Compiler Construction and Formal Languages: 1 Static or Dynamic?

Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

Compiler Construction and Formal Languages

Exercises Berthold Homann Informatik, Universit at Bremen, Germany hof@informatik.uni-bremen.de January 2013
Abstract By doing the following exercises, the students attending the course Compiler Construction and Formal Languages shall apply the concepts and algorithms taught in the course. Some of these exercises will be indicated as assignments, to be done by students, either alone or in groups of at most two, in order to get credit points and marks for the course. Some of the other exercises will be done in class.

Static or Dynamic?
static if it can be checked already while the program is compiled, and dynamic if it can, in general, only be checked at the time when the compiled program is executed.

A property P of a programming language is called

Please consider the programming languages C, Java, and Haskell (or other languages that you know). Find out which of the following properties is static, and dynamic in these languages, respectively: 1. The binding of names (identiers) to the declarations introducing them. 2. The type of a variable. 3. The bounds of an array. 4. The value of a variable. Please compare the languages wrt. your ndings.

Bootstrapping

Implementing Java The language Java is implemented by a compiler and an interpreter: The compiler transforms Java programs into Java Byte Code (JBC). JBC-programs are executed by an interpreter, known as the Java Virtual Machine (JVM).

A Portable Java Implementation Portable implementations of Java like the Java Developer Kit by Sun systems are based on the following components: 1. A master compiler from Java to JBC, written in Java itself. This compiler can be assumed to use only a small portion of the JVM library. 2. A compiler from Java to JBC, written JBC, which is derived from the master compiler. (You get this compiler for free as soon as you are able to execute Java on some platform.) 3. A Java Virtual Machine, an interpreter for JBC, written in Java. Tasks 1. Port to machine M . How can Java be ported to a new machine M , given the components above? Do this by implementing a component that is as simple and small as possible. You may assume that a native C-compiler, producing M machine code, and written in M machine code, exists for M . The resulting implementation need not run eciently it may involve an interpreter. 2. Bootstrap on machine M . How can the compiler be made native? Again, do this by implementing a small component.

3
3.1

Regular Denitions
Identiers

shall be formed according to the following rules: 1. Identier are strings (of arbitrary length) that consist of letters, digits, and (underscore). 2. Identiers start with a letter. 3. An underscore may appear only between two letters or digits.1 Examples: x Hallo Welt Identifier koennen lang sein Counterexamples: 1x H a l l o W e l t Name

3.2

String Literals

shall be dened as follows: 1. String literals are enclosed in quotes ". 2. They consist of an arbitrary number of printable characters. 3. They must not contain a newline character. 4. If one quote shall be made part of a string literal, it has to appear twice in it.2 Examples: The string literals "", " Hallo, Welt!", and """ Hallo, Welt!""" contain 0, 12, and 14 characters, respectively.
1 These 2 These

are the rules for identiers in the programming language Ada. are the rules for string literals in the programming language Pascal.

Task Give regular denitions for identiers and string literals. Tip PrintableChar shall be the set of all printable characters, including newline. The regular expression\n shall represent newline. The regular expression [c1 , . . . , cn ] shall denote the set of characters that contains all printable characters, except for c1 bis cn ; i.e., the set PrintableChar \ {c1 , . . . , cn }.

Finite Automata

Use the rules discussed in the course to construct, for the regular denitions of identiers and string literals, resp., 1. a non-deterministic automaton, 2. the equivalent deterministic automaton, by the quotient set construction, and 3. the minimal deterministic automaton.

Recognition of Comments
1. Dene comments by a regular denition. 2. Construct a minimal nite automaton for comments from this denition. You may do this intuitively, without taking the formal steps discussed in the course. 3. Check whether your denition gets simpler if you use a lookahead operator R1 /R2 , which generates the expression R1 only if it is followed by the expression R2 .

Comments shall be enclosed in /* und */; they may include newlines.

Recognition of Float Literals


String literals shall be dened by num = digit . digit+ exp?

1. Construct a minimal nite automaton for comments from this regular denition. You may do this intuitively, without taking the formal steps discussed in the course.

Lexical Analysis of OOPS-0 with lex

The language oops, which is used in the compiler project at Universit at Bremen, has the following lexemes: non-empty sequences of layout characters blank, tabulator, and newline, comments, also extending over several lines, enclosed in (* and *), identiers, 3

integer literals, string literals, the delimiters ,, :, ;, (, and ), the operator symbols <=, >=, :=, *, /, +, -, =, #, <, >, and ., as well as the keywords AND, ATTRIBUTE, BEGIN, CLASS, ELSE, ELSEIF, END, EXTENDS, IF, IS, METHOD, MOD, NEW, NOT, OR, OVERRIDE, READ, RETURN, THEN, WITH, WHILE, and WRITE. Tasks 1. Dene the scanner for oops in lex. Here, we are not interested in the translation of lexemes into tokens, so the return statements can be omitted.) The regular denitions developed in the course can be used for the lexemes. 2. Produce tow versions of the scanner: one without keywords, and one including them (Then their denitions should go before that of identiers.) 3. Compare the size of the scanner tables? So, should be keywords rather be recognized by the screener?

Ambiguity of Context-Free Grammar


R | | | | R| R RR R+ | R | R? ( R) a | b | c

Consider the following context-free rules, where terminals are underlined, and R is the start symbol:

Tasks 1. Which language is dened here? 2. Show that the grammar is ambiguous. 3. Make it unambiguous. Derivations and derivation trees shall respect that the precedences of the operators increases from top to bottom The operators , + und ? have the same precedence.

SLL(1)-Property for a Grammar of Commands


program statements ::= statement ::= statement | statement statement ; statements

Let the syntax of commands be dened as follows:

::= if expression then statements end if | | | | if expression then statements else statements end if while expression do statements end while id := expression id

expression Tasks

::= e

1. Check whether the grammar satises the SLL(1) condition. 2. Determine the First and Follower sets for that purpose. 3. If the condition is violated, transform the grammar by left factorization until it does satisfy it.

Recursive-Descent Parsing of Commands

Consider the SLL(1) syntax for commands from the previous exercise. Tasks 1. Generate a recursive descent parser. 2. Construct an abstract syntax for commands. 3. Add tree-constructing instructions to the parser. 4. Incorporate error handling into the parser, based on deletion and insertion of symbols.

10

A Grammar Violating the SLR(1) Condition


S E E$ | | | E+E EE (E) id

Consider the following grammar for expressions:

Z 0 1 2 3 4 5 6 7 8 9

$ acc r5

+ s4 r5

Action ( s5 r5 s2 s2 s2 s2 s2

id s3 s3 s3

Goto E 1 6 7 8

r5 s3 s3 s9 r2 r3 r4

r2 r3 r4

s4 r2 r3 r4

s5 s5 r3 r4

Figure 1: An action and goto table for parsing expressions Tasks 1. Construct the characteristic nite automaton for the grammar. 2. Show that the grammar violates the SLL(k ) condition, by considering its action table.

11

Bottom-Up SLR(1)-Parsing
S E | | | E$ E+E EE (E) id (1) (2) (3) (4) (5)

The grammar

has a parse table as in Figure 1 (Precedences and associativities of the operators have been used to resolve the shift-reduce conicts for this grammar.) The parser interpretes this table. States and rules are represented by their numbers, and the start state is 0: pop(0); rst symbol; while Action[top, symbol] = acc do case Action[top, symbol] of si push(symbol); push(i); nextsymbol; | rn for i:= 1 to 2* length(rule(n)) do pop; z:= top; push(left-hand-side(rule(n))); push(Goto[z]) otherwise syntaxerror( ); end case end while; if Action[top, symbol] = acc symbol = $ then print(yes!) else print(no!) end if. Tasks Execute the parser with the following programs, and non-programs:

abc ( a b) c (a b)c ( a b) c Here a, b, c are numbers. Represent the congurations of the parser as s0 v1 sk1 vk sk t tn w 1 where w is the prex read, the s0 , si are states, the vi are symbols (for 1 terminal symbols (for 1 j n). i k ), and the tj are

12

Identication with Linear Visibility

Consider the following program: declare x:k1; begin declare x:k2; declare y:k3; use y: use x end; use x Use the attribute grammar presented in the course in the following steps: Construct an abstract syntax tree. Determine the attribution rules for the program. Which declarations are associated with the identiers x and y? Recall the attribute rules:
v init v St n P v nest B n unnest v St n v St n Id enter Kd Id def Kd v L n v D n v U n

v St n

13

Attribute Grammar for Global Visibility

Dene, for the known abstract syntax and semantical basis, an attribute grammar that denes identication according to simultaneous visibility. The rules of the underlying grammar dene the following syntax trees:
P B L D U

St

St

St

St

Id

Kd

Id

Kd

The semantical basis should be extended by one operation: 7

The predicate isLocal can be useful to determine whether an identier has already be declared within the current block. In Haskell, the extended interface would be: module DeclarationTable where type TAB = [[(Id,Kd)]] type Id = String data Kd = Unbound | Typ | Const | Var | Proc deriving (Eq, Text) initial :: TAB nest, unnest :: TAB -> TAB enter :: TAB -> (Id,Kd) -> TAB def :: TAB -> Id -> Kd isLocal :: TAB -> Id -> Bool

14

Attribution with Simultaneous Visibility


P

Determine the attribution of the following syntax tree in a language with simultaneous visibility.

Id

Kd

Id

Kd

Id

Kd

Id

Kd

Id

Kd

Id

Kd

15

Declaration Tables in Java

Implement the semantic basis for identication, the declaration table, in Java. The signature is: module DeclarationTable where type TAB = [[(Id,Kd)]] type Id = String data Kd = Unbound | Typ | Const | Var | Proc deriving (Eq, Text) initial :: TAB nest, unnest :: TAB -> TAB enter :: TAB -> (Id,Kd) -> TAB def :: TAB -> Id -> Kd isLocal :: TAB -> Id -> Bool

16

Declaration Tables for Overloading

Until now, the grid-like declaration table copes with declarations that are monomorphic so that an identier is associated with at most declaration. In a polymorphic language, a set of declarations can be associated with certain identiers (of operations and functions). Then the function def yields a list (Kinds) of Kd. The function enter, when adding a declaration (x,k) to a table, has to decide whether 1. the entry can just be added, or 2. whether a similar entry is hidden by the new one, or 3. whether the new declaration is illegal, since the entry to be hidden appeared in the same block. The predicate former called isLocal is not helpful any more. A predicate distinct should be provided, which determines whether two declarations (their Kd, that is) are distinguishable, and can hence be overloaded. 3 This predicate can be used in the new implementation of enter. In Haskell the changed interface is:: module DeclarationTable where type TAB = ... type Kinds = [Kd] type Kd = ... type Id = ... initial :: TAB nest, unnest :: TAB -> TAB enter :: TAB -> (Id,Kd) -> TAB def :: TAB -> Id -> Kinds distinct :: Kd -> Kd -> Bool Tasks 1. Extend the ecient implementation of declaration tables so that it copes with overloaded declarations. 2. Does the complexity of the operations change?

17

Type Compatibility
u expressing that a type u

For the type analysis of expressions, we have introduced a relation t is compatible with a type t. Tasks

Consider a programming language you know (Java, Haskell, C, C# . . . ). 1. At which program places, type compatibility is required? 2. How is type compatibility dened in the language? 3. Does a single notion of type compatibility hold throughout the language, or are there dierent notions, depending on the context?
3 Languages dier wrt. the conditions under which functions or operations are considered to be distinguishable. This will be discussed in the course.

18

Transforming and Evaluating Expressions

Determine the transformation c = codeW (3 + (x 4)). Assume that (x) = 3, and that dop = mul has the meaning: S [SP ] := S [SP 1] S [SP ]; SP := SP 1 Execute c with the P-Machine. Assume the code c starts at location 10; S [3] shall contain the value 10 , and SP shall have the value 3. The equations for transformation are on slide 169/171. The sketch of the P-machine is on page 166. [There is a minor error on this page. Do you nd it?] A simple form of numerical choice takes the following form: C ::= case E of u => C u ; . . . ; o => C o default Cd

The value v of the expression E is used to select the command Cv labeled with the corresponding value, from a contiguous integer range u . . . o. If v is not in the range u . . . o, the default command Cd is selected. (The pre-historic switch of Algol-60, C, C++, and Java does not necessarily perform a unique selection (only if all commands are terminated with break). Numerical choice can be transformed, on the source level, into an equivalent cascade of Boolean choices (if then else ). It can also be implemented with a goto table: n unconditional jumps ujp li are inserted in the code not in the data and a computed goto jumps to the right place in that table. This instruction takes the following form: ijp P C := P C + S [SP ]; SP := SP 1

The instruction expects the goto table to be placed in the directly succeeding code locations. Following instructions might be useful. ljp u l gjp o l Tasks 1. Translate numeric choice, either to equivalent if -cascades, or with a goto table. 2. Which of the implementations do you consider to be better? More precisely, under which circumstances could one be better than the other one? 3. If the same command shall be selected for dierent numerical values, the simple form of case forces to duplicate code, which should be avoided for several reasons. How could you change the transformation, if a list of literals can be placed in front of the commands? if s[SP ] < u then P C := P C + l if s[SP ] > o then P C := P C + l

19

Transforming Generalized Expressions

1. Predened Functions. Allow calls to standard functions of the form f (e1 , . . . ek ) in expressions, and extend the transformations to handle these calls. Assume that that (f ) contains the address of code(f ). Transform function calls into machine code. You may wish to introduce an instruction that jumps to the code of standard functions. This instruction could use a register, say OPC , to save the program counter. An instruction for returning from a standard function concludes the code of standard functions. Assume that this instruction restores the program counter with P C := OP C . 10

2. Compound Operands. Allow that the inx operations and standard functions in an expression may have operands and results that occupy several storage cells, not just one. Extend the transformation scheme accordingly.

20

Array Transformation

Consider the following Pascal program: program Feld; type Complex = record re, im: Real; end; var a: array [-3..+3] of Complex; var i: Integer; begin i:= 1; a[i] := a[i+1] end. Determine the size of the types, and the address environment of the program. Translate the program and execute the translated P-code. Take care: the transformation presented in the course does only cover array ranges 0 . . . k . Invent an extension for the more general case (or just look it up in the presentations).

21

Storage and Selection of Dynamic Arrays

So far assumed that the range [u..o] of an array is static. Then its size is static as well, like of any other type discussed so far. Values of static arrays can be allocated on the stack. An array is dynamic if its range is [Eu ..Eo ] is given by the values of expressions. The size of a dynamic array is dynamic (hence the name), so that its values cannot be stored on the stack like values of static size. Many modern languages allocate dynamic arrays on the heap. This has the drawback that the space for an array has to be recovered by garbage collection if it is no longer alive. It would be better to allocate it on the stack where the storage is reclaimed automatically. For this purpose, he representation of a dynamic array is divided into two parts: 1. Its descriptor contains all information that is needed to perform the operations on it: The upper bound og (the value of Eo ). The lower bound ug (the value of Eu ). The subtrahend su (the dierence to the 0th element). The size gr, as the number of its storage cells. The ctive address fa (the address of the 0th element). 2. Its proper values are kept in a dedicated area DYN of the stack, between the area (LOC ) for values of static size and the stack for intermediate calculations in expressions (ZWERG ). The size of a dynamic array is just the (static) size of its descriptor: size (row [Eu ..Eo ] of t) = 5 Then the address environment can be determined as before. Furthermore, the elaboration of a dynamic array declaration must produce code that, at runtime, denes the values of the descriptor, and allocates space for its proper value in DYN. 11

Table 1: Instructions allocating and accessing dynamic arrays Command Meaning Explanation sad a g S [a + 4] := S [SP ]; (og) S [a + 3] := S [SP 1]; (ug) S [a + 2] := S [a + 3] g ; (su) S [a + 1] := S [a + 4] S [a + 3] + 1; (gr) S [a] := SP S [a 2] 1; (fa) SP := SP + S [a + 1] 2 (allocate space) dpl SP := SP + 1; S [SP ] := S [SP 1]; ind S [SP ] := S [S [SP ]] chd if S [SP ] < S [S [SP 2] + 3] or S [SP ] > S [S [SP 2] + 4] then error index error; ixa g S [SP 1] := S [SP 1] + (S [SP ] g ); SP := SP 1; We assume that the elaboration of local declarations calls an instruction isp g that increases register SP by the size g of all local declarations including the descriptors of dynamic arrays. Then ever declaration of a dynamic array produces the following code: [[x : row [Eu ..Eo ] of t]] = [[V ]]A ; [[Eu ]]W ; [[Eo ]]W ; sad (x) size (t); The access to an element of a dynamic array is translated as follows: [[V [E ]]] = [[V ]]A ; dpl ; ind ; [[E ]]W ; chd ; ixa size (type (x[E ])) ; sli The code [[V ]]A yields the descriptor address of a dynamic array x, not the address of the values themselves. The descriptor address is duplicated with the instruction dpl rst, and then used to calculate the address of the array values with ind (their ctive address, rather), before the Index is calculated. The original of the descriptor is needed to check array bounds with the command chd . If the variable access V is not just an identier x, but a component, say y.s, of a product, the address of the descriptor cannot be passed to the instruction directly. The original descriptor is two cells below SP above is the address of the array values and the values of the index. Only when the address of the array element is calculated, it will be slided down on the stack, overwriting the original descriptor, by the instruction sli. In the following Pascal program: procedure Feld; var a: array [-n..+n] of Complex; var i: Integer; begin i:= 1; a[i] := a[i+1] end; the global variable n shall have the value 3. Let size (Complex) = 2 and = {n 5, a 10, i 15}. After elaboration of the declarations, the stack pointer shall point to the last cell occupied by local variables, namely 15. The code for allocating the array looks as follows: [[x: array [-n..+n] of Complex]] = [[x]]A ; [[n]]W ; [[n]]W ; sad (x) size (Complex); lda (n) ; ind ; sad (x) size (Complex); = lda 10 ; lda 5 ; ind ; neg ; lda 5 ; ind ; sad 10 2; 12

= lda (x) ; lda (n) ; ind ; neg ;

By these instructions, the store is manipulated as shown in Figure2. 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 29

(n)

(x)

SP (i)

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

(n)

[[n]]W

[[n]]W

(x)

(i) SP

0 1 2 3 4 3 5 6 7 8 9 10 11 12 13 14 15 3 16 3 17

(n)

=
sad (x)2

(x)SP su 1 (og ug + 1) 2 (ug 2) ug og (i) SP

22 14 6 3 3 . . .

Figure 2: Allocation of an array descriptor

Task 1. Dene the code for [[x[i]]], assuming that i has the value 1. 2. Demonstrate how the code is executed.

22

Static Predecessors

In the course, we have discussed how static predecessors can be organized as a linked list. This leads, with deeply nested blocks, to a certain overhead in the access to non-local identiers. This can be avoided if static predecessors are organized in a display vector. Here the static predecessors are held in a (small) global stack at the start of the storage area. As the depth of block nesting is a static value, the maximal extension of the display vector can be determined in advance. Tasks Organize static predecessors in a display vector. 1. What has to be done when entering and leaving a procedure? 2. Redene the instruction lda. You may change the meaning of the modier for that purpose.

3. If a variable is global, i.e., declared on level = 0, or local, i.e., declared on = current level , the display vector is not needed at all. Dene the instructions ldg and ldl for access to global and local variables, respectively.

13

23

Translating Recursive Procedures

Translate the following Pascal function: function fac (n: Integer): Integer; begin if n <= 0 then fac := 1 else fac := fac(n-1) end Task The statement fac := E denes the result of the function fac (instead of return E in other languages). Execute the code.

24

Parameter Passing

This is about endangered ways of parameter passing. Let proc p(x : t); B be a procedure that is called as p(E ). Constant Parameters If x is handled as a constant parameter, x acts as a constant name for the value of E while p is executed, as if a declaration const x = E ; would be elaborated before entering p. Questions 1. In what diers this passing mode to value parameters (call by value )? 2. Consider the advantages and drawbacks of constant wrt. value parameters. 3. What code has to be generated for passing a dynamic array as a constant parameter? Name Parameters This way of parameter passing resembles textual substitution, and has been used in the lambda calculus. It is dened as follows: If x is passed as a name parameter, it denotes the expression E while the body B of p is executed. Whenever x is used, the expression E will be evaluated; this is done in the context of the block B , considering all declarations for B . Questions 1. Consider advantages and drawbacks of name parameters. 2. Sketch how name parameters could be transformed into code. 3. What is the dierence of name parameters to pure textual substitution?

25

Translating a Complete Program

Translate the following program:

14

program fakultaet; var x : Integer; function fac (n: Integer): Integer; begin if n <= 0 then fac := 1 else fac := n * fac(n-1) end begin x := fac(2) end.

26

Multiple Inheritance

Consider how the organization of object-oriented code must be changed in the presence of proper multiple inheritance (not only of interfaces, as in Java). 1. How can method tables be organized? 2. How can dynamic binding of methods be done?

15

You might also like