can, also, use the range() function to create multiples of p, and create a set from this sequence of multiples.
17.9. Set Exercises
197
198
Chapter 17. Sets
CHAPTER
EIGHTEEN
EXCEPTIONS
The try, except, nally and raise statements
A well-written program should produce valuable results even when exceptional conditions occur. A program depends on numerous resources: memory, les, other packages, input-output devices, to name a few. Sometimes it is best to treat a problem with any of these resources as an exception, which interrupts the normal sequential ow of the program. In Exception Semantics we introduce the semantics of exceptions. Well show the basic exception-handling features of Python in Basic Exception Handling and the way exceptions are raised by a program in Raising Exceptions. Well look at a detailed example in An Exceptional Example. In Complete Exception Handling and The nally Clause, we cover some additional syntax thats sometimes necessary. In Exception Functions, well look at a few standard library functions that apply to exceptions. We descibe most of the built-in exceptions in Built-in Exceptions. In addition to exercises in Exception Exercises, we also include style notes in Style Notes and a digression on problems that can be caused by poor use of exceptions in A Digression.
18.1 Exception Semantics

An exception is an event that interrupts the ordinary sequential processing of a program. When an exception is raised, Python will handle it immediately. Python does this by examining except clauses associated with try statements to locate a suite of statements that can process the exception. If there is no except clause to handle the exception, the program stops running, and a message is displayed on the standard error le. An exception has two sides: the dynamic change to the sequence of execution and an object that contains information about the exceptional situation. The dynamic change is initiated by the raise statement, and can nish with the handlers that process the raised exception. If no handler matches the exception, the programs execution eectively stops at the point of the raise. In addition to the dynamic side of an exception, an object is created by the raise statement; this is used to carry any information associated with the exception. Consequences. The use of exceptions has two important consequences. First, we need to clarify where exceptions can be raised. Since various places in a program will raise exceptions, and these can be hidden deep within a function or class, their presence should be announced by specifying the possible exceptions in the docstring. Second, multiple parts of a program will have handlers to cope with various exceptions. These handlers should handle just the meaningful exceptions. Some exceptions (like RuntimeError or MemoryError) generally cant 199
be handled within a program; when these exceptions are raised, the program is so badly broken that there is no real recovery. Exceptions are a powerful tool for dealing with rare, atypical conditions. Generally, exceptions should be considered as dierent from the expected or ordinary conditions that a program handles. For example, if a program accepts input from a person, exception processing is not appropriate for validating their inputs. Theres nothing rare or uncommon about a person making mistakes while attempting to enter numbers or dates. On the other hand, an unexpected disconnection from a network service is a good candidate for an exception; this is a rare and atypical situation. Examples of good exceptions are those which are raised in response to problems with physical resources like les and networks. Python has a large number of built-in exceptions, and a programmer can create new exceptions. Generally, it is better to create new exceptions rather than attempt to stretch or bend the meaning of existing exceptions.
18.2 Basic Exception Handling

Exception handling is done with the try statement. The try statement encapsulates several pieces of information. Primarily, it contains a suite of statements and a group of exception-handling clauses. Each exception-handling clause names a class of exceptions and provides a suite of statements to execute in response to that exception. The basic form of a try statement looks like this:
try: suite except exception , target : suite except: suite
Each suite is an indented block of statements. Any statement is allowed in the suite. While this means that you can have nested try statements, that is rarely necessary, since you can have an unlimited number of except clauses on a single try statement. If any of the statements in the try suite raise an exception, each of the except clauses are examined to locate a clause that matches the exception raised. If no statement in the try suite raises an exception, the except clauses are silently ignored. The rst form of the except clause provides a specic exception class which is used for matching any exception which might be raised. If a target variable name is provided, this variable will have the exception object assigned to it. The second form of the except clause is the catch-all version. This will match all exceptions. If used, this must be provided last, since it will always match the raised exception. Well look at the additional nally clause in a later sections. Important: Python 3 The except statement cant easily handle a list of exception classes. The Python 2 syntax for this is confusing because it requires some additional () around the list of exceptions.
except ( exception, ... ) , target :
200
Chapter 18. Exceptions
The Python 3 syntax wil be slightly simpler. Using the keyword as will remove the need for the additional () around the list of exceptions.
except exception, ... as target
Overall Processing. The structure of the complete try statement summarizes the philosophy of exceptions. First, try the suite of statements, expecting them work. In the unlikely event that an exception is raised, nd an exception clause and execute that exception clause suite to recover from or work around the exceptional situation. Except clauses include some combination of error reporting, recovery or work-around. For example, a recovery-oriented except clause could delete useless les. A work-around exception clause could returning a complex result for square root of a negative number. First Example. Heres the rst of several related examples. This will handle two kinds of exceptions, ZeroDivisionError and ValueError.
exception1.py
def avg( someList ): """Raises TypeError or ZeroDivisionError exceptions.""" sum= 0 for v in someList: sum = sum + v return float(sum)/len(someList) def avgReport( someList ): try: m= avg(someList) print "Average+15%=", m*1.15 except TypeError, ex: print "TypeError:", ex except ZeroDivisionError, ex: print "ZeroDivisionError:", ex
This example shows the avgReport() function; it contains a try clause that evaluates the avg() function. We expect that there will be a ZeroDivisionError exception if an empty list is provided to avg(). Also, a TypeError exception will be raised if the list has any non-numeric value. Otherwise, it prints the average of the values in the list. In the try suite, we print the average. For certain kinds of inappropriate input, we will print the exceptions which were raised. This design is generally how exception processing is handled. We have a relatively simple, clear function which attempts to do the job in a simple and clear way. We have a application-specic process which handles exceptions in a way thats appropriate to the overall application. Nested :command:try Statements. In more complex programs, you may have many function denitions. If more than one function has a try statement, the nested function evaluations will eectively nest the try statements inside each other. This example shows a function solve(), which calls another function, quad(). Both of these functions have a try statement. An exception raised by quad() could wind up in an exception handler in solve().
18.2. Basic Exception Handling
201
exception2.py
def sum( someList ): """Raises TypeError""" sum= 0 for v in someList: sum = sum + v return sum def avg( someList ): """Raises TypeError or ZeroDivisionError exceptions.""" try: s= sum(someList) return float(s)/len(someList) except TypeError, ex: return "Non-Numeric Data" def avgReport( someList ): try: m= avg(someList) print "Average+15%=", m*1.15 except TypeError, ex: print "TypeError: ", ex except ZeroDivisionError, ex: print "ZeroDivisionError: ", ex
In this example, we have the same avgReport() function, which uses avg() to compute an average of a list. Weve rewritten the avg() function to depend on a sum() function. Both avgReport() and avg() contain try statements. This creates a nested context for evaluation of exceptions. Specically, when the function sum is being evaluated, an exception will be examined by avg() rst, then examined by avgReport(). For example, if sum() raises a TypeError exception, it will be handled by avg(); the avgReport() function will not see the TypeError exception. Function Design. Note that this example has a subtle bug that illustrates an important point regarding function design. We introduced the bug when we dened avg() to return either an answer or an error status code in the form of a string. Generally, things are more complex when we try to mix return of valid results and return of error codes. Status codes are the only way to report errors in languages that lack exceptions. C, for example, makes heavy use of status codes. The POSIX standard API denitions for operating system services are oriented toward C. A program making OS requests must examing the results to see if it is a proper values or an indication that an error occurred. Python, however, doesnt have this limitation. Consequently many of the OS functions available in Python modules will raise exceptions rather than mix proper return values with status code values. In our case, our design for avg() attepts to return either a valid numeric result or a string result. To be correct we would have to do two kinds of error checking in avgReport(). We would have to handle any exceptions and we would also have to examine the results of avg() to see if they are an error value or a proper answer. Rather than return status codes, a better design is to simply use exceptions for all kinds of errors. IStatus codes have no real purposes in well-designed programs. In the next section, well look at how to dene and raise our own exceptions.
202
18.3 Raising Exceptions

The raise statement does two things: it creates an exception object, and immediately leaves the expected program execution sequence to search the enclosing try statements for a matching except clause. The eect of a raise statement is to either divert execution in a matching except suite, or to stop the program because no matching except suite was found to handle the exception. The Exception object created by raise can contain a message string that provides a meaningful error message. In addition to the string, it is relatively simple to attach additional attributes to the exception. Here are the two forms for the raise satement.
raise exceptionClass , value raise exception
The rst form of the raise statement uses an exception class name. The optional parameter is the additional value that will be contained in the exception. Generally, this is a string with a message, however any object can be provided. Heres an example of the raise statement.
raise ValueError, "oh dear me"
This statement raises the built-in exception ValueError with an amplifying string of "oh dear me". The amplifying string in this example, one might argue, is of no use to anybody. This is an important consideration in exception design. When using a built-in exception, be sure that the arguments provided pinpoint the error condition. The second form of the raise statement uses an object constructor to create the Exception object.
raise ValueError( "oh dear me" )
Heres a variation on the second form in which additional attributes are provided for the exception.
ex= MyNewError( "oh dear me" ) ex.myCode= 42 ex.myType= "O+" raise ex
In this case a handler can make use of the message, as well as the two additional attributes, myCode and myType. Dening Your Own Exception. You will rarely have a need to raise a built-in exception. Most often, you will need to dene an exception which is unique to your application. Well cover this in more detail as part of the object oriented programming features of Python, in Classes . Heres the short version of how to create your own unique exception class.
class MyError( Exception ): pass
This single statement denes a subclass of Exception named MyError. You can then raise MyError in a raise statement and check for MyError in except clauses. Heres an example of dening a unique exception and raising this exception with an amplifying string.
18.3. Raising Exceptions
203
quadratic.py
import math class QuadError( Exception ): pass def quad(a,b,c): if a == 0: ex= QuadError( "Not Quadratic" ) ex.coef= ( a, b, c ) raise ex if b*b-4*a*c < 0: ex= QuadError( "No Real Roots" ) ex.coef= ( a, b, c ) raise ex x1= (-b+math.sqrt(b*b-4*a*c))/(2*a) x2= (-b-math.sqrt(b*b-4*a*c))/(2*a) return (x1,x2)
Additional raise Statements. Exceptions can be raised anywhere, including in an except clause of a try statement. Well look at two examples of re-raising an exception. We can use the simple raise statement in an except clause. This re-raises the original exception. We can use this to do standardized error handling. For example, we might write an error message to a log le, or we might have a standardized exception clean-up process.
try: attempt something risky except Exception, ex: log_the_error( ex ) raise
This shows how we might write the exception to a standard log in the function log_the_error() and then re-raise the original exception again. This allows the overall application to choose whether to stop running gracefully or handle the exception. The other common technique is to transform Python errors into our applications unique errors. Heres an example that logs an error and transforms the built-in FloatingPointError into our application-specic error, MyError.
class MyError( Exception ): pass try: attempt something risky except FloatingPointError, e: do something locally, perhaps to clean up raise MyError("something risky failed: %s" % ( e, ) )
This allows us to have more consistent error messages, or to hide implementation details.
18.4 An Exceptional Example

The following example uses a uniquely named exception to indicate that the user wishes to quit rather than supply input. Well dene our own exception, and dene function which rewrites a built-in exception to be our own exception.
204
Well dene a function, ckyorn(), which does a Check for Y or N. This function has two parameters, prompt and help, that are used to prompt the user and print help if the user requests it. In this case, the return value is always a Y or N. A request for help (?) is handled automatically. A request to quit is treated as an exception, and leaves the normal execution ow. This function will accept Q or end-of-le (usually ctrl-D, but also ctrl-Z on Windows) as the quit signal.
interaction.py
class UserQuit( Exception ): pass def ckyorn( prompt, help="" ): ok= 0 while not ok: try: a=raw_input( prompt + " [y,n,q,?]: " ) except EOFError: raise UserQuit if a.upper() in [ 'Y', 'N', 'YES', 'NO' ]: ok= 1 if a.upper() in [ 'Q', 'QUIT' ]: raise UserQuit if a.upper() in [ '?' ]: print help return a.upper()[0]
We can use this function as shown in the following example.

import interaction answer= interaction.ckyorn( help= "Enter Y if finished entering data", prompt= "All done?")
This function transforms an EOFError into a UserQuit exception, and also transforms a user entry of Q or q into this same exception. In a longer program, this exception permits a short-circuit of all further processing, omitting some potentially complex if statements. Details of the ckyorn() Function Our function uses a loop that will terminate when we have successfully interpreted an answer from the user. We may get a request for help or perhaps some uninterpretable input from the user. We will continue our loop until we get something meaningful. The post condition will be that the variable ok is set to True and the answer, a is one of ("Y", "y", "N", "n"). Within the loop, we surround our raw_input() function with a try suite. This allows us to process any kind of input, including user inputs that raise exceptions. The most common example is the user entering the end-of-le character on their keyboard. We handle the built-in EOFError by raising our UserQuit exception. When we get end-of-le from the user, we need to tidy up and exit the program promptly. If no exception was raised, we examine the input character to see if we can interpret it. Note that if the user enters Q or QUIT, we treat this exactly like as an end-of-le; we raise the UserQuit exception so that the program can tidy up and exit quickly. We return a single-character result only for ordinary, valid user inputs. A user request to quit is considered extraordinary, and we raise an exception for that.
18.4. An Exceptional Example
205
18.5 Complete Exception Handling and The nally Clause

A common use case is to have some nal processing that must occur irrespective of any exceptions that may arise. The situation usually arises when an external resource has been acquired and must be released. For example, a le must be closed, irrespective of any errors that occur while attempting to read it. With some care, we can be sure that all exception clauses do the correct nal processing. However, this may lead to a some redundant programming. The nally clause saves us the eort of trying to carefully repeat the same statement(s) in a number of except clauses. This nal step will be performed before the try block is nished, either normally or by any exception. The complete form of a try statement looks like this:
try: suite except exception , target : suite except: suite finally: suite
Each suite is an indented block of statements. Any statement is allowed in the suite. While this means that you can have nested try statements, that is rarely necessary, since you can have an unlimited number of except clauses. The nally clause is always executed. This includes all three possible cases: if the try block nishes with no exceptions; if an exception is raised and handled; and if an exception is raised but not handled. This last case means that every nested try statement with a nally clause will have that nally clause executed. Use a nally clause to close les, release locks, close database connections, write nal log messages, and other kinds of nal operations. In the following example, we use the nally clause to write a nal log message.
def avgReport( someList ): try: print "Start avgReport" m= avg(someList) print "Average+15%=", m*1.15 except TypeError, ex: print "TypeError: ", ex except ZeroDivisionError, ex: print "ZeroDivisionError: ", ex finally: print "Finish avgReport"
18.6 Exception Functions

The sys module provides one function that provides the details of the exception that was raised. Programs with exception handling will occasionally use this function.
206
The sys.exc_info() function returns a 3- tuple with the exception, the exceptions parameter, and a traceback object that pinpoints the line of Python that raised the exception. This can be used something like the following not-very-good example.
exception2.py
import sys import math a= 2 b= 2 c= 1 try: x1= (-b+math.sqrt(b*b-4*a*c))/(2*a) x2= (-b-math.sqrt(b*b-4*a*c))/(2*a) print x1, x2 except: e,p,t= sys.exc_info() print e,p
This uses multiple assignment to capture the three elements of the sys.exc_info() tuple , the exception itself in e, the parameter in p and a Python traceback object in t. This catch-all exception handler in this example is a bad policy. It may catch exceptions which are better left uncaught. Well look at these kinds of exceptions in Built-in Exceptions. For example, a RuntimeError is something you should not bother catching.
18.7 Exception Attributes

Exceptions have one interesting attribute. In the following example, well assume we have an exception object named e. This would happen inside an except clause that looked like except SomeException, e:. Traditionally, exceptions had a message attribute as well as an args attribute. These were used inconsistently. When you create a new Exception instance, the argument values provided are loaded into the args attribute. If you provide a single value, this will also be available as message; this is a property name that references args[0]. Heres an example where we provided multiple values as part of our Exception.
>>> a=Exception(1,2,3) >>> a.args (1, 2, 3) >>> a.message __main__:1: DeprecationWarning: BaseException.message has been deprecated as of Python 2.6 ''
Heres an example where we provided a single value as part of our Exception; in this case, the message attribute is made available.
>>> b=Exception("Oh dear") >>> b.message 'Oh dear'
18.7. Exception Attributes
207
>>> b.args ('Oh dear',)
18.8 Built-in Exceptions

The following exceptions are part of the Python environment. There are three broad categories of exceptions. Non-error Exceptions. These are exceptions that dene events and change the sequence of execution. Run-time Errors. These exceptions can occur in the normal course of events, and indicate typical program problems. Internal or Unrecoverable Errors. These exceptions occur when compiling the Python program or are part of the internals of the Python interpreter; there isnt much recovery possible, since it isnt clear that our program can even continue to operate. Problems with the Python source are rarely seen by application programs, since the program isnt actually running. Here are the non-error exceptions. Generally, you will never have a handler for these, nor will you ever raise them with a raise statement. exception StopIteration This is raised by an iterator when there is no next value. The for statement handles this to end an iteration loop cleanly. exception GeneratorExit This is raised when a generator is closed by having the close() method evaluated. exception KeyboardInterrupt This is raised when a user hits ctrl-C to send an interrupt signal to the Python interpreter. Generally, this is not caught in application programs because its the only way to stop a program that is misbehaving. exception SystemExit This exception is raised by the sys.exit() function. Generally, this is not caught in application programs; this is used to force a program to exit. Here are the errors which can be meaningfully handled when a program runs. exception AssertionError Assertion failed. See the assert statement for more information in The assert Statement exception AttributeError Attribute not found in an object. exception EOFError Read beyond end of le. exception FloatingPointError Floating point operation failed. exception IOError I/O operation failed. exception IndexError Sequence index out of range. exception KeyError Mapping key not found.
208
exception OSError OS system call failed. exception OverflowError Result too large to be represented. exception TypeError Inappropriate argument type. exception UnicodeError Unicode related error. exception ValueError Inappropriate argument value (of correct type). exception ZeroDivisionError Second argument to a division or modulo operation was zero. The following errors indicate serious problems with the Python interepreter. Generally, you cant do anything if these errors should be raised. exception MemoryError Out of memory. exception RuntimeError Unspecied run-time error. exception SystemError Internal error in the Python interpreter. The following exceptions are more typically returned at compile time, or indicate an extremely serious error in the basic construction of the program. While these exceptional conditions are a necessary part of the Python implementation, theres little reason for a program to handle these errors. exception ImportError Import cant nd module, or cant nd name in module. exception IndentationError Improper indentation. exception NameError Name not found globally. exception NotImplementedError Method or function hasnt been implemented yet. exception SyntaxError Invalid syntax. exception TabError Improper mixture of spaces and tabs. exception UnboundLocalError Local name referenced but not bound to a value. The following exceptions are part of the implementation of exception objects. Normally, these never occur directly. These are generic categories of exceptions. When you use one of these names in a catch clause, a number of more more specialized exceptions will match these. exception Exception Common base class for all user-dened exceptions.
18.8. Built-in Exceptions
209
exception StandardError Base class for all standard Python errors. Non-error exceptions (StopIteration, GeneratorExit, KeyboardInterrupt and SystemExit) are not subclasses of StandardError. exception ArithmeticError Base class for arithmetic errors. This is the generic exception class that includes OverflowError, ZeroDivisionError, and FloatingPointError. exception EnvironmentError Base class for errors that are input-output or operating system related. This is the generic exception class that includes IOError and OSError. exception LookupError Base class for lookup errors in sequences or mappings, it includes IndexError and KeyError.
18.9 Exception Exercises

1. Input Helpers. There are a number of common character-mode input operations that can benet from using exceptions to simplify error handling. All of these input operations are based around a loop that examines the results of raw_input and converts this to expected Python data. All of these functions should accept a prompt, a default value and a help text. Some of these have additional parameters to qualify the list of valid responses. All of these functions construct a prompt of the form:
your prompt [ valid input hints ,?,q]:
If the user enters a ?, the help text is displayed. If the user enters a q, an exception is raised that indicates that the user quit. Similarly, if the KeyboardInterrupt or any end-of-le exception is received, a user quit exception is raised from the exception handler. Most of these functions have a similar algorithm.
User Input Function

(a) Construct Prompt. Construct the prompt with the hints for valid values, plus ? and q. (b) While Not Valid Input. Loop until the user enters valid input. Try the following suite of operations. Prompt and Read. Use raw_input() to prompt for and read a reply from the user. Help?. If the user entered ?, provide the help message. Quit?. If the user entered q or Q, raise a UserQuit exception. Other. Try the following suite of operations Convert. Attempt any conversion. Some inputs will involve numeric, or datetime conversions. Validate. If necessary, do any validation checks checks. For some prompts, there will be a xed list of valid answers. There may be a numeric range or a date range. For other prompts, there is no checking required. If the input passes the validation, break out of the loop. This is our hoped-for answer. 210 Chapter 18. Exceptions
In the event of an exception, the user input was invalid. Nothing?. If the user entered nothing, and there is a default value, return the default value. In the event of any other exceptions, this function should generally raise a UserQuit exception. (c) Result. Return the validated user input. Functions to implement ckdate Prompts for and validates a date. The basic version would require dates have a specic format, for example mm/dd/yy. A more advanced version would accept a string to specify the format for the input. Much of this date validation is available in the time module, which will be covered in Dates and Times: the time and datetime Modules. This function not return bad dates or other invalid input. ckint Display a prompt; verify and return an integer value ckitem Build a menu; prompt for and return a menu item. A menu is a numbered list of alternative values, the user selects a value by entering the number. The function should accept a sequence of valid values, generate the numbers and return the actual menu item string. An additional help prompt of "??" should be accepted, this writes the help message and redisplays the menu. ckkeywd Prompts for and validates a keyword from a list of keywords. This is similar to the menu, but the prompt is simply the list of keywords without numbers being added. ckpath Display a prompt; verify and return a pathname. This can use the os.path module for information on construction of valid paths. This should use fstat to check the user input to conrm that it actually exists. ckrange Prompts for and validates an integer in a given range. The range is given as separate values for the lowest allowed and highest allowed value. If either is not given, then that limit doesnt apply. For instance, if only a lowest value is given, the valid input is greater than or equal to the lowest value. If only a highest value is given, the input must be less than or equal to the highest value. ckstr Display a prompt; verify and return a string answer. This is similar to the basic raw_input(), except that it provides a simple help feature and raises exceptions when the user wants to quit. cktime Display a prompt; verify and return a time of day. This is similar to ckdate; a more advanced version would use the time module to validate inputs. The basic version can simply accept a hh:mm:ss time string and validate it as a legal time. ckyorn Prompts for and validates yes/no. This is similar to ckkeywd, except that it tolerates a number of variations on yes (YES, y, Y) and a number of variations on no (NO, n, N). It returns the canonical forms: Y or N irrespective of the input actually given.
18.10 Style Notes

Built-in exceptions are all named with a leading upper-case letter. This makes them consistent with class names, which also begin with a leading upper-case letter. Most modules or classes will have a single built-in exception, often called Error. This exception will be imported from a module, and can then be qualied by the module name. Modules and module qualication is covered in Components, Modules and Packages. It is not typical to have a complex hierarchy of exceptional conditions dened by a module.
18.10. Style Notes
211
18.11 A Digression
Readers with experience in other programming languages may equate an exception with a kind of goto statement. It changes the normal course of execution to a (possibly hard to nd) exception-handling suite. This is a correct description of the construct, which leads to some diicult decision-making. Some exception-causing conditions are actually predictable states of the program. The notable exclusions are I/O Error, Memory Error and OS Error. These three depend on resources outside the direct control of the running program and Python interpreter. Exceptions like Zero Division Error or Value Error can be checked with simple, clear if statements. Exceptions like Attribute Error or Not Implemented Error should never occur in a program that is reasonably well written and tested. Relying on exceptions for garden-variety errors those that are easily spotted with careful design or testing is often a sign of shoddy programming. The usual story is that the programmer received the exception during testing and simply added the exception processing try statement to work around the problem; the programmer made no eort to determine the actual cause or remediation for the exception. In their defense, exceptions can simplify complex nested if statements. They can provide a clear escape from complex logic when an exceptional condition makes all of the complexity moot. Exceptions should be used sparingly, and only when they clarify or simplify exposition of the algorithm. A programmer should not expect the reader to search all over the program source for the relevant exception-handling clause. Future examples, which use I/O and OS calls, will benet from simple exception handling. However, exception laden programs are a problem to interpret. Exception clauses are relatively expensive, measured by the time spent to understand their intent.
212
CHAPTER
NINETEEN
ITERATORS AND GENERATORS

The yield Statement
Weve made extensive use of the relationship between the for statement and various kinds of iterable containers without looking too closely at how this really works. In this chapter, well look at the semantics of iterators in Iterator Semantics; this includes their close relationshp to an iterable container, and the for statement. We can then look at the semantics of generator functions in Generator Function Semantics, and talk about the syntax for dening a generator function in Dening a Generator Function. Well look at other built-in functions we use with iterators in Generator Functions. Well review statements related to the use of iterators in Generator Statements. Well provide more places where iterators are used in Iterators Everywhere, as well as an in-depth example in Generator Function Example. When we see how to dene our own classes of objects, well look at creating our own iterators in Creating or Extending Data Types.
19.1 Iterator Semantics

The easiest way to dene an iterator (and the closely-related concept of generator function) is to look at the for statement. The for statement makes use of a large number of iterator features. This statement is the core use case for iterators, and well use it to understand the interface an iterator must provide. Lets look at the following snippet of code.
for i in ( 1, 2, 3, 4, 5 ): print i
Under the hood, the for statement engages in the following sequence of interactions with an iterable object (the tuple (1,2,3,4,5)). 1. The for statement requests an iterator from the object. The for statement does this by evaluating the iter() function on the object in the in clause. The working denition of iterable is that the object responds to the iter() function by returning an iterator. All of the common containers (str, list, tuple, dict, set) will respond to the iter() function by returning an iterator over the items in the container. A dict iterator will yield the keys in the mapping.
213
2. The for statement uses the next() function to evaluate the the iterators next() method and assigns the resulting object to the target variable. In this case, the variable i is assigned to each object. 3. The for statement evaluates the suite of statements. In this case, the suite is just a print statement. 4. The for statement continues steps 2 and 3 until an exception is raised. If the exception is a StopIteration, this is handled to indicate that the loop has nished normally. The for statement is one side of the interface; the other side is the iterator object itself. From the above scenario, we can see that an iterator must dene a __next__() method that the for statement can use. This method does one of two things. Returns the next item from a sequence (or other container) or Raises the StopIteration exception. To do this, an iterator must also maintain some kind of internal state to know which item in the sequence will be delivered next. When we describe a container as iterable, we mean that it responds to the built-in iter() function by returning an iterator object that can be used by the for statement. All of the sequence containers return iterators; set, dict and les also return iterators. In the case of a dict, the iterator returns the dict keys in no particular order. Iterators in Python. As noted above, all of the containers weve seen so far have the iterable interface. This means that the container will return an iterator object that will visit all the elements (or keys) in the container. It turns out that there are many other uses of iterators in Python. Many of the functions weve looked at work with iterators. Well return to this in Iterators Everywhere. Dening Our Own Iterators. There are two ways to dene our own iterators. We can create an object that has the iterator interface, or we can dene a generator function. Under the hood, a generator function will have the iterator interface, but were saved from having to create a class with all the right method function names. Well look at Generator Functions in Generator Function Semantics. Well look at dening an Iterator class in Data + Processing = Objects.
19.2 Generator Function Semantics

A generator function is a function that can be used by the for statement as if it were an iterator. A generator looks like a conventional function, with one important dierence: a generator includes the yield statement. The essential relationship between a generator function and the for statement is the following. 1. The for statement calls the generator. The generator begins execution and executes statements in the suite up to the rst yield statement. The yield creates the initial value for the for statement to assign. 2. The for statement applies the built-in next() function to the generator functions hidden next() method. The value that was returned by the yield statement is assigned to the target variable. 3. The for statement evaluates its suite of statements. 4. The for statement applies the built-in next() function to the generator functions hidden next() method. The generator resumes execution after the yield statement. When the generator function gets to another yield statement, this value creates a value for the for statement. 214 Chapter 19. Iterators and Generators
5. The for statement continues steps 3 and 4 until the generator executes a return statement (or runs past the end of its suite). Either situation will raise the StopIteration exception. When a StopIteration is raised, it is handled by the for statement as a normal termination of the loop. What we Provide. Generator function denition is similar to function denition (see Functions). We provide three pieces of information: the name of the generator, a list of zero or more parameters, and a suite of statements that yields the output values. The suite of statements must include at least one yield statement. We evaluate a generator in a for statement by following the functions name with () enclosing any argument values. The Python interpreter evaluates the argument values, then applies the generator. This will start execution of the the generators suite. Once the generator has started, the generator and the for statement pass control back and forth. The generator will yield objects and the for statement consumes those objects. This back-and-forth between the for statement and the generator means that the generators local variables are all preserved. In other words, a generator function has a peer relationship with the for statement; its local variables are kept when it yields a value. The for suite and the generator suite could be called coroutines. Example: Using a Generator to Consolidate Information. Lexical scanning and parsing are both tasks that compilers do to discover the higher-level constructs that are present in streams of lower-level elements. A lexical scanner discovers punctuation, literal values, variables, keywords, comments, and the like in a le of characters. A parser discovers expressions and statements in a sequence of lexical elements. Lexical scanning and parsing algorithms consolidate a number of characters into tokens or a number of tokens into a statement. A characteristic of these algorithms is that some state change is required to consolidate the inputs prior to creating each output. A generator provides these characteristics by preserving the generators state each time an output is yielded. In both lexical scanning and parsing, the generator function will be looping through a sequence of input values, discovering a high-level element, and then yielding that element. The yield statement returns the sequence of results from a generator function, and also saves all the local variables, and even the location of the yield statement so that the generators next request will resume processing right after the yield .
19.3 Dening a Generator Function

The presence of the yield statement in a function means that the function is actually a generator object, and will have the an iterator-like interface built automatically. In eect it becomes a stateful object with a next() method dened so it will work with the next() function and for statement and it will raise the StopIteration exception when it returns. The syntax for a function denition is in Function Denition: The def and return Statements ; a generator is similar.
def name ( parameter , ... ): suite
The suite of statements must include at least one yield statement. The yield statement species the values emitted by the generator. Note that the expression is required.
yield expression
19.3. Dening a Generator Function
215
If a return statement is used in the function, it ends the generator by raising the StopIteration exception to alert the for statement. For obvious reasons, the return statement cannot return a value. Heres a complete, but silly example of a generator.
def primes(): yield 2 yield 3 yield 5 yield 7 yield 11 return
In this case, we simply yield a xed sequence of values. After yielding ve values, it exits. Heres how this generator is used by a for statement.
>>> for p in primes(): ... print p 2 3 5 7 11
19.4 Generator Functions

The iter() function can be used to acquire an iterator object associated with a container like a sequence, set, le or dict. We can then manipulate this iterator explicitly to handle some common situations. iter(iterable) Returns the iterator for an object. This iterator interacts with built-in types in obvious ways. For sequences, this will return each element in order. For sets, it will return each element in no particular order. For dictionaries, it will return the keys in no particular order. For les, it will return each line in order. Gettimg an explicit iterator outside a for statement is handy for dealing with data structures (like les) which have a head-body structure. In this case, there are one or more elements (the head) which are processed one way and the remaining elements (the body) which are processed another way. Well return to this in detail in Files. For now, heres a small example.
>>> >>> >>> 2 >>> ... ... 4 6 someSeq = range(2,20,2) seq_iter = iter(someSeq) next(seq_iter) for value in seq_iter: print value, 8 10 12 14 16 18
1. We create a sequence, someSeq. Any iterable object would work here; any sequence, set, dict or le. 2. We create the iterator for this sequence, and assign it to seq_iter. This object has a next() method which is used by the next() function and the for statement. 3. We call next(seq_iter) explicitly to get past one heading item in the sequence.
216
Chapter 19. Iterators and Generators
4. We then provide the iterator to the for statement. The for statement repeatedly evaluates the next() function on the iterator object and executes its suite of statements.
19.5 Generator Statements

What the for statement really does
In Iterative Processing: The for Statement , we dened a for statement using the following summary:
for variable in iterable : suite
We glossed over the iterable, showing how to create simple sequence objects using the range() function or explicit list literals. At this point, we can use a number of data structures that are iterable: they respond the the iter() function by creating an iterator. Also, we can dene generator functions, which are also iterable. The Secret World of for. Once weve looked at generator functions and iterators, we can see what the for statement really does. The purpose of the for statement is to visit each value yielded by an iterator, assigning each value to the variable. Note that there are two parts to this. First, the for variable in object evalates iter(object) to get an iterator. Objects will return an an iterator all primed and ready to yield. A generator function will eectively return itself, all primed and ready to yield. Second, the iterator object (or generator function) must yield a sequence of values. Looking forward, well see many additional applications of the way the for statement works. As we look at designing our own objects in Data + Processing = Objects, well want to assure that our objects work well with the for statement, also.
19.6 Iterators Everywhere

Iterators are ubiquitous. We have up to this point been breezy and casual about all the places iterators are used. Weve looked at many functions (max(), min(), any(), all(), sum(), sorted(), reversed(), and enumerate()) which apply to all the various container classes. Actually, these functions all apply to iterators and our containers return the iterators these functions expect. As a simple example, we can dene our own version of enumerate().
def enumerate( iterable, start=0 ): for item in iterable: yield start, item start += 1
Thats all thats required to write a function which works with an iterable, and is itself an iterator. This style of programming is called functional, and is beyond the scope of this book. Additionally, weve looked at the range() function without looking too closely at its sibling xrange(). 19.5. Generator Statements 217
Important: Python 3 and Range In Python 3, the range() function which created a list object will be replaced with an iterator. To create a list object, youll do this
someList = list( range( 6 ) )
In eect, the xrange() iterator will be renamed range(). The legacy range() function will go away. It turns out that weve been making heavy use of iterators. Functions like sorted(), reversed(), any(), all(), and sum() all work with iterators not simply list objects. Well look at how to create our own iterator objects in Collection Special Method Names for Iterators and Iterable.
19.7 Generator Function Example

Lets look at an example which summarizes some details, yielding the summaries. Assume we have the list of tuples named spins. We want to know how many red spins separate a pair of black spins, on average. We need a function which will yield the count of gaps as it examines the spins. We can then call this function repeatedly to get the gap information.
generator.py
spins = [('red', '18'), ('black', '13'), ('red', '7'), ('red', '5'), ('black', '13'), ('red', '25'), ('red', '9'), ('black', '26'), ('black', '15'), ('black', '20'), ('black', '31'), ('red', '3')] def countReds( aList ): count= 0 for color,number in aList: if color == 'black': yield count count= 0 else: count += 1 yield count gaps= [ gap for gap in countReds(spins) ] print gaps
1. The spins variable denes our sample data. This might be an actual record of spins. 2. We dene our gapCount() function. This function initializes count to show the number of non-blacks before a black. It then steps through the individual spins, in the order presented. For non-blacks, the count is incremented. 3. For black spins, however, we yield the length of the gap between the last black. When we yield a result, the generator produces a result value, and also saves all the other processing information about this function so that it can be continued from this point. When the function is continued, it resumes right after the yield statement: the count will be reset, and the for loop will advance to examine the next number in the sequence.
218
4. When the sequence is exhausted, we also yield the nal count. The rst and last gap counts may have to be discarded for certain kinds of statistical analysis. 5. This gaps statement shows how we use the generator. In this case, we create a list comprehension from the results; the for clause will step through the values yielded by the generator until the it exits normally. This sequence of values is collected into a list that we can the use for statistical analysis.
19.8 Generator Exercises

1. The Sieve of Eratosthones (Again). Look at The Sieve of Eratosthones and The Sieve of Eratosthones. We created a list or a set of candidate prime numbers. This exercise has three parts: initialization, generating the list (or set) or prime numbers, then reporting. In the list version, we had to lter the sequence of boolean values to determine the primes. In the set version, the set contained the primes. Within the Generate step, there is a point where we know that the value of p is prime. At this point, we can yield p. If we yield each value as we discover it, we eliminate the entire report step from the function. 2. The Generator Version of range(). The range() function creates a sequence. For very large sequences, this consumes a lot of memory. You can write a version of range which does not create the entire sequence, but instead yields the individual values. Using a generator will have the same eect as iterating through a sequence, but wont consume as much memory. Dene a generator, genrange(), which generates the same sequence of values as range(), without creating a list object. Check the documentation for the built-in function xrange(). 3. Prime Factors. There are two kinds of positive numbers: prime numbers and composite numbers. A composite number is the product of a sequence of prime numbers. You can write a simple function to factor numbers and yield each prime factor of the number. Your factor() function can accept a number, n, for factoring. The function will test values, f, between 2 and the square root of n to see if the expression n % f == 0 is true. If this is true. then the factor, f, divides n with no remainder; f is a factor. Dont use a simple-looking for -loop; the prime factor of 128 is 2, repeated 7 times. Youll need to use a while loop.
19.8. Generator Exercises
219
220
CHAPTER
TWENTY
FILES
The file Class; The with Statement
Programs often deal with external data; data outside of volatile primary memory. This external data could be persistent data on a le system or transient data on an input-output device. Most operating systems provide a simple, uniform interface to external data via objects of the file class. In File Semantics, we provide an overview of the semantics of les. We cover the most important of Pythons built-in functions for working with les in Built-in Functions. Well review statements for dealing with les in File Statements. In File Methods, we describe some method functions of le objects. Files are a deep, deep subject. Well touch on several modules that are related to managing les in Components, Modules and Packages. These include File Handling Modules and File Formats: CSV, Tab, XML, Logs and Others.
20.1 File Semantics

In one sense a le is a container for a sequence of bytes. A more useful view, however, is that a file is a container of data objects, encoded as a sequence of bytes. Files can be kept on persistent but slow devices like disks. Files can also be presented as a stream of bytes owing through a network interface. Even the users keyboard can be processed as if it was a le; in this case the le forces our software to wait until the person types something. Our operating systems use the abstraction of le as a way to unify access to a large number of devices and operating system services. In the Linux world, all external devices, plus a large number of in-memory data structures are accessible through the le interface. The wide variety of things with le-like interfaces is a consequence of how Unix was originally designed. Since the number and types of devices that will be connected to a computer is essentially innite, device drivers were designed as a simple, exible plug-in to the operating system. For more information on the ubiquity of les, see Additional Background . Files include more than disk drives and network interfaces. Kernel memory, random data generators, semaphores, shared memory blocks, and other things have le interfaces, even though they arent strictly speaking devices. Our OS applies the le abstraction to many things. Python, similarly, extends the le interface to include certain kinds of in-memory buers. All GNU/Linux operating systems make all devices available through a standard le-oriented interface. Windows makes most devices available through a reasonably consistent le interface. Pythons file class provides access to the OS le APIs, giving our applications the same uniform access to a variety of devices. Important: Terminology
221
The terminology is sometimes confusing. We have physical les on our disk, the le abstraction in our operating system, and file objects in our Python program. Our Python file object makes use of the operating system le APIs which, in turn, manipulate the les on a disk. Well try to be clear, but with only one overloaded word for three dierent things, this chapter may sometimes be confusing. We rarely have a reason to talk about a physical le on a disk. Generally well talk about the OS abstraction of le and the Python class of file. Standard Files. Consistent with POSIX standards, all Python programs have three les available: sys.stdin, sys.stdout, sys.stderr. These les are used by certain built-in statements and functions. The print statement (and print() function), for example, writes to sys.stdout by default. The raw_input() function writes the prompt to sys.stdout and reads the input from sys.stdin. These standard les are always available, and Python assures that they are handled consistently by all operating systems. The sys module makes these les available for explict use. Newbies may want to check File Redirection for Newbies for some additional notes on these standard les. File Redirection for Newbies The presence of a standard input le and two standard output les is a powerful tool for simplifying programs and making them more reusable. Programs which read from standard input and write to standard output can be combined into processing sequences, called pipelines. The POSIX-standard syntax for creating a pipeline is shown below.
$ ./step1.py <source.dat | ./step2.py >result.dat
This pipeline has two steps: step1.py reads input from stdin and writes to stdout. Weve told the shell to redirect stdin so that it reads the le source.dat. Weve also told the shell to connect step1.py standard output to step2.py standard input. Weve also redirected step2.py standard output to a le, result.dat. We can simplify our programs when we make the shell responsible for le name and pipeline connections.
20.2 File Organization and Structure

Some operating systems provide support for a large variety of le organizations. Dierent le organizations will include dierent record termination rules, possibly with record keys, and possibly xed length records. The POSIX standard, however, considers a le to be nothing more than a sequence of bytes. It becomes entirely the job of the application program, or libraries outside the operating system to impose any organization on those bytes. The basic file objects in Python consider a le to be a sequence of text characters (ASCII or Unicode) or bytes. The characters can be processed as a sequence of variable length lines; each line terminated with a newline character. Files moved from a Windows environment may contain lines which appear to have an extraneous ASCII carriage return character (r), which is easily removed with the string strip() method. Ordinary text les can be managed directly with the built-in file objects and their methods for reading and writing lines of data. We will cover this basic text le processing in the rest of this chapter. Files which are a sequence of bytes dont properly have line boundaries. Byte-oriented les could include characters (in ASCII or a Unicode encoding) or other data objects encoded as bytes. Well address some byte-oriented les with library modules like pickle and csv.
222
Chapter 20. Files
20.3 Additional Background

The GNU/Linux view of les can be surprising for programmers with a background that focuses on mainframe Z/OS or Windows. This is additional background information for programmers who are new to the POSIX use of the le abstraction. This POSIX view informs how Python works. In the Z/OS world, les are called data sets, and can be managed by the OS catalog or left uncataloged. Uncataloged data sets are fairly common. In the GNU/Linux world, the catalog (called a directory) is seamless, silent and automatic. Files are almost always cataloged In the GNU/Linux world, uncataloged, temporary les are atypical, rarely used, and require a special API call. In the Z/OS world, les are generally limited to disk les and nothing else. This is dierent from the GNU/Linux use of le to mean almost any kind of external device or service. Block Mode Files. File devices can be organized into two dierent kinds of structures: block mode and character mode. Block mode devices are exemplied by magnetic disks: the data is structured into blocks of bytes that can be accessed in any order. Both the media (disk) and read-write head can move; the device can be repositioned to any block as often as necessary. A disk provides direct (sometimes also called random) access to each block of data. Character mode devices are exemplied by network connections: the bytes come pouring into the processor buers. The stream cannot be repositioned. If the buer lls up and bytes are missed, the lost data are gone forever. Operating system support for block mode devices includes le directories and le management utilities for deleting, renaming and copying les. Modern operating systems include le navigators (Finders or Explorers), iconic representations of les, and standard GUI dialogs for opening les from within application programs. The operating system also handles moving data blocks from memory buers to disk and from disk to memory buers. All of the device-specic vagaries are handled by having a variety of device drivers so that a range of physical devices can be supported in a uniform manner by a single operating system software interface. Files on block mode devices are sometimes called seekable. They support the operating system seek() function that can begin reading from any byte of the le. If the le is structured in xed-size blocks or records, this seek function can be very simple and eective. Typically, database applications are designed to work with xed-size blocks so that seeking is always done to a block, from which database rows are manipulated. Character Mode Devices and Keyboards. Operating systems also provide rich support for character mode devices like networks and keyboards. Typically, a network connection requires a protocol stack that interprets the bytes into packets, and handles the error correction, sequencing and retransmission of the packets. One of the most famous protocol stacks is the TCP/IP stack. TCP/IP can make a streaming device appear like a sequential le of bytes. Most operating systems come with numerous client programs that make heavy use of the netrowk, examples include sendmail, ftp, and a web browser. A special kind of character mode le is the console; it usually provides input from the keyboard. The POSIX standard allows a program to be run so that input comes from les, pipes or the actual user. If the input le is a TTY (teletype), this is the actual human users keyboard. If the le is a pipe, this is a connection to another process running concurrently. The keyboard console or TTY is dierent from ordinary character mode devices, pipes or les for two reasons. First, the keyboard often needs to explicitly echo characters back so that a person can see what they are typing. Second, pre-processing must often be done to make backspaces work as expected by people. The echo feature is enabled for entering ordinary data or disabled for entering passwords. The echo feature is accomplished by having keyboard events be queued up for the program to read as if from a le. These same keyboard events are automatically sent to update the GUI if echo is turned on.
20.3. Additional Background
223
The pre-processing feature is used to allow some standard edits of the input before the application program receives the buer of input. A common example is handling the backspace character. Most experienced computer users expect that the backspace key will remove the last character typed. This is handled by the OS: it buers ordinary characters, removes characters from the buer when backspace is received, and provides the nal buer of characters to the application when the user hits the Return key. This handling of backspaces can also be disabled; the application would then see the keyboard events as raw characters. The usual mode is for the OS to provide cooked characters, with backspace characters handled before the application sees any data. Typically, this is all handled in a GUI in modern applications. However, Python provides some functions to interact with Unix TTY console software to enable and disable echo and process raw keyboard input. File Formats and Access Methods. In Z/OS (and Open VMS, and a few other operating systems) les have very specic formats, and data access is mediated by the operating system. In Z/OS, they call these access methods, and they have names like BDAM or VSAM. This view is handy in some respects, but it tends to limit you to the access methods supplied by the OS vendor. The GNU/Linux view is that les should be managed minimally by the operating system. At the OS level, les are just bytes. If you would like to impose some organization on the bytes of the le, your application should provide the access method. You can, for example, use a database management system (DBMS) to structure your bytes into tables, rows and columns. The C-language standard I/O library (stdio) can access les as a sequence of individual lines; each line is terminated by the newline character, n . Since Python is built in the C libraries, Python can also read les as a sequence of lines.
20.4 Built-in Functions

There are two built-in functions that create a new le or open an existing le. open(lename, [mode], [buering] ) Create a Python le object associated with an operating system le. filename is the name of the le. mode can be r, w or a for reading (default), writing or appending. The le will be created if it doesnt exist when opened for writing or appending; it will be truncated when opened for writing. Add a b to the mode for binary les. Add a + to the mode to allow simultaneous reading and writing. If the buffering argument is given, 0 means unbuered, 1 means line buered, and larger numbers specify the buer size. file(lename, [mode], [buering] ) Does the same thing as the open(). This is present so that the name of this factory function matches the class of the object being created. The open() function is more descriptive of what is really going on in the program. The file() function is used for type comparisons. Creating File Name Strings. A lename string can be given as a standard name, or it can use OS-specic punctuation. The standard is to use / to separate elements of a le path; Python can do OS-specic translation. Windows, however, uses \ for most levels of the path, but has a leading device character separated by a : . Rather than force your program to implement the various operating system punctuation rules, Python provides modules to help you construct and process le names. The os.path module should be used to construct le names. Best practice is to use the os.path.join() function to make le names from sequences of strings. Well look at this in File Handling Modules.
224
Chapter 20. Files
The filename string can be a simple le name, also called a relative path string, where the OS rules of applying a current working directory are used to create a full, absolute path. Or the lename string can be a full absolute path to the le. File Mode Strings. The mode string species how the le will be accessed by the program. There are three separate issues addressed by the mode string: opening, text handling and operations. Opening. For the opening part of the mode string, there are three alternatives: r Open for reading. Start at the beginning of the le. If the le does not exist, raise an IOError exception. This is implied if nothing else is specied. w Open for writing. Start at he beginning of the le, overwriting an existing le. If the le does not exist, create the le. a Open for appending. Start at the end of the le. If the le does not exist, create the le. Text Handling. For the text handling part of the mode string, there are two alternatives: b Interpret the le as bytes, not text. (nothing) The default, if nothing is specied is to interpret the content as text: a sequence of characters with newlines at the end of each line. U The capital U mode (when used with r) enables universal newline reading. This allows your program to cope with the non-standard line-ending characters present in some Windows les. The standard end-of-line is a single newline character, \n. In Windows, an additional \r character may also be present. Operations. For the additional operations part of the mode string, there are two alternatives: + Allow both read and write operations. (nothing) If nothing is specied, allow only reads for les opened with r; allow only writes for les opened with w or a. Typical combinations include "rb" to read data as bytes and "w+" to create a text le for reading and writing. Examples. The following examples create le objects for further processing:
myLogin = open( ".login", "r" ) newSource = open( "somefile.c", "w" ) theErrors = open( "error.log", "a" ) someData = open( 'source.dat', 'rb' )
myLogin A text le, opened for reading. newSource A text le, opened for writing. If the le exists, it is overwritten. theErrors A text le, opened for appending. If the le doesnt exist, its created. someData A binary le, opened for reading. Buering les is typically left as a default, specifying nothing. However, for some situations buering can improve performance. Error logs, for instance, are often unbuered, so the data is available immediately. Large input les may have large buer numbers specied to encourage the operating system to optimize input operations by reading a few large chunks of data instead of a large number of smaller chunks.
20.4. Built-in Functions
225
20.5 File Statements

There are a number of statements that have specic features related to tuple objects. The for Statement. Principally, we use the for statement to work with les. Text les are iterable, making them a natural t with the for statement. The most common pattern is the following:
source = open( "someFile.txt", "r" ) for line in source: # process line source.close()
Additionally, we use the with statement with les. This assures that we have without exception closed the le when were done using it. The with Statement. The with statement is used to be absolutely sure that we have closed a le (or other resource) when were done using it. The with statement uses an object called a context manager. This manager object can be assigned to a temporary variable and used in the with statements suite. See Managing Contexts: the with Statement for more information on creating a context manager. The two central features are 1. The context manager will be closed at the end of the with statement. This is guaranteed, irrespective of any exceptions that are raised. 2. A file is a context manager. It will be closed. with Statement Syntax. The with statement has the following syntax.
with expression as variable : suite
A file object conforms to the context manager interface. It has an __enter__() and a __exit__() method. It will be closed at the end of the with statement. Generally, we use this as follows.
with file("somefile","r") as source: for line in source: print line
At the end of the with statement, irrespective of any exceptions which are handled or not handled the le will be closed and the relevant resources released.
20.6 File Methods

The built-in file() function creates a file object. The resulting object has a number of operations that change the state of the le, read or write data, or return information about the le. Read Methods. The following methods read from a le. As data is read, the le position is advanced from the beginning to the end of the le. The le must be opened with a mode that includes or implies 'r' for these methods to work.
226
Chapter 20. Files
read([size] ) Read as many as size characters or bytes from the le. If size is negative or omitted, the rest of the le is read. readline([size] ) Read the next line. If size is negative or omitted, the next complete line is read. If the size is given and positive, read as many as size characters from the le; an incomplete line can be read. If a complete line is read, it includes the trailing newline character, \n. If the le is at the end, this will return a zero length string. If the le has a blank line, the blank like will be a string of length 1 (the newline character at the end of the line.) readlines([hint] ) Read the next lines or as many lines from the next hint characters from le. The value of hint may be rounded up to match an internal buer size. If hint is negative or omitted, the rest of the le is read. All lines will include the trailing newline character, \n. If the le is at the end, this returns a zero length list. Write Methods. The following methods write to a le. As data is written, the le position is advanced, possibly growing the le. If the le is opened for write, the position begins at the beginning of the le. If the le is opened for append, the position begins at the end of the le. If the le does not already exist, both writing and appending are equivalent. The le must be opened with a mode that includes 'a' or 'w' for these methods to work. flush() Flush all accumulated data from the internal buers to the OS le. Depending on your OS, this may also force all the data to be written to the device. write(string) Write the given string to the le. Buering may mean that the string does not appear on any console until a close() or flush() method is used. writelines(list ) Write the list of strings to the le. Buering may mean that the strings do not appear on any console until a close() or flush() operation is used. truncate([size] ) Truncate the le. If size is not given, the le is truncated at the current position. If size is given, the le will be truncated at size. If the le isnt as large as the given size, the results vary by operating system. This function is not available on all platforms. Position Control Methods. The current position of a le can be examined and changed. Ordinary reads and writes will alter the position. These methods will report the position, and allow you to change the position that will be used for the next operation. seek(oset, [whence] ) Change the position from which the le will be processed. There are three values for whence which determine the direction of the move. If whence is zero (or omitted), move to the absolute position given by offset. f.seek(0) will rewind le f. If whence is 1, move relative to the current position by offset bytes. If oset is negative, move backwards; otherwise move forward. If whence is 2, move relative to the end of le. f.seek(0,2) will advance le f to the end, making it possible to append to the le. tell() Return the position from which the le will be processed. This is a partner to the seek() method; any
20.6. File Methods
227
position returned by the tell() method can be used as an argument to the seek() method to restore the le to that position. Other Methods. These are additional useful methods of a le object. close() Close the le, ushing all data. The closed ag is set. Any further operations (except a redundant close) raise an IOError exception. fileno() Return the internal le descriptor (FD) used by the OS library when working with this le. A number of Python modules provide functions that use the OS libraries; the OS libraries need the FD. isatty() Return True if the le is connected to the console or keyboard. Some handy attributes of a le. file.closed -> boolean This attribute is True if the le is closed. file.mode -> string This attribute is the mode argument to the open() function that was used to create the le object. file.name -> string This attribute of is the lename argument to the open() function that was used to create the le object. file.encoding -> string This is the encoding for the le. Many Unicode les will have a Byte Order Mark (BOM) that provides the encoding.
20.7 Several Examples

Well look at four examples of le processing. In all cases, well read simple text les. Well show some traditional kinds of le processing programs and how those can be implemented using Python.
20.7.1 Reading a Text File

The following program will examine a standard unix password le. Well use the explicit readline() method to show the processing in detail. Well use the split() method of the input string as an example of parsing a line of input.
readpswd.py
pswd = file( "/etc/passwd", "r" ) for aLine in pswd fields= aLine.split( ":" ) print fields[0], fields[1] pswd.close()
1. This program creates a file object, pswd, that represents the /etc/passwd le, opened for reading. 2. A file is an iterator over lines of text. We can use a file in the for statement; the file object will return each individual line in response to the iterator.next() method.
228
Chapter 20. Files
3. The input string is split into individual elds using :literal>:: boundaries. Two particular elds are printed. Field 0 is the username and eld 1 is the password. 4. Closing the le releases any resources used by the le processing. For non-unix users, a password le looks like the following:
root:q.mJzTnu8icF.:0:10:Sysadmin:/:/bin/csh fred:6k/7KCFRPNVXg:508:10:% Fredericks:/usr2/fred:/bin/csh
The : separated elds inlcude the user name, password, user id, group id, user name, home directory and shell to run upon login.
20.7.2 Reading a CSV File the Hard Way

This le will have a CSV (Comma-Separated Values) le format that we will parse. The csv module does a far better job than this little program. Well look at that module in Comma-Separated Values: The csv Module. A popular stock quoting service on the Internet will provide CSV les with current stock quotes. The les have comma-separated values in the following format:
stock, lastPrice, date, time, change, openPrice, daysHi, daysLo, volume
The stock, date and time are typically quoted strings. The other elds are numbers, typically in dollars or percents with two digits of precision. We can use the Python eval() function on each column to gracefully evaluate each value, which will eliminate the quotes, and transform a string of digits into a oating-point price value. Well look at dates in Dates and Times: the time and datetime Modules. This is an example of the le:
"^DJI",10623.64,"6/15/2001","4:09PM",-66.49,10680.81,10716.30,10566.55,N/A "AAPL",20.44,"6/15/2001","4:01PM",+0.56,20.10,20.75,19.35,8122800 "CAPBX",10.81,"6/15/2001","5:57PM",+0.01,N/A,N/A,N/A,N/A
The rst line shows a quote for an index: the Dow-Jones Industrial average. The trading volume doesnt apply to an index, so it is N/A. The second line shows a regular stock (Apple Computer) that traded 8,122,800 shares on June 15, 2001. The third line shows a mutual fund. The detailed opening price, days high, days low and volume are not reported for mutual funds. After looking at the results on line, we clicked on the link to save the results as a CSV le. We called it quotes.csv. The following program will open and read the quotes.csv le after we download it from this service.
readquotes.py
qFile= file( "quotes.csv", "r" ) for q in qFile: try: stock, price, date, time, change, opPrc, dHi, dLo, vol\ = q.strip().split( "," ) print eval(stock), float(price), date, time, change, vol except ValueError: pass qFile.close()
20.7. Several Examples
229
1. We open our quotes le, quotes.csv, for reading, creating an object named qFile. 2. We use a for statement to iterate through the sequence of lines in the le. 3. The quotes le typically has an empty line at the end, which splits into zero elds, so we surround this with a try statement. The empty line will raise a ValueError exception, which is caught in the except clause and ignored. 4. Each stock quote, q, is a string. We use the string.strip() method to remove whitespace; on the resulting string we use the string.split() method to split the string on ",". This transforms the input string into a list of individual elds. We use multiple assignment to assign each eld to a relevant variable. Note that we strip this le into nine elds, leading to a long statement. We put a \ to break the statement into two lines. 5. The name of the stock is a string which includes extra quotes. In order to gracefully remove the quotes, we use the eval() function. The price is a string. We could also use eval function to evaluate this string as a Python value. Instead, we use the float() function to convert the price string to a proper numeric value for further processing. As a practical matter, this is a currency value, and we need to use a Decimal value, not a oat value. The decimal module handles currency very nicely.
20.7.3 Read, Sort and Write

For COBOL expatriates, heres an example that shows a short way to read a le into an in-memory sequence, sort that sequence and print the results. This is a very common COBOL design pattern, and it tends to be rather long and complex in COBOL. This example looks forward to some slightly more advanced techniques like list sorting. Well delve into sorting in Functional Programming with Collections.
sortquotes.py
data= [] with file( "quotes.csv", "r" ) as qFile: for q in qFile: fields= tuple( q.strip().split( "," ) ) if len(fields) == 9: data.append( fields ) def priceVolume(a): return a[1], a[8] data.sort( key=priceVolume ) for stock, price, date, time, change, opPrc, dHi, dLo, vol in data: print stock, price, date, time, change, volume
1. We create an empty sequence, data, to which we will append tuples created from splitting each line into elds. 2. We create le object, qFile that will read all the lines of our CSV-format le. 3. This for loop will set q to each line in the le. 4. The variable fields is created by stripping whitespace from the line, q, breaking it up on the "," boundaries into separate elds, and making the resulting sequence of eld values into a tuple.
230
Chapter 20. Files
If the line has the expected nine elds, the tuple is appended to the data, sequence. Lines with the wrong number of elds are typically the blank lines at the beginning or end of the le. 5. To prepare for the sort, we dene a key function. This will extract elds 1 and 8, price and volume. If we dont use a key function, the tuple will be sorted by elds in order. The rst eld is stock name. 6. We can then sort the data sequence. Note that the list.sort() method does not return a value. It mutates the list. The sort method will use our priceVolume() function to extract keys for comparing records. This kind of sort is covered in depth in Advanced List Sorting. 7. Once the sequence of data elements is sorted, we can then print a report showing our stocks ranked by price, and for stocks of the same price, ranked by volume. We could expand on this by using the % operator to provide a nicer-looking report format. Note that we arent obligated to sort the sequence. We can use the sorted() function here, also.
for stock, price, date, time, change, opPrc, dHi, dLo, vol \ in sorted( data, key=priceVolume ): print stock, price, date, time, change, volume
This does not update the data list, but is otherwise identical.
20.7.4 Reading Records

In languages like C or COBOL a record or struct will describe the contents of a le. The advantage of a record is that the elds have names instead of numeric positions. In Python, we can acheive the same level of clarity using a dict for each line in the le. For this, well download les from a web-based portfolio manager. This portfolio manager gives us stock information in a le called display.csv. Here is an example.
+/-,Ticker,Price,Price Change,Current Value,Links,# Shares,P/E,Purchase Price, -0.0400,CAT,54.15,-0.04,2707.50,CAT,50,19,43.50, -0.4700,DD,45.76,-0.47,2288.00,DD,50,23,42.80, 0.3000,EK,46.74,0.30,2337.00,EK,50,11,42.10, -0.8600,GM,59.35,-0.86,2967.50,GM,50,16,53.90,
This le contains a header line that names the data columns, making processing considerably more reliable. We can use the column titles to create a dict for each line of data. By using each data line along with the column titles, we can make our program quite a bit more exible. This shows a way of handling this kind of well-structured information.
readportfolio.py
invest= 0 current= 0 with open( "display.csv", "rU" ) as quotes: titles= quotes.next().strip().split( ',' ) for q in quotes: values= q.strip().split( ',' ) data= dict( zip(titles,values) ) print data invest += float(data["Purchase Price"])*float(data["# Shares"])
20.7. Several Examples
231
current += float(data["Price"])*float(data["# Shares"]) print invest, current, (current-invest)/invest
1. We open our portfolio le, display.csv, for reading, creating a le object named quotes. 2. The rst line of input, varname.next(), is the set of column titles. We strip any extraneous whitespace characters from this line, and then perform a split to create a list of individual column title strs. This list is tagged with the variable titles. 3. We also initialize two counters, invest and current to zero. These will accumulate our initial investment and the current value of this portfolio. 4. We use a for statement to iterate through the remaining lines in quotes le. Each line is assigned to q. 5. Each stock quote, q, is stripped and split to separate the elds into a list. We assign this list to the variable values. 6. We create a dict, data; the column titles in the titles list are the keys. The data elds from the current record, in values are used to ll this dict. The built-in zip() function is designed for precisely this situation: it interleaves values from each list to create a new list of tuples. Now, we have access to each piece of data using its proper column tile. The number of shares is in the column titled "# Shares". We can nd this information in data["# Shares"]. 7. We perform some simple calculations on each dict. In this case, we convert the purchase price to a number, convert the number of shares to a number and multiply to determine how much we spent on this stock. We accumulate the sum of these products into invest. We also convert the current price to a number and multiply this by the number of shares to get the current value of this stock. We accumulate the sum of these products into current. 8. When the loop has terminated, we can write out the two numbers, and compute the percent change.
20.8 File Exercises

1. File Structures. What is required to process variable length lines of data in an arbitrary (random) order? How is the application program to know where each line begins? 2. Device Structures. Some disk devices are organized into cylinders and tracks instead of blocks. A disk may have a number of parallel platters; a cylinder is the stack of tracks across the platters available without moving the read-write head. A track is the data on one circular section of a single disk platter. What advantages does this have? What (if any) complexity could this lead to? How does an application program specify the tracks and sectors to be used? Some disk devices are described as a simple sequence of blocks, in no particular order. Each block has a unique numeric identier. What advantages could this have? Some disk devices can be partitioned into a number of logical devices. Each partition appears to be a separate device. What (if any) relevance does this have to le processing? 3. Portfolio Position. We can create a simple CSV le that contains a description of a block of stock. Well call this the portfolio le. If we have access to a spreadsheet, we can create a simple le with four columns: stock, shares, purchase date and purchase price. We can save this as a CSV le. If we dont have access to a spreadsheet, we can create this le in IDLE. Heres an example line.
232
Chapter 20. Files
stock,shares,"Purchase Date","Purchase Price" "AAPL", 100, "10/1/95", 14.50 "GE", 100, "3/5/02", 38.56
We can read this le, multiply shares by purchase price, and write a simple report showing our initial position in each stock. Note that each line will be a simple string. When we split this string on the ,s (using the string split() method) we get a list of strings. Well still need to convert the number of shares and the purchase price from strings to numbers in order to do the multiplication. 4. Aggregated Portfolio Position. In Porfolio Position we read a le and did a simple computation on each row to get the purchase price. If we have multiple blocks of a given stock, these will be reported as separate lines of detail. Wed like to combine (or aggregate) any blocks of stock into an overall position. Programmers familiar with COBOL (or RPG) or similar languages often use a ControlBreak reporting design which sorts the data into order by the keys, then reads the lines of data looking for break in the keys. This design uses very little memory, but is rather slow and complex. Its far simpler to use a Python dictionary than it is to use the Control-Break algorithm. Unless the number of distinct key values is vast (on the order of hundreds of thousands of values) most small computers will t the entire summary in a simple dictionary. A program which produces summaries, then, would have the following design pattern. (a) Create an empty dictionary for retaining aggregates. (b) Open and read the header line from the le. This has the eld names. (c) Read the portfolio le. For each line in the le, do the following. i. Create a tuple from the data elds. You can create a row dictionary from a zip() of the header elds zipped with a row. ii. If the stock name key does not exist in the aggregate dictionary, insert the necessary element, and provide a suitable initial value. iii. Locate the stock name in the dictionary, accumulate a new aggregate value. (d) Write the aggregate dictionary keys and values as the nal report. Some people like to see the aggregates sorted into order. This is a matter of using sorted() to iterate through the dictionary keys in the desired order to write the nal report. 5. Portfolio Value. In Reading a CSV File the Hard Way, we looked at a simple CSV-format le with stock symbols and prices. This le has the stock symbol and last price, which serves as a daily quote for this stocks price. Well call this the stock-price le. We can now compute the aggregate value for our portfolio by extracting prices from the stock price le and number of shares from the portfolio le. If youre familiar with SQL, this is called a join operation ; and most databases provide a number of algorithms to match rows between two tables. If youre familiar with COBOL, this is often done by creating a lookup table, which is an in-memory array of values. Well create a dictionary from the stock-price le. We can then read our portfolio, locate the price in our dictionary, and write our nal report of current value of the portfolio. This leads to a program with the following design pattern. (a) Load the price mapping from the stock-price le. i. Create an empty stock price dictionary.
20.8. File Exercises
233
ii. Read the stock price le. For each line in the le, populate the dictionary, using the stock name as the key, and the most recent sale price is the value. (b) Process the position information from the portfolio le. See Aggregated Portfolio Position and Porfolio Position for the skeleton of this process. In the case of a stock with no price, the program should produce a no price quote line in the output report. It should not produce a KeyError exception.
234
Chapter 20. Files
CHAPTER
TWENTYONE
FUNCTIONAL PROGRAMMING WITH COLLECTIONS

This chapter presents some advanced collection concepts. In Lists of Tuples we describe the relatively common Python data structure built from a list of tuples. Well cover a powerful list construction method called a list comprehension in List Comprehensions. We can use this to simplify some common list-processing patterns. In Sequence Processing Functions: map(), lter() and reduce() well cover three functions that can simplify some list processing. map(), filter() and reduce() provide features that overlap with list comprehensions. In Advanced List Sorting we cover some advanced sequence sorting techniques. In particular, well look closely at how to provide a suitable key function to control sorting. Well look at lambda forms in The Lambda. These arent essential for Python programming, but theyre handy for clarifying a piece of code in some rare cases. In Multi-Dimensional Arrays or Matrices we cover simple multidimensional sequences. These are sometimes called matrices or arrays. Even more complex data structures are available, of course. Numerous modules handle the sophisticated representation schemes described in the Internet standards (called Requests for Comments, RFCs). Well touch on these in The Python Library.
21.1 Lists of Tuples

The list of tuple structure is remarkably useful. In other languages, like Java or C++, we are forced to either use built-in arrays or create an entire class denition to simply keep a few values togther. One common situation is processing list of simple coordinate pairs for 2-dimensional or 3-dimensional geometries. Additional examples might include examining a list of tuples that contain the three levels for red, green and blue that dene a color. Or for printing the values for cyan, magenta, yellow and black that dene a color. As an example of using a red, green, blue tuple, we may have a list of individual colors that looks like the following.
colorScheme = [ (0,0,0), (0x20,0x30,0x20), (0x10,0xff,0xff) ]
Weve already seen how dictionaries (Mappings and Dictionaries) have an dict.items() method that provides the dictionary keys and values as a list of 2-tuples. Additionally, the zip() built-in function interleaves two or more sequences to create a list of tuples.
235
The for Statement. A interesting form of the for statement is one that exploits multiple assignment to work with a list of tuples. Consider the following examples:
for c,f in [ ("red",18), ("black",18), ("green",2) ]: print "%s occurs %f " % (c, f/38.0) for r, g, b in [ (0,0,0), (0x20,0x30,0x20), (0x10,0xff,0xff) ]: print "red: %x, green: %x, blue: %x" % ( 255-r, 255-g, 255-b )
In these examples, we have created a list of tuples. The for statement uses a form of multiple assignment to split up each tuple into a xed number of variables. The rst example is equivalent to the following.
for p in [ ("red",18), ("black",18), ("green",2) ]: c,f = p print "%s occurs %f " % (c, f/38.0)
This technique works because tuples are expected to have a xed, known number of elements. Heres an example using dict.items(). We looked at dictionaries in Mappings and Dictionaries.
d = { 'red':18, 'black':18, 'green':2 } for c,f in d.items(): print "%s occurs %f " % (c, f/38.0)
21.2 List Comprehensions

Python provides several list literals or displays. The most common list display is the simple literal value shown in List Literal Values: values are written as follows:
[ expression , ... ]
Python has a second kind of list display, based on a list comphrehension. A list comprehension is an expression that combines a function, a for statement and an optional if statement into one tidy package. This allows a simple, clear expression of the processing that will build up an iterable sequence. List Comprehension Semantics. The most important thing about a list comprehension is that it is an iterable that applies a calculation to another iterable. A list display can use a list comprehension iterable to create a new list. When we write a list comprehension, we will provide an iterable, a variable and an expression. Python will process the iterator as if it was a for-loop, iterating through a sequence of values. It evaluates the expression, once for each iteration of the for-loop. The resulting values can be collected into a fresh, new list, or used anywhere an iterator is used. List Comprehension Syntax. A list comprehension is technically a complex expression. Its often used in list displays, but can be used in a variety of places where an iterator is expected.
expr for-clause
The expr is any expression. It can be a simple constant, or any other expression (including a nested list comprehension). The for-clause mirrors the for statement:
236
Chapter 21. Functional Programming with Collections
for variable in sequence
A common use for this is in a list display. Well show some list comprehension examples used to create new lists.
even = [ 2*x for x in range(18) ] hardways = [ (x,x) for x in (2,3,4,5) ] samples = [ random.random() for x in range(10) ]
even This is a list of values [0, 2, 4, ..., 14]. hardways This is a list of 2-tuples. Each 2-tuple is built from the values in the given sequence. The result is [(2,2), (3,3), (4,4), (5,5)]. samples This is a list of 10 random numbers. A list display that uses a list comprehension behaves like the following loop:
r= [] for variable in sequence : r.append( expr )
The basic process, then, is to iterate through the sequence in the for-clause, evaluating the expression, expr. The values that result are assembled into the list. If the expression depends on the for-clause, each value in the list can be dierent. If the expression doesnt depend on the for-clause, each value will be the same. Heres an example where the expression depends on the for-clause.
>>> [ v*2+1 for v in range(10) ] [1, 3, 5, 7, 9, 11, 13, 15, 17, 19] >>> sum(_) 100
This creates the rst 10 odd numbers. It starts with the sequence created by range(10). The for-clause assigns each value in this sequence to the local variable v. The expression, v*2+1, is evaluated for each distinct value of v. The expression values are assembled into the resulting list. Heres an example where the expression doesnt depend on the for-clause.
b= [ 0 for i in range(10) ]
b will be a list of 10 zeroes. Comprehensions Outside List Displays. A list comprehension can be used outside of a list display. When we write a list display (using [ and ]) were using an iterable (the list comprehension) to create a new list. We can use the iterable list comprehension in other contexts that expect an iterator.
square = sum( (2*a+1) for a in range(10) ) column_1 = tuple( 3*b+1 for b in range(12) ) rolls = ( (random.randint(1,6),random.randint(1,6)) for u in range(100) ) hardways = any( d1==d2 for d1,d2 in rolls )
square This is the sum of odd numbers. The list comprehension ((2*a+1) for a in range(10)) is an iterable which the sum() function can use. column_1 This creates a tuple of 12 values using a a list comprehension. 21.2. List Comprehensions 237
rolls This creates a generator object that will iterate over 100 values using a list comprehension. This does not actually create 100 values it denes an iterator that will produce 100 values. Note that this generator has an internal state: it can only be used once. Once it has generated all its values, it will not generate any other values. A statement like r1 = list(rolls) will use the iterator object, rolls with the list() function to actually create an object that has 100 random values. hardways This iterates through the iterator named rolls to see if any of the pairs had the number rolled the hard way with both values equal. The if Clause. A list comprehension can also have an if-clause. The basic syntax is as follows:
expr for-clause if-clause
The for-clause mirrors the for statement:

for variable in sequence
The if-clause mirrors the if statement:

if filter
Here is an example of a complex list comprehension in a list display.

hardways = [ (x,x) for x in range(1,7) if x+x not in (2, 12) ]
This more complex list comprehension behaves like the following loop:
r= [] for variable in sequence : if filter: r.append( expr )
The basic process, then, is to iterate through the sequence in the for-clause, evaluating the if-clause. When the if-clause is True, evaluate the expression, expr. The values that result are assembled into the list. Heres another example.
>>> [ (x,2*x+1) for x in range(10) if x%3 == 0 ] [(0, 1), (3, 7), (6, 13), (9, 19)]
This works as follows: 1. The for-clause iterates through the 10 values given by range(10), assigning each value to the local variable x. 2. The if-clause evaluates the lter function, x%3==0. If it is False, the value is skipped. If it is True, the expression, at (x,2*x+1), is evaluated and retained. 3. The sequence of 2-tuples are assembled into a list. A list comprehension can have any number of for-clauses and if-clauses, freely-intermixed. A for-clause must be rst. The clauses are evaluated from left to right.
238
21.3 Sequence Processing Functions: map(), filter() and reduce()

The map(), filter(), and reduce() built-in functions are handy functions for processing sequences without writing lengthy for-loops. These owe much to the world of functional programming languages. The idea of each is to take a small function you write and apply it to all the elements of a sequence, saving you from writing an explicit loop. The implicit loop within each of these functions may be faster than an explicit for loop. Additionally, each of these is a pure function, returning a result value. This allows the results of the functions to be combined into complex expressions relatively easily. Processing Pipeline. It is very, very common to apply a single function to every value of a list. In some cases, we may apply multiple simple functions in a kind of processing pipeline. Heres an example.
>>> def ftin_2_in( ftin ): ... feet, inches = ftin ... return 12.0*feet + inches ... >>> heights = [ (5,8), (5,9), (6,2), (6,1), (6,7) ] >>> map( ftin_2_in, heights ) [68.0, 69.0, 74.0, 73.0, 79.0] >>> >>> def in_2_m( inches ): ... return inches * 0.0254 ... >>> map( in_2_m, map( ftin_2_in, heights ) ) [1.7271999999999998, 1.7525999999999999, 1.8795999999999999, 1.8541999999999998, 2.0065999999999997]
1. We dened a simple function, ftin_2_in() that converts a distance in the form (ft,in) into a distance measured in inches. 2. We dened a set of peoples heights in heights. 3. We used the map() function to apply our ftin_2_in() function to each value in heights, creating a new list of values. 4. We dened a simple function, in_2_m() that converts a distance in inches into a distance in meters. 5. We did a fairly complex calculation where we applied ftin_2_in() to each value in heights. We then applied in_2_m() to each of those values. Weve converted a list of values from English (ft,in) to proper metric units by applying two simple functions to each value. This concept can be used extensively using these functions and list comprehensions to create complex and sophisticated software from a series of simple transformations. Denitions. Each of the map(), filter() and reduce() functions transform an iterable (a sequence or generator function). The map() and filter() each apply some function to a sequence to create a new sequence. The reduce() function applies a function which will reduce the sequence to a single value. There are a number of special-purpose reduce functions that weve already seen. These reductions include sum(), any(), all(). The map() and filter() functions have no internal state, they simply apply the function to each individual value of the sequence. The reduce() function, in contrast, maintains an internal state which is seeded from an initial value, passed to the function along with each value of the sequence and returned as the nal result. Here are the formal denitions. 21.3. Sequence Processing Functions: map(), filter() and reduce() 239
map(function, sequence, [...] ) Create a new list from the results of applying the given function to the items of the the given sequence.
>>> map( int, [ "10", "12", "14", 3.1415926, 5L ] ) [10, 12, 14, 3, 5]
This function behaves as if it had the following denition.

def map( aFunction, aSequence ): return [ aFunction(v) for v in aSequence ]
It turns out that more than one sequence can be given. In this case, the function must accept multiple arguments, and there must be as many sequences as arguments to the function. The corresponding items from each sequence are provided as arguments to the function. If any sequence is too short, None is used for missing the values. If the function is None, map() will create tuples from corresponding items in each list, much like the zip() function. filter(function, sequence) Return a list containing those items of sequence for which the given function is True. If the function is None, return a list of items that are equivalent to True. This function behaves as if it had the following denition.
def filter( aFunction, aSequence ): return [ v for v in aSequence if aFunction(v) ]
Heres an example
>>> import random >>> rolls = list( (random.randint(1,6),random.randint(1,6)) for u in range(100) ) >>> def hardways( pair ): ... d1, d2 = pair ... return d1 == d2 and d1+d2 in ( 4, 6, 8, 10 ) >>> filter( hardways, rolls ) [(4, 4), (5, 5), (2, 2), (5, 5), (4, 4), (5, 5), (5, 5), (3, 3), (2, 2), (2, 2), (5, 5), (4, 4)] >>> len(_) 12
1.We created 100 random rolls. 2.We dened a function that evaluates a roll to see if the point was made the hard way. 3.We applied our lter to our rolls to see that 12/100 rolls are hardway rolls. Heres anther example
>>> def over_2( m ): ... return m > 2.0 >>> filter( over_2, map( in_2_m, map( ftin_2_in, heights ) ) ) [2.0065999999999997]
1.We dened a lter function which returns True if the argument value is greater than 2.0. 2.We ltered our list of heights to locate any heights over 2.0 meters.
240
reduce(function, sequence, [initial=0] ) The given function must accept two argument values. This function will apply that function to an internal accumulator and each item of a sequence, from left to right, so as to reduce the sequence to a single value. The internal accumulator is initialized with the given initial value (or 0 if no value is provided.) This behaves as if it had the following denition.
def reduce( aFunction, aSequence, init= 0 ): r= init for s in aSequence: r= aFunction( r, s ) return r
Heres an example.
>>> def plus( a, b ): ... return a+b >>> reduce( plus, [1, 3, 5, 7, 9] ) 25
Note that Python has a number of built-in reductions: sum(), any() and all() are kinds of reduce functions. Costs and Benets. What are the advantages? First, the functional version can be clearer. Its a single line of code that summarizes the processing. Second, and more important, Python can execute the sequence processing functions far faster than the equivalent explicit loop. You can see that map() and filter() are equivalent to simple list comprehensions. This gives you two ways to specify these operations, both of which have approximately equivalent performance. This also means that map() and filter() arent essential to Python, but they are widely used. The reduce() function is a bit of a problem. It can have remarkably bad performance if it is misused. Consequently, there is some debate about the value of having this function. Another Example. Heres an interesting example that combines reduce() and map(). This uses two functions dened in earlier examples, add() and oddn().
def plus( a, b ): return a+b def oddn( n ): return 2*n+1 for i in range(10): sq=reduce( plus, map(oddn, range(i)), 0 ) print i, sq
Lets look at the evaluation of sq from innermost to outermost. 1. The range(i) generates a sequence of numbers from 0 to i -1. 2. A map() function applies oddn() to the sequence created by range(i), creating i odd values. The oddn() function returns the n th odd value. 3. A reduce() function applies plus() to the sequence of odd values, creating a sum. The zip() Function. The zip() function interleaves values from two or more sequences to create a new sequence. The new sequence is a sequence of tuples. Each item of a tuple is the corresponding values from from each sequence.
21.3. Sequence Processing Functions: map(), filter() and reduce()
241
zip(sequence, [sequence...] ) Interleave values from the various sequences to create tuples. If any sequence is too long, truncate it. Heres an example.
>>> zip( range(5), range(1,12,2) ) [(0, 1), (1, 3), (2, 5), (3, 7), (4, 9)]
In this example, we zipped two sequences together. The rst sequence was range(5), which has ve values. The second sequence was range(1,12,2) which has 6 odd numbers from 1 to 11. Since zip() truncates to the shorter list, we get ve tuples, each of which has the matching values from both lists. The map() function behaves a little like zip() when there is no function provided, just sequences. However, map() does not truncate, it lls the shorter list with None values.
>>> map( None, range(5), range(1,12,2) ) [(0, 1), (1, 3), (2, 5), (3, 7), (4, 9), (None, 11)]
21.4 Advanced List Sorting

Consider a list of tuples. We could get such a list when processing information that was extracted from a spreadsheet program. For example, if we had a spreadsheet with raw census data, we can easily transform it into a sequence of tuple s that look like the following.
jobData= [ (001,'Albany','NY',162692), (003,'Allegany','NY',11986), ... (121,'Wyoming','NY',8722), (123,'Yates','NY',5094) ]
We can also save the spreadsheet data in csv format and use the csv module to read it. Well return to the csv module in Components, Modules and Packages. List of Tuples from Spreadsheet Rows To create a list of tuples from a spreadsheet, you can do the following. In each row of the spreadsheet, put in a formula that creates a tuple from the various cells. This formula will have to include the necessary additional quotes. If youre using an Open Oice .ORG spreadsheet, it might look something like this
="(" & CONCATENATE( """" & A1 & ""","; """" & B1 & ""","; """" & C1 & """," ) & ")"
Once we have each row as a tuple, we can put some [] in front of the rst tuple and after the last tuple to make a proper list display. We can also slap an Python assignment statement onto this list of rows and turn our spreadsheet into a Python statement. We can copy and paste this data into our Python script. Sorting this list can be done trivially with the list.sort() method.
jobData.sort()
242
Recall that this updates the list in place. The sort() method specically does not return a result. A common mistake is to say something like: a= b.sort(). The sort method always returns None. This kind of sort will simply compare the tuple items in the order presented in the tuple. In this case, the county number is rst. What if we want to sort by some other column, like state name or jobs? Lets say we wanted to sort by state name, the third element in the tuple. We have two strategies for sorting when we dont want the simplistic comparison of elements in order. 1. We can provide a key extraction function to the sort() method. This will locate the key value (or a tuple of key values) within the given objects. 2. We can use the decorate - sort - undecorate pattern. What we do is decorate each element in the list, making it into a new kind of 2-tuple with the elds on which we want to sort as the rst element of this tuple and the original data as the second element of the tuple. This has the side eect of creating a second copy of the original list. Sorting With Key Extraction. The sort() method of a list can accept a keyword parameter, key, that provides a key extraction function. This function returns a value which can be used for comparison purposes. To sort our jobData by the third eld, we can use a function like the following.
def byState( a ): return a[2] jobData.sort( key=byState )
This byState() function returns the selected key value, which is then used by sort to order the tuples in the original list. If we want to sort by a multi-part key, we cna do something like the following.
def byStateJobs( a ): return ( a[2], a[3] )
This function will create a two-value tuple and use these two values for ordering the items in the list. Sorting With List Decoration. Supercially, this method appears more complex. However it is remarkably exible. This is a slightly more general solution than using a key extractor function. The idea is to transform the initial list of values into a new list of 2-tuples, with the rst item being the key and the second item being the original tuple. The rst item, only used for sorting, is a decoration placed in front of the original value. In this example, we decorate our values with a 2-tuple of state names and number of jobs. We can sort this temporary list of 2-tuples. Then we can strip o the decoration and recover the original values. Heres the example shown as three distinct steps.
deco= [ ((a[2],a[3]),a) for a in jobData ] deco.sort() state_jobs= [ v for k,v in deco ]
Heres an example shown with a single operation.

state_jobs = [ v for k,v in sorted( ((a[2],a[3]),a) for a in jobData ) ]
This works by evaluating the undecorate list comprehension, v for k,v in sorted(). That list comprehension depends on the output from the sorted() function. The sorted() function depends on the decorate list comprehension, (...,a) for a in jobData.
21.4. Advanced List Sorting
243
21.5 The Lambda

The functions map(), filter(), reduce(), and the list.sort() method all use small functions to control their operations. For example, we would write something like:
def byState( x ): return x[2] data.sort( key=byState )
In this case, we provided the byState() function to the list.sort() method. In many cases, this function is used only once, and it hardly seems necessary to dene a function object for a single use like this. Instead of dening a function, Python allows us to provide a lambda form. This is a kind of anonymous, one-use-only function body in places where we only need a very, very simple function. A lambda form is like a dened function: it has parameters and computes a value. The body of a lambda, however, can only be a single expression, limiting it to relatively simple operations. If it gets complex, youll have to dene a real function. Syntax. The syntax is relatively simple.
lambda :replaceable: parameter , ... : expression
There can be any number of parameters. The result of the expression is the value when the lambda is applied to arguments. No actual return statement is required. No statements can be used, just an expression. Note that a lambda form is not a statement; its an expression thats used within other expressions. The lambda form does not dene an object with a long life-time. The lambda form object generally exists just in the scope of a single statements execution. Generally, a lambda will look like this.
lambda a: a[0]*2+a[1]
This is a lambda which takes a tuple argument value and returns a value based on the rst two elements of the tuple. Examples. Heres an example of using a lambda form and applying it directly to its arguments.
>>> from math import pi >>> (lambda x: pi*x*x)(5) 78.539816339744831
1. The (lambda x: pi*x*x) is a function with no name; it accepts a single argument, x, and computes pi*x*x. 2. We apply the lambda to an argument value of 5. The value of applying the lambda to 5 is the value 52 = 25 . Heres a lambda form used in the map function.
>>> map( lambda x: pi*x*x, range(8) ) [0.0, 3.1415926535897931, 12.566370614359172, 28.274333882308138, 50.26548245743669, 78.539816339744831, 113.09733552923255, 153.93804002589985]
244
This map() function applies a lambda form to the values from 0 to 7 as created by the range(). The input sequence is mapped to the output sequence by having the lambda object applied to each value. Parameterizing a Lambda. Sometimes we want to have a lambda with an argument dened by the context or scope in which the lambda is being used. This is very similar to a closure, in which a free variable is bound into the lambda expression. Heres the basic form thats commonly used to create a closure in Python. This is a function that returns a lambda for later use.
>>> ... ... >>> >>> 10 >>> >>> 15 def timesX( x ): return lambda a: x*a t2= timesX(2) t2(5) t3= timesX(3) t3(5)
We can use this kind of thing as follows. We call our closure function to create a lambda thats got a constant bound into it. We can then use the resulting lambda. In this case, were using it in a map() function evaluation.
>>> map( timesX(3), range(5) ) [0, 3, 6, 9, 12]
Consider this more complex example.

>>> spins = [ (23,"red"), (21,"red"), (0,"green"), (24,"black") ] >>> def byColor( color ): ... return lambda t: color == t[1] ... >>> filter( byColor("red"), spins ) [(23, 'red'), (21, 'red')] >>> filter( byColor("green"), spins ) [(0, 'green')]
1. We have four sample spins of a roulette wheel, spins. 2. We have a function, byColor(), that creates a closure. This function binds the name of a color into a simple lambda, lambda t: color == t[1]. The resulting lambda can be used anywhere a function is required. 3. We use the byColor() function to create a lambda what we use for ltering our collection of spins. We create a lambda with byColor("red") that will return True for spins that have the color of "red". 4. We use the byColor() function to create another lambda what we use for ltering our collection of spins. We create a lambda with byColor("green") that will return True for spins that have the color of "green". As an alternative to creating lists with the filter() function, similar results can be created with a list comprehension.
>>> byRed= byColor("red") >>> [ s for s in spins if byRed(s) ] [(23, 'red'), (21, 'red')]
21.5. The Lambda
245
21.6 Multi-Dimensional Arrays or Matrices

There are situations that demand multi-dimensional arrays or matrices. In many languages (Java, COBOL, BASIC) this notion of multi-dimensionality is handled by pre-declaring the dimensions (and limiting the sizes of each dimension). In Python, these are handled somewhat more simply. If you have a need for more sophisticated processing than we show in this section, youll need to get the Python Numeric module, also known as NumPy. This is a Source Forge project, and can be found at http://numpy.sourceforge.net/. Lets look at a simple two-dimensional tabular summary. When rolling two dice, there are 36 possible outcomes. We can tabulate these in a two-dimensional table with one die in the rows and one die in the columns: 1 2 3 4 5 6 1 2 3 4 5 6 7 2 3 4 5 6 7 8 3 4 5 6 7 8 9 4 5 6 7 8 9 10 5 6 7 8 9 10 11 6 7 8 9 10 11 12
In Python, a multi-dimensional table like this can be implemented as a sequence of sequences. A table is a sequence of rows. Each row is a sequence of individual cells. This allows us to use mathematical-like notation. Where the mathematician might say Ai,j , in Python we can say A[i][j]. In Python, we want the row i from table A, and column j from that row. This is essentiall the like the list of tuples, yet again. See Lists of Tuples. List of Lists Example. We can build a table using a nested list comprehension. The following example creates a table as a sequence of sequences and then lls in each cell of the table.
table= [ [ 0 for i in range(6) ] for j in range(6) ] print table for d1 in range(6): for d2 in range(6): table[d1][d2]= d1+d2+2 print table
This program produced the following output.

[[0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0]] [[2, 3, 4, 5, 6, 7], [3, 4, 5, 6, 7, 8], [4, 5, 6, 7, 8, 9], [5, 6, 7, 8, 9, 10], [6, 7, 8, 9, 10, 11], [7, 8, 9, 10, 11, 12]]
1. First, we created a six by six table of zeroes, named table. Each item in the table is a 6-item list of zeroes. We used a list comprehension to create an object for each value of j in the range of 0 to 6. Each of the objects is a list of zeroes, one for each value of i in the range of 0 to 6. 2. We printed that list of lists. 3. We then lled this with each possible combination of two dice. We iterate over all combinations of two dice, lling in each cell of the table. This is done as two nested loops, one loop for each of the two dice. The outer enumerates all values of one die, d1. The loop enumerates all values of a second die, d2.
246
Updating each cell involves selecting the row with table[d1]; this is a list of 6 values. The specic cell in this list is selected by [d2]. We set this cell to the number rolled on the dice, d1+d2+2. Additional Examples. The printed list of list structure is a little hard to read. The following loop would display the table in a more readable form.
>>> ... ... [2, [3, [4, [5, [6, [7, for row in table: print row 3, 4, 5, 6, 7, 8, 4, 5, 6, 7, 8, 9, 5, 6, 7] 6, 7, 8] 7, 8, 9] 8, 9, 10] 9, 10, 11] 10, 11, 12]
As an exercise, well leave it to the reader to add some features to this to print column and row headings along with the contents. As a hint, the "%2d" % value string operation might be useful to get xed-size numeric conversions. Explicit Index Values. Lets summarize our matrix of die rolls, and accumulate a frequency table. Well use a simple list with 13 buckets (numbered from 0 to 12) to accumulate the frequency of each die roll.
fq= 13*[0] for i in range(6): for j in range(6): c= table[i][j] fq[ c ] += 1
1. We initialize the frequency table, fq, to be a list of 13 zeroes. 2. The outer loop sets the variable i to the values from 0 to 5. 3. The inner loop sets the variable j to the values from 0 to 5. 4. We use the index value of i to select a row from the table, and the index value of j to select a column from that row. This is the value, c . We then accumulate the frequency occurances in the frequency table, fq . This looks very mathematical and formal. However, Python gives us an alternative, which can be somewhat simpler. Using List Iterators Instead of Index Values. Since our table is a list of lists, we can make use of the power of the for statement to step through the elements without using an index.
fq= 13*[0] print fq for row in table: for c in row: fq[c] += 1
1. We initialize the frequency table, fq, to be a list of 13 zeroes. 2. The outer loop sets the variable row to each element of the original table variable. This decomposes the table into individual rows, each of which is a 6-element list. 3. The inner loop sets the variable c to each columns value within the row. This decomposes the row into the individual values.
21.6. Multi-Dimensional Arrays or Matrices
247
4. We count the actual occurances of each value, c by using the value as an index into the frequency table, fq . The increment the frequency value by 1. Mathematical Matrices. We use the explicit index technique for managing the mathematically-dened matrix operations. Matrix operations are done more clearly with this style of explicit index operations. Well show matrix addition as an example, here, and leave matrix multiplication as an exercise in a later section.
m1 = [ [1, 2, 3, 0], [4, 5, 6, 0], [7, 8, 9, 0] ] m2 = [ [2, 4, 6, 0], [1, 3, 5, 0], [0, -1, -2, 0] ] m3= [ 4*[0] for i in range(3) ] for i in range(3): for j in range(4): m3[i][j]= m1[i][j]+m2[i][j]
1. In this example we created two input matrices, m1 and m2, each three by four. 2. We initialized a third matrix, m3, to three rows of four zeroes, using a comprehension. 3. We iterated through all rows (using the i variable), and all columns (using the j variable) and computed the sum of m1 and m2. Python provides a number of modules for handling this kind of processing. In Components, Modules and Packages well look at modules for more sophisticated matrix handling.
21.7 Exercises
1. All Dice Combinations. Write a list comprehension that uses nested for-clauses to create a single list with all 36 dierent dice combinations from (1,1) to (6,6). 2. Temperature Table. Write a list comprehension that creates a list of tuple s. Each tuple has two values, a temperature in Farenheit and a temperature in Celsius. Create one list for Farenheit values from 0 to 100 in steps of 5 and the matching Celsius values. Create another list for Celsius values from -10 to 50 in steps of 2 and the matching Farenheit values. 3. Dene max() and min(). Use reduce() to create versions of the built-ins max() and min(). You may nd this diicult to do this with a simple lambda form. However, consider the following. We can pick a value from a tuple like this: (a,b)[0] == a, and (a,b)[1] == b. What are the values of (a,b)[a<b] and (a,b)[a>b]? 4. Compute the Average or Mean. A number of standard descriptive statistics can be built with reduce(). The basic formulae are given in Tuples. Write a function based on reduce() which computes the mean. Mean is a simple add-reduction of the values in a sequence divided by the length. In essence, youre inventing a version of sum() based on reduce(). 5. Compute the Variance and Standard Deviation. A number of standard descriptive statistics can be built with reduce(). The basic formulae are given in Tuples. Given a sequence of values A = {a0 , a1 , ..., an }, the standard deviation has a number of alternative denitions. One approach is to sum the values and square this number, as well as sum the squares of
248
each number. Summing squares can be done as a map() to compute squares and then use the sum() function. s1 (A2 i)
0i<n
s2 (
Ai )2
0i<n
s1 s2 v n A v Also the standard deviation can be dened as the square root of the average variance. 0i<n Ai m n 2 0i<n (Ai m) v n1 A v 6. Distinct Values In A Sequence. In List Exercises, one of the exercises looked at accumulating the distinct values in a sequence. Given an input sequence, seq, we can easily sort this sequence. This will put all equal-valued elements together. The comparison for unique values is now done between adjacent values, instead of a lookup in the resulting sequence.
Unique Values of a Sequence, seq , using sort()

Initalize Set result [ ] an empty sequence. Sort the input sequence, seq. Loop. For each value, v, in the sorted seq. Already in result? Is v the last element in result? If so, ignore it. If not, append v to the sequence result. Result. Return array result, which has distinct values from seq. While this is appealing in its simplicity, the sort operation makes it relatively ineicient. 7. Compute the Median. The median function arranges the values in sorted order. It locates either the mid-most value (if there are an odd number) or it averages two adjacent values (if there are an even number). If len(data) % 2 == 1, there is an odd number of values, and (len(data)+1)/2 is the midmost value. Otherwise there is an even number of values, and the len(data)/2 and len(data)/2-1 are the two mid-most values which must be averaged. 8. Portfolio Reporting. In Tuple Exercises, one of the exercises presented a stock portfolio as a sequence of tuples. Plus, we wrote two simple functions to evaluate purchase price and total gain or loss for this portfolio. Develop a function (or a lambda form) to sort this porfolio into ascending order by current value (current price number of shares). This function (or lambda) will require comparing the products of two elds instead of simply comparing two elds. 21.7. Exercises 249
9. Matrix Formatting. Given a 6 6 matrix of dice rolls, produce a nicely formatted result. Each cell should be printed with a format like "| %2s" so that vertical lines separate the columns. Each row should end with an |. The top and bottom should have rows of "----" printed to make a complete table. 10. Three Dimensions. If the rolls of two dice can be expressed in a two-dimensional table, then the rolls of three dice can be expressed in a three-dimensional table. Develop a three dimensional table, 6 x 6 x 6, that has all 216 dierent rolls of three dice. Write a loop that extracts the dierent values and summarizes them in a frequency table. The range of values will be from 3 to 18.
250
CHAPTER
TWENTYTWO
ADVANCED MAPPING TECHNIQUES

This chapter presents two advanced map concepts. In Default Dictionaries we show how to use the collections.defaultdict class. In Inverting a Dictionary we discuss inverting a dictionary to process it by value instead of by key.
22.1 Default Dictionaries

Python has a module, called collections, that includes a number of more advanced collection classes. Well look at one particularly useful one, defaultdict. The point of a defaultdict is to handle the dict.get() method in a dierent way. When the builtin dict class is confronted with a missing key, it raises a KeyError exception. When the collections.defaultdict class is confronted with a missing key, it will use a supplied function to create a default entry in the dictionary. Consider this snippet.
frequency = {} for i in range(100): d1, d2 = random.randint(1,6), random.randint(1,6) frequency[d1+d2] += 1
This is elegantly simply, but all it does is crash in a KeyError exception. We can add code, using dict.setdefault() to handle missing keys. However, defaultdict allows us to write this simple thing and have it work as expected. Importing defaultdict. We can make the defaultdict container available via the following:
from collections import defaultdict
Well look at the import statement in detail in Components, Modules and Packages. Creating a defaultdict. We create a new defaultdict by providing a function which will create a default value for us. Most applications of defaultdict will be used for counting or accumulating lists, so the most common initializations are the following two.
from collections import defaultdict frequency_table = defaultdict( int ) search_index = defaultdict( list )
251
frequency_table We create a default dict using the built-in int() function to create default values. Calling int() returns a zero, which is a useful default for frequency tables. This dict allows you to say frequency_table[someKey] += 1 without worrying about KeyError exceptions. If the key value is in the dictionary, then that value is incremented. If the key value is not in the dictionary, the initialization function (int) will evaluated to create the default value. search_index We create a default dict using the built-in list() function to create default values. Calling list() returns a an empty list, which is a useful default for a non-unique search index. This dict allows you to say search_index[someKey].append( someValue ) without worrying about KeyError exceptions. If the key value is in the dictionary, then list value is append to. If the key value is not in the dictionary, the initialization function (list) will evaluated to create the default value, an empty list. This can then be appended to. Example. The following example accumulates a frequency table of 1000 dice rolls.
from collections import defaultdict from random import randint frequency = defaultdict( int ) for i in range(100): d1, d2 = randint(1,6), randint(1,6) frequency[d1+d2] += 1
22.2 Inverting a Dictionary

An inverted database has one copy of each distinct value. This single copy is then associated the list of record identiers that share this value. Rather than this: key 1 2 3 name KaDiMa Dekkan Swell year 1972 2000 1972
We would have something like this. name KaDiMa Dekkan Swell And this year 1972 2000 record list [1, 3] [2] record list [1] [2] [3]
This inversion technique often to mappings. We may want to search a dictionary for a particular value instead of a particular key. In some cases, we may have multiple searches over the same base collection of information. Lets look at the simplest case rst. We have a frequency table like the following:
252
Chapter 22. Advanced Mapping Techniques
frequency= {2: 2, 3: 1, 4: 4, 5: 10, 6: 12, 7: 22, 8: 19, 9: 11, 10: 12, 11: 4, 12: 3}
Wed like to invert this dictionary, and display the dictionary in ascending order by value. Ths issue is that the values may not be unique.
>>> byFreq = defaultdict(list) >>> for k in frequency: ... byFreq[frequency[k]].append( k ) ... >>> byFreq defaultdict(<type 'list'>, {1: [3], 2: [2], 3: [12], 4: [4, 11], 10: [5], 11: [9], 12: [6, 10], 19: [8], 22: [7]})
This technique creats a new dictionary with the original values as keys and a list of the original keys as values.
22.3 Exercises
1. Compute the Mode. The mode function nds the most common value in a data set. This can be done by computing the frequency with which each unique value occurs. Youll need to invert the dictionary so that you work with the values (the frequencies) instead of the keys. The simplest solution is to nd the maximum frequency, the associated key is the mode. However, there may be ties for rst. This may mean thay you have a bimodal or even multi-modal distribution. Because of this its best to sort and check the top two frequencies to be sure there isnt a tie. If there is a tie, there are several possible response. One is to dene the mode function to return a tuple with all the values tied for most frequency. Generally, there will only be a single frequent value, but in the event of ties, all top values will be in the tuple. Another choice is to raise an exception. 2. Stock Purchases by Year. In Dictionary Exercises, we looked at blocks of stock where each block was represented as a simple tuple.
purchases = [ ( 'GE', 100, '10-sep-2001', 48 ), ( 'CAT', 100, '1-apr-1999', 24 ), ( 'GE', 200, '1-jul-1999', 56 ) ]
Create a dictionary where the index is the year of purchase, and the value is a list of purchases. Parsing the date can be simplied to date[-4:] for now. Later in Components, Modules and Packages, well address date-time parsing. 3. Eiciency. What is the typical complexity of a sort algorithm? What is the complexity of a hash table? Compare and contrast sorting and using a dictionary to summarize data according to distinct values. In SQL, we specify a dictionary-like summary of data using a GROUP BY clause. Why does a database use a sort?
22.3. Exercises
253
254
Chapter 22. Advanced Mapping Techniques
Part IV
Data + Processing = Objects
255
Encapsulating Data and Processing into Class Denitions

In Language Basics, we examined the core statements in the Python language. In Data Structures, we examined the built-in data structures available to us as programmers. Using these data structures gave us some hands-on experience with a number of classes. After using this variety of built-in objects, we are better prepared to design our own objects. Classes introduces basics of class denitions and Advanced Class Denition introduces simple inheritance. We extend this discussion further to include several common design patterns that use polymorphism. In Some Design Patterns we cover a few common design patterns. Creating or Extending Data Types describes the mechanism for adding types to Python that behave like the built-in types. We will spend some time on Pythons exible notion of attribute in Attributes, Properties and Descriptors. It turns out that an attribute can be a simple instance variable or it can be a method function that manages an instance variable. Well look at Pythons decorators in Decorators; this is handy syntax for assuring that specic aspects of a family of classes are implemented consistently. Well look at the various ways to dene properties in objects.properties. Additionally, well look how we can manage more sophisticated object protocols in Managing Contexts: the with Statement . Data Types. Weve looked at most of these data types in Data Structures. This is a kind of road-map of some of the most important built-in features. None. A unique constant, handy as a placeholder when no other value is appropriate. A number of built-in functions return values of None to indicate that no useful work can be done. NotImplemented. A unique constant, returned by special methods to indicate that the method is not implemented. See Numeric Type Special Methods for more information. Numeric types have relatively simple values. These are immutable objects: they cannot have their values changed, but they can participate in numerous arithmetic and comparison operations like +, -, *, /, //, **. Boolean. (bool) A variety of values are treated as logically false: False, 0, None, "", (), [], {}, set(). All other values are logically True. Integer. (int) Typically 32-bit numbers with a range of -2,147,483,648 through 2,147,483,647. Long. (long) These are specially coded integers of arbitrary length. They grow as needed to accurately represent numeric results. Literals end with L. In Python 3, the integer and long types will be unied and the remaining distinctions removed. Float. (float) These are oating point, scientic notation numbers. They are represented using the platforms oating point notation, so ranges and precisions vary. Typically these are called double precision in other languages, and are often 64-bits long. Complex. (complex) These are a pair of oating point numbers of the form a + bj , where a is the real part and b is the imaginary part. j = 1. Sequence. Collections of objects identied by their order or position. Immutable sequences are created as needed and can be used but never changed. * String. (str) A sequence of individual ASCII characters. * Unicode. (unicode) A sequence of individual Unicode characters. In Python 3, String and Unicode will be unied into a single class named str that includes features of unicode.
257
* Tuple. (tuple) A sequence of a xed number of Python items. Literals look like ( expression , ... ) Mutable sequences can be created, appended-to, changed, and have elements deleted. * List. (list) A sequence Python items. Literals look like [ expression , ... ] Operations like append(), pop() and sort() can be used to change lists. Set and Frozenset. (set, frozenset) Collections of objects. The collection is neither ordered nor keyed. Each item stands for itself. A set is mutable; we can append, change and delete elements from a set. A frozenset is immutable. Mapping. Collections of objects identied by keys instead of order. Dictionary. (dict) A collection of objects which are indexed by other objects. It is like a sequence of key:value pairs, where keys can be found eiciently. Any Python object can be used as the value. Keys have a small restriction: mutable lists and other mappings cannot be used as keys. Literals look like { key : value , ... } File. (classname:le) Python supports several operations on les, most notably reading, writing and closing. Python also provides numerous modules for interacting with the operating systems management of les. Callable. When we create a function with the def statement, we create a callable object. We can also dene our own classes with a special method of __call__() to make a callable object that behaves like a function. Class. What well cover in this part. There are numerous additional data structures that are part of Pythons implementation internals; they are beyond the scope of this book. One of the most powerful and useful features of Python, is its ability to dene new classes. The next chapters will introduce the class and the basics of object-oriented programming.
258
CHAPTER
TWENTYTHREE
CLASSES
Object-oriented programming permits us to organize our programs around the interactions of objects. A class provides the denition of the structure and behavior of the objects; each object is an instance of a class. Consequently, a typical program is a number of class denitions and a nal main function. The main function creates the objects that will perform the job of the program. This chapter presents the basic techniques of dening classes. In Semantics we dene the semantics of objects and the classes which dene their attributes (instance variables) and behaviors. In Class Denition: the class Statement we show the syntax for creating class denitions; we cover the use of objects in Creating and Using Objects. Python has some system-dened names that classes can exploit to make them behave like built-in Python classes, a few of these are introduced in Special Method Names. We provide some examples in Some Examples. Perhaps the most important part of working with objects is how they collaborate to do useful work; we introduce this in Object Collaboration
23.1 Semantics
Object-oriented programming focuses software design and implementation around the denitions of and interactions between individual objects. An object is said to encapsulate a state of being and a set of behaviors; it is both data and processing. Each instance of a class has individual copies of attributes which are tightly coupled with the class-wide operations. We can understand objects by looking at four features, adapted from [Rumbaugh91]. Identity. Each object is unique and is distinguishable from all other objects. In the real world, two otherwise identical coee cups can be distiguished as distinct objects. For example, they occupy dierent locations on our desk. In the world of a computers memory, objects could be identied by their address, which would make them unique. Classication. This is sometimes called Encapsulation. Objects with the same attributes and behavior belong to a common class. Each individual object has unique attribute values. We saw this when we looked at the various collection classes. Two dierent list objects have the same general structure, and the same behavior. Both lists respond to append(), pop(), and all of the other methods of a list. However, each list object has a unique sequence of values. Inheritance. A class can inherit methods from a parent class, reusing common features. A superclass is more general, a subclass overrides superclass features, and is more specic. With the built-in Python classes, weve looked at the ways in which all immutable sequences are alike. Polymorphism. A general operation can have variant implementation methods, depending on the class of the object. We saw this when we noted that almost every class on Python has a + operation.
259
Between two oating-point numbers the + operation adds the numbers, between two lists, however, the + operation concatenates the lists. Pythons Implementation. A class is the Python-language denition of the features of individual objects: the names of the attributes and denitions of the operations. Python implements the general notion of attribute as a dictionary of instance variables for an object. Python implements the general idea of an operation through a collection of methods or method functions of an objects class. Note that all Python objects are instances of some class. This includes something as simple as None or True.
>>> type(None) <type 'NoneType'> >>> type(True) <type 'bool'>
Additionally, a class also constructs new object instances for us. Once weve dened the class, we can then use it as a kind of factory to create new objects. Class Denition. Python class denitions require us to provide a number of things. We must provide a distinct name to the class. We list the superclasses from which a subclass inherits features. In Python 2, classes should explicitly be dened as subclasses of object. In Python 3, this will be the default. We have multiple inheritance available in Python. This diers from the single-inheritance approach used by languages like Java. We provide method functions which dene the operations for the class. We dene the behavior of each object through its method functions. Note that the attributes of each object are created by an initialization method function (named __init__()) when the object is created. We can dene attributes as part of the class denition. If we do, these will be class-level attributes, shared by all instances of the class. Python provides the required mechanism for unique identity. You can use the id() function to interrogate the unique identier for each object. Technically, a class denition creates a new class object. This Python object contains the denitions of the method functions. Additionally, a class object can also own class-level variables; these are, in eect, attributes which are shared by each individual object of that class. We can use this class object to create class instance objects. Its the instances that do the real work of our programs. The class is simply a template or factory for creating the instance objects. Duck Typing. Note that our instance variables are not a formal part of the class denition. This diers from Java or C++ where the instance variables must be statically declared. Another consequence of Pythons dynamic nature is that polymorphism is based on simple matching of method names. This is distinct from languages like Java or C++ where polymorphism depends on inheritance and precise class (or interface) relationships. Pythons approach to polymorphism is sometimes called duck typing: if it quacks like a duck and walks like a duck it is a duck. If several objects have the common method names, they are eectively polymorphic with respect to those methods.
260
Chapter 23. Classes
Were All Adults. The best programming practice is to treat each object as if the internal implementation details where completely opaque. Often, a class will have public methods that are a well-dened and supported interface, plus it will have private methods that are implementation details which can be changed without notice. All other objects within an application should use only the methods and attributes that comprise the class interface. Some languages (like C++ or Java) have a formal distinction between interface and implementation. Python has a limited mechanism for making a distinction between the dened interface and the private implementation of a class. The Python philosophy is sometimes called Were All Adults: theres little need for the (childish) formality between interface and implementation. Programmers can (and should) be trusted to read the documentation for a class and use the methods appropriately. Python oers two simple technique sfor separating interface from implementation. We can use a leading _ on an instance variable or method function name to make it more-or-less private to the class. We can use properties or descriptors to create more sophisticated protocols for accessing instance variables. Well wait until Attributes, Properties and Descriptors to cover these more advanced techniques. An Objects Lifecycle. Each instance of every class has a lifecycle. The following is typical of most objects. 1. Denition. The class denition is read by the Python interpreter (or it is a builtin class). Class denitions are created by the class statement. Examples of built-in classes include file, str, int, etc. 2. Construction. An instance of the class is constructed: Python allocates the namespace for the objects instance variables and associating the object with the class denition. The __init__() method is executed to initialize any attributes of the newly created instance. 3. Access and Manipulation. The instances methods are called (similar to function calls we covered in Functions), by client objects or the main script. There is a considerable amount of collaboration among objects in most programs. Methods that report on the state of the object are sometimes called accessors; methods that change the state of the object are sometimes called mutators or manipulators. 4. Garbage Collection. Eventually, there are no more references to this instance. Typically, the variable with an object reference was part of the body of a function that nished, the namespace is dropped, and the variables no longer exist. Python detects this, and removes the referenced object from memory, freeing up the storage for subsequent reuse. This freeing of memory is termed garbage collection, and happens automatically. See Garbage Collection for more information.
23.1. Semantics
261
Garbage Collection It is important to note that Python counts references to objects. When object is no longer in use, the reference count is zero, the object can be removed from memory. This is true for all objects, especially objects of built-in classes like str. This frees us from the details of memory management as practiced by C++ programmers. When we do something like the following:
s= "123" s= s+"456"
The following happens. 1. Python creates the string "123" and puts a reference to this string into the variable s. 2. Python creates the string "456". 3. Python performs the string concatenation method between the string referenced by s and the string "456", creating a new string "123456". 4. Python assigns the reference to this new "123456" string into the variable s. 5. At this point, strings "123" and "456" are no longer referenced by any variables. These objects will be destroyed as part of garbage collection.
23.2 Class Denition: the class Statement

We create a class denition with a class statement. We provide the class name, the parent classes, and the method function denitions.
class name ( parent ) : suite
The name is the name of the class, and this name is used to create new objects that are instances of the class. Traditionally, class names are capitalized and class elements (variables and methods) are not capitalized. The parent is the name of the parent class, from which this class can inherit attributes and operations. For simple classes, we dene the parent as object. Failing to list object as a parent class is not strictly speaking a problem; using object as the superclass does make a few of the built-in functions a little easier to use. Important: Python 3.0 In Python 3.0, using object as a parent class will no longer be necessary. In Python 2.6, however, it is highly recommended. The suite is a series of function denitions, which dene the class. All of these function denitions must have a rst positional argument, self, which Python uses to identify each objects unique attribute values. The suite can also contain assignment statements which create instance variables and provide default values. The suite typically begins with a comment string (often a triple-quoted string) that provides basic documentation on the class. This string becomes a special attribute, called __doc__. It is available via the help() function. For example:
import random class Die(object): """Simulate a 6-sided die.""" def roll( self ): self.value= random.randint(1,6)
262
Chapter 23. Classes
return self.value def getValue( self ): return self.value
1. We imported the random module to provide the random number generator. 2. We dened the simple class named Die, and claimed it as a subclass of object. The indented suite contains three elements. The docstring, which provides a simple denition of the real-world thing that this class represents. As with functions, the docstring is retrieved with the help() function. We dened a method function named roll(). This method function has the mandatory positional parameter, self, which is used to qualifty the instance variables. The self variable is a namespace for all of the attributes and methods of this object. We dened a method function named getValue(). This function will return the last value rolled. When the roll() method of a Die object is executed, it sets that objects instance variable, self.value, to a random value. Since the variable name, value, is qualied by the instance variable, self, the variable is local to the specic instance of the object. If we omitted the self qualier, Python would create a variable in the local namespace. The local namespace ceases to exist at the end of the method function execution, removing the local variables.
23.3 Creating and Using Objects

Once we have a class denition, we can make objects which are instances of that class. We do this by evaluating the class as if it were a function: for example, Die(). When we make one of these class calls, two things will happpen. A new object is created. This object has a reference to its class denition. The objects initializer method, __init__(), is called. Well look at how you dene this method function in th next section. Lets create two instances of our Die class.
>>> d1= Die() >>> d2= Die() >>> d1.roll(), d2.roll() (6, 5) >>> d1.getValue(), d2.getValue() (6, 5) >>> d1, d2 (<__main__.Die object at 0x607bb0>, <__main__.Die object at 0x607b10>) >>> d1.roll(), d2.roll() (1, 3) >>> d1.value, d2.value (1, 3)
1. We use the Die class object to create two variables, d1, and d2; both are new objects, instances of Die. 2. We evaluate the roll() method of d1; we also evaluate the roll() method of d2. Each of these calls sets the objects value variable to a unique, random number. Theres a pretty good chance (1 in 6) that both values might happen to be the same. If they are, simply call d1.roll() and d2.roll() again to get new values.
23.3. Creating and Using Objects
263
3. We evaluate the getValue() method of each object. The results arent too surprising, since the value attribute was set by the roll() method. This attribute will be changed the next time we call the roll() method. 4. We also ask for a representation of each object. If we provide a method named __str__() in our class, that method is used; otherwise Python shows the memory address associated with the object. All we can see is that the numbers are dierent, indicating that these instances are distinct objects.
23.4 Special Method Names

There are several special methods that are essential to the implementation of a class. Each of them has a name that begins and ends with double underscores. These method names are used implicitly by Python. Section 3.3 of the Python Language Reference provides the complete list of these special method names. Well look at the special method names in depth in Creating or Extending Data Types. Until then, well look at a few special method names that are used heavily. __init__(self, ...) The __init__() method of a class is called by Python to initialize a newly-created object. Note that The __init__() method can accept parameters, but does not return anything. It sets the internal state of the object. __str__(self ) The __str__() method of a class is called whenever Python needs a string representation of an object. This is the method used by the str() built-in function. When printing an object, the str() is called implicitly to get the value that is printed. __repr__(self ) The __repr__() method of a class is used when we want to see the details of an objects values. This method is used by the repr() function. Initializing an Object with __init__(). When you create an object, Python will both create the object and also call the objects __init__() method. This method function can create the objects instance variables and perform any other one-time initialization. There are, typically, two kinds of instance variables that are created by the __init__() method: variables based on parameters and variables that are independent of any parameters. Heres an example of a company description that might be suitable for evaluating stock performance. In this example, all of the instance variables (self.name, self.symbol, self.price) are based on parameters to the __init__() method.
class Company( object ): def __init__( self, name, symbol, stockPrice ): self.name= name self.symbol= symbol self.price= stockPrice def valueOf( self, shares ): return shares * self.price
When we create an instance of Company, we use code like this.

c1= Company( "General Electric", "GE", 30.125 )
This will provide three values to the parameters of __init__().
264
Chapter 23. Classes
String value of an object with __str__(). The __str__() method function is called whenever an instance of a class needs to be converted to a string. Typically, this occus when we use the str() function on an object. Also, when we reference object in a print statement, the str() function is evaluated. Consider this denition of the class Card.
class Card( object ): def __init__( self, rank, suit ): self.rank= rank self.suit= suit self.points= rank def hard( self ): return self.points def soft( self ): return self.points
When we try to print an instance of the class, we get something like the following.
>>> c = Card( 3, "D" ) >>> c <__main__.Card object at 0x607fb0> >>> str(c) '<__main__.Card object at 0x607fb0>'
This is the default behavior for the __str__() method. We can, however, override this with a function that produces a more useful-looking result.
def __str__( self ): return "%2d%s" % (self.rank, self.suit)
Adding this method function converts the current value of the die to a string and returns this. Now we get something much more useful.
>>> d = Card( 4, "D" ) >>> d <__main__.Card object at 0x607ed0> >>> str(d) ' 4D' >>> print d 4D
Representation details with __repr__(). While the __str__() method produces a human-readable string, we sometimes want the nitty-gritty details. The __repr__() method function is evaluated whenever an instance of a class must have its detailed representation shown. This is usually done in response to evaluating the repr() function. Examples include the following:
>>> repr(c) '<__main__.Card object at 0x607fb0>'
If we would like to produce a more useful result, we can override the __repr__() function. The objective is to produce a piece of Python programming that would reconstruct the original object.
def __repr__( self ): return "Card(%d,%r )" % (self.rank,self.suit)
We use __repr__() to produce a clear denition of how to recreate the given object.
23.4. Special Method Names
265
>>> c = Card( 5, "D" ) >>> repr(c) "Card(5,'D')"
Special Attribute Names. In addition to the special method names, each object has a number of special attributes. These are documented in section 2.3.10 of the Python Library Reference. Well look at just a few, including __dict__, __class__ and __doc__. __dict__ The attribute variables of a class instance are kept in a special dictionary object named __dict__. As a consequence, when you say self.attribute= value, this has almost identical meaning to self.__dict__['attribute']= value. Combined with the % string formatting operation, this feature is handy for writing __str__() and __repr__() functions.
def __str__( self ): return "%(rank)2s%(suit)s" % self.__dict__ def __repr__( self ): return "Card(%(rank)r ,%(suit)r )" % self.__dict__
__class__ This is the class to which the object belongs. __doc__ The docstring from the class denition.
23.5 Some Examples

Well look at two examples of class denitions. In the both examples, well write a script which denes a class and then uses the class.
die.py
#!/usr/bin/env python """Define a Die and simulate rolling it a dozen times.""" import random class Die(object): """Simulate a generic die.""" def __init__( self ): self.sides= 6 self.roll() def roll( self ): """roll() -> number Updates the die with a random roll.""" self.value= 1+random.randrange(self.sides) return self.value def getValue( self ): """getValue() -> number Return the last value set by roll().""" retur self.value def main(): d1, d2 = Die(), Die() for n in range(12):
266
Chapter 23. Classes
print d1.roll(), d2.roll() main()
1. This version of the Die class contains a doc string and three methods: __init__(), roll() and getValue(). 2. The __init__() method, called a constructor, is called automatically when the object is created. We provide a body that sets two instance variables of a Die object. It sets the number of sides, sides to 6 and it then rolls the die a rst time to set a value. 3. The roll() method, called a manipulator, generates a random number, updating the value instance variable. 4. The getValue() method, called a getter or an accessor, returns the value of the value instance variable, value. Why write this kind of function? Why not simply use the instance variable? Well address this in the FAQs at the end of this chapter.< 5. The main() function is outside the Die class, and makes use of the class denition. This function creates two Die, d1 and d2, and then rolls those two Die a dozen times. 6. This is the top-level script in this le. It executes the main() function, which in turn then creates Die objects. The __init__() method can accept arguments. This allows us to correctly initialize an object while creating it. For example:
point.py
#!/usr/bin/env python """Define a geometric point and a few common manipulations.""" class Point( object ): """A 2-D geometric point.""" def __init__( self, x, y ): """Create a point at (x,y).""" self.x, self.y = x, y def offset( self, xo, yo ): """Offset the point by xo parallel to the x-axis and yo parallel to the y-axis.""" self.x += xo self.y += yo def offset2( self, val ): """Offset the point by val parallel to both axes.""" self.offset( val, val ) def __str__( self ): """Return a pleasant representation.""" return "(%g,%g)" % ( self.x, self.y )
1. This class, Point, initializes each point object with the x and y coordinate values of the point. It also provides a few member functions to manipulate the point. 2. The __init__() method requires two argument values. A client program would use Point( 640, 480 ) to create a Point object and provide arguments to the __init__() method function. 3. The offset() method requires two argument values. This is a manipulator which changes the state of the point. It moves the point to a new location based on the oset values.
23.5. Some Examples
267
4. The offset2() method requires one argument value. This method makes use of the offset() method. This kind of reuse assures that both methods are perfectly consistent. 5. Weve added a __str__() method, which returns the string representation of the object. When we print any object, the print statement (or print() function) automatically calls the str() built-in function. The str() function uses the __str__() method of an object to get a string representation of the object.
def main(): obj1_corner = Point( 12, 24 ) obj2_corner = Point( 8, 16 ) obj1_corner.offset( -4, -8 ) print obj1_corner print obj2_corner main()
1. We construct a Point, named obj1_corner. 2. We manipulate the obj1_corner Point to move it a few pixels left and up. 3. We access the obj1_corner object by printing it. This will call the str() function, which will use the __str__() method to get a string representation of the Point. The self Variable. These examples should drive home the ubiquirty of the self variable. Within a class, we must be sure to use self. in front of the method function names as well as attribute names. For example, our offset2() function accepts a single value and calls the objects offset() function using self.offset( val, val ). The self variable is so important, well highlight it. Important: The self variable In Python, the self qualier is simply required all the time. Programmers experienced in Java or C++ may object to seeing the explicit self. in front of all variable names and method function names. In Java and C++, there is a this. qualier which is assumed by the compiler. Sometimes this qualier is required to disambiguate names, other times the compiler can work out what you meant. Some programmers complain that self is too much typing, and use another variable name like my . This is unusual, generally described as a bad policy, but it is not unheard of. An object is a namespace; it contains the attributes. We can call the attributes instance variables to distinguish them from global variables and free variables. Instance Variables These are part of an objects namespace. Within the method functions of a class, these variables are qualied by self. Outside the method functions of the class, these variables are qualied by the objects name. In die.py , the main() function would refer to d1.value to get the value attribute of object d1. Global Variables Global variables are pare of a special global namespace. The global statement creates the variable name in the global namespace instead of the local namespace. See The global Statement for more information. While its easy to refer to global variables, its not as easy to create them. Free Variables Within a method function, a variable that is not qualied by self., nor marked by global is a free variable. Python checks the local namespace, then the global namespace for this variable. This ambiguity is, generally, not a good idea.
268
Chapter 23. Classes
23.6 Object Collaboration

Object-oriented programming helps us by encapsulating data and processing into a tidy class denition. This encapsulation assures us that our data is processed correctly. It also helps us understand what a program does by allowing us to ignore the details of an objects implementation. When we combine multiple objects into a collaboration, we exploit the power of ecapsulation. Well look at a simple example of creating a composite object, which has a number of detailed objects inside it. Dening Collaboration. Dening a collaboration means that we are creating a class which depends on one or more other classes. Heres a new class, Dice, which uses instances of our Die class. We can now work with a Dice collection, and not worry about the details of the individual Die objects.
dice.py - part 1
class Dice( object ): """Simulate a pair of dice.""" def __init__( self ): "Create the two Die objects." self.myDice = ( Die(), Die() ) def roll( self ): "Return a random roll of the dice." for d in self.myDice: d.roll() def getTotal( self ): "Return the total of two dice." t= 0 for d in self.myDice: t += d.getValue() return t def getTuple( self ): "Return a tuple of the dice values." return tuple( [d.getValue() for d in self.myDice] ) def hardways( self ): "Return True if this is a hardways roll." return self.myDice[0].getValue() == self.myDice[1].getValue()
1. Were building on the denition of a single Die, from the die.py example. We didnt repeat it here to save some space in the example. 2. This class, Dice, denes a pair of Die instances. 3. The __init__() method creates an instance variable, myDice, which has a tuple of two instances of the Die class. 4. The roll() method changes the overall state of a given Dice object by changing the two individual Die objects it contains. This manipulator uses a for loop to assign each of the internal Die objects to d. In the loop it calls the roll() method of the Die object, d. This technique is called delegation: a Dice object delegates the work to two individual Die objects. We dont know, or care, how each Die computes its next value. 5. The getTotal() method computes a sum of all of the Die objects. It uses a for loop to assign each of the internal Die objects to d. It then uses the getValue() method of d. This is the oicial interface method; by using it, we can remain blissfully unaware of how Die saves its state. 6. The getTuple() method returns the values of each Die object. It uses a list comprehension to create a list of the value instance variables of each Die object. The built-in function tuple() converts the 23.6. Object Collaboration 269
list into an immutable tuple. 7. The hardways() method examines the value of each Die objec to see if they are the same. If they are, the total was made the hard way. The getTotal() and getTuple() methods return basic attribute information about the state of the object. These kinds of methods are often called getters because their names start with get. Collaborating Objects. The following function exercises an instance this class to roll a Dice object a dozen times and print the results.
def test2(): x= Dice() for i in range(12): x.roll() print x.getTotal(), x.getTuple()
This function creates an instance of Dice, called x. It then enters a loop to perform a suite of statements 12 times. The suite of statements rst manipulates the Dice object using its roll() method. Then it accesses the Dice object using getTotal() and getTuple() method. Heres another function which uses a Dice object. This function rolls the dice 1000 times, and counts the number of hardways rolls as compared with the number of other rolls. The fraction of rolls which are hardways is ideally 1/6, 16.6%.
def test3(): x= Dice() hard= 0 soft= 0 for i in range(1000): x.roll() if x.hardways(): hard += 1 else: soft += 1 print hard/1000., soft/1000.
Independence. One point of object collaboration is to allow us to modify one class denition without breaking the entire program. As long as we make changes to Die that dont change the interface that Die uses, we can alter the implementation of Die all we want. Similarly, we can change the implementation of Dice, as long as the basic set of methods are still present, we are free to provide any alternative implementation we choose. We can, for example, rework the denition of Die condent that we wont disturb Dice or the functions that use Dice ( test2() and test3() ). Lets change the way it represents the value rolled on the die. Heres an alternate implemetation of Die. In this case, the private instance variable, value, will have a value in the range 0 value < 5. When getValue() adds 1, the value is in the usual range for a single die, 1 value < 6.
class Die(object): """Simulate a 6-sided die.""" def __init__( self ): self.roll() def roll( self ): self.value= random.randint(0,5) retuen self.value def getValue( self ): return 1+self.value
270
Chapter 23. Classes
Since this version of Die has the same interface as other versions of Die in this chapter, it is polymorphic with them. There could be performance dierences, depending on the performance of random.randint() and random.randrange() functions. Since random.randint() has a slightly simpler denition, it may process more quickly. Similarly, we can replace Die with the following alternative. Depending on the performance of choice(), this may be faster or slower than other versions of Die.
class Die(object): """Simulate a 6-sided die.""" def __init__( self ): self.domain= range(1,7) def roll( self ): self.value= random.choice(self.domain) return self.value def getValue( self ): return self.value
23.7 Class Denition Exercises

These exercises are considerably more sophisticated then the exercises in previous parts. Each of these sections describes a small project that requires you to create a number of distinct classes which must collaborate to produce a useful result. When we document a method function, we dont mention the self variable. This is required when you actually write the class denition. However, we dont show it in the documentation.
23.7.1 Stock Valuation

A Block of stock has a number of attributes, including a purchase price, purchase date, and number of shares. Commonly, methods are needed to compute the total spent to buy the stock, and the current value of the stock. A Position is the current ownership of a company reected by all of the blocks of stock. A Portfolio is a collection of Positions ; it has methods to compute the total value of all Blocks of stock. When we purchase stocks a little at a time, each Block has a dierent price. We want to compute the total value of the entire set of Block s, plus an average purchase price for the set of Block s. The StockBlock class. First, dene a StockBlock class which has the purchase date, price per share and number of shares. Here are the method functions this class should have. __init__(self, date, price, number_of_shares) Populate the individual elds of date, price and number of shares. This is information which is part of the Position, made up of individual blocks. Dont include the company name or ticker symbol. __str__(self ) Return a nicely formatted string that shows the date, price and shares. getPurchValue(self ) Compute the value as purchase price per share shares. getSaleValue(self, salePrice) Use salePrice to compute the value as sale price per share shares.
23.7. Class Denition Exercises
271
getROI(self, salePrice) Use salePrice to compute the return on investment as (sale value - purchase value) purchase value. Note that this is not the annualized ROI. Well address this issue below. We can load a simple database with a piece of code the looks like the following. The rst statement will create a sequence with four blocks of stock. We chose variable name that would remind us that the ticker symbols for all four is GM. The second statement will create another sequence with four blocks.
blocksGM = [ StockBlock( StockBlock( StockBlock( StockBlock( ] blocksEK = [ StockBlock( StockBlock( StockBlock( StockBlock( ]
purchDate='25-Jan-2001', purchDate='25-Apr-2001', purchDate='25-Jul-2001', purchDate='25-Oct-2001',
purchPrice=44.89, purchPrice=46.12, purchPrice=52.79, purchPrice=37.73,
shares=17 shares=17 shares=15 shares=21
), ), ), ),
purchDate='25-Jan-2001', purchDate='25-Apr-2001', purchDate='25-Jul-2001', purchDate='25-Oct-2001',
purchPrice=35.86, purchPrice=37.66, purchPrice=38.57, purchPrice=27.61,
shares=22 shares=21 shares=20 shares=28
), ), ), ),
The Position class. A separate class, Position, will have an the name, symbol and a sequence of StockBlocks for a given company. Here are some of the method functions this class should have. Position.() __init__(self, name, symbol, * blocks) Accept the company name, ticker symbol and a collection of StockBlock instances. __str__(self ) Return a string that contains the symbol, the total number of shares in all blocks and the total purchse price for all blocks. getPurchValue(self ) Sum the purchase values for all of the StockBlocks in this Position. It delegates the hard part of the work to each StockBlocks getPurchValue() method. getSaleValue(self, salePrice) The getSaleValue() method requires a salePrice; it sums the sale values for all of the StockBlocks in this Position. It delegates the hard part of the work to each StockBlocks getSaleValue() method. getROI(self, salePrice) The getROI() method requires a salePrice; it computes the return on investment as (sale value purchase value) purchase value. This is an ROI based on an overall yield. Note that this is not the annualized ROI. Well address this issue below. We can create our Position objects with the following kind of initializer. This creates a sequence of three individual Position objects; one has a sequence of GM blocks, one has a sequence of EK blocks and the third has a single CAT block.
portfolio= [ Position( "General Motors", "GM", blocksGM ), Position( "Eastman Kodak", "EK", blocksEK ) Position( "Caterpillar", "CAT", [ StockBlock( purchDate='25-Oct-2001', purchPrice=42.84, shares=18 ) ] ) ]
272
Chapter 23. Classes
An Analysis Program. You can now write a main program that writes some simple reports on each Position object in the portfolio. One report should display the individual blocks purchased, and the purchase value of the block. This requires iterating through the Positions in the portfolio, and then delegating the detailed reporting to the individual StockBlocks within each Position. Another report should summarize each position with the symbol, the total number of shares and the total value of the stock purchased. The overall average price paid is the total value divided by the total number of shares. In addition to the collection of StockBlock objects that make up a Position, one additional piece of information that is useful is the current trading price for the Position. First, add a currentPrice attribute, and a method to set that attribute. Then, add a getCurrentValue() method which computes a sum of the getSaleValue() method of each StockBlock, using the trading price of the Position. Annualized Return on Investment. In order to compare portfolios, we might want to compute an annualized ROI. This is ROI as if the stock were held for eactly one year. In this case, since each block has dierent ownership period, the annualized ROI of each block has to be computed. Then we return an average of each annual ROI weighted by the sale value. The annualization requires computing the duration of stock ownership. This requires use of the time module. Well cover that in depth in Dates and Times: the time and datetime Modules. The essential feature, however, is to parse the date string to create a time object and then get the number of days between two time objects. Heres a code snippet that does most of what we want.
>>> import datetime >>> dt1="25-JAN-2001" >>> tm1= datetime.datetime.strptime( dt1, "%d-%b-%Y" ).date() >>> tm1 datetime.date(2001, 1, 25) >>> dt2= "25-JUN-2001" >>> tm2= datetime.datetime.strptime( dt2, "%d-%b-%Y" ).date() >>> tm2 datetime.date(2001, 6, 25) >>> tm2-tm1 datetime.timedelta(151) >>> (tm2-tm1).days/365.25 0.4134154688569473
In this example, tm1 and tm2 are datetime.date objects with details parsed from the date string by datetime.datetime.strptime(). We can subtract two datetime.date objects and get a datetime.timedelta that has the number of days between the two dates. A timedelta can be used on datetime.datetime objects to get days and seconds between two date-time stamps. In this case, there are 151 days between the two dates. When we divide by the number of days in a year (including leap days) we get the fraction of a year between the two dates. ownedFor(self, saleDate) This method computes the number of years the stock was owned. annualROI(self, salePrice, saleDate) This methods divides the gross ROI by the duration in years to return the annualized ROI. Once weve added the necessary support to StockBlock, we can then add to Position. annualROI(self, salePrice, saleDate) Given the StockBlock.annualROI() method, we can then compute a weighted average of each blocks ROI. This is the annualized ROI for the entire position.
273
23.7.2 Dive Logging and Surface Air Consumption Rate

The Surface Air Consumption Rate is used by SCUBA divers to predict air used at a particular depth. If we have a sequence of Dive objects with the details of each dive, we can do some simple calculations to get averages and ranges for our air consumption rate. For each dive, we convert our air consumption at that dives depth to a normalized air consumption at the surface. Given depth (in feet), d, tarting tank pressure (psi), s, nal tank pressure (psi), f, and time (in minutes) of t, the SACR, c, is given by the following formula. c= 33(s f ) t(d + 33)
Typically, you will average the SACR over a number of similar dives. The Dive Class. You will want to create a Dive class that contains attributes which include start pressure, nish pressure, time and depth. Typical values are a starting pressure of 3000, ending pressure of 700, depth of 30 to 80 feet and times of 30 minutes (at 80 feet) to 60 minutes (at 30 feet). SACRs are typically between 10 and 20. Your Dive class should have a method named Dive.getSACR() which returns the SACR for that dive. To make life a little simpler putting the data in, well treat time as string of HH:MM, and use string functions to pick this apart into hours and minutes. We can save this as tuple of two intgers: hours and minutes. To compute the duration of a dive, we need to normalize our times to minutes past midnight, by doing hh*60+mm. Once we have our times in minutes past midnight, we can easily subtract to get the number of minutes of duration for the dive. Youll want to create a method function Dive.getDuration() to do just this computation for each dive. __init__(self, pressure_start, pressure_nish, time_start, time_nish, depth) The __init__() method will initialize a Dive with the start and nish pressure in PSI, the in and out time as a string, and the depth as an integer. This method should parse both the time_start string and time_finish string and normalize each to be minutes after midnight so that it can compute the duration of the dive. Note that a practical dive log would have additional information like the date, the location, the air and water temperature, sea state, equipment used and other comments on the dive. __str__(self ) The __str__() method should return a nice string representation of the dive information. getSACR(self ) The getSACR() method can compute the SACR value from the starting pressure, nal pressure, time and depth information. The DiveLog Class. Well want to initialize our dive log as follows:
log = [ Dive( Dive( Dive( Dive( ]
start=3100, start=2700, start=2800, start=2800,
finish=1300, finish=1000, finish=1200, finish=1150,
in="11:52", in="11:16", in="11:26", in="11:54",
out="12:45", out="12:06", out="12:06", out="12:16",
depth=35 depth=40 depth=60 depth=95
), ), ), ),
Rather than use a simple sequence of Dive objects, you can create a DiveLog class which has a sequence of Dive objects plus a DiveLog.getAvgSACR() method. Your DiveLog method can be initiatlized with a sequence of dives, and can have an append method to put another dive into the sequence. Exercising the Dive and DiveLog Classes. Heres how the nal application could look. Note that were using an arbitrary number of argument values to the Dive.__init__() function, therefore, it has to be declared as def __init__( self, *listOfDives ) 274 Chapter 23. Classes
log= DiveLog( Dive( start=3100, finish=1300, Dive( start=2700, finish=1000, Dive( start=2800, finish=1200, Dive( start=2800, finish=1150, ) print log.getAvgSACR() for d in log.dives: print d
in="11:52", in="11:16", in="11:26", in="11:54",
out="12:45", out="12:06", out="12:06", out="12:16",
depth=35 depth=40 depth=60 depth=95
), ), ), ),
23.7.3 Multi-Dice
If we want to simulate multi-dice games like Yacht, Kismet, Yatzee, Zilch, Zork, Greed or Ten Thousand, well need a collection that holds more than two dice. The most common conguration is a ve-dice collection. In order to be exible, well need to dene a Dice object which will use a tuple, list or Set of individual Die instances. Since the number of dice in a game rarely varies, we can also use a FrozenSet. Once you have a Dice class which can hold a collection of dice, you can gather some statistics on various multi-dice games. These games fall into two types. In both cases, the players turn starts with rolling all the dice, the player can then elect to re-roll or preserve selected dice. Scorecard Games. In Yacht, Kismet and Yatzee, ve dice are used. The rst step in a players turn is a roll of all ve dice. This can be followed by up to two additional steps in which the player decides which dice to preserve and which dice to roll. The player is trying to make a scoring hand. A typical scorecard for these games lists a dozen or more hands with associated point values. Each turn must ll in one line of the scorecard; if the dice match a hand which has not been scored, the player enters a score. If a turn does not result in a hand that matches an unscored hand, then a score of zero is entered. The game ends when the scorecard is lled. A typical score card has spaces for 3-of-a-kinds (1 through 6) worth the sum of the scoring dice; a four-of-a-kind and full house (3 of a kind and a pair) worth the sum of the dice; a small straight (4 in a row) worth 25 points; a long straight (all 5 in a row) worth 30 points; a chance (sum of the dice), plus 5-of-a-kind worth 50 points. Some games award a 35 point bonus for getting all six 3-of-a-kind scores. Point Games. In Zilch, Zork, Green or Ten Thousand, ve dice are typical, but there are some variations. The player in this game has no limit on the number of steps in their turn. The rst step is to roll all the dice and determine a score. Their turn ends when they perceive the risk of another step to be too high, or theyve made a roll which gives them a score of zero (or zilch) for the turn. Typically, if the newly rolled dice are non-scoring, their turn is over with a score of zero. At each step, the player is looking at newly rolled dice which improve their score. The game ends when someone has a score of 10,000. A tyipcal set of rules awards a straight 1000. Three-of-a-kind scores 100 the dies value (except three ones is 1000 points). After removing any three-of-a-kinds, each die showing 1 scores 100, each die showing 5 scores 50. Additionally, some folks will score 1000 the dies value for ve-of-a-kind. Our MultiDice class will be based on the example of Dice in this chapter. In addition to a collection of Die instances (a sequence, Set or FrozenSet), the class will have the following methods.
275
__init__(self, dice) When initializing an instance of MultiDice, youll create a collection of individual Die instances. You can use a sequence of some kind, a Set or a FrozenSet. roll(self ) Roll all dice in the sequence or Set. Note that your choice of collection doesnt materially alter this method. Thats a cool feature of Python. getDice(self ) This method returns the collection of dice as a tuple so that a client class can examine them and potentialy re-roll some or all of the dice. reroll(self, * dice) Roll just the given dice. Remember that the MultiDice.getDice() returned the actual dice objects from our set. When the client program gives these objects back to us, we dont need to search through our sequence or set to locate the underlying objects. Weve been given the objects. score(self ) This method will score the hand, returning a list of two-tuples. Each two-tuple will have the name of the hand and the point value for the particular game. In some cases, there will be multiple ways to score a hand, and the list will reect all possible scorings of the hand, in order from most valuable to least valuable. In other cases, the list will only have a single element. It isnt practical to attempt to write a universal MultiDice class that covers all variations of dice games. Rather than write a gigantic does-everything class, the better policy is to create a family of classes that build on each other using inheritance. Well look at this in Inheritance. For this exercise, youll have to pick one game and compute the score for that particular game. Later, well see how to create an inheritance hierarchy that can cover all of these multi-dice games. For the scorecard games (Yacht, Kismet, Yatzee), we want to know if this set of dice matches any of the scorecard hands. In many cases, a set of dice can match a number of potential hands. A hand of all ve dice showing the same value (e.g, a 6) is matches the sixes, three of a kind, four of a kind, ve of a kind and wild-card rows on most game score-sheets. A sequence of ve dice will match both a long straight and a short straight. Common Scoring Methods. No matter which family of games you elect to pursue, youll need some common method functions to help score a hand. The following methods will help to evaluate a set of dice to see which hand it might be. matchDie(self, die) Give a Die, use matchValue() to partition the dice based on the value of the given Dies value. matchValue(self, number ) Given a numeric value, partition the dice into two sets: the dice which have a value that matches the given Die, and the remaining Die which do not match the value. Return both sets. NOfAKind(self, n) This functions will evaluate MultiDice.matchDie() for each Die in the collection. If any given Die has a matchDie() with a match set that contains n matching dice, the hand as a whole matches the template. This method can be used for 3 of a kind, 4 of a kind and 5 of a kind. This method returns the matching dice or None if the hand did not have N-of-a-kind. The matching dice set can then be summed (for the hands that count only scoring dice) or the entire set of dice can be summed (for the hands that count all dice.)
276
Chapter 23. Classes
largeStraight(self ) This function must establish that all ve dice form a sequence of values from 1 to 5 or 2 to 6. There must be no gaps and no duplicated values. smallStraight(self ) This function must establish that four of the ve dice form a sequence of values. There are a variety of ways of approaching this; it is actually a challenging algorithm. Heres one approach: create a sequence of dice, and sort them into order. Look for an ascending sequence with one irrelevant die in it. This irrelevant die must be either (a) a gap at the start of the sequence (1, 3, 4, 5, 6) or (b) a gap at the end of the sequence (1, 2, 3, 4, 6 ) or (c) a single duplicated value (1, 2, 2, 3, 4, 5) within the sequence. chance(self ) The sum of the dice values. It is a number between 5 and 30. Design Notes This isnt the best way to handle scoring. A better way is to use the Strategy design pattern and dene a separate class for scoring a game. The overall game can then use a simple, generic MultiDice instance. Each games scoring class, in turn, would use Strategy to delegate the rules for matching and scoring a hand to separate objects. We would dene a class for each kind of hand. Each various kinds of hand objects can interrogate the dice to determine if the dice matched its distinct pattern. Each hand object, besides checking for a match, can also encapsulate the score for the hand. This is something well look at in Strategy. Scoring Yacht, Kismet and Yatzee. For scoring these hands, your overall score() method function will step through the candidate hands in a specic order. Generally, youll want to check for fiveOfAKind() rst, since fourOfAKind() and threeOfAKind() will also be true for this hand. Similarly, youll have to check for largeStraight() before smallStraight(). Your score() method will evaluate each of the scoring methods. If the method matches, your method will append a two-tuple with the name and points to the list of scores. Scoring Zilch, Zork and 10,000. A scoring hands description can be relatively complex in these games. For example, you may have a hand with three 2s, a 1 and a 5. This is worth 350. The description has two parts: the three-of-a-kind and the extra 1s and 5s. Here are the steps for scoring this game. 1. Evaluate the largeStraight() method. If the hand matches, then return a list with an appropriate 2-tuple. 2. If youre building a game variation where ve of a kind is a scoring hand, then evaluate fiveOfAKind(). If the hand matches, then return a list with an appropriate 2-tuple. 3. Three of a kind. Evaluate the threeOfAKind() method. This will create the rst part of the hands description. If the matching set has exactly three dice, then the set of unmatched dice must be examined for additional 1s and 5s. The rst part of the hands description string is three-of-a-kind. If the matching has four or ve dice, then one or two dice must be popped from the matching set and added to the non-matching set. The set of unmatched dice must be examined for addtional 1s and 5s. The rst part of the hands description string is three-of-a-kind. If there was no set of three matching dice, then all the dice are in the non-matching set, which is checked for 1s and 5s. The string which describes the hand has no rst part, since there was no three-of-a-kind. 23.7. Class Denition Exercises 277
4. 1-5s. Any non-matching dice from the threeOfAKind() test are then checked using matchValue() to see if there are 1s or 5s. If there are any, this is the second part of the hands description. If there are none, then theres no second part of the description. 5. The nal step is to assemble the description. There are four cases: nothing, three-of-a-kind with no 1-5s, 1-5s with no three-of-a-kind, and three-of-a-kind plus 1-5s. In the nothing case, this is a non-scoring hand. In the other three cases, it is a scoring hand, and you can assign point values to each part of the description. Exercising The Dice. Your main script should create a MultiDice object, execute an initial roll and score the result. It should then pick three dice to re-roll and score the result. Finally, it should pick one die, re-roll this die and score the result. This doesnt make sophisticated strategic decisions, but it does exercise your MultiDice object thoroughly. When playing a scorecard game, the list of potential hands is examined to ll in another line on the scorecard. When playing a points game, each throw must result in a higher score than the previous throw or the turn is over. The Players Decision When playing these games, a human player will generally glance at the dice, form a pattern, and decide if the dice are close to one of the given hands. This is a challenging deision process to model. To create a proper odds-based judgement of possible outcomes, one would have to enumerate all possible games trees. Consider that there are 7,776 possible ways to roll the initial ve dice. From here one can reroll from 0 to all ve (7,776 outcomes) dice. For scorecard games, its possible to enumerate all possible game trees because there are only three total rolls. While there are a lot of dierent ways for the game to evolve, there are only a few scoring hands as the nal result. Each scoring hand has a value and a count of alternative trees that lead to that hand.
23.7.4 Rational Numbers

A Rational number is a ratio of two integers. Examples include 1/2, 2/3, 22/7, 355/113. We can do arithmetic operations on rational numbers. We can display them as proper fractions (3 1/7), improper fractions (22/7) or decimal expansions (3.1428571428571428). The essence of this class is to save a rational number and perform arithmetic operations on this number or between two rational numbers. Its also important to note that all arithmetic operations will create a new Rational number computed from the inputs. This makes our object mostly immutable, which is the expected behavior for numbers. Well start by dening methods to add and multiply two rational values. Later, well delve into the additional methods youd need to write to create a robust, complete implementation. Youll write __add__() and __mul__() methods that will perform their processing with two Rational values: self and other. In both cases, the variable other has to be another Rational number instance. You can check this with a simple assert statement. assert type(self) == type(other). Generally, however, this extra checking isnt essential. If you try to use a non-Rational number, youll get an AttributeError exception when you try to access the various parts of the Rational number. Design. A Rational class has two attributes: the numerator and the denominator of the value. These are both integers. Here are the various methods you should create.
278
Chapter 23. Classes
__init__(self, numerator, denominator=1) Accept the numerator and denominator values. It can have a default value for the denominator of 1. This gives us two constructors: Rational(2,3) and Rational(4). The rst creates the fraction 2/3. The second creates the fraction 4/1. This method should call the Rational._reduce() method to assure that the fraction is properly reduced. For example, Rational(8,4) should automatically reduce to a numerator of 2 and a denominator of 1. __str__(self ) Return a nice string representation for the rational number, typically as an improper fraction. This gives you the most direct view of your Rational number. You should provide a separate method to provide a proper fraction string with a whole number and a fraction. This other method would do additional processing to extract a whole name and remainder. __float__(self ) Return the oating-point value for the fraction. This method is called when a program does float( rational ). __add__(self, other ) Create and return a new Rational number that is the sum of self and other. Sum of S + O where S and O are Rational numbers,
Sn Sd
and
On Od .
Sn On Sn Od + On Sd + = Sd Od Sd Od Example: 3/5 + 7/11 = (33 + 35)/55 = 71/55. __mul__(self, other ) Create and returns a new Rational number that is the product of self and other. This new fraction that has a numerator of (self.numerator other.numerator), and a denominator of ( self.denominator other.denominator ). Product of S + O where S and O are Rational numbers,
Sn Sd
and
On Od .
Sn On Sn On = Sd Od Sd Od Example: 3/5 7/11 = 21/55. _reduce(self ) Reduce this Rational number by removing the greatest common divisor from the numerator and the denominator. This is called by Rational.__init__(), Rational.__add__(), Rational.__mul__(), assure that all fractions are reduced. Reduce is a two-step operation. First, nd the greatest common divisor between the numerator and denominator. Second, divide both by this divisor. For example 8/4 has a GCD of 4, and reduces to 2/1. The Greatest Common Divisor (GCD) algorithm is given in Greatest Common Divisor and Greatest Common Divisor . Note that weve given this method a name that begins with _ to make it private. Its a mutator and updates the object, something that may violate the expectation of immutability.
279
23.7.5 Playing Cards and Decks

Standard playing cards have a rank (ace, two through ten, jack, queen and king) and suit (clubs, diamonds, hearts, spades). These form a nifty Card object with two simple attributes. We can add a few generally useful functions. Here are the methods for your Card class. __init__(self, rank, suit ) Set the rank and suit of the card. The suits can be coded with a single character (C, D, H, S), and the ranks can be coded with a number from 1 to 13. The number 1 is an ace. The numbers 11, 12, 13 are Jack, Queen and King, respectively. __str__(self ) Return the rank and suit in the form 2C or AS or JD. A rank of 1 would become A, a rank of 11, 12 or 13 would become J, Q or K, respectively. __eq__(self, other ) Compare this card with other card considering both rank and suit. __ne__(self, other ) This can be implemented as not self == other. __lt__(self, other ) Compare this card with other, return True if this cards rank is less than the other. If the ranks are equal compare the suits. __le__(self, other ) This can be implemented as self < other or self == other. __gt__(self, other ) This can be implemented as not self <= other. __ge__(self, other ) This can be implemented as not self < other. Dealing and Decks. Card s are dealt from a Deck; a collection of Card s that includes some methods for shuing and dealing. Here are the methods that comprise a Deck. __init__(self ) Create all 52 cards. It can use two loops to iterate through the sequence of suits ("C", "D", "H", "S") and iterate through the ranks range(1,14) . After creating each Card, it can append each Card to a sequence like a list. deal(self ) This method needs to do two things. First it must shue the cards. The random module has a shuffle() function which does precisely this. Second, it should deal the shued cards. Dealing is best done with a generator method function. The deal() method function should have a simple for-loop that yields each individual Card; this can be used by a client application to generate hands. The presence of the yield statement will make this method function usable by a for statement in a client application script. Basic Testing. You should do some basic tests of your Card objects to be sure that they respond appropriately to comparison operations. For example,
>>> x1= Card(11,"C") >>> x1 JC >>> x2= Card(12,"D")
280
Chapter 23. Classes
>>> x1 < x2 True
You can write a simple test script which can the do the following to deal Cards from a Deck. In this example, the variable dealer will be the iterator object that the for statement uses internally.
d= Deck() dealer= d.deal() c1= dealer.next() c2= dealer.next()
Hands. Many card games involve collecting a hand of cards. A Hand is a collection of Card s plus some addition methods to score the hand in way thats appropriate to the given game. We have a number of collection classes that we can use: list, tuple, dictionary and set. In Blackjack, the Hand will have two Cards assigned initially; it will then be scored. After this, the player must choose among accepting another card (a hit), using this hand against the dealer (standing), doubling the bet and taking one more card, or splitting the hand into two hands. Ignoring the split option for now, its clear that the collection of Cards has to grow and then get scored again. What are the pros and cons of list, tuple, set and dictionary for a hand which grows? When considering Poker, we have to contend with the innumerable variations on poker; well look at simple ve-card draw poker. Games like seven-card stud require you to score potential hands given only two cards, and as many as 21 alternative ve-card hands made from seven cards. Texas Hold-Em has from three to ve common cards plus two private cards, making the scoring rather complex. For ve-card draw, the Hand will have ve cards assigned initially, and it will be scored. Then some cards can be removed and replaced, and the hand scored again. Since a valid poker hand is an ascending sequence of cards, called a straight, it is handy to sort the collection of cards. What are the pros and cons of list, tuple, set and dictionary?
23.7.6 Blackjack Hands

For Blackjack, well extend our Card class to score hands. When used in Blackjack, a Card has a point value in addition to a rank and suit. Aces are either 1 or 11; two through ten are worth 2-10; the face cards are all worth 10 points. When an ace is counted as 1 point, the total is called the hard total. When an ace is counted as 11 points, the total is called a soft total. You can add a point attribute to your card class. This can be set as part of __init__() processing. In that case, the following methods simple return the point value. As an alternative, you can compute the point value each time it is requested. This has the obvious disadvantage of being slower. However, it is considerably simpler to add methods to a class without revising the existing __init__() method. Here are the methods youll need to add to your Card class in order to handle Blackjack hands. getHardValue(self ) Returns the points. For most ranks, the points are the rank. For ranks of 11, 12 and 13 return a point value of 10. getSoftValue(self ) Returns the points. For most ranks, the points are the rank. For ranks of 11, 12 and 13 return a point value of 10. A rank of 1 returns a point value of 11. As a teaser for the next chapter, well note that these methods should be part of a Blackjack-specic subclass of the generic Card class. For now, however, well just update the Card class denition.When we look at inheritance in Inheritance, well see that a class hierarchy can be simpler than the if-statements in the getHardValue() and getSoftValue() methods.
281
Scoring Blackjack Hands. The objective of Blackjack is to accumulate a Hand with a total point value that is less than or equal to 21. Since an ace can count as 1 or 11, its clear that only one of the aces in a hand can have a value of 11, and any other aces must have a value of 1. Each Card produces a hard and soft point total. The Hand as a whole also has hard and soft point totals. Often, both hard and soft total are equal. When there is an ace, however, the hard and soft totals for the hand will be dierent. We have to look at two cases. No Aces. The hard and soft total of the hand will be the same; its the total of the hard value of each card. If the hard total is less than 21 the hand is in play. If it is equal to 21, it is a potential winner. If it is over 21, the hand has gone bust. Both totals will be computed as the hard value of all cards. One or more Aces. The hard and soft total of the hand are dierent. The hard total for the hand is the sum of the hard point values of all cards. The soft total for the hand is the soft value of one ace plus the hard total of the rest of the cards. If the hard or soft total is 21, the hand is a potential winner. If the hard total is less than 21 the hand is in play. If the hard total is over 21, the hand has gone bust. The Hand class has a collection of Cards, usually a sequence, but a Set will also work. Here are the methods of the Hand class. __init__(self, * cards) This method should be given two instances of Card to represent the initial deal. It should create a sequence or Set with these two initial cards. __str__(self ) Return a string with all of the individual cards. A construct like the following works out well: ",".join( map(str, self.cards ) ). This gets the string representation of each card in the self.cards collection, and then uses the strings join() method to assemble the nal display of cards. hardTotal(self ) Sums the hard value of each Card. softTotal(self ) Check for any Card with a dierent hard and soft point value (this will be an ace). The rst such card, if found, is the softSet. The remaining cards form the hardSet. Its entirely possible that the softSet will be empty. Its also entirely possible that there are multiple cards which could be part of the softSet. The value of this function is the total of the hard values for all of the cards in the hardSet plus the soft value of the card in the softSet. add(self, card ) This method will add another Card() to the Hand(). Exercising Card, Deck and Hand. Once you have the Card, Deck and Hand classes, you can exercise these with a simple function to play one hand of blackjack. This program will create a Deck and a Hand; it will deal two Card s into the Hand. While the Hand s total is soft 16 or less, it will add Cards. Finally, it will print the resulting Hand. There are two sets of rules for how to ll a Hand. The dealer is tightly constrained, but players are more free to make their own decisions. Note that the players hands which go bust are settled immediately, irrespective of what happens to the dealer. On the other hand, the players hands which total 21 arent resolved until the dealer nishes taking cards. The dealer must add cards to a hand with a soft 16 or less. If the dealer has a soft total between 17 and 21, they stop. If the dealer has a soft total which is over 21, but a hard total of 16 or less, they will take cards. If the dealer has a hard total of 17 or more, they will stop.
282
Chapter 23. Classes
A player may add cards freely until their hard total is 21 or more. Typically, a player will stop at a soft 21; other than that, almost anything is possible. Additional Plays. Weve avoided discussing the options to split a hand or double the bet. These are more advanced topics that dont have much bearing on the basics of dening Card, Deck and Hand. Splitting simply creates additional Hands. Doubling down changes the bet and gets just one additional card.
23.7.7 Poker Hands

Well extend the Card class we created in Playing Cards and Decks to score hands in Poker, where both the rank and suit are used to determine the hand that is held. Poker hands are ranked in the following order, from most desirable (and least likely) down to least desirable (and all too common). 1. Straight Flush. Five cards of adjacent ranks, all of the same suit. 2. Four of a Kind. Four cards of the same rank, plus another card. 3. Full House. Three cards of the same rank, plus two cards of the same rank. 4. Flush. Five cards of the same suit. 5. Straight. Five cards of adjacent ranks. In this case, Ace can be above King or below 2. 6. Three of a Kind. Three cards of the same rank, plus two cards of other ranks. 7. Two Pair. Two cards of one rank, plus two cards of another rank, plus one card of a third rank. 8. Pair. Two cards of one rank, plus three cards of other ranks. 9. High Card. The highest ranking card in the hand. Note that a straight ush is both a straight and a ush; four of a kind is also two pair as well as one pair; a full house is also two pair, as well as a one pair. It is important, then, to evaluate poker hands in decreasing order of importance in order to nd the best hand possible. In order to distinguish between two straights or two full-houses, it is important to also record the highest scoring card. A straight with a high card of a Queen, beats a straight with a high card of a 10. Similarly, a full house or two pair is described as queens over threes, meaning there are three queens and two threes comprising the hand. Well need a numeric ranking that includes the hands rank from 9 down to 1, plus the cards in order of importance to the scoring of the hand. The importance of a card depends on the hand. For a straight or straight ush, the most important card is the highest-ranking card. For a full house, the most important cards are the three-of-a kind cards, followed by the pair of cards. For two pair, however, the most important cards are the high-ranking pair, followed by the low-ranking pair. This allows us to compare two pair 10s and 4s against two pair 10s and 9s. Both hands have a pair of 10s, meaning we need to look at the third card in order of importance to determine the winner. Scoring Poker Hands. The Hand class should look like the following. This denition provides a number of methods to check for straight, ush and the patterns of matching cards. These functions are used by the score() method, shown below.
class PokerHand: def __init__( self, cards ): self.cards= cards self.rankCount= {} def straight( self ): all in sequence def straight( self ): all of one suit
283
def matches( self ): tuple with counts of each rank in the hand def sortByRank( self ): sort into rank order def sortByMatch( self ): sort into order by count of each rank, then rank
This function to score a hand checks each of the poker hand rules in descending order.
def score( self ): if self.straight() and self.flush(): self.sortByRank() return 9 elif self.matches() == ( 4, 1 ): self.sortByMatch() return 8 elif self.matches() == ( 3, 2 ): self.sortByMatch() return 7 elif self.flush(): self.sortByRank() return 6 elif self.straight(): self.sortByRank() return 5 elif self.matches() == ( 3, 1, 1 ): self.sortByMatch() return 4 elif self.matches() == ( 2, 2, 1 ): self.sortByMatchAndRank() return 3 elif self.matches() == ( 2, 1, 1, 1 ): self.sortByMatch() return 2 else: self.sortByRank() return 1
Youll need to add the following methods to the PokerHand class. straight(self ) True if the cards form a straight. This can be tackled easily by sorting the cards into descending order by rank and then checking to see if the ranks all dier by exactly one. flush(self ) True if all cards have the same suit. matches(self ) Returns a tuple of the counts of cards grouped by rank. This can be done iterating through each card, using the cards rank as a key to the self.rankCount dictionary; the value for that dictionary entry is the count of the number of times that rank has been seen. The values of the dictionary can be sorted, and form six distinct patterns, ve of which are shown above. The sixth is simply (1, 1, 1, 1, 1) , which means no two cards had the same rank. sortByRank(self ) Sorts the cards by rank. sortByMatch(self ) Uses the counts in the self.rankCount dictionary to update each card with its match count, and then
284
Chapter 23. Classes
sorts the cards by match count. sortByMatchAndRank(self ) Uses the counts in the self.rankCount dictionary to update each card with its match count, and then sorts the cards by match count and rank as two separate keys. Exercising Card, Deck and Hand. Once you have the Card, Deck and Hand classes, you can exercise these with a simple function to play one hand of poker. This program will create a Deck and a Hand; it will deal ve Cards into the Hand. It can score the hand. It can replace from zero to three cards and score the resulting hand.
285
286
Chapter 23. Classes
CHAPTER
TWENTYFOUR
ADVANCED CLASS DEFINITION

This section builds up some additional class denition techniques. We describe the basics of inheritance in Inheritance. We turn to a specic inheritance technique, polymorphism in Polymorphism. There are some class-related functions, which we describe in Built-in Functions. Well look at some specic class initializer technique in Initializer Techniques. We include a digression on design approaches in Design Approaches. In Class Variables we provide information on class-level variables, dierent from instance variables. We conclude this chapter with some style notes in Style Notes.
24.1 Inheritance
In Semantics we identied four important features of objects. Identity. Classication. Inheritance. Polymorphism. The point of inheritance is to allow us to create a subclass which inherits all of the features of a superclass. The subclass can add or replace method functions of the superclass. This is typically used by dening a general-purpose superclass and creating specialized subclasses that all inherit the general-purpose features but add special-purposes features of their own. We do this by specifying the parent class when we create a subclass.
class subclass ( superclass ) : suite
All of the methods of the superclass are, by denition, also part of the subclass. Often the suite of method functions will add to or override the denition of a parent method. If we omit providing a superclass, we create a classical class denition, where the Python type is instance; we have to do additional processing to determine the actual type. Generally, we should avoid this kind of class denition. It works, but isnt ideal. When we use object as the superclass, the Python type is reported more simply as the appropriate class object. As a general principle, every class denition should be a subclass of object, either directly or indirectly. Important: Python 3 In Python 3, this distinction will be removed. A class with no explicit superclass will still be a subclass of object. 287
Extending a Class. There are two trivial subclassing techniques. One denes a subclass which adds new methods to the superclass. The other overrides a superclass method. The overriding technique leads to two classes which are polymorphic because they have the same interface. Well return to polymorphism in Polymorphism. Heres a revised version of our basic Dice class and a subclass to create CrapsDice.
crapsdice.py
#!/usr/bin/env python """Define a Die, Dice and CrapsDice.""" class Die(object): """Simulate a 6-sided die.""" def __init__( self ): self.domain= range(1,7) def roll( self ): self.value= random.choice(self.domain) return self.value def getValue( self ): return self.value class Dice( object ): """Simulate a pair of dice.""" def __init__( self ): "Create the two Die objects." self.myDice = ( Die(), Die() ) def roll( self ): "Return a random roll of the dice." for d in self.myDice: d.roll() def getTotal( self ): "Return the total of two dice." return self.myDice[0].value + self.myDice[1].value def getTuple( self ): "Return a tuple of the dice." return self.myDice class CrapsDice( Dice ): """Extends Dice to add features specific to Craps.""" def hardways( self ): """Returns True if this was a hardways roll?""" return self.myDice[0].value == self.myDice[1].value def isPoint( self, value ): """Returns True if this roll has the given total""" return self.getTotal() == value
The CrapsDice class contains all the features of Dice as well as the additional features we added in the class declaration. We can, for example, evaluate the roll() and hardways() methods of CrapsDice. The roll() method is inherited from Dice, but the hardways() method is a direct part of CrapsDice. Adding Instance Variables. Adding new instance variables requires that we extend the __init__() method.
288
Chapter 24. Advanced Class Denition
In this case our subclass __init__() function must start out doing everything the superclass __init__() function does, and then creates a few more attributes. Python provides us the super() function to help us do this. We can use super() to distinguish between method functions with the same name dened in the superclass and extended in a subclass. super(type, variable) This will do two things: locate the superclass of the given type, and it then bind the given variable to create an object of the superclass. This is often used to call a superclass method from within a subclass: super( classname ,self).method() Heres a template that shows how a subclass __init__() method uses super() to evaluate the superclass __init__() method.
class Subclass( Superclass ): def __init__( self ): super(Subclass,self)__init__() # Subclass-specific stuff follows
This will bind our self variable to the parent class so that we can evaluate the parent class __init__() method. After that, we can add our subclass initialization. Well look at additional techniques for creating very exible __init__() methods in Initializer Techniques. Various Kinds of Cards. Lets look closely at the problem of cards in Blackjack. All cards have several general features: they have a rank and a suit. All cards have a point value. However, some cards use their rank for point value, other cards use 10 for their point value and the aces can be either 1 or 11, depending on the the rest of the cards in the hand. We looked at this in the Playing Cards and Decks exercise in Classes. We can model this very accurately by creating a Card class that encapsulates the generic features of rank, suit and point value. Our class will have instance variables for these attribites. The class will also have two functions to return the hard value and soft value of this card. In the case of ordinary non-face, non-ace cards, the point value is always the rank. We can use this Card class for the number cards, which are most common.
class Card( object ): """A standard playing card for Blackjack.""" def __init__( self, r, s ): self.rank, self.suit = r, s self.pval= r def __str__( self ): return "%2d%s" % ( self.rank, self.suit ) def getHardValue( self ): return self.pval def getSoftValue( self ): return self.pval
We can create a subclass of Card which is specialized to handle the face cards. This subclass simply overrides the value of self.pval, using 10 instead of the rank value. In this case we want a FaceCard.__init__() method that uses the parents Card.__init__() method, and then does additional processing. The existing denitions of getHardValue() and getSoftValue() method functions, however, work ne for this subclass. Since Card is a subclass of object, so is FaceCard. Additionally, wed like to report the card ranks using letters (J, Q, K) instead of numbers. We can override the __str__() method function to do this translation from rank to label.
class FaceCard( Card ): """A 10-point face card: J, Q, K."""
24.1. Inheritance
289
def __init__( self, r, s ): super(FaceCard,self).__init__( r, s ) self.pval= 10 def __str__( self ): label= ("J","Q","K")[self.rank-11] return "%2s%s" % ( label, self.suit )
We can also create a subclass of Card for Aces. This subclass inherits the parent class __init__() function, since the work done there is suitable for aces. The Ace class, however, provides a more complex algorithms for the getHardValue() and getSoftValue() method functions. The hard value is 1, the soft value is 11.
class Ace( Card ): """An Ace: either 1 or def __str__( self ): return "%2s%s" % ( def getHardValue( self return 1 def getSoftValue( self return 11
11 points.""" "A", self.suit ) ): ):
Deck and Shoe as Collections of Cards. In a casino, we can see cards handled in a number of dierent kinds of collections. Dealers will work with a single deck of 52 cards or a multi-deck container called a shoe. We can also see the dealer putting cards on the table for the various players hands, as well as a dealers hand. Each of these collections has some common features, but each also has unique features. Sometimes its diicult to reason about the various classes and discern the common features. In these cases, its easier to dene a few classes and then refactor the common features to create a superclass with elements that have been removed from the subclasses. Well do that with Decks and Shoes. We can dene a Deck as a sequence of Cards. The deck.__init__() method function of Deck creates appropriate Cards of each subclass. These are Card objects in the range 2 to 10, FaceCard obejcts with ranks of 11 to 13, and Ace objects with a rank of 1.
class Deck( object ): """A deck of cards.""" def __init__( self ): self.cards= [] for suit in ( "C", "D", "H", "S" ): self.cards+= [Card(r,suit) for r in range(2,11)] self.cards+= [TenCard(r,suit) for r in range(11,14)] self.cards+= [Ace(1,suit)] def deal( self ): for c in self.cards: yield c
In this example, we created a single instance variable self.cards within each Deck instance. For dealing cards, weve provided a generator function which yields the Card objects in a random order. Weve omitted the randomization from the deal() function; well return to it in the exercises. For each suit, we created the Cards of that suit in three steps. 1. We created the number cards with a list comprehension to generate all ranks in the range 2 through 10. 2. We created the face cards with a similar process, except we use the TenCard class constructor, since blackjack face cards all count as having ten points.
290
3. Finally, we created a one-item list of an Ace instance for the given suit. We can use Deck objects to create an multi-deck shoe. (A shoe is what dealers use in casinos to handle several decks of slippery playing cards.) The Shoe class will create six separate decks, and then merge all 312 cards into a single sequence.
class Shoe( object ): """Model a multi-deck shoe of cards.""" def __init__( self, decks=6 ): self.cards= [] for i in range(decks): d= Deck() self.cards += d.cards def deal( self ): for c in self.cards: yield c
For dealing cards, weve provided a generator function which yields the Cards in a random order. Weve omitted the randomization from the deal() function; well return to it in the exercises. Factoring Out Common Features. When we compare Deck and Shoe, we see two obviously common features: they both have a collection of Cards, called self.cards. Also, they both have a deal() method which yields a sequence of cards. We also see things which are dierent. The most obvious dierences are details of initializing self.cards. It turns out that the usual procedure for dealing from a shoe involves shuing all of the cards, but dealing from only four or ve of the six available decks. This is done by inserting a marker one or two decks in from the end of the shoe. In factoring out the common features, we have a number of strategies. One of our existing classes is already generic-enough to be the superclass. In the Card example, we used the generic Card class as superclass for other cards as well as the class used to implement the number cards. In this case we will make concrete object instances from the superclass. We may need to create a superclass out of our subclasses. Often, the superclass isnt useful by itself; only the subclasses are really suitable for making concrete object instances. In this case, the superclass is really just an abstraction, it isnt meant to be used by itself. Heres an abstract CardDealer from which we can subclass Deck and Shoe. Note that it does not create any cards. Each subclass must do that. Similarly, it cant deal properly because it doesnt have a proper shuffle() method dened.
class CardDealer( object ): def __init__( self ): self.cards= [] def deal( self ): for c in self.shuffle(): yield c def shuffle( self ): ...to be done in the exercises...
Python does not have a formal notation for abstract or concrete superclasses. When creating an abstract superclass it is common to return NotImplemented or raise NotImplementedError to indicate that a method must be overridden by a subclass. We can now rewrite Deck as subclasses of CardDealer.
24.1. Inheritance
291
class Deck( CardDealer ): def __init__( self ): super(Deck,self).__init__() for s in ("C","D","H","S"): for suit in ( "C", "D", "H", "S" ): self.cards+= [Card(r,suit) for r in range(2,11)] self.cards+= [TenCard(r,suit) for r in range(11,14)] self.cards+= [Ace(1,suit)]
We can also rewrite Shoe as subclasses of CardDealer.

class Shoe( CardDealer ): def __init__( self, decks=6 ): CardDealer.__init__( self ) for i in range(decks): d= Deck() self.cards += d.cards
The benet of this is to assure that Deck and Shoe actually share common features. This is not cut and paste sharing. This is by denition sharing. A change to CardDealer will change both Deck and Shoe, assuring complete consistency.
24.2 Polymorphism
In Semantics we identied four important features of objects. Identity. Classication. Inheritance. Polymorphism. Polymorphism exists when we dene a number of subclasses which have commonly named methods. A function can use objects of any of the polymorphic classes without being aware that the classes are distinct. In some languages, it is essential that the polymorphic classes have the same interface (or be subinterfaces of a common parent interface), or be subclasses of a common superclass. This is sometimes called strong, hierarchical typing, since the type rules are very rigid and follow the subclass/subinterface hierarchy. Python implements something that is less rigid, often called duck typing. The phrase follows from a quote attributed to James Whitcomb Riley: When I see a bird that walks like a duck and swims like a duck and quacks like a duck, I call that bird a duck. In short, two objects are eectively of the class Duck if they have a common collection of methods (walk, swim and quack, for example.) When we look at the examples for Card, FaceCard, Ace in Inheritance, we see that all three classes have the same method names, but have dierent implementations for some of these methods. These three classes are polymorphic. A client class like Hand can contain individual objects of any of the subclasses of Card. A function can evaluate these polymorphic methods without knowing which specic subclass is being invoked. In our example, both FaceCard and Ace were subclasses of Card. This subclass relationship isnt necesary for polymorphism to work correctly in Python. However, the subclass relationship is often an essential ingredient in an overall design that relies on polymorphic classes.
292
Whats the Benet? If we treat all of the various subclasses of Card in a uniform way, we eectively delegate any special-case processing into the relevant subclass. We concentrate the implementation of a special case in exactly one place. The alternative is to include if statements all over our program to enforce special-case processing rules. This diusing of special-case processing means that many components wind up with an implicit relationship. For example, all portions of a program that deal with Cards would need multiple if statements to separate the number card points, face card points and ace points. By making our design polymorphic, all of our subclasses of Card have ranks and suits, as well as hard and soft point values. We we can design the Deck and Shoe classes to deal cards in a uniform way. We can also design a Hand class to total points without knowing which specic class to which a Card object belongs. Similarly, we made our design for Deck and Shoe classes polymorphic. This allows us to model one-deck blackjack or multi-deck blackjack with no other changes to our application. The Hand of Cards. In order to completely model Blackjack, well need a class for keeping the player and dealers hands. There are some dierences between the two hands: the dealer, for example, only reveals their rst card, and the dealer cant split. There are, however, some important similarities. Every kind of Hand must determine the hard and soft point totals of the cards. The hard point total for a hand is simply the hard total of all the cards. The soft total, on the other hand, is not simply the soft total of all cards. Only the Ace cards have dierent soft totals, and only one Ace can meaningfully contribute its soft total of 11. Generally, all cards provide the same hard and soft point contributions. Of the cards where the hard and soft values dier, only one such card needs to be considered. Note that we are using the values of the getHardValue() and getSoftValue() methods. Since this test applies to all classes of cards, we preserve polymorphism by checking this property of every card. Well preserving just one of the cards with a soft value that is dierent from the hard value. At no time do use investigate the class of a Card to determine if the card is of the class Ace. Examining the class of each object needlessly constrains our algorithm. Using the polymorphic methods means that we can make changes to the class structure without breaking the processing of the Hand class. Important: Pretty Poor Polymorphism The most common indicator of poor use polymorphism is using the type(), isinstance() and issubclass() functions to determine the class of an object. These should used rarely, if at all. All processing should be focused on what is dierent about the objects, not the class to which an object belongs. We have a number of ways to represent the presence of a Card with a distinct hard and soft value. An attribute with the point dierence (usually 10). A collection of all Cards except for one Card with a point dierence, and a single attribute for the extra card. Well choose the rst implementation. We can use use a sequence to hold the cards. When cards are added to the hand, the rst card that returns distinct values for the hard value and soft value will be used to set a variable has keeps the hard vs. soft point dierence.
hand.py
class Hand( object ): """Model a player's hand.""" def __init__( self ): self.cards = [ ] self.softDiff= 0 def addCard( self, aCard ):
24.2. Polymorphism
293
self.cards.append( aCard ) if aCard.getHardValue() != aCard.getSoftValue(): if self.softDiff == 0: self.softDiff= aCard.getSoftValue()-aCard.getHardValue() def points( self ): """Compute the total points of cards held.""" p= 0 for c in self.cards: p += c.getHardValue() if p + self.softDiff <= 21: return p + self.softDiff else: return p
1. The __init__() special function creates the instance variable, self.cards, which we will use to accumulate the Card objects that comprise the hand. This also sets self.softDiff which is the dierence in points between hard and soft hands. Until we have an Ace, the dierence is zero. When we get an Ace, the dierence will be 10. 2. We provide an addCard() method that places an additional card into the hand. At this time, we examine the Card to see if the soft value is dierent from the hard value. If so, and we have not set the self.softDiff yet, we save this dierence. 3. The points() method evaluates the hand. It initializes the point count, p, to zero. We start a for-loop to assign each card object to c. We could, as an alternative, use a sum() function to do this. If the total with the self.softDiff is 21 or under, we have a soft hand, and these are the total points. If the total with the self.softDiff is over 21, we have a hard hand. The hard hand may total more than 21, in which case, the hand is bust.
24.3 Built-in Functions

There are two built in functions of some importance to object oriented programming. These are used to determine the class of an object, as well as the inheritance hierarchy among classes. isinstance(object, type) True if object is an instance of the given type or any of the subclasses of type. issubclass(parameter, base) True if class class is a subclass of class base. This question is usually moot, because most programs are designed to provide the expected classes of objects. There are some occasions for deep paranoia; when working with untrusted software, your classes may need to be sure that other programmers are following the rules. In Java and C++, the compiler can check these situations. In Python, the compiler doesnt check this, so we may want to include run-time checks. super(type) This will return the superclass of the given type. All of the basic factory functions (str, int, float, long, complex, unicode, tuple, list, dict, set) are eectively class names. You can, therefore, use a test like isinstance( myParam ,int) to conrm that the argument value provided to this parameter is an integer. An additional class, basestring is the parent class of both str and unicode.
294
The following example uses the isinstance() function to validate the type of argument values. First, well dene a Roulette wheel class, Wheel, and two subclasses, Wheel1 with a single zero and Wheel2 with zero and double zero.
wheel.py
import random class Wheel( object ): def value( self ): return NotImplemented class Wheel1( Wheel ): def value( self ): spin= random.randrange(37) return str(spin) class Wheel2( Wheel ): def __init__( self ): self.values= ['00'] + map( str, range(37) ) def value( self ): return random.randchoice( self.values )
1. The Wheel class denes the interface for Roulette wheels. The actual class denition does nothing except show what the expected method functions should be. We could call this an abstract denition of a Wheel. 2. The Wheel1 subclass uses a simple algorithm for creating the spin of a wheel. The value() method chooses a number between 0 and 36. It returns a string representation of the number. This has only a single zero. 3. The Wheel2 subclass creates an instance variable, values, to contain all possible results. This includes the 37 values from 0 to 36, plus an additional 00 value. The value() method chooses one of these possible results. The following function expects that its parameter, w, is one of the subclasses of Wheel.
def simulate( w ): if not isinstance( w, Wheel ): raise TypeError( "Must be a subclass of Wheel" ) for i in range(10): print w.value()
In this case, the simulate function checks its argument, w to be sure that it is a subclass of Wheel. If not, the function raises the built in TypeError. An alternative is to use an assertion for this.
def simulate( w ): assert isinstance( w, Wheel ) for i in range(10): print w.value()
24.3. Built-in Functions
295
24.4 Collaborating with max(), min() and sort()

The min() and max() functions can interact with our classes in relatively simple ways. Similarly, the sort() method of a list can also interact with our new class denitions. In all three cases, a keyword parameter of key can be used to control which attributes are used for determining minimum, maximum or sort order. The key parameter must be given a function, and that function is evaluated on each item that is being compared. Heres a quick example.
class Boat( object ): def __init__( self, name, loa ): self.name= name self.loa= loa def byName( aBoat ): return aBoat.name def byLOA( aBoat ): return aBoat.loa fleet = [ Boat("KaDiMa", 18 ), Boat( "Emomo", 21 ), Boat("Leprechaun", 30 ) ] first= min( fleet, key=byName ) print "Alphabetically First:", first longest= max( fleet, key=byLOA ) print "Longest:", longest
As min(), max() or sort traverse the sequence doing comparisons among the objects, they evaluate the key() function we provided. In this example, the provided function simply selects an attribute. Clearly the functions could do calculations or other operations on the objects.
24.5 Initializer Techniques

When we dene a subclass, we are often extending some superclass to add features. One common design pattern is to do this by dening a subclass to use parameters in addition to those expected by the superclass. We must reuse the superclass constructor properly in our subclass. Referring back to our Card and FaceCard example in Inheritance, we wrote an initializer in FaceCard that referred to Card. The FaceCard.__init__() method evaluates super(FaceCard,self).__init__( rank, suit ). It passed the same arguments to the Card.__init__() method. Note: Older Programs In older programs, youll see an alternative to the super() function. Some classes will have an explicit call to Card.__init__( self, r, s ). Were naming the class (not an object of the class) and explicitly providing the self variable. We can make use of the techniques covered in Advanced Parameter Handling For Functions to simplify our subclass initializer.
class FaceCard( Card ): """Model a 10-point face card: J, Q, K.""" def __init__( self, * args ): super(FaceCard,self).__init__( * args ) self.label= ("J","Q","K")[self.rank-11] self.pval= 10
296
def __str__( self ): return "%2s%s" % ( self.label, self.suit )
In this case we use the def __init__( self, *args ) to capture all of the positional parameters in a single sequence, named args. We then give that entire sequence of positional parameters to Card.__init__(). By using the * operator, we tell Python to explode the list into individual positional parameters. Lets look at a slightly more sophisticated example.
boat.py
class Boat( object ): def __init__( self, name, loa ): """Create a new Boat( name, loa )""" self.name= name self.loa= loa class Catboat( Boat ) def __init__( self, sailarea, * args ): """Create a new Catboat( sail_area, name, loa )""" super(Catboat,self).__init__( * args ) self.main_area= sail_area class Sloop( CatBoat ): def __init__( self, jib_area, * args ); """Create a new Sloop( jib_area, main_area, name, loa )""" super(Sloop,self).__init__( * args ) self.jib_area= jib_area
1. The Boat class denes essential attributes for all kinds of boats: the name and the length overall (LOA). 2. In the case of a Catboat, we add a single sail area to be base denition of Boat. We use the superclass initialization to prepare the basic name and length overall attributes. Then our subclass adds the sailarea for the single sail on a catboat. 3. In the case of a Sloop, we add another sail to the denition of a Catboat. We add the new parameter rst in the list, and the remaining parameters are simply given to the superclass for its initialization.
24.6 Class Variables

The notion of object depends on having instance variables (or attributes) which have unique values for each object. We can extend this concept to include variables that are not unique to each instance, but shared by every instance of the class. Class level variables are created in the class denition itself; instance variables are created in the individual class method functions (usually __init__()). Class level variables are usually variables with values that dont change; these are sometimes called manifest constants or named constants. In Python, theres no formal declaration for a named constant. A class level variable that changes will be altered for all instances of the class. This use of class-level variables is often confusing to readers of your program. Class-level variables with state changes need a complete explanation. This is an example of the more usual approach with class-level constants. These are variables whose values dont vary; instead, they exist to clarify and name certain values or codes. 24.6. Class Variables 297
wheel.py
import random class Wheel( object ): """Simulate a roulette wheel.""" green, red, black= 0, 1, 2 redSet= [1,3,5,7,9,12,14,16,18,19,21,23,25,27,30,32, 34,36] def __init__( self ): self.lastSpin= ( None, None ) def spin( self ): """spin() -> ( number, color ) Spin a roulette wheel, return the number and color.""" n= random.randrange(38) if n in [ 0, 37 ]: n, color= 0, Wheel.green elif n in Wheel.redSet: color= Wheel.red else: color= Wheel.black self.lastSpin= ( n, color ) return self.lastSpin
1. Part of denition of the class Wheel includes some class variables. These variables are used by all instances of the class. By dening three variables, green, red and black, we can make our programs somewhat more clear. Other parts of our program that use the Wheel class can then reference the colors by name, instead of by an obscure numeric code. A program would use Wheel.green to refer to the code for green within the Wheel class. 2. The Wheel class also creates a class-level variable called redSet. This is the set of red positions on the Roulette wheel. This is dened at the class level because it does not change, and there is no benet to having a unique copy within each instance of Wheel. 3. The __init__() method creates an instance variable called lastSpin. If we had multiple wheel objects, each would have a unique value for lastSpin. They all would all, however, share a common denition of green, red, black and redSet. 4. The spin() method updates the state of the wheel. Notice that the class level variables are referenced with the class name: Wheel.green. The instance level variables are referenced with the instance parameter: self.lastSpin. The class level variables are also available using the instance parameter, Wheel.green is the same object as self.green. The spin() method determines a random number between 0 and 37. The numbers 0 and 37 are treated as 0 and 00, with a color of green; a number in the Wheel.redSet is red, and any other number is black. We also update the state of the Wheel by setting self.lastSpin. Finally, the spin() method returns a tuple with the number and the code for the color. Note that we cant easily tell 0 from 00 with this particular class denition. The following program uses this Wheel class denition. It uses the class-level variables red and black to clarify the color code that is returned by spin().
w= Wheel() n,c= w.spin() if c == Wheel.red: print n, "red" elif c == Wheel.black: print n, "black" else: print n
Our sample program creates an instance of Wheel, called w. The program calls the spin() method of Wheel , which updates w.lastSpin and returns the tuple that contains the number and color. We use multiple
298
assignment to separate the two parts of the tuple. We can then use the class-level variables to decode the color. If the color is Wheel.red, we can print "red".
24.7 Static Methods and Class Method

In a few cases, our class may have methods which depend on only the argument values, or only on class variables. In this case, the self variable isnt terribly useful, since the method doesnt depend on any attributes of the instance. Objects which depend on argument values instead of internal status are called Lightweight or Flyweight objects. A method which doesnt use the self variable is called a static method. These are dened using a built-in function named staticmethod(). Python has a handy syntax, called a decorator, to make it easier to apply the staticmethod() function to our method function denition. Well return to decorators in Decorators . Heres the syntax for using the staticmethod() decorator.
@staticmethod def name ( param , ... ) : suite
To evaluate a static method function, we simply reference the method of the class: Class.method() instead of using a specic instance of the class. Example of Static Method. Heres an example of a class which has a static method. Weve dened a deck shuer. It doesnt have any attributes of its own. Instead, it applies its shuffle() algorithm to a Deck object.
class Shuffler( object ): @staticmethod def shuffle( aDeck ): for i in range(len(aDeck)): card= aDeck.get( random.randrange(len(aDeck)) ) aDeck.put( i, card ) d1= Deck() Shuffler.shuffle( d1 )
Class Method. The notion of a class method is relatively specialized. A class method applies to the class itself, not an instance of the class. A class method is generally used for introspection on the structure or denition of the class. It is commonly dened by a superclass so that all subclasses inherit the necessary introspection capability. Generally, class methods are dened as part of sophisticated, dynamic frameworks. For our gambling examples, however, we do have some potential use for class methods. We might want to provide a base Player class who interacts with a particular Game to make betting decisions. Our superclass for all players can dene methods that would be used in a subclass.
24.8 Design Approaches

When we consider class design, we have often have a built-in or library class which does some of the job we want. For example, we want to be able to accumulate a list of values and then determine the average: this is a very list-like behavior, extended with a new feature. There are two overall approaches for extending a class: wrapping and inheritance.
24.7. Static Methods and Class Method
299
Wrap an existing class (for example, a tuple, list, set or map) in a new class which adds features. This allows you to redene the interface to the existing class, which often involves removing features. Inherit from an existing class, adding features as part of the more specialized subclass. This may require you to read more of the original class documentation to see a little of how it works internally. Both techniques work extremely well; there isnt a profound reason for making a particular choice. When wrapping a collection, you can provide a new, focused interface on the original collection; this allows you to narrow the choices for the user of the class. When subclassing, however, you often have a lot of capabilities in the original class you are extending. Duck Typing. In Polymorphism, we mentioned Duck Typing. In Python, two classes are practically polymorphic if they have the same inteface methods. They do not have to be subclasses of the same class or interface (which is the rule in Java.) This principle means that the distinction between wrapping and inheritance is more subtle in Python than in other languages. If you provide all of the appropriate interface methods to a class, it behaves as if it was a proper subclass. It may be a class that is wrapped by another class that provides the same interface. For example, say we have a class Dice, which models a set of individual Die objects.
class Dice( object ): def __init__( self ): self.theDice= [ Die(), Die() ] def roll( self ): for d in self.theDice: d.roll() return self.theDice
In essence, our class is a wrapper around a list of dice, named theDice. However, we dont provide any of the interface methods that are parts of the built-in list class. Even though this class is a wrapper around a list object, we can add method names based on the built-in list class: append(), extend(), count(), insert(), etc.
class Dice( object ): def __init__( self ): self.theDice= [ Die(), Die() ] def roll( self ): for d in self.theDice: d.roll() return self.theDice def append( self, aDie ): self.theDice.append( aDie ) def __len__( self ): return len( self.theDice )
Once weve dened these list-like functions we have an ambiguous situation. We could have a subclass of list, which initializes itself to two Die objects and has a roll() method. We could have a distinct Dice class, which provides a roll() method and a number of other methods that make it look like a list. For people who will read your Python, clarity is the most important feature of the program. In making design decisions, one of your rst questions has to be what is the real thing that Im modeling? Since many alternatives will work, your design should reect something that claries the problem youre solving.
300
24.9 Advanced Class Denition Exercises

24.9.1 Sample Class with Statistical Methods
We can create a Samples class which holds a collection of sample values. This class can have functions for common statistics on the objects samples. For additional details on these algorithms, see the exercises in Tuples and Sequence Processing Functions: map(), lter() and reduce(). Well look at subclassing the built-in list class, by creating a class, Samples, which extends list. Youll need to implement the following methods in your new class. __init__(self, * args) Save a sequence of samples. It could, at this time, also precompute a number of useful values, like the sum, count, min and max of this set of data. When no data is provided, these values would be set to None. __str__(self ) Return string with a summary of the data. An example is a string like "%d values, min %g, max %g, mean %g" with the number of data elements, the minimum, the maximum and the mean. The superclass, list, __repr__() function will return the raw data. mean(self ) Returns the sum divided by the count. mode(self ) Return the most popular of the sample values. Below weve provided an algorithm that can be used to locate the mode of a sequence of samples. For information on computing the mode, see Exercises. median(self ) The median is the value in the middle of the sequence. First, sort the sequence. If there is an odd number of elements, pick the middle-most element. If there is an even number of elements, average the two elements that are mid-most. variance(self ) For each sample, compute the dierence between the sample and the mean, square this value, and sum these squares. The number of samples minus 1 is the degrees of freedom. The sum, divided by the degrees of freedom, is the variance. Note that you need two samples to meaningfully compute a variance. stdev(self ) The square root of the variance. Note that the list superclass already works correctly with the built-in min() and max() functions. In this case, this consequence of using inheritance instead of wrapping turns out to be an advantage.
24.9.2 Shuing Method for the Deck class

Shuing is a matter of taking existing cards and putting them into other positions. There are a many ways of doing this. Well need to try both to see which is faster. In essence, we need to create a polymorphic family of classes that we can use to measure performance.
Shuing Variation 1 - Exchange

For i in range 0 to the number of cards 24.9. Advanced Class Denition Exercises 301
Generate a random number r in the range 0 to the number of cards. Use Multiple Assignement to swap cards at position i and r.
Shuing Variation 2 - Build

Create an empty result sequence, s. While there are cards in the source self.cards sequence. Generate a random number r in the range 0 to the number of cards. Append card r to the result sequence; delete object r from the source self.cards sequence. The pop() method of a sequence can return a selected element and delete it from a sequence nicely. Replace self.cards with the result sequence, s .
Shuing Variation 3 - Sort

Create a key function which actually returns a random value. Use the sort() method of a list with this random key-like function.
self.cards.sort( key=aRandomKeyFunction )
Shuing Variation 4 - random.shue

The random module has a shuffle() method which can be used as follows.
random.shuffle( self.cards )
Of these four algorithms, which is fastest? The best way to test these is to create four separate subclasses of Deck, each of which provides a dierent implementation of the shuffle() method. A main program can then create an instance of each variation on Deck and do several hundred shues. We can create a timer using the time module. The time.clock() function will provide an accurate time stamp. The dierence between two calls to time.clock() is the elapsed time. Because shuing is fast, well do it 100 times to get a time thats large enough to be accurate. Because all of our variations on Deck are polymorphic, our main program should look something like the following.
d1= d2= d3= d4= for DeckExch() DeckBuild() DeckSortRandom() DeckShuffle() deck in ( d1, d2, d3, d4 ): start= time.clock() for i in range(100): d.shuffle() finish= time.clock()
302
24.9.3 Encapsulation
The Shuing exercise built several alternate solutions to a problem. We are free to implement an algorithm with no change to the interface of Deck. This is a important eect of the principal of encapsulation: a class and the clients that use that class are only coupled together by an interface dened by method functions. There are a variety of possible dependencies between a class and its clients. Interface Method Functions. A client can depend on method functions specically designated as an interface to a class. In Python, we can dene internal methods by prexing their names with _. Other names (without the leading _ ) dene the public interface. All Method Functions. A client can depend on all method functions of a class. This removes the complexity of hidden, internal methods. Instance Variables. A client can depend on instance variables in addition to method functions. This can remove the need to write method functions that simply return the value of an instance variable. Global Variables. Both classes share global variables. The Python global statement is one way to accomplish this. Implementation. A client can depend on the specic algorithm being executed by a class. A client method can have expectations of how a class is implemented. What are the advantages and disadvantages of each kind of dependency?
24.9.4 Class Responsibilities

Assigning responsibility to class can be challenging. A number of reasons can be used to justify the functions and instance variables that are combined in a single class. Convenience. A class is dened to do things because well its convenient to write the program that way. Similar Operations. A class is dened because it does all input, all output, or all calculations. Similar Time. A class is dened to handle all initialization, all processing or all nal cleanup. Sequence. We identify some operations which are performed in a simple sequence and bundle these into a single class. Common Data. A class is dened because it has the operations which isolate a data structure or algorithm. What are the possible dierences between theses? What are the advantages and disadvantages of each?
24.10 Style Notes

Classes are perhaps the most important organizational tool for Python programming. Python software is often designed as a set of interacting classes. There are several conventions for naming and documenting class denitions. It is important to note that the suite within a class denition is typically indented four spaces. It is often best to set your text editor with tab stops every four spaces. This will usually yield the right kind of layout. Each functions suite is similarly indented four spaces, as are the suites within compound statements. Blank lines are used sparingly; most typically a single blank line will separate each function denition within the class. A lengthy class denition, with a number of one-liner set-get accessor functions may group the accessors together without any intervening blank lines. 24.10. Style Notes 303
Class names are typically MixedCase with a leading uppercase letter. Members of the class (method functions and attributes) typically begin with a lowercase letter. Class names are also, typically singular nouns. We dont dene People , we dene Person. A collection might be a PersonList or PersonSet. Note that the following naming conventions are honored by Python: single_trailing_underscore_ Used to make a variable names dierent from a similar Python reserved word. For example: range_ is a legal variable name, where range would not be legal. _single_leading_underscore Used to make variable or method names hidden. This conceals them from the dir() function. __double_leading_underscore Class-private names. Use this to assure that a method function is not used directly by clients of a class. __double_leading_and_trailing_underscore__ These are essentialy reserved by Python for its own internals. Docstring Recommendations. The rst line of a class body is the docstring; this provides an overview of the class. It should summarize the responsibilities and collaborators of the class. It should summarize the public methods and instance variables. Individual method functions are each documented in their own docstrings. Tools like Sphinx and Epydoc will look for the __init__() docstring as part of summarizing a class. When dening a subclass, be sure to mention the specic features added (or removed) by the subclass. There are two basic cases: overriding and extending. When overriding a superclass method function, the subclass has replaced the superclass function. When extending a superclass function, the subclass method will call the superclass function to perform some of the work. The override-extend distinctions must be made clear in the docstring. When initializing instance variables in the __init__() function, a string placed after the assignment statement can serve as a denition of the variable. RST Docstring. The most widely-used technique is to write reStructuredText (RST) markup in the docstrings. This is extracted and formatted by tools like Sphinx and epydoc. For information on RST formatting, see PEP 287, as well as http://docutils.sourceforge.net/.
class Dice( object ): """Model two dice used for craps.
Relies on Die class.
:ivar theDice: tuple with two Die instances .. method:: roll
roll dice and return total """ def __init__(self): """Create a new pair of dice.""" self.theDice = ( Die(), Die() ) def roll(self): """Roll the dice and return the sum. :returns: number """ [ d.roll() for d in self.theDice ] t = sum( theDice ) return t
304
Generally, we have been omitting a complete docstring header on each class in the interest of saving some space for the kind of small examples presented in the text.
24.10. Style Notes
305
306
CHAPTER
TWENTYFIVE
SOME DESIGN PATTERNS

The best way to learn object-oriented design is to look at patterns for common solutions to ubiquitous problems. These patterns are often described with a synopsis that gives you several essential features. The writer of a pattern will describe a programming context, the specic problem, the forces that lead to various kinds of solutions, a solution that optimizes the competing forces, and any consequences of choosing this solution. There are a number of outstanding books on patterns. Well pick a few key patterns from one of the books, and develop representative classes in some depth. The idea is to add a few additional Python programming techniques, along with a number of class design techniques. Most of these patterns come from the Gang of Four design patterns book, Design Patterns: Elements of Reusable Object-Oriented Software [Gamma95]. Well look at a few design patterns that illustrate some useful Python programming techniques. Factory This is a pattern for designing a class which is used as a factory for a family of other classes. This allows a client program to use a very exible and extensible Factory to create necessary objects. State This is a pattern for desiging a hierarchy of classes that describes states (or status) and state-specic processing or data. Strategy This is a pattern that helps create a class that supports an extension in the form of alternative strategies for implementing methods.
25.1 Factory
When we add subclasses to a class hierarchy, we may also need to rearrange the statements where objects are created. To provide a exible implementation, it is generally a good idea to centralize all of the object creation statements into a single extendable class. When we extend the subclass hierarchy we can also create a relevant subclass of the centralized object creation class. The design pattern for this kind of centralized object creator can be called a Factory. It contains the details for creating an instance of each of the various subclasses. In the next section, Components, Modules and Packages, well look at how to package a class hierarchy in a module. Often the classes and the factory object are bundled into one seamless module. Further, as the module evolves and improves, we need to preserve the factory which creates instances of other classes in the module. Creating a class with a factory method helps us control the evolution of a module. If we omit the Factory class, then everyone who uses our module has to rewrite their programs when we change our class hierarchy.
307
Extending the Card Class Hierarchy. Well extend the Card class hierarchy, introduced in Inheritance. That original design had three classes: Card, FaceCard and AceCard. While this seems complete for basic Blackjack, we may need to extend these classes. For example, if we are going to simulate a common card counting technique, well need to separate 2-6 from 7-9, leading to two more subclasses. Adding subclasses can easily ripple through an application, leading to numerous additional, sometimes complex changes. We would have to look for each place where the various subclasses of cards were created. The Factory design pattern, however, provides a handy solution to this problem. An object of a class based on the Factory pattern creates instances of other classes. This saves having to place creation decisions throughout a complex program. Instead, all of the creation decision-making is centralized in the factory class. For our card example, we can dene a CardFactory that creates new instances of Card (or the appropriate subclass.)
class CardFactory( object ): def newCard( self, rank, suit ): if rank == 1: return Ace( rank, suit ) elif rank in [ 11, 12, 13 ]: return FaceCard( rank, suit ) else: return Card( rank, suit )
We can simplify our version of Deck using this factory.

class Deck( object ): def __init__( self ): factory= CardFactory() self.cards = [ factory.newCard( rank+1, suit ) for suit in range(4) for rank in range(13) ] Rest of the class is the same
Tip: Centralized Object Creation While it may seem like overhead to centralize object creation in factory objects, it has a number of benets. First, and foremost, centralizing object creation makes it easy to locate the one place where objects are constructed, and x the constructor. Having object construction scattered around an application means that time is spent searching for and xing things that are, in a way, redundant. Additionally, centralized object creation is the norm for larger applications. When we break down an application into the data model, the view objects and the control objects, we nd at least two kinds of factories. The data model elements are often created by fetching from a database, or parsing an input le. The control objects are part of our application that are created during initialization, based on conguration parameters, or created as the program runs based on user inputs. Finally, it makes evolution of the application possible when we are creating a new version of a factory rather than tracking down numerous creators scattered around an application. We can assure ourselves that the old factory is still available and still passes all the unit tests. The new factory creates the new objects for the new features of the application software. Extending the Factory. By using this kind of Factory method design pattern, we can more easily create new subclasses of Card. When we create new subclasses, we do three things: 1. Extend the Card class hierarchy to dene the additional subclasses.
308
Chapter 25. Some Design Patterns
2. Extend the CardFactory creation rules to create instances of the new subclasses. This is usually done by creating a new subclass of the factory. 3. Extend or update Deck to use the new factory. We can either create a new subclass of Deck, or make the factory object a parameter to Deck. Lets create some new subclasses of Card for card counting. These will subdivide the number cards into low, neutral and high ranges. Well also need to subclass our existing FaceCard and Ace classes to add this new method.
class CardHi( Card ): """Used for 10.""" def count( self ): return -1 class CardLo( Card ): """Used for 3, 4, 5, 6, 7.""" def count( self ): return +1 class CardNeutral( Card ): """Used for 2, 8, 9.""" def count( self ): return 0 class FaceCount( FaceCard ): """Used for J, Q and K""" def count( self ): return -1 class AceCount( Ace ): """Used for A""" def count( self ): return -1
A counting subclass of Hand can sum the count() values of all Card instances to get the count of the deck so far. Once we have our new subclasses, we can create a subclass of CardFactory to include these new subclasses of Card. Well call this new class HiLoCountFactory. This new subclass will dene a new version of the newCard() method that creates appropriate objects. By using default values for parameters, we can make this factory option transparent. We can design Deck to use the original CardFactory by default. We can also design Deck to accept an optional CardFactory object, which would tailor the Deck for a particular player strategy.
class Deck( object ): def __init__( self, factory=CardFactory() ): self.cards = [ factory.newCard( rank+1, suit ) for suit in range(4) for rank in range(13) ] Rest of the class is the same
The Overall Main Program. Now we can have main programs that look something like the following.
d1 = Deck() d2 = Deck(HiLoCountFactory())
In this case, d1 is a Deck using the original denitions, ignoring the subclasses for card counting. The d2 Deck is built using a dierent factory and has cards that include a particular card counting strategy. We can now introduce variant card-counting schemes by introducing further subclasses of Card and CardFactory. To pick a particular set of card denitions, the application creates an instance of one of the available subclasses of CardFactory. Since all subclasses have the same newCard() method, the various objects are interchangeable. Any CardFactory object can be used by Deck to produce a valid deck of cards.
25.1. Factory
309
This evolution of a design via new subclasses is a very important technique of object-oriented programming. If we add features via subclasses, we are sure that the original denitions have not been disturbed. We can be completely condent that adding a new feature to a program will not break old features.
25.2 State
Objects have state changes. Often the processing that an object performs depends on the state. In nonobject-oriented programming languages, this state-specic processing is accomplished with long, and sometimes complex series of if statements. The State design pattern gives us an alternative design. As an example, the game of Craps has two states. A players rst dice roll is called a come out roll. Depending on the number rolled, the player immediately wins, immediately loses, or the game transitions to a point roll. The game stays in the point roll state until the player makes their point or crap out with a seven. The following table provides a complete picture of the state changes and the dice rolls that cause those changes. Table 25.1: Craps State Change Rules State Point O; the Come Out Roll; only Pass and Dont Pass bets allowed. Roll 2, 3, 12 7, 11 4, 5, 6, 8, 9, 10 Point On; any additional bets may be placed. 2, 3, 12 11 7 Point, p Non-p number Bet Resolution Craps: Pass bets lose, Dont Pass bets win. Winner: Pass bets win, Dont Pass bets lose. No Resolution Next State Point O Point O Point On the number rolled, p. Point still on Point still on Point O Point O Point still on
No Resolution No Resolution Loser: all bets lose. The table is cleared. Winner: point is made, Pass bets win, Dont Pass bets lose. Nothing; Come bets are activated
The State design pattern is essential to almost all kinds of programs. The root cause of the hideous complexity that characterizes many programs is the failure to properly use the State design pattern. The Craps Class. The overall game of craps can be represented in an object of class Craps. A Craps object would have a play1Round() function to initialize the game in the come out roll state, roll dice, pay o bets, and possibly change states. Following the State design pattern, we will delegate state-specic processing to an object that represents just attributes and behaviors unique to each state of the game. We pan to create a CrapsState class with two subclasses: CrapsStateComeOutRoll and CrapsStatePointRoll. The overall Craps object will pass the dice roll to the CrapsState object for evaluation. The CrapsState object calls methods in the original Craps object to pay or collect when there is a win or loss. The CrapsState object can also return an object for the next state. Additionally, the CrapsState object will have to indicate then the game actually ends. Well look at the Craps object to see the context in which the various subclasses of CrapsState must operate.
310
craps.py
import dice class Craps( object ): """Simple game of craps.""" def __init__( self ): self.state= None self.dice= dice.Dice() self.playing= False def play1Round( self ): """Play one round of craps until win or lose.""" self.state= CrapsStateComeOutRoll() self.playing= True while self.playing: self.dice.roll() self.state= self.state.evaluate( self, self.dice ) def win( self ): """Used by CrapsState when the roll was a winner.""" print "winner" self.playing= False def lose( self ): """Used by CrapsState when the roll was a loser.""" print "loser" self.playing= False
1. The Craps class constructor, __init__(), creates three instance variables: state, dice and playing. The state variable will contain an instance of CrapsState, either a CrapsStateComeOutRoll or a CrapsStatePointRoll. The dice variable contains an instance of the class Dice, dened in Class Denition: the class Statement . The playing variable is a simple switch that is True while we the game is playing and False when the game is over. 2. The play1Round() method sets the state to CrapsStateComeOutRoll, and sets the playing variable to indicate that the game is in progress. The basic loop is to roll the dice and the evaluate the dice. This method calls the state-specic evaluate() function of the current CrapsState object. We give this method a reference to overall game, via the Craps object. That reference allows the CrapsState to call the win() or lose() method in the Craps object. The evaluate() method of CrapsState is also given the Dice object, so it can get the number rolled from the dice. Some propositions (called hardways) require that both dice be equal; for this reason we pass the actual dice to evaluate(), not just the total. 3. When the win() or lose() method is called, the game ends. These methods can be called by the the evaluate() method of the current CrapsState. The playing variable is set to False so that the games loop will end. The CrapsState Class Hierarchy. Each subclass of CrapsState has a dierent version of the evaluate() operation. Each version embodies one specic set of rules. This generally leads to a nice simplication of those rules; the rules can be stripped down to simple if statements that evaluate the dice in one state only. No additional if statements are required to determine what state the game is in.
class CrapsState( object ): """Superclass for states of a craps game.""" def evaluate( self, crapsGame, dice ): raise NotImplementedError def __str__( self ): return self.__doc__
25.2. State
311
The CrapsState superclass denes any features that are common to all the states. One common feature is the denition of the evaluate() method. The body of the method is uniquely dened by each subclass. We provide a denition here as a formal place-holder for each subclass to override. In Java, we would declare the class and this function as abstract. Python lacks this formalism, but it is still good practice to include a placeholder. Subclasses for Each State. The following two classes dene the unique evaluation rules for the two game states. These are subclasses of CrapsState and inherit the common operations from the superclass.
class CrapsStateComeOutRoll ( CrapsState ): """Come out roll rules.""" def evaluate( self, crapsGame, dice ): if dice.total() in [ 7, 11 ]: crapsGame.win() return self elif dice.total() in [ 2, 3, 12 ]: crapsGame.lose() return self return CrapsStatePointRoll( dice.total() )
The CrapsStateComeOutRoll provides an evaluate() function that denes the come out roll rules. If the roll is an immediate win (7 or 11), it calls back to the Craps object to use the win() method. If the roll is an immediate loss (2, 3 or 12), it calls back to the Craps object to use the lose() method. In all cases, it returns an object which is the next state; this might be the same instance of CrapsStateComeOutRoll or a new instance of CrapsStatePointRoll.
class CrapsStatePointRoll ( CrapsState ): """Point roll rules.""" def __init__( self, point ): self.point= point def evaluate( self, crapsGame, dice ): if dice.total() == 7: crapsGame.lose() return None if dice.total() == self.point: crapsGame.win() return None return self
The CrapsStatePointRoll provides an evaluate() method that denes the point roll rules. If a seven was rolled, the game is a loss, and this method calls back to the Craps object to use the lose() method, which end the game. If the point was rolled, the game is a winner, and this method calls back to the Craps object to use the win() method. In all cases, it returns an object which is the next state. This might be the same instance of CrapsStatePointRoll ` or a new instance of :class:`CrapsStateComeOutRoll. Extending the State Design. While the game of craps doesnt have any more states, we can see how additional states are added. First, a new state subclass is dened. Then, the main object class and the other states are updated to use the new state. An additional feature of the state pattern is its ability to handle state-specic conditions as well as statespecic processing. Continuing the example of craps, the only bets allowed on the come out roll are pass and dont pass bets. All other bets are allowed on the point rolls. We can implement this state-specic condition by adding a validBet() method to the Craps class. This will return True if the bet is valid for the given game state. It will return False if the bet is not valid. Since this is a state-specic condition, the actual processing must be delegated to the CrapsState subclasses.
312
25.3 Strategy
Objects can often have variant algorithms. The usual textbook example is an object that has two choices for an algorithm, one of which is slow, but uses little memory, and the other is fast, but requires a lot of storage for all that speed. In our examples, we can use the Strategy pattern to isolate the details of a betting strategy from the rest of a casino game simulation. This will allow us to freely add new betting strategies without disrupting the simulation. One strategy in Roulette is to always bet on black. Another strategy is to wait, counting red spins and bet on black after weve seen six or more reds in a row. These are two alternate player strategies. We can separate these betting decision algorithms from other features of player. We dont want to create an entire subclass of player to reect this choice of algorithms. The Strategy design pattern helps us break something rather complex, like a Player, into separate pieces. The essential features are in one object, and the algorithm(s) that might change are in separate strategy object(s). The essential features are dened in the core class, the other features are strategies that are used by the core class. We can then create many alternate algorithms as subclasses of the plug-in Strategy class. At run time, we decide which strategy object to plug into the core object. The Two Approaches. As mentioned in Design Approaches, we have two approaches for extending an existing class: wrapping and inheritance. From an overall view of the collection of classes, the Strategy design emphasizes wrapping. Our core class is a kind of wrapper around the plug-in strategy object. The strategy alternatives, however, usually form a proper class hierarchy and are all polymorphic. Lets look at a contrived, but simple example. We have two variant algorithms for simulating the roll of two dice. One is quick and dirty and the other more exible, but slower. First, we create the basic Dice class, leaving out the details of the algorithm. Another object, the strategy object, will hold the algorithm
class Dice( object ): def __init__( self, strategy ): self.strategy= strategy self.lastRoll= None def roll( self ): self.lastRoll= self.strategy.roll() return self.lastRoll def total( self ): return reduce( lambda a,b:a+b, self.lastRoll, 0 )
The Dice class rolls the dice, and saves the roll in an instance variable, lastRoll, so that a client object can examine the last roll. The total() method computes the total rolled on the dice, irrespective of the actual strategy used. The Strategy Class Hierarchy. When an instance of the Dice class is created, it must be given a strategy object to which we have delegated the detailed algorithm. A strategy object must have the expected interface. The easiest way to be sure it has the proper interface is to make each alternative a subclass of a strategy superclass.
import random class DiceStrategy( object ): def roll( self ): raise NotImplementedError
The DiceStrategy class is the superclass for all dice strategies. It shows the basic method function that all subclasses must override. Well dene two subclasses that provide alternate strategies for rolling dice. The rst, DiceStrategy1 is simple. 25.3. Strategy 313
class DiceStrategy1( DiceStrategy ): def roll( self ): return ( random.randrange(6)+1, random.randrange(6)+1 )
This DiceStrategy1 class simply uses the random module to create a tuple of two numbers in the proper range and with the proper distribution. The second alternate strategy, DiceStrategy2, is quite complex.
class DiceStrategy2( DiceStrategy ): class Die: def __init__( self, sides=6 ): self.sides= sides def roll( self ): return random.randrange(self.sides)+1 def __init__( self, set=2, faces=6 ): self.dice = tuple( DiceStrategy2.Die(faces) for d in range(set) ) def roll( self ): return tuple( x.roll() for x in self.dice )
This DiceStrategy2 class has an internal class denition, Die that simulates a single die with an arbitrary number of faces. An instance variable, sides shows the number of sides for the die; the default number of sides is six. The roll() method returns are random number in the correct range. The DiceStrategy2 class creates a number of instances of Die objects in the instance variable dice. The default is to create two instances of Die objects that have six faces, giving us the standard set of dice for craps. The roll() function creates a tuple by applying a roll() method to each of the Die objects in self.dice. Creating Dice with a Plug-In Strategy. We can now create a set of dice with either of these strategies.
dice1= Dice( DiceStrategy1() ) dice2 = Dice( DiceStrategy2() )
The dice1 instance of Dice uses an instance of the DiceStrategy1 class. This strategy object is used to constuct the instance of Dice. The dice2 variable is created in a similar manner, using an instance of the DiceStrategy2 class. Both dice1 and dice2 are of the same class, Dice, but use dierent algorithms to achieve their results. This technique gives us tremendous exibility in designing a program. Multiple Patterns. Construction of objects using the strategy pattern works well with a Factory pattern, touched on in Factory. We could, for instance, use a Factory Method to decode input parameters or command-line options. This give us something like the following.
class MakeDice( object ): def newDice( self, strategyChoice ): if strategyChoice == 1: strat= DiceStrategy1() else: strat= DiceStrategy2() return Dice( strat )
This allows a program to create the Dice with something like the following.
dice = MakeDice().newDice( :replaceable:`someInputOption` )
314
When we add new strategies, we can also subclass the MakeDice class to include those new strategy alternatives.
25.4 Design Pattern Exercises

25.4.1 Alternate Counting Strategy
A simple card counting strategy in Blackjack is to score +1 for cards of rank 3 to 7, 0 for cards of rank 2, 8 and 9 and -1 for cards 10 to King and Ace. The updates to the Card class hierarchy are shown in the text. Create a subclass of CardFactory, which replaces newCard() with a version that creates the correct subclass of Card, based on the rank.
25.4.2 Six Reds

A common strategy in Roulette is to wait until six reds in a row are spun and then start betting on only black. There are three player betting states: waiting, counting and betting. A full simulation will require a RouletteGame class to spin the wheel and resolve bets, a Wheel object to produce a sequence of random spins, and a Table to hold the individual bets. Wed also need a class to represent the Bet s. Well skim over the full game and try to focus on the player and player states. A Player has a stake which is their current pool of money. The RouletteGame oers the Player an opportunity to bet, and informs the player of the resulting spin of the wheel. The Player uses a SixRedsState to count reds and bet on black. The various SixRedsState classes have two methods, a bet() method decides to bet or not bet, and an outcome() method that records the outcome of the previous spin. Each class implements these methods dierently, because each class represents a dierent state of the players betting policy. counting In the counting state, the player is counting the number of reds in a row. If a red was spun and the count is < 6, add one to a red counter and stay in this state. If a red is spun and the count is = 6, add one to a red counter and transition to the betting state. If black or green is spun, reset the count to zero. :betting [In the betting state, the player is] betting on black. In a more advanced version, the player would also increase their bet for each red count over six. If a red was spun, add one to a red counter and stay in the betting state. If black was spun, transition to the counting state. If green was spun, transition to the counting state. Caution: Caution Well focus on the SixRedsState design. We wont spend time on the actual betting or payouts. For now, we can simply log wins and losses with a print statement. First, build a simple Player class, that has the following methods. class Player() __init__(self, stake) Sets the players initial stake. For now, we wont do much with this. In other player strategies, however, this may inuence the betting. More importantly, this will set the initial betting state of Counting. __str__(self ) Returns a string that includes the current stake and state information for the player. 25.4. Design Pattern Exercises 315
getBet(self ) This will use the current state to determine what bet (if any) to place. outcome(self, spin) This will provide the color information to the current state. It will also update the players betting state with a state object returned the current state. Generally, each state will simple return a copy of itself. However, the counting state object will return a betting state object when six reds have been seen in a row. Second, create a rudimentary RouletteGame that looks something like the following. class RouletteGame() __init__(self, player ) A RouletteGame is given a Player instance when it is constructed. __str__(self ) Its not clear what wed display. Maybe the player? Maybe the last spin of the wheel? play1Round(self ) The play1Round() method gets a bet from the Player object, spins the wheel, and reports the spin back to the Player object. A more complete simulation would also resolve the bets, and increase the players stake with any winnings. Note that calling the Players outcome() method does two things. First, it provides the spin to the player playRounds(self, rounds=12 ) A simple loop that calls self.play1Round as many times as required. For guidance on designing the Wheel used by the RouletteGame, see Class Variables and Function Denition: The def and return Statements. State Class Hierarchy. The best approach is to get the essential features of RouletteGame, Wheel and Player to work. Rather than write a complete version of the players getBet() and outcome() methods, we can use simple place-holder methods that simply print out the status information. Once we have these objects collaborating, then the three states can be introduced. The superclass, SixRedsState, as well as the two subclasses, would all be similar to the following. class SixRedsState() __init__(self, player ) The superclass initialization saves the player object. Some subclasses will initialize a count to zero. __str__(self ) The superclass __str__() method should return a NotImplemented value, to indicate that the superclass was used improperly. getBet(self ) The getBet() method either returns None in the waiting and counting states, or returns a bet on red in the betting state. The superclass can return None, since thats a handy default behavior. outcome(self, spin) The outcome() method is given a tuple with a number and a color. Based on the rules given above, each subclass of SixRedsState will do slightly dierent things. The superclass can return NotImplemented. We need to create two subclasses of SixRedState : SixRedCounting In this state, the getBet() method returns None ; this behavior is dened by the superclass, so we dont need to implement this method. The outcome() method checks the spin. If it is Red, this object increments the count by one. If it is black it resets the
316
count to zero. If the count is six, return an instance of SixRedBetting . Otherwise, return self as the next state. SixRedBetting In this state, the getBet() method returns a bet on Black; for now, this can be the string "Black". The outcome() method checks the spin. If it is Red, this object increments the count by one and returns self. If the spin is black it returns an instance of SixRedCounting. This will stop the betting and start counting. Once we have the various states designed, the Player can then be revised to initialize itself with an instance of the wating class, and then delegate the getBet() request from the game to the state object, and delegate the outcome() information from the game to the state object.
25.4.3 Roulette Wheel Alternatives

There are several possible implementations of the basic Roulette wheel. One variation simply uses random.randrange() to generate numbers in the range of 0 to 37, and treats 37 as double zero. To separate double zero from zero, its best to use character string results. Another strategy is to create a sequence of 38 strings (00, 0, 1, etc.) and use random.choice() to pick a number from the sequence. Create a superclass for WheelStrategy and two subclasses with these variant algorithms. Create a class for Wheel which uses an instance of WheelStrategy to get the basic number. This class for Wheel should also determine whether the number is red, black or green. The Wheel class spin() method should return a tuple with the number and the color. Create a simple test program to create an instance of Wheel with an instance of WheelStrategy. Collect 1000 spins and print the frequency distribution.
25.4.4 Shuing Alternatives

Shuing rearranges a deck or shoe of multiple decks; there are many possible algorithms. First, you will need a Card class to keep basic rank and suit information. Next, you will need a basic Deck class to hold cards. See Playing Cards and Decks for additional details. We looked at shuing in an earlier exercise, but packaged it as part of the Deck class, not as a separate strategy. See Advanced Class Denition Exercises. We can rework those exercises to separate shuing from Deck. The Deck class must have a shuffle() function; but this should simply call a method of the shue strategy object. Because the Deck is a collection of Card s, the Deck object should be passed to the shue strategy. The call would like something like this:
class Deck( object ): Other parts of Deck def shuffle( self ): self.shuffleStrategy.shuffle( self )
Create a superclass for shue strategies. Create a subclass for each of the following algorithms: For each card position in the deck, exchange it with a card position selected randomly For even-numbered card position (positions 0, 2, 4, 6, etc.) exchange it with an odd-numbered card position, selected randomly (random.choice from 1, 3, 5, 7, 9, etc.) Swap two randomly-selected positions; do this 52 times Create a simple test program that creates a Deck with each of the available a Shuffle strategy objects. 25.4. Design Pattern Exercises 317
25.4.5 Shuing Quality

An open issue in the shuing exercise is the statistical quality of the shuing actually performed. Statistical tests of random sequences are subtle, and more advanced than we can cover in this set of exercises. What we want to test is that each card is equally likely to land in each position of the deck. We can create a dictionary, with the key of each card, and the item associated with that key is a list of positions in which the card occured. We can evaluate a shue algorithm as follows.
Test A Shue
Setup. Create a Deck . Create an empty dictionary, positions for recording card positions. For each Card in the Deck, insert the Card in the positions dictionary; the value associated with the Card is a unique empty list used to record the positions at which this Card is found. For Each Strategy. Perform the following evaluation for an instance of each Shuffle class, s. Create Deck. Set the Decks current shue strategy to s. Shue. Shue the Deck. Record All Positions. For each card in the deck, d. Record Card Position. Locate the Cards position list in the positions dictionary; append the position of this Card to the list in the positions dictionary. Chi-Squared. The chi-squared statistical test can be used to compare the actual frequency histogram to the expected frequency histogram. If you shue each deck 520 times, a given card should appear in each of the positions approximately 10 times. Ideally, the distribution is close to at, but not exactly. The chi-squared test compares sequence of actual frequencies, a, and a sequence of expected frequencies, e. It returns the chi-squared metric for the comparison of these two sequences. Both sequences must be the same length and represent frequencies in the same order. 2 = (ai ei )2 ei
0in
We can use the built-in zip() function to interleave the two lists, creating a sequence of tuples of ( actual, expected ). This sequence of tuples can be used with the multiple-assignment for loop to assign a value from actual to one variable, and a value from expected to another variable. This allows a simple, elegant for statement to drive the basic comparison function.
318
CHAPTER
TWENTYSIX
CREATING OR EXTENDING DATA TYPES

When we use an operator, like + or *, what happens depends on the types of the objects involved. When we say c*2, the value depends on the type of c . If c is numeric, then 2 may have to be converted to the same type of number as c, and the answer will be a number. If, however, c is a sequence, the result is a new sequence.
>>> c=8.0 >>> c*2 16.0 >>> c="8.0" >>> c*2 '8.08.0' >>> c=(8,0) >>> c*2 (8, 0, 8, 0)
The selection of appropriate behavior is accomplished by the relatively simple mechanism of special method names within Python. Each class of objects, either built-in or created by a programmer, can provide the required special method names to create the intimate relationship between the class, the built-in functions and the mathematical operators. Well look at the general principles in Semantics of Special Methods. If you provide special methods, you can make your class behave like a built-in class. Your class can participate seamlessly with built-in Python functions like str(), len(), repr(). These are basic special methods, covered in Basic Special Methods. Well look at some special attributes in Special Attribute Names. Your class can also participate with the usual mathematical operators like + and *. Well look at this in Numeric Type Special Methods. Additionally, your class could also use the collection operators in a manner similar to a dict, set, tuple or list. Well look at this in Collection Special Method Names, Collection Special Method Names for Iterators and Iterable, Collection Special Method Names for Sequences, Collection Special Method Names for Sets and Collection Special Method Names for Mappings. Well look at a few examples in Mapping Example and Iterator Examples. Well look at special names for handling attributes are handled in Attributes, Properties and Descriptors. Additionally, we can extend built-in classes. We do this by extending some of the special methods to do additional or dierent things. Well examine this in Extending Built-In Classes.
319
26.1 Semantics of Special Methods

Python has a number of language features that interact with the built-in data types. For example, objects of all built-in types can be converted to strings. You can use the built-in str() function to perform these conversions. The str() function invokes the __str__() special method of the given object. In eect, str(a) is evaluated as a.__str__(). When you create your own class, you must supply the specially-named method, __str__(), that the built-in str() function can use to successfully convert your classes values to strings. The default implementation of __str__() is pretty lame; youll always want to override it. In Special Method Names we introduced a few special method names. We looked at __init__(), which is evaluated implicitly when an object is created. We looked at __str__(), which is used by the str() function and __repr__() that is used by the repr() function. A huge number of Python features work through these special method names. When you provide appropriate special methods for your class, it behaves more like a built-in class. You may be suspicious that the special method name __str__() matches the built-in function str(). There is no simple, obvious rule. Many of the built-in functions invoke specially-named methods of the class that are similar. The operators and other special symbols, however, cant have a simple rule for pairing operators with special method names. Youll have to actually read the documentation for built-in functions (Library Reference, section 2.1) and special method names (Language Reference, section 3.3) to understand all of the relationships. Categories of Special Method Names. The special methods fall into several broad categories. The categories are dened by the kind of behavior your class should exhibit. Basic Object Behaviors A number of special method names make your object behave like oher built-in objects. These special methods make your class respond to str(), repr() and various comparison operators. This also includes methods that allow your object to respond to the hash() function, which allows instances of your class to be a key to a mapping. Numeric Behaviors These special methods allow your class to respond to the artithmetic operators: +, -, *, /, %, **, <<, >>, and, or and not. When you implement these special methods, your class will behave like the built-in numeric types. Container Behaviors If your new class is a container (or a collection), there are a number of methods required so that your class can behave like the built-in collection types (sequence, set, mapping). Section 3.4.5 of the Python Language Reference calls them containers. However, well call them collections below. Iterator Behavior An iterator has a unique protcol. The for statement requires an __iter__() method to product an iterator for an iterable object. The iterator must provide the next() method to actually do the iteration. While this isnt tied to a collection, its commonly used with collections, so well show this with the collection special names below. Attribute Handling Behavior Some special methods customize how your class responds to the . operator for manpulating attributes. For example, when you evaluate object.attr. This is commonly used when attribute manipulation is more complex than simply locating an attribute that was dened by __init__(). Function Behavior You can make your object behave like a function. When you dene the method __call__(), your object is callable , and can be used as if it was a function. Statement Interaction There are a few special methods required by statements. The for statement requires an __iter__() method to product an iterator for an iterable object. The iterator object must implement a next() method.
320
Chapter 26. Creating or Extending Data Types
The with statement requires __enter__() and __exit__() methods.
26.2 Basic Special Methods

In addition to __init__() and __str__() there are a number of methods which are appropriate for classes of all kinds. __init__(self, args...) Called when a new instance of the class is created. Note that this overrides any superclass __init__() method; to do superclass initialization rst, you must evaluate the superclass __init__() like this: super( class, self ).__init__( args... ). The super() function identies the superclass of your class, class. __del__(self ) Called when the this object is no longer referenced anywhere in the running program; the object is about to be removed by garbage collection. This is rarely used. Note that this is called as part of Python garbage collection; it is not called by the del statement. __repr__(self ) Called by the repr() built-in function. Typically, the string returned by this will look like a valid Python expression to reconstruct the object. __str__(self ) Called by the str() built-in function. This is called implicitly by the print statement (and print() function) to convert an object to a convenient, pretty string representation. __eq__(self, other ) Called by ==. __ne__(self, other ) Called by !=. __lt__(self, other ) Called by <. __le__(self, other ) Called by <=. __gt__(self, other ) Called by >. __ge__(self, other ) Called by >=. __hash__(self ) Called during dictionary operations, and by the built-in function hash() to transform an object to a unique 32-bit integer hash value. Objects which compare equal should also have the same hash value. If a class does not dene a __eq__() method it should not dene a __hash__() operation either. Classes with mutable objects can dene __eq__() but should not dene __hash__(), or objects would move around in the dictionary. __nonzero__(self ) Called during truth value testing; must return 0 or 1. If this method is not dened, and __len__() is dened, then __len__() is called based on the assumption that this is a collection. If neither function is dened, all values are considered True.
26.2. Basic Special Methods
321
26.3 Special Attribute Names

As part of creating a class denition, Python adds a number of special attributes. These are informational in nature, and cannot not be easily be changed except by redening the class or function, or reimporting the module. __class__ This objects class name. The class has a __name__ attribute which is the name of the class. __module__ The module in which the class was dened. __dict__ The dictionary which contains the objects attributes and methods. __bases__ The base classes for this class. These are also called superclasses. __doc__ The documentation string. This is part of the response produced by the help() function. Heres an example of how the class docstring is used to produce the help() results for a class.
import random print random.Random.__doc__ help(random.Random)
26.4 Numeric Type Special Methods

When creating a new numeric data type, you must provide denitions for the essential mathematical and logical operators. When we write an expression using the usual +, -, *, /, //, %, or **, Python transforms this to method function calls. Consider the following:
v1= MyClass(10,20) v2= MyClass(20,40) x = v1 + v2
In this case, Python will evaluate line 3 as if you had written:

x = v1.__add__( v2 )
Every arithmetic operator is transformed into a method function call. By dening the numeric special methods, your class willwork with the built-in arithmetic operators. There are, however, some subtleties to this. Forward, Reverse and In-Place Method Functions. There are as many as three variant methods required to implement each operation. For example, * is implemented by __mul__(), __rmul__() and __imul__(). There are forward and reverse special methods so that you can assure that your operator is properly commutative. There is an in-place special method so that you can implement augmented assignment eiciently (see Augmented Assignment ). You dont need to implement all three versions. If you implement just the forward version, and your program does nothing too odd or unusual, everything will work out well. The reverse name is used for special situations that involve objects of multiple classes. Python makes two attempts to locate an appropriate method function for an operator. First, it tries a class based on the left-hand operand using the forward name. If no suitable special method is found, it tries the the right-hand operand, using the reverse name. 322 Chapter 26. Creating or Extending Data Types
Additionally, we can return the special value NotImplemented to indicate that a specic version of a method function is not implemented. This will cause Python to skip this method function and move on to an alternative. Consider the following:
v1= MyClass(10,20) x = v1 * 14 y = 28 * v1
Both lines 2 and 3 require conversions between the built-in integer type and MyClass. For line 2, the forward name is used. The expression v1*14 is evaluated as if it was
x = v1.__mul__( 14 )
For line 3, the reverse name is used. The expression 28*v1 is evaluated as if it was
y = v1.__rmul__( 28 )
Note: Python 3 and Coercion Historically, as Python has evolved, so have the ins and outs of argument coercion from data type to data type. Weve omitted the real details of the coercion rules. In Python 3.0, the older notion of type coercion and the coerce() function will be dropped altogether, so well focus on the enduring features that will be preserved. Section 3.4.8 of the Python Language Reference covers this in more detail; along with the caveat that the Python 2 rules have gotten too complex. The Operator Algorithm. The algorithm for determing what happens with x op y is approximately as follows. Note that a special method function can return the value NotImplemented. This indicates that the operation cant work directly only the values, and another operation should be chosen. The rules provide for a number of alternative operations, this allows a class to be designed in a way that will cooperate successfully with potential future subclasses. 1. The expression string % anything is a special case and is handled rst. This assures us that the value of anything is left untouched by any other rules. Generally, it is a tuple or a dictionary, and should be left as such. 2. If this is an augmented assignment statement (known as an in-place operator, e.g., variable $= anything) for some operator, $. If the left operand implements __iXopX__(), then that __iXopX__() special method is invoked without any coercion. These in-place operators permit you to do an eicient udpate the left operand object instead of creating a new object. 3. As a special case, the two operators are superclass XopX subclass, then the right operand (the subclass) __rXopX__() method is tried rst. If this is not implemented or returns NotImplemented then the the left operand (the superclass) __XopX__() method is used. This is done so that a subclass can completely override binary operators, even for built-in types. 4. For x op y, x.__op__(y) is tried rst. If this is not implemented or returns NotImplemented , y.__rop__(x) is tried second. The following functions are the forward operations, used to implement the associated expressions.
26.4. Numeric Type Special Methods
323
method function __add__(other)() __sub__(other)() __mul__(other)() __div__(other)() __truediv__(other)() __floordiv__(other)() __mod__(other)() __divmod__(other)() __pow__(other [, modulo] )() __lshift__(other)() __rshift__(other)() __and__(other)() __or__(other)() __xor__(other)()
original expression self + other self - other self * other self / other classical Python 2 division self / other when from __future__ import division or Python 3 self // other self % other divmod( self, other ) self ** other self << other self >> other self & other self | other self ^ other
The method functions in this group are used to resolve operators by attempting them using a reversed sense. method function __radd__(other)() __rsub__(other)() __rmul__(other)() __rdiv__(other)() __rtruediv__(other)() __rfloordiv__(other)() __rmod__(other)() __rdivmod__(other)() __rpow__(other [,modulo])() __rlshift__(other)() __rrshift__(other)() __rand__(other)() __ror__(other)() __rxor__(other)() original expression other + self other - self other * self other / self classical Python 2 division other / self when from __future__ import division or Python 3 other // self other % self divmod( other, self ) other ** self other << self other >> self other & self other | self other ^ self
The method functions in this group are used to resolve operators by attempting them using an incremental sense. method function __iadd__(other)() __isub__(other)() __imul__(other)() __idiv__(other)() __itruediv__(other)() __ifloordiv__(other)() __imod__(other)() __ipow__(other [,modulo])() __ilshift__(other)() __irshift__(other)() __iand__(other)() __ior__(other)() __ixor__(other)() original expression self self self self self self self self self self self self self += other -= other *= other /= other classical Python 2 division /= other when from __future__ import division or Python 3 //= other %= other **= other <<= other >>= other &= other |= other ^= other
The method functions in the following group implement the basic unary operators.
324
method function __neg__() __pos__() __abs__() __invert__() __complex__() __int__() __long__() __float__() __oct__() __hex__() __index__()
original expression - self + self abs( self ) ~ self complex( self ) int( self ) long( self ) float( self ) oct( self ) hex( self ) sequence[self], usually as part of a slicing operation which required integers
Rational Number Example. Consider a small example of a number-like class. The Rational Numbers exercise in Classes describes the basic structure of a class to handle rational math, where every number is represented as a fraction. Well add some of the special methods required to make this a proper numeric type. Well nish this in the exercises.
class Rational( object ): def __init__( self, num, denom= 1L ): self.n= long(num) self.d= long(denom) def __add__( self, other ): return Rational( self.n*other.d + other.n*self.d, self.d*other.d ) def __str__( self ): return "%d/%d" % ( self.n, self.d )
This class has enough methods dened to allow us to add fractions as follows:
>>> x = Rational( 3, 4 ) >>> y = Rational( 1, 3 ) >>> x + y 7/12
In order to complete this class, we would need to provide most of the rest of the basic special method names (there is almost never a need to provide a denition for __del__()). We would also complete the numeric special method names. Additionally, we would have to provide correct algorithms that reduced fractions, plus an additional conversion to respond with a mixed number instead of an improper fraction. Well revisit this in the exercises. Conversions From Other Types. For your class to be used successfully, your new numeric type should work in conjunction with existing Python types. You will need to use the isinstance() function to examine the arguments and make appropriate conversions. Consider the following expressions:
x = Rational( 22, 7 ) y = x+3 z = x+0.5
Our original __add__() method assumed that the other object is a Rational object. But in this case, weve provided int and float values for other. Generally, numeric classes must be implemented with tests for various other data types and appropriate conversions.
26.4. Numeric Type Special Methods
325
We have to use the isinstance() function to perform checks like the following: isinstance( other, int ). This allows us to detect the various Python built-in types. Function Reference vs. Function Call In this case, we are using a reference to the int function; we are not evaluating the int() function. If we incorrectly said isinstance( other, int() ), we would be attempting to evaluate the int() function without providing an argument; this is clearly illegal. If the result of isinstance( other, types ) is True in any of the following cases, some type of simple conversion should be done, if possible. complex. isinstance( other, complex ). You may want to raise an exception here, since its hard to see how to make rational fractions and complex numbers conformable. If this is a common situation in your application, you might need to write an even more sophisticated class that implements complex numbers as a kind of rational fraction. Another choice is to write a version of the abs() function of the complex number, which creates a proper rational fraction for the complex magnitude of the given value. oat. isinstance( other, float ). One choice is to truncate the value of other to long , using the built-in long() function and treat it as a whole number, the other choice is to determine a fraction that approximates the oating point value. int or long. isinstance( other, (int,long) ). Any of these means that the other value is clearly the numerator of a fraction, with a denominator of 1. string. isinstance( other, basestring ). We might try to convert the other value to a long using the built-in long() function. If the conversion fails, we could try a float. The exception thats thrown from any of the attempted conversions will make the error obvious. The basestring type, by the way, is the superclass for ASCII strings ( str ) and Unicode strings ( unicode ). Rational. isinstance( other, Rational ). This indicates that the other value is an instance of our Rational class; we can do the processing as expected, knowing that the object has all the methods and attributes we need. Here is a version of __sub__() with an example of type checking. If the other argument is an instance of the class Rational, we can perform the subtract operation. Otherwise, we attempt to convert the other argument to an instance of Rational and attempt the subtraction between two Rationals.
def __sub__( self, other ): if isinstance(other,Rational): return Rational( self.n*other.d - other.n*self.d, self.d*other.d ) else: return self - Rational(long(other))
An alternative to the last line of code is the following.

return Rational( self.n-long(other)*self.d, self.d )
While this second version performs somewhat quicker, it expresses the basic rational addition algorithm twice, once in the if: suite and again in the else: suite. A principle of object oriented programming is to maximize reuse and minimize restating an algorithm. My preference is to state the algorithm exactly once and reuse it as much as possible. Reverse Operators. In many cases, Python will reverse the two operands, and use a function like __rsub__() or __rdiv__(). For example: 326 Chapter 26. Creating or Extending Data Types
def __rsub__( self, other ): if isinstance(other,Rational): return Rational( other.n*self.d - self.n*other.d, self.d*other.d ) else: return Rational(long(other)) - self
You can explore this behavior with short test programs like the following:
x = Rational( 3,4 ) print x-5 print 5-x
26.5 Collection Special Method Names

The various collection special method names can be organized several dierent ways. Above, in Semantics of Special Methods we claimed that a bunch of special method names were related to container and iterator behavior. These categories from the language reference dont tell the whole story. Python gives us additional tools to create classes that behave like the built-in collection classes. We can use the abstract base classes in the collections module to jump-start our denition of new types of collections. Each abstract base class (ABC) in the collections module provides a common feature (or set of features) with the method functions that are required to implement that feature. In some cases, the features build on each other, and a number of method functions are required. Since each of the ABC classes is abstract, theyre missing the implementation of one or more methods. To use these classes, youll have to provide the necessary methods. One very important consequence of using the collections base classes is that it creates standardized names for the various features. This simplies the assertions that might be required when checking the argument values to a function or method function. For more information, see section 9.3.1 of the Python Library Reference, as well as PEP 3119. Some Foundational Denitions. Well look at some foundational abstract classes rst. Each of these denes a small group of fundamental features. Well use this in the next section to build more sophisticated classes. Well look at the following: Container. What makes a container? The in test for membership. Extend this class to make objects of your class a container. Hashable. This makes something usable as a key for mappings or sets. Extend this class to make objects of your class a hashable key. Sized. This makes something report how many elements it has. Extend this class to make objects of your class respond to the len() function with a size. Callable. Extend this class to make objects of your class behave like a function. class Container() To be a Container object, the class must provide the __contains__() method. __contains__(self, value) Return true if the value is in the container.
26.5. Collection Special Method Names
327
This example is a little silly, but it shows a tuple-like container that secretly adds an additional item.
class BonusContainer( collections.Container ): def __init__( self, \*members ): self.members= members + ( 'SecretKey', ) def __contains__( self, value ): return value in self.members
class Hashable() To be a Hashable object, the class must provide the __hash__() method. This is a requirement for any object we might want to use as a key to a dictionary. __hash__(self ) Return a hash for this Hashable object. The easiest wasy to compute a hash is to sume the hashes of the various elements. Heres an example of a class that creates a hash, suitable for use in a dictionary.
class StockBlock( collections.Hashable ): def __init__( self, name, price, shares ): self.name= name self.price= price self.shares= shares self._hash= hash(self.name)+hash(self.price)+hash(self.shares) def __hash__( self ): return self._hash
class Sized() To be a Sized object, the class must provide the __len__() method. __len__(self ) Return the size of this collection. This is generally understood to be the nunber of items in the collection. class Callable() To be a Callable object, the class must provie the __call__() method. Functions are callable objects, but we can also dene a class that creates callable objects, similar to a function denition. __call__(self, parameters...) This method is called when the object is used as if it were a function. We might create and use callable object with something like the following.
callable_object = MyCallable() callable_object( argument, values )
Heres an example of a callable class denition. We can use this to create callable objects that are essentially functions.
class TimesX( collections.Callable ): def __init__( self, factor ): self.factor= factor def __call__( self, argument ): return argument * self.factor
We can use this class to create callable objects as follows:
328
>>> times_2= TimesX(2) >>> times_2( 5 ) 10 >>> import math >>> times_pi= TimesX(math.pi) >>> times_pi( 3*3 ) 28.274333882308138
1.We created a callable object, times_2, as an instance of TimesX with a factor of 2. 2.We applied our times_2 function to a value of 5. 3.We created a callable object, times_pi, as an instance of TimesX with a factor of math.pi. 4.We applied our times_i function to a value of 3*3.
26.6 Collection Special Method Names for Iterators and Iterable

The following collection class denitions introduce special method names to make your class respond to the iterator protocols used by the for statement. Well look at dening iterable contains and iterators in depth in Collection Special Method Names for Iterators and Iterable. class Iterable() To be an Iterable object, the class must provide the __iter__() method. __iter__(self ) Returns an Iterator for this Iterable object. Generally, this will look like the following.
import collections class MyIterable( collections.Iterable ): # all the other methods of the MyIterable collection def __iter__( self ): return MyIteratorClass( self )
This means that we have a class that extends collections.Iterator which will control the iteration through the given MyIterable collection. These two classes are sometimes called friends, since the the Iterator often has deeper access to the Iterable. class Iterator() To be an Iterator object, the class must provide the __next__() method. Additionally, an Iterator is itself Iterable, so it must provide a __iter__() method. Generally, the __iter__() method just does return self. __next__(self ) Advance the iterator and either return the next object from the iterable container or raise an StopIteration exception.
26.6. Collection Special Method Names for Iterators and Iterable
329
Important: Python 3 The the Iterator method name of __next__() is focused on Python 3. For Python 2 compatability, you might want to also dened next().
def next( self ): return self.__next__()
__iter__(self ) Additionally an iterator will also provide a denition of the __iter__() special method name. This will simply return the iterator itself (return self). This prevents small problems with redundant calls to the iter() built-in function. Note there are still more features of iterators. The Python Enhancement Proposal (PEP 342) describes some considerably more advanced features of iterators and the yield statement.
26.7 Collection Special Method Names for Sequences

The following collection class denitions introduce special method names to make your class behave like a Sequence (tuple) or a Mutable Sequence (list). class Sequence() A Sequence object (essentially a tuple) is based on three more fundamental denitions: Sized, Iterable and Container. As such, the class must dene a number of methods including __contains__(), __len__(), __iter__(). A Sequence must dene a __getitem__() method. Additionally, methods like __reversed__(), index() and count() are also sensible for this class. The collections.Sequence provides defaults for these methods. Since were talking about an object thats like a tuple or frozenset, there arent any methods for updating or mutating the contents. Note also, that Hashable isnt part of a Sequence. You might want to add Hashable if your Sequence can support a xed hash value, suitable for use as dictionary key. This class provides a default implementation of __iter__() that is based on using a range of values and calling __getitem__(). __getitem__(self, index ) Return the value at the given index in the Sequence. The __getitem__() method function should be prepared for the key to be either a simple integer or a slice object. When called with an integer, it returns an element of the sequence or raises IndexError. A slice is a simple object with three attributes: start, stop and step. When called with a slice object, it returns another Sequence. The following examples show common slice situations. The expression someSequence[1:5] is transformed to someSequence.__getitem__( slice(1,5) ). The slice object has the following attribute values: key.start = 1, key.stop = 5, key.step = None. The expression someSequence[2:8:2] is transformed to someSequence.__getitem__( slice(2,8,2) ). The slice object has the following attribute values: key.start = 2, key.stop = 8, key.step = 2.
330
The expression someSequence[1:3,5:8] is transformed into someSequence.__getitem__( ( slice(1,3), slice(5,8) ) ). The key argument will be a tuple of slice objects. Heres an example of a simple class that behaves like a tuple, with some restrictions.
import collections class Point( collections.Sequence ): def __init__( self, x, y ): self.x= x self.y= y def __len__( self ): return 2 def __contains__( self, key ): return self.x==key or self.y==key def __getitem__( self, position ): if position in ( 0, -2 ): return self.x elif position in ( 1, -1 ): return self.y else: raise IndexError
class MutableSequence() A MutableSequence exteds a Sequence. It requires all of the Sequence methods to be dened. Plus it has some additional methods. To be mutable, it must have a way to update and remove items. The methods __setitem__(), __delitem__(), and insert() must be dened to update a MutableSequence. There are of course, numerous additional methods that are provided by default. Any of the optional methods from Sequence, plus append(), reverse(), extend(), pop(), remove(), and __iadd__(). You can override these denitions if you want to improve their performance. __getitem__(self, index ) Return the value at the given index in the Sequence. __setitem__(self, index, value) Replace the value at the given index in the Sequence __delitem__(self, index, value) Remove the value at the given index in the Sequence. The __getitem__(), __setitem__() and __delitem__() method function should be prepared for the index to be either a simple integer, a slice object, or a tuple of slice objects. insert(self, index, value) Insert a new value before the given index in the Sequence.
26.8 Collection Special Method Names for Sets

The following collection class denitions introduce special method names to make your class behave like a Set (frozenset) or a MutableSet (set). class Set() A Set, like a basic Sequence, is based on three more fundamental denitions: Sized, Iterable, Container. The basic collections.Set is an immutable set; its the basis for the built-in frozenset. A mutable set will build on this denition. As such, the class must dene a number of methods including __contains__(), __len__(), __iter__().
26.8. Collection Special Method Names for Sets
331
Since, generally, a set simply checks for membership, we dont need too much more. The comparison operations (__le__(), __lt__(), __eq__(), __ne__(), __gt__(), __ge__(), __and__(), __or__(), __sub__(), __xor__(), and isdisjoint()) have default denitions. You can override these for performance reasons. class MutableSet() A MutableSet extends a Set. It requires all of the Set methods to be dened. Plus it has some additional methods. To be mutable, it must have the methods add() and discard() to update the MutableSet. The collections.MutableSet is is the basis for the built-in set. The collections.MutableSet provides the following method functions clear(), pop(), remove(). These are based on our supplied add() and discard() Also, the following operators are provided so that a MutableSet can be updated with another set: __ior__(), __iand__(), __ixor__(), and __isub__(). add(self, item) Updates the set to add the item, if it was not already a member. discard(self, item) Updates the set to remove the item, if it was a member. Does nothing if the member was not already in the set.
26.9 Collection Special Method Names for Mappings

The following collection class denitions introduce special method names to make your class behave like a Mapping or a MutableMapping (dict). class Mapping() A Mapping, like a basic Sequence, is based on three more fundamental denitions: Sized, Iterable, Container. The basic collections.Mapping is the denition of an immutable mapping. A mutable mapping (like a dict) will build on this denition. As such, the class must dene a number of methods including __contains__(), __len__(), __iter__(). A Mapping must dene a __getitem__() method. Additionally, default methods will be provided for __contains__(), keys(), items(), values(), get(), __eq__(), and __ne__(). The equality test compares the list created by items() to assure that each item tuple has the same key and value. __getitem__(self, key ) Returns the value corresponding to key, or raises KeyError. class MutableMapping() A MutableMapping exteds a Mapping. It requires all of the Set methods to be dened. Plus it has some additional methods. To be mutable, the methods __setitem__() and __setitem__() must be dened. Also, methods are provided for pop(), popitem(), clear(), update(), and setdefault() __getitem__(self, key ) Returns the value corresponding to key, or raises KeyError. A collections.defaultdict does not raise an exception. For keys that dont exist, this version of a MutableMapping creates a default value and then returns that.
332
__setitem__(self, key, value) Puts an item into the mapping; the item has the given key and value. If the key did not exist it is added. If the key did exist, the previous value is replaced. __delitem__(self, key ) Removes an item from the mapping, or raises a KeyError if the item did not exist. Beyond these two base classes, there are some additional classes that help you to dene a view thats based on a mapping. class KeysView() An object of this class is built from an existing Mapping, and behaves like a Set that contains the keys from the exsting Mapping. The methods from Sized (__len__()) and Set (__contains__() and __iter__()) are dened. class ValuesView() An object of this class is built from an existing Mapping, and behaves like a Set that contains the values from the exsting Mapping. Unlike a proper Set, however, there may appear to be multiple copies of a given value. The methods from Sized (__len__()) and Set (__contains__() and __iter__()) are dened. class ItemsView() An object of this class is built from an existing Mapping, and behaves like a Set that contains a sequence of ( key, value ) tuples from the exsting Mapping. The methods from Sized (__len__()) and Set (__contains__() and __iter__()) are dened. Note that __contains__() checks for the presence of a ( key, value ) 2-tuple.
26.10 Mapping Example

An immutable mapping is a kind of translation from key to value. The mapping is xed when the object is created and cannot be updated. Heres an example of a small class to dene an immutable mapping that were calling a Translation. Note that our immutable mapping happens to have a plain old dict under the hood.
class Translation( collections.Mapping ): def __init__( self, ** kw ): self._map= kw def __len__( self ): return len(self._map) def __contains__( self, key ): return key in self._map def __iter__( self ): return iter(self._map) def __getitem__( self, key ): return self._map[key]
Heres a transcript of using our Translation class to create an object that translates some names to numeric values.
>>> c = Translation( red=0, green=1, blue=2 ) >>> c['red'] 0
26.10. Mapping Example
333
>>> c['white'] Traceback (most recent call last): File "<stdin>", line 1, in <module> File "<stdin>", line 11, in __getitem__ KeyError: 'white' >>> c['black']= 3 Traceback (most recent call last): File "<stdin>", line 1, in <module> TypeError: 'Translation' object does not support item assignment >>> for nm in c: ... print nm ... blue green red
26.11 Iterator Examples

The built-in sequence types (list, tuple, string) all produce iterator objects for use by the for statement. The set, frozenset, dict and file objects also produce an iterator. In addition to dening ordinary generator methods by using the yield statement, your classes can also produce iterator objects. This can make a program slightly simpler to read by assuring that loops are simple, obvious for statements. Easy Iterators. When writing a collection-like class, you can simply write method functions that include the yield statement.
class Deck( object ): def __init__( self ): # Create the self.cards container. def deal( self ): self.shuffle() for c in self.cards: yield c
The deal() method is an iterator. We can use this iterator as follows.

d= Deck() for card in d.deal(): print c
This is the same technique covered in Iterators and Generators, except used with method functions instead of stand-alone functions. Unique Iterator Classes. Generally, an iterator is an object weve designed to help us use a more complex container. Consequently, the container will usually contain a factory method which creates iterators associated with the container. A container will implement the special method __iter__() to emit an iterator properly congured to handle the container. When we evaluate iter( someList ), the object, someList, must return an iterator object ready to be used with a for statement. The way this works is the iter() function evaluates the __iter__() method function of the object, someList. The objects __iter__() method function creates the object to be used as an iterator. Well do something similar in our own classes.
334
In the following example classes, well create a class which wraps a list and provides and a specialized iterator that yields only non-zero values of the collection.
import collections class DataSamples( collections.Iterable, collections.Sized ): def __init__( self, aList=None ): self.values= [] if aList is None or aList def __iter__( self ): return NonZeroIter( self ) def __len__( self ): return len( self.values ) def __getitem__( self, index ): return self.values[index]
1. When we initialize a DataSamples instance, we save any provided sequence of values. This class behaves like a collection. We havent provided all of the methods, however, in order to keep the example short. Clearly, to be list-like, well need to provide an append() method. 2. When we evaluate the iter() function for a DataSamples object, the DataSamples object will create a new, initialized NonZeroIter. Note that we provide the DataSamples object to the new NonZeroIter, this allows the iterator to process the collection properly.
class NonZeroIter( collections.Iterator ): def __init__( self, aDataSamples ): self.ds= aDataSamples self.pos= -1 def __next__( self ): while self.pos+1 != len(self.ds) and self.ds[self.pos+1] == 0: self.pos += 1 if self.pos+1 == len( self.ds ): raise StopIteration self.pos += 1 return self.ds[self.pos] def next( self ): return self.__next__() def __iter__( self ): return self
1. When initialized, the NonZeroIter saves the collection that it works with. It also sets its current state; in this instance, we have pos set to -1, just prior to the element well return. 2. The next() function of the iterator locates the next non-zero value. If there is no next value or no next non-zero value, it raises StopIteration to notify the for statement. Otherwise, it returns the next non-zero value. It updates its state to reect the value just returned. 3. The __iter__() function of the iterator typically returns self. We can make use of this iterator as follows.
ds = DataSamples( [0,1,2,0,3,0] ) for value in ds: print value
The for statement calls iter(ds) implicitly, which calls ds.__iter__(), which creates the NonZeroIter instance. The for statement then calls the next() method of this iterator object to get the non-zero values from the DataSamples object. When the iterator nally raises the StopIteration exception, the for statement nishes normally.
26.11. Iterator Examples
335
26.12 Extending Built-In Classes

We can extend all of Pythons built-in classes. This allows us to add or modify features of the data types that come with Python. This may save us from having to build a program from scratch. Heres a quick example of a class which extends the built-in list class by adding a new method.
>>> ... ... ... >>> >>> [1, >>> 8 >>> >>> 13 class SumList( list ): def sum( self ): return sum( self ) x= SumList( [1, 3, 4] ) x 3, 4] x.sum() x.append( 5 ) x.sum()
Clearly, we can extend a class like list with additiona; methods for various statistical calculations. Each function can be added as easily as the SumList.sum() function in the above example. Heres an exmaple of extending a dictionary with simple statistical functions. We based on dictionary on collections.defaultdict because it makes it very simple to create a frequency table. See Default Dictionaries for more information on default dictionaries.
>>> >>> ... ... ... ... ... >>> >>> 13 >>> >>> 18 >>> >>> 23 from collections import defaultdict class StatDict( defaultdict ): def __init__( self, valueList=None ): super( StatDict, self ).__init__( int, valueList ) def sum( self ): return sum( k*self[k] for k in self ) x = StatDict( [ (2,1), (3,1), (4,2) ] ) x.sum() x[5] += 1 x.sum() x[5] += 1 x.sum()
Each time we increment the frequency of a numeric value, the defaultdict will add the integer count automatically.
26.13 Special Method Name Exercises

26.13.1 Geometric Points
A 2-dimensional point is a coordinate pair, an x and y value. If we limit the range to the range 0 to 2**16, we can do a few extra operations quickly.
336
Develop the basic routines for __init__(), __repr__(), __str__(). The __hash__() function can simply combine x and y via x<<16+y. Develop a test routine that creates a sequence of points. Also, be sure to develop a test that uses points as keys in a dictionary.
26.13.2 Rational Numbers

Finish the Rational number class by adding all of the required special methods. The Rational Numbers exercise in Classes describes the basic structure of a class to handle rational math, where every number is represented as a fraction.
26.13.3 Currency and the Cash Drawer

Currency comes in denominations. For instance, US currency comes in $100, $50, $20, $10, $5, $1, $.50, $.25, $.10, $.05, and $.01 denominations. Parker Brothers Monopolygame has currency in 500, 100, 50, 20, 10, 5 and 1. Prior to 1971, English currency had 50, 20, 10, 5, 1, shillings (1/12 of a pound) and pence (1/20 of a shilling). An amount of money can be represented as an appropriate tuple of integers, each of which represents the specic numbers of each denomination. For instance, one representation for $12.98 in US currency is (0, 0, 0, 1, 0, 2, 0, 3, 2, 0, 3). Each subclass of Currency has a specic mix of denominations. We might dene subclasses for US currency, Monopoly currency or old English currency. These classes would dier in the list of currencies. An object of class currency would be created with a specic mix of denominatioons. The superclass should include operations to add and subtract Currency objects. An __iadd__( currency )() method, for example would add the denominations in currency to this objects various denominations. An __isub__( currency )() method, for example would subtract the denominations in currency to this objects various denominations; in the event of attempting to subtract more than is available, the object would raise an exception. Be sure to dene the various conversions to oat, int and long so that the total value of the collection of bills and coins can be reported easily. An interesting problem is to translate a decimal amount into appropriate currency. Note that numbers like 0.10 dont have a precise oating-point representation; oating point numbers are based on powers of 2, and 0.10 can only be approximated by a nite-precision binary fraction. For US currency, its best to work in pennies, representing $1.00 as 100. Develop a method which will translate a given target amount, t, into an appropriate mixture of currency denominations. In this case, we can iterate through the denominations from largest to smallest, determining the largest quantity, q of a denomination, d, such that q d t. This version doesnt depend on the current value of the Currency object. More Adanced Solution. A more advanced version is to create a Currency object with a given value; this would represent the money in a cash drawer, for example. A method of this object would make an amount of money from only the available currency in the cash drawer, or raise an exception if it could not be done. In this case, we iterate through the denominations, d, from largest to smallest, determining the largest quantity, q, such that q d t, consistent with available money in the cash drawer. If we dont have enough of a given denomination, it means that we will be using more of the smaller denominations. One basic test case is to create a currency object with a large amount of money available for making change. In the following example, we create a cash drawer with $804.55. We accept a payment of $10 as 1 $5, 4 $1, 3 $.25, 2 $.10 and 1 $.05. Then we accept a payment of $20, for a bill of $15.24, meaning we need to pay out $4.76 in change. 26.13. Special Method Name Exercises 337
drawer = USCurrency( (5,2,6,5,5,5,5,5,5,5,5) ) drawer += USCurrency((0,0,0,0,1,4,0,3,2,1,0)) drawer += USCurrency((0,0,1,0,0,0,0,0,0,0,0)) drawer.payMoney( 4.76 )
Interestingly, if you have $186.91 (one of each bill and coin) you can nd it almost impossible to make change. Confronted with impossible situations, this class should raise an UnableToMakeChange exception.
26.13.4 Sequences with Statistical Methods

Create a sequence class, StatSeq that can hold a sequence of data values. This class must be a subclass of collections.MutableSequence and dene all of the usual sequence operators, including __add__(), __radd__(), __iadd__(), but not __mul__(). The __init__() function should accept a sequence to initialize the collection. The various __add__() functions should append the values from a StatSeq instance as well as from any subclass of collections.Sequence (which includes list and tuple.) Most importantly, this class shold dene the usual statistical functions like mean and standard deviation, described in the exercises after Tuples, Sequence Processing Functions: map(), lter() and reduce() and Sample Class with Statistical Methods in Classes. Since this class can be used everywhere a sequence is used, interface should match that of built-in sequences, but extra features are now readily available. For a test, something like the following should be used:
import random samples = StatSeq( [ random.randrange(6) for i in range(100) ] ) print samples.mean() s2= StatSeq() for i in range(100): ss.append( random.randrange(6) ) # Also allow s2 += [ random.randrange(6) ] print s2.mean()
There are two approaches to this, both of which have pros and cons. Dene a subclass of list with a few additional methods. This will be dened as class StatSeq( list ):. Dene a new class (a subclass of object) that contains an internal list, and provides all of the sequence special methods. Some (like append) will be delegated to the internal list object. Others (like mean) will be performed by the StatSeq class itself. This will be dened as class StatSeq( object ):. Note that the value of mean() does not have to be computed when it is requested. It is possible to simply track the changing sum of the sequence and length of the sequence during changes to the values of the sequence. The sum and length are both set to zero by __init__(). The sum and length are incremented by every __add__(), append(), insert(). They are decremented by pop(), remove(), and and __delitem__(). Finally, theres a two-part change for __setitem__(): the old value is deducted and the new value is added. This way the calculation of mean is simply a division operation. Keeping track of sums and counts can also optimize mode and standard deviation. A similar optimization for median is particularly interesting, as it requires that the sample data is retained by this class in sorted order. This means that each insert must preserve the sorted data set so that the median value can be retrieved without rst sorting the entire sequence of data. You can use the bisect module to do this.
338
There are a number of algorithms for maintaining the data set in sorted order. You can refer to Knuths The Art of Computer Programming [Knuth73], and Cormen, Leiserson, Rivest Introduction to Algorithms [Cormen90] which cover this topic completely.
26.13.5 Chessboard Locations

A chessboard can be thought of as a mapping from location names to pieces. There are two common indexing schemes from chessboards: algebraic and descriptive. In algebraic notation the locations have a rank-le address of a number and a letter. In descriptive notation the le is given by the starting pieces le, rank and players color. See Chess Game Notation for an extension to this exercise. The algebraic description of the chess board has les from a-h going from whites left to right. It has ranks from 1-8 going from whites side (1) to blacks side (8). Boards are almost always shown with position a1 in the lower left-hand corner and h8 in the upper right, white starts at the bottom of the picture and black starts at the top. In addition to the simplied algebraic notation, there is also a descriptive notation, which reects each players unique point of view. The descriptive board has a queens side (whites left, les a-d) and a kings side (whites right, les e-h). Each rank is numbered starting from the player. White has ranks 1-8 going from white to black. Black, at the same time as ranks 1-8 going back toward white. Each of the 64 spaces on the board has two names, one from whites point of view and one from blacks. Translation from descriptive to algebraic is straight-forward. Given the players color and a descriptive location, it can be translated to an algebraic location. The les translate through a relatively simple lookup to transform QR to a, QKt to b, QB to c, Q to d, K to e, KB to f, KKt to g, KR to h. The ranks translate through a simple calculation: whites ranks are already in algebraic notation; for blacks rank of r, 9 r is the algebraic location. Create a class to represent a chess board. Youll need to support the special function names to make this a kind of mapping. The __getitem__() function will locate the contents of a space on the board. The __setitem__() function will place a piece at a space on the board. If the key to either function is algebraic (2 characters, lower case le from a-h and digit rank from 1-8), locate the position on the board. If the key is not algebraic, it should be translated to algebraic. The codes for pieces include the piece name and color. Piece names are traditionally p or nothing for pawns, R for rook, N for knight, B for bishop, Q for queen and K for king. Pawns would be simply the color code w or b. Other pieces would have two-character names: Rb for a black rook, Qw for the white queen. The __init__() method should set the board in the standard starting position:
26.13. Special Method Name Exercises
339
Table 26.1: Chess Starting Position piece white rooks white knights white bishop white queen white king white pawns black rooks black knights black bishops black queen black king black pawns algebraic a1, h1 b1, g1 c1, f1 d1 e1 a2-h2 a8, h8 b8, g8 c8, f8 d8 e8 a7-a7 descriptive wQR1, wKR1 wQKt1, wKKt1 wQB1, wKB wQ1 wK1 wQR2-wKR2 bQR1, bKR1 bQKt1, bKKt1 bQB1, bKB1 bQ1 bK1 bQR2-bKR2 piece code Rw Nw Bw Qw Kw w Rb Nb Bb Qb Kb b
Heres a sample ve-turn game. It includes a full description of each move, and includes the abbreviated chess game notation. 1. white pawn from e2 to e4; K2 to K5 black pawn from e7 to e5; K2 to K5 2. white knight from g1 to f3; KKt1 to KB3 black pawn from d7 to d6; Q2 to Q3 3. white pawn from d2 to d4; Q2 to Q4 black bishop from c8 to g4; QB1 to KKt5 4. white pawn at d4 takes pawn at e5; Q4 to K5 black bishop at g4 takes knight at f3; KKt5 to KB6 5. white Q at d1 takes bishop at f3; Q1 to KB3 black pawn at d6 takes e5; Q3 to K4 The main program should be able to place and remove pieces with something like the following:
chess= Board() # move pawn from white King 2 to King 5 chess['wK5']= chess['wK2']; chess['wK2']= '' # move pawn from black King 2 to King 5 chess['bK5']= chess['bK2']; chess['bK2']= '' # algebraic notation to print the board for rank in [ '8', '7', '6', '5', '4', '3', '2', '1']: for file in [ 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h']: print "%5s" % board[file+rank], print
The algebraic output can be changed to the following, which some people nd simpler.
for rank in ('8','7','6','5','4','3','2','1'): print "".join( [ "%5s" % board[file+rank] for file in ('a','b','c','d','e','f','g','h') ] )
You should also write a move() function to simplify creating the test game. A move typically consists of the piece name, the from position, the to position, plus optional notes regarding check and pawn promotions. 340 Chapter 26. Creating or Extending Data Types
26.13.6 Relative Positions on a Chess Board

When decoding a log of a chess game in Short Algebraic Notation (SAN), it is often necessary to search for a piece that made a given move. Well look at this problem in detail in Chess Game Notation. There are actually a number of search algorithms, each constrained by the rules for moving a particular piece. For example, the knight makes a short L-shaped move and there are only 8 positions on the board from which a knight can start to end up at a given spot. The queen, on the other hand, moves horizontally, vertically or diagonally any distance, and there are as many as 24 starting positions for the queen to end up at a given spot. This search is simplied by having iterators that know a few rules of chess and an give us a sequence of appropriate rank and le values. Wed like to be able to say something like the following.
piece, move, toPos = ( "Q", "x", "f3" ) for fromPos in aBoard.queenIter( toPos ): if aBoard[fromPos] == 'Q': print "Queen from", fromPos, "takes", aBoard[toPos], "at", toPos
Well review a few chess denitions for this problem. You can also see Chessboard Locations in Collection Special Method Names for some additional background. The algebraic description of the chess board has les from a-h going from whites left to right. It has ranks from 1-8 going from whites side (1) to blacks side (8). Boards are almost always shown with position a1 in the lower left-hand corner and h8 in the upper right, white starts at the bottom of the picture and black starts at the top. We need the following collection of special-purpose iterators. The kingIter() method has to enumerate the eight positions that surround the king. The queenIter() method has to enumerate all the positions in the same rank, the same le, and on the diagonals. Each of these must be examined from the queens position moving toward the edge of the board. This search from the queen outward allows us to locate blocking pieces that would prevent the queen from making a particular move. The bishopIter() method has to enumerate all the positions on the diagonals. Each of these must be examined from the bishops position moving toward the edge of the board. The knightIter() method has to enumerate the eight positions that surround the knight, reecting the knights peculiar move of 2 spaces on one axis and 1 space on the other axis. There are four combinations of two ranks and one le and four more combinations of two les and one rank from the ending position. As with the king, no piece can block a knights move, so order doesnt matter. The rookIter() method has to enumerate all the positions in the same rank and the same le. Each of these must be examined from the rooks position moving toward the edge of the board. The pawnIter() method has to enumerate a fairly complex set of positions. Most pawn moves are limited to going forward one rank in the same le. Since we need to know which direction is forward, we need to know the color of the pawn. For white pawns, forward means the ranks increase from 2 to 8. For black pawns, then, forward means the ranks decrease from 7 down to 1. Pawn captures involve going forward one rank in an adjacent le. Further complicating the analysis is the ability for a pawns rst move to be two ranks instead of one. We note that the queens iterator is really a combination of the bishop and the rook. Well look at the rooks iterator, because it is can be adapted to be a bishop iterator, and then those two combined to create the queen iterator.
26.13. Special Method Name Exercises
341
Given a starting position with a rank of r and a le of f, well need to examine all ranks starting from r and moving toward the edge of the board. These are r 1, r + 1, r 2, r + 2, r 3, r + 3, .... Similarly, we need to examine all of the les starting from f and moving toward the edge of the board. These are f 1, f + 1, f 2, f + 2, f 3, f + 3, .... Before doing an comparison, we need to lter the le and rank combinations to assure that they are legal positions. Additionally, we need to stop looking when weve encountered a piece of our own color or an opposing piece that isnt the one were searching for. These intervnening pieces block the intentended move.
342
CHAPTER
TWENTYSEVEN
ATTRIBUTES, PROPERTIES AND DESCRIPTORS

When we reference an attribute of an object with something like someObject.someAttr, Python uses several special methods to get the someAttr attribute of the object. In the simplest case, attributes are simply instance variables. But thats not the whole story. To see how we can control the meaning of attributes, we have to emphasize a distinction: An attribute is a name that appears after an object name. This is the syntactic construct. For example, someObj.name. An instance variable is an item in the internal __dict__ of an object. The default semantics of an attribute reference is to provide access to the instance variable. When we say someObj.name, the default behavior is eectively someObj.__dict__['name']. This is not the only meaning an attribute name can have. In Semantics of Attributes well look at the various ways we can control what an attribute name means. The easiest technique for controlling the meaning of an attribute name is to dene a property. Well look at this in Properties. The most sophisticated technique is to dene a descriptor. Well look at this in Descriptors. The most exible technique is to override the special method names and take direct control over attribute processing. Well look at this in Attribute Access Exercises.
27.1 Semantics of Attributes

Fundamentally, an object encapsulates data and processing via its instance variables and method functions. Because of this encapsulation, we can think of a class denition as providing both an interface denition and an implementation that supports the dened interface. In some languages, the attributes are the instance variables; an attribute can name expose the private state of the object. Consequently, some languages suggest that attributes should not be part of the interface. Further, some languages (C, C#, Java are examples) provide syntax to distinguish the public interface from the private implementation. In Python, this distinction between interface and implementation is not heavily emphasized in the syntax, since it can often lead to wordy, complex programs. Most well-designed classes, however, tend to have a set of interface methods that form the interface for collaborating with objects of that class.
343
In Python, the method names and instance variables which begin with _ are treated as part of the private implementation of the class. These names arent shown in the help() function, for example. The remaining elements (without a leading _) form the public interface. Encapsulation There are two commonly-used design patterns for encapsulating an objects instance variables. Getters and Setters. This design pattern can insulate each instance variable with some method functions to get and set the value of that instance variable. When we do this, each access to an attribute of the object is via an explicit function: anObject.setPrice( someValue ) and anObject.getPrice(). Attributes. We can insulate an instance variable with more sophisticated attribute access. In the case of Python, we have several techniques for handling attribute access. Python Attributes. In Python, the attributes do not have to be instance variables. Properties. We can bind getter, setter (and deleter) functions with an attribute name, using the builtin property() function or @property decorator. When we do this, each reference to an attribute has the syntax of direct access to an instance variable, but it invokes the given method function. Well look at this in Properties. Descriptors. We can implement getter, setter (and deleter) functions into a separate descriptor class. We then create an instance of this class as the object named by an attribute. When we do this, each reference to an attribute has the syntax of direct access to an instance variable, but it invokes a method function of the Descriptor object. Well look at this in Descriptors. Special Method Names. We can override denitions for several special method names. There are several methods which plug into the standard algorithm. A fourth method, __getattribute__(), allows you to change attribute access in a fundamental way. Well look at this in Attribute Access Exercises. Warning: Caution Changing attribute access in radical ways can interfere with how people understand the operation of your classes and objects. The default assumption is that an attribute is an instance variable. While we can fundamentally alter the meaning of a Python attribute, we need to be cautious about violating the default assumptions of people reading our software.
27.2 Properties
The property() function gives us a handy way to implement an attribute that is more than a simple reference to an instance variable. Through the property function, we can assign a simple attribute name to parameter-less getter and setter method functions. This allows us to create programs that look like the following example.
>>> oven= Temperature() >>> oven.farenheit= 450 >>> oven.celsius 232.22222222222223 >>> oven.celsius= 175
344
Chapter 27. Attributes, Properties and Descriptors
>>> oven.farenheit 347.0
In this example, we set one attribute and the value of another attribute changes to mirror it precisely. We can do this by dening some method functions and binding them to attribute names. Property Design Pattern. The Property design pattern has a number of method functions which are bound together with a single property name. The method functions can include any combination of a getter, a setter and a deleter. To create a property, we dene the instance variable and one or more method functions. This is identical with the Getter and Setter design pattern. To make a property, we provide these method functions to the property() function to bind the various methods to an attribute name. Heres the denition of the property() function. property(fget, [fset, fdel, doc] ) This binds the given method functions into a property denition. Usually the result value is assigned to an attribute of a class. The fget parameter must be a getter function, of the form function(self)->value. The fset parameter must be a setter function, of the form function(self,value). The fdel parameter must be a deleter function, of the form function(self). This can be used to delete the attribute, and is evaluated in response to del object.attribute. The doc parameter becomes the docstring for the attribute. If this is not provided the docstring from the getter function becomes the docstring for the property.
class SomeClass( object ): def getThis( self ): return self._hidden_variable * 2 def setThis( self, value ): self._hidden_variable = float(value) / 2 this= property( getThis, setThis )
This creates a property, named this: dened by two method functions getThis() and setThis(). The functions do a fairly silly calculation, but this shows how an attribute can embody a calculation. The property function can also be used as a decorator. Well look at decorators in detail in Decorators. Here is a quick sample of the using the the property() with the decorato syntax instead of the function syntax.
class SomeClass( object ): @property def this( self ): return self._hidden_variable * 2 @this.setter def this( self, value ): self._hidden_variable = float(value) / 2
Property Example. The following example shows a class denition with four method functions that are used to dene two properties.
27.2. Properties
345
property.py
class Temperature( object ): def fget( self ): return self.celsius * 9 / 5 + 32 def fset( self, value ): self.celsius= (float(value)-32) * 5 / 9 farenheit= property( fget, fset ) def cset( self, value ): self.cTemp= float(value) def cget( self ): return self.cTemp celsius= property( cget, cset, doc="Celsius temperature" )
1. We create the farenheit property from the fget() and fset() method functions. When we use the farenheit attribute on the left side of an assignment statement, Python will use the setter method. When we use this attribute in an expression, Python will use the getter method. We dont show a deleter method; it would be used when the attribute is used in a del statement. 2. We create the celsius property from the cget() and cset() method functions. When we use the celsius attribute on the left side of an assignment statement, Python will use the setter method. When we use this attribute in an expression, Python will use the getter method. The doc string provided for the celsius attribute is available as Temperature.celsius.__doc__.
27.3 Descriptors
A Descriptor is a class which provides the detailed get, set and delete control over an attribute of another object. This allows you to dene attributes which are fairly complex objects in their own right. The idea is that we can use simple attribute references in a program, but those simple references are actually method functions of a descriptor object. This allows us to create programs that look like the following example.
>>> oven= Temperature() >>> oven.farenheit= 450 >>> oven.celsius 232.22222222222223 >>> oven.celsius= 175 >>> oven.farenheit 347.0
In this example, we set one attribute and the value of another attribute changes to mirror it precisely. A common use for descriptors is in an object-oriented database (or an object-relational mapping). In a database context, getting an attribute value may require fetching data objects from the le system; this may involve creating and executing a query in a database. This kind of related-item fetch operation will be shared widely among the persistent classes in an application. Rather than attempt to manage this shared functionality via inheritance of method functions, its simpler to split it into a separate Descriptor class and use this descriptor to manage the access of related objects. Descriptor Design Pattern. The Descriptor design pattern has two parts: an Owner and an attribute Descriptor. The Owner is usually a relatively complex object that uses one or more Descriptors for its
346
attributes. A Descriptor class denes get, set and delete methods for a class of attributes of the Owner. Each Descriptor object manages a specic attribute. Note that Descriptors will tend to be reusable, generic classes of attributes. The Owner can have multiple instances of each Descriptor class to manage attributes with similar behaviors. Each Descriptor object is a unique instance of a Descriptor class, bound to a distinct attribute name when the Owner class is dened. To be recognized as a Descriptor, a class must implement some combination of the following three methods. __get__(self, instance, owner ) The instance parameter is the self variable of object being accessed. The owner parameter is the owning class object. This method of the descriptor must return this attributes value. If this descriptor implements a class level variable, the instance parameter can be ignored; the instance is irrelevant and may be None. The owner parameter refers to the class. __set__(self, instance, value) The instance argument is the self variable of the object being updated. This method of the descriptor must set this attributes value. __delete__(self, instance) The instance argument is the self variable of the object being accessed. This method of the descriptor must delete this attributes value. Sometimes, a Descriptor class will also need an __init__() method function to initialize the descriptors internal state. Less commonly, the descriptor may also need __str__() or __repr__() method functions to display the instance variable correctly. You must also make a design decision when dening a descriptor. You must determine where the underlying instance variable is contained. You have two choices. The Descriptor object has the instance variable. In this case, the descriptor objects self variable has the instance variable. This is common for attributes that can be updated. The Owner object contains the instance variable. In this case, the descriptor object must use the instance parameter to reference a value in the owning object. This is common for attributes that are get-only. Descriptor Example. Heres a simple example of an object with two attributes dened by descriptors. One descriptor (Celsius) contains its own value. The other desriptor (Farenheit), depends on the Celsius value, showing how attributes can be linked so that a change to one directly changes the other.
descriptor.py
class Celsius( object ): """Fundamental Temperature Descriptor.""" def __init__( self, value=0.0 ): self.value= float(value) def __get__( self, instance, owner ): return self.value def __set__( self, instance, value ): self.value= float(value) class Farenheit( object ): """Requires that the owner have a ``celsius`` attribute."""
27.3. Descriptors
347
def __get__( self, instance, owner ): return instance.celsius * 9 / 5 + 32 def __set__( self, instance, value ): instance.celsius= (float(value)-32) * 5 / 9 class Temperature( object ): celsius= Celsius() farenheit= Farenheit()
1. Weve dened a Celsius descriptor. The Celsius descriptor has an __init__() method which denes the attributes value. The Celsius descriptor implements the __get__() method to return the current value of the attribute, and a __set__() method to change the value of this attribute. 2. The Farenheit descriptor implements a number of conversions based on the value of the celsius attribute. The __get__() method converts the internal value from Celsius to Farenheit. The __set__() method converts the supplied value (in Farenheit) to Celsius. In a way, the Farenheit descriptor is an observer of the owning objectss celsius attribute. 3. The owner class, Temperature has two attributes, both of which are managed by descriptors. One attribute, celsius, uses an instance of the Celsius descriptor. The other attribute, farenheit, uses an instance of the Fareheit descriptor. When we use one of these attributes in an assignment statement, the descriptors __set__() method is used. When we use one of these attributes in an expression, the descriptors __get__() method is used. We didnt show a __delete__() method; this would be used when the attribute is used in a del statement. Lets look at what happens when we set an attribute value, for example, using oven.farenheit= 450 . In this case, the farenheit attribute is a Descriptor with a __set__() method. This __set__() method is evaluated with instance set to the object which is being modied (the oven variable) and owner set to the Temperature class. The __set__() method computes the celsius value, and provides that to the celsius attribute of the instance. The Celsius descriptor simply saves the value. When we get an attribute value, for example, using oven.celsius, the following happens. Since celsius is a Descriptor with a __get__() method, this method is evaluated, and returns the celsius temperature.
27.4 Attribute Handling Special Method Names

Fundamentally, attribute access works through a few special method names. Python has a default approach: it checks the object for an instance variable that has the attributes name before using these attribute handling methods. Because Python uses these methods when an attribute is not already an instance variable, you can easily create innite recursion. This can happen if you write self.someAttr inside the body of a __getattr__() or __setattr__() method and the attribute is not in the objects __dict__, Python will use the __getattr__() or __setattr__() method to resolve the name. Oops. Within __getattr__() and __setattr__(), you have to use the internal __dict__ explicitly. These are the low-level attribute access methods. __getattr__(self, name) Returns a value for an attibute when the name is not an instance attribute nor is it found in any of the parent classes. name is the attribute name. This method returns the attribute value or raises an AttributeError exception. __setattr__(self, name, value) Assigns a value to an attribute. name is the attribute name, value is the value to assign to it.
348
Note that if you naively do self.name= value in this method, you will have an innite recursion of __setattr__() calls. To access the internal dictionary of attributes, __dict__, you have to use the following: self.__dict__[name] = value. __delattr__(self, name) Delete the named attribute from the object. name is the attribute name. __getattribute__(self, name) Low-level access to a named attribute. If you provide this, it replaces the default approach of searching for an attribute and then using __getattr__() if the named attribute isnt an instance variable of the class. If you want to extend the default approach, you must explicitly evaluate the superclass __getattribute__() method with super( Class,self ).__getattribute__( name ). This only works for classes which are derived from object.
27.5 Attribute Access Exercises

1. Rework Previous Exercises. Refer to exercises for previous chapters (Class Denition Exercises, Advanced Class Denition Exercises, Design Pattern Exercises, Special Method Name Exercises). Rework these exercises to manage attributes with getters and setters. Use the property function to bind a pair of getter and setter functions to an attribute name. The following examples show the before and after of this kind of transformation.
class SomeClass( object ): def __init__( self, someValue ): self.myValue= someValue
When we introduce the getter and setter method functions, we should also rename the original attribute to make it private. When we dene the property, we can use the original attributes name. In eect, this set of transformations leaves the class interface unchanged. We have added the ability to do additional processing around attribute get and set operations.
class SomeClass( object ): def __init__( self, someValue ): self._myValue= someValue def getMyValue( self ): return self._myValue def setMyvalue( self, someValue ): self._myValue= someValue myValue= property( getMyValue, setMyValue )
The class interface should not change when you replace an attibute with a property. The original unit tests should still work perfectly. 2. Rework Previous Exercises. Refer to exercises for previous chapters (Class Denition Exercises, Advanced Class Denition Exercises, Design Pattern Exercises, Special Method Name Exercises). Rework these exercises to manage attributes with Descriptors. Dene a Descriptor class with __get__() and __set__() methods for an attribute. Replace the attribute with an instance of the Descriptor. When we introduce a descriptor, our class should look something like the following.
27.5. Attribute Access Exercises
349
class ValueDescr( object ): def __set__( self, instance, value ): instance.value= value def __get__( self, instance, owner ): return instance.value class SomeClass( object ): def __init__( self, someValue ): self.myValue= ValueDescr()
The class interface should not change when you replace an attibute with a descriptor. The original unit tests should still work perfectly. 3. Tradeos and Design Decisions. What is the advantage of Pythons preference for referring to attributes directly instead of through getter and setter method functions? What is the advantage of having an attribute bound to a property or descriptor instead of an instance variable? What are the potential problems with the indirection created by properties or descriptors?
350
CHAPTER
TWENTYEIGHT
DECORATORS
In addition to object-oriented programming, Python also supports an approach called Aspect-Oriented Programming . Object-oriented programming focuses on structure and behavior of individual objects. Aspectoriented programming renes object design techniques by dening aspects which are common across a number of classes or methods. The focus of aspect-oriented programming is consistency. Toward this end Python allows us to dene decorators which we can apply to class denitions and method denitions and create consistency. We have to note that decorators can easily be overused. The issue is to strike a balance between the obvious programming in the class denition and the not-obvious programming in the decorator. Generally, decorators should be transparently simple and so obvious that they hardly bear explanation. Well look at what a decorator is in Semantics of Decorators. Well look at some built-in decorators in Built-in Decorators. In Dening Decorators well look at dening our own decorators. It is possible to create some rather sophisticated decorators. Well look at the issues surrounding this in Dening Complex Decorators.
28.1 Semantics of Decorators

Essentially, a decorator is a function that is applied to another function. The purpose of a decorator is to transform the function denition we wrote (the argument function) into another (more complex) function denition. When Python applies a decorator to a function denition, a new function object is returned by the decorator. The idea of decorators is to allow us to factor out some common aspects of several functions or method functions. We can then write a simpler form of each function and have the common aspect inserted into the function by the decorator. When we say
@theDecorator def someFunction( anArg ): pass # some function body
We are doing the following: 1. We dened an argument function, someFunction().
351
2. Python applied the decorator function, theDecorator(), to our argument function. The decorator function will return a value; this should be some kind of callable object, either a class with a __call__() method or a function. 3. Python binds the result of the decorator evaluation to the original function name, someFunction(). In eect, we have a more sophisticated version of someFunction() created for us by the theDecorator() function. Cross-Cutting Concerns. The aspects that makes sense for decorators are aspects that are truly common. These are sometimes called cross-cutting concerns because they cut across multiple functions or multiple classes. Generally, decorators fall into a number of common categories. Simplifying Class Denitions. One common need is to create a method function which applies to the class-level attributes, not the instance variables of an object. For information on class-level variables, see Class Variables. The @staticmethod decorator helps us build method functions that apply to the class, not a specic object. See Static Methods and Class Method . Additionally, we may want to create a class function which applies to the class as a whole. To declare this kind of method function, the built-in @classmethod decorator can be used. If you look at the Python Wiki page for decorators (http://wiki.python.org/moin/PythonDecoratorLibrary), you can nd several examples of decorators that help dene properties for managing attributes. Debugging. There are several popular decorators to help with debugging. Decorators can be used to automatically log function arguments, function entrance and exit. The idea is that the decorator wraps your method function with additional statements to record details of the method function. One of the more interesting uses for decorators is to introduce some elements of type safety into Python. The Python Wiki page shows decorators which can provide some type checking for method functions where this is essential. Additionally, Python borrows the concept of deprecation from Java. A deprecated function is one that will be removed in a future version of the module, class or framework. We can dene a decorator that uses the Python warnings module to create warning messages when the deprecated function is used. Handling Database Transactions. In some frameworks, like Django (http://www.djangoproject.org), decorators are used to simplify denition of database transactions. Rather than write explicit statements to begin and end a transaction, you can provide a decorator which wraps your method function with the necessary additional processing. Authorization. Web Security stands on several legs; two of those legs are authentication and authorization. Authentication is a serious problem involving transmission and validation of usernames and passwords or other credentials. Its beyond the scope of this book. Once we know who the user is, the next question is what are they authorized to do? Decorators are commonly used web frameworks to specify the authorization required for each function.
28.2 Built-in Decorators

Python has a few built-in decorators. staticmethod(function) The @staticmethod decorator modies a method function so that it does not use any self variable. The method function will not have access to a specic instance of the class.
352
Chapter 28. Decorators
This kind of method is part of a class, but can only be used when qualied by the class name or an instance variable. For an example of a static method, see Static Methods and Class Method . classmethod(function) The @classmethod decorator modies a method function so that it receives the class object as the rst parameter instead of an instance of the class. This method function wil have access to the class object itself. property(fget, [fset, fdel, doc] ) The @property decorator modies from one to three method functions to be a properties of the class. The returned method functions invokes the given getter, setter and/or deleter functions when the attribute is referenced. Heres a contrived example of using introspection to display some features of a objects class.
introspection.py
import types class SelfDocumenting( object ): @classmethod def getMethods( aClass ): return [ (n,v.__doc__) for n,v in aClass.__dict__.items() if type(v) == types.FunctionType ] def help( self ): """Part of the self-documenting framework""" print self.getMethods() class SomeClass( SelfDocumenting ): attr= "Some class Value" def __init__( self ): """Create a new Instance""" self.instVar= "some instance value" def __str__( self ): """Display an instance""" return "%s %s" % ( self.attr, self.instVar )
1. We import the types module to help us distinguish among the various elements of a class denition. 2. We dene a superclass that includes two methods. The classmethod, getMethods(), introspects a class, looking for the method functions. The ordinary instance method, help(), uses the introspection to print a list of functions dened by a class. 3. We use the @classmethod decorator to modify the getMethods() function. Making the getMethods() into a class method means that the rst argument will be the class object itself, not an instance. 4. Every subclass of SelfDocumenting can print a list of method functions using a help() method. Heres an example of creating a class and calling the help method we dened. The result of the getMethods() method function is a list of tuples with method function names and docstrings.
>>> ac= SomeClass() >>> ac.help() [('__str__', 'Display an instance'), ('__init__', 'Create a new Instance')]
28.2. Built-in Decorators
353
28.3 Dening Decorators

A decorator is a function which accepts a function and returns a new function. Since its a function, we must provide three pieces of information: the name of the decorator, a parameter, and a suite of statements that creates and returns the resulting function. The suite of statements in a decorator will generally include a function def statement to create the new function and a return statement. A common alternative is to include a class denition statement . If a class denition is used, that class must dene a callable object by including a denition for the __call__() method and (usually) being a subclass of collections.Callable. There are two kinds of decorators, decorators without arguments and decorators with arguments. In the rst case, the operation of the decorator is very simple. In the case where the decorator accepts areguments the denition of the decorator is rather obscure, well return to this in Dening Complex Decorators. A simple decorator has the following outline:
def myDecorator( argumentFunction ): def resultFunction( \*args, \*\*keywords ): enhanced processing including a call to argumentFunction resultFunction.__doc__= argumentFunction.__doc__ return resultFunction
In some cases, we may replace the result function denition with a result class denition to create a callable class. Heres a simple decorator that we can use for debugging. This will log function entry, exit and exceptions.
trace.py
def trace( aFunc ): """Trace entry, exit and exceptions.""" def loggedFunc( \*args, \*\*kw ): print "enter", aFunc.__name__ try: result= aFunc( \*args, \*\*kw ) except Exception, e: print "exception", aFunc.__name__, e raise print "exit", aFunc.__name__ return result loggedFunc.__name__= aFunc.__name__ loggedFunc.__doc__= aFunc.__doc__ return loggedFunc
1. The result function, loggedFunc(), is built when the decorator executes. This creates a fresh, new function for each use of the decorator. 2. Within the result function, we evaluate the original function. Note that we simply pass the argument values from the evaluation of the result function to the original function. 3. We move the original functions docstring and name to the result function. This assures us that the result function looks like the original function. Heres a class which uses our @trace decorator.
354
trace_client.py
class MyClass( object ): @trace def __init__( self, someValue ): """Create a MyClass instance.""" self.value= someValue @trace def doSomething( self, anotherValue ): """Update a value.""" self.value += anotherValue
Our class denition includes two traced function denitions. Heres an example of using this class with the traced functions. When we evaulate one of the traced methods it logs the entry and exit events for us. Additionally, our decorated function usees the original method function of the class to do the real work.
>>> mc= MyClass( 23 ) enter __init__ exit __init__ >>> mc.doSomething( 15 ) enter doSomething exit doSomething >>> mc.value 38
28.4 Dening Complex Decorators

A decorator transforms an argument function denition into a result function denition. In addition to a function, we can also provide argument values to a decorator. These more complex decorators involve a two-step dance that creates an intermediate function as well as the nal result function. The rst step evaluates the abstract decorator to create a concrete decorator. The second step applies the concrete decorator to the argument function. This second step is what a simple decorator does. Assume we have some qualied decorator, for example @debug( flag ), where ag can be True to enable debugging and False to disable debugging. Assume we provide the following function denition.
debugOption= True class MyClass( object ): @debug( debugOption ) def someMethod( self, args ): real work
Heres what happens when Python creates the denition of the someMethod() function. 1. Denes the argument function, someMethod(). 2. Evaluate the abstract decorator debug( debugOption ) to create a concrete decorator based on the argument value. 3. Apply the concrete decorator the the argument function, someMethod(). 4. The result of the concrete decorator is the result function, which is given the name someMethod(). Heres an example of one of these more complex decorators. Note that these complex decorators work by creating and return a concrete decorators. Python then applies the concrete decorators to the argument function; this does the work of transforming the argument function to the result function. 28.4. Dening Complex Decorators 355
debug.py
def debug( theSetting ): def concreteDescriptor( aFunc ): if theSetting: def debugFunc( *args, **kw ): print "enter", aFunc.__name__ return aFunc( *args, **kw ) debugFunc.__name__= aFunc.__name__ debugFunc.__doc__= aFunc.__doc__ return debugFunc else: return aFunc return concreteDescriptor
1. This is the concrete decorators, which is created from the argument, theSetting. 2. If theSetting is True, the concrete decorator will create the result function named debugFunc(), which prints a message and then uses the argument function. 3. If theSetting is False, the concrete descriptor will simply return the argument function without any overhead.
28.5 Decorator Exercises

1. Merge the @trace and @debug decorators. Combine the features of the @trace decorator with the parameterization of the @debug decorator. This should create a better @trace decorator which can be enabled or disabled simply. 2. Create a @timing decorator. Similar to the parameterized @debug decorator, the @timing decorator can be turned on or o with a single parameter. This decorator prints a small timing summary.
356
CHAPTER
TWENTYNINE
MANAGING CONTEXTS: THE WITH STATEMENT

Many objects manage resources, and must impose a rigid protocol on use of that resource. For example, a le object in Python acquires and releases OS les, which may be associated with devices or network interfaces. We looked at the way that file objects can be managed by the with statement in File Statements. In this section, well look at ways in which the new with statement will simplify le or database processing. We will look at the kinds of object design considerations which are required to create your own objects that work well with the with statement. Important: Legacy In versions of Python prior to 2.6, we must enable the with statement by using the following statement.
from __future__ import with_statement
29.1 Semantics of a Context

While most use of the with statement involve acquiring and releasing specic resources like OS les the statement can be applied somewhat more generally. To make the statement more widely applicable, Python works with a context. A context is not limited to acquiring and releasing a le or database connection. A context could be a web transaction, a users logged-in session, a particular transaction or any other stateful condition. Generally, a context is a state which must endure for one or more statements, has a specic method for entering the state and has a specic method for exiting the state. Further, a contexts exit must be done with the dened method irrespective of any exceptions that might occur within the context. Database operations often center on transactions which must either be completed (to move the database to a new, iternally consistent state,) or rolled back to reset the database to a prior consistent state. In this case, exceptions must be tolerated so that the database server can be instructed to commit the transaction or roll it back. Well also use a context to be sure that a le is closed, or a lock is released. We can also use a context to be sure that the user interface is reset properly when a user switches their focus or an error occurs in a complex interaction. The design pattern has two elements: a Context Manager and a Working Object. The Context Manager is used by the with statement to enter and exit the context. One thing that can happen when entering a
357
context is that a Working Object is created as part of the entry process. The Working Object is often used for les and databases where we interact with the context. The Working Object isnt always necessary; for example acquiring and releasing locks is done entirely by the Context Manager.
29.2 Using a Context

There are a few Python library classes which provide context information that is used by the with statement. The most commonly-used class is the file class. There are two forms of the with statement. In the rst, the context object does not provide a context-specic object to work with. In the second, the context provides us an object to be used within the context.
with context : suite with context as variable : suite
Well look at the second form, since that is how the file class works. A file object is a kind of context manager, and responds to the protocol dened by the with statement. When we open a le for processing, we are creating a context. When we leave that context, we want to be sure that the le is properly closed. Heres the standard example of how this is used.
with file('someData.txt','r') as theFile: for aLine in theFile: print aLine # the file was closed by the context manager
1. We create the le, which can be used as a context manager. The with statement enters the context, which returns a le object that we can use for input and output purposes. The as clause species that the working object is assigned to theFile. 2. This is a pretty typical for statement that reads each line of a le. 3. The with statement also exits the context, irrespective of the presence or absence of exceptions. In the case of a file context manager, this will close the le. In the previous example, we saw that the le factory function is used to create a context manager. This is possible because a le has several interfaces: it is a context manager as well as being a working le object. This is potentially confusing because it conate le context manager with the working le object. However, it also This has the advantage of making the with statement optional. In some simple applications, improperly closed les have few real consequences, and the carefully managed context of a with statement isnt necessary.
29.3 Dening a Context Manager Function

One easy way to create a context manager is to write a function that handles the acquisition and release of an important resource. The contextlib module provides a handy decorator that transforms a simple generator function into a context manager.
358
Chapter 29. Managing Contexts: the with Statement
Conceptually, youll break your function into two phases: The __enter__() phase acquires the resources. This is written as statements from the start of the functions suite to up to the one and only yield statement. The value that is yielded is the value thats used in the as clause. Once the function yields this value, then the with statements suite starts processing. If an exception occurs anywhere in the with statement, it will be raised by the yield statement, and must be handled by the function to assure that resources are released properly. The __exit__() phase releases resources. This is written as statements after the yield statement. Heres an example of a context manager that manages a subclass of file. The FileWithCount has an extra method that will append a summary line that shows if everything has gone properly. Well manage this object so that we are assured that le either has a summary, or is removed from the directory.
import os class FileWithCount( file ): def __init__( self, *args, **kw ): super( FileWithCount, self ).__init__( *args, **kw ) self.count= 0 def write( self, data ): super( FileWithCount, self ).write( data ) self.count += data.count('\n' ) def writelines( self, dataList ): for d in dataList: self.write( d ) def summarize( self ): super( FileWithCount, self ).write( "\n:count:`%d`\n" % self.count )
1. We dened a FileWithCount class that adds a line-counting feature to the built-in file type. 2. Note that were dening a subclass of file that adds features. For the most part, we simply pass the arguments to the superclass method functions. 3. The write() method counts the number of newline characters written to the le. 4. The summarize() method appends a label with nal count to the end of the le. Heres a context manager that uses our FileWithCount class.
from contextlib import contextmanager @contextmanager def file_with_count( *args, **kw ): # The __enter__ processing try: counted_file= FileWithCount( *args, **kw ) yield counted_file # The __exit__ processing -- if everything's ok counted_file.summarize() counted_file.close() except: # The __exit__ processing -- if there as an exception counted_file.close() os.remove( counted_file.name ) raise
29.3. Dening a Context Manager Function
359
1. We dened a context manager function, file_with_count() that builds an instance of FileWithCount and yields it to a with statement. 2. If everything works normally, the counted_file.summarize() statement is executed. 3. If there is an exception, then the counted_file.close() statement is executed, and the le is removed via os.remove(). The is removed so that incomplete les are not left around to confuse other application programs or users. Heres an example of using this context manager to create a temporary directory.
with file_with_count( "file.data", "w" ) as results: results.write( "this\nthat\n" )
This yields a le with the following content.

MacBook-5:Python-2.6 slott$ cat file.data this that :count:`2`
The nal :count: line is written automatically by the context manager, simplifying our application.
29.4 Dening a Context Manager Class

In some cases, a class is both the working object and the context manager. The file class is the central example of this. The two do not have to be tied together. Its clearer if well look at creating a context manager that is separate from the working object rst. A Context Manager must implement two methods to collaborate properly with the with statement. __enter__(self ) This method is called on entry to the with statement. The value returned by this method function will be the value assigned to the as variable. __exit__(self, exc_type, exc_val, exc_tb) This method is called on exit from the with statement. If the exc_type, exc_val or exc_tb parameters have values other than None, then an exception occured. A return value of False will propagate the exception after __exit__() nishes. A return value of True will suppress any exception and nish normally. If the exc_type, exc_val or exc_tb parameters have values of None, then this is a normal conclusion of the with statement. The method should return True. Here is a context manager class, named FileCountManager, which incorporates the FileWithCount class, shown above. To be a context manager, this class implements the required __enter__() and __exit__() methods.
class FileCountManager( object ): def __init__( self, *args, **kw ): self.theFile= FileWithCount( *args, **kw ) def __enter__( self ): return self.theFile def __exit__( self, exc_type, exc_val, exc_tb ): if exc_type is not None:
360
# Exception occurred self.theFile.close() os.remove( self.theFile.name ) return False # Will raise the exception self.theFile.summarize() self.theFile.close() return True # Everything's okay
1. The __enter__() method creates the FileWithCount and returns it so that the as clause will assign this to a variable. 2. The __exit__() method checks to see if it is ending with an exception. In case of an exception, we close the le and remove it. We also return False to permit the exception to propagate outside the with statement. If there was no exception, then the :class:FileWithCount.summarize is used to write the summary and the le is closed. The overall main program can have the following structure. We dont need to make special arrangements in the main program to be sure that the log is nalized correctly. We have delegated those special arrangements to the context manager object, leaving us with an uncluttered main program.
with FileCountManager( "file.data", "w" ) as results: results.write( "this\nthat\n" )
29.5 Context Manager Exercises

1. Build a class with its own manager. Merge the methods from FileCountManager into FileWithCount to create a single class which does both sets of features. 2. List with a Checksum. A crypto application works with lists, but only lists that have a checksum of the values in a list of numbers. Dene a class, ManageChecksum, which removes and replaces the last element in a non-empty list. The __init__() method accepts a single parameter which is a list or MutableSequence object. On __entry__(), given a zero-length list, return the list object. On __entry__(), given a list with 1 or more elements, pop the last element. Assert that this is the cryptological checksum of the values of the other elements. Return this updated list. On __exit__(), comput cryptological checksum of the elements and append this checksum to the list. For now, our cryptological checksum can be the simple sum, created with sum(). As an advanced exercise, look at using hashlib to put a better checksum on the list. You should be able to do the following kinds of test.
crypto_list = [] with ManageChecksum( crypto_list ) as theList: theList.extend( [5, 7, 11] ) assert theList == [5, 7, 11] # no checksum in the with statement theList.append( 13 ) assert theList[:-1] == [5, 7, 11, 13] # checksum was added to prevent tampering with ManageChecksum( crypto_list) as theList:
29.5. Context Manager Exercises
361
theList.pop( 0 ) assert theList[:-1] == [7, 11, 13]
Inside the with statement, the list is an ordinary list. Outside the with statement, the list has an anti-tampering checksum appended.
362
Part V
Components, Modules and Packages
363
Organization and Deployment

The basic Python language is rich with features. These include several sophisticated built-in data types (Data Structures), numerous basic statments (Language Basics), a variety of common arithmetic operators and a library of built-in functions. In order to keep the basic Python kernel small, relatively feature features are built-in. A small kernel means that Python interpreters can be provided in a variety of software application, extending functionality of the application without bloating due to a large and complex command language. The more powerful and sophisticated features of Python are separated into extension modules. There are several advantages to this. First, it allows each program to load only the relevant modules, speeding start-up. Second, it allows additional modules to be added easily. Third, it allows a module to be replaced, allowing you to choose among competing solutions to a problem. The second point above, easily adding modules, is something that needs to be emphasized. In the Python community, this is called the batteries included principle. The ideal is to make Python directly applicable to just about any practical problem you may have. Some modules have already been covered in other chapters. In The math Module we covered math and random modules. In Strings we covered the string module. Overview of this part. This part will cover selected features of a few modules. The objective is to introduce some of the power of key Python modules and show how the modules are used to support software development. This isnt a reference, or even a complete guide to these modules. The standard Python Library documentation and other books describe all available modules in detail. Remember that Python is an open-source project: in some cases, youll have to read the modules source to see what it really does and how it works. This part provides a general overview of how to create Python modules in Modules. Well distinguish package and module in Packages. Well overview the Python Library in The Python Library. Module Details. We cover several essential modules in some detail. Complex Strings: the re Module covers regular expressions, which you can use to do string matching and parsing. Dates and Times: the time and datetime Modules covers how to handle the vagaries of our calendar with the time and datetime modules. Well cover many aspects of le handling in File Handling Modules; this includes modules like: sys, glob, fnmatch, fileinput, os, os.path, tempfile, and shutil. Well also look at modules for reading and writing les in various formats in File Formats: CSV, Tab, XML, Logs and Others. Comma-Separated Values: The csv Module will cover Comma Separated Values (CSV) les. Tab-delimited les, however, are a simpler problem, and dont require a separate module. Property Files and Conguration (or .INI ) Files: The CongParser Module will cover parsing Conguration les, sometimes called .INI les. An .INI le is not the best way to handle congurations, but this technique is common enough that we need to show it. Fixed Format Files, A COBOL Legacy: The codecs Module will cover ways to handle COBOL les which are in a xed length format, using EBCDIC data instead of the more common ASCII or Unicode. XML Files: The xml.etree and xml.sax Modules will cover the techniques for parsing XML les.
365
Programs The Ultimate Modules. In a sense a top-level program is a module that does something useful. Its important understand programs as being reusable modules. Eventually most really useful programs get rewritten and merged into larger, more sophisticated programs. In Programs: Standing Alone this part covers modules essential for creating polished, complete stand-alone programs. This includes the getopt and optparse modules. The nal chapter, Architecture: Clients, Servers, the Internet and the World Wide Web covers integration among programs using the client-server programming model. This includes a number of modules that are essential for creating networked programs. We can use the HTTP protocol with a number of modules covered in The World Wide Web and the HTTP protocol . We can use the REST to create or use web services. This is covered in Web Services. Using Writing Web Clients: The urllib2 Module, we can leverage a number of protocols to read a le located anywhere on the World-Wide Web via their Uniform Resource Locator (URL). If none of these protocols are suitable, we can invent our own, using the low-level socket module, covered in Socket Programming.
366
CHAPTER
THIRTY
MODULES
A module allows us to group Python classes, functions and global variables. Modules are one level of composition of a large program from discrete components. The modules we design in one context can often be reused to solve other problems. In Module Semantics we describe the basic semantics of modules. In Module Denition we describe how to dene a module. Well show how to use a module with the import statement in Module Use: The import Statement . A module must be found on the search path, well briey talk about ways to control this in Finding Modules: The Path. There are number of variations on the import statement; well look at these in Variations on An import Theme. Well also look at the exec statement in The exec Statement . This chapter ends with some style notes in Style Notes.
30.1 Module Semantics

A module is a le that contains Python programming. A module can be brought into another program via the import statement, or it can be executed directly as the main script of an application program. There are two purposes for modules; some les may do both. A library module is expected to contain denitions of classes, functions and module variables. If it does anything beyond this, it is generally hard to understand and harder to use properly. A script (or application or main module) does the useful work of an application. It will have up to three disctinct elements: imports of any modules on which it depeds, main function denitions, a script that does the real work. It generally uses library modules. Modules give us a larger-scale structure to our programs. We see the following levels of Python programming: The individual statement. A statement makes a specic state change by changing the value of a variable. State changes are what advance our program from its initial state to the desired ending state. Multiple statements are combined in a function. Functions are designed to be an indivisible, atomic unit of work. Functions can be easily combined with other functions, but never taken apart. Functions may be as simple as a single statement. While functions could be complex, its important that they be easily understood, and this limits their complexity. A well-chosen function name helps to clarify the processing. Multiple functions and related data are used to dene a class of objects. To be useful, a class must have a narrowly dened set of responsibilities. These responsibilities are characterized by the class attributes and behaviors.
367
Multiple closely-related classes, functions and variables are combined into modules. The module name is the le name. A module should provide a closely related set of classes and functions. Sometimes, a module will package a set of closely related functions. Modules can be combined into packages. The directory structure denes the packages and their constituent modules. Additionally, packages contain some additional les that Python uses to locate all of the elements of the package. The user-oriented application program depend on modules or packages. The application the functionality that the user perceives is usually the top-most executable script that does the useful work. The relationship between shell commands (or desktop icons, or web links) that a user sees and the packaging of components that implement those commands can be murky. For example, a single application le may have multiple aliases, making it appear like independed commands. A single script can process multiple command-line options. The application-level view, since it is presented to the user, must focused on usability: the shell commands or icons the user sees. The design of modules and packages sould be focused on maintenance and adaptability. The modules are the les that you, the developer, must use to keep the software understandable. Components: Class and Module. A class is a container for attributes, method functions, and nested class denitions. Similarly, a module can also contain attributes, functions and class denitions. A module is dierent from a class in several ways. First, a module is a physical le; it is the unit of software construction and conguration management. A class is dened within a le. Additionally, a class denition allows us to create a number of class instances, or objects. A module, on other hand, can only have a single instance. Python will only import a module one time; any additional requests to import a module have no eect. Any variables dened by the module, similarly, will only have a single instance. Beyond this technical distinction, we generally understand modules to be the big, easy-to-understand components out of which are applications are built. A class is a ner-grained piece of functionality, which usually captures a small, very tightly bound collection of attribute and operations.
30.2 Module Denition

A module is a le; the name of the module is the le name. The .py extension on the le name is required by the operating system, and gracefully ignored by Python. We can create a le named roulette.py, include numerous denitions of classes and functions related to the game of roulette. Note that a module name must be a valid Python name. Operating system le names dont have the same restrictions on their names. The rules for variable names are in Variables. A modules name is limited to letters, digits and _s. Standard le systems are case-sensitive. Filenames are traditionally all lower-case. Consequently, most Python module le names are all lower-case letters. Note that the Windows lesystem has very complex le naming rules in which there are long versions of names and short versions of names. Further, Windows is exible regarding case matching of le names. It helps to avoid these problems by using relatively short, all lower-case lenames in Windows, also. The rst line of a module is usually a #! comment; this is typically #!/usr/bin/env python. The next few lines are a triple-quoted module doc string that denes the contents of the module le. As with other Python doc strings, the rst line of the string is a summary of the module. This is followed by a more complete denition of the modules contents, purpose and usage. Also, a module may include the __all__ variable. This lists the names which are created by an from module import *. Absent the __all__ variable, all names that dont begin with _ are created. Example. For example, we can create the following module, called dice.py. 368 Chapter 30. Modules
dice.py
#!/usr/bin/env python """dice - basic definitions for Die and Dice. Die - a single die Dice - a collection of one or more dice roll - a function to roll a pair of dice""" from random import * class Die( object ): """Simulate a 6-sided die.""" def __init__( self ): self.value= None def roll( self ): self.value= randrange(6)+1 def total( self ): return self.value class Dice: """Simulate a pair of 6-sided dice.""" def __init__( self ): self.value = ( Die(), Die() ) def roll( self ): map( lambda d: d.roll(), self.value ) def dice( self ): return tuple( [d.value for d in self.value] ) pair= Dice() def roll(): pair.roll() return pair.dice()
1. A main script le must include the shell escape to run nicely in a Posix or Mac OS environment. Other les, even if they arent main scripts, can include this to mark them as Python. 2. Our docstring is a minimal summary. Well-written docstrings provide more information on the classes, variables and functions that are dened by the module. 3. Many modules depends on other modules. Note that Python optimizes these imports; if some other module has already imported a given module, it is simply made available to our module. If the module has not been imported already, it is imported for use by our module. 4. As is typical of many modules, this module provides some class denitions. 5. This module denes a module-global variable, pair. This variable is part of the module; it appears global to all classes and functions within this module. It is also available to every client of this module. Since this variable is part of the module, every client is sharing a single variable. 6. This module denes a handy function, roll(), which uses the module global variable, pair. Conspicuous by its absence is any main script to do anything more useful than create a single module global variable. This module is a pure library module.
30.2. Module Denition
369
30.3 Module Use: The import Statement

Since a module is just a Python le, there are two ways to use a module. We can import the module, to make use of its denitions, or we can execute it as a script le to have it do useful work. We started looking at execution of scripts back in Script Mode, and have been using it heavily. We looked briey at the import statement in Using Modules. There are several variations on this statement that well look at in the next section. In this section, well look at more features of the import statement. The essential import statement has the following syntax:
import module
The module name is the Python le name with the .py le extension removed. Python does the following. 1. Search the global namespace for the module. If the module exists, it had already been imported; for the basic import, nothing more needs to be done. 2. If the module doesnt exist, search the Python path for a le; the le name is the module name plus the .py extension. The search path has a default value, and can be modied by command-line arguments and by environment variables. If the module name cant be found anywhere on the path, an ImportError exception is raised. 3. If the le was found, create the modules new, unique namespace; this is the container in which the modules denitions and module-level variables will be created. Execute the statements in the module, using the modules namespace to store the variables and denitions. Well look a little more closely at namespaces below. The most important eect of importing a module is that the Python denitions from the module are now part of the running Python environment. Each class, function or variable dened in the module is available for use. Since these objects are contained in the modules namespace, The names of those elements must be qualied by the modules name. In the following example, we import the dice module. Python will search for module dice, then for the le dice.py from which to create the module. After importing, create an instance of the Dice class and called that instance craps. We qualied the class name with the module name: dice.Dice.
>>> >>> >>> >>> (3, import dice craps= dice.Dice() craps.roll() craps.dice() 5)
Namespaces. Python maintains a number of local namespaces and one global namespace. A unique local namespace is used when evaluating each function or method function. In eect, a variable created in a functions namespace is private to that function; it only exists only while the function executes. Typically, when a module is imported, the modules namespace is the only thing created in the global namespace. All of the modules objects are inside the modules namespace. You can explore this by using the built-in dir() function. Do the following sequence of steps. 1. Create a small module le (like the dice.py example, above). 2. Start a fresh command-line Python interpreter in the same directory as your module le. Starting the intepreter in the same directory is the simplest way to be sure that your module will be found by the import statement.
370
Chapter 30. Modules
3. Evalute dir() to see what is in the initial global namespace. 4. Import your module. 5. Evaluate dir() to see what got added to the namespace. 6. Evaluate dir(your module) to see whats in your modules namespace. Scripts and Modules. There are two ways to use a Python le. We can execute it as a script or we can import it as a library module. We need to keep this distinction clear when we create our Python applications. The le that a user executes will do useful work, and must be a script of some kind. This script can be an icon that the user double-clicked, or a command that the user typed at a command prompt; in either case, a single Python script initiates the processing. A le that is imported will provide denitions. Well emphasize this distinction. Important: Bad Behavior The standard expectation is that a library module will contain only denitions. Some modules create module global variables; this must be fully documented. It is bad behavior for an imported module to attempt to do any real work beyond creating denitions. Any real work that a library module does makes reuse of that module nearly impossible. Importing a module means the module le is executed. This creates is an inherent, but important ambiguity. A given le can be used as a script or used as a library; any le can be used either way. Heres the complete set of alternatives. A top-level script. You execute a script with the Run Module menu item in IDLE. You can also execute a script from your operating system command prompt. For example, python file.py will execute the given le as a script. Also, you can set up most operating systems so that entering le.py at the command prompt will execute the le as a script. Also, you can set up most GUIs so that double-clicking the file.py icon will launch Python and execute the given le as a script. import. You can import a library module. As described above, Python will not import a module more than once. If the module was not previously imported, Python creates a namespace and executes the le. The namespace is saved. exec. Pythons exec statement is similar to the import statement, with an important dierence: The exec statement executes a le in the current namespace. The exec statement doesnt create a new namespace. Well look at this in The exec Statement . The Main-Import Switch. Since a le can be used as script or library, we can intentionally create les that are both. We can create a script which can also be used as a library. And we can create a library which has a useful behavior when used as a script. This promotes reuse of libraries. Python provides a global variable that helps to dierentiate between a main program script and a module library module. The global __name__ variable is equal to "__main__" when the initial (or top-level or outermost or main) le is being processed. When you have an executable script, and you run that script from the command line, that script sees __name__ equal to "__main__". However, when an import is in process, the __name__ variable is the name of the module being imported. As an example, we can make use of this to provide stand-alone unit testing for a library module. When we write a module that is primarily denitional, we can have it execute its own unit tests when it is used as a main program. This makes testing a library module simple: we import it and it runs its unit test. We do this by examining the __name__ variable.
if __name__ == "__main__": unittest.main()
30.3. Module Use: The import Statement
371
When some other script imports a module (for example, named dice.py), the __name__ variable is "dice" and nothing special is done. When testing, however, we can execute the module by itself; in this case the __name__ variable is "__main__" and the test function is executed.
30.4 Finding Modules: The Path

For modules to be available for use, the Python interpreter must be able to locate the module le. Python has a set of directories in which it looks for module les. This set of directories is called the search path, and is analogous to the PATH environment variable used by an operating system to locate an executable le. Pythons search path is built from a number of sources: PYTHONHOME is used to dene directories that are part of the Python installation. If this environment variable is not dened, then a standard directory structure is used. For Windows, the standard location is based on the directory into which Python is installed. For most Linux environments, Python is installed under /usr/local, and the libraries can be found there. For Mac OS, the home directory is under /Library/Frameworks/Python.framework. PYTHONPATH is used to add directories to the path. This environment variable is formatted like the OS PATH variable, with a series of lenames separated by :s (or ;s for Windows). Script Directory. If you run a Python script, that scripts directory is placed rst on the search path so that locally-dened moules will be used instead of built-in modules of the same name. The site modules locations are also added. (This can be disabled by starting Python with the -S option.) The site module will use the PYTHONHOME location(s) to create up to four additional directories. Generally, the most interesting one is the site-packages directory. This directory is a handy place to put additional modules youve downloaded. Additionally, this directory can contain .pth les. The site module reads .pth les and puts the named directories onto the search path. The search path is dened by the path variable in the sys module. If we import sys, we can display sys.path. This is very handy for debugging. When debugging shell scripts, it can help to run python -c 'import sys; print sys.path' just to see parts of the Python environment settings. Installing a module, then, is a matter of assuring that the module appears on the search path. There are four central methods for doing this. Some packages will suggest you create a directory and place the package in that directory. This may be done by downloading and unzipping a le. It may be done by using Subversion and sychronizing your subversion copy with the copy on a server. Either way, you will likely only need to create an operating system link to this directory and place that link in site-packages directory. Some packages will suggest you download (or use subversion) to create a temporary copy. They will provide you with a script typically based on setup.py which moves les into the correct locations. This is called the distutils distribution. This will generally copy the module les to the site-packages directory. Some packages will rely on setuptools. This is a package from the http://peak.telecommunity.com/DevCenter/setuptools Python Enterprise Application Kit that extends distuils to further automates download and installation. This tool, also, works by moving the working library modules to the site-packages directory. Extending the search path. Either set the PYTHONPATH environment variable, or put .pth les in the site-packages directory. Tip: Windows Environment In the Windows environment, the Python_Path symbol in the Windows registry is also used to locate modules. This, however, is not portable or standardized, so try to avoid using it. 372 Chapter 30. Modules
30.5 Variations on An import Theme

There are several variations on the import statement. We looked at these briey in The math Module. In this section, well cover the variations available on the import statement. Basic Import. This is covered in Module Use: The import Statement . Import As. This allows us to import a module, and assign it a new name. From Module Import Names. This allows us to import a module, making some names part of the global namespace. Combined From and As import.
30.5.1 Import As
A useful variation in the import statement is to rename a module using the as clause.
import module as name
This module renaming is used in two situations. We have two or more interchangable versions of a module. The module name is rather long and painful to type. There are number of situations where we have interchangable versions of a module. One example is the built-in os module. This module gives us a number of functions that behave identically on most major operating systems. The way this is done is to create a number of variant implementations of these functions, and then use as appropriate as clause to give them a platform-neutral name. Heres a summary of how the os module uses import as.
if 'posix' in _names: import posixpath as path elif 'nt' in _names: import ntpath as path elif 'mac' in _names: import macpath as path
After this if -statement, one of the various platform-specic modules will have been imported, and it will have the platform-independent name of os.path. In the case of some modules, the name is rather long. For example, sqlalchemy is long and easy to misspell. Its somewhat simpler to use the following technique.
import sqlalchemy as sa db= sa.create_engine('sqlite:///file.db')
This allows us to use sa as the module name.
30.5.2 From Module Import Names

Two other variations on the import statement introduce selected names from the module into the local namespace. One form picks specic names to make global.
30.5. Variations on An import Theme
373
from module import name , ...
This version of import adds a step after the module is imported. It adds the given list of names into the local namespace, making them available without using the module name as a qualier. For example:
from math import sin, cos, tan print dir(math) ['__doc__', '__name__', 'acos', 'asin', 'atan', 'atan2', 'ceil', 'cos', 'cosh', 'e', 'exp', 'fabs', 'floor', 'fmod', 'frexp', 'hypot', 'ldexp', 'log', 'log10', 'modf', 'pi', 'pow', 'sin', 'sinh', 'sqrt', 'tan', 'tanh'] print locals() {'math': <module 'math' (built-in)>, '__doc__': None, '__version__': '1.0', '__file__': 'Macintosh HD:SWdev:Jack:Python:Mac:Tools:IDE:PythonIDE.py', '__name__': '__main__', '__builtins__': <module '__builtin__' (built-in)>, 'inspect': <function inspect at 0x0d084310>, 'sin': <built-in function sin>, 'cos': <built-in function cos>, 'tan': <built-in function tan>}
In this example, the locals() value shows that the sin(), cos() and tan() functions are now directly part of the namespace. We can use these functions without referring to the math module. We can evaluate sin(0.7854), rather than having to say math.sin(0.7854). This is discouraged because it tends to conceal the origin of objects. Another variation on import makes all names in the module part of the local namespace. This import has the form:
from module import *
This makes all names from the module available in the local namespace.
30.5.3 Import and Rename

Finally, we can combine the from and as options to both import selected items and provide more understandable names for them. We can say things like:
from module importname as name
In this case, were both concealing the source of the item and its original name. Wed best have a very good reason for this. Think of the confusion that can be caused by
from math import sqrt as sin
This must be used cautiously to prevent creating more problems than it appears to solve.
374
Chapter 30. Modules
30.6 The exec Statement

The import statement, in eect, executes the module le. Typically, the les we import are dened as sequences of denitions. Since our main program often begins with a series of import statements, these modules are imported into the global namespace. Python also optimizes the modules brought in by the import statement so that they are only imported once. The exec statement can execute a le, a string of Python code, as well as a code created by the compile() function. Unlike the import statement, it doesnt optimize module denitions or create and save a new namespace.
exec expression
The functions eval() and execfile() do essentially the same thing. Warning: Fundamental Assumptions The exec statement, eval() function and execfile() functions are dangerous tools. These break one of the Fundamental Assumptions: the source you are reading is the source that is being executed. A program that uses the exec statement or eval() function is incorprating other source statements into the program dynamically. This can be hard to follow, maintain or enhance. Generally, the exec statement is something that must be used with some care. The most common use is to bring in a set of conguration parameters written as simple Python assignment statements. For example, we might use a le like the following as the conguration paramaters for a program.
db_server= "dbs_prod_01" db_port= "3306" db_name= "PROD"
30.7 Module Exercises

30.7.1 Refactor a Script
A very common situation is to take a script le apart and create a formal module of the denitions and a separate module of the script. If you refer back to your previous exercise scripts, youll see that many of your les have denitions followed by a main script which demonstrates that your denitions actually work. When refactoring these, youll need to separate the denitions from the test script. Lets assume you have the following kind of script as the result of a previous exercise.
# Some Part 3 Exercise. class X( object ): does something class Y( X ): does something a little different x1= X() x1.someMethod() y2= Y() y2.someOtherMethod()
30.6. The exec Statement
375
Youll need to create two les from this. The module will be the simplest to prepare, assume the le name is myModule.py
#!/usr/bin/env python class X( object ): does something class Y( X ): does something a little different
Your new new demonstation application will look like this because you will have to qualify the class and function names that are created by the module.
#!/usr/bin/env python import myModule x1= myModule.X() x1.someMethod() y2= myModule.Y() y2.someOtherMethod()
Your original test script had an implicit assumption that the denitions and the test script were all in the same namespace. This will no longer be true. While you can nesse this by using from myNewModule import *, this is not the best programming style. It is better to rewrite the test script to explicitly qualify names with the module name. There are a number of related class denitions in previous exercises that can be used to create modules. Any of the exercises in Class Denition Exercises contains a number of related classes. The exercise in Shuing Method for the Deck class has two parts: denitions of Deck and related material, and a procedure for comparing dierent shuing algorithms. This should be repackaged to separate the performance measurement script from the basic denitions. You should be able to separate the Deck and the various shuing strategies in a module separate from the performance measurement script. The simulation built in State can be formalized into two modules. The lowest-level module denes the basic game of Roulette including the Wheel and RouletteGame. Another module imports this and denes the Player and the states. Finally, the main script imports the game, the player and runs the simulation to produce a log of wins and losses. The rational number class, built in Numeric Type Special Methods can be formalized into a module. A script can import this module and demonstrate the various operations on rational numbers. The sequence with additional statistical methods, built in Sample Class with Statistical Methods can be formalized into a module. A script can import this module and demonstrate the various statistical operations on sample data.
30.7.2 Install a New Module

Create a simple module le with some denitions. Preferrably, this is a solution to Refactor a Script . Install the denitional part into the PYTHONPATH. Be sure to rename or remove the local version of this le. Be sure to use each installation method. 1. Move the module le to the site-packages directory. Be sure that it is removed (or renamed) in your local directory. 2. Move the module le to another directory and create a hard link (using the Linux ln or equivalent Windows utility) from site-packages to the other directory you created. 376 Chapter 30. Modules
3. Remove the hard link and put a .pth le in the site-packages directory. 4. Remove the .pth le and update the PYTHONPATH environment variable to reference the new directory.
30.7.3 Planning for Maintenance and Upgrades

There are a number of module installation scenarios; each of these will require a dierent technique. Compare and contrast these techniques from several points of view: cost to deploy, security of the deployment, ease of debugging, and control over what the user experiences. You have to provide modules and an application on a number of desktop PCs. Python must be installed on each individual desktop. However, the application that uses Python could be put on a shared network drive. What are the pros and cons of installing a Python-based application locally versus on a network drive? How would you handle the initial setup? How would you handle an upgrade to Python itself? For example, how would you install Python 3.1 so as to preserve your modules and application? How would you control an upgrade to a Python-based application? For example, you have a new module le that needs to be made available to all users. You have to provide modules and an application on a server, shared by a number of users. Python is installed on the server, as is the Python-based application. What security considerations should be put into place? How would you handle initial installation of Python and your server-based application? How would you handle an upgrade to Python on this shared server? How would you control an upgrade to the Python-based application on this shared server?
30.8 Style Notes

Modules are a critical organizational tool for nal delivery of Python programming. Python software is delivered as a set of module les. Often a large application will have one or more module les plus a main script that initiates the application. There are several conventions for naming and documenting module les. Module names are python identiers as well as le names. Consequently, they can only use _ as a punctuation mark. Most modules have names that are mixedCase, beginning with lowercase letters. This conforms to the useage on most le systems. MacOS users would do well to keep their module names to a single word, and end with .py. This promotes portability to operating systems where le names are more typically single words. Windows, in particular, can have trouble with spaces in le names. Some Python modules are a wrapper for a C-language module. In this case the C/C++ module is named in all lowercase and has a leading underscore (e.g. _socket ). A modules contents start with a docstring. After the docstring comes any version control information. The bulk of a module is typically a series of denitions for classes, exceptions and functions. A modules docstring should begin with a one-line pithy summary of the module. This is usually followed by in inventory of the public classes, exceptions and functions this module creates. Detailed change history does not belong here, but in a separate block of comments or an additional docstring. If you use CVS or svn to track the versions of your source les, following style is recommended. This makes the version information available as a string, and visible in the .py source as well as any .pyc working les.
30.8. Style Notes
377
"""My Demo Module. This module is just a demonstration of some common styles.""" __version__ = "$Revision: 1.3$"
Note that the modules name will qualift everything created in the module, it is never necessary to have a prex in front of each name inside the module to show its origin. For example, consider a module that contains classes and functions related to statistical analysis, called stats.py. The stats module might contain a class for tracking individual samples. Poor Names. We dont include extra name prexes like statsSample or stats_sample. Better Names. We would call our internal sample class Sample. A client application that contains an import stats statement, would refer to the class as stats.Sample. This needless over-qualication of names sometimes devolves to sillines, with class names beginning with c_, function names beginning with f_, the expected data type indicated with a letter, and the scope (global variables, local variables and function parameters) all identied with various leading and trailing letters. This is not done in Python programming. Class names begin with uppercase letters, functions begin with lowercase. Global variables are identied explicitly in global statements. Most functions are kept short enough that that the parameter names are quite obvious. Any element of a module with a name that begins with _single_leading_underscore is not created in the namespace of the client module. When we use from stats import *, these names that begin with _ are not inserted in the global namespace. While usable within the module, these names are not visible to client modules, making them the equivalent of Javas private declaration. Errors and Exceptions. A common feature of modules is to create a module-wide exception class. The usuall approach looks like the following. Within a module, you would dene an Error class as follows:
class Error( Exception ): pass
You can then raise your module-specic exception with the following.
raise Error, "additional notes"
A client module or program can then reference the modules exception in a try statement as module.Error. For example:
import aModule try: aModule.aFunction() except: aModule.Error, ex: print "problem", ex raise
With this style, the origin of the error is shown clearly.
378
Chapter 30. Modules
CHAPTER
THIRTYONE
PACKAGES
A package is a collection of Python modules. Packages allow us to structure a collection of modules. In Package Semantics we describe the basic semantics of packages. In Package Denition we describe how to dene a package. Well look at using a package in Package Use.
31.1 Package Semantics

A package is a directory that contains modules. Having a directory of modules allows us to have modules contained within other modules. This allows us to use qualied module names, clarifying the organization of our software. We can, for example, have several simulations of casino games. Rather than pile all of our various les into a single, at directory, we might have the following kind of directory structure. (This isnt technically complete, it needs a few additional les.)
casino/ craps/ dice.py game.py player.py roulette/ wheel.py game.py player.py blackjack/ cards.py game.py player.py srategy/ basic.py martingale.py bet1326.py cancellation.py
Given this directory structure, our overall simulation might include statements like the following.
import craps.game, craps.player import strategy.basic as betting class MyPlayer( craps.player.Player ): def __init__( self, stake, turns ): betting.initialize(self)
379
We imported the game and player modules from the craps package. We imported the basic module from the strategy package. We dened a new player based on a class named Player in the craps.player package. We have a number of alternative betting strategies, all collected under the strategy package. When we import a particular betting strategy, we name the module betting. We can then change to a dierent betting strategy by changing the import statement. There are two reasons for using a package of modules. There are a lot of modules, and the package structure claries the relationships among the modules. If we have several modules related to the game of craps, we might have the urge to create a craps_game.py module and a craps_player.py module. As soon as we start structuring the module names to show a relationship, we can use a package instead. There are alternative implementations, and the package contains polymorphic modules. In this case, we will often use an import package.alternative as interface kind of import statement. This is often used for interfaces and drivers to isolate the interface details and provide a uniform API to the rest of the Python application. It is possible to go overboard in package structuring. The general rule is to keep the package structure relatively at. Having only one module at the bottom of deeply-nested packages isnt really very informative or helpful.
31.2 Package Denition

In order for Python to make use of a directory as package, the directory must have a name that is a valid Python identier and contain a special module named __init__. Valid Python names are composed of letters, digits and underscores. See Variables for more information. The __init__ module is often an empty le, __init__.py in the package directory. Nothing else is required to make a directory into a package. The __init__.py le, however, is essential. Without it, youll get an ImportError. For example, consider a number of modules related to the denition of cards. We might have the following les.
cards/ __init__.py standard.py blackjack.py poker.py
The cards.standard module would provide the base denition of card as an object with suit and rank. The cards.blackjack module would provide the subclasses of cards that we looked at in Blackjack Hands. The cards.poker module would provided the subclasses of cards that we looked at in Poker Hands. The cards.blackjack module and the cards.poker module should both import the cards.standard module to get the base denition for the Card and Deck classes. The __init__ module. The __init__ module is the initialization module in a package. It is processed the rst time that a package name is encountered in an import statement. In eect, it initializes Pythons understanding of the package. Since it is always loaded, it is also eectively the default module in a package. There are a number of consequences to this. We can import the package, without naming a specic module. In this case weve imported just the initialization module, __init__. If this is part of our design, well put some kind of default or top-level denitions in the __init__ module. 380 Chapter 31. Packages
We can import a specic module from the package. In this case, we also import the initialization module along with the requested module. In this case, the __init__ module can provide additional denitions; or it can simply be empty. In our cards example, above, we would do well to make the __init__ module dene the basic Card and Deck classes. If we import the package, cards, we get the default module, __init__ , which gives us cards.Card and cards.Deck. If we import a specic module like cards.blackjack , we get the __init__ module plus the named module within the package.
31.3 Package Use

If the __init__ module in a package is empty, the package is little more than a collection of module les. In this case, we dont generally make direct use of a package. We merely mention it in an import statement: import cards.poker. On other hand, if the __init__ module has some denitions, we can import the package itself. Importing a package just imports the __init__ module from the package directory. In this case, we mention the package in an import statement: import cards. Even if the __init__ module has some denitions in it, we can always import a specic module from within the package. Indeed, it is possible for the __init__ module in a package is to do things like adjust the search path prior to locating individual module les.
31.4 Package Exercises

1. Create a Package. In the previous chapters exercises (see Refactor a Script ) are some suggestions for creating a simple module. The modules decribed in that exercise are so small that adding modules does not seem necessary. However, its the best example of how packages arise when solving practical problems. Pick a module that youve already created. Add a second le with a few simple classes that do little real work. These are best described as Hello World classes, since they dont do anything more useful than provide a simple response to indicate that the module was imported correctly. Create a packge directory with the necessary __init__.py le. Create a demonstration script which imports and exercises both modules from this package.
31.5 Style Notes

Since a package, like a module, is both a le system location and a Python construct, the name must be a valid Python name, using just letters, numbers and _s. Additionally, some le systems are cavalier about maintaining the original case of the lename. The old Mac OS (pre Mac OS X), and many of the old Windows variants would casually alter the case of lenames. Consequently, package and module names should be all lower case. This way, there is no ambiguity about the intended case of the module name. Package structures should be relatively at. The general rule is to keep the package structure relatively at. Having only one module at the bottom of deeply-nested packages isnt really very informative or helpful. While it may seem like import casino.games.definitions.tablegames.dicegames.craps leaves lots of room for expansion, there just arent enough casino games to make this immensely deep structure usable.
31.3. Package Use
381
Packages are generally used for two things: Collecting related modules together in a directory to simplify installation, maintenenace and documentation. Dening alternative implementations as simply as possible. If all of your modules are more-or-less unique, then a package structure isnt going to help. Similarly, if you dont have alternate implementations of some driver or interface, a package structure isnt useful.
382
Chapter 31. Packages
CHAPTER
THIRTYTWO
THE PYTHON LIBRARY

Consistent with the Pythonic Batteries Included philopsophy of Python, there are hundreds of extension modules. It can be diicult to match a programming need with a specic module. The Python Library Reference document can be hard to pick through to locate an appropriate module. Well start at the top of the library organization and work our way down to a useful subset of the tremendous wealth that is Python. In Overview of the Python Library well take a very high level overview of whats in the Python library. Well closely at the 50 or so most useful modules in Most Useful Library Sections.
32.1 Overview of the Python Library

The Python Library Reference organizes modules into the following sections. The current version of the Library documentation strives to prsent the modules with the most useful near the front of the list. The rst 23 chapters, plus chapter 26 are the most useful. From chapter 24 and below (except for chapter 26), the modules are too highly specialized to cover in this book. 1. Introduction 2. Built-in Objects. This chapter provides complete documentation of the built-in functions, exceptions and constants. 3. Built-in Types. All of the data types weve looked at are documented completely in this chapter of the library reference. Of course, there are additional types in the Python reference that we havent looked at. 4. String Services. This chapter includes almost a dozen modules for various kinds of string and text handling. This includes regular expression pattern matching, Unicode codecs and other string-processing modules. 5. Data Types. This chapter has almost 20 modules providing additional data types, including datetime. 6. Numeric and Mathematical Modules. This chapter describes math, decimal and random modules. 7. Internet Data Handling. One secret behind the internet is the use of standardized sophisticated data objects, like email messages with attachments. This chapter covers over a dozen modules for handling data passed through the internet. 8. Structured Markup Processing Tools. XML, HTML and SGML are all markup languages. This chapter covers tools for parsing these languages to separate the content from the markup. 9. File Formats. This chapter covers modules for parsing les in format like Comma Separated Values (CSV). 10. Cryptographic Services. This chapter has modules which can be used to develop and compare secure message hashes. 383
11. File and Directory Access. This chapter of the Library Reference covers many of the modules well look at in File Handling Modules. 12. Data Compression and Archiving. This chapter describes modules for reading and writing zip les, tar les and BZ2 les. Well cover these modules in File Handling Modules, also. 13. Data Persistence. Objects can be written to les, sockets or databases so that they can persist beyond the processing of one specic program. This chapter covers a number of packages for pickling objects so they are preserved. The SQLite 3 relational database is also described in this module. 14. Generic Operating System Services. An Operating System provides a number of services to our application programs, including access to devices and les, consistent notions of time, ways to handle command-line options, logging, and handling operating system errors. Well look some of these modules in Programs: Standing Alone. 15. Optional Operating System Services. This section includes operating system services that are common to most Linux variants, but not always available in Windows. 16. Unix Specic Services. There are a number of Unix and Linux-specic features provided by these modules. 17. Interprocess Communication and Networking. Larger and more complex application programs often consist of multiple, cooperating components. The World Wide Web, specically, is based on the interaction between client and server programs. This chapter describes modules that provide a basis for communcating among the OS processes that execute our programs. 18. Internet Protocols and Support. This chapter describes over two dozen modules that process a wide variety of internet-related data structures. This varies from the relatively simple processing of URLs to the relatively complex processing of XML-based Remote Procedure Calls (XML-RPC). 19. Multimedia Services. Multimedia includes sounds and images; these modules can be used to manipulate sound or image les. 20. Graphical User Interfaces with Tk. The Tkinter module is one way to build a graphical desktop application. The GTK libraries are also widely used to build richly interactive desktop applications; to make use of them, youll need to download the pyGTK package. 21. Internationalization. These packages help you separating your message strings from the rest of your application program. You can then translate your messages and provide language-specic variants of your software. 22. Program Frameworks. These are modules to help build command-line applications. 23. Development Tools. These modules are essential to creating polished, high-quality software: they support the creation of usable documents and reliable tests for Python programs. 24. The Python Debugger 25. The Python Prolers 26. Python Runtime Services. This chapter describes the sys module, which provides a number of useful objects. 27. Custom Python Interpreters 28. Restricted Execution 29. Importing Modules 30. Python Language Services 31. Python compiler package 32. Abstract Syntax Trees
384
Chapter 32. The Python Library
33. Miscellaneous Services 34. SGI IRIX Specic Services 35. SunOS Specic Services 36. MS Windows Specic Services
32.2 Most Useful Library Sections

This section will overview about 50 of the most useful libary modules. These modules are proven technology, widely used, heavily tested and constantly improved. The time spent learning these modules will reduce the time it takes you to build an application that does useful work. Well dig more deeply into just a few of these modules in subsequent chapters. Tip: Lessons Learned As a consultant, weve seen far too many programmers writing modules which overlap these. There are two causes: ignorance and hubris. In this section, we hope to tackle the ignorance cause. Python includes a large number of pre-built modules. The more you know about these, the less programming you have to do. Hubris sometimes comes from the feeling that the library module doesnt t our unique problem well enough to justify studying the library module. In other languages we cant read the library module to see what it really does. In Python, however, the documentation is only an introduction; were encouraged to actually read the library module. This is called the Use the Source, Luke principle. We nd that hubris is most closely associated with calendrical calcuations. It isnt clear why programmers invest so much time and eort writing buggy calendrical calculations. Python provides many modules for dealing with times, dates and the calendar. 8. String Services. The String Services modules contains string-related functions or classes. See Strings for more information on strings. re The re module is the core of text pattern recognition and processing. A regular expression is a formula that species how to recognize and parse strings. The re module is described in detail in Complex Strings: the re Module. struct The avowed purpose of the struct module is to allow a Python program to access Clanguage APIs; it packs and unpacks C-language struct object. It turns out that this module can also help you deal with les in packed binary formats. diib The difflib module contains the essential algorithms for comparing two sequences, usually sequences of lines of text. This has algorithms similar to those used by the Unix di command (the Window COMP command). StringIO cStringIO There are two variations on StringIO which provide le-like objects that read from or write to a string buer. The StringIO module denes the class StringIO , from which subclasses can be derived. The cStringIO module provides a high-speed C-language implementation that cant be subclassed. Note that these modules have atypical mixed-case names. textwrap This is a module to format plain text. While the word-wrapping task is sometimes handled by word processors, you may need this in other kinds of programs. Plain text les are still the most portable, standard way to provide a document.
32.2. Most Useful Library Sections
385
codecs This module has hundreds of text encodings. This includes the vast array of Windows code pages and the Macintosh code pages. The most commonly used are the various Unicode schemes (utf-16 and utf-8). However, there are also a number of codecs for translating between strings of text and arrays of bytes. These schemes include base-64, zip compression, bz2 compression, various quoting rules, and even the simple rot_13 substitution cipher. 9. Data Types. The Data Types modules implement a number of widely-used data structures. These arent as useful as sequences, dictionaries or strings which are built-in to the language. These data types include dates, general collections, arrays, and schedule events. This module includes modules for searching lists, copying structures or producing a nicely formatted output for a complex structure. datetime The datetime handles details of the calendar, including dates and times. Additionally, the time module provides some more basic functions for time and date processing. Well cover both modules in detail in Dates and Times: the time and datetime Modules. These modules mean that you never need to attempt your own calendrical calculations. One of the important lessons learned in the late 90s was that many programmers love to tackle calendrical calculations, but their eorts had to be tested and reworked prior to January 1, 2000, because of innumerable small problems. calendar This module contains routines for displaying and working with the calendar. This can help you determine the day of the week on which a month starts and ends; it can count leap days in an interval of years, etc. collections This package contains some handy data types, plus the Abstract Base Classes that we use for dening our own collections. Data types include the collections.deque a double-ended queue that can be used as stack (LIFO) or queue (FIFO). The collections.defaultdict class, which can return a default value instead of raising an exception for missing keys. The collections.namedtuple function helps us to create a small, specialized class that is a tuple with named positions. We made use of this library in Creating or Extending Data Types. bisect The bisect module contains the bisect() function to search a sorted list for a specic value. It also contains the insort() fucntion to insert an item into a list maintaining the sorted order. This module performs faster than simply appending values to a list and calling the sort() method of a list. This modules source is instructive as a lesson in well-crafted algorithms. array The array module gives you a high-performance, highly compact collection of values. It isnt as exible as a list or a tuple, but it is fast and takes up relatively little memory. This is helpful for processing media like image or sound les. sched The sched module contains the denition for the scheduler class that builds a simple task scheduler. When a scheduler is contructed, it is given two user-supplied functions: one returns the time and the other executes a delay waiting for the time to arrive. For real-time scheduling, the time module time() and sleep() functions can be used. The scheduler has a main loop that calls the supplied time function and compares the current time with the time for scheduled tasks; it then calls the supplied a delay function for the dierence in time. It runs the scheduled task, and calls the delay function with a duration of zero to release any resources. Clearly, this simple algorithm is very versatile. By supplying custom time functions that work in minutes instead of seconds, and a delay function that does additional background processing while waiting for the scheduled time, a exible task manager can be constructed. copy The copy module contains functions for making copies of complex objects. This module contains a function to make a shallow copy of an object, where any objects contained within the parent are not copied, but references are inserted in the parent. It also contains a
386
function to make a deep copy of an object, where all objects contained within the parent object are duplicated. Note that Pythons simple assignment only creates a variable which is a label (or reference) to an object, not a duplicate copy. This module is the easiest way to create an independent copy. pprint The pprint module contains some useful functions like pprint.pprint() for printing easy-to-read representations of nested lists and dictionaries. It also has a PrettyPrinter class from which you can make subclasses to customize the way in which lists or dictionaries or other objects are printed. 10. Numeric and Mathematical Modules. These modules include more specialized mathemathical functions and some additional numeric data types. decimal The decimal module provides decimal-based arithmetic which correctly handles significant digits, rounding and other features common to currency amounts. math The math module was covered in The math Module. It contains the math functions like sine, cosine and square root. random The random module was covered in The math Module. 11. File and Directory Access. Well look at many of these modules in File Handling Modules. These are the modules which are essential for handling data les. os.path The os.path module is critical for creating portable Python programs. The popular operating systems (Linux, Windows and MacOS) each have dierent approaches to le names. A Python program that depends on os.path will behave more consistently in all environments. leinput The fileinput module helps your progam process a large number of les smoothly and simply. glob fnmatch The glob and fnmatch modules help a Windows program handle wild-card le names in a standard manner. shutil The shutil module provides shell-like utilities for le copy, le rename, directory moves, etc. This module lets you write short, eective Pytthon programs that do things that are typically done by shell scripts. Why use Python instead of the shell? Python is far easier to read, far more eicient, and far more capable of writing moderately sophisticated programs. Using Python saves you from having to write long, painful shell scripts. 12. Data Persistence. There are several issues related to making objects persistent. In Chapter 12 of the Python Reference, there are several modules that help deal with les in various kinds of formats. Well talk about these modules in detail in File Formats: CSV, Tab, XML, Logs and Others. There are several additional techniques for managing persistence. We can pickle or shelve an object. In this case, we dont dene our le format in detail, instead we leave it to Python to persist our objects. We can map our objects to a relational database. In this case, well use the SQL language to dene our storage, create and retrieve our objects. pickle shelve The pickle and shelve modules are used to create persistent objects; objects that persist beyond the one-time execution of a Python program. The pickle module produces a serial text representation of any object, however complex; this can reconstitute an object from its text representation. The shelve module uses a dbm database to store and retrieve objects. 32.2. Most Useful Library Sections 387
The shelve module is not a complete object-oriented database, as it lacks any transaction management capabilities. sqlite3 This module provides access to the SQLite relational database. This database provides a signicant subset of SQL language features, allowing us to build a relational database thats compatible with products like MySQL or Postgres. 13. Data Compression and Archiving. These modules handle the various le compression algorithms that are available. Well look at these modules in File Handling Modules. tarle ziple These two modules create archive les, which contain a number of les that are bound together. The TAR format is not compressed, where the ZIP format is compressed. Often a TAR archive is compressed using GZIP to create a .tar.gz archive. zlib gzip bz2 These modules emplioye dierent compression algorithms. They all have similar features to compress or uncompress les. 14. File Formats. These are modules for reading and writing les in a few of the amazing variety of le formats that are in common use. In addition to these common formats, modules in chapter 20, Structured Markup Processig Tools are also important. csv The csv module helps you parse and create Comma-Separated Value (CSV) data les. This helps you exchange data with many desktop tools that produce or consume CSV les. Well look at this in Comma-Separated Values: The csv Module. CongParser Conguration les can take a number of forms. The simplest approach is to use a Python module as the conguration for a large, complex program. Sometimes congurations are encoded in XML. Many Windows legacy programs use .INI les. The CongParser can gracefully parse these les. 15. Cryptographic Services. These modules arent specically encryption modules. Many popular encryption algorithms are protected by patents. Often, encryption requires compiled modules for performance reasons. These modules compute secure digests of messages using a variety of algorithms. hashlib Compute a secure hash or digest of a message to ensure that it was not tampered with. The hashlib.md5 class creates an MD5 hash, which is often used for validating that a downloaded le was recieved correctly and completely. 16. Generic Operating System Services. The following modules contain basic features that are common to all operating systems. Most of this commonality is acheived by using the C standard libraries. By using this module, you can be assured that your Python application will be portable to almost any operating system. os The os (and os.path) modules provide access to a number of operating system features. The os module provides control over Processes, Files and Directories. Well look at os and os.path in The os Module and The os.path Module. time The time module provides basic functions for time and date processing. Additionally datetime handles details of the calendar more gracefully than time does. Well cover both modules in detail in Dates and Times: the time and datetime Modules. Having modules like datetime and time mean that you never need to attempt your own calendrical calculations. One of the important lessons learned in the late 90s was that many
388
programmers love to tackle calendrical calculations, but their eorts had to be tested and reworked because of innumerable small problems. getopt optparse A well-written program makes use of the command-line interface. It is congured through options and arguments, as well as properties les. Well cover optparse in Programs: Standing Alone. Command-line programs for Windows will also need to use the glob module to perform standard le-name globbing. logging Often, you want a simple, standardized log for errors as well as debugging information. Well look at logging in detail in Log Files: The logging Module. 17. Optional Operating System Services. This section includes less-common modules for handling threading other features that are more-or-less unavailable in Windows. 18. Interprocess Communication and Networking. This section includes modules for creating processes and doing simple interprocess communication (IPC) using the standard socket abstraction. subprocess The subprocess module provides the class required to create a separate process. The standard approach is called forking a subprocess. Under Windows, similar functionality is provided. Using this, you can write a Python program which can run any other program on your computer. This is very handy for automating complex tasks, and it allows you to replace clunky, diicult shell scripts with Python scripts. socket This is a Python implementation of the standard socket library that supports the TCP/IP protocol. 19. Internet Data Handling. The Internet Data Handling modules contain a number of handy algorithms. A great deal of data is dened by the Internet Request for Comments (RFC). Since these eectively standardize data on the Internet, it helps to have modules already in place to process this standardized data. Most of these modules are specialized, but a few have much wider application. mimify base64 binascii binhex quopri uu These modules all provide various kinds of conversions, ecapes or quoting so that binary data can be manipulated as safe, universal ASCII text. The number of these modules reects the number of dierent clever solutions to the problem of packing binary data into ordinary email messages. 20. Structured Markup Processing Tools. The following modules contain algorithms for working with structured markup: Standard General Markup Lanaguage (SGML), Hypertext Markup Language (HTML) and Extensible Markup Language (XML). These modules simplify the parsing and analysis of complex documents. In addition to these modules, you may also need to use the CSV module for processing les; thats in chapter 9, File Formats. htmllib Ordinary HTML documents can be examined with the htmllib module. This module based on the sgmllib module. The basic HTMLParser class denition is a superclass; you will typically override the various functions to do the appropriate processing for your application.
389
One problem with parsing HTML is that browsers in order to conform with the applicable standards must accept incorrect HTML. This means that many web sites publish HTML which is tolerated by browsers, but cant easily be parsed by htmllib. When confronted with serious horrors, consider downloading the Beautiful Soup module (http://www.crummy.com/software/BeautifulSoup/). This handles erroneous HTML more gracefully than htmllib. xml.sax xml.dom xml.dom.minidom The xml.sax and xml.dom modules provide the classes necessary to conveniently read and process XML documents. A SAX parser separates the various types of content and passes a series of events the handler objects attached to the parser. A DOM parser decomposes the document into the Document Object Model (DOM). The xml.dom module contains the classes which dene an XML documents structure. The xml.dom.minidom module contains a parser which creates a DOM object. Additionally, the formatter module, in chapter 24 (Miscellaneous Modules) goes along with these. 21. Internet Protocols and Support. The following modules contain algorithms for responding the several of the most common Internet protocols. These modules greatly simplify developing applications based on these protocols. cgi The cgi module can be used for web server applications invoked as Common Gateway Interface (CGI) scripts. This allows you to put Python programming in the cgi-bin directory. When the web server invokes the CGI script, the Python interpreter is started and the Python script is executed. wsgiref The Web Services Gateway Interface (WSGI) standard provides a much simpler framework for web applications and web services. See PEP 333 for more information. Essentially, this subsumes all of CGI, plus adds several features and a systematic way to compose larger applications from smaller components. urllib urllib2 urlparse These modules allow you to write relatively simple application programs which open a URL as if it were a standard Python le. The content can be read and perhaps parsed with the HTML or XML parser modules, described below. The urllib module depends on the httplib, ftplib and gopherlib modules. It will also open local les when the scheme of the URL is file:. The urlparse module includes the functions necessary to parse or assemble URLs. The urllib2 module handles more complex situations where there is authentication or cookies involved. httplib ftplib gopherlib The httplib, ftplib and gopherlib modules include relatively complete support for building client applications that use these protocols. Between the html module and httplib module, a simple character-oriented web browser or web content crawler can be built. poplib imaplib The poplib and imaplib modules allow you to build mail reader client applications. The poplib module is for mail clients using the Post-Oice Protocol, POP3 (RFC 1725), to
390
extract mail from a mail server. The imaplib module is for mail servers using the Internet Message Access Protocol, IMAP4 (RFC 2060) to manage mail on an IMAP server. nntplib The nntplib module allows you to build a network news reader. The newsgroups, like comp.lang.python, are processed by NNTP servers. You can build special-purpose news readers with this module. SocketServer The SocketServer module provides the relatively advanced programming required to create TCP/IP or UDP/IP server applications. This is typically the core of a stand-alone application server. SimpleHTTPServer CGIHTPPServer BaseHTTPServer The SimpleHTTPServer and CGIHTTPServer modules rely on the basic BaseHTTPServer and SocketServer modules to create a web server. The SimpleHTTPServer module provides the programming to handle basic URL requests. The CGIHTTPServer module adds the capability for running CGI scripts; it does this with the fork() and exec() functions of the os module, which are not necessarily supported on all platforms. asyncore asynchat The asyncore (and asynchat) modules help to build a time-sharing application server. When client requests can be handled quickly by the server, complex multi-threading and multi-processing arent really necessary. Instead, this module simply dispatches each client communication to an appropriate handler function. 22. Multimedia Services. This is beyond the scope of this book. 23. Internationalization. A well-written application avoids including messages as literal strings within the program text. Instead, all messages, prompts, labels, etc., are kept as a separate resource. These separate string resources can then be translated. locale The locale module fetches the current locales date, time, number and currency formatting rules. This provides functions which will format and parse dates, times, numbers and currency amounts. A user can change their locale with simple operating system settings, and your application can work consistently with all other programs. 24. Program Frameworks. Well talk about a number of program-related issues in Programs: Standing Alone and Architecture: Clients, Servers, the Internet and the World Wide Web. Much of this goes beyond the standard Python library. Within the library are two modules that can help you create large, sophisticated command-line application programs. cmd The cmd module contains a superclass useful for building the main command-reading loop of an interactive program. The standard features include printing a prompt, reading commands, providing help and providing a command history buer. A subclass is expected to provide functions with names of the form do_command(). When the user enters a line beginning with command, the appropriate do_command() function is called. shlex The shlex module can be used to tokenize input in a simple language similar to the Linux shell languages. This module denes a basic shlex class with parsing methods that can separate words, quotes strings and comments, and return them to the requesting program. 25. Graphical User Interfaces with Tk. This is beyond the scope of this book. 26. Development Tools. The testing tools are central to creating reliable, complete and correct software.
391
doctest When a function or a class docstring includes a snippet of interactive Python, the doctest module can use this snippet to conrm that the function or class works as advertised. For example:
def myFunction( a, b ): """>>> myFunction( 2, 3 ) 6 >>> myFunction( 5.0, 7.0 ) 35.0 """ return a * b
The >>> myFunction( 2, 3 ) lines are parsed by doctest. They are evaluated, and the actual result compared with the docstring comment. unittest This is more sophisticated testing framework in which you create TestCases which dene a xture, an operation and expected results. 2to3 This module is used to convert Python 2 les to Python 3. Prior to using this, you should run your Python programs with the -3 option to identify any potential incompatibilities. Once youve xed all of the incompatibilities, you can condently convert your program to Python 3. Do not tweak the output from this conversion. If your converted program doesnt work under Python 3, its almost always a problem with your original program playing fast and loose with Python rules. In the unlikely event that this module cannot convert your program, you should probably rewrite your program to eliminate the features that are causing problems. 27. Debugging and Proling. Debugging is an important skill, as is performance proling. Much of this is beyond the scope of this book. timeit This is a handy module that lets you get timing information to compare performance of alternative implementations of an algorithm. 28. Python Runtime Services. The Python Runtime Services modules are considered to support the Python runtime environment. sys The sys module contains execution context information. It has the command-line arguments (in sys.argv) used to start the Python interpreter. It has the standard input, output and error le denitions. It has functions for retrieving exception information. It denes the platform, byte order, module search path and other basic facts. This is typically used by a main program to get run-time environment information. Most of the remaining sections of the library, with one exception, are too advanced for this book. 1. Custom Python Interpreters 1. Restricted Execution 1. Importing Modules 1. Python Language Services 1. Python compiler package
34. Miscellaneous Services. This is a vague catch-all that only has one module. formatter The formatter module can be used in conjunction with the HTML and XML parsers. A formatter instance depends on a writer instance that produces the nal (formatted) output. It can also be used on its own to format text in dierent ways. 392 Chapter 32. The Python Library
The HTML parser can produce a plain-text version of a web page. To do this, it uses the formatter module.
32.3 Library Exercises

1. Why are there multiple versions of some packages? Look at some places where there are two modules which clearly do the same or almost the same things. Examples include time and datetime, urllib and urllib2, pickle and cPickle, StringIO and cStringIO, subprocess and popen2, getopt and optparse. Why allow this duplication? Why not pick a best module and discard the others? 2. Is it better to build an application around the library or simply design the application and ignore the library? Assuming that we have some clear, detailed requirements, what is the benet of time spent searching through the library? What if most library modules are a near-miss? Should we alter our design to leverage the library, or just write the program without considering the library? 3. Which library modules are deprecated or disabled? Why are these still documented in the library?
32.3. Library Exercises
393
394
CHAPTER
THIRTYTHREE
COMPLEX STRINGS: THE RE MODULE

There are a number of related problems when processing strings. When we get strings as input from les, we need to recognize the input as meaningful. Once were sure its in the right form, we need to parse the inputs, sometimes well ahve to convert some parts into numbers (or other objects) for further use. For example, a le may contain lines which are supposed to be like "Birth Date: 3/8/85" . We may need to determine if a given string has the right form. Then, we may need to break the string into individual elements for date processing. We can accomplish these recognition, parsing and conversion operations with the re module in Python. A regular expression (RE) is a rule or pattern used for matching strings. It diers from the fairly simple wild-card rules used by many operating systems for naming les with a pattern. These simple operating system le-name matching rules are embodied in two simpler packages: fnmatch and glob. Well look at the semantics of a regular expression in Semantics. Well look at the syntax for dening a RE in Creating a Regular Expression. In Using a Regular Expression well put the regular expression to use.
33.1 Semantics
One way to look at regular expressions is as a production rule for constructing strings. In principle, such a rule could describe an innite number of strings. The real purpose is not to enumerate all of the strings described by the production rule, but to match a candidate string against the production rule to see if the rule could have constructed the given string. For example, a rule could be "aba". All strings of the form "aba" would match this simple rule. This rule produces only a single string. Determining a match between a given string and the one string produced by this rule is pretty simple. A more complex rule could be "ab*a". The b* means zero or more copies of b. This rule produces an innite set of strings including "aa", "aba", "abba", etc. Its a little more complex to see if a given string could have been produced by this rule. The Python re module includes Python constructs for creating regular expressions (REs), matching candidate strings against REs, and examining the details of the substrings that match. There is a lot of power and subtlety to this package. A complete treatment is beyond the scope of this book.
395
33.2 Creating a Regular Expression

There are a lot of options and clauses that can be used to create regular expressions. We cant pretend to cover them all in a single chapter. Instead, well cover the basics of creating and using REs. The full set of rules is given in section 8.2.1 Regular Expression Syntax of the Python Library Reference document. Additionally, there are many ne books devoted to this subject. Any ordinary character, by itself, is an RE. Example: "a" is an RE that matches the character a in the candidate string. While trivial, it is critical to know that each ordinary character is a stand-alone RE. Some characters have special meanings. We can escape that special meaning by using a \ in front of them. For example, * is a special character, but \* escapes the special meaning and creates a single-character RE that matches the character *. Additionally, some ordinary characters can be made special with \. For instance \d is any digit, \s is any whitespace character. \D is any non-digit, \S is any non-whitespace character. The character . is an RE that matches any single character. Example: "x.z" is an RE that matches the strings like xaz or xbz, but doesnt match strings like xabz. The brackets, "[...]", create a RE that matches any character between the [ ]s. Example: "x[abc]z" matches any of xaz, xbz or xcz. A range of characters can be specied using a -, for example "x[1-9]z". To include a -, it must be rst or last. ^ cannot be rst. Multiple ranges are allowed, for example "x[A-Za-z]z" . Heres a common RE that matches a letter followed by a letter, digit or _: "[A-Za-z][A-Za-z1-9_]". The modied brackets, "[^...]", create a regular expression that matches any character except those between the [ ]s. Example: "a[^xyz]b" matches strings like a9b and a$b, but dont match axb. As with [ ], a range can be specied and multiple ranges can be specied. A regular expression can be formed from concatenating regular expressions. Example: "a.b" is three regular expressions, the rst matches a, the second matches any character, the third matches b. A regular expression can be a group of regular expressions, formed with ()s. Example: "(ab)c" is a regular expression composed of two regular expressions: "(ab)" (which, in turn, is composed of two REs) and "c". ()s also group REs for extraction purposes. The elements matched within () are remembered by the regular expression processor and set aside in a Match object. A regular expression can be repeated. Several repeat constructs are available: "x*" repeats "x" zero or more times; "x+" repeats "x" 1 or more times; "x?" repeats "x" zero or once. Example: "1(abc)*2" matches 12 or 1abc2 or 1abcabc2, etc. The rst match, against 12, is often surprising; but there are zero copies of abc between 1 and 2. The above (*, + and ?) are called greedy repeats because they will match the longest string of repeats. There are versions which are not greedy: *?, +? will match the shortest string of repeats . The character "^" is an RE that only matches the beginning of the line, "$" is an RE that only matches the end of the line. Example: "^$" matches a completely empty line. Here are some examples.
r"[_A-Za-z][_A-Za-z1-9]*"
396
Chapter 33. Complex Strings: the re Module
Matches a Python identier. This embodies the rule of starting with a letter or _, and containing any number of letters, digits or _s. Note that any number includes 0 occurances, so a single letter or _ is a valid identier.
r"^\s*import\s"
Matches a simple import statement. It matches the beginning of the line with ^ , zero or more whitespace characters with \s* , the sequence of letters import ; and one more whitespace character. This pattern will ignore the rest of the line.
r"^\s*from\s+[_A-Za-z][_A-Za-z1-9]*\s+import\s"
Matches a from module import statement. As with the simple import, it matches the beginning of the line (^), zero or more whitespace characters (\s*), the sequence of letters from, a Python module name, one or more whitespace characters (\s+), the sequence import, and one more whitespace character.
r"(\d+):(\d+):(\d+\.?\d*)"
Matches a one or more digits, a :, one or more digits, a :, and digits followed by optional . and zero or more other digits. For example 20:07:13.2 would match, as would 13:04:05 Further, the ()s group the digit strings for conversion and further processing. The resulting Match object will have four groups. group(0) is the entire match; group(1) will be the rst string of digits.
r"def\s+([_A-Za-z][_A-Za-z1-9]*)\s+$[^)]*$:"
Matches Python function denition lines. It matches the letters def; a string of 1 or more whitespace characters (s); an identier, surrounded by ()s to capture the entire identier as a match. It matches a ( ; weve used ( to escape the meaning of ( and make it an ordinary character. It matches a string of non) characters, which would be the parameter list. The parameter list ends with a ) ; weve used ) to make escape the meaning of ) and make it an ordinary character. Finally, we need tyo see the :.
33.3 Using a Regular Expression

There are several methods which are commonly used with regular expressions. The most common rst step is to compile the RE denition string to make an Pattern object. The resulting Pattern object can then be used to match or search candidate strings. A successful match returns a Match object with details of the matching substring. The re module provides the compile() function. compile(expr ) Create a Pattern object from an RE string, expr. The Pattern is used for all subsequent searching or matching operations. A Pattern has several methods, including match() and search(). Generally, raw string notation (r"pattern") is used to write a RE. This simplies the \s required. Without the raw notation, each \ in the string would have to be escaped by a \ , making it \\ . This rapidly gets cumbersome. There are some other options available for re.compile(), see the Python Library Reference, section 4.2, for more information. The following methods are part of a Pattern object created by the re.compile() function. match(string) Match the candidate string against the compiled regular expression (an instance of Pattern). Matching
33.3. Using a Regular Expression
397
means that the regular expression and the candidate string must match, starting at the beginning of the candidate string. A Match object is returned if there is match, otherwise None is returned. search(string) Search a candidate string for the compiled regular expression (an instance of Pattern). Searching means that the regular expression must be found somewhere in the candidate string. A Match object is returned if the pattern is found, otherwise None is returned. If search() or match() nds the pattern in the candidate string, a Match object is created to describe the part of the candidate string which matched. The following methods are part of a Match object. group(number ) Retrieve the string that matched a particular () grouping in the regular expression. Group zero is a tuple of everything that matched. Group 1 is the material that matched the rst set of ()s. Heres a more complete example. This sample decodes a complex input value into elds and then computes a single result.
>>> import re >>> raw_input= "20:07:13.2" >>> hms_pat = re.compile( r'(\d+):(\d+):(\d+\.?\d*)' ) >>> hms_match= hms_pat.match( raw_input ) >>> hms_match.group( 0, 1, 2, 3 ) ('20:07:13.2', '20', '07', '13.2') >>> >>> h,m,s= map( float, hms_match.group(1,2,3) ) >>> h 20.0 >>> m 7.0 >>> >>> s 13.199999999999999 >>> seconds= ((h*60)+m)*60+s >>> seconds 72433.199999999997
1. The import statement incorporates the re module. 2. The raw_input variable is sample input, perhaps from a le, perhaps from input(). 3. The hms_pat variable is the compiled regular expression object which matches three numbers, using "(d+)", separated by :s. The digit-sequence REs are surround by ()s so that the material that matched is returned as a group. This will lead to four groups: group 0 is everything that matched, groups 1, 2, and 3 are successive digit strings. 4. The hms_match variable is a Match object that indicates success or failure in matching. If hms_match is None, no match occurred. Otherwise, the hms_match.group() method will reveal the individually matched input items. 5. The statement that sets h, m, and s does three things. First is uses hms_match.group() to create a tuple of requested items. Each item in the tuple will be a string, so the map() function is used to apply the built-in float() function against each string to create a tuple of three numbers. Finally, this statement relies on the multiple-assignment feature to set all three variables at once. Finally, seconds is computed as the number of seconds past midnight for the given time stamp.
398
33.4 Regular Expression Exercises

1. Parse Old Stock prices. Create a function that will decode the old-style fractional stock price. The price can be a simple oating point number or it can be a fraction, for example, 4 5/8. Develop two patterns, one for numbers with optional decimal places and another for a number with a space and a fraction. Write a function that accepts a string and checks both patterns, returning the correct decimal price for whole numbers (e.g., 14), decimal prices (e.g., 5.28) and fractional prices (27 1/4). 2. Parse Dates. Create a function that will decode a few common American date formats. For example, 3/18/87 is March 18, 1987. You might want to do 18-Mar-87 as an alternative format. Stick to two or three common formats; otherwise, this can become quite complex. Develop the required patterns for the candidate date formats. Write a function that accepts a string and checks all patterns. It will return the date as a tuple of ( year, month, day ).
33.4. Regular Expression Exercises
399
400
CHAPTER
THIRTYFOUR
DATES AND TIMES: THE TIME AND DATETIME MODULES

When processing dates and times, we have a number of problems. Most of these problems stem from the irregularities and special cases in the units we use to measure time. We generally measure time in a number of compatible as well as incompatible units. For example, weeks, days, hours, minutes and seconds are generally compatible, with the exception of leap-second handling. Months, and years, however are incompatible with days and require sophisticated conversion. Problems which mix month-oriented dates and numbers of days are particularly diicult. The number of days between two dates, or a date which is 90 days in the future are notoriously diicult to compute correctly. We need to represent a point in time, a date, a time of day or a date-time stamp. We need to be able to do arithmetic on this point in time. And, we need to represent this point in time as a properly-punctuated string. The time module contains a number of portable functions needed to format times and dates. The datetime module builds on this to provide a representation that is slightly more convenient for some things. Well look at the denition of a moment in time in Semantics: What is Time? .
34.1 Semantics: What is Time?

The Gregorian calendar is extremely complex. Some of that complexity stems from trying to impose a xed year on the wobbly, irregular orbit of our planet. There are several concesssions required to impose a calendar year with integer numbers of days that will match the astronomial year of approximately 365.2425 days. The Gregorian calendars concession is the periodic addition of a leap day to approximate this fractional day. The error is just under .25, so one leap day each four years gets close to the actual duration of the year. Some additional complexity stems from trying to break the year into a sequence of months. Fixed-length months dont work well because the year is 73 5 days long: there arent many pleasant factors that divide this length. If the year were only 360 days long, wed be able to create xed-length months. The Gregorian calendars concession to the individibility of the year is 12 months of 28, 29, 30 or 31 days in length. We have several systems of units available to us for representing a point in time. Seconds are the least common denominator. We can easily derive hours and minutes from seconds. There are 24 60 60 = 86,400 seconds in a day. (Astronomers will periodically add a leap second, so this is not absolutely true.) We can use seconds as a simple representation for a point in time. We can pick some epoch and represent any other point in time as the number of seconds after the epoch. This makes arithmetic very simple. However, its hard to read; what month contains second number 1,190,805,137?
401
Days are another common denominator in the calendar. There are seven days in a week, and (usually) 86,400 seconds in day, so those conversions are simple. We can pick some epoch and represent any other point in time as the number of days after the epoch. This also makes arithmetic very simple. However, its hard to read; what month contains day number 732,945? Months are serious problem. If we work with the conventional date triple of year, month, and day, we cant compute intervals between dates very well at all. We cant locate a date 90 days in the future without a rather complex algorithm. Local Time. We also to have acknowledge the subtlety of local time and the potential dierences between local standard time and local daylight time (or summer time). Since the clock shifts, some time numbers (1:30 AM, for example) will appear to repeat, this can require the timezone hint to decode the time number. The more general solution is to do all work in UTC. Input is accepted and displayed in the current local time for the convenience of users. This has the advantage of being timezone neutral, and it makes time values monotonically increasing with no confusing repeats of a given time of day during the hour in which the clock is shifted. The time Solution. The time module uses two dierent representations for a point in time, and provides numerous functions to help us convert back and forth between the two representations. A float seconds number. This is the UNIX internal representation for time. (The number is seconds past an epoch; the epoch happens to January 1st, 1970.) In this representation, a duration between points in time is also a float number. A struct_time object. This object has nine attributes for representing a point in time as a Gregorian calendar date and time. Well look at this structure in detail below. In this representation, there is no representation for the duration between points in time. You need to convert back and forth between struct_time and seconds representations. The datetime Solution. The datetime module contain all of the objects and methods required to correctly handle the sometimes obscure rules for the Gregorian calendar. Additionally, it is possible to use date information in the datetime to usefully conver among the worlds calendars. For details on conversions between calendar systems, see Calendrical Calculations [Dershowitz97]. The datetime module has just one representation for a point in time. It assigns an ordinal number to each day. The numbers are based on an epochal date, and algorithms to derive the year, month and day information for that ordinal day number. Similarly, this class provides algorithms to convert a calendar date to an ordinal day number. (Note: the Gregorian calendar was not dened until 1582, all dates before the oicial adoption are termed proleptic. Further, the calendar was adopted at dierent times in dierent countries.) There are four classes in this module that help us handle dates and times in a uniform and correct manner. Well skip the more advanced topic of the datetime.tzinfo class. datetime.time An instance of datetime.time has four attributes: hour, minute, second and microsecond. Additionally, it can also carry an instance of tzinfo which describes the time zone for this time. datetime.date An instance of datetime.date has three attributes: year, month and day. There are a number of methods for creating datetime.date objects, and converting datetime.date objectss to various other forms, like oating-point timestamps, 9-tuples for use with the time module, and ordinal day numbers. datetime.datetime An instance of datetime.datetime combines datetime.date and datetime.time. There are a number of methods for creating datetime.datetime objects, and converting datetime.datetime s to various other forms, like oating-point timestamps, 9-tuples for use with the time module, and ordinal day numbers. datetime.timedelta A datetime.timedelta is the duration between two dates, times or datetimes. It has a value in days, seconds and microseconds. These can be added to or subtracted from dates, times
402
Chapter 34. Dates and Times: the time and datetime Modules
or datetimes to compute new dates, times or datetimes.
34.2 Some Class Denitions

A time.struct_time object behaves like an object as well as a tuple. You can access the attributes of the structure by position as well as by name. Note that this class has no methods of its own; you manipulate these objects using functions in the time module. tm_year The year. This will be a full four digit year, e.g. 1998. tm_mon The month. This will be in the range of 1 to 12. tm_mday The day of the month. This will be in the range of 1 to the number of days in the given month. tm_hour The hour of the day, in the range 0 to 23. tm_min The minutes of the hour, in the range 0 to 59. tm_sec The seconds of the minute, in the range 0 to 61 because leap seconds may be included. Not all platforms support leap seconds. tm_wday The day of the week. This will be in the range of 0 to 6. 0 is Monday, 6 is Sunday. tm_yday The day of the year, in the range 1 to 366. tm_isdst Is the time in local daylight savings time. 0 means that this is standard time; 1 means daylight time. If you are creating this object, you can provide -1; the time.mktime() can then determine DST based on the date and time. Well focus on the datetime.datetime class, since it includes datetime.date and datetime.time. This class has the following attributes. MINYEAR MAXYEAR These two attributes bracket the time span for which datetime works correctly. This is year 1 to year 9999, which covers the forseeable future as well as a proleptic past the predates the invention of the Gregorian calendar in 1582. min max The earliest and laterst representable datetimes. In eect these are datetime(MINYEAR, 1, 1, tzinfo=None) and (MAXYEAR, 12, 31, 23, 59, 59, 999999, tzinfo=None). resolution The smallest dierences between datetimes. timedelta(microseconds=1). This is typically equaly to
year The year. This will be a full four digit year, e.g. 1998. It will always be between MINYEAR and MAXYEAR, inclusive. month The month. This will be in the range 1 to 12. day The day. This will be in the range 1 to the number of days in the given month. hour The hour. This will be in the range 0 to 23. minute The minute. This will be in the range 0 to 59. second The second. This will be in the range 0 to 59. microsecond The microsecond (millionths of a second). This will in the range 0 to 999,999. Some platforms dont have a system clock which is this accurate. However, the SQL standard imposes this resolution on most date time values.
34.2. Some Class Denitions
403
To get fractions of a second, youll compute second+microsecond/1000000. tzinfo The datetime.tzinfo object that was provided to the initial datetime.datetime constructor. Otherwise it will be None.
34.3 Creating a Date-Time

There are two use cases for creating date , time, struct_time or datetime instances. In the simplest case, were asking our operating system for the current date-time or the date-time associated with some resource or event. In the more complex case, we are asking a user for input (perhaps on an interactive GUI, a web form, or reading a le prepared by a person); we are parsing some user-supplied values to see if they are a valid date-time and using that value. From The OS. We often get time from the OS when we want the current time, or we want one of the timestamps associated with a system resource like a le or directory. Heres a sampling of techniques for getting a date-time. time() Returns the current moment in time as a float seconds number. See File Handling Modules for examples of getting le timestamps; these are always a float seconds value. Well often need to convert this to a struct_time or datetime object so that we can provide formatted output for users. The functions time.localtime() or time.gmtime() will convert this value to a struct_time. The class methods datetime.datetime.fromtimestamp(), and datetime.datetime.utcfromtimestamp() will create a datetime object from this time value. Then, we can use time.strftime() or time.asctime() to format and display the time. ctime() Returns a string representation of the current time. These values arent terribly useful for further calculation, but they are handy, standardized timestamp strings. asctime() Returns a string representation of the current time. These values arent terribly useful for further calculation, but they are handy, standardized timestamp strings. localtime() When evaluated with no argument value, this will create a struct_time object from the current time. Since we cant do arithmetic with these values, we often need to convert them to something more useful. This time is converted to localtime using your current locale settings. gmtime() When evaluated with no argument value, this will create a struct_time object from the current time. Since we cant do arithmetic with these values, we often need to convert them to something more useful. The time.mktime() function will convert the struct_time to a float seconds time. We have to use the datetime.datetime constructor to create a datetime from a struct_time. This can be long-winded, it will look like datetime.date( ts.tm_year, ts.tm_month, ts.tm_day ). today() now() utcnow() All of these are class methods of the datetime class; the create a datetime object. The today() function uses the simple time.time() notion of the current moment and returns local time. The 404 Chapter 34. Dates and Times: the time and datetime Modules
now() function may use a higher-precision time, but it will be local time. The utcnow() function uses high-precision time, and returns UTC time, not local time. We cant directly get a float seconds number number from a datetime value. However, we can do arithmetic directly with datetime values, making the float seconds value superous. We can get the struct_time value from a datetime, using the timetuple() or utctimetuple() method functions of the datetime object. Getting Time From A User. Human-readable time information generally has to be parsed from one or more string values. Human-readable time can include any of the endless variety of formats in common use. This will include some combination of years, days, months, hours, minutes and seconds, and timezone names. There are two general approaches to parsing time. In most cases, it is simplest to use datetime.strptime() to parse a string and create a datetime object. In other cases, we can use time.strptime(). In the most extreme case, we have to either use the re module (Complex Strings: the re Module), or some other string manipulation, and then create the date-time object directly. strptime(string, [format] ) This function will use the given format to attempt to parse the input string. If the value doesnt match the format, it will raise a ValueError exception. If the format is not a complete datetime, then defaults are lled in. The default year is 1900, the default month is 1 the default day is 1. The default time values are all zero. Well look at the format string under the time.strftime() function, below. strptime(string, [format] ) This function will use the given format to attempt to parse the input string. If the value doesnt match the format, it will raise a ValueError exception. If the format is not a complete time, then defaults are lled in. The default year is 1900, the default month is 1 the default day is 1. The default time values are all zero. Well look at the format string under the time.strftime() function, below. struct_time(9-tuple) Creates a struct_time from a 9-valued tuple: (year, month, day, hour, minute, second, day of week, day of year, dst-flag). Generally, you can supply 0 for day of week and day of year. The dst ag is 0 for standard time, 1 for daylight (or summer) time, and -1 when the date itself will dene if the time is standard or daylight. This constructor does no validation; it will tolerate invalid values. If we use the time.mktime() function to do a conversion, this may raise an OverflowError if the time value is invalid. Typically, youll build this 9-tuple from user-supplied inputs. We could parse a string using the re module, or we could be collecting input from elds in a GUI or the values entered in a web-based form. Then you attempt a time.mktime() conversion to see if it is valid. datetime(year, month, day, [hour, minute, second, microsecond, timezone] ) Creates a datetime from individual parameter values. Note that the time elds are optional; if omitted the time value is 0:00:00, which is midnight. This constructor will not tolerate a bad date. It will raise a ValueError for an invalid date.
34.4 Date-Time Calculations and Manipulations

There are two common classes of date-time calculations:
34.4. Date-Time Calculations and Manipulations
405
Duration or interval calculations in days (or seconds), where the month, week and year boundaries dont matter. The time representation as a single oating-point number of seconds works well for this. Also, the datetime provides a datetime.timedelta that works well for this. Calendar calculations where the month, week of month and day of week matter. The time representation as a 9-element struct_time structure works well for this. Generally, we use datetime.datetime for this. Duration or interval calculations in days (or seconds). In this case, the month, week and year boundaries dont matter. For example, the number of hours, days or weeks between two dates doesnt depend on months or year boundaries. Similarly, calculating a date 90 days in the future or past doesnt depen on month or year considerations. Even the dierence between two times is properly a date-time calculation so that we can allow for rollover past midnight. We have two ways to do this. We can use float seconds information, as produced by the time module. When were using this representation, a day is 24 60 60 = 86, 400 seconds, and a week is 7 24 60 60 = 604, 800 seconds. For the following examples, t1 and t2 and float seconds times. (t2-t1)/3600 is the number of hours between two times. (t2-t1)/86400 is the days between two dates. t1+90*86400 is the date 90 days in the future. We can also use datetime objects for this, since datetime objects correctly handle arithmetic operations. When were using this representation, well also work with datetime.timedelta objects; these have days, seconds and microseconds attributes. For the following examples, t1 and t2 and datetime objects. In a relatively simple case, the hours between two datetimes is (t2-t1).seconds/3600. This works when were sure that there sill be less than 24 hours between the two datetimes. In the more general case, we have a two-part calculation: First we get the timedelta between two datetimes with td = t2-t1. td= (t2-t1); seconds= td.days*86400+td.seconds is seconds between two dates. td= (t2-t1); secdonds= td.days*86400+td.seconds+td.microseconds/1000000.0 is seconds down to the datetime.resolution. td= (t2-t1); seconds= td.days+td.seconds/86400.0 is days between two dates. Calendar calculations where the month, week of month and day of week matter. For example, the number of months between two dates rarely involves the day of the month. A date that is 3 months in the future, will land on the same day of the month, except in unusual cases where it would be the 30th of February. For these situations, highly problem-specic rules have to be applied; theres no general principle. We have two ways to do this. We can use struct_time objects as produced by the time module. We can replace any struct_time elds, and possibly create an invalid date. We may need to use time.mktime() to validate the resulting struct_time object. In the following examples, t1 is a struct_time object. Adding some offset in months, correctly allowing for year-end rollover, is done as follows.
monthSequence= (t1.tm_year*12 + t1.tm_mon-1) + offset futureYear, futureMonth0 = divmod( monthSequence, 12 ) t1.tm_year= futureYear t1.tm_month= futureMonth0 + 1
406
We can also use datetime objects for this. In this case, we well use the replace() method to replace a value in a datetime with other values. In the following examples, t1 is a datetime object. Adding some offset in months, correctly allowing for year-end rollover, is done as follows.
monthSequence= (t1.year*12 + t1.month-1) + offset futureYear, futureMonth0 = divmod( monthSequence, 12 ) t1= t1.replace( year=futureYear, month=futureMonth0+1 )
The following methods return information about a given datetime object. In the following denitions, dt is a datetime object. date() Return a new date object from the date elds of this datetime object. time() Return a new time object from the time elds of this datetime object. replace([year, month, day, hour, minute, second, microsecond] ) Return a new datetime object from the current datetime object after replacing any values provided by the keyword arguments. toordinal() Return the ordinal day number for this datetime. This a unique day number. weekday() Return the day of the week. Monday = 0, Sunday = 6. isoweekday() Return the ISO day of the week. Monday = 1, Sunday = 7. isocalendar() Return a tuple with ( ISO year, ISO week, ISO week day ).
34.5 Presenting a Date-Time

To format human-readable time, we have a number of functions in the time module, and methods of a datetime object. Here are the functions in the time module. strftime(format, time_struct ) Convert a struct_time to a string according to a format specication. The specication rules are provided below. This is an example of how to produce a timestamp with the fewest implicit assumptions.
time.strftime( "%x %X ", time.localtime( time.time() ) )
This line of code shows a standardized and portable way to produce a time stamp. The time.time() function produces the current time in UTC (Coordinated Universal Time). Time is represented as a oating point number of seconds after an epoch. asctime(time_struct ) Convert a struct_time to a string, e.g. Sat Jun 06 16:26:11 1998. This is the same as a the format string "%a %b %d %H:%M:%S %Y". ctime(seconds) Convert a time in seconds since the Epoch to a string in local time. asctime(localtime(seconds)). This is equivalent to
34.5. Presenting a Date-Time
407
A datetime object has the following methods for producting formatted output. In the following denitions, dt is a datetime object. isoformat([separator] ) Return string representing this date in ISO 8601 standard format. The separator string is used between the date and the time; the default value is T.
>>> d= datetime.datetime.now() >>> d.isoformat() '2009-11-08T09:34:17.945641'
ctime() Return string representing this date and time. This is equivalent to time.ctime(time.mktime( dt.timetuple())) for some datetime object, dt. strftime(format ) Return string representing this date and time, formatted using the given format string. This is equivalent to time.strftime( format, time.mktime( dt.timetuple() ) for some datetime object, dt.
34.6 Formatting Symbols

The time.strftime() and time.strptime() functions use the following formatting symbols to convert between time_struct or datetime.datetime and strings. Formatting symbols like %c, %x and %X produce standard formats for whole date-time stamps, dates or time. Other symbols format parts of the date or time value. The following examples show a particular date (Satuday, August 4th) formatted with each of the formatting strings.
408
Table 34.1: Time Formatting Symbols "%a" Locales "%A" Locales "%b" Locales "%B" Locales "%c" Locales 3-letter abbreviated weekday name. full weekday name. 3-letter abbreviated month name. full month name. appropriate full date and time representation. Sat Saturday Aug August Saturday August 04 17:11:20 2001 04 17 05 216 08 11 pm 20 30
"%d" Day of the month as a 2-digit decimal number. "%H" Hour (24-hour clock) as a 2-digit decimal number. "%I" Hour (12-hour clock) as a 2-digit decimal number. "%j" Day of the year as a 3-digit decimal number. "%m" Month as a 2-digit decimal number. "%M" Minute as a 2-digit decimal number. "%p" Locales equivalent of either AM or PM. "%S" Second as a 2-digit decimal number. "%U" Week number of the year (Sunday as the rst day of the week) as a 2-digit decimal number. All days in a new year preceding the rst Sunday are considered to be in week 0. "%w" Weekday as a decimal number, 0 = Sunday. "%W" Week number of the year (Monday as the rst day of the week) as a 2-digit decimal number. All days in a new year preceding the rst Monday are considered to be in week 0. "%x" Locales appropriate date representation. "%X" Locales appropriate time representation. "%y" Year without century as a 2-digit decimal number. "%Y" Year with century as a decimal number. "%Z" Time zone name (or if no time zone exists). "%%" A literal % character.
6 31
Saturday August 04 2001 17:11:20 01 2001 %
34.7 Time Exercises

1. Return on Investment. Return on investment (ROI) is often stated on an annual basis. If you buy and sell stock over shorter or longer periods of time, the ROI must be adjusted to be a full years time period. The basic calculation is as follows: Given the sale date, purchase date, sale price, sp, and purchase price, pp. Compute the period the asset was held: use time.mktime() to create oating point time values for sale date, s, and purchase date, p. The weighting, w, is computed as
w = (86400*365.2425) / ( s - p )
Write a program to compute ROI for some ctitious stock holdings. Be sure to include stocks held both more than one year and less than one year. See Stock Valuation in Classes for some additional information on this kind of calculation. 2. Surface Air Consumption Rate. When doing SACR calculations (see Surface Air Consumption Rate, and Dive Logging and Surface Air Consumption Rate) weve treated the time rather casually. In the event of a night dive that begins before midnight and ends after midnight the next day, our quick and dirty time processing doesnt work correctly.
34.7. Time Exercises
409
Revise your solution to use a more complete date-time stamp for the start and end time of the dive. Use the time module to parse those date-time stamps and compute the actual duration of the dive.
34.8 Additional time Module Features

Here are some additional functions in the time module. sleep(seconds) Delay execution for a given number of seconds. The argument may be a oating point number for subsecond precision. accept2dyear(arg) If non-zero, 2-digit years are accepted. 69-99 is treated as 1969 to 1999, 0 to 68 is treated as 2000 to 2068. This is 1 by default, unless the PYTHONY2K environment variable is set; then this variable will be zero. timezone Dierence in seconds between UTC and local time. Often a multiple of 3600 (all US timezones are in whole hours). For example, if your locale is US Eastern Time, this is 18000 (5 hours). altzone Dierence in seconds between UTC and local alternate time (Daylight Savings Time). Often a multiple of 3600 (US time zones are in whole hours). For example, if your locale was set to US Eastern Time, this would be 14400 (4 hours). daylight Non-zero if the current locale uses daylight savings time. Zero if it does not. tzname The name of the timezone.
410
CHAPTER
THIRTYFIVE
FILE HANDLING MODULES

There are a number of operations closely related to le processing. Deleting and renaming les are examples of operations that change the directory information that the operating system maintains to describe a le. Python provides numerous modules for these operating system operations. We cant begin to cover all of the various ways in which Python supports le handling. However, we can identify the essential modules that may help you avoid reinventing the wheel. Further, these modules can provide you a view of the Pythonic way of working with data from les. The following modules have features that are essential for supporting le processing. Well cover selected features of each module that are directly relevant to le processing. Well present these in the order youd nd them in the Python library documentation. Chapter 11 File and Directory Access. Chapter 11 of the Library reference covers many modules which are essential for reliable use of les and directories. Well look closely at the following modules. os.path This module has functions for numerous common pathname manipulations. Use these functions to split and join full directory path names. This is operating-system neutral, with a correct implementation for all operating systems. One of the most obvious dierences among operating systems is the way that les are named. In particular, the path separator can be either the POSIX standard /, or the windows \. Rather than make each program aware of the operating system rules for path construction, Python provides the os.path module to make all of the common lename manipulations completely consistent. OS-Specic Paths Programmers are faced with a dilemma between writing a simple hack to strip paths or extensions from le names and using the os.path module. Some programmers argue that the os.path module is too much overhead for such a simple problem as removing the .html from a le name. Other programmers recognize that most hacks are a false economy: in the long run they do not save time, but rather lead to costly maintenance when the program is expanded or modied. fileinput This module has functions which will iterate over lines from multiple input streams. This allows you to write a single, simple loop that processes lines from any number of input les. tempfile This module has a class and functions to generate temporary les and temporary le names. This is the most secure way to generate a temporary le. glob This module provides shell style pathname pattern expansion. Unix shells translate name patterns like *.py into a list of les. This is called globbing. The glob module implements this within Python, which allows this feature to work even in Windows where it isnt supported by the OS itself.
411
fnmatch This module provides UNIX shell style lename pattern matching. This implements the glob-style rules using *, ? and []. * matches any number of characters, ? matches any single character, [chars]. encloses a list of allowed characters, [!chars] encloses a list of disallowed characters. shutil This module has usefule, high-level le operations, including copy, rename and remove. These are the kinds of things that the shell handles with simple commands like cp or rm. This module makes these features available to a Python program or script. Chapter 12 Data Persistence. There are several issues related to making objects persistent. Well look at these modules in detail in File Formats: CSV, Tab, XML, Logs and Others. pickle shelve The pickle and shelve modules are used to create persistent objects; objects that persist beyond the one-time execution of a Python program. The pickle module produces a serial text representation of any object, however complex; this can reconstitute an object from its text representation. The shelve module uses a dbm database to store and retrieve objects. The shelve module is not a complete object-oriented database, as it lacks any transaction management capabilities. sqlite3 This module provides access to the SQLite relational database. This database provides a signicant subset of SQL language features, allowing us to build a relational database thats compatible with products like MySQL or Postgres. Chapter 13 Data Compression and Archiving. Data Compression is covered in Chapter 12 of the Library referece. Well look closely at the following modules. tarfile zipfile These modules helps you read and write archive les; les which are an archive of a complex directory structure. This includes GNU/Linux tape archive (.tar) les, compressed GZip tar les (.tgz les or .tar.gz les) sometimes called tarballs, and ZIP les. zlib gzip bz2 These modules are all variations on a common theme of reading and writing les which are compressed to remove redundant bytes of data. The zlib and bz2 modules have a more sophisticated interface, allowing you to use compression selectively within a more complex application. The gzip module has a dierent (and simpler) interface that only applies only to complete les. Chapter 14 File Formats. These are modules for reading and writing les in a few of the amazing variety of le formats that are in common use. Well focus on just a few. csv The csv module helps you parse and create Comma-Separated Value (CSV) data les. This helps you exchange data with many desktop tools that produce or consume CSV les. Well look at this in Comma-Separated Values: The csv Module. Chapter 16 - Generic Operating System Services. The following modules contain basic features that are common to all operating systems os Miscellaneous OS interfaces. This includes parameters of the current process, additional le object creation, manipulations of le descriptors, managing directories and les, managing subprocesses, and additional details about the current operating system. time The time module provides basic functions for time and date processing. Additionally datetime handles details of the calendar more gracefully than time does. Well cover both modules in detail in Dates and Times: the time and datetime Modules. logging Often, you want a simple, standardized log for errors as well as debugging information. Well look at logging in detail in Log Files: The logging Module. Chapter 28 - Python Runtime Services. These modules described in Chapter 26 of the Library reference include some that are used for handling various kinds of les. Well look closely as just one. sys This module has several system-specic parameters and functions, including denitions of the three standard les that are available to every program.
412
Chapter 35. File Handling Modules
35.1 The os.path Module

The os.path module contains more useful functions for managing path and directory names. A serious mistake is to use ordinary string functions with literal stringsfor the path separators. A Windows program using \ as the separator wont work anywhere else. A less serious mistake is to use os.pathsep instead of the routines in the os.path module. The os.path module contains the following functions for completely portable path and lename manipulation. basename(path) Return the base lename, the second half of the result created by os.path.split( path ). dirname(path) Return the directory name, the rst half of the result created by os.path.split( path ). exists(path) Return True if the pathname refers to an existing le or directory. getatime(path) Return the last access time of a le, reported by os.stat(). See the time module for functions to process the time value. getmtime(path) Return the last modication time of a le, reported by os.stat(). See the time module for functions to process the time value. getsize(path) Return the size of a le, in bytes, reported by os.stat(). isdir(path) Return True if the pathname refers to an existing directory. isfile(path) Return True if the pathname refers to an existing regular le. join(string, ...) Join path components using the appropriate path separator. split(path) Split a path into two parts: the directory and the basename (the lename, without path separators, in that directory). The result (s, t) is such that os.path.join( s, t ) yields the original path. splitdrive(path) Split a pathname into a drive specication and the rest of the path. Useful on DOS/Windows/NT. splitext(path) Split a path into root and extension. The extension is everything starting at the last dot in the last component of the pathname; the root is everything before that. The result (r, e) is such that r+e yields the original path. The following example is typical of the manipulations done with os.path.
import sys, os.path def process( oldName, newName ): Some Processing... for oldFile in sys.argv[1:]: dir, fileext= os.path.split(oldFile) file, ext= os.path.splitext( fileext ) if ext.upper() == '.RST':
35.1. The os.path Module
413
newFile= os.path.join( dir, file ) + '.HTML' print oldFile, '->', newFile process( oldFile, newFile )
1. This program imports the sys and os.path modules. 2. The process() function does something interesting and useful to the input le. It is the real heart of the program. 3. The for statement sets the variable oldFile to each string (after the rst) in the sequence sys.argv. 4. Each le name is split into the path name and the base name. The base name is further split to separate the le name from the extension. The os.path does this correctly for all operating systems, saving us having to write platform-specic code. For example, splitext() correctly handles the situation where a linux le has multiple .s in the le name. 5. The extension is tested to be .RST. A new le name is created from the path, base name and a new extension (.HTML). The old and new le names are printed and some processing, dened in the process(), uses the oldFile and newFile names.
35.2 The os Module

The os module contains an interface to many operating system-specic functions to manipulate processes, les, le descriptors, directories and other low level features of the OS. Programs that import and use os stand a better chance of being portable between dierent platforms. Portable programs must depend only on functions that are supported for all platforms (e.g., unlink() and opendir()), and perform all pathname manipulation with os.path. The os module exports the following variables that characterize your operating system. name A name for the operating system, for example 'posix', 'nt', 'dos', 'os2', 'mac', or 'ce'. Note that Mac OS X has an os.name of 'posix'; but sys.platform is 'darwin'. Windows XP has an os.name of 'nt'. curdir The string which represents the current directory ('.' , generally). pardir The string which represents the parent directory ('..' , generally). sep altsep The (or a most common) pathname separator ('/' or ':' or '\') and the alternate pathname separator (None or '/'). Most of the Python library routines will translate '/' to the correct value for the operating system (typically, '\' on Windows.) It is best to always use os.path rather than these low-level constants. pathsep The component separator used in the OS environment variable $PATH (':' is the standard, ';' is used for Windows). linesep The line separator in text les ('n' is the standard; 'rn' is used for Windowss). This is already part of the readlines() function.
414
defpath The default search path for executables. The standard is ':/bin:/usr/bin' or '.;C:\bin' for Windows. The os module has a large number of functions. Many of these are not directly related to le manipulation. However, a few are commonly used to create and remove les and directories. Beyond these basic manipulations, the shutil module supports a variety of le copy operations. chdir(path) Change the current working directory to the given path. This is the directory which the OS uses to transform a relative le name into an absolute le name. getcwd() Returns the path to the current working directory. This is the directory which the OS use to transform a relative le name into an absolute le name. listdir(path) Returns a list of all entries in the given directory. mkdir(path, [mode] ) Create the given directory. In GU/Linux, the mode can be given to specify the permissions; usually this is an octal number. If not provided, the default of 0777 is used, after being updated by the OS umask value. rename(source, destination) Rename the source lename to the destination lename. There are a number of errors that can occur if the source le doesnt exist or the destination le already exists or if the two paths are on dierent devices. Each OS handles the situations slightly dierently. remove(le) Remove (also known as delete or unlink) the le. If you attempt to remove a directory, this will raise OSError. If the le is in use, the standard behavior is to remove the le when it is nally closed; Windows, however, will raise an exception. rmdir(path) Remove (also known as delete or unlink) the directory. if you attempt to remove an ordinary le, this will raise OSError. Heres a short example showing some of the functions in the os module.
>>> import os >>> os.chdir("/Users/slott") >>> os.getcwd() '/Users/slott' >>> os.listdir(os.getcwd()) ['.bash_history', '.bash_profile', '.bash_profile.pysave', '.DS_Store', '.filezilla', '.fonts.cache-1', '.fop', '.idlerc', '.isql.history', '.lesshst', '.login_readpw', '.monitor', '.mozilla', '.sh_history', '.sqlite_history', '.ssh', '.subversion', '.texlive2008', '.Trash', '.viminfo', '.wxPyDemo', '.xxe', 'Applications', 'argo.user.properties', 'Burn Folder.fpbf', 'Desktop', 'Documents', 'Downloads', 'Library', 'Movies', 'Music', 'Pictures', 'Public', 'Sites']
35.2. The os Module
415
35.3 The fileinput Module

The fileinput module interacts with sys.argv. The fileinput.input() function opens les based on all the values of sys.argv[1:]. It carefully skips sys.argv[0], which is the name of the Python script le. For each le, it reads all of the lines as text, allowing a program to read and process multiple les, like many standard Unix utilities. The typical use case is:
import fileinput for line in fileinput.input(): process(line)
This iterates over the lines of all les listed in sys.argv[1:], with a default of sys.stdin if the list is empty. If a lename is - it is also replaced by sys.stdin at that position in the list of les. To specify an alternative list of lenames, pass it as the argument to input(). A single le name is also allowed in addition to a list of le names. While processing input, several functions are available in the fileinput module: input() Iterator over all lines from the cumulative collection of les. filename() The lename of the line that has just been read. lineno() The cumulative line number of the line that has just been read. filelineno() The line number in the current le. isfirstline() True if the line just read is the rst line of its le. isstdin() True if the line was read from sys.stdin. nextfile() Close the current le so that the next iteration will read the rst line from the next le (if any); lines not read from the le will not count towards the cumulative line count; the lename is not changed until after the rst line of the next le has been read. close() Closes the iterator. All les are opened in text mode. If an I/O error occurs during opening or reading a le, the IOError exception is raised. Example. We can use leinput to write a Python version of the common Unix utility, grep. The grep utility searches a list of les for a given pattern. For non-Unix users, the grep utility looks for the given regular expression in any number of les. The name grep is an acronym of Global Regular Expression Print. This is similar to the Windows command nd.
416
greppy.py
#!/usr/bin/env python import sys import re import fileinput pattern= re.compile( sys.argv[1] ) for line in fileinput.input(sys.argv[2:]): if pattern.match( line ): print fileinput.filename(), fileinput.filelineno(), line
1. The sys module provides access to the command-line parameters. The re module provides the pattern matching. The fileinput module makes searching an arbitrary list of les simple. For more information on re, see Complex Strings: the re Module. 2. The rst command line argument ( sys.argv[0] ) is the name of the script, which this program ignores. The second command-line argument is the pattern that denes the target of the search. The remaining command-line arguments are given to fileinput.input() so that all the named les will be examined. 3. The pattern regular expression is matched against each individual input line. If match() returns None, the line did not match. If match() returns an object, the program prints the current le name, the current line number of the le and the actual input line that matched. After we do a chmod +x greppy.py, we can use this program as follows. Note that we have to provide quotes to prevent the shell from doing globbing on our pattern string.
$ greppy.py 'import.*random' *.py demorandom.py 2 import random dice.py 1 import random functions.py 2 import random
Windows Windows users will be disappointed because the GNU/Linux shell languages all handle le wild-card processing (globbing) automatically. The shell uses the le-name patterns to create a complete list of le names that match the pattern to the application. Windows does not supply lists of le names that match patterns to application programs. Therefore, we have to use the glob module to transform sys.argv[2:] from a pattern to lists of les. Also, Windows users will have to use "" around the pattern, where Unix and Mac OS shell users will typically use '' . This is a dierence between the Unix shell quoting rules and the Windows quoting rules.
35.4 The glob and fnmatch Modules

The glob module adds a necessary Unix shell capability to Windows programmers. The glob module includes the following function
35.4. The glob and fnmatch Modules
417
glob(wildcard ) Return a list of lenames that match the given wild-card pattern. The fnmatch module is used for the wild-card pattern matching. A common use for glob is something like the following under Windows.
import glob, sys for arg in sys.argv[1:]: for f in glob.glob(arg): process( f )
This makes Windows programs process command line arguments somewhat like Unix programs. Each argument is passed to glob.glob() to expand any patterns into a list of matching les. If the argument is not a wild-card pattern, glob simply returns a list containing this one le name. The fnmatch module has the algorithm for actually matching a wild-card pattern against specic le names. This module implements the Unix shell wild-card rules. These are not the same as the more sophisticated regular expression rules. The module contains the following function: fnmatch(lename, pattern) Return True if the lename matches the pattern. The patterns use * to match any number of characters, ? to match any single character. [letters] matches any of these letters, and [!letters] matches any letter that is not in the given set of letters.
>>> import fnmatch >>> fnmatch.fnmatch('greppy.py','*.py') True >>> fnmatch.fnmatch('README','*.py') False
35.5 The tempfile Module

One common problem is to open a unique temporary le to hold intermediate results. The tempfile module provides us a handy way to create temporary les that we can write to and read from. The tempfile module creates a temporary le in the most secure and reliable manner. The tempfile module includes an internal function, mkstemp() which dioes the hard work of creating a unique temporary le. TemporaryFile(mode=w+b, bufsize=-1, suix=, prex=tmp, dir=None) This function creates a le which is automatically deleted when it is closed. All of the parameters are optional. By default, the mode is 'w+b': write with read in binary mode. The bufsize parameter has the same meaning as it does for the built-in open() function. The keyword parameters suffix, prefix and dir provide some structure to the name assigned to the le. The suffix should include the dot, for example suffix='.tmp'. NamedTemporaryFile(mode=w+b, bufsize=-1, suix=, prex=tmp, dir=None, delete=True) This function is similar to TemporaryFile() ; it creates a le which can be automatically deleted when it is closed. The temporary le, however, is guaranteed to be visible on the le system while the le is open. The parameters are the same as a tempfile.TemporaryFile(). If the delete parameter is False, the le is not automatically deleted.
418
mkstemp(suix=, prex=tmp, dir=None, text=False) -> ( fd, name) This function does the essential job of creating a temporary le. It returns the interanl le descriptor as well as the name of the le. The le is not automatically deleted. If necessary, the le created by this function can be explicitly deleted with os.remove().
import tempfile, os fd, tempName = tempfile.mkstemp( '.d1' ) temp= open( tempName, 'w+' ) Some Processing...
This fragment will create a unique temporary le name with an extension of .d1. Since the name is guaranteed to be unique, this can be used without fear of damaging or overwriting any other le.
35.6 The shutil Module

The shutil module helps you automate copying les and directories. This saves the steps of opening, reading, writing and closing les when there is no actual processing, simply moving les. copy(source, destination) Copy data and mode bits; basically the unix command cp src dst. If destination is a directory, a le with the same base name as source is created. If destination is a full le name, this is the destination le. copyfile(source, destination) Copy data from source to destination. Both names must be les. copytree(source, destination) Recursively copy the entire directory tree rooted at source to destination. destination must not already exist. Errors are reported to standard output. rmtree(path) Recursively delete a directory tree rooted at path. Note that removing a le is done with os.remove() (or os.unlink()). This module allows us to build Python applications that are like shell scripts. There are a lot of advantages to writing Python programs rather than shell scripts to automate mundane tasks. First, Python programs are easier to read than shell scripts. This is because the language did not evolve in way that emphasized tersness; the shell script languages use a minimum of punctuation, which make them hard to read. Second, Python programs have a more sophisticated programming model, with class denitions, and numerous sophisticated data structures. The shell works with simple argument lists; it has to resort to running the test or expr programs to process string s or numbers. Finally, Python programs have direct access to more of the operating systems features than the shell. Generally, many of the basic GNU/Linux API calls are provided via innumerable small programs. Rather than having the shell run a small program to make an API call, Python can simply make the API call.
35.7 The File Archive Modules: tarfile and zipfile

An archive le contains a complex, hierarchical le directory in a single sequential le. The archive le includes the original directory information as well as a the contents of all of the les in those directories. There are a number of archive le formats, Python directory supports two: tar and zip archives.
35.6. The shutil Module
419
The tar (Tape Archive) format is widely used in the GNU/Linux world to distribute les. It is a POSIX standard, making it usable on a wide variety of operating systems. A tar le can also be compressed, often with the GZip utility, leading to .tgz or .tar.gz les which are compressed archives. The Zip le format was invented by Phil Katz at PKWare as a way to archive a complex, hierarchical le directory into a compact sequential le. The Zip format is widely used but is not a POSIX standard. Zip le processing includes a choice of compression algorithms; the exact algorithm used is encoded in the header of the le, not in the name of le. Creating a TarFile or a ZipFile. Since an archive le is still essentially a single le, it is opened with a variation on the open() function. Since an archive le contains directory and le contents, it has a number of methods above and beyond what a simple le has. open(name=None, mode=r, leobj=None, bufsize=10240 ) This module-level function opens the given tar le for processing. The name is a le name string; it is optional because the leobj can be used instead. The leobject is a conventional le object, which can be used instead of the name; it can be a standard le like sys.stdin. The buersize is like the built-in open() function. The mode is similar to the built-in open() function; it has numerous additional characters to specify the compression algorithms, if any. ZipFile(le, mode=r, compression=ZIP_STORED, allowZip64=False) This class constructor opens the given zip le for processing. The name is a le name string. The mode is similar to the built-in open() function. The compression is the compression code. It can be zipfile.ZIP_STORED or zipfile.ZIP_DEFLATED. A compression of ZIP_STORED uses no compression; a value of ZIP_DEFLATED uses the Zlib compression algorithms. The allowZip64 option is used when creating new, empty Zip Files. If this is set to True, then this will create les with the ZIP64 extensions. If this is left at False, any time a ZIP64 extension would be required will raise an exception. The open function can be used to read or write the archive le. It can be used to process a simple disk le, using the lename. Or, more importantly, it can be used to process a non-disk le: this includes tape devices and network sockets. In the non-disk case, a le object is given to tarfile.open(). For tar les, the mode information is rather complex because we can do more than simply read, write and append. The mode string adresses three issues: the kind of opening (reading, writing, appending), the kind of access (block or stream) and the kind of compression. For zip les, however, the mode is simply the kind of opening that is done. Opening - Both zip and tar les. A zip or tar le can be opened in any of three modes. r Open the le for reading only. w Open the le for writing only. a Open an existing le for appending. Access - tar les only. A tar le can have either of two fundamentally dierent kinds of access. If a tar le is a disk le, which supports seek and tell operations, then you we access the tar le in block mode. If the tar le is a stream, network connection or a pipeline, which does not support seek or tell operations, then we must access the archive in stream mode. : Block mode. The tar le is an disk le, and seek and tell operations are supported. This is the assumed default, if neither : or | are specied. | Stream mode. The tar le is a stream, socket or pipeline, and cannot respond to seek or tell operations. Note that you cannot append to a stream, so the 'a|' combination is illegal. This access distinction isnt meaningful for zip les.
420
Compression - tar les only. A tar le may be compressed with GZip or BZip2 algorithms, or it may be uncompressed. Generally, you only need to select compression when writing. It doesnt make sense to attempt to select compression when appending to an existing le, or when reading a le. (nothing) The tar le will not be compressed. gz The tar le will be compressed with GZip. bz2 The tar le will be compressed with BZip2. This compression distinction isnt meaningful for zip les. zipfile.ZipFile constructor. Zip le compression is specied in the
Tar File Examples. The most common block modes for tar les are r, a, w:, w:gz, w:bz2. Note that read and append modes cannot meaningfully provide compression information, since its obvious from the le if it was compressed, and which algorithm was used. For stream modes, however, the compression information must be provided. The modes include all six combinations: r|, r|gz, r|bz2, w|, w|gz, w|bz2. Directory Information. Each individual le in a tar archive is described with a TarInfo object. This has name, size, access mode, ownership and other OS information on the le. A number of methods will retrieve member information from an archive. getmember(name) Reads through the archive index looking for the given member name. Returns a TarInfo object for the named member, or raises a KeyError exception. getmembers() Returns a list of TarInfo objects for all of the members in the archive. next() Returns a TarInfo object for the next member of the archive. getNames() Returns a list of member names. Each individual le in a zip archive is described with a ZipInfo object. This has name, size, access mode, ownership and other OS information on the le. A number of methods will retrieve member information from an archive. getinfo(name) Locates information about the given member name. Returns a ZipInfo object for the named member, or raises a KeyError exception. infolost() Returns a list of ZipInfo objects for all of the members in the archive. namelist() Returns a list of member names. Extracting Files From an Archive. If a tar archive is opened with r, then you can read the archive and extract les from it. The following methods will extract member les from an archive. extract(member, [path] ) The member can be either a string member name or a TarInfo for a member. This will extract the les contents and reconstruct the original le. If path is given, this is the new location for the le. extractfile(member ) The member can be either a string member name or a TarInfo for a member. This will open a simple le for access to this members contents. The member access le has only read-oriented methods, limited to read(), readline(), readlines(), seek(), tell().
35.7. The File Archive Modules: tarfile and zipfile
421
If a zip archive is opened with r, then you can read the archive and extract the contents of a le from it. read(member ) The member is a string member name. This will extract the members contents, decompress them if necessary, and return the bytes that consitute the member. Creating or Extending an Archive. If a tar archive is opened with w or a, then you can add les to it. The following methods will add member les to an archive. add(name, [arcname, recursive=True] ) Adds the le with the given name to the current archive le. If arcname is provided, this is the name the le will have in the archive; this allows you to build an archive which doesnt reect the source structure. Generally, directories are expanded; using recursive=False prevents expanding directories. addfile(tarinfo, leobj ) Creates an entry in the archive. The description comes from the tarinfo, an instance of TarInfo, created with the gettarinfo() function. The leobj is an open le, from which the content is read. Note that the TarInfo.size eld can override the actual size of the le. For a given lename, fn, this might look like the following: tf.addfile( tf.gettarinfo(fn), open(fn,"r") ). close() Closes the archive. For archives being written or appended, this adds the block of zeroes that denes the end of the le. gettarinfo(name, [arcname, leobj] ) Creates a TarInfo object for a le based either on name , or the leobj. If a name is given, this is a local lename. The arcname is the name that will be used in the archive, allowing you to modify local lesystem names. If the leobj is given, this le is interrogated to gather required information. If a zip archive is opened with w or a, then you can add les to it. write(lename, [arcname, compress] ) The lename is a string le name. This will read the le, compress it, and write it to the archive. If the arcname is given, this will be the name in the archive; otherwise it will use the original lename. The compress parameter overrides the default compression specied when the ZipFile was created. writestr(arcname, bytes) The arcname is a string le name or a ZipInfo object that will be used to create a new member in the archive. This will write the given bytes to the archive. The compression used is specied when the ZipFile is created. A tarfile Example. Heres an example of a program to examine a tarle, looking for documentation like .html les or README les. It will provide a list of .html les, and actually show the contents of the README les.
readtar.py
#!/usr/bin/env python """Scan a tarfile looking for *.html and a README.""" import tarfile import fnmatch archive= tarfile.open( "SQLAlchemy-0.3.5.tar.gz", "r" ) for mem in archive.getmembers(): if fnmatch.fnmatch( mem.name, "*.html" ): print mem.name elif fnmatch.fnmatch( mem.name.upper(), "*README*" ): print mem.name
422
docFile= archive.extractfile( mem ) print docFile.read()
A zipfile Example. Heres an example of a program to create a ziple based on the .xml les in a particular directory.
writezip.py
import zipfile, os, fnmatch bookDistro= zipfile.ZipFile( 'book.zip', 'w', zipfile.ZIP_DEFLATED ) for nm in os.listdir('..'): if fnmatch.fnmatch(nm,'*.xml'): full= os.path.join( '..', nm ) bookDistro.write( full ) bookDistro.close()
35.8 The sys Module

The sys module provides access to some objects used or maintained by the interpreter and to functions that interact strongly with the interpreter. The sys module also provides the three standard les used by Python. sys.stdin Standard input le object; used by raw_input() and input(). Also available via sys.stdin.read() and related methods of the le object. sys.stdout Standard output le object; used by the print statement. sys.stdout.write() and related methods of the le object. Also available via
sys.stderr Standard error object; used for error messages, typically unhandled exceptions. Available via sys.stderr.write() and related methods of the le object. A program can assign another le object to one of these global variables. When you change the le for these globals, this will redirect all of the interpreters I/O. Command-Line Arguments. One important object made available by this module is the variable sys.argv. This variable has the command line arguments used to run this script. For example, if we had a python script called portfolio.py, and executed it with the following command:
python portfolio.py -xvb display.csv
Then the sys.argv list would be ["portfolio.py", "-xvb", "display.csv"]. Sophisticated argument processing is done with the optparse module. A few other interesting objects in the sys module are the following variables. sys.version The version of this interpreter as a string. For example, '2.6.3 (r263:75184, Oct 2 2009, 07:56:03) n[GCC 4.0.1 (Apple Inc. build 5493)]' sys.version_info Version information as a tuple, for example: (2, 6, 3, 'final', 0). sys.hexversion Version information encoded as a single integer. Evaluating hex(sys.hexversion) yields '0x20603f0'. Each byte of the value is version information. sys.copyright Copyright notice pertaining to this interpreter. 35.8. The sys Module 423
sys.platform Platform identier, for example, 'darwin', 'win32' or 'linux2'. sys.prex File Path prex used to nd the Python library, for example '/usr', '/Library/Frameworks/Python.framework/Versions/2.5' or 'c:\Python25'.
35.9 Additional File-Processing Modules

There are several other chapters of the Python Library Reference that cover with even more le formats. Well identify them briey here. Chapter 7 Internet Data Handling. Reading and processing les of Internet data types is very common. Internet data types have formal denitions governed by the internet standards, called Requests for Comment (RFCs). The following modules are for handling Internet data structures. These modules and the related standards are beyond the scope of this book. email Helps you handle email MIME attachments. mailcap Mailcap le handling. mailbox Read various mailbox formats. mhlib Manipulate MH mailboxes from Python. mimetools Tools for parsing MIME-style message bodies. mimetypes Mapping of lename extensions to MIME types. MimeWriter Generic MIME le writer. mimify Mimication and unmimication of mail messages. multifile Support for reading les which contain distinct parts, such as some MIME data. rfc822 Parse RFC 822 style mail headers. base64 Encode and decode les using the MIME base64 data. binhex Encode and decode les in binhex4 format. binascii Tools for converting between binary and various ASCII-encoded binary representations. quopri Encode and decode les using the MIME quoted-printable encoding. uu Encode and decode les in uuencode format. Chapter 13 - Data Persistence. Many Python programs will also deal with Python objects that are exported from memory to external les or retrieved from les to memory. Since an external le is more persistent than the volatile working memory of a computer, this process makes an object persistent or retrieves a persistent object. One mechanism for creating a persistent object is called serialization, and is supported by several modules, which are beyond the scope of this book. pickle Convert Python objects to streams of bytes and back. cPickle Faster version of pickle, but not subclassable. copy_reg Register pickle support functions. shelve Python object persistence. marshal Convert Python objects to streams of bytes and back (with dierent constraints). More complex le structures can be processed using the standard modules available with Python. The widelyused DBM database manager is available, plus additional modules are available on the web to provide ODBC
424
access or to connect to a platform-specic database access routine. The following Python modules deal with these kinds of les. These modules are beyond the scope of this book. anydbm Generic interface to DBM-style database modules. whichdb Guess which DBM-style module created a given database. dbm The standard database interface, based on the ndbm library. gdbm GNUs reinterpretation of dbm. dbhash DBM-style interface to the BSD database library. bsddb Interface to Berkeley DB database library dumbdbm Portable implementation of the simple DBM interface. Additionally, Python provides a relational database module. sqlite3 A very pleasant, easy-to-use relational database (RDBMS). This handles a wide variety of SQL statements.
35.10 File Module Exercises

1. Source Lines of Code. One measure of the complexity of an application is the count of the number of lines of source code. Often, this count discards comment lines. Well write an application to read Python source les, discarding blank lines and lines beginning with #, and producing a count of source lines. Well develop a function to process a single le. Well use the glob module to locate all of the *.py les in a given directory. Develop a fileLineCount( name )() which opens a le with the given name and examines all of the lines of the le. Each line should have strip() applied to remove leading and trailing spaces. If the resulting line is of length zero, it was eectively blank, and can be skipped. If the resulting line begins with # the line is entirely a comment, and can be skipped. All remaining lines should be counted, and fileLineCount() returns this count. Develop a directoryLineCount( path )() function which uses the path with the glob.glob() to expand all matching le names. Each le name is processed with fileLineCount( name )() to get the number of non-comment source lines. Write this to a tab-delimited le; each line should have the form filenametlines. For a sample application, look in your Python distribution for Lib/idelib/*.py. 2. Summarize a Tab-Delimited File. The previous exercise produced a le where each line has the form filenametlines. Read this tab-delimited le, producing a nicer-looking report that has column titles, le and line counts, and a total line count at the end of the report. 3. File Processing Pipeline. The previous two exercises produced programs which can be part of a processing pipeline. The rst exercise should p should produce its output on sys.stdout. The second exercise should gather its input from sys.stdin. Once this capability is in place, the pipeline can be invoked using a command like the following:
$ python lineCounter.py | python lineSummary.py
35.10. File Module Exercises
425
426
CHAPTER
THIRTYSIX
FILE FORMATS: CSV, TAB, XML, LOGS AND OTHERS

We looked at general features of the le system in Files. In this chapter well look at Python techniques for handling les in a few of the innumeraable formats that are in common use. Most le formats are relatively easy to handle with Python techniques weve already seen. Comma-Separated Values (CSV) les, XML les and packed binary les, however, are a little more sophisticated. This only the tip of the iceberg in the far larger problem called persistence. In addition to simple le system persistence, we also have the possibility of object persistence using an object database. In this case, the databse processing lies between our program and the le system on which the database resides. This area also includes object-relational mapping, where our program relies on a mapper; the mapper uses to database, and the database manages the le system. We cant explore the whole persistence problem in this chapter. In this chapter well present a conceptual overview of the various approaches to reading and writing les in Overview . Well look at reading and writing CSV les in Comma-Separated Values: The csv Module, tab-delimited les in Tab Files: Nothing Special . Well look reading property les in Property Files and Conguration (or .INI ) Files: The CongParser Module. Well look at the subleties of processing legacy COBOL les in Fixed Format Files, A COBOL Legacy: The codecs Module. Well cover the basics of reading XML les in XML Files: The xml.etree and xml.sax Modules. Most programs need a way to write sophisticated, easy-to-control log les what contain status and debugging information. For simple one-page programs, the print statement is ne. As soon as we have multiple modules, where we need more sophisticated debugging, we nd a need for the logging module. Of course, any program that requires careful auditing will benet from the logging module. Well look at creating standard logs in Log Files: The logging Module.
36.1 Overview
When we introduced the concept of le we mentioned that we could look at a le on two levels. A le is a sequence of bytes. This is the OSs view of views, as it is the lowest-common denominator. A le is a sequence of data objects, represented as sequences of bytes. A le format is the processing rules required to translate between usable Python objects and sequences of bytes. People have invented innumerable distinct le formats. Well look at some techniques which should cover most of the bases. Well look at three broad families of les: text, binary and pickled objects. Each has some advantages and processing complexities.
427
Text les are designed so that a person can easily read and write them. Well look at several common text le formats, including CSV, XML, Tab-delimited, property-format, and xed position. Since text les are intended for human consumption, they are diicult to update in place. Binary les are designed to optimize processing speed or the overall size of the le. Most databases use very complex binary le formats for speed. A JPEG le, on the other hand, uses a binary format to minimize the size of the le. A binary-format le will typically place data at known osets, making it possible to do direct access to any particular byte using the seek() method of a Python le object. Pickled Objects are produced by Pythons pickle or shelve modules. There are several pickle protocols available, including text and binary alternatives. More importantly, a pickled le is not designed to be seen by people, nor have we spent any design eort optimizng performace or size. In a sense, a pickled object requires the least design eort.
36.2 Comma-Separated Values: The csv Module

Often, we have data that is in Comma-Separated Value (CSV) format. This used by many spreadsheets and is a widely-used standard for data les. In Reading a CSV File the Hard Way we parsed CSV les using simple string manipulations. The csv module does a far better job at parsing and creating CSV les than the programming we showed in those examples. About CSV Files. CSV les are text les organized around data that has rows and columns. This format is used to exchange data between spread-sheet programs or databases. A CSV le uses a number of punctuation rules to encode the data. Each row is delimited by a line-ending sequence of characters. This is usually the ASCII sequence rn. Since this may not be the default way to process text les on your platform, you have to open les using the rb and wb modes. Within a row, columns are delimited by a ,. To handle the situation where a columns data contains a ,, the column data may be quoted; surrounded by " characters. If the column contains a ", there are two common rules used. One CSV dialect uses an escape character, usually \". The other dialect uses double "". In the ideal case, a CSV le will have the same number of columns in each row, and the rst row will be column titles. Almost as pleasant is a le without column titles, but with a known sequence of columns. In the more complex cases, the number of columns per row varies. The csv Module. The CSV module provides you with readers or writers; these are objects which use an existing file object, created with the file() or open() function. A CSV reader will read a le, parsing the commas and quotes, delivering you the data elements of each row in a sequence or mapping. A CSV writer will create a le, adding the necessary commas and quotes to create a valid CSV le. The following constructors within the csv module are used to create a reader, DictReader, writer or DictWriter. reader(csvle) Creates a reader object which can parse the given le, returning a sequence of values for each line of the le. The csvfile can be any iterable object. This can be used as follows.
rdr= csv.reader( open( "file.csv", "rb" ) ) for row in rdr: print row
428
Chapter 36. File Formats: CSV, Tab, XML, Logs and Others
writer(csvle) Creates a writer object which can format a sequence of values and write them to a line of the le. The csvfile can be any object which supports a write() method. This can be used as follows.
target= open( "file.csv", "wb" ) wtr= csv.writer( target ) wtr.writerow( ["some","list","of","values"] ) target.close()
Its very handy to use the with statement to assure that the le is properly closed.
with open( "file.csv", "wb" ) as target: wtr= csv.writer( target ) wtr.writerow( ["some","list","of","values"] )
DictReader(csvle, [eldnames] ) Creates a DictReader object which can parse the given le, returning a dictionary of values for each line of the le. The dictionary keys are typically the rst line of the le. You can, optionally, provide the eld names if they are not the rst line of the le. The csvfile can be any iterable object. DictWriter(csvle, [eldnames] ) Creates a DictWriter object which can format a dictionary of values and write them to a line of the le. You must provide a sequence of eld names which is used to format each individual dictionary entry. The csvfile can be any object which supports a write() method. Reader Functions. The following functions within a reader (or DictReader) object will read and parse the CSV le. Writer Functions. The following functions with a writer (or DictWriter) object will format and write a CSV le. Basic CSV Reading Example. The basic CSV reader processing treats each line of the le as data. This is typical for les which lack column titles, or les which have such a complex format that special parsing and analysis is required. In some cases, a le has a simple, regular format with a single row of column titles, which can be processed by a special reader well look at below. Well revise the readquotes.py program from Reading a CSV File the Hard Way. This will properly handle all of the quoting rules, eliminating a number of irritating problems with the example in the previous chapter.
readquotes2.py
import csv qFile= file( "quotes.csv", "rb" ) csvReader= csv.reader( qFile ) for q in csvReader: try: stock, price, date, time, change, opPrc, dHi, dLo, vol = q print stock, float(price), date, time, change, vol except ValueError: pass qFile.close()
1. We open our quotes le, quotes.csv, for reading, creating an object named qFile.
36.2. Comma-Separated Values: The csv Module
429
2. We create a csv.reader object which will parse this le for us, transforming each line into a sequence of individual column values. 3. We use a for statement to iterate through the sequence of lines in the le. 4. In the unlikely event of an invalid number for the price, we surround this with a try statement. The invalid number line will raise a ValueError exception, which is caught in the except clause and quietly ignored. 5. Each stock quote, q, is a sequence of column values. We use multiple assignment to assign each eld to a relevant variable. We dont need to strip whitespace, split the string, or handle quotes; the reader already did this. 6. Since the price is a string, we use the float() function to convert this string to a proper numeric value for further processing. Column Headers as Dictionary Keys In some cases, you have a simple, regular le with a single line of column titles. In this case, you can transform each line of the le into a dictionary. The key for each eld is the column title. This can lead to programs which are more clear, and more exible. The exibility comes from not assuming a specic order to the columns. Well revise the readportfolio.py program from Reading Records . This will properly handle all of the quoting rules, eliminating a number of irritating problems with the example in the previous chapter. It will make use of the column titles in the le.
readportfolio2.py
import csv quotes=open( "display.csv", "rb" ) csvReader= csv.DictReader( quotes ) invest= 0 current= 0 for data in csvReader: print data invest += float(data["Purchase Price"])*float(data["# Shares"]) current += float(data["Price"])*float(data["# Shares"]) print invest, current, (current-invest)/invest
1. We open our portfolio le, display.csv, for reading, creating a le object named quotes. 2. We create a csv.DictReader object from our quotes le. This will read the rst line of the le to get the column titles; each subsequent line will be parsed and transformed into a dictionary. 3. We initialize two counters, invest and current to zero. These will accumulate our initial investment and the current value of this portfolio. 4. We use a for statement to iterate through the lines in quotes le. Each line is parsed, and the column titles are used to create a dictionary, which is assigned to data. 5. Each stock quote, q, is a string. We use the strip() operation to remove excess whitespace characters; the string which is created then performs the split() method to separate the elds into a list. We assign this list to the variable values. 6. We perform some simple calculations on each dict. In this case, we convert the purchase price to a number, convert the number of shares to a number and multiply to determine how much we spent on this stock. We accumulate the sum of these products into invest. We also convert the current price to a number and multiply this by the number of shares to get the current value of this stock. We accumulate the sum of these products into current.
430
7. When the loop has terminated, we can write out the two numbers, and compute the percent change. Writing CSV Files The most general case for writing CSV is shown in the following example. Assume weve got a list of objects, named someList. Further, lets assume that each object has three attributes: this, that and aKey.
import csv myFile= open( " :replaceable:`result` ", "wb" ) wtr= csv.writer( myFile ) for someData in :replaceable:`someList` : aRow= [ someData.this, someData.that, someData.aKey, ] wtr.writerow( aRow ) myFile.close()
In this case, we assemble the list of values that becomes a row in the CSV le. In some cases we can provide two methods to allow our classes to participate in CSV writing. We can dene a csvRow() method as well as a csvHeading() method. These methods will provide the necessary tuples of heading or data to be written to the CSV le. For example, lets look at the following class denition for a small database of sailboats. This class shows how the csvRow() and csvHeading() methods might look.
class Boat( object ): csvHeading= [ "name", "rig", "sails" ] def __init__( aBoat, name, rig, sails ): self.name= name self.rig= rig self.sails= sails def __str__( self ): return "%s (%s, %r )" % ( self.name, self.rig, self.sails ) def csvRow( self ): return [ self.name, self.rig, self.sails ]
Including these methods in our class denitions simplies the loop that writes the objects to a CSV le. Instead of building each row as a list, we can do the following: wtr.writerow( someData.csvRow() ) . Heres an example that leverages each objects internal dictionary (__dict__) to dump objects to a CSV le.
db= [ Boat( "KaDiMa", "sloop", ( "main", "jib" ) ), Boat( "Glinda", "sloop", ( "main", "jib", "spinnaker" ) ), Boat( "Eillean Glas", "sloop", ( "main", "genoa" ) ), ] test= file( "boats.csv", "wb" ) wtr= csv.DictWriter( test, Boat.csvHeading ) wtr.writerow( dict( zip( Boat.csvHeading, Boat.csvHeading ) ) ) for d in db: wtr.writerow( d.__dict__ ) test.close()
36.3 Tab Files: Nothing Special

Tab-delimited les are text les organized around data that has rows and columns. This format is used to exchange data between spread-sheet programs or databases. A tab-delimited le uses just rwo punctuation 36.3. Tab Files: Nothing Special 431
rules to encode the data. Each row is delimited by an ordinary newline character. This is usually the standard n. If you are exchanging les across platforms, you may need to open les for reading using the rU mode to get universal newline handling. Within a row, columns are delimited by a single character, often t. The column punctuation character that is chosen is one that will never occur in the data. It is usually (but not always) an unprintable character like t. In the ideal cases, a CSV le will have the same number of columns in each row, and the rst row will be column titles. Almost as pleasant is a le without column titles, but with a known sequence of columns. In the more complex cases, the number of columns per row varies. When we have a single, standard punctuation mark, we can simply use two operations in the string and list classes to process les. We use the split() method of a string to parse the rows. We use the join() method of a list to assemble the rows. We dont actually need a separate module to handle tab-delimited les. Reading. The most general case for reading Tab-delimited data is shown in the following example.
myFile= open( " :replaceable:`somefile` ", "rU" ) for aRow in myFile: print aRow.split('\t') myFile.close()
Each row will be a list of column values. Writing. The writing case is the inverse of the reading case. Essentially, we use a "t".join( someList ) to create the tab-delimeted row. Heres our sailboat example, done as tab-delimited data.
test= file( "boats.tab", "w" ) test.write( "\t".join( Boat.csvHeading ) ) test.write( "\n" ) for d in db: test.write( "\t".join( map( str, d.csvRow() ) ) ) test.write( "\n" ) test.close()
Note that some elements of our data objects arent string values. In this case, the value for sails is a tuple, which needs to be converted to a proper string. The expression map(str, someList ) applies the str() function to each element of the original list, creating a new list which will have all string values. See Sequence Processing Functions: map(), lter() and reduce().
36.4 Property Files and Conguration (or .INI ) Files: ConfigParser Module
The
A property le, also known as a conguration (or .INI) le denes property or conguration values. It is usually just a collection of settings. The essential property-le format has a simple row-oriented format with only two values in each row. A conguration (or .INI) le organizes a simple list of properties into one or more named sections. A property le uses a few punctuation rules to encode the data. Lines begining with # or ; are ignored. In some dialects the comments are # and !.
432
Each property setting is delimited by an ordinary newline character. This is usually the standard n. If you are exchanging les across platforms, you may need to open les for reading using the rU mode to get universal newline handling. Each property is a simple name and a value. The name is a string characters that does not use a separator character of : or =. The value is everything after the punctuation mark, with leading and trailing spaces removed. In some dialects space is also a separator character. Some property le dialects allow a value to continue on to the next line. In this case, a line that ends with \ (the cwo-character sequence \ \n) escapes the usual meaning of \n. Rather being the end of a line, \n is demoted to just another whitespace character. A property le is an extension to the basic tab-delimited le. It has just two columns per line, and some space-stripping is done. However, it doesnt have a consistent separator, so it is slightly more complex to parse. The extra feature introduced in a conguration le is named sections. A line beginning with [, ending with ], is the beginning of a section. The []s surround the section name. All of the lines from here to the next section header are collected together. Reading a Simple Property File. Heres an example of reading the simplest kind of property le. In this case, well turn the entire le into a dictionary. Python doesnt provide a module for doing this. The processing is a sequence string manipulations to parse the le.
propFile= file( r"C:\Java\jdk1.5.0_06\jre\lib\logging.properties", "rU" ) propDict= dict() for propLine in propFile: propDef= propLine.strip() if len(propDef) == 0: continue if propDef[0] in ( '!', '#' ): continue punctuation= [ propDef.find(c) for c in ':= ' ] + [ len(propDef) ] found= min( [ pos for pos in punctuation if pos != -1 ] ) name= propDef[:found].rstrip() value= propDef[found:].lstrip(":= ").rstrip() propDict[name]= value propFile.close() print propDict print propDict['handlers']
The input line is subject to a number of processing steps. 1. First the leading and trailing whitespace is removed. If the line is empty, nothing more needs to be done. 2. If the line begins with ! or # ( ; in some dialects) it is ignored. 3. We nd the location of all relevant puncuation marks. In some dialects, space is not permitted. Note that we through the length of the line on the end to permit a single word to be a valid property name, with an implicit value of a zero-length string. 4. By discarding punction positions of -1, we are only processing the positions of punctuation marks which actually occur in the string. The smallest of these positions is the left-most punctuation mark. 5. The name is everything before the punctuation mark with whitespace remove. 6. The value is everything after the punctuaion mark. Any additional separators are removed, and any trailing whitespace is also removed.
36.4. Property Files and Conguration (or .INI ) Files: The ConfigParser Module
433
Reading a Cong File. The ConfigParser module has a number of classes for processing conguration les. You initialize a CongParse object with default values. The object can the read one or more a conguration les. You can then use methods to determine what sections were present and what options were dened in a given section.
import ConfigParser cp= ConfigParser.RawConfigParser( ) cp.read( r"C:\Program Files\Mozilla Firefox\updater.ini" ) print cp.sections() print cp.options('Strings') print cp.get('Strings','info')
Eschewing Obfuscation. While a property le is rather simple, it is possible to simplify property les further. The essential property denition syntax is so close to Pythons own syntax that some applications use a simple le of Python variable settings. In this case, the settings le would look like this.
settings.py
# Some Properties TITLE = "The Title String" INFO = """The information string. Which uses Python's ordinary techniques for long lines."""
This le can be introduced in your program with one statement: import settings . This statement will create module-level variables, settings.TITLE and settings.INFO.
36.5 Fixed Format Files, A COBOL Legacy: The codecs Module

Files that come from COBOL programs have three characteristic features: The le layout is dened positionally. There are no delimiters or separators on which to base le parsing. The le may not even have n characters at the end of each record. Theyre usually encoded in EBCDIC, not ASCII or Unicode. They may include packed decimal elds; these are numeric values represented with two decimal digits (or a decimal digit and a sign) in each byte of the eld. The rst problem requires guring the starting position and size of each eld. In some cases, there are no gaps (or ller) between elds; in this case the sizes of each eld are all that are required. Once we have the position and size, however, we can use a string slice operation to pick those characters out of a record. The code is simply aLine[start:start+size]. We can tackle the second problem using the codecs module to decode the EBCDIC characters. The result of codecs.getdecoder('cp037') is a function that you can use as an EBCDIC decoder. The third problem requires that our program know the data type as well as the position and oset of each eld. If we know the data type, then we can do EBCDIC conversion or packed decimal conversion as appropriate. This is a much more subtle algorithm, since we have two strategies for converting the data elds. See Strategy for some reasons why wed do it this way. In order to mirror COBOLs largely decimal world-view, we will need to use the decimal module for all numbers and airthmetic.
434
We note that the presence of packed decimal data changes the le from text to binary. Well begin with techniques for handling a text le with a xed layout. However, since this often slides over to binary le processing, well move on to that topic, also. Reading an All-Text File. If we ignore the EBCDIC and packed decimal problems, we can easily process a xed-layout le. The way to do this is to dene a handy structure that denes our record layout. We can use this structure to parse each record, transforming the record from a string into a dictionary that we can use for further processing. In this example, we also use a generator function, yieldRecords(), to break the le into individual records. We separate this functionality out so that our processing loop is a simple for statement, as it is with other kinds of les. In principle, this generator function can also check the length of recBytes before it yields it. If the block of data isnt the expected size, the le was damaged and an exception should be raised.
layout = [ ( 'field1', 0, 12 ), ( 'field2', 12, 4 ), ( 'anotherField', 16, 20 ), ( 'lastField', 36, 8 ), ] reclen= 44 def yieldRecords( aFile, recSize ): recBytes= aFile.read(recSize) while recBytes: yield recBytes recBytes= aFile.read(recSize) cobolFile= file( 'my.cobol.file', 'rb' ) for recBytes in yieldRecords(cobolFile, reclen): record = dict() for name, start, size in layout: record[name]= recBytes[start:start+len]
Reading Mixed Data Types. If we have to tackle the complete EBCDIC and packed decimal problem, we have to use a slightly more sophisticated structure for our le layout denition. First, we need some data conversion functions, then we can use those functions as part of picking apart a record. We may need several conversion functions, depending on the kind of data thats present in our le. Minimally, well need the following two functions. display(bytes) This function is used to get character data. In COBOL, this is called display data. It will be in EBCDIC if our les originated on a mainframe.
def display( bytes ): return bytes
packed(bytes) This function is used to get packed decimal data. In COBOL, this is called COMP-3 data. In our example, we have not dealt with the insert of the decimal point prior to the creation of a decimal.Decimal object.
import codecs display = codecs.getdecoder('cp037') def packed( bytes ): n= [ '' ] for b in bytes[:-1]: hi, lo = divmod( ord(b), 16 )
36.5. Fixed Format Files, A COBOL Legacy: The codecs Module
435
n.append( str(hi) ) n.append( str(lo) ) digit, sign = divmod( ord(bytes[-1]), 16 ) n.append( str(digit) ) if sign in (0x0b, 0x0d ): n[0]= '-' else: n[0]= '+' return n
Given these two functions, we can expand our handy record layout structure.
layout = [ ( 'field1', 0, 12, display ), ( 'field2', 12, 4, packed ), ( 'anotherField', 16, 20, display ), ( 'lastField', 36, 8, packed ), ] reclen= 44
This changes our record decoding to the following.

cobolFile= file( 'my.cobol.file', 'rb' ) for recBytes in yieldRecords(cobolFile, reclen): record = dict() for name, start, size, convert in layout: record[name]= convert( recBytes[start:start+len] )
This example underscores some of the key values of Python. Simple things can be kept simple. The layout structure, which describes the data, is both easy to read, and written in Python itself. The evolution of this example shows how adding a sophisticated feature can be done simply and cleanly. At some point, our record layout will have to evolve from a simple tuple to a proper class denition. Well need to take this evolutionary step when we want to convert packed decimal numbers into values that we can use for further processing.
36.6 XML Files: The xml.etree and xml.sax Modules

XML les are text les, intended for human consumption, that mix markup with content. The markup uses a number of relatively simple rules. Additionally, there are structural requirements that assure that an XML le has a minimal level of validity. There are additional rules (either a Document Type Dention, DTD, or an XML Schema Denition, XSD) that provide additional structural rules. There are several XML parsers available with Python. xml.expat Well ignore this parser, not for any particularly good reason. xml.sax Well look at the SAX parser because it provides us a way to break gigantic XML les into more manageable chunks. xml.dom This is a document object model (DOM) for XML. xml.minidom This is a stripped-down implementation of the XML document object model, along with a parser to build the document objects from XML. xml.pulldom This module uses SAX to create a document objects from a portion of a larger XML document.
436
xml.etree This is a more useful DOM-oriented parser that allows sophisticated XPATH-like searching through the resulting document objects. xml.sax Parsing. The Standard API for XML (SAX) parser is described as an event parser. The parser recognizes dierent elements of an XML document and invokes methods in a handler which you provide. Your handler will be given pieces of the document, and can do appropriate processing with those pieces. For most XML processing, your program will have the following outline: This parser will then use your ContentHandler as it parses. 1. Dene a subclass of xml.sax.ContentHandler. The methods of this class will do your unique processing will happen. 2. Request the module to create an instance of an xml.sax.Parser. 3. Create an instance of your handler class. Provide this to the parser you created. 4. Set any features or options in the parser. 5. Invoke the parser on your document (or incoming stream of data from a network socket). Heres a short example that shows the essentials of building a simple XML parser with the xml.sax module. This example denes a simple ContentHandler that prints the tags as well as counting the occurances of the <informaltable> tag.
import xml.sax class DumpDetails( xml.sax.ContentHandler ): def __init__( self ): self.depth= 0 self.tableCount= 0 def startElement( self, aName, someAttrs ): print self.depth*' ' + aName self.depth += 1 if aName == 'informaltable': self.tableCount += 1 def endElement( self, aName ): self.depth -= 1 def characters( self, content ): pass # ignore the actual data p= xml.sax.make_parser() myHandler= DumpDetails() p.setContentHandler( myHandler ) p.parse( "../p5-projects.xml" ) print myHandler.tableCount, "tables"
Since the parsing is event-driven, your handler must accumulate any context required to determine where the individual tags occur. In some content models (like XHTML and DocBook) there are two levels of markup: structural and semantic. The structural markup includes books, parts, chapters, sections, lists and the like. The semantic markup is sometimes called inline markup, and it includes tags to identify function names, class names, exception names, variable names, and the like. When processing this kind of document, youre application must determine the which tag is which. A ContentHandler Subclass. The heart of a SAX parser is the subclass of ContentHandler that you dene in your application. There are a number of methods which you may want to override. Minimally, youll override the startElement() and characters() methods. There are other methods of this class described in section 20.10.1 of the Python Library Reference. setDocumentLocator(locator ) The parser will call this method to provide an xml.sax.Locator object. This object has the XML
36.6. XML Files: The xml.etree and xml.sax Modules
437
document ID information, plus line and column information. The locator will be updated within the parser, so it should only be used within these handler methods. startDocument() The parser will call this method at the start of the document. It can be used for initialization and resetting any context information. endDocument() This method is paired with the startDocument() method; it is called once by the parser at the end of the document. startElement(name, attrs) The parser calls this method with each tag that is found, in non-namespace mode. The name is the string with the tag name. The attrs parameter is an xml.sax.Attributes object. This object is reused by the parser; your handler cannot save this object. The xml.sax.Attributes object behaves somewhat like a mapping. It doesnt support the [] operator for getting values, but does support get(), has_key(), items(), keys(), and values() methods. endElement(name) The parser calls this method with each tag that is found, in non-namespace mode. The name is the string with the tag name. startElementNS(name, qname, attrs) The parser calls this method with each tag that is found, in namespace mode. You set namesace mode by using the parsers p.setFeature( xml.sax.handler.feature_namespaces, True ). The name is a tuple with the URI for the namespace and the tag name. The qname is the fully qualied text name. The attrs is described above under ContentHandler.startElementNS(). endElementNS(name, qname) The parser calls this method with each tag that is found, in namespace mode. The name is a tuple with the URI for the namespace and the tag name. The qname is the fully qualied text name. characters(content ) The parser uses this method to provide character data to the ContentHandler. The parser may provide character data in a single chunk, or it may provide the characters in several chunks. ignorableWhitespace(whitespace) The parser will use this method to provide ignorable whitespace to the ContentHandler. This is whitespace between tags, usually line breaks and indentation. The parser may provide whitespace in a single chunk, or it may provide the characters in several chunks. processingInstructions(target, data) The parser will provide all <? target data ?> processing instructions to this method. Note that the initial <?xml version="1.0" encoding="UTF-8"?> is not reported. xml.etree Parsing. The Document Object Model (DOM) parser creates a document object model from your XML document. The parser transforms the text of an XML document into a DOM object. Once your program has the DOM object, you can examine that object. Heres a short example that shows the essentials of building a simple XML parser with the xml.etree module. This example locates all instances of the <informaltable> tag in the XML document and prints parts of this tags content.
#!/usr/bin/env python from xml.etree import ElementTree
438
dom1 = ElementTree.parse("../PythonBook-2.5/p5-projects.xml") for t in dom1.getiterator("informaltable"): print t.attrib for row in t.find('thead').getiterator('tr'): print "head row" for header_col in row.getiterator('th'): print header_col.text for row in t.find('tbody').getiterator('tr'): for body_col in row.getiterator('td'): print body_col.text
The DOM Object Model. The heart of a DOM parser is the DOM class hierarchy. There is a widely-used XML Document Object Model denition. This standard applies to both Java programs as well as Python. The xml.dom package provides denitions which meet this standard. The standard doesnt address how XML is parsed to create this structure. Consequently, the xml.dom package has no oicial parser. You could, for example, use a SAX parser to produce a DOM structure. Your handler would create objects from the classes dened in xml.dom. The xml.dom.minidom package is an implementation of the DOM standard, which is slightly simplied. This implementation of the standard is extended to include a parser. The essential class denitions, however, come from xml.dom. The standard element hierarchy is rather complex. Theres an overview of the DOM model in The DOM Class Hierarchy. The ElementTree Document Object Model. When using xml.etree your program will work with a number of xml.etree.ElementTree objects. Well look at a few essential classes of the DOM. There are other classes in this model, described in section 20.13 of the Python Library Reference. Well focus on the most commonly-used features of this class. class ElementTree() parse(source) Generally, ElementTree processing starts with parsing an XML document. The source can either be a lename or an object that contains XML text. The result of parsing is an object that ts the ElementTree interface, and has a number of methods for examining the structure of the document. getroot() Return the root Element of the document. find(match) Return the rst child element matching match. self.getroot().find(match). See Element.find(). This is a handy shortcut for
findall(match) Locate all child elements matching match. This self.getroot().findall(match). See Element.findall().
is
handy
shortcut
for
Returns an iterable yielding all matching elements in document order. findtext(condition, [default=None] ) Locate the rst child element matching match. This is a handy shortcut for self.getroot().findtext(match). See Element.findtext(). getiterator([tag=None] ) Creates a tree iterator with the current element as the root. The iterator iterates over this element
36.6. XML Files: The xml.etree and xml.sax Modules
439
and all elements below it, in document (depth rst) order. If tag is not None or *, only elements whose tag equals tag are returned from the iterator. See Element.getiterator(). class Element() The ElementTree is a collection of individual Elements. Each Element is either an Element, a Comment, or a Processing Instruction. Generall, Comments and Processing Instructions behave like Elements. tag The tag for this element in the XML stucture. text Generally, this is the text found between the element tags. tail This holds the text found after an elements end tag and before the next tag. Often this is simply the whitespace between tags. attrib A mutable mapping containing the elements attributes. get(name, [default=None] ) Fetch the value of an attribute. items() Return all attributes in a list as ( name, value ) tuples. keys() Return a list of all attribute names. find(match) Return the rst child element matching match. The match may be a simple tag name or and XPath expression. Returns an Element instance or None. findall(match) Locate all child elements matching match. The match may be a simple tag name or and XPath expression. Returns an iterable yielding all matching elements in document order. findtext(condition, [default=None] ) Locate the rst child element matching match. The match may be a simple tag name or and XPath expression. Returns the text value of the rst matching element. If the element is empty, the text will be a zero-length string. Return default if no element was found. getiterator([tag=None] ) Creates a tree iterator with the current element as the root. The iterator iterates over this element and all elements below it, in document (depth rst) order. If tag is not None or *, only elements whose tag equals tag are returned from the iterator. getchildren() Iterate through all children. The elements are returned in document order. When using Element.find(), Element.findall() and Element.findtext(), a simple XPATH-like syntax can be used. Match queries can have the form "tag/tag/tag" to specify a specic grant-parent-parent-child nesting of tags. Additionally, * can be used as a wildcard. For example, heres a query that looks for a specic nesting of tags.
440
from xml.etree import ElementTree dom1 = ElementTree.parse("../PythonBook-2.5/p5-projects.xml") for t in dom1.findall("chapter/section/informaltable"): print t
Note that full XPATH syntax is accepted, but most of it is ignored.
36.7 Log Files: The logging Module

Most programs need a way to write sophisticated, easy-to-control log les what contain status and debugging information. Any program that requires careful auditing will benet from using the logging module to create an easy-to-read permanent log. Also, when we have programs with multiple modules, and need more sophisticated debugging, well nd a need for the logging module. There are several closely related concepts that dene a log. 1. Your program will have a hierarchical tree of Loggers. Each Logger is used to do two things. It creates LogRecord object with your messages about errors, or debugging information. It provides these LogRecords to Handlers. Generally, each major component will have its own logger. The various loggers can have separate lter levels so that debugging or warning messages can be selectively enabled or disabled. 2. Your program will have a small number of Handlers, which are given LogRecords. A Handler can ignore the records, write them to a le or insert them into a database. Its common to have a handler which creates a very detailed log in a persistent le, and a second handler that simply reports errors and exceptions to the systems stderr le. 3. Each Handler can make use of a Formatter to provide a nice, readable version of each LogRecord message. 4. Also, you can build sophisticated Filters if you need to handle complex situations. The default conguration gives you a single Logger , named "", which uses a StreamHandler congured to write to standard error le, stderr. Advantages. While the logging module can appear complex, it gives us a number of distinct advatages. Multiple Loggers. We can easily create a large number of separate loggers. This helps us to manage large, complex programs. Each component of the program can have its own, indepenent logger. We can congure the collection of loggers centrally, however, supporting sophisticated auditing and debugging which is independent of each individual component. Also, all the loggers can feed a single, common log le. Each logger can also have a severity level lter. This allows us to selectively enable debugging or disable warnings on a logger-by-logger basis. Hierarchy of Loggers. Each Logger instance has a name, which is a .-separated string of names. For example, 'myapp.stock', 'myapp.portfolio'. This forms a natural hierarchy of Loggers. Each child inherits the conguration from its parent, which simplies conguration. If, for example, we have a program which does stock portfolio analysis, we might have a component which does stock prices and another component which does overall portfolio calculations. Each component, then, could have a separate Logger which uses component name. Both of these Loggers are children of the "" Logger ; the conguration for the top-most Logger would apply to both children. 36.7. Log Files: The logging Module 441
Some components dene their own Loggers. For example SQLAlchemy, has a set of Loggers with 'sqlalchemy' as the rst part of their name. You can congure all of them by using that top-level name. For specic debugging, you might alter the conguration of just one Logger, for example, 'sqlalchemy.orm.sync'. Multiple Handlers. Each Logger can feed a number of Handlers. This allows you to assure that a single important log messages can go to multiple destinations. A common setup is to have two Handlers for log messages: a FileHandler which records everything, and a StreamHandler which writes only severe error messages to stderr. For some kinds of applications, you may also want to add the SysLogHandler (in conjunction with a Filter) to send some messages to the operating system-maintained system log as well as the applications internal log. Another example is using the SMTPHandler to send selected log messages via email as well as to the applications log and stderr. Level Numbers and Filters. Each LogRecord includes a message level number, and a destination Logger name (as well as the text of the message and arguments with values to insert into the message). There are a number of predened level numbers which are used for ltering. Additionally, a Filter object can be created to lter by destination Logger name, or any other criteria. The predened levels are CRITICAL, ERROR, WARNING, INFO, and DEBUG. These are coded with numeric values from 50 to 10. Critical messages usually indicate a complete failure of the application, they are often the last message sent before it stops running; error messages indicate problems which are not fatal, but preclude the creation of usable results; warnings are questions or notes about the results being produced. The information messages are the standard messages to describe successful processing, and debug messages provide additional details. By default, all Loggers will show only messages which have a level number greater than or equal to WARNING, which is generally 30. When enabling debugging, we rarely want to debug an entire application. Instead, we usually enable debugging on specic modules. We do this by changing the level of a specic Logger. You can create additional level numbers or change the level numbers. Programmers familiar with Java, for example, might want to change the levels to SEVERE, WARNING, INFO, CONFIG, FINE, FINER, FINEST, using level numbers from 70 through 10. Module-Level Functions. The following module-level functions will get a Logger that can be used for logging. Additionally, there are functions can also be used to create Handlers, Filters and Formatters that can be used to congure a Logger. getLogger(name) Returns a Logger with the given name. The name is a .-separated string of names (e.g., "x.y.z" ) If the Logger already exists, it is returned. If the Logger did not exist, it is created and returned. addLevelName(level, name) Denes (or redenes) a level number, proving a name that will be displayed for the given level number Generally, you will parallel these denitions with your own constants. For example, CONFIG=20; logging.addLevelName(CONFIG,"CONFIG") basicConfig(...) Congures the logging system. By default this creates a StreamHandler directed to stderr, and a default Formatter. Also, by default, all Loggers show only WARNING or higher messages. There are a number of keyword parameters that can be given to basicConfig(). Parameters
442
lename This keyword provides the lename used to create a FileHandler instead of a StreamHandler. The log will be written to the given le. lemode If a lename is given, this is the mode to open the le. By default, a le is opened with 'a', appending the log le. format This is the format string for the Handler that is created. A Formatter object has a format() method which expects a dictionary of values; the format string uses "%(key)s" conversion specications. See String Formatting with Dictionaries for more information. The dictionary provided to a Formatter is the LogRecord , which has a number of elds that can be interpolated into a log string. datefmt The date/time format to use for the asctime attribute of a LogRecord. This is a format string based on the time package time.strftime() function. See Dates and Times: the time and datetime Modules for more information on this format string. level This is the default message level for all loggers. The default is WARNING, 30. Messages with a lower level (i.e., INFO and DEBUG) are not show. stream This is a stream that will be used to initialize a StreamHandler instead of a FileHandler. This is incompatible with filename. If both filename and stream are provided, stream is ignored. Typically, youll use this in the following form: logging.basicConfig( level=logging.INFO ). fileConfig(le) Congures the logging system. This will read a conguration le, which denes the loggers, handlers and formatters that will be built initially. Once the loggers are built by the conguration, then the logging.getLogger() function will return one of these pre-built loggers. shutdown() Finishes logging by ushing all buers and closing all handlers, which generally closes any internally created les and streams. An application must do this last to assure that all log messages are properly recorded in the log. Logger Method Functions. The following functions are used to create a LogRecord in a Logger; a LogRecord is then processed by the Handlers associated with the Logger. Many of these functions have essentially the same signature. They accept the text for a message as the rst argument. This message can have string conversion specications, which are lled in from the various arguments. In eect, the logger does message % ( args ) for you. You can provide a number of argument values, or you can provide a single argument which is a dictionary. This gives us two principle methods for producing log messages. log.info( "message %s, %d", "some string", 2 ) log.info( "message "anotherpart": 2 ) %(part1)s, %(anotherpart)d", "part1" : "some string",
These functions also have an optional argument, exc_info , which can have either of two values. You can provide the keyword argument exc_info= sys.exc_info(). As an alternative, you can provide exc_info=True, in which case the logging module will call sys.exc_info() for you. debug(message, args, ...) Creates a LogRecord with level DEBUG, then processes this LogRecord on this Logger. The message is the message text; the args are the arguments which are provided to the formatting operator, %. info(message, args, ...) Creates a LogRecord with level INFO on this logger. The positional arguments ll in the message; a single positional argument can be a dictionary. 36.7. Log Files: The logging Module 443
warning(message, args, ...) Creates a LogRecord with level WARNING on this logger. The positional arguments ll in the message; a single positional argument can be a dictionary. error(message, args, ...) Creates a LogRecord with level ERROR on this logger. The positional arguments ll in the message; a single positional argument can be a dictionary. critical(message, args, ...) Creates a LogRecord with level CRITICAL on this logger. The positional arguments ll in the message; a single positional argument can be a dictionary. log(level, message, args, ...) Creates a LogRecord with the given lvl on this logger. The positional arguments ll in the message; a single positional argument can be a dictionary. The exc_info keyword argument can provide exception information. exception(message, args, ...) Creates a LogRecord with level ERROR on this logger. The positional arguments ll in the message; a single positional argument can be a dictionary. Exception info is added to the logging message, as if the keyword parameter exc_info=True. This method should only be called from an exception handler. isEnabledFor(level ) Returns True if this Logger will handle messages of this level or higher. This can be handy to prevent creating really complex debugging output that would only get ignored by the logger. This is rarely needed, and is used in the following structure:
if log.isEnabledFor(logging.DEBUG): log.debug( "some complex message" )
The following method functions are used to congure a Logger. Generally, youll congure Loggers using the module level basicConfig() and fileConfig() functions. However, in some specialized circumstances (like unit testing), you may want ner control without the overhead of a conguration le. propagte When set to True, all the parents of a given Logger must also handle the message. This assures consistency for audit purposes. When False, the parents will not handle the message. A False value might be used for keeping debugging messages separate from other messages. By default this is a True value. setLevel(level ) Sets the level for this Logger ; messages less severe are ignored. Messages of this severity or higher are handled. The special value of logging.NOTSET indicates that this Logger inherits the setting from the parent. The root logger has a default value of logging.WARNING. getEffectiveLevel() ets the level for this Logger. If this Logger has a setting of logging.NOTSET (the default for all Loggers ) then it inherits the level from its parent. addFilter(lter ) Adds the given Filter object to this Logger. removeFilter(lter ) Removes the given Filter object from this Logger. addHandler(handler ) Adds the given Handler object to this Logger.
444
removeHandler(handler ) Removes the given Handler object from this Logger. There are also some functions which would be used if you were creating your own subclass of Logger for more specialized logging purposes. These methods include log.filter(), log.handle() and log.findCaller(). Using a Logger. Generally, there are a number of ways of using a Logger. In a module that is part of a larger application, we will get an instance of a Logger, and trust that it was congured correctly by the overall application. In the top-level application we may both congure and use a Logger. This example shows a simple module le which uses a Logger.
logmodule.py
import logging, sys logger= logging.getLogger(__name__) def someFunc( a, b ): logger.debug( "someFunc( %d, %d )", a, b ) try: return 2*int(a) + int(b) except ValueError, e: logger.warning( "ValueError in someFunc( %r , %r )", a, b, exc_info=True ) def mainFunc( *args ): logger.info( "Starting mainFunc" ) z= someFunc( args[0], args[1] ) print z logger.info( "Ending mainFunc" ) if __name__ == "__main__": logging.fileConfig( "logmodule_log.ini" ) mainFunc( sys.argv[1:] ) logging.shutdown()
1. We import the logging module and the sys module. 2. We ask the logging module to create a Logger with the given name. We use the Python assigned __name__ name. This work well for all imported library modules and packages. We do this through a factory function to assure that the logger is congured correctly. The logging module actually keeps a pool of Loggers, and will assure that there is only one instance of each named Logger. 3. This function has a debugging message and a warning message. This is typical of most function denitions. Ordinarily, the debug message will not show up in the log; we can only see it if we provide a conguration which sets the log level to DEBUG for the root logger or the logmodule Logger. 4. This function has a pair of informational messages. This is typical of main functions which drive an overall application program. Applications which have several logical steps might have informational messages for each step. Since informational messages are lower level than warnings, these dont show up by default; however, the main program that uses this module will often set the overall level to logging.INFO to enable informational messages.
36.7. Log Files: The logging Module
445
36.8 File Format Exercises

1. Create An Oice Suite Result. Back in Iteration Exercises we used the for statement to produce tabular displays of data in a number of exercises. This included How Much Eort to Produce Software?, Wind Chill Table, Celsius to Fahrenheit Conversion Tables and Dive Planning Table. Update one of these programs to produce a CSV le. If you have a desktop oice suite, be sure to load the CSV le into a spreadsheet program to be sure it looks correct. 2. Proper File Parsing. Back in File Module Exercises we built a quick and dirty CSV parser. Fix these programs to use the CSV module properly. 3. Conguration Processing. In Stock Valuation, we looked at a program which processed blocks of stock. One of the specic programs was an analysis report which showed the value of the portfolio on a given date at a given price. We make this program more exible by having it read a conguration le with the current date and stock prices. 4. Oice Suite Extraction. Most oice suite software can save les in XML format as well as their own proprietary format. The XML is complex, but you can examine it in pieces using Python programs. It helps to work with highly structured data, like an XML version of a spreadsheet. For example, your spreadsheet may use tags like <Table>, <Row> and <Cell> to organize the content of the spreadsheet. First, write a simple program to show the top-level elements of the document. It often helps to show the text within those elements so that you can correlate the XML structure with the original document contents. Once you can display the top-level elements, you can focus on the elements that have meaningful data. For example, if you are parsing spreadsheet XML, you can assembled the values of all of the <Cell>s in a <Row> into a proper row of data, perhaps using a simple Python list.
36.9 The DOM Class Hierarchy

This is some supplemental information on the xml.dom and xml.minidom object models for XML documents. class Node() The Node class is the superclass for all of the various DOM classes. It denes a number of attributes and methods which are common to all of the various subclasses. This class should be thought of as abstract: it is not used directly; it exists to provide common features to all of the subclasses. Here are the attributes which are common to all of the various kinds of Node objects. nodeType This is an integer code that discriminates among the subclasses of Node. There are a number of helpful symbolic constants which are class variables in xml.dom.Node. These constants dene the various types of Nodes. ELEMENT_NODE, ATTRIBUTE_NODE, TEXT_NODE, PROCESSING_INSTRUCTION_NODE, COMMENT_NODE, NOTATION_NODE. CDATA_SECTION_NODE, ENTITY_NODE, DOCUMENT_NODE, DOCUMENT_TYPE_NODE,
attributes This is a map-like collection of attributes. It is an instance of xml.dom.NamedNodeMap . It has method functions including get() , getNamedItem() , getNamedItemNS() , has_key() , item() , items() , itemsNS() , keys() , keysNS() , length() , removeNamedItem() , removeNamedItemNS() , setNamedItem() , setNamedItemNS() , values() . The item() and length() methods are dened by the standard and provided for Java compatibility. 446 Chapter 36. File Formats: CSV, Tab, XML, Logs and Others
localName If there is a namespace, then this is the portion of the name after the colon. If there is no namespace, this is the entire tag name. prefix If there is a namespace, then this is the portion of the name before the colon. If there is no namespace, this is an empty string. namespaceURI If there is a namespace, this is the URI for that namespace. If there is no namespace, this is None . parentNode This is the parent of this Node. The Document Node will have None for this attribute, since it is the parent of all Nodes in the document. For all other Node s, this is the context in which the Node appears. previousSibling Sibling Nodes share a common parent. This attribute of a Node is the Node which precedes it within a parent. If this is the rst Node under a parent, the previousSibling will be None . Often, the preceeding Node will be a Text containing whitespace. nextSibling Sibling Nodes share a common parent. This attribute of a Node is the Node which follows it within a parent. If this is the last Node under a parent, the nextSibling will be None . Often, the following Node will be Text containing whitespace. childNodes The list of child Nodes under this Node. Generally, this will be a xml.dom.NodeList instance, not a simple Python list . A NodeList behaves like a list , but has two extra methods: item() and length() , which are dened by the standard and provided for Java compatibility. firstChild The rst Node in the childNodes list, similar to childNodes[:1]. It will be None if the childNodes list is also empty. lastChild The last Node in the childNodes list, similar to childNodes[-1:]. It will be None if the childNodes list is also empty. Here are some attributes which are overridden in each subclass of Node. They have slightly dierent meanings for each node type. nodeName A string with the name for this Node. For an Element, this will be the same as the tagName attribute. In some cases, it will be None. nodeValue A string with the value for this Node. For an Text , this will be the same as the data attribute. In some cases, it will be None. Here are some methods of a Node. hasAttributes() This function returns True if there are attributes associated with this Node. hasChildNodes() This function returns True if there child Node s associated with this Node. class Document(Node) This is the top-level document, the object returned by the parser. It is a subclass of Node, so it inherits
36.9. The DOM Class Hierarchy
447
all of those attributes and methods. The Document class adds some attributes and method functions to the Node denition. documentElement This attribute refers to the top-most Element in the XML document. A Document may contain DocumentType, ProcessingInstruction and Comment Nodes, also. This attribute saves you having to dig through the childNodes list for the top Element. getElementsByTagName(tagName) This function returns a NodeList with each Element in this Document that has the given tag name. getElementsByTagNameNS(namespaceURI, tagName) This function returns a NodeList with each Element in this Document that has the given namespace URI and local tag name. class Element(Node) This is a specic element within an XML document. An element is surrounded by XML tags. In <para id="sample">Text</para>, the tag is <para>, which provides the name for the Element. Most Elements will have children, some will have Attributes as well as children. The Element class adds some attributes and method functions to the Node denition. tagName The full name for the tag. If there is a namesace, this will be the complete name, including colons. This will also be in nodeValue . getElementsByTagName(tagName) This function returns a NodeList with each Element in this Element that has the given tag name. getElementsByTagNameNS(namespaceURI, tagName) This function returns a NodeList with each Element in this Element that has the given namespace URI and local tag name. hasAttribute(name) Returns True if this Element has an Attr with the given name. hasAttribute(namespaceURI, localName) Returns True if this Element has an Attr with the given name based on the namespace and localName. getAttribute(name) Returns the string value of the Attr with the given name. If the attribute doesnt exist, this will return a zero-length string. getAttributeNS(namespaceURI, localName) Returns the string value of the Attr with the given name. If the attribute doesnt exist, this will return a zero-length string. getAttributeNode(name) Returns the Attr with the given name. If the named attribute doesnt exist, this method returns None. getAttributeNodeNS(namespaceURI, localName) Returns the Attr with the given name. If the named attribute doesnt exist, this method returns None. class Attr(Node) This is an attribute, within an Element. In <para id="sample">Text</para>, the tag is <para>; this tag has an attribute of id with a value of sample . Generally, the nodeType, nodeName and nodeValue attributes are all that are used. The Attr class adds some attributes to the Node denition.
448
name The full name of the attribute, which may include colons. The Node class denes localName, prefix and namespaceURI which may be necessary for correctly processing this attribute. value The string value of the attribute. Also note that nodeValue will have a copy of the attributes value. class Text(Node) class CDATASection(Node) This is the text within an element. In <para id="sample">Text</para> , the text is Text . Note that end of line characters and indentation also count as Text nodes. Further, the parser may break up a large piece of text into a number of smaller Text nodes. The Text class adds an attribute to the Node denition. data The text. Also note that nodeValue will have a copy of the text. class Comment(Node) This is the text within a comment. The  characters are not included. The Comment class adds an attribute to the Node denition. data The comment. Also note that nodeValue will have a copy of the comment.
36.9. The DOM Class Hierarchy
449
450
CHAPTER
THIRTYSEVEN
PROGRAMS: STANDING ALONE

This chapter will cover additional aspects of creating some common kinds of programs in Python. Well survey the landscape in Kinds of Programs. Then, in Command-Line Programs: Servers and Batch Processing will the essence of program startup using command-line options and operands. Well look at parsing command line options in The optparse Module Interactive graphical user interfaces are beyond the scope of this book. There are several handy graphic frameworks, including Tkinter and GTK that help you write graphical user interfaces. However, GUI programs are still started from the command line, so this section is relevant for those kinds of programs.
37.1 Kinds of Programs

There are many design patterns for our application programs. We can identify a number of features that distinguish dierent kinds of programs. We can create a taxonomy of program designs based on how we interact with them. We could also create a taxonomy based on the programs internal structure or its interfaces. We can look at programs from a number of perspectives. The type of interaction (command-line vs. user interaction). The type of architecture (stand-alone vs. client-server). Interaction. We can look at a program based on the type of interaction that it has with a person. Theres a spectrum of interaction. A program can be started from the command line and have no further interaction with the human user. We can call these batch programs because they usually process a batch of individual transactions. We can also call them command-line programs because our only interaction is at the command prompt. A large number of data analysis and business-oriented programs work with batches of data. Additionally, we can describe servers as being similar to batch programs. This is a focus for this chapter. A program can have very sophisticated interaction with the human user. The interaction may be character-oriented, or it can have a graphic user interface (GUI) and be started by double-clicking an icon. Whats important is that the user drives the processing, not the batch of data. T ypically, a program with rich user interaction will be a client of one or more services. These programs are beyond the scope of this book. Architecture. We can also look at programs based on their architecture and how they interact with other programs.
451
Some programs stand alone. They have an executable le which starts things o, and perhaps includes some libraries. Often a client program is a stand-alone program that runs on someones desktop. This is a focus for this chapter. Some programs plug into a larger and more sophisticated frameworks. The framework is, essentially, a closely related collection of libraries and interfaces. Most web applications are built as programs which plug into a web server framework. There is a tremendous amount of very common processing in handling a web transaction. Theres little value in repeating this programming, so we inherit it from the framework. We can distinguish programs in how they interact with other programs to create a larger system. Well turn to this topic in the next chapter, Architecture: Clients, Servers, the Internet and the World Wide Web. Some programs are clients. They rely on services provided by other programs. The service it relies on might be a web server or a database server. In some cases, the client program has rich user interaction and stands alone. Some programs are servers. They provide services to other programs. The service might be domain names, time, or any of a myriad of services that are an essential part of Linux and other operating systems. Some programs are both servers and clients of other services. Most servers have no interaction; they are command-line programs which are clients of other command-line programs. A web server typically has plug in web applications which use database servers. A database server may make use of other services within an operating system. Combinations. Many programs combine interaction with being a client of one or more services. Most browsers, like Firefox, are clients for servers which use a number of protocols, including HTTP, POP3, IMAP4, FTP, NNTP, and GOPHER. Besides being a client, a browser also provides graphics, handling numerous MIME data types for dierent kinds of images and sounds. These interactive, client-side applications are the most complex, and we cant begin to cover them in this book. In order to cover the basics, we have to focus on command-line programs which stand-alone. From there we can branch out to command-line clients and servers. Command-Line Subspecies. Stand-alone, command-line programs have a number of design patterns. Some programs are lters that read an input le, perform an extract or a calculation and produce a result le that is derived from the input. Programs can be compilers, performing extremely complex transformations from one or more input les to create an output le. Programs can be interpreters, where statements in a language are read and processed. Some programs, like the Unix awk utility, combine ltering and interpreting. Some command-line programs are clients of services. An FTP client program may display contents of an FTP server, accepting user commands through a graphical user interface (GUI) and transferring les. An IMAP client program may extract data from mailboxes on a mail server, accepting commands and transferring or displaying mail messages. Yet another common type of command-line program is a server. These programs are also interactive, but they interact with client programs, not a person through a GUI. An HTTP server like Apache, for instance, responds to browser requests for web pages. An FTP server responds to FTP client requests for le transfers. A server is often a kind of batch program, since it is left running for indenite periods of time, and has no user interaction.
452
Chapter 37. Programs: Standing Alone
37.2 Command-Line Programs: Servers and Batch Processing

Many programs have minimal or no user interaction at all. They are run from a command-line prompt, perform their function, and exit gracefully. They may produce a log; they may return a status code to the operating system to indicate success for failure. Almost all of the core Linux utilities (cp, rm, mv, ln, ls, df, du, etc.) are programs that decode commandline parameters, perform their processing function and return a status code. Except for a few explicitly interactive programs like editors (ex, vi, emacs, etc.), almost all of the core elements of Linux are lter-like programs. There are two critical features that make a command-line program well-behaved. First, the program should accept the arguments in a standard manner. Second the program should generally limit output to the standard output and standard error les created by the environment. When any other les are written it must be by user request and possibly require interactive conrmation. Command Line Options and Operands. The standard handling of command-line arguments is given as 13 rules for UNIX commands, as shown in the intro section of UNIX man pages. These rules describe the program names (rules 1-2), simple options (rules 3-5), options that take argument values (rules 6-8) and operands (rules 9 and 10) for the program. 1. The program name should be between two and nine characters. This is consistent with most le systems where the program name is a le name. In the Python environment, the program le must have extension of .py. 2. The program name should include only lower-case letters and digits. The objective is to keep names relatively simple and easy to type correctly. Mixed-case names and names with punctuation marks can introduce diiculties in typing the program name correctly. To be used as a module or package in Python, the program le name must be just letters, digits and _s. 3. Option names should be one character long. This is diicult to achieve in complex programs. Often, options have two forms: a single-character short form and a multi-character long form. 4. Single-character options are preceded by -. Multiple-character options are preceeded by --. All options have a ag that indicates that this is an option, not an operand. Single character options, again, are easier to type, but may be hard to remember for new users of a program. 5. Options with no arguments may be grouped after a single -. This allows a series of one-character options to be given in a simple cluster, for example ls -ldai bin clusters the -l, -d, -a and -i options. 6. Options that accept an argument value use a space separator. The option arguments are not run together with the option. Without this rule, it might be diicult to tell a option cluster from an option with arguments. Without this rule cut -ds could be an argument value of s for the -d option, or it could be clustered single-character options -d and -s. 7. Option-arguments cannot be optional. If an option requires an argument value, presence of the option means that an argument value will follow. If the presence of an option is somehow dierent from supplying a value for the option, two separate options must be used to specify these various conditions. 8. Groups of option-arguments following an option must be a single word; either separated by commas or quoted. For example: -d "9,10,56". A space would mean another option or the beginning of the operands. 9. All options must precede any operands on the command line. This basic principle assures a simple, easy to understand uniformity to command processing. 10. The string -- may be used to indicate the end of the options. This is particularly important when any of the operands begin with - and might be mistaken for an option.
37.2. Command-Line Programs: Servers and Batch Processing
453
11. The order of the options relative to one another should not matter. Generally, a program should absorb all of the options to set up the processing. 12. The relative order of the operands (or arguments) may be signicant. This depends on what the operands mean and what the program does. 13. The operand - preceded and followed by a space character should only be used to mean standard input. This may be passed as an operand, to indicate that the standard input le is processed at this time. For example, cat file1 - file2 will process le1, standard input and le2. These rules are handled by the optparse module. Output Control. A well-behaved program does not overwrite data without an explicit demand from a user. Programs with a assumed, default or implicit output le are a pronblem waiting to happen. A well-behaved program should work as follows. 1. A well-designed program has an obvious responsibility that is usually tied to creating one specic output. This can be a report, or a le of some kind. In a few cases we may nd it necessary to optimize processing so that a number of unrelated outputs are produced by a single program. 2. The best policy for this output is to write the resulting le to standard output (sys.stdout, which is the destination for the print statement.) Any logging, status or error reporting is sent to sys.stderr. If this is done, then simple shell redirection operators can be used to collect this output in an obvious way.
python someProgram.py >this_file_gets_written
3. In some cases, there are actually two outputs: details and a useful summary. In this case, the summary should go to standard output, and an option species the destination of the details.
python aProgram.py -o details.dat >summary.txt
Program Startup and the Operating System Interface. The essential operating system interface to our programs is relatively simple. The operating system will start the Python program, providing it with the three standard les (stdin, stdout, stderr; see File Semantics for more information), and the command line arguments. In response, Python provides a status code back to the operating system. Generally a status code of 0 means things worked perfectly. Status codes which are non-zero indicate some kind of problem or failure. When we run something like
python casinosim.py -g craps
The operating system command processor (the Linux shell or Windows cmd.exe) breaks this line into a command (python) and a sequence of argument values. The shell nds the relevant executable le by searching its PATH, and then starts the program, providing the rest of the command line as argument values to that program. A Python program will see the command line arguments assigned to sys.argv as ["casinosim.py", "-g", "craps"]. argv[0] is the name of the main module, the script Python is currently running. When the script in casinosym.py nishes running, the Python interpreter also nishes, and returns a status code of 0 to the operating system. To return a non-zero status code, use the sys.exit() function. Reuse and The Main-Import Switch. In Module Use: The import Statement we talked about the Main-Import switch. The global __name__ variable is essential for determing the context in which a module is used.
454
A well-written application module often includes numerous useful class and function denitions. When combining modules to create application programs, it may be desirable to take a module that had been originally designed as a stand-alone program and combine it with others to make a larger and more sophisticated program. In some cases, a module may be both a main program for some use cases and a library module for other use cases. The __name__ variable denes the context in which a module is being used. During evaluation of a le, when __name__ == "__main__", this module is the main module, started by the Python interpreter. Otherwise, __name__ will be the name of the le being imported. If __name__ is not the string "__main__", this module is being imported, and should take no action of any kind. This test is done with the as follows:
if __name__ == "__main__": main()
This kind of reuse assures that programming is not duplicated. It is notoriously diicult to maintain two separate les that are supposed to contain the same program text. This kind of cut and paste reuse is a terrible burden on programmers. Python encourages reuse through both classes and modules. All modules can be cogured as importable and reusable programming.
37.3 The optparse Module

The command line arguments from the operating system are put into the sys.argv variable as a sequence of strings. Looking at the syntax rules for command line options and operands in the previous section we can see that it can be challenging to parse this sequence of strings. The optparse module helps us parse the options and operands that are provided to our program on the command line. This module has two very important class denitions: OptionParser() and Values. An OptionParser object does two things: It contais a complete map of your options, operands and any help strings or documentation. This module can, therefore, produce a complete, standard-looking command synopsis. The -h and --help options will do this by default. It parses the sys.argv[1:] list of strings and creates a Values object with the resulting option values. The OptionParser has the following methods and attributes. There are a number of features which are used by applications which need to create a specialized subclass. Well focus on the basic use case and ignore some of the features focused on extensibility. class OptionParser(keywords...) The constructor for an optparse.OptionParser has a number of keyword arguments that can be used to dene the programs options. Parameters usage This keyword parameter sets the usage summary that will print when the options cannot be parsed, or help is requested. If you dont provide this, then your programs name will be taken from sys.argv[0] . You can suppress the usage information by setting this to the special constant optparse.SUPPRESS_USAGE . version This keyword parameter provides a version string. It also adds the option of -version which displays this string. This string can contain the formatting characters %prog which will be replaced with the programs name.
37.3. The optparse Module
455
description A paragraph of text with an overview of your program. This is displayed in respose to a help request. add_help_option This is True by default; it adds the -h and -help options. You can set this to False to prevent adding the help options. prog The name of your program. Use this if you dont want sys.argv[0] to be used. add_option(string, keywords...) This method of an OptionParser denes an option. The positional argument values are the various option strings for this option. There can be any mixture of short (-X) and long (--long) option strings. This is followed by any number of keyword arguments to provide additional details about this option. It is rare, but possible to have multiple short option strings for the same option. Heres an example:
import optparse parser= optparse.OptionParser() parser.add_option( "-o", "--output", "-w", dest="output_file", metavar="output" )
Thisdenes three dierent variations that set a single destination value, output_file. In the help text, the string output will be used to identify the three alternative option strings. Parameters action This keyword parameter takes a string. It denes what to do when this option appears on the command line. The default action is "store". Choices include "store", "store_const", "store_true", "store_false", "append", "count", "callback" and "help". The store actions store the options value. The append action accumulates a list of values. The count action counts occurances of the option. Count is often used so that -v is verbose and -vv is even more verbose. type This keyword parameter takes a string. It denes what type of value this option uses. The default type is string. Choices include "string", "int", "long", "choice", "float" and "complex". For an action of "count", the type is dened as "int"; you dont need to specify a type. dest This keyword parameter takes a string. It denes the attribute name in the OptionParse object that will have the nal value. If you dont provide a value, then the rst long option name will be used to create the attribute. If you didnt provide any long option names, then the rst short option name will be used. nargs This keyword parameter takes an integer. It denes how many values are permitted for this option. The default value is 1. If this value is more than 1, then a tuple is created to contain the sequence of values. const If the action parameter was "store_const", this keyword parameter provides the constant value which is stored. choice If the type parameter was "choice", this is a list of strings that contain the valid choices. If the options value is not in this list, then this is a run-time error. This set of choices is displayed as the help for this option. help This keyword parameter provides the help text for this option.
456
metavar This keyword parameter provides the options name as shown to the user in the help documentation. This may be dierent than the abbreviations chosen for the option. callback If the action parameter was "callback", this is a callable (either a function or a class with a __call__() method) that is called. This is called with four positional values: the Option object which requested the callback, the command line option string, the options value string, and the overall OptionParser object. callback_args callback_kwargs If the action was "callback", these keyword parameters provide the additional arguments and keywords used when calling the given function or object. set_defaults(keywords...) This method of an OptionParser provides all of the options default values. Each keyword parameter is a destination name. These must match the dest names (or the the option string) for each option that you are providing a default value. parse_args([args], [values=None], ) -> ( options, operands) This method will parse the provided command-line argument strings and update a given optparse.Values object. By default, this will parse sys.argv[1:] so you dont need to provide a value for the args parameter. Also, this will create and populate a fresh optparse.Values object, so you dont need to provide a value for the values parameter. The usual approach is options, operands = myParser.parse_args(). A Values object is created by an OptionParser. It has the attribute values built from defaults and actual parsed arguments. The attributes are dened by the options seen during parsing and any default settings that were provided to the OptionParser. A Complete Example. Heres a more complete example of using optparse. Assume we have a program with the following synopsis. -v-h-d mm/dd/yy-s symbolfile
portfolio.py
This program has two single-letter options: -v and -h. It has two options which take argument values, -d and -s. Finally, it accepts an operand of a le name. These options can be processed as follows:
"""portfolio.py -- examines a portfolio file """ import optparse class Portfolio( object ): def __init__( self, date, symbol ): ... def process( self, aFile ): ... def main(): oparser= optparse.OptionParser( usage=__doc__ ) oparser.add_option( "-v", action="count", dest="verbose" ) oparser.add_option( "-d", dest="date" ) oparser.add_option( "-s", dest="symbol" ) oparser.set_defaults( verbose=0, date=None, symbol="GE" ) options, operands = oparser.parse_args()
37.3. The optparse Module
457
portfolio= Portfolio( options.date, options.symbol ) for f in operands: portfolio.process( f ) if __name__ == "__main__": main()
The programs options are added to the parser. The default values, similarly are set in the parser. The parse_args() function separates the the options from the arguments, and builds the options object with the defaults and the parsed options. The process() function performs the real work of the program, using the options and operands extracted from the command line.
37.4 Command-Line Examples

Lets look at a simple, but complete program le. The program simulates several dice throws. Weve decided that the command-line synopsis should be: -v-s samples
dicesim.py
The -v option leads to verbose output, where every individual toss of the dice is shown. Without the -v option, only the summary statistics are shown. The -s option tells how many samples to create. If this is omitted, 100 samples are used. Here is the entire le. This program has a ve-part design pattern that weve grouped into three sections.
dicesim.py
#!/usr/bin/env python """dicesim.py .. program:: dicesym.py Simulate rolls of dice. .. cmdoption:: -v Produce verbose output, show each sample .. cmdoption:: -s samples The number of samples (default 100) """ import dice import optparse import sys def dicesim( samples=100, verbosity=0 ): """Simulate the roll of two dice by producing the requested samples. :param samples: the number of samples, default is 100 :param verbose: the level of detail to show """
458
d= dice.Dice() t= 0 for s in range(samples): n= d.roll() if verbosity != 0: print n t += n print "%s samples, average is %s" % ( samples, t/float(samples) ) def main(): parser= optparse.OptionParser() parser.add_option( '-v', dest='verbosity', action='count' ) parser.add_option( '-s', dest='samples', type='int' ) parser.set_defaults( verbosity=0, samples=100 ) opts, args = parser.parse_args() dicesim( samples=opts.samples, verbosity=opts.verbosity ) if __name__ == "__main__": main()
1. The docstring provides the synopsis of the program, plus any other relevant documentation. This should be reasonably complete. Each element of the documentation is separated by blank lines. Several standard document extract utilities expect this kind of formatting. Note that the docstring uses Restuctured Text markup with the Sphinx extensions. This will allow Sphinx to produce good-looking documentation for our program. 2. The imports line lists the other modules on which this program depends. Each of these modules might have the main-import switch and a separate main program. Our objective is to reuse the imported classes and functions, not the main function. 3. The dicesym() function is is the actual heart of the program. It is a function that does the essential work. Its designed so that it can be imported by some other program and reused. 4. The main() function is the interface between the operating system that initiates this program and the actual work in dicesym(). This does not have much reuse potential. 5. The top-level if statement makes the determination if this is a main program or an import. If it is an import, then __name__ is not "__main__", and no additional processing happens beyond the denitions. If it is the main program, the __name__ is "__main__"; the arguments are parsed by main(), which calls dicesym() to do the real work. This is a typical layout for a complete Python main progam. There are two clear objecives. First, keep the main() program focused; second, provide as many opportunities for reuse as possible.
37.5 Other Command-Line Features

Python, primarily, is a programming language. However, Python is also a family of related programs which interpret the Python language. While we can generally assume that the Python language is the same as the Python interpreter, there are some subtleties that are features of the interpreter, separate from the language. Generally, the CPython interpreter is the baseline against which others are compared, and from which others are derived. Other interpreters include Jython and Iron Python. The Python interpreter has a fairly simple command-line interface. We looked at it briey in Script Mode. In non-Windows environments, you can use the man command to see the full set of command-line options. In all cases, you can run python -h or python help to get a summary of the options.
37.5. Other Command-Line Features
459
Generally there are several kinds of command-line options. Identify The Program. This is done with -c, -m, - and the le argument. The -c option provides the Python program on the command line as a quoted string. This isnt terribly useful. However, we can use it for things like the following.
python -c 'import sys; print sys.version'
Note the rarely-used ; to terminate a statement. The -m option will locate a module on the PYTHONPATH and execute that module. This allows you to install a complete application in the Python library and execute the top-level main program script. As we noted in Script Mode, the command-line argument to the Python interpreter is expected to be a Python program le. Additionally, we can provide a Python program on standard input and use python - to read and process that program. Select the Division Operator Semantics. This is done with -Q. As we noted in Division Operators, there are two senses for division. You can control the meaning of / using -Qnew and -Qold. You can also debug problems with -Qwarn or -Qwarnall. Rather than rely on -Qnew, you should include from __future__ import division in every program that uses the new // operator and the new sense of the / operator. Optimization. This is done with -O and -OO to permit some optimization of the Python bytecode. This may lead to small performance improvements. Generally, there are two sources for performance improvements that are far more important than optimization. First, and most fundamentally, correct choices of data structures and algorithms have the most profound inuence on performance. Second, modules can be written in C and use the Python APIs. These C-language modules can dramatically improve performance, also. Startup and Loading. The -S and -E options control the way Python starts and which modules it loads. The -E option ignores all environment variables ( PYTHONPATH is the most commonly used environment variable.) Ordinarily Python executes an implicit import site when it starts executing. The site module populates sys.path with standard locations for packages and modules. The -S option will suppress this behavior. Debugging. The -d, -i, -v and -u options provide some debugging help. Python has some additional debugging information that you can access with the -d option. The -i option will allow you to execute a script and then interact with the Python intrpreter. The -v option will display verbose information on the import processing. Sometimes it will help to remove the automatic buering of standard output. If you use the -u option, mixed stderr and stdout streams may be easier to read. Indentation Problem-Solving. The -t option gives warning on inconsistent use of tabs and spaces. The -tt option makes these warnings into errors, stopping your program from running. The -x option skips the rst line of a le. This can be used for situations where the rst line of a Python le cant be a simple #! line. If the rst line cant be a comment to Python, this will skip that line. Additionally, Python comes with a Tabnanny script that can help resolve tab and space indentation issues. These problems can be prevented by using spaces instead of tabs. 460 Chapter 37. Programs: Standing Alone
There are a number of environment variables that Python uses. Well look at just a few. PYTHONPATH This denes the set of directories searched for modules. This is in addition to the directories placed on to sys.path by the site module. PYTHONSTARTUP This le is executed when you start Python for interactive use. You can use the script executed at startup time to import useful modules, dene handy functions or alter your working environment in other ways.
37.6 Command-Line Exercises

1. Create Programs. Refer back to exercises in Language Basics. See sections Numeric Types and Expressions, Condition Exercises, Iteration Exercises, Function Exercises. Modify these scripts to be stand-alone programs. In particular, they should get their input via optparse from the command line instead of input() or other mechanism. 2. Larger Programs. Refer back to exercises in Data Structures. See sections String Exercises, Tuple Exercises, List Exercises, Dictionary Exercises, Exception Exercises. Modify these scripts to be standalone programs. In many cases, these programs will need input from les. The le names should be taken from the command line using optparse. 3. Object-Oriented Programs. Refer back to exercises in Class Denition Exercises, Advanced Class Denition Exercises. Modify these scripts to be stand-alone programs.
37.7 The getopt Module

This is additional reference material on the getopt module for parsing command-line options and arguments. The command line arguments from the operating system are put into the sys.argv variable as a sequence of strings. Looking at the syntax rules for command line options and operands in the previous section we can see that it can be challenging to parse this sequence of strings. The getopt module helps us parse the options and operands that are provided to our program on the command line. This module has one very important function, also named getopt(). getopt(args, options, [long_options] ) Decode the given sequence of arguments, args, using the given set of options and long_options. Returns a tuple. The rst value is a sequence of normalized (option, value) pairs. The scond value is a sequence of the programs operand values. The args value should not include sys.argv[0], the program name. Therefore, the argument value for args is almost always sys.argv[1:]. The options value is a string of the one-letter options. Any options which require argument values are followed by a :. For example, "ab:c" means that the program will accept -a, -c , -ac , -b value as options. The long_options value is optional, if present it is a list of the long options. If a long option requires a parameter value, its name must end in =. For example, ("silent","debug","log=") means that options like --silent, --debug, and --log=myfile.log are accepted as options. There are two results of getopt(): the options and the operands. The options is a list of ( :replaceable:`name , value ) pairs. The operands is the list of names which follow the last option. In most cases, this list is a list of le names to be used as input.
37.6. Command-Line Exercises
461
There are several ways to handle the options list. We can iterate through this list, setting global variables, or conguring some processing object. This works well when we have both short and long option names for the same conguration setting.
options, operands = getopt.getopt( sys.argv[1:], ... ) for name, value in options: if name == "-X" or name == "--long": set some global variable
We can dene our conguration as a dictionary. We can then update this dictionary with the options. This forces the rest of our program to handle the -X or --long names for coguration parameters.
config = { "-X" : default, "--long": default } options, operands = getopt.getopt( sys.argv[1:], ... ) config.update( dict(options) )
We can dene our conguration as a dictionary. We can initialize that conguration dictionary with the given options then fold in default values. While pleasantly obvious, it still makes the -X and --long options visible throughout our program.
options, operands = getopt.getopt( sys.argv[1:], ... ) config= dict(options) config.setdefault( "-X", value ) config.setdefault( "--long", value )
One very adaptable and reusable structure is the following.

class MyApplication( object ): def __init__( self ): self.someProperty= some_default def process( self, aFileName ): """ The Real Work. """ This is the real work of this applications def main(): theApp= MyApplication() options, operands = getopt.getopt( sys.argv[1:], "..." ) for name, value in options: if name == "-X" or name == "--long": set properties in theApp for fileName in operaands: theApp.process( aFileName )
A Complete Example. Heres a more complete example of using getopt. Assume we have a program with the following synopsis. -v-h-d mm/dd/yy-s symbolfile
portfolio.py
This program has two single-letter options: -v and -h. It has two options which take argument values, -d and -s. Finally, it accepts an operand of a le name. These options can be processed as follows:
462
"""portfolio.py -- examines a portfolio file """ import getopt class Portfolio( object ): ... def main(): portfolio= Portfolio() opts, operands = getopt( sys.argv[1:], "vhd:s:" ) for o,v in opts: if o == "-v": portfolio.verbose= True elif o == "-h": print __doc__ return elif o == "-d": portfolio.date= v elif o == "-s": portfolio.symbol= v for f in operands: portfolio.process( f ) if __name__ == "__main__": main()
The programs options are coded as "vhd:s:": the single-letter options (-v and -h) and the value options (-d and -s). The getopt() function separates the the options from the arguments, and returns the options as a sequence of option ags and values. The process() function performs the real work of the program, using the options and operands extracted from the command line.
37.7. The getopt Module
463
464
CHAPTER
THIRTYEIGHT
ARCHITECTURE: CLIENTS, SERVERS, THE INTERNET AND THE WORLD WIDE WEB
The World-Wide Web is a metaphorical description for the sophisticated interactions among computers. The core technology that creates this phenomenon is the Internetworking Protocol suite, sometimes called The Internet. Fundamentally, the internetworking protocols dene a relationship between pieces of software called the client-server model. In this case some programs (like browsers) are clients. Other programs (like web servers, databases, etc.) are servers. This client-server model of programming is very powerful and adaptable. It is powerful because it makes giant, centralized servers available to large numbers of remote, widely distributed users. It is adaptable because we dont need to send software to everyones computer to make a change to the centralized service. Essentially, every client-server application involves a client application program, a server application, and a protocol for communication betweem the two processes. In most cases, these protocols are part of the popular and enduring suite of internetworking protocols based on TCP/IP. For more information in TCP/IP, see Internetworking with TCP/IP [Comer95]. Well digress into the fundamentals of TCP/IP in About TCP/IP . Well look at whats involved in a web server in The World Wide Web and the HTTP protocol . Well look briey at web services in Web Services . Well look at slightly lower-level protocols in Writing Web Clients: The urllib2 Module. Finally, well show how you can use low-level sockets in Socket Programming. Generally, you can almost always leverage an existing protocol; but its still relatively simple to invent your own.
38.1 About TCP/IP

The essence of TCP/IP is a multi-layered view of the world. This view separates the mechanics of operating a simple Local Area Network (LAN) from the interconnection between networks, called internetworking. Hardware. The lowest level of network services are provided by mechanisms like Ethernet (see the IEEE 802.3 standards), which covers wiring between computers. The Ethernet standards include alternatives like 10BaseT (for twisted pairs of thin wires), 10Base2 (for thicker coaxial cabling). Network services may also be wireless, using the IEEE 802.11 standards. In all cases, though, these network services provide for simple naming of devices and moving bits from device to device. What makes these low level is that these services are limited by having to know the hardware name of the receiving device; usually called the MAC address. When you buy a new network card for your computer, you eectively change your computers hardware name.
465
The TCP/IP standards put several layers of control on top of these data passing mechanisms. While these additional layers allow interconnection between networks, they also provide a standard library for using all of the various kinds of network hardware that is available. Internetworking Protocol. First, the Internet Protocol (IP) standard species addresses that are independent of the underlying hardware. The IP also breaks messages into packets and reassembles the packets in order to be independent of any network limitations on transmission lengths. Additionally, the IP standard species how to route packets among networks, allowing packets to pass over bridges and routers between networks. This is the fundamental reason why internetworking was created in the rst place. Finally, IP provides a formal Network Interface Layer to divorce IP and all higher level protocols from the mechanics of the actual network. This allows for independent evolution of the application software (like the World Wide Web) and the various network alternatives (wired, wirelss, broadband, dial-up, etc.) Transport Control Protocol. The Transport Control Protocol (TCP) protocol relies on IP. It provides a reliable stream of bytes from one application process to another. It does this by breaking the data into packets and using IP to route those packets from source to receiver. It also uses IP to send status information and retry lost or corrupted packets. TCP keeps complete control so that the bytes that are sent are recieved exactly once and in the correct order. Many applications, in turn, depend on the TCP/IP protocol capabilities. The Hypertext Transport Protocol (HTTP), used to view a web page, works by creating a TCP/IP connection (called a socket ) between browser and web server. A request is sent from browser to web server. The web server responds to the browser request. When the web page content is complete, the socket is closed and the socket connection can be discarded. Python Modules. Python provides a number of complete client protocols that are built on TCP/IP in the following modules: urllib, httplib, ftplib, gopherlib, poplib, imaplib, nntplib, smtplib, telnetlib. Each of these exploits one or more protocols in the TCP/IP family, including HTTP, FTP, GOPHER, POP, IMAP, NNTP, SMTP and Telnet. The urllib and urllib2 modules make use of multiple protocols, including HTTP and FTP, which are commonly provided by web servers. Well look into the details of just one of these higher-level procotols built on TCP/IP. Well look at HTTP and how this serves web pages for people. Well look at using this to create a web service, also. Protocols, like SMTP, POP and IMAP are used to route and read email. One can argue that SMTP is perhaps the most used protocol ever invented, since every email on the internet is pushed around by SMTP.
38.2 The World Wide Web and the HTTP protocol

One of the most widely-used protocol built on top of TCP/IP is probably HTTP. It is the backbone of the World Wide Web. The HTTP protocol denes two parties: the client (or browser) and the server. The browser is generally some piece of software like FireFox, Opera or Safari. The web server is usually based on the Apache web server, but there are several others in common use. The HyperText Transfer Protocol (HTTP) species a request and a reply. Our client (usually a browser) sends a request. The web server sends us a reply. [And yes, the World Wide Web is that simple. the sophistication comes from all the clever things that browsers and servers do with this simple protocol.] Requests. An HTTP request includes a number of pieces of information. A few of these pieces of information are of particular interest to a web application. operation The operation (or method) is generally GET or POST. There are other commands specied in the protocol (like PUT or DELETE), but they arent provided by browsers.
466
Chapter 38. Architecture: Clients, Servers, the Internet and the World Wide Web
This isnt visible. Generally, any URL you enter into a browser is accessed with a GET method. When you ll in a form and click a button, then the form is often sent as a POST request. url The URL locates tthe resource. It includes a scheme, a path, a query, and other optional information like a query. When we browse http://homepage.mac.com/s_lott, the //homepage.mac.com/s_lott is the path. The http: is the scheme (or protocol) being used. headers There are a number of headers which are included in the query; these describe the browser, and what the browser is capable of. The headers summarize some of the browsers preferences, like the language and locale. They also describe any additional data that is attached to the request. The content-length header, in particular, tells you that form input or a le upload is attached. Reply. An HTTP reply includes a number of pieces of information. It always begins with a MIMEtype string that tells the browser what kind of document will follow. This string us often text/html or text/plain. The reply also includes the status code and a number of headers. Often the headers are version infromation that the browser can reveal via the Page Info menu item in the browser. Finally, the reply includes the actual document, either plain text, HTML or an image. There are a number of HTTP status codes. Generally, a successful request has a status code of 200, indicating that request is complete, and the page is being sent. The 30x status codes indicate that the page was moved, the "Location" header provides the URL to which the browser will redirect. The 40x status codes indicate problems with the request. Generally, the resource was not found. The 50x status codes indicate problems with the server or the fundamental syntax of the request.
38.3 Writing Web Clients: The urllib2 Module

Since the World Wide Web is a client-server protocol, we can create clients or servers (or both). Generally, the clients are web browsers. There are, however, numerous applications where we want to get software from a server on the web, but we dont want to use a browser. We might have a daily extract of data, or an hourly summary of Twitter postings. These can be done by writing a web client. Fundamentally, a web client engages in an HTTP request and processes the reply that comes from the web server. When the response is in a structured markup language (like HTML or XML), then well need to parse this resulting le format. We looked at XML parsing in XML Files: The xml.etree and xml.sax Modules. HTML parsing is similar. Resources. A central piece of the design for the World-Wide Web is the concept of a Uniform Resource Locator (URL) and Uniform Resource Identier (URI). A URL provides several pieces of information for getting at a piece of data located somewhere on the internet. A URL has several data elements. Heres an example URL: http://www.python.org/download/ A scheme or protocol (http)
38.3. Writing Web Clients: The urllib2 Module
467
A location which provides the service. This includes the location (www.python.org) and an optional port number. The default port number depends on the scheme. Port 80 is for http. A path (download) An operation (GET is generally used) It turns out that we have a choice of several schemes for accessing data, making it very pleasant to use URLs. The protocols include ftp. The File Transfer Protocol, FTP. This will send a single le from an FTP server to our client. For example, ftp://aeneas.mit.edu/pub/gnu/dictionary/cide.a is the identier for a specic le. http. The Hypertext Transfer Protocol, HTTP. Amongst other things that HTTP can do, it can send a single le from a web server to our client. For example, http://www.crummy.com/software/BeautifulSoup/download/BeautifulSoup.py retrieves the current release of the Beautiful Soup module. le. The local le protocol. We can use a URL beginning with file:/// to access les on our local computer. HTTP Interaction. A great deal of information on the World Wide Web is available using simple URIs. In any well-design web site, we can simply GET the resource that the URL identies. A large number of transactions are available through HTTP requests. Many web pages provide HTML that will be presented to a person using a browser. In some cases, a web page provides an HTML form to a person. The person may ll in a form and click a button. This executes an HTTP POST transaction. The urllib2 module allows us to write Python programs which, in eect, ll in the blanks on a form and submit that request to a web server. Also note that some web sites manage interaction with people via cookies. This, too, can be handled with urllib2. Example. By using URLs in our programs, we can write software that reads local les as well as it reads remote les. Well show just a simple situation where a le of content can be read by our application. In this case, we located a le provided by an HTTP server and an FTP server. We can download this le and read it from our own local computer, also. As an example, well look at the Collaborative International Dictionary of English, CIDE. Here are three places that these les can be found, each using dierent protocols. However, using the urrllb2 module, we can read and process this le using any protocol and any server. FTP ftp://aeneas.mit.edu/pub/gnu/dictionary/cide.a This URL describes the aeneas.mit.edu server that has the CIDE les, and will respond to the FTP protocol. HTTP http://ftp.gnu.org/gnu/gcide/gcide-0.46/cide.a This URL names ftp.gnu.org server that has the CIDE les, and responds to the HTTP protocol. the
FILE file:///Users/slott/Documents/dictionary/cide.a This URL names a le on my local computer. Your computer may not have this path or this le.
urlreader.py
#!/usr/bin/env python """Get the "A" section of the GNU CIDE Collaborative International Dictionary of English """ import urllib2
468
#baseURL= "ftp://aeneas.mit.edu/pub/gnu/dictionary/cide.a" baseURL= "http://ftp.gnu.org/gnu/gcide/gcide-0.46/cide.a" #baseURL= "file:///Users/slott/Documents/dictionary/cide.a" dictXML= urllib2.urlopen( baseURL, "r" ) print len(dictXML.read()) dictXML.close()
1. We import the urllib2 module. 2. We name the URLs well be reading. In this case, any of these URLs will provide the le. 3. When we open the URL, we can read the le.
38.4 Writing Web Applications

A web application is usually embedded in a web server. The point of a web application is to respond to HTTP requests with appropriate replies. The HTTP protocol is fairly simple, making it possible in principle to write a complete web server in Python. In the long run, however, a web server written entirely in Python doesnt scale well. To provide reasonable levels of service to large numbers of users, there are a great many optimizations that are essential. One of the most important optimizations relates to the nature of the various downloads from a web server. When we request a page, the initial download in response to the GET request is usually an HTML document. Embedded in the HTML are references to numerous other les, including style sheets, Javascript libraries, images and other media. The HTML is often built dynamically and requires a sophisticated Python-based application. The rest of the content, however, is more-or-less static, and does not require deep sophistication. The static media needs to be sent as simply as possible. This dichotomy between small, complex dynamic HTML content and large, simple static content leads us to a two-part design. We want to use Python only for the HTML, and use some other, faster, application for the static content. It works out best if we embed our Python application in a web server like Apache. We can delegate the static content to Apache. We reserve the dynamic HTML creation for our Python web application programs. We usually use a component called mod_wsgi to extend Apache with Python. The idea is to congure Apache to separate requests for static media content from the requests for the dynamic HTML pages. Apache serves the static content from local les. Apache delegates (via mod_wsgi) some web requests to our Python application. Priviledge. Note that web servers usually listen on port 80. Writing applications that use this port (or any other port numbered below 1024) requires special operating system privileges. Writing priviledged applications is beyond the scope of this book. For that reason, well focus on writing applications which do one of two things. Use a higher-numbered port. 8000, 8008 or some other non-priviledged port is typical. Plug into the Apache server. Apache is a priviledged process, and can open port 80, handing requests to our applications through some kind of gateway.
38.4. Writing Web Applications
469
38.4.1 About Web Servers

A web server handles half of the HTTP conversation. We have a number of choices of ways to implement this half of the protocol. We can write our own from scratch. Python provides us some seed modules from which we can build a working server. In some applications, where the volume is low, this is entirely appropriate. See the BaseHTTPServer, SimpleHTTPServer and CGIHTTPServer modules for simple web servers. Also see the wsgiref package for a more sophisticated web server. As noted above, this is relatively ineicient because well be using the vast power of Python to serve a lot of static content les. Also, its diicult to listen for web requests on port 80 using a Python application. We can plug into the Apache server. Apache supports a wide variety of Gateway Interface technologies, including CGI and SCGI. Using the Python cgi module, we can create a CGI or SCGI script. This is an ineicient use of system resources because each request starts a complete, fresh Python interpreter. We can plug into the Apache server with mod_python. This Apache module embeds a Python interpreter directly in Apache. This embedded interpreter then runs your Python programs as part of Apaches response to HTTP requests. This is very secure and very fast. This is a relatively direct connection with Apache. One of the most popular (and exible) connections to Apache is mod_wsgi. We can use the mod_wsgi Apache extension in one of two ways. We can embed Python into Apache, or we can have Python running as a separate daemon process. Using Python as a separate deamon means that the Apache process is free to serve other web requests while our Python process is doing the complex work of creating the HTML. Generally, using cgi or mod_wsgi is still rather complex. There are numerous details of parsing requests, handing sessions, identifying users, managing logs, etc., which are common problems with common solutions. Web Frameworks. Rather than invent all of the supporting technology for a web site, its easiest to use a web application framework. If we use a framework, we can focus on the content and presentation of our web site and leave the housekeeping to the folks who write the framework. A web framework will connect to Apache; it will handle the details of parsing a web request and providing a suitable response. Using a web framework means that we do much, much less programming. Python has dozens of popular, successful web frameworks. You can look at Zope, Pylons, Django and TurboGears for some examples of dozens of ways that the Python community has simplied the construction of web applications. We cant easily cover any of the web frameworks in this book. BaseHTTPServer, just to show whats involved in HTTP. But we can take a quick look at
38.4.2 Using BaseHTTPServer

Fundamentally, a web server is an application that listens for and handles requests sent using the HTTP protocol. The handler is required to formulate a suitable response. This listen and handle loop is implemented by an instance of the BaseHTTPServer.HTTPServer class. We construct the server by providing a handler class. Each HTTP request will lead to creation of an instance of the handler class. The BaseHTTPServer.HTTPServer class has two methods to provide the overall main loop of a web server.
470
class HTTPServer() __init__(address, handlerClass) Parameters address A two-tuple, with server name and port number, usually something like ('',8008). handlerClass A subclass of BaseHTTPServer.BaseHTTPRequestHandler. This is the class itself, not an instance of the class. The server will create objects of this class and invoke that objects methods. handle_request() This method of a server will handle just one request. Its handy for debugging. Or, you could create your own serve forever loop. serve_forever() This method of a server will handle requests until the server is stopped forcibly. A forcible stop is usually an external kill signal (or the equivalent in Windows). An HTTPServer object requires a subclass of BaseHTTPServer.BaseHTTPRequestHandler. The base class does a number of standard operations related to handling web service requests. Generally, youll need to override just a few methods. Since most browsers will only send GET or POST requests, you only need to provide do_GET() and do_POST() methods. class YourRequestHandler(BaseHTTPRequestHandler ) do_GET(self ) Handle a GET request from a browser. The request is available in a number of attribute values. do_POST(self ) Handle a POST request from a browser. The request is available in a number of attribute values. This class has a number of instance variables which characterize the specic request that is currently being handled. client_address An internet address as used by Python. This is a 2-tuple: (host address, port number). command The command in the request. This will usually be GET or POST. path The requested path. request_version The protocol version string sent by the browser. 'HTTP/1.1'. Generally it will be 'HTTP/1.0' or
headers This is a collection of headers, usually an instance of mimetools.Message. This is a mapping-like class that gives you access to the individual headers in the request. The header "cookie", for instance, will have the cookies being sent back by the browser. You will need to decode the value of the cookie, usually using the Cookie module.
471
rfile If there is an input stream, this is a le-like object that can read that stream. Do not read this without providing a specic size to read. Generally, you want to get headers['Content-Length'] and read this number of bytes. If you do not specify the number of bytes to read, and there is no supplemental data, your program will wait for data on the underlying socket. Data which will never appear. wfile This is the response socket, which the browser is reading. The response protocol requires that it be used as follows: 1.Use self.send_response( number ) or self.send_response( number, text ). Usually you simply send 200. 2.Use self.send_header( header, value ) to send specic headers, like "Content-type" or "Content-length". The "Set-cookie" header provides cookie values to the browser. The "Location" header is used for a 30x redirect response. 3.Use self.end_headers() to nish sending headers and start sending the resulting page. 4.Then (and only then) you can use self.wfile.write() to send the page content. 5.Use self.wfile.close() if this is a HTTP/1.0 connection. Your class should provide some class level values which are provided to the browser. server_version A string to identify your server and version. This string can have multiple clauses, each separated by whitespace. Each clause is of the form product/version. The default is 'BaseHTTP/0.3'. error_message_format This is the web page to send back by the send_error method. The send_error method uses the error code to create a dictionary with three keys: "code", "message" and "explain". The "code" item in the dictionary has the numeric error code. The "message" item is the short message from the self.responses dictionary. The explain method is the long message from the self.responses dictionary. Since a dictionary is provided, the formatting string for his error message must include dictionary-oriented conversion strings: %(code)d, %(message)s and %(explain)s. protocol_version This is the HTTP version being used. This defaults to 'HTTP/1.0'. If you set this to 'HTTP/1.1', then you should also use the "Content-Length" header to provide the browser with the precise size of the page being sent. responses A dictionary, keyed by status code. Each entry is a two-tuple with a short message and a long explanation. These two values become the message and the explain in an error message. The message for status code 200, for example, is 'OK'. The explanation is somewhat longer. This class has a number of methods which youll want to use from within your do_GET() and do_POST() methods. send_error(code, [message] ) Send an error response. By default, this is a complete, small page that shows the code, message and explanation. If you do not provide a message, the short message from the self.responses[code] mapping will be used.
472
send_response(code, [message] ) Sends a response in pieces. If you do not provide a message, the short message from the self.responses[code] mapping will be used. This method is the rst step in sending a response. by self.send_header() if any headers are present. self.end_headers(). Then the page content can be sent. This must be followed It must be followed by
send_header(name, value) Send one HTTP header and its value. Use this to send specic headers, like "Content-type" or "Content-length". If you are doing a redirect, youll need to include the "Location" header. end_headers() Finish sending the headers; get ready to send the page content. Generally, this is followed by writing to self.wfile. log_request(code, [size] ) Uses self.log_message() to write an entry into the log le for a normal response. This is done automatically by send_headers(). log_error(format, args...) Uses self.log_message() to write an entry into the log le for an error response. This is done automatically by send_error(). log_message(format, args...) Writes an entry into the log le. You might want to override this if you want a dierent format for the error log, or you want it to go to a dierent destination than sys.stderr. Example. The following example shows the skeleton for a simple HTTP server. This sever merely displays the GET or POST request that it receives. A Python-based web server cant ever be fast enough to replace Apache. However, for some applications, you might nd it convenient to develop a small, simple application which handles HTTP.
webserver.py
import BaseHTTPServer class MyHandler( BaseHTTPServer.BaseHTTPRequestHandler ): server_version= "MyHandler/1.1" def do_GET( self ): self.log_message( "Command: %s Path: %s Headers: %r " % ( self.command, self.path, self.headers.items() ) ) self.dumpReq( None ) def do_POST( self ): self.log_message( "Command: %s Path: %s Headers: %r " % ( self.command, self.path, self.headers.items() ) ) if self.headers.has_key('content-length'): length= int( self.headers['content-length'] ) self.dumpReq( self.rfile.read( length ) ) else: self.dumpReq( None ) def dumpReq( self, formInput=None ): response= "<html><head></head><body>" response+= "<p>HTTP Request</p>" response+= "<p>self.command= <tt>%s</tt></p>" % ( self.command ) response+= "<p>self.path= <tt>%s</tt></p>" % ( self.path )
473
response+= "</body></html>" self.sendPage( "text/html", response ) def sendPage( self, type, body ): self.send_response( 200 ) self.send_header( "Content-type", type ) self.send_header( "Content-length", str(len(body)) ) self.end_headers() self.wfile.write( body ) def httpd(handler_class=MyHandler, server_address = ('', 8008), ): srvr = BaseHTTPServer.HTTPServer(server_address, handler_class) srvr.serve_forever() if __name__ == "__main__": httpd( )
1. You must create a subclass of BaseHTTPServer.BaseHTTPRequestHandler. Since most browsers will only send GET or POST requests, we only provide do_GET() and do_POST() methods. Additionally, we provide a value of server_version which will be sent back to the browser. 2. The HTTP protocol allows our application to put the input to a form either in the URL or in a separate data stream. Generally, a forms will use a POST request; the data is available. 3. This is the start of a debugging routine that dumps the complete request. This is handy for learning how HTTP works. 4. This shows the proper sequence for sending a simple page back to a browser. Thi s technique will work for les of all types, including images. This method doesnt handle complex headers, particularly cookies, very well. 5. This creates the server, srvr, as an instance of BaseHTTPServer.HTTPServer which uses MyHandler to process each request.
38.4.3 Using WSGI-based Web Servers

Fundamentally, an web server is an application that listens for and handles requests sent using the HTTP protocol. The handler is required to formulate a suitable response. Python Enhance Proposal PEP 333 denes a standard approach to handling web requests, called the Web Services Gateway Interface, WSGI. This standard allows us to build large, sophisticated web sites as a composition of many smaller components. Its best to think of WSGI as a system of pipes for routing requests and responses. To make this composition work, each WSGI application must adhere to a standardized denition. A WSGI application is a function or a callable object. The request is given to the application in the form of a dictionary. The response from the application is broken into two parts: the status plus headers are separated from the response body. A WSGI application is given a function that is must use to send the status and headers. The return value of the WSGI application is the body of the response. A WSGI application must have the following signature. wsgi_app(environ, start_response) Parameters
474
environ A dictionary with the entire HTTP request environment. This includes the OS environment, the HTTP headers, a number of items that dene the request, plus some additioanl WSGI-specic items. start_response A function that is used to start sending the status and headers. start_response(status, header_list ) The start_response() function is what your application uses to start sending an HTTP response. This includes the status and the various headers. Parameters status The status number header_list A list of 2-tuples. Each item has a header name and value. All WSGI-compatible applications must do two things. They must see to it that the start_response() function is called. They must return a list of strings. When we think of a WSGI application as a pipe, we see that an application will accomplish the above requirements one of two ways. Some WSGI applications will call another WSGI application, and return that applications list of strings. Other WSGI applications will call the start_response() function and return a list of strings that form the body of the response. The WSGI environment includes the following items that dene the request. REQUEST_METHOD The HTTP request method, generally "GET" or "POST". SCRIPT_NAME The initial portion of the request URL path. This may be empty, depending on the structure of your applications. PATH_INFO The remainder of the request URL path, designating the resource within your application. QUERY_STRING The portion of the request URL that follows the ?. CONTENT_TYPE The value of any Content-Type header in the HTTP request. If an upload is being done, this may have a value. CONTENT_LENGTH The value of any Content-Length header in the HTTP request. If an upload is being done, this may have a value. SERVER_NAME SERVER_PORT The host name and port number SERVER_PROTOCOL The protocol the client used; either "HTTP/1.0" or "HTTP/1.1". The WSGI environment includes the following WSGI-specic items. wsgi.version The tuple (1,0), representing WSGI version 1.0. wsgi.url_scheme The scheme portion of the URL; either "http" or "https". wsgi.input An input le from which the HTTP request body can be read. Generally, the body of a POST request will contain the input elds from the associated HTML form. wsgi.errors An output le to which error output can be written. This generally the main log le for the server. wsgi.multithread True if the application object may be simultaneously invoked by another thread in the same process.
475
An application might use this information to determine how to manage database connections or other resources. wsgi.multiprocess True if an equivalent application object may be simultaneously invoked by another process. wsgi.run_once True if the server or gateway expects that the application will only be invoked once by the containing process; i.e., is this a one-shot CGI-style script. There are numerous WSGI-based applications and frameworks. Well look at some components based on the wsgiref implementation. A good alternative is the werkzeug implementation. For more information, see http://werkzeug.pocoo.org/. Heres an example of a WSGI application that dumps its environment as the response page.
import cgi def dump_all_app(environ, start_response): status = '200 OK' headers = [('Content-type', 'text/html')] start_response(status, headers) env_dump= [ "<tt>%s=%r </tt><br/>" % (k,cgi.escape(str(environ[k]))) for k in environ ] return [ "<html>", "<head><title>dump_all</title></head>", "<body><p>"] + env_dump + ["</p></body>", "</html>" ]
1. We import the cgi library to make use of the cgi.escape() function. This function replaces "<" with "<", ">" with ">", and "&" with "&" to allow the value of an environment value to contain HTML. 2. We dene our application according to the WSGI standard. We accept an environment and a function that will start our response. 3. We dene a simple status and the single mandatory header, which we send via the supplied start_response function. 4. We then return a sequence of strings that is the body of page. WSGI Server. Separate from the WSGI applications is the WSGI server. This is built around a single application that will respond to requests on a specic port. This example uses the wsgiref implementation.
from wsgiref.simple_server import make_server httpd = make_server('', 8008, dump_all_app) print "Serving HTTP on port 8008..." # Respond to requests until process is killed httpd.serve_forever()
Composite Applications. The beauty of WSGI is that it allows the construction of Composite Applications. There are two general design patterns. Dispatching or Routing. In this case, a WSGI application selects among other applications and forwards the request to one or more other applications.
476
A URL parsing application, for example, can use wsgiref.util.shift_path_info() as part of transforming a URL into an application. Middleware or Pipe. In this case, a WSGI application enriches the environment and passes the request to another application. For example, authorization and authentication is a kind of pipe. The authorization application forwards valid requests with user information or reponds with an error. Each individual aspect of a complex web application can be separated into a distinct WSGI application. This individual aspects include things like the following. Authentication. An fork-style application can handle the HTTP_Authentication header. If the request lacks a proper header, this application can respond with an status 401. It can delegate basic authentication to one application and digest authentication to another application. One authenticated, an application can enrich the environment with the authenticated user information. Perhaps fetching any saved session information. Authorization. A pipeline application can determine if the user is actually allowed to perform the requested function. If the user is not authorized, it can produce a redirection to a login page. If the user is authorized, it can redirect to another application that does real work. Caching. A pipeline application can check for a given URL and return a previous result for known URLs that havent expired. For new, unknown URLs (or expired URLs) the request can be passed on to application that does the real work. Form Data Parsing. A pipeline application can parse the form data and enrich the environment with data from the various form elds. After parsing, another application can be called to process the form input. Upload Storing. A pipeline application can capture the uploaded le and save it in an upload directory for further processing. It can enrich the environment with information about the uploaded le. After saving, another application can be called to process the uploaded le.
38.5 Sessions and State

The HTTP protocol is dened as being stateless. Each request-reply transaction is independent, with no memory of any prior transaction. If a web server is only providing access to static pages of content, this stateless transaction is precisely what we expect. However, if we want a richer, more sophisticated, data processing application, we expect the application to be stateful. Indeed, one of the primary reasons for using computers is to store and retrieve information. Stored information represents the state of a database or le. Also, an individual transaction often involves the server retaining state as we enter data, correct that data, and nally commit the change to the database. The core issue is this. Given the stateless HTTP transactions and numerous concurrent clients, how do we distinguish the sequence of requests for a single uiser? Cookies. The HTTP/1.1 standard introduced the concept of a cookie. A cookie is a small packet of data that is sent to a browser as part of a response header. The browser must then include the cookie as part of each subsequent request. This permits a web server to recognize a specic browser session, and assure that the users interactions are stateful. By making the HTTP session stateful, a web application can respond in more meaningful ways. Sessions. To create stateful web applications, we need to introduce the concept of a session. The web application must do the following kinds of things. 38.5. Sessions and State 477
1. New Session? If the request has no cookie, it represents a new session. Create a distinct session object, put that objects unique key into a cookie and put the cookie into the response header. 2. Existing Session? If the request has a cookie, it represents an existing session. Locate the distinct session object that matches the cookie. All data that must be reected back to the user must be kept in an object that is unique to each session. Clearly, these session objects will accumulate as a web server runs. For speed of access, the sessions are kept in a simple dictionary. Periodically, the web server must examine the sessions and discard any that are older than some reasonable threshold. For private information (like nancial or medical records) 20 minutes is deemed old enough. For other things, session objects may last for several hours.
38.6 Handling Form Inputs

While the full extent of web applications is beyond the scope of this book, we can look at the essential ingredients in processing form input in a web server. Heres an example of a simple form. This form will send a POST request to the path . when the user clicks the Convert button. The input will include three name-value pairs with keys of fahrenheit (from the <input type="text">), celsius (from the other <input> tag) and action (from the <button type="submit">).
<html><head><title>Conversion</title><head> <body><form action="." method="POST"> <label>Fahrenheit</label> <input type="text" name="fahrenheit"/> <br/> <label>Celsius</label> <input type="text" name="celsius"/> <br/> <button type="submit" name="action" value="submit">Convert</button> </form> </body> </html>
Browser Processing. Given a form, a browser displays the elements. It then allows the user to interact with the form. When the user clicks submit, the contents for the form are transformed into a HTTP request. The method attribute of the form determines what request method is used and how the forms data is packaged for transmission to the web server. For method="GET", the request is a GET, and the contents of the form are URL-encoded and put into the URL after a ?. The request might look like this.
http://localhost:8008/?fahrenheit=&celsius=12.0&action=submit
A WSGI application will nd this data in environ["QUERY_STRING"]. The easiest way to handle this data is to use cgi.parse() in the cgi module.
data = cgi.parse( environ["QUERY_STRING"], environ )
478
For method="POST", the request is a POST, and then contents of the form are URL-encoded and put into the request as a stream of data. A WSGI application will nd this data in the le-like object environ["wsgi.input"]. This object has the data associated with the request. The number of bytes is given by environ["CONTENT_LENGTH"]. The easiest way to handle this data is to use cgi.parse() in the cgi module.
data = cgi.parse( environ["wsgi.input"], environ )
Application Processing. Generally, the best design pattern is to build applications that have the following outline. This isnt complete, but it is a useful starting point. Well add to this below. 1. When the user clicks a URL, the browser sends a GET request. The application responds with an empty form. 2. The user lls in the form, clicks the submit button. The browser sends a POST request, often to the same URL. The application validates the form input. If the input is valid, the application responds with the resulting page. If the input is not valid, the application responds with the form and any error messages. The forms data is parsed with cgi.parse(). Heres an example WSGI application that shows the POST and GET processing.
form ="""\ <html><head><title>title</title><head> <body> <p>%(messages)s</p> <form action="." method="GET"> <label>label1</label> <input type="text" name="field1" value="%(field1)s"/> <br/> <label>label2</label> <input type="text" name="field2" value="%(field2)s"/> <br/> <button type="submit" name="action" value="submit">Convert</button> </form> </body> </html> """ def conversion( environ, start_response ): # For a GET, display the empty form. if environ['REQUEST_METHOD'] == "GET": status = '200 OK' # HTTP Status headers = [('Content-type', 'text/html')] # HTTP Headers start_response(status, headers) return [ form % { 'field1' : '', 'field2' : '', 'messages':'' } ] # For a POST, parse the input, validate it, and try to process it. else: data= cgi.parse( environ['wsgi.input'], environ ) try: if 'field1' in data: field1= data.get('field1',[""])[0] if 'field2' in data: field2= data.get('celsius',[""])[0] # Validate... # Do processing... status = '200 OK' # HTTP Status headers = [('Content-type', 'text/html')] # HTTP Headers
38.6. Handling Form Inputs
479
start_response(status, headers) return [ form % { 'field1' : field1, 'field2' : field2, 'messages':'' } ] except Exception, e: status = '400 ERROR' # HTTP Status headers = [('Content-type', 'text/html')] # HTTP Headers start_response(status, headers, exc_info=e) return [ form % { 'field1' : '', 'field2' : '', 'messages':repr(data) } ]
The Post and Back Problem. Note that if you submit the form as a POST and click your browsers back button, after looking at the next page, the form gets submitted again. Your browser will conrm that you want to submit form data again. This behavior is usually prevented by using the Redirect-after-Post (also called the Post-Redirect-Get) design pattern. The response to a page processed with POST, is a status 301 (Redirect) response. This response must include a header with a label of Location and a value that is a URL to which the browser will address a GET request. This makes the back button behave nicely. The complete overview, then, is the following. 1. When the user clicks a URL, the browser sends a GET request. If theres no cookie, the application creates a new, unique sesions and builds the appropriate cookie. If theres a cookie, the application retrieves the session and any saved state in that session. The saved state may include error messages, status messages, and previously entered values on the form. The application responds with the form, including any messages. 2. The user lls in the form, clicks the submit button. The browser sends a POST request, often to the same URL. If theres no cookie, something has gone awry. This is handled like the GET request the cookie must be added and the form sent. There should be a cookie with session information that provides any needed context. The application validates the form input. If the input is valid, the application does the expected processing. The session is updated with completion messages. The application sends a "301 REDIRECT" response. This causes the browser to do a GET to the given location. If the input is not valid, the application responds with the form and any error messages. In more complex applications, there may be multiple pages, or multiple-step transactions. There may also be a conrm page at the end which summarizes the transaction before the real work is done. This requires accumulating considerable information in the session.
38.7 Web Services

When we looked at HTTP in The World Wide Web and the HTTP protocol , we were interested in its original use case of serving web pages for people. We can build on HTTP, creating an interface between software components, something called a web service. A web service leverages the essential request-reply nature of HTTP, but takes the elaborate human-centric HTML web page out of the response. Instead of sending back something for people to read, web services send just the facts without a sophisticated presentation.
480
Web services allow us to have a multi-server architecture. A central web server provides interaction with people. When a persons browser makes a request, this central web server can make web service requests of other servers to gather the information. After gathering the information, the central server aggregates it and builds the nal HTML-based presentation, which is the reply sent to the human user. Web services are an adaptation of HTTP; see The World Wide Web and the HTTP protocol for a summary. Web services rely on a number of other technologies. There are several competing alternatives, and well look at web services in general before looking at a specic technique.
38.7.1 Web Services Overview

There are a number of ways of approaching the problem of coordinating work between clients and servers. All of these alternatives have their advantages and disadvantages. XML-RPC. The XML-RPC protcol uses XML notation to make a remote procedure call (RPC). It works by sending an HTTP request that contains the name of the procedure to call and the arguments to that procedure. This protocol uses HTTP POST requests to provide the XML document. SOAP. There are two variations on the Simple Object Access Protocol (SOAP): remote procedure call variation and document. The RPC variant is basically the next generation of XML-RPC, where an XML document encodes the name of the procedure and the arguments. The document variant merely sends an XML document; the document provides all the information required by the server. This protocol is supported by additional standards like Web Services Denition Language (WSDL). REST. The Representational State Transfer (REST) protcol uses the HTTP operations (POST, GET, PUT, DELETE) and Uniform Resource Identiers (URI) to manipulate remote objects. This protocol is perhaps the simplest of the web services protocols; for this reason it is very popular. Well focus on REST because it can be done largely using urllib2 features (see Writing Web Clients: The urllib2 Module) and the JSON library. RESTful Web Services. The essence of REST is that we are accessing a resource that resides on another, remote computer. In order to do this, we must transfer a representation of that objects state. We have, therefore, three separate issues that we have to address. 1. Representing an objects state. We can use XML for this. There are other notations including JSON and YAML which are also used to represent an objects state. Well focus on JSON because its widely used and very simple. This representation issue happens on both client and server side of the transaction. When the client wants to create or update a resource, it must represent the object. When the server wants to provide a resource, it must represent the object, also. 2. Making the client request. This means marshalling the arguments, making the request, and unmarshalling the response. Since REST is based on HTTP, this is a kind of HTTP client access using one of the four methods: GET, POST, PUT, DELETE. 3. Serving requests. This means unmarshalling arguments, doing something useful, and marshalling a response. Since this is based on HTTP, this is a kind of HTTP server. A GET request, generally, doesnt have any arguments; it identies a resource, which is marshalled and returned. A POST request creates a new resource. The associated data is an URL-encoded version of the resources to create. Often the created resource is marshalled and returned as a kind conrmation. A PUT request will replace a resource. The associated data is an URL-encoded version of the resources to replace or update the existing resource.
38.7. Web Services
481
A DELETE request will remove a resource. Generally, this doesnt have any arguments; it identies a resource, which is removed.
38.7.2 Web Services Server

Lets imagine that weve built a an extremely good simulation of a roulette wheel. Wed like to package this as a web service so that many people can share this in their simulations of Roulette. In some cases, a web service is built into a more complete web application framework. Often the server will have a human interface as well as a web service interface. The human interface will use HTML. The web service interface will use JSON. Well simplify things slightly, and create a family of WSGI applications to route requests, handle the JSON replies and handle HTML replies. The Resource. The resource were serving is a roulette wheel. We created this Python module that denes the wheel. Each spin creates a dictionary that shows a number of bets which are won by this spin. This is a separate module that includes just the class denition that well be serving.
wheel.py
import random class Wheel( object ): redNumbers= set( [1,3,5,7,9,12,14,16,18,19,21,23,25,27,30,32,34,36] ) domain = range(1,37) + [ "0", "00" ] def __init__( self ): self.rng= random.Random() self.last_spin= None self.count= 0 def spin( self ): n = random.choice( Wheel.domain ) if n in ( "0", "00" ): self.last_spin= { "number": n, "color": "green", "even": False, "high": None, "twelve": None, "column": None, } else: color = "red" if n in Wheel.redNumbers else "black" self.last_spin= { "number": n, "color": color, "even": n%2==0, "high": n>=18, "twelve": n//12, "column": n%3, } self.count += 1 return self.last_spin
482
The WSGI Applications. Well dene several WSGI Applications that will create a comprehensive wheel web service. This rst example is the top-level routing application that parses the URL and delegates the work to another application. This is not a very exible design. There are numerous better examples of very exible routing using regular expression matching and other techniques.
wheelservice.py, part 1
import import import import import import wsgiref.util cgi sys traceback json wheel
# A global object so we can maintain state. theWheel= wheel.Wheel() def routing( environ, start_response ): """Route based on top-level name of URL.""" try: top_name = wsgiref.util.shift_path_info( environ ) if top_name == "html": return person( environ, start_response ) elif top_name == "json": return service( environ, start_response ) else: start_response( '404 NOT FOUND', [("content-type","text/plain")] ) return [ "Resource not found." ] except Exception, e: environ['wsgi.errors'].write( "Exception %r " % e ) traceback.print_exc( file=environ['wsgi.errors'] ) status = "500 ERROR" response_headers = [("content-type","text/plain")] start_response(status, response_headers, sys.exc_info()) return ["Application Problems. ", repr(e) ]
1. We dene some global state information for these applications to share. In this case, all the application share theWheel. 2. The routing() function is a WSGI application, and has the proper arguments. 3. We use wsgiref.util.shift_path_info() to parse out the rst level of the URL path, and use this to distiguish between the HTML-oriented human interface and the JSON-oriented web interface. 4. In case of an exception, we print the traceback to the log, and return an error page. A common extension to this routing is to respond to the /favicon.ico request by providing a graphic image le that can be displayed in the URL box of the browser. This second example is the next-level person and service applications. These do the real work of the overall service. They either respond to a person (using HTML) or to a web services client (using JSON).
38.7. Web Services
483
def person( environ, start_response, exc_info=None ): """Print some information about the stateful wheel.""" global theWheel status = '200 OK' headers = [('Content-type', 'text/html')] start_response(status, headers, exc_info) return [ "<html>", "<head><title>Wheel Service</title></head>", "<body>", "<p>Wheel service is spinning.</p>" "<p>Served %d spins.</p>" % (theWheel.count,), "</body>", "</html>" ] def service( environ, start_response ): """Update the stateful wheel.""" global theWheel spin= theWheel.spin() status= '200 OK' headers = [("content-type","text/plain")] start_response( status, headers ) return [ json.dumps(spin) ]
1. The person() function is a WSGI application, and has the proper arguments. 2. The person() function refers to the global state information, theWheel. It creates an HTML page with some status information. 3. The service() function is a WSGI application, and has the proper arguments. 4. The service() function refers to the global state information, theWheel. It uses json to formulate a reply in JSON notation. The WSGI Service. The WSGI service simply wraps our composite application routing() and serves it.
from wsgiref.simple_server import make_server httpd = make_server('', 8008, routing) print "Serving HTTP on port 8008..." # Respond to requests until process is killed httpd.serve_forever()
38.7.3 Web Services Client

Lets imagine that a colleague has built a web service which provides us with an extremely good simulation of a roulette wheel. Our colleague has provided us with the following summary of this web service.
484
host 10.0.1.5. While IP address numbers are the lowest-common denominator in naming, some people will create Domain Name Servers (DNS) which provide interesting names instead of numeric addresses. If you are testing on a single computer, you will use localhost. port number 8008. While the basic HTTP service is dened to run on port 80, you may have other web services which, for security reasons, arent available on port 80. Port numbers from 1024 and up may be allocated for other purposes, so port numbers are often changed as part of the conguration of a program. path /json/ for the basic web services request. This isnt the best denition for this resource, since it cant easily be expanded. method GET. response JSON-encoded dictionary with attributes of the spin. This gives us a nal URL of http://localhost:8008/json/ for access to this service. To create a web services client, we can use the urllib2 module to access this service.
wheelclient.py
import urllib2 import json def get_a_spin( ): result= urllib2.urlopen( "http://localhost:8008/json/" ) assert result.code == 200 assert result.msg == "OK" # print result.headers # to see information about the service data= result.read() return json.loads( data ) spin= get_a_spin() print spin
1. We import the urllib2 library, which allows us to do HTTP GET and POST as part of a RESTful web service. 2. We import the json library, which well use to decode the response from the web service. 3. We dene a function which will use the web service to get a spin of a roulette wheel. This function uses urllib2.urlopen() to make a GET request to the given URL. The server will respond with a JSON-encoded response. 4. We load use json.loads() to parse the response and build a dictionary.
38.8 Client-Server Exercises

38.8.1 Create a Customized FTP Client
Write a simple, special-purpose FTP client that establishes a connection with an FTP server, gets a directory and ends the connection. The FTP directory commands are DIR and LS. The responses may be long and complex, so this program must be prepared to read many lines of a response.
38.8. Client-Server Exercises
485
For more information, RFC 959 has complete information on all of the commands an FTP server should handle. Generally, the DIR or LS command, the GET and PUT commands are suicient to do simple FTP transfers. Your client will need to open a socket on port 21. It will send the command line, and then read and print all of the reply information. In many cases, you will need to provide additional header elds in order to get a satisfactory response from a web server. To test this, youll need to either activate an FTP server on your computer, or locate another computer that oers FTP services.
38.8.2 Desktop Web Application

You can easily write a desktop application that uses web technology, but doesnt use the Internet. Heres how it would work. Your application is built as a very small web server, based on BaseHTTPServer.HTTPServer or wsgiref.simple_server. This application prepares HTML pages and forms for the user. The user will interact with the application through a standard browser like Firefox, Opera or Safari. Rather than connect to a remote web server somewhere in the Internet, this browser will connect to a small web server running on your desktop. The URL will be http://localhost:8008/. You can package your application with a simple shell script (or .BAT le) which does two things. (This can also be done as a simple Python program using the subprocess module.) 1. It starts a subprocess to run the HTTP server side of your application. 2. It starts another subprocess to run the browser, providing an 'http://localhost:8008' to point the browser at your application server. initial link of
Since this is a single-user application, there wont be multiple, concurrent sessions, which greatly simplies web application implementation. Example Application. We could, for example, write a small application that did Fahrenheit to Celsius conversion. We would create a Python web server and a wrapper script that launched the server and launched a browser. The Input Form. While the full power of HTML is beyond the scope of this book, well provide a simple form using the <form>, <label>, <button> and <input> tags. Heres an example of a simple form. This form will send a POST request to the path . when the user clicks the Convert button. The input will include name-value pairs with keys of fahrenheit, celsius and action. The value will be a list of strings. Since the form only has a simple text eld with a given name, there will be a single string in each list. The cgi.parse() function can parse the encoded form input.
<html><head><title>Conversion</title><head> <body><form action="." method="POST"> <label>Fahrenheit</label> <input type="text" name="fahrenheit"/> <br/> <label>Celsius</label> <input type="text" name="celsius"/> <br/> <button type="submit" name="action" value="submit">Convert</button> </form>
486
</body> </html>
The WSGI Applications. Youll need to write at least one WSGI application to handle the form input. If the environ['REQUEST_METHOD'] is "GET", this is an initial request, the form should be returned. If the environ['REQUEST_METHOD'] is "POST", this is is the form, as lled in by the user. Extract the elds provided by the user. Attempt to convert the input to a oating-point number. If this fails, or if both elds are empty, present the form with error messages. Calculate the value of the empty eld from the completed eld. Present the form with values lled in. An Overview of the WSGI Server. Youll use wsgiref.make_server() to create a server from your form-handling application. Youll need to provide an address like ('', 8008) , and the name of your application. This objects serve_forever() method will then handle HTTP requests on port 8008.
38.8.3 Complete Roulette Server

Well create an alternative implementation of the simple Roulette server shown in Web Services. Well dene a simple REST-based protocol for placing bets, spinning the wheel and retrieving the results of the placed bets. Each BET will be considered a resource. Well use POST to create the resource, and GET to check on the resource after the spin. The spin is a subtle issue. In a sense, were merely getting a value. However, executing the spin, changes the state of the various bets. Therefore, the spin should be a kind of POST transaction. Session and State. A web site that interacts with a browser generally uses cookies to maintain state so that the person doesnt have to be aware of how state is maintained. For web services applications, its cosiderably simpler to maintain state explicitly. A WS client program can use explicit session identication. Well handle this thorough a simple GET request which provides a unique session identier. A more secure method would include HTTP Digest Authentication. However, thats beyond the scope of this book. Resource Details. Our top-level application will examine the request URL. Well consider the top-level URL as the resource type. Our top-level application can then route the request based on this resource name. session A POST request to /session/ will allocate a new session. The response is a JSON document with the session identier. The data sent is a JSON document with information like the bettors name. The session identier must be used in all further transactions to identify the specic bettor. bet A POST request to /bet/session/ will create a new bet. The processing could look like the following. The data sent is a JSON document with a list of bets with a specic proposition (red, black, even, odd, etc.) and an amount.
487
Roulette tables often have a minimum and a maximum bet amount. These might be $10 and $500. In addition to valid names and amounts, the total of the bets must also conform to these limits to be considered valid. If a bet (or the total) is invalid, the server should respond with a 501 INVALID status. This message body should include details on which bet was rejected and why. If the entire sequence of bets and amounts are valid, the server should respond with a 200 OK status. The response body is a JSON document that includes a conrmation. This can be a simple sequential number or a UUID or some similar secure hand-shake. A GET request to /bet/session/confirmation/ will return the status of the requested bet. The response is a JSON document with the bet and the outcome. If the wheel has not been spun, the outcome is None. Otherwise, the outcome is the bets payout multiplied by the amount of the bet. spin A POST request to /spin/session/ indicates that the bets and placed and the wheel can be spun. This will respond with a JSON document that has the wheel spin conrmation number. Currently, theres not much use for this information except to acknowledge that the wheel was spun. After a POST request to /spin/session/, a client will have to do a GET request to retrieve bet results.
38.8.4 Roulette Client

Write a simple client which places a number of bets on the Roulette server, using the Martingale betting strategy. The Martingale strategy is relatively simple. A bet is placed on just one 2:1 proposition (Red, Black, Even, Odd, High or Low). A base betting amount (the table minimum) is used. If the outcome of the spin is a winner, the betting amount is reset to the base. If the outcome of the spin is a loser, the betting amount is doubled. Note that this double up on a loss betting will lead to situations where the ideal bet is beyond the table maximum. In that case, your simulation must adjust the bets to be the table maximum. Also note that a proper simulation has a budget for betting. When the budget is exhausted, the simulation has to stop playing.
38.8.5 Roulette Client

We can write a web application which uses our Roulette Server. This will lead to a fairly complex (but typical) architecture, with two servers and a client. Well have the Roulette Server from the Complete Roulette Server exercise, running on some nonpriviledged port (like 36000). This server accepts bets and spins the wheel on behalf of a client process. It has no user interaction, it simply maintains state, in the form of bets placed. Well have a web server, similar to the Desktop Web Application exercise, running on port 8008. This application can present a simple form for placing a bet or spinning the wheel.
488
The user can ll in the elds to dene a bet. When the user clicks the Bet button, the web application will make a request to the Roulette Server and present the results in the HTML page that is returned to the user. If the bet is valid, the web application will make a request to the Roulette Server to spin the wheel. It will present the results in the HTML page that is returned to the user. This interaction between web application and Roulette server can all be done with urllib2. Well can then use a browser to contact our web server. This client will browse http://localhost:8008 to get a web page with a simple form for placing a bets and spinning the wheel. A simple HTML form might look like the following.
<html><head><title>Roulette</title></head> <body> <p>Results from previous request go here</p> <form action="." method="POST"> <label>Amount</label> <input type="text" name="amount"/> <br/> <label>Proposition</label> <select name="proposition"> <option>Red</option> <option>Black</option> <option>Even</option> <option>Odd</option> <option>High</option> <option>Low</option> </select> <br/> <button type="submit" name="action" value="bet">Bet</button> </form> </body> </html>
A GET request can present the form. A POST request must parse the input to nd the values of the two elds ("proposition" and "amount"). It must validate input on the form, make requests to the server, and present the results.
38.8.6 Chess Server

In Chessboard Locations we described some of the basic mechanics of chess play. A chess server would allow exactly two clients to establish a connection. It would then a chess moves from each client and respond to both clients with the new board position. We can create a simple web service that has a number of methods for handling chess moves. To do this, well need to create a basic ChessBoard class which has a number of methods that establish players, move pieces, and report on the boards status. Its essential that the ChessBoard be a single object that maintains the state of the game. When two players are connected, each will need to see a common version of the chessboard. Here are some of the methods that are essential to making this work. class ChessBoard() __init__(self ) Initialize the chessboard with all pieces in the starting position. It will also create two variables
489
to store the player names. These two player names will initially be None, since no player has connected. connect(self ) Allow a player to connect to the chess server. If two players are already connected, this will return an Error message. Otherwise, this will return an acceptance message that includes the players assigned color (White or Black.) move(self, player, from, to) Handle a request by a player to move a piece from one location to another location. Both locations are simply le (a-h) and rank (1-8). If the move is allowed, the response is an acknowledgement. Otherwise, an error response is sent. The chess server will need to track the moves to be sure they players are alternating moves properly. board(self ) The response is a 128-character string that reports the state of the board. It has each rank in order from 8 to 1; within each rank is each le from a to h. Each position is two characters with a piece code or spaces if the position is empty. A web service will oer a simple RESTful resource for interacting. game A POST request to /game should include a JSON document that identies the player. The successful response is a JSON document which provides a game identication for the player along with the players color. A POST request to /game/game/ includes a JSON document with the player identication and their move. If this is valid, the reponse is a JSON document with the move number and board position. If this is invalid, the response is a JSON document with the move number, board position and the error message. A GET request to /game/game/ will respond with the history of moves and the board position. A client application for this web service can use urllib2 to make the various POST and GET requests of the server. Generally the ue case for the client would have the following outline. 1. The client process will attempt a connection. If that fails, the server is somehow unable to start a new game. 2. The client will display the board state and wait for the user to make a move. Once the move is entered, the client will make web services requests to provide moves from the given player and display the resulting board status. 3. Also, the client must poll the server to see if the other player has entered their move. If a web page includes the following HTML, it will periodically refresh itself, polling the server. <meta http-equiv="refresh" content="60">. This is included within the <head> tags. This will poll once every 60 seconds. For a desktop applicaiton, the polling is usually done by waiting a few seconds via time.sleep().
490
38.9 Socket Programming

Socket-level programming isnt our rst choice for solving client-server problems. Sockets are nicely supported by Python, however, giving us a way to create a new protocol when the vast collection of existing internetworking protocols are inadequate. Client-server applications include a client-side program, a server, a connection and a protocol for communication betweem the two processes. One of the most popular and enduring suite of client-server protocols is based on the Internetworking protocol: TCP/IP. For more information in TCP/IP, see Internetworking with TCP/IP [ Comer95 ] . All of the TCP/IP protocols are based on the basic socket . A socket is a handy metaphor for the way that the Transport Control Protocol (TCP) reliably moves a stream of bytes between two processes. The socket module includes a number of functions to create and connect sockets. Once connected, a socket behaves essentially like a le: it can be read from and written to. When we are nished with a socket, we can close it, releasing the network resources that were tied up by our processing.
38.9.1 Client Programs

When a client application communicates with a server, the client does three things: it establishes the connection, it sends the request and it reads the reply from the server. For some client-server relationships, like a databsae server, there may be multiple requests and replies. For other client-server requests, for example, the HTTP protocol, a single request may involve a number of replies. To establish a connection, the client needs two basic facts about the server: the IP address and a port number. The IP address identies the specic computer (or host) that will handle the request. The port number identies the application program that will process the request on that host. A typical host will respond to requests on numerous ports. The port numbers prevent requests from being sent to the wrong application program. Port numbers are dened by several standards. Examples include FTP (port 21) and HTTP (port 80). A client program makes requests to a server by using the following outline of processing. 1. Develop the servers address. Fundamentally, an IP address is a 32-bit host number and a 16-bit port number. Since these are diicult to manage, a variety of coding schemes are used. In Python, an address is a 2-tuple with a string and a number. The string representes the IP address in dotted notation ( "194.109.137.226" ) or as a domain name ( "www.python.org" ); the number is the port number from 0 to 65535. 2. Create a socket and connect it to this address. This is a series of function calls to the socket module. When this is complete, the socket is connected to the remote IP address and port and the server has accepted the connection. 3. Send the request. Many of the standard TCP/IP protocols expect the commands to be sent as strings of text, terminated with the n character. Often a Python le object is created from the socket so that the complete set of le method functions for reading and writing are available. 4. Read the reply. Many of the standard protocols will respond with a 3-digit numeric code indicating the status of the request. Well review some common variations on these codes, below. Developing an Address. An IP address is numeric. However, the Internet provides domain names, via Domain Name Services (DNS). This permits useful text names to be associated with numeric IP addresses. Were more used to "www.python.org". DNS resolves this to an IP address. The socket module provides functions for DNS name resolution. The most common operation in developing an address is decoding a host name to create the numeric IP address. The socket module provides several functions for working with host names and IP addresses. 38.9. Socket Programming 491
gethostname() Returns the current host name. gethostbyname(host ) Returns the IP address (a string of the form 255.255.255.255) for a host. socket::gethostbyaddr( address ) -> ( name, aliasList, addressList )() Return the true host name, a list of aliases, and a list of IP addresses, for a host. The host argument is a string giving a host name or IP number. socket::getservbyname( servicename, protocolname ) -> integer() Return a port number from a service name and protocol name. The protocol name should be tcp or udp. Typically, the socket.gethostbyname() function is used to develop the IP address of a specic server name. It does this by makig a DNS inquiry to transform the host name into an IP address. Port Numbers. The port number is usually dened by your application. For instance, the FTP application uses port number 21. Port numbers from 0 to 1023 are assigned by RFC 1700 standard and are called the well known ports. Port numbers from 1024 to 49151 are available to be registered for use by specic applications. The Internet Assigned Numbers Authority (IANA) tracks these assigned port numbers. See http://www.iana.org/assignments/port-numbers. You can use the private port numbers, from 49152 to 65535, without fear of running into any conicts. Port numbers above 1024 may conict with installed software on your host, but are generally safe. Port numbers below 1024 are restricted so that only priviledged programs can use them. This means that you must have root or administrator access to run a program which provides services on one of these ports. Consequently, many application programs which are not run by root, but run by ordinary users, will use port numbers starting with 1024. It is very common to use ports from 8000 and above for services that dont require root or administrator privileges to run. Technically, port 8000 has a dened use, and that use has nothing to do with HTTP. Port 8008 and 8080 are the oicial alternatives to port 80, used for developing web applications. However, in spite of an oicial use, port 8000 is often used for web applications. The usual approach is to have a standard port number for your application, but allow users to override this in the event of conicts. This can be a command-line parameter or it can be in a conguration le. Generally, a client program must accept an IP address as a command-line parameter. A network is a dynamic thing: computers are brought online and oine constantly. A hard-wired IP address is an inexcusable mistake. Create and Connect a Socket. A socket is one end of a network connection. Data passes bidirectionally through a socket between client and server. The socket module denes the SocketType, which is the class for all sockets. The socket() function creates a socket object. socket(family, [type, protocol] ) Open a socket of the given type. The family argument species the address family; it is normally socket.AF_INET. The type argument species whether this is a TCP/IP stream (socket.SOCK_STREAM) or UDP/IP datagram (socket.SOCK_DGRAM) socket. The protocol argument is not used for standard TCP/IP or UDP/IP. A SocketType object has a number of method functions. Some of these are relevant for server-side processing and some for client-side processing. The client side method functions for establishing a connection include the following. connect(address) Connect the socket to a remote address; the address is usually a (host address, port #) tuple. In the event of a problem, this will raise an exception.
492
connect_ex(address) Connect the socket to a remote address; the address is usually a (host address, port #) tuple. This will return an error code instead of raising an exception. A value of 0 means success. fileno() Return underlying le descriptor, usable by the select module or the os.read() and os.write() functions. getpeername() Return the remote address bound to this socket; not supported on all platforms. getsockname() Return the local address bound to this socket. getsockopt(level, opt, [buen] ) Get socket options. See the UNIX man pages for more information. The level is usually SOL_SOCKET. The option names all begin with SO_ and are dened in the module. You will have to use the struct module to decode results. setblocking(ag) Set or clear the blocking I/O ag. setsockopt(level, opt, value) Set socket options. See the UNIX man pages for more information. The level is usual SOL_SOCKET. The option names all begin with SO_ and are dened in the module. You will have to use the struct module to encode parameters. shutdown(how ) Shutdown traic on this socket. If how is 0, receives are disallowed; if how is 1, sends are disallowed. Usually this is 2 to disallow both reads and writes. Generally, this should be done before the close() . close() Close the socket. Its usually best to use the shutdown() method before closing the socket. Sending the Request and Receiving the Reply. Sending requests and processing replies is done by writing to the socket and reading data from the socket. Often, the response processing is done by reading the file object that is created by a sockets makefile() method. Since the value returned by makefile() is a conventional le, then readlines() and writelines() methods can be used on this le object. A SocketType object has a number of method functions. Some of these are relevant for server-side processing and some for client-side processing. The client side method functions for sending (and receiving) data include the following. recv(bufsize, [ags] ) Receive data, limited by bufsize. flags are MSG_OOB (read out-of-band data) or MSG_PEEK (examine the data without consuming it; a subsequent recv() will read the data again). recvfrom(bufsize, [ags], ) -> ( string, address) Receive data and senders address, arguments are the same as recv(). send(string, [ags] ) Send data to a connected socket. The MSG_OOB ag is supported for sending out-of-band data. sendto(string, [ags] ) Send data to a given address, using an unconnected socket. The ags option is the same as send(). Return value is the number of bytes actually sent. makefile(mode, bufsize) Return a le object corresponding to this socket. The mode and bufsize options are the same as used in the built in file() function.
38.9. Socket Programming
493
Example. The following examples show a simple client application using the socket module. This is the Client class denition.
#!/usr/bin/env python import socket class Client( object ): rbufsize= -1 wbufsize= 0 def __init__( self, address=('localhost',7000) ): self.server=socket.socket( socket.AF_INET, socket.SOCK_STREAM ) self.server.connect( address ) self.rfile = self.server.makefile('rb', self.rbufsize) self.wfile = self.server.makefile('wb', self.wbufsize) def makeRequest( self, text ): """send a message and get a 1-line reply""" self.wfile.write( text + '\n' ) data= self.rfile.read() self.server.close() return data print "Connecting to Echo Server" c= Client() response= c.makeRequest( "Greetings" ) print repr(response) print "Finished"
A Client object is initialized with a specic server name. The host ( "localhost" ) and port number ( 8000 ) are default values in the class __init__() function. The address of "localhost" is handy for testing a client and a server on your PC. First the socket is created, then it is bound to an address. If no exceptions are raised, then an input and output le are created to use this socket. The makeRequet() function sends a message and then reads the reply.
38.9.2 Server Programs

When a server program starts, it creates a socket on which it listens for requests. The server has a three-step response to a client. First, it accepts the connection, then it reads and processes the clients request. Finally, it sends a reply to the client. For some client-server relationships, like a database server, there may be multiple requests and replies. Since database requests may take a long time to process, the server must be multi-threaded in order to handle concurrent requests. In the case of HTTP, a single request will lead to multiple replies. A server program handles requests from a client by using the following outline of processing. 1. Create a Listener Socket. A listener socket is waiting for client connection requests. 2. Accept a Client Connection. When a client attempts a connection, the sockets accept() method will return a daughter socket connected to the client. This daughter socket is used for all subsequent processing. 3. Read the request. Many of the standard TCP/IP protocols expect the commands to be sent as strings of text, terminated with the n character. Often a Python le object is created from the socket so that the complete set of le method functions for reading and writing are available. 4. Send the reply. Many of the standard protocols will respond with a 3-digit numeric code indicating the status of the request. Well review some common variations on these codes, below.
494
Create and Listen on a Socket. The following methods are relevant when creating server-side sockets. These server side method functions are used for establishing the public socket that is waiting for client connections. In each denition, the variable s is a socket object. bind(address) Bind the socket to a local address tuple of ( IP Address and port number ). This tuple is the address and port that will be used by clients to connect with this server. Generally, the rst part of the tuple is simply to indicate that this server uses the address of the computer on which it is running. listen(queueSize) Start listening for incoming connections, queueSize species the number of queued connections. accept() -> ( socket, address) Accept a client connection, returning a socket connected to the client and client address. The original bound socket, which was set in listen mode is left alone, and is still listening for the next connection. Once the socket connection has been accepted, processing is a simple matter of reading and writing on the daughter socket. We wont show an example of writing a server program using simple sockets. The best way to make use of server-side sockets is to use the SocketServer module.
38.9.3 Practical Server Programs with SocketServer

Generally, we use the SocketServer module for simple socket processing. Usually, we create a TCPSocket using this module. This can simplify the processing of requests and replies. The SocketServer module is the basis for the SimpleHTTPServer (see The World Wide Web and the HTTP protocol ). Much of server-side processing is encapsulated in two classes of the SocketServer module. You will subclass the StreamRequestHandler class to process TCP/IP requests. This subclass will include the methods that do the essential work of the program. You will then create an instance of the TCPServer class and give it your RequestHandler subclass. The instance of TCPServer will to manage the public socket, and all of the basic processing. For each connection, it will create an instance of your subclass of StreamRequestHandler to handle the connection. Dene a RequestHandler. Dening a handler is done by creating a subclass of StreamRequestHandler or BaseRequestHandler and adding a handle() method function. The BaseRequestHandler denes a simple framework that TCPServer can use when data is received on a socket. Generally, we use a subclass of StreamRequestHandler. This class has methods that create les from the socket. This alliows the handle() method function to simply read and write les. Specically, the superclass will assure that the variables self.rfile and self.wfile are available. For example, the echo service runs in port 7. The echo service simply reads the data provided in the socket, and echoes it back to the sender. Many Linux boxes have this service enabled by default. We can build the basic echo handler by creating a subclass of StreamRequestHandler.
#!/usr/bin/env python """My Echo""" import SocketServer class EchoHandler( SocketServer.StreamRequestHandler ): def handle(self): input= self.request.recv(1024) print "Input: %r " % ( input, ) self.request.send("Heard: %r \n" % ( input, ) ) server= SocketServer.TCPServer( ("",7000), EchoHandler )
495
print "Starting Server" server.serve_forever()
This class can be used by a TCPServer instance to handle requests. In this, the TCPServer instance named server creates an instance of EchoHandler each time a connection is made on port 7. The derived socket is given to the handler instance, as the instance variable self.request. A more sophisticated handler might decode input commands and perform unique processing for each command. For example, if we were building an on-line Roulette server, there might be three basic commands: a place bet command, a show status command and a spin the wheel command. There might be additional commands to join a table, chat with other players, perform credit checks, etc. Methods of TCPServer. In order to process requests, there are two methods of a TCPServer that are of interest. handle_request() Handle a single request: wait for input, create the handler object to process the request. serve_forever() Handle requests in an innite loop. Runs until the loop is broken with an exception.
38.9.4 Protocol Design Notes

Generally, HTTP-based web services do almost everything we need; and they do this kind of thing in a simple and standard way. Using sockets is done either to invent something new or to cope with something very old. Using web services is often a better choice than inventing your own protocol. If you cant, for some reason, make suitable use of web services, here are some lessons gleaned from the reading the Internetworking Requests for Comments (RFCs). Many protocols involve a request-reply conversational style. The client connects to the server and makes requests. The server replies to each request. Some protocols (for example, FTP) may involve a long conversation. Other protocols (for example, HTTP) involve a single request and (sometimes) a single reply. Many web sites leverate HTTPs ability to send multiple replies, but some web sites send a single, tidy response. Many of the Internet standard requests are short 1- to 4-character commands. The syntax is kept intentionally very simple, using spaces for delimeters. Complex syntax with optional clauses and sophisticated punctuation is often an aid for people. In most web protocols, a sequence of simple commands are used instead of a single, complex statement. The responses are often 3-digit numbers plus explanatory comments. The application depends on the 3-digit number. The explanatory comments can be written to a log or displayed for a human user. The status numbers are often coded as follows: 1yz Preliminary reply, more replies will follow. 2yz Completed. 3yz More information required. In the case of FTP, this is typically the start of a dialog. In the case of HTTP, it is often a redirect. 4yz Request not completed; trying again makes sense. This is a transient problem like a deadlock, timeout, or le system problem. In the case of HTTP, this is also used for an authentication problem. 5yz Request not completed because its in error; trying again doesnt make sense. This a syntax problem or other error with the request. The middle digit within the response provides some additional information.
496
x0z The response message is syntax-related. x1z The response message is informational. x2z The response message is about the connection. x3z The response message is about accounting or authentication. x5z The response message is le-system related. These codes allow a program to specify multi-part replies using 1 yz codes. The status of a client-server dialog is managed with 3 yz codes that request additional information. 4 yz codes are problems that might get xed. 5 yz codes are problems that can never be xed (the request doesnt make sense, has illegal options, etc.) Note that protocols like FTP (RFC 959) provide a useful convention for handling multi-line replies: the rst line has a - after the status number to indicate that additional lines follow; each subsequent lines are indented. The nal line repeats the status number. This rule allows us to detect the rst of many lines, and absorb all lines until the matching status number is read.
497
498
Part VI
Projects
499
Projects to Build Skills

Programming language skills begin with the basic syntax and semantics of the language. They advance through the solution of small exercises and are rened through solving more complete problems. Real-world applications, used every day in business and research, are less than ideal for learning a programming language. The business-oriented problems often have a very narrow in focus; the solutions are dictated by odd budgetary constraints or departmental politics. Reasearch problems are also narrowly focused, often lacking a nal application to surround the interesting parts of the programming and create a nal, nished product. This part provides several large exercises that provide for more advanced programming than the smaller exercises at the end of each section. These arent real-world in scope, but they are quite a bit larger than the small exercises at the end of each chapter. These are ranked in order of diiculty.
501
502
CHAPTER
THIRTYNINE
AREAS OF THE FLAG

From Robert Banks, Slicing Pizzas, Racing Turtles, and Further Adventures in Applied Mathematics [Banks02]. This project is simple: it does not use loops, if-statements or any data structures. This exercise focuses on expressions and assignment statements. Facts about the American ag. The standard dimension is the width (or hoist). This is the basic unit that all others are multiplied by. A ag that is 30 inches wide will simply have 30 inches multiplied by all of the other measurements. Width (or hoist ) Wf = 1.0 Length (or y ) Lf = 1.9 Wf Width of union (or canton) Wu = Length of union Lu = 0.76 Wf Radius of a star R = 0.0308 Wf These are other facts; they are counts, not measurements. Number of red stripes Sr = 7 Number of white stripes Sw = 6 Number white stars Ns = 50 Width of a stripe Ws =
1 Sr +Sw 7 13
Wf
Wf
39.1 Basic Red, White and Blue

Red Area. There are 4 red stripes which abut the blue eld and 3 red stripes run the full length of the ag. We can compute the area of the red, since it is 4 short rectangles and 3 long rectangles. The short rectangle areas are the width of a stripe (Ws ) times the whole length (length of the y, Lf ) less the width of the blue union (Wu ). The long rectangle areas are simply the width of the stripe (Ws ) times the length of the y (Lf ). Red = 4 Ws (Lf Lu ) + 3 Ws Lf . White Area. There are 3 white stripes which abut the blue eld, 3 whie stripes run the full length of the ag, plus there are the 50 stars.
503
We can compute the basic area of the white using a similar analysis as area of the red, and adding in the areas of the stars, 50S . Well return to the area of the stars, last. W hite = 3 Ws (Lf Lu ) + 3 Ws Lf + 50S. Blue Area. The blue area is the area of the union, less the area of the stars, 50S . Blue = (Lu Wu ) 50S.
39.2 The Stars

Area of the Stars. A 5-sided star (pentagram) can be analyzed as 5 kites of 2 triangles. The area of each kite, K, is computed as follows.
The inner angles of all ve kites ll the inside of the pentagram, and the angles must sum to 360 , therefore each inner angle is 360 5 = 72. Angles of any triangle sum to 180 . The lower triangle is symmetric, therefore, the other two angles must sum to 180. The lower triangle has two side angles of (180 72)/2 = 54. We see that straight lines contain an outer triangle and two inner triangles. We know the inner triangles add to 54 + 54 = 108; a straight line is 180. Therefore, the outer triangle has two 72 corners and a 36 peak. The area of the two triangles can be computed from these two angles. a = 36 b = 72 Recall that the radius of a star, R is 0.0308 Wf . Heres one version of the area of the kite. K=
b sin a 2 sin 2 R2 1 sin ( a + b ) 2
Heres the other version of the area of the kite. K=

b sin 2 R2 2
Note that the math library math.sin() and math.cos() functions operate in radians, not degrees. The conversion rule is radians = 180 degrees. Therefore, we often see something like sin(a /180) for an angle, a, in degrees. The Golden Ratio is = 504
1+ 5 2
(about 1.61803). Chapter 39. Areas of the Flag
The total area of a star is S =5K Given a specic width of a ag (in feet), we can compute the actual areas of each color. Check Your Results. Blue is 18.73% of the total area.
39.2. The Stars
505
506
Chapter 39. Areas of the Flag
CHAPTER
FORTY
BOWLING SCORES
Bowling is played in ten frames, each of which allows one or two deliveries. If all ten pins are bowled over in the rst delivery, there is no second delivery. Each frame has a score based on the delivery in that frame, as well as the next one or two deliveries. This means that the score for a frame may not necessarily be posted at the end of the frame. It also means that the tenth frame may require a total of three deliveries to resolve the scoring. Rule A. The score for a frame is the total pins bowled over during that frame, if the number is less than ten (an open frame, or error or split depending some other rules beyond the scope of this problem). Rule B. If all ten pins are bowled over on the rst delivery (a strike), the score for that frame is 10 + the next two deliveries. Rule C. If all ten pins are bowled over between the rst two deliveries (a spare), the score for that frame is 10 + the next delivery. A game can be as few as twelve deliveries: ten frames of strikes require two additional deliveries in the tenth frame to resolve the rule B scoring. A game can be as many as twenty-one deliveries: nine open frames of less than 10 pins bowled over during the frame, and a spare in the tenth frame requiring one extra delivery to resolve the rule C scoring. There is a relatively straight-forward annotation for play. Each frame has two characters to describe the pins bowled during the delivery. The nal frame has three characters for a total of 21 characters. Rule A: If the frame is open, the two characters are the two deliveries; the total will be less than 10. If a delivery fails to bowl over any pins, a "-" is used instead of a number. Rule B: If the frame is strike, the two characters are "X ". No second delivery was made. Rule C: If the frame is a spare, the rst character is the number of pins on the rst delivery. The second character is a "/". For example:
"8/9-X X 6/4/X 8-X XXX"
This can be analyzed into ten frames as follows:
507
Frame First delivery 1 8 2 3 4 5 6 7 8 9 10 9 10 10 6 4 10 8 10 10
Second delivery /, must have been 2 -, 0 (not taken) (not taken) /, must have been 4 /, must have been 6 (not taken) -, 0 (not taken) 10 and 10
Scoring rule C- spare = 10 + next delivery A- open = 9 B- strike = 10 + next 2 deliveries B- strike = 10 + next 2 deliveries C- spare = 10 + next delivery C- spare = 10 + next delivery B- strike = 10 + next 2 deliveries A- open = 8 B- strike = 10 + next 2 deliveries B- strike = 10 + next 2 deliveries, two extra deliveries are taken during this 10th frame.
Frame Score 19 9 26 20 14 20 18 8 30 30
Total 19 28 54 74 88 108 126 134 164 194
Each of the rst nine frames has a two-character code for each delivery. There are three forms: "X ". "n/" where n is - or 1-9. "nm" where n and m are - or 1-9. The two values cannot total to 10. The tenth frame has a three-character code for each of the deliveries. There are three forms: "XXX". "n/r" where n is -, 1-9 and r is X, -, 1-9. "nm " where n and m are - or 1-9. The two values cannot total to 10. Write a valid(game)() function that will validate a 21-character string as describing a legal game. Write a scoring method, scores(game)(), that will accept the 21-character scoring string and produce a sequence of frame-by-frame totals. Write a reporting method, scoreCard(game)(), that will use the validation and scoring functions to produce a scorecard. The scorecard shows three lines of output with 5 character positions for each frame. The top line has the ten frame numbers: 2 digits and 3 spaces for each frame. The second line has the character codes for the delivery: 2 or 3 characters and 3 or 2 spaces for each of the ten frames. The third line has the cumulative score for each frame: 3 digit number and 2 spaces. The game shown above would have the following output.
1 8/ 19 2 928 3 X 54 4 X 74 5 6/ 88 6 4/ 108 7 X 126 8 8134 9 X 164 10 XXX 194
508
Chapter 40. Bowling Scores
CHAPTER
FORTYONE
MUSICAL PITCHES
A musician will call the frequency of a sound its pitch. When the frequencies of two pitches dier by a factor of two, we say they harmonize. We call this interval between the two pitches an octave. This perception of harmony happens because the two sounds reinforce each other completely. Indeed, some people have trouble telling two notes apart when they dier by an octave. This trivial example of harmony is true for all powers of two, including ..., 1/8, 1/4, 1/2, 1, 2, 4, 8, .... Classical European music divided that perfectly harmonious factor of two interval into eight asymmetric steps; for this historical reason, it is called an octave. Other cultures divide this same interval into dierent numbers of steps with dierent intervals. More modern European music further subdivides the octave, creating a 12-step system. The most modern version of this system has 12 equally spaced intervals, a net simplication over the older 8-step system. The pitches are assigned names using ats () and sharps (), leading to each pitch having several names. Well simplify this system slightly, and use the following 12 names for the pitches within a single octave: A A B C C D D E F F G G The eight undecorated names (A through G and an A with a frequency double the original A) form our basic octave; the additional notes highlight the interesting asymetries. For example, the interval from A to B is called a whole step or a second, with A being half-way between. The interval from B to C, however is only a half step to begin with. Also, it is common to number the various octaves as though the octaves begin with the C, not the A. So, some musicians consider the basic scale to be C, D, E, F, G, A, B, with a C in the next higher octave. The higher C is twice the frequency of the lower C.
509
The tuning of an instrument to play these pitches is called its temperament. A check on the web for reference material on tuning and temperament will reveal some interesting ways to arrive at the tuning of a musical instrument. It is suprising to learn that there are many other ways to arrive at the 12 steps of the scale. This demonstrates that our ear is either remarkably inaccurate or remarkably forgiving of errors in tuning. Well explore a number of alternate systems for deriving the 12 pitches of a scale. Well use the simple equal-temperament rules, plus well derive the pitches from the overtones we hear, plus a more musical rule called the circle of fths, as well as a system called Pythagorean Tuning. Interesting side topics are the questions of how accurate the human ear really is, and can we really hear the dierences? Clearly, professional musicians will spend time on ear training to spot ne gradations of pitch. However, even non-musicians have remarkably accurate hearing and are easily bothered by small discrepencies in tuning. Musicians will divide the octave into 1200 cents. Errors on the order of 50 cents, 1/24 of the octave, are noticable even to people who claim they are tone deaf. When two tunings produce pitches with a ratio larger than 1.0293, it is easily recognized as out of tune. These exercises will make extensive use of loops and the list data structure.
41.1 Equal Temperament

The equal temperament tuning divides the octave into twelve equally sized steps. Moving up the scale is done by multiplying the base frequency by some amount between 1 and 2. If we multiply a base frequency by 2 or more, we have jumped to another octave. If we multiply a base frequency by a value between 0 and 0.5, we have jumped into a lower octave. When we multiply a frequency by values between 0.5 and 1, we are computing lower pitches in the same octave. Similarly, multiplying a frequency by values between 1 and 2 computes a higher pitch in the same octave. We want to divide the octave into twelve steps: when we do a sequence of twelve multiplies by this step, we should arrive at an exact doubling of the base frequency. The steps of the octave, then, would be b, b s, b s s, ..., b s12 . This step value, therefore is the following value. s=e
log 2 12
If we multiply this 12 times for each of the 12 steps, we nd the following. s12 = e12
log 2 12
=2
For a given pitch number, p, from 0 to 88, the following formula gives us the frequency. We can plug in a base frequency, b of 27.5 Hz for the low A on a piano and get the individual pitches for each of the 88 keys. f = b 2p = b ep
log 2 12
(41.1)
Actual piano tuning is a bit more subtle than this, but these frequencies are very close to the ideal modern piano tuning. Equal Temperament Pitches. Develop a loop to generate these pitches and their names. If you create a simple tuple of the twelve names shown above (from A to G ), you can pick out the proper name from the tuple for a given step, s, using int( s % 12 ). Check Your Results. You should nd that an A has a pitch of 440, and the G ten steps above it will be 783.99 Hz. This 440 Hz A is the most widely used reference pitch for tuning musical instruments.
510
Chapter 41. Musical Pitches
41.2 Overtones
A particular musical sound consists of the fundamental pitch, plus a sequence of overtones of higher frequency, but lower power. The distribution of power among these overtones determines the kind of instrument we hear. We can call the overtones the spectrum of frequencies created by an instrument. A violins frequency spectrum is distinct from the frequency spectrum of a clarinet. The overtones are usually integer multiples of the base frequency. When any instrument plays an A at 440 Hz, it also plays As at 880 Hz, 1760 Hz, 3520 Hz, and on to higher and higher frequencies. While we are not often consciously aware of these overtones, they are profound, and determine the pitches that we nd harmonious and discordant. If we expand the frequency spectrum through the rst 24 overtones, we nd almost all of the musical pitches in our equal tempered scale. Some pitches (the octaves, for example) match precisely, while other pitches dont match very well at all. This is a spread of almost ve octaves of overtones, about the limit of human hearing. Even if we use a low base frequency, b, of 27.5 Hz, it isnt easy to compare the pitches for the top overtone, b 24, with a lower overtone like b 8: theyre in two dierent octaves. However, we can divide each frequency by a power of 2, which will normalize it into the lowest octave. Once we have the lowest octave version of each overtone pitch, we can compare them against the equal temperament pitch for the same octave. The following equation computes the highest power of 2, p2 , for a given overtone multiplier, m, such that f p2 f. 2 <b2 log m p2 = (41.2) log 2 Given this highest power of highest power of 2, p2 , we can normalize a frequency by simple division to create what we could call the rst octave pitch, f0 . f0 = f 2p2 (41.3)
The list of rst octave pitches arrives in a peculiar order. Youll need to collect the values into a list and sort that list. You can then produce a table showing the 12 pitches of a scale using the equal temperament and the overtones method. They dont match precisely, which leads us to an interesting musical question of which sounds better to most listeners. Overtone Pitches. Develop a loop to multiply the base frequency of 27.5 Hz by values from 3 to 24, compute the highest power of 2 required to normalize this back into the rst octave, p2 , and compute the rst octave values, f0 . Save these rst octave values in a list, sort it, and produce a report comparing these values with the closest matching equal temperament values. Note that you will be using 22 overtone multipliers to compute twelve scale values. You will need to discard duplicates from your list of overtone frequencies. Check Your Results. You should nd that the 6th overtone is 192.5 Hz, which noralizes to 48.125 in the st octave. The nearest comparable equal-tempered pitch is 48.99 Hz. This is an audible dierence to some people; the threshold for most people to say something sounds wrong is a ratio of 1.029, these two dier by 1.018.
41.3 Circle of Fifths

When we look at the overtone analysis, the second overtone is three times the base frequency. When we normalize this back into the rst octave, it produces a note with the frequency ratio of 3/2. This is almost 41.2. Overtones 511
as harmonious as the octave, which had a frequency ratio of exactly 2. In the original 8-step scale, this was the 5th step; the interval is called a fth for this historical reason. It is also called a dominant. Looking at the names of our notes, this is E, the 7th step of the more modern 12-step scale that starts on A. This pitch has an interesting mathematical property. When we look at the 12-step tuning, we see that numbers like 1, 2, 3, 4, and 6 divide the 12-step octave evenly. However, numbers like 5 and 7 dont divide the octave evenly. This leads to an interesting cycle of notes that are separated by seven steps: A, E, B, F , C , .... We can see this clearly by writing the 12 names of notes around the outside of a circle. Put each note in the position of an hour with A in the 12-oclock position. You can then walk around the circle in groups of seven pitches. This is called the Circle of Fifths because we see all 12 pitches by stepping through the names in intervals of a fth. This also works for the 5th step of the 12-step scale; the interval is called a fourth in the old 8-step scale. Looking at our note names, it is the D. If we use this interval, we create a Circle of Fourths. Write two loops to step around the names of notes in steps of 7 and steps of 5. You can use something like range( 0, 12*7, 7 ) or range( 0, 12*5, 5 ) to get the steps, s. You can then use names[s % 12] to get the specic names for each pitch. Youll know these both work when you see that the two sequences are the same things in opposite orders. Circle of Fifths Pitches. Develop a loop similar to the one in the overtones exercise; use multipliers based on 3/2: 3/2, 6/2, 9/2, .... to compute the 12 pitches around the circle of fths. Youll need to compute the highest power of 2, using (41.2), and normalize the pitches into the rst octave using (41.3). Save these rst octave values in a list, indexed by s % 12; you dont need to sort this list, since the pitch can be computed directly from the step. Rational Circle of Fifths. Use the Python rational number module, fractions to do these calculations, also. Check Your Results. Using this method, youll nd that G could be dened as 49.55 Hz. The overtone analysis suggested 48.125 Hz. The equal temperament suggested 48.99 Hz.
41.4 Pythagorean Tuning

When we do the circle of fths calculations using rational numbers instead of oating point numbers, we nd a number of simple-looking fractions like 3/2, 4/3, 9/8, 16/9 in our results. These fractions lead to a geometrical interpretation of the musical intervals. These fractions correspond with some early writings on music by the mathematician Pythagoras. Well provide one set of commonly-used list of fractions for Pythagorean tuning. These can be compared
512
with other results to make the whole question of scale tuning even more complex. A A B C C D D E F F G G 1:1 256 : 243 9:8 32 : 27 81 : 64 4:3 729 : 512 3:2 128 : 81 27 : 16 16 : 9 243 : 128
Pythagorean Pitches. Develop a simple representation for the above ratios. A list of tuples works well, for example. Use the ratio to compute the frequencies for the various pitches, using 27.5 Hz for the base frequency of the low A. Compare these values with equal temperament, overtones and circle of fths tuning. Check Your Results. The value for G is 27.5 16 9 = 48.88 Hz.
41.5 Five-Tone Tuning

The subject of music is rich with cultural and political overtones. Well try to avoid delving too deeply into anything outside the basic accoustic properties of pitches. One of the most popular alternative scales divides the octave into ve equally-spaced steps. This tuning produces pitches that are distinct from those in the 12 pitches available in European music. The original musical tradition behind the blues once used a ve step scale. You can revise the formula in (41.1) to use ve steps instead of twelve. This will provide a new table of frequencies. The intervals should be called something distinctive like V, W, X, Y, Z and V in the second octave. Five-Tone Pitches. Develop a loop similar to the 12-tone Equal Temperament (Equal Temperament ) to create the 5-tone scale pitches. Note that the 12-tone scale leads to 88 distinct pitches on a piano; this 5-tone scale only needs 36. Compare 12-Tone and 5-Tone Scales. Produce a three column table with the 12-tone pitch names and frequencies aligned with the 5-tone frequencies. You will have to do some clever sorting and matching. The frequencies for V will match the frequencies for A precisely. The other pitches, however, will fall into gaps. The resulting table should look like the following name 12 A A# B freq. 440.0 466.16 493.88 name 5 V freq. 440.0
505.42
41.5. Five-Tone Tuning
513
514
CHAPTER
FORTYTWO
WHAT CAN BE COMPUTED?

This project will look at Finite State Machines, one of the foundations of computing. Well provide some background on why these are important in Background. In The Turing Machine well discuss the Turing Machine in some detail. Well provide an example in Example Machine. The essential exercise is to write the necessary Python software to emulate a Turing Machine. Well look at the implementation closely in Turing Machine Implementation. Exercise 1 is the rst set of programming exercises. In Test Machines well look at some basic Turing Machine specications that we can use for testing. Exercise 2 will implemnet some of those additional machines to act as conrmation and as unit tests. As a digression, well provide additional background in Consequences. This background answers some of the questions raised in the Background section. We can then look at further applications of Finite State Machines in Other Applications. This includes some simple lexical scanning and parsing applications. Well look at some variations on the implementation of our Turning Machine in Alternative Specications
42.1 Background
In the 1930s at the very dawn of digital computation mathematician Alan Turing wondered what numbers could be computed by a digital computer. Indirectly, he was wondering what should a computer be designed to do. What was a sensible set of constraints or parameters around computer and programming language? At that time computer was a persons job title. A mathematician, like Alan Turing, would create worksheets for people to use in computing a particular result. Turing worked on code breaking during World War II, and his computers were people who carried out rather complex analyses of coded messages. Turing prepared the instructions and worksheets for his computers to decode intercepted messages. The question of computable numbers to modern programmers can seem a bit silly. The unthinking answer is that digital computers can represent any number. But we have a bit of a problem. Mathematicians have identied a pretty wide variety of numbers. Can we represent all rational fractions? What about all irrational decimals? What about all complex numbers? Are there yet other kinds of numbers that a computer cant represent? The essential question is what numbers should we be able to compute with Python? Python. This essential computability question denes how we design computer hardware and how we design programming languages. The rst step, then is to dene what is computable. Clearly, a language like Python must have some minimal set of features. Further, there must be some standard of computability that a
515
language designer must respect in creating the language. In the next section, well look at the subject of computability. Once we agree on what is computable, we can then match a given programming language (e.g., Python) against the denition to see if our programming language is complete. This sounds like a daunting, heavyweight mathematical proof, but actually, its relatively simple. In this section, well construct the proof that Python is a complete language. Really. The denition of computable is based on a hypothetical model of a computer. We can (and will) build that hypothetical computer model in Python. The Python program that implements the hypothetical computer is the constructive proof that the Python language completely ts the hypothetical model of computability. The essential notion of computability is based on the idea of computable numbers. So well look at those briey. Numbers. A common mistake is to claim that there are just numbers. This is clearly untrue. We can easily dene natural numbers with a pleasant set of axioms. We can say that we have a rst number, call it 0. We can dene a successor numbers, and the properties of equality and create an innite set of numbers. Yes, there are an innite set of numbers. But we have this concept of successor, so in principle we can count this set. Interestingly, this set doesnt include negative numbers. Okay, ne, well add negative numbers to our innite set of natural numbers, and we have a larger innite set of numbers. Turns out, we can work out a mapping from natural numbers to signed numbers. This makes the set of signed integers countably innite. The long class is a physical implementation of this abstact concept of signed natural number. Also, the decimal.Decimal class, with a xed decimal point, is essentially the same kind of number. This set of integers doesnt include any fractions. We can easily dene an innite set of fractions using any two of our innite set of natural numbers. This gives us a set of rational numbers thats obviously also innite. It seems to be a larger set, but we can work out a mapping from natural numbers to rational fractions. This makes the rational numbers countably innite, also. The Rational class (see Rational Numbers) is an implementation of this abstract concept of rational number. Yet More Numbers. The set of rational numbers, however, still isnt the nal word. We can dene some algebraic numbers, which are solutions to equations that arent rational nunmbers. For instance, solving this polynomial for x, x2 2 = 0. The answer is x = 2; no rational number can satisfy this. This means that there must be irrational numbers. We can combine the irrational and rational numbers into a big set of numbers, commonly called real numbers. If we start writing down numbers as long strings of decimal digits (or binary bits) we can with some cleverness create numbers that we cannot easily map to the natural numbers. This means that innite decimal fractions form a larger set of numbers thats uncountably innite. Rats. The float class is a nite implementation these innite binary fractions. Since the float class is actually nite, it suers from some practical limitations, but, in spirit, it is intended to follow this kind of number. Were still not done. solving this polynomial for x, x2 + 1 = 0, leads us to complex numbers. Okay. Fine, there are a lot of numbers. The rst question is, are all of these computable? Essentially, we are also asking can we dene computable? Wed really like a constructive denition of computable that matches our vague notions of what a computer should do. Profound Limitations. For purposes of simplication, we need to set aside the nite part of physical computers and discuss a hypothetical innite computer. Why should talk about theoretical computers? The idea behind this is to divorce a simple, formal denition of computer from any specic and complex piece of hardware. If we try to discuss specic pieces of hardware, we have lots of weird limitations and
516
Chapter 42. What Can be Computed?
gotchas based on limited memory, available power, limited desktop space, dissipation of heat, radio-frequency interference, laws of physicas, etc. The point is to rst talk about whats possible before talking about whats practical. Before we even start, we need to know if there any profound limitations on whats possible. Since the answer is yes, there are theoretical limitations, we need to know exactly what those limitations are. Goals. Fundamentally, the overall goal has two parts. 1. Show that there is a Turing Machine which can compute any particular mathematical function. This denes the theoretical possibilities for computable numbers. Well skip over this. 2. Create a Python implementation of a Turing machine. This will establish that the Python language is in principle capable of all the theoretical computations. Well trust the mathematicians that the Turing Machines computable numbers are a good match for what we understand numbers to be. It turns out that the match isnt perfect; well look at this briey in Consequences.
42.2 The Turing Machine

To talk about computability, we need a general-purpose (and simple) model for a computer. We need something that we can dene clearly and completely. We also need something that captures what we mean by compute. We need input, output, state change, and some kind of program to control those state changes. Well call our theoretical computer a Turing Machine, naming it after mathematician Alan Turing. The essential feature of all computation is state change. We introduced the idea in Variables, Assignment and Input . Therefore, a hypothetical computer must have some current internal state and some rules for changing the internal state based on inputs. This internal state is as simple as ag or marker that identies which transition rules are active. If we were to number each rule, the state would be the current rule number. If we had a person following instructions on a worksheet, we might ask them to put a removable sticky-note on the step they were following so they wouldnt lose their place. In addition to internal state, we need some form of input, output and (possibly) working memory. Turings idea was to use an innitely long tape for this. [Innite? Yes. Its an abstraction; we need to dene the set of countably innite numbers, so we need an innite tape.] The idea is that we can mark the tape with some values, start the machine processing. When it stops, we can look at the tape to see the answer. Alphabet. Our machine must be able to record state on the tape. We need to dene some alphabet of symbols. A minimal alphabet is two distinct symbols and nothing more. We could use X and O or and . We could use a larger alphabet of the 128 ASCII characters that Python uses. Or we could use the vast library of Unicode symbols. The only restriction is that the symbols are distinct and the alphabet of symbols is limited and known in advance. For now, well stick to just two symbols. The Tape. Our Turing machine, then, has as its visible component, a tape which shuttles back and forth under a read/write station. Or, we could think of a little read/write device that walks up and down a long tape laid out on the oor of a vast building. There are some great YouTube videos of Turning Machines. Feel free to watch those for a better sense of that a Turing Machine might be like.
42.2. The Turing Machine
517
The interesting part is the current position of the read/write device relative to the tape itself. Heres a sample tape with a few symbols. Well use brackets to show where the read/write device is positioned. [] ... Clearly, this alphabet includes at least the symbols: , , . Also, the current position is the fourth one on the tape. The Rules. Our Turning Machine has an internal state to remember where it is, an innite tape that is input, output and working memory. It also has some rules for deciding what to do. To keep things simple, each rule includes the following things: Symbol. This is the symbol to be matched against the current symbol on the tape. Since the alphabet is nite, well have one rule per symbol. Write. Write a new symbol or do no writing, leaving the tape unchanged. A purist could argue that writing the matched symbol back to the tape is the same as not writing. Move. The machine can move the tape forward one, backward one, or not at all. Next State. This next state is the same or dierent collection of rules. This can be simplied. It can also be made more complex. Well look at some alternatives below. For now, this will give us enough to work with. For a give state, we have a set of rules that species the moving and writing for each possible symbol. If we have a small alphabet of just two symbols, each state has only two rules. If we have a larger alphabet, each state must have a larger number of rules. Halting. Its sometimes helpful to have a special rule to stop the machine from further processing. This halt rule is simply a special case rule that doesnt have a next state. Instead of going to a next state, the machine stops working. A rule that doesnt write, doesnt move and species no state change must halt, since the machine will stay in that state forever.
42.3 Example Machine

Lets use a very small alphabet of and .
Well dene a very simple machine to check the parity of a tape. We want to be sure that number of symbols is even. If its odd, well write an extra symbol. If its even, well write an extra . This kind of algorithm is used with 7-bit ASCII characters to create 8-bit even-parity for transmission across noisy wires. If the received 8-bit value does not have even parity, the character is ignored. Here are the two states. 1. Even. If : write nothing, move tape forward, change to state 1. If : write nothing, move tape forward, stay in state 0. If blank: write , change to state 2. 2. Odd. If : write nothing, move tape forward, change to state 0. If : write nothing, move tape forward, stay in state 1. If blank: write , change to state 2. 3. Halt. For all symbols, write nothing, do not move, change to state 2.
518
Lets look at how this works with a sample tape. 1. The machine starts in the State 0. The tape has ve symbols, and were positioned at the start. This is the starting condition. [ ] 2. The state is State 0. The symbol is 1. ...
, so the machine moves the tape forward and switches to State [ ] ...
3. The state is State 1. The symbol is 0.
, so the machine moves the tape forward and switches to State [] ...
4. The state is State 0. The symbol is , so the machine moves the tape forward and stays in State 0. [ ] ... 5. The state is State 0. The symbol is 1. , so the machine moves the tape forward and switches to State [] ...
6. The state is State 1. The symbol is , so the machine moves the tape forward and stays in State 1. [ ] ... and changes
7. The state is State 1 and the tape is blank: the machine moves th etape forward, writes to State 0. [ ] ...
8. The state is State 0 and the tape is blank: the machine writes a and changes to State 2. [] ... on the tape.
9. In State 2, the machine halts. There are an even number of [] ...
42.4 Turing Machine Implementation

Our job is to write a Python script that simulates this kind of Turing Machine. The existence of such a script is proof that the Python language does everything a Turing Machine does. (Separately, we need to prove that a Turing Machine does everything that a mathematician could ever want.) Well need a model of the tape. A Python list can accomplish this nicely, we can populate the list with symbols like this:
42.4. Turing Machine Implementation
519
tape = [ 'X', 'X', 'O', 'X', 'O', None, None, None ]
By adding a bunch of blank cells at the end of the tape, we can simplify our implementation. Theoretically, we have an innite number of blank cells on the tape; therefore a well-written Turing Machine program will simply appned empty cells to the list as long as necessary. To get started, though, its easy to simply pad the tap with blank cells. Well need a model of the rules. Each rule can be encoded as a tuple (or namedtuple) that has the required state change: the tape movement, symbol to write and the next state.
Rule = namedtuple( 'Rule', ['write', 'move', 'state'] )
For the write attribute, we can provide the symbol to be written, or None if nothing is to be written. For the move attribute, we can use the values +1, -1 or None. For the state attribute, we can provide the number of the next state, or None if the machine should halt. We can then code each state of our machine as a dictionary of rules that map a symbol on the tape to a state change tuple that shows what to do.
rule_0_map = { 'X': Rule(None,+1,1), 'O': Rule(None,+1,0), None: Rule(None,None,None) }
We can then code our transition rules as a whole as a dictionary that maps rule numbers to rules.
transitions = { 0: rule_0_map, 1: rule_1_map, }
In order to simulate a Turing Machine, we need the following things: The initial tape. The transitions, which is a dictionary that maps a state to a set of rules. Each rule maps a tape symbol to a state change. A starting state number and starting tape position. A relatively simply Python function can then step through this Turning machine cycle until the machine halts.
Turing Machine Function

Let T be a tape, a list of symbols. Let M be the complete set of machine states and rules, M = {S0 , S1 , S2 , ...}. Each state is set of rules mapping from a symbol, C, to a state change, Sn = {C0 w, m, r0 , C1 w, m, r1 , ....}. Each state change, w, m, r species what symbol to write, w, what direction to move, m, and the next state, r. 1. Initialize. Set the initial tape position, P to 0, the start of the tape. Set the current state, S to state 0. 1. Cycle. While the machines state is not the halt state of None.
520
(a) Eval. Set C to the symbol at the current position, P, on the tape, T. Given current state, S, and current symbol, C, determine which state change to use. (b) Apply. Given the state change, w, m, r, do the following. If indicated, write the requested symbol, w, in the new position. This updates the tape, T and the current position, P. Move the tape in the appropriate direction. This updates the current position, P : next is m = +1, previous is m = 1. No movement is None. Moving the tape to the next position may require extending the tapes list object to add a blank cell. Set the current state, S, to the next current state, r. (c) Log. Optionally, print the current state of the machine to a log. 2. Display. Print the nal state and nal content of the tape.
42.5 Exercise 1
1. Dene a function that implements the basic Turing Machine. turing(transitions, tape) Accepts a dictionary of transition rules and an input tape. Starts in rule 0 with tape at position 0. Executes the various transitions until the machine reaches a transition where the next state is None, then it halts. Prints a log showing the tape as it moves and changes. 2. Create a set of transitions to do the Even X processing. If the number of Xs on the tape even, halt. If the number of Xs on the tape is odd, append an X and halt.
42.6 Test Machines

The point of dening this Turing Machine is to dene whats Computable. Lets start by computing various kinds of natural numbers. Lets say the input will encode numbers of as a sequence of "X" terminated by at least one "O". For example, Zero is a tape with ["O"]. One is a tape with ["X", "O"]. First, consider creating a machine to do a simple add 1 operation.
Add 1 Machine
1. If symbol is ["O"]: dont move, write ["X"], change to state 2. If symbol is ["X"]: move to the next cell, stay in state 1. 2. For all symbols, move to the next cell, write O, change to state 3. 3. For all symbols, halt. 42.5. Exercise 1 521
Second, consider creating a machine to do a simple subtract 1 operation.
Subtract 1 Machine
1. If symbol is ["O"]: Move to the previous cell, switch to state 2. Note that if the tape has zero, this will fall of the front of the tape, breaking the machine. If symbol is ["X"]: move to the next cell, stay in state 1. 2. For all symbols, write ["O"], change to state 3. 3. For all symbols, halt. We can see that with an add 1 machine and a subtract 1 machine, we can do a great deal of mathematics on natural numbers. Add Two Numbers. Consider a machine that will add two numbers. Wed start with a tape that encoded two numbers, for example ["X", "X", "O", "X", "X", "X", "O"]. This machine would terminate with a sum with some extra "O" symbols. A result like ["O", "O", "O", "X", "X", "X", "X", "X", "O"] would be expected. We can implement this by combining our add 1 and subtract 1 machines. We need to an add 1 to the right-hand number and a subtract 1 on the left-hand number until the left-hand number vanishes in a urry of "O" symbols. Subtract Two Numbers. We can also subtract two numbers. Wed start with a tape that encoded two numbers, for example ["X", "X", "O", "X", "X", "X", "O"]. This machine would do subtract 1 from each number until one of them has been reduced to zero. Depending on the locations of the "O" symbols we have a positive or negative result. Yet More Math. Multiplication is repeated addition. Given two numbers, call them, l and r, wed do this via (1) copy r to the end of the tape and (2) subtract 1 from l until r was reduced to zero. You can also think of this machine a having three numbers on the tape: l and r and p. The machine would start by creating an extra "O" at the end of the tape to, eectively set p to zero. Then the cycle is add r to p and subtract 1 from l, both of which weve already implemented. Division, similarly, is repeated subtraction. Given two numbers, call them, l and r, create a third number, q at the end of the tape. To divide l and r, subtract r from l and add 1 to q. This is repeated until l is less than r. Note that comparison is easily done via subtraction. However, the subtraction dened above is destructive, in that it updates both numbers. One way to handle this is to make a complete copy of l and r for comparison purposes. Optimization. Note that these naive repeated additional or repeated subtraction algorithms are ineicient. The point is to dene whats computable in principle. Once we know whats computable, we can work at optimization later.
42.7 Exercise 2
1. Build the add 1 and subtract 1 machines. 2. Build a test driver program. This will exerise a given machine with particular inputs and check the outputs.
522
Youll need a function to create an input tape from a simple tuple. A tuple like (2, 3) should create a tape of ["X", "X", "O", "X", "X", "X", "O"]. Youll need a function to summarize the resulting tape into a simple tuple. A tape like ["X", "X", "X", "O", "X", "X", "O"] should produce the tuple (2, 3). You can then dene a machine, and an input tape. The resulting tape can be easily checked for the right answer. Once we have the test driver, we can explore some alternate implementations. We can also explore some alternate denitions of Turing Machine processing rules.
42.8 Better Implementations

The initial implementation (using various kinds of strings and integers) suers from a number of problems. We can (and will) use the State design pattern to fundamentally restructure the denition of a machine. Machine. A Machine is a class denition that includes several attributes and methods. The Tape. The tape on which the Machine is currently working. This object can be responisble for tapes current position, plus the next and previous movements. Additionally, this class can handle growing the tape so that the next operation always works. The collection of states. Each state is an instance of the State class. Each State object is a collection of Rule objects. The current State. This is a variable that contains the machines current state. Machine.next(), Machine.prev() methods to update the position of the tape. Machine.write() method to change the symbol on the tape. A Machine.run() method that accepts a tape, a state collection, and an identier for a starting state. It steps through the machines operating cycle with the given tape, printing a log of the tape at each step. State. A State is a class denition that includes a rule for each symbol. We used a simple dictionary in the rst implementation. A subclass of dictionary will be ne. We dont need to extend it, so we can have something like
class State( dict ): pass
State Change. We can use the Command design pattern to dene some reusable command objects. Well need a bunch of Command classes and instances. Each StateChange object is given a particular machine to update. We can dene an abstract super class called StateChange. Each specic state change is a polymorphic subclass to make specic changes to the Machines Tape. Write. A StateChange subclass with a Next.__call__( self, machine )() method which calls machine.write(self.symbol). No_Write. A StateChange subclass with a Next.__call__( self, machine )() method which does nothing. Next. A StateChange subclass with a Next.__call__( self, machine )() method which updates the machine by calling machine.next().
42.8. Better Implementations
523
Prev. A StateChange subclass with a Next.__call__( self, machine )() method which updates the machine by calling machine.prev(). Stay. A StateChange subclass with a Next.__call__( self, machine )() method which does nothing. Rule. Each Rule has several parts that we initially encoded as strings. Given these Command class denitions above, we can create a bunch of individual objects that slightly simplify our machine denition.
write_X = Write("X") write_O = Write("O") nothing = No_Write() next = Next() prev = Prev() stay = Stay() halt = None
Now we can create Rule objects using these objects instead of string literals. Instead of checking the rules attributes for string values, we can directly invoke the rules attributes, since theyre callable objects. A rule would now look like this.
rule_1_map = { 'X': Rule(nothing, next, 0), 'O': Rule(nothing, next, 1), None: Rule(write_x, next, halt) }
In our machine, we can then revise how we evaluate the rules. Instead something like this:
if rule.move == "next": machine.next() elif rule.move == "next": machine.next() else: assert rule.move is None if rule.write is not None: machine.write( rule.write )
We can do the following:

rule.move( machine ) rule.write( machine )
Eliminating the if statements will result in a net speedup as well as a simplication. Tape. The Tape class has dene the tape as a list. It can also implement the next and previous, including a feature to extend the tape as necessary. This class can handle current position of the tape, also.
42.9 Exercise 3
1. Dene the required classes for an object-based Turing Machine implementation. Machine State StateChange Rule
524
Tape 2. Dene a function that implements this new Turing Machine. turing_obj(transitions, tape) Accepts a dictionary of transition rules built using objects instead of strings and an input tape. Starts in rule 0 with tape at position 0. Executes the various transitions until the machine reaches a transition where the next state is None, then it halts. Prints a log showing the tape as it moves and changes. 3. Using your existing test driver from Exercise 2, assure yourself that the new machine works as well as the old machine.
42.10 Consequences
As we noted above, the set of real numbers is uncountably innite. However, this Turing Machine denition of numbers can be shown to be countably innite. This means that there are some real numbers which cannot be computed. In spite of this gap between the two orders of innity, there are enough real numbers that can be computed that its fairly diicult to dene the kinds of numbers which are not computable. Also, the set of computable numbers is closed under ordinary arithmetic operations: given two computable numbers the sum, dierence, product or quotient are also computable. Whats more important is that our two implementations of Turing Machines prove that Python can easily work with all computable numbers. Consequently, the Python language (like all other programming languages) is Turing Complete and suitable for use doing anything that involves numbers. Since we can encode almost anything as a number, we can create Python programs to process this domain of almost anything. There are at the fringes of mathematics some things which cannot be encoded as numbers. Thats way beyond the scope of this book. If youre interested, though, you can research the Halting Problem, Chaitins Constant and The Busy Beaver Problem for some things which are clearly not computable and things which may not be computable.
42.11 Other Applications

Some simple variations on the basic Turing Machine act as a pattern recognizers. These Finite State Machines are the typical solution to a large number of problems. As an example, consider a machine to recognize simple dates in the form mm/dd/yyyy. Well tolerate a single digit month or a single digit day. But we must have all four digits for the year. Clearly, well have to expand our set of symbols to include all Unicode characters. Also, its simpler if we work with character classes (i.e., digits) instead of patiently listing each symbol. We can if necessary prove that a list of symbols is equivalent to a single symbol by patiently building the equivalent Turing Machine. The individual rule is simply repeated for each symbol in the class. Also, were not so much interested in processing a vague, general Tape. Were interested in processing a specic string. However, a string for our purposes is a sequence and therefore is almost equivalent to a list. Finally, were never going to write. Its a read-only tape of symbols. Instead of writing, were going to holt in one of two states: an accepting state or a rejecting state. If we halt in an accepting state, the string of characters matched the desired pattern.
42.10. Consequences
525
Short-Hand. Well use a short-hand to dene the Turing Machine that matches the above string. Well just write the symbol, and the next state like this: digit: 1. Well assume that we always move to next position on the input tape. Further, for all other characters, not given explicitly in the rules, the machine simply halts in a rejecting state.
Pattern Matcher
1. digit: 1. 2. digit: 2. /: 3. 3. /: 3. 4. digit: 4. 5. digit: 5. /: 6. 6. /: 6. 7. digit: 7. 8. digit: 8. 9. digit: 9. 10. digit: 10. 11. At this point, were at the end of the input string. We can halt, accepting the string as matching the pattern. Building the Machine. One can see that the state transitions in this kind of simple pattern-matching machine are so simple that a short-hand notation could be used. For example: "\d\d?/\d\d?/\d\d\d\d" would describe the pattern. In this case, "\d" stands for match a digit and advance to the next state. And "\d?" stands for match a digit and advance to the next state or look ahead to the next character and advance to the next state if it will match. Refer back to the section on Complex Strings: the re Module. In essence, a regular expression pattern denes a kind of nishte-state machine to match the given string as if it was a specialized one-way tape.
42.12 Alternative Specications

There are number of alternative formulations for the Turing Machine. Two-way innite tapes. Above, we assumed that the tape had an end and went innitely far to the right. This has a nice parallel with the Python list structure. When the transition rule attempted a next at the end of the tape, we simply appended a new, empty cell to the list. We can change this to allow the tape to stretch innitely far to right or the left. To allow the tape to go innitely far to the left, we have several choices: Continue to model the the tape as a dict. However, when we attempt to go to the previous cell from position 0, we need to insert a new, empty cell at position 0 and leave the position number alone. Model the tape as a dict and use dict.setdefault() to be sure that a new position has a proper empty cell before processing.
526
This is a little bit like creating a tape with a bunch of empty cells before and after the relevant data for the problem at hand. Two-dimensional tapes. Instead of using a one-dimensional innite tape, we could consider a twodimensional magnetic surface, which can be moved left, right, up and down. This changes our tape dramatically. Rather than a simple list, we should probably use a dict with a two-tuple as the key. Our initial state would be at position (0,0) on the surface. Our movement rules, would look something like this.
class Right( StateChange ): def __call__( self, machine ): x, y = machine.position machine.position = x+1, y class Left( StateChange ): def __call__( self, machine ): x, y = machine.position machine.position = x-1, y
The contents of the surface, then, would be available as surface.setdefault( position, None ). This would return any value that was on the surface, and create an empty cell if necessary. Multi-Step movement of the head. Above, we simplied the tape movements to be one-step. Either previous and next, or up, down, left and right. We can also change the denition to permit multiple-step movements instead of single-step movements. This would extend our basic state change class denitions to accept a parameter with the distance to move. In most cases, there isnt an easy way to make use of this extension. Most of the processing we can imagine requires stepping over each symbol one at time. We might, however, slightly simpliify some pattern recognition with multiple-position skips. Multiple tapes. An interesting extension is to add a second tape. In this case, we have two input symbols and two sets of transitions. Either tape can be moved and either tape can write a new symbol. You might use one tape for input and one tape for the results. In this case, and Add 1 operation would copy all the input symbols to the output tape, and then write one extra symbol to the output tape. Arbitrary Alphabet. As we noted in Other Applications, its sensible to allow a much larger alphabet than just two symbols. Simpler State Changes. We dened our machine to do both a tape movement and a write. We specied that the tape movement could be next, previous or none. We also said that there could be an optional write of a new symbol. This can be broken down into ner steps. We can, for example, separate the moving and the writing and say that a rule has the following attributes: A current symbol on the tape to match. A change: either move the tape, or write. Next State. This would create additional states in some machines because we would have to separate writing from moving the tape.
42.12. Alternative Specications
527
42.13 Exercise 4
Pick one of these alternate formalation for the Turing Machine. Two-way innite tapes. Two-dimensional tapes. Multi-Step movement of the head. Multiple tapes. Simpler State Changes. 1. Implement a new Turing Machine with this alternate formulation. 2. Revise your test cases from Exercise 2 to t your new machine formulation. 3. Create an additional test simply to exercise each feature of your new machine. There is an important question raised by the alternative formulations of the Turing Machine. Do these alternatives allow more computable numbers? It turns out we can nd a way to build any of these machines using any other machine. They are all provably identical.
528
CHAPTER
FORTYTHREE
MAH JONGG HANDS

The game of Mah Jongg is played with a deck of tiles with three suits with ranks from one to nine. There are four sets of these 27 tiles. Additionally there are four copies of the four winds and three dragons. This gives a deck of 136 tiles. The three suits are dots, bamboo and characters (wan, simplied or traditional). The ranks are the numbers one to nine. [One story is that the dots are chinese coins, the bamboo are stacks of 100 coins that and the wan represents 10,000 coins.] Since these tiles have ranks, they can form a variety of interesting combinations including matches and sequences. The winds and dragons are collectively called honors. There are four winds: East (), South (), West (), and North (). There are three dragons: White (), Red (), and Green (). These honors tiles dont have ranks, merely names. Since there are four of each, these tiles can only participate in matching; theres no sequence of winds combination. In some variations of the game there are also jokers, seasons and owers. Well leave these out of our analysis for the moment.
43.1 Tile Class Hierarchy

We can dene a parent class of Tile, and two subclasses: SuitTile and HonorTile. These have slightly dierent attributes. The SuitTile class has suit and rank information. The HonorTile class merely has a unique name. The superclass can dene a basic comparison function, __eq__(), that compares self.getName() to other.getName() to see if the other tile has the same name. For SuitTile, the name includes rank and suit. The SuitTile class, however, needs to dene methods for __lt__(), __le__(), __gt__() and __ge__() to compare rank and suit. The HonorTile class can simply return False for the various __lt__() and __gt__(). The implementation of __ge__() and __le__() must simply use __eq__(). class Tile() __init__(self, name) Build this tile from the given name. __str__(self ) Returns the Tile.getName() value. 529
__eq__(self, other ) This is simply self.getName() == other.getName(). __ne__(self, other ) This is simply self.getName() != other.getName(). getSuit(self ) Returns NotImplemented; each subclass must override this. getRank(self ) Returns NotImplemented; each subclass must override this. getName(self ) Returns the tiles name. class SuitTile(Tile) __init__(self, rank, suit ) Initializes a tile with rank and suit instead of name. getSuit(self ) Returns this tiles suit getRank(self ) Returns this tiles rank getName(self ) Returns this tiles full name, including suit and rank. __lt__(self, other ) Compares rank and suit. __le__(self, other ) Compares rank and suit. __gt__(self, other ) Compares rank and suit. __ge__(self, other ) Compares rank and suit. class HonorTile(Tile) getSuit(self ) Returns None. getRank(self ) Returns None. getName(self ) Returns this tiles full name. __lt__(self, other ) Returns False. __le__(self, other ) Returns the value of HonorTile.__eq__(). __gt__(self, other ) Returns False.
530
Chapter 43. Mah Jongg Hands
__ge__(self, other ) Returns the value of HonorTile.__eq__(). If we use the names "Bamboo", "Character" and "Dots", this makes the suits occur alphabetically in front of the honors without any further special processing. If, on the other hand, we want to use Unicode characters for the suits, we should add an additional sort key to the Tile that can be overridden by SuitTile and HonorsTile to force a particular sort order. Note that the two ranks of one and nine have special status among the suit tiles. These are called terminals; ranks two through eight are called simples. Currently, we dont have a need for this distinction. Build The Tile Class Hierarchy. First, build the tile class hierarchy. This includes the Tile, SuitTile, HonorTile classes. Write a short test that will be sure that the equality tests work correctly among tiles.
43.2 Wall Class

You hould also dene a Mah Jongg Wall class which holds the initial set of 136 tiles. We can create additional subclasses to add as many as a dozen more tiles to include jokers, owers and seasons. Shuing and Dealing Mah Jongg tiles are too large to manipulate like playing cards. They are shued by stirring them face down in the middle of the table. Then the tiles are stacked to make a wall with four sides. Each side is a row 17 tiles long and two tiles tall. Since this is a gambling game, there are fairly colorful procedures for establishing where people will sit, who will deal rst, which of the four sides will be dealt from and where along the side the dealing will begin. class Wall() __init__(self ) Create the set of 136 tiles. This is four copies of the following tiles: The twenty-seven combinations of each suit (dot, bamboo, and character) and each rank (one through nine). The seven honor tiles (east, south, west, north, red, white, green). __str__(self ) Display information about the wall. shuffle(self ) Shues the wall tiles. deal(self ) Return the next undealt tile. This will not enumerate through all of the tiles. Generally, six tiles will remain undealt. The wall is nearly identical with a deck of playing cards. See Advanced Class Denition Exercises for more guidance on this class design. Build The Wall Class. Design and implement the Wall. Write a short test that will be sure that it shues and deals tiles properly.
43.2. Wall Class
531
43.3 TileSet Class Hierarchy

A winning Mah Jongg hand generally as 14 tiles in ve scoring sets. Exactly one of these stes must be a pair. The remaining sets are generally four groups of three tiles. Under some circumstances, there can one ore more 4-of-a-kind sets, and the hand will also be larger. This can happen when you are holding three-of-a-kind and draw the fourth. Your hand must be extended by one tile to become 15 tiles in size. Clearly, this can only happen four times, leading to an upper limit of a four groups of four tiles. There are four varieties of set: Pair. Two matching tiles; either honor tiles with the same name, or suit tiles with the same rank and suit. Three of a kind. Three matching tiles. Sequence of three in a row of the same suit. Only suit tiles can participate in a sequence. The same suit is an essential feature. Four of a kind. The most common winning hands have 14 tiles: 4 sets of three and a pair. A Mah Jongg Hand object, then, is a list of Tiles. This class needs a method, mahjongg() that returns True is the hand is a winning hand. The evaluation is rather complex, because a tile can participate in a number of sets, and several alternative interpretations may be necessary to determine the appropriate use for a given tile. Consider a hand with 2, 2, 2, 3, 4 of bamboo. This is either a set of three 2s and a non-scoring 3 and 4, or it is a pair of 2s and a sequence of 2, 3, 4. The mahjongg() method, then must create ve TileSet objects, assigning individual Tiles to the TileSets until all of the Tiles nd a home. The hand is a winning hand if all sets are full, there are ve sets, and one set is a pair. Well cover the design of the TileSet classes in this section, and return to the design of the Hand class in the next section. The Set Class Hierachy. We can create a class hierarchy around the four varieties of TileSet: pairs, threes, fours and straights. A PairSet holds two of a kind: both of the tiles have the same name or the same suit and rank. A ThreeSet and FourSet are similar, but have a dierent expectation for being full. A SequenceSet holds three suit tiles of the same suit with adjacent ranks. Since we will sort the tiles into ascending order, this set will be built in ascending order, making the comparison rules slightly simpler. Well dene a TileSet superclass to hold a sequence of Tiles. We will be able to add new tiles to a TileSet, as well as check to see if a tile could possibly belong to a TileSet. Finally, we can check to see if the TileSet is full. The superclass, TileSet, is abstract and returns NotImplemented for the full() method. The sublasses will override this methods with specic rules appropriate to the kind of set. class TileSet() __init__(self ) Create a new, empty set of Tiles. __str__(self ) Representation of the contents of this set. canContain(self, aTile) The superclass canContain() method returns True if the list is empty; it returns False if the list
532
is full. Otherwise it compares the new tile against the last tile in the list to see if they are equal. Since most of the subclasses must match exactly, this rule will be used. The sequence subclass must override this to compare suit and rank correctly. add(self, aTile) The add() method appends the new tile to the internal list. A pop() function can remove the last tile appended to the list. full(self ) The superclass will return NotImplemented. Each subclass must override this. fallback(self, tileStack ) The superclass fallback() pushes all of the tiles from the TileSet back onto the given stack. The superclass version pushes the tiles and then returns None. Each subclass must override this to return a dierent fallback TileSet instance. pair(self ) The superclass pair() method returns False. The PairSet subclass must override this to return True. An important note about the fallback() method is that the stack that will be given as an argument in tileStack is part of the Hand, and is maintainted by doing tileStack.pop(0) to get the rst tile, and tileStack.insert( 0, aTile ) to push a tile back onto the front of the hand of tiles. Well need the following four subclasses of TileSet. class FourSet(TileSet ) Specializes TileSet for sets of four matching tiles. The full() method returns True when there are four elements in the list. The fallback() method pushes the sets tiles onto the given tileStack; it returns a new ThreeSet with the rst tile from the tileStack. class ThreeSet(TileSet ) Specializes TileSet for sets of three matching tiles. The full() method returns True when there are three elements in the list. The fallback() method pushes the sets tiles onto the given tileStack; it returns a new SequenceSet with the rst tile from the tileStack. class SequenceSet(TileSet ) Specializes TileSet for sets of three tiles of the same suit and ascending rank. The canContain() returns True for an empty list of tiles, False for a full list of tiles, otherwise it compares the suit and rank of the last tile in the list with the new tile to see if the suits match and the new tiles rank is one more than the last tile in the list. The full() method returns True when there are three elements in the list. The fallback() method pushes the sets tiles onto the given tileStack ; it returns a new PairSet with the rst tile from the tileStack. class PairSet(TileSet ) The full() method returns True when there are two elements in the list. The fallback() method is inherited from the superclass method in TileSet; this method returns None, since there is no fallback from a pair. This subclass also returns True for the pair() method. The idea of these class denitions is that the Hand can attempt to use a FourSet to collect a group of tiles. If this doesnt work out, we put the tiles back into the hand, and try a ThreeSet. If this doesnt work out,
43.3. TileSet Class Hierarchy
533
we put the tiles back and try a SequenceSet. The last resort is to try a PairSet. There is no fallback after a pair set, and the hand cannot be a winner.
43.4 Hand Class

A Mah Jongg Hand object, then, is a list of Tiles The mahjongg() creates an assignment of individual TileSets It checks these sets to see if all of them are full, if there are ve of them and if one of the ve is a pair. If so, it returns True because the hand is a winning hand. If we sort the tiles by name or suit, we can more eectively assign tiles to sets. The rst step in the mahjongg() method is to sort the tiles into order. Then the tiles can be broken into sets based on what matches between the tiles.
Hand Scoring
The mahjongg() function examines a hand to determine if the tiles can be assigned to ve scoring sets, one of which is a pair. 1. Sort Tiles. Sort the tiles by name (or suit) and by rank for suit tiles where the suit matches. We will treat the hand of tiles as a tile stack, popping and pushing tiles from position 0 using pop(0) and insert(0,tile). 2. Stack of TileSets. The candidate set denition is a stack of TileSet objects. Create an empty list to be used as the candidate stack. Create a new, empty FourSet and push this onto the top of the candidate stack. 3. Examine Tiles. Use the examine() function to examine the tiles of the hand, assigning tiles to TileSets in the candidate stack. When this operation is complete, we may have a candidate assignment that will contain a number of TileSets, some of which are full, and some are incomplete. We may also have an empty stack because we have run out of fallback TileSets. 4. While Not A Winner. While we have TileSets in the candidate stack, use the allFull() to see if all TileSets are full, there are ve sets, and there is exactly one pair. If we do not have ve full TileSets and a single pair, then we must fallback to another subclass of TileSet. (a) Retry. Use the retry() method to pop the last candidate TileSet, and use that TileSets fallback() to create a dierent TileSet for examination. Save this TileSet in n. (b) Any More Assignments? If the result of retry() is None, there are no more fallbacks; we can return False. (c) Examine Tiles. Append the TileSet, n, returned by retry() to the candidate stack. Use the examine() function to examine the tiles of the hand, assigning tiles to TileSets. When this operation is complete, we may have a candidate assignment that will contain a number of TileSets, some of which are full, and some are incomplete. 5. Winner? If we nish the loop normally, it means we have a candidate set assignment which has ve full sets, one of which is a pair. For some hands, there can be multiple winners; however, we wont continue the examination to locate additional winning assignments. The allFull() function checks three conditions: all TileSets are full, there are ve TileSets, and is one TileSet is a pair. The rst test, all TileSets are full, can be done with the built-in all() function, looking something like the following all( s.full() for s in sets ).
534
Examine All Tiles

The examine() method requires a non-empty stack of candidate TileSets, created by the mahjongg() method. It assigns all of the remaining tiles beginning with the top-most candidate TileSet. Initially, the entire hand is examined. After each retry, some number of tiles will have been pushed back into the hand for re-examination. 1. While More Tiles. If the tile stack in the Hand is empty, we are done, all tiles have been assigned to Sets. (a) Next Tile. Pop the next unexamined tile from the tile stack, assigning it to the variable t. (b) Topmost Set Full? If the topmost set on the set stack is full, push a new, empty FourSet onto the top of the set stack. (This is also a handy place to use a print statement to watch the progress of the evaluation.) (c) Topmost Set Can Contain? If the top-most TileSet can contain Tile t, add this tile to the set. Were done examining this tile. (d) Topmost Set Cant Contain. Put the tile t back into the stack of tiles to be examined. Use the retry() function to pop the TileSet from the stack, and fallback to another subclass of TileSet. (e) Another Retry? If the result of the retry() is None, weve run out of alternatives, return from this function. Otherwise, append the new TileSet created by retry() to the stack of candidate sets.
Retry a TileSet Assignment

The retry() method requires at least one TileSet in the assignments. This will pop that TileSet, pushing the tiles back into the hand. It will then use the popped TileSets fallback() method to get another avor of TileSet to try. 1. Pop. Pop the top-most TileSet from the TileSet stack, assign it to s. Call s fallback() method to get a new top-most TileSet, assign this to n. 2. Out Of Fallbacks? While the TileSet stack is not empty and n is None, there was no fallback. (a) Pop Another. Pop the top-most set from the set stack, assign it to s. Call s fallback() method to get a new top-most TileSet, assign this to n. 3. Done? If n is None and the set stack is empty, the hand is incomplete and we are out of fallback sets. Otherwise, append n to the stack of TileSets.
43.5 Some Test Cases

The following test case is typical.
Bamboo: 2, 2, 2, 3, 4, 5, 5, 5 Dots: 2, 2, 2 Green Dragon |times| 3
In this case, we will attempt to put the 2 Bamboo tiles into a TileSet of four. No other tile will ll this TileSet. After looking at the remaining tiles, well pop that incomplete TileSet, and put them into a TileSet of three. This will be full, so well move on.
43.5. Some Test Cases
535
The 3 Bamboo will be put into a set of four. No other tile can ll this set, so well pop it, and put the 3 Bamboo into a set of three. Again, no other tile can ll this set, so well pop that, and ll back to a Sequence. This set will be lled with the 4 and 5. The remaining two 5 Bamboo tiles will be put into a set of four (which wont be lled), a set of three (which wont be lled), a straight (which wont be lled) and nally a pair. The three 2 Dots tiles will be put into a set of le (which wont be lled) and a set of three. The fate awaits the three green dragon tiles. The nal set stack will have a three set, a straight, a pair, and two three sets. This meets the rules for ve full sets, one of which is a pair.
def testHand1(): t1= [ SuitTile( 2, "Bamboo" ), SuitTile( 2, "Bamboo" ), SuitTile( 2, "Bamboo" ), SuitTile( 3, "Bamboo" ), SuitTile( 4, "Bamboo" ), SuitTile( 5, "Bamboo" ), SuitTile( 5, "Bamboo" ), SuitTile( 5, "Bamboo" ), SuitTile( 2, "Dot" ), SuitTile( 2, "Dot" ), SuitTile( 2, "Dot" ), HonorsTile( "Green" ), HonorsTile( "Green" ), HonorsTile( "Green" ), ] h1= Hand( *t1 ) print h1.mahjongg()
More Complex. The following test case is a little more diicult.

Bamboo: 2, 2, 2, 2, 3, 4 Green Dragon |times| 3 Red Dragon |times| 3 North Wind |times| 2
The initial run of four 2 Bamboo tiles will be put into a set of four. The next 3 Bamboo and 4 Bamboo tiles will be put into a four set (which wont be lled), a three set and straight. None if this will be successful. We then pop the initial set of four 2 Bamboo tiles and replace that with a set of three. The 2 Bamboo will be tried in a set of four, and a set of three before it winds up in a sequence. This sequence will allow the 3 Bamboo and the 4 Bamboo to be added. The remaining honors will be tried against four sets and then three sets before the hand is found to be valid. Another Test. Heres a challenging test case with two groups of tiles that require multiple retries. Heres a summary of the hand.
Bamboo: 2, 2, 2, 3, 4, 5, 5, 5 Dots: 2, 2, 2, 2, 3, 4
Heres the test xture.

def testHand2(): t2= [ SuitTile( 2, "Bamboo" ), SuitTile( 2, "Bamboo" ), SuitTile( 2, "Bamboo" ), SuitTile( 3, "Bamboo" ), SuitTile( 4, "Bamboo" ), SuitTile( 5, "Bamboo" ), SuitTile( 5, "Bamboo" ), SuitTile( 5, "Bamboo" ), SuitTile( 2, "Dot" ), SuitTile( 2, "Dot" ), SuitTile( 2, "Dot" ), SuitTile( 2, "Dot" ), SuitTile( 3, "Dot" ), SuitTile( 4, "Dot" ), ]
536
h2= Hand( *t2 ) print h2.mahjongg()
Ideally, your overall unit test looks something like the following.
import unittest class TestHand_1(unittest.TestCase): def setUp( self ): Create the hand def testHand1_should_mahjongg( self ): self.assert_( h1.mahjongg() ) self.assertEqual( str(h1.sets[0]), self.assertEqual( str(h1.sets[1]), self.assertEqual( str(h1.sets[2]), self.assertEqual( str(h1.sets[3]), self.assertEqual( str(h1.sets[4]), class TestHand_2(unittest.TestCase): def setUp( self ): Create the hand def testHand2_should_mahjongg( self ): self.assert_( h2.mahjongg() ) ... check individual TileSets if __name__ == "__main__": unittest.main()
"ThreeSet['2B', '2B', '2B']" ) "SequenceSet['3B', '4B', '5B']" ) "PairSet['5B', '5B']" ) "ThreeSet['2D', '2D', '2D']" ) "ThreeSet['Green', 'Green', 'Green']" )
A set of nine interesting test cases can be built around the following set of tiles: 3 1s, 2, 3, 4, 5, 6, 7, 8, and 3 9s all of the same suit. Adding any number tile of the same suit to this set of 13 will create a winning hand. Develop a test function that iterates through the nine possible hands and prints the results.
43.6 Hand Scoring - Points

A hand has a point value, based on the mixture of TileSets. This point value is used to resolve the amount owed to the winner by the losers in the game. There is a subtlety to this evaluation that we have to gloss over, and that is the rules about for concealed and exposed or melded TileSets. For now, we will assume that all TileSets are concealed. Exposed and Melded Sets During the play of Mah Jongg, a player will drawn a tile from the Wall and evaluate their hand. If the hand is a winner, the game is over. Otherwise, the player will discard a tile. With some minor restrictions, a player can draw the last tile discarded by another player to make a complete set; the player must expose the set to do draw the discard. If the player completes a set with a tile drawn from the wall, they do not have to expose it, so the set is concealed. Concealed sets are worth more than exposed sets. There are number of variations in the rules for drawing discarded tiles and exposing sets. Well avoid much of this complexity, and focus on assigning point values to the sets using just the rules for concealed sets. We need to expand our denition of SuitTile. There are two dierent score values for SuitTiles: the terminals (one and nine) have one score, and the simples (two through eight) have a dierent score. This will lead to two subclasses of SuitTile: TerminalSuitTile and SimpleSuitTile.
43.6. Hand Scoring - Points
537
A winning hand has a base value of 20 points plus points assigned for each of the four scoring sets and the pair. TileSet SequenceSet ThreeSet FourSet Simples 0 4 16 Terminals or Honors 0 8 32
The PairSet is typically worth zero points. However, the following kinds of pairs can add points to a hand. A pair of dragons is worth 2 points. A pair of winds associated with your seat at the table is worth 2 points. A full game consists of four rounds. Each round has a prevailing wind. Within each round, each of the players will be the dealer. A pair of the rounds prevailing winds is worth 2 points. A double wind pair occurs when your seats wind is also the prevailing wind. A pair of this wind is worth 4 points. There are a few more ways to add points, all related to the mechanics of play, not to the hand itself. Update the Tile Class Hierarchy. TerminalSuitTile and SimpleSuitTile. You will need to add two new subclass of SuitTile:
You will want to upgrade Wall to correctly generate the various HonorsTile, TerminalSuitTile and SimpleSuitTile instances. You may also want to create a Generator for tiles. A function similar to the following can make programs somewhat easier to read.
def tile( *args ): """tile(name) -> HonorsTile tile( rank, suit ) -> SuitTile """ if len(args) == 1: return HonorsTile( *args ) elif args[0] in ( 1, 9 ): return TerminalSuitTile( *args ) else: return SimpleSuitTile( *args )
Update the TileSet Class Hierarchy. You will need to add at least one new method to the TileSet classes. class TileSet() points(self, prevailingWind, myWind ) Examine the rst Tile of the TileSet to see if it is simple() or not, and return the proper number of points. The two wind parameters arent used for most TileSet subclasses. In the case of PairSet, however, the rst Tile must be checked against two rules. If prevailingWind is the same as myWind and the same as the tiles name, this is worth 4 points. If the tiles lucky() method is True (a dragon, or one of the two winds), then the value is 2 points. Update the Hand Class. Youll want to add at least one new method to the Hand class.
538
class Hand() points(self, prevailingWind, myWind ) Compute the total number of points for a hand. pointReport(self, prevailingWind, myWind ) Print a small scorecard for the hand, showing each set and the points awarded. You will want to revise your unit tests, also, to reect these changes. Youll also need to add additional unit tests to check the number of points in each hand. Test Cases. For the rst test cases in the previous Some Test Cases, here are the scores. Set Winning ThreeSet[2B, 2B, 2B] StraightSet[3B, 4B, 5B] PairSet[5B, 5B] ThreeSet[2D, 2D, 2D] ThreeSet[Green, Green, Green] Points Points 20 4 0 0 4 8 36
For the second test cases in Some Test Cases, here are the scores. The assumption here is that were not sitting at North, and were not playing the nal four hands (where the prevailing wind is North.) Set Winning ThreeSet[2B, 2B, 2B] StraightSet[2B, 3B, 4B] ThreeSet[Green, Green, Green] ThreeSet[Red, Red, Red] PairSet[N, N] Points Points 20 4 0 8 8 0 36
For the third test cases in Some Test Cases, here are the scores. Set Winning ThreeSet[2B, 2B, 2B] StraightSet[3B, 4B, 5B] PairSet[5B, 5B] ThreeSet[2D, 2D, 2D] StraightSet[2D, 3D, 4D] Points Points 20 4 0 0 4 0 28
Be sure to add a test case with lucky tiles (dragons or winds) as the pair.
43.7 Hand Scoring - Doubles

The point value for a hand can be doubled a number of times for a variety of rare achievements. Most of these rules of these are additional properties of TileSets that are summarized by the Hand. Each non-pair TileSet that contains lucky tiles is worth 1 double (2 ). In the case of the players wind being the prevailing wind, a TileSet of this wind is worth 2 doubles (4 ) A hand of four ThreeSet or FourSet (i.e., no SequenceSet) merits a double. Depending on how the hand was played and how many of these triples were concealed or melded, the hand can have a second double, or
43.7. Hand Scoring - Doubles
539
possibly even pay the house limit, something well look into in Limit Hands. For now, we are ignoring these mechanics of play issues, and will simply double the score if there are no SequenceSets. A hand that has three consecutive SequenceSets in the same suit is doubled. There are many rule variations on this theme, including same-rank sequences from all three suits, same-rank ThreeSet s or FourSets from all three suits. Well focus on the three-consecutive rule for now. If the hand is worth exactly 20 points (it is all SequenceSets and an unlucky PairSet), then it merits one double for being all non-scoring sets. There are six dierent consistency tests. These are exclusive and at most one of these conditions will be true. 1. If the hand is all terminals and honors (no simples), it is doubled. 2. If each set in the hand has one terminal or an honor in it, the hand is doubled. A hand could have four SequenceSet s, each of which begins with one or ends with nine, and a pair of honors or terminals to qualify for this double. 3. If the hand is all simples (no terminals or honors), it is doubled. 4. If all of the SuitTiles are of the same suit, and all other tiles are HonorsTiles, this is doubled. 5. If all of the SuitTiles are of the same suit, and there are no HonorsTiles, this is doubled four times (16?). 6. If the hand contains TileSets of all three dragons, and one of those sets is a PairSet, this is called the Little Three Dragons, and the hands points are doubled. There are a few more ways to add doubles, all related to the mechanics of play, not to the hand itself. Update TileSet Class Hierarchy. Youll need to add the following functions to the TileSet classes. class TileSet() lucky(self, prevailingWind, myWind ) For the ThreeSet or FourSet subclasses, this returns True for a complete set with lucky tiles (dragons, the prevailing wind or the players wind.) Other subclasses of TileSet return False. triplet(self ) Returns True for a ThreeSet or FourSet. Other subclasses of TileSet return False. This is used to see if a hand is all threes or fours, which merits a double. The name triplet isnt literally true; a literally true function name would be cumbersome. sequenceSuit(self ) Returns the suit of a SequenceSet. For this class only, it should be based on the suit() method described below. Other subclasses of of TileSet return None. sequenceRank(self ) Returns the lowest rank of a SequenceSet. Other subclasses of of TileSet return None. allSimple(self ) Returns True if the TileSet contains only simple tiles. noSimple(self ) Returns True if the TileSet contains no simple tiles. This is an all terminals and honors TileSet.
540
This is not the opposite of allSimple(). A SequenceSet could have a mixture of Terminal and Simple tiles, so there are three cases: allSimple, noSimple and a mixed bag. oneTermHonor(self ) Returns True if the TileSet contains one terminal or honor. Since we only have a simple() function, this means there is one non-simple Tile in the TileSet. suit(self ) If all tiles have the same value for Tile.getSuit(), return that value. If there is a mixture of suits, or suits of None, return None. bigDragon(self ) For the ThreeSet class or FourSet class, return True if all tiles are Dragons. Other subclasses of TileSet return False. littleDragon(self ) For the PairSet class, return True if all tiles are Dragons. Other subclasses of TileSet return False. Update Hand Class. Youll need to add the following functions to the Hand classes. class Hand() luckySets(self, prevailingWind, myWind ) Returns the number of lucky sets. This function also checks for the double wind conditions where prevailWind is the same as myWind and one of the TileSets has this condition and throws an additional doubling in for this. groups(self ) Returns 1 if all TileSets have the triple() property True. sequences(self ) Returns 1 if three of the TileSets have the same value for sequenceSuit(), and the values for sequenceRank() are 1, 4, and 7. noPoints(self ) Returns 1 all of the TileSets are worth zero points. consistency(self ) Returns 1 or 4 after checking for the following conditions: If allSimple() is true for all TileSets, return 1. If noSimple() is true for all TileSets, return 1. If oneTermHonor() is true for all TileSets, return 1. If every TileSet has the same value for suit() and there is no TileSet where suit() is None, return 4. If every TileSet has the same value for suit() or the value for suit() is None, return 1 . If there are two TileSets for which bigDragon() is true, and one TileSet for which littleDragon() is true, return 1. The sum of the double functions is the total number of doubles for the hand. This is given by code something like the following.
doubles = hand.luckySets( wind_prevail, wind_me ) + hand.groups() + hand.sequences() + hand.noPoints() + hand.consi final_score = 2**doubles * base_points
43.7. Hand Scoring - Doubles
541
An amazing hand of all one suit with three consecutive sequences leads to 5 doubles, 32 the base number of points. class Hand() doubleReport(self ) Prints a small scorecard for the hand, showing each double that was awarded. You can then write a scoreCard() which produces the pointReport(), the doubleReport() and the nal score of 2d p, where d is the number of doubles and p is the base number of points. The total score is often rounded to the nearest 10, as well as limited to 500 or less to produce a nal score. This nal score is used to settle up the payments at the end of the game. Losers Pay There are a number of variations on the payments at the end of the game. The simplest version has all losers paying an equal amount to the winner. Generally, if the dealer wins, the payments to the dealer are doubled, and when the dealer loses, the payment to the winner is doubled. There are some popular scoring variations that penalize the player whos discard allowed another player to win.
43.8 Limit Hands

At the end of a hand of Mah Jongg, the winner is paid based on the nal score of the hand. Generally, the nal score is limited to 500 points. There are, however, some extraordinary hands which simply score this limit amount. These conditions are checked rst; if none of these are true, then the normal hand scoring is performed. The Big Three Dragons hand has three TileSets for which the bigDragon() function is true. The Little Four Winds hand has three ThreeSets or FourSetss for which the wind() function is true and a PairSet for which wind() is true. The Big Four Winds hand has four ThreeSets or FourSets s for which the wind() function is true. The All Honors hand has all TileSets composed of HonorsTiles; these will all have either wind() or dragon() true. The All Terminals hand has all TileSets composed of TerminalSuitTiles. An additional hand that pays the limit also breaks many of the rules for a winning hand. This is the Thirteen Orphans hand, which is one each of the various terminals and honors: three dragons, four winds, three ones, three nines and any other of the thirteen terminal and honor tiles. This requires a special-case test in Hand that short-cuts all of the evaluation algorithm. An interesting limit hand is the Nine Gates hand, which is 3?1s, 2, 3, 4, 5, 6, 7, 8, and 3?9s all of the same suit. Any other tile of this suit will create a winning hand that pays the limit. Just considering the hand outside the mechanics of play, it would get four doubles because it is all one suit, plus the possibility of an additional double for consecutive sequences. The Nine Gates hand is only a limit hand if the player draws it as a completely concealed hand. There are a few other limit hands, including all concealed triplets, or being dealt a winning hand. These, however, depend on the mechanics of play, not the hand itself. Update Set Class Hierarchy. Youll want to add wind() and dragon() methods to the TileSet hierarchy. These return True if all Tiles in the TileSet are a wind or a dragon, respectively.
542
Update Hand Class. You can add six additional methods to Hand to check for each of these limit hands. The nal step is to update the finalScore() to check for limit hands prior to computing points and doubles.
43.8. Limit Hands
543
544
CHAPTER
FORTYFOUR
CHESS GAME NOTATION

See Chessboard Locations for some additional background. Chess is played on an 8x8 board. One player has white pieces, one has black pieces. Each players pieces include eight pawns, two rooks, two knights, two bishops, a king and a queen. The various pieces have dierent rules for movement. Players move alternately until one players king is in a position from which it cannot escape but must be taken, or there is a draw. There are a number of rules that lead to a draw, all beyond the scope of this problem. White moves rst. A game is recorded as a log of the numbered moves of pieces, rst white then black. The Portable Game Notation (PGN) standard includes additional descriptive information about the players and venue. There are two notations for logging a chess game. The newer, algebraic notation and the older descriptive notation. We will write a program that will process a log in either notation and play out the game, showing the chess board after each of blacks moves. It can also be extended to convert logs to completely standard PGN notation.
44.1 Algebraic Notation

Well present the formal denition of algebraic notation including Algebraic Notation (LAN) and Short Algebraic Notation (SAN). Well follow this with a summary and some examples. This section will end with some Algorithm R, used to resolve which of the available pieces could perform a legal move. Denition. Algebraic notations uses letters a-h for the les (columns across the board) from whites left to right, and numbers for the ranks (rows of the board) from white (1) to black (8). Piece symbols in the log are as follows: Piece Pawn Rook Knight Bishop Queen King Symbol (omitted) R N B Q K Move Summary 1 or 2 spaces forward anywhere in the same rank or same le 2 in one direction and 1 in the other (L-shaped) diagonally, any distance horizontal, vertical or diagonal, any distance 1 space in any direction
The game begins in the following starting position.
545
White a1 b1 c1 d1 e1 f1 g1 h1 a2-h2
Black a8 b8 c8 d8 e8 f8 g8 h8 a7-h7
Piece rook knight bishop queen king bishop knight rook pawns
There are two forms of algebraic notation: short (or standard) algebraic notation (SAN), where only the destination is shown, and long algebraic notation (LAN) that shows the piece, the starting location and the destination. Long Notation. The basic syntax for LAN is as follows:
[ P ] f r m f r [ n ]
P The piece name (omitted for pawns). f The le (a-h) moving from and to. r The rank (1-8) moving from and to. m The move (- or x). n any notes about the move (+, #, !, !!, ?, ??). The notes may include = and a piece letter when a pawn is promoted. Short Notation. Short notation omits the starting le and rank unless they are essential for disambiguating a move. The basic syntax is as follows:
[ P ] [ m ] [ d ] f r [ n ]
P The piece name (omitted for pawns). m The move is only specied for a capture (x). d The discriminator: either a le (preferred) or a rank or both used to choose which piece moved when there are multiple possibilities. f The le (a-h) moving to. r The rank (1-8) moving to. n any notes about the move (+, #, !, !!, ?, ??). The notes may include = and a piece letter when a pawn is promoted. Additional Syntax. In both notations, the castle moves are written O-O or O-O-O (capital letters, not numbers). Similarly, the end of a game is often followed with a 1-0 (white wins), 0-1 (black wins), 1/2-1/2 (a draw), or * for a game that is unknown or abandoned. Each turn is preceeded by a turn number. Typically the number and a . preceeds the white move. Sometimes (because of commentary), the number and ... preceed the black move. The PGN standard for notes is $ and a number, common numbers are as follows: $1 good move (traditional ! ) $2 poor move (traditional ? ) $3 very good move (traditional !! )
546
Chapter 44. Chess Game Notation
$4 very poor move (traditional ?? ) Legal Moves. Each piece has a legal move. This is a critical part of processing abbreviated notation where the log gives the name of the piece and where it wound up. The legal moves to determine which of two (or eight) pieces could have made the requested move. This requires a simple search of pieces to see which could have made the move. Pawn. A pawn moves forward only, in its same le. For white, the rank number must increase by 1 or 2. It can increase by 2 when it is in its starting position; rank 2. For black, the rank number must decrease by 1 or 2. It can decrease by 2 when the pawn is in its starting position of rank 7. A pawn captures on the diagonal: it will move into an adjacent le and forward one rank, replacing the piece that was there. There is one origin for all but the opening pawn moves: one rank back on the le in which the pawn ended its move. There is one origin for an opening pawn move that lands in rank 4 or 5: two ranks back on the le where the pawn ended its move. There are two possible origins for any pawn capture (one position on a le adjacent to the one in which the pawn ended its move). Rook. A rook moves in ranks or les only, with no limit on distance. There are 16 possible origins for any rook move, including the entire rank or the entire le in which the rook ended its move. Knight. A knight makes an L-shaped move. It moves two spaces in one direction, turns 90-degrees and moves one more space. From g1, a knight can move to either f3 or h3. The rank changes by 2 and the le by 1; or the le changes by 2 and the rank changes by 1. There are 8 places a knight could start from relative to its nal location. Bishop. A bishop moves diagonally. The amount of change in the rank must be the same as the change in the le. There are 16 places a bishop can start from on the two diagonals that intersect the nal location. Queen. The queens move combines bishop and rook: any number of spaces diagonally, horizontally or vertically. There are 16 places on the diagonals, plus 16 more places on the horizontals and verticals where the queen could originate. Pawns that reach the opposite side of the board are often promoted to queens, meaning there can be multiple queens late in the game. King. The king is unique, there is only one. The king can only move one space hoizontally, vertically or diagonally. King and Rook. The king and a rook can also engage in a move called castling: both pieces move. When the king and the closest rook (the one in le h) castle, this is king side and annoted O-O . The king moves from le e to le g and the rook from el h to le f. When the king and the queens side rook (the one in le a) castle, this is annotated O-O-O . The king moves from le e to le c and the rook move from le a to le d. Castling can only be accomplished if (a) neither piece has moved and (b) space between them is unoccupied by other pieces. Part a of this rule requires that the game remember when a king or rook moves, and eliminate that side from available castling moves. Moving the rook in le a eliminates queen-side castling; moving the rook in le h eliminates king-side castling. Moving the king eliminates all castling. Summary and Examples. Heres a suymmary of the algebraic notation symbols used for annotating chess games. This is followed by some examples.
44.1. Algebraic Notation
547
Symbol a-h 1-8 R, N, B, Q, K x + # ++ = 0-0 0-0-0 e.p.
Meaning le from whites left to right rank from white to black rook, knight, bishop, king, queen move (non-SAN) capture; the piece that was at this location is removed check, a note that the king is threatened checkmate, a note that this is the reason for the end of the game checkmate (non-SAN), a note that this is the reason for the end of the game promoted to; a pawn arriving at the opposite side of the board is promoted to another piece, often a queen. castle on the kings side; swap the king and the rook in positions e1 and h1 (if neither has moved before this point in the game) castle on the queens side; swap the king and the rook in positions e1 and a1 (if neither has moved before this point in the game) en passant capture (non-SAN), a note that a pawn was taken by another pawn passing it. When a pawns rst move is a two space move (from 7 to 5 for black or 2 to 4 for white) it can be captured my moving behind it to the 6th rank (white taking black) or 3rd rank (black taking white). en passant capture (non-SAN), see e.p. editorial comments (non-SAN), weak, very weak, strong, very strong
ep ?, ??, !, !!
Heres parts of an example game in abbreviated notation: 1. e4 e5. White pawn moves to e4 (search e3 and e2 for the pawn that could do this); black pawn moves to e5 (search e6 and e7 for a pawn that could do this) 2. Nf3 d6. White knight moves to f3 (search 8 positions: g1, h2, h4, g5, e5, d4, d2, e1 and g1 for the knight that could do this); black pawn moves to d6 (search d7 and d8 for the pawn). 3. d4 Bg4. White pawn moves from d4 (search d3 and d2 for the pawn); black bishop moves to g4 (search the four diagonals: f5, e6, d7, c8; h5; h3; f3, e3, and d3 for the bishop that could do this). 4. dxe5 Bxf3. A white pawn in d takes a piece at e5, the pawn must have been at d4, the black pawn at e5 is removed; a black bishop takes a piece at f3 (search the four radiating diagonals from f3: e4, d5, c6, b7, a8; g4, h5; g2, h1; e2, d1). 5. Qxf3 dxe5. The white queen takes the piece at f3; the black pawn in d4 takes the piece in e5. Heres a typical game in abbreviated notation:
1.c4 e6 2.Nf3 d5 3.d4 Nf6 4.Nc3 Be7 5.Bg5 0-0 6.e3 h6 7.Bh4 b6 8.cxd5 Nxd5 9.Bxe7 Qxe7 10.Nxd5 exd5 11.Rc1 Be6 12.Qa4 c5 13.Qa3 Rc8 14.Bb5 a6 15.dxc5 bxc5 16.0-0 Ra7 17.Be2 Nd7 18.Nd4 Qf8 19.Nxe6 fxe6 20.e4 d4 21.f4 Qe7 22.e5 Rb8 23.Bc4 Kh8 24.Qh3 Nf8 25.b3 a5 26.f5 exf5 27.Rxf5 Nh7 28.Rcf1 Qd8 29.Qg3 Re7 30.h4 Rbb7 31.e6 Rbc7 32.Qe5 Qe8 33.a4 Qd8 34.R1f2 Qe8 35.R2f3 Qd8 36.Bd3 Qe8 37.Qe4 Nf6 38.Rxf6 gxf6 39.Rxf6 Kg8 40.Bc4 Kh8 41.Qf4 1-0
Heres a small game in full notation:

1.f2-f4 e7-e5 2.f4xe5 d7-d6 3.e5xd6 Bf8xd6 4.g2-g3 Qd8-g5 5.Ng1-f3 Qg5xg3+ 6.h2xg3 Bd6xg3#
548
44.2 Algorithms for Resolving Moves

Algebraic notation is terse because it is focused on a human student of chess. It contains just enough information for a person to follow the game. Each individual move cannot be interpreted as a stand-alone (or context-free statement). Each moves description only makes sense in the context of the game state established by all the moves that came before it. Therefore, in order to interpret a log of chess moves, we also need to maintain the state of the chess board. Given that we have a model of the chess board, Algorithm G can locate the pieces and execute the moves an entire game.
Algorithm G
(Resolve chess moves in SAN notation, playing out the entire game.) We are given a block of text with a sequence of chess turns. Assume that line breaks have been removed and the game ending marker has been removed from the block of text. 1. Parse turn into moves. Locate move number, white move and black move. Lines that dont have this form are some kind of commentary and can be ignored. 2. Parse each move. For each move, parse the move. Identify the piece (R, B, N, Q, K, pawn if none of these). Identify the optional le (a - h) or rank (1 - 8) for the source. Identifty the optional x for a capture. Identify the destination le (a - h) and rank (1 - 8 ). Identify any other material like + or # for checks, = x for promotions, or !, !! , ? , ?? for editorial comments. 3. Castling? If the move is simply 0-0 or O-O-O, move both the king (in le e) and the appropriate rook. For 0-0 it is the rook in le h moves to f, the king moves from e to g. For O-O-O it is the rook in le a moves to d, the king moves from e to c. 4. Fully specied from location? If a two-character from-position is given, this is the starting location. Execute the move with step 7. 5. Partially specied from location? If a one-character from-position is given (a-h or 1-8), restrict the search for the source to this rank for le. Use the piece-specic version of Algorithm S with rank or le restrictions for the search. After the starting location is found, execute the move with step 7. 6. Omitted from location? Search all possible origins for the from-position for this piece. Each piece has a unique search pattern based on the pieces movement rules. Use the piece-specic version of Algorithm S with no restrictions for the search. After the starting location is found, execute the move with step 7. 7. Execute move. Move the piece, updating the state of the board, removing captured pieces. Periodically during game processing, print the board position. The board, by the way, is always oriented so that a1 is a dark square in the lower-left. 8. Next move. Loop to step 2, processing the black move after the white move in this turn. If the black move is omitted or is one of the ending strings, skip the black move. 9. Next turn. Loop to step 1, processing the next turn. If the turn number is omitted or is one of the ending strings, this is the end of the game. We have to design six dierent kinds of searches for possible starting pieces. These searches include pawns, rooks, knights, bishops queens and the king. Well provide formal algorithms for pawns and rooks, and informal specications for the other pieces.
44.2. Algorithms for Resolving Moves
549
Algorithm P
(Search for possible pawn starting locations.) Given a destination location, piece color, and optional restrictions on starting rank or starting le. 1. First move. If the destination is rank 4 (white) or rank 5 (black), search two spaces back for the rst move of a pawn (from rank 7 or rank 2). If moving this piece will not put the king into check, this is the stating location. 2. Previous space. Search the previous space (rank -1 for white, rank +1 for black) for a move. If moving this piece will not put the same color king into check, this is the stating location. 3. Capture. Search the adjacent les one space back for a pawn which performed a capture. If moving this piece will not put the same color king into check, this is the stating location. 4. Error. If no source can be found, the game notation is incorrect.
Algorithm R
(Search for possible rook starting locations.) Given a destination location, piece color, and optional restrictions on starting rank or starting le. 1. To Right. 1. Initialize. Set r +1. 2. Loop. Target position has le oset by r from destination. 3. On Board. If this is o the board, or a non-rook was found, break from this loop. If moving this piece will not put the king into check, return this position as the stating location. 4. Loop. Increment r. Continue this loop. 1. To Left. (a) Initialize. Set r 1. (b) Loop. Target position has le oset by r from destination. (c) On Board. If this is o the board, or a non-rook was found, break from this loop. If moving this piece will not put the king into check, return this position as the stating location. (d) Loop. Decrement r. Continue this loop. 2. Toward Black. (a) Initialize. Set r +1. (b) Loop. Target position has rank oset by r from destination. (c) On Board. If this is o the board, or a non-rook was found, break from this loop. If moving this piece will not put the king into check, return this position as the stating location. (d) Loop. Increment r. Continue this loop. 3. Toward White. (a) Initialize. Set r 1. (b) Loop. Target position has rank oset by r from destination.
550
(c) On Board. If this is o the board, or a non-rook was found, break from this loop. If moving this piece will not put the king into check, return this position as the stating location. (d) Loop. Decrement r. Continue this loop. 4. Error. If no source can be found, the game notation is incorrect.
Algorithm N
(Search for possible knight starting locations.) Given a destination location, piece color, and optional restrictions on starting rank or starting le. There are as many as eight possible starting positions for a knights move. 1. Adjacent File. Four of the starting positions have a le oset of +1 or -1 and rank osets of +2 or -2. If the position is on the board, the piece is a knight, and moving this piece will not put the king into check, then this is the origin for this move. 2. Adjacent Rank. Four of the starting positions have le osets of +2 or -2 and rank osets of +1 or -1. If the position is on the board, the piece is a knight, and moving this piece will not put the king into check, then this is the origin for this move.
Algorithm B
(Search for possible bishop starting locations.) Given a destination location, piece color, and optional restrictions on starting rank or starting le. Search radially out the diagonals until edge of board or an intervening piece or the correct bishop was found. This algorithm is similar to the rook algorithm, except the osets apply to both rank and le. Applying +1 to both rank and le moves north-east; applying -1 to both rank and le moves south-west. Applying +1 to rank and -1 to le moves south east; applying -1 to rank and +1 to le moves north west. When an opposing piece is found, the search along that diagonal ends. If the position is on the board, the piece is a bishop, and moving this piece will not put the king into check, then this is the origin for this move.
Algorithm Q
Search for possible queen starting locations.) Given a destination location, piece color, and optional restrictions on starting rank or starting le. Search radially out the ranks, les and diagonals until edge of board or an intervening piece or the correct queen was found. This combines the bishop and rook algorithms. When an opposing piece is found, the search along that rank, le or diagonal ends. If the position is on the board, the piece is a queen, and moving this piece will not put the king into check, then this is the origin for this move.
44.2. Algorithms for Resolving Moves
551
Algorithm K
Search for possible king starting locations.) Given a destination location, piece color, and optional restrictions on starting rank or starting le. Search the 8 adjacent locations. These are all combinations of -1, 0, +1 for rank oset and -1, 0, +1 for le oset. Omit checking the combination with a 0 oset to rank and a 0 oset to le. If the position is on the board, the piece is the king, this is the starting position.
44.3 Descriptive Notation

Descriptive notation uses a dierent scheme for identifying locations on the board. Each le is named for the pieces at its top and bottom ends as the game begins. The board is divided into Kings side and Queens side. The les are KR, KKt, KB, K, Q, QB, QKt, QR. These are known as a, b, c, d, e, f, g, h in Algebraic notation. The ranks are counted from the players point of view, from their back row to the far row. Consequently, whites row 1 is blacks row 8. Whites Q1 is Blacks Q8; Blacks KB5 is Whites KB4. The notation has the following format:
piece [ (file rank) ] move [file rank] [note]
The symbol for the piece to be moved is one of p, B, N, R, Q, K. If capturing, the move is x, followed by the symbol of the captured piece. Examples: pxp, NxQ. A search is required to determine which piece can be taken. If not capturing, the move is -, followed by le rank to name the square moved to, from the perspective of whoever is moving, black or white If 2 pieces are both be described by a move or capture, write the location of the intended piece in parentheses. Examples: p(Q4)xR means pawn at queens rook four takes Rook, N(KB3)-K5 means knight at KB3 moves to K5 Special moves include kings side castling, designated O-O, Queens side castling, designated O-O-O. Notes. If a pawn captures en passant or in passing it is designated ep in the note. A move resulting in a check of the king is followed by ch in the note. ! means good move; ? means bad move in the note . If the pawn in front of the king is moved forward two spaces, it is described as P-K4. If the pawn in front of the queenside knight is moved forward one space, it is P-QN3. If a knight at K5 captures a rook on Q7, it would be NxR or if clarication is needed, NxR(Q7) or N(K5)xR.
44.4 Game State

In order to process a log, a model of the chess boards current position is essential. In addition to the basic 64 squares containing the pieces, several additional facts are necessary to capture the game state. The current state of a chess game is a 6-tuple of the following items: Piece Placement. An 8-tuple shows pieces in each rank from 8 down to 1. Pieces are shown as single letters, upper case for white (PRNBQK), lower case for black (prnbqk). Pieces are coded P for pawn, R for rook, N for knight, B for bishop, Q for queen and K for king. Empty spaces are shown as the number of contiguous spaces.
552
The entire rank can be coded as a 1-8 character string. 8 means no pieces in this rank. 4p3 means four empty spaces (a-d), a black pawn in le e, and 3 empty spaces (f-h). The entire 8-tuple of strings can be joined to make a string delimited by / s. rnbqkbnr/pp1ppppp/8/2p5/4P3/5N2/PPPP1PPP/RNBQKB1R. For example
Active Color. A value (w or b) showing whos turn to move is next. This color is active because this player is contemplating their move. The starting position, for instance, has an active color of w, because white moves rst. Castling Availability. A string with 1 to 4 characters showing which castling moves are still allowed. If none are allowed, a - is shown. White codes are capital letters, black are lower case. When king side castling is possible, a K (white) or k (black) is included. When queen side caslting is possible a Q (white) or q (black) is included. At the start of the game, there are four characters: KQkq. As the game progress and kings castle or are moved for other reason, or rooks are moved, the string reduces in size to -. En Passant Target. Either - or a square in rank 6 or 3. When a pawns rst move advances two spaces (from 7 to 5 for black or 2 to 4 for white), the skipped-over space is named here on the next turn only. If an opposing pawn moves to this space, an En Passant capture has occured. If no En Passant vulnerability, a - is given. Half Move Count. How many 1/2 moves since a pawn was advanced or a piece captures. This is zero after a pawn moves or a piece is captured. This is incremented after each 1/2 move (white or black) where no pawn moves and no piece is captured. When this reaches 50, the game is technically a draw. Turn. This is the turn count, it increments from 1, by 1, after blacks move.
44.5 PGN Processing Specications

There are three parts to a PGN processing program: the parsing of a PGN input le, the resolution of moves, and maintenance of the game state. Each can be dealt with separately with suitable interfaces. Each of these modules can be built and tested in isolation. First, some preliminaries. In order to resolve moves, the game state must be kept. This is a dictionary of locations and pieces, plus the ve other items of information that characterize the game state: active color (w or b), castling availability, en passant target, half-move draw count and turn number. The board has an interface that accepts a move and executes that move, updating the various elements of board state. Moves can use the Command design pattern to separate king-side castle, queen-side castle, moves, captures and promotions. The Board object will require a fully-specied move with source location and destination location. The source location is produced by the source resolution algorithm. A well-dened Board object could be used either for a single-player game (against the computer) or as part of a chess game server for two-player games. Second, the hard part. Resolution of short notation moves requires several algorithms to locate the piece that made the move. Based on input in algebraic notation, a move can be transformed from a string into a 7-tuple of color, piece, fromHint, moveType, toPosition, checkIndicator and promotionIndicator. The color is either w or b. The piece is omitted for pawns, or one of RNBQK for the other pieces. The fromHint is the from position, either a le and rank or a le alone or a rank alone. The various search algorithms are required to resolve the starting piece and location from an incomplete hint. The moveType is either omitted for a simple move or x for a capturing move. The toPosition is the rank and le at which the piece arrives.
44.5. PGN Processing Specications
553
The checkIndicator is either nothing, + or #. The promotionIndicator is either nothing or a new piece name from QBRK. This information is used by Algorithm G to resolve the full starting position information for the move, and then execute the move, updating the board position. Finally, input parsing and reporting. A PGN le contains a series of games. Each game begins with identication tags of the form [Label "value"]. The labels include names like Event, Site, Date, Round, White, Black, Result. Others labels may be present. After the identication tags is a blank line followed by the text of the moves, called the movetext. The movetext is supposed to be SAN (short notation), but some les are LAN (long notation). The moves should end with the result ( 1-0, 0-1, *, or 1/2-1/2), followed by 1 or more blank lines. Heres a sample game from a recent tournament.
[Event "25th European Club Cup 2009"] [Date "2009.10.04"] [Round "1"] [Board "1"] [White "Aronian, Levon"] [Black "Docx, Stefan"] [Result "1-0"] 1.d4 d6 2.Nf3 g6 3.e4 Bg7 4.Bc4 Nf6 5.Qe2 O-O 6.O-O Bg4 7.Rd1 Nc6 8.h3 Bxf3 9.Qxf3 Nd7 10.c3 e5 11.Be3 a6 12.Na3 exd4 13.cxd4 Qh4 14.Rac1 Nf6 15.Bd3 Rfe8 16.Rc4 Nd7 17.Bb1 Rad8 18.b4 Nb6 19.Rcc1 d5 20.e5 Qe7 21.Nc2 Nc4 22.a3 b5 23.Ba2 Nb8 24.Re1 c6 25.Qg3 Qf8 26.h4 Nd7 27.h5 Ra8 28.Rcd1 a5 29.Bc1 Qe7 30.Bb1 Qe6 31.hxg6 fxg6 32.Rd3 axb4 33.Nxb4 c5 34.Nc2 Qc6 35.f4 Qb6 36.Qf3 Rad8 37.Ne3 Nxe3 38.dxc5 Nxc5 39.Bxe3 Qa5 40.Bd2 Qb6 41.Qf2 Re6 42.Rh3
Design Considerations. In order to handle various forms for the movetext, there have to be two move parsing classes with identical interfaces. These polymorphic classes implement long-notation and shortnotation parsing. In the event that a short-notation parser object fails, then the long-notation parser object can be used instead. If both fail, the le is invalid. A PGN processing program should be able to read in a le of games, execute the moves, print logs in various forms (SAN, LAN and Descriptive), print board positions in various forms. The program should also be able to convert les from LAN or Descriptive to SAN. Additionally, the processor should be able to validate logs, and produce error messages when the chess notation is invalid. Additionally, once the basic PGN capabilities are in place, a program can be adapted to do analysis of games. For instance it should be able to report only games that have specic openings, piece counts at the end, promotions to queen, castling, checks, etc.
554
Part VII
Back Matter
555
CHAPTER
FORTYFIVE
BIBLIOGRAPHY
45.1 Use Cases 45.2 Computer Science 45.3 Design Patterns 45.4 Languages 45.5 Problem Domains
557
558
Chapter 45. Bibliography
CHAPTER
FORTYSIX
INDICES AND TABLES

Index Module Index Search Page
559
560
Chapter 46. Indices and Tables
CHAPTER
FORTYSEVEN
PRODUCTION NOTES
The following toolset was used for production of this book. Python 2.6.3. Sphinx 0.6.3. Docutils 0.5. Komodo Edit 5.2.2. pyPDF 1.12. MacTeX-2008.
561
562
Chapter 47. Production Notes
BIBLIOGRAPHY
[Jacobson92] Ivar Jacobson, Magnus Christerson, Patrik Jonsson, Gunnar vergaard. Object-Oriented Software Engineering. A Use Case Driven Approach. 1992. Addison-Wesley. 0201544350. [Jacobson95] Ivar Jacobson, Maria Ericsson, Agenta Jacobson. The Object Advantage. Business Process Reengineering with Object Technology. 1995. Addison-Wesley. 0201422891. [Boehm81] Barry Boehm. Software Engineering Economics. 1981. Prentice-Hall PTR. 0138221227. [Comer95] Douglas Comer. Internetworking with TCP/IP. Principles, Protocols, and Architecture. 3rd edition. 1995. Prentice-Hall. 0132169878. [Cormen90] Thomas H. Cormen, Charles E. Leiserson, Ronald L. Rivest. Introduction To Algorithms. 1990. MIT Press. 0262031418. [Dijkstra76] Edsger Dijkstra. A Discipline of Programming. 1976. Prentice-Hall. 0613924118. [Gries81] David Gries. The Science of Programming. 1981. Springer-Verlag. 0387964800. [Holt78] R. C. Holt, G. S. Graham, E. D. Lazowska, M. A. Scott. Structured Concurrent Programming with Operating Systems Applications. 1978. Addison-Wesley. 0201029375. [Knuth73] Donald Knuth. The Art of Computer Programming. Fundamental Algorithms. 1973. AddisonWesley. 0201896834. [Meyer88] Bertrand Meyer. Object-Oriented Software Construction. 1988. Prentice Hall. 0136290493. [Parnas72] D. Parnas. On the Criteria to Be Used in Decomposing Systems into Modules. 1053-1058. 1972. Communications of the ACM.. [Gamma95] Erich Gamma, Richard Helm, Ralph Johnson, John Vlissides. Design Patterns. Elements of Object-Oriented Software. 1995. Addison-Wesley Professional. 0201633612. [Larman98] Craig Larman. Applying UML and Patterns. An Introduction to Object-Oriented Analysis and Design. 1998. Prentice-Hall. 0137488807. [Lott05] Steven Lott. Building Skills in Object-Oriented Design. Step-by-Step Construction of A Complete Application. 2005. Steven F. Lott [Rumbaugh91] James Rumbaugh, Michael Blaha, William Premerlani, Frederick Eddy, William Lorensen. Object-Oriented Modeling and Design. 1991. Prentice Hall. 0136298419. [Geurts91] Leo Geurts, Lambert Meertens, Steven Pemberton. The ABC Programmers Handbook. 1991. Prentice-Hall. 0-13-000027-2 [Gosling96] Gosling, McGilton. Java Language Environment White Paper. 1996. Sun Microsystems.. [Harbison92] Samuel P. Harbison. Modula-3. 1992. Prentice-Hall. 0-13-596396-6.
563
[vanRossum04] Guido van Rossum, Fred L. Drake. Python Documentation. 2004. Python Labs.. [Wirth74] Proceedings of the IFIP Congress 74. 1974. North-Holland.. On the Design of Programming Languages. 386-393. [Banks02] Robert B. Banks. Slicing Pizzas, Racing Turtles, and Further Adventures in Applied Mathematics. 2002. Princeton University Press. 0-691-10284-8. [Dershowitz97] Nachum Dershowitz, Edward M. Reingold. Calendrical Calculations. 1997. Cambridge University Press. 0-521-56474-3 [Latham98] Lance Latham. Standard C Date/Time Library. 1998. Miller-Freeman. 0-87930-496-0. [Meeus91] Jean Meeus. Astronomical Algorithms. 1991. Willmann-Bell Inc.. 0-943396-35-2. [Neter73] John Neter, William Wasserman, G. A. Whitmore. Fundamental Statistics for Business and Economics. 4th edition. 1973. Allyn and Bacon, Inc.. 020503853. [OBeirne65] T. M. OBeirne. Puzzles and Paradoxes. 1965. Oxford University Press.. [Shackleford04] Michael Shackleford. The Wizard Of Odds. 2004. [Silberstang05] Edwin Silberstang. The Winners Guide to Casino Gambling. 4th edition. 2005. Owl Books. 0805077650. [Skiena01] Steven Skiena. Calculated Bets. Computers, Gambling, and Mathematical Modeling to Win. 2001. Cambridge University Press. 0521009626.
564
Bibliography

Building Skills in Python

Uploaded by

Copyright:

Available Formats

Building Skills in Python

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Building Skills in Python

Uploaded by

Copyright:

Available Formats

Building Skills in Python

April 20, 2010

5.4 5.5 5.6

11 Additional Notes On Functions 125 11.1 Functions and Namespaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 ii

Data + Processing = Objects

Components, Modules and Packages

41.1 41.2 41.3 41.4 41.5

Equal Temperament Overtones . . . . . . Circle of Fifths . . . Pythagorean Tuning Five-Tone Tuning .

Building Skills in Python, Release 2.6.5

A Programmers Introduction to Python

Building Skills in Python, Release 2.6.5

1.1 Why Read This Book?

Building Skills in Python, Release 2.6.5

Building Skills in Python, Release 2.6.5

1.3 Organization of This Book

Building Skills in Python, Release 2.6.5

Building Skills in Python, Release 2.6.5

1.5 Programming Style

1.6 Conventions Used in This Book

1.5. Programming Style

Building Skills in Python, Release 2.6.5

Building Skills in Python, Release 2.6.5

The Processing View

Building Skills in Python, Release 2.6.5

BACKGROUND AND HISTORY

2.2 Features of Python

Building Skills in Python, Release 2.6.5

2.3.1 The Java Yardstick

Building Skills in Python, Release 2.6.5

Building Skills in Python, Release 2.6.5

2.3.2 The Modula-2 Yardstick

Chapter 2. Background and History

Building Skills in Python, Release 2.6.5

2.4 Some Jargon

2.4. Some Jargon

Building Skills in Python, Release 2.6.5

Chapter 2. Background and History

3.1 Windows Installation

Building Skills in Python, Release 2.6.5

3.1.1 Windows Pre-Installation

3.1.2 Windows Installation

Chapter 3. Python Installation

Building Skills in Python, Release 2.6.5

3.1.3 Windows Post-Installation

3.1. Windows Installation

Building Skills in Python, Release 2.6.5

3.2 Macintosh Installation

3.2.1 Macintosh Pre-Installation

3.2.2 Macintosh Installation

Chapter 3. Python Installation

Building Skills in Python, Release 2.6.5

3.2.3 Macintosh Post-Installation

Enter end-of-le ctrl-D to exit from Python.

3.3 GNU/Linux and UNIX Overview

3.3. GNU/Linux and UNIX Overview

Building Skills in Python, Release 2.6.5

3.3.1 Checking for Python

3.3.2 YUM Installation

Chapter 3. Python Installation

Building Skills in Python, Release 2.6.5