\input texinfo @c -*- texinfo -*- @comment ======================================================== @comment %**start of header @setfilename m4.info @include version.texi @settitle GNU M4 @value{VERSION} macro processor @setchapternewpage odd @ifnothtml @setcontentsaftertitlepage @end ifnothtml @finalout @c @tabchar{} @c ---------- @c The testsuite expects literal tab output in some examples, but @c literal tabs in texinfo lead to formatting issues. @macro tabchar @ @c @end macro @c @ovar{ARG} @c ------------------- @c The ARG is an optional argument. To be used for macro arguments in @c their documentation (@defmac). @macro ovar{varname} @r{[}@var{\varname\}@r{]}@c @end macro @c @dvar{ARG, DEFAULT} @c ------------------- @c The ARG is an optional argument, defaulting to DEFAULT. To be used @c for macro arguments in their documentation (@defmac). @macro dvar{varname, default} @r{[}@var{\varname\} = @samp{\default\}@r{]}@c @end macro @comment %**end of header @comment ======================================================== @copying This manual (@value{UPDATED}) is for GNU M4 (version @value{VERSION}), a package containing an implementation of the m4 macro language. Copyright @copyright{} 1989-1994, 2004-2011 Free Software Foundation, Inc. @quotation Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.3 or any later version published by the Free Software Foundation; with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts. A copy of the license is included in the section entitled ``GNU Free Documentation License.'' @end quotation @end copying @dircategory Text creation and manipulation @direntry * M4: (m4). A powerful macro processor. @end direntry @titlepage @title GNU M4, version @value{VERSION} @subtitle A powerful macro processor @subtitle Edition @value{EDITION}, @value{UPDATED} @author by Ren@'e Seindal, Fran@,{c}ois Pinard, @author Gary V. Vaughan, and Eric Blake @author (@email{bug-m4@@gnu.org}) @page @vskip 0pt plus 1filll @insertcopying @end titlepage @contents @ifnottex @node Top @top GNU M4 @insertcopying @end ifnottex GNU @code{m4} is an implementation of the traditional UNIX macro processor. It is mostly SVR4 compatible, although it has some extensions (for example, handling more than 9 positional parameters to macros). @code{m4} also has builtin functions for including files, running shell commands, doing arithmetic, etc. Autoconf needs GNU @code{m4} for generating @file{configure} scripts, but not for running them. GNU @code{m4} was originally written by Ren@'e Seindal, with subsequent changes by Fran@,{c}ois Pinard and other volunteers on the Internet. All names and email addresses can be found in the files @file{m4-@value{VERSION}/@/AUTHORS} and @file{m4-@value{VERSION}/@/THANKS} from the GNU M4 distribution. This is release @value{VERSION}. It is now considered stable: future releases in the 1.4.x series are only meant to fix bugs, increase speed, or improve documentation. However@dots{} An experimental feature, which would improve @code{m4} usefulness, allows for changing the syntax for what is a @dfn{word} in @code{m4}. You should use: @comment ignore @example ./configure --enable-changeword @end example @noindent if you want this feature compiled in. The current implementation slows down @code{m4} considerably and is hardly acceptable. In the future, @code{m4} 2.0 will come with a different set of new features that provide similar capabilities, but without the inefficiencies, so changeword will go away and @emph{you should not count on it}. @menu * Preliminaries:: Introduction and preliminaries * Invoking m4:: Invoking @code{m4} * Syntax:: Lexical and syntactic conventions * Macros:: How to invoke macros * Definitions:: How to define new macros * Conditionals:: Conditionals, loops, and recursion * Debugging:: How to debug macros and input * Input Control:: Input control * File Inclusion:: File inclusion * Diversions:: Diverting and undiverting output * Text handling:: Macros for text handling * Arithmetic:: Macros for doing arithmetic * Shell commands:: Macros for running shell commands * Miscellaneous:: Miscellaneous builtin macros * Frozen files:: Fast loading of frozen state * Compatibility:: Compatibility with other versions of @code{m4} * Answers:: Correct version of some examples * Copying This Package:: How to make copies of the overall M4 package * Copying This Manual:: How to make copies of this manual * Indices:: Indices of concepts and macros @detailmenu --- The Detailed Node Listing --- Introduction and preliminaries * Intro:: Introduction to @code{m4} * History:: Historical references * Bugs:: Problems and bugs * Manual:: Using this manual Invoking @code{m4} * Operation modes:: Command line options for operation modes * Preprocessor features:: Command line options for preprocessor features * Limits control:: Command line options for limits control * Frozen state:: Command line options for frozen state * Debugging options:: Command line options for debugging * Command line files:: Specifying input files on the command line Lexical and syntactic conventions * Names:: Macro names * Quoted strings:: Quoting input to @code{m4} * Comments:: Comments in @code{m4} input * Other tokens:: Other kinds of input tokens * Input processing:: How @code{m4} copies input to output How to invoke macros * Invocation:: Macro invocation * Inhibiting Invocation:: Preventing macro invocation * Macro Arguments:: Macro arguments * Quoting Arguments:: On Quoting Arguments to macros * Macro expansion:: Expanding macros How to define new macros * Define:: Defining a new macro * Arguments:: Arguments to macros * Pseudo Arguments:: Special arguments to macros * Undefine:: Deleting a macro * Defn:: Renaming macros * Pushdef:: Temporarily redefining macros * Indir:: Indirect call of macros * Builtin:: Indirect call of builtins Conditionals, loops, and recursion * Ifdef:: Testing if a macro is defined * Ifelse:: If-else construct, or multibranch * Shift:: Recursion in @code{m4} * Forloop:: Iteration by counting * Foreach:: Iteration by list contents * Stacks:: Working with definition stacks * Composition:: Building macros with macros How to debug macros and input * Dumpdef:: Displaying macro definitions * Trace:: Tracing macro calls * Debug Levels:: Controlling debugging output * Debug Output:: Saving debugging output Input control * Dnl:: Deleting whitespace in input * Changequote:: Changing the quote characters * Changecom:: Changing the comment delimiters * Changeword:: Changing the lexical structure of words * M4wrap:: Saving text until end of input File inclusion * Include:: Including named files * Search Path:: Searching for include files Diverting and undiverting output * Divert:: Diverting output * Undivert:: Undiverting output * Divnum:: Diversion numbers * Cleardivert:: Discarding diverted text Macros for text handling * Len:: Calculating length of strings * Index macro:: Searching for substrings * Regexp:: Searching for regular expressions * Substr:: Extracting substrings * Translit:: Translating characters * Patsubst:: Substituting text by regular expression * Format:: Formatting strings (printf-like) Macros for doing arithmetic * Incr:: Decrement and increment operators * Eval:: Evaluating integer expressions Macros for running shell commands * Platform macros:: Determining the platform * Syscmd:: Executing simple commands * Esyscmd:: Reading the output of commands * Sysval:: Exit status * Mkstemp:: Making temporary files Miscellaneous builtin macros * Errprint:: Printing error messages * Location:: Printing current location * M4exit:: Exiting from @code{m4} Fast loading of frozen state * Using frozen files:: Using frozen files * Frozen file format:: Frozen file format Compatibility with other versions of @code{m4} * Extensions:: Extensions in GNU M4 * Incompatibilities:: Facilities in System V m4 not in GNU M4 * Other Incompatibilities:: Other incompatibilities Correct version of some examples * Improved exch:: Solution for @code{exch} * Improved forloop:: Solution for @code{forloop} * Improved foreach:: Solution for @code{foreach} * Improved copy:: Solution for @code{copy} * Improved m4wrap:: Solution for @code{m4wrap} * Improved cleardivert:: Solution for @code{cleardivert} * Improved capitalize:: Solution for @code{capitalize} * Improved fatal_error:: Solution for @code{fatal_error} How to make copies of the overall M4 package * GNU General Public License:: License for copying the M4 package How to make copies of this manual * GNU Free Documentation License:: License for copying this manual Indices of concepts and macros * Macro index:: Index for all @code{m4} macros * Concept index:: Index for many concepts @end detailmenu @end menu @node Preliminaries @chapter Introduction and preliminaries This first chapter explains what GNU @code{m4} is, where @code{m4} comes from, how to read and use this documentation, how to call the @code{m4} program, and how to report bugs about it. It concludes by giving tips for reading the remainder of the manual. The following chapters then detail all the features of the @code{m4} language. @menu * Intro:: Introduction to @code{m4} * History:: Historical references * Bugs:: Problems and bugs * Manual:: Using this manual @end menu @node Intro @section Introduction to @code{m4} @cindex overview of @code{m4} @code{m4} is a macro processor, in the sense that it copies its input to the output, expanding macros as it goes. Macros are either builtin or user-defined, and can take any number of arguments. Besides just doing macro expansion, @code{m4} has builtin functions for including named files, running shell commands, doing integer arithmetic, manipulating text in various ways, performing recursion, etc.@dots{} @code{m4} can be used either as a front-end to a compiler, or as a macro processor in its own right. The @code{m4} macro processor is widely available on all UNIXes, and has been standardized by POSIX. Usually, only a small percentage of users are aware of its existence. However, those who find it often become committed users. The popularity of GNU Autoconf, which requires GNU @code{m4} for @emph{generating} @file{configure} scripts, is an incentive for many to install it, while these people will not themselves program in @code{m4}. GNU @code{m4} is mostly compatible with the System V, Release 3 version, except for some minor differences. @xref{Compatibility}, for more details. Some people find @code{m4} to be fairly addictive. They first use @code{m4} for simple problems, then take bigger and bigger challenges, learning how to write complex sets of @code{m4} macros along the way. Once really addicted, users pursue writing of sophisticated @code{m4} applications even to solve simple problems, devoting more time debugging their @code{m4} scripts than doing real work. Beware that @code{m4} may be dangerous for the health of compulsive programmers. @node History @section Historical references @cindex history of @code{m4} @cindex GNU M4, history of @code{GPM} was an important ancestor of @code{m4}. See C. Strachey: ``A General Purpose Macro generator'', Computer Journal 8,3 (1965), pp.@: 225 ff. @code{GPM} is also succinctly described into David Gries classic ``Compiler Construction for Digital Computers''. The classic B. Kernighan and P.J. Plauger: ``Software Tools'', Addison-Wesley, Inc.@: (1976) describes and implements a Unix macro-processor language, which inspired Dennis Ritchie to write @code{m3}, a macro processor for the AP-3 minicomputer. Kernighan and Ritchie then joined forces to develop the original @code{m4}, as described in ``The M4 Macro Processor'', Bell Laboratories (1977). It had only 21 builtin macros. While @code{GPM} was more @emph{pure}, @code{m4} is meant to deal with the true intricacies of real life: macros can be recognized without being pre-announced, skipping whitespace or end-of-lines is easier, more constructs are builtin instead of derived, etc. Originally, the Kernighan and Plauger macro-processor, and then @code{m3}, formed the engine for the Rational FORTRAN preprocessor, that is, the @code{Ratfor} equivalent of @code{cpp}. Later, @code{m4} was used as a front-end for @code{Ratfor}, @code{C} and @code{Cobol}. Ren@'e Seindal released his implementation of @code{m4}, GNU @code{m4}, in 1990, with the aim of removing the artificial limitations in many of the traditional @code{m4} implementations, such as maximum line length, macro size, or number of macros. The late Professor A. Dain Samples described and implemented a further evolution in the form of @code{M5}: ``User's Guide to the M5 Macro Language: 2nd edition'', Electronic Announcement on comp.compilers newsgroup (1992). Fran@,{c}ois Pinard took over maintenance of GNU @code{m4} in 1992, until 1994 when he released GNU @code{m4} 1.4, which was the stable release for 10 years. It was at this time that GNU Autoconf decided to require GNU @code{m4} as its underlying engine, since all other implementations of @code{m4} had too many limitations. More recently, in 2004, Paul Eggert released 1.4.1 and 1.4.2 which addressed some long standing bugs in the venerable 1.4 release. Then in 2005, Gary V. Vaughan collected together the many patches to GNU @code{m4} 1.4 that were floating around the net and released 1.4.3 and 1.4.4. And in 2006, Eric Blake joined the team and prepared patches for the release of 1.4.5, 1.4.6, 1.4.7, and 1.4.8. More bug fixes were incorporated in 2007, with releases 1.4.9 and 1.4.10. Eric continued with some portability fixes for 1.4.11 and 1.4.12 in 2008, 1.4.13 in 2009, 1.4.14 and 1.4.15 in 2010, and 1.4.16 in 2011. Meanwhile, development has continued on new features for @code{m4}, such as dynamic module loading and additional builtins. When complete, GNU @code{m4} 2.0 will start a new series of releases. @node Bugs @section Problems and bugs @cindex reporting bugs @cindex bug reports @cindex suggestions, reporting If you have problems with GNU M4 or think you've found a bug, please report it. Before reporting a bug, make sure you've actually found a real bug. Carefully reread the documentation and see if it really says you can do what you're trying to do. If it's not clear whether you should be able to do something or not, report that too; it's a bug in the documentation! Before reporting a bug or trying to fix it yourself, try to isolate it to the smallest possible input file that reproduces the problem. Then send us the input file and the exact results @code{m4} gave you. Also say what you expected to occur; this will help us decide whether the problem was really in the documentation. Once you've got a precise problem, send e-mail to @email{bug-m4@@gnu.org}. Please include the version number of @code{m4} you are using. You can get this information with the command @kbd{m4 --version}. Also provide details about the platform you are executing on. Non-bug suggestions are always welcome as well. If you have questions about things that are unclear in the documentation or are just obscure features, please report them too. @node Manual @section Using this manual @cindex examples, understanding This manual contains a number of examples of @code{m4} input and output, and a simple notation is used to distinguish input, output and error messages from @code{m4}. Examples are set out from the normal text, and shown in a fixed width font, like this @comment ignore @example This is an example of an example! @end example To distinguish input from output, all output from @code{m4} is prefixed by the string @samp{@result{}}, and all error messages by the string @samp{@error{}}. When showing how command line options affect matters, the command line is shown with a prompt @samp{$ @kbd{like this}}, otherwise, you can assume that a simple @kbd{m4} invocation will work. Thus: @comment ignore @example $ @kbd{command line to invoke m4} Example of input line @result{}Output line from m4 @error{}and an error message @end example The sequence @samp{^D} in an example indicates the end of the input file. The sequence @samp{@key{NL}} refers to the newline character. The majority of these examples are self-contained, and you can run them with similar results by invoking @kbd{m4 -d}. In fact, the testsuite that is bundled in the GNU M4 package consists of the examples in this document! Some of the examples assume that your current directory is located where you unpacked the installation, so if you plan on following along, you may find it helpful to do this now: @comment ignore @example $ @kbd{cd m4-@value{VERSION}} @end example As each of the predefined macros in @code{m4} is described, a prototype call of the macro will be shown, giving descriptive names to the arguments, e.g., @deffn Composite example (@var{string}, @dvar{count, 1}, @ @ovar{argument}@dots{}) This is a sample prototype. There is not really a macro named @code{example}, but this documents that if there were, it would be a Composite macro, rather than a Builtin. It requires at least one argument, @var{string}. Remember that in @code{m4}, there must not be a space between the macro name and the opening parenthesis, unless it was intended to call the macro without any arguments. The brackets around @var{count} and @var{argument} show that these arguments are optional. If @var{count} is omitted, the macro behaves as if count were @samp{1}, whereas if @var{argument} is omitted, the macro behaves as if it were the empty string. A blank argument is not the same as an omitted argument. For example, @samp{example(`a')}, @samp{example(`a',`1')}, and @samp{example(`a',`1',)} would behave identically with @var{count} set to @samp{1}; while @samp{example(`a',)} and @samp{example(`a',`')} would explicitly pass the empty string for @var{count}. The ellipses (@samp{@dots{}}) show that the macro processes additional arguments after @var{argument}, rather than ignoring them. @end deffn @cindex numbers All macro arguments in @code{m4} are strings, but some are given special interpretation, e.g., as numbers, file names, regular expressions, etc. The documentation for each macro will state how the parameters are interpreted, and what happens if the argument cannot be parsed according to the desired interpretation. Unless specified otherwise, a parameter specified to be a number is parsed as a decimal, even if the argument has leading zeros; and parsing the empty string as a number results in 0 rather than an error, although a warning will be issued. This document consistently writes and uses @dfn{builtin}, without a hyphen, as if it were an English word. This is how the @code{builtin} primitive is spelled within @code{m4}. @node Invoking m4 @chapter Invoking @code{m4} @cindex command line @cindex invoking @code{m4} The format of the @code{m4} command is: @comment ignore @example @code{m4} @r{[}@var{option}@dots{}@r{]} @r{[}@var{file}@dots{}@r{]} @end example @cindex command line, options @cindex options, command line @cindex @env{POSIXLY_CORRECT} All options begin with @samp{-}, or if long option names are used, with @samp{--}. A long option name need not be written completely, any unambiguous prefix is sufficient. POSIX requires @code{m4} to recognize arguments intermixed with files, even when @env{POSIXLY_CORRECT} is set in the environment. Most options take effect at startup regardless of their position, but some are documented below as taking effect after any files that occurred earlier in the command line. The argument @option{--} is a marker to denote the end of options. With short options, options that do not take arguments may be combined into a single command line argument with subsequent options, options with mandatory arguments may be provided either as a single command line argument or as two arguments, and options with optional arguments must be provided as a single argument. In other words, @kbd{m4 -QPDfoo -d a -df} is equivalent to @kbd{m4 -Q -P -D foo -d -df -- ./a}, although the latter form is considered canonical. With long options, options with mandatory arguments may be provided with an equal sign (@samp{=}) in a single argument, or as two arguments, and options with optional arguments must be provided as a single argument. In other words, @kbd{m4 --def foo --debug a} is equivalent to @kbd{m4 --define=foo --debug= -- ./a}, although the latter form is considered canonical (not to mention more robust, in case a future version of @code{m4} introduces an option named @option{--default}). @code{m4} understands the following options, grouped by functionality. @menu * Operation modes:: Command line options for operation modes * Preprocessor features:: Command line options for preprocessor features * Limits control:: Command line options for limits control * Frozen state:: Command line options for frozen state * Debugging options:: Command line options for debugging * Command line files:: Specifying input files on the command line @end menu @node Operation modes @section Command line options for operation modes Several options control the overall operation of @code{m4}: @table @code @item --help Print a help summary on standard output, then immediately exit @code{m4} without reading any input files or performing any other actions. @item --version Print the version number of the program on standard output, then immediately exit @code{m4} without reading any input files or performing any other actions. @item -E @itemx --fatal-warnings @cindex errors, fatal @cindex fatal errors Controls the effect of warnings. If unspecified, then execution continues and exit status is unaffected when a warning is printed. If specified exactly once, warnings become fatal; when one is issued, execution continues, but the exit status will be non-zero. If specified multiple times, then execution halts with non-zero status the first time a warning is issued. The introduction of behavior levels is new to M4 1.4.9; for behavior consistent with earlier versions, you should specify @option{-E} twice. @item -i @itemx --interactive @itemx -e Makes this invocation of @code{m4} interactive. This means that all output will be unbuffered, and interrupts will be ignored. The spelling @option{-e} exists for compatibility with other @code{m4} implementations, and issues a warning because it may be withdrawn in a future version of GNU M4. @item -P @itemx --prefix-builtins Internally modify @emph{all} builtin macro names so they all start with the prefix @samp{m4_}. For example, using this option, one should write @samp{m4_define} instead of @samp{define}, and @samp{m4___file__} instead of @samp{__file__}. This option has no effect if @option{-R} is also specified. @item -Q @itemx --quiet @itemx --silent Suppress warnings, such as missing or superfluous arguments in macro calls, or treating the empty string as zero. @item --warn-macro-sequence@r{[}=@var{regexp}@r{]} Issue a warning if the regular expression @var{regexp} has a non-empty match in any macro definition (either by @code{define} or @code{pushdef}). Empty matches are ignored; therefore, supplying the empty string as @var{regexp} disables any warning. If the optional @var{regexp} is not supplied, then the default regular expression is @samp{\$\(@{[^@}]*@}\|[0-9][0-9]+\)} (a literal @samp{$} followed by multiple digits or by an open brace), since these sequences will change semantics in the default operation of GNU M4 2.0 (due to a change in how more than 9 arguments in a macro definition will be handled, @pxref{Arguments}). Providing an alternate regular expression can provide a useful reverse lookup feature of finding where a macro is defined to have a given definition. @item -W @var{regexp} @itemx --word-regexp=@var{regexp} Use @var{regexp} as an alternative syntax for macro names. This experimental option will not be present in all GNU @code{m4} implementations (@pxref{Changeword}). @end table @node Preprocessor features @section Command line options for preprocessor features @cindex macro definitions, on the command line @cindex command line, macro definitions on the @cindex preprocessor features Several options allow @code{m4} to behave more like a preprocessor. Macro definitions and deletions can be made on the command line, the search path can be altered, and the output file can track where the input came from. These features occur with the following options: @table @code @item -D @var{name}@r{[}=@var{value}@r{]} @itemx --define=@var{name}@r{[}=@var{value}@r{]} This enters @var{name} into the symbol table. If @samp{=@var{value}} is missing, the value is taken to be the empty string. The @var{value} can be any string, and the macro can be defined to take arguments, just as if it was defined from within the input. This option may be given more than once; order with respect to file names is significant, and redefining the same @var{name} loses the previous value. @item -I @var{directory} @itemx --include=@var{directory} Make @code{m4} search @var{directory} for included files that are not found in the current working directory. @xref{Search Path}, for more details. This option may be given more than once. @item -s @itemx --synclines @cindex synchronization lines @cindex location, input @cindex input location Generate synchronization lines, for use by the C preprocessor or other similar tools. Order is significant with respect to file names. This option is useful, for example, when @code{m4} is used as a front end to a compiler. Source file name and line number information is conveyed by directives of the form @samp{#line @var{linenum} "@var{file}"}, which are inserted as needed into the middle of the output. Such directives mean that the following line originated or was expanded from the contents of input file @var{file} at line @var{linenum}. The @samp{"@var{file}"} part is often omitted when the file name did not change from the previous directive. Synchronization directives are always given on complete lines by themselves. When a synchronization discrepancy occurs in the middle of an output line, the associated synchronization directive is delayed until the next newline that does not occur in the middle of a quoted string or comment. @comment options: -s @example define(`twoline', `1 2') @result{}#line 2 "stdin" @result{} changecom(`/*', `*/') @result{} define(`comment', `/*1 2*/') @result{}#line 5 @result{} dnl no line hello @result{}#line 7 @result{}hello twoline @result{}1 @result{}#line 8 @result{}2 comment @result{}/*1 @result{}2*/ one comment `two three' @result{}#line 10 @result{}one /*1 @result{}2*/ two @result{}three goodbye @result{}#line 12 @result{}goodbye @end example @item -U @var{name} @itemx --undefine=@var{name} This deletes any predefined meaning @var{name} might have. Obviously, only predefined macros can be deleted in this way. This option may be given more than once; undefining a @var{name} that does not have a definition is silently ignored. Order is significant with respect to file names. @end table @node Limits control @section Command line options for limits control There are some limits within @code{m4} that can be tuned. For compatibility, @code{m4} also accepts some options that control limits in other implementations, but which are automatically unbounded (limited only by your hardware and operating system constraints) in GNU @code{m4}. @table @code @item -g @itemx --gnu Enable all the extensions in this implementation. In this release of M4, this option is always on by default; it is currently only useful when overriding a prior use of @option{--traditional}. However, having GNU behavior as default makes it impossible to write a strictly POSIX-compliant client that avoids all incompatible GNU M4 extensions, since such a client would have to use the non-POSIX command-line option to force full POSIX behavior. Thus, a future version of M4 will be changed to implicitly use the option @option{--traditional} if the environment variable @env{POSIXLY_CORRECT} is set. Projects that intentionally use GNU extensions should consider using @option{--gnu} to state their intentions, so that the project will not mysteriously break if the user upgrades to a newer M4 and has @env{POSIXLY_CORRECT} set in their environment. @item -G @itemx --traditional Suppress all the extensions made in this implementation, compared to the System V version. @xref{Compatibility}, for a list of these. @item -H @var{num} @itemx --hashsize=@var{num} Make the internal hash table for symbol lookup be @var{num} entries big. For better performance, the number should be prime, but this is not checked. The default is 509 entries. It should not be necessary to increase this value, unless you define an excessive number of macros. @item -L @var{num} @itemx --nesting-limit=@var{num} @cindex nesting limit @cindex limit, nesting Artificially limit the nesting of macro calls to @var{num} levels, stopping program execution if this limit is ever exceeded. When not specified, nesting defaults to unlimited on platforms that can detect stack overflow, and to 1024 levels otherwise. A value of zero means unlimited; but then heavily nested code could potentially cause a stack overflow. The precise effect of this option is more correctly associated with textual nesting than dynamic recursion. It has been useful when some complex @code{m4} input was generated by mechanical means, and also in diagnosing recursive algorithms that do not scale well. Most users never need to change this option from its default. @cindex rescanning This option does @emph{not} have the ability to break endless rescanning loops, since these do not necessarily consume much memory or stack space. Through clever usage of rescanning loops, one can request complex, time-consuming computations from @code{m4} with useful results. Putting limitations in this area would break @code{m4} power. There are many pathological cases: @w{@samp{define(`a', `a')a}} is only the simplest example (but @pxref{Compatibility}). Expecting GNU @code{m4} to detect these would be a little like expecting a compiler system to detect and diagnose endless loops: it is a quite @emph{hard} problem in general, if not undecidable! @item -B @var{num} @itemx -S @var{num} @itemx -T @var{num} These options are present for compatibility with System V @code{m4}, but do nothing in this implementation. They may disappear in future releases, and issue a warning to that effect. @item -N @var{num} @itemx --diversions=@var{num} These options are present only for compatibility with previous versions of GNU @code{m4}, and were controlling the number of possible diversions which could be used at the same time. They do nothing, because there is no fixed limit anymore. They may disappear in future releases, and issue a warning to that effect. @end table @node Frozen state @section Command line options for frozen state GNU @code{m4} comes with a feature of freezing internal state (@pxref{Frozen files}). This can be used to speed up @code{m4} execution when reusing a common initialization script. @table @code @item -F @var{file} @itemx --freeze-state=@var{file} Once execution is finished, write out the frozen state on the specified @var{file}. It is conventional, but not required, for @var{file} to end in @samp{.m4f}. @item -R @var{file} @itemx --reload-state=@var{file} Before execution starts, recover the internal state from the specified frozen @var{file}. The options @option{-D}, @option{-U}, and @option{-t} take effect after state is reloaded, but before the input files are read. @end table @node Debugging options @section Command line options for debugging Finally, there are several options for aiding in debugging @code{m4} scripts. @table @code @item -d@r{[}@var{flags}@r{]} @itemx --debug@r{[}=@var{flags}@r{]} Set the debug-level according to the flags @var{flags}. The debug-level controls the format and amount of information presented by the debugging functions. @xref{Debug Levels}, for more details on the format and meaning of @var{flags}. If omitted, @var{flags} defaults to @samp{aeq}. @item --debugfile@r{[}=@var{file}@r{]} @itemx -o @var{file} @itemx --error-output=@var{file} Redirect @code{dumpdef} output, debug messages, and trace output to the named @var{file}. Warnings, error messages, and @code{errprint} output are still printed to standard error. If these options are not used, or if @var{file} is unspecified (only possible for @option{--debugfile}), debug output goes to standard error; if @var{file} is the empty string, debug output is discarded. @xref{Debug Output}, for more details. The option @option{--debugfile} may be given more than once, and order is significant with respect to file names. The spellings @option{-o} and @option{--error-output} are misleading and inconsistent with other GNU tools; for now they are silently accepted as synonyms of @option{--debugfile} and only recognized once, but in a future version of M4, using them will cause a warning to be issued. @ignore @comment not worth including in the manual, but provides a good test @comment examples @comment options: -Dbar=hello -tbar --debugfile= foo --debugfile - @example $ @kbd{m4 -d -Iexamples -Dbar=hello -tbar --debugfile= foo --debugfile - @result{}hello errprint(`hi ')dnl @error{}hi bar @error{}m4trace: -1- bar -> `hello' @result{}hello @end example @end ignore @item -l @var{num} @itemx --arglength=@var{num} Restrict the size of the output generated by macro tracing to @var{num} characters per trace line. If unspecified or zero, output is unlimited. @xref{Debug Levels}, for more details. @item -t @var{name} @itemx --trace=@var{name} This enables tracing for the macro @var{name}, at any point where it is defined. @var{name} need not be defined when this option is given. This option may be given more than once, and order is significant with respect to file names. @xref{Trace}, for more details. @end table @node Command line files @section Specifying input files on the command line @cindex command line, file names on the @cindex file names, on the command line The remaining arguments on the command line are taken to be input file names. If no names are present, standard input is read. A file name of @file{-} is taken to mean standard input. It is conventional, but not required, for input files to end in @samp{.m4}. The input files are read in the sequence given. Standard input can be read more than once, so the file name @file{-} may appear multiple times on the command line; this makes a difference when input is from a terminal or other special file type. It is an error if an input file ends in the middle of argument collection, a comment, or a quoted string. The options @option{--define} (@option{-D}), @option{--undefine} (@option{-U}), @option{--synclines} (@option{-s}), and @option{--trace} (@option{-t}) only take effect after processing input from any file names that occur earlier on the command line. For example, assume the file @file{foo} contains: @comment ignore @example $ @kbd{cat foo} bar @end example The text @samp{bar} can then be redefined over multiple uses of @file{foo}: @comment options: -Dbar=hello foo -Dbar=world foo @example $ @kbd{m4 -Dbar=hello foo -Dbar=world foo} @result{}hello @result{}world @end example If none of the input files invoked @code{m4exit} (@pxref{M4exit}), the exit status of @code{m4} will be 0 for success, 1 for general failure (such as problems with reading an input file), and 63 for version mismatch (@pxref{Using frozen files}). If you need to read a file whose name starts with a @file{-}, you can specify it as @samp{./-file}, or use @option{--} to mark the end of options. @ignore @comment Test that 'm4 file/' detects that file is not a directory; we @comment can assume that the current directory contains a Makefile. @comment mingw fails with EINVAL rather than ENOTDIR. @comment status: 1 @comment xerr: ignore @comment options: Makefile/ @example @error{}m4: cannot open `Makefile/': Not a directory @end example @comment Test that closed stderr does not cause a crash. Not all @comment systems have the same message for EBADF. @comment xerr: ignore @example ifdef(`__unix__', , `errprint(` skipping: syscmd does not have unix semantics ')m4exit(`77')')dnl changequote(`[', `]')dnl syscmd([echo | ']__program__[' >&-])dnl @error{}m4: write error: Bad file descriptor sysval @result{}1 @end example @example ifdef(`__unix__', , `errprint(` skipping: syscmd does not have unix semantics ')m4exit(`77')')dnl changequote(`[', `]')dnl syscmd([echo 'esyscmd(echo hi >&2 && echo err"print(bye )d"nl)dnl' > tmp.m4 \ && ']__program__[' tmp.m4 <&- >&- \ && rm tmp.m4])sysval @error{}hi @error{}bye @result{}0 @end example @comment Test that we obey POSIX semantics with -D interspersed with @comment files, even with POSIXLY_CORRECT (BSD getopt gets it wrong). $ @kbd{m4 } @example ifdef(`__unix__', , `errprint(` skipping: syscmd does not have unix semantics ')m4exit(`77')')dnl changequote(`[', `]')dnl syscmd([POSIXLY_CORRECT=1 ']__program__[' -Dbar=hello foo -Dbar=world foo])dnl @result{}hello @result{}world sysval @result{}0 @end example @end ignore @node Syntax @chapter Lexical and syntactic conventions @cindex input tokens @cindex tokens As @code{m4} reads its input, it separates it into @dfn{tokens}. A token is either a name, a quoted string, or any single character, that is not a part of either a name or a string. Input to @code{m4} can also contain comments. GNU @code{m4} does not yet understand multibyte locales; all operations are byte-oriented rather than character-oriented (although if your locale uses a single byte encoding, such as @sc{ISO-8859-1}, you will not notice a difference). However, @code{m4} is eight-bit clean, so you can use non-@sc{ascii} characters in quoted strings (@pxref{Changequote}), comments (@pxref{Changecom}), and macro names (@pxref{Indir}), with the exception of the @sc{nul} character (the zero byte @samp{'\0'}). @menu * Names:: Macro names * Quoted strings:: Quoting input to @code{m4} * Comments:: Comments in @code{m4} input * Other tokens:: Other kinds of input tokens * Input processing:: How @code{m4} copies input to output @end menu @node Names @section Macro names @cindex names @cindex words A name is any sequence of letters, digits, and the character @samp{_} (underscore), where the first character is not a digit. @code{m4} will use the longest such sequence found in the input. If a name has a macro definition, it will be subject to macro expansion (@pxref{Macros}). Names are case-sensitive. Examples of legal names are: @samp{foo}, @samp{_tmp}, and @samp{name01}. @node Quoted strings @section Quoting input to @code{m4} @cindex quoted string @cindex string, quoted A quoted string is a sequence of characters surrounded by quote strings, defaulting to @samp{`} and @samp{'}, where the nested begin and end quotes within the string are balanced. The value of a string token is the text, with one level of quotes stripped off. Thus @comment ignore @example `' @result{} @end example @noindent is the empty string, and double-quoting turns into single-quoting. @comment ignore @example ``quoted'' @result{}`quoted' @end example The quote characters can be changed at any time, using the builtin macro @code{changequote}. @xref{Changequote}, for more information. @node Comments @section Comments in @code{m4} input @cindex comments Comments in @code{m4} are normally delimited by the characters @samp{#} and newline. All characters between the comment delimiters are ignored, but the entire comment (including the delimiters) is passed through to the output---comments are @emph{not} discarded by @code{m4}. Comments cannot be nested, so the first newline after a @samp{#} ends the comment. The commenting effect of the begin-comment string can be inhibited by quoting it. @example $ @kbd{m4} `quoted text' # `commented text' @result{}quoted text # `commented text' `quoting inhibits' `#' `comments' @result{}quoting inhibits # comments @end example The comment delimiters can be changed to any string at any time, using the builtin macro @code{changecom}. @xref{Changecom}, for more information. @ignore @comment Detect regression in 1.4.10b in regards to reparsing comments. @comment Not worth including in the manual. @example define(`e', `$@@')define(`q', ``$@@'')define(`foo', `bar') @result{} q(e(`one ',#two ' foo )) @result{}`one @result{}',`#two bar @result{}'' changecom(`<', `>')define(`n', `$#') @result{} n(e(<`>, <'>)) @result{}1 len(e(<`>, ,<'>)) @result{}12 @end example @end ignore @node Other tokens @section Other kinds of input tokens @cindex tokens, special Any character, that is neither a part of a name, nor of a quoted string, nor a comment, is a token by itself. When not in the context of macro expansion, all of these tokens are just copied to output. However, during macro expansion, whitespace characters (space, tab, newline, formfeed, carriage return, vertical tab), parentheses (@samp{(} and @samp{)}), comma (@samp{,}), and dollar (@samp{$}) have additional roles, explained later. @node Input processing @section How @code{m4} copies input to output As @code{m4} reads the input token by token, it will copy each token directly to the output immediately. The exception is when it finds a word with a macro definition. In that case @code{m4} will calculate the macro's expansion, possibly reading more input to get the arguments. It then inserts the expansion in front of the remaining input. In other words, the resulting text from a macro call will be read and parsed into tokens again. @code{m4} expands a macro as soon as possible. If it finds a macro call when collecting the arguments to another, it will expand the second call first. This process continues until there are no more macro calls to expand and all the input has been consumed. For a running example, examine how @code{m4} handles this input: @comment ignore @example format(`Result is %d', eval(`2**15')) @end example @noindent First, @code{m4} sees that the token @samp{format} is a macro name, so it collects the tokens @samp{(}, @samp{`Result is %d'}, @samp{,}, and @samp{@w{ }}, before encountering another potential macro. Sure enough, @samp{eval} is a macro name, so the nested argument collection picks up @samp{(}, @samp{`2**15'}, and @samp{)}, invoking the eval macro with the lone argument of @samp{2**15}. The expansion of @samp{eval(2**15)} is @samp{32768}, which is then rescanned as the five tokens @samp{3}, @samp{2}, @samp{7}, @samp{6}, and @samp{8}; and combined with the next @samp{)}, the format macro now has all its arguments, as if the user had typed: @comment ignore @example format(`Result is %d', 32768) @end example @noindent The format macro expands to @samp{Result is 32768}, and we have another round of scanning for the tokens @samp{Result}, @samp{@w{ }}, @samp{is}, @samp{@w{ }}, @samp{3}, @samp{2}, @samp{7}, @samp{6}, and @samp{8}. None of these are macros, so the final output is @comment ignore @example @result{}Result is 32768 @end example As a more complicated example, we will contrast an actual code example from the Gnulib project@footnote{Derived from a patch in @uref{http://lists.gnu.org/archive/html/bug-gnulib/@/2007-01/@/msg00389.html}, and a followup patch in @uref{http://lists.gnu.org/archive/html/bug-gnulib/@/2007-02/@/msg00000.html}}, showing both a buggy approach and the desired results. The user desires to output a shell assignment statement that takes its argument and turns it into a shell variable by converting it to uppercase and prepending a prefix. The original attempt looks like this: @example changequote([,])dnl define([gl_STRING_MODULE_INDICATOR], [ dnl comment GNULIB_]translit([$1],[a-z],[A-Z])[=1 ])dnl gl_STRING_MODULE_INDICATOR([strcase]) @result{} @w{ } @result{} GNULIB_strcase=1 @result{} @w{ } @end example Oops -- the argument did not get capitalized. And although the manual is not able to easily show it, both lines that appear empty actually contain two trailing spaces. By stepping through the parse, it is easy to see what happened. First, @code{m4} sees the token @samp{changequote}, which it recognizes as a macro, followed by @samp{(}, @samp{[}, @samp{,}, @samp{]}, and @samp{)} to form the argument list. The macro expands to the empty string, but changes the quoting characters to something more useful for generating shell code (unbalanced @samp{`} and @samp{'} appear all the time in shell scripts, but unbalanced @samp{[]} tend to be rare). Also in the first line, @code{m4} sees the token @samp{dnl}, which it recognizes as a builtin macro that consumes the rest of the line, resulting in no output for that line. The second line starts a macro definition. @code{m4} sees the token @samp{define}, which it recognizes as a macro, followed by a @samp{(}, @samp{[gl_STRING_MODULE_INDICATOR]}, and @samp{,}. Because an unquoted comma was encountered, the first argument is known to be the expansion of the single-quoted string token, or @samp{gl_STRING_MODULE_INDICATOR}. Next, @code{m4} sees @samp{@key{NL}}, @samp{ }, and @samp{ }, but this whitespace is discarded as part of argument collection. Then comes a rather lengthy single-quoted string token, @samp{[@key{NL}@ @ @ @ dnl comment@key{NL}@ @ @ @ GNULIB_]}. This is followed by the token @samp{translit}, which @code{m4} recognizes as a macro name, so a nested macro expansion has started. The arguments to the @code{translit} are found by the tokens @samp{(}, @samp{[$1]}, @samp{,}, @samp{[a-z]}, @samp{,}, @samp{[A-Z]}, and finally @samp{)}. All three string arguments are expanded (or in other words, the quotes are stripped), and since neither @samp{$} nor @samp{1} need capitalization, the result of the macro is @samp{$1}. This expansion is rescanned, resulting in the two literal characters @samp{$} and @samp{1}. Scanning of the outer macro resumes, and picks up with @samp{[=1@key{NL}@ @ ]}, and finally @samp{)}. The collected pieces of expanded text are concatenated, with the end result that the macro @samp{gl_STRING_MODULE_INDICATOR} is now defined to be the sequence @samp{@key{NL}@ @ @ @ dnl comment@key{NL}@ @ @ @ GNULIB_$1=1@key{NL}@ @ }. Once again, @samp{dnl} is recognized and avoids a newline in the output. The final line is then parsed, beginning with @samp{ } and @samp{ } that are output literally. Then @samp{gl_STRING_MODULE_INDICATOR} is recognized as a macro name, with an argument list of @samp{(}, @samp{[strcase]}, and @samp{)}. Since the definition of the macro contains the sequence @samp{$1}, that sequence is replaced with the argument @samp{strcase} prior to starting the rescan. The rescan sees @samp{@key{NL}} and four spaces, which are output literally, then @samp{dnl}, which discards the text @samp{ comment@key{NL}}. Next comes four more spaces, also output literally, and the token @samp{GNULIB_strcase}, which resulted from the earlier parameter substitution. Since that is not a macro name, it is output literally, followed by the literal tokens @samp{=}, @samp{1}, @samp{@key{NL}}, and two more spaces. Finally, the original @samp{@key{NL}} seen after the macro invocation is scanned and output literally. Now for a corrected approach. This rearranges the use of newlines and whitespace so that less whitespace is output (which, although harmless to shell scripts, can be visually unappealing), and fixes the quoting issues so that the capitalization occurs when the macro @samp{gl_STRING_MODULE_INDICATOR} is invoked, rather then when it is defined. It also adds another layer of quoting to the first argument of @code{translit}, to ensure that the output will be rescanned as a string rather than a potential uppercase macro name needing further expansion. @example changequote([,])dnl define([gl_STRING_MODULE_INDICATOR], [dnl comment GNULIB_[]translit([[$1]], [a-z], [A-Z])=1dnl ])dnl gl_STRING_MODULE_INDICATOR([strcase]) @result{} GNULIB_STRCASE=1 @end example The parsing of the first line is unchanged. The second line sees the name of the macro to define, then sees the discarded @samp{@key{NL}} and two spaces, as before. But this time, the next token is @samp{[dnl comment@key{NL}@ @ GNULIB_[]translit([[$1]], [a-z], [A-Z])=1dnl@key{NL}]}, which includes nested quotes, followed by @samp{)} to end the macro definition and @samp{dnl} to skip the newline. No early expansion of @code{translit} occurs, so the entire string becomes the definition of the macro. The final line is then parsed, beginning with two spaces that are output literally, and an invocation of @code{gl_STRING_MODULE_INDICATOR} with the argument @samp{strcase}. Again, the @samp{$1} in the macro definition is substituted prior to rescanning. Rescanning first encounters @samp{dnl}, and discards @samp{ comment@key{NL}}. Then two spaces are output literally. Next comes the token @samp{GNULIB_}, but that is not a macro, so it is output literally. The token @samp{[]} is an empty string, so it does not affect output. Then the token @samp{translit} is encountered. This time, the arguments to @code{translit} are parsed as @samp{(}, @samp{[[strcase]]}, @samp{,}, @samp{ }, @samp{[a-z]}, @samp{,}, @samp{ }, @samp{[A-Z]}, and @samp{)}. The two spaces are discarded, and the translit results in the desired result @samp{[STRCASE]}. This is rescanned, but since it is a string, the quotes are stripped and the only output is a literal @samp{STRCASE}. Then the scanner sees @samp{=} and @samp{1}, which are output literally, followed by @samp{dnl} which discards the rest of the definition of @code{gl_STRING_MODULE_INDICATOR}. The newline at the end of output is the literal @samp{@key{NL}} that appeared after the invocation of the macro. The order in which @code{m4} expands the macros can be further explored using the trace facilities of GNU @code{m4} (@pxref{Trace}). @node Macros @chapter How to invoke macros This chapter covers macro invocation, macro arguments and how macro expansion is treated. @menu * Invocation:: Macro invocation * Inhibiting Invocation:: Preventing macro invocation * Macro Arguments:: Macro arguments * Quoting Arguments:: On Quoting Arguments to macros * Macro expansion:: Expanding macros @end menu @node Invocation @section Macro invocation @cindex macro invocation @cindex invoking macros Macro invocations has one of the forms @comment ignore @example name @end example @noindent which is a macro invocation without any arguments, or @comment ignore @example name(arg1, arg2, @dots{}, arg@var{n}) @end example @noindent which is a macro invocation with @var{n} arguments. Macros can have any number of arguments. All arguments are strings, but different macros might interpret the arguments in different ways. The opening parenthesis @emph{must} follow the @var{name} directly, with no spaces in between. If it does not, the macro is called with no arguments at all. For a macro call to have no arguments, the parentheses @emph{must} be left out. The macro call @comment ignore @example name() @end example @noindent is a macro call with one argument, which is the empty string, not a call with no arguments. @node Inhibiting Invocation @section Preventing macro invocation An innovation of the @code{m4} language, compared to some of its predecessors (like Strachey's @code{GPM}, for example), is the ability to recognize macro calls without resorting to any special, prefixed invocation character. While generally useful, this feature might sometimes be the source of spurious, unwanted macro calls. So, GNU @code{m4} offers several mechanisms or techniques for inhibiting the recognition of names as macro calls. @cindex GNU extensions @cindex blind macro @cindex macro, blind First of all, many builtin macros cannot meaningfully be called without arguments. As a GNU extension, for any of these macros, whenever an opening parenthesis does not immediately follow their name, the builtin macro call is not triggered. This solves the most usual cases, like for @samp{include} or @samp{eval}. Later in this document, the sentence ``This macro is recognized only with parameters'' refers to this specific provision of GNU M4, also known as a blind builtin macro. For the builtins defined by POSIX that bear this disclaimer, POSIX specifically states that invoking those builtins without arguments is unspecified, because many other implementations simply invoke the builtin as though it were given one empty argument instead. @example $ @kbd{m4} eval @result{}eval eval(`1') @result{}1 @end example There is also a command line option (@option{--prefix-builtins}, or @option{-P}, @pxref{Operation modes, , Invoking m4}) that renames all builtin macros with a prefix of @samp{m4_} at startup. The option has no effect whatsoever on user defined macros. For example, with this option, one has to write @code{m4_dnl} and even @code{m4_m4exit}. It also has no effect on whether a macro requires parameters. @comment options: -P @example $ @kbd{m4 -P} eval @result{}eval eval(`1') @result{}eval(1) m4_eval @result{}m4_eval m4_eval(`1') @result{}1 @end example Another alternative is to redefine problematic macros to a name less likely to cause conflicts, @xref{Definitions}. If your version of GNU @code{m4} has the @code{changeword} feature compiled in, it offers far more flexibility in specifying the syntax of macro names, both builtin or user-defined. @xref{Changeword}, for more information on this experimental feature. Of course, the simplest way to prevent a name from being interpreted as a call to an existing macro is to quote it. The remainder of this section studies a little more deeply how quoting affects macro invocation, and how quoting can be used to inhibit macro invocation. Even if quoting is usually done over the whole macro name, it can also be done over only a few characters of this name (provided, of course, that the unquoted portions are not also a macro). It is also possible to quote the empty string, but this works only @emph{inside} the name. For example: @example `divert' @result{}divert `d'ivert @result{}divert di`ver't @result{}divert div`'ert @result{}divert @end example @noindent all yield the string @samp{divert}. While in both: @example `'divert @result{} divert`' @result{} @end example @noindent the @code{divert} builtin macro will be called, which expands to the empty string. @cindex rescanning The output of macro evaluations is always rescanned. In the following example, the input @samp{x`'y} yields the string @samp{bCD}, exactly as if @code{m4} has been given @w{@samp{substr(ab`'cde, `1', `3')}} as input: @example define(`cde', `CDE') @result{} define(`x', `substr(ab') @result{} define(`y', `cde, `1', `3')') @result{} x`'y @result{}bCD @end example @ignore @comment Similar, but with argument references, to ensure good test @comment coverage. @example define(`x1', `len(`$1'') @result{} define(`y1', ``$1')') @result{} x1(`01234567890123456789')y1(`98765432109876543210') @result{}40 @end example @end ignore Unquoted strings on either side of a quoted string are subject to being recognized as macro names. In the following example, quoting the empty string allows for the second @code{macro} to be recognized as such: @example define(`macro', `m') @result{} macro(`m')macro @result{}mmacro macro(`m')`'macro @result{}mm @end example Quoting may prevent recognizing as a macro name the concatenation of a macro expansion with the surrounding characters. In this example: @example define(`macro', `di$1') @result{} macro(`v')`ert' @result{}divert macro(`v')ert @result{} @end example @noindent the input will produce the string @samp{divert}. When the quotes were removed, the @code{divert} builtin was called instead. @node Macro Arguments @section Macro arguments @cindex macros, arguments to @cindex arguments to macros When a name is seen, and it has a macro definition, it will be expanded as a macro. If the name is followed by an opening parenthesis, the arguments will be collected before the macro is called. If too few arguments are supplied, the missing arguments are taken to be the empty string. However, some builtins are documented to behave differently for a missing optional argument than for an explicit empty string. If there are too many arguments, the excess arguments are ignored. Unquoted leading whitespace is stripped off all arguments, but whitespace generated by a macro expansion or occurring after a macro that expanded to an empty string remains intact. Whitespace includes space, tab, newline, carriage return, vertical tab, and formfeed. @example define(`macro', `$1') @result{} macro( unquoted leading space lost) @result{}unquoted leading space lost macro(` quoted leading space kept') @result{} quoted leading space kept macro( divert `unquoted space kept after expansion') @result{} unquoted space kept after expansion macro(macro(` ')`whitespace from expansion kept') @result{} @result{}whitespace from expansion kept macro(`unquoted trailing whitespace kept' ) @result{}unquoted trailing whitespace kept @result{} @end example @cindex warnings, suppressing @cindex suppressing warnings Normally @code{m4} will issue warnings if a builtin macro is called with an inappropriate number of arguments, but it can be suppressed with the @option{--quiet} command line option (or @option{--silent}, or @option{-Q}, @pxref{Operation modes, , Invoking m4}). For user defined macros, there is no check of the number of arguments given. @example $ @kbd{m4} index(`abc') @error{}m4:stdin:1: Warning: too few arguments to builtin `index' @result{}0 index(`abc',) @result{}0 index(`abc', `b', `ignored') @error{}m4:stdin:3: Warning: excess arguments to builtin `index' ignored @result{}1 @end example @comment options: -Q @example $ @kbd{m4 -Q} index(`abc') @result{}0 index(`abc',) @result{}0 index(`abc', `b', `ignored') @result{}1 @end example Macros are expanded normally during argument collection, and whatever commas, quotes and parentheses that might show up in the resulting expanded text will serve to define the arguments as well. Thus, if @var{foo} expands to @samp{, b, c}, the macro call @comment ignore @example bar(a foo, d) @end example @noindent is a macro call with four arguments, which are @samp{a }, @samp{b}, @samp{c} and @samp{d}. To understand why the first argument contains whitespace, remember that unquoted leading whitespace is never part of an argument, but trailing whitespace always is. It is possible for a macro's definition to change during argument collection, in which case the expansion uses the definition that was in effect at the time the opening @samp{(} was seen. @example define(`f', `1') @result{} f(define(`f', `2')) @result{}1 f @result{}2 @end example It is an error if the end of file occurs while collecting arguments. @comment status: 1 @example hello world @result{}hello world define( ^D @error{}m4:stdin:2: ERROR: end of file in argument list @end example @node Quoting Arguments @section On Quoting Arguments to macros @cindex quoted macro arguments @cindex macros, quoted arguments to @cindex arguments, quoted macro Each argument has unquoted leading whitespace removed. Within each argument, all unquoted parentheses must match. For example, if @var{foo} is a macro, @comment ignore @example foo(() (`(') `(') @end example @noindent is a macro call, with one argument, whose value is @samp{() (() (}. Commas separate arguments, except when they occur inside quotes, comments, or unquoted parentheses. @xref{Pseudo Arguments}, for examples. It is common practice to quote all arguments to macros, unless you are sure you want the arguments expanded. Thus, in the above example with the parentheses, the `right' way to do it is like this: @comment ignore @example foo(`() (() (') @end example @cindex quoting rule of thumb @cindex rule of thumb, quoting It is, however, in certain cases necessary (because nested expansion must occur to create the arguments for the outer macro) or convenient (because it uses fewer characters) to leave out quotes for some arguments, and there is nothing wrong in doing it. It just makes life a bit harder, if you are not careful to follow a consistent quoting style. For consistency, this manual follows the rule of thumb that each layer of parentheses introduces another layer of single quoting, except when showing the consequences of quoting rules. This is done even when the quoted string cannot be a macro, such as with integers when you have not changed the syntax via @code{changeword} (@pxref{Changeword}). The quoting rule of thumb of one level of quoting per parentheses has a nice property: when a macro name appears inside parentheses, you can determine when it will be expanded. If it is not quoted, it will be expanded prior to the outer macro, so that its expansion becomes the argument. If it is single-quoted, it will be expanded after the outer macro. And if it is double-quoted, it will be used as literal text instead of a macro name. @example define(`active', `ACT, IVE') @result{} define(`show', `$1 $1') @result{} show(active) @result{}ACT ACT show(`active') @result{}ACT, IVE ACT, IVE show(``active'') @result{}active active @end example @node Macro expansion @section Macro expansion @cindex macros, expansion of @cindex expansion of macros When the arguments, if any, to a macro call have been collected, the macro is expanded, and the expansion text is pushed back onto the input (unquoted), and reread. The expansion text from one macro call might therefore result in more macros being called, if the calls are included, completely or partially, in the first macro calls' expansion. Taking a very simple example, if @var{foo} expands to @samp{bar}, and @var{bar} expands to @samp{Hello}, the input @comment options: -Dbar=Hello -Dfoo=bar @example $ @kbd{m4 -Dbar=Hello -Dfoo=bar} foo @result{}Hello @end example @noindent will expand first to @samp{bar}, and when this is reread and expanded, into @samp{Hello}. @ignore @comment not worth documenting, but test that the command line can @comment define macros that take parameters @comment options: -Dfoo -Decho=$@ @example $ @kbd{m4 -Dfoo -Decho='$@'} foo @result{} foo(`silently ignored') @result{} echo(`1', `2') @result{}1,2 @end example @end ignore @node Definitions @chapter How to define new macros @cindex macros, how to define new @cindex defining new macros Macros can be defined, redefined and deleted in several different ways. Also, it is possible to redefine a macro without losing a previous value, and bring back the original value at a later time. @menu * Define:: Defining a new macro * Arguments:: Arguments to macros * Pseudo Arguments:: Special arguments to macros * Undefine:: Deleting a macro * Defn:: Renaming macros * Pushdef:: Temporarily redefining macros * Indir:: Indirect call of macros * Builtin:: Indirect call of builtins @end menu @node Define @section Defining a macro The normal way to define or redefine macros is to use the builtin @code{define}: @deffn Builtin define (@var{name}, @ovar{expansion}) Defines @var{name} to expand to @var{expansion}. If @var{expansion} is not given, it is taken to be empty. The expansion of @code{define} is void. The macro @code{define} is recognized only with parameters. @end deffn The following example defines the macro @var{foo} to expand to the text @samp{Hello World.}. @example define(`foo', `Hello world.') @result{} foo @result{}Hello world. @end example The empty line in the output is there because the newline is not a part of the macro definition, and it is consequently copied to the output. This can be avoided by use of the macro @code{dnl}. @xref{Dnl}, for details. The first argument to @code{define} should be quoted; otherwise, if the macro is already defined, you will be defining a different macro. This example shows the problems with underquoting, since we did not want to redefine @code{one}: @example define(foo, one) @result{} define(foo, two) @result{} one @result{}two @end example @cindex GNU extensions GNU @code{m4} normally replaces only the @emph{topmost} definition of a macro if it has several definitions from @code{pushdef} (@pxref{Pushdef}). Some other implementations of @code{m4} replace all definitions of a macro with @code{define}. @xref{Incompatibilities}, for more details. As a GNU extension, the first argument to @code{define} does not have to be a simple word. It can be any text string, even the empty string. A macro with a non-standard name cannot be invoked in the normal way, as the name is not recognized. It can only be referenced by the builtins @code{indir} (@pxref{Indir}) and @code{defn} (@pxref{Defn}). @cindex arrays Arrays and associative arrays can be simulated by using non-standard macro names. @deffn Composite array (@var{index}) @deffnx Composite array_set (@var{index}, @ovar{value}) Provide access to entries within an array. @code{array} reads the entry at location @var{index}, and @code{array_set} assigns @var{value} to location @var{index}. @end deffn @example define(`array', `defn(format(``array[%d]'', `$1'))') @result{} define(`array_set', `define(format(``array[%d]'', `$1'), `$2')') @result{} array_set(`4', `array element no. 4') @result{} array_set(`17', `array element no. 17') @result{} array(`4') @result{}array element no. 4 array(eval(`10 + 7')) @result{}array element no. 17 @end example Change the @samp{%d} to @samp{%s} and it is an associative array. @node Arguments @section Arguments to macros @cindex macros, arguments to @cindex arguments to macros Macros can have arguments. The @var{n}th argument is denoted by @code{$n} in the expansion text, and is replaced by the @var{n}th actual argument, when the macro is expanded. Replacement of arguments happens before rescanning, regardless of how many nesting levels of quoting appear in the expansion. Here is an example of a macro with two arguments. @deffn Composite exch (@var{arg1}, @var{arg2}) Expands to @var{arg2} followed by @var{arg1}, effectively exchanging their order. @end deffn @example define(`exch', `$2, $1') @result{} exch(`arg1', `arg2') @result{}arg2, arg1 @end example This can be used, for example, if you like the arguments to @code{define} to be reversed. @example define(`exch', `$2, $1') @result{} define(exch(``expansion text'', ``macro'')) @result{} macro @result{}expansion text @end example @xref{Quoting Arguments}, for an explanation of the double quotes. (You should try and improve this example so that clients of @code{exch} do not have to double quote; or @pxref{Improved exch, , Answers}). As a special case, the zeroth argument, @code{$0}, is always the name of the macro being expanded. @example define(`test', ``Macro name: $0'') @result{} test @result{}Macro name: test @end example If you want quoted text to appear as part of the expansion text, remember that quotes can be nested in quoted strings. Thus, in @example define(`foo', `This is macro `foo'.') @result{} foo @result{}This is macro foo. @end example @noindent The @samp{foo} in the expansion text is @emph{not} expanded, since it is a quoted string, and not a name. @cindex GNU extensions @cindex nine arguments, more than @cindex more than nine arguments @cindex arguments, more than nine @cindex positional parameters, more than nine GNU @code{m4} allows the number following the @samp{$} to consist of one or more digits, allowing macros to have any number of arguments. The extension of accepting multiple digits is incompatible with POSIX, and is different than traditional implementations of @code{m4}, which only recognize one digit. Therefore, future versions of GNU M4 will phase out this feature. To portably access beyond the ninth argument, you can use the @code{argn} macro documented later (@pxref{Shift}). POSIX also states that @samp{$} followed immediately by @samp{@{} in a macro definition is implementation-defined. This version of M4 passes the literal characters @samp{$@{} through unchanged, but M4 2.0 will implement an optional feature similar to @command{sh}, where @samp{$@{11@}} expands to the eleventh argument, to replace the current recognition of @samp{$11}. Meanwhile, if you want to guarantee that you will get a literal @samp{$@{} in output when expanding a macro, even when you upgrade to M4 2.0, you can use nested quoting to your advantage: @example define(`foo', `single quoted $`'@{1@} output') @result{} define(`bar', ``double quoted $'`@{2@} output'') @result{} foo(`a', `b') @result{}single quoted $@{1@} output bar(`a', `b') @result{}double quoted $@{2@} output @end example To help you detect places in your M4 input files that might change in behavior due to the changed behavior of M4 2.0, you can use the @option{--warn-macro-sequence} command-line option (@pxref{Operation modes, , Invoking m4}) with the default regular expression. This will add a warning any time a macro definition includes @samp{$} followed by multiple digits, or by @samp{@{}. The warning is not enabled by default, because it triggers a number of warnings in Autoconf 2.61 (and Autoconf uses @option{-E} to treat warnings as errors), and because it will still be possible to restore older behavior in M4 2.0. @comment options: --warn-macro-sequence @example $ @kbd{m4 --warn-macro-sequence} define(`foo', `$001 $@{1@} $1') @error{}m4:stdin:1: Warning: definition of `foo' contains sequence `$001' @error{}m4:stdin:1: Warning: definition of `foo' contains sequence `$@{1@}' @result{} foo(`bar') @result{}bar $@{1@} bar @end example @node Pseudo Arguments @section Special arguments to macros @cindex special arguments to macros @cindex macros, special arguments to @cindex arguments to macros, special There is a special notation for the number of actual arguments supplied, and for all the actual arguments. The number of actual arguments in a macro call is denoted by @code{$#} in the expansion text. @deffn Composite nargs (@dots{}) Expands to a count of the number of arguments supplied. @end deffn @example define(`nargs', `$#') @result{} nargs @result{}0 nargs() @result{}1 nargs(`arg1', `arg2', `arg3') @result{}3 nargs(`commas can be quoted, like this') @result{}1 nargs(arg1#inside comments, commas do not separate arguments still arg1) @result{}1 nargs((unquoted parentheses, like this, group arguments)) @result{}1 @end example Remember that @samp{#} defaults to the comment character; if you forget quotes to inhibit the comment behavior, your macro definition may not end where you expected. @example dnl Attempt to define a macro to just `$#' define(underquoted, $#) oops) @result{} underquoted @result{}0) @result{}oops @end example The notation @code{$*} can be used in the expansion text to denote all the actual arguments, unquoted, with commas in between. For example @example define(`echo', `$*') @result{} echo(arg1, arg2, arg3 , arg4) @result{}arg1,arg2,arg3 ,arg4 @end example Often each argument should be quoted, and the notation @code{$@@} handles that. It is just like @code{$*}, except that it quotes each argument. A simple example of that is: @example define(`echo', `$@@') @result{} echo(arg1, arg2, arg3 , arg4) @result{}arg1,arg2,arg3 ,arg4 @end example Where did the quotes go? Of course, they were eaten, when the expanded text were reread by @code{m4}. To show the difference, try @example define(`echo1', `$*') @result{} define(`echo2', `$@@') @result{} define(`foo', `This is macro `foo'.') @result{} echo1(foo) @result{}This is macro This is macro foo.. echo1(`foo') @result{}This is macro foo. echo2(foo) @result{}This is macro foo. echo2(`foo') @result{}foo @end example @noindent @xref{Trace}, if you do not understand this. As another example of the difference, remember that comments encountered in arguments are passed untouched to the macro, and that quoting disables comments. @example define(`echo1', `$*') @result{} define(`echo2', `$@@') @result{} define(`foo', `bar') @result{} echo1(#foo'foo foo) @result{}#foo'foo @result{}bar echo2(#foo'foo foo) @result{}#foobar @result{}bar' @end example @ignore @comment Not worth putting in the manual, but this example is needed for @comment good test coverage of copying large strings across recursion @comment levels. @example define(`echo', `$@@')dnl echo(echo(`01234567890123456789', `01234567890123456789') echo(`98765432109876543210', `98765432109876543210')) @result{}01234567890123456789,01234567890123456789 @result{}98765432109876543210,98765432109876543210 len((echo(`01234567890123456789', `01234567890123456789')echo(`98765432109876543210', `98765432109876543210'))) @result{}84 indir(`echo', indir(`echo', `01234567890123456789', `01234567890123456789') indir(`echo', `98765432109876543210', `98765432109876543210')) @result{}01234567890123456789,01234567890123456789 @result{}98765432109876543210,98765432109876543210 define(`argn', `$#')dnl define(`echo1', `-$@@-')define(`echo2', `,$@@,')dnl echo1(`1', `2', `3') argn(echo1(`1', `2', `3')) @result{}-1,2,3- 3 echo2(`1', `2', `3') argn(echo2(`1', `2', `3')) @result{},1,2,3, 5 @end example @end ignore A @samp{$} sign in the expansion text, that is not followed by anything @code{m4} understands, is simply copied to the macro expansion, as any other text is. @example define(`foo', `$$$ hello $$$') @result{} foo @result{}$$$ hello $$$ @end example @cindex rescanning @cindex literal output @cindex output, literal If you want a macro to expand to something like @samp{$12}, the judicious use of nested quoting can put a safe character between the @code{$} and the next character, relying on the rescanning to remove the nested quote. This will prevent @code{m4} from interpreting the @code{$} sign as a reference to an argument. @example define(`foo', `no nested quote: $1') @result{} foo(`arg') @result{}no nested quote: arg define(`foo', `nested quote around $: `$'1') @result{} foo(`arg') @result{}nested quote around $: $1 define(`foo', `nested empty quote after $: $`'1') @result{} foo(`arg') @result{}nested empty quote after $: $1 define(`foo', `nested quote around next character: $`1'') @result{} foo(`arg') @result{}nested quote around next character: $1 define(`foo', `nested quote around both: `$1'') @result{} foo(`arg') @result{}nested quote around both: arg @end example @node Undefine @section Deleting a macro @cindex macros, how to delete @cindex deleting macros @cindex undefining macros A macro definition can be removed with @code{undefine}: @deffn Builtin undefine (@var{name}@dots{}) For each argument, remove the macro @var{name}. The macro names must necessarily be quoted, since they will be expanded otherwise. The expansion of @code{undefine} is void. The macro @code{undefine} is recognized only with parameters. @end deffn @example foo bar blah @result{}foo bar blah define(`foo', `some')define(`bar', `other')define(`blah', `text') @result{} foo bar blah @result{}some other text undefine(`foo') @result{} foo bar blah @result{}foo other text undefine(`bar', `blah') @result{} foo bar blah @result{}foo bar blah @end example Undefining a macro inside that macro's expansion is safe; the macro still expands to the definition that was in effect at the @samp{(}. @example define(`f', ``$0':$1') @result{} f(f(f(undefine(`f')`hello world'))) @result{}f:f:f:hello world f(`bye') @result{}f(bye) @end example It is not an error for @var{name} to have no macro definition. In that case, @code{undefine} does nothing. @node Defn @section Renaming macros @cindex macros, how to rename @cindex renaming macros @cindex macros, displaying definitions @cindex definitions, displaying macro It is possible to rename an already defined macro. To do this, you need the builtin @code{defn}: @deffn Builtin defn (@var{name}@dots{}) Expands to the @emph{quoted definition} of each @var{name}. If an argument is not a defined macro, the expansion for that argument is empty. If @var{name} is a user-defined macro, the quoted definition is simply the quoted expansion text. If, instead, there is only one @var{name} and it is a builtin, the expansion is a special token, which points to the builtin's internal definition. This token is only meaningful as the second argument to @code{define} (and @code{pushdef}), and is silently converted to an empty string in most other contexts. Combining a builtin with anything else is not supported; a warning is issued and the builtin is omitted from the final expansion. The macro @code{defn} is recognized only with parameters. @end deffn Its normal use is best understood through an example, which shows how to rename @code{undefine} to @code{zap}: @example define(`zap', defn(`undefine')) @result{} zap(`undefine') @result{} undefine(`zap') @result{}undefine(zap) @end example In this way, @code{defn} can be used to copy macro definitions, and also definitions of builtin macros. Even if the original macro is removed, the other name can still be used to access the definition. The fact that macro definitions can be transferred also explains why you should use @code{$0}, rather than retyping a macro's name in its definition: @example define(`foo', `This is `$0'') @result{} define(`bar', defn(`foo')) @result{} bar @result{}This is bar @end example Macros used as string variables should be referred through @code{defn}, to avoid unwanted expansion of the text: @example define(`string', `The macro dnl is very useful ') @result{} string @result{}The macro@w{ } defn(`string') @result{}The macro dnl is very useful @result{} @end example @cindex rescanning However, it is important to remember that @code{m4} rescanning is purely textual. If an unbalanced end-quote string occurs in a macro definition, the rescan will see that embedded quote as the termination of the quoted string, and the remainder of the macro's definition will be rescanned unquoted. Thus it is a good idea to avoid unbalanced end-quotes in macro definitions or arguments to macros. @example define(`foo', a'a) @result{} define(`a', `A') @result{} define(`echo', `$@@') @result{} foo @result{}A'A defn(`foo') @result{}aA' echo(foo) @result{}AA' @end example On the other hand, it is possible to exploit the fact that @code{defn} can concatenate multiple macros prior to the rescanning phase, in order to join the definitions of macros that, in isolation, have unbalanced quotes. This is particularly useful when one has used several macros to accumulate text that M4 should rescan as a whole. In the example below, note how the use of @code{defn} on @code{l} in isolation opens a string, which is not closed until the next line; but used on @code{l} and @code{r} together results in nested quoting. @example define(`l', `<[>')define(`r', `<]>') @result{} changequote(`[', `]') @result{} defn([l])defn([r]) ]) @result{}<[>]defn([r]) @result{}) defn([l], [r]) @result{}<[>][<]> @end example @cindex builtins, special tokens @cindex tokens, builtin macro Using @code{defn} to generate special tokens for builtin macros outside of expected contexts can sometimes trigger warnings. But most of the time, such tokens are silently converted to the empty string. @example $ @kbd{m4 -d} defn(`defn') @result{} define(defn(`divnum'), `cannot redefine a builtin token') @error{}m4:stdin:2: Warning: define: invalid macro name ignored @result{} divnum @result{}0 len(defn(`divnum')) @result{}0 @end example Also note that @code{defn} with multiple arguments can only join text macros, not builtins, although a future version of GNU M4 may lift this restriction. @example $ @kbd{m4 -d} define(`a', `A')define(`AA', `b') @result{} traceon(`defn', `define') @result{} defn(`a', `divnum', `a') @error{}m4:stdin:3: Warning: cannot concatenate builtin `divnum' @error{}m4trace: -1- defn(`a', `divnum', `a') -> ``A'`A'' @result{}AA define(`mydivnum', defn(`divnum', `divnum'))mydivnum @error{}m4:stdin:4: Warning: cannot concatenate builtin `divnum' @error{}m4:stdin:4: Warning: cannot concatenate builtin `divnum' @error{}m4trace: -2- defn(`divnum', `divnum') @error{}m4trace: -1- define(`mydivnum', `') @result{} traceoff(`defn', `define') @result{} @end example @node Pushdef @section Temporarily redefining macros @cindex macros, temporary redefinition of @cindex temporary redefinition of macros @cindex redefinition of macros, temporary @cindex definition stack @cindex pushdef stack @cindex stack, macro definition It is possible to redefine a macro temporarily, reverting to the previous definition at a later time. This is done with the builtins @code{pushdef} and @code{popdef}: @deffn Builtin pushdef (@var{name}, @ovar{expansion}) @deffnx Builtin popdef (@var{name}@dots{}) Analogous to @code{define} and @code{undefine}. These macros work in a stack-like fashion. A macro is temporarily redefined with @code{pushdef}, which replaces an existing definition of @var{name}, while saving the previous definition, before the new one is installed. If there is no previous definition, @code{pushdef} behaves exactly like @code{define}. If a macro has several definitions (of which only one is accessible), the topmost definition can be removed with @code{popdef}. If there is no previous definition, @code{popdef} behaves like @code{undefine}. The expansion of both @code{pushdef} and @code{popdef} is void. The macros @code{pushdef} and @code{popdef} are recognized only with parameters. @end deffn @example define(`foo', `Expansion one.') @result{} foo @result{}Expansion one. pushdef(`foo', `Expansion two.') @result{} foo @result{}Expansion two. pushdef(`foo', `Expansion three.') @result{} pushdef(`foo', `Expansion four.') @result{} popdef(`foo') @result{} foo @result{}Expansion three. popdef(`foo', `foo') @result{} foo @result{}Expansion one. popdef(`foo') @result{} foo @result{}foo @end example If a macro with several definitions is redefined with @code{define}, the topmost definition is @emph{replaced} with the new definition. If it is removed with @code{undefine}, @emph{all} the definitions are removed, and not only the topmost one. However, POSIX allows other implementations that treat @code{define} as replacing an entire stack of definitions with a single new definition, so to be portable to other implementations, it may be worth explicitly using @code{popdef} and @code{pushdef} rather than relying on the GNU behavior of @code{define}. @example define(`foo', `Expansion one.') @result{} foo @result{}Expansion one. pushdef(`foo', `Expansion two.') @result{} foo @result{}Expansion two. define(`foo', `Second expansion two.') @result{} foo @result{}Second expansion two. undefine(`foo') @result{} foo @result{}foo @end example @cindex local variables @cindex variables, local Local variables within macros are made with @code{pushdef} and @code{popdef}. At the start of the macro a new definition is pushed, within the macro it is manipulated and at the end it is popped, revealing the former definition. It is possible to temporarily redefine a builtin with @code{pushdef} and @code{defn}. @node Indir @section Indirect call of macros @cindex indirect call of macros @cindex call of macros, indirect @cindex macros, indirect call of @cindex GNU extensions Any macro can be called indirectly with @code{indir}: @deffn Builtin indir (@var{name}, @ovar{args@dots{}}) Results in a call to the macro @var{name}, which is passed the rest of the arguments @var{args}. If @var{name} is not defined, an error message is printed, and the expansion is void. The macro @code{indir} is recognized only with parameters. @end deffn This can be used to call macros with computed or ``invalid'' names (@code{define} allows such names to be defined): @example define(`$$internal$macro', `Internal macro (name `$0')') @result{} $$internal$macro @result{}$$internal$macro indir(`$$internal$macro') @result{}Internal macro (name $$internal$macro) @end example The point is, here, that larger macro packages can have private macros defined, that will not be called by accident. They can @emph{only} be called through the builtin @code{indir}. One other point to observe is that argument collection occurs before @code{indir} invokes @var{name}, so if argument collection changes the value of @var{name}, that will be reflected in the final expansion. This is different than the behavior when invoking macros directly, where the definition that was in effect before argument collection is used. @example $ @kbd{m4 -d} define(`f', `1') @result{} f(define(`f', `2')) @result{}1 indir(`f', define(`f', `3')) @result{}3 indir(`f', undefine(`f')) @error{}m4:stdin:4: undefined macro `f' @result{} @end example When handed the result of @code{defn} (@pxref{Defn}) as one of its arguments, @code{indir} defers to the invoked @var{name} for whether a token representing a builtin is recognized or flattened to the empty string. @example $ @kbd{m4 -d} indir(defn(`defn'), `divnum') @error{}m4:stdin:1: Warning: indir: invalid macro name ignored @result{} indir(`define', defn(`defn'), `divnum') @error{}m4:stdin:2: Warning: define: invalid macro name ignored @result{} indir(`define', `foo', defn(`divnum')) @result{} foo @result{}0 indir(`divert', defn(`foo')) @error{}m4:stdin:5: empty string treated as 0 in builtin `divert' @result{} @end example @node Builtin @section Indirect call of builtins @cindex indirect call of builtins @cindex call of builtins, indirect @cindex builtins, indirect call of @cindex GNU extensions Builtin macros can be called indirectly with @code{builtin}: @deffn Builtin builtin (@var{name}, @ovar{args@dots{}}) Results in a call to the builtin @var{name}, which is passed the rest of the arguments @var{args}. If @var{name} does not name a builtin, an error message is printed, and the expansion is void. The macro @code{builtin} is recognized only with parameters. @end deffn This can be used even if @var{name} has been given another definition that has covered the original, or been undefined so that no macro maps to the builtin. @example pushdef(`define', `hidden') @result{} undefine(`undefine') @result{} define(`foo', `bar') @result{}hidden foo @result{}foo builtin(`define', `foo', defn(`divnum')) @result{} foo @result{}0 builtin(`define', `foo', `BAR') @result{} foo @result{}BAR undefine(`foo') @result{}undefine(foo) foo @result{}BAR builtin(`undefine', `foo') @result{} foo @result{}foo @end example The @var{name} argument only matches the original name of the builtin, even when the @option{--prefix-builtins} option (or @option{-P}, @pxref{Operation modes, , Invoking m4}) is in effect. This is different from @code{indir}, which only tracks current macro names. @comment options: -P @example $ @kbd{m4 -P} m4_builtin(`divnum') @result{}0 m4_builtin(`m4_divnum') @error{}m4:stdin:2: undefined builtin `m4_divnum' @result{} m4_indir(`divnum') @error{}m4:stdin:3: undefined macro `divnum' @result{} m4_indir(`m4_divnum') @result{}0 @end example Note that @code{indir} and @code{builtin} can be used to invoke builtins without arguments, even when they normally require parameters to be recognized; but it will provoke a warning, and result in a void expansion. @example builtin @result{}builtin builtin() @error{}m4:stdin:2: undefined builtin `' @result{} builtin(`builtin') @error{}m4:stdin:3: Warning: too few arguments to builtin `builtin' @result{} builtin(`builtin',) @error{}m4:stdin:4: undefined builtin `' @result{} builtin(`builtin', ``' ') @error{}m4:stdin:5: undefined builtin ``' @error{}' @result{} indir(`index') @error{}m4:stdin:7: Warning: too few arguments to builtin `index' @result{} @end example @ignore @comment This example is not worth putting in the manual, but it is @comment needed for full coverage. Autoconf's m4_include relies heavily @comment on this feature. @example builtin(`include', `foo')dnl @result{}bar @end example @comment And this example triggers a regression present in 1.4.10b. @example define(`s', `builtin(`shift', $@@)')dnl define(`loop', `ifelse(`$2', `', `-', `$1$2: $0(`$1', s(s($@@)))')')dnl loop(`1') @result{}- loop(`1', `2') @result{}12: - loop(`1', `2', `3') @result{}12: 13: - loop(`1', `2', `3', `4') @result{}12: 13: 14: - loop(`1', `2', `3', `4', `5') @result{}12: 13: 14: 15: - @end example @end ignore @node Conditionals @chapter Conditionals, loops, and recursion Macros, expanding to plain text, perhaps with arguments, are not quite enough. We would like to have macros expand to different things, based on decisions taken at run-time. For that, we need some kind of conditionals. Also, we would like to have some kind of loop construct, so we could do something a number of times, or while some condition is true. @menu * Ifdef:: Testing if a macro is defined * Ifelse:: If-else construct, or multibranch * Shift:: Recursion in @code{m4} * Forloop:: Iteration by counting * Foreach:: Iteration by list contents * Stacks:: Working with definition stacks * Composition:: Building macros with macros @end menu @node Ifdef @section Testing if a macro is defined @cindex conditionals There are two different builtin conditionals in @code{m4}. The first is @code{ifdef}: @deffn Builtin ifdef (@var{name}, @var{string-1}, @ovar{string-2}) If @var{name} is defined as a macro, @code{ifdef} expands to @var{string-1}, otherwise to @var{string-2}. If @var{string-2} is omitted, it is taken to be the empty string (according to the normal rules). The macro @code{ifdef} is recognized only with parameters. @end deffn @example ifdef(`foo', ``foo' is defined', ``foo' is not defined') @result{}foo is not defined define(`foo', `') @result{} ifdef(`foo', ``foo' is defined', ``foo' is not defined') @result{}foo is defined ifdef(`no_such_macro', `yes', `no', `extra argument') @error{}m4:stdin:4: Warning: excess arguments to builtin `ifdef' ignored @result{}no @end example @node Ifelse @section If-else construct, or multibranch @cindex comparing strings @cindex discarding input @cindex input, discarding The other conditional, @code{ifelse}, is much more powerful. It can be used as a way to introduce a long comment, as an if-else construct, or as a multibranch, depending on the number of arguments supplied: @deffn Builtin ifelse (@var{comment}) @deffnx Builtin ifelse (@var{string-1}, @var{string-2}, @var{equal}, @ @ovar{not-equal}) @deffnx Builtin ifelse (@var{string-1}, @var{string-2}, @var{equal-1}, @ @var{string-3}, @var{string-4}, @var{equal-2}, @dots{}, @ovar{not-equal}) Used with only one argument, the @code{ifelse} simply discards it and produces no output. If called with three or four arguments, @code{ifelse} expands into @var{equal}, if @var{string-1} and @var{string-2} are equal (character for character), otherwise it expands to @var{not-equal}. A final fifth argument is ignored, after triggering a warning. If called with six or more arguments, and @var{string-1} and @var{string-2} are equal, @code{ifelse} expands into @var{equal-1}, otherwise the first three arguments are discarded and the processing starts again. The macro @code{ifelse} is recognized only with parameters. @end deffn Using only one argument is a common @code{m4} idiom for introducing a block comment, as an alternative to repeatedly using @code{dnl}. This special usage is recognized by GNU @code{m4}, so that in this case, the warning about missing arguments is never triggered. @example ifelse(`some comments') @result{} ifelse(`foo', `bar') @error{}m4:stdin:2: Warning: too few arguments to builtin `ifelse' @result{} @end example Using three or four arguments provides decision points. @example ifelse(`foo', `bar', `true') @result{} ifelse(`foo', `foo', `true') @result{}true define(`foo', `bar') @result{} ifelse(foo, `bar', `true', `false') @result{}true ifelse(foo, `foo', `true', `false') @result{}false @end example @cindex macro, blind @cindex blind macro Notice how the first argument was used unquoted; it is common to compare the expansion of a macro with a string. With this macro, you can now reproduce the behavior of blind builtins, where the macro is recognized only with arguments. @example define(`foo', `ifelse(`$#', `0', ``$0'', `arguments:$#')') @result{} foo @result{}foo foo() @result{}arguments:1 foo(`a', `b', `c') @result{}arguments:3 @end example For an example of a way to make defining blind macros easier, see @ref{Composition}. @cindex multibranches @cindex switch statement @cindex case statement The macro @code{ifelse} can take more than four arguments. If given more than four arguments, @code{ifelse} works like a @code{case} or @code{switch} statement in traditional programming languages. If @var{string-1} and @var{string-2} are equal, @code{ifelse} expands into @var{equal-1}, otherwise the procedure is repeated with the first three arguments discarded. This calls for an example: @example ifelse(`foo', `bar', `third', `gnu', `gnats') @error{}m4:stdin:1: Warning: excess arguments to builtin `ifelse' ignored @result{}gnu ifelse(`foo', `bar', `third', `gnu', `gnats', `sixth') @result{} ifelse(`foo', `bar', `third', `gnu', `gnats', `sixth', `seventh') @result{}seventh ifelse(`foo', `bar', `3', `gnu', `gnats', `6', `7', `8') @error{}m4:stdin:4: Warning: excess arguments to builtin `ifelse' ignored @result{}7 @end example @ignore @comment Stress tests, not worth documenting. @comment Ensure that references compared to strings work regardless of @comment similar prefixes. @example define(`e', `$@@')define(`long', `01234567890123456789') @result{} ifelse(long, `01234567890123456789', `yes', `no') @result{}yes ifelse(`01234567890123456789', long, `yes', `no') @result{}yes ifelse(long, `01234567890123456789-', `yes', `no') @result{}no ifelse(`01234567890123456789-', long, `yes', `no') @result{}no ifelse(e(long), `01234567890123456789', `yes', `no') @result{}yes ifelse(`01234567890123456789', e(long), `yes', `no') @result{}yes ifelse(e(long), `01234567890123456789-', `yes', `no') @result{}no ifelse(`01234567890123456789-', e(long), `yes', `no') @result{}no ifelse(-e(long), `-01234567890123456789', `yes', `no') @result{}yes ifelse(-`01234567890123456789', -e(long), `yes', `no') @result{}yes ifelse(-e(long), `-01234567890123456789-', `yes', `no') @result{}no ifelse(`-01234567890123456789-', -e(long), `yes', `no') @result{}no ifelse(-e(long)-, `-01234567890123456789-', `yes', `no') @result{}yes ifelse(-`01234567890123456789-', -e(long)-, `yes', `no') @result{}yes ifelse(-e(long)-, `-01234567890123456789', `yes', `no') @result{}no ifelse(`-01234567890123456789', -e(long)-, `yes', `no') @result{}no ifelse(`-'e(long), `-01234567890123456789', `yes', `no') @result{}yes ifelse(-`01234567890123456789', `-'e(long), `yes', `no') @result{}yes ifelse(`-'e(long), `-01234567890123456789-', `yes', `no') @result{}no ifelse(`-01234567890123456789-', `-'e(long), `yes', `no') @result{}no ifelse(`-'e(long)`-', `-01234567890123456789-', `yes', `no') @result{}yes ifelse(-`01234567890123456789-', `-'e(long)`-', `yes', `no') @result{}yes ifelse(`-'e(long)`-', `-01234567890123456789', `yes', `no') @result{}no ifelse(`-01234567890123456789', `-'e(long)`-', `yes', `no') @result{}no @end example @end ignore Naturally, the normal case will be slightly more advanced than these examples. A common use of @code{ifelse} is in macros implementing loops of various kinds. @node Shift @section Recursion in @code{m4} @cindex recursive macros @cindex macros, recursive There is no direct support for loops in @code{m4}, but macros can be recursive. There is no limit on the number of recursion levels, other than those enforced by your hardware and operating system. @cindex loops Loops can be programmed using recursion and the conditionals described previously. There is a builtin macro, @code{shift}, which can, among other things, be used for iterating through the actual arguments to a macro: @deffn Builtin shift (@var{arg1}, @dots{}) Takes any number of arguments, and expands to all its arguments except @var{arg1}, separated by commas, with each argument quoted. The macro @code{shift} is recognized only with parameters. @end deffn @example shift @result{}shift shift(`bar') @result{} shift(`foo', `bar', `baz') @result{}bar,baz @end example An example of the use of @code{shift} is this macro: @cindex reversing arguments @cindex arguments, reversing @deffn Composite reverse (@dots{}) Takes any number of arguments, and reverses their order. @end deffn It is implemented as: @example define(`reverse', `ifelse(`$#', `0', , `$#', `1', ``$1'', `reverse(shift($@@)), `$1'')') @result{} reverse @result{} reverse(`foo') @result{}foo reverse(`foo', `bar', `gnats', `and gnus') @result{}and gnus, gnats, bar, foo @end example While not a very interesting macro, it does show how simple loops can be made with @code{shift}, @code{ifelse} and recursion. It also shows that @code{shift} is usually used with @samp{$@@}. Another example of this is an implementation of a short-circuiting conditional operator. @cindex short-circuiting conditional @cindex conditional, short-circuiting @deffn Composite cond (@var{test-1}, @var{string-1}, @var{equal-1}, @ @ovar{test-2}, @ovar{string-2}, @ovar{equal-2}, @dots{}, @ovar{not-equal}) Similar to @code{ifelse}, where an equal comparison between the first two strings results in the third, otherwise the first three arguments are discarded and the process repeats. The difference is that each @var{test-} is expanded only when it is encountered. This means that every third argument to @code{cond} is normally given one more level of quoting than the corresponding argument to @code{ifelse}. @end deffn Here is the implementation of @code{cond}, along with a demonstration of how it can short-circuit the side effects in @code{side}. Notice how all the unquoted side effects happen regardless of how many comparisons are made with @code{ifelse}, compared with only the relevant effects with @code{cond}. @example define(`cond', `ifelse(`$#', `1', `$1', `ifelse($1, `$2', `$3', `$0(shift(shift(shift($@@))))')')')dnl define(`side', `define(`counter', incr(counter))$1')dnl define(`example1', `define(`counter', `0')dnl ifelse(side(`$1'), `yes', `one comparison: ', side(`$1'), `no', `two comparisons: ', side(`$1'), `maybe', `three comparisons: ', `side(`default answer: ')')counter')dnl define(`example2', `define(`counter', `0')dnl cond(`side(`$1')', `yes', `one comparison: ', `side(`$1')', `no', `two comparisons: ', `side(`$1')', `maybe', `three comparisons: ', `side(`default answer: ')')counter')dnl example1(`yes') @result{}one comparison: 3 example1(`no') @result{}two comparisons: 3 example1(`maybe') @result{}three comparisons: 3 example1(`feeling rather indecisive today') @result{}default answer: 4 example2(`yes') @result{}one comparison: 1 example2(`no') @result{}two comparisons: 2 example2(`maybe') @result{}three comparisons: 3 example2(`feeling rather indecisive today') @result{}default answer: 4 @end example @cindex joining arguments @cindex arguments, joining @cindex concatenating arguments Another common task that requires iteration is joining a list of arguments into a single string. @deffn Composite join (@ovar{separator}, @ovar{args@dots{}}) @deffnx Composite joinall (@ovar{separator}, @ovar{args@dots{}}) Generate a single-quoted string, consisting of each @var{arg} separated by @var{separator}. While @code{joinall} always outputs a @var{separator} between arguments, @code{join} avoids the @var{separator} for an empty @var{arg}. @end deffn Here are some examples of its usage, based on the implementation @file{m4-@value{VERSION}/@/examples/@/join.m4} distributed in this package: @comment examples @example $ @kbd{m4 -I examples} include(`join.m4') @result{} join,join(`-'),join(`-', `'),join(`-', `', `') @result{},,, joinall,joinall(`-'),joinall(`-', `'),joinall(`-', `', `') @result{},,,- join(`-', `1') @result{}1 join(`-', `1', `2', `3') @result{}1-2-3 join(`', `1', `2', `3') @result{}123 join(`-', `', `1', `', `', `2', `') @result{}1-2 joinall(`-', `', `1', `', `', `2', `') @result{}-1---2- join(`,', `1', `2', `3') @result{}1,2,3 define(`nargs', `$#')dnl nargs(join(`,', `1', `2', `3')) @result{}1 @end example Examining the implementation shows some interesting points about several m4 programming idioms. @comment examples @example $ @kbd{m4 -I examples} undivert(`join.m4')dnl @result{}divert(`-1') @result{}# join(sep, args) - join each non-empty ARG into a single @result{}# string, with each element separated by SEP @result{}define(`join', @result{}`ifelse(`$#', `2', ``$2'', @result{} `ifelse(`$2', `', `', ``$2'_')$0(`$1', shift(shift($@@)))')') @result{}define(`_join', @result{}`ifelse(`$#$2', `2', `', @result{} `ifelse(`$2', `', `', ``$1$2'')$0(`$1', shift(shift($@@)))')') @result{}# joinall(sep, args) - join each ARG, including empty ones, @result{}# into a single string, with each element separated by SEP @result{}define(`joinall', ``$2'_$0(`$1', shift($@@))') @result{}define(`_joinall', @result{}`ifelse(`$#', `2', `', ``$1$3'$0(`$1', shift(shift($@@)))')') @result{}divert`'dnl @end example First, notice that this implementation creates helper macros @code{_join} and @code{_joinall}. This division of labor makes it easier to output the correct number of @var{separator} instances: @code{join} and @code{joinall} are responsible for the first argument, without a separator, while @code{_join} and @code{_joinall} are responsible for all remaining arguments, always outputting a separator when outputting an argument. Next, observe how @code{join} decides to iterate to itself, because the first @var{arg} was empty, or to output the argument and swap over to @code{_join}. If the argument is non-empty, then the nested @code{ifelse} results in an unquoted @samp{_}, which is concatenated with the @samp{$0} to form the next macro name to invoke. The @code{joinall} implementation is simpler since it does not have to suppress empty @var{arg}; it always executes once then defers to @code{_joinall}. Another important idiom is the idea that @var{separator} is reused for each iteration. Each iteration has one less argument, but rather than discarding @samp{$1} by iterating with @code{$0(shift($@@))}, the macro discards @samp{$2} by using @code{$0(`$1', shift(shift($@@)))}. Next, notice that it is possible to compare more than one condition in a single @code{ifelse} test. The test of @samp{$#$2} against @samp{2} allows @code{_join} to iterate for two separate reasons---either there are still more than two arguments, or there are exactly two arguments but the last argument is not empty. Finally, notice that these macros require exactly two arguments to terminate recursion, but that they still correctly result in empty output when given no @var{args} (i.e., zero or one macro argument). On the first pass when there are too few arguments, the @code{shift} results in no output, but leaves an empty string to serve as the required second argument for the second pass. Put another way, @samp{`$1', shift($@@)} is not the same as @samp{$@@}, since only the former guarantees at least two arguments. @cindex quote manipulation @cindex manipulating quotes Sometimes, a recursive algorithm requires adding quotes to each element, or treating multiple arguments as a single element: @deffn Composite quote (@dots{}) @deffnx Composite dquote (@dots{}) @deffnx Composite dquote_elt (@dots{}) Takes any number of arguments, and adds quoting. With @code{quote}, only one level of quoting is added, effectively removing whitespace after commas and turning multiple arguments into a single string. With @code{dquote}, two levels of quoting are added, one around each element, and one around the list. And with @code{dquote_elt}, two levels of quoting are added around each element. @end deffn An actual implementation of these three macros is distributed as @file{m4-@value{VERSION}/@/examples/@/quote.m4} in this package. First, let's examine their usage: @comment examples @example $ @kbd{m4 -I examples} include(`quote.m4') @result{} -quote-dquote-dquote_elt- @result{}---- -quote()-dquote()-dquote_elt()- @result{}--`'-`'- -quote(`1')-dquote(`1')-dquote_elt(`1')- @result{}-1-`1'-`1'- -quote(`1', `2')-dquote(`1', `2')-dquote_elt(`1', `2')- @result{}-1,2-`1',`2'-`1',`2'- define(`n', `$#')dnl -n(quote(`1', `2'))-n(dquote(`1', `2'))-n(dquote_elt(`1', `2'))- @result{}-1-1-2- dquote(dquote_elt(`1', `2')) @result{}``1'',``2'' dquote_elt(dquote(`1', `2')) @result{}``1',`2'' @end example The last two lines show that when given two arguments, @code{dquote} results in one string, while @code{dquote_elt} results in two. Now, examine the implementation. Note that @code{quote} and @code{dquote_elt} make decisions based on their number of arguments, so that when called without arguments, they result in nothing instead of a quoted empty string; this is so that it is possible to distinguish between no arguments and an empty first argument. @code{dquote}, on the other hand, results in a string no matter what, since it is still possible to tell whether it was invoked without arguments based on the resulting string. @comment examples @example $ @kbd{m4 -I examples} undivert(`quote.m4')dnl @result{}divert(`-1') @result{}# quote(args) - convert args to single-quoted string @result{}define(`quote', `ifelse(`$#', `0', `', ``$*'')') @result{}# dquote(args) - convert args to quoted list of quoted strings @result{}define(`dquote', ``$@@'') @result{}# dquote_elt(args) - convert args to list of double-quoted strings @result{}define(`dquote_elt', `ifelse(`$#', `0', `', `$#', `1', ```$1''', @result{} ```$1'',$0(shift($@@))')') @result{}divert`'dnl @end example It is worth pointing out that @samp{quote(@var{args})} is more efficient than @samp{joinall(`,', @var{args})} for producing the same output. @cindex nine arguments, more than @cindex more than nine arguments @cindex arguments, more than nine One more useful macro based on @code{shift} allows portably selecting an arbitrary argument (usually greater than the ninth argument), without relying on the GNU extension of multi-digit arguments (@pxref{Arguments}). @deffn Composite argn (@var{n}, @dots{}) Expands to argument @var{n} out of the remaining arguments. @var{n} must be a positive number. Usually invoked as @samp{argn(`@var{n}',$@@)}. @end deffn It is implemented as: @example define(`argn', `ifelse(`$1', 1, ``$2'', `argn(decr(`$1'), shift(shift($@@)))')') @result{} argn(`1', `a') @result{}a define(`foo', `argn(`11', $@@)') @result{} foo(`a', `b', `c', `d', `e', `f', `g', `h', `i', `j', `k', `l') @result{}k @end example @node Forloop @section Iteration by counting @cindex for loops @cindex loops, counting @cindex counting loops Here is an example of a loop macro that implements a simple for loop. @deffn Composite forloop (@var{iterator}, @var{start}, @var{end}, @var{text}) Takes the name in @var{iterator}, which must be a valid macro name, and successively assign it each integer value from @var{start} to @var{end}, inclusive. For each assignment to @var{iterator}, append @var{text} to the expansion of the @code{forloop}. @var{text} may refer to @var{iterator}. Any definition of @var{iterator} prior to this invocation is restored. @end deffn It can, for example, be used for simple counting: @comment examples @example $ @kbd{m4 -I examples} include(`forloop.m4') @result{} forloop(`i', `1', `8', `i ') @result{}1 2 3 4 5 6 7 8@w{ } @end example For-loops can be nested, like: @comment examples @example $ @kbd{m4 -I examples} include(`forloop.m4') @result{} forloop(`i', `1', `4', `forloop(`j', `1', `8', ` (i, j)') ') @result{} (1, 1) (1, 2) (1, 3) (1, 4) (1, 5) (1, 6) (1, 7) (1, 8) @result{} (2, 1) (2, 2) (2, 3) (2, 4) (2, 5) (2, 6) (2, 7) (2, 8) @result{} (3, 1) (3, 2) (3, 3) (3, 4) (3, 5) (3, 6) (3, 7) (3, 8) @result{} (4, 1) (4, 2) (4, 3) (4, 4) (4, 5) (4, 6) (4, 7) (4, 8) @result{} @end example The implementation of the @code{forloop} macro is fairly straightforward. The @code{forloop} macro itself is simply a wrapper, which saves the previous definition of the first argument, calls the internal macro @code{@w{_forloop}}, and re-establishes the saved definition of the first argument. The macro @code{@w{_forloop}} expands the fourth argument once, and tests to see if the iterator has reached the final value. If it has not finished, it increments the iterator (using the predefined macro @code{incr}, @pxref{Incr}), and recurses. Here is an actual implementation of @code{forloop}, distributed as @file{m4-@value{VERSION}/@/examples/@/forloop.m4} in this package: @comment examples @example $ @kbd{m4 -I examples} undivert(`forloop.m4')dnl @result{}divert(`-1') @result{}# forloop(var, from, to, stmt) - simple version @result{}define(`forloop', `pushdef(`$1', `$2')_forloop($@@)popdef(`$1')') @result{}define(`_forloop', @result{} `$4`'ifelse($1, `$3', `', `define(`$1', incr($1))$0($@@)')') @result{}divert`'dnl @end example Notice the careful use of quotes. Certain macro arguments are left unquoted, each for its own reason. Try to find out @emph{why} these arguments are left unquoted, and see what happens if they are quoted. (As presented, these two macros are useful but not very robust for general use. They lack even basic error handling for cases like @var{start} less than @var{end}, @var{end} not numeric, or @var{iterator} not being a macro name. See if you can improve these macros; or @pxref{Improved forloop, , Answers}). @node Foreach @section Iteration by list contents @cindex for each loops @cindex loops, list iteration @cindex iterating over lists Here is an example of a loop macro that implements list iteration. @deffn Composite foreach (@var{iterator}, @var{paren-list}, @var{text}) @deffnx Composite foreachq (@var{iterator}, @var{quote-list}, @var{text}) Takes the name in @var{iterator}, which must be a valid macro name, and successively assign it each value from @var{paren-list} or @var{quote-list}. In @code{foreach}, @var{paren-list} is a comma-separated list of elements contained in parentheses. In @code{foreachq}, @var{quote-list} is a comma-separated list of elements contained in a quoted string. For each assignment to @var{iterator}, append @var{text} to the overall expansion. @var{text} may refer to @var{iterator}. Any definition of @var{iterator} prior to this invocation is restored. @end deffn As an example, this displays each word in a list inside of a sentence, using an implementation of @code{foreach} distributed as @file{m4-@value{VERSION}/@/examples/@/foreach.m4}, and @code{foreachq} in @file{m4-@value{VERSION}/@/examples/@/foreachq.m4}. @comment examples @example $ @kbd{m4 -I examples} include(`foreach.m4') @result{} foreach(`x', (foo, bar, foobar), `Word was: x ')dnl @result{}Word was: foo @result{}Word was: bar @result{}Word was: foobar include(`foreachq.m4') @result{} foreachq(`x', `foo, bar, foobar', `Word was: x ')dnl @result{}Word was: foo @result{}Word was: bar @result{}Word was: foobar @end example It is possible to be more complex; each element of the @var{paren-list} or @var{quote-list} can itself be a list, to pass as further arguments to a helper macro. This example generates a shell case statement: @comment examples @example $ @kbd{m4 -I examples} include(`foreach.m4') @result{} define(`_case', ` $1) $2=" $1";; ')dnl define(`_cat', `$1$2')dnl case $`'1 in @result{}case $1 in foreach(`x', `(`(`a', `vara')', `(`b', `varb')', `(`c', `varc')')', `_cat(`_case', x)')dnl @result{} a) @result{} vara=" a";; @result{} b) @result{} varb=" b";; @result{} c) @result{} varc=" c";; esac @result{}esac @end example The implementation of the @code{foreach} macro is a bit more involved; it is a wrapper around two helper macros. First, @code{@w{_arg1}} is needed to grab the first element of a list. Second, @code{@w{_foreach}} implements the recursion, successively walking through the original list. Here is a simple implementation of @code{foreach}: @comment examples @example $ @kbd{m4 -I examples} undivert(`foreach.m4')dnl @result{}divert(`-1') @result{}# foreach(x, (item_1, item_2, ..., item_n), stmt) @result{}# parenthesized list, simple version @result{}define(`foreach', `pushdef(`$1')_foreach($@@)popdef(`$1')') @result{}define(`_arg1', `$1') @result{}define(`_foreach', `ifelse(`$2', `()', `', @result{} `define(`$1', _arg1$2)$3`'$0(`$1', (shift$2), `$3')')') @result{}divert`'dnl @end example Unfortunately, that implementation is not robust to macro names as list elements. Each iteration of @code{@w{_foreach}} is stripping another layer of quotes, leading to erratic results if list elements are not already fully expanded. The first cut at implementing @code{foreachq} takes this into account. Also, when using quoted elements in a @var{paren-list}, the overall list must be quoted. A @var{quote-list} has the nice property of requiring fewer characters to create a list containing the same quoted elements. To see the difference between the two macros, we attempt to pass double-quoted macro names in a list, expecting the macro name on output after one layer of quotes is removed during list iteration and the final layer removed during the final rescan: @comment examples @example $ @kbd{m4 -I examples} define(`a', `1')define(`b', `2')define(`c', `3') @result{} include(`foreach.m4') @result{} include(`foreachq.m4') @result{} foreach(`x', `(``a'', ``(b'', ``c)'')', `x ') @result{}1 @result{}(2)1 @result{} @result{}, x @result{}) foreachq(`x', ```a'', ``(b'', ``c)''', `x ')dnl @result{}a @result{}(b @result{}c) @end example Obviously, @code{foreachq} did a better job; here is its implementation: @comment examples @example $ @kbd{m4 -I examples} undivert(`foreachq.m4')dnl @result{}include(`quote.m4')dnl @result{}divert(`-1') @result{}# foreachq(x, `item_1, item_2, ..., item_n', stmt) @result{}# quoted list, simple version @result{}define(`foreachq', `pushdef(`$1')_foreachq($@@)popdef(`$1')') @result{}define(`_arg1', `$1') @result{}define(`_foreachq', `ifelse(quote($2), `', `', @result{} `define(`$1', `_arg1($2)')$3`'$0(`$1', `shift($2)', `$3')')') @result{}divert`'dnl @end example Notice that @code{@w{_foreachq}} had to use the helper macro @code{quote} defined earlier (@pxref{Shift}), to ensure that the embedded @code{ifelse} call does not go haywire if a list element contains a comma. Unfortunately, this implementation of @code{foreachq} has its own severe flaw. Whereas the @code{foreach} implementation was linear, this macro is quadratic in the number of list elements, and is much more likely to trip up the limit set by the command line option @option{--nesting-limit} (or @option{-L}, @pxref{Limits control, , Invoking m4}). Additionally, this implementation does not expand @samp{defn(`@var{iterator}')} very well, when compared with @code{foreach}. @comment examples @example $ @kbd{m4 -I examples} include(`foreach.m4')include(`foreachq.m4') @result{} foreach(`name', `(`a', `b')', ` defn(`name')') @result{} a b foreachq(`name', ``a', `b'', ` defn(`name')') @result{} _arg1(`a', `b') _arg1(shift(`a', `b')) @end example It is possible to have robust iteration with linear behavior and sane @var{iterator} contents for either list style. See if you can learn from the best elements of both of these implementations to create robust macros (or @pxref{Improved foreach, , Answers}). @node Stacks @section Working with definition stacks @cindex definition stack @cindex pushdef stack @cindex stack, macro definition Thanks to @code{pushdef}, manipulation of a stack is an intrinsic operation in @code{m4}. Normally, only the topmost definition in a stack is important, but sometimes, it is desirable to manipulate the entire definition stack. @deffn Composite stack_foreach (@var{macro}, @var{action}) @deffnx Composite stack_foreach_lifo (@var{macro}, @var{action}) For each of the @code{pushdef} definitions associated with @var{macro}, invoke the macro @var{action} with a single argument of that definition. @code{stack_foreach} visits the oldest definition first, while @code{stack_foreach_lifo} visits the current definition first. @var{action} should not modify or dereference @var{macro}. There are a few special macros, such as @code{defn}, which cannot be used as the @var{macro} parameter. @end deffn A sample implementation of these macros is distributed in the file @file{m4-@value{VERSION}/@/examples/@/stack.m4}. @comment examples @example $ @kbd{m4 -I examples} include(`stack.m4') @result{} pushdef(`a', `1')pushdef(`a', `2')pushdef(`a', `3') @result{} define(`show', ``$1' ') @result{} stack_foreach(`a', `show')dnl @result{}1 @result{}2 @result{}3 stack_foreach_lifo(`a', `show')dnl @result{}3 @result{}2 @result{}1 @end example Now for the implementation. Note the definition of a helper macro, @code{_stack_reverse}, which destructively swaps the contents of one stack of definitions into the reverse order in the temporary macro @samp{tmp-$1}. By calling the helper twice, the original order is restored back into the macro @samp{$1}; since the operation is destructive, this explains why @samp{$1} must not be modified or dereferenced during the traversal. The caller can then inject additional code to pass the definition currently being visited to @samp{$2}. The choice of helper names is intentional; since @samp{-} is not valid as part of a macro name, there is no risk of conflict with a valid macro name, and the code is guaranteed to use @code{defn} where necessary. Finally, note that any macro used in the traversal of a @code{pushdef} stack, such as @code{pushdef} or @code{defn}, cannot be handled by @code{stack_foreach}, since the macro would temporarily be undefined during the algorithm. @comment examples @example $ @kbd{m4 -I examples} undivert(`stack.m4')dnl @result{}divert(`-1') @result{}# stack_foreach(macro, action) @result{}# Invoke ACTION with a single argument of each definition @result{}# from the definition stack of MACRO, starting with the oldest. @result{}define(`stack_foreach', @result{}`_stack_reverse(`$1', `tmp-$1')'dnl @result{}`_stack_reverse(`tmp-$1', `$1', `$2(defn(`$1'))')') @result{}# stack_foreach_lifo(macro, action) @result{}# Invoke ACTION with a single argument of each definition @result{}# from the definition stack of MACRO, starting with the newest. @result{}define(`stack_foreach_lifo', @result{}`_stack_reverse(`$1', `tmp-$1', `$2(defn(`$1'))')'dnl @result{}`_stack_reverse(`tmp-$1', `$1')') @result{}define(`_stack_reverse', @result{}`ifdef(`$1', `pushdef(`$2', defn(`$1'))$3`'popdef(`$1')$0($@@)')') @result{}divert`'dnl @end example @node Composition @section Building macros with macros @cindex macro composition @cindex composing macros Since m4 is a macro language, it is possible to write macros that can build other macros. First on the list is a way to automate the creation of blind macros. @cindex macro, blind @cindex blind macro @deffn Composite define_blind (@var{name}, @ovar{value}) Defines @var{name} as a blind macro, such that @var{name} will expand to @var{value} only when given explicit arguments. @var{value} should not be the result of @code{defn} (@pxref{Defn}). This macro is only recognized with parameters, and results in an empty string. @end deffn Defining a macro to define another macro can be a bit tricky. We want to use a literal @samp{$#} in the argument to the nested @code{define}. However, if @samp{$} and @samp{#} are adjacent in the definition of @code{define_blind}, then it would be expanded as the number of arguments to @code{define_blind} rather than the intended number of arguments to @var{name}. The solution is to pass the difficult characters through extra arguments to a helper macro @code{_define_blind}. When composing macros, it is a common idiom to need a helper macro to concatenate text that forms parameters in the composed macro, rather than interpreting the text as a parameter of the composing macro. As for the limitation against using @code{defn}, there are two reasons. If a macro was previously defined with @code{define_blind}, then it can safely be renamed to a new blind macro using plain @code{define}; using @code{define_blind} to rename it just adds another layer of @code{ifelse}, occupying memory and slowing down execution. And if a macro is a builtin, then it would result in an attempt to define a macro consisting of both text and a builtin token; this is not supported, and the builtin token is flattened to an empty string. With that explanation, here's the definition, and some sample usage. Notice that @code{define_blind} is itself a blind macro. @example $ @kbd{m4 -d} define(`define_blind', `ifelse(`$#', `0', ``$0'', `_$0(`$1', `$2', `$'`#', `$'`0')')') @result{} define(`_define_blind', `define(`$1', `ifelse(`$3', `0', ``$4'', `$2')')') @result{} define_blind @result{}define_blind define_blind(`foo', `arguments were $*') @result{} foo @result{}foo foo(`bar') @result{}arguments were bar define(`blah', defn(`foo')) @result{} blah @result{}blah blah(`a', `b') @result{}arguments were a,b defn(`blah') @result{}ifelse(`$#', `0', ``$0'', `arguments were $*') @end example @cindex currying arguments @cindex argument currying Another interesting composition tactic is argument @dfn{currying}, or factoring a macro that takes multiple arguments for use in a context that provides exactly one argument. @deffn Composite curry (@var{macro}, @dots{}) Expand to a macro call that takes exactly one argument, then appends that argument to the original arguments and invokes @var{macro} with the resulting list of arguments. @end deffn A demonstration of currying makes the intent of this macro a little more obvious. The macro @code{stack_foreach} mentioned earlier is an example of a context that provides exactly one argument to a macro name. But coupled with currying, we can invoke @code{reverse} with two arguments for each definition of a macro stack. This example uses the file @file{m4-@value{VERSION}/@/examples/@/curry.m4} included in the distribution. @comment examples @example $ @kbd{m4 -I examples} include(`curry.m4')include(`stack.m4') @result{} define(`reverse', `ifelse(`$#', `0', , `$#', `1', ``$1'', `reverse(shift($@@)), `$1'')') @result{} pushdef(`a', `1')pushdef(`a', `2')pushdef(`a', `3') @result{} stack_foreach(`a', `:curry(`reverse', `4')') @result{}:1, 4:2, 4:3, 4 curry(`curry', `reverse', `1')(`2')(`3') @result{}3, 2, 1 @end example Now for the implementation. Notice how @code{curry} leaves off with a macro name but no open parenthesis, while still in the middle of collecting arguments for @samp{$1}. The macro @code{_curry} is the helper macro that takes one argument, then adds it to the list and finally supplies the closing parenthesis. The use of a comma inside the @code{shift} call allows currying to also work for a macro that takes one argument, although it often makes more sense to invoke that macro directly rather than going through @code{curry}. @comment examples @example $ @kbd{m4 -I examples} undivert(`curry.m4')dnl @result{}divert(`-1') @result{}# curry(macro, args) @result{}# Expand to a macro call that takes one argument, then invoke @result{}# macro(args, extra). @result{}define(`curry', `$1(shift($@@,)_$0') @result{}define(`_curry', ``$1')') @result{}divert`'dnl @end example Unfortunately, with M4 1.4.x, @code{curry} is unable to handle builtin tokens, which are silently flattened to the empty string when passed through another text macro. This limitation will be lifted in a future release of M4. @cindex renaming macros @cindex copying macros @cindex macros, copying Putting the last few concepts together, it is possible to copy or rename an entire stack of macro definitions. @deffn Composite copy (@var{source}, @var{dest}) @deffnx Composite rename (@var{source}, @var{dest}) Ensure that @var{dest} is undefined, then define it to the same stack of definitions currently in @var{source}. @code{copy} leaves @var{source} unchanged, while @code{rename} undefines @var{source}. There are only a few macros, such as @code{copy} or @code{defn}, which cannot be copied via this macro. @end deffn The implementation is relatively straightforward (although since it uses @code{curry}, it is unable to copy builtin macros, such as the second definition of @code{a} as a synonym for @code{divnum}. See if you can design a version that works around this limitation, or @pxref{Improved copy, , Answers}). @comment examples @example $ @kbd{m4 -I examples} include(`curry.m4')include(`stack.m4') @result{} define(`rename', `copy($@@)undefine(`$1')')dnl define(`copy', `ifdef(`$2', `errprint(`$2 already defined ')m4exit(`1')', `stack_foreach(`$1', `curry(`pushdef', `$2')')')')dnl pushdef(`a', `1')pushdef(`a', defn(`divnum'))pushdef(`a', `2') @result{} copy(`a', `b') @result{} rename(`b', `c') @result{} a b c @result{}2 b 2 popdef(`a', `c')c a @result{} 0 popdef(`a', `c')a c @result{}1 1 @end example @node Debugging @chapter How to debug macros and input @cindex debugging macros @cindex macros, debugging When writing macros for @code{m4}, they often do not work as intended on the first try (as is the case with most programming languages). Fortunately, there is support for macro debugging in @code{m4}. @menu * Dumpdef:: Displaying macro definitions * Trace:: Tracing macro calls * Debug Levels:: Controlling debugging output * Debug Output:: Saving debugging output @end menu @node Dumpdef @section Displaying macro definitions @cindex displaying macro definitions @cindex macros, displaying definitions @cindex definitions, displaying macro @cindex standard error, output to If you want to see what a name expands into, you can use the builtin @code{dumpdef}: @deffn Builtin dumpdef (@ovar{names@dots{}}) Accepts any number of arguments. If called without any arguments, it displays the definitions of all known names, otherwise it displays the definitions of the @var{names} given. The output is printed to the current debug file (usually standard error), and is sorted by name. If an unknown name is encountered, a warning is printed. The expansion of @code{dumpdef} is void. @end deffn @example $ @kbd{m4 -d} define(`foo', `Hello world.') @result{} dumpdef(`foo') @error{}foo:@tabchar{}`Hello world.' @result{} dumpdef(`define') @error{}define:@tabchar{} @result{} @end example The last example shows how builtin macros definitions are displayed. The definition that is dumped corresponds to what would occur if the macro were to be called at that point, even if other definitions are still live due to redefining a macro during argument collection. @example $ @kbd{m4 -d} pushdef(`f', ``$0'1')pushdef(`f', ``$0'2') @result{} f(popdef(`f')dumpdef(`f')) @error{}f:@tabchar{}``$0'1' @result{}f2 f(popdef(`f')dumpdef(`f')) @error{}m4:stdin:3: undefined macro `f' @result{}f1 @end example @xref{Debug Levels}, for information on controlling the details of the display. @node Trace @section Tracing macro calls @cindex tracing macro expansion @cindex macro expansion, tracing @cindex expansion, tracing macro @cindex standard error, output to It is possible to trace macro calls and expansions through the builtins @code{traceon} and @code{traceoff}: @deffn Builtin traceon (@ovar{names@dots{}}) @deffnx Builtin traceoff (@ovar{names@dots{}}) When called without any arguments, @code{traceon} and @code{traceoff} will turn tracing on and off, respectively, for all currently defined macros. When called with arguments, only the macros listed in @var{names} are affected, whether or not they are currently defined. The expansion of @code{traceon} and @code{traceoff} is void. @end deffn Whenever a traced macro is called and the arguments have been collected, the call is displayed. If the expansion of the macro call is not void, the expansion can be displayed after the call. The output is printed to the current debug file (defaulting to standard error, @pxref{Debug Output}). @example $ @kbd{m4 -d} define(`foo', `Hello World.') @result{} define(`echo', `$@@') @result{} traceon(`foo', `echo') @result{} foo @error{}m4trace: -1- foo -> `Hello World.' @result{}Hello World. echo(`gnus', `and gnats') @error{}m4trace: -1- echo(`gnus', `and gnats') -> ``gnus',`and gnats'' @result{}gnus,and gnats @end example The number between dashes is the depth of the expansion. It is one most of the time, signifying an expansion at the outermost level, but it increases when macro arguments contain unquoted macro calls. The maximum number that will appear between dashes is controlled by the option @option{--nesting-limit} (or @option{-L}, @pxref{Limits control, , Invoking m4}). Additionally, the option @option{--trace} (or @option{-t}) can be used to invoke @code{traceon(@var{name})} before parsing input. @comment The explicit -dp neutralizes the testsuite default of -d. @comment options: -dp -L3 -tifelse @comment status: 1 @example $ @kbd{m4 -L 3 -t ifelse} ifelse(`one level') @error{}m4trace: -1- ifelse @result{} ifelse(ifelse(ifelse(`three levels'))) @error{}m4trace: -3- ifelse @error{}m4trace: -2- ifelse @error{}m4trace: -1- ifelse @result{} ifelse(ifelse(ifelse(ifelse(`four levels')))) @error{}m4:stdin:3: recursion limit of 3 exceeded, use -L to change it @end example Tracing by name is an attribute that is preserved whether the macro is defined or not. This allows the selection of macros to trace before those macros are defined. @example $ @kbd{m4 -d} traceoff(`foo') @result{} traceon(`foo') @result{} foo @result{}foo defn(`foo') @result{} define(`foo', `bar') @result{} foo @error{}m4trace: -1- foo -> `bar' @result{}bar undefine(`foo') @result{} ifdef(`foo', `yes', `no') @result{}no indir(`foo') @error{}m4:stdin:9: undefined macro `foo' @result{} define(`foo', `blah') @result{} foo @error{}m4trace: -1- foo -> `blah' @result{}blah traceoff @result{} foo @result{}blah @end example Tracing even works on builtins. However, @code{defn} (@pxref{Defn}) does not transfer tracing status. @example $ @kbd{m4 -d} traceon(`traceon') @result{} traceon(`traceoff') @error{}m4trace: -1- traceon(`traceoff') @result{} traceoff(`traceoff') @error{}m4trace: -1- traceoff(`traceoff') @result{} traceoff(`traceon') @result{} traceon(`eval', `m4_divnum') @result{} define(`m4_eval', defn(`eval')) @result{} define(`m4_divnum', defn(`divnum')) @result{} eval(divnum) @error{}m4trace: -1- eval(`0') -> `0' @result{}0 m4_eval(m4_divnum) @error{}m4trace: -2- m4_divnum -> `0' @result{}0 @end example @xref{Debug Levels}, for information on controlling the details of the display. The format of the trace output is not specified by POSIX, and varies between implementations of @code{m4}. @ignore @comment not worth including in the manual, but this tests a trace code @comment path that was temporarily broken @comment options: -de --trace ifelse @example $ @kbd{m4 -de --trace ifelse} define(`e', `ifelse(`$1', `$2', `ifelse(`$1', `$2', `e(shift($@@))')')') @result{} e(`1', `1') @error{}m4trace: -1- ifelse -> ifelse(`1', `1', `e(shift(`1',`1'))') @error{}m4trace: -1- ifelse -> e(shift(`1',`1')) @error{}m4trace: -1- ifelse @result{} @end example @end ignore @node Debug Levels @section Controlling debugging output @cindex controlling debugging output @cindex debugging output, controlling The @option{-d} option to @code{m4} (or @option{--debug}, @pxref{Debugging options, , Invoking m4}) controls the amount of details presented in three categories of output. Trace output is requested by @code{traceon} (@pxref{Trace}), and each line is prefixed by @samp{m4trace:} in relation to a macro invocation. Debug output tracks useful events not associated with a macro invocation, and each line is prefixed by @samp{m4debug:}. Finally, @code{dumpdef} (@pxref{Dumpdef}) output is affected, with no prefix added to the output lines. The @var{flags} following the option can be one or more of the following: @table @code @item a In trace output, show the actual arguments that were collected before invoking the macro. This applies to all macro calls if the @samp{t} flag is used, otherwise only the macros covered by calls of @code{traceon}. Arguments are subject to length truncation specified by the command line option @option{--arglength} (or @option{-l}). @item c In trace output, show several trace lines for each macro call. A line is shown when the macro is seen, but before the arguments are collected; a second line when the arguments have been collected and a third line after the call has completed. @item e In trace output, show the expansion of each macro call, if it is not void. This applies to all macro calls if the @samp{t} flag is used, otherwise only the macros covered by calls of @code{traceon}. The expansion is subject to length truncation specified by the command line option @option{--arglength} (or @option{-l}). @item f In debug and trace output, include the name of the current input file in the output line. @item i In debug output, print a message each time the current input file is changed. @item l In debug and trace output, include the current input line number in the output line. @item p In debug output, print a message when a named file is found through the path search mechanism (@pxref{Search Path}), giving the actual file name used. @item q In trace and dumpdef output, quote actual arguments and macro expansions in the display with the current quotes. This is useful in connection with the @samp{a} and @samp{e} flags above. @item t In trace output, trace all macro calls made in this invocation of @code{m4}, regardless of the settings of @code{traceon}. @item x In trace output, add a unique `macro call id' to each line of the trace output. This is useful in connection with the @samp{c} flag above. @item V A shorthand for all of the above flags. @end table If no flags are specified with the @option{-d} option, the default is @samp{aeq}. The examples throughout this manual assume the default flags. @cindex GNU extensions There is a builtin macro @code{debugmode}, which allows on-the-fly control of the debugging output format: @deffn Builtin debugmode (@ovar{flags}) The argument @var{flags} should be a subset of the letters listed above. As special cases, if the argument starts with a @samp{+}, the flags are added to the current debug flags, and if it starts with a @samp{-}, they are removed. If no argument is present, all debugging flags are cleared (as if no @option{-d} was given), and with an empty argument the flags are reset to the default of @samp{aeq}. The expansion of @code{debugmode} is void. @end deffn @comment The explicit -dp neutralizes the testsuite default of -d. @comment options: -dp @example $ @kbd{m4} define(`foo', `FOO') @result{} traceon(`foo') @result{} debugmode() @result{} foo @error{}m4trace: -1- foo -> `FOO' @result{}FOO debugmode @result{} foo @error{}m4trace: -1- foo @result{}FOO debugmode(`+l') @result{} foo @error{}m4trace:8: -1- foo @result{}FOO @end example The following example demonstrates the behavior of length truncation, when specified on the command line. Note that each argument and the final result are individually truncated. Also, the special tokens for builtin functions are not truncated. @comment options: -l6 @example $ @kbd{m4 -d -l 6} define(`echo', `$@@')debugmode(`+t') @result{} echo(`1', `long string') @error{}m4trace: -1- echo(`1', `long s...') -> ``1',`l...' @result{}1,long string indir(`echo', defn(`changequote')) @error{}m4trace: -2- defn(`change...') @error{}m4trace: -1- indir(`echo', ) -> ``'' @result{} @end example This example shows the effects of the debug flags that are not related to macro tracing. @comment examples @comment options: -dip @example $ @kbd{m4 -dip -I examples} @error{}m4debug: input read from stdin include(`foo')dnl @error{}m4debug: path search for `foo' found `examples/foo' @error{}m4debug: input read from examples/foo @result{}bar @error{}m4debug: input reverted to stdin, line 1 ^D @error{}m4debug: input exhausted @end example @node Debug Output @section Saving debugging output @cindex saving debugging output @cindex debugging output, saving @cindex output, saving debugging @cindex GNU extensions Debug and tracing output can be redirected to files using either the @option{--debugfile} option to @code{m4} (@pxref{Debugging options, , Invoking m4}), or with the builtin macro @code{debugfile}: @deffn Builtin debugfile (@ovar{file}) Sends all further debug and trace output to @var{file}, opened in append mode. If @var{file} is the empty string, debug and trace output are discarded. If @code{debugfile} is called without any arguments, debug and trace output are sent to standard error. This does not affect warnings, error messages, or @code{errprint} output, which are always sent to standard error. If @var{file} cannot be opened, the current debug file is unchanged, and an error is issued. The expansion of @code{debugfile} is void. @end deffn @example $ @kbd{m4 -d} traceon(`divnum') @result{} divnum(`extra') @error{}m4:stdin:2: Warning: excess arguments to builtin `divnum' ignored @error{}m4trace: -1- divnum(`extra') -> `0' @result{}0 debugfile() @result{} divnum(`extra') @error{}m4:stdin:4: Warning: excess arguments to builtin `divnum' ignored @result{}0 debugfile @result{} divnum @error{}m4trace: -1- divnum -> `0' @result{}0 @end example @node Input Control @chapter Input control This chapter describes various builtin macros for controlling the input to @code{m4}. @menu * Dnl:: Deleting whitespace in input * Changequote:: Changing the quote characters * Changecom:: Changing the comment delimiters * Changeword:: Changing the lexical structure of words * M4wrap:: Saving text until end of input @end menu @node Dnl @section Deleting whitespace in input @cindex deleting whitespace in input @cindex discarding input @cindex input, discarding The builtin @code{dnl} stands for ``Discard to Next Line'': @deffn Builtin dnl All characters, up to and including the next newline, are discarded without performing any macro expansion. A warning is issued if the end of the file is encountered without a newline. The expansion of @code{dnl} is void. @end deffn It is often used in connection with @code{define}, to remove the newline that follows the call to @code{define}. Thus @example define(`foo', `Macro `foo'.')dnl A very simple macro, indeed. foo @result{}Macro foo. @end example The input up to and including the next newline is discarded, as opposed to the way comments are treated (@pxref{Comments}). Usually, @code{dnl} is immediately followed by an end of line or some other whitespace. GNU @code{m4} will produce a warning diagnostic if @code{dnl} is followed by an open parenthesis. In this case, @code{dnl} will collect and process all arguments, looking for a matching close parenthesis. All predictable side effects resulting from this collection will take place. @code{dnl} will return no output. The input following the matching close parenthesis up to and including the next newline, on whatever line containing it, will still be discarded. @example dnl(`args are ignored, but side effects occur', define(`foo', `like this')) while this text is ignored: undefine(`foo') @error{}m4:stdin:1: Warning: excess arguments to builtin `dnl' ignored See how `foo' was defined, foo? @result{}See how foo was defined, like this? @end example If the end of file is encountered without a newline character, a warning is issued and dnl stops consuming input. @example m4wrap(`m4wrap(`2 hi ')0 hi dnl 1 hi') @result{} define(`hi', `HI') @result{} ^D @error{}m4:stdin:1: Warning: end of file treated as newline @result{}0 HI 2 HI @end example @node Changequote @section Changing the quote characters @cindex changing quote delimiters @cindex quote delimiters, changing @cindex delimiters, changing The default quote delimiters can be changed with the builtin @code{changequote}: @deffn Builtin changequote (@dvar{start, `}, @dvar{end, '}) This sets @var{start} as the new begin-quote delimiter and @var{end} as the new end-quote delimiter. If both arguments are missing, the default quotes (@code{`} and @code{'}) are used. If @var{start} is void, then quoting is disabled. Otherwise, if @var{end} is missing or void, the default end-quote delimiter (@code{'}) is used. The quote delimiters can be of any length. The expansion of @code{changequote} is void. @end deffn @example changequote(`[', `]') @result{} define([foo], [Macro [foo].]) @result{} foo @result{}Macro foo. @end example The quotation strings can safely contain eight-bit characters. @ignore @comment Yuck. I know of no clean way to render an 8-bit character in @comment both info and dvi. This example uses the `open-guillemot' and @comment `close-guillemot' characters of the Latin-1 character set. @example define(`a', `b') @result{} «a» @result{}«b» changequote(`«', `»') @result{} «a» @result{}a @end example @end ignore If no single character is appropriate, @var{start} and @var{end} can be of any length. Other implementations cap the delimiter length to five characters, but GNU has no inherent limit. @example changequote(`[[[', `]]]') @result{} define([[[foo]]], [[[Macro [[[[[foo]]]]].]]]) @result{} foo @result{}Macro [[foo]]. @end example Calling @code{changequote} with @var{start} as the empty string will effectively disable the quoting mechanism, leaving no way to quote text. However, using an empty string is not portable, as some other implementations of @code{m4} revert to the default quoting, while others preserve the prior non-empty delimiter. If @var{start} is not empty, then an empty @var{end} will use the default end-quote delimiter of @samp{'}, as otherwise, it would be impossible to end a quoted string. Again, this is not portable, as some other @code{m4} implementations reuse @var{start} as the end-quote delimiter, while others preserve the previous non-empty value. Omitting both arguments restores the default begin-quote and end-quote delimiters; fortunately this behavior is portable to all implementations of @code{m4}. @example define(`foo', `Macro `FOO'.') @result{} changequote(`', `') @result{} foo @result{}Macro `FOO'. `foo' @result{}`Macro `FOO'.' changequote(`,) @result{} foo @result{}Macro FOO. @end example There is no way in @code{m4} to quote a string containing an unmatched begin-quote, except using @code{changequote} to change the current quotes. If the quotes should be changed from, say, @samp{[} to @samp{[[}, temporary quote characters have to be defined. To achieve this, two calls of @code{changequote} must be made, one for the temporary quotes and one for the new quotes. Macros are recognized in preference to the begin-quote string, so if a prefix of @var{start} can be recognized as part of a potential macro name, the quoting mechanism is effectively disabled. Unless you use @code{changeword} (@pxref{Changeword}), this means that @var{start} should not begin with a letter, digit, or @samp{_} (underscore). However, even though quoted strings are not recognized, the quote characters can still be discerned in macro expansion and in trace output. @example define(`echo', `$@@') @result{} define(`hi', `HI') @result{} changequote(`q', `Q') @result{} q hi Q hi @result{}q HI Q HI echo(hi) @result{}qHIQ changequote @result{} changequote(`-', `EOF') @result{} - hi EOF hi @result{} hi HI changequote @result{} changequote(`1', `2') @result{} hi1hi2 @result{}hi1hi2 hi 1hi2 @result{}HI hi @end example Quotes are recognized in preference to argument collection. In particular, if @var{start} is a single @samp{(}, then argument collection is effectively disabled. For portability with other implementations, it is a good idea to avoid @samp{(}, @samp{,}, and @samp{)} as the first character in @var{start}. @example define(`echo', `$#:$@@:') @result{} define(`hi', `HI') @result{} changequote(`(',`)') @result{} echo(hi) @result{}0::hi changequote @result{} changequote(`((', `))') @result{} echo(hi) @result{}1:HI: echo((hi)) @result{}0::hi changequote @result{} changequote(`,', `)') @result{} echo(hi,hi)bye) @result{}1:HIhibye: @end example However, if you are not worried about portability, using @samp{(} and @samp{)} as quoting characters has an interesting property---you can use it to compute a quoted string containing the expansion of any quoted text, as long as the expansion results in both balanced quotes and balanced parentheses. The trick is realizing @code{expand} uses @samp{$1} unquoted, to trigger its expansion using the normal quoting characters, but uses extra parentheses to group unquoted commas that occur in the expansion without consuming whitespace following those commas. Then @code{_expand} uses @code{changequote} to convert the extra parentheses back into quoting characters. Note that it takes two more @code{changequote} invocations to restore the original quotes. Contrast the behavior on whitespace when using @samp{$*}, via @code{quote}, to attempt the same task. @example changequote(`[', `]')dnl define([a], [1, (b)])dnl define([b], [2])dnl define([quote], [[$*]])dnl define([expand], [_$0(($1))])dnl define([_expand], [changequote([(], [)])$1changequote`'changequote(`[', `]')])dnl expand([a, a, [a, a], [[a, a]]]) @result{}1, (2), 1, (2), a, a, [a, a] quote(a, a, [a, a], [[a, a]]) @result{}1,(2),1,(2),a, a,[a, a] @end example If @var{end} is a prefix of @var{start}, the end-quote will be recognized in preference to a nested begin-quote. In particular, changing the quotes to have the same string for @var{start} and @var{end} disables nesting of quotes. When quote nesting is disabled, it is impossible to double-quote strings across macro expansions, so using the same string is not done very often. @example define(`hi', `HI') @result{} changequote(`""', `"') @result{} ""hi"""hi" @result{}hihi ""hi" ""hi" @result{}hi hi ""hi"" "hi" @result{}hi" "HI" changequote @result{} `hi`hi'hi' @result{}hi`hi'hi changequote(`"', `"') @result{} "hi"hi"hi" @result{}hiHIhi @end example @ignore @comment And another stress test, not worth documenting in the manual. @example define(`aaaaaaaaaaaaaaaaaaaa', `A')define(`q', `"$@@"') @result{} changequote(`"', `"') @result{} q(q("aaaaaaaaaaaaaaaaaaaa", "a")) @result{}A,a @end example @end ignore It is an error if the end of file occurs within a quoted string. @comment status: 1 @example `hello world' @result{}hello world `dangling quote ^D @error{}m4:stdin:2: ERROR: end of file in string @end example @comment status: 1 @example ifelse(`dangling quote ^D @error{}m4:stdin:1: ERROR: end of file in string @end example @node Changecom @section Changing the comment delimiters @cindex changing comment delimiters @cindex comment delimiters, changing @cindex delimiters, changing The default comment delimiters can be changed with the builtin macro @code{changecom}: @deffn Builtin changecom (@ovar{start}, @dvar{end, @key{NL}}) This sets @var{start} as the new begin-comment delimiter and @var{end} as the new end-comment delimiter. If both arguments are missing, or @var{start} is void, then comments are disabled. Otherwise, if @var{end} is missing or void, the default end-comment delimiter of newline is used. The comment delimiters can be of any length. The expansion of @code{changecom} is void. @end deffn @example define(`comment', `COMMENT') @result{} # A normal comment @result{}# A normal comment changecom(`/*', `*/') @result{} # Not a comment anymore @result{}# Not a COMMENT anymore But: /* this is a comment now */ while this is not a comment @result{}But: /* this is a comment now */ while this is not a COMMENT @end example @cindex comments, copied to output Note how comments are copied to the output, much as if they were quoted strings. If you want the text inside a comment expanded, quote the begin-comment delimiter. Calling @code{changecom} without any arguments, or with @var{start} as the empty string, will effectively disable the commenting mechanism. To restore the original comment start of @samp{#}, you must explicitly ask for it. If @var{start} is not empty, then an empty @var{end} will use the default end-comment delimiter of newline, as otherwise, it would be impossible to end a comment. However, this is not portable, as some other @code{m4} implementations preserve the previous non-empty delimiters instead. @example define(`comment', `COMMENT') @result{} changecom @result{} # Not a comment anymore @result{}# Not a COMMENT anymore changecom(`#', `') @result{} # comment again @result{}# comment again @end example The comment strings can safely contain eight-bit characters. @ignore @comment Yuck. I know of no clean way to render an 8-bit character in @comment both info and dvi. This example uses the `open-guillemot' and @comment `close-guillemot' characters of the Latin-1 character set. @example define(`a', `b') @result{} «a» @result{}«b» changecom(`«', `»') @result{} «a» @result{}«a» @end example @end ignore If no single character is appropriate, @var{start} and @var{end} can be of any length. Other implementations cap the delimiter length to five characters, but GNU has no inherent limit. Comments are recognized in preference to macros. However, this is not compatible with other implementations, where macros and even quoting takes precedence over comments, so it may change in a future release. For portability, this means that @var{start} should not begin with a letter, digit, or @samp{_} (underscore), and that neither the start-quote nor the start-comment string should be a prefix of the other. @example define(`hi', `HI') @result{} define(`hi1hi2', `hello') @result{} changecom(`q', `Q') @result{} q hi Q hi @result{}q hi Q HI changecom(`1', `2') @result{} hi1hi2 @result{}hello hi 1hi2 @result{}HI 1hi2 @end example Comments are recognized in preference to argument collection. In particular, if @var{start} is a single @samp{(}, then argument collection is effectively disabled. For portability with other implementations, it is a good idea to avoid @samp{(}, @samp{,}, and @samp{)} as the first character in @var{start}. @example define(`echo', `$#:$*:$@@:') @result{} define(`hi', `HI') @result{} changecom(`(',`)') @result{} echo(hi) @result{}0:::(hi) changecom @result{} changecom(`((', `))') @result{} echo(hi) @result{}1:HI:HI: echo((hi)) @result{}0:::((hi)) changecom(`,', `)') @result{} echo(hi,hi)bye) @result{}1:HI,hi)bye:HI,hi)bye: changecom @result{} echo(hi,`,`'hi',hi) @result{}3:HI,,HI,HI:HI,,`'hi,HI: echo(hi,`,`'hi',hi`'changecom(`,,', `hi')) @result{}3:HI,,`'hi,HI:HI,,`'hi,HI: @end example It is an error if the end of file occurs within a comment. @comment status: 1 @example changecom(`/*', `*/') @result{} /*dangling comment ^D @error{}m4:stdin:2: ERROR: end of file in comment @end example @node Changeword @section Changing the lexical structure of words @cindex lexical structure of words @cindex words, lexical structure of @cindex syntax, changing @cindex changing syntax @cindex regular expressions @quotation The macro @code{changeword} and all associated functionality is experimental. It is only available if the @option{--enable-changeword} option was given to @command{configure}, at GNU @code{m4} installation time. The functionality will go away in the future, to be replaced by other new features that are more efficient at providing the same capabilities. @emph{Do not rely on it}. Please direct your comments about it the same way you would do for bugs. @end quotation A file being processed by @code{m4} is split into quoted strings, words (potential macro names) and simple tokens (any other single character). Initially a word is defined by the following regular expression: @comment ignore @example [_a-zA-Z][_a-zA-Z0-9]* @end example Using @code{changeword}, you can change this regular expression: @deffn {Optional builtin} changeword (@var{regex}) Changes the regular expression for recognizing macro names to be @var{regex}. If @var{regex} is empty, use @samp{[_a-zA-Z][_a-zA-Z0-9]*}. @var{regex} must obey the constraint that every prefix of the desired final pattern is also accepted by the regular expression. If @var{regex} contains grouping parentheses, the macro invoked is the portion that matched the first group, rather than the entire matching string. The expansion of @code{changeword} is void. The macro @code{changeword} is recognized only with parameters. @end deffn Relaxing the lexical rules of @code{m4} might be useful (for example) if you wanted to apply translations to a file of numbers: @example ifdef(`changeword', `', `errprint(` skipping: no changeword support ')m4exit(`77')')dnl changeword(`[_a-zA-Z0-9]+') @result{} define(`1', `0')1 @result{}0 @end example Tightening the lexical rules is less useful, because it will generally make some of the builtins unavailable. You could use it to prevent accidental call of builtins, for example: @example ifdef(`changeword', `', `errprint(` skipping: no changeword support ')m4exit(`77')')dnl define(`_indir', defn(`indir')) @result{} changeword(`_[_a-zA-Z0-9]*') @result{} esyscmd(`foo') @result{}esyscmd(foo) _indir(`esyscmd', `echo hi') @result{}hi @result{} @end example Because @code{m4} constructs its words a character at a time, there is a restriction on the regular expressions that may be passed to @code{changeword}. This is that if your regular expression accepts @samp{foo}, it must also accept @samp{f} and @samp{fo}. @example ifdef(`changeword', `', `errprint(` skipping: no changeword support ')m4exit(`77')')dnl define(`foo ', `bar ') @result{} dnl This example wants to recognize changeword, dnl, and `foo\n'. dnl First, we check that our regexp will match. regexp(`changeword', `[cd][a-z]*\|foo[ ]') @result{}0 regexp(`foo ', `[cd][a-z]*\|foo[ ]') @result{}0 regexp(`f', `[cd][a-z]*\|foo[ ]') @result{}-1 foo @result{}foo changeword(`[cd][a-z]*\|foo[ ]') @result{} dnl Even though `foo\n' matches, we forgot to allow `f'. foo @result{}foo changeword(`[cd][a-z]*\|fo*[ ]?') @result{} dnl Now we can call `foo\n'. foo @result{}bar @end example @ignore @comment One more test of including newline in a macro name; but this @comment does not need to be displayed in the manual. This ensures @comment that line numbering is correct when dnl cuts across include @comment file boundaries, and when __file__ or __line__ is the last @comment token in an include file. @example ifdef(`changeword', `', `errprint(` skipping: no changeword support ')m4exit(`77')')dnl define(`bar ', defn(`dnl'))dnl define(`baz', `dnl include(`foo') ignored dnl')dnl changeword(`\([_a-zA-Z][_a-zA-Z0-9]*\|bar \)') @result{} __file__:__line__ @result{}stdin:10 include(`foo') ignored __file__:__line__ @result{}stdin:12 baz ignored __file__:__line__ @result{}stdin:14 define(`bar ', defn(`__file__')) @result{} include(`foo') @result{}examples/foo define(`bar ', defn(`__line__')) @result{} include(`foo') @result{}1 __file__:__line__ @result{}stdin:21 @end example @end ignore @code{changeword} has another function. If the regular expression supplied contains any grouped subexpressions, then text outside the first of these is discarded before symbol lookup. So: @example ifdef(`changeword', `', `errprint(` skipping: no changeword support ')m4exit(`77')')dnl ifdef(`__unix__', , `errprint(` skipping: syscmd does not have unix semantics ')m4exit(`77')')dnl changecom(`/*', `*/')dnl define(`foo', `bar')dnl changeword(`#\([_a-zA-Z0-9]*\)') @result{} #esyscmd(`echo foo \#foo') @result{}foo bar @result{} @end example @code{m4} now requires a @samp{#} mark at the beginning of every macro invocation, so one can use @code{m4} to preprocess plain text without losing various words like @samp{divert}. In @code{m4}, macro substitution is based on text, while in @TeX{}, it is based on tokens. @code{changeword} can throw this difference into relief. For example, here is the same idea represented in @TeX{} and @code{m4}. First, the @TeX{} version: @comment ignore @example \def\a@{\message@{Hello@}@} \catcode`\@@=0 \catcode`\\=12 @@a @@bye @result{}Hello @end example @noindent Then, the @code{m4} version: @example ifdef(`changeword', `', `errprint(` skipping: no changeword support ')m4exit(`77')')dnl define(`a', `errprint(`Hello')')dnl changeword(`@@\([_a-zA-Z0-9]*\)') @result{} @@a @result{}errprint(Hello) @end example In the @TeX{} example, the first line defines a macro @code{a} to print the message @samp{Hello}. The second line defines @key{@@} to be usable instead of @key{\} as an escape character. The third line defines @key{\} to be a normal printing character, not an escape. The fourth line invokes the macro @code{a}. So, when @TeX{} is run on this file, it displays the message @samp{Hello}. When the @code{m4} example is passed through @code{m4}, it outputs @samp{errprint(Hello)}. The reason for this is that @TeX{} does lexical analysis of macro definition when the macro is @emph{defined}. @code{m4} just stores the text, postponing the lexical analysis until the macro is @emph{used}. You should note that using @code{changeword} will slow @code{m4} down by a factor of about seven, once it is changed to something other than the default regular expression. You can invoke @code{changeword} with the empty string to restore the default word definition, and regain the parsing speed. @node M4wrap @section Saving text until end of input @cindex saving input @cindex input, saving @cindex deferring expansion @cindex expansion, deferring It is possible to `save' some text until the end of the normal input has been seen. Text can be saved, to be read again by @code{m4} when the normal input has been exhausted. This feature is normally used to initiate cleanup actions before normal exit, e.g., deleting temporary files. To save input text, use the builtin @code{m4wrap}: @deffn Builtin m4wrap (@var{string}, @dots{}) Stores @var{string} in a safe place, to be reread when end of input is reached. As a GNU extension, additional arguments are concatenated with a space to the @var{string}. The expansion of @code{m4wrap} is void. The macro @code{m4wrap} is recognized only with parameters. @end deffn @example define(`cleanup', `This is the `cleanup' action. ') @result{} m4wrap(`cleanup') @result{} This is the first and last normal input line. @result{}This is the first and last normal input line. ^D @result{}This is the cleanup action. @end example The saved input is only reread when the end of normal input is seen, and not if @code{m4exit} is used to exit @code{m4}. @comment FIXME: this contradicts POSIX, which requires that "If the @comment m4wrap macro is used multiple times, the arguments specified @comment shall be processed in the order in which the m4wrap macros were @comment processed." It is safe to call @code{m4wrap} from saved text, but then the order in which the saved text is reread is undefined. If @code{m4wrap} is not used recursively, the saved pieces of text are reread in the opposite order in which they were saved (LIFO---last in, first out). However, this behavior is likely to change in a future release, to match POSIX, so you should not depend on this order. It is possible to emulate POSIX behavior even with older versions of GNU M4 by including the file @file{m4-@value{VERSION}/@/examples/@/wrapfifo.m4} from the distribution: @comment examples @example $ @kbd{m4 -I examples} undivert(`wrapfifo.m4')dnl @result{}dnl Redefine m4wrap to have FIFO semantics. @result{}define(`_m4wrap_level', `0')dnl @result{}define(`m4wrap', @result{}`ifdef(`m4wrap'_m4wrap_level, @result{} `define(`m4wrap'_m4wrap_level, @result{} defn(`m4wrap'_m4wrap_level)`$1')', @result{} `builtin(`m4wrap', `define(`_m4wrap_level', @result{} incr(_m4wrap_level))dnl @result{}m4wrap'_m4wrap_level)dnl @result{}define(`m4wrap'_m4wrap_level, `$1')')')dnl include(`wrapfifo.m4') @result{} m4wrap(`a`'m4wrap(`c ', `d')')m4wrap(`b') @result{} ^D @result{}abc @end example It is likewise possible to emulate LIFO behavior without resorting to the GNU M4 extension of @code{builtin}, by including the file @file{m4-@value{VERSION}/@/examples/@/wraplifo.m4} from the distribution. (Unfortunately, both examples shown here share some subtle bugs. See if you can find and correct them; or @pxref{Improved m4wrap, , Answers}). @comment examples @example $ @kbd{m4 -I examples} undivert(`wraplifo.m4')dnl @result{}dnl Redefine m4wrap to have LIFO semantics. @result{}define(`_m4wrap_level', `0')dnl @result{}define(`_m4wrap', defn(`m4wrap'))dnl @result{}define(`m4wrap', @result{}`ifdef(`m4wrap'_m4wrap_level, @result{} `define(`m4wrap'_m4wrap_level, @result{} `$1'defn(`m4wrap'_m4wrap_level))', @result{} `_m4wrap(`define(`_m4wrap_level', incr(_m4wrap_level))dnl @result{}m4wrap'_m4wrap_level)dnl @result{}define(`m4wrap'_m4wrap_level, `$1')')')dnl include(`wraplifo.m4') @result{} m4wrap(`a`'m4wrap(`c ', `d')')m4wrap(`b') @result{} ^D @result{}bac @end example Here is an example of implementing a factorial function using @code{m4wrap}: @example define(`f', `ifelse(`$1', `0', `Answer: 0!=1 ', eval(`$1>1'), `0', `Answer: $2$1=eval(`$2$1') ', `m4wrap(`f(decr(`$1'), `$2$1*')')')') @result{} f(`10') @result{} ^D @result{}Answer: 10*9*8*7*6*5*4*3*2*1=3628800 @end example Invocations of @code{m4wrap} at the same recursion level are concatenated and rescanned as usual: @example define(`aa', `AA ') @result{} m4wrap(`a')m4wrap(`a') @result{} ^D @result{}AA @end example @noindent however, the transition between recursion levels behaves like an end of file condition between two input files. @comment status: 1 @example m4wrap(`m4wrap(`)')len(abc') @result{} ^D @error{}m4:stdin:1: ERROR: end of file in argument list @end example @node File Inclusion @chapter File inclusion @cindex file inclusion @cindex inclusion, of files @code{m4} allows you to include named files at any point in the input. @menu * Include:: Including named files * Search Path:: Searching for include files @end menu @node Include @section Including named files There are two builtin macros in @code{m4} for including files: @deffn Builtin include (@var{file}) @deffnx Builtin sinclude (@var{file}) Both macros cause the file named @var{file} to be read by @code{m4}. When the end of the file is reached, input is resumed from the previous input file. The expansion of @code{include} and @code{sinclude} is therefore the contents of @var{file}. If @var{file} does not exist, is a directory, or cannot otherwise be read, the expansion is void, and @code{include} will fail with an error while @code{sinclude} is silent. The empty string counts as a file that does not exist. The macros @code{include} and @code{sinclude} are recognized only with parameters. @end deffn @comment status: 1 @example include(`none') @error{}m4:stdin:1: cannot open `none': No such file or directory @result{} include() @error{}m4:stdin:2: cannot open `': No such file or directory @result{} sinclude(`none') @result{} sinclude() @result{} @end example The rest of this section assumes that @code{m4} is invoked with the @option{-I} option (@pxref{Preprocessor features, , Invoking m4}) pointing to the @file{m4-@value{VERSION}/@/examples} directory shipped as part of the GNU @code{m4} package. The file @file{m4-@value{VERSION}/@/examples/@/incl.m4} in the distribution contains the lines: @comment ignore @example $ @kbd{cat examples/incl.m4} @result{}Include file start @result{}foo @result{}Include file end @end example Normally file inclusion is used to insert the contents of a file into the input stream. The contents of the file will be read by @code{m4} and macro calls in the file will be expanded: @comment examples @example $ @kbd{m4 -I examples} define(`foo', `FOO') @result{} include(`incl.m4') @result{}Include file start @result{}FOO @result{}Include file end @result{} @end example The fact that @code{include} and @code{sinclude} expand to the contents of the file can be used to define macros that operate on entire files. Here is an example, which defines @samp{bar} to expand to the contents of @file{incl.m4}: @comment examples @example $ @kbd{m4 -I examples} define(`bar', include(`incl.m4')) @result{} This is `bar': >>bar<< @result{}This is bar: >>Include file start @result{}foo @result{}Include file end @result{}<< @end example This use of @code{include} is not trivial, though, as files can contain quotes, commas, and parentheses, which can interfere with the way the @code{m4} parser works. GNU @code{m4} seamlessly concatenates the file contents with the next character, even if the included file ended in the middle of a comment, string, or macro call. These conditions are only treated as end of file errors if specified as input files on the command line. In GNU @code{m4}, an alternative method of reading files is using @code{undivert} (@pxref{Undivert}) on a named file. @ignore @comment Test that include(`file/') detects that file is not a @comment directory; we can assume that the current directory contains a @comment Makefile. mingw fails with EINVAL rather than ENOTDIR. @comment status: 1 @comment xerr: ignore @example include(`Makefile/') @error{}m4:stdin:1: cannot open `Makefile/': Not a directory @result{} @end example @comment POSIX allows, but doesn't require, failure on reading @comment directories. But since they aren't text files, it never makes @comment sense, so we globally forbid it even if fopen doesn't. mingw @comment fails with EACCES rather than EISDIR. @comment status: 1 @comment xerr: ignore @example include(`.') @error{}m4:stdin:1: cannot open `.': Is a directory @result{} @end example @comment Meanwhile, ignore errors with sinclude. @example sinclude(`Makefile/') @result{} sinclude(`.') @result{} @end example @end ignore @node Search Path @section Searching for include files @cindex search path for included files @cindex included files, search path for @cindex GNU extensions GNU @code{m4} allows included files to be found in other directories than the current working directory. @cindex @env{M4PATH} If the @option{--prepend-include} or @option{-B} command-line option was provided (@pxref{Preprocessor features, , Invoking m4}), those directories are searched first, in reverse order that those options were listed on the command line. Then @code{m4} looks in the current working directory. Next comes the directories specified with the @option{--include} or @option{-I} option, in the order found on the command line. Finally, if the @env{M4PATH} environment variable is set, it is expected to contain a colon-separated list of directories, which will be searched in order. If the automatic search for include-files causes trouble, the @samp{p} debug flag (@pxref{Debug Levels}) can help isolate the problem. @node Diversions @chapter Diverting and undiverting output @cindex deferring output Diversions are a way of temporarily saving output. The output of @code{m4} can at any time be diverted to a temporary file, and be reinserted into the output stream, @dfn{undiverted}, again at a later time. @cindex @env{TMPDIR} Numbered diversions are counted from 0 upwards, diversion number 0 being the normal output stream. GNU @code{m4} tries to keep diversions in memory. However, there is a limit to the overall memory usable by all diversions taken together (512K, currently). When this maximum is about to be exceeded, a temporary file is opened to receive the contents of the biggest diversion still in memory, freeing this memory for other diversions. When creating the temporary file, @code{m4} honors the value of the environment variable @env{TMPDIR}, and falls back to @file{/tmp}. Thus, the amount of available disk space provides the only real limit on the number and aggregate size of diversions. @ignore @comment We need to test spilled diversions, but don't need to expose @comment this highly repetitive test in the manual. @example divert(`-1')define(`f', `.') define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) divert`'dnl len(f) @result{}1048576 divert(`1') f divert(`2') f divert(`-1')undivert divert(`1')bye ^D @result{}bye @end example @comment Another test of spilled diversions. @example divert(`-1')define(`f', `.') define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) define(`f', defn(`f')defn(`f')) divert`'dnl len(f) @result{}1048576 divert(`1') f m4exit @end example @comment Catch regression in 1.4.10 with spilled diversions. @example ifdef(`__unix__', , `errprint(` skipping: syscmd does not have unix semantics ')m4exit(`77')')dnl changequote(`[', `]')dnl syscmd([echo 'divert(1)hi format(%1000000d, 1)' | ']__program__[' | sed -n 1p])dnl @result{}hi sysval @result{}0 @end example @comment Avoid quadratic copying time when transferring diversions; @comment test both in-memory and spilled to file. @comment examples @example $ @kbd{m4 -I examples} include(`forloop2.m4')dnl divert(`1')format(`%10000s', `')dnl forloop(`i', `1', `10000', `divert(incr(i))undivert(i)')dnl divert(`9001')format(`%1000000s', `')dnl forloop(`i', `9001', `10000', `divert(incr(i))undivert(i)')dnl divert(`-1')undivert @end example @end ignore Diversions make it possible to generate output in a different order than the input was read. It is possible to implement topological sorting dependencies. For example, GNU Autoconf makes use of diversions under the hood to ensure that the expansion of a prerequisite macro appears in the output prior to the expansion of a dependent macro, regardless of which order the two macros were invoked in the user's input file. @menu * Divert:: Diverting output * Undivert:: Undiverting output * Divnum:: Diversion numbers * Cleardivert:: Discarding diverted text @end menu @node Divert @section Diverting output @cindex diverting output to files @cindex output, diverting to files @cindex files, diverting output to Output is diverted using @code{divert}: @deffn Builtin divert (@dvar{number, 0}) The current diversion is changed to @var{number}. If @var{number} is left out or empty, it is assumed to be zero. If @var{number} cannot be parsed, the diversion is unchanged. The expansion of @code{divert} is void. @end deffn When all the @code{m4} input will have been processed, all existing diversions are automatically undiverted, in numerical order. @example divert(`1') This text is diverted. divert @result{} This text is not diverted. @result{}This text is not diverted. ^D @result{} @result{}This text is diverted. @end example Several calls of @code{divert} with the same argument do not overwrite the previous diverted text, but append to it. Diversions are printed after any wrapped text is expanded. @example define(`text', `TEXT') @result{} divert(`1')`diverted text.' divert @result{} m4wrap(`Wrapped text precedes ') @result{} ^D @result{}Wrapped TEXT precedes diverted text. @end example @cindex discarding input @cindex input, discarding If output is diverted to a negative diversion, it is simply discarded. This can be used to suppress unwanted output. A common example of unwanted output is the trailing newlines after macro definitions. Here is a common programming idiom in @code{m4} for avoiding them. @example divert(`-1') define(`foo', `Macro `foo'.') define(`bar', `Macro `bar'.') divert @result{} @end example @cindex GNU extensions Traditional implementations only supported ten diversions. But as a GNU extension, diversion numbers can be as large as positive integers will allow, rather than treating a multi-digit diversion number as a request to discard text. @example divert(eval(`1<<28'))world divert(`2')hello ^D @result{}hello @result{}world @end example Note that @code{divert} is an English word, but also an active macro without arguments. When processing plain text, the word might appear in normal text and be unintentionally swallowed as a macro invocation. One way to avoid this is to use the @option{-P} option to rename all builtins (@pxref{Operation modes, , Invoking m4}). Another is to write a wrapper that requires a parameter to be recognized. @example We decided to divert the stream for irrigation. @result{}We decided to the stream for irrigation. define(`divert', `ifelse(`$#', `0', ``$0'', `builtin(`$0', $@@)')') @result{} divert(`-1') Ignored text. divert(`0') @result{} We decided to divert the stream for irrigation. @result{}We decided to divert the stream for irrigation. @end example @node Undivert @section Undiverting output Diverted text can be undiverted explicitly using the builtin @code{undivert}: @deffn Builtin undivert (@ovar{diversions@dots{}}) Undiverts the numeric @var{diversions} given by the arguments, in the order given. If no arguments are supplied, all diversions are undiverted, in numerical order. @cindex file inclusion @cindex inclusion, of files @cindex GNU extensions As a GNU extension, @var{diversions} may contain non-numeric strings, which are treated as the names of files to copy into the output without expansion. A warning is issued if a file could not be opened. The expansion of @code{undivert} is void. @end deffn @example divert(`1') This text is diverted. divert @result{} This text is not diverted. @result{}This text is not diverted. undivert(`1') @result{} @result{}This text is diverted. @result{} @end example Notice the last two blank lines. One of them comes from the newline following @code{undivert}, the other from the newline that followed the @code{divert}! A diversion often starts with a blank line like this. When diverted text is undiverted, it is @emph{not} reread by @code{m4}, but rather copied directly to the current output, and it is therefore not an error to undivert into a diversion. Undiverting the empty string is the same as specifying diversion 0; in either case nothing happens since the output has already been flushed. @example divert(`1')diverted text divert @result{} undivert() @result{} undivert(`0') @result{} undivert @result{}diverted text @result{} divert(`1')more divert(`2')undivert(`1')diverted text`'divert @result{} undivert(`1') @result{} undivert(`2') @result{}more @result{}diverted text @end example When a diversion has been undiverted, the diverted text is discarded, and it is not possible to bring back diverted text more than once. @example divert(`1') This text is diverted first. divert(`0')undivert(`1')dnl @result{} @result{}This text is diverted first. undivert(`1') @result{} divert(`1') This text is also diverted but not appended. divert(`0')undivert(`1')dnl @result{} @result{}This text is also diverted but not appended. @end example Attempts to undivert the current diversion are silently ignored. Thus, when the current diversion is not 0, the current diversion does not get rearranged among the other diversions. @example divert(`1')one divert(`2')two divert(`3')three divert(`2')undivert`'dnl divert`'undivert`'dnl @result{}two @result{}one @result{}three @end example @cindex GNU extensions @cindex file inclusion @cindex inclusion, of files GNU @code{m4} allows named files to be undiverted. Given a non-numeric argument, the contents of the file named will be copied, uninterpreted, to the current output. This complements the builtin @code{include} (@pxref{Include}). To illustrate the difference, assume the file @file{foo} contains: @comment ignore @example $ @kbd{cat foo} bar @end example @noindent then @example define(`bar', `BAR') @result{} undivert(`foo') @result{}bar @result{} include(`foo') @result{}BAR @result{} @end example If the file is not found (or cannot be read), an error message is issued, and the expansion is void. It is possible to intermix files and diversion numbers. @example divert(`1')diversion one divert(`2')undivert(`foo')dnl divert(`3')diversion three divert`'dnl undivert(`1', `2', `foo', `3')dnl @result{}diversion one @result{}bar @result{}bar @result{}diversion three @end example @node Divnum @section Diversion numbers @cindex diversion numbers The current diversion is tracked by the builtin @code{divnum}: @deffn Builtin divnum Expands to the number of the current diversion. @end deffn @example Initial divnum @result{}Initial 0 divert(`1') Diversion one: divnum divert(`2') Diversion two: divnum ^D @result{} @result{}Diversion one: 1 @result{} @result{}Diversion two: 2 @end example @node Cleardivert @section Discarding diverted text @cindex discarding diverted text @cindex diverted text, discarding Often it is not known, when output is diverted, whether the diverted text is actually needed. Since all non-empty diversion are brought back on the main output stream when the end of input is seen, a method of discarding a diversion is needed. If all diversions should be discarded, the easiest is to end the input to @code{m4} with @samp{divert(`-1')} followed by an explicit @samp{undivert}: @example divert(`1') Diversion one: divnum divert(`2') Diversion two: divnum divert(`-1') undivert ^D @end example @noindent No output is produced at all. Clearing selected diversions can be done with the following macro: @deffn Composite cleardivert (@ovar{diversions@dots{}}) Discard the contents of each of the listed numeric @var{diversions}. @end deffn @example define(`cleardivert', `pushdef(`_n', divnum)divert(`-1')undivert($@@)divert(_n)popdef(`_n')') @result{} @end example It is called just like @code{undivert}, but the effect is to clear the diversions, given by the arguments. (This macro has a nasty bug! You should try to see if you can find it and correct it; or @pxref{Improved cleardivert, , Answers}). @node Text handling @chapter Macros for text handling There are a number of builtins in @code{m4} for manipulating text in various ways, extracting substrings, searching, substituting, and so on. @menu * Len:: Calculating length of strings * Index macro:: Searching for substrings * Regexp:: Searching for regular expressions * Substr:: Extracting substrings * Translit:: Translating characters * Patsubst:: Substituting text by regular expression * Format:: Formatting strings (printf-like) @end menu @node Len @section Calculating length of strings @cindex length of strings @cindex strings, length of The length of a string can be calculated by @code{len}: @deffn Builtin len (@var{string}) Expands to the length of @var{string}, as a decimal number. The macro @code{len} is recognized only with parameters. @end deffn @example len() @result{}0 len(`abcdef') @result{}6 @end example @node Index macro @section Searching for substrings @cindex substrings, locating Searching for substrings is done with @code{index}: @deffn Builtin index (@var{string}, @var{substring}) Expands to the index of the first occurrence of @var{substring} in @var{string}. The first character in @var{string} has index 0. If @var{substring} does not occur in @var{string}, @code{index} expands to @samp{-1}. The macro @code{index} is recognized only with parameters. @end deffn @example index(`gnus, gnats, and armadillos', `nat') @result{}7 index(`gnus, gnats, and armadillos', `dag') @result{}-1 @end example Omitting @var{substring} evokes a warning, but still produces output; contrast this with an empty @var{substring}. @example index(`abc') @error{}m4:stdin:1: Warning: too few arguments to builtin `index' @result{}0 index(`abc', `') @result{}0 index(`abc', `b') @result{}1 @end example @ignore @comment Expose a bug in the strstr() algorithm present in glibc @comment 2.9 through 2.12 and in gnulib up to Sep 2010. @example index(`;:11-:12-:12-:12-:12-:12-:12-:12-:12.:12.:12.:12.:12.:12.:12.:12.:12-', `:12-:12-:12-:12-:12-:12-:12-:12-') @result{}-1 @end example @comment Expose a bug in the gnulib replacement strstr() algorithm @comment present from Jun 2010 to Feb 2011, including m4 1.4.15. @example index(`..wi.d.', `.d.') @result{}4 @end example @end ignore @node Regexp @section Searching for regular expressions @cindex basic regular expressions @cindex regular expressions @cindex expressions, regular @cindex GNU extensions Searching for regular expressions is done with the builtin @code{regexp}: @deffn Builtin regexp (@var{string}, @var{regexp}, @ovar{replacement}) Searches for @var{regexp} in @var{string}. The syntax for regular expressions is the same as in GNU Emacs, which is similar to BRE, Basic Regular Expressions in POSIX. @ifnothtml @xref{Regexps, , Syntax of Regular Expressions, emacs, The GNU Emacs Manual}. @end ifnothtml @ifhtml See @uref{http://www.gnu.org/@/software/@/emacs/@/manual/@/emacs.html#Regexps, Syntax of Regular Expressions} in the GNU Emacs Manual. @end ifhtml Support for ERE, Extended Regular Expressions is not available, but will be added in GNU M4 2.0. If @var{replacement} is omitted, @code{regexp} expands to the index of the first match of @var{regexp} in @var{string}. If @var{regexp} does not match anywhere in @var{string}, it expands to -1. If @var{replacement} is supplied, and there was a match, @code{regexp} changes the expansion to this argument, with @samp{\@var{n}} substituted by the text matched by the @var{n}th parenthesized sub-expression of @var{regexp}, up to nine sub-expressions. The escape @samp{\&} is replaced by the text of the entire regular expression matched. For all other characters, @samp{\} treats the next character literally. A warning is issued if there were fewer sub-expressions than the @samp{\@var{n}} requested, or if there is a trailing @samp{\}. If there was no match, @code{regexp} expands to the empty string. The macro @code{regexp} is recognized only with parameters. @end deffn @example regexp(`GNUs not Unix', `\<[a-z]\w+') @result{}5 regexp(`GNUs not Unix', `\a-c-a') @result{}<;>abcba translit(`abcdef', `aabdef', `bcged') @result{}bgced @end example In the @sc{ascii} encoding, the first example deletes all uppercase letters, the second converts lowercase to uppercase, and the third `mirrors' all uppercase letters, while converting them to lowercase. The two first cases are by far the most common, even though they are not portable to @sc{ebcdic} or other encodings. The fourth example shows a range ending in @samp{-}, as well as back-to-back ranges. The final example shows that @samp{a} is mapped to @samp{b}, not @samp{c}; the resulting @samp{b} is not further remapped to @samp{g}; the @samp{d} and @samp{e} are swapped, and the @samp{f} is discarded. @ignore @comment No need to fight 8-bit characters, as it is difficult to get @comment rendering right in both info and dvi. @example translit(`«abc~', `~-»') @result{}abc @end example @comment Stress test short arguments, since they use a different code @comment path. @example translit(`abcdeabcde', `a') @result{}bcdebcde translit(`abcdeabcde', `ab') @result{}cdecde translit(`abcdeabcde', `a', `f') @result{}fbcdefbcde translit(`abcdeabcde', `a', `f') @result{}fbcdefbcde translit(`abcdeabcde', `a', `fg') @result{}fbcdefbcde translit(`abcdeabcde', `ab', `f') @result{}fcdefcde translit(`abcdeabcde', `ab', `fg') @result{}fgcdefgcde translit(`abcdeabcde', `ab', `ba') @result{}bacdebacde translit(`abcdeabcde', `e', `f') @result{}abcdfabcdf translit(`abc', `', `cde') @result{}abc translit(`', `a', `bc') @result{} @end example @end ignore Omitting @var{chars} evokes a warning, but still produces output. @example translit(`abc') @error{}m4:stdin:1: Warning: too few arguments to builtin `translit' @result{}abc @end example @node Patsubst @section Substituting text by regular expression @cindex basic regular expressions @cindex regular expressions @cindex expressions, regular @cindex pattern substitution @cindex substitution by regular expression @cindex GNU extensions Global substitution in a string is done by @code{patsubst}: @deffn Builtin patsubst (@var{string}, @var{regexp}, @ovar{replacement}) Searches @var{string} for matches of @var{regexp}, and substitutes @var{replacement} for each match. The syntax for regular expressions is the same as in GNU Emacs (@pxref{Regexp}). The parts of @var{string} that are not covered by any match of @var{regexp} are copied to the expansion. Whenever a match is found, the search proceeds from the end of the match, so a character from @var{string} will never be substituted twice. If @var{regexp} matches a string of zero length, the start position for the search is incremented, to avoid infinite loops. When a replacement is to be made, @var{replacement} is inserted into the expansion, with @samp{\@var{n}} substituted by the text matched by the @var{n}th parenthesized sub-expression of @var{patsubst}, for up to nine sub-expressions. The escape @samp{\&} is replaced by the text of the entire regular expression matched. For all other characters, @samp{\} treats the next character literally. A warning is issued if there were fewer sub-expressions than the @samp{\@var{n}} requested, or if there is a trailing @samp{\}. The @var{replacement} argument can be omitted, in which case the text matched by @var{regexp} is deleted. The macro @code{patsubst} is recognized only with parameters. @end deffn @example patsubst(`GNUs not Unix', `^', `OBS: ') @result{}OBS: GNUs not Unix patsubst(`GNUs not Unix', `\<', `OBS: ') @result{}OBS: GNUs OBS: not OBS: Unix patsubst(`GNUs not Unix', `\w*', `(\&)') @result{}(GNUs)() (not)() (Unix)() patsubst(`GNUs not Unix', `\w+', `(\&)') @result{}(GNUs) (not) (Unix) patsubst(`GNUs not Unix', `[A-Z][a-z]+') @result{}GN not@w{ } patsubst(`GNUs not Unix', `not', `NOT\') @error{}m4:stdin:6: Warning: trailing \ ignored in replacement @result{}GNUs NOT Unix @end example Here is a slightly more realistic example, which capitalizes individual words or whole sentences, by substituting calls of the macros @code{upcase} and @code{downcase} into the strings. @deffn Composite upcase (@var{text}) @deffnx Composite downcase (@var{text}) @deffnx Composite capitalize (@var{text}) Expand to @var{text}, but with capitalization changed: @code{upcase} changes all letters to upper case, @code{downcase} changes all letters to lower case, and @code{capitalize} changes the first character of each word to upper case and the remaining characters to lower case. @end deffn First, an example of their usage, using implementations distributed in @file{m4-@value{VERSION}/@/examples/@/capitalize.m4}. @comment examples @example $ @kbd{m4 -I examples} include(`capitalize.m4') @result{} upcase(`GNUs not Unix') @result{}GNUS NOT UNIX downcase(`GNUs not Unix') @result{}gnus not unix capitalize(`GNUs not Unix') @result{}Gnus Not Unix @end example Now for the implementation. There is a helper macro @code{_capitalize} which puts only its first word in mixed case. Then @code{capitalize} merely parses out the words, and replaces them with an invocation of @code{_capitalize}. (As presented here, the @code{capitalize} macro has some subtle flaws. You should try to see if you can find and correct them; or @pxref{Improved capitalize, , Answers}). @comment examples @example $ @kbd{m4 -I examples} undivert(`capitalize.m4')dnl @result{}divert(`-1') @result{}# upcase(text) @result{}# downcase(text) @result{}# capitalize(text) @result{}# change case of text, simple version @result{}define(`upcase', `translit(`$*', `a-z', `A-Z')') @result{}define(`downcase', `translit(`$*', `A-Z', `a-z')') @result{}define(`_capitalize', @result{} `regexp(`$1', `^\(\w\)\(\w*\)', @result{} `upcase(`\1')`'downcase(`\2')')') @result{}define(`capitalize', `patsubst(`$1', `\w+', `_$0(`\&')')') @result{}divert`'dnl @end example While @code{regexp} replaces the whole input with the replacement as soon as there is a match, @code{patsubst} replaces each @emph{occurrence} of a match and preserves non-matching pieces: @example define(`patreg', `patsubst($@@) regexp($@@)')dnl patreg(`bar foo baz Foo', `foo\|Foo', `FOO') @result{}bar FOO baz FOO @result{}FOO patreg(`aba abb 121', `\(.\)\(.\)\1', `\2\1\2') @result{}bab abb 212 @result{}bab @end example Omitting @var{regexp} evokes a warning, but still produces output; contrast this with an empty @var{regexp} argument. @example patsubst(`abc') @error{}m4:stdin:1: Warning: too few arguments to builtin `patsubst' @result{}abc patsubst(`abc', `') @result{}abc patsubst(`abc', `', `\\-') @result{}\-a\-b\-c\- @end example @node Format @section Formatting strings (printf-like) @cindex formatted output @cindex output, formatted @cindex GNU extensions Formatted output can be made with @code{format}: @deffn Builtin format (@var{format-string}, @dots{}) Works much like the C function @code{printf}. The first argument @var{format-string} can contain @samp{%} specifications which are satisfied by additional arguments, and the expansion of @code{format} is the formatted string. The macro @code{format} is recognized only with parameters. @end deffn Its use is best described by a few examples: @comment This test is a bit fragile, if someone tries to port to a @comment platform without infinity. @example define(`foo', `The brown fox jumped over the lazy dog') @result{} format(`The string "%s" uses %d characters', foo, len(foo)) @result{}The string "The brown fox jumped over the lazy dog" uses 38 characters format(`%*.*d', `-1', `-1', `1') @result{}1 format(`%.0f', `56789.9876') @result{}56790 len(format(`%-*X', `5000', `1')) @result{}5000 ifelse(format(`%010F', `infinity'), ` INF', `success', format(`%010F', `infinity'), ` INFINITY', `success', format(`%010F', `infinity')) @result{}success ifelse(format(`%.1A', `1.999'), `0X1.0P+1', `success', format(`%.1A', `1.999'), `0X2.0P+0', `success', format(`%.1A', `1.999')) @result{}success format(`%g', `0xa.P+1') @result{}20 @end example Using the @code{forloop} macro defined earlier (@pxref{Forloop}), this example shows how @code{format} can be used to produce tabular output. @comment examples @example $ @kbd{m4 -I examples} include(`forloop.m4') @result{} forloop(`i', `1', `10', `format(`%6d squared is %10d ', i, eval(i**2))') @result{} 1 squared is 1 @result{} 2 squared is 4 @result{} 3 squared is 9 @result{} 4 squared is 16 @result{} 5 squared is 25 @result{} 6 squared is 36 @result{} 7 squared is 49 @result{} 8 squared is 64 @result{} 9 squared is 81 @result{} 10 squared is 100 @result{} @end example The builtin @code{format} is modeled after the ANSI C @samp{printf} function, and supports these @samp{%} specifiers: @samp{c}, @samp{s}, @samp{d}, @samp{o}, @samp{x}, @samp{X}, @samp{u}, @samp{a}, @samp{A}, @samp{e}, @samp{E}, @samp{f}, @samp{F}, @samp{g}, @samp{G}, and @samp{%}; it supports field widths and precisions, and the flags @samp{+}, @samp{-}, @samp{ }, @samp{0}, @samp{#}, and @samp{'}. For integer specifiers, the width modifiers @samp{hh}, @samp{h}, and @samp{l} are recognized, and for floating point specifiers, the width modifier @samp{l} is recognized. Items not yet supported include positional arguments, the @samp{n}, @samp{p}, @samp{S}, and @samp{C} specifiers, the @samp{z}, @samp{t}, @samp{j}, @samp{L} and @samp{ll} modifiers, and any platform extensions available in the native @code{printf}. For more details on the functioning of @code{printf}, see the C Library Manual, or the POSIX specification (for example, @samp{%a} is supported even on platforms that haven't yet implemented C99 hexadecimal floating point output natively). Unrecognized specifiers result in a warning. It is anticipated that a future release of GNU @code{m4} will support more specifiers, and give better warnings when various problems such as overflow are encountered. Likewise, escape sequences are not yet recognized. @example format(`%p', `0') @error{}m4:stdin:1: Warning: unrecognized specifier in `%p' @result{} @end example @ignore @comment Expose a crash with a bad format string fixed in 1.4.15. @comment Unfortuntely, 8-bit bytes are hard to check for; but the @comment exit status is enough to sniff the crash in broken versions. @comment xerr: ignore @example format(`%'format(`%c', `128')) @result{} @end example @end ignore @node Arithmetic @chapter Macros for doing arithmetic @cindex arithmetic @cindex integer arithmetic Integer arithmetic is included in @code{m4}, with a C-like syntax. As convenient shorthands, there are builtins for simple increment and decrement operations. @menu * Incr:: Decrement and increment operators * Eval:: Evaluating integer expressions @end menu @node Incr @section Decrement and increment operators @cindex decrement operator @cindex increment operator Increment and decrement of integers are supported using the builtins @code{incr} and @code{decr}: @deffn Builtin incr (@var{number}) @deffnx Builtin decr (@var{number}) Expand to the numerical value of @var{number}, incremented or decremented, respectively, by one. Except for the empty string, the expansion is empty if @var{number} could not be parsed. The macros @code{incr} and @code{decr} are recognized only with parameters. @end deffn @example incr(`4') @result{}5 decr(`7') @result{}6 incr() @error{}m4:stdin:3: empty string treated as 0 in builtin `incr' @result{}1 decr() @error{}m4:stdin:4: empty string treated as 0 in builtin `decr' @result{}-1 @end example @node Eval @section Evaluating integer expressions @cindex integer expression evaluation @cindex evaluation, of integer expressions @cindex expressions, evaluation of integer Integer expressions are evaluated with @code{eval}: @deffn Builtin eval (@var{expression}, @dvar{radix, 10}, @ovar{width}) Expands to the value of @var{expression}. The expansion is empty if a problem is encountered while parsing the arguments. If specified, @var{radix} and @var{width} control the format of the output. Calculations are done with 32-bit signed numbers. Overflow silently results in wraparound. A warning is issued if division by zero is attempted, or if @var{expression} could not be parsed. Expressions can contain the following operators, listed in order of decreasing precedence. @table @samp @item () Parentheses @item + - ~ ! Unary plus and minus, and bitwise and logical negation @item ** Exponentiation @item * / % Multiplication, division, and modulo @item + - Addition and subtraction @item << >> Shift left or right @item > >= < <= Relational operators @item == != Equality operators @item & Bitwise and @item ^ Bitwise exclusive-or @item | Bitwise or @item && Logical and @item || Logical or @end table The macro @code{eval} is recognized only with parameters. @end deffn All binary operators, except exponentiation, are left associative. C operators that perform variable assignment, such as @samp{+=} or @samp{--}, are not implemented, since @code{eval} only operates on constants, not variables. Attempting to use them results in an error. However, since traditional implementations treated @samp{=} as an undocumented alias for @samp{==} as opposed to an assignment operator, this usage is supported as a special case. Be aware that a future version of GNU M4 may support assignment semantics as an extension when POSIX mode is not requested, and that using @samp{=} to check equality is not portable. @comment status: 1 @example eval(`2 = 2') @error{}m4:stdin:1: Warning: recommend ==, not =, for equality operator @result{}1 eval(`++0') @error{}m4:stdin:2: invalid operator in eval: ++0 @result{} eval(`0 |= 1') @error{}m4:stdin:3: invalid operator in eval: 0 |= 1 @result{} @end example Note that some older @code{m4} implementations use @samp{^} as an alternate operator for the exponentiation, although POSIX requires the C behavior of bitwise exclusive-or. The precedence of the negation operators, @samp{~} and @samp{!}, was traditionally lower than equality. The unary operators could not be used reliably more than once on the same term without intervening parentheses. The traditional precedence of the equality operators @samp{==} and @samp{!=} was identical instead of lower than the relational operators such as @samp{<}, even through GNU M4 1.4.8. Starting with version 1.4.9, GNU M4 correctly follows POSIX precedence rules. M4 scripts designed to be portable between releases must be aware that parentheses may be required to enforce C precedence rules. Likewise, division by zero, even in the unused branch of a short-circuiting operator, is not always well-defined in other implementations. Following are some examples where the current version of M4 follows C precedence rules, but where older versions and some other implementations of @code{m4} require explicit parentheses to get the correct result: @example eval(`1 == 2 > 0') @result{}1 eval(`(1 == 2) > 0') @result{}0 eval(`! 0 * 2') @result{}2 eval(`! (0 * 2)') @result{}1 eval(`1 | 1 ^ 1') @result{}1 eval(`(1 | 1) ^ 1') @result{}0 eval(`+ + - ~ ! ~ 0') @result{}1 eval(`2 || 1 / 0') @result{}1 eval(`0 || 1 / 0') @error{}m4:stdin:9: divide by zero in eval: 0 || 1 / 0 @result{} eval(`0 && 1 % 0') @result{}0 eval(`2 && 1 % 0') @error{}m4:stdin:11: modulo by zero in eval: 2 && 1 % 0 @result{} @end example @cindex GNU extensions As a GNU extension, the operator @samp{**} performs integral exponentiation. The operator is right-associative, and if evaluated, the exponent must be non-negative, and at least one of the arguments must be non-zero, or a warning is issued. @example eval(`2 ** 3 ** 2') @result{}512 eval(`(2 ** 3) ** 2') @result{}64 eval(`0 ** 1') @result{}0 eval(`2 ** 0') @result{}1 eval(`0 ** 0') @result{} @error{}m4:stdin:5: divide by zero in eval: 0 ** 0 eval(`4 ** -2') @error{}m4:stdin:6: negative exponent in eval: 4 ** -2 @result{} @end example Within @var{expression}, (but not @var{radix} or @var{width}), numbers without a special prefix are decimal. A simple @samp{0} prefix introduces an octal number. @samp{0x} introduces a hexadecimal number. As GNU extensions, @samp{0b} introduces a binary number. @samp{0r} introduces a number expressed in any radix between 1 and 36: the prefix should be immediately followed by the decimal expression of the radix, a colon, then the digits making the number. For radix 1, leading zeros are ignored, and all remaining digits must be @samp{1}; for all other radices, the digits are @samp{0}, @samp{1}, @samp{2}, @dots{}. Beyond @samp{9}, the digits are @samp{a}, @samp{b} @dots{} up to @samp{z}. Lower and upper case letters can be used interchangeably in numbers prefixes and as number digits. Parentheses may be used to group subexpressions whenever needed. For the relational operators, a true relation returns @code{1}, and a false relation return @code{0}. Here are a few examples of use of @code{eval}. @example eval(`-3 * 5') @result{}-15 eval(`-99 / 10') @result{}-9 eval(`-99 % 10') @result{}-9 eval(`99 % -10') @result{}9 eval(index(`Hello world', `llo') >= 0) @result{}1 eval(`0r1:0111 + 0b100 + 0r3:12') @result{}12 define(`square', `eval(`($1) ** 2')') @result{} square(`9') @result{}81 square(square(`5')` + 1') @result{}676 define(`foo', `666') @result{} eval(`foo / 6') @error{}m4:stdin:11: bad expression in eval: foo / 6 @result{} eval(foo / 6) @result{}111 @end example As the last two lines show, @code{eval} does not handle macro names, even if they expand to a valid expression (or part of a valid expression). Therefore all macros must be expanded before they are passed to @code{eval}. Some calculations are not portable to other implementations, since they have undefined semantics in C, but GNU @code{m4} has well-defined behavior on overflow. When shifting, an out-of-range shift amount is implicitly brought into the range of 32-bit signed integers using an implicit bit-wise and with 0x1f). @example define(`max_int', eval(`0x7fffffff')) @result{} define(`min_int', incr(max_int)) @result{} eval(min_int` < 0') @result{}1 eval(max_int` > 0') @result{}1 ifelse(eval(min_int` / -1'), min_int, `overflow occurred') @result{}overflow occurred min_int @result{}-2147483648 eval(`0x80000000 % -1') @result{}0 eval(`-4 >> 1') @result{}-2 eval(`-4 >> 33') @result{}-2 @end example If @var{radix} is specified, it specifies the radix to be used in the expansion. The default radix is 10; this is also the case if @var{radix} is the empty string. A warning results if the radix is outside the range of 1 through 36, inclusive. The result of @code{eval} is always taken to be signed. No radix prefix is output, and for radices greater than 10, the digits are lower case. The @var{width} argument specifies the minimum output width, excluding any negative sign. The result is zero-padded to extend the expansion to the requested width. A warning results if the width is negative. If @var{radix} or @var{width} is out of bounds, the expansion of @code{eval} is empty. @example eval(`666', `10') @result{}666 eval(`666', `11') @result{}556 eval(`666', `6') @result{}3030 eval(`666', `6', `10') @result{}0000003030 eval(`-666', `6', `10') @result{}-0000003030 eval(`10', `', `0') @result{}10 `0r1:'eval(`10', `1', `11') @result{}0r1:01111111111 eval(`10', `16') @result{}a eval(`1', `37') @error{}m4:stdin:9: radix 37 in builtin `eval' out of range @result{} eval(`1', , `-1') @error{}m4:stdin:10: negative width to builtin `eval' @result{} eval() @error{}m4:stdin:11: empty string treated as 0 in builtin `eval' @result{}0 @end example @node Shell commands @chapter Macros for running shell commands @cindex UNIX commands, running @cindex executing shell commands @cindex running shell commands @cindex shell commands, running @cindex commands, running shell There are a few builtin macros in @code{m4} that allow you to run shell commands from within @code{m4}. Note that the definition of a valid shell command is system dependent. On UNIX systems, this is the typical @command{/bin/sh}. But on other systems, such as native Windows, the shell has a different syntax of commands that it understands. Some examples in this chapter assume @command{/bin/sh}, and also demonstrate how to quit early with a known exit value if this is not the case. @menu * Platform macros:: Determining the platform * Syscmd:: Executing simple commands * Esyscmd:: Reading the output of commands * Sysval:: Exit status * Mkstemp:: Making temporary files @end menu @node Platform macros @section Determining the platform @cindex platform macros Sometimes it is desirable for an input file to know which platform @code{m4} is running on. GNU @code{m4} provides several macros that are predefined to expand to the empty string; checking for their existence will confirm platform details. @deffn {Optional builtin} __gnu__ @deffnx {Optional builtin} __os2__ @deffnx {Optional builtin} os2 @deffnx {Optional builtin} __unix__ @deffnx {Optional builtin} unix @deffnx {Optional builtin} __windows__ @deffnx {Optional builtin} windows Each of these macros is conditionally defined as needed to describe the environment of @code{m4}. If defined, each macro expands to the empty string. For now, these macros silently ignore all arguments, but in a future release of M4, they might warn if arguments are present. @end deffn When GNU extensions are in effect (that is, when you did not use the @option{-G} option, @pxref{Limits control, , Invoking m4}), GNU @code{m4} will define the macro @code{@w{__gnu__}} to expand to the empty string. @example $ @kbd{m4} __gnu__ @result{} __gnu__(`ignored') @result{} Extensions are ifdef(`__gnu__', `active', `inactive') @result{}Extensions are active @end example @comment options: -G @example $ @kbd{m4 -G} __gnu__ @result{}__gnu__ __gnu__(`ignored') @result{}__gnu__(ignored) Extensions are ifdef(`__gnu__', `active', `inactive') @result{}Extensions are inactive @end example On UNIX systems, GNU @code{m4} will define @code{@w{__unix__}} by default, or @code{unix} when the @option{-G} option is specified. On native Windows systems, GNU @code{m4} will define @code{@w{__windows__}} by default, or @code{windows} when the @option{-G} option is specified. On OS/2 systems, GNU @code{m4} will define @code{@w{__os2__}} by default, or @code{os2} when the @option{-G} option is specified. If GNU @code{m4} does not provide a platform macro for your system, please report that as a bug. @example define(`provided', `0') @result{} ifdef(`__unix__', `define(`provided', incr(provided))') @result{} ifdef(`__windows__', `define(`provided', incr(provided))') @result{} ifdef(`__os2__', `define(`provided', incr(provided))') @result{} provided @result{}1 @end example @node Syscmd @section Executing simple commands Any shell command can be executed, using @code{syscmd}: @deffn Builtin syscmd (@var{shell-command}) Executes @var{shell-command} as a shell command. The expansion of @code{syscmd} is void, @emph{not} the output from @var{shell-command}! Output or error messages from @var{shell-command} are not read by @code{m4}. @xref{Esyscmd}, if you need to process the command output. Prior to executing the command, @code{m4} flushes its buffers. The default standard input, output and error of @var{shell-command} are the same as those of @code{m4}. By default, the @var{shell-command} will be used as the argument to the @option{-c} option of the @command{/bin/sh} shell (or the version of @command{sh} specified by @samp{command -p getconf PATH}, if your system supports that). If you prefer a different shell, the @command{configure} script can be given the option @option{--with-syscmd-shell=@var{location}} to set the location of an alternative shell at GNU @code{m4} installation; the alternative shell must still support @option{-c}. The macro @code{syscmd} is recognized only with parameters. @end deffn @example define(`foo', `FOO') @result{} syscmd(`echo foo') @result{}foo @result{} @end example Note how the expansion of @code{syscmd} keeps the trailing newline of the command, as well as using the newline that appeared after the macro. The following is an example of @var{shell-command} using the same standard input as @code{m4}: @comment ignore @example $ @kbd{echo "m4wrap(\`syscmd(\`cat')')" | m4} @result{} @end example @ignore @comment If the user types the example below with stdin being an @comment interactive terminal, then cat will hang waiting for additional @comment input after m4 has exited. But the testsuite is using a pipe @comment for stdin. Hence, we have two versions - the one we feed the @comment testsuite below, and the one we display to the user above that @comment more accurately shows what the testsuite is really doing but @comment which the testsuite cannot parse. @example m4wrap(`syscmd(`cat')') @result{} ^D @end example @end ignore It tells @code{m4} to read all of its input before executing the wrapped text, then hand a valid (albeit emptied) pipe as standard input for the @code{cat} subcommand. Therefore, you should be careful when using standard input (either by specifying no files, or by passing @samp{-} as a file name on the command line, @pxref{Command line files, , Invoking m4}), and also invoking subcommands via @code{syscmd} or @code{esyscmd} that consume data from standard input. When standard input is a seekable file, the subprocess will pick up with the next character not yet processed by @code{m4}; when it is a pipe or other non-seekable file, there is no guarantee how much data will already be buffered by @code{m4} and thus unavailable to the child. @node Esyscmd @section Reading the output of commands @cindex GNU extensions If you want @code{m4} to read the output of a shell command, use @code{esyscmd}: @deffn Builtin esyscmd (@var{shell-command}) Expands to the standard output of the shell command @var{shell-command}. Prior to executing the command, @code{m4} flushes its buffers. The default standard input and standard error of @var{shell-command} are the same as those of @code{m4}. The error output of @var{shell-command} is not a part of the expansion: it will appear along with the error output of @code{m4}. By default, the @var{shell-command} will be used as the argument to the @option{-c} option of the @command{/bin/sh} shell (or the version of @command{sh} specified by @samp{command -p getconf PATH}, if your system supports that). If you prefer a different shell, the @command{configure} script can be given the option @option{--with-syscmd-shell=@var{location}} to set the location of an alternative shell at GNU @code{m4} installation; the alternative shell must still support @option{-c}. The macro @code{esyscmd} is recognized only with parameters. @end deffn @example define(`foo', `FOO') @result{} esyscmd(`echo foo') @result{}FOO @result{} @end example Note how the expansion of @code{esyscmd} keeps the trailing newline of the command, as well as using the newline that appeared after the macro. Just as with @code{syscmd}, care must be exercised when sharing standard input between @code{m4} and the child process of @code{esyscmd}. @node Sysval @section Exit status @cindex UNIX commands, exit status from @cindex exit status from shell commands @cindex shell commands, exit status from @cindex commands, exit status from shell @cindex status of shell commands To see whether a shell command succeeded, use @code{sysval}: @deffn Builtin sysval Expands to the exit status of the last shell command run with @code{syscmd} or @code{esyscmd}. Expands to 0 if no command has been run yet. @end deffn @example sysval @result{}0 syscmd(`false') @result{} ifelse(sysval, `0', `zero', `non-zero') @result{}non-zero syscmd(`exit 2') @result{} sysval @result{}2 syscmd(`true') @result{} sysval @result{}0 esyscmd(`false') @result{} ifelse(sysval, `0', `zero', `non-zero') @result{}non-zero esyscmd(`echo dnl && exit 127') @result{} sysval @result{}127 esyscmd(`true') @result{} sysval @result{}0 @end example @code{sysval} results in 127 if there was a problem executing the command, for example, if the system-imposed argument length is exceeded, or if there were not enough resources to fork. It is not possible to distinguish between failed execution and successful execution that had an exit status of 127, unless there was output from the child process. On UNIX platforms, where it is possible to detect when command execution is terminated by a signal, rather than a normal exit, the result is the signal number shifted left by eight bits. @comment This test has difficulties being portable, even on platforms @comment where syscmd invokes /bin/sh. Kill is not portable with signal @comment names. According to autoconf, the only portable signal numbers @comment are 1 (HUP), 2 (INT), 9 (KILL), 13 (PIPE) and 15 (TERM). But @comment all shells handle SIGINT, and ksh handles HUP (as in, the shell @comment exits normally rather than letting the signal terminate it). @comment Also, TERM is flaky, as it can also kill the running m4 on @comment systems where /bin/sh does not create its own process group. @comment And PIPE is unreliable, since people tend to run with it @comment ignored, with m4 inheriting that choice. That leaves KILL as @comment the only signal we can reliably test. @example dnl This test assumes kill is a shell builtin, and that signals are dnl recognizable. ifdef(`__unix__', , `errprint(` skipping: syscmd does not have unix semantics ')m4exit(`77')')dnl syscmd(`kill -9 $$') @result{} sysval @result{}2304 syscmd() @result{} sysval @result{}0 esyscmd(`kill -9 $$') @result{} sysval @result{}2304 @end example @node Mkstemp @section Making temporary files @cindex temporary file names @cindex files, names of temporary Commands specified to @code{syscmd} or @code{esyscmd} might need a temporary file, for output or for some other purpose. There is a builtin macro, @code{mkstemp}, for making a temporary file: @deffn Builtin mkstemp (@var{template}) @deffnx Builtin maketemp (@var{template}) Expands to the quoted name of a new, empty file, made from the string @var{template}, which should end with the string @samp{XXXXXX}. The six @samp{X} characters are then replaced with random characters matching the regular expression @samp{[a-zA-Z0-9._-]}, in order to make the file name unique. If fewer than six @samp{X} characters are found at the end of @code{template}, the result will be longer than the template. The created file will have access permissions as if by @kbd{chmod =rw,go=}, meaning that the current umask of the @code{m4} process is taken into account, and at most only the current user can read and write the file. The traditional behavior, standardized by POSIX, is that @code{maketemp} merely replaces the trailing @samp{X} with the process id, without creating a file or quoting the expansion, and without ensuring that the resulting string is a unique file name. In part, this means that using the same @var{template} twice in the same input file will result in the same expansion. This behavior is a security hole, as it is very easy for another process to guess the name that will be generated, and thus interfere with a subsequent use of @code{syscmd} trying to manipulate that file name. Hence, POSIX has recommended that all new implementations of @code{m4} provide the secure @code{mkstemp} builtin, and that users of @code{m4} check for its existence. The expansion is void and an error issued if a temporary file could not be created. The macros @code{mkstemp} and @code{maketemp} are recognized only with parameters. @end deffn If you try this next example, you will most likely get different output for the two file names, since the replacement characters are randomly chosen: @comment ignore @example $ @kbd{m4} define(`tmp', `oops') @result{} maketemp(`/tmp/fooXXXXXX') @result{}/tmp/fooa07346 ifdef(`mkstemp', `define(`maketemp', defn(`mkstemp'))', `define(`mkstemp', defn(`maketemp'))dnl errprint(`warning: potentially insecure maketemp implementation ')') @result{} mkstemp(`doc') @result{}docQv83Uw @end example @cindex GNU extensions Unless you use the @option{--traditional} command line option (or @option{-G}, @pxref{Limits control, , Invoking m4}), the GNU version of @code{maketemp} is secure. This means that using the same template to multiple calls will generate multiple files. However, we recommend that you use the new @code{mkstemp} macro, introduced in GNU M4 1.4.8, which is secure even in traditional mode. Also, as of M4 1.4.11, the secure implementation quotes the resulting file name, so that you are guaranteed to know what file was created even if the random file name happens to match an existing macro. Notice that this example is careful to use @code{defn} to avoid unintended expansion of @samp{foo}. @example $ @kbd{m4} define(`foo', `errprint(`oops')') @result{} syscmd(`rm -f foo-??????')sysval @result{}0 define(`file1', maketemp(`foo-XXXXXX'))dnl ifelse(esyscmd(`echo \` foo-?????? \''), ` foo-?????? ', `no file', `created') @result{}created define(`file2', maketemp(`foo-XX'))dnl define(`file3', mkstemp(`foo-XXXXXX'))dnl ifelse(len(defn(`file1')), len(defn(`file2')), `same length', `different') @result{}same length ifelse(defn(`file1'), defn(`file2'), `same', `different file') @result{}different file ifelse(defn(`file2'), defn(`file3'), `same', `different file') @result{}different file ifelse(defn(`file1'), defn(`file3'), `same', `different file') @result{}different file syscmd(`rm 'defn(`file1') defn(`file2') defn(`file3')) @result{} sysval @result{}0 @end example @ignore @c Not worth documenting, but make sure we don't leave trailing NUL in @c the expansion. @example syscmd(`rm -rf foodir')sysval @result{}0 syscmd(`mkdir foodir')sysval @result{}0 len(mkstemp(`foodir/fooXXXXX')) @result{}16 syscmd(`rm -r foodir')sysval @result{}0 @end example @c Likewise, and ensure that traditional mode leaves the result unquoted @c without creating a file. @comment options: -G @example syscmd(`rm -f foo-*')sysval @result{}0 len(maketemp(`foo-XXXXX')) @error{}m4:stdin:2: recommend using mkstemp instead @result{}9 define(`abc', `def') @result{} maketemp(`foo-abc') @result{}foo-def @error{}m4:stdin:4: recommend using mkstemp instead syscmd(`test -f foo-*')ifelse(sysval, `0', `0', `1') @result{}1 @end example @end ignore @node Miscellaneous @chapter Miscellaneous builtin macros This chapter describes various builtins, that do not really belong in any of the previous chapters. @menu * Errprint:: Printing error messages * Location:: Printing current location * M4exit:: Exiting from @code{m4} @end menu @node Errprint @section Printing error messages @cindex printing error messages @cindex error messages, printing @cindex messages, printing error @cindex standard error, output to You can print error messages using @code{errprint}: @deffn Builtin errprint (@var{message}, @dots{}) Prints @var{message} and the rest of the arguments to standard error, separated by spaces. Standard error is used, regardless of the @option{--debugfile} option (@pxref{Debugging options, , Invoking m4}). The expansion of @code{errprint} is void. The macro @code{errprint} is recognized only with parameters. @end deffn @example errprint(`Invalid arguments to forloop ') @error{}Invalid arguments to forloop @result{} errprint(`1')errprint(`2',`3 ') @error{}12 3 @result{} @end example A trailing newline is @emph{not} printed automatically, so it should be supplied as part of the argument, as in the example. Unfortunately, the exact output of @code{errprint} is not very portable to other @code{m4} implementations: POSIX requires that all arguments be printed, but some implementations of @code{m4} only print the first. Furthermore, some BSD implementations always append a newline for each @code{errprint} call, regardless of whether the last argument already had one, and POSIX is silent on whether this is acceptable. @node Location @section Printing current location @cindex location, input @cindex input location To make it possible to specify the location of an error, three utility builtins exist: @deffn Builtin __file__ @deffnx Builtin __line__ @deffnx Builtin __program__ Expand to the quoted name of the current input file, the current input line number in that file, and the quoted name of the current invocation of @code{m4}. @end deffn @example errprint(__program__:__file__:__line__: `input error ') @error{}m4:stdin:1: input error @result{} @end example Line numbers start at 1 for each file. If the file was found due to the @option{-I} option or @env{M4PATH} environment variable, that is reflected in the file name. The syncline option (@option{-s}, @pxref{Preprocessor features, , Invoking m4}), and the @samp{f} and @samp{l} flags of @code{debugmode} (@pxref{Debug Levels}), also use this notion of current file and line. Redefining the three location macros has no effect on syncline, debug, warning, or error message output. This example reuses the file @file{incl.m4} mentioned earlier (@pxref{Include}): @comment examples @example $ @kbd{m4 -I examples} define(`foo', ``$0' called at __file__:__line__') @result{} foo @result{}foo called at stdin:2 include(`incl.m4') @result{}Include file start @result{}foo called at examples/incl.m4:2 @result{}Include file end @result{} @end example The location of macros invoked during the rescanning of macro expansion text corresponds to the location in the file where the expansion was triggered, regardless of how many newline characters the expansion text contains. As of GNU M4 1.4.8, the location of text wrapped with @code{m4wrap} (@pxref{M4wrap}) is the point at which the @code{m4wrap} was invoked. Previous versions, however, behaved as though wrapped text came from line 0 of the file ``''. @example define(`echo', `$@@') @result{} define(`foo', `echo(__line__ __line__)') @result{} echo(__line__ __line__) @result{}4 @result{}5 m4wrap(`foo ') @result{} foo(errprint(__line__ __line__ )) @error{}8 @error{}9 @result{}8 @result{}8 __line__ @result{}11 m4wrap(`__line__ ') @result{} ^D @result{}12 @result{}6 @result{}6 @end example The @code{@w{__program__}} macro behaves like @samp{$0} in shell terminology. If you invoke @code{m4} through an absolute path or a link with a different spelling, rather than by relying on a @env{PATH} search for plain @samp{m4}, it will affect how @code{@w{__program__}} expands. The intent is that you can use it to produce error messages with the same formatting that @code{m4} produces internally. It can also be used within @code{syscmd} (@pxref{Syscmd}) to pick the same version of @code{m4} that is currently running, rather than whatever version of @code{m4} happens to be first in @env{PATH}. It was first introduced in GNU M4 1.4.6. @node M4exit @section Exiting from @code{m4} @cindex exiting from @code{m4} @cindex status, setting @code{m4} exit If you need to exit from @code{m4} before the entire input has been read, you can use @code{m4exit}: @deffn Builtin m4exit (@dvar{code, 0}) Causes @code{m4} to exit, with exit status @var{code}. If @var{code} is left out, the exit status is zero. If @var{code} cannot be parsed, or is outside the range of 0 to 255, the exit status is one. No further input is read, and all wrapped and diverted text is discarded. @end deffn @example m4wrap(`This text is lost due to `m4exit'.') @result{} divert(`1') So is this. divert @result{} m4exit And this is never read. @end example A common use of this is to abort processing: @deffn Composite fatal_error (@var{message}) Abort processing with an error message and non-zero status. Prefix @var{message} with details about where the error occurred, and print the resulting string to standard error. @end deffn @comment status: 1 @example define(`fatal_error', `errprint(__program__:__file__:__line__`: fatal error: $* ')m4exit(`1')') @result{} fatal_error(`this is a BAD one, buster') @error{}m4:stdin:4: fatal error: this is a BAD one, buster @end example After this macro call, @code{m4} will exit with exit status 1. This macro is only intended for error exits, since the normal exit procedures are not followed, i.e., diverted text is not undiverted, and saved text (@pxref{M4wrap}) is not reread. (This macro could be made more robust to earlier versions of @code{m4}. You should try to see if you can find weaknesses and correct them; or @pxref{Improved fatal_error, , Answers}). Note that it is still possible for the exit status to be different than what was requested by @code{m4exit}. If @code{m4} detects some other error, such as a write error on standard output, the exit status will be non-zero even if @code{m4exit} requested zero. If standard input is seekable, then the file will be positioned at the next unread character. If it is a pipe or other non-seekable file, then there are no guarantees how much data @code{m4} might have read into buffers, and thus discarded. @node Frozen files @chapter Fast loading of frozen state Some bigger @code{m4} applications may be built over a common base containing hundreds of definitions and other costly initializations. Usually, the common base is kept in one or more declarative files, which files are listed on each @code{m4} invocation prior to the user's input file, or else each input file uses @code{include}. Reading the common base of a big application, over and over again, may be time consuming. GNU @code{m4} offers some machinery to speed up the start of an application using lengthy common bases. @menu * Using frozen files:: Using frozen files * Frozen file format:: Frozen file format @end menu @node Using frozen files @section Using frozen files @cindex fast loading of frozen files @cindex frozen files for fast loading @cindex initialization, frozen state @cindex dumping into frozen file @cindex reloading a frozen file @cindex GNU extensions Suppose a user has a library of @code{m4} initializations in @file{base.m4}, which is then used with multiple input files: @comment ignore @example $ @kbd{m4 base.m4 input1.m4} $ @kbd{m4 base.m4 input2.m4} $ @kbd{m4 base.m4 input3.m4} @end example Rather than spending time parsing the fixed contents of @file{base.m4} every time, the user might rather execute: @comment ignore @example $ @kbd{m4 -F base.m4f base.m4} @end example @noindent once, and further execute, as often as needed: @comment ignore @example $ @kbd{m4 -R base.m4f input1.m4} $ @kbd{m4 -R base.m4f input2.m4} $ @kbd{m4 -R base.m4f input3.m4} @end example @noindent with the varying input. The first call, containing the @option{-F} option, only reads and executes file @file{base.m4}, defining various application macros and computing other initializations. Once the input file @file{base.m4} has been completely processed, GNU @code{m4} produces in @file{base.m4f} a @dfn{frozen} file, that is, a file which contains a kind of snapshot of the @code{m4} internal state. Later calls, containing the @option{-R} option, are able to reload the internal state of @code{m4}, from @file{base.m4f}, @emph{prior} to reading any other input files. This means instead of starting with a virgin copy of @code{m4}, input will be read after having effectively recovered the effect of a prior run. In our example, the effect is the same as if file @file{base.m4} has been read anew. However, this effect is achieved a lot faster. Only one frozen file may be created or read in any one @code{m4} invocation. It is not possible to recover two frozen files at once. However, frozen files may be updated incrementally, through using @option{-R} and @option{-F} options simultaneously. For example, if some care is taken, the command: @comment ignore @example $ @kbd{m4 file1.m4 file2.m4 file3.m4 file4.m4} @end example @noindent could be broken down in the following sequence, accumulating the same output: @comment ignore @example $ @kbd{m4 -F file1.m4f file1.m4} $ @kbd{m4 -R file1.m4f -F file2.m4f file2.m4} $ @kbd{m4 -R file2.m4f -F file3.m4f file3.m4} $ @kbd{m4 -R file3.m4f file4.m4} @end example Some care is necessary because not every effort has been made for this to work in all cases. In particular, the trace attribute of macros is not handled, nor the current setting of @code{changeword}. Currently, @code{m4wrap} and @code{sysval} also have problems. Also, interactions for some options of @code{m4}, being used in one call and not in the next, have not been fully analyzed yet. On the other end, you may be confident that stacks of @code{pushdef} definitions are handled correctly, as well as undefined or renamed builtins, and changed strings for quotes or comments. And future releases of GNU M4 will improve on the utility of frozen files. @ignore @c This example is not worth putting in the manual, but caused core @c dumps in all versions prior to 1.4.11. @comment options: -F /dev/null @example traceon(`undefined')dnl @end example @c Make sure freezing is successful. @example ifdef(`__unix__', , `errprint(` skipping: syscmd does not have unix semantics ')m4exit(`77')')dnl changequote(`[', `]')dnl syscmd([echo 'changequote([,])pushdef([divnum],[hi])dnl' \ | ']__program__[' -F in.m4f \ && echo 'divnum popdef([divnum])divnum' \ | ']__program__[' -R in.m4f \ && rm in.m4f])status sysval @result{}hi 0 @result{}status 0 @end example @c Detect inability to freeze. @c Some systems harden /, and fail with EACCES rather than ENOENT. @comment options: -F /none/such @comment xerr: ignore @comment status: 1 @example $ @kbd{m4 -F /none/such} ^D @error{}m4: cannot open `/none/such': No such file or directory @end example @end ignore When an @code{m4} run is to be frozen, the automatic undiversion which takes place at end of execution is inhibited. Instead, all positively numbered diversions are saved into the frozen file. The active diversion number is also transmitted. A frozen file to be reloaded need not reside in the current directory. It is looked up the same way as an @code{include} file (@pxref{Search Path}). If the frozen file was generated with a newer version of @code{m4}, and contains directives that an older @code{m4} cannot parse, attempting to load the frozen file with option @option{-R} will cause @code{m4} to exit with status 63 to indicate version mismatch. @node Frozen file format @section Frozen file format @cindex frozen file format @cindex file format, frozen file Frozen files are sharable across architectures. It is safe to write a frozen file on one machine and read it on another, given that the second machine uses the same or newer version of GNU @code{m4}. It is conventional, but not required, to give a frozen file the suffix of @code{.m4f}. These are simple (editable) text files, made up of directives, each starting with a capital letter and ending with a newline (@key{NL}). Wherever a directive is expected, the character @samp{#} introduces a comment line; empty lines are also ignored if they are not part of an embedded string. In the following descriptions, each @var{len} refers to the length of the corresponding strings @var{str} in the next line of input. Numbers are always expressed in decimal. There are no escape characters. The directives are: @table @code @item C @var{len1} , @var{len2} @key{NL} @var{str1} @var{str2} @key{NL} Uses @var{str1} and @var{str2} as the begin-comment and end-comment strings. If omitted, then @samp{#} and @key{NL} are the comment delimiters. @item D @var{number}, @var{len} @key{NL} @var{str} @key{NL} Selects diversion @var{number}, making it current, then copy @var{str} in the current diversion. @var{number} may be a negative number for a non-existing diversion. To merely specify an active selection, use this command with an empty @var{str}. With 0 as the diversion @var{number}, @var{str} will be issued on standard output at reload time. GNU @code{m4} will not produce the @samp{D} directive with non-zero length for diversion 0, but this can be done with manual edits. This directive may appear more than once for the same diversion, in which case the diversion is the concatenation of the various uses. If omitted, then diversion 0 is current. @item F @var{len1} , @var{len2} @key{NL} @var{str1} @var{str2} @key{NL} Defines, through @code{pushdef}, a definition for @var{str1} expanding to the function whose builtin name is @var{str2}. If the builtin does not exist (for example, if the frozen file was produced by a copy of @code{m4} compiled with changeword support, but the version of @code{m4} reloading was compiled without it), the reload is silent, but any subsequent use of the definition of @var{str1} will result in a warning. This directive may appear more than once for the same name, and its order, along with @samp{T}, is important. If omitted, you will have no access to any builtins. @item Q @var{len1} , @var{len2} @key{NL} @var{str1} @var{str2} @key{NL} Uses @var{str1} and @var{str2} as the begin-quote and end-quote strings. If omitted, then @samp{`} and @samp{'} are the quote delimiters. @item T @var{len1} , @var{len2} @key{NL} @var{str1} @var{str2} @key{NL} Defines, though @code{pushdef}, a definition for @var{str1} expanding to the text given by @var{str2}. This directive may appear more than once for the same name, and its order, along with @samp{F}, is important. @item V @var{number} @key{NL} Confirms the format of the file. @code{m4} @value{VERSION} only creates and understands frozen files where @var{number} is 1. This directive must be the first non-comment in the file, and may not appear more than once. @end table @node Compatibility @chapter Compatibility with other versions of @code{m4} @cindex compatibility This chapter describes the many of the differences between this implementation of @code{m4}, and of other implementations found under UNIX, such as System V Release 3, Solaris, and BSD flavors. In particular, it lists the known differences and extensions to POSIX. However, the list is not necessarily comprehensive. At the time of this writing, POSIX 2001 (also known as IEEE Std 1003.1-2001) is the latest standard, although a new version of POSIX is under development and includes several proposals for modifying what @code{m4} is required to do. The requirements for @code{m4} are shared between SUSv3 and POSIX, and can be viewed at @uref{http://www.opengroup.org/onlinepubs/@/000095399/@/utilities/@/m4.html}. @menu * Extensions:: Extensions in GNU M4 * Incompatibilities:: Facilities in System V m4 not in GNU M4 * Other Incompatibilities:: Other incompatibilities @end menu @node Extensions @section Extensions in GNU M4 @cindex GNU extensions @cindex POSIX This version of @code{m4} contains a few facilities that do not exist in System V @code{m4}. These extra facilities are all suppressed by using the @option{-G} command line option (@pxref{Limits control, , Invoking m4}), unless overridden by other command line options. @itemize @bullet @item In the @code{$@var{n}} notation for macro arguments, @var{n} can contain several digits, while the System V @code{m4} only accepts one digit. This allows macros in GNU @code{m4} to take any number of arguments, and not only nine (@pxref{Arguments}). This means that @code{define(`foo', `$11')} is ambiguous between implementations. To portably choose between grabbing the first parameter and appending 1 to the expansion, or grabbing the eleventh parameter, you can do the following: @example define(`a1', `A1') @result{} dnl First argument, concatenated with 1 define(`_1', `$1')define(`first1', `_1($@@)1') @result{} dnl Eleventh argument, portable define(`_9', `$9')define(`eleventh', `_9(shift(shift($@@)))') @result{} dnl Eleventh argument, GNU style define(`Eleventh', `$11') @result{} first1(`a', `b', `c', `d', `e', `f', `g', `h', `i', `j', `k') @result{}A1 eleventh(`a', `b', `c', `d', `e', `f', `g', `h', `i', `j', `k') @result{}k Eleventh(`a', `b', `c', `d', `e', `f', `g', `h', `i', `j', `k') @result{}k @end example @noindent Also see the @code{argn} macro (@pxref{Shift}). @item The @code{divert} (@pxref{Divert}) macro can manage more than 9 diversions. GNU @code{m4} treats all positive numbers as valid diversions, rather than discarding diversions greater than 9. @item Files included with @code{include} and @code{sinclude} are sought in a user specified search path, if they are not found in the working directory. The search path is specified by the @option{-I} option and the @env{M4PATH} environment variable (@pxref{Search Path}). @item Arguments to @code{undivert} can be non-numeric, in which case the named file will be included uninterpreted in the output (@pxref{Undivert}). @item Formatted output is supported through the @code{format} builtin, which is modeled after the C library function @code{printf} (@pxref{Format}). @item Searches and text substitution through basic regular expressions are supported by the @code{regexp} (@pxref{Regexp}) and @code{patsubst} (@pxref{Patsubst}) builtins. Some BSD implementations use extended regular expressions instead. @item The output of shell commands can be read into @code{m4} with @code{esyscmd} (@pxref{Esyscmd}). @item There is indirect access to any builtin macro with @code{builtin} (@pxref{Builtin}). @item Macros can be called indirectly through @code{indir} (@pxref{Indir}). @item The name of the program, the current input file, and the current input line number are accessible through the builtins @code{@w{__program__}}, @code{@w{__file__}}, and @code{@w{__line__}} (@pxref{Location}). @item The format of the output from @code{dumpdef} and macro tracing can be controlled with @code{debugmode} (@pxref{Debug Levels}). @item The destination of trace and debug output can be controlled with @code{debugfile} (@pxref{Debug Output}). @item The @code{maketemp} (@pxref{Mkstemp}) macro behaves like @code{mkstemp}, creating a new file with a unique name on every invocation, rather than following the insecure behavior of replacing the trailing @samp{X} characters with the @code{m4} process id. @item POSIX only requires support for the command line options @option{-s}, @option{-D}, and @option{-U}, so all other options accepted by GNU M4 are extensions. @xref{Invoking m4}, for a description of these options. The debugging and tracing facilities in GNU @code{m4} are much more extensive than in most other versions of @code{m4}. @end itemize @node Incompatibilities @section Facilities in System V @code{m4} not in GNU @code{m4} The version of @code{m4} from System V contains a few facilities that have not been implemented in GNU @code{m4} yet. Additionally, POSIX requires some behaviors that GNU @code{m4} has not implemented yet. Relying on these behaviors is non-portable, as a future release of GNU @code{m4} may change. @itemize @bullet @item POSIX requires support for multiple arguments to @code{defn}, without any clarification on how @code{defn} behaves when one of the multiple arguments names a builtin. System V @code{m4} and some other implementations allow mixing builtins and text macros into a single macro. GNU @code{m4} only supports joining multiple text arguments, although a future implementation may lift this restriction to behave more like System V@. The only portable way to join text macros with builtins is via helper macros and implicit concatenation of macro results. @item POSIX requires an application to exit with non-zero status if it wrote an error message to stderr. This has not yet been consistently implemented for the various builtins that are required to issue an error (such as @code{eval} (@pxref{Eval}) when an argument cannot be parsed). @item Some traditional implementations only allow reading standard input once, but GNU @code{m4} correctly handles multiple instances of @samp{-} on the command line. @item POSIX requires @code{m4wrap} (@pxref{M4wrap}) to act in FIFO (first-in, first-out) order, but GNU @code{m4} currently uses LIFO order. Furthermore, POSIX states that only the first argument to @code{m4wrap} is saved for later evaluation, but GNU @code{m4} saves and processes all arguments, with output separated by spaces. @item POSIX states that builtins that require arguments, but are called without arguments, have undefined behavior. Traditional implementations simply behave as though empty strings had been passed. For example, @code{a`'define`'b} would expand to @code{ab}. But GNU @code{m4} ignores certain builtins if they have missing arguments, giving @code{adefineb} for the above example. @item Traditional implementations handle @code{define(`f',`1')} (@pxref{Define}) by undefining the entire stack of previous definitions, and if doing @code{undefine(`f')} first. GNU @code{m4} replaces just the top definition on the stack, as if doing @code{popdef(`f')} followed by @code{pushdef(`f',`1')}. POSIX allows either behavior. @item POSIX 2001 requires @code{syscmd} (@pxref{Syscmd}) to evaluate command output for macro expansion, but this was a mistake that is anticipated to be corrected in the next version of POSIX. GNU @code{m4} follows traditional behavior in @code{syscmd} where output is not rescanned, and provides the extension @code{esyscmd} that does scan the output. @item At one point, POSIX required @code{changequote(@var{arg})} (@pxref{Changequote}) to use newline as the close quote, but this was a bug, and the next version of POSIX is anticipated to state that using empty strings or just one argument is unspecified. Meanwhile, the GNU @code{m4} behavior of treating an empty end-quote delimiter as @samp{'} is not portable, as Solaris treats it as repeating the start-quote delimiter, and BSD treats it as leaving the previous end-quote delimiter unchanged. For predictable results, never call changequote with just one argument, or with empty strings for arguments. @item At one point, POSIX required @code{changecom(@var{arg},)} (@pxref{Changecom}) to make it impossible to end a comment, but this is a bug, and the next version of POSIX is anticipated to state that using empty strings is unspecified. Meanwhile, the GNU @code{m4} behavior of treating an empty end-comment delimiter as newline is not portable, as BSD treats it as leaving the previous end-comment delimiter unchanged. It is also impossible in BSD implementations to disable comments, even though that is required by POSIX. For predictable results, never call changecom with empty strings for arguments. @item Most implementations of @code{m4} give macros a higher precedence than comments when parsing, meaning that if the start delimiter given to @code{changecom} (@pxref{Changecom}) starts with a macro name, comments are effectively disabled. POSIX does not specify what the precedence is, so this version of GNU @code{m4} parser recognizes comments, then macros, then quoted strings. @item Traditional implementations allow argument collection, but not string and comment processing, to span file boundaries. Thus, if @file{a.m4} contains @samp{len(}, and @file{b.m4} contains @samp{abc)}, @kbd{m4 a.m4 b.m4} outputs @samp{3} with traditional @code{m4}, but gives an error message that the end of file was encountered inside a macro with GNU @code{m4}. On the other hand, traditional implementations do end of file processing for files included with @code{include} or @code{sinclude} (@pxref{Include}), while GNU @code{m4} seamlessly integrates the content of those files. Thus @code{include(`a.m4')include(`b.m4')} will output @samp{3} instead of giving an error. @item Traditional @code{m4} treats @code{traceon} (@pxref{Trace}) without arguments as a global variable, independent of named macro tracing. Also, once a macro is undefined, named tracing of that macro is lost. On the other hand, when GNU @code{m4} encounters @code{traceon} without arguments, it turns tracing on for all existing definitions at the time, but does not trace future definitions; @code{traceoff} without arguments turns tracing off for all definitions regardless of whether they were also traced by name; and tracing by name, such as with @option{-tfoo} at the command line or @code{traceon(`foo')} in the input, is an attribute that is preserved even if the macro is currently undefined. Additionally, while POSIX requires trace output, it makes no demands on the formatting of that output. Parsing trace output is not guaranteed to be reliable, even between different releases of GNU M4; however, the intent is that any future changes in trace output will only occur under the direction of additional @code{debugmode} flags (@pxref{Debug Levels}). @item POSIX requires @code{eval} (@pxref{Eval}) to treat all operators with the same precedence as C@. However, earlier versions of GNU @code{m4} followed the traditional behavior of other @code{m4} implementations, where bitwise and logical negation (@samp{~} and @samp{!}) have lower precedence than equality operators; and where equality operators (@samp{==} and @samp{!=}) had the same precedence as relational operators (such as @samp{<}). Use explicit parentheses to ensure proper precedence. As extensions to POSIX, GNU @code{m4} gives well-defined semantics to operations that C leaves undefined, such as when overflow occurs, when shifting negative numbers, or when performing division by zero. POSIX also requires @samp{=} to cause an error, but many traditional implementations allowed it as an alias for @samp{==}. @item POSIX 2001 requires @code{translit} (@pxref{Translit}) to treat each character of the second and third arguments literally. However, it is anticipated that the next version of POSIX will allow the GNU @code{m4} behavior of treating @samp{-} as a range operator. @item POSIX requires @code{m4} to honor the locale environment variables of @env{LANG}, @env{LC_ALL}, @env{LC_CTYPE}, @env{LC_MESSAGES}, and @env{NLSPATH}, but this has not yet been implemented in GNU @code{m4}. @item POSIX states that only unquoted leading newlines and blanks (that is, space and tab) are ignored when collecting macro arguments. However, this appears to be a bug in POSIX, since most traditional implementations also ignore all whitespace (formfeed, carriage return, and vertical tab). GNU @code{m4} follows tradition and ignores all leading unquoted whitespace. @item @cindex @env{POSIXLY_CORRECT} A strictly-compliant POSIX client is not allowed to use command-line arguments not specified by POSIX. However, since this version of M4 ignores @env{POSIXLY_CORRECT} and enables the option @code{--gnu} by default (@pxref{Limits control, , Invoking m4}), a client desiring to be strictly compliant has no way to disable GNU extensions that conflict with POSIX when directly invoking the compiled @code{m4}. A future version of @code{GNU} M4 will honor the environment variable @env{POSIXLY_CORRECT}, implicitly enabling @option{--traditional} if it is set, in order to allow a strictly-compliant client. In the meantime, a client needing strict POSIX compliance can use the workaround of invoking a shell script wrapper, where the wrapper then adds @option{--traditional} to the arguments passed to the compiled @code{m4}. @end itemize @node Other Incompatibilities @section Other incompatibilities There are a few other incompatibilities between this implementation of @code{m4}, and the System V version. @itemize @bullet @item GNU @code{m4} implements sync lines differently from System V @code{m4}, when text is being diverted. GNU @code{m4} outputs the sync lines when the text is being diverted, and System V @code{m4} when the diverted text is being brought back. The problem is which lines and file names should be attached to text that is being, or has been, diverted. System V @code{m4} regards all the diverted text as being generated by the source line containing the @code{undivert} call, whereas GNU @code{m4} regards the diverted text as being generated at the time it is diverted. The sync line option is used mostly when using @code{m4} as a front end to a compiler. If a diverted line causes a compiler error, the error messages should most probably refer to the place where the diversion was made, and not where it was inserted again. @comment options: -s @example divert(2)2 divert(1)1 divert`'0 @result{}#line 3 "stdin" @result{}0 ^D @result{}#line 2 "stdin" @result{}1 @result{}#line 1 "stdin" @result{}2 @end example The current @code{m4} implementation has a limitation that the syncline output at the start of each diversion occurs no matter what, even if the previous diversion did not end with a newline. This goes contrary to the claim that synclines appear on a line by themselves, so this limitation may be corrected in a future version of @code{m4}. In the meantime, when using @option{-s}, it is wisest to make sure all diversions end with newline. @item GNU @code{m4} makes no attempt at prohibiting self-referential definitions like: @example define(`x', `x') @result{} define(`x', `x ') @result{} @end example @cindex rescanning There is nothing inherently wrong with defining @samp{x} to return @samp{x}. The wrong thing is to expand @samp{x} unquoted, because that would cause an infinite rescan loop. In @code{m4}, one might use macros to hold strings, as we do for variables in other programming languages, further checking them with: @comment ignore @example ifelse(defn(`@var{holder}'), `@var{value}', @dots{}) @end example @noindent In cases like this one, an interdiction for a macro to hold its own name would be a useless limitation. Of course, this leaves more rope for the GNU @code{m4} user to hang himself! Rescanning hangs may be avoided through careful programming, a little like for endless loops in traditional programming languages. @end itemize @node Answers @chapter Correct version of some examples Some of the examples in this manuals are buggy or not very robust, for demonstration purposes. Improved versions of these composite macros are presented here. @menu * Improved exch:: Solution for @code{exch} * Improved forloop:: Solution for @code{forloop} * Improved foreach:: Solution for @code{foreach} * Improved copy:: Solution for @code{copy} * Improved m4wrap:: Solution for @code{m4wrap} * Improved cleardivert:: Solution for @code{cleardivert} * Improved capitalize:: Solution for @code{capitalize} * Improved fatal_error:: Solution for @code{fatal_error} @end menu @node Improved exch @section Solution for @code{exch} The @code{exch} macro (@pxref{Arguments}) as presented requires clients to double quote their arguments. A nicer definition, which lets clients follow the rule of thumb of one level of quoting per level of parentheses, involves adding quotes in the definition of @code{exch}, as follows: @example define(`exch', ``$2', `$1'') @result{} define(exch(`expansion text', `macro')) @result{} macro @result{}expansion text @end example @node Improved forloop @section Solution for @code{forloop} The @code{forloop} macro (@pxref{Forloop}) as presented earlier can go into an infinite loop if given an iterator that is not parsed as a macro name. It does not do any sanity checking on its numeric bounds, and only permits decimal numbers for bounds. Here is an improved version, shipped as @file{m4-@value{VERSION}/@/examples/@/forloop2.m4}; this version also optimizes overhead by calling four macros instead of six per iteration (excluding those in @var{text}), by not dereferencing the @var{iterator} in the helper @code{@w{_forloop}}. @comment examples @example $ @kbd{m4 -d -I examples} undivert(`forloop2.m4')dnl @result{}divert(`-1') @result{}# forloop(var, from, to, stmt) - improved version: @result{}# works even if VAR is not a strict macro name @result{}# performs sanity check that FROM is larger than TO @result{}# allows complex numerical expressions in TO and FROM @result{}define(`forloop', `ifelse(eval(`($2) <= ($3)'), `1', @result{} `pushdef(`$1')_$0(`$1', eval(`$2'), @result{} eval(`$3'), `$4')popdef(`$1')')') @result{}define(`_forloop', @result{} `define(`$1', `$2')$4`'ifelse(`$2', `$3', `', @result{} `$0(`$1', incr(`$2'), `$3', `$4')')') @result{}divert`'dnl include(`forloop2.m4') @result{} forloop(`i', `2', `1', `no iteration occurs') @result{} forloop(`', `1', `2', ` odd iterator name') @result{} odd iterator name odd iterator name forloop(`i', `5 + 5', `0xc', ` 0x`'eval(i, `16')') @result{} 0xa 0xb 0xc forloop(`i', `a', `b', `non-numeric bounds') @error{}m4:stdin:6: bad expression in eval (bad input): (a) <= (b) @result{} @end example One other change to notice is that the improved version used @samp{_$0} rather than @samp{_foreach} to invoke the helper routine. In general, this is a good practice to follow, because then the set of macros can be uniformly transformed. The following example shows a transformation that doubles the current quoting and appends a suffix @samp{2} to each transformed macro. If @code{foreach} refers to the literal @samp{_foreach}, then @code{foreach2} invokes @code{_foreach} instead of the intended @code{_foreach2}, and the mixing of quoting paradigms leads to an infinite recursion loop in this example. @comment options: -L9 @comment status: 1 @comment examples @example $ @kbd{m4 -d -L 9 -I examples} define(`arg1', `$1')include(`forloop2.m4')include(`quote.m4') @result{} define(`double', `define(`$1'`2', arg1(patsubst(dquote(defn(`$1')), `[`']', `\&\&')))') @result{} double(`forloop')double(`_forloop')defn(`forloop2') @result{}ifelse(eval(``($2) <= ($3)''), ``1'', @result{} ``pushdef(``$1'')_$0(``$1'', eval(``$2''), @result{} eval(``$3''), ``$4'')popdef(``$1'')'') forloop(i, 1, 5, `ifelse(')forloop(i, 1, 5, `)') @result{} changequote(`[', `]')changequote([``], ['']) @result{} forloop2(i, 1, 5, ``ifelse('')forloop2(i, 1, 5, ``)'') @result{} changequote`'include(`forloop.m4') @result{} double(`forloop')double(`_forloop')defn(`forloop2') @result{}pushdef(``$1'', ``$2'')_forloop($@@)popdef(``$1'') forloop(i, 1, 5, `ifelse(')forloop(i, 1, 5, `)') @result{} changequote(`[', `]')changequote([``], ['']) @result{} forloop2(i, 1, 5, ``ifelse('')forloop2(i, 1, 5, ``)'') @error{}m4:stdin:12: recursion limit of 9 exceeded, use -L to change it @end example One more optimization is still possible. Instead of repeatedly assigning a variable then invoking or dereferencing it, it is possible to pass the current iterator value as a single argument. Coupled with @code{curry} if other arguments are needed (@pxref{Composition}), or with helper macros if the argument is needed in more than one place in the expansion, the output can be generated with three, rather than four, macros of overhead per iteration. Notice how the file @file{m4-@value{VERSION}/@/examples/@/forloop3.m4} rearranges the arguments of the helper @code{_forloop} to take two arguments that are placed around the current value. By splitting a balanced set of parantheses across multiple arguments, the helper macro can now be shared by @code{forloop} and the new @code{forloop_arg}. @comment examples @example $ @kbd{m4 -I examples} include(`forloop3.m4') @result{} undivert(`forloop3.m4')dnl @result{}divert(`-1') @result{}# forloop_arg(from, to, macro) - invoke MACRO(value) for @result{}# each value between FROM and TO, without define overhead @result{}define(`forloop_arg', `ifelse(eval(`($1) <= ($2)'), `1', @result{} `_forloop(`$1', eval(`$2'), `$3(', `)')')') @result{}# forloop(var, from, to, stmt) - refactored to share code @result{}define(`forloop', `ifelse(eval(`($2) <= ($3)'), `1', @result{} `pushdef(`$1')_forloop(eval(`$2'), eval(`$3'), @result{} `define(`$1',', `)$4')popdef(`$1')')') @result{}define(`_forloop', @result{} `$3`$1'$4`'ifelse(`$1', `$2', `', @result{} `$0(incr(`$1'), `$2', `$3', `$4')')') @result{}divert`'dnl forloop(`i', `1', `3', ` i') @result{} 1 2 3 define(`echo', `$@@') @result{} forloop_arg(`1', `3', ` echo') @result{} 1 2 3 include(`curry.m4') @result{} forloop_arg(`1', `3', `curry(`pushdef', `a')') @result{} a @result{}3 popdef(`a')a @result{}2 popdef(`a')a @result{}1 popdef(`a')a @result{}a @end example Of course, it is possible to make even more improvements, such as adding an optional step argument, or allowing iteration through descending sequences. GNU Autoconf provides some of these additional bells and whistles in its @code{m4_for} macro. @node Improved foreach @section Solution for @code{foreach} The @code{foreach} and @code{foreachq} macros (@pxref{Foreach}) as presented earlier each have flaws. First, we will examine and fix the quadratic behavior of @code{foreachq}: @comment examples @example $ @kbd{m4 -I examples} include(`foreachq.m4') @result{} traceon(`shift')debugmode(`aq') @result{} foreachq(`x', ``1', `2', `3', `4'', `x ')dnl @result{}1 @error{}m4trace: -3- shift(`1', `2', `3', `4') @error{}m4trace: -2- shift(`1', `2', `3', `4') @result{}2 @error{}m4trace: -4- shift(`1', `2', `3', `4') @error{}m4trace: -3- shift(`2', `3', `4') @error{}m4trace: -3- shift(`1', `2', `3', `4') @error{}m4trace: -2- shift(`2', `3', `4') @result{}3 @error{}m4trace: -5- shift(`1', `2', `3', `4') @error{}m4trace: -4- shift(`2', `3', `4') @error{}m4trace: -3- shift(`3', `4') @error{}m4trace: -4- shift(`1', `2', `3', `4') @error{}m4trace: -3- shift(`2', `3', `4') @error{}m4trace: -2- shift(`3', `4') @result{}4 @error{}m4trace: -6- shift(`1', `2', `3', `4') @error{}m4trace: -5- shift(`2', `3', `4') @error{}m4trace: -4- shift(`3', `4') @error{}m4trace: -3- shift(`4') @end example @cindex quadratic behavior, avoiding @cindex avoiding quadratic behavior Each successive iteration was adding more quoted @code{shift} invocations, and the entire list contents were passing through every iteration. In general, when recursing, it is a good idea to make the recursion use fewer arguments, rather than adding additional quoted uses of @code{shift}. By doing so, @code{m4} uses less memory, invokes fewer macros, is less likely to run into machine limits, and most importantly, performs faster. The fixed version of @code{foreachq} can be found in @file{m4-@value{VERSION}/@/examples/@/foreachq2.m4}: @comment examples @example $ @kbd{m4 -I examples} include(`foreachq2.m4') @result{} undivert(`foreachq2.m4')dnl @result{}include(`quote.m4')dnl @result{}divert(`-1') @result{}# foreachq(x, `item_1, item_2, ..., item_n', stmt) @result{}# quoted list, improved version @result{}define(`foreachq', `pushdef(`$1')_$0($@@)popdef(`$1')') @result{}define(`_arg1q', ``$1'') @result{}define(`_rest', `ifelse(`$#', `1', `', `dquote(shift($@@))')') @result{}define(`_foreachq', `ifelse(`$2', `', `', @result{} `define(`$1', _arg1q($2))$3`'$0(`$1', _rest($2), `$3')')') @result{}divert`'dnl traceon(`shift')debugmode(`aq') @result{} foreachq(`x', ``1', `2', `3', `4'', `x ')dnl @result{}1 @error{}m4trace: -3- shift(`1', `2', `3', `4') @result{}2 @error{}m4trace: -3- shift(`2', `3', `4') @result{}3 @error{}m4trace: -3- shift(`3', `4') @result{}4 @end example Note that the fixed version calls unquoted helper macros in @code{@w{_foreachq}} to trim elements immediately; those helper macros in turn must re-supply the layer of quotes lost in the macro invocation. Contrast the use of @code{@w{_arg1q}}, which quotes the first list element, with @code{@w{_arg1}} of the earlier implementation that returned the first list element directly. Additionally, by calling the helper method immediately, the @samp{defn(`@var{iterator}')} no longer contains unexpanded macros. The astute m4 programmer might notice that the solution above still uses more memory and macro invocations, and thus more time, than strictly necessary. Note that @samp{$2}, which contains an arbitrarily long quoted list, is expanded and rescanned three times per iteration of @code{_foreachq}. Furthermore, every iteration of the algorithm effectively unboxes then reboxes the list, which costs a couple of macro invocations. It is possible to rewrite the algorithm for a bit more speed by swapping the order of the arguments to @code{_foreachq} in order to operate on an unboxed list in the first place, and by using the fixed-length @samp{$#} instead of an arbitrary length list as the key to end recursion. The result is an overhead of six macro invocations per loop (excluding any macros in @var{text}), instead of eight. This alternative approach is available as @file{m4-@value{VERSION}/@/examples/@/foreach3.m4}: @comment examples @example $ @kbd{m4 -I examples} include(`foreachq3.m4') @result{} undivert(`foreachq3.m4')dnl @result{}divert(`-1') @result{}# foreachq(x, `item_1, item_2, ..., item_n', stmt) @result{}# quoted list, alternate improved version @result{}define(`foreachq', `ifelse(`$2', `', `', @result{} `pushdef(`$1')_$0(`$1', `$3', `', $2)popdef(`$1')')') @result{}define(`_foreachq', `ifelse(`$#', `3', `', @result{} `define(`$1', `$4')$2`'$0(`$1', `$2', @result{} shift(shift(shift($@@))))')') @result{}divert`'dnl traceon(`shift')debugmode(`aq') @result{} foreachq(`x', ``1', `2', `3', `4'', `x ')dnl @result{}1 @error{}m4trace: -4- shift(`x', `x @error{}', `', `1', `2', `3', `4') @error{}m4trace: -3- shift(`x @error{}', `', `1', `2', `3', `4') @error{}m4trace: -2- shift(`', `1', `2', `3', `4') @result{}2 @error{}m4trace: -4- shift(`x', `x @error{}', `1', `2', `3', `4') @error{}m4trace: -3- shift(`x @error{}', `1', `2', `3', `4') @error{}m4trace: -2- shift(`1', `2', `3', `4') @result{}3 @error{}m4trace: -4- shift(`x', `x @error{}', `2', `3', `4') @error{}m4trace: -3- shift(`x @error{}', `2', `3', `4') @error{}m4trace: -2- shift(`2', `3', `4') @result{}4 @error{}m4trace: -4- shift(`x', `x @error{}', `3', `4') @error{}m4trace: -3- shift(`x @error{}', `3', `4') @error{}m4trace: -2- shift(`3', `4') @end example In the current version of M4, every instance of @samp{$@@} is rescanned as it is encountered. Thus, the @file{foreachq3.m4} alternative uses much less memory than @file{foreachq2.m4}, and executes as much as 10% faster, since each iteration encounters fewer @samp{$@@}. However, the implementation of rescanning every byte in @samp{$@@} is quadratic in the number of bytes scanned (for example, making the broken version in @file{foreachq.m4} cubic, rather than quadratic, in behavior). A future release of M4 will improve the underlying implementation by reusing results of previous scans, so that both styles of @code{foreachq} can become linear in the number of bytes scanned. Notice how the implementation injects an empty argument prior to expanding @samp{$2} within @code{foreachq}; the helper macro @code{_foreachq} then ignores the third argument altogether, and ends recursion when there are three arguments left because there was nothing left to pass through @code{shift}. Thus, each iteration only needs one @code{ifelse}, rather than the two conditionals used in the version from @file{foreachq2.m4}. @cindex nine arguments, more than @cindex more than nine arguments @cindex arguments, more than nine So far, all of the implementations of @code{foreachq} presented have been quadratic with M4 1.4.x. But @code{forloop} is linear, because each iteration parses a constant amount of arguments. So, it is possible to design a variant that uses @code{forloop} to do the iteration, then uses @samp{$@@} only once at the end, giving a linear result even with older M4 implementations. This implementation relies on the GNU extension that @samp{$10} expands to the tenth argument rather than the first argument concatenated with @samp{0}. The trick is to define an intermediate macro that repeats the text @code{m4_define(`$1', `$@var{n}')$2`'}, with @samp{n} set to successive integers corresponding to each argument. The helper macro @code{_foreachq_} is needed in order to generate the literal sequences such as @samp{$1} into the intermediate macro, rather than expanding them as the arguments of @code{_foreachq}. With this approach, no @code{shift} calls are even needed! Even though there are seven macros of overhead per iteration instead of six in @file{foreachq3.m4}, the linear scaling is apparent at relatively small list sizes. However, this approach will need adjustment when a future version of M4 follows POSIX by no longer treating @samp{$10} as the tenth argument; the anticipation is that @samp{$@{10@}} can be used instead, although that alternative syntax is not yet supported. @comment examples @example $ @kbd{m4 -I examples} include(`foreachq4.m4') @result{} undivert(`foreachq4.m4')dnl @result{}include(`forloop2.m4')dnl @result{}divert(`-1') @result{}# foreachq(x, `item_1, item_2, ..., item_n', stmt) @result{}# quoted list, version based on forloop @result{}define(`foreachq', @result{}`ifelse(`$2', `', `', `_$0(`$1', `$3', $2)')') @result{}define(`_foreachq', @result{}`pushdef(`$1', forloop(`$1', `3', `$#', @result{} `$0_(`1', `2', indir(`$1'))')`popdef( @result{} `$1')')indir(`$1', $@@)') @result{}define(`_foreachq_', @result{}``define(`$$1', `$$3')$$2`''') @result{}divert`'dnl traceon(`shift')debugmode(`aq') @result{} foreachq(`x', ``1', `2', `3', `4'', `x ')dnl @result{}1 @result{}2 @result{}3 @result{}4 @end example For yet another approach, the improved version of @code{foreach}, available in @file{m4-@value{VERSION}/@/examples/@/foreach2.m4}, simply overquotes the arguments to @code{@w{_foreach}} to begin with, using @code{dquote_elt}. Then @code{@w{_foreach}} can just use @code{@w{_arg1}} to remove the extra layer of quoting that was added up front: @comment examples @example $ @kbd{m4 -I examples} include(`foreach2.m4') @result{} undivert(`foreach2.m4')dnl @result{}include(`quote.m4')dnl @result{}divert(`-1') @result{}# foreach(x, (item_1, item_2, ..., item_n), stmt) @result{}# parenthesized list, improved version @result{}define(`foreach', `pushdef(`$1')_$0(`$1', @result{} (dquote(dquote_elt$2)), `$3')popdef(`$1')') @result{}define(`_arg1', `$1') @result{}define(`_foreach', `ifelse(`$2', `(`')', `', @result{} `define(`$1', _arg1$2)$3`'$0(`$1', (dquote(shift$2)), `$3')')') @result{}divert`'dnl traceon(`shift')debugmode(`aq') @result{} foreach(`x', `(`1', `2', `3', `4')', `x ')dnl @error{}m4trace: -4- shift(`1', `2', `3', `4') @error{}m4trace: -4- shift(`2', `3', `4') @error{}m4trace: -4- shift(`3', `4') @result{}1 @error{}m4trace: -3- shift(``1'', ``2'', ``3'', ``4'') @result{}2 @error{}m4trace: -3- shift(``2'', ``3'', ``4'') @result{}3 @error{}m4trace: -3- shift(``3'', ``4'') @result{}4 @error{}m4trace: -3- shift(``4'') @end example It is likewise possible to write a variant of @code{foreach} that performs in linear time on M4 1.4.x; the easiest method is probably writing a version of @code{foreach} that unboxes its list, then invokes @code{_foreachq} as previously defined in @file{foreachq4.m4}. In summary, recursion over list elements is trickier than it appeared at first glance, but provides a powerful idiom within @code{m4} processing. As a final demonstration, both list styles are now able to handle several scenarios that would wreak havoc on one or both of the original implementations. This points out one other difference between the list styles. @code{foreach} evaluates unquoted list elements only once, in preparation for calling @code{@w{_foreach}}, similary for @code{foreachq} as provided by @file{foreachq3.m4} or @file{foreachq4.m4}. But @code{foreachq}, as provided by @file{foreachq2.m4}, evaluates unquoted list elements twice while visiting the first list element, once in @code{@w{_arg1q}} and once in @code{@w{_rest}}. When deciding which list style to use, one must take into account whether repeating the side effects of unquoted list elements will have any detrimental effects. @comment examples @example $ @kbd{m4 -I examples} include(`foreach2.m4') @result{} include(`foreachq2.m4') @result{} dnl 0-element list: foreach(`x', `', `') / foreachq(`x', `', `') @result{} /@w{ } dnl 1-element list of empty element foreach(`x', `()', `') / foreachq(`x', ``'', `') @result{}<> / <> dnl 2-element list of empty elements foreach(`x', `(`',`')', `') / foreachq(`x', ``',`'', `') @result{}<><> / <><> dnl 1-element list of a comma foreach(`x', `(`,')', `') / foreachq(`x', ``,'', `') @result{}<,> / <,> dnl 2-element list of unbalanced parentheses foreach(`x', `(`(', `)')', `') / foreachq(`x', ``(', `)'', `') @result{}<(><)> / <(><)> define(`ab', `oops')dnl using defn(`iterator') foreach(`x', `(`a', `b')', `defn(`x')') /dnl foreachq(`x', ``a', `b'', `defn(`x')') @result{}ab / ab define(`active', `ACT, IVE') @result{} traceon(`active') @result{} dnl list of unquoted macros; expansion occurs before recursion foreach(`x', `(active, active)', ` ')dnl @error{}m4trace: -4- active -> `ACT, IVE' @error{}m4trace: -4- active -> `ACT, IVE' @result{} @result{} @result{} @result{} foreachq(`x', `active, active', ` ')dnl @error{}m4trace: -3- active -> `ACT, IVE' @error{}m4trace: -3- active -> `ACT, IVE' @result{} @error{}m4trace: -3- active -> `ACT, IVE' @error{}m4trace: -3- active -> `ACT, IVE' @result{} @result{} @result{} dnl list of quoted macros; expansion occurs during recursion foreach(`x', `(`active', `active')', ` ')dnl @error{}m4trace: -1- active -> `ACT, IVE' @result{} @error{}m4trace: -1- active -> `ACT, IVE' @result{} foreachq(`x', ``active', `active'', ` ')dnl @error{}m4trace: -1- active -> `ACT, IVE' @result{} @error{}m4trace: -1- active -> `ACT, IVE' @result{} dnl list of double-quoted macro names; no expansion foreach(`x', `(``active'', ``active'')', ` ')dnl @result{} @result{} foreachq(`x', ```active'', ``active''', ` ')dnl @result{} @result{} @end example @ignore @comment Not worth putting in the manual, but make sure that foreach @comment implementations behave, and that final implementation is @comment linear. @comment boxed recursion @comment examples @comment options: -Dlimit=10 -Dverbose @example $ @kbd {m4 -I examples -Dlimit=10 -Dverbose} include(`loop.m4')dnl @result{} 1 2 3 4 5 6 7 8 9 10 @end example @comment unboxed recursion @comment examples @comment options: -Dlimit=10 -Dverbose -Dalt @example $ @kbd {m4 -I examples -Dlimit=10 -Dverbose -Dalt} include(`loop.m4')dnl @result{} 1 2 3 4 5 6 7 8 9 10 @end example @comment foreach via forloop recursion @comment examples @comment options: -Dlimit=10 -Dverbose -Dalt=4 @example $ @kbd {m4 -I examples -Dlimit=10 -Dverbose -Dalt=4} include(`loop.m4')dnl @result{} 1 2 3 4 5 6 7 8 9 10 @end example @comment examples @comment options: -Dlimit=2500 -Dalt=4 @example $ @kbd {m4 -I examples -Dlimit=2500 -Dalt=4} include(`loop.m4')dnl @end example @comment examples @comment options: -Dlimit=10000 -Dalt=4 @example $ @kbd {m4 -I examples -Dlimit=10000 -Dalt=4} define(`foo', `divert`'len(popdef(`_foreachq')_foreachq($@@))')dnl define(`debug', `pushdef(`_foreachq', defn(`foo'))') @result{} include(`loop.m4')dnl @result{}48894 @end example @end ignore @node Improved copy @section Solution for @code{copy} The macro @code{copy} presented above is unable to handle builtin tokens with M4 1.4.x, because it tries to pass the builtin token through the macro @code{curry}, where it is silently flattened to an empty string (@pxref{Composition}). Rather than using the problematic @code{curry} to work around the limitation that @code{stack_foreach} expects to invoke a macro that takes exactly one argument, we can write a new macro that lets us form the exact two-argument @code{pushdef} call sequence needed, so that we are no longer passing a builtin token through a text macro. @deffn Composite stack_foreach_sep (@var{macro}, @var{pre}, @var{post}, @ @var{sep}) @deffnx Composite stack_foreach_sep_lifo (@var{macro}, @var{pre}, @ @var{post}, @var{sep}) For each of the @code{pushdef} definitions associated with @var{macro}, expand the sequence @samp{@var{pre}`'definition`'@var{post}}. Additionally, expand @var{sep} between definitions. @code{stack_foreach_sep} visits the oldest definition first, while @code{stack_foreach_sep_lifo} visits the current definition first. The expansion may dereference @var{macro}, but should not modify it. There are a few special macros, such as @code{defn}, which cannot be used as the @var{macro} parameter. @end deffn Note that @code{stack_foreach(`@var{macro}', `@var{action}')} is equivalent to @code{stack_foreach_sep(`@var{macro}', `@var{action}(', `)')}. By supplying explicit parentheses, split among the @var{pre} and @var{post} arguments to @code{stack_foreach_sep}, it is now possible to construct macro calls with more than one argument, without passing builtin tokens through a macro call. It is likewise possible to directly reference the stack definitions without a macro call, by leaving @var{pre} and @var{post} empty. Thus, in addition to fixing @code{copy} on builtin tokens, it also executes with fewer macro invocations. The new macro also adds a separator that is only output after the first iteration of the helper @code{_stack_reverse_sep}, implemented by prepending the original @var{sep} to @var{pre} and omitting a @var{sep} argument in subsequent iterations. Note that the empty string that separates @var{sep} from @var{pre} is provided as part of the fourth argument when originally calling @code{_stack_reverse_sep}, and not by writing @code{$4`'$3} as the third argument in the recursive call; while the other approach would give the same output, it does so at the expense of increasing the argument size on each iteration of @code{_stack_reverse_sep}, which results in quadratic instead of linear execution time. The improved stack walking macros are available in @file{m4-@value{VERSION}/@/examples/@/stack_sep.m4}: @comment examples @example $ @kbd{m4 -I examples} include(`stack_sep.m4') @result{} define(`copy', `ifdef(`$2', `errprint(`$2 already defined ')m4exit(`1')', `stack_foreach_sep(`$1', `pushdef(`$2',', `)')')')dnl pushdef(`a', `1')pushdef(`a', defn(`divnum')) @result{} copy(`a', `b') @result{} b @result{}0 popdef(`b') @result{} b @result{}1 pushdef(`c', `1')pushdef(`c', `2') @result{} stack_foreach_sep_lifo(`c', `', `', `, ') @result{}2, 1 undivert(`stack_sep.m4')dnl @result{}divert(`-1') @result{}# stack_foreach_sep(macro, pre, post, sep) @result{}# Invoke PRE`'defn`'POST with a single argument of each definition @result{}# from the definition stack of MACRO, starting with the oldest, and @result{}# separated by SEP between definitions. @result{}define(`stack_foreach_sep', @result{}`_stack_reverse_sep(`$1', `tmp-$1')'dnl @result{}`_stack_reverse_sep(`tmp-$1', `$1', `$2`'defn(`$1')$3', `$4`'')') @result{}# stack_foreach_sep_lifo(macro, pre, post, sep) @result{}# Like stack_foreach_sep, but starting with the newest definition. @result{}define(`stack_foreach_sep_lifo', @result{}`_stack_reverse_sep(`$1', `tmp-$1', `$2`'defn(`$1')$3', `$4`'')'dnl @result{}`_stack_reverse_sep(`tmp-$1', `$1')') @result{}define(`_stack_reverse_sep', @result{}`ifdef(`$1', `pushdef(`$2', defn(`$1'))$3`'popdef(`$1')$0( @result{} `$1', `$2', `$4$3')')') @result{}divert`'dnl @end example @ignore @comment Not worth putting in the manual, but make sure that @comment stack_foreach_sep has linear performance. @comment examples @example $ @kbd {m4 -I examples} include(`forloop3.m4')include(`stack_sep.m4')dnl forloop(`i', `1', `10000', `pushdef(`s', i)') @result{} define(`colon', `:')define(`dash', `-') @result{} len(stack_foreach_sep(`s', `dash', `', `colon')) @result{}58893 @end example @end ignore @node Improved m4wrap @section Solution for @code{m4wrap} The replacement @code{m4wrap} versions presented above, designed to guarantee FIFO or LIFO order regardless of the underlying M4 implementation, share a bug when dealing with wrapped text that looks like parameter expansion. Note how the invocation of @code{m4wrap@var{n}} interprets these parameters, while using the builtin preserves them for their intended use. @comment examples @example $ @kbd{m4 -I examples} include(`wraplifo.m4') @result{} m4wrap(`define(`foo', ``$0:'-$1-$*-$#-')foo(`a', `b') ') @result{} builtin(`m4wrap', ``'define(`bar', ``$0:'-$1-$*-$#-')bar(`a', `b') ') @result{} ^D @result{}bar:-a-a,b-2- @result{}m4wrap0:---0- @end example Additionally, the computation of @code{_m4wrap_level} and creation of multiple @code{m4wrap@var{n}} placeholders in the original examples is more expensive in time and memory than strictly necessary. Notice how the improved version grabs the wrapped text via @code{defn} to avoid parameter expansion, then undefines @code{_m4wrap_text}, before stripping a level of quotes with @code{_arg1} to expand the text. That way, each level of wrapping reuses the single placeholder, which starts each nesting level in an undefined state. Finally, it is worth emulating the GNU M4 extension of saving all arguments to @code{m4wrap}, separated by a space, rather than saving just the first argument. This is done with the @code{join} macro documented previously (@pxref{Shift}). The improved LIFO example is shipped as @file{m4-@value{VERSION}/@/examples/@/wraplifo2.m4}, and can easily be converted to a FIFO solution by swapping the adjacent invocations of @code{joinall} and @code{defn}. @comment examples @example $ @kbd{m4 -I examples} include(`wraplifo2.m4') @result{} undivert(`wraplifo2.m4')dnl @result{}dnl Redefine m4wrap to have LIFO semantics, improved example. @result{}include(`join.m4')dnl @result{}define(`_m4wrap', defn(`m4wrap'))dnl @result{}define(`_arg1', `$1')dnl @result{}define(`m4wrap', @result{}`ifdef(`_$0_text', @result{} `define(`_$0_text', joinall(` ', $@@)defn(`_$0_text'))', @result{} `_$0(`_arg1(defn(`_$0_text')undefine(`_$0_text'))')dnl @result{}define(`_$0_text', joinall(` ', $@@))')')dnl m4wrap(`define(`foo', ``$0:'-$1-$*-$#-')foo(`a', `b') ') @result{} m4wrap(`lifo text m4wrap(`nested', `', `$@@ ')') @result{} ^D @result{}lifo text @result{}foo:-a-a,b-2- @result{}nested $@@ @end example @node Improved cleardivert @section Solution for @code{cleardivert} The @code{cleardivert} macro (@pxref{Cleardivert}) cannot, as it stands, be called without arguments to clear all pending diversions. That is because using undivert with an empty string for an argument is different than using it with no arguments at all. Compare the earlier definition with one that takes the number of arguments into account: @example define(`cleardivert', `pushdef(`_n', divnum)divert(`-1')undivert($@@)divert(_n)popdef(`_n')') @result{} divert(`1')one divert @result{} cleardivert @result{} undivert @result{}one @result{} define(`cleardivert', `pushdef(`_num', divnum)divert(`-1')ifelse(`$#', `0', `undivert`'', `undivert($@@)')divert(_num)popdef(`_num')') @result{} divert(`2')two divert @result{} cleardivert @result{} undivert @result{} @end example @node Improved capitalize @section Solution for @code{capitalize} The @code{capitalize} macro (@pxref{Patsubst}) as presented earlier does not allow clients to follow the quoting rule of thumb. Consider the three macros @code{active}, @code{Active}, and @code{ACTIVE}, and the difference between calling @code{capitalize} with the expansion of a macro, expanding the result of a case change, and changing the case of a double-quoted string: @comment examples @example $ @kbd{m4 -I examples} include(`capitalize.m4')dnl define(`active', `act1, ive')dnl define(`Active', `Act2, Ive')dnl define(`ACTIVE', `ACT3, IVE')dnl upcase(active) @result{}ACT1,IVE upcase(`active') @result{}ACT3, IVE upcase(``active'') @result{}ACTIVE downcase(ACTIVE) @result{}act3,ive downcase(`ACTIVE') @result{}act1, ive downcase(``ACTIVE'') @result{}active capitalize(active) @result{}Act1 capitalize(`active') @result{}Active capitalize(``active'') @result{}_capitalize(`active') define(`A', `OOPS') @result{} capitalize(active) @result{}OOPSct1 capitalize(`active') @result{}OOPSctive @end example First, when @code{capitalize} is called with more than one argument, it was throwing away later arguments, whereas @code{upcase} and @code{downcase} used @samp{$*} to collect them all. The fix is simple: use @samp{$*} consistently. Next, with single-quoting, @code{capitalize} outputs a single character, a set of quotes, then the rest of the characters, making it impossible to invoke @code{Active} after the fact, and allowing the alternate macro @code{A} to interfere. Here, the solution is to use additional quoting in the helper macros, then pass the final over-quoted output string through @code{_arg1} to remove the extra quoting and finally invoke the concatenated portions as a single string. Finally, when passed a double-quoted string, the nested macro @code{_capitalize} is never invoked because it ended up nested inside quotes. This one is the toughest to fix. In short, we have no idea how many levels of quotes are in effect on the substring being altered by @code{patsubst}. If the replacement string cannot be expressed entirely in terms of literal text and backslash substitutions, then we need a mechanism to guarantee that the helper macros are invoked outside of quotes. In other words, this sounds like a job for @code{changequote} (@pxref{Changequote}). By changing the active quoting characters, we can guarantee that replacement text injected by @code{patsubst} always occurs in the middle of a string that has exactly one level of over-quoting using alternate quotes; so the replacement text closes the quoted string, invokes the helper macros, then reopens the quoted string. In turn, that means the replacement text has unbalanced quotes, necessitating another round of @code{changequote}. In the fixed version below, (also shipped as @file{m4-@value{VERSION}/@/examples/@/capitalize2.m4}), @code{capitalize} uses the alternate quotes of @samp{<<[} and @samp{]>>} (the longer strings are chosen so as to be less likely to appear in the text being converted). The helpers @code{_to_alt} and @code{_from_alt} merely reduce the number of characters required to perform a @code{changequote}, since the definition changes twice. The outermost pair means that @code{patsubst} and @code{_capitalize_alt} are invoked with alternate quoting; the innermost pair is used so that the third argument to @code{patsubst} can contain an unbalanced @samp{]>>}/@samp{<<[} pair. Note that @code{upcase} and @code{downcase} must be redefined as @code{_upcase_alt} and @code{_downcase_alt}, since they contain nested quotes but are invoked with the alternate quoting scheme in effect. @comment examples @example $ @kbd{m4 -I examples} include(`capitalize2.m4')dnl define(`active', `act1, ive')dnl define(`Active', `Act2, Ive')dnl define(`ACTIVE', `ACT3, IVE')dnl define(`A', `OOPS')dnl capitalize(active; `active'; ``active''; ```actIVE''') @result{}Act1,Ive; Act2, Ive; Active; `Active' undivert(`capitalize2.m4')dnl @result{}divert(`-1') @result{}# upcase(text) @result{}# downcase(text) @result{}# capitalize(text) @result{}# change case of text, improved version @result{}define(`upcase', `translit(`$*', `a-z', `A-Z')') @result{}define(`downcase', `translit(`$*', `A-Z', `a-z')') @result{}define(`_arg1', `$1') @result{}define(`_to_alt', `changequote(`<<[', `]>>')') @result{}define(`_from_alt', `changequote(<<[`]>>, <<[']>>)') @result{}define(`_upcase_alt', `translit(<<[$*]>>, <<[a-z]>>, <<[A-Z]>>)') @result{}define(`_downcase_alt', `translit(<<[$*]>>, <<[A-Z]>>, <<[a-z]>>)') @result{}define(`_capitalize_alt', @result{} `regexp(<<[$1]>>, <<[^\(\w\)\(\w*\)]>>, @result{} <<[_upcase_alt(<<[<<[\1]>>]>>)_downcase_alt(<<[<<[\2]>>]>>)]>>)') @result{}define(`capitalize', @result{} `_arg1(_to_alt()patsubst(<<[<<[$*]>>]>>, <<[\w+]>>, @result{} _from_alt()`]>>_$0_alt(<<[\&]>>)<<['_to_alt())_from_alt())') @result{}divert`'dnl @end example @node Improved fatal_error @section Solution for @code{fatal_error} The @code{fatal_error} macro (@pxref{M4exit}) is not robust to versions of GNU M4 earlier than 1.4.8, where invoking @code{@w{__file__}} (@pxref{Location}) inside @code{m4wrap} would result in an empty string, and @code{@w{__line__}} resulted in @samp{0} even though all files start at line 1. Furthermore, versions earlier than 1.4.6 did not support the @code{@w{__program__}} macro. If you want @code{fatal_error} to work across the entire 1.4.x release series, a better implementation would be: @comment status: 1 @example define(`fatal_error', `errprint(ifdef(`__program__', `__program__', ``m4'')'dnl `:ifelse(__line__, `0', `', `__file__:__line__:')` fatal error: $* ')m4exit(`1')') @result{} m4wrap(`divnum(`demo of internal message') fatal_error(`inside wrapped text')') @result{} ^D @error{}m4:stdin:6: Warning: excess arguments to builtin `divnum' ignored @result{}0 @error{}m4:stdin:6: fatal error: inside wrapped text @end example @c ========================================================== Appendices @node Copying This Package @appendix How to make copies of the overall M4 package @cindex License, code This appendix covers the license for copying the source code of the overall M4 package. This manual is under a different set of restrictions, covered later (@pxref{Copying This Manual}). @menu * GNU General Public License:: License for copying the M4 package @end menu @node GNU General Public License @appendixsec License for copying the M4 package @cindex GPL, GNU General Public License @cindex GNU General Public License @cindex General Public License (GPL), GNU @include gpl-3.0.texi @node Copying This Manual @appendix How to make copies of this manual @cindex License, manual This appendix covers the license for copying this manual. Note that some of the longer examples in this manual are also distributed in the directory @file{m4-@value{VERSION}/@/examples/}, where a more permissive license is in effect when copying just the examples. @menu * GNU Free Documentation License:: License for copying this manual @end menu @node GNU Free Documentation License @appendixsec License for copying this manual @cindex FDL, GNU Free Documentation License @cindex GNU Free Documentation License @cindex Free Documentation License (FDL), GNU @include fdl-1.3.texi @node Indices @appendix Indices of concepts and macros @menu * Macro index:: Index for all @code{m4} macros * Concept index:: Index for many concepts @end menu @node Macro index @appendixsec Index for all @code{m4} macros This index covers all @code{m4} builtins, as well as several useful composite macros. References are exclusively to the places where a macro is introduced the first time. @printindex fn @node Concept index @appendixsec Index for many concepts @printindex cp @bye @c Local Variables: @c coding: iso-8859-1 @c fill-column: 72 @c ispell-local-dictionary: "american" @c indent-tabs-mode: nil @c whitespace-check-buffer-indent: nil @c End: