Library of Congress Cataloging in Publication Data Main entry under title: Postscript language program design. Includes index.

1. PostScript (Computer program language) I. Adobe Systems. QA ???? ISBN 0-201-14396-8

Copyright 1988 by Adobe Systems Incorporated. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior written permission of the publisher. Printed in the United States of America. Published simultaneously in Canada. PostScript is a registered trademark of Adobe Systems Incorporated. The information in this book is furnished for informational use only, is subject to change without notice, and should not be construed as a commitment by Adobe Systems Incorporated. Adobe Systems Incorporated assumes no responsibility or liability for any errors or inaccuracies that may appear in this book. The software described in this book is furnished under license and may only be used or copied in accordance with the terms of such license. ISBN 0-201-14396-8 ABCDEFGHIJ-HA-898 First printing: February 1988

iv

CONTENTS
PREFACE.................................................................................. xi chapter 1 THE POSTSCRIPT LANGUAGE: OVERVIEW ............................1
1.1 INTRODUCTION...................................................................................1 1.2 THE LANGUAGE MODEL.......................................................................2 DICTIONARIES AND DATA STRUCTURES...............................................4 STACKS ...........................................................................................4 BUILT-IN POSTSCRIPT LANGUAGE OPERATORS .....................................5 1.3 THE IMAGING MODEL..........................................................................5 COORDINATE SYSTEMS .....................................................................6 PATHS AND PAINT ............................................................................7 FONTS ............................................................................................7 1.4 ELECTRONIC PUBLISHING AND PRINTING ...........................................7 1.5 PROGRAM DESIGN GUIDELINES ...........................................................9

chapter 2 THE EXECUTION MODEL ......................................................11

2.1 INTRODUCTION.................................................................................11 2.2 THE PRINTING JOB MODEL ................................................................11 2.3 THE OPERAND STACK.........................................................................13 OBJECTS IN THE POSTSCRIPT LANGUAGE ..........................................13 THE STACK AS A DATA STRUCTURE ...................................................15 2.4 THE DICTIONARY STACK ....................................................................16 DICTIONARY OBJECTS .....................................................................17 USING THE DICTIONARY STACK .......................................................18 2.5 OPERATORS AND NAME LOOKUP ......................................................19 THE BIND OPERATOR ......................................................................21 2.6 THE INTERPRETER AND THE SCANNER ...............................................23 RECOGNITION OF OBJECTS .............................................................24 2.7 PROCEDURES......................................................................................25 VERY LARGE PROCEDURE BODIES ......................................................29

2.8 THE EXECUTION STACK ..................................................................... 32 2.9 THE SERVER LOOP .............................................................................. 34

chapter 3 THE IMAGING MODEL.......................................................... 37

3.1 INTRODUCTION................................................................................. 37 3.2 APPLYING THE METAPHOR................................................................. 38 3.3 CONSTRUCTION OF PATHS ............................................................... 40 THE GRAPHICS STATE AND PATHS .................................................... 40 3.4 PAINTING OPERATIONS ..................................................................... 42 3.5 WHAT HAPPENS TO THE CURRENT PATH? ......................................... 43 3.6 PROCEDURES FOR CONSTRUCTING PATHS ....................................... 44 RECTANGLES ................................................................................. 44 CIRCLES AND ARCS ........................................................................ 46 ARROWHEADS ............................................................................... 46 3.7 TEXT OPERATIONS ............................................................................. 49 CHARACTER WIDTHS ...................................................................... 51 3.8 CLIPPING............................................................................................ 53 COMPLEXITY AND PERFORMANCE ................................................... 53 3.9 RASTERIZATION.................................................................................. 54 3.10 SAVE AND RESTORE............................................................................ 54 3.11 THE FONT CACHE .............................................................................. 55

chapter 4 EMULATORS AND TRANSLATORS ........................................ 57

4.1 INTRODUCTION................................................................................. 57 4.2 EMULATING ANOTHER PRINTER......................................................... 58 THE STRINGWIDTH OPERATOR ......................................................... 61 TEXT JUSTIFICATION IN AN EMULATOR ............................................. 62 4.3 TRANSLATING EXISTING FILE FORMATS............................................. 68 UNITS ........................................................................................... 68 FONTS .......................................................................................... 69 4.4 FONT DIFFERENCES............................................................................ 69 4.5 USING THE IMAGING MODEL ............................................................ 71 PRESERVING HIGH-LEVEL INFORMATION ........................................... 72 RENDERING ................................................................................... 72 4.6 OPTIMIZING TRANSLATOR OUTPUT .................................................. 72 4.7 COMPUTATION AND DECISION-MAKING ......................................... 73

vi

chapter 5 DESIGNING THE PAGE AND THE PROGRAM .......................77

5.1 INTRODUCTION.................................................................................77 5.2 PAGE LAYOUT CONSIDERATIONS.......................................................77 PAGE NESTING AND INDEPENDENCE ................................................78 5.3 PRODUCING POSTSCRIPT LANGUAGE OUTPUT .................................79 5.4 ROUND-OFF AND COORDINATE SYSTEMS.........................................80 5.5 EFFICIENCY.........................................................................................81 DATA TRANSMISSION OVERHEAD .....................................................82 COMPUTATION..............................................................................83 INTERPRETATION TIME ....................................................................85

chapter 6 PROGRAM STRUCTURE .........................................................87

6.1 INTRODUCTION.................................................................................87 6.2 THE PROLOGUE AND SCRIPT MODEL.................................................87 6.3 MODULARITY AND PAGE STRUCTURE ................................................89 GROUND STATE .............................................................................90 THE OPERAND STACK .....................................................................92 FUNCTIONAL AND GRAPHIC INDEPENDENCE ....................................92 SAVE AND RESTORE ........................................................................93 PAGE ELEMENTS AND THEIR PROPERTIES ...........................................94 6.4 DOCUMENT STRUCTURING CONVENTIONS......................................96

chapter 7 THE MECHANICS OF SETTING TEXT ....................................99

7.1 INTRODUCTION.................................................................................99 7.2 CHARACTER WIDTHS ........................................................................100 7.3 MARGINS AND JUSTIFICATION.........................................................102 JUSTIFICATION .............................................................................104 7.4 HANDLING DIFFERENT FONTS .........................................................106 7.5 LEADING AND POINT SIZE ...............................................................108 7.6 KERNING AND LIGATURES................................................................109 7.7 ENCODING AND CHARACTER SETS..................................................114 7.8 COMPOSITE CHARACTERS AND ACCENTS .......................................117 7.9 NON-ROMAN FONTS.......................................................................117 CHARACTER WIDTHS AND ORIGINS ................................................118

vii

chapter 8 SCANNED IMAGES AND HALFTONES ................................ 123

8.1 INTRODUCTION............................................................................... 123 8.2 THE IMAGE OPERATOR..................................................................... 123 HOW IT WORKS ........................................................................... 124 THE IMAGE MATRIX ...................................................................... 124 DATA ACQUISITION PROCEDURES .................................................. 126 SMALL AMOUNTS OF DATA ........................................................... 127 LARGE AMOUNTS OF DATA ........................................................... 127 A COMMON ERROR AND ITS CAUSE ............................................... 129 SYNTHETIC DATA ......................................................................... 129 8.3 DATA COMPRESSION....................................................................... 131 8.4 HALFTONE SCREENS ........................................................................ 131 HALFTONING IN THE POSTSCRIPT LANGUAGE ................................. 132 CHANGING THE HALFTONE SCREEN ............................................... 132 8.5 THE SPOT FUNCTION....................................................................... 134

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING .......................... 137

9.1 INTRODUCTION............................................................................... 137 9.2 PATTERN FILLS.................................................................................. 137 9.3 LOGOS, GRIDS, FORMS, AND SPECIAL FONTS ................................. 143 GRIDS ......................................................................................... 146 9.4 TRANSFORMATION MATRICES......................................................... 149 INVERTED COORDINATE SYSTEMS .................................................. 151 9.5 COLOR AND COLOR SEPARATIONS ................................................. 151 COLOR SEPARATIONS ................................................................... 152 SPOT COLOR ............................................................................... 154

chapter 10 FILE INTERCHANGE STANDARDS ....................................... 157

10.1 INTRODUCTION............................................................................... 157 10.2 CONFORMING DOCUMENTS .......................................................... 157 10.3 HANDLING PRINTER-SPECIFIC FEATURES.......................................... 158 10.4 SPECIFYING PAPER SIZES .................................................................. 159 10.5 PRINTER QUERIES ............................................................................. 162 10.6 CONDITIONAL EXECUTION ............................................................. 163 10.7 FONT AVAILABILITY .......................................................................... 164 10.8 PUTTING IT ALL TOGETHER.............................................................. 165

viii

chapter 11 MERGING FILES FROM DIFFERENT SOURCES.....................167

11.1 INTRODUCTION...............................................................................167 11.2 USING EXISTING CONTEXT..............................................................168 11.3 ERROR RECOVERY .............................................................................169 11.4 HANDLING SHOWPAGE ...................................................................170 11.5 SCREEN REPRESENTATIONS..............................................................171

chapter 12 WRITING A PRINT SPOOLER ...............................................173

12.1 INTRODUCTION...............................................................................173 12.2 PRINTER MANAGEMENT...................................................................174 12.3 COMMUNICATIONS ........................................................................174 MESSAGES ...................................................................................175 12.4 USING EXITSERVER ...........................................................................176 12.5 MANAGING FILES AND FONTS.........................................................178 %%DocumentFonts ......................................................................178 %%IncludeFont ............................................................................179 %%BeginFont, %%EndFont ...........................................................180 12.6 DETERMINING WHAT FONTS ARE AVAILABLE ...................................181 12.7 HANDLING RESOURCE SHORTAGES .................................................182 12.8 PRINTER DESCRIPTION FILES ............................................................183

chapter 13 MEMORY AND FILE RESOURCE MANAGEMENT ................185

13.1 MEMORY STRUCTURE.......................................................................185 13.2 MEMORY ALLOCATION ....................................................................186 13.3 SAVE AND RESTORE..........................................................................187 SAVE OBJECTS ..............................................................................187 THE INVALIDRESTORE ERROR..........................................................188 13.4 DOWNLOADABLE FONT PROGRAMS ...............................................190 13.5 PACKED ARRAYS ...............................................................................190 13.6 RASTER MEMORY..............................................................................191 13.7 FILE SYSTEMS AND DISK MANAGEMENT..........................................192 13.8 POSTSCRIPT LANGUAGE FILE OPERATIONS ......................................192 THE STANDARD INPUT STREAM ......................................................193

ix

chapter 14 ERROR HANDLING .............................................................. 197

14.1 INTRODUCTION............................................................................... 197 14.2 STRATEGIES ...................................................................................... 197 NON-STANDARD OPERATORS ........................................................ 197 IMPLEMENTATION LIMITS EXCEEDED ............................................. 198 14.3 THE STOPPED OPERATOR................................................................. 199 14.4 THE ERROR HANDLING MECHANISM............................................... 203 14.5 REDEFINING ERROR PROCEDURES.................................................... 204 14.6 HANDLING ERROR MESSAGES.......................................................... 206

chapter 15 DEBUGGING TECHNIQUES................................................. 207

15.1 INTRODUCTION............................................................................... 207 15.2 ESTABLISHING TWO-WAY COMMUNICATION ................................. 207 SERIAL COMMUNICATIONS ........................................................... 208 PARALLEL COMMUNICATIONS ....................................................... 209 PACKET NETWORK COMMUNICATIONS .......................................... 209 15.3 UNDERSTANDING POSTSCRIPT LANGUAGE ERRORS........................ 209 ERROR: UNDEFINED ...................................................................... 210 ERROR: TYPECHECK ...................................................................... 211 15.4 REDEFINING BUILT-IN OPERATORS .................................................. 212 15.5 STACK TRACES ................................................................................. 214 15.6 INTERACTIVE TECHNIQUES .............................................................. 214 15.7 COORDINATE SYSTEM TRANSFORMATIONS.................................... 215 15.8 DEBUGGING MESSAGES................................................................... 216

appendix A ERROR HANDLER ................................................................ 217 INDEX.................................................................................. 221

PREFACE
ABOUT THIS BOOK
PostScript Language Program Design is intended to provide a solid background for developing software in the PostScript languagenot just getting a program to work, but thoroughly designing it from top to bottom. The book is organized into fteen chapters, each of which addresses a specic aspect of program design or problem-solving. The ideas in this book are targeted for sophisticated software development, although they should be useful at any level. The goal of this book is to teach the fundamentals of designing PostScript programs and to show how the language works, so that your programs will be fast, well-behaved, easy to understand, and portable. The PostScript language is somewhat unusual in the realm of programming languages in that it is interpreted and stack-based. Learning to program effectively in the PostScript language is not as easy as just learning another procedural high-level language. This book should help you learn how to think in the language as well as provide pragmatic advice and examples for accomplishing specic tasks. The sample programs contained in each chapter are intended to be directly usable. They are examples, but they are not trivial ones. Each program has been carefully designed and debugged and should provide an excellent foundation for an industrialstrength printer driver. Feel free to use the codethat is why it is there. This volume is a companion volume to two other books by Adobe Systems Incorporated: PostScript Language Reference Manual and PostScript Language Tutorial and Cookbook, both published by and available from the Addison-Wesley Publishing Company, Inc. We recommend that you have at least the Reference Manual handy when writing programs. If you have not

xi

been exposed to the language before, the Tutorial and Cookbook is a good place to start.

PRODUCTION NOTE
Glenn Reid, of Adobe Systems Incorporated, is the principal author of this book. Glenn conceived and executed the text, examples, and illustrations, with feedback from many people at Adobe. John Deubert helped with preliminary outlines of the book and supplied the rst draft of Chapter 8. Doug Brotz, Ed Taft, Ann Robinson, Ming Lau, and many others proofread the manuscript, found bugs, and offered valuable advice. Thanks to Carole Alden of Addison-Wesley as editor for providing guidance and encouragement. Special thanks to Pat Marriott, who was a paragon of patience and wisdom throughout the project, and to Linda Gass, who made it all possible. The book was produced using Frame Maker 1.1 from Frame Technologies, Inc., San Jose, California. It was composed and written on a Sun Microsystems 3/50 workstation running Sun Unix. All illustrations were produced in the PostScript language directly with the Adobe Illustrator software package on an Apple Macintosh Plus computer and imported into Frame Maker for nal printing. Proofs were printed on a Digital Equipment Corporation ScriptPrinter and nal camera-ready copy was set on a Linotype L-300 typesetter at Adobe Systems.

PROGRAMMING STYLE AND NOTATION

Programming style is an important part of programming in any language. The PostScript language has a very unstructured syntax and can be somewhat difcult to read, depending on how it is written. For software development, there are three goals for programming style: the program text should be readable (meaning that the structure and execution of the le should be easy to follow), it should be easily edited, and it should be consistent (especially when more than one programmer is working on the same project). Many reasonable approaches are possible. The style chosen for the listings in this book is designed to be as readable and

xii

maintainable as possible without requiring too much white space, which tends to make the listings long and more difcult to read.

PROCEDURE BODIES
Procedure bodies are probably the most common construct in any program written in the PostScript language. Procedure bodies are delimited by braces, also frequently called curly braces. It is important to be able to determine at a glance where the beginning and end of a procedure body are found. PostScript procedures can be used as arguments for many different operators, including loop, forall, kshow, image, ifelse, def, and others. The notation used for procedures in this book uses a simple PostScript language comment to indicate the operator for which the procedure body is an operand: usertime dup 1000 add { %loop dup usertime lt { pop exit } if } bind loop Short procedure bodies that t entirely on one line do not use this comment convention, since their use is more obvious.

INDENTATION
In general, any construct that has a clear beginning and end has its contents indented, including gsave and grestore, begin and end, procedure brace delimiters, and so on: /$arithdict 12 dict def $arithdict begin /avg { %def add 2 div } bind def /graph { %def gsave 0 0 moveto 1 3 avg lineto 2 setlinecap stroke grestore } bind def end % $arithdict

xiii

The horizontal alignment (indentation) of the beginning of the line containing an open-procedure brace delimiter matches the horizontal alignment of the close brace: currentpoint exch 500 gt { %ifelse 72 exch 12 sub moveto }{ %else pop } ifelse The slightly unusual technique of placing brace delimiters on a line by themselves is to make it easier to edit. You can add lines to or subtract lines from procedure bodies without worrying about the delimiters becoming unbalanced. There are, of course, other styles that work well. This one has been adopted throughout this text so that it is at least consistent, which is the last of the three goals.

THE NAME POSTSCRIPT

The name PostScript is a registered trademark of Adobe Systems Incorporated. This book, PostScript Language Program Design, is concerned with programs written in the PostScript language. All instances of the name PostScript in the text are references to the PostScript language as dened by Adobe Systems Incorporated, unless otherwise stated. The name PostScript also is used as a product trademark for Adobe Systems implementation of the PostScript language interpreter. Any references to a PostScript printer, a PostScript le, or a PostScript driver, refer to printers, les and driver programs (respectively) which are written in or support the PostScript language. The sentences in this book that use PostScript language as an adjective phrase are so constructed to reinforce that the name refers to the standard language denition as set forth by Adobe.

xiv

chapter 1 THE POSTSCRIPT LANGUAGE: OVERVIEW

Things should be made as simple as possible, but no simpler. Albert Einstein

1.1 INTRODUCTION
PostScript is the name of the computer programming language developed originally by Adobe Systems Incorporated to communicate high-level graphic information to digital laser printers. It is a exible, compact, and powerful language both for expressing graphic images and for performing general programming tasks. As is true with many programming languages, the PostScript language has been designed for a specic purposeto express complex digital graphics in a device-independent manner. Powerful typesetting features are built into the language for sophisticated handling of letterforms as graphics. The PostScript programming language is an easy one to learn, and graphics programs may be written by hand to produce high quality text and images. However, the language is intended for machine generation. That is, PostScript language programs are generally produced by other software rather than by programmers. This chapter is an overview of some of the more important features of the PostScript language. It is intended to provide a basis for understanding the why behind the programming examples given in the rest of the text. The full specication of the PostScript programming language can be found in a companion volume, PostScript Language Reference Manual (Addison-Wesley).

chapter 1 THE POSTSCRIPT LANGUAGE: OVERVIEW

It is assumed that the reader of this book is a programmer or has some experience with programming languages. Previous experience with the PostScript language is recommended, but not required. The book is targeted at the specic tasks that need to be performed to implement a driver for any device containing a PostScript interpreter. The examples in the book are intended to be cut and paste routines. They are optimized for efciency wherever possible, and should be appropriate for direct inclusion in a product-level driver. Rather than making the examples simple and relying on the text to illustrate the point, the examples show what you should do, and the text attempts to explain why. In theory, at least, it should be possible to paste all the examples in the book together without necessarily understanding how they work and have a very functional program when you are through. In practice, of course, it is best to understand why an example is written the way it is, so that you can adapt the programming style to your own implementation. Much of the material contained in this book is useful to developers of drivers for devices containing PostScript interpreters. A printer driver is typically the part of a software application that produces output specically intended for a printer. A printer driver for a device with a PostScript interpreter simply produces a PostScript language page description as output, rather than supplying escape sequences for a particular printing device. These page descriptions are partly generated by the application software and partially written by hand. This book is dedicated to the process of design which must go into producing high-quality PostScript language software.

1.2 THE LANGUAGE MODEL

The three most important aspects of the PostScript programming language are that it is interpreted, that it is stack-based, and that it uses a unique data structure called a dictionary. The dictionary mechanism gives the PostScript language a exible, extensible base, and the fact that the language is interpreted and uses a stack model means that programs can be of arbitrary length and complexity. Since very little overhead is necessary to execute the programs, they can be interpreted directly from the

chapter 1 THE POSTSCRIPT LANGUAGE: OVERVIEW

input stream, which means that no memory restriction is placed on a PostScript language program other than memory allocated by the program itself. The PostScript language is based heavily on context, or state. This context dependency is very powerfulby modifying the coordinate system transformation, any page can be easily scaled, rotated, or translated. The program executes in the context established for it beforehand. By redening the context of the dictionary stack mechanism, built-in PostScript operators can easily be replaced by procedures with additional functionality. Correctly using the context provided by the PostScript imaging model is a challenging aspect of learning the language, since it behaves much differently than most other programming languages in this respect. The PostScript language is designed to be interpreted, not compiled. User programs do not execute directly on the CPUthey are read and interpreted by another program, the PostScript interpreter. The PostScript language is a high-level programming language and it can be used to implement algorithms and process data, but a careful understanding is necessary to use it effectively. The difference in execution speed between a poorly written C program and well written one may be a factor of two or three at best, if the code is improved dramatically. More often, improvements of 10 percent are considered quite good. In general, speed improvements to software written in compiled languages come from improving the algorithms used, not from changing ones programming style. Simply improving the style of a poorly written PostScript language program can yield substantial execution speed improvements, frequently by a factor of ten or more. The nature of the interpreted language model and the execution semantics require more attention to detail than most compiled languages. Learning to use the PostScript language well can have an immediate and dramatic effect on the efciency and reliability of your software.

chapter 1 THE POSTSCRIPT LANGUAGE: OVERVIEW

DICTIONARIES AND DATA STRUCTURES

The PostScript language provides a unique data structure known as a dictionary. A dictionary is a structure in which are stored key value pairs, where the key and the value can be any valid PostScript language objects. Presenting a key (usually a name object) to the PostScript interpreter will cause it to perform a name lookup, and execute the resulting object found in the dictionary. This mechanism provides an extremely exible foundation for the language. New procedures can easily be created, and existing operator names can be redened. The PostScript language also provides several other standard data structures: strings, arrays (of any PostScript objects, including other arrays), integers, reals, names (analogous to identiers in other languages), and les (with standard le operations dened). All of these data types are maintained as PostScript objects, which are manipulated on the operand stack. A dictionary can contain any of these data types as a value. The notion of a variable usually means a dictionary entry in the PostScript language. A PostScript dictionary is the only way to store PostScript objects for later recall (other than leaving a reference to them on the operand stack). The standard operator for making a dictionary entry is the def operator, which makes a denition in the current dictionary. This mechanism is used for conventional purposes such as dening and using variables or procedures, but it can also be used for more complex data structures.

STACKS
The PostScript language is based on stacks. A stack is a xed data structure in which objects are placed temporarily, one on top of another. By nature, stacks are last in, rst out data structures. If you put three items on a stack, the third one is on the top, (and is the rst item available). Operators in the PostScript language are preceded by their operands. The operands are placed on the operand stack as they are encountered by the interpreter, and when the operator itself is encountered and executed, its arguments are found on the stack. This is often referred to as post-x notation.

chapter 1 THE POSTSCRIPT LANGUAGE: OVERVIEW

There are several stacks used by the PostScript interpreteran operand stack, an execution stack, a dictionary stack, and a graphics state stack. The operand stack is in constant use, while the others are more specialized. The operand stack is frequently referred to simply as the stack. Programming in the PostScript language is primarily an exercise in stack manipulation. The most common PostScript errors result from incorrect manipulation of the stack, and many optimization strategies consist of using the stack as efciently as possible. Further discussion of stack mechanics is provided in Chapter 2. It is worthwhile to become quite familiar with the mechanics of stack operations in the PostScript language.

BUILT-IN POSTSCRIPT LANGUAGE OPERATORS

The standard PostScript interpreter has more than 250 built-in operators. It is an extremely rich language. The nature of the PostScript languages execution model makes it very easy to add new procedures that behave like operators, since the interface between them is just the operand stack. It is difcult to learn how to use 262 operators at rst. It is even harder to remember all of them. However, using the appropriate operator for the task at hand can be very important in terms of performance and program design As a rule of thumb, keep the following in mind: If you are having difculty with some aspect of your program, it is likely that the problem has been encountered before. There may be a specic PostScript language operator that directly addresses the problem. It is a good idea to briey review a list of operators grouped by function when you encounter problems. (See chapter 6 of the PostScript Language Reference Manual.)

1.3 THE IMAGING MODEL

The imaging model of the PostScript language includes standard operations for typesetting and rendering graphics. The model is device- and resolution-independent, and provides an abstrac-

chapter 1 THE POSTSCRIPT LANGUAGE: OVERVIEW

tion for raster imaging that frees application software from having to make device-specic rendering decisions. An important aspect of using the PostScript language correctly is to make proper use of the imaging model. The purpose of the PostScript language is to provide a uniform way to represent visual elements on any raster device. An application program typically needs to communicate visually with the user in terms of representing text, graphics, and aspects of the user interface. Whatever internal representation or data structure is used, some imaging model is necessary when presenting the information visually. When printing on a device with a PostScript interpreter, some translation is usually necessary between the applications internal representation for graphics and the PostScript languages imaging model, unless the screen display is also driven by a PostScript interpreter. The PostScript language provides both an execution model and an imaging model to applications programs. An important consideration is to adapt naturally to the imaging model when producing output for the PostScript interpreter, rather than using the programming power of the language to emulate another graphic imaging model.

COORDINATE SYSTEMS
The user coordinate system in the PostScript imaging model provides a consistent and predictable environment for constructing graphics. This coordinate system, referred to as user space, is the same for all printers with PostScript interpretersit is only the mapping from user space to device space that is different from one raster device to another, depending on the resolution and orientation of the device. To produce truly device-independent programs with the PostScript language, it is important to think only in terms of the user space abstraction, and never about any particular output device. Similarly, page layout and sizes should be expressed only in terms of where they fall in user space. The coordinate system transformation from user space into device space can be modied easily to scale an entire page when it is actually ren-

chapter 1 THE POSTSCRIPT LANGUAGE: OVERVIEW

dered, and it is good not to make any specic assumptions at page composition time about the size of the nal page.

PATHS AND PAINT

Graphics in the PostScript language are rendered by constructing algorithmic paths consisting of lines, arcs of circles, or Bezir curves. These paths are then painted with a current color or grayscale (the default color is black). Paths may also be used for clipping or masking graphics, and they may be either stroked with a line, lled as regions, or both. Paint is applied to the current page as opaque color, completely painting over anything that may be beneath it on the current page. There is a more detailed discussion of this in Chapter 3.

FONTS
The PostScript language has a completely integrated model for quality typesetting. PostScript fonts are executable programs that draw the character shapes using the same path construction and painting mechanism as all other graphics. The text can be thought of as integrated graphics, and can be transformed in any way that graphics can be. There is extensive support in the PostScript language (beyond the standard graphics commands) for manipulating font programs and for setting text characters. Fonts may be created from existing graphic shapes. Once a program or procedure has been devised that renders a particular shape, that code can be packaged into a font dictionary and placed on the page through the text-handling mechanisms. It is the representation of typefaces as graphic descriptions that realizes the full integration of text and graphics on a single page. It is not that the text alone may be scaled and rotated that is signicantit is that the entire page is completely integrated and may be treated as a single graphic entity, and the page itself may be printed half-size or rotated to any angle.

1.4 ELECTRONIC PUBLISHING AND PRINTING

Electronic publishing is the business of composing and printing documents, from books to magazines and even newsletters.

chapter 1 THE POSTSCRIPT LANGUAGE: OVERVIEW

Computers are now used extensively to simplify the editing, composition, and typesetting of books and documents. The PostScript language is a system solution to the perplexing problem of composing, proong, and printing documents. It is intended chiey for high quality typesetting, graphic design, and book production. However it is also used for more standard computer printing, including everything from program listings to business correspondence using typewriter-like typefaces. There are many, many different levels at which a computer environment might support PostScript technology. A PostScript printer may be used to get hard copy of something on the computer screen, it may be used to print text in much the same way that a line printer does, or the PostScript language typesetting and graphic models may be used fully in the preparation of camera-ready output for major book production. There is a great difference between printed output intended as a nal product and simply a hard copy of some information that already exists in some other form. When designing software, it is useful to consider what the nal product will be. If the program is a word processor, for instance, is the nal product considered to be the words themselves, their arrangement, the typeface that they are set in, or the book in which they will eventually appear? To enter even slightly into the level of presentation means a heavy commitment to the nal form. For example, if the arrangement of words is part of the product, then it becomes a visual product, and most likely a printed visual product. When designing software for printing and publishing, the nalform technology must be considered. The graphic imaging model used by the printing technology, the available typefaces, the units of measure, and the available paper sizes are all part of the nal product. A common problem in bridging from traditional computer printing into the realm of electronic publishing is to muddy the waters between composing the document and printing it. For example, when printing on PostScript devices, one must compose text using the character widths of the fonts that will ulti-

chapter 1 THE POSTSCRIPT LANGUAGE: OVERVIEW

mately be used on the printer. This requires some advance knowledge of the printing technology. But it is the abstraction that is used, and not a particular output device. The resolution and printing characteristics of the device should not be propagated into the document designonly the ideals of the layout.

1.5 PROGRAM DESIGN GUIDELINES

There are a few items that may be kept in mind while implementing a driver for a PostScript device. As with most software development, the most difcult part of writing programs in the PostScript language is the design of the program. If the design is good, implementing it is easy. If the design is poor, it may not even be possible to correctly implement it. Below are some helpful items to keep in mind when writing your software. All of them are explained more fully within the text of this book; this is only an attempt to prime the pump before you start reading: Use the operand stack efciently. Pay particular attention to the order of the elements on the stack and how they are used. Avoid unnecessary redundancy. When a program is produced, check for many repetitive steps that perhaps could be condensed. Keep identiers short if they are to be transmitted many times. Use the PostScript imaging model effectively. When printing, a document must be translated into the language of the printer. This includes a philosophical adjustment to the nature of the PostScript imaging model. It is better to translate into the PostScript imaging model than to maintain another set of graphics primitives using the PostScript language for emulation.

chapter 1 THE POSTSCRIPT LANGUAGE: OVERVIEW

chapter 2 THE EXECUTION MODEL

It is more of a job to interpret the interpretations than to interpret the things, and there are more books about books than about any other subject.... Michel Eyquem de Montaigne

2.1 INTRODUCTION
The manner in which the PostScript interpreter executes a program has a profound effect on the efciency and design of software. This chapter takes a close look at the execution semantics and details of the PostScript interpreter, with the idea that a full understanding of how a program is being executed will provide a solid foundation for good programming skills. Probably the most important concept to glean from this chapter is the notion of PostScript language objects and how they are stored and manipulated. Understanding the mechanics of the scanner and interpreter will also help shed light on issues of efciency.

2.2 THE PRINTING JOB MODEL

PostScript interpreters are most frequently used in batch mode. That is, a document is prepared by applications software and is then submitted to a PostScript printer for interpretation. A document may consist of any number of pages and any amount of text and graphical information, since interpretation is carried out directly as the le is received by the interpreter. For example, it is typical to start with a page composition program that permits users to construct pages, graphics, and text interactively. The PostScript language may be used for display purposes during the document composition, but it is likely to be an interactive session. When the document is to be printed, the

chapter 2 THE EXECUTION MODEL

11

application invokes a particular section of code usually known as a printer driver to produce the output for the printing device. This page description must then be transmitted to the printer as a batch job, usually by an intermediate communications filter. The print job is then scanned and interpreted by the PostScript interpreter for that printer, and the resulting pages are printed. This model of execution is mostly invisible when one creates and executes PostScript language programs. It is presented here as a framework of reference, since it is the outermost level of interpretation in a device with a PostScript interpreter. A PostScript interpreter (in batch mode) executes a single job from the time it receives the initial byte over a communications channel through the next end-of-le (EOF) indication. All legal PostScript language sequences are interpreted and executed by the interpreter until the EOF is reached. At the end of each job, the interpreter resets the state which existed at the beginning of the job. This is actually accomplished by the save and restore operators. The interpreter then returns to an idle state until the next job is received. This behavior is discussed in more detail in Section 2.9, THE SERVER LOOP. This job execution model has two important consequences: The PostScript interpreter is completely restored to its original state after each job nishes execution. Each print job must be a contiguous stream of bytes. Anything that is to be part of a particular document must be present in the print le. The main advantage of this mechanism is that it guarantees certain initial state conditions to all print jobs, regardless of whether the job is the rst or the forty-rst one interpreted since the machine was powered on. A print job should always assume that the initial state of the PostScript interpreter is the default state, which is dened in Section 6.3, MODULARITY AND PAGE STRUCTURE. This initial state is guaranteed by the job server loop. For reasons that will become apparent, it is important to trust the initial state of the interpreter. A common practice is to initialize the interpreter at the beginning of each job, in an

12

chapter 2 THE EXECUTION MODEL

attempt to guarantee some execution state for the job. In most instances, this accomplishes nothing. (The state has already been initialized by the server loop). In instances where the state is not the default state, there is likely to be a good reason for it. For example, the transformation matrix may have been modied to print the page as a small illustration rather than as a full-size page. Initializing the matrix will reverse this effect. For a more involved discussion of this principle, see Section 6.3 and most of Chapter 11.

2.3 THE OPERAND STACK

The operand stack is, in a sense, the global communications area for execution and parameter passing in the PostScript language. All data objects that are to be operated on by any procedure or operator must be present on the operand stack. Objects are placed on the operand stack by the PostScript interpreter. As a rule of thumb, any object that is not directly executable is placed on the operand stack. There are a few special exceptions to this, which are detailed throughout this chapter.

OBJECTS IN THE POSTSCRIPT LANGUAGE

An object is actually a unied representation used for all entities in the PostScript language. Each object is a xed-size package (in many implementations it is 8 bytes, for example). Since these objects are all the same size, they are easily manipulated on the operand stack, placed into dictionaries, and passed through the internal mechanisms of the PostScript interpreter. There are two distinct kinds of objects, simple and composite. A simple object is one whose value can be stored within the 8byte packet that represents the object. These include the data types integer, mark, real, boolean, name, and null. Each of these data types can be represented in a self-contained object. A composite object is one whose value is stored elsewhere in memory. The object itself contains a pointer to the value, which in turn can consist of other objects. Only the data types string, dictionary, and array are composite objects. A procedure body is actually just an executable array, so procedure bodies also qualify as composite objects.

chapter 2 THE EXECUTION MODEL

13

To understand the distinction between a simple and a composite object, consider what happens when the dup operator is executed. The dup operator makes a copy of (duplicates) the top entry on the operand stack. Since the operand stack contains only objects, the result of executing dup is to duplicate the object itself. If the topmost object on the operand stack is a simple object, dup produces a new object with a copy of the original value, since the object is completely self-contained. With a composite object, however, the new object produced by dup contains a pointer to the same value as the original. The value of the object is stored elsewhere in memory, and is not duplicated. This means that any changes made to the value of the duplicated copy are also made to the original. You can obtain a new, distinct copy by allocating a new composite object (with the dict, string, or array operators, for example) and using the copy operator to copy the original objects value to the new one. Another way to think of it is that the value stored with an object either ts into the object (simple) or it doesnt. If it doesnt t, then a pointer to the value is stored instead (composite). See Figure 2.1.

gure 2.1
duplicating a simple object: 9 dup duplicating a composite object: (PostScript) dup

()

PostScript

9 operand stack

()

operand stack

14

chapter 2 THE EXECUTION MODEL

Each object has several attributes associated with it. These include a literal or executable ag, an access attribute, which controls read-write-execute permissions for the object, and a length attribute (used only with composite objects). There is also a type associated with every object as one of its attributes. Some of these attributes are easily changed with the following operators: Attributes literal/executable access Operators cvlit, cvx readonly executeonly noaccess cvi, cvr, cvrs cvs, cvn

type

For the most part, there is no need to convert objects from one type to another or to modify their attributes. However, it is good to remember that PostScript language objects are typed and maintained internally with this representation.

THE STACK AS A DATA STRUCTURE

The operand stack is often overlooked as a data structure for program development. In fact, it is the fastest, most direct, cheapest, and easiest-to-use data structure available in the PostScript language. The techniques for using the stack effectively are many. Almost without exception, the sample programs in this book use the operand stack directly instead of dening local storage. Since all objects are initially constructed on the operand stack when they are read from the input stream, the most natural approach is to use them directly from the stack. Resist the temptation to dene variables for procedure operands, unless those values are needed more than once or twice. This is related to the fundamental design issues of how and when to pass values on the stack, when to use the graphics state stack, and when to use global variables. For further discussion on program organization, please see Chapter 5. Objects that are placed on the operand stack should be in a natural order appropriate for their intended use. To understand this,

chapter 2 THE EXECUTION MODEL

15

consider the two (equivalent) PostScript language fragments given in listing 2-1; each takes the same values on the operand stack, but each in a different order. Compare the procedures and how they use the data.

listing 2-1
%!PS-Adobe-2.0 %%Title: stack example /procGOOD { %def findfont exch scalefont setfont } bind def /procOK { %def exch findfont exch scalefont setfont } bind def /procBAD { %def /ptsize exch def /fontname exch def fontname findfont ptsize scalefont setfont } bind def %%EndProlog 12 /StoneSerif procGOOD /StoneSerif 12 procOK /StoneSerif 12 procBAD %%Trailer

The procedure calls themselves look much the same. The data passed is identical in each case. The difference between the procGOOD and procOK procedures is only a single exch instruction. However, if the data is supplied in the natural order, the exch is unnecessary. The only difference between procGOOD and procBAD is that the operand stack is used naturally in the good example, whereas the operands are dened into a dictionary (and subsequently retrieved from the dictionary and put back on the operand stack) in the bad example.

2.4 THE DICTIONARY STACK

The dictionary stack is a very special feature of the PostScript language. It is the context in which name lookup and denitions

16

chapter 2 THE EXECUTION MODEL

occur. It can be manipulated directly, but it is best to think of it as an environment, not a data structure. The dictionary stack contains objects, just as the operand stack does. However, the only type of object which can be placed on the dictionary stack is, naturally, a dictionary object.

DICTIONARY OBJECTS
A dictionary is a composite object, much like an array. It is represented by a dictionary object, and it is always referenced through that object. The object can be placed onto one of the stacks, or it can be stored in a dictionary, just like any other object. However, certain special operations can be performed only with dictionaries. In particular, entries in a dictionary are ordered pairs. Each entry has a key and a value (actually an object, not to be confused with the value of the object itself). Either of these objects can itself be a composite object, including a dictionary. Dictionaries are the foundation of the storage and retrieval mechanism in the PostScript language. Data can actually exist in only one of two places: On the operand stack (or as a component of a composite object on the operand stack) In a dictionary (or in a component of a dictionary, such as an array). The reason is that unless an object is on one of the stacks, the only way to call it up is by name. If you dont store something explicitly into a dictionary (with def, for instance) and the object is removed from the operand stack, it cannot be recovered. That alone is not quite reason enough to place something in a dictionary, but it is something to be aware of. Dictionary objects can be created only by the dict operator. The dict operator requires an integer value as its argument, and it creates an empty dictionary object with that number of elements allocated for later use. The dictionary object is returned on the operand stack. For example, the dictionary created by the sequence 3 dict is just a composite object. It is not the current dictionary, nor does it have a name. It is just an empty data

chapter 2 THE EXECUTION MODEL

17

structure on the operand stack, which (in this instance) can contain up to three entries.

USING THE DICTIONARY STACK

The dictionary stack starts out with two entries. These are known as systemdict and userdict. These dictionaries cannot be removed from the dictionary stack. The systemdict dictionary contains predened names for all the standard PostScript language operators. The other dictionary, userdict, is a writable dictionary that typically has about 200 elements available. userdict is on the top of the dictionary stack when each job begins, and systemdict is right below it. The dictionaries on the dictionary stack are the context in which name lookup takes place. You can use this mechanism to redene built-in PostScript language operators or to dene new procedures, as appropriate. The name-lookup mechanism is described in Section 2.5. Other dictionary objects can be placed onto the dictionary stack with the begin operator. It is called begin because it affects the context in which dictionary operations take place. Its effect is to place a dictionary object on top of the dictionary stack, making it the current dictionary. The current dictionary is simply the one on top of the dictionary stack. The current dictionary is where denitions are placed when the def operator is executed. It is also the rst place the interpreter looks when a name lookup takes place. Lets look at a PostScript language sequence that creates and uses a dictionary: 3 dict begin /proc1 { pop } def /two 2 def /three (trois) def currentdict end This example creates a dictionary with three available slots, places it on the dictionary stack (with begin, making it the current dictionary), and then makes three denitions (with def). The denitions are placed into the newly created dictionary only because it was made the current dictionary, not because it was recently created. The call to currentdict makes a copy of

18

chapter 2 THE EXECUTION MODEL

the dictionary object (the one currently on top of the dictionary stack) and places the copy on the operand stack. Notice that the dictionary object that was initially created was also on the operand stack until begin moved it to the dictionary stack. The end operator simply pops one element off the dictionary stack (and discards it). If currentdict had not been executed before end, the dictionary (along with its three entries) would have been lost. This is because the object created by the 3 dict statement was never stored into a dictionaryit was simply used by the begin operator. Here is an equivalent (though substantively different) PostScript language fragment: 3 dict dup /proc1 { pop } put dup /two 2 put dup /three (trois) put This examples uses put instead of def. It never places the dictionary onto the dictionary stack, and it makes each entry explicitly by presenting the dictionary object to the put operator each time. (The dup is used to avoid losing the object when put is performed.) However, the resulting dictionary on the operand stack is identical to the one created in the previous example. The dictionary stack is an environment. It is most important as a context for name lookup, which is discussed in depth in the next section. The previous discussion, although cumbersome, is meant to illustrate two concepts: Dictionaries are represented as objects, which can either be left on one of the stacks or given a name and dened as an entry in another dictionary. Dictionary objects do not necessarily have names and are much like arrays in their use. They are special mostly in the context of name lookup.

2.5 OPERATORS AND NAME LOOKUP

All built-in PostScript language operators are stored in dictionaries as executable objects of type operatortype. The keys under

chapter 2 THE EXECUTION MODEL

19

which these operators are stored are the names by which we think of them: moveto, showpage, or add, for example. The names themselves are not special in any way. They are used simply as keys for dictionary entries. (For the most part, they are keys in systemdict.) The actual execution is carried out after these names have been looked up and an executable object has been found in systemdict. This mechanism is known as automatic name lookup. Name lookup is invoked whenever the PostScript interpreter encounters an executable name object. The difference between a literal name and an executable name in the PostScript language is the presence (or absence) of a leading slash character: /moveto moveto % literal name % executable name

When the PostScript interpreter encounters an executable name, it looks up the name in the current dictionary context and interprets whatever it nds as a result of that name lookup.

gure 2.2

3 dict begin

userdict

systemdict

dictionary stack

Name lookup is done from the top down, in the context of the current dictionary stack. When a name is encountered

20

chapter 2 THE EXECUTION MODEL

(whether it is in the original input le or found in a procedure body executed later) it is looked up in all dictionaries that are currently on the dictionary stack. (See gure 2.2.) If the interpreter nds an executable object (like a procedure body) or a built-in PostScript language operator, it executes it directly, rather than placing it on the operand stack. This is how both operators and procedures are executed: They are just looked up in a dictionary as names, and the object found in the dictionary is interpreted further.

THE BIND OPERATOR

The bind operator is used to perform early name binding on procedure bodies. What this means, effectively, is that it is possible to look up all the names in a procedure ahead of time and, if a real operator is found (as opposed to an array, a procedure, or any other type of object), to replace the executable name with the operator itself. This is a very important concept in making PostScript programs more efcient. The resulting procedure body (after bind is applied) may actually have its contents modied. Here is a sample use: %!PS-Adobe-2.0 /F { %def findfont exch scalefont setfont } bind def %%EndProlog 12 /StoneInformal F When bind is executed, there are two things on the operand stack: a literal name, /F, and a procedure body, { findfont exch scalefont setfont }. The bind operator returns an array object on the stack when it is nished. In this case, however, since none of the operators in the procedure have previously been redened, bind nds real operators (objects of type /operatortype) for every name lookup except findfont, which is actually a procedure. The result looks something like this: { findfont --exch-- --scalefont-- --setfont-- } The --setfont-- notation with the hyphens is a convention used by the == operator to indicate an object of type /operatortype

chapter 2 THE EXECUTION MODEL

21

(as distinguished from an executable name). It means that the name objects have been removed and replaced by operator objects. The array as it is stored in memory actually looks like gure 2.3:

gure 2.3
value: findfont type: nametype executable: executable value: --exch-type: operatortype executable: executable value: --scalefont-type: operatortype executable: executable value: --setfont-type: operatortype executable: executable

value: type: arraytype executable: executable length: 4 value: F type: nametype executable: literal

operand stack

procedure body (in memory)

The procedure body is actually an array object with its executable ag set, and the contents of the array are, in turn, other objects. After bind is applied, the original name objects are replaced by their respective operator objects, as you can see in the gure. (See Section 2.7 for more detailed information about procedure bodies and how they are constructed.) It is useful to execute bind on any procedure body that will be used many times. The result of binding a procedure body is that the name-lookup operations can be distilled out, requiring less time to execute the procedure. The semantics of bind dictate that if an operator has been redened prior to being used in a procedure, then bind leaves the name in the procedure body and will not replace it with the operator object. /showpage { } def /ENDPAGE { %def gsave showpage % name is not replaced grestore } bind def

22

chapter 2 THE EXECUTION MODEL

/ENDPAGE { %def gsave showpage % name is replaced by operator grestore } bind def /showpage { } def % has no effect in ENDPAGE proc Because bind replaces only those items that are found to be of type operatortype, it will not replace an object that is of type arraytype (which is the case when an operator name is redened with a user procedure). This means simply that it is all right to bind procedures even if some of their elements might be redened, as long as the denitions are done before the procedures are called.

2.6 THE INTERPRETER AND THE SCANNER

The PostScript interpreter deals only with objects. A separate entity exists between the interpreter and the standard input stream known as the scanner. The scanner actually converts incoming bytes into objects and hands them to the interpreter. It is invoked directly by the interpreter whenever more information is needed from the input stream. (Note that the scanner does not pre-scan any of the incoming bytes, since they could be used as data and not read by the interpreter.) The scanner actually tokenizes the input stream into PostScript objects. This includes conversion of names into name objects and strings into string objects. It is the scanner that makes the distinction between a literal and an executable name by recognizing the leading slash character (/). Thus the distinction between /findfont and findfont is made by the scanner, which sets the literal or executable ag on the resulting name object. The scanner recognizes strings (with either the (string) or <string> notation) and converts them into string objects. The scanner also recognizes the backslash notation in string objects (including \\, \(, \), and \octal notation). For example, if it encounters a sequence like the following: (\101B\103D \\\(ABCD\) (balanced))

chapter 2 THE EXECUTION MODEL

23

the scanner will recognize and replace the backslash notation with the appropriate characters and it will produce a string object as follows. The parentheses are omitted because they are not part of the string, but indications to the scanner of where the string starts and end (although this string contains some parentheses as data, as you can see): ABCD \(ABCD) (balanced) The scanner also is responsible for issuing the PostScript language syntaxerror, which can only result from mismatched (unbalanced) string delimiters and unexpected ends-of-le. There are no other syntax errors possible in the PostScript language! The PostScript interpreter invokes the scanner each time it needs a token from the object currently being executed. The scanner reads from a le or string with exactly the same semantics as the token operator (see the PostScript Language Reference Manual), returning exactly one object to the interpreter.

RECOGNITION OF OBJECTS
It is the scanners responsibility to recognize objects. There are only four kinds of objects that it knows how to recognize: Literal strings: any string object delimited by either < > or ( ). The scanner watches for \ notation and attempts to nd the closing delimiter. Mismatched delimiters result in syntaxerror. Numbers: the scanner tries rst to make a number out of a token. This includes reals, integers, and radix notation. Names: anything that cannot be interpreted as one of the above is returned as a name object. The literal/executable ag is set depending on the presence of a leading slash (/). Procedure bodies: Procedure bodies delimited by curly braces { } are recognized by the scanner and returned to the interpreter as executable array objects.

24

chapter 2 THE EXECUTION MODEL

There are a few special cases. For instance, the leading slash (indicating a literal name) overrides the scanners ability to recognize a token as a number: /1000 type == nametype It is worth noting that any type of object can be used as a key in a dictionary. Names are used most often, but they are not required. The following is a perfectly legal, although somewhat confusing, PostScript language sequence (shown as though typed in interactive mode): PS> 13 (thirteen) def PS> 13 load == (thirteen) PS> It is necessary to use load in this example because the token 13 is recognized rst as an integer, not as a name. Therefore, automatic name lookup is not invoked. The load operator explicitly invokes key lookup for any object, in the context of the current dictionary stack. Note: When a name is encountered in a program, the scanner returns a name object. Name objects require a fixed amount of overhead, regardless of the number of bytes in the name. However, time required to recognize a name is proportional to its length. With this in mind, it is always best to use short procedure names if the procedure will be invoked many times (as in the script, but not in the prologuesee Section 6.2).

2.7 PROCEDURES
A PostScript language procedure is simply an executable array of PostScript objects. Normally, procedures are given a name and stored in a dictionary, so that they may be called up by presenting the appropriate executable name to the PostScript interpreter. However, procedures can exist without having names at all (as can any object until it is stored into a dictionary).

chapter 2 THE EXECUTION MODEL

25

A procedure that exists (perhaps on the operand stack) but is not referenced by name is typically referred to as a procedure body. Procedure bodies are usually stored in a dictionary under a particular name and used in the same manner as the built-in operators, although they are also required (or permitted) as arguments to several operators, including image, imagemask, setscreen, settransfer, if, ifelse, loop, and forall. The PostScript languages mechanism for building and dening a procedure is worth examining more closely: /F { %def findfont exch scalefont setfont } bind def This is a typical procedure denition. It builds a procedure body (in this case consisting mostly of operators), executes bind to distill out the name lookup, and then denes the procedure in the current dictionary under the name F. Lets look at this sequence as the PostScript interpreter would, token by token: /F The scanner digests this as a literal name object and hands it to the interpreter. The PostScript interpreter places all literal objects directly on the operand stack and moves on. { This is a special PostScript language delimiter (informally known as a curly brace, or open-curly). It marks the beginning of a procedure body. The scanner enters deferred execution mode when this character is encountered. All objects through the closing } token are scanned but not executed, regardless of their type. When the closing delimiter is found, an executable array is created containing all the collected objects between the { } delimiters. In most implementations, the scanner uses the operand stack to construct procedure bodies, placing a mark on the stack and then placing each object on the operand stack until the closing

26

chapter 2 THE EXECUTION MODEL

brace is found, at which point the array is constructed from the objects on the stack. This places a practical limitation on the number of objects (which may be composite objects) which can be put into a procedure. In most implementations, this number is 499, which is the depth of the operand stack less 1 for the mark (listing 2-3 contains a method for building much larger procedure bodies when necessary). Here is the state of the stack just before the } character is encountered, which causes the objects on the stack (down through the mark) to be collapsed into a single array object. setfont scalefont exch findfont --mark-/F

These are all objects of type /nametype, with the exception of the mark, which is of /marktype. The scanner performs no name lookups during the scanning of a procedure body. When the scanner recognizes the } delimiter it performs several functions. First, it decrements the level counter that keeps track of the nesting of curly braces, and if the level is back to 0, the interpreter reverts back to immediate execution mode (which is the normal state of the PostScript interpreter). Next, the scanner performs the equivalent of counttomark array astore: It allocates an array of the appropriate length, lls it with the objects on the stack down through the mark, removes the mark, and leaves a freshly created array object on the stack. The nal step is to make the new array an executable array. (It does this by performing the equivalent of cvx.) Here is the state of the operand stack right after the } character is processed:

chapter 2 THE EXECUTION MODEL

27

--executable array-/F

The name /F is still on the bottom of the operand stack, where it was at the beginning of this example. The entire procedure body is now stored in memory as a sequence of objects represented as an executable array object on the operand stack. Note that there is still no procedure named F. There are merely two objects on the operand stack. They could be eliminated very easily by executing the clear or pop operators, for example. It is not until the def operator is nally executed that this array object is safely stored in the current dictionary under the name F. The remaining part of the last line of the program is bind def These are two more executable names. Each of these is looked up and executed (since the interpreter is now in immediate execution mode). The effect of executing bind is to perform early name binding on any elements of the array object that may be executable operator names. (See Section 2.5, OPERATORS AND NAME LOOKUP, for a discussion of bind.) The bind operator expects to find an executable array on the operand stack, and it leaves one there (possibly containing different objects) when it is through. The effect of executing def, of course, is to take the top two elements from the operand stack and make a key-value pair association in the current dictionary. The F procedure is now dened. It can be used like any of the built-in operators in the PostScript language, as long as the dictionary in which it is stored is available on the dictionary stack.

listing 2-2
%!PS-Adobe-2.0 /F { %def findfont exch scalefont setfont } bind def %%EndProlog

28

chapter 2 THE EXECUTION MODEL

24 /Helvetica F 288 72 moveto (Green Book) show showpage

Listing 2-2 would result in something like this being printed on the page:

Green Book
VERY LARGE PROCEDURE BODIES
Occasionally one needs to dene a very large procedure body, perhaps as a character description in a font or as part of a logotype. The operand stack depth limit is typically about 500 elements, which limits the total size of a procedure created while in deferred execution mode, since all the elements in the procedure are rst placed on the operand stack. There are a few ways to cope with this problem. The easiest way, if the host application can keep track of how many elements are being generated, is to break out pieces of the procedure and make them separate procedures, invoked by name within the large procedure body: /small1 { one two three four } def /small2 { five six seven eight } def /small3 { nine ten eleven twelve } def /large { small1 small2 small3 } def If the procedure is being generated automatically, rather than by hand, this may not be a feasible approach. For one thing, the number of new (unique) names required cannot be predicted easily. In listing 2-3 is a method for collapsing a large procedure into many small component pieces.

chapter 2 THE EXECUTION MODEL

29

listing 2-3
%!PS-Adobe-2.0 %%Title: collapseproc.ps %%EndComments /EXEC { /exec load } bind def /mainproc [ %def { gsave 400 36 moveto } EXEC { /Times-Italic findfont 24 } EXEC { scalefont setfont (from Listing 2-3) } EXEC { show grestore } EXEC ] cvx bind def %%EndProlog mainproc showpage %%Trailer

Notice the use of exec in this example. It is necessary since the component procedure bodies will not be immediately executed when encountered directly (they are simply placed on the operand stack). Also, the relationship between building square bracket arrays and curly-brace procedure bodies is interesting (and crucial to this example). Since the outermost brackets are actually square brackets, all the intervening PostScript code is executed, not deferred. For this reason, in order to place the exec operator in the procedure body, its value must be obtained with load (see the EXEC procedure in the example). Here is what the /mainproc in listing 2-3 looks like after it is built (a little interactive dialog is shown to show the length of the array, too): PS> /mainproc load dup == { { gsave 400 36 moveto } --exec-- { /Times-Italic findfont 24 } --exec-- { scalefont setfont (from Listing 2-3) } --exec-- { show grestore } --exec-- } PS> length == 8 PS>

30

chapter 2 THE EXECUTION MODEL

Each procedure body within the square brackets is built individually, and when the } token is encountered, each one is collapsed into a single (executable) array object on the operand stack. This way the operand stack does not overflow, since it is effectively segmented into smaller pieces. An array of arrays is built, and each sub-array is explicitly executed by exec. The outermost array is simply made executable and can be used like any other procedure.

listing 2-4
%!PS-Adobe-2.0 %%Title: addproc.ps %%EndComments /F { findfont exch scalefont setfont } bind def /S { moveto show } bind def /addprocs { %def dup length 2 add array cvx dup 3 -1 roll 2 exch putinterval dup 0 4 -1 roll put dup 1 /exec load put } bind def %%EndProlog gsave currenttransfer { 1 exch sub } addprocs settransfer 36 /Helvetica-Oblique F 1 setgray % normally white... (Black) 36 670 S % but the inverse transfer (Words) 36 634 S % function yields black grestore showpage %%Trailer

Another application of this technique is a space-efcient way to concatenate two procedure bodies, as can be seen in listing 2-4. It denes a procedure called addprocs which concatenates two existing procedure bodies. It creates one new procedure body that is the length of the second procedure plus two. The array object representing the first procedure is inserted into the 0th position in the new array, the exec operator is placed in the sec-

chapter 2 THE EXECUTION MODEL

31

ond slot, and the second procedure is copied into the rest of the new procedure body, resulting in a new procedure like this: { one two three } { four five six } addprocs => { { one two three } exec four five six } This preserves at least the first of the procedures without having to copy it (it is simply executed with exec).

2.8 THE EXECUTION STACK

The PostScript interpreter always looks at the top of the execution stack when it has nished executing any single operator. The execution stack contains the object currently being executed, whether it is a procedure body, the standard input file, some other le object, or a string. During the normal execution of a print job, there is a le object on the execution stack that represents the standard input le, and as the program is executed, other procedure bodies may also be placed on top of the execution stack temporarily until they have been completely executed. Objects can explicitly be placed on the execution stack with the exec, run, or stopped, for, ifelse, and forall operators. Each of these causes an object to be executed directly. If the object is a file or string, it is tokenized rst by the scanner. If the object is an executable array (a procedure body), each object in the array is executed consecutively. If the object is a le, it is scanned and interpreted and the le object is popped from the execution stack when the end-of-le indication is reached. There is an innite loop on the bottom of the execution stack, known as the server loop. (See the next section for more details.) The current job is typically represented by a le object on top of the server loop. As the le is executed, other objects may be executed as well. For instance, a common occurrence is for a procedure to be dened and executed as in the following example (which is borrowed from listing 2-3):

32

chapter 2 THE EXECUTION MODEL

%!PS-Adobe-2.0 /F { %def findfont exch scalefont setfont } bind def %%EndProlog 24 /Helvetica F 0 0 moveto (Red Book) show showpage This program is represented by a le object on the execution stack that the scanner reads and interprets token by token. We have seen the execution of the scanner; now lets look at what happens when the procedure is executed. When the F token is scanned and interpreted from the le object on top of the execution stack, it is seen to be an executable name. This name is looked up in the current dictionary stack, and an executable array (procedure body) is found. This procedure body is then pushed onto the execution stack:

{ findfont --exch-- --scalefont-- --setfont--} -file{ --job server-- }

Remember that the array is represented by an array object, even when it is on the execution stack. Figure 2.4 shows what this array object looks like on the execution stack. The PostScript interpreter executes array objects one element at a time, leaving the unused part of the array on the execution stack. The interpreter always checks the remainder of the procedure body before it executes the rst element. If the procedure body is empty, the interpreter discards it. This permits innitely deep tail recursion, since the empty procedure bodies will not accumulate on the execution stack. (Tail recursion means that the recursive procedure call is the very last element (the tail) of the procedure body.)

chapter 2 THE EXECUTION MODEL

33

gure 2.4
value: --findfont-type: operatortype executable: executable value: --exch-type: operatortype executable: executable value: --scalefont-type: operatortype executable: executable value: --setfont-type: operatortype executable: executable

value: type: arraytype executable: executable length: 4 value: --filestream-type: filetype executable: executable

execution stack

procedure body (in memory)

2.9 THE SERVER LOOP

The execution stack starts out with a procedure that is an innite loop. The interpreter is always busy executing this loop, even when the interpreter is in an idle state. This loop is known as the server loop, and each time around the loop it checks to see if there is a new job to execute (if one of the communications ports has become active). When a new job is discovered, a le object representing the standard input file is opened and placed on the execution stack. This le object is executed until the end-of-le indication is reached, at which point it is popped off the execution stack, and the interpreter reenters the server loop. The primary task of the server loop is to isolate one batch job from another. Before each job starts, the server loop executes the save operator. When a job has nished executing, the server loop is re-entered, and a restore is performed, restoring the state of the interpreter back to exactly the state that was in effect at the start of the job. Errors are caught by using the stopped operator. There is a discussion of this technique in Section 14.3.

34

chapter 2 THE EXECUTION MODEL

In listing 2-5 is an idea of what the server loop looks like. A few pieces of pseudo-code are included to show some functionality without providing all the nuts and bolts. These are represented in italics. This example does not match any real server loop code, but it provides the theoretical model for how it operates. The server loop procedures for any given interpreter can be seen by simply executing this code: 128 array execstack ==

listing 2-5
/server save def % save the initial state { %loop server restore % when entering loop { %loop any port active { %if exit } if } loop % until new job is encountered /server save def % before each job (%stdin) (r) file cvx stopped { % execute the file object errordict /handleerror get exec currentfile flushfile } if } bind loop % main server loop

chapter 2 THE EXECUTION MODEL

35

chapter 3 THE IMAGING MODEL

Painting isnt an sthetic operation; its a form of magic designed as a mediator between this strange hostile world and us.... Pablo Picasso

3.1 INTRODUCTION
The PostScript language imaging model is the metaphor through which graphics are rendered on output devices. The imaging model consists of specic rules and mechanisms by which a picture is described, and its behavior is exactly predictable. The imaging model for silk-screening, for example, might be loosely stated like this: Ink is forced through a mask of lm that adheres to a stretched piece of silk. Ink ows through the mask only where it has been cut away. The imaging model for the PostScript language might be stated: Algorithmic paths are created to dene an area or a series of lines, and ink is applied to those paths through one of several methods, including stroking, lling, halftoning, text rendering, and clipping. In the world of computers, all graphics application programs have a graphic imaging model, whether it is explicitly stated or simply assumed. Even text editors have a paradigm for the lines of text being edited. Although video screens are two-dimensional, the model used by a drawing program may permit objects to overlap on the screen, as though there were a third dimension. There is no correct imaging modelit is just a consistent way of thinking about graphics that is useful to a computer program. Writing a driver for an application to print through a PostScript interpreter is almost entirely the task of adapting one graphic imaging model to another. The act of printing a page is a translation of some computer graphics representation into a printed

chapter 3 THE IMAGING MODEL

37

reality. If the software permits the user to draw a circle, for example, the circle must be stored in some way (the internal representation), and ultimately printed with whatever means are available.

3.2 APPLYING THE METAPHOR

The PostScript language uses a path and paint model for imaging. A path is constructed out of simple graphics primitives such as moveto, lineto, curveto, and closepath. The resulting path may then have paint applied to it with one of the painting operators (such as fill or stroke). Paint is opaque in the PostScript imaging modelthat is, any color of paint will completely obscure any paint that may already have been laid on the page. Path construction and painting operations in the PostScript language are state-dependent. All of the elements used in graphic operations are maintained by the PostScript interpreter in what is known as the graphics state. This includes the current path, the current line weight, the current color, the current transformation matrix, and other elements used in rendering graphics. At any moment there is only one current path. The path can have additional elements added to it, it may be used by one of the painting operators, or it may even be initialized (with newpath). There can be many disconnected subpaths that together make up the current path. The approach used for rendering graphics consists basically of the repetition of the following steps: Save the current state of the interpreter VM (memory). Establish a current point (with moveto). Construct a path, if needed (it is not necessary, for example, with show). Paint the path (or show characters), resulting in marks on the page. restore to the previously saved state. (Optional) Execute showpage to print the current page.

38

chapter 3 THE IMAGING MODEL

gure 3.1

chapter 3 THE IMAGING MODEL

User Space

Device Space ( frame buffer)

Physical Page
gure 3.1

Path Construction

Painting (Rendering)
eofill imagemask ashow kshow fill stroke image show widthshow erasepage

Printing
showpage copypage

39

moveto lineto curveto arc, arcto closepath show widthshow charpath clip clippath strokepath

rmoveto rlineto rcurveto arcn reversepath ashow kshow flattenpath eoclip initclip newpath

3.3 CONSTRUCTION OF PATHS

Paths are constructed using path construction operators. Here is a list of all of them: moveto rmoveto arc closepath clip initclip show ashow grestore lineto rlineto arcn clippath eoclip strokepath widthshow kshow restore curveto rcurveto arcto charpath reversepath flattenpath newpath initgraphics grestoreall

Each of these adds to (or creates, or modies) the current path. The newpath, initgraphics, restore, and grestoreall operators can be thought of as path construction operators, even though they are really path destruction operators. The variations on the show operator are path construction operators only in that they affect the current point, which is part of the current path. The current path is part of the standard graphics state, and it is preserved (and restored or overwritten) by both the gsave and grestore operators and the save and restore operators. The clip operator uses the current path (appending it to the current clip path) but does not destroy it. This is often the cause of strange program behavior when using clip. Using the newpath operator after the clip usually fixes these problems. Construction of a path alone is not sufcient to cause marks to be made on the page. A path must be painted in order to see it.

THE GRAPHICS STATE AND PATHS

There is an interesting relationship between building the current path and changing the graphics state. For example, consider the short PostScript program given in listing 3-1. Notice that the constructed path has three elements: an initial moveto and two instances of rlineto, each of which appends a line segment to the current path. However, there are also three changes to the graphic state. The setlinewidth operator is invoked twice

40

chapter 3 THE IMAGING MODEL

in this example, and scale is invoked once. What does this path look like on the nal page, after it has been stroked?

listing 3-1
%!PS-Adobe-2.0 %%Title: pathconstruct.ps %%EndComments /F { findfont exch scalefont setfont } bind def %%EndProlog gsave 400 300 moveto 100 0 rlineto % horizontal line 2 setlinewidth 4 4 scale 0 100 rlineto % vertical line 1 setlinewidth gsave % make the picture more interesting... 14 /Times-Roman F (A) show 2 -12 rmoveto -20 rotate (A) show grestore stroke grestore showpage %%Trailer

To answer this, one must look more carefully at the graphics state (and at figure 3.2). Here are the elements of the graphics state: Current point and current path Current transformation matrix (CTM) Current font Current color (or gray level) Line weight, line cap, and line join Line miter limit and dash pattern Current raster device Current clipping region All halftone parameters The entries set in bold are the only elements of the graphics state that affect path construction directly. The others affect

chapter 3 THE IMAGING MODEL

41

only the painting of the path. In listing 3-1, the 4 4 scale does affect the subsequent execution of rlineto, causing the line to be (effectively) four times as long when it is appended to the current path. The path construction operators use the current transformation as the path is being constructed, but once an element of the current path exists, it is not affected by further changes to the coordinate system.

gure 3.2

Another way to think of it is that changes to the current transformation matrix and the current font affect future elements of the path, but not existing elements. Changes to the line width or the current color take effect only when stroke or fill is executed, and they apply to the entire current path.

3.4 PAINTING OPERATIONS

A path, once constructed, must be painted in order to make marks on the page. There are several painting operators, each of which causes some marks to be made in device space (assuming

42

chapter 3 THE IMAGING MODEL

that some part of the path is within the clipping region, of course). Here is a list of painting operators: fill image show ashow eofill imagemask widthshow kshow stroke

When marks are made in device space by one of these painting operators, they stay there. Even the showpage, erasepage, and framedevice operators, which cause device space to be cleared, operate by painting the entire page with white. In the steps for path construction and painting outlined in Section 3.2, the use of save and restore guarantees the functional independence of each page element, although the marks on the page persist until showpage is nally executed.

3.5 WHAT HAPPENS TO THE CURRENT PATH?

There is only one current path in the graphics state. A large percentage of the execution of a document involves the construction and use of paths. It denitely helps to understand this mechanism and use it efciently. In particular, keep the following in mind: It is inefcient to destroy a path explicitly (with newpath) if it will happen as a natural consequence of using the path (as it will with stroke or fill). Dont preserve the path (with gsave and grestore) if it isnt needed for anything. It is unfortunately fairly common to see sequences like the following one in production software: 0 0 moveto 100 100 lineto gsave stroke grestore newpath The path is carefully preserved, stroked, and restored, only to be destroyed immediately with newpath. In this case three unnecessary operators are executed for every line drawn.
chapter 3 THE IMAGING MODEL 43

3.6 PROCEDURES FOR CONSTRUCTING PATHS

It is often found that the kinds of shapes rendered by a graphics application require a little more than simple calls to lineto or curveto. In these cases, after careful consideration, it may be determined that writing a procedure to implement a particular graphic primitive is advantageous. For example, rectangles, circles, arrowheads, and other shapes may be fundamental parts of your application. It is important to construct efcient procedures. Dening a procedure and using it instead of a number of PostScript language commands is only a useful technique if the procedure is faster than the unrolled code would have been. There are a couple of considerations that make it easier to generate fast procedures: What is the least amount of critical data needed from the host application to construct the required shape? What is the natural order and format in which to supply the data, based on the PostScript operators to be used? How should the operands be presented on the operand stack to minimize the computation required of the procedure itself? For example, if the arc operator is used (as it is to construct a circle), the center and radius should be specied, not the corners of a bounding rectangle (see Section 5.5 and Figure 5.1). If an arrowhead is to be drawn, it should be carefully worked into the line-drawing primitives like lineto and curveto, making use of the way paths are naturally constructed.

RECTANGLES
A common task performed by a graphics application is to draw a rectangle, often one that is shaded and/or outlined with a border. There are many ways to represent a rectangle. The goal, when writing a PostScript driver, is to minimize the computation required and the data transmitted. For example, if rectangles are always oriented either vertically or horizontally (never at an angle), they can be represented by two points (for example, the lower left and upper right corners), or by one corner

44

chapter 3 THE IMAGING MODEL

and the width and height of the rectangle. However, if the rectangle may be rendered at an angle, four points are necessary, or two points and an angle (although that would require more computation). A rectangle procedure may require several painting attributes (like a gray level and a line weight) as well as the geometric information to construct the rectangle. If so, this data should be passed to the procedure on the operand stack. Listing 3-2 provides a good rectangle procedure.

listing 3-2
%!PS-Adobe-2.0 %%Title: rectangle.ps %%EndComments % "R" takes the following arguments: % lineweight linegray fillgray % width height LLx LLy /R { %def gsave moveto 1 index 0 rlineto 0 exch rlineto neg 0 rlineto closepath gsave setgray fill grestore setgray setlinewidth stroke grestore } bind def %%EndProlog 3 .5 .9 200 300 100 100 R showpage %%Trailer

This example draws a rectangle that is 200 points wide, 300 points tall, is lled with 10 percent gray (0.9) and stroked with a 3-point line at 50 percent gray (0.5). Notice that the procedure uses the values directly from the stack (rather than dening them into a dictionary and calling them up again), since they are only needed once.

chapter 3 THE IMAGING MODEL

45

CIRCLES AND ARCS

Circles are another simple graphic shape for which there are many, many possible representations. The PostScript language has one in the arc operator. To draw a circle in the PostScript language, use the method provided in the imaging model.

listing 3-3
%!PS-Adobe-2.0 %%Title: circle.ps %%EndComments % "C" takes the following arguments: % linewidth linegray fillgray % X Y r ang1 ang2 /C { %def gsave newpath 0 360 arc gsave setgray fill grestore setgray setlinewidth stroke grestore } bind def %%EndProlog 3 .5 .9 300 500 200 C showpage %%Trailer

Listing 3-3 draws a circle of radius 200 at the point 300, 500, lls it with 10 percent gray (0.9), and strokes it with a 3-point line at 50 percent gray.

ARROWHEADS
Lines terminating with arrow heads are commonplace in many graphics applications. In listing 3-4 is a procedure that will draw an arrow head at the current point, at a specied orientation and line width. (The line width is used only as a size referencethe arrow head is scaled to an appropriate size for lines of the specied width.) The orientation is supplied by a previous point. The arrow head is drawn parallel to the line implied by

46

chapter 3 THE IMAGING MODEL

the previous point and the current point, which should be the tangent of the line at the end point. See Listing 3-4.

listing 3-4
%!PS-Adobe-2.0 %%Title: arrowhead.ps %%EndComments %%BeginProcSet: arrows 1.0 0 % "arrowhead" takes these arguments: % lineweight prevX prevY /arrowhead { %def gsave currentpoint 4 2 roll exch 4 -1 roll exch sub 3 1 roll sub exch atan rotate dup scale -1 2 rlineto 7 -2 rlineto -7 -2 rlineto closepath fill grestore newpath } bind def /l^ { %def % lineto-arrow currentlinewidth currentpoint 5 3 roll lineto currentpoint stroke moveto arrowhead } bind def /rl^ { %def % rlineto-arrow currentlinewidth currentpoint 5 3 roll rlineto currentpoint stroke moveto arrowhead } bind def /arc^ { %def % arc-arrow 5 copy arc currentpoint stroke moveto % stroke arc % getting the correct orientation for the arrowhead % is tricky. This procedure uses the arguments to % "arc" to determine the tangent of the curve at the % endpoint, and it orients the arrowhead along that % tangent line. It leaves an X-Y point that is just % behind the arrowhead along the tangent.

chapter 3 THE IMAGING MODEL

47

% newX = X + radius * cos(endAngle-1) % newY = Y + radius * sin(endAngle-1) exch pop 1 sub % endAngle - 1 degree dup cos 2 index mul 4 index add % arrowX exch sin 2 index mul 3 index add % arrowY currentlinewidth 2 add 3 1 roll % thickness arrowhead pop pop pop % draw -> } bind def %%EndProcSet: arrows 1.0 0 %%EndProlog %%Page: 1 1 % line sample: 200 600 moveto 0 10 360 { currentlinewidth .1 add setlinewidth gsave dup cos 100 mul exch sin 100 mul rl^ grestore } bind for newpath % curve sample: /radius 10 def .1 setlinewidth 0 30 360 { /radius radius 10 add def 280 230 radius 0 5 -1 roll arc^ } bind for showpage %%Trailer

Listing 3-4 contains a fairly sophisticated set of procedures. There is one procedure to draw an arrowhead at the current point, and there are three other procedures (rl^, l^, and arc^) that work like rlineto, lineto, and arc, respectively, except that each draws an arrowhead at the end of the line segment (or arc) using the arrowhead procedure. In the arc^ procedure, the tangent to the curve is computed and a point is passed to the arrowhead procedure for orienting it properly. It is more likely that the host application would provide this simple x-y location to orient the arrowhead, to avoid this largely unnecessary calculation. There are also other possible orientation schemes.

48

chapter 3 THE IMAGING MODEL

Note: Keep track of the current path, and use it naturally. Avoid unnecessary path operations or safeguards that may seem at rst like good, defensive programming. The language is designed to work the way you probably need it to work, and it is best to use it that way. The fill, eofill, and stroke operators effectively use up the current path, destroying it when they are through (by executing the equivalent of newpath). The image and imagemask operators do not disturb the current path in any way. The show operator and its siblings make changes only to the current point (corresponding to the width of the string being shown), and otherwise do not disturb the current path.

3.7 TEXT OPERATIONS

In the PostScript language, text is rendered with the show operator (or one of its relatives). The show operator is somewhat special in the PostScript imaging model, in that there are very specic requirements for setting text that may not apply when rendering other graphic elements. In particular, text characters are normally set one after another, along a baseline. The show operator leaves the current point when it is done, rather than destroying the path as other path painting operators do. Furthermore, show requires not only a current point but a current font in order to render text (unlike other painting operators). In order to effectively make use of the PostScript imaging model, one should understand just what effect the show operator has, and when it is appropriate, for instance, to use gsave and grestore to preserve the current point (and the current font). Let us consider two variations that print a couple of lines of 10point text (listing 3-5).

chapter 3 THE IMAGING MODEL

49

listing 3-5
%!PS-Adobe-2.0 %%Title: text examples %%EndComments /F { findfont exch scalefont setfont } bind def /S1 { moveto show } bind def /S21 { moveto gsave show grestore } bind def /S22 { %def 0 exch rmoveto gsave show grestore } bind def %%EndProlog % method 1: 10 /Times-Roman F (method 1:) 72 512 S1 (Viva Nuestra Senora de Guadelupe!) 72 500 S1 (Viva la Independencia!) 72 488 S1 % method 2: 10 /Times-Roman F (method 2:) 72 412 S21 ( There was a young lady named Bright,) -12 S22 ( Whose speed was far faster than light;) -12 S22 ( She set out one day) -12 S22 ( In a relative way) -12 S22 ( And returned home the previous night.) -12 S22 ( -Arthur Buller) -12 S22 showpage %%Trailer

The rst method in listing 3-5 (labeled method 1 in the comments) has a very simple procedure (called S1) which does a moveto and a show. The x,y location for the text is passed on the stack along with the string, and the current font is understood already to exist. The second example (called method 2) uses a different approach. The rst line uses an x,y pair, as does method 1, except that it uses gsave and grestore to preserve the current point after show is nished. Subsequent lines of text are set with just a relative quantity (an amount to move down before

50

chapter 3 THE IMAGING MODEL

printing the line). This enables the driver to generate and transmit less data, but there is a greater computation cost in doing the gsave, grestore, exch, and rmoveto for each line. The two examples above are inherently based on the behavior of the show operator: It always leaves the current point at an adjusted position (usually to the right of the text shown) based on the character widths. In method 2, the second line (and subsequent lines) of text are dependent on the rst, in the sense that the position is relative to the rst line. Another approach might be to have a procedure for setting an entire paragraph, where the inter-line spacing is constant for the whole paragraph, and the moveto can be generated easily from within the code. The rest of this book is dedicated to the decision-making process. The PostScript language execution and imaging models provide great freedom in setting text, but in any environment there are constraints or advantages that may dictate using one method over another. Understanding the workings of the mechanism will help you to make these decisions.

CHARACTER WIDTHS
Character widths are stored within PostScript language font programs as displacements in coordinate space. The best way to think of text setting is to view each character as an origin, a shape, and a width. The relationship between the shape and the origin is xedthe origin is the 0,0 point in character space (described later) and the character draws itself relative to that location. The width is essentially an implicit rmoveto that occurs after the character shape is rendered. (See gure 3.3). The character width is not related to the actual execution of the character shape (the shape may even be in the font cache, and not get executed at all). This is not important unless you are building font characters yourself. (See also Section 9.3). Remember that characters widths have both an x and a y component. A common bug is to call stringwidth but to forget that there are two values returned. Typically, the extra number stays on the stack until a typecheck occurs later.

chapter 3 THE IMAGING MODEL

51

gure 3.3

character width (displacement after show) origin (current point when show executes)

The important thing to remember is to use the way The PostScript language renders text instead of ghting it. Below are some aspects of the execution and imaging models that commonly present difculty. Each of these is a symptom of not having used the language effectively: When you try to use save and restore at the page boundaries, you discover that it destroys the current font at the bottom of the page, and it needs to be explicitly re-set at the beginning of the next page. You find that changing fonts in the middle of a line of text makes it hard to justify the line, since the character widths are different for the two fonts. You are using stringwidth to determine where text is going to be placed. (The application should always control placement of text, and should know the character widths beforehand.) When you relocate or remove graphics from a le, the text does not print in the correct position on the page, since it relies on the current point being in the right place.

52

chapter 3 THE IMAGING MODEL

Some thought is required to build a complex program, and using the PostScript language is no exception. Learning the characteristics of each text operator and remembering the semantics of the imaging model will help enormously in designing a simple and efcient program.

3.8 CLIPPING
Clipping is a mechanism used to limit the area in which paint can be applied by the PostScript language painting operators. It behaves like a stencil. There is always a clipping path in the graphics state, which by default is the current page boundary. The clipping path may have further restrictions appended to it or removed from it (with grestore, for example)but it cant be replaced entirely. The clip operator appends the current path to the current clipping path (but does not destroy the current path).

COMPLEXITY AND PERFORMANCE

Clipping requires computation. When the PostScript interpreter paints a path, it must determine whether or not the marks fall within the current clipping path. This computation can be quite expensive, depending upon the implementation. Additionally, the clip operator reduces the current path to a attened path as it is added to the clipping path. This means that it is broken down into line segments. There is a practical limit on how many path elements may be stored, both in the clipping path and in the current path in the graphics state. Exceeding this limit will result in a limitcheck error. Depending on the path and the resolution of the output device, a attened path may contain hundreds or even thousands of elements, and care must be exercised when using clip. One source of unexpected behavior stems from the fact that the clip operator does not destroy the current path. Remember to execute newpath after clip unless the current path is to be used for something else. The clip operator does not initialize the current path.

chapter 3 THE IMAGING MODEL

53

3.9 RASTERIZATION
Rasterization is a term for the process of converting the original high-level PostScript language description into rasters, or plain old bits of black or white (or plain old 16 bits of color or plain old 8 bits of grayscale, or whatever is the appropriate fodder for the output device). In particular, rasterization is the end result of all the painting operators such as stroke, fill, and show. The real device independence of the PostScript language lies in the fact that the interpreter performs the rasterization at the resolution of the output device. Only high-level descriptions of graphics are sent to the printer. (Even scanned images are scalable, and not resolution-dependent.) In many devices, the PostScript interpreter can rasterize in parallel with the interpretation of the program, and may be deferred when the original painting operator is executed. This is especially true in high-speed printing devices, where some of the nal rasterization may be done on the y during showpage. Most programs need not be concerned with the rasterization process itselfthe purpose of the PostScript language is that the user program is removed from the level of rasterization to ensure true device independence.

3.10 SAVE AND RESTORE

The save and restore mechanism is a fundamental part of the PostScript imaging model. It affects all aspects of a program, including its execution, the graphics state, and its use of memory and resources. As a starting place for this discussion, here is a rule of thumb: If you nd it difcult to get save and restore to work correctly in your program, you should redesign your program. A PostScript program should be designed from the very start to work easily with the state-dependent execution model of the PostScript interpreter. Page elements should be independent of one another. Each page should be functionally independent

54

chapter 3 THE IMAGING MODEL

of all others. (See Chapter 6). The graphics state stack should naturally parallel the execution of your program. Simply stated, save and restore (they are always used in pairs) provide a mechanism for saving the current state of the world and then returning to it. This mechanism affects everything (except the contents of the stacks and the marks on the current page). It saves the contents of dictionaries, the current line weight, the current halftone screen, the current font, the path, the clipping region, the existence of string bodies, array objects, and all objects in memory. There is nothing special required to use save and restore effectively. They should be designed in from the beginning into any procedures that use memory. If memory conservation is not an issue, gsave and grestore will probably work just as well, and are more efcient. In general, every procedure that affects the graphics state in some way should contain either gsave and grestore or save and restore. Chapter 6 provides more detail on how to structure procedures and programs to take full advantage of these operators. Section 13.3 discusses the use of save and restore for memory management.

3.11 THE FONT CACHE

The font cache is an optimization technique. The best way to use the font caching mechanism is to be glad that it exists. There is no direct means for manipulating the font cache, and it is not intended for general use. However, it is useful to have some background on how it works. Characters go into the font cache whenever they are executed by the show operator, as long as they will t (based on the font caching limits currently in effect, not on how many characters are already in the cache) and as long as the font itself executes the setcachedevice operator (the only other choice is setcharwidth, which explicitly avoids the font cache).

chapter 3 THE IMAGING MODEL

55

Characters are removed from the font cache on a least-recentlyused basis whenever the available space in the cache is exhausted. The font cache occupies a xed amount of space in most implementations (although not all of them), and characters always are added to the cache if they qualify. If there is not enough room, the least recently used characters are removed. Entries are made into the font cache for a single character, and are not pre-cached for an entire font. The entry is made based on the current font, the size and orientation of the character (based on the current transformation matrix), the character name, and its width.

56

chapter 3 THE IMAGING MODEL

chapter 4 EMULATORS AND TRANSLATORS

Nature has ... some sort of arithmetical-geometrical coordinate system, because nature has all kinds of models. Richard Buckminster Fuller

4.1 INTRODUCTION
The PostScript language can be thought of in two distinct ways: it provides an imaging model for describing and printing complex text and graphics, and it is a complete and general programming language. A printer emulator is software that uses the PostScript programming language to implement some other imaging model. The simplest example of this might be a line printer emulator, in which the ASCII text sent to the printer is simply listed out on the page rather than being interpreted. A print format translator is a host-based program which converts a le created for another printing device (and its accompanying imaging model) into a PostScript le. For instance, the program might take line printer input les and produce PostScript output les which may then be transmitted to a device for printing. There are many problems which can arise in translating other print les or emulating other printers, most of which generally have to do with fonts. When setting text, the widths of the individual characters are very important in making decisions about line breaks and for centering text. If text has been formatted for a different printer (perhaps with different character widths for its fonts) the translation process becomes more difcult. In addition, many print les may have been decomposed from their

chapter 4 EMULATORS AND TRANSLATORS

57

original editable form into an output format appropriate for a particular printer, and what used to be text or labels might be vectors or bit images by the time the print le is produced. At that stage, there is no reasonable way to reverse-engineer the print le to produce the clean typography which the PostScript language provides. Whenever possible, it is best to compose pages originally with PostScript devices in mind, and to use font width tables that correspond to the fonts that will be used on the PostScript printer. It is also best to consider the PostScript imaging model fully, and to adapt the editable representation to the PostScript language model when producing the print le, rather than translating it after the fact, or programming the interpreter to emulate another printer. This chapter outlines some of the best methods for writing emulators and translators when it is not possible to modify the software which produces the print les. Chapter 5 is devoted to the subject of creating PostScript language page descriptions directly from the composing application.

4.2 EMULATING ANOTHER PRINTER

To successfully emulate another printer in the PostScript language, the software must be able to print any le that would produce results on the original printer. This means that any control sequences, condensed printing modes, or other features of the printer must be mimicked completely. From an implementation point of view, an emulator is simply a program that executes in the printer and reads directly from the input (superseding the interpreter). As input is received, the emulator looks for control codes, keeps track of line positions, reproduces condensed or overstrike modes, and prints pages as required. Because the PostScript language is interpreted, the program loop is actually re-interpreted each time around, and must be written carefully in order to make performance reasonable. Listing 4-1 is a simple loop that reads a xed-length string from the current le (normally the standard input stream, %stdin)

58

chapter 4 EMULATORS AND TRANSLATORS

and prints it, quitting only when the readstring operator detects an end-of-le indication. Note that no parsing of the string is attempted, no control codes are recognized, and there is no capability to change fonts. In fact, with this example even line breaks (carriage return characters) are ignored.

listing 4-1
%!PS-Adobe-2.0 %%Title: emulate1 /buff 128 string def /emulate1 { %def { %loop currentfile buff readstring exch show not { exit } if } loop showpage } bind def %%EndProlog 72 750 moveto /Courier findfont 10 scalefont setfont emulate1 So spake the Fiend, and with necessity, The tyrants plea, excused his devilish deeds. -Milton

The readline operator may be used to automatically recognize newline characters, but with even this simple change the program becomes much more complex. For instance, there is immediately the possibility of trying to read in a line that is longer than the allocated buffer size (this is not a problem with the readstring operator, since it does not look for newline characters, it simply lls the buffer). Additionally, depending on the communications channel, newlines may be handled differently in different environments and by different PostScript interpreters. In the purest sense, it is better not to use readline, but instead to use search to parse for various escape sequences and newline characters explicitly. Listing 4-2 is a revised form of the previous example. It searches for linefeed characters.

chapter 4 EMULATORS AND TRANSLATORS

59

listing 4-2
%!PS-Adobe-2.0 %%Title: simple line printer emulator /buff 5000 string def /leftmargin 72 def /topmargin 72 def /bottom 72 def /top 792 topmargin sub def /ptsize 10 def /lead 10 def /EOLchar (\n) def % line feed /F { findfont exch scalefont setfont } bind def /newline { %def currentpoint exch pop lead sub dup bottom lt { %if showpage pop top } if leftmargin exch moveto } bind def /emulate { %def { %loop currentfile buff readstring exch EOLchar { %loop (innermost) search { %ifelse show newline }{ %else show exit } ifelse } loop not { %if readstring found EOF exit } if } loop showpage leftmargin top moveto } bind def %%EndProlog %%BeginSetup ptsize /Courier F leftmargin top moveto %%EndSetup emulate I wasted time, and now doth time waste me; For now hath time made me his numbering clock; My thoughts are minutes. - William Shakespeare

60

chapter 4 EMULATORS AND TRANSLATORS

There are several items of note in this program. The innermost loop, for instance, takes advantage of the syntax of the search operator. After calling search, there is a boolean on the top of the operand stack which indicates the success or failure of the search. This boolean is used immediately by the ifelse statement, and one of the two procedure bodies is conditionally executed. If the search failed, the original string is still on the stack, and it is merely printed (show) and the loop is exited. If the search succeeded, there are three strings on the stack. The topmost one is the portion of the string just before the newline (the newline was the match string provided to search), which is then printed (with show). The remaining two strings are already in the correct position for another call to the search operator, so it is okay to just drop through the loop and re-enter at the top. The search operator was designed very carefully to return on the stack the information that is most likely to be needed, in an appropriate order, and it is easy to write a program that takes advantage of this design. This use of search can easily be extended to search for other characters as well as the newline character, although the inner loop would become more complicated. For instance, the tab character could be caught, and perhaps the escape character, in order to interpret control sequences.

THE STRINGWIDTH OPERATOR

If the printer emulator is likely to be faced with printing text, be sure to read Chapter 7, which presents several approaches to setting text, including font changes, justication, leading, and underlining. In general, however, all text-setting algorithms (whether implemented in an emulator or in a print driver) need to know the widths of character strings in order to make positioning judgements. Widths are normally made available on the host system for computing line breaks, but an emulator is forced to do the calculations in the PostScript interpreter. Character widths can be obtained using the stringwidth operator. This operator requires a current font (and a current point) and a string object on the operand stack. It returns two values, which represent the width of the string in both x and y. The width of a string is dened to be the distance by which the cur-

chapter 4 EMULATORS AND TRANSLATORS

61

rent point is modied if the given string were printed in the current font. Here is an example of its use (a session from an interactive PostScript interpreter): PS> 0 0 moveto PS> /Times-Roman findfont PS> 12 scalefont setfont PS> (TRANSLATING EXISTING FILE FORMATS) PS> dup stringwidth exch == == 230.316 0 PS> The stringwidth operator can take almost as long to execute as show, since it does the same work of imaging the characters into the font cache and computing the offset to the current point that would occur. It is not too unreasonable to say that printing time doubles when the width of a string must be computed before it is printed. Although the characters are already in the font cache when show is executed, the overhead in interpretation and execution time required to compute the strings width and to adjust the current point tend to offset this. Whenever possible, an emulator written in the PostScript language should avoid computing the width of a character string. For instance, if line breaks have already been determined (as is usually the case in line printer output) and the text is not justied, there is no reason to measure the string.

TEXT JUSTIFICATION IN AN EMULATOR

Performing full justication on a line of text with more than one font presents some tough problems for an emulator. First the line of text must be collected, and each substring (per font used) must be measured with stringwidth. Then the number of spaces must be counted, as well as the total number of characters in the line. Finally a subtraction and division problem are performed to gure out how much the line must be stretched to t the margins, and the error divided evenly among the spaces or characters in the string and widthshow (or ashow) is called. The program in listing 4-3 is a printer emulator. It emulates a ctitious printer that is reasonably close to several real-world

62

chapter 4 EMULATORS AND TRANSLATORS

printers. It takes ASCII input data already broken into lines (it uses readline, in fact) and prints it with a variety of possible fonts. It is designed to use a default font (which you can specify) and to search for any of several escape sequences which permit changing fonts and going into or out of justication mode. The following are the recognized escape sequences: ESC Sequence ESC Y ESC N ESC 0 ESC 1 ESC 2 ESC 3 Function justication on justication off set font 0 set font 1 set font 2 set font 3 Result

(roman) (bold) (italic) (bold italic)

The escape character and the individual commands are soft. The fonts mapped to the font codes are also soft. The 0, 1, 2, and 3 in the font codes are, in this example, actually the ASCII versions of those numbers (48, 49, and so on) so that the example program is readable. Similarly, the escape character is set to the caret character (^). In a real emulator, these would probably be changed. This emulator prints at roughly 17 pages per minute on current implementations, with a good mix of fonts and in full justication mode.

chapter 4 EMULATORS AND TRANSLATORS

63

listing 4-3
%!PS-Adobe-2.0 %%Title: justify.ps %%Creator: Glenn Reid %%EndComments %%BeginProcSet: general 1.0 0 /FS { findfont exch scalefont } bind def %%EndProcSet: general 1.0 0 %%BeginProcSet: justify 1.0 0 /spacecount { %def 0 exch (\040) { %loop (space) search { %ifelse pop 3 -1 roll 1 add 3 1 roll }{ pop exit } ifelse } loop } bind def /justify { %def % mark (text) <font> (text) <font> /spaces 0 def /currentwidth 0 def /currlength 0 def counttomark /num exch def 2 2 num { %for % roll through stack and count -2 roll setfont dup spacecount spaces add /spaces exch def dup stringwidth pop currentwidth add /currentwidth exch def dup length currlength add /currlength exch def currentfont } for currlength 1 le { setfont show } { %ifelse /adjust %def columnwidth currentwidth sub spaces 0 eq { %ifelse currlength 1 sub }{ %else spaces } ifelse div def num 2 idiv %repeat spaces 0 eq { %ifelse { %repeat setfont adjust 0 3 -1 roll ashow } }{ %else

64

chapter 4 EMULATORS AND TRANSLATORS

{ %repeat setfont adjust 0 32 4 -1 roll widthshow } } ifelse repeat } ifelse } bind def %%EndProcSet: justify 1.0 0 %%BeginProcSet: linemulator 1.1 0 /buff 1000 string def % longest line for "readline % for recognizing escape sequences to change fonts % and enter/leave "justify" mode: /ESC (^) def % escape character /justON (Y) 0 get def % ESC-Y /justOFF (N) 0 get def % ESC-N /QUIT (Q) 0 get def % ESC-Q /zero (0) 0 get def % for subtracting ASCII offset /newline { %def currentpoint exch pop lead sub dup bottom lt { %if showpage pop top } if leftmargin exch moveto } bind def /linetoolong { %def newline (%%[Error: line too long.]%%) show showpage } bind def /printline { %def JUST { %ifelse counttomark 2 ge { justify } if }{ counttomark 1 ge { show } if } ifelse newline } bind def /emulate { %def % read currentfile one line at a time. Search for % ESC sequences. If not found, just "show" with % current font and move on. If justification is % turned on, then LEAVE all the strings and the % correct font dictionaries on the stack, and call % "justify" at the end of the line: /fmax fonts length def /top 792 topmargin sub def /JUST justification def leftmargin top moveto

chapter 4 EMULATORS AND TRANSLATORS

65

/sv save def { %loop mark currentfile buff { readline } stopped { linetoolong stop } if not /done exch def % use the boolean later { %loop % handle escape sequences ESC search { %ifelse JUST { %ifelse exch pop exch currentfont exch }{ show pop } ifelse dup length 0 eq { pop ESC JUST { %ifelse exch currentfont exch } if printline exit } if dup 0 get % escape argument true exch % look for ESC-fontchange? dup justON eq { /JUST true def exch pop false exch } if dup justOFF eq { /JUST false def exch pop false exch } if dup QUIT eq { /done true def exit } if exch { %ifelse zero sub % subtract ASCII offset dup dup fmax le exch 0 ge and { %ifelse fonts exch get setfont dup length 1 sub 1 exch getinterval }{ %else pop ESC JUST { %ifelse exch currentfont exch }{ show } ifelse } ifelse }{ %else pop dup length 1 sub 1 exch getinterval } ifelse }{ %else

66

chapter 4 EMULATORS AND TRANSLATORS

JUST { %ifelse dup length 0 ne { %ifelse currentfont printline }{ pop printline } ifelse }{ %else dup length 0 ne { %ifelse printline }{ %else pop newline } ifelse } ifelse exit } ifelse } loop cleartomark done { exit } if currentpoint /JUST JUST sv restore /sv save def def moveto } loop showpage sv restore } bind def %%EndProcSet: linemulator 1.1 0 %%EndProlog %%BeginSetup /columnwidth 432 def /justification false def /leftmargin 108 def /topmargin 72 def /bottom 72 def /lead 14 def /fonts [ %def 12 /Times-Roman FS % ESC-0 12 /Times-Bold FS % ESC-1 12 /Times-Italic FS % ESC-2 12 /Times-BoldItalic FS % ESC-3 ] def fonts 0 get setfont %%EndSetup emulate (turn on justification after this line)^Y Here is ^1bold^0, here is ^2italic^0, and here is ^3BoldItalic^0. ^NNo other justification is necessary.^Q

chapter 4 EMULATORS AND TRANSLATORS

67

4.3 TRANSLATING EXISTING FILE FORMATS

A description of how to write a print format translator is not directly within the scope of this book, since it most often means writing an application on a host computer to accomplish the translation. However, the PostScript language output which is produced by the translator should be designed carefully, and there are several issues that are likely to be common to many different kinds of translators.

UNITS
One of the rst issues that must be addressed when translating a print le intended for another printer is to determine what units or measurements were used in the original le, and to convert or scale them into an appropriate PostScript coordinate system. Conversion between number systems can result in signicant round-off error in computation. When setting text and producing graphics, exact tolerances are needed to preserve the quality of the page image. It is best to study the relationship between the host units used by the host application and the default PostScript coordinate system and to perform the simplest possible conversion. One way to map an applications coordinate system into the units that the PostScript language uses is simply to use the scale operator. However, the following issues should be considered if this technique is chosen: Fonts are scaled separately from the user coordinate system, but they are rendered through the current transformation matrix. If the user-space transformation is modied, then the numbers provided to scalefont must be adjusted accordingly. The implicit mapping performed by the current transformation matrix represents a calculation. This calculation is performed whether or not the mapping is used as a means to reconcile coordinate system differences; taking advantage of this mechanism is essentially free.

68

chapter 4 EMULATORS AND TRANSLATORS

 showpage performs an implicit initgraphics. It will be necessary to use gsave and grestore to preserve the coordinate system across pages, or to set it explicitly for each page. For example, if the base coordinate system in a layout application uses inches as its unit of measurement, it is much more efcient to scale the PostScript coordinate system to match this than it is to programmatically convert numbers into points: /in. { 72 mul } def 1 in. 2 in. moveto This requires three name lookups, the execution of a procedure body, and two multiplication operations (as well as 6 more bytes of information to be received and parsed by the PostScript interpreter) to position the current point. Compare this to the following bit of code, which has the same function and more than twice as fast: 72 72 scale 1 2 moveto

FONTS
The most basic problem encountered in translating other print formats into the PostScript language is incompatibility among font libraries. If a xed-pitch font is used (one in which all character widths are the same) the problem is simplied quite a bit, but in most cases involving proportional fonts, there will be incompatibilities. Deciding on a philosophical approach to these kinds of differences is perhaps the most difcult part of writing a good emulator or translator.

4.4 FONT DIFFERENCES

For both emulators and translators, it is a tough problem to print a document that was formatted for a different printer. Most formatting and typesetting decisions contain very typefacespecic assumptions. Line layout, justication, centering, and amount of text on a page are all contingent upon the fonts for which the document was originally formatted.

chapter 4 EMULATORS AND TRANSLATORS

69

When the fonts available on the original printer do not match the fonts available on the PostScript device in question, the rst step is to introduce a font substitution strategy. Essentially, a table is built that maps the generic Roman font to Times-Roman on the PostScript printer, and so on for all the possible fonts. Some of these substitutions will be better than others (depending on the similarities or differences between the original and substituted fonts), but they provide a good starting point. A renement to this approach is to dene additional fonts in the PostScript language that may more closely match the fonts on the original printer, rather than trying to use the existing PostScript fonts. In either case, when a font is encountered in the original print le, some PostScript font must be used. At the second level, a decision must reached as to where compromises can be made. For instance, for text which is not justied, it may be best to allow the normal letter spacing of the PostScript font to prevail, as long as the lines will be approximately the correct length. However, with justied text (like the text you are reading, with the margins aligned on both sides) the difference in line lengths between the original line of text (set in one font) and the line set in Times-Roman on a PostScript printer must be accounted for. The widthshow and awidthshow operators can be used to stretch a line of text to any arbitrary size, although the result may not necessarily look like the original text. Typographers usually prefer to modify the word spacing rather than the letter spacing (this means using widthshow and stretching only the space character). If the text is not justied, then the line can simply be printed with show and the result can be quite good, depending on the match of the substituted font. (Frequently the results are much better than the original printer may have been able to produce, if the font style is not particularly important.) When the decisions have been made as to roughly how to deal with the original imaging model and font library, there are implementation decisions to be made. For printing text, switching from one font to another tends to be the most difcult task, especially when the text is justied. Depending on the informa-

70

chapter 4 EMULATORS AND TRANSLATORS

tion provided in the original print le, there are two schools of thought: Print all instances of any given font at once, for any particular page, regardless of their positions on the page. Proceed left to right, top to bottom, switching fonts as necessary. If the print format you are translating has location information for each word (or letter, or string) independently, the rst approach may be preferable, especially if there are many font changes. If the existing print format relies on the current position of one piece of text in order to set the next, this is probably not feasible. Sometimes this kind of positioning information can be computed from the original print le, even if it is not inherent in the original format. For instance, the positions of the lines of text (or words of text, given many font changes) must be determined at some point in order to print them. Unfortunately, they are often computed in the printer simply because it seems easier. However, by keeping a width table for all characters of a font, it is fairly easy to compute the position of any word on a page simply by adding together the character widths of preceding characters. The resulting PostScript program can be enough faster to make the rst of these approaches worth considering when the translator is being designed.

4.5 USING THE IMAGING MODEL

To translate a graphics data le into a PostScript print le, it is necessary to correctly translate all aspects of the graphic imaging model used by the original creator software. Whenever graphic print les are produced, assumptions are made about the capabilities of the output device and the semantics of the individual graphic objects (for example, the notion of an object being a circle, not just a series of vectors). Whenever these semantics can be preserved at a high level, the resulting PostScript program is more efcient (and the results are usually better).

chapter 4 EMULATORS AND TRANSLATORS

71

PRESERVING HIGH-LEVEL INFORMATION

If the captions on an illustration or other text matter are preserved as character information in the print le, the strings can be passed directly to the printer and set as text. However, if the text was reduced to a graphic representation such as vectors or pixels as the print le was produced, the semantic notion of text is lost. Similarly, if a line of text in the print le is originally set justied, it means that some computation was made to determine where the words would be placed. If a print le translator can determine that all the words on the line are evenly spaced and were intended to be set justied, an invocation of widthshow may be generated for the entire line of text, rather than many instances of moveto and show. This would result in a signicant performance gain.

RENDERING
The physical properties of a display or marking device heavily inuence the way in which graphics are rendered. PostScript interpreters are currently designed to drive raster devices. In particular, the nal page image is typically represented as a large pixel array. Other devices, notably pen plotters or vector displays, may require an input le containing only the equivalent of moveto and lineto operations. The capabilities of the target device often factor heavily into the graphic imaging assumptions in the print le. For further discussion on adapting to the PostScript imaging model, read Chapter 3.

4.6 OPTIMIZING TRANSLATOR OUTPUT

There are several areas where the code produced by a translator can be optimized. As an example, consider a print le intended for a pen plotter which is restricted in movement along one axis at a time. It is common to see sequences like the following for positioning the pen in a plotting device: 0 150 rmoveto 300 0 rmoveto

72

chapter 4 EMULATORS AND TRANSLATORS

This represents movement rst in the y direction, then in the x direction. By recognizing this sequence in the original print le, it can easily be optimized by combining the motion into a single command: 300 150 rmoveto This is quite a bit shorter than the original, and it will therefore transmit faster and be interpreted faster. The key is to recognize sequences or patterns in the original le that might be represented more compactly in the PostScript language. Another important optimization relates to setting text. Strings should be printed in as long a chunk as is practical; entire lines at a time are best. As an example, consider two approaches to setting a line of text in which the word spacing needs to be modied slightly, as in the following example: % first example: 0 800 moveto (Entire) show 4 0 rmoveto (line) show 4 0 rmoveto (at) show 4 0 rmoveto (a) show 4 0 rmoveto (time.) show % second example: 0 800 moveto 4 0 32 (Entire line at a time) widthshow Even if the rst of these is optimized by putting the rmoveto operation into a short procedure, it is still much slower to reposition between words than to print the whole line, modifying the width of the space using widthshow, as is done in the second part of the example. Chapter 7 contains more detailed information on text setting algorithms.

4.7 COMPUTATION AND DECISION-MAKING

In general, the goal when writing a PostScript language driver is to distill out as much computation as possible from the PostScript program. The PostScript language is extremely

chapter 4 EMULATORS AND TRANSLATORS

73

exible, but it is designed as a nal-form output representation, not as a layout or computation tool. Making decisions at the PostScript interpreter level can also be quite expensive. As usual, it is best to make the decisions on the host computer if possible. Consider the mechanism presented in listing 4-4 for performing line wrapping. This is not a recommended approach, given the amount of computation performed.

listing 4-4
%!PS-Adobe-2.0 %%Title: example of what not do to /SH { %def dup stringwidth pop currentpoint 3 1 roll add 300 gt { %if 72 exch 48 sub dup 36 lt { %if showpage pop 700 } if moveto }{ %else pop } ifelse show } bind def %%EndProlog 72 700 moveto /Times-Roman findfont 48 scalefont setfont (As ) SH (usual, ) SH (it ) SH (is ) SH (best ) SH (to ) SH (make ) SH (the ) SH (decisions ) SH (on ) SH (the ) SH (host ) SH (computer ) SH (if ) SH (possible.) SH showpage %%Trailer

This program takes input text one word at a time and places each string on the page. Right before printing each word, its

74

chapter 4 EMULATORS AND TRANSLATORS

width is computed and added to the current point on the page. If the resulting position would exceed the right margin, that word is moved to the next line. If that new line is below the bottom margin (36), then showpage is rst executed for the current page, and the current point is then moved to the top of the next page. There is a great deal of computation involved in doing this, and it may cause a PostScript printer to run at less than one fourth the speed which it might otherwise achieve if the same page were composed carefully on the host system.

chapter 4 EMULATORS AND TRANSLATORS

75

chapter 5 DESIGNING THE PAGE AND THE PROGRAM

Computers are at a double disadvantage in the production of documents. They are not clever, like secretaries, nor can they read penciled proofreaders marks, like typographers. Brian K. Reid

5.1 INTRODUCTION
An important aspect of designing printer driver software is to adapt the applications graphic data structures to the PostScript language imaging model. If the application has a representation for text, exactly how is it maintained? How are graphics edited, stored, and represented? What might be the relationship between lled regions in the PostScript language and the graphic texturing in the applications own representation? As is the case with any output device, these internal graphic imaging models must be translated into a format appropriate for the printer: the PostScript language. The subject of this chapter is the task of designing the page description as a software application in its own right.

5.2 PAGE LAYOUT CONSIDERATIONS

One of the basic ideas in the PostScript language model is the coordinate system. It is best to think in terms of user space, and not about the actual paperthe job of getting a piece of paper under the coordinate system for printing is almost the last consideration, and the PostScript language is innitely exible about the exact mapping of the user coordinate system onto the physical paper.

chapter 5 DESIGNING THE PAGE AND THE PROGRAM

77

Frequently, one of the rst questions that arises in the early stages of laying out a PostScript language driver is: Where is the top of the page? Many imaging models work from the top of the page with positive y coordinates growing downward, rather than using the lower left corner as the origin. The way to think about this in the PostScript imaging model is to answer: There is no page at the composition stage, only an innite real number coordinate system. A good way to think about page composition is to imagine the desired page at any size: 11 inches by 18 inches, 36 inches by 48 inches, or 8-1/2 inches by 11 inches, it doesnt matter. What matters is the relative positioning of page elements, the overall proportions of the page, and its design. The page can easily be scaled to print at any size and orientation, and on any kind of paper (or on many sheets instead of one) without much trouble. It is most important to consider how the page elements will interact with one another. In fact, the individual page elements should probably interact very little. That is, each page element should be self-contained, and be placed and executed individually. Columns of text (or perhaps even individual lines of text) might be considered to be independent page elements. In this context, independent means that there should be no ripple effects from, say, changing the location of one page element, or selecting a thicker line weight or a different font. This kind of context should be isolated from one page element to another.

PAGE NESTING AND INDEPENDENCE

It is important that all of the individual page descriptions within a document print le maintain functional independence from one another. This means, informally, that you should not dene a procedure in the midst of page 1 and then use it again on page 2. It also means that there should be contextual independence between the pages. There should be no leftover graphics state, no reliance on the contents of a variable that may have been dened on a previous page, and no carry-over of the current font. The motivations for this are many. Most important among them are the ability to reverse the pages for printing, and the ability to print pages in parallel. There is a much fuller coverage

78

chapter 5 DESIGNING THE PAGE AND THE PROGRAM

of this in Chapter 6, but it is mentioned here because it affects page layout design directly. All programs in the PostScript language should be written as if they were subroutines rather than the main program. You should avoid PostScript language operators which are absolute (for instance, initgraphics, grestoreall, defaultmatrix, and so on). For example, a document that is composed to print on standard letter-sized paper should be written in such a way that it can easily be printed 2-up (two half-size page images on a single sheet of paper, side by side). In order to accomplish this, the coordinate system must be scaled to half its original size. If the document executes initgraphics, this effect is nullied, since the default transformation matrix is installed. Even simply printing a page in landscape mode will not work if the le contains a call to initgraphics or grestoreall. There is further discussion of this in Chapter 6. Note: Never initialize or replace the existing state of the interpreter. All changes should be modications to the existing state, including changes to the transformation matrix, the halftone mechanism, the transfer function, and redenitions for operator names. Always concatenate the new procedures or matrices with the existing ones.

5.3 PRODUCING POSTSCRIPT LANGUAGE OUTPUT

The fundamental level of a PostScript driver is the print statements in the host application which actually emit code fragments. This level is often overlooked in the overall design of a PostScript driver, but it is an extremely important area to consider, especially in terms of efciency. It is at this level that the basic translation from the applications internal data structures to the PostScript print format takes place. It is an important place to consider some of the higher-level concerns: Analyze the geometry or properties of the object being generated, and how it relates to the inherent PostScript imaging model.

chapter 5 DESIGNING THE PAGE AND THE PROGRAM

79

 Identify the aspects of this page element that can be considered data, and the order in which these data are needed at the PostScript level. Perform any calculations necessary to convert data from the programs internal representation into the desired PostScript language representation. Decide whether to generate native PostScript code at this stage, or to generate a procedure call to a procedure dened in the prologue. The machine-generated part of the data stream (the script) should be as compact, modular, and efcient as possible. The script segment of a PostScript program is tightly linked to procedure denitions in the prologue, and a large part of the design process for a PostScript language program is to decide what the distribution of labor should be between the prologue and the script. Note: Remember that the PostScript language is interpreted, and that it is a programming language only as one aspect of its design. It is best not to defer calculation (such as division problems, computing the diameter of a circle, or guring the length of some text) to the interpreter. Instead, perform these calculations on the host system as the script is being generated, providing the data to the procedures in the format expected by the PostScript language and the individual operators used.

5.4 ROUND-OFF AND COORDINATE SYSTEMS

Raster output devices are digital. That is, the pixels that make up device space are discrete, and you either hit one of them or you dont. The user space coordinate system in the PostScript language model, however, is a real number coordinate system, based on the printers point (1/72 of an inch). This permits very accurate construction of paths, since the paths are maintained as points in an ideal coordinate space. When the paths are painted, the decision is made as to which of the pixels should be painted.

80

chapter 5 DESIGNING THE PAGE AND THE PROGRAM

Essentially, there are two possible approaches in a digital device such as a laser printer: 1. A moveto or lineto instruction falls on the nearest pixel boundary in device space. 2. A path operation (like moveto) is maintained as a real number. Either of these methods could be adopted. The rst method would help ensure uniformity of line weights, for example, but it would prove far less accurate for constructing curves and other complex shapes (especially character descriptions in fonts). The second method is the one chosen for the PostScript language. One side effect of this model is that stroked lines (or lled shapes) may vary by as much as one device pixel, depending on where the paths fall in device space. There is an easy way to coerce the path construction operators into behaving like the rst example, if desired. (See Section 9.3.) Be careful not to make any assumptions about the resolution or orientation of device space when you use the PostScript language, or your program may not be device-independent.

5.5 EFFICIENCY
The overall efciency of a PostScript program is affected primarily by the following three factors: 1. Data transmission (and tokenizing) time 2. Computation overhead 3. Interpretation time All three of these are of roughly equal importance, although one or the other of them may become more signicant in a particular application or environment. For instance, at low baud rates across a serial connection, data transmission time is likely to be the primary bottleneck, whereas across a high bandwidth connection, it is often less signicant. However, reducing the

chapter 5 DESIGNING THE PAGE AND THE PROGRAM

81

time the scanner takes to tokenize the input stream is benecial at any data transfer rate.

DATA TRANSMISSION OVERHEAD

Transmission and tokenization overhead can be minimized by dening very short procedure names for use in the script section of the document (which is usually the largest part of any PostScript le). Avoiding the retransmission of redundant information can also play a large role (for instance, long font names or extra unnecessary moveto instructions). Aside from the time it takes to transmit the data to the PostScript device, the scan time is proportional to the number of bytes in the data stream. Short identiers can be scanned more quickly than long ones. Compare the following two PostScript language fragments: %!PS-Adobe-2.0 %%EndProlog gsave 10 10 moveto 20 10 lineto 20 20 lineto 10 20 lineto closepath fill grestore showpage %%Trailer %!PS-Adobe-2.0 /gs /gsave load def /gr /grestore load def /m /moveto load def /l /lineto load def /cf { closepath setgray fill } bind def /S /showpage load def %%EndProlog gs 10 10 m 20 10 l 20 20 l 10 20 l 0.5 cf gr S %%Trailer

82

chapter 5 DESIGNING THE PAGE AND THE PROGRAM

The second example is actually longer than the rst, but the initial procedure denitions (the prologue) are only executed once. For a very short document like this, the rst example is faster. However, it would not take very many more boxes to change the balance in favor of the second example, since the amount of information transmitted and scanned for each box is minimized. Notice that, in the previous example, each time a box is drawn a very predictable data stream is produced. In particular, there is always the sequence of moveto lineto lineto lineto closepath fill. An extremely useful approach in program design is to look for patterns like this and to factor out repeated procedure calls from the script. The following fragment denes a procedure called B that combines the functions of m, l, and cf in the previous example: %!PS-Adobe-2.0 /B { %def gsave moveto lineto lineto lineto closepath setgray fill grestore } bind def /S /showpage load def %%EndProlog 0.5 10 10 20 10 20 20 10 20 B S %%Trailer

COMPUTATION
Consider the following illustration of computation overhead. The task at hand is to draw an arc of a circle, but the stored representation for this arc maintained by the composition software is much different than the model used by the native arc operator in the PostScript language. In the original representation, an arc is represented by storing two corners of the bounding rectangle of the arc, and two angles representing the starting and ending points for the arc. (See gure 5.1).

chapter 5 DESIGNING THE PAGE AND THE PROGRAM

83

gure 5.1

300, 300 90 0

100, 100

The arc is stored in the host application simply as these six integers This representation must be converted to the syntax required by the arc operator, which uses the center of the arc, its radius, and the beginning and ending angles to represent the arc segment. Listing 5-1 is a procedure that draws this arc merely by passing in the original six numbers as parameters. Incidentally, this listing is presented as an example of what it is best not to do.

listing 5-1
%! example of how not to draw an arc: /origmtx matrix def /drawarc { %def /EndAngle exch def /StartAngle exch def /URy exch def /URx exch def /LLy exch def /LLx exch def gsave LLx URx add 2 div LLy URy add 2 div translate

84

chapter 5 DESIGNING THE PAGE AND THE PROGRAM

newpath URx LLx sub URy LLy sub origmtx currentmatrix 3 1 roll scale newpath 0 0 0.5 StartAngle EndAngle arc setmatrix stroke grestore } bind def % end of prologue 100 100 300 300 0 90 drawarc 35 10 translate % as opposed to this: 300 200 100 0 90 arc stroke showpage

Contrast this to a simple call to the built-in PostScript arc operator, without any of the attendant computation. The resulting mark on the page is the same: 300 200 100 0 90 arc stroke Refer to Section 3.6 for a good procedure to construct and paint a circle with the arc operator.

INTERPRETATION TIME
Interpretation time is the time spent in the PostScript interpreter when no work is being done directly. For example, the scan time, stack manipulations and name lookup required to execute 0 0 moveto are interpretation time necessary for establishing a current point. Minimizing interpretation time is probably the most effective way to improve the performance of a PostScript program. The shorter the procedure is, and the fewer PostScript operators executed, the more efcient the program will be. This can make an order of magnitude difference in execution time under some circumstances, given the interpreted nature of the language.

chapter 5 DESIGNING THE PAGE AND THE PROGRAM

85

There are several ways to go about improving your PostScript programs efciency, especially the interpreter overhead: Study the listings in this book. The examples throughout the text have been carefully designed to be efcient. Look carefully at the balance between the data passed in the script and the procedures dened in the prologue. Study your script. After you have gotten your program working, take a critical look at the script. Forget the constraints that you are working under, and forget the prologue procedures themselves. Are there any patterns in the output? Can you factor anything out of the script? There are many clues to the performance of your driver locked up in the script for an average document. Look at your prologue procedures. As a rule of thumb, if they are more than a few lines long, they are too complicated. If there are more than four or ve instances of roll, dup, exch, or index, you may want to look more carefully at the data itself. Can it be rearranged? Can a division be performed by the host application? On the other hand, if there are a lot of instances of /variable exch def, look carefully to see if these dictionary entries need to be made. If the data is completely used up by the procedure, there is usually no reason to store the data in a dictionary, and it tends to be much slower to do so. Do a timing test. It is an extremely rare page that should take more than about ve seconds to print. Without exception. This is true even when writing emulators. If your PostScript program is taking longer than that to print a page with text and graphics on it, start simplifying things. The usertime operator can be used for precise timings. Implement special cases. For the most part, procedures that are generalized to handle several possibilities are a bad idea in PostScript programming. It is better to make several simple, slightly different variations and invoke the appropriate one.

86

chapter 5 DESIGNING THE PAGE AND THE PROGRAM

chapter 6 PROGRAM STRUCTURE

Anything thats designed to do more than one thing cant do any of them well. Anonymous

6.1 INTRODUCTION
A PostScript language program contains several levels of structure. The program le conventionally consists of a prologue, a script, and a trailer. The prologue may be composed of several procedure sets, each of which is an independent package of procedures appropriate for a specic task. The script is divided into pages, and each page may contain many page elements. Within the execution of page elements, there are many procedure calls, and there is a structured interaction between these procedures and the PostScript language operators. The trailer section often is empty, but it can be used to restore the environment to its original state at the end of a document. The program structure is important for several reasons. First, the structure affects the execution, as is true with most programming languages. The performance and robustness of the program are, in turn, linked to its execution. PostScript les also should be well-formed, according to specic structuring conventions (see Section 6.4) to cooperate effectively with print spoolers, page reversal programs, and applications wishing to import PostScript les as illustrations. There is a thorough discussion in Chapter 11 of the issues of embedding a complete PostScript document le within another PostScript le.

6.2 THE PROLOGUE AND SCRIPT MODEL

A good model of organization for a PostScript le is to divide it into two basic pieces, known as the prologue and the script. The

chapter 6 PROGRAM STRUCTURE

87

prologue consists of procedure denitions and constants, and should contain only denitions (it should not, for instance, contain any modications to the graphics state). The script is program-generated data along with procedure invocations (calls either to native PostScript language operators or to procedures dened in the prologue). A software developer typically writes the prologue by hand, but the script is generated by software and is usually different for each document printed. The PostScript interpreter does not require the clean break in functionality found in the prologueand-script model, but it is easier to maintain the software, and, more importantly, many kinds of pre- or post-processing of PostScript les depend on this structure. For instance, if the script can be broken cleanly into pages then host software can reverse the pages even after the application has produced a nal print le. (See Section 6.3, MODULARITY AND PAGE STRUCTURE.) Here is a simple example of a simple program written in the PostScript language that has an inherent prologue and script: /F { findfont exch scalefont setfont } bind def 12 /Optima-Oblique F 10 10 moveto (Fairly Oblique) show showpage There is only a single procedure denition (the F procedure), and it must, of course, be dened before it can be executed. It is signicant, however, that the denition precedes all of the executable code. For instance, if the 10 10 moveto were performed rst, the le would still execute correctly, but the clean distinction between the prologue denitions and the executable script would be lost. What follows is the same program with PostScript language comments included to indicate the structure of the le. These are known as structure comments. (See section 6.4.) These comments

88

chapter 6 PROGRAM STRUCTURE

are ignored by the PostScript interpreter, but may be parsed by other software to determine the programs structure:

listing 6-1
%!PS-Adobe-2.0 %%Title: example %%Pages: 1 %%BeginProcSet: TextProcs 1.0 0 /F { findfont exch scalefont setfont } bind def %%EndProcSet %%EndProlog %%Page: one 1 12 /Optima-Oblique F 100 100 moveto (Fairly Oblique) show showpage %%Trailer

These structure comments may be used, for example, as guidelines to reverse the pages or to print only one page of many. The last section in this chapter is devoted to these structure comment conventions, but their use here points out the implicit prologue/script structure in the original example program. Adding the comments identies the program as a conforming PostScript language program. These comments should not be included unless the le adheres carefully to the prologue and script model and conforms to the PostScript Document Structuring Conventions. The version number to which it conforms is included in the very rst line as part of the %!PS-Adobe- comment; in this case, the version is 2.0.

6.3 MODULARITY AND PAGE STRUCTURE

The PostScript language is context-dependent. This means that execution usually depends on previously dened context. The

chapter 6 PROGRAM STRUCTURE

89

nature of the execution model can be reduced to the following operations, which are typically performed in sequence: Set (or add to) the state of the interpreter. This could consist of setting the current font, adding to the current path, or making a dictionary entry. Execute a graphic operation. This includes, for instance, using show to print characters, applying fill to the current path, or executing showpage to print the current page. Note that the execution of PostScript operators depends to a large extent on the current state of the interpreter (in particular, the graphics state). This is an extremely powerful mechanism, but one which must be understood and used carefully. It is not necessary to program defensively, but it is usually good to isolate the interdependency between page elements as much as possible. Modularity is an important aspect of program design. It means isolating pieces of code and making them perform independently, with a rigorously dened interface between them. In the PostScript language, this might seem to be hard to accomplish, given the strong dependency upon context. However, by carefully looking at the ground state of the interpreter, the contextdependency can be used to advantage.

GROUND STATE
The ground state in a PostScript interpreter is the default context in which all jobs are executed. The ground state of the interpreter is guaranteed to have the following properties: The coordinate system (user space) is set up with printers points as the default measure (72 points to a linear inch), with the origin of the coordinate system placed in the lower left hand of the current page. There are no marks on the current page (it need not be erased).

90

chapter 6 PROGRAM STRUCTURE

 There is no current path. (This means that there is no current point, either.) There is no current font. The dictionary stack contains userdict and systemdict (userdict is on the top of the stack). All elements of the graphics state have some default setting (for example, butted line end caps, mitered line joins, and black as the current color). The default graphics state is enumerated fully in the PostScript Language Reference Manual. The current transformation matrix is in its default state (it is machine-dependent). This context is the starting place for all programs executing on the PostScript interpreter. In order to cooperate effectively with the execution environment, the ground state should be taken to be inviolable. That is, since the page is blank, it should not be erased. Since the graphic state is in its default state, one should not execute initgraphics. Only those elements of the interpreters state which need to differ from the ground state should explicitly be set. There are several reasons for this: the programs are simpler; there is usually nothing accomplished by unnecessarily setting default states; if the state of the machine differs from the documented default, it is probably done deliberately. For example, to print pages 2-up requires modication to the current transformation matrix, which initgraphics would reset. Some parts of the graphics state are restored naturally to their default state, and gsave and grestore may not be necessary. For example, the current point is destroyed naturally by the path painting operators. If a procedure constructs a path and paints it without modifying any other aspects of the ground state, then it need not use gsave and grestore. Modularity is best achieved by carefully redening the ground state and returning to it between elements (modules). That way, every page element, every document page, and every proce-

chapter 6 PROGRAM STRUCTURE

91

dure can expect the same initial state. It may be convenient in many cases to use the default state of the PostScript interpreter as the ground state for an application, although a slightly different state can be dened at the beginning of the document and reestablished with save and restore.

THE OPERAND STACK

The PostScript language provides a uniform, rigorously dened interface between all modules of a program, and it is best to use this system to provide functional independence among modules. This mechanism is simply the operand stack. All parameters are passed via the operand stack, and all operators have very carefully dened semantics in terms of what they expect on the stack and what they will return on the stack. This is inherently the way programmers write PostScript procedures and functions, too, but it is important to realize that it is a denition of the interface between program modules, in addition to being the only reasonable way to pass information. Use of the operand stack as both an interface and a data structure is covered in Section 2.3. For purposes of examining program modularity, it is enough to think of it as the communications medium between modules of a program.

FUNCTIONAL AND GRAPHIC INDEPENDENCE

Two kinds of failures can result when program modules are not sufciently independent. They are actually quite different, and must be considered individually: Incorrect functional context (execution errors) Incorrect graphic interdependency (an unintended picture) Most programmers are familiar with the rst of these: if you dont dene a module correctly or pass it the proper parameters, the program fails. the interpreter will usually alert the programmer to this kind of difculty by executing the typecheck or stackunderflow errors, for example, although occasionally the program will continue to execute, even though the operands on the stack may not be the right ones.

92

chapter 6 PROGRAM STRUCTURE

The second kind of failure mode above, which has been termed Incorrect graphic interdependency, is typically a conceptual error on the part of the programmer, and can be paraphrased: It printed, but it did not print as I intended. This is usually the result of a kind of ripple effect that occurs when some aspects of the graphics state carry over from one graphic element to another. For example, if the caption under a drawing is linked to the current point after the drawing is executed, the caption may end up in the wrong place if the drawing is changed. Similarly, if the current font at the end of one paragraph is bold and it is not explicitly re-set to the roman face at the beginning of the next paragraph, the following text may incorrectly print in the bold font. Unfortunately, many programs which are poorly designed only get worse as bugs like this get xed. The resulting program, after debugging, is typically much more complex than it should be, and is often characterized by unnecessary calls to gsave and grestore or explicitly executing 0 setgray at the beginning of every procedure. These kinds of problems usually can be traced to a single procedure that failed to restore the state correctly when it nished executing. This second kind of graphic modularity is perhaps the more important one in terms of designing a driver for an application. Because the order of evaluation for page elements is not enforced when building a page in the PostScript language, it is the programmers responsibility to determine how (and in what order) to execute the various graphic elements on a page. This typically depends upon the composition softwares imaging model, but it must be carefully considered in order to write a modular PostScript language program.

SAVE AND RESTORE

Perhaps the best way to ensure independence between graphic modules is by using the save and restore mechanism provided in the PostScript language. The save and restore operators capture a particular state of the interpreter and restore it exactly as it was. In particular, a well-dened ground state may be saved and reclaimed between each graphic element on a page. Each individual component is then free to modify the graphics state in any way which seems appropriate (including scaling, rotating, changing fonts, and so forth), but these changes are guaranteed

chapter 6 PROGRAM STRUCTURE

93

to be isolated from other page elements by the save/restore mechanism. Since save and restore are somewhat more expensive than gsave and grestore, they should only be used where the contents of VM need to be restored as well as the graphics state. They are appropriate at a fairly coarse level, either at each page boundary, or between page elements that consume memory (like text blocks, since the strings require storage space). There is further information on save and restore in Section 13.3 and in Section 3.10.

PAGE ELEMENTS AND THEIR PROPERTIES

When designing programs in the PostScript language, characteristic graphic elements should be isolated, each of which may have a set of properties associated with it. These properties should be dened to be the set of characteristics that may differ from the ground state. For instance, a line-drawing primitive will have the line weight dened to be one of its properties, since it is useful to be able to provide different line thicknesses. On the other hand, a line-drawing routine need not modify the current font in order to perform correctly. This approach is a formalization of the principle of functional independence. By carefully isolating which aspects of the interpreters state needs to be changed for each graphic procedure, and by making sure the state is returned afterward, the necessary independence can be achieved simply and effectively. With this approach, each page element will likely have a particular position on the page as one of its properties. This corresponds to establishing a current point in the graphics state before the code is executed. In many instances, it seems as though the position of a page element depends (or should depend) on some other page element. Part of the work in designing a driver for an application involves determining the interdependency between page elements, both graphically and programmatically. Listing 6-2 is an example of two routines that have some graphic interdependency. One is called circle, the other square. Each

94

chapter 6 PROGRAM STRUCTURE

of them calls the fill operator, which makes it implicitly dependent upon the current color. In the example, it can be seen that setting the gray value affects all subsequently drawn circles or squares.

listing 6-2
%!PS-Adobe-2.0 %%Title: graphic independence %%EndComments /circle { %def 0 360 arc fill } bind def /square { %def moveto 1 index 0 rlineto 0 exch rlineto neg 0 rlineto closepath fill } bind def %%EndProlog 0.5 setgray 72 100 200 circle 200 300 100 100 square 0 setgray 200 300 300 300 square showpage %%Trailer

Listing 6-3 shows the same two procedures, slightly rewritten to have the color (gray value) of each shape passed as one of its arguments. This effectively removes the interdependency between the objects. If the shapes are likely to be painted with various gray levels, it is appropriate to pass the shade of gray as one of the attributes for those page elements. This also ensures that the ground state is preserved between page elements, and other procedures can rely on the current color to be as originally dened. For each graphic element on a page, for each module of the PostScript language driver, you must decide exactly how the ele-

chapter 6 PROGRAM STRUCTURE

95

ments will behave with respect to one another. A ground state can be dened that is appropriate for all the various elements on a page, and each element can provide any alternate properties that are appropriate.

listing 6-3
%!PS-Adobe-2.0 %%Title: graphic dependence %%EndComments /circle { %def gsave arc setgray fill grestore } bind def /square { %def gsave moveto 1 index 0 rlineto 0 exch rlineto neg 0 rlineto closepath setgray fill grestore } bind def %%EndProlog 0.5 72 100 200 0 360 circle 0.5 200 300 100 100 square 0 200 300 300 300 square showpage %%Trailer

6.4 DOCUMENT STRUCTURING CONVENTIONS

The PostScript Document Structuring Conventions are a formal set of guidelines for PostScript language programs. Program les that observe the guidelines are known as conforming les. The purpose of having conforming le standards is to facilitate document interchange and promote the development of print spoolers and font servers. Many documents are composed on a workstation and printed later on one of perhaps several different printing devices. Being able to identify the structure of the

96

chapter 6 PROGRAM STRUCTURE

le permits greater exibility to adapt the needs of the document to a particular printer. The structuring conventions have a tangible representation in the form of PostScript language comments that are embedded in the program le, which delineate its structure. These comments are intended to be parsed and understood by software, rather than humans. The format of the comments should be strictly followed for this reason. It is also understood that a le which claims to be a conforming le (by having the appropriate comments embedded) should indeed conform to the structuring conventions. The PostScript document structuring conventions are fully documented in a separate document available from Adobe Systems entitled PostScript Document Structuring Conventions. They can also be seen in context in most of the program listings in this book, although their use here is minimal, to keep the listings from being lengthy and hard to read. The purpose of the comments is to convey information about the program to parsing software that may be interested in determining its structure (or its resource requirements, such as downloadable fonts). The structuring rules, including page modularity, are required for parallel processing, page reversal, and other document handling needs. Please refer to Chapter 10 for further information on PostScript le interchange standards.

chapter 6 PROGRAM STRUCTURE

97

chapter 7 THE MECHANICS OF SETTING TEXT

You shall see them on a beautiful quarto page, where a neat rivulet of text shall meander through a meadow of margin. Richard Brinsley Sheridan

7.1 INTRODUCTION
Setting text is one of the most common operations performed by many printing applications, and one of the most difcult. There are many aspects of typography and graphic design that can be difcult to accommodate in a computer program. For example, a typesetter may expect justied text, even when it contains different fonts or point sizes, kerning of the text based both on the point size and on specic pairs of letters, and all at the full rated speed of the printer. Setting non-Roman text or using font characters as symbols presents another class of problems but uses the same mechanisms in the PostScript language. Many of the requirements of setting text are in fact document formatting issues. For instance, establishing margins, laying out lines, or choosing a text point size are decisions that are typically made by a word processing application or some other layout software. Writing a PostScript language driver is the task of converting the document formatters representation of text into a printable form. This chapter is devoted to the mechanics of setting text in the PostScript language. A primary focus is efciency: how to get the job done as fast as possible. Most of the topics are specic to proportional fonts and traditional typography, although the concepts should apply as well to monospaced fonts and special requirements.

chapter 7 THE MECHANICS OF SETTING TEXT

99

Unfortunately, optimizing the text-setting portion of a driver depends to some extent on the nature of the text being set. It may be that an algorithm that works well for setting many lines of text in the same font becomes too slow when fonts are changed frequently, or that a method that assumes line spacing is uniform may be difcult to adapt to display typography. Several different approaches and algorithms are presented in this chapter, and the software designer must choose which ones are appropriate for a particular task. In many instances, it is best to build in several mechanisms, and to invoke one or the other of them depending on the text being set. Note: There is one principle to keep in mind when deciding upon an algorithm for setting text. The longer the string presented to one of the show operators, the more efcient the system is likely to be. This is because the PostScript language built-in operators, such as show, widthshow, and ashow, operate essentially at compiled speed once they have been invoked. Each moveto or div operation performed must rst be interpreted, which is signicantly slower.

7.2 CHARACTER WIDTHS

Formatting text requires access to character width information. In order to decide where to break a line of text, one needs to be able to compute the length of the line of text by adding up the widths of the individual characters in the line. In order to center a heading on a page, one needs to know what the width of the heading is. Line breaks and centering are formatting issues. Making a decision to break a line is a document-wide decision, since it can affect page breaks or paragraphs, which in turn can affect the rest of the document. Under most circumstances, the formatting of a document is left to document composition systems, and is not appropriate at the printing stage. The standard approach to document formatting is to make character metric information available in a format that can be used by

100

chapter 7 THE MECHANICS OF SETTING TEXT

host composition software. All a word processor needs to be able to use a new font is a table with the widths of all the characters in the font. In a screen display system, a set of screen fonts may be required, as well, although the pixel widths of the screen characters may or may not be accurate enough for careful typesetting. Usually both width tables and screen fonts are required. Adobe Systems typefaces are made available with separate metrics les in a le format designed to be parsed by application software. These les are known as AFM les (for Adobe Font Metrics). A complete description of the format of these les is found in a separate document available from Adobe Systems, entitled AFM Files: An Interchange Format for POSTSCRIPT Font Metrics. These les also contained detailed kerning and ligature information. (Their use is discussed in Section 7.6.) The width of a character is dened to be the amount by which the current point is modied after printing the character. For Roman character sets, the width typically includes the left and right sidebearings, which are a comfortable amount of space on each side of the character for good letter spacing. Figure 7.1 is an illustration of four characters (including a space character) from the font StoneSerif (which is also the font you are reading), that shows the character widths and sidebearings.

gure 7.1
character bounding box

character width (displacement)

w at
sidebearings
101

Notice that the physical width of the two characters is not quite the same as the optical width of the characters, due to the side-

chapter 7 THE MECHANICS OF SETTING TEXT

bearings. This is often ignored in ordinary typography, although it can be handled by obtaining the bounding box of the characters at each end of the string to determine what the side-bearings are. The bounding box provides the corners of an imaginary rectangle that just touches all the extremes of the character shape (hence bounding). This information is given as an offset from the character origin, and therefore provides the left side-bearing directly, and can be subtracted from the width of the character to determine the right side-bearing. The bounding box of each character is also provided in the AFM les.

7.3 MARGINS AND JUSTIFICATION

In setting columns of text (or body copy) there are often margins observed on each side of the lines of text. These margins are usually specied as the amount of white space between the text and the edges of the paper, but this is not a very specic measurement when a document may be printed on more than one size of paper. A better measure probably is to specify the starting point of the text, and the line length of the text. These starting point for the text is certainly needed for setting with the PostScript language. There are four basic types of line layout that are in common use: centered, ush left, ush right, and justied. These are also sometimes referred to as centered, leftjustied, right-justied, and fully-justied. In Figure 7.2 is an example with four text blocks, each of which is set in one of these styles. One might also want to indent the rst line of a paragraph of text or provide unusual borders around which to ow text. These can all be handled relatively easily if both the starting point of the text and the particular line length are maintained by the composing application.

102

chapter 7 THE MECHANICS OF SETTING TEXT

gure 7.2
an excerpt from
Then this ebony bird beguiling my sad fancy into smiling, By the grave and stern decorum of the countenance it wore, Though thy crest be shorn and shaven, thou, I said, art sure no craven, Ghastly grim and ancient Raven wandering from the Nightly shore Tell me what thy lordly name is on the Nights Plutonian shore! Quoth the Raven Nevermore.

The Raven
by Edgar Allan Poe February, 1845
Much I marvelled this ungainly fowl to hear discourse so plainly, Though its answer little meaning little relevancy bore; For we cannot help agreeing that no sublunary being Ever yet was blessed with seeing bird above his chamber door Bird or beast upon the sculptured bust above his chamber door, With such name as Nevermore.

But the Raven, sitting lonely on the placid bust, spoke only, That one word, as if his soul in that one word he did outpour. Nothing farther then he utterednot a feather then he uttered Till I scarcely more than muttered Other friends have own before On the morrow he will leave me, as my Hopes have own before. Quoth the Raven Nevermore.

In listing 7-1 some procedures are dened that implement rightshow and centershow. These procedures do a fair amount of calculation, and are only appropriate if the widths of the strings cannot be determined at the host (for instance, they might be used in a printer emulator). Generally, the host application will already have determined precisely where the text should be placed, and can simply use moveto and show for all of the above styles. These procedures combine the moveto and show operations. Both the x, y values of the location on the page and the string to be printed are passed on the operand stack. This is faster than doing separate name lookup and execution on both moveto and show. Notice that the margins are maintained by the host application, and an explicit location on the page is passed for each string.

chapter 7 THE MECHANICS OF SETTING TEXT

103

listing 7-1
%!PS-Adobe-2.0 %%Title: margin procedures %%BeginProcSet: text-procs 1.0 0 % moveto-show /SH { %def moveto show } bind def % rightshow /RS { %def moveto dup stringwidth neg exch neg exch rmoveto show } bind def %centershow /CS { %def moveto dup stringwidth 2 div neg exch 2 div neg exch rmoveto show } bind def %%EndProcSet: text-procs 1.0 0 %%EndProlog /Times-Roman findfont 14 scalefont setfont (Centered about + current point) 306 140 CS (+Flush Left at current point) 72 120 SH (Flush Right at current point+) 560 120 RS showpage %%Trailer

JUSTIFICATION
For standard text-setting, the natural letter spacing and word spacing should be used. Letters may be kerned to improve their visual spacing (see section 7.6), but the lines of text should usually not be stretched in any way unless it is necessary to justify the text. There are two levels of renement for justifying a column of text:

104

chapter 7 THE MECHANICS OF SETTING TEXT

 Provide careful hyphenation and word breaking to get the lines of text as close as possible to the correct length. Modify the word spacing as necessary to stretch the text to t the necessary width of the line. Under normal circumstances, the inter-letter spacing should not be modied at all. The legibility of text tends to suffer greatly when letter spacing is compromised. If hyphenation is not available, or when it cannot be avoided, both word and letter spacing can be modied simultaneously with the awidthshow operator. The appropriate way to modify word spacing is to use the widthshow operator, which was designed with this purpose in mind. widthshow prints text just as show does, except that it modies the character width of a particular character by the specied amount while it is printing it. By supplying a small amount by which to change the width of the space character, word spacing can be modied quite efciently and easily. Listing 7-2 contains an example of its use.

listing 7-2
%!PS-Adobe-2.0 %%Title: widthshow example /F { findfont exch scalefont setfont } bind def /W /widthshow load def %%EndProlog %%BeginSetup /sp 32 def % ASCII space %%EndSetup 36 745 moveto 24 /Times-Italic F 6 0 sp (PostScript Language Program Design) W showpage %%Trailer

chapter 7 THE MECHANICS OF SETTING TEXT

105

The computation that needs to be performed is to determine the amount by which to modify the width of the space character. For justied text, the space modier (SpaceMod) is the intended line width (LineWidth) less the width of the existing string (StringWidth) divided by the number of spaces in the string (Spaces). This computation can easily be performed by the host application if the character widths are known. SpaceMod = (LineWidth - StringWidth) Spaces This SpaceMod quantity is passed directly to the widthshow operator each time it is called. Notice that this value will be different for every line of text, since there are likely to be different numbers of spaces (and different string widths) for each line.

7.4 HANDLING DIFFERENT FONTS

Justifying lines of text in which there may be font changes (or point size changes) requiring several instances of show require more careful computation to determine the exact position of each of the components. The larger the string that is provided to show, the more efcient the driver will be. It is best to break each line into as few pieces as necessary to accomplish the task. For instance, using widthshow to modify word spacing is superior to using repetitive moveto and show operations. Listing 7-3 contains an approach for setting a single line of text with multiple font changes. The method used is to show all parts of the line that are in the same font at the same time. This means rst nding, say, all the roman parts of the string and showing them (put each substring on the stack with its individual x position). Then showing all the bold parts (they could as well be a completely different font), and so on. This is a somewhat unusual approach, and requires some careful setup work on the host system, but it will likely be quite fast when there are many font changes. It also provides a natural framework for handling downloadable fonts, since the text is grouped by font used, not
106 chapter 7 THE MECHANICS OF SETTING TEXT

by context on a line. Figure 7.3 shows the output of this program.

listing 7-3
%!PS-Adobe-2.0 %%Title: mixed-font text setting /FS { %def findfont exch scalefont } bind def /SF { %def Fonts exch get setfont } bind def /MS { %multiple "show" counttomark 2 idiv { %repeat 0 moveto show } repeat pop mark } bind def /LINE { %def gsave translate mark % mark is for MS procedure } bind def /S { %def moveto show } bind def /END { %def pop grestore } bind def %%EndProlog %%BeginSetup /Fonts [ %def 12 /Times-Roman FS 12 /Times-Bold FS 12 /Times-Italic FS ] def %%EndSetup %%BeginObject: text_block 0 SF save % text block ground state % first line of text 72 312 LINE % first line start (Here is a line with ) 0

chapter 7 THE MECHANICS OF SETTING TEXT

107

(words and ) 140.315 (words.) 238.991 MS 2 SF (two ) 192.971 (italic ) 112.644 MS 1 SF (bold ) 213.311 (two ) 90.984 MS END % lines 2, 3, and 4 in the text: (You shall see them on a beautiful quarto) 72 288 S (page, where a neat rivulet of text shall) 72 276 S (meander through a meadow of margin.) 72 264 S % credit for quotation: 160 240 LINE 2 SF (Richard Brinsley Sheridan) 0 MS END restore % to ground state %%EndObject showpage %%Trailer

gure 7.3 Here is a line with two italic words and two bold words. You shall see them on a beautiful quarto page, where a neat rivulet of text shall meander through a meadow of margin. Richard Brinsley Sheridan

108

chapter 7 THE MECHANICS OF SETTING TEXT

7.5 LEADING AND POINT SIZE

Leading is the distance between lines of text. In the days of metal (lead) type, it referred to actual strips of lead placed between two rows of type to increase the spacing between them. In contemporary computer typesetting, the value typically refers to the baseline-to-baseline distance between two lines of text, rather than the distance from the bottom of one line to the top of the next. Within the context of this book, leading refers to the distance between baselines. There are practically no rules governing the height of individual letterforms in digital fonts. The point size of a font is the metric by which the font is chosen and set. Originally, the point size of a lead font was a measure of the slug in which the letterform was cast. Today, digital outline fonts are not restricted at all in terms of point size, and the size does not necessarily reflect the height of the characters, the bounding box, or anything else. Although information is typically provided for the capital height, the x-height, and perhaps the ascender and descender heights, these metrics are not guaranteed to be consistent even across any individual font, and especially not from one font to another. In general, no assumptions should be made about the relationship between the actual height of characters a font and the point size. (See gure 7.4.)

gure 7.4

capital height

Hjx
descender

point size
x-height

chapter 7 THE MECHANICS OF SETTING TEXT

109

7.6 KERNING AND LIGATURES

Kerning is a modication to inter-letter spacing of text. Kerning is applied with proportionally spaced typefaces for purely visual reasons: either because it looks better, or because it improves the legibility of the text. There are two varieties of kerning: pairwise kerning and track kerning. Pairwise kerning is an adjustment to the inter-letter spacing between a particular pair of letters. There is usually a table of values stored for each font. Since the motivation for kerning is purely visual, there is no easy way to do it algorithmically. Track kerning is a change to the inter-letter spacing of an entire line of text, based on the point size of the font used. For instance, smaller type should be set slightly wider than normal, and very large type (display type) should be spaced closer together. Again, the reasons are optical. (See gure 7.5.)

gure 7.5

AWAY A AY W
110

standard letter spacing

tightly kerned letter spacing

Kerning, at its simplest, involves an adjustment of the current point after the rst character of a pair. This is effectively the same as temporarily changing the character width, depending on the following character. The second character is set normally, although it may itself be a candidate for beginning another kern pair. The rmoveto operator is suited to adjusting the current point, as in the following example. However, it should be worked carefully into whatever text-setting algorithm is used.

chapter 7 THE MECHANICS OF SETTING TEXT

%!PS-Adobe-2.0 %%EndComments /S /show load def /r /rmoveto load def %%EndProlog /StoneSerif findfont 48 scalefont setfont 100 100 moveto (A) S -5.5 0 r (W) S -6.8 0 r (A) S -5 0 r (Y) S showpage There is a bit of unnecessary interpreter overhead for each character printed, in this example. This can be tightened up a bit, using the kshow operator; notice that the values for rmoveto are based on the user-space coordinate system. If the text were set at a different point size, these values would be different, as in the following example: %!PS-Adobe-2.0 %%EndComments /K { %def { pop pop 0 rmoveto } exch kshow } bind def %%EndProlog /StoneSerif findfont 48 scalefont setfont 100 100 moveto -5 -6.8 -5.5 (AWAY) K showpage This procedure minimizes some of the interpreter overhead, since fewer name lookups are performed, and since bind is performed on the procedure body. The main drawback of this approach is that kshow will always execute the procedure between every pair of characters in the string. If no kerning is desired, 0 must still be supplied on the stack (resulting in the unnecessary execution of 0 0 rmoveto). The viability of this approach depends on the amount of kerning that needs to be done. If only an occasional pair of characters is kerned, this is not a good approach. However, if every letter is positioned individually, the K procedure above might be a good choice.

chapter 7 THE MECHANICS OF SETTING TEXT

111

Another approach to kerning is to use a negative width space character, and actually put the character in between the pair of letters to be kerned. The show operator will then automatically adjust the position based on the width of the kern character. The only difculty with this is that in order to provide enough exibility in kern values, many space characters of varying widths are required. It is also more difcult for the composing application to insert the extra space characters wherever kerning is required. Listing 7-4 contains a good hybrid approach. A line of text is set by breaking it into pieces that are as long as possible, and placing all the strings on the operand stack along with some extra information. There are three variations on the theme: stackshow, which prints several strings on the stack, each with a different font (supplied on the stack); kernstackshow, which performs the same operation but permits a value for rmoveto between strings; kernshow, which permits kerning values but not font changes. Each of them provides for a fast approach to setting text, where the component strings are made as long as possible, depending on the lines of text in the host application.

listing 7-4
%!PS-Adobe-2.0 %%Title: text-procs.ps %%Creator: Glenn Reid, Adobe Systems %%EndComments %%BeginProcSet: general 1.0 0 /F { findfont exch scalefont setfont } bind def /Fdef { findfont exch scalefont def } bind def /M /moveto load def %%EndProcSet: general 1.0 0 %%BeginProcSet: text-procs 1.0 0 % mark (text) fontdict (text) fontdict ... stackshow /stackshow { %def % reverse stack order first: 2 2 counttomark 2 sub { -2 roll } for counttomark 2 idiv { %repeat setfont show } repeat pop } bind def

112

chapter 7 THE MECHANICS OF SETTING TEXT

% mark (Txt) font rX rY ... kernstackshow /kernstackshow { %def 4 4 counttomark 2 sub { -4 roll } for counttomark 4 idiv { %repeat rmoveto setfont show } repeat pop } bind def % "kernshow" is like "stackshow" except that it does % not use a font dictionary for each string. It % should be used when the kerned text is all set in % the same font. % mark (text) rX rY (text) rX rY .... kernshow /kernshow { %def 3 3 counttomark 2 sub { -3 roll } for counttomark 3 idiv { %repeat rmoveto show } repeat pop } bind def /MM { moveto mark } bind def %%EndProcSet: text-procs 1.0 0 %%EndProlog %%BeginSetup /F1 50 /StoneSerif Fdef /F2 50 /StoneSerif-Semibold Fdef /F3 50 /StoneSerif-Italic Fdef %%EndSetup %%Page: 1 1 F1 setfont 36 430 MM (S) 0 0 (W) -4 0 (EPT A) -4 0 (W) -8.0 0 (A) -9.25 0 (Y) -6.75 0 kernshow 36 380 MM (S) F1 0 0 (W) F1 -4 0 (EPT ) F1 -4 0 (A) F3 -4 0 (W) F3 -8.0 0 (A) F3 -9.25 0 (Y) F3 -6.75 0 kernstackshow showpage %%Trailer

chapter 7 THE MECHANICS OF SETTING TEXT

113

Ligatures are also used to improve the visual quality of text. These are specially designed characters that comprise two (or even three) sequential letterforms. For instance, the two letters f and i, when set together, may be combined into a single shape: . This can affect composition software since the width of the ligature is usually different than the sum of the widths of the f and i characters. (See gure 7.6; the font is Stone Serif-Italic.) Ligatures cannot usually be added to a fontthey are part of the original font design. However, the application should recognize and replace sequential instances of, say, f and i, with the single ligature character. Spelling checkers, hyphenation tables, and justication algorithms need to know about them, as well.

gure 7.6

fig. ve
7.7 ENCODING AND CHARACTER SETS

Use of the ligature

A typical PostScript language font program is a collection of procedures, each with a particular name, stored into a dictionary. Each of these procedures, when executed, draws one character shape. There are other entries in a font dictionary, but all of them are used to set up or execute one of the character-drawing procedures. The collection of character procedures that are stored in a font is known as the character set of that font. There can be any number of these character procedures stored in the font dictionary. There is no limitation (other than available memory) on the total number of characters. There is, however, a limitation on how many of them can be accessed at any particular moment.

114

chapter 7 THE MECHANICS OF SETTING TEXT

Access to font characters is controlled through the encoding mechanism. The characters in a font currently must be accessed by character code. A string presented to the show operator can be thought of simply as a sequence of ASCII bytes, each of which is actually just an index into the encoding vector for the current font. The encoding vector is an array of 256 name objects, each of which should be the name of a procedure stored in the font dictionary. The show operator actually presents the integer character code to the BuildChar procedure in each font, and it is up to that procedure to invoke the correct character description. This mechanism was deliberately set up to provide an easy way to change the mapping between the codes in the string and the characters that are actually selected from the font. Each font dictionary is required (by definefont) to have an array named /Encoding as one of its top-level entries. To reencode a font, copy the font dictionary, put in a different Encoding vector, and rename the font. Listing 7-5 contains a good reencoding algorithm. It is designed to avoid having to provide redundant data. A block of sequentially encoded names can be specied by providing only the initial encoding valuethe others are assigned in sequence.

listing 7-5
%!PS-Adobe-2.0 %%Title: reencode.ps %%EndComments /F { %def findfont exch scalefont setfont } bind def %%BeginProcSet: reencode 1.0 0 % This file defines a procedure called "R" which % reencodes a font. It expects three objects on the % stack: % % [ array ] /NewName /OldName % % The array should contain pairs of % <number> <name>, % like "32 /space", each of which defines a slot in the % encoding and the name to put in that slot. Only

chapter 7 THE MECHANICS OF SETTING TEXT

115

% those names that are needed to over-ride the % existing ones should be specified. An encoding % value (number) may be specified followed by more % than one name, like "128 /name1 /name2". % In this case, the names will be sequentially stored % in the encoding starting at the initial number % given (128). /RE { %def findfont begin currentdict dup length dict begin { %forall 1 index /FID ne {def} {pop pop} ifelse } forall /FontName exch def dup length 0 ne { %if /Encoding Encoding 256 array copy def 0 exch { %forall dup type /nametype eq { %ifelse Encoding 2 index 2 index put pop 1 add }{ %else exch pop } ifelse } forall } if pop currentdict dup end end /FontName get exch definefont pop } bind def %%EndProcSet: reencode 1.0 0 %%EndProlog %%BeginSetup /stdencoding [ 39/quotesingle 96/grave 128/Adieresis/Aring/Ccedilla/Eacute/Ntilde /Odieresis/Udieresis/aacute/agrave/acircumflex /adieresis/atilde/aring/ccedilla/eacute/egrave/ecircumflex /edieresis/iacute/igrave/icircumflex/idieresis/ntilde/oacute /ograve/ocircumflex/odieresis/otilde/uacute/ugrave /ucircumflex/udieresis/dagger/.notdef/cent/sterling/section /bullet/paragraph/germandbls/registered/copyright /trademark/acute/dieresis/.notdef/AE/Oslash/.notdef /.notdef/.notdef/.notdef/yen/.notdef/.notdef/.notdef/.notdef /.notdef/.notdef/ordfeminine/ordmasculine/.notdef/ae/oslash /questiondown/exclamdown/logicalnot/.notdef/florin /.notdef/.notdef/guillemotleft/guillemotright/ellipsis/.notdef /Agrave/Atilde/Otilde/OE/oe/endash/emdash/quotedblleft /quotedblright/quoteleft/quoteright/.notdef/.notdef /ydieresis/Ydieresis/fraction/currency/guilsinglleft /guilsinglright/fi/fl/daggerdbl/periodcentered /quotesinglbase/quotedblbase/perthousand/Acircumflex /Ecircumflex/Aacute/Edieresis/Egrave/Iacute/Icircumflex /Idieresis/Igrave/Oacute/Ocircumflex/.notdef/Ograve /Uacute/Ucircumflex/Ugrave/dotlessi/circumflex/tilde

116

chapter 7 THE MECHANICS OF SETTING TEXT

/macron/breve/dotaccent/ring/cedilla/hungarumlaut /ogonek/caron ] def stdencoding /_StoneSerif /StoneSerif RE stdencoding /_StoneSerif-Italic /StoneSerif-Italic RE %%EndSetup 90 /_StoneSerif-Italic F 50 220 moveto (fig. \336ve) show showpage %%Trailer

7.8 COMPOSITE CHARACTERS AND ACCENTS

Many languages require accented characters that may not be included in the standard character set. Most of the necessary characters can be derived from combining existing glyphs in the font. These are known as composite characters. Some composite characters are already included in fonts from Adobe Systems, but are not encoded. (They are not present in the StandardEncoding vector that is used in most fonts.) These can be accessed by merely reencoding the font. (See the previous example program). However, for character combinations which are not already included in the font, some extra steps must be taken to set up the fonts.

7.9 NON-ROMAN FONTS

Many of the assumptions that are made about setting text can be challenged by a font that does not contain a standard Roman alphabet as its character set. In particular, a font may in fact be a collection of graphic objects, each of which can call upon the entire richness of the PostScript language during its execution. Notions of point size, leading, and kerning can take on a different meaning when fonts are used for such purposes. In many ways, display typography requires placement of letterforms as graphic objects, and some of the generalized rules for placement given in this section apply directly to display type.

chapter 7 THE MECHANICS OF SETTING TEXT

117

CHARACTER WIDTHS AND ORIGINS

The correct way to think about a character in a PostScript font is that it is anchored at the character origin, and that the current point is displaced by the character width vector after the glyph has been drawn. Remember that character widths in PostScript fonts have both an x and a y component. The current point may be modied as appropriate for specic character sequences. For example, let us look at a few characters from a music notation font, Sonata. There are characters in the font which are designed much like accents are designed in other fonts: They are intended to be used as composite elements to build up other characters. Figure 7.7 shows three characters from Sonata: the quarternotehead, the upstem, and the sxflagup. Alongside them is the composite character that can be built by painting these characters sequentially. The sixteenth ag character and the upward stem both have a width displacement of 0 in both x and y. That means that the current point does not move at all after the character has been printed. This permits characters to be superimposed one on top of another. It also provides perfect registration of the shapes, with no possibility of position roundoff error. The nal notehead character does have a width, and the current point is left just to the right of the notehead character after it is printed. Notice also the offset from the origin that is inherent to the two composite characters. They are designed to register exactly with the character.

118

chapter 7 THE MECHANICS OF SETTING TEXT

gure 7.7

\ k \k
upstem sxflagup quarternotehead sixteenth note When the three characters are set together, a sixteenth note results. Using one additional character (extendflagup) provides the capability of building arbitrarily complicated notes. In this case, the current point is adjusted between characters, but is restricted to vertical movement. The alignment of the characters is guaranteed by their design. (See gure 7.8.) Working with text characters which are in fact used more as graphic symbols than as conventional text requires some careful thought. The same design guidelines apply for using setfont and show with Sonata as they would with StoneSerif. The fewer operations (and the more characters that can be printed at once with a single call to show), the better the performance will be. There are applications for which specialized fonts can be a big advantage. For instance, CAD/CAM applications may build a library of symbols that may be used within a diagram. If they are used as font characters, then the speed of the font cache can dramatically decrease printing time for diagrams with many symbols. The font should be carefully designed so that it can be used with maximum efciency, and the character widths and origins planned around where the current point should be.

chapter 7 THE MECHANICS OF SETTING TEXT

119

gure 7.8

\ k
extendflagup composite sixty-fourth note

Arranging characters on a page is the realm of word-processing or page composition software. In order to work with non-Latin character sets, many of the built-in assumptions about text must be relaxed. For example, word wrapping, line spacing, and hyphenation are very closely tied to languages which read left to right, which set characters on a common baseline, and which have very little variation in vertical placement of characters. Setting text in Arabic, Kanji, Urdu, Mathematics, or other languages require much more sophisticated placement algorithms and judgement. Working with these languages in the PostScript language can be reduced to two characteristic issues: Knowing the origin of the characters is crucial to placing it in the correct position. The character width affects the placement of the following character. Adjusting the current point slightly after printing a character is called kerning when setting most languages with the Roman alphabet. Kerning is often considered optional and it is a renement to typesetting. In other languages, the current point may need to be adjusted between every two characters on the page, depending on the context, and it is not considered optional. This is not really a kerning problem, but an issue of placement. But the same principles hold as with all typesetting

120

chapter 7 THE MECHANICS OF SETTING TEXT

done with the PostScript languageif more than one character can be passed to show at once, there will be a signicant performance gain over individually placing each character. Some of the burden for this lies on the type designer, and some on the careful placement algorithms used by the layout software.

chapter 7 THE MECHANICS OF SETTING TEXT

121

chapter 8 SCANNED IMAGES AND HALFTONES

This affair must all be unraveled from within. He tapped his forehead. These little gray cells.... Agatha Christie

8.1 INTRODUCTION
The image operator is the mechanism in the PostScript language for printing sampled images of any size in up to 256 shades of gray. These images may be printed at any size and orientation with a minimum of calculation. The PostScript language also has a built-in digital halftoning system that permits full control of the rendering of gray shades with halftone screens. This chapter takes an in-depth look at the image operator and the halftone mechanism.

8.2 THE IMAGE OPERATOR

The image operator takes data representing an image made up of many small samples and renders it on the current page at a specied location, size, and rotation. The data represents an image that is a certain number of samples wide and high (represented herein by the width and height, respectively). Each sample is represented by a one-, two-, four-, or eight-bit value that assigns a shade of gray to that sample. Each sample will be eventually printed as a black, white, or gray rectangle on the output page. (Color PostScript devices will have additional operators to implement full color images.) The image operator requires ve arguments on the operand stack. These are as follows:

chapter 8 SCANNED IMAGES AND HALFTONES

123

 The width of the image, in samples. For an 8-bit deep image, this number is the same as the number of bytes in one row of source data. The height of the image, in samples (or scan lines). The number of bits per sample in the original data. A black-and-white scanned image has 1 bit per sample, a grayscale image may have 2, 4, or 8 bits per sample. The image matrix, which is a transformation matrix that maps user space onto the coordinate system implied by the scanning process in the original data. The data acquisition procedure, which fetches sample data and leaves it on the operand stack as a string object.

HOW IT WORKS
The image operator multiplies width times height times bits per sample to determine how many bits of data are needed to render the image. It then executes the data acquisition procedure as many times as necessary, imaging each string directly into device space as it is produced by the procedure. The mapping provided by the image matrix is applied as the data is being imaged.

THE IMAGE MATRIX

Sampled data (perhaps from a digitized photograph) has an implicit coordinate system imposed on it by the scanning process. The rst sample scanned is, in effect, in position 0,0. The second sample, assuming a horizontal scan, would be in position 1,0 and so on. When the image operator prints this sampled image in user space, it assigns a particular position in user space to each sample. Thus, there is a coordinate transformation between user space and sample space that is involved in the printing of a sampled image. The image matrix that is passed to the image operator is the transformation matrix that provides the transformation

124

chapter 8 SCANNED IMAGES AND HALFTONES

between the region in user space the nal printed image should occupy and the region in sample space that the image originally occupied. Since either the user space transformation or the image matrix may be modied independently, there are an innite number of combinations that will produce the desired image. It is best to hold one of the matrices as a constant entity, and adjust the other one based on the image being rendered. Here are some steps that will simplify use of the image matrix, and provide a generally correct mapping for any image data into the user coordinate space. Use translate, rotate, and scale to alter user space so that a one-unit square at the origin would occupy a space on the current page exactly where you want the image to be printed. That is, if the lower-left corner of the image is to be 3 inches to the right and 10 inches above the current user space origin, and the image is to occupy a space 1 inch wide and 2 inches tall, precede the call to the image operator with the lines save 216 720 translate 72 144 scale Invoke the image operator, supplying an image matrix that prints the sampled image in the one-unit square at the origin. This matrix is a very easy one to calculate. Images scanned bottom-up and top-down require the following image matrices, respectively: [ w 0 0 h 0 0 ] if scanned from the bottom [ w 0 0 -h 0 h ] if scanned from the top where w and h are the width and height of the sample data in samples (not points). For a more complete discussion of the image matrix, see Section 4.7 of the PostScript Language Reference Manual. Provide a data acquisition procedure. This is covered more fully in the next section.

chapter 8 SCANNED IMAGES AND HALFTONES

125

DATA ACQUISITION PROCEDURES

The data acquisition procedure is a PostScript language procedure body used by the image operator. It is the responsibility of this procedure to obtain some amount of sampled data, put this data into a string, and leave the string on the operand stack. The image operator will call the data acquisition procedure as many times as necessary, each time using the string left on the operand the stack, until it has received an amount of data equal to the (width x height x bits-per-sample ) as provided to the image operator. The image operator processes the data within the string most signicant bit rst. Thus, the string (T) left on the stack would be interpreted by the image operator as the string of bits 01010100 This is the binary representation of the decimal number 84, the ASCII character code of the T character. These bits would be interpreted 1, 2, 4, or 8 at a time to represent the gray value of successive samples (as specied by the bits-per-sample value passed to the image operator). The data acquisition procedure used in a particular instance depends upon the amount and nature of the data making up the image. When using strings to represent binary data, careful calculation is necessary to match the amount of data provided to the amount of data that the image operator expects to receive. This is especially tricky if 2 or 4 bits per sample are used. For example, let us suppose an image is 10 samples wide and 8 scan lines high, and each sample consists of 4 bits of grayscale information. The call to image will look something like this: 10 8 4 [ 10 0 0 8 0 0 ] { datastring } image

126

chapter 8 SCANNED IMAGES AND HALFTONES

The width and height of the image are handed directly to the image operator on the operand stack. The datastring, however, must be of the appropriate length to match the dimensions of the image. Strings contain 8-bit bytes. To represent 10 x 8 x 4 bits of image data, 320 bits of information are needed, which requires a 40 byte string to contain the 320 bits.

SMALL AMOUNTS OF DATA

If the image is made up of a very small amount of dataa few dozen bytes or lessit may be most convenient to let the data acquisition procedure place the string directly on the stack. That is, the procedure may look something like this: { <013f8e40ae183d022f> } The angle brackets in PostScript language syntax denote a hexadecimal (base 16) string. Each pair of characters represents, in hexadecimal, eight bits of information, or a single byte in the string. Thus, the above example would leave a nine-character data string on the operand stack, or 72 bits worth of image data. This form is most appropriate for small images such as printed representations of screen characters or icon bit maps. The PostScript Language Tutorial and Cookbook has several examples of this kind of data acquisition procedure.

LARGE AMOUNTS OF DATA

Printing a scanned photograph or other large sampled image requires much more data than can be placed in virtual memory at once. In these cases, the data should be sent to the printer in line with the PostScript program, immediately following the call to the image operator. The data acquisition procedure must arrange to read the input stream, placing the incoming data into a string in conveniently sized increments. The following program fragment incorporates such a data acquisition procedure: /DataString 512 string def 512 340 8 [ 512 0 0 -340 0 340 ] {

chapter 8 SCANNED IMAGES AND HALFTONES

127

currentfile DataString readhexstring pop } bind image f6b94f... This program fragment prints an image consisting of 512 columns of 340 rows each, where each image pixel has a gray scale specied by an 8-bit value. The image matrix indicates that the data represents an image scanned from the top down. The readhexstring operator takes a string object and a le object (in this case the current le, which is the le object that is currently being read by the PostScript interpreter). It reads ASCII hexadecimal data from the le and places the data into the string until either the string is full or the end of le has been reached. The readhexstring operator returns a string object on the operand stack containing data read from the le, and a boolean object which is false if the end of le was reached before the string was lled and true otherwise. In the previous example, the boolean value returned by readhexstring was popped from the stack for efciency, since the procedure will be called many times by the image operator. For debugging purposes, or for better error recovery, it is best to interrogate the resulting boolean and provide some sort of error message if the end-of-le indication was reached prematurely. The readhexstring operator takes the incoming data two characters at a time, interpreting each pair and producing a single byte in the resulting the data string. It ignores all characters that are not valid hexadecimal digits, so that the data may include tabs, spaces, newline characters, and so forth. The ASCII hex representation of the sample data should immediately follow the call to the image operator, as in our example above. The data acquisition procedure is called repeatedly, taking characters from the input stream until the entire 512-by340-by-8-bit image has been processed. At that point, the input stream will be passed back to the PostScript interpreter and the original program can resume.

128

chapter 8 SCANNED IMAGES AND HALFTONES

A COMMON ERROR AND ITS CAUSE

A common problem when using the image operator results from a mismatch between the amount of data supplied and the size of the buffer used by the readhexstring operator. In particular, if the height times width of the image is not evenly divisible by the size of the buffer, readhexstring will keep reading the input stream, looking for as much hexadecimal as it can nd (remember that it just skips anything that is not valid hexadecimal data). This usually means that the word showpage (or any other part of the program) is treated as data, and the a and e characters from showpage are read into the string buffer as hexadecimal values (the others are ignored). If readhexstring encounters an end-of-le indication, it returns the boolean false. However, as in the example above, this value is ignored (by popping it), and the premature end-of-le condition is not recognized. The symptom of this is that the page simply does not print. Nothing in this scenario provokes an execution error, because each component of the system is behaving as it is supposed tobut the result is not what was intended.

SYNTHETIC DATA
The data handed to the image operator can be generated directly by the program. For example, listing 8-1 supplies a synthetic data string to the image operator each time the data acquisition procedure is called; in this case, the string is the same each time and represents a gray-scale fountain with 256 possible gray levels (although not all of them will be attainable on low-resolution devices). This somewhat unusual approach builds a string of 256 bytes, each of which ranges from 0 to 255 sequentially (the for loop accomplishes this, although cryptically). When the image operator uses this string, each image sample is an 8-bit value, and the values range continuously from 0 to 255 each of these image samples is interpreted as a gray value which is then rendered with the halftone mechanism. The actual number of gray levels achieved in the fountain will depend on the constraints of the current halftone screen frequency and the resolution of the output device, but the transition between them will be as smooth as possible.

chapter 8 SCANNED IMAGES AND HALFTONES

129

listing 8-1
%!PS-Adobe-2.0 %%EndComments /DataString 256 string def /IM { %def gsave translate scale image grestore } bind def %%EndProlog 0 1 255 { DataString exch dup put } bind for 1 256 8 [1 0 0 256 0 0] { DataString } 72 72 144 36 IM showpage %%Trailer

The translate and scale operations in the IM procedure locate the image and stretch it to the desired size. The image is rendered as a single sample in width, with a height of 256 samples. The image matrix provided maps this into the unit square in user space, and the scale stretches it to be exactly a 1-inch square on the nal page. (See gure 8.1.)

gure 8.1

Note that, although this image is 1 sample wide and 256 samples high, the printed size of the image is 1 inch square. Remember that the size of the printed image is determined by the arguments passed to the scale operator, not by the number of rows and columns in the sampled data.

130

chapter 8 SCANNED IMAGES AND HALFTONES

8.3 DATA COMPRESSION

Since the image operator provides a generalized mechanism for acquiring data (the data acquisition procedure), the data may be compressed on the host system and decompressed by the PostScript interpreter after it has been transmitted to the PostScript device. This may drastically reduce transmission time, but it is quite difcult to make the procedure fast enough to realize a net gain in overall evaluation time.

8.4 HALFTONE SCREENS

Traditional printing methods cannot vary the tone or shade of ink placed on paper. That is, printing technology allows only one shade of ink to be placed on the paper at a time. Multi-color jobs are printed in several passes, one for each color, or they simulate shades of color by halftoning. Halftoning is familiar to anyone who has looked closely at newspaper photographs. This technique reproduces shades of gray in a nal print with small black dots whose sizes are determined by the desired shade of gray. The larger the dots, the darker the gray that the eye will perceive. Another technique is to use very thin lines that grow thicker with increasingly dark gray levels. Traditionally the way this was accomplished was to lay a halftone screen across the image and photograph the original through the screen. A halftone screen had a certain number of black dots or lines per inch, and were referred to by this number (for instance, a 60 line screen). This number is reected in the frequency parameter that is provided in the PostScript language halftoning mechanism.

HALFTONING IN THE POSTSCRIPT LANGUAGE

The PostScript language halftone mechanism parallels the traditional methods both in terminology and in technology. The screen may be specied at various frequencies and angles, and the shape of the spot can be specied through an executable function.

chapter 8 SCANNED IMAGES AND HALFTONES

131

The PostScript interpreter prints shades of gray with a digital approximation of true halftoning. The dots that a PostScript printer produces are grouped into a square called a halftone cell. On a 300 dot-per-inch printer using a 60-line halftone screen and an angle of 0 degrees, this cell is 5 device pixels on a side. Each gray level is rendered by turning more pixels on in the halftone cell, and a gray area is provided by tiling the area with these halftone cells. Most PostScript interpreters by default print black within the halftone cells in such a way that each halftone cell looks like a single dot that expands as the gray gets darker, as in the fountain in gure 8.1. This is similar to the results of traditional halftoning; it is known informally as a dot pattern.

CHANGING THE HALFTONE SCREEN

The PostScript language allows you to change the halftone screen by which it represents grays with the setscreen operator. This operator takes three arguments: Frequency: the number of halftone cells per inch. Angle: the angle from the horizontal by which the halftone cells are rotated. Spot Function: a procedure body that denes the shape of the halftone spot. In listing 8-3 is a short program that changes the frequency of the halftone mechanism to 20 lines per linear inch.

132

chapter 8 SCANNED IMAGES AND HALFTONES

listing 8-2
%!PS-Adobe-2.0 %%EndComments /F { findfont exch scalefont setfont } bind def /setF { %def currentscreen 4 -2 roll pop % remove existing frequency 3 1 roll setscreen } bind def %%EndProlog gsave 20 setF .4 setgray 250 90 moveto 300 /StoneSerif-Italic F (&) show grestore showpage %%Trailer

Figure 8.2 shows a gray character rendered the screen frequency was set to 20 as in listing 8-2.

chapter 8 SCANNED IMAGES AND HALFTONES

133

gure 8.2

8.5 THE SPOT FUNCTION

&
0,1 -1,0 1,0 0,-1

The setscreen operator considers the halftone cell to be centered on a set of coordinate axes, with the cell extending for one unit in each direction. (See gure 8.3.) The spot function supplied to the setscreen operator must calculate a priority for each location within the halftone cell.

gure 8.3

134

chapter 8 SCANNED IMAGES AND HALFTONES

The spot function is handed the x and y coordinates on the operand stack that represent the location of the center of each device pixel found in the halftone cell. The function must calculate a priority value in the range -1 to 1 for each position. The halftone mechanism will turn on spots within the halftone cell in order from high to low priority. For example, the following call to setscreen causes the PostScript interpreter to render gray levels with a line screen, instead of a dot screen: %!PS-Adobe-2.0 %%EndProlog /DataString 256 string def 0 1 255 { DataString exch dup put } bind for 40 45 { pop } setscreen 72 72 scale 1 256 8 [1 0 0 256 0 0] {DataString} image showpage The spot function in this case consists only of a pop operation. The function is called with the stack holding the x and y coordinates (with y on top) of a position within the halftone cell. The pop removes the y coordinate from the stack so that the priority of any position in the halftone cell is its x coordinate. That is, the farther to the right a position is within the halftone cell, the higher its x coordinate and the sooner the printer will print a spot at that position.

gure 8.4

-.8 -.4 -.8 -.4 -.8 -.4 -.8 -.4 -.8 -.4

0 0 0 0 0

.4 .4 .4 .4 .4

.8 .8 .8 .8 .8

chapter 8 SCANNED IMAGES AND HALFTONES

135

The resulting priorities for a 5-by-5 halftone cell (the default in many PostScript interpreters) are shown in gure 8.4. These priorities render gray areas as a series of lines of varying thicknesses. The gradual blackening of the halftone cell is performed pixel by pixel by the PostScript interpreter, with pixels of highest priority (as determined by this spot function) being blackened rst as the gray level is varied from white to black.

136

chapter 8 SCANNED IMAGES AND HALFTONES

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

If the only tool you have is a hammer, you tend to see every problem as a nail. Anon.

9.1 INTRODUCTION
This chapter focuses on some of the more complex graphic imaging problems that may be encountered when implementing a driver for a graphics application. Some of the topics covered in this chapter involve graphic operations that are not indigenous to the PostScript imaging model. Nonetheless, there are graphics applications that need to render these effects on PostScript printers. Emphasis is placed on the most efcient methods for solving these graphics problems with the standard PostScript language imaging model.

9.2 PATTERN FILLS

Filling a region with a particular pattern currently presents some difculty in the PostScript language. The fill operator invokes the halftone mechanism to ll a region with some particular color, but the halftone machinery is not intended for generalized pattern lls. In particular, halftone screens are rendered directly in device space, which makes it difcult to make the screens device-independent. The halftone mechanism is the fastest method of providing a pattern ll as long as the pattern is not changed frequently. Since the halftone screen is intended to be set once and forgotten, much of the overhead is done when setscreen is invoked. Even if only one gray level is needed, they may all be computed

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

137

(by calling the spot function as many times as necessary to build each possible halftone cell). This makes it fairly expensive to change the halftone screen. Halftone screens are also very device-dependent by nature, and do not rotate or scale with the current transformation matrix. Also, the screens are oriented in device space, and they probably will not work the same way on different devices. Another approach to pattern lls is to use a pattern font. A character (or a number of characters) in a font may be dened to represent the ll pattern, and a region may be tiled by showing text within the boundaries of the region. Unfortunately, under most circumstances this requires use of the clip operator to establish the boundaries of the region to be lled before the pattern text is painted. This presents two difculties: Clipping can be quite slow if the clipping region is complex (the region to be lled with a pattern, in this case). The available path space can overow when clip is used on a complex path (yielding a limitcheck error). However, the pattern font provides a general solution to deviceindependent pattern lls, and is a recommended approach. In listing 9-1 are some sample PostScript procedures which implement pattern lling algorithms. The approach is to dene a font with pattern characters in it, then select one of those characters for tiling. The actual ll process uses clip to establish a clipping region, then uses show with rows of pattern characters to tile the area. Care is taken to assure that the pattern will lock into device space. This keeps the pattern from scaling or rotating, and helps guarantee that there will be no stitching problems between rows of the pattern.

listing 9-1
%!PS-Adobe-2.0 %%Title: patternfill.ps %%EndComments %%BeginProcSet: patternfill 1.0 0

138

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

% width height matrix proc key cache % definepattern -> font /definepattern { %def 7 dict begin /FontDict 9 dict def FontDict begin /cache exch def /key exch def /proc exch cvx def /mtx exch matrix invertmatrix def /height exch def /width exch def /ctm matrix currentmatrix def /ptm matrix identmatrix def /str (12345678901234567890123456789012) def end /FontBBox [ %def 0 0 FontDict /width get FontDict /height get ] def /FontMatrix FontDict /mtx get def /Encoding StandardEncoding def /FontType 3 def /BuildChar { %def pop begin FontDict begin width 0 cache { %ifelse 0 0 width height setcachedevice }{ %else setcharwidth } ifelse 0 0 moveto width 0 lineto width height lineto 0 height lineto closepath clip newpath gsave proc grestore end end } def FontDict /key get currentdict definefont end } bind def

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

139

% dict patternpath % dict matrix patternpath /patternpath { %def dup type /dicttype eq { %ifelse begin FontDict /ctm get setmatrix }{ %else exch begin FontDict /ctm get setmatrix concat } ifelse currentdict setfont FontDict begin FontMatrix concat width 0 dtransform round width div exch round width div exch 0 height dtransform round height div exch round height div exch 0 0 transform round exch round exch ptm astore setmatrix pathbbox height div ceiling height mul 4 1 roll width div ceiling width mul 4 1 roll height div floor height mul 4 1 roll width div floor width mul 4 1 roll 2 index sub height div ceiling cvi exch 3 index sub width div ceiling cvi exch 4 2 roll moveto FontMatrix ptm invertmatrix pop { %repeat gsave ptm concat dup str length idiv { %repeat str show } repeat dup str length mod str exch 0 exch getinterval show grestore 0 height rmoveto } repeat pop end end } bind def

140

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

% dict patternfill % dict matrix patternfill /patternfill { %def gsave clip patternpath grestore newpath } bind def % dict patterneofill % dict matrix patterneofill /patterneofill { %def gsave eoclip patternpath grestore newpath } bind def % dict patternstroke % dict matrix patternstroke /patternstroke { %def gsave strokepath clip patternpath grestore newpath } bind def % dict ax ay string patternashow % dict matrix ax ay string patternashow /patternashow { %def (0) exch { %forall 2 copy 0 exch put pop dup false charpath currentpoint 5 index type /dicttype eq { %ifelse 5 index patternfill }{ %else 6 index 6 index patternfill } ifelse moveto 3 copy pop rmoveto } forall pop pop pop dup type /dicttype ne { pop } if pop } bind def

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

141

% dict string patternshow % dict matrix string patternshow /patternshow { %def 0 exch 0 exch patternashow } bind def /opaquepatternfill { %def gsave 1 setgray fill grestore patternfill } bind def /square { %def gsave moveto dup 0 rlineto dup 0 exch rlineto neg 0 rlineto closepath findfont % a pattern font patternfill grestore } bind def %%EndProcSet %%EndProlog %%BeginSetup 15 15 [300 72 div 0 0 300 72 div 0 0] { %definepattern 2 setlinecap 7.5 0 moveto 15 7.5 lineto 0 7.5 moveto 7.5 15 lineto 2 setlinewidth stroke } bind /RIGHTdiagonal true definepattern pop 15 15 [300 72 div 0 0 300 72 div 0 0] { %definepattern 2 setlinecap 7.5 0 moveto 0 7.5 lineto 15 7.5 moveto 7.5 15 lineto 2 setlinewidth stroke } bind /LEFTdiagonal true definepattern pop

142

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

30 30 [300 72 div 0 0 300 72 div 0 0] { %definepattern 2 2 scale 2 setlinecap 7.5 0 moveto 15 7.5 lineto 0 7.5 moveto 7.5 15 lineto 7.5 0 moveto 0 7.5 lineto 15 7.5 moveto 7.5 15 lineto 0.5 setlinewidth stroke } bind /crosshatch true definepattern pop %%EndSetup %%Page: 1 1 /RIGHTdiagonal 72 460 100 square /LEFTdiagonal 72 480 120 square /crosshatch 72 500 140 square showpage %%Trailer

9.3 LOGOS, GRIDS, FORMS, AND SPECIAL FONTS

It is often desirable to create a graphic object that will be printed on every page of a document, such as a company logotype, a printed form (like an income tax form), or perhaps a background grid of some sort. Depending on the complexity of the image, this can be accomplished in one of several ways. The rst (and simplest) approach to this problem is to place the necessary PostScript code to draw the image into a procedure, and to invoke the procedure at the start of each page. This is the most general solution, and if the complexity of the image is not too great, this is the best approach. Listing 9-2 provides a procedure that will produce a simple logotype. The start of each page can then begin with an invocation of this logo procedure. This will help save the transmission and scanning overhead for each page, but the object will still need to be executed on each page.

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

143

listing 9-2
%!PS-Adobe-1.0 %%Title: logo procedure %%EndComments /logo { %def % draw at current point gsave /Times-Roman findfont 48 scalefont setfont (G) show -13.5 -14.5 rmoveto (R) show grestore -4 -8 rmoveto /Helvetica findfont 11 scalefont setfont (glenn) show 22 3 rmoveto (reid) show } bind def %%EndProlog gsave 100 100 moveto logo grestore showpage %%Trailer

The copypage operator is designed to print the current page without disturbing the contents of the frame buffer into which the page has been imaged. This permits reusing the image that already exists on the page. A form of electronic mask white may be used from one page to the next to erase the elds on the form and print new data in those areas. (This can be accomplished simply by setting the current color to white and using fill.) The only difculty with using copypage is that it doesnt allow fast PostScript interpreters to take advantage of parallel processing. In most implementations of the PostScript interpreter there is a provision for writing bits into one part of the frame buffer while the other end is being imaged onto paper. In extremely fast printers there may be many different frame buffers and a high degree of parallel processing. If the user program forces reuse of the same frame buffer by using copypage,

144

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

this parallelism may be defeated (and the printer will run much slower than its rated capacity). There is a trade-off between using copypage and executing the procedure on each page. The use of copypage is discouraged in all but the most unusual situations. The third approach that can be used to create a reusable graphic image is to make it a character in a user-dened font. The font may contain only a single character, or perhaps many such images. Each image is simply a procedure that is invoked by the font machinery as if to draw a character shape. The advantage of this method is that if the image is an appropriate size to t into the font cache, then it will be extremely fast. In listing 9-3 is the logo procedure from the listing 9-2, placed into font format and invoked by a single-character show operation:

listing 9-3
%!PS-Adobe-2.0 %%Title: logo font example %%EndComments /F { findfont exch scalefont setfont } bind def %%EndProlog %%BeginSetup %%BeginFont: Logo-Font 1.0 0 12 dict begin % leave one empty slot for FID /FontName /Logo-Font def /FontMatrix [ 1 0 0 1 0 0 ] def /FontType 3 def /FontBBox [ -5 -12 85 50 ] def /BuildChar { %def exch begin 70 0 -5 -12 85 50 setcachedevice Encoding exch get load exec end } bind def /Encoding 256 array %def dup 0 1 255 { /.notdef put dup } for pop dup 71 /logo put % "G" character def /logo { %def % draw at current point 20 0 moveto

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

145

gsave /Times-Roman findfont 48 scalefont setfont (G) show -13.5 -14.5 rmoveto (R) show grestore -4 -8 rmoveto /Helvetica findfont 11 scalefont setfont (glenn) show 22 3 rmoveto (reid) show } bind def currentdict end dup /FontName get exch definefont pop %%EndFont: Logo-Font %%EndSetup %%Page: 1 1 gsave 100 100 moveto 1.5 /Logo-Font F (G) show grestore showpage %%Trailer

GRIDS
Producing a grid of lines can result in some slightly unexpected results, due to the rounding of line widths as lines are stroked. (See also Section 5.4.) The following sequence produces a thin line on the page: .1 setlinewidth 72 72 moveto 572 72 lineto stroke However, the exact thickness of the line depends on where the moveto and lineto coordinates fall in device space. PostScript maintains the full precision of coordinates all the way through to device space, and it is not until the execution of stroke or fill that decisions are made for painting individual device pixels. Figure 9.1 is an illustration with a moveto lineto stroke sequence for a short line segment. It shows the interior of the stroked path and the pixels that are blackened by stroke.

146

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

gure 9.1
actual line on paper stroke path current path device space

72 72 moveto

72 90 lineto stroke

Two procedures are dened in listing 9-4 (redenitions of the moveto and lineto operators) that provide lines of uniform thickness, although their spacing cannot also be guaranteed. The crux of these procedures is the use of the transform and itransform operators. The coordinates for moveto and lineto are actually transformed into device space and rounded off to the nearest device pixel boundary. The coordinates are then inverse-transformed back into user space, and the subsequent moveto or lineto is executed. The points are then guaranteed to land on pixel boundaries, which will cause stroke to produce a line of a uniform thickness regardless of where on the page the lines are located. Notice that the coordinates in device space are not inspected in any way; they are merely rounded off and returned to user space. This keeps the code from becoming device-dependent, since no assumptions are made as to the resolution. Figure 9.2 shows the line that results when the M and L procedures are used instead of moveto and lineto. Since the locations are rounded to the nearest device pixel, the stroke operation always encompasses the same number of pixels to either side of the path, resulting in an even line.

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

147

listing 9-4
%!PS-Adobe-2.0 %%Title: line thickness example %%EndComments /M { %def transform round exch round exch itransform moveto } bind def /L { %def transform round exch round exch itransform lineto } bind def %%EndProlog gsave .1 setlinewidth 10 10 800 { %for pop 0 10 translate 550 0 M 612 0 L stroke } bind for grestore showpage %%Trailer

148

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

gure 9.2
actual line on paper stroke path current path device space

72 72 moveto

72 90 lineto stroke

9.4 TRANSFORMATION MATRICES

Coordinate system transformations, including translate, rotate, scale, and concat, are matrix operations that affect the current transformation matrix that exists between user coordinate space and device coordinate space. This mapping is represented by a six-element array (or matrix). All coordinate system transformations are combined with the existing transformation simply by multiplying the existing matrix by the elements of the new transformation. Scaling and translating are easy. There are specic elements of the matrix which directly affect these operations: [ Xscale 0 0 Yscale Xtranslate Ytranslate ] In the following code fragments, the rst line in each pair is equivalent to the second line in the pair: 3 5 scale [ 3 0 0 5 0 0 ] concat 100 300 translate [ 1 0 0 1 100 300 ] concat

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

149

100 200 translate 2 2 scale [ 2 0 0 2 100 200 ] concat The concat operator allows you to supply a complete matrix which is to be concatenated with the existing transformation matrix. Each of the scale and translate operations in the previous example is easily represented by specifying the appropriate matrix and using concat. The scale, rotate, and translate operators are provided for convenience, but are not strictly necessary. Rotation of the coordinate space is less straightforward. The entries in the matrix are actually derived by taking the sin or cos of the desired angle of rotation: [ cos(angle) sin(angle) -sin(angle) cos(angle) 0 0 ] The following is an example of the combined effects of translation and rotation. In order to set landscape mode, the coordinate axes must be rotated 90 degrees and translated so that the origin remains in the lower left-hand corner of the page. Both of the following PostScript segments will place a standard letter sized paper in landscape mode: 90 rotate 0 -612 translate 612 0 translate 90 rotate Notice that in both cases the sequence 90 rotate is used, but the translation component is not the same. Note: These coordinate system transformations are not commutative. That is, if the order of evaluation is changed, different results may be achieved. By convention, landscape pages should always be effected by rst rotating (positive angle) then translating (negative Y value): 90 rotate 0 -612 translate

150

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

As a convention, it is best always to accomplish rotation of a page into a landscape orientation in this manner (positive 90 degrees for the rotation component, then a negative translation component along the Y axis). This is so that if the page is nested in another page or scaled, the landscape orientation can be preserved easily.

INVERTED COORDINATE SYSTEMS

It is possible to invert the coordinate system so that positive y values grow downward, for instance. This particular transformation can be accomplished as follows: 0 792 translate % top of page 1 -1 scale This moves the origin to the upper left corner of the (lettersized) page, and ips the y axis. Note that although graphics can be easily inverted this way, text requires some special attention to keep it from being inverted, as well. One way to address this is to use the makefont operator instead of scalefont: /Times-Roman findfont [ 12 0 0 -12 0 0 ] makefont setfont The -12 in the scale matrix inverts the text about the y axis, which will counteract the inversion of the entire user space coordinate system, causing the text to be right-side up. Inverting the coordinate system is a specialized approach, and probably should be used only when it is difcult or impossible to use the standard orientation.

9.5 COLOR AND COLOR SEPARATIONS

Printing in color using the PostScript language model provides some very powerful capabilities. Colors can be specied with a reectance model using percentages of using cyan, magenta, yellow, and black (using the setcmykcolor operator), which correspond to the four colors used in full process color printing. Color can also be specied using a luminance model, using either the hue, saturation, and brightness model (with the sethsbcolor oper-

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

151

ator) or the red, green, blue color model (using the setrgbcolor operator). For color printing, the reectance model is generally used, with cyan, magenta, yellow, and black as the component colors. Color PostScript printers provide direct support for this color model through the setcmykcolor operator and through related operators to control four separate halftone screens and transfer functions. A colorimage operator is also provided for rendering full color images. Once a color has been specied in the user program, it is rendered through a PostScript language abstraction similar to the black-and-white halftoning mechanism. A current color is established in the graphics state, and all the painting operators use the current color as they apply paint to the current page. The PostScript language interpreter is responsible for producing that color on a given output device. If the device is a screen, the color must be displayed through available screen phosphors. With color printing devices, the color may be produced by successive passes of cyan, magenta, yellow, and black, where each pass uses the appropriate percentage of color as specied by the user program.

COLOR SEPARATIONS
If a PostScript program contains color specication, it can be printed on any PostScript interpreter, and the device will render the color to the best of its ability. However, for full process color presses, you may wish to separate the image into its four components of cyan, magenta, yellow, and black. This can be done by producing four photographic masters (camera-ready plates), one for each process color. These plates are actually printed in black and white using the standard halftoning mechanism, and each plate is then overprinted with the appropriate process color ink in the nal press run. Color separations may be obtained simply by redening the color operators to essentially throw away all but a single process color. The le must then be printed in four passes, where each pass sifts out a different process color. (See listing 9-5.)

152

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

listing 9-5
%!PS-Adobe-2.0 %%Title: separation.ps /separations 24 dict def separations begin /cmykprocs [ %def % cyan { pop pop pop 1 exch sub setgray } % magenta { pop pop exch pop 1 exch sub setgray } % yellow { pop 3 1 roll pop pop 1 exch sub setgray } % black { 4 1 roll pop pop pop 1 exch sub setgray } ] def /screenangles [ %def 105 % cyan 75 % magenta 0 % yellow 45 % black ] def end % separations % setupcolor takes 0, 1, 2, or 3 as its argument, % for cyan, magenta, yellow, and black. /CYAN 0 def /MAGENTA 1 def /YELLOW 2 def /BLACK 3 def /setupcolor { %def userdict begin dup separations /cmykprocs get exch get /setcmykcolor exch def separations /screenangles get exch get currentscreen exch pop 3 -1 roll exch setscreen /setscreen {} def end } bind def %%EndProlog %%BeginSetup CYAN setupcolor %%EndSetup %%BeginDocument: originalfile.ps % entire file is placed here. %%EndDocument %%Trailer

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

153

The halftone screen angles used are extremely important to good color separations. Each process color component should be printed with a different screen angle to minimize color interference in the nal press run. The exact angles used are devicedependent and should be determined separately for each different PostScript device. The goal is to provide placement of color spots that will not interfere with one another, and which will avoid moir patterns as the screens are overlaid.

SPOT COLOR
Spot color is much different than four-color process printing. If you imagine real printing presses, the kind you put ink into instead of toner, you can imagine using a colored ink instead of using black. That is a good model of spot color. It is a single color of ink, rather than a color that is composed as some mix of the four process colors cyan, magenta, yellow, and black. There are two sides of spot color: specifying the color and actually printing it. If and when there are devices which permit printing with two or three (or more) individual colors on a single device, then those colors must be specied by the user program in much the same way that shades of gray are specied with the setgray operator. More commonly, spot color will be mastered on a black-and-white device and printed on a high-speed press with colored inks. From the PostScript interpreters point of view, it will be printing only in shades of black, although the masters may eventually be printed with a colored ink when taken to press. Probably the simplest way to specify spot color is to be able to selectively print components of a document. For example, if a user decides that all chapter headings in a book are to be printed in a particular color of green, and the rest of the book is to be printed in black, he or she needs some way to print the document in two individual passes: the green pass and the black pass. To produce the green pass, only those objects in the document which are specied as green should be printed at all (the black material is simply not printed). An application program can permit tagging of objects to be one color or another, and provide a mechanism for printing only selected objects by naming them. In this way the user can

154

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

selectively print some part of the document to obtain a separation. The PostScript interpreter, of course, is simply interpreting and printing graphics in some shade of black or white, and nothing at all is required at the PostScript level. If the application produces PostScript code that is sufciently modular, each graphic element on the page can be bracketed by the special comments %%BeginObject and %%EndObject. These comments simply provide a way to name the graphic elements on the page. A post-processing program could then selectively extract certain elements of the page for printing. (See the PostScript Document Structuring Conventions document from Adobe Systems for more information on these comments). The PostScript imaging model, however, can complicate the real world of spot color. For instance, if a black rectangle is drawn and lled and a green circle is printed on top of it, the PostScript imaging model completely overwrites the black with the green. However, if the page is printed in two passes, with one pass containing only the black elements and the other pass only the green ones, the black rectangle will not have a white circle cut out of it where the green will go. When green ink is actually printed on top of the black ink on a regular press, some muddying of the color will result. To be done correctly, the black pass should have color removed wherever another color would be overlaid on top of it. This is a much more difcult thing to accomplish, and cannot be done by simply printing individual page elements selectively. This kind of separation problem can also occur in four-color process printing, if the overlaid colors are printed without rst removing any colors which may precede them. This nuance of color printing can be accommodated at the PostScript language level simply by redening some of the color specication operators. Rather than extracting and printing individual objects on a page, all of them may be sent to the printer. To print a separation, the color operators are redened to print black wherever, say, green is specied, and to print with white if any other color is called for. Then, in any place where the green would be obscured by another color above it, that second color would be printed in white, effectively erasing that part of the green image. The technique for this is much like that

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

155

found in listing 9-5. The difference is the selection processrather than selecting only the cyan pass, some method of selection by name must be implemented in order to specify the spot color. The full details of this specication and separation technique are beyond the scope of this book. For more detailed information, please contact someone at Adobe Systems.

156

chapter 9 COMPLEX GRAPHIC PROBLEM-SOLVING

chapter 10 FILE INTERCHANGE STANDARDS

Many of the greatest things man has achieved are not the result of consciously directed thought, and still less the product of a deliberately coordinated effort of many individuals, but of a process in which the individual plays a part which he can never fully understand. Friedrich August von Hayek

10.1 INTRODUCTION
PostScript language print les that are intended to migrate between different environments should observe standard structuring conventions, and they should be device independent. The standardized structure permits document managers to incorporate the documents into other systems and to print them in the appropriate environment. It also makes parallel processing possible and provides a mechanism for ensuring that fonts and other resources are supplied as needed.

10.2 CONFORMING DOCUMENTS

A conforming document is a PostScript document le that correctly observes the structuring conventions. There are two aspects to a conforming document le: Comments: The structure of the le should be delineated (as appropriate) by structure comment conventions. These are PostScript comments in a specic format that can be parsed by applications software to determine the structure and contents of a PostScript le. (See Section 6.4.) The comments include an introductory com-

chapter 10 FILE INTERCHANGE STANDARDS

157

ment that marks a le as a conforming le (including a version number). Structure: The structure of the le should be in conformance with the le structuring guidelines. This includes page modularity (pages should be functionally independent) and the avoidance (or careful use) of system-level PostScript language operators such as initgraphics and erasepage. The rules for conformance are not particularly limiting. Conforming les are, for the most part, ordinary PostScript language program les. The notion of conformance is introduced as a guarantee to applications software that the le plays by the rules.

10.3 HANDLING PRINTER-SPECIFIC FEATURES

In order for PostScript les to be truly portable, any PostScript code that invokes a specic device attribute (such as a paper tray or frame buffer) must be clearly segregated in the structure of the le. Print le managers (or spoolers) can take action to replace a device-dependent module from a program le, but only if these sections can be identied easily. In order to provide a general mechanism for invoking printerspecic features, Adobe Systems has provided PostScript Printer Description Files, which collectively set forth a machine-parsable le format containing a representation for the PostScript program fragments needed to invoke particular features on each printer. These les are arranged by keywords for each feature, and document management software can easily extract the necessary PostScript to set a paper tray, invoke manual feed, or determine whether or not a certain font is present in ROM. Document composition software should produce a PostScript document le that contains an indication of what printerspecic features are needed (or what features have been invoked in that le already, and might need to be changed if the file is sent to a different printer). This indication is in the form of PostScript comments:

158

chapter 10 FILE INTERCHANGE STANDARDS

%%BeginPaperSize: Ledger statusdict begin ledgertray end %%EndPaperSize This PostScript language sequence correctly invokes the ledgertray operator on any printer that has this operator in statusdict. If a document manager wishes to send the le to a different printer than was originally intended (that may have a slightly different method for invoking the Ledger paper size), it should remove the code between these comments, and use the keyword Ledger to nd the appropriate PostScript language segment from the printer description le, with which it will replace the previous code sequence. Even if it happens to be sent to another printer for which ledgertray is the appropriate operator, it doesnt hurt to replace it.

10.4 SPECIFYING PAPER SIZES

Formatting a document is different from printing a document. A document formatting program makes decisions about where page elements will fall on the printed page. However, in the PostScript language imaging model, the document formatter lays pages out in user space. It is easy enough to modify the mapping between user space and device space to achieve almost any kind of printing need. Because of this exibility in printing, there are really two different decisions that need to be made to choose a page size. The rst decision is made when the PostScript print le is formatted. If the intended page size is, say, 11 by 17 inches (ledger paper), then the numbers in the PostScript le address user space coordinates anywhere from 0 to 1224 in the x direction. Let us draw a simple box around the intended page to illustrate this. The box represents a collection of marks that could be an entire document on ledger-size paper. The important thing is to have coordinates in the program that are in the range of 1224 in the x direction. (See listing 10-1.) Notice the use of the comments to provide information about the program, especially the %%BoundingBox comment.

chapter 10 FILE INTERCHANGE STANDARDS

159

listing 10-1
%!PS-Adobe-2.0 %%BoundingBox: 0 0 1224 792 % % <width> <height> <llX> <llY> <thick> box /box { %def gsave setlinewidth moveto dup 0 exch rlineto exch 0 rlineto 0 exch neg rlineto closepath stroke grestore } bind def %%EndProlog 1152 720 36 36 3 box showpage %%Trailer

This program prints a rectangular border (with a line thickness of 3 points) one half inch in each direction from the edges of an 11-by-17 inch page. The rectangle is 10 inches high and 16 inches long. The coordinates in this le represent a decision to lay out the box on a ledger-sized sheet of paper, in a sense. The second decision is to print the document. Notice the distinction between the composition of the document and the act of printing it. At this stage, a particular printer is chosen. It is typical to inspect a proof of the document on the screen rst, then perhaps on a medium-resolution printer, and use a high resolution image-setter for nal output. Each device may have a different capability for rendering a page size as large as 11 x 17 (including, perhaps, not being able to support it at all). The actual invocation of a specic paper size should be done at print time. In listing 10-2 is the same example with specic printing instructions added. In particular, the ledgertray operator is invoked to print on 11-by-17 inch paper. This code is put between a %%BeginPaperSize and %%EndPaperSize comment since the code is (by necessity) device dependent.

160

chapter 10 FILE INTERCHANGE STANDARDS

listing 10-2
%!PS-Adobe-2.0 %%BoundingBox: 0 0 1224 792 %%DocumentPaperSizes: Ledger %%EndComments % <width> <height> <llX> <llY> <thick> box /box { %def gsave setlinewidth moveto dup 0 exch rlineto exch 0 rlineto 0 exch neg rlineto closepath stroke grestore } bind def %%EndProlog %%BeginSetup %%BeginPaperSize: Ledger statusdict begin /ledgertray where { %ifelse pop ledgertray }{ %else 90 rotate 0 -612 translate 792 1224 div dup scale % scale down for proofing on letter paper } ifelse end %%EndPaperSize %%EndSetup %%Page: 1 1 1152 720 36 36 3 box showpage %%Trailer

Notice that the PaperSize code checks for the existence of the ledgertray operator (using known) and if it is not found, the page is printed much smaller to t on letter-size paper. This technique can help provide fallbacks for devices which may contain unknown features (see also Section 10.6, CONDITIONAL EXECUTION).

chapter 10 FILE INTERCHANGE STANDARDS

161

10.5 PRINTER QUERIES

A PostScript language printer has the capability of providing a great deal of information back to the host computer. After all, PostScript is a complete programming language with full input/output support across the communications channel. However, the process of issuing a query to a PostScript interpreter can present some difculties. A query is dened to be a PostScript program whose sole purpose is to return some information from the PostScript interpreter back to the host computer. An example of this might be a short program to enumerate the names of all the fonts currently stored in the printer. Queries are typically issued even before the print le is produced, since some decisions may have to be made based on the available paper sizes and on the resident fonts. In an interactive environment, the notion of a query is a little less strict, since virtually every transaction will provide some information back to the host application. By querying the printer (or more generally, the PostScript interpreter), information can be obtained about the state of the interpreter. However, if the issuing software expects a particular bit of returned information, it may become impossible to introduce a print spooler between the composition software and the printer (unless the spooler could somehow interpret and answer the queries). See Chapter 12 for more details on writing print spoolers for PostScript interpreters. Since the need for querying is acknowledged under some circumstances, a specic protocol has been establish whereby queries are isolated into a single job, followed by an end-of-le indication. In addition, the queries should have well-formed PostScript comments to provide information to spoolers as to the nature of the query. In any case, the queries should be kept to an absolute minimum. Listing 10-3 is an example of a query job (two individual queries are contained within the job). Notice the rst line of the le, which contains a special notation to identify the job as a query job: %!PS-Adobe-2.0 Query

162

chapter 10 FILE INTERCHANGE STANDARDS

This is so that software reading the comments can quickly ascertain that it is a query job. See the PostScript Document Structuring Conventions for more detail on this.

listing 10-3
%!PS-Adobe-2.0 Query %%Title: sample query job %%?BeginFontQuery: Palatino-Roman /Palatino-Roman dup FontDirectory exch known { %ifelse 1 % yes }{ 0 % no } ifelse == flush pop %%?EndFontQuery %%?BeginVMStatus vmstatus exch sub == flush pop %%?EndVMStatus %%EOF

10.6 CONDITIONAL EXECUTION

When invoking a PostScript operator that may not be present in all implementations, it is best to put it in a context of conditional execution. This way the program can provide a reasonable alternative if the feature is not present on a given printer. For example, listing 10-4 is an invocation of the ledger-size paper tray that prints the page scaled down on letter-size paper if ledger is not available.

listing 10-4
%!PS-Adobe-2.0 %%EndComments %%EndProlog %%BeginSetup %%BeginPaperSize: Ledger statusdict /ledgertray known { %if

chapter 10 FILE INTERCHANGE STANDARDS

163

statusdict begin ledgertray end }{ %else (Ledger not found. Scaling...\n) print 792 1224 div dup scale } ifelse %%EndPaperSize %%EndSetup %%Page: 1 1 showpage %%Trailer

10.7 FONT AVAILABILITY

One of the most common difculties encountered when porting PostScript les from one environment to another is the availability of fonts. If a document is composed to be printed with a particular font, it basically should not be printed at all if the font is unavailable (there are some circumstances under which font substitution is appropriatesee Chapter 4 for more on this). A document le should communicate its font needs to other software through PostScript structure comments. There are a number of comments which may apply when fonts are used: %%DocumentFonts: Optima StoneSans Courier %%DocumentNeededFonts: Optima %%DocumentSuppliedFonts: StoneSans %%EndComments ... %%IncludeFont: Optima ... %%BeginFont: StoneSans % font description here %%EndFont The %%DocumentFonts comment provides a list of all the fonts that are actually referenced in the document (a findfont

164

chapter 10 FILE INTERCHANGE STANDARDS

is done). The %%DocumentNeededFonts comment is used to indicate that there is explicit use of the %%IncludeFont comment. It is an indication that the le should be parsed to satisfy a font request embedded in the le. The %%DocumentSuppliedFonts comment is used where an entire font description is contained within the body of the document le (as a downloadable font). The font itself is delimited by %%BeginFont and %%EndFont comments, as in the example. Again, the header comment is provided for information about what is contained in the body of the le. When a le is imported, either from another system or as an illustration in a document, these comments should be parsed to determine the font needs of the document being imported. When appropriate, the importing application must locate the needed fonts and place them in-line in the le, and/or update its own structure comments to reect the aggregate font needs of its own document le and the imported one.

10.8 PUTTING IT ALL TOGETHER

PostScript document les can have many complex requirements. This is partly because the possibilities are so rich, and because the PostScript language is complete. There are large sets of manuals provided for most programming environments with convention after convention that need to be followed for compatibility even with different versions of a single machine. The PostScript language spans many systems and architectures and the conventions for its use are correspondingly relatively simple. However, they are extremely important, in much the same way that all programming conventions are important: they help ensure compatibility across many kinds of products and environments. If they are observed properly, everyone benets. Here are some brief guidelines for structuring documents properly, and for using the conventions appropriately, and to reduce some of the complexity to simple rules of thumb: If a structuring convention or comment doesnt make sense in the environment that you are working in, dont use it. For example, omit the %%BoundingBox

chapter 10 FILE INTERCHANGE STANDARDS

165

comment if you are issuing a query, and dont put in the %%Page comments if the pages arent functionally independent of one another. If you include PostScript code that is not clearly documented as being a standard part of the language, then mark this code with some kind of %%Begin and %%End comments, according to the comment conventions. If the entire PostScript job is somehow special (a downloaded font, a query, or an exitserver job, for example), mark it at the beginning of the job according to the conventions. If you are unsure of whether your le is conforming, try importing it into another document as an illustration, and see if it works. Dont execute any of the initialization operators (any operator that contains init or default, among others). As a rule of thumb, dont execute anything in the prologue (including save), and dont dene anything new in the script (unless it is a state variable). Dont try to put save and restore around your whole job, prologue and all. That is done for you. It also makes the le non-conforming if you do. Do put save and restore around the individual pages of the script, or even more frequently, if appropriate.

166

chapter 10 FILE INTERCHANGE STANDARDS

chapter 11 MERGING FILES FROM DIFFERENT SOURCES

Life being all inclusion and confusion, and art being all discrimination and selection, the latter, in search of the hard latent value with which it alone is concerned, sniffs round the mass as instinctively and unerringly as a dog suspicious of some buried bone. Henry James

11.1 INTRODUCTION
One goal of software integration is to be able to produce complex documents. For instance, a report may need to contain a graph of data stored in a company database. This involves integrating data into a graphics application and merging it with text. Often the result of this integration is a printed page or a transparency for an overhead presentation. In a PostScript environment, this can be accomplished at print time in much the same way that photographs are added directly to the plates (stripped in) for an offset printer. If the software used to produce the graphs can produce a PostScript le as output, it can be merged with the PostScript output from a word processor to print the graph on the same page as the text. Essentially, it means the ability to paste in illustrations composed by other software. The nature of the PostScript language makes this work quite well. The coordinate system can be scaled and translated before the illustration is executed, and the entire graph can be printed at some percentage of its original size, and even rotated. However, some cooperation between the composing software and the importing software is necessary to allow it to work smoothly.

chapter 11 MERGING FILES FROM DIFFERENT SOURCES

167

This chapter is devoted to the relationship that should exist between the composing software and the importing software. It is closely related to Chapter 10, in that the best way to successfully merge PostScript les from different sources is to make sure that all the les are conforming PostScript les.

11.2 USING EXISTING CONTEXT

Every PostScript program executes in the context of an already executing PostScript program. At the start, a print le executes within the context of the server loop, which is just another piece of code. Any given PostScript le may have another complete program le embedded within it, which then executes in whatever context is presented to it by the outer program. In order to keep track of who is who, let us dene some terms: Document File: The document le is a PostScript program that may have another program embedded within it. It is typically simply the output of a page layout program or other document-producing software. Illustration: This is a program which executes within the context of another PostScript program. It need not truly be an illustration, but it is a good way to think of the relationship. Print Job: This refers to the PostScript le that is at the outermost level of execution. Note that this does not include the server loop. It is the rst user-level print le. Every PostScript program should be written as though it were an illustration, even if it is printed directly rather than imported into another document. A document le and an illustration do not differ at the PostScript language level, except that an illustration typically will not consist of more than one page. They differ at the conceptual level and in terms of how they are being used. Please refer to Chapter 6 for more details on program modularity and independence.

168

chapter 11 MERGING FILES FROM DIFFERENT SOURCES

A PostScript program should only change those elements of the ground state of the interpreter that are necessary for its own execution, and it should return the state of the interpreter to that same condition when it is through executing. The recovery can be accomplished simply by using save and restore, but it is the programmers responsibility not to disturb any parts of the machines state that do not need to be modied for the task at hand. In particular, it is important never to initialize the state of the interpreter.

11.3 ERROR RECOVERY

There is a discussion of error recovery principles in Chapter 14 of this book. In general, it is okay for a document program to merely report an error in a reasonable fashion, rather than being expected to recover from it gracefully. For instance, if a mechanism is put in place that assigns a job name to each illustration that is embedded in it, then any execution errors which might result from the execution of those les can be identied easily by that name. In some cases, a print job cannot do any better than this at error recovery. For instance, if the illustration is actually reading from the current le (as with the image operator), it may not quit reading at the appropriate place in the le (if it is not written correctly), and could read past the end of the illustration and into the rest of the original document. The stopped operator can be used to catch (almost all) execution errors in an embedded illustration. In Chapter 14 (listing 14-2) there is a procedure called userexec which is used much like exec. It takes any executable object (for instance, the le object returned by currentfile cvx) and executes it in an errorrecovery context. This can be used quite effectively for included illustrations, if needed. For most purposes, it is better to identify the included job by some name, and trust it to execute correctly. If it doesnt, the user will know what happened if you supply the name of the le that was executing when the error occurred.

chapter 11 MERGING FILES FROM DIFFERENT SOURCES

169

11.4 HANDLING SHOWPAGE

Since the showpage operator is common to almost all PostScript programs, it is not fair to make it illegal in a conforming PostScript le. Instead, the burden is placed on the including document to disable showpage when appropriate. In listing 111 is a bit of code that might be wrapped around a program le that is to be embedded into the current le.

listing 11-1
%!PS-Adobe-2.0 %%Title: include.ps %%EndComments %%BeginProcSet: include.ps 1.0 0 /beginexecute { %def /level0 save def /showpage { } def /jobname exch def } bind def /endexecute { %def level0 restore } bind def %%EndProcSet %%EndProlog % including "Illustration.ps" (Illustration.ps) beginexecute 0 0 translate % if needed 1 1 scale % if needed %%BeginDocument: Illustration.ps %!PS-Adobe-2.0 %%Title: Illustration.ps jobname = flush %%Trailer %%EndDocument endexecute showpage %%Trailer

This disables the showpage operator for the duration of the execution of Illustration.ps, and reinstates it to its original value

170

chapter 11 MERGING FILES FROM DIFFERENT SOURCES

(whatever it may have been) after the restore. Notice also that the name of the le is made available in the jobname entry, which can be reported if an error occurs.

11.5 SCREEN REPRESENTATIONS

In visual page composition systems, it is typically possible to see on the screen what will eventually print. It is an unusual approach to be able to print what you cannot see (or edit). This situation arises whenever printing technology advances more rapidly than front-end composition systems. In general, it is powerful to build front-end systems that permit inclusion of unknown elements that are specic to the printing technology (a bit image, perhaps). In most environments, some indication must be given on the host system wherever an imported le is added by the application. As a specic example, a visual page layout program may have the capability to merge a PostScript le into the output from the program, but it needs some mechanism for representing the image on the front end. There are three approaches to this problem. The rst is the simplest, and the last is probably the best: 1. Represent the included le by name. One example of this is simply an include statement in the input le: #include "illustration.ps" In a markup-language approach to document layout, this is usually the way les are included. A slightly more sophisticated approach might include displaying a rectangle on the screen where the drawing will be placed (or permitting it to be scaled or moved on the page as it is being imported). 2. Display a simulated bit image of the illustration on the screen, but do not print it (send the PostScript le to the printer instead). This simulated image could be prepared by the software that produced the illustration in

chapter 11 MERGING FILES FROM DIFFERENT SOURCES

171

the rst place, and must be represented by a convention to which each application agrees. 3. Interpret the PostScript Code. If the PostScript le can be executed and displayed directly, no simulation is required. This can be thought of as a preview capability if the software does not use the PostScript language as its display model already. In each of these approaches, the importing application can handle the le better if it knows that it is a PostScript language illustration. In particular, it can consult the bounding box information in the le to display the appropriate rectangle, or it can display the simulated bit image on the screen if it knows where to look for it. A current convention in the industry is for a drawing application to produce conforming PostScript les with a preview bit image along with them, in a standard format which importing applications can understand. It is then up to the importing program to display the bit image on the screen and keep the PostScript program to send to the printer. As the PostScript language becomes more commonplace for screen drivers, the applications may not need to use the previewed bit image stored with the le, but may choose to interpret the PostScript code directly.

172

chapter 11 MERGING FILES FROM DIFFERENT SOURCES

chapter 12 WRITING A PRINT SPOOLER

12.1 INTRODUCTION
A print spooler is a program that collects output for a given printer or network and saves it for later printing. The le may then be put in line behind others waiting to print, or it may be printed immediately, but the host application which originally sent the print job can return to what it was doing, knowing that the job has been queued (or has printed) successfully. The advantages of this mechanism are well known. A job may be reprinted or rescheduled, it may be cancelled, and (best of all) the host application doesnt have to wait for it to print. The difculties result from the fact that the link between the application and the printer is made more indirect. Certainly a front-end program that tries to query the printer to nd out about available resources would be surprised if the device did not answer. Also, the possibility arises that a job might be queued for one printer but be actually printed on another device. Cooperation is required, as usual, on both ends of an exchange like this. The host applications must behave appropriately so that they do not make assumptions about the printer that would make it difcult or impossible to write a print spooler. The spooler must be written in such a way that it is completely transparentif the job would have printed without the spooler, it should print through the spooler as well. The assumptions in this chapter apply almost exclusively to PostScript interpreters in batch mode. Any interactive systems will probably not use print spoolers.

chapter 12 WRITING A PRINT SPOOLER

173

12.2 PRINTER MANAGEMENT

PostScript is a programming language. It is conceivable that a one-line program sent to a standard PostScript interpreter could cause 2 megabytes of information to be sent back to the host computer. This is (today) a little unusual for a printer. In some instances, it is a temptation: Why should I take the time to compute something that the printer already knows; Ill just ask the printer. In other instances, the information may be a part of the standard communication with the printer (for instance, if an execution error occurs, an error message is sent back to the host system). The task of managing a PostScript printer is a larger issue than simply print spooling. There are resource management issues (is Optima a resident font, is it on the printers local disk, is it available on a font server, or should it be downloaded?). Handling and reporting errors, out-of-paper conditions, and paper jams is the responsibility of the printer management software. Perhaps print job arbitration and le spooling is another function of this software, although they are usually separate entities.

12.3 COMMUNICATIONS
The low-level considerations of maintaining two-way communication with a PostScript device are a critical part of printer management. There are some aspects of this communication which differ greatly from other printing devices, and which need attention in the design phase. The batch job model under which PostScript printers normally operate is founded on individual print les. That is, a single job is dened from the initial byte through the nal end-of-le indication. The end-of-le (EOF) should never be contained in the document le itself; it is part of the communication dynamics. In fact, the EOF marker for, say, a serial port is much different than the EOF indication on a packet network. The PostScript interpreter will echo an EOF to match each EOF that it receives. This is a conrmation that the entire le has been received. A second job should never be sent until the rst

174

chapter 12 WRITING A PRINT SPOOLER

one has completely nished executing (including echoing back the EOF indication).

MESSAGES
PostScript printers tend to generate ASCII text messages back to the host system. Some of these are in response to status requests, and some of them appear whether or not you ask for them. For example, on many devices findfont will return the font Courier if it cannot nd the font asked for. When this substitution is performed, a message is generated back to the host computer that looks like this: StoneSans not found, using Courier. If a PostScript error is triggered, the standard error handler produces a message back to the host system containing the name of the error and the token that triggered it: %%[ Error: typecheck; OffendingCommand: if ]%% Printer management software should be prepared for any kind of text that may be sent back from the printer. Here is a reasonable approach for dealing with these messages: If the text is thought to be an error message, it should be communicated back to whatever software submitted the job to be printed, or displayed for the user to see. If the text is not an error message, then it should be compared to a list of known standard messages that may be issued by that printer (as listed in the Printer Description File). Some of these may safely be ignored. If the text is not recognized, it should be communicated back to the application which submitted the job, and/or saved in a disk le and made accessible to the user. Under many circumstances, a disk le containing everything that the printer sent back to the host may be appropriate, both as an error log and as a way to extract specic bits of information from the printer.

chapter 12 WRITING A PRINT SPOOLER

175

There are messages which are more or less standard depending on the physical capabilities of the printing device. For example, following are a number of relatively common (self-explanatory) messages. These are all very device-specic, and the Printer Description File should be consulted for a precise list of them for a given printer, as in the following list: %%[ PrinterError: out of paper ]%% %%[ PrinterError: cover open ]%% %%[ PrinterError: warming up ]%% %%[ PrinterError: out of toner ]%% %%[ PrinterError: paper jam ]%% %%[ PrinterError: recorder not responding ]%% These messages are similar in syntax to the PostScript execution error messages, but are only issued when the marking engine is not ready to print a page for some reason. As an aside, it is generally not a good idea to issue the executive operator (available on most devices) while in batch mode. The executive operator is intended for interactive use; it echoes back everything typed, permits simple line editing, and presents a prompt. Batch jobs are not guaranteed to run precisely the same way in executive mode (due to the nature of command-line editing) and it is therefore not recommended for tracking the execution of a le. It is, however, quite handy for typing short PostScript language sequences and observing the results.

12.4 USING EXITSERVER

exitserver is a specialized and environment-dependent system operator available in most devices which support batch PostScript jobs (it is not part of the standard language denition). Its purpose is to exit the server loop (described in section 2.9). The exitserver operator requires a password, and will result in an invalidpassword error if the incorrect password is given. Since the job server loop is generally responsible for cleaning up the state of the interpreter between jobs, any changes that

176

chapter 12 WRITING A PRINT SPOOLER

are made outside the server loop by using exitserver will remain as part of the permanent state of the interpreter for all subsequent jobs. This only applies to changes to VM (like procedure denitions), since the stacks and graphics state are cleared after each job. Note: The exitserver operator is a system-level command. It is not intended for general use by applications programs. Using exitserver is forbidden in conforming documents. See section 10.1 for further information on conformance specications. The exitserver mechanism can be used to download fonts or procedure bodies that will stay resident in the printer until the device is powered off. exitserver also is required for changing any of the persistent parameters such as communications protocols or default conditions (such as the default job timeout value). Using exitserver initiates a new PostScript job. It should always be the rst line of code in the program. The condition of being outside the server loop persists until the next end-of-le indication is read by the interpreter. Any changes made to the state of VM during the job will remain until the interpreter is restarted. Here is the necessary PostScript language sequence to exit the server loop: serverdict begin 0 exitserver The exitserver operator resides in the serverdict dictionary, and this dictionary must be made available in order to execute exitserver. It is not necessary to use end to remove it from the dictionary stackwhen the new job is started the contents of all of the stacks are cleared. In this example, 0 is placed on the operand stack as the password required by exitserver. This password generally may be changed by using the setpassword operator.

chapter 12 WRITING A PRINT SPOOLER

177

12.5 MANAGING FILES AND FONTS

A print spooler may be called upon to manage external resources like downloadable fonts or les. For instance, when a print le is spooled, it typically will not have embedded font descriptions (downloadable fonts) in the document le itself. Certainly in a highly distributed environment each workstation should not have local copies of the fonts. When a document is submitted for printing, the font requirements (resource needs) are communicated along with the document, usually in the form of document structuring comments (see chapter 10 and section 6.4). This list of needed fonts (or les) is compared to the list of available fonts in the printer, and any fonts that are not already resident in the device should be downloaded along with the document le. There are several specic structuring conventions which apply in this situation, and it is worth going through them in detail.

%%DocumentFonts
The %%DocumentFonts comment is used in the header section of the comments (or it may be deferred to the trailer). It should give a list of all fonts that are used by the document (if a findfont is performed on the font name anywhere in the document, it should be in this list). Here is an example of its use (notice the %%+ syntax for continuation; the lines should not exceed 255 characters in length): %!PS-Adobe-2.0 %%DocumentFonts: Courier Times-Roman %%+ StoneSerif %%+ StoneSerif-SemiboldItalic %%+ Optima %%EndComments This comment may be used rst to determine if there are any font needs that are not resolved within the document. If all the fonts in this list are determined to be resident in the PostScript device, no further attention is necessary.

178

chapter 12 WRITING A PRINT SPOOLER

%%IncludeFont
This comment can appear anywhere within the body of a PostScript document le. It is a resource request. The named font should be inserted at exactly the point in the le that the comment is found (if it is determined that the font is not resident; otherwise %%IncludeFont can be ignored). The list of fonts for which the %%IncludeFont comment is used is made available in the documents header comments section through the comment %%DocumentNeededFonts. This has the same syntax as the %%DocumentFonts comment (see listing 12-1).

listing 12-1
%!PS-Adobe-2.0 %%DocumentFonts: Courier Times-Roman %%+ StoneSerif %%+ StoneSerif-SemiboldItalic %%+ Optima %%DocumentNeededFonts: Optima %%+ StoneSerif %%EndComments %%EndProlog %%Page: 1 1 save %%IncludeFont: Optima restore %%Page: 2 2 save %%IncludeFont: Optima %%IncludeFont: StoneSerif restore %%Trailer

These comments are directives. If they name fonts that are determined not to be resident in the printer, then the requests need to be satised, or the document will not print successfully.

chapter 12 WRITING A PRINT SPOOLER

179

%%BeginFont, %%EndFont
When a downloadable font le is embedded in a document le (either before it reaches the print spooler, or by the spooler itself), the font should be bracketed with %%BeginFont and %%EndFont comments. The %%DocumentSuppliedFonts comment should be used to indicate the names of any fonts supplied within the body of the document: %!PS-Adobe-2.0 %%DocumentFonts: Sonata %%DocumentSuppliedFonts: Sonata %%EndComments %%EndProlog save %%BeginFont: Sonata % Sonata font le is embedded here %%EndFont restore %%Trailer The le should execute correctly without any attention by the print spooler, since the font is supplied within the document. However, there are a few ways in which the spooler might benet by using these comments: The fonts may be collected and stored in the spoolers font area for later use with %%IncludeFont comments. The fonts may be removed from the print stream if they are determined to already be resident in the printer. However, the fonts should always be replaced by an instance of %%IncludeFont so that it is reversible. Whenever a font is added or deleted from a print le, the comments should be adjusted accordingly. %%IncludeFont and %%BeginFont/%%EndFont are essentially inverses of one another (each should be replaced by the other, as appropriate). There are exactly analogous comments and operations dened for les. Details on these can be found in the specication of the PostScript Document Structuring Conventions.

180

chapter 12 WRITING A PRINT SPOOLER

12.6 DETERMINING WHAT FONTS ARE AVAILABLE

PostScript printers have many options for storing fonts. They may be built into the interpreters initial VM (in ROM, for instance), they may be available on a built-in le system, in font cartridges (a special case of the le system) or available on a network font server. Fonts can also be downloaded to printers on a semi-permanent basis into user VM (with exitserver). It is not a trivial task to determine what fonts are available on a given printer. Host-resident software should keep track of the fonts that are resident in a printer, and at best should make its list visible to (and editable by) the user. Fonts that are built into a printer (in ROM) are always available, and are detailed in the Printer Description File for that printer (see Section 12.8).

listing 12-2
%!PS-Adobe-2.0 ExitServer %%EndComments /$timestamp where { %if pop (time is: ) print $timestamp = flush stop % stop the current job } if %%BeginExitServer: 0 serverdict begin 0 exitserver %%EndExitServer /$timestamp (Tue Oct 27 17:53:43 PST 1987) def %%Trailer %%EOF

A good approach to managing the list of currently available fonts is to query the printer for the initial list of fonts, and to keep that list available. If a font is downloaded to the printer, the list should be updated. If the printer is restarted, the list should be re-generated. Noticing that the printer has restarted can be accomplished by placing a time stamp in the printer, then

chapter 12 WRITING A PRINT SPOOLER

181

checking for it each time a new connection to the printer is opened (listing 12-2).

12.7 HANDLING RESOURCE SHORTAGES

It is possible that a spooler will be presented with a le to print for which it cannot nd (or invoke) all the necessary resources. For instance, the requested paper size may not be available, or a needed font. There are basically three ways to handle this situation: 1. Reject the le (issue an error condition of some kind). 2. Substitute another resource (scale to t on letter-sized paper, for instance, or default to a different font). 3. Trust me. Sometimes the user knows more than the software, and it should always be possible to force the spooler to pass the le through untouched. For instance, a new font storage device might be in place, but the spooler may not know how to nd those fonts and will want to substitute Courier. The substitution option is the most difcult to accomplish, and may or may not be advisable, depending on the circumstances. Most people will not be happy if they ask for Univers and you give them Helvetica. The pages may wind up in the trash anyway, and it might have been better simply to reject the job in the rst place (and save paper). The software at least should announce its intentions in a loud voice if it is going to substitute one resource for another. In any case, the method used to determine whether or not the resources are available may not allow for every option and something like the trust me in the list above should be made possible.

12.8 PRINTER DESCRIPTION FILES

The problem of keeping track of different PostScript printers and their various features, fonts, and capabilities is not an easy

182

chapter 12 WRITING A PRINT SPOOLER

task. Beyond the core of the PostScript language, there are many differences between one product and another. The most important of these differences are formalized in Printer Description Files (available from printer vendors and from Adobe Systems). These les contain a list of the features of the printers and the necessary PostScript code to invoke them properly. Here are the kinds of features which are supported through these les: Paper Sizes. Printers may support various different paper sizes, paper trays, and even continuous-feed rolls of paper, which can virtually be any size. These are detailed in Printer Description les by keywords indicating the paper size. There is also a great deal of information on the dimensions and imageable regions of each paper size. Resident Fonts. A list of all the fonts that come standard in the printer, including the version number and whether or not it has a standard encoding vector. Query Strings: these are miniature PostScript programs that will query the device for the presence or state of many of the features found in the le (for example, whether a particular font is available, or the current paper tray). Available Memory. This provides the amount of memory available in the initial conguration of the printer. Special Resources: this includes the availability of a resident le system, variable paper sizing, the default exitserver password, and others. The details on the format of Printer Description Files are kept in a separate document entitled PostScript Printer Description Files, which is available from Adobe Systems.

chapter 12 WRITING A PRINT SPOOLER

183

chapter 13 MEMORY AND FILE RESOURCE MANAGEMENT

While memory holds a seat In this distracted globe. Remember thee! Yea, from the table of my memory Ill wipe away all trivial fond records. William Shakespeare

13.1 MEMORY STRUCTURE

Writable memory in a PostScript interpreter is referred to as virtual memory, or VM. It is the memory that is used for all aspects of program execution that require information storage. VM is used for storing dictionary entries, downloadable fonts, frame buffers, the operand stack, string bodies, and anything else that requires memory. When a PostScript interpreter starts up, it allocates a large amount of the available memory for internal processing. The details of this allocation are device-specic, but in many implementations they are roughly as follows: area of virtual memory operand stack dictionary stack execution stack userdict 8.5 x 11 inch frame buffer size in bytes 4,000 160 2,000 4,000 1,051,875

There is typically anywhere from about 200 kilobyes to several megabytes of usable VM after the interpreter is initialized. The vmstatus operator can be used to determine the amount of free memory in the interpreter.

chapter 13 MEMORY AND FILE RESOURCE MANAGEMENT

185

13.2 MEMORY ALLOCATION

Memory can be allocated either implicitly or explicitly. Essentially all memory is allocated by the creation of composite objects. There are several PostScript operators that explicitly allocate VM: string array dict } ] The } and ] operators allocate memory in the same way that array does. Each of them counts down to the topmost mark on the stack and is equivalent to the PostScript sequence counttomark array astore. Memory is implicitly allocated primarily by the creation of literal string bodies. For example, the following PostScript segment allocates approximately 24 bytes of VM: (This is a literal string) show Since the show operator consumes the string object on the operand stack, the string object is lost and the string body is oating in VM somewhere. The memory allocated is not reclaimed except by the save and restore operators. The amount of VM used by any particular PostScript data structure is implementation-specic. In current PostScript interpreters, VM usage is approximately as follows: operator name array string dict bytes per element 8 1 20

For example, each of the following allocates 80 bytes of memory in a current implementation: 80 string 4 dict

186

chapter 13 MEMORY AND FILE RESOURCE MANAGEMENT

10 array [ 1 2 3 4 5 6 7 8 9 10 ] true { %if exch pop 3 -1 roll exch show 0 10 rmoveto } if It is good to understand the ways in which memory is allocated in order to be able to conserve it. It is not generally true that you can rst get a program to work and then go back and add some kind of memory management. Since save and restore are the mechanism for controlling memory use, and since they affect more than just the use of memory, this must be designed in from the start.

13.3 SAVE AND RESTORE

The only way to manage memory is with save and restore. All VM used since the previous save is reclaimed when restore is executed, including the contents of dictionaries and arrays. The save and restore mechanism should be used as often as necessary to reclaim memory. The save and restore operators affect all other aspects of the state of the interpreter, including the graphics state. It is best to use save and restore in an integrated fashion to preserve both graphics state and the contents of VM between logically independent elements of a PostScript program. There is a discussion of the use of save and restore in Section 6.3 relating to modular program design.

SAVE OBJECTS
The save and restore mechanism works in a somewhat unusual fashion. The save operator produces what is known as a save object. This object is like a coat check couponin order to get your save level back, you must present the save object to the clerk. A save object contains a generation number that is assigned sequentially. It is possible to restore to any save level by pre-

chapter 13 MEMORY AND FILE RESOURCE MANAGEMENT

187

senting the appropriate save object to the restore operator, but restoring level 1 makes levels 2 and 3 no longer valid (for obvious reasons). Since save produces an object and restore requires one, they can be used somewhat like gsave and grestore. However, the save object must stay on the operand stack: save count 1 eq { (save object is on stack) = } if restore Since the save object must be presented to restore, it is usually better to store it in a dictionary than to leave it on the operand stack for an extended period of time: /saveobj save def count 1 eq { %ifelse (save object is on stack) = }{ %else (save object is NOT on stack) = } ifelse saveobj restore The sequence /saveobj save def seems a little unusual at rst. Since save produces an object on the stack, and since /saveobj is just a name object resting on the operand stack, it is a simple denition. The save does not affect the operand stack, and since the name object is created before the save, it does not affect the state of VM that is saved. The object is then retrieved by that name when it is needed later.

THE INVALIDRESTORE ERROR

The restore operator can trigger one error condition which can be quite difcult to debug: the invalidrestore error. There is only one condition which will provoke this error, but there are several ways to achieve it. The restore operator restores the entire state of memory back to a previously saved state. All objects in memory which have been created more recently than that save state are discarded. The condition which provokes an invalidrestore error is a com-

188

chapter 13 MEMORY AND FILE RESOURCE MANAGEMENT

posite object on one of the stacks (operand stack, dictionary stack, or execution stack) that points to a value in memory somewhere that would be reclaimed by the restore operation. Since restore does not affect any of the stacks, it is considered to be an error condition for an object there to have its composite value removed from under it. The invalidrestore error can be prevented simply by never leaving a composite object on any of the stacks that was created more recently than the last save level (at least, dont have it on one of the stacks if you are about to do a restore). The most common of these, actually, is to leave a dictionary on the dictionary stack. For example, the following PostScript fragment provokes the invalidrestore error: save /mydict 28 dict def mydict begin restore It is difcult to get a composite object onto the execution stack that would cause an invalidrestore, but it is possible. Since the most likely object to nd on the execution stack is a procedure body, this is the place to look (particularly innocent procedure bodies like those presented to ifelse). The following example provokes an invalidrestore error because the remainder of the procedure body in which the restore is found is still on the execution when restore is executed, but it was created more recently than the save object. save Ypos 36 lt { restore showpage } if The only way to prevent this is to dene the procedure before the save is executed, or not to execute restore from within the procedure. For example, the following short procedure denition xes the problem: /restorestate { Ypos 36 lt { restore showpage } if } bind def save restorestate

chapter 13 MEMORY AND FILE RESOURCE MANAGEMENT

189

13.4 DOWNLOADABLE FONT PROGRAMS

Downloadable fonts are simply PostScript programs. In general the programs build a data structure and then register it as a font. A downloadable font program is conceptually like this: 127 dict begin % allocate a font dictionary /FontName /TestFont def /BuildChar { } def % ... currentdict end /TestFont exch definefont pop The definefont operator takes a font dictionary and registers it in the list of fonts available to findfont. This includes making an entry in a dictionary known as FontDirectory which contains the association between the font name and the font dictionary. Font dictionaries are allocated from VM in exactly the same way that any other dictionaries are allocated, and use memory from the same pool. The only way to reclaim the memory used by a downloadable font is with save and restore, since the space occupied by a font is in the form of a dictionary. Each element of a font dictionary may require more memory, as well. For instance, the entry in a font which provides the font matrix requires memory to store the matrix itself: /FontMatrix [ .001 0 0 .001 0 0 ] def This requires about (8*6 = 48) bytes of memory. Note that the 8 bytes or so required for the array object itself was already allocated to the dictionary in which it is stored. Its existence on the operand stack requires no further VM. A typical Adobe downloadable font dictionary requires approximately 30 kbytes of VM.

13.5 PACKED ARRAYS

In all but the very rst implementation of the PostScript interpreter, there is a feature known as packed arrays. This feature is a storage optimization that permits arrays to be stored in much

190

chapter 13 MEMORY AND FILE RESOURCE MANAGEMENT

less space than standard arrays. A packed array is read-only. All array operators other than put and putinterval will work with packed arrays. In general, sequential access to packed arrays is as fast as with normal arrays, but random access is slower. Packed arrays are typically about 50 to 75 percent smaller than their unpacked counterparts. Packed arrays are perfectly suited for procedure bodies, which are read-only arrays. A typical way to use packed arrays is to turn the feature on at the beginning of the prologue denitions, and turn it back off at the end of the prologue, so that all the procedure bodies created in the prologue will be stored as packed arrays, saving space: %!PS-Adobe-2.0 %%EndComments /setpacking where { /currpack currentpacking def pop true setpacking } if /prologue { packed array procedures } bind def /setpacking where { pop currpack setpacking } if %%EndProlog %%Trailer All procedure bodies constructed while packing is set to true will be packed arrays. There is also an operator called packedarray which is analogous to the array operator, except that it constructs a packed array object instead of an array object. This is a more specic operator that can be used to allocate a single packed array object; the setpacking operator is more appropriate for general use.

13.6 RASTER MEMORY

Frame buffers (or, in some devices, band buffers) are large chunks of memory set aside as raster memory (or display memory). Frame buffers are used essentially as device space for some PostScript interpreters. Frame buffers are typically allocated by the framedevice operator, although there is already typically a

chapter 13 MEMORY AND FILE RESOURCE MANAGEMENT

191

default frame buffer in place for each device. The framedevice operator takes a specied width and height and allocates the entire chunk of memory necessary for rendering that page size. It is not a recommended operator, however. The amount of memory required for a frame buffer can be determined easily. One byte of memory contains 8 bits of information, each of which is used as a pixel in device space (unless it is a grayscale or color device, in which case there may be many bits per pixel). An 8-1/2 inch by 11 inch frame buffer on a 300 dotper-inch printer requires the following amount of memory: 300 (8.5 x 11) 8 This is 1,051,875 bytes of memory (just over one megabyte). Changing page sizes therefore can affect the amount of usable memory that is available for a user job.

13.7 FILE SYSTEMS AND DISK MANAGEMENT

PostScript interpreters that have a le system available can use this secondary storage for many different purposes. Typically the font cache is permitted to occupy space on the disk as well as in the xed partition of memory normally used for the character bitmaps. Additionally, a high-resolution PostScript devices may use the disk for raster storage when it runs out of space in memory. Finally, downloadable font programs can usually reside on the le system. For the most part, disk management is limited to the user partition of the le system. This is the space available for downloaded fonts and other user-dened disk les. The system partition is used for the font cache and perhaps for raster storage. The PostScript language supplement for any particular device should provide details on these operations.

13.8 POSTSCRIPT LANGUAGE FILE OPERATIONS

The PostScript language has full software support for le systems. There are PostScript operators for opening, reading, writ-

192

chapter 13 MEMORY AND FILE RESOURCE MANAGEMENT

ing, and closing le streams present in all implementations of PostScript interpreters. Many PostScript devices do not have a le system available to them directly (unless there is a disk and le system built into the printer, for instance). However, the input stream and output stream are treated as le objects, and the le operators may be used to read and write these streams.

THE STANDARD INPUT STREAM

The input stream is directly accessible to user programs as a le object. The currentfile operator returns a le object representing the current input stream. The PostScript interpreter uses the same le object for interpreting your program, but there is only one le pointer. This means that the le can contain both data and executable PostScript language statements. Listing 13-1 is a short example of how a program can read data from the current le.

listing 13-1
%!PS-Adobe-2.0 %%Title: currentfile example /buff 128 string def /F { findfont exch scalefont setfont } def %%EndProlog 12 /Helvetica F 10 10 moveto currentfile buff readline This line is actually data pop show showpage %%Trailer

The interpreter scans and executes the currentfile operator, which produces a le object and leaves it on the operand stack. The interpreter then executes buff, which places a string object on the stack. Then readline is scanned and executed.

chapter 13 MEMORY AND FILE RESOURCE MANAGEMENT

193

The le pointer for the input stream is left at the point just after the readline token. When readline is executed, the le object from which it reads is the one left by currentfile (the standard input stream). The readline operator picks up exactly where the interpreter left off, and reads one line of input (In this case, it reads the string This line is actually data into the input buffer buff). When readline nishes, it has advanced the le pointer to a position just after the line of data, and the interpreter picks up where readline left off. The interpreter then scans and executes show, which prints the data string at location 10 10 on the current page. This technique of reading from the current input stream is used in many ways. It can be used by a printer emulator to read data intended for a different printer. It may be used to read bitmap data for the image operator directly from the input stream (using readhexstring rather than readline). Listing 13-2 contains a short program that reads a font program from the current le and write it to a disk le, assuming that a le system is available to the PostScript interpreter. The le name needs to be changed for each different font downloaded (this is done by a front-end program). Since the program reads through end-of-le, it must be in a job by itself. Subsequent jobs can then access the font from the le system simply by calling findfont. This example uses the font from listing 9-3.

listing 13-2
%!PS-Adobe-2.0 %%Title: writetodisk.ps %%Creator: Glenn Reid %%EndComments /buff 128 string def /fd (fonts/Logo-Font) (w) file def { %loop currentfile buff readstring { %ifelse fd exch writestring }{ %else dup length 0 gt { %ifelse fd exch writestring }{ %else

194

chapter 13 MEMORY AND FILE RESOURCE MANAGEMENT

pop } ifelse fd closefile exit } ifelse } bind % file should follow the "loop" token.... loop %%BeginFont: Logo-Font 1.0 0 12 dict begin % leave one empty slot for FID /FontName /Logo-Font def /FontMatrix [ 1 0 0 1 0 0 ] def /FontType 3 def /FontBBox [ -5 -12 85 50 ] def /BuildChar { %def exch begin 70 0 -5 -12 85 50 setcachedevice Encoding exch get load exec end } bind def /Encoding 256 array %def dup 0 1 255 { /.notdef put dup } for pop dup 71 /logo put % "G" character def /logo { %def % draw at current point 20 0 moveto gsave /Times-Roman findfont 48 scalefont setfont (G) show -13.5 -14.5 rmoveto (R) show grestore -4 -8 rmoveto /Helvetica findfont 11 scalefont setfont (glenn) show 22 3 rmoveto (reid) show } bind def currentdict end dup /FontName get exch definefont pop %%EndFont: Logo-Font

chapter 13 MEMORY AND FILE RESOURCE MANAGEMENT

195

chapter 14 ERROR HANDLING

Reason and experiment have been indulged, and error has ed before them. Thomas Jefferson

14.1 INTRODUCTION
Programs written in the PostScript language are extremely portable. They are high-level source code intended to be executed on any number of different PostScript interpreter, even if the software that produces the program cannot physically be hooked up to the particular device. This chapter addresses some tactics for making PostScript programs as robust as possible. It does not specically address printer differences and portability issues, which are dealt with in other chapters. The focus is placed instead on the error handling capabilities of the PostScript language and how this mechanism can be used to advantage in applications programs.

14.2 STRATEGIES
The kind of errors that can be caught by an error handling strategy are typically execution errors. Assuming that the program produced by the application has been debugged to a reasonable degree, there are only a few possible failure modes which can arise in the PostScript language environment.

NON-STANDARD OPERATORS
One possible source of error is to make reference to a specialized operator that takes advantage of a feature that may not be present in all implementations of the PostScript interpreter (typically giving rise to the undefined error). For instance,

chapter 14 ERROR HANDLING

197

invoking a printer-specic feature like manual paper feed or a particular paper tray will not work on a printer which does not offer that feature. A simple mechanism to avoid this is to use the known operator to check for the existence of the operator before you use it: statusdict /ledgertray known { %ifelse statusdict begin ledgertray end }{ %else 90 rotate 0 -612 translate 612 792 div dup scale } ifelse This tests to see if the ledgertray operator exists in statusdict. If it does, it is invoked. If it does not, then the standard letter-size page is placed into landscape mode and scaled down to simulate the layout of ledgertray. The where operator may also be used in a similar fashion.

IMPLEMENTATION LIMITS EXCEEDED

A limitcheck error can arise if an internal limitation has been exceeded by a user program. This can occur, for instance, when using clip, charpath, or flattenpath. These operators are all sensitive to the resolution of the PostScript device to some extent. Both clip and charpath provide a attened path, which means that all curve segments in the path are reduced to line segments. The current atness affects the closeness of the approximation. There will typically be many more line segments when attening a path on a high resolution device than on a low resolution device. For example, gure 14.1 shows a curved path stroked rst with gray at the normal atness, then in black with a atness of ve user units (about 100 device units at 1270 dpi). There are other circumstances and operators which can provoke the limitcheck error, of course. Operations which are sensitive to the resolution or amount of available memory may be susceptible to provoking a limitcheck error on one PostScript device but not another. With extremely complex paths, operators like stroke and fill can provoke limitcheck errors, although it is rare. This typically only happens when printing at

198

chapter 14 ERROR HANDLING

especially high resolution. The next section describes the use of the stopped operator, which may be used to catch this kind of execution error. Another error which can arise when implementation limits are exceeded is VMerror. This means simply that the PostScript interpreter has run out of memory. Unfortunately, this error is not always caught by all versions of the PostScript interpreter (depending on what was being executed when it ran out of memory). One symptom of exceeding the memory of a PostScript device may be that the interpreter will restart without explanation. VMerrors can be avoided by proper use of save and restore.

gure 14.1

14.3 THE STOPPED OPERATOR

The stopped operator provides a method for regaining control of the program when the stop operator is executed. As discussed in section 14.4, the default behavior of all error operators is to store some of the state of the machine in an error dictionary and to execute stop. The stopped operator takes any executable object on the stack (usually a procedure body, but it could also be an executable le

chapter 14 ERROR HANDLING

199

object). It copies that object to the execution stack (like exec) and executes it. The stopped operator determines whether or not stop was executed, places a boolean on the top of the stack, and returns control to the interpreter. Listing 14-1 is an example of using stopped to catch a limitcheck operation on a flattenpath operation.

listing 14-1
%!PS-Adobe-2.0 %%Title: stopped-example %%EndComments /STROKE { %def currentflat { %loop { flattenpath } stopped { %ifelse currentflat 1 add setflat ("limitcheck" on "flattenpath": ) print (trying again with flatness of ) print currentflat ( ) cvs = flush }{ %else exit } ifelse } bind loop stroke setflat % restore original flatness } bind def %%EndProlog %%Page: 1 1 % build a complicated path: 0 1 30 { %for 12 exch 2 mul translate 20 0 moveto 0 0 20 0 360 arc } for STROKE showpage %%Trailer

This program will loop until the flattenpath operation succeeds. When it fails, the atness is increased slightly and the

200

chapter 14 ERROR HANDLING

procedure tries flattenpath again. This approach can be used to catch any kind of errors, although it is complicated enough that it is only encouraged under some circumstances. Another use of the stopped operator is to catch possible execution errors in included PostScript les. For instance, if your application contains a feature for including arbitrary PostScript les from other applications (or potentially hand-written code), it might be worthwhile to trap the errors and take some suitable action. A procedure called userexec is dened in listing 14-2 that will execute any object in the stopped context and return control to the procedure when it has nished (even if it terminates with an error).

listing 14-2
%!PS-Adobe-2.0 %%Title: userexec.ps %%Creator: Glenn Reid %%EndComments /buff 512 string def /p /print load def /bp { buff cvs print } bind def /jobdef { %def errordict begin /jobname exch def end } bind def /userexec { %def stopped { %ifelse $error /newerror get { %if $error begin (%%[Error: ) print /errorname load bp (; OffendingCommand: ) p /command load bp (; JobName: ) p errordict /jobname known { %ifelse errordict /jobname get }{ (stdin) } ifelse bp (]%%\n) print flush end

chapter 14 ERROR HANDLING

201

{ %loop currentfile buff readline { %ifelse (%ENDuserexec) eq { exit } if }{ %else /UnexpectedEOF % report error errordict /rangecheck get exec } ifelse } loop } if % if newerror is true }{ %else % end of file must have been reached. Is this an % error? Sometimes it means the improper use of % "currentfile readstring" and/or "image." } ifelse $error /newerror false put } bind def %%EndProlog (ORIGjob) jobdef % execute some included job directly in-line: /svobj save def (USERjob) jobdef currentfile cvx userexec %%BeginDocument: included.ps %!PS %%Title: included.ps %%For: triggering an error /var1 10 def /var2 20 def %%EndProlog %%Page: 1 1 var1 var2 triggererror more program statements which should be ignored showpage %%Trailer %%EndDocument % should never get this far: (USERjob completed.\n) print flush stop %ENDuserexec % here is where ORIGjob picks up again svobj restore % will restore ORIGjob name, too (Continuing with ) print errordict /jobname get print (\n) print flush %%Trailer

202

chapter 14 ERROR HANDLING

When an error is encountered, the userexec procedure must ush the rest of the included illustration. This is done by searching the rest of the le for a line containing only the string %ENDuserexec. This string would have been inserted at the end of the included code by the document composition software. The search is accomplished in this case with the readline and eq operators.

14.4 THE ERROR HANDLING MECHANISM

The PostScript language has a built-in procedure dened for each possible error condition. The procedure is invoked by the error handling mechanism in the PostScript interpreter. The following are the steps that occur when the interpreter triggers an execution error: The error occurs (say it is a limitcheck error). The PostScript interpreter nds a procedure named limitcheck in a dictionary known as errordict and executes it. The limitcheck procedure captures the contents of the operand stack, the execution stack, and the dictionary stack and places these into array objects in a subdictionary of errordict known as $error. The name of the error (in this case limitcheck) is stored in the $error dictionary under the name errorname. Similarly, the OffendingCommand (the object that was executing at the time of the error) is stored in $error under the name command. The limitcheck procedure then executes stop, and its job is done. Whenever stop is executed, control returns to the innermost stopped context. You can nest stopped contexts. In fact, the job server loop always executes user jobs with the stopped operator so that it can regain control when an error occurs.

chapter 14 ERROR HANDLING

203

The job server loop contains a standard error handling mechanism that looks for and invokes another standard procedure called handleerror (it expects to nd it in errordict) whenever the stopped operator returns true. The handleerror procedure produces the familiar error message (by looking in $error for the information stored there by the error operators): %%[Error: limitcheck; OffendingCommand: flattenpath]%% See the next section for an example of how to redene the handleerror procedure.

14.5 REDEFINING ERROR PROCEDURES

Any of the individual error procedures can be redened to be any arbitrary PostScript procedure. In addition, the higher level handleerror procedure can be redened to provide more information when an error condition is encountered. One widely used error handler prints the error message and the contents of the operand stack on the current page, for example. Listing 14-3 is a sample program that will redene an individual error procedure (in this case, the limitcheck procedure):

listing 14-3
%!PS-Adobe-2.0 %%EndComments errordict begin /*limitcheck /limitcheck load def /limitcheck { %def (Free VM: ) print vmstatus exch sub == flush pop *limitcheck } bind def end %errordict %%Trailer

204

chapter 14 ERROR HANDLING

This procedure produces a message back to the host computer indicating the amount of free memory in the printer when the limitcheck error occurred, and then invokes the original limitcheck procedure as well (it has been saved under the name *limitcheck in this example). In listing 14-4 is an example redenition of the handleerror procedure that produces a stack trace back to the host system when a PostScript error is triggered:

listing 14-4
%!PS-Adobe-2.0 %%Title: errstack.ps %%EndComments %%EndProlog %%BeginSetup /orighandleerror errordict /handleerror get def errordict begin /handleerror { %def orighandleerror errordict begin $error begin (STACK:) = flush currentdict /ostack known { ostack aload pop pstack flush clear } if end end } bind def end %%EndSetup %%Trailer

This procedure calls the existing error handler, and also returns a readable representation of the operand stack back to the host system.

chapter 14 ERROR HANDLING

205

14.6 HANDLING ERROR MESSAGES

Software used to drive a PostScript device has a certain responsibility to handle error messages (or other information) sent back by the PostScript interpreter. The standard (default) condition of the error handling mechanism in most printers is to produce a well-dened message back to the host computer and to ignore everything it receives until the next end-of-le indication. If the error handler is not redened, something reasonable must be done with the error messages. See Chapter 12 for further information on printer management issues. Depending on the system conguration, it may not be possible to pass the error message all the way back to the originator of the print le. The correct way to notify a user that his or her print job failed will be very environment-specic. In some cases, it may be appropriate to display the error on the screen of the host workstation. In others, perhaps the message can be written into a printer log le, or an electronic mail message sent. In still others instances, perhaps an error break page can be printed by the interpreter and sent back to the original source along with whatever pages of the document might have been printed correctly. In many cases, it is probably best to locally redene the behavior of the error-handling mechanism in the PostScript interpreter to suit a particular environment.

206

chapter 14 ERROR HANDLING

chapter 15 DEBUGGING TECHNIQUES

If a string has one end, then it must have another. /usr/games/fortune

15.1 INTRODUCTION
Since the PostScript language is interpreted, it is fairly easy to debug programs written in the language. However, the fact that the programs typically are executed on a remote processor (for example, in a laser printer) can make the process a little more difcult. In this chapter, it is assumed that the working environment consists of a host computer connected through some kind of communications interface to a printer containing a PostScript interpreter. There are other possible situations, but this is a common environment and one in which the debugging issues are perhaps the most difcult.

15.2 ESTABLISHING TWO-WAY COMMUNICATION

The rst step in reliably debugging a PostScript application is to build a solid connection between the host computer and the printer. There are several levels of communication that need to be set up and tested. PostScript language printers should be thought of as computers when setting up communications channels. In particular, full bidirectional communication should be expected. At the simplest level, the printer will send back error messages when an execution error occurs. It is also possible for a user program to generate arbitrary amounts of information at the printer and transmit it back to the host.

chapter 15 DEBUGGING TECHNIQUES

207

Since a large part of debugging consists of trying to determine what really happened, establishing solid communications is always a good rst step. This may consist of purchasing or writing a downloading utility program, or it may mean implementing the communications layer of your application rst.

SERIAL COMMUNICATIONS
There are three main aspects to setting up good serial communications: Get the cable congured correctly. Get ow control working properly. Be prepared to receive information at any moment from the PostScript device. Flow control may be implemented in software (the XON/XOFF protocol for serial connections, or a network protocol such as AppleTalk) or it may be handled by the hardware (the DTR/ DSR protocol, for example). Most PostScript devices are initially congured for software handshaking, and need explicitly to be set to handled other kinds of protocols. The following is an example that will work on most PostScript devices with serial communications (please consult the particular printer manual for possible differences): serverdict begin 0 exitserver statusdict begin 25 9600 4 setsccbatch end In this example, 25 refers to the 25-pin connector (as opposed to the 9-pin connector, for example). 9600 is the baud rate of the communications. 4 is a number from 0 to 7 that represents a combination of parity setting and the handshaking mechanism. Values from 0 to 3 use software handshaking (XON/XOFF), and values from 4 to 7 will turn on hardware handshaking instead (which may not be available on all devices). The setsccbatch operator establishes the default serial (scc) communications for the given port in batch mode (as opposed to interactive mode).

208

chapter 15 DEBUGGING TECHNIQUES

PARALLEL COMMUNICATIONS
Parallel ports are not available on all PostScript devices. Parallel communication ports in general do not support bi-directional transmissions, so if the parallel port is used, some special care must be used to receive any messages. Some devices may allow you to open the serial port explicitly to send information back to the host, but you have to know about it and it is not supported on all devices (please consult the documentation for the individual printer for more specic information). Parallel communications probably are not the best mode to use for general debugging of PostScript programs.

PACKET NETWORK COMMUNICATIONS

Network connections like AppleTalk or DECnet are available on some PostScript devices. The details of packet-switched network communications are too complex to be detailed here. Probably the best approach is to use existing network software to establish communication for debugging with this kind of network connection.

15.3 UNDERSTANDING POSTSCRIPT ERRORS

When a PostScript execution error occurs, the default action taken is to issue an error message. This message consists of two elements: The Error name. This is the name of the PostScript error that occurredfor example, stackunderflow. The OffendingCommand. This is the PostScript object that was being executed at the time of the error. A PostScript error message is very specic about what happened, but may not tell you why. For example, the moveto operator expects to nd two numbers on the operand stack, which it assumes are x and y coordinates in user space. There are only two possible execution errors that can arise from the moveto operator (although moveto will happily take any two numbers that it nds on the stack and make that the current point, regardless of what the numbers might bethis is a third kind of error):

chapter 15 DEBUGGING TECHNIQUES

209

%%[ Error: typecheck; OffendingCommand: moveto ]%% %%[ Error: stackunderflow; OffendingCommand: moveto ]%% Either too few items are available on the operand stack (causing stackunderflow) or those items are not numbers (causing a typecheck). Its that simple. Understanding why some PostScript error occurred usually requires a certain amount of context. For instance, it is helpful to know some of the following things: What items were on the operand stack at the moment the error occurred? At what point in the input le did the error occur? What dictionaries were on the dictionary stack? What was the current value of some variable or aspect of the interpreters state? Probably the most useful of these is to determine what objects are on the operand stack. The contents of the stack usually provide all the information that is necessary to establish what the context of the error was (although a stackunderflow error is certainly an exception to this, since there is nothing on the operand stack). See Section 15.5 for more details on getting stack traces.

ERROR: UNDEFINED
One of the more common errors in PostScript programming is the undefined error. This error is generated by the PostScript interpreter if a name lookup fails or if an explicit get, or load cannot nd its argument in the context of the current dictionary stack. Here are some possible causes of the undefined error: The name is spelled wrong. PostScript names are case sensitive. This is probably the most common problem.

210

chapter 15 DEBUGGING TECHNIQUES

 A procedure name was used before it was dened. This sometimes results when procedures are inadvertently moved in the le, or when the slash is missing from a name (see Section 2.4). If the name given as the OffendingCommand appears to be only part of one of the names in your program le, this may mean that the communications are not solid. If part of the le is lost on its way to the printer, the interpreter may pick up in the middle of a token, and give an error. Check especially for ow control problems. If the get operator is being used at all in the program, perhaps its arguments are in the wrong order or just incorrect. Oddly, the OffendingCommand will not be get, but will be the key that is passed to get. If the name being looked up is in a special dictionary (for instance, statusdict, serverdict, or a local dictionary dened by the program), that dictionary may not be on the dictionary stack. Automatic name lookup works only in the context of the dictionary stack. Use the begin operator to push dictionaries onto the dictionary stack. ^D: A control-D character (decimal ASCII 04), is the end-of-le indication over most serial connections. However, it is not an end-of-le over networks like AppleTalk. These should be removed from the le stream. Actually, they should never be put there to begin with. (See Section 12.3.) If the byte gets through to the PostScript interpreter (across AppleTalk, for instance) it will provoke an undefined error.

ERROR: TYPECHECK
This is probably the most common error in PostScript programming. Typically it results from the incorrect manipulation of objects on the operand stack. Sometimes this is a result of not realizing that a given PostScript operator might return something to you on the stack. Here are some common causes of the typecheck error:

chapter 15 DEBUGGING TECHNIQUES

211

 stringwidth returns both an x and a y value. For most fonts, the y component is 0. If this value is not needed, it should be removed from the stack with the pop operator. Very often an extra integer turns up on the stack from forgetting how the stringwidth operator works. Some operators require several operands. Be careful to supply all of them, and in the correct order. There is a difference between an integer and a real number. Make sure you supply each operator with the correct type. This is especially important with the idiv operator, since early versions of the PostScript interpreter graciously permitted real operands as arguments when they should not have. Do not supply an integer where a boolean is required. 0 and 1 are not equivalent to true and false, which are genuine PostScript objects of type booleantype. Use the pstack operator to help you debug problems with stack manipulation. Sometimes the roll or exch operators can help when operators are in the wrong order, although typically it means that the operands should have been transmitted in a different order.

15.4 REDEFINING BUILT-IN OPERATORS

Any of the names of PostScript operators may be redened. This is done simply by dening a new value for that name in a dictionary other than systemdict. This can be useful for debugging purposes. (See listing 15-1.) The redenitions of moveto and lineto in this example print messages back to the host each time they are called. The original denitions of moveto and lineto are still invoked (as *moveto and *lineto), so the execution of the le is not be disturbed in any way. The redenitions are local to the debug dictionary. As long as that dictionary is not on the dictionary stack, these denitions

212

chapter 15 DEBUGGING TECHNIQUES

are not available. Placing debug begin at the beginning of a document (and end at the end, of course, if the program gets that far) will cause the le to execute with these helpful debugging procedures available.

listing 15-1
%!PS-Adobe-2.0 %%Title: pathtrace.ps %%Creator: Glenn Reid %%EndComments /debug 10 dict def debug begin /str 128 string def /*moveto /moveto load def /*lineto /lineto load def /moveto { %def (moveto: ) print exch dup str cvs print % X-val ( ) print exch dup == % Y-val flush *moveto % execute real moveto } bind def /lineto { %def (lineto: ) print exch dup str cvs print % X-val ( ) print exch dup == % Y-val flush *lineto % execute real lineto } bind def end %%EndProlog debug begin 100 100 translate 100 100 moveto 200 200 lineto stroke showpage end %%Trailer

chapter 15 DEBUGGING TECHNIQUES

213

15.5 STACK TRACES

Getting a trace of the operand stack when an error occurs provides very useful debugging information. The only reliable way to accomplish this is to redene the error handling procedure to dump the stack when any error occurs. There is an example in Section 14.5 (listing 14-4) that redenes the handleerror procedure to produce a stack trace. There is also a widely available PostScript le known as ehandler.ps that provides a slightly different redenition of this procedure to actually print the error and stack trace on the current page when an error occurs. This is especially useful in that the page image that is already built can be seen even if showpage has not been reached in the program le. This can help narrow down exactly where the error might have occurred. A listing of ehandler.ps is included in Appendix B. A stack trace of the operand stack is useful for two reasons. It can help nd the spot in the le that the error occurred by providing some of the context in which the error was triggered. It also helps to discover why the error happened. One of the most common errors in PostScript programs is incorrect manipulation of the operand stack. Seeing a snapshot of the stack when an error occurs helps solve this kind of error quickly.

15.6 INTERACTIVE TECHNIQUES

Under some circumstances, it is possible to interact with a PostScript interpreter directly. In some devices, there is an operator called executive which will provide an interactive context for PostScript programming. A prompt is provided, and simple line-editing primitives are supported (such as backspace and delete-line). This can be a particularly useful environment for testing simple algorithms or exploring the nature of the PostScript interpreter. If a direct connection can be established with the PostScript interpreter (for instance, by hooking up a simple terminal to the printers communications port), an interactive session can be initiated by typing ^D (control-D, or end-of-le) followed by the word executive (and a carriage return). The word executive

214

chapter 15 DEBUGGING TECHNIQUES

will not echo back onto your screen as you type it, but the characters will be received by the PostScript interpreter. If you type it correctly, you will see a brief message and a PS> prompt. (If you dont, you will get an error.) From that point on (through the next end-of-le), you will be in interactive mode. PostScript language commands received by the interpreter in interactive mode will be interpreted in the same way that batch les are interpreted, including executing procedures, printing characters, using showpage; the whole language is at your ngertips.

15.7 COORDINATE SYSTEM TRANSFORMATIONS

A common error in PostScript language programming that is difcult to debug is the off-the-page error. That is, the PostScript program may execute correctly, and a page will be printed, but the page is completely blank. Sometimes the reason for this is that the coordinate system mapping has been modied in such a way that the image is actually painted off the page somewhere. This is not an error condition to the PostScript interpreter. Any part of any image may be off the page. Remember that, conceptually, it is user space that rotates, not device space. To simulate this, tape a sheet of paper to the oor in front of you, and focus your eyes on the lower left corner of the page. That is the origin of the coordinate system. If you were to perform a 90 rotate, think of user space rotation 90 degrees counterclockwise (positive rotation is always counterclockwise; negative rotation is clockwise). Imagine that your head is user space. That is, rotate yourself 90 degrees counterclockwise over the page. Notice that the Y axis is now along the short edge of the paper, and it is behind you (if you are still focusing on that corner). Translations take effect along the current axes. For instance, to nish setting landscape mode from there, it would be necessary to add a 0 -612 translate: 90 rotate 0 -612 translate If the previous discussion makes no sense to you, dont worry; it is just one way to visualize the transformations.

chapter 15 DEBUGGING TECHNIQUES

215

The relationship between rotate and scale is not a commutative one. That is, you cannot reverse the order of evaluation and get the same result. There is a further discussion of coordinate system transformations in Section 9.4.

15.8 DEBUGGING MESSAGES

A helpful technique for locating the place in a program le where an error occurred is to insert messages at key points in the le. Listing 15-2 is an example of this technique. These messages may also be put into procedure bodies, as seen in listing 151, where operators are redened to return status information.

listing 15-2
%!PS-Adobe-2.0 %%EndComments /msg { %def print (\n) print flush } bind def %%EndProlog %%Page: 1 1 (Got past the prologue.) msg %%Trailer (made it to the trailer) msg

216

chapter 15 DEBUGGING TECHNIQUES

appendix A ERROR HANDLER

%!PS-Adobe-2.0 % ehandler.ps -- Downloaded Error Break-page handler % Copyright (c) 1984, 1985, 1986 Adobe Systems Incorporated. % All Rights Reserved. 0000 % exitserver password /$brkpage where { %ifelse pop pop (Error Handler in place - not loaded again\n) print flush stop }{ %else dup serverdict begin statusdict begin checkpassword { %ifelse (Error Handler downloaded.\n)print flush exitserver }{ %else pop (Bad Password on loading error handler!!!\n) print flush stop } ifelse } ifelse /$brkpage 64 dict def $brkpage begin /prnt { %def dup type /stringtype ne { =string cvs } if dup length 6 mul /tx exch def /ty 10 def currentpoint /toy exch def /tox exch def 1 setgray newpath tox toy 2 sub moveto 0 ty rlineto tx 0 rlineto 0 ty neg rlineto closepath fill tox toy moveto 0 setgray show } bind def /nl { %def currentpoint exch pop lmargin exch moveto 0 -10 rmoveto } def

appendix A

217

/== { /cp 0 def typeprint nl } def /typeprint { dup type dup currentdict exch known {exec}{ unknowntype }ifelse } readonly def /lmargin 72 def /rmargin 72 def /tprint { %def dup length cp add rmargin gt { nl /cp 0 def } if dup length cp add /cp exch def prnt } readonly def /cvsprint { =string cvs tprint ( ) tprint } readonly def /unknowntype { %def exch pop cvlit (??) tprint cvsprint } readonly def /integertype { cvsprint } readonly def /realtype { cvsprint } readonly def /booleantype { cvsprint } readonly def /operatortype { (//) tprint cvsprint } readonly def /marktype { pop (-mark- ) tprint } readonly def /dicttype { pop (-dictionary- ) tprint } readonly def /nulltype { pop (-null- ) tprint } readonly def /filetype { pop (-filestream- ) tprint } readonly def /savetype { pop (-savelevel- ) tprint } readonly def /fonttype { pop (-fontid- ) tprint } readonly def /nametype { %def dup xcheck not { (/) tprint } if cvsprint } readonly def /stringtype { %def dup rcheck { %ifelse (\() tprint tprint (\) )tprint }{ %else pop (-string- ) tprint } ifelse } readonly def /arraytype { %def dup rcheck { %ifelse dup xcheck { %ifelse ({ ) tprint { typeprint } forall ( }) tprint } { %else ([ ) tprint { typeprint } forall ( ]) tprint } ifelse }{ %else pop (-array- ) tprint

218

appendix A

} ifelse } readonly def /packedarraytype { %def dup rcheck { %ifelse dup xcheck { %ifelse ({ ) tprint { typeprint } forall ( }) tprint }{ %else ([ ) tprint { typeprint } forall ( ]) tprint } ifelse }{ %else pop (-packedarray- ) tprint } ifelse } readonly def /courier /Courier findfont 10 scalefont def /OLDhandleerror errordict /handleerror get def end %$brkpage /handleerror { %put systemdict begin $error begin $brkpage begin newerror { %ifelse /newerror false store vmstatus pop pop 0 ne { grestoreall } if initgraphics courier setfont lmargin 720 moveto (ERROR: ) prnt errorname prnt nl (OFFENDING COMMAND: ) prnt /command load prnt $error /ostack known { %if nl nl (STACK:) prnt nl nl $error /ostack get aload length { == } repeat } if systemdict /showpage get exec /newerror true store /OLDhandleerror load end end end exec }{ %else end end end } ifelse } dup 0 systemdict put % replace name by actual dict object dup 4 $brkpage put % replace name by dict object bind readonly errordict 3 1 roll put % put proc in errordict as /handleerror

appendix A

219

INDEX
Non-Alphabetic
%!PS-Adobe- 91 %%BeginFont 166, 182 %%BoundingBox 167 %%DocumentFonts 166, 181 %%DocumentNeededFonts 166, 181 %%DocumentSuppliedFonts 166 %%EndFont 166, 182 %%IncludeFont 166, 181 %%Page 168 %stdin 59 conceptual error 95 conforming files 91, 98, 159 context graphic 78 context dependencies 91 coordinate system 69, 217 coordinate systems 77 copypage 146 curly brace 26 current dictionary 18 current file 59 current font 49 current path 43, 49 current point 75 currentfile operator 59, 171, 195

A
addprocs 31 Adobe Illustrator xii AFM files 103, 104 arc operator 85 array 188 arrowhead procedure 46 ascender 111 ashow 102 attributes 15 awidthshow 70

D
data acquisition procedure 126, 128 data compression 133 data transmission time 81 data, program-generated 90 debugging 209 defaultmatrix 79 defaults coordinate system 92 current font 93 current page 92 current path 93 default context 92 dictionary stack 93 error handler 208 exitserver password 179 deferred execution 26 definefont 192 descender 111 device space 217 dictionary 2 dictionary stack 213 disk user partition 194 disk management 194 displays 173 Downloadable fonts 192 driver 79

B
band buffers 193 batch mode 11 baud rate 210 begin operator 213 bi-directional communication 209 bind 28 bounding box 104, 174

C
capital height 111 case sensitive 212 character metric 102 character set 116 character width 103 character widths 51, 102 circle procedure 46 clip operator 53, 140, 200 clipping region 140 comments, structure 91, 98, 159 communications and spoolers 176 packet network 211 parallel 211 serial 210 composite characters 119 computation 81 computation in PostScript 73 concat 151 concatenating procedures 31

E
efficiency 11, 81, 83, 85 calculations 80 imaging model translation 79 Einstein, Albert 1 emulator 57 escape sequences 59, 61 fonts 57 line printer 57 using search 59 using stringwidth 61

INDEX

223

encoding vector 117 end-of-file 34, 176, 179 error execution 199, 211 tactics 199 typecheck 213 error handler 207 error handling mechanism 205 error messages 209 error procedures 206 errordict 206 escape character 61 escape sequences 63 exec operator 31 executable array 27 executable name 20 execution environment 93 execution error 199 execution stack 32, 34 execution, deferred 26 executive operator 178, 216 exitserver 178

graphics problem solving 139 graphics state 92, 189 gray shades 133 grestoreall 79 grids 148 ground state 92

H
halftone angle 133, 134 cells 134 device dependencies 139 frequency 133, 134 pattern fills 139 pixel priority 137 setscreen operator 134 spot function 133, 134, 137 halftoning 125, 133 handleerror 206, 216 high resolution devices 200

F
features 160 file formats, existing 68 file object 32, 34, 195 file system 194 flattenpath 200 font cache 55 font management 183 FontDirectory 192 FontMatrix 192 fonts accented characters 119 and emulators 69 ascender height 111 capital height 111 character bounding box 104 character widths 61 descender height 111 dictionaries 192 downloadable 192 encoding 116 handling different 108 in emulators 57 leading 111 memory use 192 metrics 103 non-Roman 119 point size 111 screen fonts 103 substitution 70 x-height 111 Frame buffers 193 framedevice 193 functional independence 78

I
illustrations 170 images bits per pixel 126 data acquisition procedure 126 image matrix 126 matrix 126 parameters 125 premature end-of-file 131 scanned 127 synthetic data 131 using image operator 127 image operator 125 imaging model 37, 57, 71 designing the driver 77 efficiency 79 immediate execution 27 implementation limits 200 income tax form 145 indentation xiii index 221, 225 initgraphics 79, 93 input stream 32, 195 integration 169 interactive PostScript 216 interpretation overhead 81 interpretation time 85 invalidpassword 178 invalidrestore error 190, 191

J
job 34 job execution 11 jobname 173, 203 justified text 62, 70

G
get operator 212 graphic design 101 graphic imaging model 71 graphics internal representation 38

K
kerning 101, 106, 112 kernshow procedure 114 kernstackshow procedure 114 key-value pair 28 kshow operator 113

224

INDEX

L
landscape mode 79 leading 111 ledger paper 161 ledgertray 161 legibility 107 letter spacing 106 ligatures 116 limitcheck 53, 140, 200 line weight 148, 149 line wrapping 74 literal name 20 load operator 212 logotypes 145

M
machine-generated code 79 margins 101, 104 matrix operations 151 memory array object 192 strings 188 VMerror 201 memory, virtual 187 merging files 169 messages, printer 177 modularity 92, 93 graphic modularity 95 save and restore 95 moveto 105

N
name lookup, automatic 20 names binding 28 nesting, program file 170 newpath 38, 49

O
object properties 96 objects access attribute 15 attributes 15 composite 13 executable 15 length 15 literal 15 simple 13 type 15 OffendingCommand 211 operand stack 13, 94, 212 operators 19, 21 operators, non-standard 199 optimization 73 rmoveto vs. widthshow 73 output stream 195

page layout 77 page, top of 78 pages page independence 90 painting operators 38 paper sizes 161 parallel processing 146 parameters 94 pasteup 169 path construction 38 paths 37, 38 pattern fills 139 pattern font 140 performance 85 pixel locking 149 Poe, Edgar Allan 105 point size 111 PostScript Document Structuring Conventions 98 PostScript error 211 PostScript Interpreter 3 PostScript Printer Description Files 160 preview 174 print flush 218 print format translator 57, 68 printer emulator 57, 196 printer management 176 problems 139 procedure 25, 145 procedure bodies xiii procedure body 26 procedure definitions 90 procedure sets 89 procedures concatenating 31 designing prologue 80 short names 82 very large 29 program design 77 programming style xii prologue 89 related to script 80 prologue procedures 80 properties, object 96 pstack operator 214

Q
queries, printer 164 query 164 font 183 query comments 164 queueing 175

R
raster devices 72 rasterization 54, 193 readhexstring 130, 196 readline operator 59, 195 readstring 59 rectangle procedure 44 redefining operators 214 restore 189 and execution stack 191 and the dictionary stack 191

P
packed arrays 192 page element 78 page elements 89 page independence 78

INDEX

225

robustness 89 rotate 151, 217 round-off, numeric 80, 149

S
sampled images 125 save save objects 189 versus gsave 190 save & restore server loop 34 save & restore 52, 54, 188, 189 scale 151 scan time 82 scanner 23 strings 23 screen fonts 103 script 89 machine-generated 80 search 59, 61 server loop 34, 178 serverdict 179, 213 setcharwidth 55 setsccbatch 210 setscreen 134 show 49, 62, 102, 105, 117 showpage 54, 172 sidebearings 103 spooler 175 stack 4 stack trace 216 stack, data on 44 stack-based 2 stackshow procedure 114 stackunderflow 94, 211, 212 state using existing 79 statusdict 213 stopped context 205 stopped operator 171, 201 string objects 23 string, literal 188 strings 23 stringwidth 52, 61, 108 stroke 148 style ease of editing xii indentation xiii readability xii switching fonts 70 syntaxerror 24 systemdict 214

line length 104 margins 104, 105 token operator 24 tokenization 23, 81 trailer 89 transform 149 transformation matrices 151 transformations 217 translate 151 translator optimization 72 typecheck 94, 212, 213 typography 101

U
undefined 212 units 68 user space 217 user-defined font 147 userexec procedure 203

V
variables 15 virtual memory 187 VM 187 dictionary stack 187 execution stack 187 operand stack 187 userdict 187 VMerror 201 vmstatus 187

W
well-formed 89 where operator 200 widths, character 61, 62, 103 widthshow 70, 73, 102, 107 word spacing 70, 107

X
x-height 111

T
tab character 61 text baseline 49 centered 104 columns 78 flow 104 flush left 104 flush right 104 justified 104 leading 111

226

INDEX

BOOK DESIGN
This book design is borrowed from an earlier design by Robert Ishi for the PostScript Language Reference Manual and was adapted by Glenn Reid to the technology at hand and the unique demands of the book. The type used is entirely from the Stone family, designed at Adobe Systems by Sumner Stone, Director of Typography. Chapter headings are set in Stone Sans Semibold 24 point, section headings are set in Stone Sans 12 point, and the body text is set in 10 on 12 point Stone Serif with Italic and SemiBold. All example PostScript language programs and listings are set in 10 point Stone Informal, using the regular and Semibold weights.