xref: /netbsd-src/external/gpl3/gcc.old/dist/gcc/doc/cpp.texi (revision fdd524d4ccd2bb0c6f67401e938dabf773eb0372)
1\input texinfo
2@setfilename cpp.info
3@settitle The C Preprocessor
4@setchapternewpage off
5@c @smallbook
6@c @cropmarks
7@c @finalout
8
9@include gcc-common.texi
10
11@copying
12@c man begin COPYRIGHT
13Copyright @copyright{} 1987-2015 Free Software Foundation, Inc.
14
15Permission is granted to copy, distribute and/or modify this document
16under the terms of the GNU Free Documentation License, Version 1.3 or
17any later version published by the Free Software Foundation.  A copy of
18the license is included in the
19@c man end
20section entitled ``GNU Free Documentation License''.
21@ignore
22@c man begin COPYRIGHT
23man page gfdl(7).
24@c man end
25@end ignore
26
27@c man begin COPYRIGHT
28This manual contains no Invariant Sections.  The Front-Cover Texts are
29(a) (see below), and the Back-Cover Texts are (b) (see below).
30
31(a) The FSF's Front-Cover Text is:
32
33     A GNU Manual
34
35(b) The FSF's Back-Cover Text is:
36
37     You have freedom to copy and modify this GNU Manual, like GNU
38     software.  Copies published by the Free Software Foundation raise
39     funds for GNU development.
40@c man end
41@end copying
42
43@c Create a separate index for command line options.
44@defcodeindex op
45@syncodeindex vr op
46
47@c Used in cppopts.texi and cppenv.texi.
48@set cppmanual
49
50@ifinfo
51@dircategory Software development
52@direntry
53* Cpp: (cpp).                  The GNU C preprocessor.
54@end direntry
55@end ifinfo
56
57@titlepage
58@title The C Preprocessor
59@versionsubtitle
60@author Richard M. Stallman, Zachary Weinberg
61@page
62@c There is a fill at the bottom of the page, so we need a filll to
63@c override it.
64@vskip 0pt plus 1filll
65@insertcopying
66@end titlepage
67@contents
68@page
69
70@ifnottex
71@node Top
72@top
73The C preprocessor implements the macro language used to transform C,
74C++, and Objective-C programs before they are compiled.  It can also be
75useful on its own.
76
77@menu
78* Overview::
79* Header Files::
80* Macros::
81* Conditionals::
82* Diagnostics::
83* Line Control::
84* Pragmas::
85* Other Directives::
86* Preprocessor Output::
87* Traditional Mode::
88* Implementation Details::
89* Invocation::
90* Environment Variables::
91* GNU Free Documentation License::
92* Index of Directives::
93* Option Index::
94* Concept Index::
95
96@detailmenu
97 --- The Detailed Node Listing ---
98
99Overview
100
101* Character sets::
102* Initial processing::
103* Tokenization::
104* The preprocessing language::
105
106Header Files
107
108* Include Syntax::
109* Include Operation::
110* Search Path::
111* Once-Only Headers::
112* Alternatives to Wrapper #ifndef::
113* Computed Includes::
114* Wrapper Headers::
115* System Headers::
116
117Macros
118
119* Object-like Macros::
120* Function-like Macros::
121* Macro Arguments::
122* Stringification::
123* Concatenation::
124* Variadic Macros::
125* Predefined Macros::
126* Undefining and Redefining Macros::
127* Directives Within Macro Arguments::
128* Macro Pitfalls::
129
130Predefined Macros
131
132* Standard Predefined Macros::
133* Common Predefined Macros::
134* System-specific Predefined Macros::
135* C++ Named Operators::
136
137Macro Pitfalls
138
139* Misnesting::
140* Operator Precedence Problems::
141* Swallowing the Semicolon::
142* Duplication of Side Effects::
143* Self-Referential Macros::
144* Argument Prescan::
145* Newlines in Arguments::
146
147Conditionals
148
149* Conditional Uses::
150* Conditional Syntax::
151* Deleted Code::
152
153Conditional Syntax
154
155* Ifdef::
156* If::
157* Defined::
158* Else::
159* Elif::
160
161Implementation Details
162
163* Implementation-defined behavior::
164* Implementation limits::
165* Obsolete Features::
166* Differences from previous versions::
167
168Obsolete Features
169
170* Obsolete Features::
171
172@end detailmenu
173@end menu
174
175@insertcopying
176@end ifnottex
177
178@node Overview
179@chapter Overview
180@c man begin DESCRIPTION
181The C preprocessor, often known as @dfn{cpp}, is a @dfn{macro processor}
182that is used automatically by the C compiler to transform your program
183before compilation.  It is called a macro processor because it allows
184you to define @dfn{macros}, which are brief abbreviations for longer
185constructs.
186
187The C preprocessor is intended to be used only with C, C++, and
188Objective-C source code.  In the past, it has been abused as a general
189text processor.  It will choke on input which does not obey C's lexical
190rules.  For example, apostrophes will be interpreted as the beginning of
191character constants, and cause errors.  Also, you cannot rely on it
192preserving characteristics of the input which are not significant to
193C-family languages.  If a Makefile is preprocessed, all the hard tabs
194will be removed, and the Makefile will not work.
195
196Having said that, you can often get away with using cpp on things which
197are not C@.  Other Algol-ish programming languages are often safe
198(Pascal, Ada, etc.) So is assembly, with caution.  @option{-traditional-cpp}
199mode preserves more white space, and is otherwise more permissive.  Many
200of the problems can be avoided by writing C or C++ style comments
201instead of native language comments, and keeping macros simple.
202
203Wherever possible, you should use a preprocessor geared to the language
204you are writing in.  Modern versions of the GNU assembler have macro
205facilities.  Most high level programming languages have their own
206conditional compilation and inclusion mechanism.  If all else fails,
207try a true general text processor, such as GNU M4.
208
209C preprocessors vary in some details.  This manual discusses the GNU C
210preprocessor, which provides a small superset of the features of ISO
211Standard C@.  In its default mode, the GNU C preprocessor does not do a
212few things required by the standard.  These are features which are
213rarely, if ever, used, and may cause surprising changes to the meaning
214of a program which does not expect them.  To get strict ISO Standard C,
215you should use the @option{-std=c90}, @option{-std=c99} or
216@option{-std=c11} options, depending
217on which version of the standard you want.  To get all the mandatory
218diagnostics, you must also use @option{-pedantic}.  @xref{Invocation}.
219
220This manual describes the behavior of the ISO preprocessor.  To
221minimize gratuitous differences, where the ISO preprocessor's
222behavior does not conflict with traditional semantics, the
223traditional preprocessor should behave the same way.  The various
224differences that do exist are detailed in the section @ref{Traditional
225Mode}.
226
227For clarity, unless noted otherwise, references to @samp{CPP} in this
228manual refer to GNU CPP@.
229@c man end
230
231@menu
232* Character sets::
233* Initial processing::
234* Tokenization::
235* The preprocessing language::
236@end menu
237
238@node Character sets
239@section Character sets
240
241Source code character set processing in C and related languages is
242rather complicated.  The C standard discusses two character sets, but
243there are really at least four.
244
245The files input to CPP might be in any character set at all.  CPP's
246very first action, before it even looks for line boundaries, is to
247convert the file into the character set it uses for internal
248processing.  That set is what the C standard calls the @dfn{source}
249character set.  It must be isomorphic with ISO 10646, also known as
250Unicode.  CPP uses the UTF-8 encoding of Unicode.
251
252The character sets of the input files are specified using the
253@option{-finput-charset=} option.
254
255All preprocessing work (the subject of the rest of this manual) is
256carried out in the source character set.  If you request textual
257output from the preprocessor with the @option{-E} option, it will be
258in UTF-8.
259
260After preprocessing is complete, string and character constants are
261converted again, into the @dfn{execution} character set.  This
262character set is under control of the user; the default is UTF-8,
263matching the source character set.  Wide string and character
264constants have their own character set, which is not called out
265specifically in the standard.  Again, it is under control of the user.
266The default is UTF-16 or UTF-32, whichever fits in the target's
267@code{wchar_t} type, in the target machine's byte
268order.@footnote{UTF-16 does not meet the requirements of the C
269standard for a wide character set, but the choice of 16-bit
270@code{wchar_t} is enshrined in some system ABIs so we cannot fix
271this.}  Octal and hexadecimal escape sequences do not undergo
272conversion; @t{'\x12'} has the value 0x12 regardless of the currently
273selected execution character set.  All other escapes are replaced by
274the character in the source character set that they represent, then
275converted to the execution character set, just like unescaped
276characters.
277
278Unless the experimental @option{-fextended-identifiers} option is used,
279GCC does not permit the use of characters outside the ASCII range, nor
280@samp{\u} and @samp{\U} escapes, in identifiers.  Even with that
281option, characters outside the ASCII range can only be specified with
282the @samp{\u} and @samp{\U} escapes, not used directly in identifiers.
283
284@node Initial processing
285@section Initial processing
286
287The preprocessor performs a series of textual transformations on its
288input.  These happen before all other processing.  Conceptually, they
289happen in a rigid order, and the entire file is run through each
290transformation before the next one begins.  CPP actually does them
291all at once, for performance reasons.  These transformations correspond
292roughly to the first three ``phases of translation'' described in the C
293standard.
294
295@enumerate
296@item
297@cindex line endings
298The input file is read into memory and broken into lines.
299
300Different systems use different conventions to indicate the end of a
301line.  GCC accepts the ASCII control sequences @kbd{LF}, @kbd{@w{CR
302LF}} and @kbd{CR} as end-of-line markers.  These are the canonical
303sequences used by Unix, DOS and VMS, and the classic Mac OS (before
304OSX) respectively.  You may therefore safely copy source code written
305on any of those systems to a different one and use it without
306conversion.  (GCC may lose track of the current line number if a file
307doesn't consistently use one convention, as sometimes happens when it
308is edited on computers with different conventions that share a network
309file system.)
310
311If the last line of any input file lacks an end-of-line marker, the end
312of the file is considered to implicitly supply one.  The C standard says
313that this condition provokes undefined behavior, so GCC will emit a
314warning message.
315
316@item
317@cindex trigraphs
318@anchor{trigraphs}If trigraphs are enabled, they are replaced by their
319corresponding single characters.  By default GCC ignores trigraphs,
320but if you request a strictly conforming mode with the @option{-std}
321option, or you specify the @option{-trigraphs} option, then it
322converts them.
323
324These are nine three-character sequences, all starting with @samp{??},
325that are defined by ISO C to stand for single characters.  They permit
326obsolete systems that lack some of C's punctuation to use C@.  For
327example, @samp{??/} stands for @samp{\}, so @t{'??/n'} is a character
328constant for a newline.
329
330Trigraphs are not popular and many compilers implement them
331incorrectly.  Portable code should not rely on trigraphs being either
332converted or ignored.  With @option{-Wtrigraphs} GCC will warn you
333when a trigraph may change the meaning of your program if it were
334converted.  @xref{Wtrigraphs}.
335
336In a string constant, you can prevent a sequence of question marks
337from being confused with a trigraph by inserting a backslash between
338the question marks, or by separating the string literal at the
339trigraph and making use of string literal concatenation.  @t{"(??\?)"}
340is the string @samp{(???)}, not @samp{(?]}.  Traditional C compilers
341do not recognize these idioms.
342
343The nine trigraphs and their replacements are
344
345@smallexample
346Trigraph:       ??(  ??)  ??<  ??>  ??=  ??/  ??'  ??!  ??-
347Replacement:      [    ]    @{    @}    #    \    ^    |    ~
348@end smallexample
349
350@item
351@cindex continued lines
352@cindex backslash-newline
353Continued lines are merged into one long line.
354
355A continued line is a line which ends with a backslash, @samp{\}.  The
356backslash is removed and the following line is joined with the current
357one.  No space is inserted, so you may split a line anywhere, even in
358the middle of a word.  (It is generally more readable to split lines
359only at white space.)
360
361The trailing backslash on a continued line is commonly referred to as a
362@dfn{backslash-newline}.
363
364If there is white space between a backslash and the end of a line, that
365is still a continued line.  However, as this is usually the result of an
366editing mistake, and many compilers will not accept it as a continued
367line, GCC will warn you about it.
368
369@item
370@cindex comments
371@cindex line comments
372@cindex block comments
373All comments are replaced with single spaces.
374
375There are two kinds of comments.  @dfn{Block comments} begin with
376@samp{/*} and continue until the next @samp{*/}.  Block comments do not
377nest:
378
379@smallexample
380/* @r{this is} /* @r{one comment} */ @r{text outside comment}
381@end smallexample
382
383@dfn{Line comments} begin with @samp{//} and continue to the end of the
384current line.  Line comments do not nest either, but it does not matter,
385because they would end in the same place anyway.
386
387@smallexample
388// @r{this is} // @r{one comment}
389@r{text outside comment}
390@end smallexample
391@end enumerate
392
393It is safe to put line comments inside block comments, or vice versa.
394
395@smallexample
396@group
397/* @r{block comment}
398   // @r{contains line comment}
399   @r{yet more comment}
400 */ @r{outside comment}
401
402// @r{line comment} /* @r{contains block comment} */
403@end group
404@end smallexample
405
406But beware of commenting out one end of a block comment with a line
407comment.
408
409@smallexample
410@group
411 // @r{l.c.}  /* @r{block comment begins}
412    @r{oops! this isn't a comment anymore} */
413@end group
414@end smallexample
415
416Comments are not recognized within string literals.
417@t{@w{"/* blah */"}} is the string constant @samp{@w{/* blah */}}, not
418an empty string.
419
420Line comments are not in the 1989 edition of the C standard, but they
421are recognized by GCC as an extension.  In C++ and in the 1999 edition
422of the C standard, they are an official part of the language.
423
424Since these transformations happen before all other processing, you can
425split a line mechanically with backslash-newline anywhere.  You can
426comment out the end of a line.  You can continue a line comment onto the
427next line with backslash-newline.  You can even split @samp{/*},
428@samp{*/}, and @samp{//} onto multiple lines with backslash-newline.
429For example:
430
431@smallexample
432@group
433/\
434*
435*/ # /*
436*/ defi\
437ne FO\
438O 10\
43920
440@end group
441@end smallexample
442
443@noindent
444is equivalent to @code{@w{#define FOO 1020}}.  All these tricks are
445extremely confusing and should not be used in code intended to be
446readable.
447
448There is no way to prevent a backslash at the end of a line from being
449interpreted as a backslash-newline.  This cannot affect any correct
450program, however.
451
452@node Tokenization
453@section Tokenization
454
455@cindex tokens
456@cindex preprocessing tokens
457After the textual transformations are finished, the input file is
458converted into a sequence of @dfn{preprocessing tokens}.  These mostly
459correspond to the syntactic tokens used by the C compiler, but there are
460a few differences.  White space separates tokens; it is not itself a
461token of any kind.  Tokens do not have to be separated by white space,
462but it is often necessary to avoid ambiguities.
463
464When faced with a sequence of characters that has more than one possible
465tokenization, the preprocessor is greedy.  It always makes each token,
466starting from the left, as big as possible before moving on to the next
467token.  For instance, @code{a+++++b} is interpreted as
468@code{@w{a ++ ++ + b}}, not as @code{@w{a ++ + ++ b}}, even though the
469latter tokenization could be part of a valid C program and the former
470could not.
471
472Once the input file is broken into tokens, the token boundaries never
473change, except when the @samp{##} preprocessing operator is used to paste
474tokens together.  @xref{Concatenation}.  For example,
475
476@smallexample
477@group
478#define foo() bar
479foo()baz
480     @expansion{} bar baz
481@emph{not}
482     @expansion{} barbaz
483@end group
484@end smallexample
485
486The compiler does not re-tokenize the preprocessor's output.  Each
487preprocessing token becomes one compiler token.
488
489@cindex identifiers
490Preprocessing tokens fall into five broad classes: identifiers,
491preprocessing numbers, string literals, punctuators, and other.  An
492@dfn{identifier} is the same as an identifier in C: any sequence of
493letters, digits, or underscores, which begins with a letter or
494underscore.  Keywords of C have no significance to the preprocessor;
495they are ordinary identifiers.  You can define a macro whose name is a
496keyword, for instance.  The only identifier which can be considered a
497preprocessing keyword is @code{defined}.  @xref{Defined}.
498
499This is mostly true of other languages which use the C preprocessor.
500However, a few of the keywords of C++ are significant even in the
501preprocessor.  @xref{C++ Named Operators}.
502
503In the 1999 C standard, identifiers may contain letters which are not
504part of the ``basic source character set'', at the implementation's
505discretion (such as accented Latin letters, Greek letters, or Chinese
506ideograms).  This may be done with an extended character set, or the
507@samp{\u} and @samp{\U} escape sequences.  The implementation of this
508feature in GCC is experimental; such characters are only accepted in
509the @samp{\u} and @samp{\U} forms and only if
510@option{-fextended-identifiers} is used.
511
512As an extension, GCC treats @samp{$} as a letter.  This is for
513compatibility with some systems, such as VMS, where @samp{$} is commonly
514used in system-defined function and object names.  @samp{$} is not a
515letter in strictly conforming mode, or if you specify the @option{-$}
516option.  @xref{Invocation}.
517
518@cindex numbers
519@cindex preprocessing numbers
520A @dfn{preprocessing number} has a rather bizarre definition.  The
521category includes all the normal integer and floating point constants
522one expects of C, but also a number of other things one might not
523initially recognize as a number.  Formally, preprocessing numbers begin
524with an optional period, a required decimal digit, and then continue
525with any sequence of letters, digits, underscores, periods, and
526exponents.  Exponents are the two-character sequences @samp{e+},
527@samp{e-}, @samp{E+}, @samp{E-}, @samp{p+}, @samp{p-}, @samp{P+}, and
528@samp{P-}.  (The exponents that begin with @samp{p} or @samp{P} are new
529to C99.  They are used for hexadecimal floating-point constants.)
530
531The purpose of this unusual definition is to isolate the preprocessor
532from the full complexity of numeric constants.  It does not have to
533distinguish between lexically valid and invalid floating-point numbers,
534which is complicated.  The definition also permits you to split an
535identifier at any position and get exactly two tokens, which can then be
536pasted back together with the @samp{##} operator.
537
538It's possible for preprocessing numbers to cause programs to be
539misinterpreted.  For example, @code{0xE+12} is a preprocessing number
540which does not translate to any valid numeric constant, therefore a
541syntax error.  It does not mean @code{@w{0xE + 12}}, which is what you
542might have intended.
543
544@cindex string literals
545@cindex string constants
546@cindex character constants
547@cindex header file names
548@c the @: prevents makeinfo from turning '' into ".
549@dfn{String literals} are string constants, character constants, and
550header file names (the argument of @samp{#include}).@footnote{The C
551standard uses the term @dfn{string literal} to refer only to what we are
552calling @dfn{string constants}.}  String constants and character
553constants are straightforward: @t{"@dots{}"} or @t{'@dots{}'}.  In
554either case embedded quotes should be escaped with a backslash:
555@t{'\'@:'} is the character constant for @samp{'}.  There is no limit on
556the length of a character constant, but the value of a character
557constant that contains more than one character is
558implementation-defined.  @xref{Implementation Details}.
559
560Header file names either look like string constants, @t{"@dots{}"}, or are
561written with angle brackets instead, @t{<@dots{}>}.  In either case,
562backslash is an ordinary character.  There is no way to escape the
563closing quote or angle bracket.  The preprocessor looks for the header
564file in different places depending on which form you use.  @xref{Include
565Operation}.
566
567No string literal may extend past the end of a line.  Older versions
568of GCC accepted multi-line string constants.  You may use continued
569lines instead, or string constant concatenation.  @xref{Differences
570from previous versions}.
571
572@cindex punctuators
573@cindex digraphs
574@cindex alternative tokens
575@dfn{Punctuators} are all the usual bits of punctuation which are
576meaningful to C and C++.  All but three of the punctuation characters in
577ASCII are C punctuators.  The exceptions are @samp{@@}, @samp{$}, and
578@samp{`}.  In addition, all the two- and three-character operators are
579punctuators.  There are also six @dfn{digraphs}, which the C++ standard
580calls @dfn{alternative tokens}, which are merely alternate ways to spell
581other punctuators.  This is a second attempt to work around missing
582punctuation in obsolete systems.  It has no negative side effects,
583unlike trigraphs, but does not cover as much ground.  The digraphs and
584their corresponding normal punctuators are:
585
586@smallexample
587Digraph:        <%  %>  <:  :>  %:  %:%:
588Punctuator:      @{   @}   [   ]   #    ##
589@end smallexample
590
591@cindex other tokens
592Any other single character is considered ``other''.  It is passed on to
593the preprocessor's output unmolested.  The C compiler will almost
594certainly reject source code containing ``other'' tokens.  In ASCII, the
595only other characters are @samp{@@}, @samp{$}, @samp{`}, and control
596characters other than NUL (all bits zero).  (Note that @samp{$} is
597normally considered a letter.)  All characters with the high bit set
598(numeric range 0x7F--0xFF) are also ``other'' in the present
599implementation.  This will change when proper support for international
600character sets is added to GCC@.
601
602NUL is a special case because of the high probability that its
603appearance is accidental, and because it may be invisible to the user
604(many terminals do not display NUL at all).  Within comments, NULs are
605silently ignored, just as any other character would be.  In running
606text, NUL is considered white space.  For example, these two directives
607have the same meaning.
608
609@smallexample
610#define X^@@1
611#define X 1
612@end smallexample
613
614@noindent
615(where @samp{^@@} is ASCII NUL)@.  Within string or character constants,
616NULs are preserved.  In the latter two cases the preprocessor emits a
617warning message.
618
619@node The preprocessing language
620@section The preprocessing language
621@cindex directives
622@cindex preprocessing directives
623@cindex directive line
624@cindex directive name
625
626After tokenization, the stream of tokens may simply be passed straight
627to the compiler's parser.  However, if it contains any operations in the
628@dfn{preprocessing language}, it will be transformed first.  This stage
629corresponds roughly to the standard's ``translation phase 4'' and is
630what most people think of as the preprocessor's job.
631
632The preprocessing language consists of @dfn{directives} to be executed
633and @dfn{macros} to be expanded.  Its primary capabilities are:
634
635@itemize @bullet
636@item
637Inclusion of header files.  These are files of declarations that can be
638substituted into your program.
639
640@item
641Macro expansion.  You can define @dfn{macros}, which are abbreviations
642for arbitrary fragments of C code.  The preprocessor will replace the
643macros with their definitions throughout the program.  Some macros are
644automatically defined for you.
645
646@item
647Conditional compilation.  You can include or exclude parts of the
648program according to various conditions.
649
650@item
651Line control.  If you use a program to combine or rearrange source files
652into an intermediate file which is then compiled, you can use line
653control to inform the compiler where each source line originally came
654from.
655
656@item
657Diagnostics.  You can detect problems at compile time and issue errors
658or warnings.
659@end itemize
660
661There are a few more, less useful, features.
662
663Except for expansion of predefined macros, all these operations are
664triggered with @dfn{preprocessing directives}.  Preprocessing directives
665are lines in your program that start with @samp{#}.  Whitespace is
666allowed before and after the @samp{#}.  The @samp{#} is followed by an
667identifier, the @dfn{directive name}.  It specifies the operation to
668perform.  Directives are commonly referred to as @samp{#@var{name}}
669where @var{name} is the directive name.  For example, @samp{#define} is
670the directive that defines a macro.
671
672The @samp{#} which begins a directive cannot come from a macro
673expansion.  Also, the directive name is not macro expanded.  Thus, if
674@code{foo} is defined as a macro expanding to @code{define}, that does
675not make @samp{#foo} a valid preprocessing directive.
676
677The set of valid directive names is fixed.  Programs cannot define new
678preprocessing directives.
679
680Some directives require arguments; these make up the rest of the
681directive line and must be separated from the directive name by
682whitespace.  For example, @samp{#define} must be followed by a macro
683name and the intended expansion of the macro.
684
685A preprocessing directive cannot cover more than one line.  The line
686may, however, be continued with backslash-newline, or by a block comment
687which extends past the end of the line.  In either case, when the
688directive is processed, the continuations have already been merged with
689the first line to make one long line.
690
691@node Header Files
692@chapter Header Files
693
694@cindex header file
695A header file is a file containing C declarations and macro definitions
696(@pxref{Macros}) to be shared between several source files.  You request
697the use of a header file in your program by @dfn{including} it, with the
698C preprocessing directive @samp{#include}.
699
700Header files serve two purposes.
701
702@itemize @bullet
703@item
704@cindex system header files
705System header files declare the interfaces to parts of the operating
706system.  You include them in your program to supply the definitions and
707declarations you need to invoke system calls and libraries.
708
709@item
710Your own header files contain declarations for interfaces between the
711source files of your program.  Each time you have a group of related
712declarations and macro definitions all or most of which are needed in
713several different source files, it is a good idea to create a header
714file for them.
715@end itemize
716
717Including a header file produces the same results as copying the header
718file into each source file that needs it.  Such copying would be
719time-consuming and error-prone.  With a header file, the related
720declarations appear in only one place.  If they need to be changed, they
721can be changed in one place, and programs that include the header file
722will automatically use the new version when next recompiled.  The header
723file eliminates the labor of finding and changing all the copies as well
724as the risk that a failure to find one copy will result in
725inconsistencies within a program.
726
727In C, the usual convention is to give header files names that end with
728@file{.h}.  It is most portable to use only letters, digits, dashes, and
729underscores in header file names, and at most one dot.
730
731@menu
732* Include Syntax::
733* Include Operation::
734* Search Path::
735* Once-Only Headers::
736* Alternatives to Wrapper #ifndef::
737* Computed Includes::
738* Wrapper Headers::
739* System Headers::
740@end menu
741
742@node Include Syntax
743@section Include Syntax
744
745@findex #include
746Both user and system header files are included using the preprocessing
747directive @samp{#include}.  It has two variants:
748
749@table @code
750@item #include <@var{file}>
751This variant is used for system header files.  It searches for a file
752named @var{file} in a standard list of system directories.  You can prepend
753directories to this list with the @option{-I} option (@pxref{Invocation}).
754
755@item #include "@var{file}"
756This variant is used for header files of your own program.  It
757searches for a file named @var{file} first in the directory containing
758the current file, then in the quote directories and then the same
759directories used for @code{<@var{file}>}.  You can prepend directories
760to the list of quote directories with the @option{-iquote} option.
761@end table
762
763The argument of @samp{#include}, whether delimited with quote marks or
764angle brackets, behaves like a string constant in that comments are not
765recognized, and macro names are not expanded.  Thus, @code{@w{#include
766<x/*y>}} specifies inclusion of a system header file named @file{x/*y}.
767
768However, if backslashes occur within @var{file}, they are considered
769ordinary text characters, not escape characters.  None of the character
770escape sequences appropriate to string constants in C are processed.
771Thus, @code{@w{#include "x\n\\y"}} specifies a filename containing three
772backslashes.  (Some systems interpret @samp{\} as a pathname separator.
773All of these also interpret @samp{/} the same way.  It is most portable
774to use only @samp{/}.)
775
776It is an error if there is anything (other than comments) on the line
777after the file name.
778
779@node Include Operation
780@section Include Operation
781
782The @samp{#include} directive works by directing the C preprocessor to
783scan the specified file as input before continuing with the rest of the
784current file.  The output from the preprocessor contains the output
785already generated, followed by the output resulting from the included
786file, followed by the output that comes from the text after the
787@samp{#include} directive.  For example, if you have a header file
788@file{header.h} as follows,
789
790@smallexample
791char *test (void);
792@end smallexample
793
794@noindent
795and a main program called @file{program.c} that uses the header file,
796like this,
797
798@smallexample
799int x;
800#include "header.h"
801
802int
803main (void)
804@{
805  puts (test ());
806@}
807@end smallexample
808
809@noindent
810the compiler will see the same token stream as it would if
811@file{program.c} read
812
813@smallexample
814int x;
815char *test (void);
816
817int
818main (void)
819@{
820  puts (test ());
821@}
822@end smallexample
823
824Included files are not limited to declarations and macro definitions;
825those are merely the typical uses.  Any fragment of a C program can be
826included from another file.  The include file could even contain the
827beginning of a statement that is concluded in the containing file, or
828the end of a statement that was started in the including file.  However,
829an included file must consist of complete tokens.  Comments and string
830literals which have not been closed by the end of an included file are
831invalid.  For error recovery, they are considered to end at the end of
832the file.
833
834To avoid confusion, it is best if header files contain only complete
835syntactic units---function declarations or definitions, type
836declarations, etc.
837
838The line following the @samp{#include} directive is always treated as a
839separate line by the C preprocessor, even if the included file lacks a
840final newline.
841
842@node Search Path
843@section Search Path
844
845GCC looks in several different places for headers.  On a normal Unix
846system, if you do not instruct it otherwise, it will look for headers
847requested with @code{@w{#include <@var{file}>}} in:
848
849@smallexample
850/usr/local/include
851@var{libdir}/gcc/@var{target}/@var{version}/include
852/usr/@var{target}/include
853/usr/include
854@end smallexample
855
856For C++ programs, it will also look in
857@file{@var{libdir}/../include/c++/@var{version}},
858first.  In the above, @var{target} is the canonical name of the system
859GCC was configured to compile code for; often but not always the same as
860the canonical name of the system it runs on.  @var{version} is the
861version of GCC in use.
862
863You can add to this list with the @option{-I@var{dir}} command line
864option.  All the directories named by @option{-I} are searched, in
865left-to-right order, @emph{before} the default directories.  The only
866exception is when @file{dir} is already searched by default.  In
867this case, the option is ignored and the search order for system
868directories remains unchanged.
869
870Duplicate directories are removed from the quote and bracket search
871chains before the two chains are merged to make the final search chain.
872Thus, it is possible for a directory to occur twice in the final search
873chain if it was specified in both the quote and bracket chains.
874
875You can prevent GCC from searching any of the default directories with
876the @option{-nostdinc} option.  This is useful when you are compiling an
877operating system kernel or some other program that does not use the
878standard C library facilities, or the standard C library itself.
879@option{-I} options are not ignored as described above when
880@option{-nostdinc} is in effect.
881
882GCC looks for headers requested with @code{@w{#include "@var{file}"}}
883first in the directory containing the current file, then in the
884directories as specified by @option{-iquote} options, then in the same
885places it would have looked for a header requested with angle
886brackets.  For example, if @file{/usr/include/sys/stat.h} contains
887@code{@w{#include "types.h"}}, GCC looks for @file{types.h} first in
888@file{/usr/include/sys}, then in its usual search path.
889
890@samp{#line} (@pxref{Line Control}) does not change GCC's idea of the
891directory containing the current file.
892
893You may put @option{-I-} at any point in your list of @option{-I} options.
894This has two effects.  First, directories appearing before the
895@option{-I-} in the list are searched only for headers requested with
896quote marks.  Directories after @option{-I-} are searched for all
897headers.  Second, the directory containing the current file is not
898searched for anything, unless it happens to be one of the directories
899named by an @option{-I} switch.  @option{-I-} is deprecated, @option{-iquote}
900should be used instead.
901
902@option{-I. -I-} is not the same as no @option{-I} options at all, and does
903not cause the same behavior for @samp{<>} includes that @samp{""}
904includes get with no special options.  @option{-I.} searches the
905compiler's current working directory for header files.  That may or may
906not be the same as the directory containing the current file.
907
908If you need to look for headers in a directory named @file{-}, write
909@option{-I./-}.
910
911There are several more ways to adjust the header search path.  They are
912generally less useful.  @xref{Invocation}.
913
914@node Once-Only Headers
915@section Once-Only Headers
916@cindex repeated inclusion
917@cindex including just once
918@cindex wrapper @code{#ifndef}
919
920If a header file happens to be included twice, the compiler will process
921its contents twice.  This is very likely to cause an error, e.g.@: when the
922compiler sees the same structure definition twice.  Even if it does not,
923it will certainly waste time.
924
925The standard way to prevent this is to enclose the entire real contents
926of the file in a conditional, like this:
927
928@smallexample
929@group
930/* File foo.  */
931#ifndef FILE_FOO_SEEN
932#define FILE_FOO_SEEN
933
934@var{the entire file}
935
936#endif /* !FILE_FOO_SEEN */
937@end group
938@end smallexample
939
940This construct is commonly known as a @dfn{wrapper #ifndef}.
941When the header is included again, the conditional will be false,
942because @code{FILE_FOO_SEEN} is defined.  The preprocessor will skip
943over the entire contents of the file, and the compiler will not see it
944twice.
945
946CPP optimizes even further.  It remembers when a header file has a
947wrapper @samp{#ifndef}.  If a subsequent @samp{#include} specifies that
948header, and the macro in the @samp{#ifndef} is still defined, it does
949not bother to rescan the file at all.
950
951You can put comments outside the wrapper.  They will not interfere with
952this optimization.
953
954@cindex controlling macro
955@cindex guard macro
956The macro @code{FILE_FOO_SEEN} is called the @dfn{controlling macro} or
957@dfn{guard macro}.  In a user header file, the macro name should not
958begin with @samp{_}.  In a system header file, it should begin with
959@samp{__} to avoid conflicts with user programs.  In any kind of header
960file, the macro name should contain the name of the file and some
961additional text, to avoid conflicts with other header files.
962
963@node Alternatives to Wrapper #ifndef
964@section Alternatives to Wrapper #ifndef
965
966CPP supports two more ways of indicating that a header file should be
967read only once.  Neither one is as portable as a wrapper @samp{#ifndef}
968and we recommend you do not use them in new programs, with the caveat
969that @samp{#import} is standard practice in Objective-C.
970
971@findex #import
972CPP supports a variant of @samp{#include} called @samp{#import} which
973includes a file, but does so at most once.  If you use @samp{#import}
974instead of @samp{#include}, then you don't need the conditionals
975inside the header file to prevent multiple inclusion of the contents.
976@samp{#import} is standard in Objective-C, but is considered a
977deprecated extension in C and C++.
978
979@samp{#import} is not a well designed feature.  It requires the users of
980a header file to know that it should only be included once.  It is much
981better for the header file's implementor to write the file so that users
982don't need to know this.  Using a wrapper @samp{#ifndef} accomplishes
983this goal.
984
985In the present implementation, a single use of @samp{#import} will
986prevent the file from ever being read again, by either @samp{#import} or
987@samp{#include}.  You should not rely on this; do not use both
988@samp{#import} and @samp{#include} to refer to the same header file.
989
990Another way to prevent a header file from being included more than once
991is with the @samp{#pragma once} directive.  If @samp{#pragma once} is
992seen when scanning a header file, that file will never be read again, no
993matter what.
994
995@samp{#pragma once} does not have the problems that @samp{#import} does,
996but it is not recognized by all preprocessors, so you cannot rely on it
997in a portable program.
998
999@node Computed Includes
1000@section Computed Includes
1001@cindex computed includes
1002@cindex macros in include
1003
1004Sometimes it is necessary to select one of several different header
1005files to be included into your program.  They might specify
1006configuration parameters to be used on different sorts of operating
1007systems, for instance.  You could do this with a series of conditionals,
1008
1009@smallexample
1010#if SYSTEM_1
1011# include "system_1.h"
1012#elif SYSTEM_2
1013# include "system_2.h"
1014#elif SYSTEM_3
1015@dots{}
1016#endif
1017@end smallexample
1018
1019That rapidly becomes tedious.  Instead, the preprocessor offers the
1020ability to use a macro for the header name.  This is called a
1021@dfn{computed include}.  Instead of writing a header name as the direct
1022argument of @samp{#include}, you simply put a macro name there instead:
1023
1024@smallexample
1025#define SYSTEM_H "system_1.h"
1026@dots{}
1027#include SYSTEM_H
1028@end smallexample
1029
1030@noindent
1031@code{SYSTEM_H} will be expanded, and the preprocessor will look for
1032@file{system_1.h} as if the @samp{#include} had been written that way
1033originally.  @code{SYSTEM_H} could be defined by your Makefile with a
1034@option{-D} option.
1035
1036You must be careful when you define the macro.  @samp{#define} saves
1037tokens, not text.  The preprocessor has no way of knowing that the macro
1038will be used as the argument of @samp{#include}, so it generates
1039ordinary tokens, not a header name.  This is unlikely to cause problems
1040if you use double-quote includes, which are close enough to string
1041constants.  If you use angle brackets, however, you may have trouble.
1042
1043The syntax of a computed include is actually a bit more general than the
1044above.  If the first non-whitespace character after @samp{#include} is
1045not @samp{"} or @samp{<}, then the entire line is macro-expanded
1046like running text would be.
1047
1048If the line expands to a single string constant, the contents of that
1049string constant are the file to be included.  CPP does not re-examine the
1050string for embedded quotes, but neither does it process backslash
1051escapes in the string.  Therefore
1052
1053@smallexample
1054#define HEADER "a\"b"
1055#include HEADER
1056@end smallexample
1057
1058@noindent
1059looks for a file named @file{a\"b}.  CPP searches for the file according
1060to the rules for double-quoted includes.
1061
1062If the line expands to a token stream beginning with a @samp{<} token
1063and including a @samp{>} token, then the tokens between the @samp{<} and
1064the first @samp{>} are combined to form the filename to be included.
1065Any whitespace between tokens is reduced to a single space; then any
1066space after the initial @samp{<} is retained, but a trailing space
1067before the closing @samp{>} is ignored.  CPP searches for the file
1068according to the rules for angle-bracket includes.
1069
1070In either case, if there are any tokens on the line after the file name,
1071an error occurs and the directive is not processed.  It is also an error
1072if the result of expansion does not match either of the two expected
1073forms.
1074
1075These rules are implementation-defined behavior according to the C
1076standard.  To minimize the risk of different compilers interpreting your
1077computed includes differently, we recommend you use only a single
1078object-like macro which expands to a string constant.  This will also
1079minimize confusion for people reading your program.
1080
1081@node Wrapper Headers
1082@section Wrapper Headers
1083@cindex wrapper headers
1084@cindex overriding a header file
1085@findex #include_next
1086
1087Sometimes it is necessary to adjust the contents of a system-provided
1088header file without editing it directly.  GCC's @command{fixincludes}
1089operation does this, for example.  One way to do that would be to create
1090a new header file with the same name and insert it in the search path
1091before the original header.  That works fine as long as you're willing
1092to replace the old header entirely.  But what if you want to refer to
1093the old header from the new one?
1094
1095You cannot simply include the old header with @samp{#include}.  That
1096will start from the beginning, and find your new header again.  If your
1097header is not protected from multiple inclusion (@pxref{Once-Only
1098Headers}), it will recurse infinitely and cause a fatal error.
1099
1100You could include the old header with an absolute pathname:
1101@smallexample
1102#include "/usr/include/old-header.h"
1103@end smallexample
1104@noindent
1105This works, but is not clean; should the system headers ever move, you
1106would have to edit the new headers to match.
1107
1108There is no way to solve this problem within the C standard, but you can
1109use the GNU extension @samp{#include_next}.  It means, ``Include the
1110@emph{next} file with this name''.  This directive works like
1111@samp{#include} except in searching for the specified file: it starts
1112searching the list of header file directories @emph{after} the directory
1113in which the current file was found.
1114
1115Suppose you specify @option{-I /usr/local/include}, and the list of
1116directories to search also includes @file{/usr/include}; and suppose
1117both directories contain @file{signal.h}.  Ordinary @code{@w{#include
1118<signal.h>}} finds the file under @file{/usr/local/include}.  If that
1119file contains @code{@w{#include_next <signal.h>}}, it starts searching
1120after that directory, and finds the file in @file{/usr/include}.
1121
1122@samp{#include_next} does not distinguish between @code{<@var{file}>}
1123and @code{"@var{file}"} inclusion, nor does it check that the file you
1124specify has the same name as the current file.  It simply looks for the
1125file named, starting with the directory in the search path after the one
1126where the current file was found.
1127
1128The use of @samp{#include_next} can lead to great confusion.  We
1129recommend it be used only when there is no other alternative.  In
1130particular, it should not be used in the headers belonging to a specific
1131program; it should be used only to make global corrections along the
1132lines of @command{fixincludes}.
1133
1134@node System Headers
1135@section System Headers
1136@cindex system header files
1137
1138The header files declaring interfaces to the operating system and
1139runtime libraries often cannot be written in strictly conforming C@.
1140Therefore, GCC gives code found in @dfn{system headers} special
1141treatment.  All warnings, other than those generated by @samp{#warning}
1142(@pxref{Diagnostics}), are suppressed while GCC is processing a system
1143header.  Macros defined in a system header are immune to a few warnings
1144wherever they are expanded.  This immunity is granted on an ad-hoc
1145basis, when we find that a warning generates lots of false positives
1146because of code in macros defined in system headers.
1147
1148Normally, only the headers found in specific directories are considered
1149system headers.  These directories are determined when GCC is compiled.
1150There are, however, two ways to make normal headers into system headers.
1151
1152The @option{-isystem} command line option adds its argument to the list of
1153directories to search for headers, just like @option{-I}.  Any headers
1154found in that directory will be considered system headers.
1155
1156All directories named by @option{-isystem} are searched @emph{after} all
1157directories named by @option{-I}, no matter what their order was on the
1158command line.  If the same directory is named by both @option{-I} and
1159@option{-isystem}, the @option{-I} option is ignored.  GCC provides an
1160informative message when this occurs if @option{-v} is used.
1161
1162The @option{-cxx-isystem} command line option adds its argument to the
1163list of C++ system headers, similar to @option{-isystem} for C headers.
1164
1165@findex #pragma GCC system_header
1166There is also a directive, @code{@w{#pragma GCC system_header}}, which
1167tells GCC to consider the rest of the current include file a system
1168header, no matter where it was found.  Code that comes before the
1169@samp{#pragma} in the file will not be affected.  @code{@w{#pragma GCC
1170system_header}} has no effect in the primary source file.
1171
1172On very old systems, some of the pre-defined system header directories
1173get even more special treatment.  GNU C++ considers code in headers
1174found in those directories to be surrounded by an @code{@w{extern "C"}}
1175block.  There is no way to request this behavior with a @samp{#pragma},
1176or from the command line.
1177
1178@node Macros
1179@chapter Macros
1180
1181A @dfn{macro} is a fragment of code which has been given a name.
1182Whenever the name is used, it is replaced by the contents of the macro.
1183There are two kinds of macros.  They differ mostly in what they look
1184like when they are used.  @dfn{Object-like} macros resemble data objects
1185when used, @dfn{function-like} macros resemble function calls.
1186
1187You may define any valid identifier as a macro, even if it is a C
1188keyword.  The preprocessor does not know anything about keywords.  This
1189can be useful if you wish to hide a keyword such as @code{const} from an
1190older compiler that does not understand it.  However, the preprocessor
1191operator @code{defined} (@pxref{Defined}) can never be defined as a
1192macro, and C++'s named operators (@pxref{C++ Named Operators}) cannot be
1193macros when you are compiling C++.
1194
1195@menu
1196* Object-like Macros::
1197* Function-like Macros::
1198* Macro Arguments::
1199* Stringification::
1200* Concatenation::
1201* Variadic Macros::
1202* Predefined Macros::
1203* Undefining and Redefining Macros::
1204* Directives Within Macro Arguments::
1205* Macro Pitfalls::
1206@end menu
1207
1208@node Object-like Macros
1209@section Object-like Macros
1210@cindex object-like macro
1211@cindex symbolic constants
1212@cindex manifest constants
1213
1214An @dfn{object-like macro} is a simple identifier which will be replaced
1215by a code fragment.  It is called object-like because it looks like a
1216data object in code that uses it.  They are most commonly used to give
1217symbolic names to numeric constants.
1218
1219@findex #define
1220You create macros with the @samp{#define} directive.  @samp{#define} is
1221followed by the name of the macro and then the token sequence it should
1222be an abbreviation for, which is variously referred to as the macro's
1223@dfn{body}, @dfn{expansion} or @dfn{replacement list}.  For example,
1224
1225@smallexample
1226#define BUFFER_SIZE 1024
1227@end smallexample
1228
1229@noindent
1230defines a macro named @code{BUFFER_SIZE} as an abbreviation for the
1231token @code{1024}.  If somewhere after this @samp{#define} directive
1232there comes a C statement of the form
1233
1234@smallexample
1235foo = (char *) malloc (BUFFER_SIZE);
1236@end smallexample
1237
1238@noindent
1239then the C preprocessor will recognize and @dfn{expand} the macro
1240@code{BUFFER_SIZE}.  The C compiler will see the same tokens as it would
1241if you had written
1242
1243@smallexample
1244foo = (char *) malloc (1024);
1245@end smallexample
1246
1247By convention, macro names are written in uppercase.  Programs are
1248easier to read when it is possible to tell at a glance which names are
1249macros.
1250
1251The macro's body ends at the end of the @samp{#define} line.  You may
1252continue the definition onto multiple lines, if necessary, using
1253backslash-newline.  When the macro is expanded, however, it will all
1254come out on one line.  For example,
1255
1256@smallexample
1257#define NUMBERS 1, \
1258                2, \
1259                3
1260int x[] = @{ NUMBERS @};
1261     @expansion{} int x[] = @{ 1, 2, 3 @};
1262@end smallexample
1263
1264@noindent
1265The most common visible consequence of this is surprising line numbers
1266in error messages.
1267
1268There is no restriction on what can go in a macro body provided it
1269decomposes into valid preprocessing tokens.  Parentheses need not
1270balance, and the body need not resemble valid C code.  (If it does not,
1271you may get error messages from the C compiler when you use the macro.)
1272
1273The C preprocessor scans your program sequentially.  Macro definitions
1274take effect at the place you write them.  Therefore, the following input
1275to the C preprocessor
1276
1277@smallexample
1278foo = X;
1279#define X 4
1280bar = X;
1281@end smallexample
1282
1283@noindent
1284produces
1285
1286@smallexample
1287foo = X;
1288bar = 4;
1289@end smallexample
1290
1291When the preprocessor expands a macro name, the macro's expansion
1292replaces the macro invocation, then the expansion is examined for more
1293macros to expand.  For example,
1294
1295@smallexample
1296@group
1297#define TABLESIZE BUFSIZE
1298#define BUFSIZE 1024
1299TABLESIZE
1300     @expansion{} BUFSIZE
1301     @expansion{} 1024
1302@end group
1303@end smallexample
1304
1305@noindent
1306@code{TABLESIZE} is expanded first to produce @code{BUFSIZE}, then that
1307macro is expanded to produce the final result, @code{1024}.
1308
1309Notice that @code{BUFSIZE} was not defined when @code{TABLESIZE} was
1310defined.  The @samp{#define} for @code{TABLESIZE} uses exactly the
1311expansion you specify---in this case, @code{BUFSIZE}---and does not
1312check to see whether it too contains macro names.  Only when you
1313@emph{use} @code{TABLESIZE} is the result of its expansion scanned for
1314more macro names.
1315
1316This makes a difference if you change the definition of @code{BUFSIZE}
1317at some point in the source file.  @code{TABLESIZE}, defined as shown,
1318will always expand using the definition of @code{BUFSIZE} that is
1319currently in effect:
1320
1321@smallexample
1322#define BUFSIZE 1020
1323#define TABLESIZE BUFSIZE
1324#undef BUFSIZE
1325#define BUFSIZE 37
1326@end smallexample
1327
1328@noindent
1329Now @code{TABLESIZE} expands (in two stages) to @code{37}.
1330
1331If the expansion of a macro contains its own name, either directly or
1332via intermediate macros, it is not expanded again when the expansion is
1333examined for more macros.  This prevents infinite recursion.
1334@xref{Self-Referential Macros}, for the precise details.
1335
1336@node Function-like Macros
1337@section Function-like Macros
1338@cindex function-like macros
1339
1340You can also define macros whose use looks like a function call.  These
1341are called @dfn{function-like macros}.  To define a function-like macro,
1342you use the same @samp{#define} directive, but you put a pair of
1343parentheses immediately after the macro name.  For example,
1344
1345@smallexample
1346#define lang_init()  c_init()
1347lang_init()
1348     @expansion{} c_init()
1349@end smallexample
1350
1351A function-like macro is only expanded if its name appears with a pair
1352of parentheses after it.  If you write just the name, it is left alone.
1353This can be useful when you have a function and a macro of the same
1354name, and you wish to use the function sometimes.
1355
1356@smallexample
1357extern void foo(void);
1358#define foo() /* @r{optimized inline version} */
1359@dots{}
1360  foo();
1361  funcptr = foo;
1362@end smallexample
1363
1364Here the call to @code{foo()} will use the macro, but the function
1365pointer will get the address of the real function.  If the macro were to
1366be expanded, it would cause a syntax error.
1367
1368If you put spaces between the macro name and the parentheses in the
1369macro definition, that does not define a function-like macro, it defines
1370an object-like macro whose expansion happens to begin with a pair of
1371parentheses.
1372
1373@smallexample
1374#define lang_init ()    c_init()
1375lang_init()
1376     @expansion{} () c_init()()
1377@end smallexample
1378
1379The first two pairs of parentheses in this expansion come from the
1380macro.  The third is the pair that was originally after the macro
1381invocation.  Since @code{lang_init} is an object-like macro, it does not
1382consume those parentheses.
1383
1384@node Macro Arguments
1385@section Macro Arguments
1386@cindex arguments
1387@cindex macros with arguments
1388@cindex arguments in macro definitions
1389
1390Function-like macros can take @dfn{arguments}, just like true functions.
1391To define a macro that uses arguments, you insert @dfn{parameters}
1392between the pair of parentheses in the macro definition that make the
1393macro function-like.  The parameters must be valid C identifiers,
1394separated by commas and optionally whitespace.
1395
1396To invoke a macro that takes arguments, you write the name of the macro
1397followed by a list of @dfn{actual arguments} in parentheses, separated
1398by commas.  The invocation of the macro need not be restricted to a
1399single logical line---it can cross as many lines in the source file as
1400you wish.  The number of arguments you give must match the number of
1401parameters in the macro definition.  When the macro is expanded, each
1402use of a parameter in its body is replaced by the tokens of the
1403corresponding argument.  (You need not use all of the parameters in the
1404macro body.)
1405
1406As an example, here is a macro that computes the minimum of two numeric
1407values, as it is defined in many C programs, and some uses.
1408
1409@smallexample
1410#define min(X, Y)  ((X) < (Y) ? (X) : (Y))
1411  x = min(a, b);          @expansion{}  x = ((a) < (b) ? (a) : (b));
1412  y = min(1, 2);          @expansion{}  y = ((1) < (2) ? (1) : (2));
1413  z = min(a + 28, *p);    @expansion{}  z = ((a + 28) < (*p) ? (a + 28) : (*p));
1414@end smallexample
1415
1416@noindent
1417(In this small example you can already see several of the dangers of
1418macro arguments.  @xref{Macro Pitfalls}, for detailed explanations.)
1419
1420Leading and trailing whitespace in each argument is dropped, and all
1421whitespace between the tokens of an argument is reduced to a single
1422space.  Parentheses within each argument must balance; a comma within
1423such parentheses does not end the argument.  However, there is no
1424requirement for square brackets or braces to balance, and they do not
1425prevent a comma from separating arguments.  Thus,
1426
1427@smallexample
1428macro (array[x = y, x + 1])
1429@end smallexample
1430
1431@noindent
1432passes two arguments to @code{macro}: @code{array[x = y} and @code{x +
14331]}.  If you want to supply @code{array[x = y, x + 1]} as an argument,
1434you can write it as @code{array[(x = y, x + 1)]}, which is equivalent C
1435code.
1436
1437All arguments to a macro are completely macro-expanded before they are
1438substituted into the macro body.  After substitution, the complete text
1439is scanned again for macros to expand, including the arguments.  This rule
1440may seem strange, but it is carefully designed so you need not worry
1441about whether any function call is actually a macro invocation.  You can
1442run into trouble if you try to be too clever, though.  @xref{Argument
1443Prescan}, for detailed discussion.
1444
1445For example, @code{min (min (a, b), c)} is first expanded to
1446
1447@smallexample
1448  min (((a) < (b) ? (a) : (b)), (c))
1449@end smallexample
1450
1451@noindent
1452and then to
1453
1454@smallexample
1455@group
1456((((a) < (b) ? (a) : (b))) < (c)
1457 ? (((a) < (b) ? (a) : (b)))
1458 : (c))
1459@end group
1460@end smallexample
1461
1462@noindent
1463(Line breaks shown here for clarity would not actually be generated.)
1464
1465@cindex empty macro arguments
1466You can leave macro arguments empty; this is not an error to the
1467preprocessor (but many macros will then expand to invalid code).
1468You cannot leave out arguments entirely; if a macro takes two arguments,
1469there must be exactly one comma at the top level of its argument list.
1470Here are some silly examples using @code{min}:
1471
1472@smallexample
1473min(, b)        @expansion{} ((   ) < (b) ? (   ) : (b))
1474min(a, )        @expansion{} ((a  ) < ( ) ? (a  ) : ( ))
1475min(,)          @expansion{} ((   ) < ( ) ? (   ) : ( ))
1476min((,),)       @expansion{} (((,)) < ( ) ? ((,)) : ( ))
1477
1478min()      @error{} macro "min" requires 2 arguments, but only 1 given
1479min(,,)    @error{} macro "min" passed 3 arguments, but takes just 2
1480@end smallexample
1481
1482Whitespace is not a preprocessing token, so if a macro @code{foo} takes
1483one argument, @code{@w{foo ()}} and @code{@w{foo ( )}} both supply it an
1484empty argument.  Previous GNU preprocessor implementations and
1485documentation were incorrect on this point, insisting that a
1486function-like macro that takes a single argument be passed a space if an
1487empty argument was required.
1488
1489Macro parameters appearing inside string literals are not replaced by
1490their corresponding actual arguments.
1491
1492@smallexample
1493#define foo(x) x, "x"
1494foo(bar)        @expansion{} bar, "x"
1495@end smallexample
1496
1497@node Stringification
1498@section Stringification
1499@cindex stringification
1500@cindex @samp{#} operator
1501
1502Sometimes you may want to convert a macro argument into a string
1503constant.  Parameters are not replaced inside string constants, but you
1504can use the @samp{#} preprocessing operator instead.  When a macro
1505parameter is used with a leading @samp{#}, the preprocessor replaces it
1506with the literal text of the actual argument, converted to a string
1507constant.  Unlike normal parameter replacement, the argument is not
1508macro-expanded first.  This is called @dfn{stringification}.
1509
1510There is no way to combine an argument with surrounding text and
1511stringify it all together.  Instead, you can write a series of adjacent
1512string constants and stringified arguments.  The preprocessor will
1513replace the stringified arguments with string constants.  The C
1514compiler will then combine all the adjacent string constants into one
1515long string.
1516
1517Here is an example of a macro definition that uses stringification:
1518
1519@smallexample
1520@group
1521#define WARN_IF(EXP) \
1522do @{ if (EXP) \
1523        fprintf (stderr, "Warning: " #EXP "\n"); @} \
1524while (0)
1525WARN_IF (x == 0);
1526     @expansion{} do @{ if (x == 0)
1527           fprintf (stderr, "Warning: " "x == 0" "\n"); @} while (0);
1528@end group
1529@end smallexample
1530
1531@noindent
1532The argument for @code{EXP} is substituted once, as-is, into the
1533@code{if} statement, and once, stringified, into the argument to
1534@code{fprintf}.  If @code{x} were a macro, it would be expanded in the
1535@code{if} statement, but not in the string.
1536
1537The @code{do} and @code{while (0)} are a kludge to make it possible to
1538write @code{WARN_IF (@var{arg});}, which the resemblance of
1539@code{WARN_IF} to a function would make C programmers want to do; see
1540@ref{Swallowing the Semicolon}.
1541
1542Stringification in C involves more than putting double-quote characters
1543around the fragment.  The preprocessor backslash-escapes the quotes
1544surrounding embedded string constants, and all backslashes within string and
1545character constants, in order to get a valid C string constant with the
1546proper contents.  Thus, stringifying @code{@w{p = "foo\n";}} results in
1547@t{@w{"p = \"foo\\n\";"}}.  However, backslashes that are not inside string
1548or character constants are not duplicated: @samp{\n} by itself
1549stringifies to @t{"\n"}.
1550
1551All leading and trailing whitespace in text being stringified is
1552ignored.  Any sequence of whitespace in the middle of the text is
1553converted to a single space in the stringified result.  Comments are
1554replaced by whitespace long before stringification happens, so they
1555never appear in stringified text.
1556
1557There is no way to convert a macro argument into a character constant.
1558
1559If you want to stringify the result of expansion of a macro argument,
1560you have to use two levels of macros.
1561
1562@smallexample
1563#define xstr(s) str(s)
1564#define str(s) #s
1565#define foo 4
1566str (foo)
1567     @expansion{} "foo"
1568xstr (foo)
1569     @expansion{} xstr (4)
1570     @expansion{} str (4)
1571     @expansion{} "4"
1572@end smallexample
1573
1574@code{s} is stringified when it is used in @code{str}, so it is not
1575macro-expanded first.  But @code{s} is an ordinary argument to
1576@code{xstr}, so it is completely macro-expanded before @code{xstr}
1577itself is expanded (@pxref{Argument Prescan}).  Therefore, by the time
1578@code{str} gets to its argument, it has already been macro-expanded.
1579
1580@node Concatenation
1581@section Concatenation
1582@cindex concatenation
1583@cindex token pasting
1584@cindex token concatenation
1585@cindex @samp{##} operator
1586
1587It is often useful to merge two tokens into one while expanding macros.
1588This is called @dfn{token pasting} or @dfn{token concatenation}.  The
1589@samp{##} preprocessing operator performs token pasting.  When a macro
1590is expanded, the two tokens on either side of each @samp{##} operator
1591are combined into a single token, which then replaces the @samp{##} and
1592the two original tokens in the macro expansion.  Usually both will be
1593identifiers, or one will be an identifier and the other a preprocessing
1594number.  When pasted, they make a longer identifier.  This isn't the
1595only valid case.  It is also possible to concatenate two numbers (or a
1596number and a name, such as @code{1.5} and @code{e3}) into a number.
1597Also, multi-character operators such as @code{+=} can be formed by
1598token pasting.
1599
1600However, two tokens that don't together form a valid token cannot be
1601pasted together.  For example, you cannot concatenate @code{x} with
1602@code{+} in either order.  If you try, the preprocessor issues a warning
1603and emits the two tokens.  Whether it puts white space between the
1604tokens is undefined.  It is common to find unnecessary uses of @samp{##}
1605in complex macros.  If you get this warning, it is likely that you can
1606simply remove the @samp{##}.
1607
1608Both the tokens combined by @samp{##} could come from the macro body,
1609but you could just as well write them as one token in the first place.
1610Token pasting is most useful when one or both of the tokens comes from a
1611macro argument.  If either of the tokens next to an @samp{##} is a
1612parameter name, it is replaced by its actual argument before @samp{##}
1613executes.  As with stringification, the actual argument is not
1614macro-expanded first.  If the argument is empty, that @samp{##} has no
1615effect.
1616
1617Keep in mind that the C preprocessor converts comments to whitespace
1618before macros are even considered.  Therefore, you cannot create a
1619comment by concatenating @samp{/} and @samp{*}.  You can put as much
1620whitespace between @samp{##} and its operands as you like, including
1621comments, and you can put comments in arguments that will be
1622concatenated.  However, it is an error if @samp{##} appears at either
1623end of a macro body.
1624
1625Consider a C program that interprets named commands.  There probably
1626needs to be a table of commands, perhaps an array of structures declared
1627as follows:
1628
1629@smallexample
1630@group
1631struct command
1632@{
1633  char *name;
1634  void (*function) (void);
1635@};
1636@end group
1637
1638@group
1639struct command commands[] =
1640@{
1641  @{ "quit", quit_command @},
1642  @{ "help", help_command @},
1643  @dots{}
1644@};
1645@end group
1646@end smallexample
1647
1648It would be cleaner not to have to give each command name twice, once in
1649the string constant and once in the function name.  A macro which takes the
1650name of a command as an argument can make this unnecessary.  The string
1651constant can be created with stringification, and the function name by
1652concatenating the argument with @samp{_command}.  Here is how it is done:
1653
1654@smallexample
1655#define COMMAND(NAME)  @{ #NAME, NAME ## _command @}
1656
1657struct command commands[] =
1658@{
1659  COMMAND (quit),
1660  COMMAND (help),
1661  @dots{}
1662@};
1663@end smallexample
1664
1665@node Variadic Macros
1666@section Variadic Macros
1667@cindex variable number of arguments
1668@cindex macros with variable arguments
1669@cindex variadic macros
1670
1671A macro can be declared to accept a variable number of arguments much as
1672a function can.  The syntax for defining the macro is similar to that of
1673a function.  Here is an example:
1674
1675@smallexample
1676#define eprintf(@dots{}) fprintf (stderr, __VA_ARGS__)
1677@end smallexample
1678
1679This kind of macro is called @dfn{variadic}.  When the macro is invoked,
1680all the tokens in its argument list after the last named argument (this
1681macro has none), including any commas, become the @dfn{variable
1682argument}.  This sequence of tokens replaces the identifier
1683@code{@w{__VA_ARGS__}} in the macro body wherever it appears.  Thus, we
1684have this expansion:
1685
1686@smallexample
1687eprintf ("%s:%d: ", input_file, lineno)
1688     @expansion{}  fprintf (stderr, "%s:%d: ", input_file, lineno)
1689@end smallexample
1690
1691The variable argument is completely macro-expanded before it is inserted
1692into the macro expansion, just like an ordinary argument.  You may use
1693the @samp{#} and @samp{##} operators to stringify the variable argument
1694or to paste its leading or trailing token with another token.  (But see
1695below for an important special case for @samp{##}.)
1696
1697If your macro is complicated, you may want a more descriptive name for
1698the variable argument than @code{@w{__VA_ARGS__}}.  CPP permits
1699this, as an extension.  You may write an argument name immediately
1700before the @samp{@dots{}}; that name is used for the variable argument.
1701The @code{eprintf} macro above could be written
1702
1703@smallexample
1704#define eprintf(args@dots{}) fprintf (stderr, args)
1705@end smallexample
1706
1707@noindent
1708using this extension.  You cannot use @code{@w{__VA_ARGS__}} and this
1709extension in the same macro.
1710
1711You can have named arguments as well as variable arguments in a variadic
1712macro.  We could define @code{eprintf} like this, instead:
1713
1714@smallexample
1715#define eprintf(format, @dots{}) fprintf (stderr, format, __VA_ARGS__)
1716@end smallexample
1717
1718@noindent
1719This formulation looks more descriptive, but unfortunately it is less
1720flexible: you must now supply at least one argument after the format
1721string.  In standard C, you cannot omit the comma separating the named
1722argument from the variable arguments.  Furthermore, if you leave the
1723variable argument empty, you will get a syntax error, because
1724there will be an extra comma after the format string.
1725
1726@smallexample
1727eprintf("success!\n", );
1728     @expansion{} fprintf(stderr, "success!\n", );
1729@end smallexample
1730
1731GNU CPP has a pair of extensions which deal with this problem.  First,
1732you are allowed to leave the variable argument out entirely:
1733
1734@smallexample
1735eprintf ("success!\n")
1736     @expansion{} fprintf(stderr, "success!\n", );
1737@end smallexample
1738
1739@noindent
1740Second, the @samp{##} token paste operator has a special meaning when
1741placed between a comma and a variable argument.  If you write
1742
1743@smallexample
1744#define eprintf(format, @dots{}) fprintf (stderr, format, ##__VA_ARGS__)
1745@end smallexample
1746
1747@noindent
1748and the variable argument is left out when the @code{eprintf} macro is
1749used, then the comma before the @samp{##} will be deleted.  This does
1750@emph{not} happen if you pass an empty argument, nor does it happen if
1751the token preceding @samp{##} is anything other than a comma.
1752
1753@smallexample
1754eprintf ("success!\n")
1755     @expansion{} fprintf(stderr, "success!\n");
1756@end smallexample
1757
1758@noindent
1759The above explanation is ambiguous about the case where the only macro
1760parameter is a variable arguments parameter, as it is meaningless to
1761try to distinguish whether no argument at all is an empty argument or
1762a missing argument.  In this case the C99 standard is clear that the
1763comma must remain, however the existing GCC extension used to swallow
1764the comma.  So CPP retains the comma when conforming to a specific C
1765standard, and drops it otherwise.
1766
1767C99 mandates that the only place the identifier @code{@w{__VA_ARGS__}}
1768can appear is in the replacement list of a variadic macro.  It may not
1769be used as a macro name, macro argument name, or within a different type
1770of macro.  It may also be forbidden in open text; the standard is
1771ambiguous.  We recommend you avoid using it except for its defined
1772purpose.
1773
1774Variadic macros are a new feature in C99.  GNU CPP has supported them
1775for a long time, but only with a named variable argument
1776(@samp{args@dots{}}, not @samp{@dots{}} and @code{@w{__VA_ARGS__}}).  If you are
1777concerned with portability to previous versions of GCC, you should use
1778only named variable arguments.  On the other hand, if you are concerned
1779with portability to other conforming implementations of C99, you should
1780use only @code{@w{__VA_ARGS__}}.
1781
1782Previous versions of CPP implemented the comma-deletion extension
1783much more generally.  We have restricted it in this release to minimize
1784the differences from C99.  To get the same effect with both this and
1785previous versions of GCC, the token preceding the special @samp{##} must
1786be a comma, and there must be white space between that comma and
1787whatever comes immediately before it:
1788
1789@smallexample
1790#define eprintf(format, args@dots{}) fprintf (stderr, format , ##args)
1791@end smallexample
1792
1793@noindent
1794@xref{Differences from previous versions}, for the gory details.
1795
1796@node Predefined Macros
1797@section Predefined Macros
1798
1799@cindex predefined macros
1800Several object-like macros are predefined; you use them without
1801supplying their definitions.  They fall into three classes: standard,
1802common, and system-specific.
1803
1804In C++, there is a fourth category, the named operators.  They act like
1805predefined macros, but you cannot undefine them.
1806
1807@menu
1808* Standard Predefined Macros::
1809* Common Predefined Macros::
1810* System-specific Predefined Macros::
1811* C++ Named Operators::
1812@end menu
1813
1814@node Standard Predefined Macros
1815@subsection Standard Predefined Macros
1816@cindex standard predefined macros.
1817
1818The standard predefined macros are specified by the relevant
1819language standards, so they are available with all compilers that
1820implement those standards.  Older compilers may not provide all of
1821them.  Their names all start with double underscores.
1822
1823@table @code
1824@item __FILE__
1825This macro expands to the name of the current input file, in the form of
1826a C string constant.  This is the path by which the preprocessor opened
1827the file, not the short name specified in @samp{#include} or as the
1828input file name argument.  For example,
1829@code{"/usr/local/include/myheader.h"} is a possible expansion of this
1830macro.
1831
1832@item __LINE__
1833This macro expands to the current input line number, in the form of a
1834decimal integer constant.  While we call it a predefined macro, it's
1835a pretty strange macro, since its ``definition'' changes with each
1836new line of source code.
1837@end table
1838
1839@code{__FILE__} and @code{__LINE__} are useful in generating an error
1840message to report an inconsistency detected by the program; the message
1841can state the source line at which the inconsistency was detected.  For
1842example,
1843
1844@smallexample
1845fprintf (stderr, "Internal error: "
1846                 "negative string length "
1847                 "%d at %s, line %d.",
1848         length, __FILE__, __LINE__);
1849@end smallexample
1850
1851An @samp{#include} directive changes the expansions of @code{__FILE__}
1852and @code{__LINE__} to correspond to the included file.  At the end of
1853that file, when processing resumes on the input file that contained
1854the @samp{#include} directive, the expansions of @code{__FILE__} and
1855@code{__LINE__} revert to the values they had before the
1856@samp{#include} (but @code{__LINE__} is then incremented by one as
1857processing moves to the line after the @samp{#include}).
1858
1859A @samp{#line} directive changes @code{__LINE__}, and may change
1860@code{__FILE__} as well.  @xref{Line Control}.
1861
1862C99 introduces @code{__func__}, and GCC has provided @code{__FUNCTION__}
1863for a long time.  Both of these are strings containing the name of the
1864current function (there are slight semantic differences; see the GCC
1865manual).  Neither of them is a macro; the preprocessor does not know the
1866name of the current function.  They tend to be useful in conjunction
1867with @code{__FILE__} and @code{__LINE__}, though.
1868
1869@table @code
1870
1871@item __DATE__
1872This macro expands to a string constant that describes the date on which
1873the preprocessor is being run.  The string constant contains eleven
1874characters and looks like @code{@w{"Feb 12 1996"}}.  If the day of the
1875month is less than 10, it is padded with a space on the left.
1876
1877If GCC cannot determine the current date, it will emit a warning message
1878(once per compilation) and @code{__DATE__} will expand to
1879@code{@w{"??? ?? ????"}}.
1880
1881@item __TIME__
1882This macro expands to a string constant that describes the time at
1883which the preprocessor is being run.  The string constant contains
1884eight characters and looks like @code{"23:59:01"}.
1885
1886If GCC cannot determine the current time, it will emit a warning message
1887(once per compilation) and @code{__TIME__} will expand to
1888@code{"??:??:??"}.
1889
1890@item __STDC__
1891In normal operation, this macro expands to the constant 1, to signify
1892that this compiler conforms to ISO Standard C@.  If GNU CPP is used with
1893a compiler other than GCC, this is not necessarily true; however, the
1894preprocessor always conforms to the standard unless the
1895@option{-traditional-cpp} option is used.
1896
1897This macro is not defined if the @option{-traditional-cpp} option is used.
1898
1899On some hosts, the system compiler uses a different convention, where
1900@code{__STDC__} is normally 0, but is 1 if the user specifies strict
1901conformance to the C Standard.  CPP follows the host convention when
1902processing system header files, but when processing user files
1903@code{__STDC__} is always 1.  This has been reported to cause problems;
1904for instance, some versions of Solaris provide X Windows headers that
1905expect @code{__STDC__} to be either undefined or 1.  @xref{Invocation}.
1906
1907@item __STDC_VERSION__
1908This macro expands to the C Standard's version number, a long integer
1909constant of the form @code{@var{yyyy}@var{mm}L} where @var{yyyy} and
1910@var{mm} are the year and month of the Standard version.  This signifies
1911which version of the C Standard the compiler conforms to.  Like
1912@code{__STDC__}, this is not necessarily accurate for the entire
1913implementation, unless GNU CPP is being used with GCC@.
1914
1915The value @code{199409L} signifies the 1989 C standard as amended in
19161994, which is the current default; the value @code{199901L} signifies
1917the 1999 revision of the C standard.  Support for the 1999 revision is
1918not yet complete.
1919
1920This macro is not defined if the @option{-traditional-cpp} option is
1921used, nor when compiling C++ or Objective-C@.
1922
1923@item __STDC_HOSTED__
1924This macro is defined, with value 1, if the compiler's target is a
1925@dfn{hosted environment}.  A hosted environment has the complete
1926facilities of the standard C library available.
1927
1928@item __cplusplus
1929This macro is defined when the C++ compiler is in use.  You can use
1930@code{__cplusplus} to test whether a header is compiled by a C compiler
1931or a C++ compiler.  This macro is similar to @code{__STDC_VERSION__}, in
1932that it expands to a version number.  Depending on the language standard
1933selected, the value of the macro is @code{199711L}, as mandated by the
19341998 C++ standard, or @code{201103L}, per the 2011 C++ standard.
1935
1936@item __OBJC__
1937This macro is defined, with value 1, when the Objective-C compiler is in
1938use.  You can use @code{__OBJC__} to test whether a header is compiled
1939by a C compiler or an Objective-C compiler.
1940
1941@item __ASSEMBLER__
1942This macro is defined with value 1 when preprocessing assembly
1943language.
1944
1945@end table
1946
1947@node Common Predefined Macros
1948@subsection Common Predefined Macros
1949@cindex common predefined macros
1950
1951The common predefined macros are GNU C extensions.  They are available
1952with the same meanings regardless of the machine or operating system on
1953which you are using GNU C or GNU Fortran.  Their names all start with
1954double underscores.
1955
1956@table @code
1957
1958@item __COUNTER__
1959This macro expands to sequential integral values starting from 0.  In
1960conjunction with the @code{##} operator, this provides a convenient means to
1961generate unique identifiers.  Care must be taken to ensure that
1962@code{__COUNTER__} is not expanded prior to inclusion of precompiled headers
1963which use it.  Otherwise, the precompiled headers will not be used.
1964
1965@item __GFORTRAN__
1966The GNU Fortran compiler defines this.
1967
1968@item __GNUC__
1969@itemx __GNUC_MINOR__
1970@itemx __GNUC_PATCHLEVEL__
1971These macros are defined by all GNU compilers that use the C
1972preprocessor: C, C++, Objective-C and Fortran.  Their values are the major
1973version, minor version, and patch level of the compiler, as integer
1974constants.  For example, GCC 3.2.1 will define @code{__GNUC__} to 3,
1975@code{__GNUC_MINOR__} to 2, and @code{__GNUC_PATCHLEVEL__} to 1.  These
1976macros are also defined if you invoke the preprocessor directly.
1977
1978@code{__GNUC_PATCHLEVEL__} is new to GCC 3.0; it is also present in the
1979widely-used development snapshots leading up to 3.0 (which identify
1980themselves as GCC 2.96 or 2.97, depending on which snapshot you have).
1981
1982If all you need to know is whether or not your program is being compiled
1983by GCC, or a non-GCC compiler that claims to accept the GNU C dialects,
1984you can simply test @code{__GNUC__}.  If you need to write code
1985which depends on a specific version, you must be more careful.  Each
1986time the minor version is increased, the patch level is reset to zero;
1987each time the major version is increased (which happens rarely), the
1988minor version and patch level are reset.  If you wish to use the
1989predefined macros directly in the conditional, you will need to write it
1990like this:
1991
1992@smallexample
1993/* @r{Test for GCC > 3.2.0} */
1994#if __GNUC__ > 3 || \
1995    (__GNUC__ == 3 && (__GNUC_MINOR__ > 2 || \
1996                       (__GNUC_MINOR__ == 2 && \
1997                        __GNUC_PATCHLEVEL__ > 0))
1998@end smallexample
1999
2000@noindent
2001Another approach is to use the predefined macros to
2002calculate a single number, then compare that against a threshold:
2003
2004@smallexample
2005#define GCC_VERSION (__GNUC__ * 10000 \
2006                     + __GNUC_MINOR__ * 100 \
2007                     + __GNUC_PATCHLEVEL__)
2008@dots{}
2009/* @r{Test for GCC > 3.2.0} */
2010#if GCC_VERSION > 30200
2011@end smallexample
2012
2013@noindent
2014Many people find this form easier to understand.
2015
2016@item __GNUG__
2017The GNU C++ compiler defines this.  Testing it is equivalent to
2018testing @code{@w{(__GNUC__ && __cplusplus)}}.
2019
2020@item __STRICT_ANSI__
2021GCC defines this macro if and only if the @option{-ansi} switch, or a
2022@option{-std} switch specifying strict conformance to some version of ISO C
2023or ISO C++, was specified when GCC was invoked.  It is defined to @samp{1}.
2024This macro exists primarily to direct GNU libc's header files to
2025restrict their definitions to the minimal set found in the 1989 C
2026standard.
2027
2028@item __BASE_FILE__
2029This macro expands to the name of the main input file, in the form
2030of a C string constant.  This is the source file that was specified
2031on the command line of the preprocessor or C compiler.
2032
2033@item __INCLUDE_LEVEL__
2034This macro expands to a decimal integer constant that represents the
2035depth of nesting in include files.  The value of this macro is
2036incremented on every @samp{#include} directive and decremented at the
2037end of every included file.  It starts out at 0, its value within the
2038base file specified on the command line.
2039
2040@item __ELF__
2041This macro is defined if the target uses the ELF object format.
2042
2043@item __VERSION__
2044This macro expands to a string constant which describes the version of
2045the compiler in use.  You should not rely on its contents having any
2046particular form, but it can be counted on to contain at least the
2047release number.
2048
2049@item __OPTIMIZE__
2050@itemx __OPTIMIZE_SIZE__
2051@itemx __NO_INLINE__
2052These macros describe the compilation mode.  @code{__OPTIMIZE__} is
2053defined in all optimizing compilations.  @code{__OPTIMIZE_SIZE__} is
2054defined if the compiler is optimizing for size, not speed.
2055@code{__NO_INLINE__} is defined if no functions will be inlined into
2056their callers (when not optimizing, or when inlining has been
2057specifically disabled by @option{-fno-inline}).
2058
2059These macros cause certain GNU header files to provide optimized
2060definitions, using macros or inline functions, of system library
2061functions.  You should not use these macros in any way unless you make
2062sure that programs will execute with the same effect whether or not they
2063are defined.  If they are defined, their value is 1.
2064
2065@item __GNUC_GNU_INLINE__
2066GCC defines this macro if functions declared @code{inline} will be
2067handled in GCC's traditional gnu90 mode.  Object files will contain
2068externally visible definitions of all functions declared @code{inline}
2069without @code{extern} or @code{static}.  They will not contain any
2070definitions of any functions declared @code{extern inline}.
2071
2072@item __GNUC_STDC_INLINE__
2073GCC defines this macro if functions declared @code{inline} will be
2074handled according to the ISO C99 standard.  Object files will contain
2075externally visible definitions of all functions declared @code{extern
2076inline}.  They will not contain definitions of any functions declared
2077@code{inline} without @code{extern}.
2078
2079If this macro is defined, GCC supports the @code{gnu_inline} function
2080attribute as a way to always get the gnu90 behavior.  Support for
2081this and @code{__GNUC_GNU_INLINE__} was added in GCC 4.1.3.  If
2082neither macro is defined, an older version of GCC is being used:
2083@code{inline} functions will be compiled in gnu90 mode, and the
2084@code{gnu_inline} function attribute will not be recognized.
2085
2086@item __CHAR_UNSIGNED__
2087GCC defines this macro if and only if the data type @code{char} is
2088unsigned on the target machine.  It exists to cause the standard header
2089file @file{limits.h} to work correctly.  You should not use this macro
2090yourself; instead, refer to the standard macros defined in @file{limits.h}.
2091
2092@item __WCHAR_UNSIGNED__
2093Like @code{__CHAR_UNSIGNED__}, this macro is defined if and only if the
2094data type @code{wchar_t} is unsigned and the front-end is in C++ mode.
2095
2096@item __REGISTER_PREFIX__
2097This macro expands to a single token (not a string constant) which is
2098the prefix applied to CPU register names in assembly language for this
2099target.  You can use it to write assembly that is usable in multiple
2100environments.  For example, in the @code{m68k-aout} environment it
2101expands to nothing, but in the @code{m68k-coff} environment it expands
2102to a single @samp{%}.
2103
2104@item __USER_LABEL_PREFIX__
2105This macro expands to a single token which is the prefix applied to
2106user labels (symbols visible to C code) in assembly.  For example, in
2107the @code{m68k-aout} environment it expands to an @samp{_}, but in the
2108@code{m68k-coff} environment it expands to nothing.
2109
2110This macro will have the correct definition even if
2111@option{-f(no-)underscores} is in use, but it will not be correct if
2112target-specific options that adjust this prefix are used (e.g.@: the
2113OSF/rose @option{-mno-underscores} option).
2114
2115@item __SIZE_TYPE__
2116@itemx __PTRDIFF_TYPE__
2117@itemx __WCHAR_TYPE__
2118@itemx __WINT_TYPE__
2119@itemx __INTMAX_TYPE__
2120@itemx __UINTMAX_TYPE__
2121@itemx __SIG_ATOMIC_TYPE__
2122@itemx __INT8_TYPE__
2123@itemx __INT16_TYPE__
2124@itemx __INT32_TYPE__
2125@itemx __INT64_TYPE__
2126@itemx __UINT8_TYPE__
2127@itemx __UINT16_TYPE__
2128@itemx __UINT32_TYPE__
2129@itemx __UINT64_TYPE__
2130@itemx __INT_LEAST8_TYPE__
2131@itemx __INT_LEAST16_TYPE__
2132@itemx __INT_LEAST32_TYPE__
2133@itemx __INT_LEAST64_TYPE__
2134@itemx __UINT_LEAST8_TYPE__
2135@itemx __UINT_LEAST16_TYPE__
2136@itemx __UINT_LEAST32_TYPE__
2137@itemx __UINT_LEAST64_TYPE__
2138@itemx __INT_FAST8_TYPE__
2139@itemx __INT_FAST16_TYPE__
2140@itemx __INT_FAST32_TYPE__
2141@itemx __INT_FAST64_TYPE__
2142@itemx __UINT_FAST8_TYPE__
2143@itemx __UINT_FAST16_TYPE__
2144@itemx __UINT_FAST32_TYPE__
2145@itemx __UINT_FAST64_TYPE__
2146@itemx __INTPTR_TYPE__
2147@itemx __UINTPTR_TYPE__
2148These macros are defined to the correct underlying types for the
2149@code{size_t}, @code{ptrdiff_t}, @code{wchar_t}, @code{wint_t},
2150@code{intmax_t}, @code{uintmax_t}, @code{sig_atomic_t}, @code{int8_t},
2151@code{int16_t}, @code{int32_t}, @code{int64_t}, @code{uint8_t},
2152@code{uint16_t}, @code{uint32_t}, @code{uint64_t},
2153@code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t},
2154@code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t},
2155@code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t},
2156@code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t},
2157@code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t},
2158@code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t} typedefs,
2159respectively.  They exist to make the standard header files
2160@file{stddef.h}, @file{stdint.h}, and @file{wchar.h} work correctly.
2161You should not use these macros directly; instead, include the
2162appropriate headers and use the typedefs.  Some of these macros may
2163not be defined on particular systems if GCC does not provide a
2164@file{stdint.h} header on those systems.
2165
2166@item __CHAR_BIT__
2167Defined to the number of bits used in the representation of the
2168@code{char} data type.  It exists to make the standard header given
2169numerical limits work correctly.  You should not use
2170this macro directly; instead, include the appropriate headers.
2171
2172@item __SCHAR_MAX__
2173@itemx __WCHAR_MAX__
2174@itemx __SHRT_MAX__
2175@itemx __INT_MAX__
2176@itemx __LONG_MAX__
2177@itemx __LONG_LONG_MAX__
2178@itemx __WINT_MAX__
2179@itemx __SIZE_MAX__
2180@itemx __PTRDIFF_MAX__
2181@itemx __INTMAX_MAX__
2182@itemx __UINTMAX_MAX__
2183@itemx __SIG_ATOMIC_MAX__
2184@itemx __INT8_MAX__
2185@itemx __INT16_MAX__
2186@itemx __INT32_MAX__
2187@itemx __INT64_MAX__
2188@itemx __UINT8_MAX__
2189@itemx __UINT16_MAX__
2190@itemx __UINT32_MAX__
2191@itemx __UINT64_MAX__
2192@itemx __INT_LEAST8_MAX__
2193@itemx __INT_LEAST16_MAX__
2194@itemx __INT_LEAST32_MAX__
2195@itemx __INT_LEAST64_MAX__
2196@itemx __UINT_LEAST8_MAX__
2197@itemx __UINT_LEAST16_MAX__
2198@itemx __UINT_LEAST32_MAX__
2199@itemx __UINT_LEAST64_MAX__
2200@itemx __INT_FAST8_MAX__
2201@itemx __INT_FAST16_MAX__
2202@itemx __INT_FAST32_MAX__
2203@itemx __INT_FAST64_MAX__
2204@itemx __UINT_FAST8_MAX__
2205@itemx __UINT_FAST16_MAX__
2206@itemx __UINT_FAST32_MAX__
2207@itemx __UINT_FAST64_MAX__
2208@itemx __INTPTR_MAX__
2209@itemx __UINTPTR_MAX__
2210@itemx __WCHAR_MIN__
2211@itemx __WINT_MIN__
2212@itemx __SIG_ATOMIC_MIN__
2213Defined to the maximum value of the @code{signed char}, @code{wchar_t},
2214@code{signed short},
2215@code{signed int}, @code{signed long}, @code{signed long long},
2216@code{wint_t}, @code{size_t}, @code{ptrdiff_t},
2217@code{intmax_t}, @code{uintmax_t}, @code{sig_atomic_t}, @code{int8_t},
2218@code{int16_t}, @code{int32_t}, @code{int64_t}, @code{uint8_t},
2219@code{uint16_t}, @code{uint32_t}, @code{uint64_t},
2220@code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t},
2221@code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t},
2222@code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t},
2223@code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t},
2224@code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t},
2225@code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t} types and
2226to the minimum value of the @code{wchar_t}, @code{wint_t}, and
2227@code{sig_atomic_t} types respectively.  They exist to make the
2228standard header given numerical limits work correctly.  You should not
2229use these macros directly; instead, include the appropriate headers.
2230Some of these macros may not be defined on particular systems if GCC
2231does not provide a @file{stdint.h} header on those systems.
2232
2233@item __INT8_C
2234@itemx __INT16_C
2235@itemx __INT32_C
2236@itemx __INT64_C
2237@itemx __UINT8_C
2238@itemx __UINT16_C
2239@itemx __UINT32_C
2240@itemx __UINT64_C
2241@itemx __INTMAX_C
2242@itemx __UINTMAX_C
2243Defined to implementations of the standard @file{stdint.h} macros with
2244the same names without the leading @code{__}.  They exist the make the
2245implementation of that header work correctly.  You should not use
2246these macros directly; instead, include the appropriate headers.  Some
2247of these macros may not be defined on particular systems if GCC does
2248not provide a @file{stdint.h} header on those systems.
2249
2250@item __SIZEOF_INT__
2251@itemx __SIZEOF_LONG__
2252@itemx __SIZEOF_LONG_LONG__
2253@itemx __SIZEOF_SHORT__
2254@itemx __SIZEOF_POINTER__
2255@itemx __SIZEOF_FLOAT__
2256@itemx __SIZEOF_DOUBLE__
2257@itemx __SIZEOF_LONG_DOUBLE__
2258@itemx __SIZEOF_SIZE_T__
2259@itemx __SIZEOF_WCHAR_T__
2260@itemx __SIZEOF_WINT_T__
2261@itemx __SIZEOF_PTRDIFF_T__
2262Defined to the number of bytes of the C standard data types: @code{int},
2263@code{long}, @code{long long}, @code{short}, @code{void *}, @code{float},
2264@code{double}, @code{long double}, @code{size_t}, @code{wchar_t}, @code{wint_t}
2265and @code{ptrdiff_t}.
2266
2267@item __BYTE_ORDER__
2268@itemx __ORDER_LITTLE_ENDIAN__
2269@itemx __ORDER_BIG_ENDIAN__
2270@itemx __ORDER_PDP_ENDIAN__
2271@code{__BYTE_ORDER__} is defined to one of the values
2272@code{__ORDER_LITTLE_ENDIAN__}, @code{__ORDER_BIG_ENDIAN__}, or
2273@code{__ORDER_PDP_ENDIAN__} to reflect the layout of multi-byte and
2274multi-word quantities in memory.  If @code{__BYTE_ORDER__} is equal to
2275@code{__ORDER_LITTLE_ENDIAN__} or @code{__ORDER_BIG_ENDIAN__}, then
2276multi-byte and multi-word quantities are laid out identically: the
2277byte (word) at the lowest address is the least significant or most
2278significant byte (word) of the quantity, respectively.  If
2279@code{__BYTE_ORDER__} is equal to @code{__ORDER_PDP_ENDIAN__}, then
2280bytes in 16-bit words are laid out in a little-endian fashion, whereas
2281the 16-bit subwords of a 32-bit quantity are laid out in big-endian
2282fashion.
2283
2284You should use these macros for testing like this:
2285
2286@smallexample
2287/* @r{Test for a little-endian machine} */
2288#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
2289@end smallexample
2290
2291@item __FLOAT_WORD_ORDER__
2292@code{__FLOAT_WORD_ORDER__} is defined to one of the values
2293@code{__ORDER_LITTLE_ENDIAN__} or @code{__ORDER_BIG_ENDIAN__} to reflect
2294the layout of the words of multi-word floating-point quantities.
2295
2296@item __DEPRECATED
2297This macro is defined, with value 1, when compiling a C++ source file
2298with warnings about deprecated constructs enabled.  These warnings are
2299enabled by default, but can be disabled with @option{-Wno-deprecated}.
2300
2301@item __EXCEPTIONS
2302This macro is defined, with value 1, when compiling a C++ source file
2303with exceptions enabled.  If @option{-fno-exceptions} is used when
2304compiling the file, then this macro is not defined.
2305
2306@item __GXX_RTTI
2307This macro is defined, with value 1, when compiling a C++ source file
2308with runtime type identification enabled.  If @option{-fno-rtti} is
2309used when compiling the file, then this macro is not defined.
2310
2311@item __USING_SJLJ_EXCEPTIONS__
2312This macro is defined, with value 1, if the compiler uses the old
2313mechanism based on @code{setjmp} and @code{longjmp} for exception
2314handling.
2315
2316@item __GXX_EXPERIMENTAL_CXX0X__
2317This macro is defined when compiling a C++ source file with the option
2318@option{-std=c++0x} or @option{-std=gnu++0x}. It indicates that some
2319features likely to be included in C++0x are available. Note that these
2320features are experimental, and may change or be removed in future
2321versions of GCC.
2322
2323@item __GXX_WEAK__
2324This macro is defined when compiling a C++ source file.  It has the
2325value 1 if the compiler will use weak symbols, COMDAT sections, or
2326other similar techniques to collapse symbols with ``vague linkage''
2327that are defined in multiple translation units.  If the compiler will
2328not collapse such symbols, this macro is defined with value 0.  In
2329general, user code should not need to make use of this macro; the
2330purpose of this macro is to ease implementation of the C++ runtime
2331library provided with G++.
2332
2333@item __NEXT_RUNTIME__
2334This macro is defined, with value 1, if (and only if) the NeXT runtime
2335(as in @option{-fnext-runtime}) is in use for Objective-C@.  If the GNU
2336runtime is used, this macro is not defined, so that you can use this
2337macro to determine which runtime (NeXT or GNU) is being used.
2338
2339@item __LP64__
2340@itemx _LP64
2341These macros are defined, with value 1, if (and only if) the compilation
2342is for a target where @code{long int} and pointer both use 64-bits and
2343@code{int} uses 32-bit.
2344
2345@item __SSP__
2346This macro is defined, with value 1, when @option{-fstack-protector} is in
2347use.
2348
2349@item __SSP_ALL__
2350This macro is defined, with value 2, when @option{-fstack-protector-all} is
2351in use.
2352
2353@item __SANITIZE_ADDRESS__
2354This macro is defined, with value 1, when @option{-fsanitize=address} is
2355in use.
2356
2357@item __TIMESTAMP__
2358This macro expands to a string constant that describes the date and time
2359of the last modification of the current source file. The string constant
2360contains abbreviated day of the week, month, day of the month, time in
2361hh:mm:ss form, year and looks like @code{@w{"Sun Sep 16 01:03:52 1973"}}.
2362If the day of the month is less than 10, it is padded with a space on the left.
2363
2364If GCC cannot determine the current date, it will emit a warning message
2365(once per compilation) and @code{__TIMESTAMP__} will expand to
2366@code{@w{"??? ??? ?? ??:??:?? ????"}}.
2367
2368@item __GCC_HAVE_SYNC_COMPARE_AND_SWAP_1
2369@itemx __GCC_HAVE_SYNC_COMPARE_AND_SWAP_2
2370@itemx __GCC_HAVE_SYNC_COMPARE_AND_SWAP_4
2371@itemx __GCC_HAVE_SYNC_COMPARE_AND_SWAP_8
2372@itemx __GCC_HAVE_SYNC_COMPARE_AND_SWAP_16
2373These macros are defined when the target processor supports atomic compare
2374and swap operations on operands 1, 2, 4, 8 or 16 bytes in length, respectively.
2375
2376@item __GCC_HAVE_DWARF2_CFI_ASM
2377This macro is defined when the compiler is emitting Dwarf2 CFI directives
2378to the assembler.  When this is defined, it is possible to emit those same
2379directives in inline assembly.
2380
2381@item __FP_FAST_FMA
2382@itemx __FP_FAST_FMAF
2383@itemx __FP_FAST_FMAL
2384These macros are defined with value 1 if the backend supports the
2385@code{fma}, @code{fmaf}, and @code{fmal} builtin functions, so that
2386the include file @file{math.h} can define the macros
2387@code{FP_FAST_FMA}, @code{FP_FAST_FMAF}, and @code{FP_FAST_FMAL}
2388for compatibility with the 1999 C standard.
2389@end table
2390
2391@node System-specific Predefined Macros
2392@subsection System-specific Predefined Macros
2393
2394@cindex system-specific predefined macros
2395@cindex predefined macros, system-specific
2396@cindex reserved namespace
2397
2398The C preprocessor normally predefines several macros that indicate what
2399type of system and machine is in use.  They are obviously different on
2400each target supported by GCC@.  This manual, being for all systems and
2401machines, cannot tell you what their names are, but you can use
2402@command{cpp -dM} to see them all.  @xref{Invocation}.  All system-specific
2403predefined macros expand to a constant value, so you can test them with
2404either @samp{#ifdef} or @samp{#if}.
2405
2406The C standard requires that all system-specific macros be part of the
2407@dfn{reserved namespace}.  All names which begin with two underscores,
2408or an underscore and a capital letter, are reserved for the compiler and
2409library to use as they wish.  However, historically system-specific
2410macros have had names with no special prefix; for instance, it is common
2411to find @code{unix} defined on Unix systems.  For all such macros, GCC
2412provides a parallel macro with two underscores added at the beginning
2413and the end.  If @code{unix} is defined, @code{__unix__} will be defined
2414too.  There will never be more than two underscores; the parallel of
2415@code{_mips} is @code{__mips__}.
2416
2417When the @option{-ansi} option, or any @option{-std} option that
2418requests strict conformance, is given to the compiler, all the
2419system-specific predefined macros outside the reserved namespace are
2420suppressed.  The parallel macros, inside the reserved namespace, remain
2421defined.
2422
2423We are slowly phasing out all predefined macros which are outside the
2424reserved namespace.  You should never use them in new programs, and we
2425encourage you to correct older code to use the parallel macros whenever
2426you find it.  We don't recommend you use the system-specific macros that
2427are in the reserved namespace, either.  It is better in the long run to
2428check specifically for features you need, using a tool such as
2429@command{autoconf}.
2430
2431@node C++ Named Operators
2432@subsection C++ Named Operators
2433@cindex named operators
2434@cindex C++ named operators
2435@cindex @file{iso646.h}
2436
2437In C++, there are eleven keywords which are simply alternate spellings
2438of operators normally written with punctuation.  These keywords are
2439treated as such even in the preprocessor.  They function as operators in
2440@samp{#if}, and they cannot be defined as macros or poisoned.  In C, you
2441can request that those keywords take their C++ meaning by including
2442@file{iso646.h}.  That header defines each one as a normal object-like
2443macro expanding to the appropriate punctuator.
2444
2445These are the named operators and their corresponding punctuators:
2446
2447@multitable {Named Operator} {Punctuator}
2448@item Named Operator @tab Punctuator
2449@item @code{and}    @tab @code{&&}
2450@item @code{and_eq} @tab @code{&=}
2451@item @code{bitand} @tab @code{&}
2452@item @code{bitor}  @tab @code{|}
2453@item @code{compl}  @tab @code{~}
2454@item @code{not}    @tab @code{!}
2455@item @code{not_eq} @tab @code{!=}
2456@item @code{or}     @tab @code{||}
2457@item @code{or_eq}  @tab @code{|=}
2458@item @code{xor}    @tab @code{^}
2459@item @code{xor_eq} @tab @code{^=}
2460@end multitable
2461
2462@node Undefining and Redefining Macros
2463@section Undefining and Redefining Macros
2464@cindex undefining macros
2465@cindex redefining macros
2466@findex #undef
2467
2468If a macro ceases to be useful, it may be @dfn{undefined} with the
2469@samp{#undef} directive.  @samp{#undef} takes a single argument, the
2470name of the macro to undefine.  You use the bare macro name, even if the
2471macro is function-like.  It is an error if anything appears on the line
2472after the macro name.  @samp{#undef} has no effect if the name is not a
2473macro.
2474
2475@smallexample
2476#define FOO 4
2477x = FOO;        @expansion{} x = 4;
2478#undef FOO
2479x = FOO;        @expansion{} x = FOO;
2480@end smallexample
2481
2482Once a macro has been undefined, that identifier may be @dfn{redefined}
2483as a macro by a subsequent @samp{#define} directive.  The new definition
2484need not have any resemblance to the old definition.
2485
2486However, if an identifier which is currently a macro is redefined, then
2487the new definition must be @dfn{effectively the same} as the old one.
2488Two macro definitions are effectively the same if:
2489@itemize @bullet
2490@item Both are the same type of macro (object- or function-like).
2491@item All the tokens of the replacement list are the same.
2492@item If there are any parameters, they are the same.
2493@item Whitespace appears in the same places in both.  It need not be
2494exactly the same amount of whitespace, though.  Remember that comments
2495count as whitespace.
2496@end itemize
2497
2498@noindent
2499These definitions are effectively the same:
2500@smallexample
2501#define FOUR (2 + 2)
2502#define FOUR         (2    +    2)
2503#define FOUR (2 /* @r{two} */ + 2)
2504@end smallexample
2505@noindent
2506but these are not:
2507@smallexample
2508#define FOUR (2 + 2)
2509#define FOUR ( 2+2 )
2510#define FOUR (2 * 2)
2511#define FOUR(score,and,seven,years,ago) (2 + 2)
2512@end smallexample
2513
2514If a macro is redefined with a definition that is not effectively the
2515same as the old one, the preprocessor issues a warning and changes the
2516macro to use the new definition.  If the new definition is effectively
2517the same, the redefinition is silently ignored.  This allows, for
2518instance, two different headers to define a common macro.  The
2519preprocessor will only complain if the definitions do not match.
2520
2521@node Directives Within Macro Arguments
2522@section Directives Within Macro Arguments
2523@cindex macro arguments and directives
2524
2525Occasionally it is convenient to use preprocessor directives within
2526the arguments of a macro.  The C and C++ standards declare that
2527behavior in these cases is undefined.
2528
2529Versions of CPP prior to 3.2 would reject such constructs with an
2530error message.  This was the only syntactic difference between normal
2531functions and function-like macros, so it seemed attractive to remove
2532this limitation, and people would often be surprised that they could
2533not use macros in this way.  Moreover, sometimes people would use
2534conditional compilation in the argument list to a normal library
2535function like @samp{printf}, only to find that after a library upgrade
2536@samp{printf} had changed to be a function-like macro, and their code
2537would no longer compile.  So from version 3.2 we changed CPP to
2538successfully process arbitrary directives within macro arguments in
2539exactly the same way as it would have processed the directive were the
2540function-like macro invocation not present.
2541
2542If, within a macro invocation, that macro is redefined, then the new
2543definition takes effect in time for argument pre-expansion, but the
2544original definition is still used for argument replacement.  Here is a
2545pathological example:
2546
2547@smallexample
2548#define f(x) x x
2549f (1
2550#undef f
2551#define f 2
2552f)
2553@end smallexample
2554
2555@noindent
2556which expands to
2557
2558@smallexample
25591 2 1 2
2560@end smallexample
2561
2562@noindent
2563with the semantics described above.
2564
2565@node Macro Pitfalls
2566@section Macro Pitfalls
2567@cindex problems with macros
2568@cindex pitfalls of macros
2569
2570In this section we describe some special rules that apply to macros and
2571macro expansion, and point out certain cases in which the rules have
2572counter-intuitive consequences that you must watch out for.
2573
2574@menu
2575* Misnesting::
2576* Operator Precedence Problems::
2577* Swallowing the Semicolon::
2578* Duplication of Side Effects::
2579* Self-Referential Macros::
2580* Argument Prescan::
2581* Newlines in Arguments::
2582@end menu
2583
2584@node Misnesting
2585@subsection Misnesting
2586
2587When a macro is called with arguments, the arguments are substituted
2588into the macro body and the result is checked, together with the rest of
2589the input file, for more macro calls.  It is possible to piece together
2590a macro call coming partially from the macro body and partially from the
2591arguments.  For example,
2592
2593@smallexample
2594#define twice(x) (2*(x))
2595#define call_with_1(x) x(1)
2596call_with_1 (twice)
2597     @expansion{} twice(1)
2598     @expansion{} (2*(1))
2599@end smallexample
2600
2601Macro definitions do not have to have balanced parentheses.  By writing
2602an unbalanced open parenthesis in a macro body, it is possible to create
2603a macro call that begins inside the macro body but ends outside of it.
2604For example,
2605
2606@smallexample
2607#define strange(file) fprintf (file, "%s %d",
2608@dots{}
2609strange(stderr) p, 35)
2610     @expansion{} fprintf (stderr, "%s %d", p, 35)
2611@end smallexample
2612
2613The ability to piece together a macro call can be useful, but the use of
2614unbalanced open parentheses in a macro body is just confusing, and
2615should be avoided.
2616
2617@node Operator Precedence Problems
2618@subsection Operator Precedence Problems
2619@cindex parentheses in macro bodies
2620
2621You may have noticed that in most of the macro definition examples shown
2622above, each occurrence of a macro argument name had parentheses around
2623it.  In addition, another pair of parentheses usually surround the
2624entire macro definition.  Here is why it is best to write macros that
2625way.
2626
2627Suppose you define a macro as follows,
2628
2629@smallexample
2630#define ceil_div(x, y) (x + y - 1) / y
2631@end smallexample
2632
2633@noindent
2634whose purpose is to divide, rounding up.  (One use for this operation is
2635to compute how many @code{int} objects are needed to hold a certain
2636number of @code{char} objects.)  Then suppose it is used as follows:
2637
2638@smallexample
2639a = ceil_div (b & c, sizeof (int));
2640     @expansion{} a = (b & c + sizeof (int) - 1) / sizeof (int);
2641@end smallexample
2642
2643@noindent
2644This does not do what is intended.  The operator-precedence rules of
2645C make it equivalent to this:
2646
2647@smallexample
2648a = (b & (c + sizeof (int) - 1)) / sizeof (int);
2649@end smallexample
2650
2651@noindent
2652What we want is this:
2653
2654@smallexample
2655a = ((b & c) + sizeof (int) - 1)) / sizeof (int);
2656@end smallexample
2657
2658@noindent
2659Defining the macro as
2660
2661@smallexample
2662#define ceil_div(x, y) ((x) + (y) - 1) / (y)
2663@end smallexample
2664
2665@noindent
2666provides the desired result.
2667
2668Unintended grouping can result in another way.  Consider @code{sizeof
2669ceil_div(1, 2)}.  That has the appearance of a C expression that would
2670compute the size of the type of @code{ceil_div (1, 2)}, but in fact it
2671means something very different.  Here is what it expands to:
2672
2673@smallexample
2674sizeof ((1) + (2) - 1) / (2)
2675@end smallexample
2676
2677@noindent
2678This would take the size of an integer and divide it by two.  The
2679precedence rules have put the division outside the @code{sizeof} when it
2680was intended to be inside.
2681
2682Parentheses around the entire macro definition prevent such problems.
2683Here, then, is the recommended way to define @code{ceil_div}:
2684
2685@smallexample
2686#define ceil_div(x, y) (((x) + (y) - 1) / (y))
2687@end smallexample
2688
2689@node Swallowing the Semicolon
2690@subsection Swallowing the Semicolon
2691@cindex semicolons (after macro calls)
2692
2693Often it is desirable to define a macro that expands into a compound
2694statement.  Consider, for example, the following macro, that advances a
2695pointer (the argument @code{p} says where to find it) across whitespace
2696characters:
2697
2698@smallexample
2699#define SKIP_SPACES(p, limit)  \
2700@{ char *lim = (limit);         \
2701  while (p < lim) @{            \
2702    if (*p++ != ' ') @{         \
2703      p--; break; @}@}@}
2704@end smallexample
2705
2706@noindent
2707Here backslash-newline is used to split the macro definition, which must
2708be a single logical line, so that it resembles the way such code would
2709be laid out if not part of a macro definition.
2710
2711A call to this macro might be @code{SKIP_SPACES (p, lim)}.  Strictly
2712speaking, the call expands to a compound statement, which is a complete
2713statement with no need for a semicolon to end it.  However, since it
2714looks like a function call, it minimizes confusion if you can use it
2715like a function call, writing a semicolon afterward, as in
2716@code{SKIP_SPACES (p, lim);}
2717
2718This can cause trouble before @code{else} statements, because the
2719semicolon is actually a null statement.  Suppose you write
2720
2721@smallexample
2722if (*p != 0)
2723  SKIP_SPACES (p, lim);
2724else @dots{}
2725@end smallexample
2726
2727@noindent
2728The presence of two statements---the compound statement and a null
2729statement---in between the @code{if} condition and the @code{else}
2730makes invalid C code.
2731
2732The definition of the macro @code{SKIP_SPACES} can be altered to solve
2733this problem, using a @code{do @dots{} while} statement.  Here is how:
2734
2735@smallexample
2736#define SKIP_SPACES(p, limit)     \
2737do @{ char *lim = (limit);         \
2738     while (p < lim) @{            \
2739       if (*p++ != ' ') @{         \
2740         p--; break; @}@}@}          \
2741while (0)
2742@end smallexample
2743
2744Now @code{SKIP_SPACES (p, lim);} expands into
2745
2746@smallexample
2747do @{@dots{}@} while (0);
2748@end smallexample
2749
2750@noindent
2751which is one statement.  The loop executes exactly once; most compilers
2752generate no extra code for it.
2753
2754@node Duplication of Side Effects
2755@subsection Duplication of Side Effects
2756
2757@cindex side effects (in macro arguments)
2758@cindex unsafe macros
2759Many C programs define a macro @code{min}, for ``minimum'', like this:
2760
2761@smallexample
2762#define min(X, Y)  ((X) < (Y) ? (X) : (Y))
2763@end smallexample
2764
2765When you use this macro with an argument containing a side effect,
2766as shown here,
2767
2768@smallexample
2769next = min (x + y, foo (z));
2770@end smallexample
2771
2772@noindent
2773it expands as follows:
2774
2775@smallexample
2776next = ((x + y) < (foo (z)) ? (x + y) : (foo (z)));
2777@end smallexample
2778
2779@noindent
2780where @code{x + y} has been substituted for @code{X} and @code{foo (z)}
2781for @code{Y}.
2782
2783The function @code{foo} is used only once in the statement as it appears
2784in the program, but the expression @code{foo (z)} has been substituted
2785twice into the macro expansion.  As a result, @code{foo} might be called
2786two times when the statement is executed.  If it has side effects or if
2787it takes a long time to compute, the results might not be what you
2788intended.  We say that @code{min} is an @dfn{unsafe} macro.
2789
2790The best solution to this problem is to define @code{min} in a way that
2791computes the value of @code{foo (z)} only once.  The C language offers
2792no standard way to do this, but it can be done with GNU extensions as
2793follows:
2794
2795@smallexample
2796#define min(X, Y)                \
2797(@{ typeof (X) x_ = (X);          \
2798   typeof (Y) y_ = (Y);          \
2799   (x_ < y_) ? x_ : y_; @})
2800@end smallexample
2801
2802The @samp{(@{ @dots{} @})} notation produces a compound statement that
2803acts as an expression.  Its value is the value of its last statement.
2804This permits us to define local variables and assign each argument to
2805one.  The local variables have underscores after their names to reduce
2806the risk of conflict with an identifier of wider scope (it is impossible
2807to avoid this entirely).  Now each argument is evaluated exactly once.
2808
2809If you do not wish to use GNU C extensions, the only solution is to be
2810careful when @emph{using} the macro @code{min}.  For example, you can
2811calculate the value of @code{foo (z)}, save it in a variable, and use
2812that variable in @code{min}:
2813
2814@smallexample
2815@group
2816#define min(X, Y)  ((X) < (Y) ? (X) : (Y))
2817@dots{}
2818@{
2819  int tem = foo (z);
2820  next = min (x + y, tem);
2821@}
2822@end group
2823@end smallexample
2824
2825@noindent
2826(where we assume that @code{foo} returns type @code{int}).
2827
2828@node Self-Referential Macros
2829@subsection Self-Referential Macros
2830@cindex self-reference
2831
2832A @dfn{self-referential} macro is one whose name appears in its
2833definition.  Recall that all macro definitions are rescanned for more
2834macros to replace.  If the self-reference were considered a use of the
2835macro, it would produce an infinitely large expansion.  To prevent this,
2836the self-reference is not considered a macro call.  It is passed into
2837the preprocessor output unchanged.  Consider an example:
2838
2839@smallexample
2840#define foo (4 + foo)
2841@end smallexample
2842
2843@noindent
2844where @code{foo} is also a variable in your program.
2845
2846Following the ordinary rules, each reference to @code{foo} will expand
2847into @code{(4 + foo)}; then this will be rescanned and will expand into
2848@code{(4 + (4 + foo))}; and so on until the computer runs out of memory.
2849
2850The self-reference rule cuts this process short after one step, at
2851@code{(4 + foo)}.  Therefore, this macro definition has the possibly
2852useful effect of causing the program to add 4 to the value of @code{foo}
2853wherever @code{foo} is referred to.
2854
2855In most cases, it is a bad idea to take advantage of this feature.  A
2856person reading the program who sees that @code{foo} is a variable will
2857not expect that it is a macro as well.  The reader will come across the
2858identifier @code{foo} in the program and think its value should be that
2859of the variable @code{foo}, whereas in fact the value is four greater.
2860
2861One common, useful use of self-reference is to create a macro which
2862expands to itself.  If you write
2863
2864@smallexample
2865#define EPERM EPERM
2866@end smallexample
2867
2868@noindent
2869then the macro @code{EPERM} expands to @code{EPERM}.  Effectively, it is
2870left alone by the preprocessor whenever it's used in running text.  You
2871can tell that it's a macro with @samp{#ifdef}.  You might do this if you
2872want to define numeric constants with an @code{enum}, but have
2873@samp{#ifdef} be true for each constant.
2874
2875If a macro @code{x} expands to use a macro @code{y}, and the expansion of
2876@code{y} refers to the macro @code{x}, that is an @dfn{indirect
2877self-reference} of @code{x}.  @code{x} is not expanded in this case
2878either.  Thus, if we have
2879
2880@smallexample
2881#define x (4 + y)
2882#define y (2 * x)
2883@end smallexample
2884
2885@noindent
2886then @code{x} and @code{y} expand as follows:
2887
2888@smallexample
2889@group
2890x    @expansion{} (4 + y)
2891     @expansion{} (4 + (2 * x))
2892
2893y    @expansion{} (2 * x)
2894     @expansion{} (2 * (4 + y))
2895@end group
2896@end smallexample
2897
2898@noindent
2899Each macro is expanded when it appears in the definition of the other
2900macro, but not when it indirectly appears in its own definition.
2901
2902@node Argument Prescan
2903@subsection Argument Prescan
2904@cindex expansion of arguments
2905@cindex macro argument expansion
2906@cindex prescan of macro arguments
2907
2908Macro arguments are completely macro-expanded before they are
2909substituted into a macro body, unless they are stringified or pasted
2910with other tokens.  After substitution, the entire macro body, including
2911the substituted arguments, is scanned again for macros to be expanded.
2912The result is that the arguments are scanned @emph{twice} to expand
2913macro calls in them.
2914
2915Most of the time, this has no effect.  If the argument contained any
2916macro calls, they are expanded during the first scan.  The result
2917therefore contains no macro calls, so the second scan does not change
2918it.  If the argument were substituted as given, with no prescan, the
2919single remaining scan would find the same macro calls and produce the
2920same results.
2921
2922You might expect the double scan to change the results when a
2923self-referential macro is used in an argument of another macro
2924(@pxref{Self-Referential Macros}): the self-referential macro would be
2925expanded once in the first scan, and a second time in the second scan.
2926However, this is not what happens.  The self-references that do not
2927expand in the first scan are marked so that they will not expand in the
2928second scan either.
2929
2930You might wonder, ``Why mention the prescan, if it makes no difference?
2931And why not skip it and make the preprocessor faster?''  The answer is
2932that the prescan does make a difference in three special cases:
2933
2934@itemize @bullet
2935@item
2936Nested calls to a macro.
2937
2938We say that @dfn{nested} calls to a macro occur when a macro's argument
2939contains a call to that very macro.  For example, if @code{f} is a macro
2940that expects one argument, @code{f (f (1))} is a nested pair of calls to
2941@code{f}.  The desired expansion is made by expanding @code{f (1)} and
2942substituting that into the definition of @code{f}.  The prescan causes
2943the expected result to happen.  Without the prescan, @code{f (1)} itself
2944would be substituted as an argument, and the inner use of @code{f} would
2945appear during the main scan as an indirect self-reference and would not
2946be expanded.
2947
2948@item
2949Macros that call other macros that stringify or concatenate.
2950
2951If an argument is stringified or concatenated, the prescan does not
2952occur.  If you @emph{want} to expand a macro, then stringify or
2953concatenate its expansion, you can do that by causing one macro to call
2954another macro that does the stringification or concatenation.  For
2955instance, if you have
2956
2957@smallexample
2958#define AFTERX(x) X_ ## x
2959#define XAFTERX(x) AFTERX(x)
2960#define TABLESIZE 1024
2961#define BUFSIZE TABLESIZE
2962@end smallexample
2963
2964then @code{AFTERX(BUFSIZE)} expands to @code{X_BUFSIZE}, and
2965@code{XAFTERX(BUFSIZE)} expands to @code{X_1024}.  (Not to
2966@code{X_TABLESIZE}.  Prescan always does a complete expansion.)
2967
2968@item
2969Macros used in arguments, whose expansions contain unshielded commas.
2970
2971This can cause a macro expanded on the second scan to be called with the
2972wrong number of arguments.  Here is an example:
2973
2974@smallexample
2975#define foo  a,b
2976#define bar(x) lose(x)
2977#define lose(x) (1 + (x))
2978@end smallexample
2979
2980We would like @code{bar(foo)} to turn into @code{(1 + (foo))}, which
2981would then turn into @code{(1 + (a,b))}.  Instead, @code{bar(foo)}
2982expands into @code{lose(a,b)}, and you get an error because @code{lose}
2983requires a single argument.  In this case, the problem is easily solved
2984by the same parentheses that ought to be used to prevent misnesting of
2985arithmetic operations:
2986
2987@smallexample
2988#define foo (a,b)
2989@exdent or
2990#define bar(x) lose((x))
2991@end smallexample
2992
2993The extra pair of parentheses prevents the comma in @code{foo}'s
2994definition from being interpreted as an argument separator.
2995
2996@end itemize
2997
2998@node Newlines in Arguments
2999@subsection Newlines in Arguments
3000@cindex newlines in macro arguments
3001
3002The invocation of a function-like macro can extend over many logical
3003lines.  However, in the present implementation, the entire expansion
3004comes out on one line.  Thus line numbers emitted by the compiler or
3005debugger refer to the line the invocation started on, which might be
3006different to the line containing the argument causing the problem.
3007
3008Here is an example illustrating this:
3009
3010@smallexample
3011#define ignore_second_arg(a,b,c) a; c
3012
3013ignore_second_arg (foo (),
3014                   ignored (),
3015                   syntax error);
3016@end smallexample
3017
3018@noindent
3019The syntax error triggered by the tokens @code{syntax error} results in
3020an error message citing line three---the line of ignore_second_arg---
3021even though the problematic code comes from line five.
3022
3023We consider this a bug, and intend to fix it in the near future.
3024
3025@node Conditionals
3026@chapter Conditionals
3027@cindex conditionals
3028
3029A @dfn{conditional} is a directive that instructs the preprocessor to
3030select whether or not to include a chunk of code in the final token
3031stream passed to the compiler.  Preprocessor conditionals can test
3032arithmetic expressions, or whether a name is defined as a macro, or both
3033simultaneously using the special @code{defined} operator.
3034
3035A conditional in the C preprocessor resembles in some ways an @code{if}
3036statement in C, but it is important to understand the difference between
3037them.  The condition in an @code{if} statement is tested during the
3038execution of your program.  Its purpose is to allow your program to
3039behave differently from run to run, depending on the data it is
3040operating on.  The condition in a preprocessing conditional directive is
3041tested when your program is compiled.  Its purpose is to allow different
3042code to be included in the program depending on the situation at the
3043time of compilation.
3044
3045However, the distinction is becoming less clear.  Modern compilers often
3046do test @code{if} statements when a program is compiled, if their
3047conditions are known not to vary at run time, and eliminate code which
3048can never be executed.  If you can count on your compiler to do this,
3049you may find that your program is more readable if you use @code{if}
3050statements with constant conditions (perhaps determined by macros).  Of
3051course, you can only use this to exclude code, not type definitions or
3052other preprocessing directives, and you can only do it if the code
3053remains syntactically valid when it is not to be used.
3054
3055GCC version 3 eliminates this kind of never-executed code even when
3056not optimizing.  Older versions did it only when optimizing.
3057
3058@menu
3059* Conditional Uses::
3060* Conditional Syntax::
3061* Deleted Code::
3062@end menu
3063
3064@node Conditional Uses
3065@section Conditional Uses
3066
3067There are three general reasons to use a conditional.
3068
3069@itemize @bullet
3070@item
3071A program may need to use different code depending on the machine or
3072operating system it is to run on.  In some cases the code for one
3073operating system may be erroneous on another operating system; for
3074example, it might refer to data types or constants that do not exist on
3075the other system.  When this happens, it is not enough to avoid
3076executing the invalid code.  Its mere presence will cause the compiler
3077to reject the program.  With a preprocessing conditional, the offending
3078code can be effectively excised from the program when it is not valid.
3079
3080@item
3081You may want to be able to compile the same source file into two
3082different programs.  One version might make frequent time-consuming
3083consistency checks on its intermediate data, or print the values of
3084those data for debugging, and the other not.
3085
3086@item
3087A conditional whose condition is always false is one way to exclude code
3088from the program but keep it as a sort of comment for future reference.
3089@end itemize
3090
3091Simple programs that do not need system-specific logic or complex
3092debugging hooks generally will not need to use preprocessing
3093conditionals.
3094
3095@node Conditional Syntax
3096@section Conditional Syntax
3097
3098@findex #if
3099A conditional in the C preprocessor begins with a @dfn{conditional
3100directive}: @samp{#if}, @samp{#ifdef} or @samp{#ifndef}.
3101
3102@menu
3103* Ifdef::
3104* If::
3105* Defined::
3106* Else::
3107* Elif::
3108@end menu
3109
3110@node Ifdef
3111@subsection Ifdef
3112@findex #ifdef
3113@findex #endif
3114
3115The simplest sort of conditional is
3116
3117@smallexample
3118@group
3119#ifdef @var{MACRO}
3120
3121@var{controlled text}
3122
3123#endif /* @var{MACRO} */
3124@end group
3125@end smallexample
3126
3127@cindex conditional group
3128This block is called a @dfn{conditional group}.  @var{controlled text}
3129will be included in the output of the preprocessor if and only if
3130@var{MACRO} is defined.  We say that the conditional @dfn{succeeds} if
3131@var{MACRO} is defined, @dfn{fails} if it is not.
3132
3133The @var{controlled text} inside of a conditional can include
3134preprocessing directives.  They are executed only if the conditional
3135succeeds.  You can nest conditional groups inside other conditional
3136groups, but they must be completely nested.  In other words,
3137@samp{#endif} always matches the nearest @samp{#ifdef} (or
3138@samp{#ifndef}, or @samp{#if}).  Also, you cannot start a conditional
3139group in one file and end it in another.
3140
3141Even if a conditional fails, the @var{controlled text} inside it is
3142still run through initial transformations and tokenization.  Therefore,
3143it must all be lexically valid C@.  Normally the only way this matters is
3144that all comments and string literals inside a failing conditional group
3145must still be properly ended.
3146
3147The comment following the @samp{#endif} is not required, but it is a
3148good practice if there is a lot of @var{controlled text}, because it
3149helps people match the @samp{#endif} to the corresponding @samp{#ifdef}.
3150Older programs sometimes put @var{MACRO} directly after the
3151@samp{#endif} without enclosing it in a comment.  This is invalid code
3152according to the C standard.  CPP accepts it with a warning.  It
3153never affects which @samp{#ifndef} the @samp{#endif} matches.
3154
3155@findex #ifndef
3156Sometimes you wish to use some code if a macro is @emph{not} defined.
3157You can do this by writing @samp{#ifndef} instead of @samp{#ifdef}.
3158One common use of @samp{#ifndef} is to include code only the first
3159time a header file is included.  @xref{Once-Only Headers}.
3160
3161Macro definitions can vary between compilations for several reasons.
3162Here are some samples.
3163
3164@itemize @bullet
3165@item
3166Some macros are predefined on each kind of machine
3167(@pxref{System-specific Predefined Macros}).  This allows you to provide
3168code specially tuned for a particular machine.
3169
3170@item
3171System header files define more macros, associated with the features
3172they implement.  You can test these macros with conditionals to avoid
3173using a system feature on a machine where it is not implemented.
3174
3175@item
3176Macros can be defined or undefined with the @option{-D} and @option{-U}
3177command line options when you compile the program.  You can arrange to
3178compile the same source file into two different programs by choosing a
3179macro name to specify which program you want, writing conditionals to
3180test whether or how this macro is defined, and then controlling the
3181state of the macro with command line options, perhaps set in the
3182Makefile.  @xref{Invocation}.
3183
3184@item
3185Your program might have a special header file (often called
3186@file{config.h}) that is adjusted when the program is compiled.  It can
3187define or not define macros depending on the features of the system and
3188the desired capabilities of the program.  The adjustment can be
3189automated by a tool such as @command{autoconf}, or done by hand.
3190@end itemize
3191
3192@node If
3193@subsection If
3194
3195The @samp{#if} directive allows you to test the value of an arithmetic
3196expression, rather than the mere existence of one macro.  Its syntax is
3197
3198@smallexample
3199@group
3200#if @var{expression}
3201
3202@var{controlled text}
3203
3204#endif /* @var{expression} */
3205@end group
3206@end smallexample
3207
3208@var{expression} is a C expression of integer type, subject to stringent
3209restrictions.  It may contain
3210
3211@itemize @bullet
3212@item
3213Integer constants.
3214
3215@item
3216Character constants, which are interpreted as they would be in normal
3217code.
3218
3219@item
3220Arithmetic operators for addition, subtraction, multiplication,
3221division, bitwise operations, shifts, comparisons, and logical
3222operations (@code{&&} and @code{||}).  The latter two obey the usual
3223short-circuiting rules of standard C@.
3224
3225@item
3226Macros.  All macros in the expression are expanded before actual
3227computation of the expression's value begins.
3228
3229@item
3230Uses of the @code{defined} operator, which lets you check whether macros
3231are defined in the middle of an @samp{#if}.
3232
3233@item
3234Identifiers that are not macros, which are all considered to be the
3235number zero.  This allows you to write @code{@w{#if MACRO}} instead of
3236@code{@w{#ifdef MACRO}}, if you know that MACRO, when defined, will
3237always have a nonzero value.  Function-like macros used without their
3238function call parentheses are also treated as zero.
3239
3240In some contexts this shortcut is undesirable.  The @option{-Wundef}
3241option causes GCC to warn whenever it encounters an identifier which is
3242not a macro in an @samp{#if}.
3243@end itemize
3244
3245The preprocessor does not know anything about types in the language.
3246Therefore, @code{sizeof} operators are not recognized in @samp{#if}, and
3247neither are @code{enum} constants.  They will be taken as identifiers
3248which are not macros, and replaced by zero.  In the case of
3249@code{sizeof}, this is likely to cause the expression to be invalid.
3250
3251The preprocessor calculates the value of @var{expression}.  It carries
3252out all calculations in the widest integer type known to the compiler;
3253on most machines supported by GCC this is 64 bits.  This is not the same
3254rule as the compiler uses to calculate the value of a constant
3255expression, and may give different results in some cases.  If the value
3256comes out to be nonzero, the @samp{#if} succeeds and the @var{controlled
3257text} is included; otherwise it is skipped.
3258
3259@node Defined
3260@subsection Defined
3261
3262@cindex @code{defined}
3263The special operator @code{defined} is used in @samp{#if} and
3264@samp{#elif} expressions to test whether a certain name is defined as a
3265macro.  @code{defined @var{name}} and @code{defined (@var{name})} are
3266both expressions whose value is 1 if @var{name} is defined as a macro at
3267the current point in the program, and 0 otherwise.  Thus,  @code{@w{#if
3268defined MACRO}} is precisely equivalent to @code{@w{#ifdef MACRO}}.
3269
3270@code{defined} is useful when you wish to test more than one macro for
3271existence at once.  For example,
3272
3273@smallexample
3274#if defined (__vax__) || defined (__ns16000__)
3275@end smallexample
3276
3277@noindent
3278would succeed if either of the names @code{__vax__} or
3279@code{__ns16000__} is defined as a macro.
3280
3281Conditionals written like this:
3282
3283@smallexample
3284#if defined BUFSIZE && BUFSIZE >= 1024
3285@end smallexample
3286
3287@noindent
3288can generally be simplified to just @code{@w{#if BUFSIZE >= 1024}},
3289since if @code{BUFSIZE} is not defined, it will be interpreted as having
3290the value zero.
3291
3292If the @code{defined} operator appears as a result of a macro expansion,
3293the C standard says the behavior is undefined.  GNU cpp treats it as a
3294genuine @code{defined} operator and evaluates it normally.  It will warn
3295wherever your code uses this feature if you use the command-line option
3296@option{-pedantic}, since other compilers may handle it differently.
3297
3298@node Else
3299@subsection Else
3300
3301@findex #else
3302The @samp{#else} directive can be added to a conditional to provide
3303alternative text to be used if the condition fails.  This is what it
3304looks like:
3305
3306@smallexample
3307@group
3308#if @var{expression}
3309@var{text-if-true}
3310#else /* Not @var{expression} */
3311@var{text-if-false}
3312#endif /* Not @var{expression} */
3313@end group
3314@end smallexample
3315
3316@noindent
3317If @var{expression} is nonzero, the @var{text-if-true} is included and
3318the @var{text-if-false} is skipped.  If @var{expression} is zero, the
3319opposite happens.
3320
3321You can use @samp{#else} with @samp{#ifdef} and @samp{#ifndef}, too.
3322
3323@node Elif
3324@subsection Elif
3325
3326@findex #elif
3327One common case of nested conditionals is used to check for more than two
3328possible alternatives.  For example, you might have
3329
3330@smallexample
3331#if X == 1
3332@dots{}
3333#else /* X != 1 */
3334#if X == 2
3335@dots{}
3336#else /* X != 2 */
3337@dots{}
3338#endif /* X != 2 */
3339#endif /* X != 1 */
3340@end smallexample
3341
3342Another conditional directive, @samp{#elif}, allows this to be
3343abbreviated as follows:
3344
3345@smallexample
3346#if X == 1
3347@dots{}
3348#elif X == 2
3349@dots{}
3350#else /* X != 2 and X != 1*/
3351@dots{}
3352#endif /* X != 2 and X != 1*/
3353@end smallexample
3354
3355@samp{#elif} stands for ``else if''.  Like @samp{#else}, it goes in the
3356middle of a conditional group and subdivides it; it does not require a
3357matching @samp{#endif} of its own.  Like @samp{#if}, the @samp{#elif}
3358directive includes an expression to be tested.  The text following the
3359@samp{#elif} is processed only if the original @samp{#if}-condition
3360failed and the @samp{#elif} condition succeeds.
3361
3362More than one @samp{#elif} can go in the same conditional group.  Then
3363the text after each @samp{#elif} is processed only if the @samp{#elif}
3364condition succeeds after the original @samp{#if} and all previous
3365@samp{#elif} directives within it have failed.
3366
3367@samp{#else} is allowed after any number of @samp{#elif} directives, but
3368@samp{#elif} may not follow @samp{#else}.
3369
3370@node Deleted Code
3371@section Deleted Code
3372@cindex commenting out code
3373
3374If you replace or delete a part of the program but want to keep the old
3375code around for future reference, you often cannot simply comment it
3376out.  Block comments do not nest, so the first comment inside the old
3377code will end the commenting-out.  The probable result is a flood of
3378syntax errors.
3379
3380One way to avoid this problem is to use an always-false conditional
3381instead.  For instance, put @code{#if 0} before the deleted code and
3382@code{#endif} after it.  This works even if the code being turned
3383off contains conditionals, but they must be entire conditionals
3384(balanced @samp{#if} and @samp{#endif}).
3385
3386Some people use @code{#ifdef notdef} instead.  This is risky, because
3387@code{notdef} might be accidentally defined as a macro, and then the
3388conditional would succeed.  @code{#if 0} can be counted on to fail.
3389
3390Do not use @code{#if 0} for comments which are not C code.  Use a real
3391comment, instead.  The interior of @code{#if 0} must consist of complete
3392tokens; in particular, single-quote characters must balance.  Comments
3393often contain unbalanced single-quote characters (known in English as
3394apostrophes).  These confuse @code{#if 0}.  They don't confuse
3395@samp{/*}.
3396
3397@node Diagnostics
3398@chapter Diagnostics
3399@cindex diagnostic
3400@cindex reporting errors
3401@cindex reporting warnings
3402
3403@findex #error
3404The directive @samp{#error} causes the preprocessor to report a fatal
3405error.  The tokens forming the rest of the line following @samp{#error}
3406are used as the error message.
3407
3408You would use @samp{#error} inside of a conditional that detects a
3409combination of parameters which you know the program does not properly
3410support.  For example, if you know that the program will not run
3411properly on a VAX, you might write
3412
3413@smallexample
3414@group
3415#ifdef __vax__
3416#error "Won't work on VAXen.  See comments at get_last_object."
3417#endif
3418@end group
3419@end smallexample
3420
3421If you have several configuration parameters that must be set up by
3422the installation in a consistent way, you can use conditionals to detect
3423an inconsistency and report it with @samp{#error}.  For example,
3424
3425@smallexample
3426#if !defined(FOO) && defined(BAR)
3427#error "BAR requires FOO."
3428#endif
3429@end smallexample
3430
3431@findex #warning
3432The directive @samp{#warning} is like @samp{#error}, but causes the
3433preprocessor to issue a warning and continue preprocessing.  The tokens
3434following @samp{#warning} are used as the warning message.
3435
3436You might use @samp{#warning} in obsolete header files, with a message
3437directing the user to the header file which should be used instead.
3438
3439Neither @samp{#error} nor @samp{#warning} macro-expands its argument.
3440Internal whitespace sequences are each replaced with a single space.
3441The line must consist of complete tokens.  It is wisest to make the
3442argument of these directives be a single string constant; this avoids
3443problems with apostrophes and the like.
3444
3445@node Line Control
3446@chapter Line Control
3447@cindex line control
3448
3449The C preprocessor informs the C compiler of the location in your source
3450code where each token came from.  Presently, this is just the file name
3451and line number.  All the tokens resulting from macro expansion are
3452reported as having appeared on the line of the source file where the
3453outermost macro was used.  We intend to be more accurate in the future.
3454
3455If you write a program which generates source code, such as the
3456@command{bison} parser generator, you may want to adjust the preprocessor's
3457notion of the current file name and line number by hand.  Parts of the
3458output from @command{bison} are generated from scratch, other parts come
3459from a standard parser file.  The rest are copied verbatim from
3460@command{bison}'s input.  You would like compiler error messages and
3461symbolic debuggers to be able to refer to @code{bison}'s input file.
3462
3463@findex #line
3464@command{bison} or any such program can arrange this by writing
3465@samp{#line} directives into the output file.  @samp{#line} is a
3466directive that specifies the original line number and source file name
3467for subsequent input in the current preprocessor input file.
3468@samp{#line} has three variants:
3469
3470@table @code
3471@item #line @var{linenum}
3472@var{linenum} is a non-negative decimal integer constant.  It specifies
3473the line number which should be reported for the following line of
3474input.  Subsequent lines are counted from @var{linenum}.
3475
3476@item #line @var{linenum} @var{filename}
3477@var{linenum} is the same as for the first form, and has the same
3478effect.  In addition, @var{filename} is a string constant.  The
3479following line and all subsequent lines are reported to come from the
3480file it specifies, until something else happens to change that.
3481@var{filename} is interpreted according to the normal rules for a string
3482constant: backslash escapes are interpreted.  This is different from
3483@samp{#include}.
3484
3485Previous versions of CPP did not interpret escapes in @samp{#line};
3486we have changed it because the standard requires they be interpreted,
3487and most other compilers do.
3488
3489@item #line @var{anything else}
3490@var{anything else} is checked for macro calls, which are expanded.
3491The result should match one of the above two forms.
3492@end table
3493
3494@samp{#line} directives alter the results of the @code{__FILE__} and
3495@code{__LINE__} predefined macros from that point on.  @xref{Standard
3496Predefined Macros}.  They do not have any effect on @samp{#include}'s
3497idea of the directory containing the current file.  This is a change
3498from GCC 2.95.  Previously, a file reading
3499
3500@smallexample
3501#line 1 "../src/gram.y"
3502#include "gram.h"
3503@end smallexample
3504
3505would search for @file{gram.h} in @file{../src}, then the @option{-I}
3506chain; the directory containing the physical source file would not be
3507searched.  In GCC 3.0 and later, the @samp{#include} is not affected by
3508the presence of a @samp{#line} referring to a different directory.
3509
3510We made this change because the old behavior caused problems when
3511generated source files were transported between machines.  For instance,
3512it is common practice to ship generated parsers with a source release,
3513so that people building the distribution do not need to have yacc or
3514Bison installed.  These files frequently have @samp{#line} directives
3515referring to the directory tree of the system where the distribution was
3516created.  If GCC tries to search for headers in those directories, the
3517build is likely to fail.
3518
3519The new behavior can cause failures too, if the generated file is not
3520in the same directory as its source and it attempts to include a header
3521which would be visible searching from the directory containing the
3522source file.  However, this problem is easily solved with an additional
3523@option{-I} switch on the command line.  The failures caused by the old
3524semantics could sometimes be corrected only by editing the generated
3525files, which is difficult and error-prone.
3526
3527@node Pragmas
3528@chapter Pragmas
3529
3530The @samp{#pragma} directive is the method specified by the C standard
3531for providing additional information to the compiler, beyond what is
3532conveyed in the language itself.  Three forms of this directive
3533(commonly known as @dfn{pragmas}) are specified by the 1999 C standard.
3534A C compiler is free to attach any meaning it likes to other pragmas.
3535
3536GCC has historically preferred to use extensions to the syntax of the
3537language, such as @code{__attribute__}, for this purpose.  However, GCC
3538does define a few pragmas of its own.  These mostly have effects on the
3539entire translation unit or source file.
3540
3541In GCC version 3, all GNU-defined, supported pragmas have been given a
3542@code{GCC} prefix.  This is in line with the @code{STDC} prefix on all
3543pragmas defined by C99.  For backward compatibility, pragmas which were
3544recognized by previous versions are still recognized without the
3545@code{GCC} prefix, but that usage is deprecated.  Some older pragmas are
3546deprecated in their entirety.  They are not recognized with the
3547@code{GCC} prefix.  @xref{Obsolete Features}.
3548
3549@cindex @code{_Pragma}
3550C99 introduces the @code{@w{_Pragma}} operator.  This feature addresses a
3551major problem with @samp{#pragma}: being a directive, it cannot be
3552produced as the result of macro expansion.  @code{@w{_Pragma}} is an
3553operator, much like @code{sizeof} or @code{defined}, and can be embedded
3554in a macro.
3555
3556Its syntax is @code{@w{_Pragma (@var{string-literal})}}, where
3557@var{string-literal} can be either a normal or wide-character string
3558literal.  It is destringized, by replacing all @samp{\\} with a single
3559@samp{\} and all @samp{\"} with a @samp{"}.  The result is then
3560processed as if it had appeared as the right hand side of a
3561@samp{#pragma} directive.  For example,
3562
3563@smallexample
3564_Pragma ("GCC dependency \"parse.y\"")
3565@end smallexample
3566
3567@noindent
3568has the same effect as @code{#pragma GCC dependency "parse.y"}.  The
3569same effect could be achieved using macros, for example
3570
3571@smallexample
3572#define DO_PRAGMA(x) _Pragma (#x)
3573DO_PRAGMA (GCC dependency "parse.y")
3574@end smallexample
3575
3576The standard is unclear on where a @code{_Pragma} operator can appear.
3577The preprocessor does not accept it within a preprocessing conditional
3578directive like @samp{#if}.  To be safe, you are probably best keeping it
3579out of directives other than @samp{#define}, and putting it on a line of
3580its own.
3581
3582This manual documents the pragmas which are meaningful to the
3583preprocessor itself.  Other pragmas are meaningful to the C or C++
3584compilers.  They are documented in the GCC manual.
3585
3586GCC plugins may provide their own pragmas.
3587
3588@ftable @code
3589@item #pragma GCC dependency
3590@code{#pragma GCC dependency} allows you to check the relative dates of
3591the current file and another file.  If the other file is more recent than
3592the current file, a warning is issued.  This is useful if the current
3593file is derived from the other file, and should be regenerated.  The
3594other file is searched for using the normal include search path.
3595Optional trailing text can be used to give more information in the
3596warning message.
3597
3598@smallexample
3599#pragma GCC dependency "parse.y"
3600#pragma GCC dependency "/usr/include/time.h" rerun fixincludes
3601@end smallexample
3602
3603@item #pragma GCC poison
3604Sometimes, there is an identifier that you want to remove completely
3605from your program, and make sure that it never creeps back in.  To
3606enforce this, you can @dfn{poison} the identifier with this pragma.
3607@code{#pragma GCC poison} is followed by a list of identifiers to
3608poison.  If any of those identifiers appears anywhere in the source
3609after the directive, it is a hard error.  For example,
3610
3611@smallexample
3612#pragma GCC poison printf sprintf fprintf
3613sprintf(some_string, "hello");
3614@end smallexample
3615
3616@noindent
3617will produce an error.
3618
3619If a poisoned identifier appears as part of the expansion of a macro
3620which was defined before the identifier was poisoned, it will @emph{not}
3621cause an error.  This lets you poison an identifier without worrying
3622about system headers defining macros that use it.
3623
3624For example,
3625
3626@smallexample
3627#define strrchr rindex
3628#pragma GCC poison rindex
3629strrchr(some_string, 'h');
3630@end smallexample
3631
3632@noindent
3633will not produce an error.
3634
3635@item #pragma GCC system_header
3636This pragma takes no arguments.  It causes the rest of the code in the
3637current file to be treated as if it came from a system header.
3638@xref{System Headers}.
3639
3640@item #pragma GCC warning
3641@itemx #pragma GCC error
3642@code{#pragma GCC warning "message"} causes the preprocessor to issue
3643a warning diagnostic with the text @samp{message}.  The message
3644contained in the pragma must be a single string literal.  Similarly,
3645@code{#pragma GCC error "message"} issues an error message.  Unlike
3646the @samp{#warning} and @samp{#error} directives, these pragmas can be
3647embedded in preprocessor macros using @samp{_Pragma}.
3648
3649@end ftable
3650
3651@node Other Directives
3652@chapter Other Directives
3653
3654@findex #ident
3655@findex #sccs
3656The @samp{#ident} directive takes one argument, a string constant.  On
3657some systems, that string constant is copied into a special segment of
3658the object file.  On other systems, the directive is ignored.  The
3659@samp{#sccs} directive is a synonym for @samp{#ident}.
3660
3661These directives are not part of the C standard, but they are not
3662official GNU extensions either.  What historical information we have
3663been able to find, suggests they originated with System V@.
3664
3665@cindex null directive
3666The @dfn{null directive} consists of a @samp{#} followed by a newline,
3667with only whitespace (including comments) in between.  A null directive
3668is understood as a preprocessing directive but has no effect on the
3669preprocessor output.  The primary significance of the existence of the
3670null directive is that an input line consisting of just a @samp{#} will
3671produce no output, rather than a line of output containing just a
3672@samp{#}.  Supposedly some old C programs contain such lines.
3673
3674@node Preprocessor Output
3675@chapter Preprocessor Output
3676
3677When the C preprocessor is used with the C, C++, or Objective-C
3678compilers, it is integrated into the compiler and communicates a stream
3679of binary tokens directly to the compiler's parser.  However, it can
3680also be used in the more conventional standalone mode, where it produces
3681textual output.
3682@c FIXME: Document the library interface.
3683
3684@cindex output format
3685The output from the C preprocessor looks much like the input, except
3686that all preprocessing directive lines have been replaced with blank
3687lines and all comments with spaces.  Long runs of blank lines are
3688discarded.
3689
3690The ISO standard specifies that it is implementation defined whether a
3691preprocessor preserves whitespace between tokens, or replaces it with
3692e.g.@: a single space.  In GNU CPP, whitespace between tokens is collapsed
3693to become a single space, with the exception that the first token on a
3694non-directive line is preceded with sufficient spaces that it appears in
3695the same column in the preprocessed output that it appeared in the
3696original source file.  This is so the output is easy to read.
3697@xref{Differences from previous versions}.  CPP does not insert any
3698whitespace where there was none in the original source, except where
3699necessary to prevent an accidental token paste.
3700
3701@cindex linemarkers
3702Source file name and line number information is conveyed by lines
3703of the form
3704
3705@smallexample
3706# @var{linenum} @var{filename} @var{flags}
3707@end smallexample
3708
3709@noindent
3710These are called @dfn{linemarkers}.  They are inserted as needed into
3711the output (but never within a string or character constant).  They mean
3712that the following line originated in file @var{filename} at line
3713@var{linenum}.  @var{filename} will never contain any non-printing
3714characters; they are replaced with octal escape sequences.
3715
3716After the file name comes zero or more flags, which are @samp{1},
3717@samp{2}, @samp{3}, or @samp{4}.  If there are multiple flags, spaces
3718separate them.  Here is what the flags mean:
3719
3720@table @samp
3721@item 1
3722This indicates the start of a new file.
3723@item 2
3724This indicates returning to a file (after having included another file).
3725@item 3
3726This indicates that the following text comes from a system header file,
3727so certain warnings should be suppressed.
3728@item 4
3729This indicates that the following text should be treated as being
3730wrapped in an implicit @code{extern "C"} block.
3731@c maybe cross reference NO_IMPLICIT_EXTERN_C
3732@end table
3733
3734As an extension, the preprocessor accepts linemarkers in non-assembler
3735input files.  They are treated like the corresponding @samp{#line}
3736directive, (@pxref{Line Control}), except that trailing flags are
3737permitted, and are interpreted with the meanings described above.  If
3738multiple flags are given, they must be in ascending order.
3739
3740Some directives may be duplicated in the output of the preprocessor.
3741These are @samp{#ident} (always), @samp{#pragma} (only if the
3742preprocessor does not handle the pragma itself), and @samp{#define} and
3743@samp{#undef} (with certain debugging options).  If this happens, the
3744@samp{#} of the directive will always be in the first column, and there
3745will be no space between the @samp{#} and the directive name.  If macro
3746expansion happens to generate tokens which might be mistaken for a
3747duplicated directive, a space will be inserted between the @samp{#} and
3748the directive name.
3749
3750@node Traditional Mode
3751@chapter Traditional Mode
3752
3753Traditional (pre-standard) C preprocessing is rather different from
3754the preprocessing specified by the standard.  When GCC is given the
3755@option{-traditional-cpp} option, it attempts to emulate a traditional
3756preprocessor.
3757
3758GCC versions 3.2 and later only support traditional mode semantics in
3759the preprocessor, and not in the compiler front ends.  This chapter
3760outlines the traditional preprocessor semantics we implemented.
3761
3762The implementation does not correspond precisely to the behavior of
3763earlier versions of GCC, nor to any true traditional preprocessor.
3764After all, inconsistencies among traditional implementations were a
3765major motivation for C standardization.  However, we intend that it
3766should be compatible with true traditional preprocessors in all ways
3767that actually matter.
3768
3769@menu
3770* Traditional lexical analysis::
3771* Traditional macros::
3772* Traditional miscellany::
3773* Traditional warnings::
3774@end menu
3775
3776@node Traditional lexical analysis
3777@section Traditional lexical analysis
3778
3779The traditional preprocessor does not decompose its input into tokens
3780the same way a standards-conforming preprocessor does.  The input is
3781simply treated as a stream of text with minimal internal form.
3782
3783This implementation does not treat trigraphs (@pxref{trigraphs})
3784specially since they were an invention of the standards committee.  It
3785handles arbitrarily-positioned escaped newlines properly and splices
3786the lines as you would expect; many traditional preprocessors did not
3787do this.
3788
3789The form of horizontal whitespace in the input file is preserved in
3790the output.  In particular, hard tabs remain hard tabs.  This can be
3791useful if, for example, you are preprocessing a Makefile.
3792
3793Traditional CPP only recognizes C-style block comments, and treats the
3794@samp{/*} sequence as introducing a comment only if it lies outside
3795quoted text.  Quoted text is introduced by the usual single and double
3796quotes, and also by an initial @samp{<} in a @code{#include}
3797directive.
3798
3799Traditionally, comments are completely removed and are not replaced
3800with a space.  Since a traditional compiler does its own tokenization
3801of the output of the preprocessor, this means that comments can
3802effectively be used as token paste operators.  However, comments
3803behave like separators for text handled by the preprocessor itself,
3804since it doesn't re-lex its input.  For example, in
3805
3806@smallexample
3807#if foo/**/bar
3808@end smallexample
3809
3810@noindent
3811@samp{foo} and @samp{bar} are distinct identifiers and expanded
3812separately if they happen to be macros.  In other words, this
3813directive is equivalent to
3814
3815@smallexample
3816#if foo bar
3817@end smallexample
3818
3819@noindent
3820rather than
3821
3822@smallexample
3823#if foobar
3824@end smallexample
3825
3826Generally speaking, in traditional mode an opening quote need not have
3827a matching closing quote.  In particular, a macro may be defined with
3828replacement text that contains an unmatched quote.  Of course, if you
3829attempt to compile preprocessed output containing an unmatched quote
3830you will get a syntax error.
3831
3832However, all preprocessing directives other than @code{#define}
3833require matching quotes.  For example:
3834
3835@smallexample
3836#define m This macro's fine and has an unmatched quote
3837"/* This is not a comment.  */
3838/* @r{This is a comment.  The following #include directive
3839   is ill-formed.}  */
3840#include <stdio.h
3841@end smallexample
3842
3843Just as for the ISO preprocessor, what would be a closing quote can be
3844escaped with a backslash to prevent the quoted text from closing.
3845
3846@node Traditional macros
3847@section Traditional macros
3848
3849The major difference between traditional and ISO macros is that the
3850former expand to text rather than to a token sequence.  CPP removes
3851all leading and trailing horizontal whitespace from a macro's
3852replacement text before storing it, but preserves the form of internal
3853whitespace.
3854
3855One consequence is that it is legitimate for the replacement text to
3856contain an unmatched quote (@pxref{Traditional lexical analysis}).  An
3857unclosed string or character constant continues into the text
3858following the macro call.  Similarly, the text at the end of a macro's
3859expansion can run together with the text after the macro invocation to
3860produce a single token.
3861
3862Normally comments are removed from the replacement text after the
3863macro is expanded, but if the @option{-CC} option is passed on the
3864command line comments are preserved.  (In fact, the current
3865implementation removes comments even before saving the macro
3866replacement text, but it careful to do it in such a way that the
3867observed effect is identical even in the function-like macro case.)
3868
3869The ISO stringification operator @samp{#} and token paste operator
3870@samp{##} have no special meaning.  As explained later, an effect
3871similar to these operators can be obtained in a different way.  Macro
3872names that are embedded in quotes, either from the main file or after
3873macro replacement, do not expand.
3874
3875CPP replaces an unquoted object-like macro name with its replacement
3876text, and then rescans it for further macros to replace.  Unlike
3877standard macro expansion, traditional macro expansion has no provision
3878to prevent recursion.  If an object-like macro appears unquoted in its
3879replacement text, it will be replaced again during the rescan pass,
3880and so on @emph{ad infinitum}.  GCC detects when it is expanding
3881recursive macros, emits an error message, and continues after the
3882offending macro invocation.
3883
3884@smallexample
3885#define PLUS +
3886#define INC(x) PLUS+x
3887INC(foo);
3888     @expansion{} ++foo;
3889@end smallexample
3890
3891Function-like macros are similar in form but quite different in
3892behavior to their ISO counterparts.  Their arguments are contained
3893within parentheses, are comma-separated, and can cross physical lines.
3894Commas within nested parentheses are not treated as argument
3895separators.  Similarly, a quote in an argument cannot be left
3896unclosed; a following comma or parenthesis that comes before the
3897closing quote is treated like any other character.  There is no
3898facility for handling variadic macros.
3899
3900This implementation removes all comments from macro arguments, unless
3901the @option{-C} option is given.  The form of all other horizontal
3902whitespace in arguments is preserved, including leading and trailing
3903whitespace.  In particular
3904
3905@smallexample
3906f( )
3907@end smallexample
3908
3909@noindent
3910is treated as an invocation of the macro @samp{f} with a single
3911argument consisting of a single space.  If you want to invoke a
3912function-like macro that takes no arguments, you must not leave any
3913whitespace between the parentheses.
3914
3915If a macro argument crosses a new line, the new line is replaced with
3916a space when forming the argument.  If the previous line contained an
3917unterminated quote, the following line inherits the quoted state.
3918
3919Traditional preprocessors replace parameters in the replacement text
3920with their arguments regardless of whether the parameters are within
3921quotes or not.  This provides a way to stringize arguments.  For
3922example
3923
3924@smallexample
3925#define str(x) "x"
3926str(/* @r{A comment} */some text )
3927     @expansion{} "some text "
3928@end smallexample
3929
3930@noindent
3931Note that the comment is removed, but that the trailing space is
3932preserved.  Here is an example of using a comment to effect token
3933pasting.
3934
3935@smallexample
3936#define suffix(x) foo_/**/x
3937suffix(bar)
3938     @expansion{} foo_bar
3939@end smallexample
3940
3941@node Traditional miscellany
3942@section Traditional miscellany
3943
3944Here are some things to be aware of when using the traditional
3945preprocessor.
3946
3947@itemize @bullet
3948@item
3949Preprocessing directives are recognized only when their leading
3950@samp{#} appears in the first column.  There can be no whitespace
3951between the beginning of the line and the @samp{#}, but whitespace can
3952follow the @samp{#}.
3953
3954@item
3955A true traditional C preprocessor does not recognize @samp{#error} or
3956@samp{#pragma}, and may not recognize @samp{#elif}.  CPP supports all
3957the directives in traditional mode that it supports in ISO mode,
3958including extensions, with the exception that the effects of
3959@samp{#pragma GCC poison} are undefined.
3960
3961@item
3962__STDC__ is not defined.
3963
3964@item
3965If you use digraphs the behavior is undefined.
3966
3967@item
3968If a line that looks like a directive appears within macro arguments,
3969the behavior is undefined.
3970
3971@end itemize
3972
3973@node Traditional warnings
3974@section Traditional warnings
3975You can request warnings about features that did not exist, or worked
3976differently, in traditional C with the @option{-Wtraditional} option.
3977GCC does not warn about features of ISO C which you must use when you
3978are using a conforming compiler, such as the @samp{#} and @samp{##}
3979operators.
3980
3981Presently @option{-Wtraditional} warns about:
3982
3983@itemize @bullet
3984@item
3985Macro parameters that appear within string literals in the macro body.
3986In traditional C macro replacement takes place within string literals,
3987but does not in ISO C@.
3988
3989@item
3990In traditional C, some preprocessor directives did not exist.
3991Traditional preprocessors would only consider a line to be a directive
3992if the @samp{#} appeared in column 1 on the line.  Therefore
3993@option{-Wtraditional} warns about directives that traditional C
3994understands but would ignore because the @samp{#} does not appear as the
3995first character on the line.  It also suggests you hide directives like
3996@samp{#pragma} not understood by traditional C by indenting them.  Some
3997traditional implementations would not recognize @samp{#elif}, so it
3998suggests avoiding it altogether.
3999
4000@item
4001A function-like macro that appears without an argument list.  In some
4002traditional preprocessors this was an error.  In ISO C it merely means
4003that the macro is not expanded.
4004
4005@item
4006The unary plus operator.  This did not exist in traditional C@.
4007
4008@item
4009The @samp{U} and @samp{LL} integer constant suffixes, which were not
4010available in traditional C@.  (Traditional C does support the @samp{L}
4011suffix for simple long integer constants.)  You are not warned about
4012uses of these suffixes in macros defined in system headers.  For
4013instance, @code{UINT_MAX} may well be defined as @code{4294967295U}, but
4014you will not be warned if you use @code{UINT_MAX}.
4015
4016You can usually avoid the warning, and the related warning about
4017constants which are so large that they are unsigned, by writing the
4018integer constant in question in hexadecimal, with no U suffix.  Take
4019care, though, because this gives the wrong result in exotic cases.
4020@end itemize
4021
4022@node Implementation Details
4023@chapter Implementation Details
4024
4025Here we document details of how the preprocessor's implementation
4026affects its user-visible behavior.  You should try to avoid undue
4027reliance on behavior described here, as it is possible that it will
4028change subtly in future implementations.
4029
4030Also documented here are obsolete features and changes from previous
4031versions of CPP@.
4032
4033@menu
4034* Implementation-defined behavior::
4035* Implementation limits::
4036* Obsolete Features::
4037* Differences from previous versions::
4038@end menu
4039
4040@node Implementation-defined behavior
4041@section Implementation-defined behavior
4042@cindex implementation-defined behavior
4043
4044This is how CPP behaves in all the cases which the C standard
4045describes as @dfn{implementation-defined}.  This term means that the
4046implementation is free to do what it likes, but must document its choice
4047and stick to it.
4048@c FIXME: Check the C++ standard for more implementation-defined stuff.
4049
4050@itemize @bullet
4051@need 1000
4052@item The mapping of physical source file multi-byte characters to the
4053execution character set.
4054
4055The input character set can be specified using the
4056@option{-finput-charset} option, while the execution character set may
4057be controlled using the @option{-fexec-charset} and
4058@option{-fwide-exec-charset} options.
4059
4060@item Identifier characters.
4061@anchor{Identifier characters}
4062
4063The C and C++ standards allow identifiers to be composed of @samp{_}
4064and the alphanumeric characters.  C++ and C99 also allow universal
4065character names, and C99 further permits implementation-defined
4066characters.  GCC currently only permits universal character names if
4067@option{-fextended-identifiers} is used, because the implementation of
4068universal character names in identifiers is experimental.
4069
4070GCC allows the @samp{$} character in identifiers as an extension for
4071most targets.  This is true regardless of the @option{std=} switch,
4072since this extension cannot conflict with standards-conforming
4073programs.  When preprocessing assembler, however, dollars are not
4074identifier characters by default.
4075
4076Currently the targets that by default do not permit @samp{$} are AVR,
4077IP2K, MMIX, MIPS Irix 3, ARM aout, and PowerPC targets for the AIX
4078operating system.
4079
4080You can override the default with @option{-fdollars-in-identifiers} or
4081@option{fno-dollars-in-identifiers}.  @xref{fdollars-in-identifiers}.
4082
4083@item Non-empty sequences of whitespace characters.
4084
4085In textual output, each whitespace sequence is collapsed to a single
4086space.  For aesthetic reasons, the first token on each non-directive
4087line of output is preceded with sufficient spaces that it appears in the
4088same column as it did in the original source file.
4089
4090@item The numeric value of character constants in preprocessor expressions.
4091
4092The preprocessor and compiler interpret character constants in the
4093same way; i.e.@: escape sequences such as @samp{\a} are given the
4094values they would have on the target machine.
4095
4096The compiler evaluates a multi-character character constant a character
4097at a time, shifting the previous value left by the number of bits per
4098target character, and then or-ing in the bit-pattern of the new
4099character truncated to the width of a target character.  The final
4100bit-pattern is given type @code{int}, and is therefore signed,
4101regardless of whether single characters are signed or not (a slight
4102change from versions 3.1 and earlier of GCC)@.  If there are more
4103characters in the constant than would fit in the target @code{int} the
4104compiler issues a warning, and the excess leading characters are
4105ignored.
4106
4107For example, @code{'ab'} for a target with an 8-bit @code{char} would be
4108interpreted as @w{@samp{(int) ((unsigned char) 'a' * 256 + (unsigned char)
4109'b')}}, and @code{'\234a'} as @w{@samp{(int) ((unsigned char) '\234' *
4110256 + (unsigned char) 'a')}}.
4111
4112@item Source file inclusion.
4113
4114For a discussion on how the preprocessor locates header files,
4115@ref{Include Operation}.
4116
4117@item Interpretation of the filename resulting from a macro-expanded
4118@samp{#include} directive.
4119
4120@xref{Computed Includes}.
4121
4122@item Treatment of a @samp{#pragma} directive that after macro-expansion
4123results in a standard pragma.
4124
4125No macro expansion occurs on any @samp{#pragma} directive line, so the
4126question does not arise.
4127
4128Note that GCC does not yet implement any of the standard
4129pragmas.
4130
4131@end itemize
4132
4133@node Implementation limits
4134@section Implementation limits
4135@cindex implementation limits
4136
4137CPP has a small number of internal limits.  This section lists the
4138limits which the C standard requires to be no lower than some minimum,
4139and all the others known.  It is intended that there should be as few limits
4140as possible.  If you encounter an undocumented or inconvenient limit,
4141please report that as a bug.  @xref{Bugs, , Reporting Bugs, gcc, Using
4142the GNU Compiler Collection (GCC)}.
4143
4144Where we say something is limited @dfn{only by available memory}, that
4145means that internal data structures impose no intrinsic limit, and space
4146is allocated with @code{malloc} or equivalent.  The actual limit will
4147therefore depend on many things, such as the size of other things
4148allocated by the compiler at the same time, the amount of memory
4149consumed by other processes on the same computer, etc.
4150
4151@itemize @bullet
4152
4153@item Nesting levels of @samp{#include} files.
4154
4155We impose an arbitrary limit of 200 levels, to avoid runaway recursion.
4156The standard requires at least 15 levels.
4157
4158@item Nesting levels of conditional inclusion.
4159
4160The C standard mandates this be at least 63.  CPP is limited only by
4161available memory.
4162
4163@item Levels of parenthesized expressions within a full expression.
4164
4165The C standard requires this to be at least 63.  In preprocessor
4166conditional expressions, it is limited only by available memory.
4167
4168@item Significant initial characters in an identifier or macro name.
4169
4170The preprocessor treats all characters as significant.  The C standard
4171requires only that the first 63 be significant.
4172
4173@item Number of macros simultaneously defined in a single translation unit.
4174
4175The standard requires at least 4095 be possible.  CPP is limited only
4176by available memory.
4177
4178@item Number of parameters in a macro definition and arguments in a macro call.
4179
4180We allow @code{USHRT_MAX}, which is no smaller than 65,535.  The minimum
4181required by the standard is 127.
4182
4183@item Number of characters on a logical source line.
4184
4185The C standard requires a minimum of 4096 be permitted.  CPP places
4186no limits on this, but you may get incorrect column numbers reported in
4187diagnostics for lines longer than 65,535 characters.
4188
4189@item Maximum size of a source file.
4190
4191The standard does not specify any lower limit on the maximum size of a
4192source file.  GNU cpp maps files into memory, so it is limited by the
4193available address space.  This is generally at least two gigabytes.
4194Depending on the operating system, the size of physical memory may or
4195may not be a limitation.
4196
4197@end itemize
4198
4199@node Obsolete Features
4200@section Obsolete Features
4201
4202CPP has some features which are present mainly for compatibility with
4203older programs.  We discourage their use in new code.  In some cases,
4204we plan to remove the feature in a future version of GCC@.
4205
4206@subsection Assertions
4207@cindex assertions
4208
4209@dfn{Assertions} are a deprecated alternative to macros in writing
4210conditionals to test what sort of computer or system the compiled
4211program will run on.  Assertions are usually predefined, but you can
4212define them with preprocessing directives or command-line options.
4213
4214Assertions were intended to provide a more systematic way to describe
4215the compiler's target system and we added them for compatibility with
4216existing compilers.  In practice they are just as unpredictable as the
4217system-specific predefined macros.  In addition, they are not part of
4218any standard, and only a few compilers support them.
4219Therefore, the use of assertions is @strong{less} portable than the use
4220of system-specific predefined macros.  We recommend you do not use them at
4221all.
4222
4223@cindex predicates
4224An assertion looks like this:
4225
4226@smallexample
4227#@var{predicate} (@var{answer})
4228@end smallexample
4229
4230@noindent
4231@var{predicate} must be a single identifier.  @var{answer} can be any
4232sequence of tokens; all characters are significant except for leading
4233and trailing whitespace, and differences in internal whitespace
4234sequences are ignored.  (This is similar to the rules governing macro
4235redefinition.)  Thus, @code{(x + y)} is different from @code{(x+y)} but
4236equivalent to @code{@w{( x + y )}}.  Parentheses do not nest inside an
4237answer.
4238
4239@cindex testing predicates
4240To test an assertion, you write it in an @samp{#if}.  For example, this
4241conditional succeeds if either @code{vax} or @code{ns16000} has been
4242asserted as an answer for @code{machine}.
4243
4244@smallexample
4245#if #machine (vax) || #machine (ns16000)
4246@end smallexample
4247
4248@noindent
4249You can test whether @emph{any} answer is asserted for a predicate by
4250omitting the answer in the conditional:
4251
4252@smallexample
4253#if #machine
4254@end smallexample
4255
4256@findex #assert
4257Assertions are made with the @samp{#assert} directive.  Its sole
4258argument is the assertion to make, without the leading @samp{#} that
4259identifies assertions in conditionals.
4260
4261@smallexample
4262#assert @var{predicate} (@var{answer})
4263@end smallexample
4264
4265@noindent
4266You may make several assertions with the same predicate and different
4267answers.  Subsequent assertions do not override previous ones for the
4268same predicate.  All the answers for any given predicate are
4269simultaneously true.
4270
4271@cindex assertions, canceling
4272@findex #unassert
4273Assertions can be canceled with the @samp{#unassert} directive.  It
4274has the same syntax as @samp{#assert}.  In that form it cancels only the
4275answer which was specified on the @samp{#unassert} line; other answers
4276for that predicate remain true.  You can cancel an entire predicate by
4277leaving out the answer:
4278
4279@smallexample
4280#unassert @var{predicate}
4281@end smallexample
4282
4283@noindent
4284In either form, if no such assertion has been made, @samp{#unassert} has
4285no effect.
4286
4287You can also make or cancel assertions using command line options.
4288@xref{Invocation}.
4289
4290@node Differences from previous versions
4291@section Differences from previous versions
4292@cindex differences from previous versions
4293
4294This section details behavior which has changed from previous versions
4295of CPP@.  We do not plan to change it again in the near future, but
4296we do not promise not to, either.
4297
4298The ``previous versions'' discussed here are 2.95 and before.  The
4299behavior of GCC 3.0 is mostly the same as the behavior of the widely
4300used 2.96 and 2.97 development snapshots.  Where there are differences,
4301they generally represent bugs in the snapshots.
4302
4303@itemize @bullet
4304
4305@item -I- deprecated
4306
4307This option has been deprecated in 4.0.  @option{-iquote} is meant to
4308replace the need for this option.
4309
4310@item Order of evaluation of @samp{#} and @samp{##} operators
4311
4312The standard does not specify the order of evaluation of a chain of
4313@samp{##} operators, nor whether @samp{#} is evaluated before, after, or
4314at the same time as @samp{##}.  You should therefore not write any code
4315which depends on any specific ordering.  It is possible to guarantee an
4316ordering, if you need one, by suitable use of nested macros.
4317
4318An example of where this might matter is pasting the arguments @samp{1},
4319@samp{e} and @samp{-2}.  This would be fine for left-to-right pasting,
4320but right-to-left pasting would produce an invalid token @samp{e-2}.
4321
4322GCC 3.0 evaluates @samp{#} and @samp{##} at the same time and strictly
4323left to right.  Older versions evaluated all @samp{#} operators first,
4324then all @samp{##} operators, in an unreliable order.
4325
4326@item The form of whitespace between tokens in preprocessor output
4327
4328@xref{Preprocessor Output}, for the current textual format.  This is
4329also the format used by stringification.  Normally, the preprocessor
4330communicates tokens directly to the compiler's parser, and whitespace
4331does not come up at all.
4332
4333Older versions of GCC preserved all whitespace provided by the user and
4334inserted lots more whitespace of their own, because they could not
4335accurately predict when extra spaces were needed to prevent accidental
4336token pasting.
4337
4338@item Optional argument when invoking rest argument macros
4339
4340As an extension, GCC permits you to omit the variable arguments entirely
4341when you use a variable argument macro.  This is forbidden by the 1999 C
4342standard, and will provoke a pedantic warning with GCC 3.0.  Previous
4343versions accepted it silently.
4344
4345@item @samp{##} swallowing preceding text in rest argument macros
4346
4347Formerly, in a macro expansion, if @samp{##} appeared before a variable
4348arguments parameter, and the set of tokens specified for that argument
4349in the macro invocation was empty, previous versions of CPP would
4350back up and remove the preceding sequence of non-whitespace characters
4351(@strong{not} the preceding token).  This extension is in direct
4352conflict with the 1999 C standard and has been drastically pared back.
4353
4354In the current version of the preprocessor, if @samp{##} appears between
4355a comma and a variable arguments parameter, and the variable argument is
4356omitted entirely, the comma will be removed from the expansion.  If the
4357variable argument is empty, or the token before @samp{##} is not a
4358comma, then @samp{##} behaves as a normal token paste.
4359
4360@item @samp{#line} and @samp{#include}
4361
4362The @samp{#line} directive used to change GCC's notion of the
4363``directory containing the current file'', used by @samp{#include} with
4364a double-quoted header file name.  In 3.0 and later, it does not.
4365@xref{Line Control}, for further explanation.
4366
4367@item Syntax of @samp{#line}
4368
4369In GCC 2.95 and previous, the string constant argument to @samp{#line}
4370was treated the same way as the argument to @samp{#include}: backslash
4371escapes were not honored, and the string ended at the second @samp{"}.
4372This is not compliant with the C standard.  In GCC 3.0, an attempt was
4373made to correct the behavior, so that the string was treated as a real
4374string constant, but it turned out to be buggy.  In 3.1, the bugs have
4375been fixed.  (We are not fixing the bugs in 3.0 because they affect
4376relatively few people and the fix is quite invasive.)
4377
4378@end itemize
4379
4380@node Invocation
4381@chapter Invocation
4382@cindex invocation
4383@cindex command line
4384
4385Most often when you use the C preprocessor you will not have to invoke it
4386explicitly: the C compiler will do so automatically.  However, the
4387preprocessor is sometimes useful on its own.  All the options listed
4388here are also acceptable to the C compiler and have the same meaning,
4389except that the C compiler has different rules for specifying the output
4390file.
4391
4392@emph{Note:} Whether you use the preprocessor by way of @command{gcc}
4393or @command{cpp}, the @dfn{compiler driver} is run first.  This
4394program's purpose is to translate your command into invocations of the
4395programs that do the actual work.  Their command line interfaces are
4396similar but not identical to the documented interface, and may change
4397without notice.
4398
4399@ignore
4400@c man begin SYNOPSIS
4401cpp [@option{-D}@var{macro}[=@var{defn}]@dots{}] [@option{-U}@var{macro}]
4402    [@option{-I}@var{dir}@dots{}] [@option{-iquote}@var{dir}@dots{}]
4403    [@option{-iremap}@var{src}:@var{dst}]
4404    [@option{-W}@var{warn}@dots{}]
4405    [@option{-M}|@option{-MM}] [@option{-MG}] [@option{-MF} @var{filename}]
4406    [@option{-MP}] [@option{-MQ} @var{target}@dots{}]
4407    [@option{-MT} @var{target}@dots{}]
4408    [@option{-P}] [@option{-fno-working-directory}]
4409    [@option{-x} @var{language}] [@option{-std=}@var{standard}]
4410    @var{infile} @var{outfile}
4411
4412Only the most useful options are listed here; see below for the remainder.
4413@c man end
4414@c man begin SEEALSO
4415gpl(7), gfdl(7), fsf-funding(7),
4416gcc(1), as(1), ld(1), and the Info entries for @file{cpp}, @file{gcc}, and
4417@file{binutils}.
4418@c man end
4419@end ignore
4420
4421@c man begin OPTIONS
4422The C preprocessor expects two file names as arguments, @var{infile} and
4423@var{outfile}.  The preprocessor reads @var{infile} together with any
4424other files it specifies with @samp{#include}.  All the output generated
4425by the combined input files is written in @var{outfile}.
4426
4427Either @var{infile} or @var{outfile} may be @option{-}, which as
4428@var{infile} means to read from standard input and as @var{outfile}
4429means to write to standard output.  Also, if either file is omitted, it
4430means the same as if @option{-} had been specified for that file.
4431
4432Unless otherwise noted, or the option ends in @samp{=}, all options
4433which take an argument may have that argument appear either immediately
4434after the option, or with a space between option and argument:
4435@option{-Ifoo} and @option{-I foo} have the same effect.
4436
4437@cindex grouping options
4438@cindex options, grouping
4439Many options have multi-letter names; therefore multiple single-letter
4440options may @emph{not} be grouped: @option{-dM} is very different from
4441@w{@samp{-d -M}}.
4442
4443@cindex options
4444@include cppopts.texi
4445@c man end
4446
4447@node Environment Variables
4448@chapter Environment Variables
4449@cindex environment variables
4450@c man begin ENVIRONMENT
4451
4452This section describes the environment variables that affect how CPP
4453operates.  You can use them to specify directories or prefixes to use
4454when searching for include files, or to control dependency output.
4455
4456Note that you can also specify places to search using options such as
4457@option{-I}, and control dependency output with options like
4458@option{-M} (@pxref{Invocation}).  These take precedence over
4459environment variables, which in turn take precedence over the
4460configuration of GCC@.
4461
4462@include cppenv.texi
4463@c man end
4464
4465@page
4466@include fdl.texi
4467
4468@page
4469@node Index of Directives
4470@unnumbered Index of Directives
4471@printindex fn
4472
4473@node Option Index
4474@unnumbered Option Index
4475@noindent
4476CPP's command line options and environment variables are indexed here
4477without any initial @samp{-} or @samp{--}.
4478@printindex op
4479
4480@page
4481@node Concept Index
4482@unnumbered Concept Index
4483@printindex cp
4484
4485@bye
4486