1 @c Copyright (C) 1988,1989,1992,1993,1994,1996,1998,1999,2000,2001 Free Software Foundation, Inc.
2 @c This is part of the GCC manual.
3 @c For copying conditions, see the file gcc.texi.
6 @chapter C Implementation-defined behavior
7 @cindex implementation-defined behavior, C language
9 A conforming implementation of ISO C is required to document its
10 choice of behavior in each of the areas that are designated
11 ``implementation defined.'' The following lists all such areas,
12 along with the section number from the ISO/IEC 9899:1999 standard.
15 * Translation implementation::
16 * Environment implementation::
17 * Identifiers implementation::
18 * Characters implementation::
19 * Integers implementation::
20 * Floating point implementation::
21 * Arrays and pointers implementation::
22 * Hints implementation::
23 * Structures unions enumerations and bit-fields implementation::
24 * Qualifiers implementation::
25 * Preprocessing directives implementation::
26 * Library functions implementation::
27 * Architecture implementation::
28 * Locale-specific behavior implementation::
31 @node Translation implementation
36 @cite{How a diagnostic is identified (3.10, 5.1.1.3).}
39 @cite{Whether each nonempty sequence of white-space characters other than
40 new-line is retained or replaced by one space character in translation
44 @node Environment implementation
47 The behavior of these points are dependent on the implementation
48 of the C library, and are not defined by GCC itself.
50 @node Identifiers implementation
55 @cite{Which additional multibyte characters may appear in identifiers
56 and their correspondence to universal character names (6.4.2).}
59 @cite{The number of significant initial characters in an identifier
63 @node Characters implementation
68 @cite{The number of bits in a byte (3.6).}
71 @cite{The values of the members of the execution character set (5.2.1).}
74 @cite{The unique value of the member of the execution character set produced
75 for each of the standard alphabetic escape sequences (5.2.2).}
78 @cite{The value of a @code{char} object into which has been stored any
79 character other than a member of the basic execution character set (6.2.5).}
82 @cite{Which of @code{signed char} or @code{unsigned char} has the same range,
83 representation, and behavior as ``plain'' @code{char} (6.2.5, 6.3.1.1).}
86 @cite{The mapping of members of the source character set (in character
87 constants and string literals) to members of the execution character
88 set (6.4.4.4, 5.1.1.2).}
91 @cite{The value of an integer character constant containing more than one
92 character or containing a character or escape sequence that does not map
93 to a single-byte execution character (6.4.4.4).}
96 @cite{The value of a wide character constant containing more than one
97 multibyte character, or containing a multibyte character or escape
98 sequence not represented in the extended execution character set (6.4.4.4).}
101 @cite{The current locale used to convert a wide character constant consisting
102 of a single multibyte character that maps to a member of the extended
103 execution character set into a corresponding wide character code (6.4.4.4).}
106 @cite{The current locale used to convert a wide string literal into
107 corresponding wide character codes (6.4.5).}
110 @cite{The value of a string literal containing a multibyte character or escape
111 sequence not represented in the execution character set (6.4.5).}
114 @node Integers implementation
119 @cite{Any extended integer types that exist in the implementation (6.2.5).}
122 @cite{Whether signed integer types are represented using sign and magnitude,
123 two's complement, or one's complement, and whether the extraordinary value
124 is a trap representation or an ordinary value (6.2.6.2).}
127 @cite{The rank of any extended integer type relative to another extended
128 integer type with the same precision (6.3.1.1).}
131 @cite{The result of, or the signal raised by, converting an integer to a
132 signed integer type when the value cannot be represented in an object of
133 that type (6.3.1.3).}
136 @cite{The results of some bitwise operations on signed integers (6.5).}
139 @node Floating point implementation
140 @section Floating point
144 @cite{The accuracy of the floating-point operations and of the library
145 functions in @code{<math.h>} and @code{<complex.h>} that return floating-point
146 results (5.2.4.2.2).}
149 @cite{The rounding behaviors characterized by non-standard values
150 of @code{FLT_ROUNDS} @gol
154 @cite{The evaluation methods characterized by non-standard negative
155 values of @code{FLT_EVAL_METHOD} (5.2.4.2.2).}
158 @cite{The direction of rounding when an integer is converted to a
159 floating-point number that cannot exactly represent the original
163 @cite{The direction of rounding when a floating-point number is
164 converted to a narrower floating-point number (6.3.1.5).}
167 @cite{How the nearest representable value or the larger or smaller
168 representable value immediately adjacent to the nearest representable
169 value is chosen for certain floating constants (6.4.4.2).}
172 @cite{Whether and how floating expressions are contracted when not
173 disallowed by the @code{FP_CONTRACT} pragma (6.5).}
176 @cite{The default state for the @code{FENV_ACCESS} pragma (7.6.1).}
179 @cite{Additional floating-point exceptions, rounding modes, environments,
180 and classifications, and their macro names (7.6, 7.12).}
183 @cite{The default state for the @code{FP_CONTRACT} pragma (7.12.2).}
186 @cite{Whether the ``inexact'' floating-point exception can be raised
187 when the rounded result actually does equal the mathematical result
188 in an IEC 60559 conformant implementation (F.9).}
191 @cite{Whether the ``underflow'' (and ``inexact'') floating-point
192 exception can be raised when a result is tiny but not inexact in an
193 IEC 60559 conformant implementation (F.9).}
197 @node Arrays and pointers implementation
198 @section Arrays and pointers
202 @cite{The result of converting a pointer to an integer or
203 vice versa (6.3.2.3).}
205 A cast from pointer to integer discards most-significant bits if the
206 pointer representation is larger than the integer type,
207 sign-extends@footnote{Future versions of GCC may zero-extend, or use
208 a target-defined @code{ptr_extend} pattern. Do not rely on sign extension.}
209 if the pointer representation is smaller than the integer type, otherwise
210 the bits are unchanged.
211 @c ??? We've always claimed that pointers were unsigned entities.
212 @c Shouldn't we therefore be doing zero-extension? If so, the bug
213 @c is in convert_to_integer, where we call type_for_size and request
214 @c a signed integral type. On the other hand, it might be most useful
215 @c for the target if we extend according to POINTERS_EXTEND_UNSIGNED.
217 A cast from integer to pointer discards most-significant bits if the
218 pointer representation is smaller than the integer type, extends according
219 to the signedness of the integer type if the pointer representation
220 is larger than the integer type, otherwise the bits are unchanged.
222 When casting from pointer to integer and back again, the resulting
223 pointer must reference the same object as the original pointer, otherwise
224 the behavior is undefined. That is, one may not use integer arithmetic to
225 avoid the undefined behavior of pointer arithmetic as proscribed in 6.5.6/8.
228 @cite{The size of the result of subtracting two pointers to elements
229 of the same array (6.5.6).}
233 @node Hints implementation
238 @cite{The extent to which suggestions made by using the @code{register}
239 storage-class specifier are effective (6.7.1).}
242 @cite{The extent to which suggestions made by using the inline function
243 specifier are effective (6.7.4).}
247 @node Structures unions enumerations and bit-fields implementation
248 @section Structures, unions, enumerations, and bit-fields
252 @cite{Whether a ``plain'' int bit-field is treated as a @code{signed int}
253 bit-field or as an @code{unsigned int} bit-field (6.7.2, 6.7.2.1).}
256 @cite{Allowable bit-field types other than @code{_Bool}, @code{signed int},
257 and @code{unsigned int} (6.7.2.1).}
260 @cite{Whether a bit-field can straddle a storage-unit boundary (6.7.2.1).}
263 @cite{The order of allocation of bit-fields within a unit (6.7.2.1).}
266 @cite{The alignment of non-bit-field members of structures (6.7.2.1).}
269 @cite{The integer type compatible with each enumerated type (6.7.2.2).}
273 @node Qualifiers implementation
278 @cite{What constitutes an access to an object that has volatile-qualified
283 @node Preprocessing directives implementation
284 @section Preprocessing directives
288 @cite{How sequences in both forms of header names are mapped to headers
289 or external source file names (6.4.7).}
292 @cite{Whether the value of a character constant in a constant expression
293 that controls conditional inclusion matches the value of the same character
294 constant in the execution character set (6.10.1).}
297 @cite{Whether the value of a single-character character constant in a
298 constant expression that controls conditional inclusion may have a
299 negative value (6.10.1).}
302 @cite{The places that are searched for an included @samp{<>} delimited
303 header, and how the places are specified or the header is
304 identified (6.10.2).}
307 @cite{How the named source file is searched for in an included @samp{""}
308 delimited header (6.10.2).}
311 @cite{The method by which preprocessing tokens (possibly resulting from
312 macro expansion) in a @code{#include} directive are combined into a header
316 @cite{The nesting limit for @code{#include} processing (6.10.2).}
319 @cite{Whether the @samp{#} operator inserts a @samp{\} character before
320 the @samp{\} character that begins a universal character name in a
321 character constant or string literal (6.10.3.2).}
324 @cite{The behavior on each recognized non-@code{STDC #pragma}
328 @cite{The definitions for @code{__DATE__} and @code{__TIME__} when
329 respectively, the date and time of translation are not available (6.10.8).}
333 @node Library functions implementation
334 @section Library functions
336 The behavior of these points are dependent on the implementation
337 of the C library, and are not defined by GCC itself.
339 @node Architecture implementation
340 @section Architecture
344 @cite{The values or expressions assigned to the macros specified in the
345 headers @code{<float.h>}, @code{<limits.h>}, and @code{<stdint.h>}
346 (5.2.4.2, 7.18.2, 7.18.3).}
349 @cite{The number, order, and encoding of bytes in any object
350 (when not explicitly specified in this International Standard) (6.2.6.1).}
353 @cite{The value of the result of the sizeof operator (6.5.3.4).}
357 @node Locale-specific behavior implementation
358 @section Locale-specific behavior
360 The behavior of these points are dependent on the implementation
361 of the C library, and are not defined by GCC itself.
364 @chapter Extensions to the C Language Family
365 @cindex extensions, C language
366 @cindex C language extensions
369 GNU C provides several language features not found in ISO standard C@.
370 (The @option{-pedantic} option directs GCC to print a warning message if
371 any of these features is used.) To test for the availability of these
372 features in conditional compilation, check for a predefined macro
373 @code{__GNUC__}, which is always defined under GCC@.
375 These extensions are available in C and Objective-C@. Most of them are
376 also available in C++. @xref{C++ Extensions,,Extensions to the
377 C++ Language}, for extensions that apply @emph{only} to C++.
379 Some features that are in ISO C99 but not C89 or C++ are also, as
380 extensions, accepted by GCC in C89 mode and in C++.
382 @c The only difference between the two versions of this menu is that the
383 @c version for clear INTERNALS has an extra node, "Constraints" (which
384 @c appears in a separate chapter in the other version of the manual).
387 * Statement Exprs:: Putting statements and declarations inside expressions.
388 * Local Labels:: Labels local to a statement-expression.
389 * Labels as Values:: Getting pointers to labels, and computed gotos.
390 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
391 * Constructing Calls:: Dispatching a call to another function.
392 * Naming Types:: Giving a name to the type of some expression.
393 * Typeof:: @code{typeof}: referring to the type of an expression.
394 * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
395 * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
396 * Long Long:: Double-word integers---@code{long long int}.
397 * Complex:: Data types for complex numbers.
398 * Hex Floats:: Hexadecimal floating-point constants.
399 * Zero Length:: Zero-length arrays.
400 * Variable Length:: Arrays whose length is computed at run time.
401 * Variadic Macros:: Macros with a variable number of arguments.
402 * Escaped Newlines:: Slightly looser rules for escaped newlines.
403 * Multi-line Strings:: String literals with embedded newlines.
404 * Subscripting:: Any array can be subscripted, even if not an lvalue.
405 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
406 * Initializers:: Non-constant initializers.
407 * Compound Literals:: Compound literals give structures, unions
409 * Designated Inits:: Labeling elements of initializers.
410 * Cast to Union:: Casting to union type from any member of the union.
411 * Case Ranges:: `case 1 ... 9' and such.
412 * Mixed Declarations:: Mixing declarations and code.
413 * Function Attributes:: Declaring that functions have no side effects,
414 or that they can never return.
415 * Attribute Syntax:: Formal syntax for attributes.
416 * Function Prototypes:: Prototype declarations and old-style definitions.
417 * C++ Comments:: C++ comments are recognized.
418 * Dollar Signs:: Dollar sign is allowed in identifiers.
419 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
420 * Variable Attributes:: Specifying attributes of variables.
421 * Type Attributes:: Specifying attributes of types.
422 * Alignment:: Inquiring about the alignment of a type or variable.
423 * Inline:: Defining inline functions (as fast as macros).
424 * Extended Asm:: Assembler instructions with C expressions as operands.
425 (With them you can define ``built-in'' functions.)
426 * Asm Labels:: Specifying the assembler name to use for a C symbol.
427 * Explicit Reg Vars:: Defining variables residing in specified registers.
428 * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
429 * Incomplete Enums:: @code{enum foo;}, with details to follow.
430 * Function Names:: Printable strings which are the name of the current
432 * Return Address:: Getting the return or frame address of a function.
433 * Vector Extensions:: Using vector instructions through built-in functions.
434 * Other Builtins:: Other built-in functions.
435 * Pragmas:: Pragmas accepted by GCC.
436 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
441 * Statement Exprs:: Putting statements and declarations inside expressions.
442 * Local Labels:: Labels local to a statement-expression.
443 * Labels as Values:: Getting pointers to labels, and computed gotos.
444 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
445 * Constructing Calls:: Dispatching a call to another function.
446 * Naming Types:: Giving a name to the type of some expression.
447 * Typeof:: @code{typeof}: referring to the type of an expression.
448 * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
449 * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
450 * Long Long:: Double-word integers---@code{long long int}.
451 * Complex:: Data types for complex numbers.
452 * Hex Floats:: Hexadecimal floating-point constants.
453 * Zero Length:: Zero-length arrays.
454 * Variable Length:: Arrays whose length is computed at run time.
455 * Variadic Macros:: Macros with a variable number of arguments.
456 * Escaped Newlines:: Slightly looser rules for escaped newlines.
457 * Multi-line Strings:: String literals with embedded newlines.
458 * Subscripting:: Any array can be subscripted, even if not an lvalue.
459 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
460 * Initializers:: Non-constant initializers.
461 * Compound Literals:: Compound literals give structures, unions
463 * Designated Inits:: Labeling elements of initializers.
464 * Cast to Union:: Casting to union type from any member of the union.
465 * Case Ranges:: `case 1 ... 9' and such.
466 * Mixed Declarations:: Mixing declarations and code.
467 * Function Attributes:: Declaring that functions have no side effects,
468 or that they can never return.
469 * Attribute Syntax:: Formal syntax for attributes.
470 * Function Prototypes:: Prototype declarations and old-style definitions.
471 * C++ Comments:: C++ comments are recognized.
472 * Dollar Signs:: Dollar sign is allowed in identifiers.
473 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
474 * Variable Attributes:: Specifying attributes of variables.
475 * Type Attributes:: Specifying attributes of types.
476 * Alignment:: Inquiring about the alignment of a type or variable.
477 * Inline:: Defining inline functions (as fast as macros).
478 * Extended Asm:: Assembler instructions with C expressions as operands.
479 (With them you can define ``built-in'' functions.)
480 * Constraints:: Constraints for asm operands
481 * Asm Labels:: Specifying the assembler name to use for a C symbol.
482 * Explicit Reg Vars:: Defining variables residing in specified registers.
483 * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
484 * Incomplete Enums:: @code{enum foo;}, with details to follow.
485 * Function Names:: Printable strings which are the name of the current
487 * Return Address:: Getting the return or frame address of a function.
488 * Vector Extensions:: Using vector instructions through built-in functions.
489 * Other Builtins:: Other built-in functions.
490 * Pragmas:: Pragmas accepted by GCC.
494 @node Statement Exprs
495 @section Statements and Declarations in Expressions
496 @cindex statements inside expressions
497 @cindex declarations inside expressions
498 @cindex expressions containing statements
499 @cindex macros, statements in expressions
501 @c the above section title wrapped and causes an underfull hbox.. i
502 @c changed it from "within" to "in". --mew 4feb93
504 A compound statement enclosed in parentheses may appear as an expression
505 in GNU C@. This allows you to use loops, switches, and local variables
506 within an expression.
508 Recall that a compound statement is a sequence of statements surrounded
509 by braces; in this construct, parentheses go around the braces. For
513 (@{ int y = foo (); int z;
520 is a valid (though slightly more complex than necessary) expression
521 for the absolute value of @code{foo ()}.
523 The last thing in the compound statement should be an expression
524 followed by a semicolon; the value of this subexpression serves as the
525 value of the entire construct. (If you use some other kind of statement
526 last within the braces, the construct has type @code{void}, and thus
527 effectively no value.)
529 This feature is especially useful in making macro definitions ``safe'' (so
530 that they evaluate each operand exactly once). For example, the
531 ``maximum'' function is commonly defined as a macro in standard C as
535 #define max(a,b) ((a) > (b) ? (a) : (b))
539 @cindex side effects, macro argument
540 But this definition computes either @var{a} or @var{b} twice, with bad
541 results if the operand has side effects. In GNU C, if you know the
542 type of the operands (here let's assume @code{int}), you can define
543 the macro safely as follows:
546 #define maxint(a,b) \
547 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
550 Embedded statements are not allowed in constant expressions, such as
551 the value of an enumeration constant, the width of a bit-field, or
552 the initial value of a static variable.
554 If you don't know the type of the operand, you can still do this, but you
555 must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
558 Statement expressions are not supported fully in G++, and their fate
559 there is unclear. (It is possible that they will become fully supported
560 at some point, or that they will be deprecated, or that the bugs that
561 are present will continue to exist indefinitely.) Presently, statement
562 expressions do not work well as default arguments.
564 In addition, there are semantic issues with statement-expressions in
565 C++. If you try to use statement-expressions instead of inline
566 functions in C++, you may be surprised at the way object destruction is
567 handled. For example:
570 #define foo(a) (@{int b = (a); b + 3; @})
574 does not work the same way as:
577 inline int foo(int a) @{ int b = a; return b + 3; @}
581 In particular, if the expression passed into @code{foo} involves the
582 creation of temporaries, the destructors for those temporaries will be
583 run earlier in the case of the macro than in the case of the function.
585 These considerations mean that it is probably a bad idea to use
586 statement-expressions of this form in header files that are designed to
587 work with C++. (Note that some versions of the GNU C Library contained
588 header files using statement-expression that lead to precisely this
592 @section Locally Declared Labels
594 @cindex macros, local labels
596 Each statement expression is a scope in which @dfn{local labels} can be
597 declared. A local label is simply an identifier; you can jump to it
598 with an ordinary @code{goto} statement, but only from within the
599 statement expression it belongs to.
601 A local label declaration looks like this:
604 __label__ @var{label};
611 __label__ @var{label1}, @var{label2}, @dots{};
614 Local label declarations must come at the beginning of the statement
615 expression, right after the @samp{(@{}, before any ordinary
618 The label declaration defines the label @emph{name}, but does not define
619 the label itself. You must do this in the usual way, with
620 @code{@var{label}:}, within the statements of the statement expression.
622 The local label feature is useful because statement expressions are
623 often used in macros. If the macro contains nested loops, a @code{goto}
624 can be useful for breaking out of them. However, an ordinary label
625 whose scope is the whole function cannot be used: if the macro can be
626 expanded several times in one function, the label will be multiply
627 defined in that function. A local label avoids this problem. For
631 #define SEARCH(array, target) \
634 typeof (target) _SEARCH_target = (target); \
635 typeof (*(array)) *_SEARCH_array = (array); \
638 for (i = 0; i < max; i++) \
639 for (j = 0; j < max; j++) \
640 if (_SEARCH_array[i][j] == _SEARCH_target) \
641 @{ value = i; goto found; @} \
648 @node Labels as Values
649 @section Labels as Values
650 @cindex labels as values
651 @cindex computed gotos
652 @cindex goto with computed label
653 @cindex address of a label
655 You can get the address of a label defined in the current function
656 (or a containing function) with the unary operator @samp{&&}. The
657 value has type @code{void *}. This value is a constant and can be used
658 wherever a constant of that type is valid. For example:
666 To use these values, you need to be able to jump to one. This is done
667 with the computed goto statement@footnote{The analogous feature in
668 Fortran is called an assigned goto, but that name seems inappropriate in
669 C, where one can do more than simply store label addresses in label
670 variables.}, @code{goto *@var{exp};}. For example,
677 Any expression of type @code{void *} is allowed.
679 One way of using these constants is in initializing a static array that
680 will serve as a jump table:
683 static void *array[] = @{ &&foo, &&bar, &&hack @};
686 Then you can select a label with indexing, like this:
693 Note that this does not check whether the subscript is in bounds---array
694 indexing in C never does that.
696 Such an array of label values serves a purpose much like that of the
697 @code{switch} statement. The @code{switch} statement is cleaner, so
698 use that rather than an array unless the problem does not fit a
699 @code{switch} statement very well.
701 Another use of label values is in an interpreter for threaded code.
702 The labels within the interpreter function can be stored in the
703 threaded code for super-fast dispatching.
705 You may not use this mechanism to jump to code in a different function.
706 If you do that, totally unpredictable things will happen. The best way to
707 avoid this is to store the label address only in automatic variables and
708 never pass it as an argument.
710 An alternate way to write the above example is
713 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
715 goto *(&&foo + array[i]);
719 This is more friendly to code living in shared libraries, as it reduces
720 the number of dynamic relocations that are needed, and by consequence,
721 allows the data to be read-only.
723 @node Nested Functions
724 @section Nested Functions
725 @cindex nested functions
726 @cindex downward funargs
729 A @dfn{nested function} is a function defined inside another function.
730 (Nested functions are not supported for GNU C++.) The nested function's
731 name is local to the block where it is defined. For example, here we
732 define a nested function named @code{square}, and call it twice:
736 foo (double a, double b)
738 double square (double z) @{ return z * z; @}
740 return square (a) + square (b);
745 The nested function can access all the variables of the containing
746 function that are visible at the point of its definition. This is
747 called @dfn{lexical scoping}. For example, here we show a nested
748 function which uses an inherited variable named @code{offset}:
752 bar (int *array, int offset, int size)
754 int access (int *array, int index)
755 @{ return array[index + offset]; @}
758 for (i = 0; i < size; i++)
759 @dots{} access (array, i) @dots{}
764 Nested function definitions are permitted within functions in the places
765 where variable definitions are allowed; that is, in any block, before
766 the first statement in the block.
768 It is possible to call the nested function from outside the scope of its
769 name by storing its address or passing the address to another function:
772 hack (int *array, int size)
774 void store (int index, int value)
775 @{ array[index] = value; @}
777 intermediate (store, size);
781 Here, the function @code{intermediate} receives the address of
782 @code{store} as an argument. If @code{intermediate} calls @code{store},
783 the arguments given to @code{store} are used to store into @code{array}.
784 But this technique works only so long as the containing function
785 (@code{hack}, in this example) does not exit.
787 If you try to call the nested function through its address after the
788 containing function has exited, all hell will break loose. If you try
789 to call it after a containing scope level has exited, and if it refers
790 to some of the variables that are no longer in scope, you may be lucky,
791 but it's not wise to take the risk. If, however, the nested function
792 does not refer to anything that has gone out of scope, you should be
795 GCC implements taking the address of a nested function using a technique
796 called @dfn{trampolines}. A paper describing them is available as
799 @uref{http://people.debian.org/~karlheg/Usenix88-lexic.pdf}.
801 A nested function can jump to a label inherited from a containing
802 function, provided the label was explicitly declared in the containing
803 function (@pxref{Local Labels}). Such a jump returns instantly to the
804 containing function, exiting the nested function which did the
805 @code{goto} and any intermediate functions as well. Here is an example:
809 bar (int *array, int offset, int size)
812 int access (int *array, int index)
816 return array[index + offset];
820 for (i = 0; i < size; i++)
821 @dots{} access (array, i) @dots{}
825 /* @r{Control comes here from @code{access}
826 if it detects an error.} */
833 A nested function always has internal linkage. Declaring one with
834 @code{extern} is erroneous. If you need to declare the nested function
835 before its definition, use @code{auto} (which is otherwise meaningless
836 for function declarations).
839 bar (int *array, int offset, int size)
842 auto int access (int *, int);
844 int access (int *array, int index)
848 return array[index + offset];
854 @node Constructing Calls
855 @section Constructing Function Calls
856 @cindex constructing calls
857 @cindex forwarding calls
859 Using the built-in functions described below, you can record
860 the arguments a function received, and call another function
861 with the same arguments, without knowing the number or types
864 You can also record the return value of that function call,
865 and later return that value, without knowing what data type
866 the function tried to return (as long as your caller expects
869 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
870 This built-in function returns a pointer to data
871 describing how to perform a call with the same arguments as were passed
872 to the current function.
874 The function saves the arg pointer register, structure value address,
875 and all registers that might be used to pass arguments to a function
876 into a block of memory allocated on the stack. Then it returns the
877 address of that block.
880 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
881 This built-in function invokes @var{function}
882 with a copy of the parameters described by @var{arguments}
885 The value of @var{arguments} should be the value returned by
886 @code{__builtin_apply_args}. The argument @var{size} specifies the size
887 of the stack argument data, in bytes.
889 This function returns a pointer to data describing
890 how to return whatever value was returned by @var{function}. The data
891 is saved in a block of memory allocated on the stack.
893 It is not always simple to compute the proper value for @var{size}. The
894 value is used by @code{__builtin_apply} to compute the amount of data
895 that should be pushed on the stack and copied from the incoming argument
899 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
900 This built-in function returns the value described by @var{result} from
901 the containing function. You should specify, for @var{result}, a value
902 returned by @code{__builtin_apply}.
906 @section Naming an Expression's Type
909 You can give a name to the type of an expression using a @code{typedef}
910 declaration with an initializer. Here is how to define @var{name} as a
911 type name for the type of @var{exp}:
914 typedef @var{name} = @var{exp};
917 This is useful in conjunction with the statements-within-expressions
918 feature. Here is how the two together can be used to define a safe
919 ``maximum'' macro that operates on any arithmetic type:
923 (@{typedef _ta = (a), _tb = (b); \
924 _ta _a = (a); _tb _b = (b); \
925 _a > _b ? _a : _b; @})
928 @cindex underscores in variables in macros
929 @cindex @samp{_} in variables in macros
930 @cindex local variables in macros
931 @cindex variables, local, in macros
932 @cindex macros, local variables in
934 The reason for using names that start with underscores for the local
935 variables is to avoid conflicts with variable names that occur within the
936 expressions that are substituted for @code{a} and @code{b}. Eventually we
937 hope to design a new form of declaration syntax that allows you to declare
938 variables whose scopes start only after their initializers; this will be a
939 more reliable way to prevent such conflicts.
942 @section Referring to a Type with @code{typeof}
945 @cindex macros, types of arguments
947 Another way to refer to the type of an expression is with @code{typeof}.
948 The syntax of using of this keyword looks like @code{sizeof}, but the
949 construct acts semantically like a type name defined with @code{typedef}.
951 There are two ways of writing the argument to @code{typeof}: with an
952 expression or with a type. Here is an example with an expression:
959 This assumes that @code{x} is an array of pointers to functions;
960 the type described is that of the values of the functions.
962 Here is an example with a typename as the argument:
969 Here the type described is that of pointers to @code{int}.
971 If you are writing a header file that must work when included in ISO C
972 programs, write @code{__typeof__} instead of @code{typeof}.
973 @xref{Alternate Keywords}.
975 A @code{typeof}-construct can be used anywhere a typedef name could be
976 used. For example, you can use it in a declaration, in a cast, or inside
977 of @code{sizeof} or @code{typeof}.
981 This declares @code{y} with the type of what @code{x} points to.
988 This declares @code{y} as an array of such values.
995 This declares @code{y} as an array of pointers to characters:
998 typeof (typeof (char *)[4]) y;
1002 It is equivalent to the following traditional C declaration:
1008 To see the meaning of the declaration using @code{typeof}, and why it
1009 might be a useful way to write, let's rewrite it with these macros:
1012 #define pointer(T) typeof(T *)
1013 #define array(T, N) typeof(T [N])
1017 Now the declaration can be rewritten this way:
1020 array (pointer (char), 4) y;
1024 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
1025 pointers to @code{char}.
1029 @section Generalized Lvalues
1030 @cindex compound expressions as lvalues
1031 @cindex expressions, compound, as lvalues
1032 @cindex conditional expressions as lvalues
1033 @cindex expressions, conditional, as lvalues
1034 @cindex casts as lvalues
1035 @cindex generalized lvalues
1036 @cindex lvalues, generalized
1037 @cindex extensions, @code{?:}
1038 @cindex @code{?:} extensions
1039 Compound expressions, conditional expressions and casts are allowed as
1040 lvalues provided their operands are lvalues. This means that you can take
1041 their addresses or store values into them.
1043 Standard C++ allows compound expressions and conditional expressions as
1044 lvalues, and permits casts to reference type, so use of this extension
1045 is deprecated for C++ code.
1047 For example, a compound expression can be assigned, provided the last
1048 expression in the sequence is an lvalue. These two expressions are
1056 Similarly, the address of the compound expression can be taken. These two
1057 expressions are equivalent:
1064 A conditional expression is a valid lvalue if its type is not void and the
1065 true and false branches are both valid lvalues. For example, these two
1066 expressions are equivalent:
1070 (a ? b = 5 : (c = 5))
1073 A cast is a valid lvalue if its operand is an lvalue. A simple
1074 assignment whose left-hand side is a cast works by converting the
1075 right-hand side first to the specified type, then to the type of the
1076 inner left-hand side expression. After this is stored, the value is
1077 converted back to the specified type to become the value of the
1078 assignment. Thus, if @code{a} has type @code{char *}, the following two
1079 expressions are equivalent:
1083 (int)(a = (char *)(int)5)
1086 An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
1087 performs the arithmetic using the type resulting from the cast, and then
1088 continues as in the previous case. Therefore, these two expressions are
1093 (int)(a = (char *)(int) ((int)a + 5))
1096 You cannot take the address of an lvalue cast, because the use of its
1097 address would not work out coherently. Suppose that @code{&(int)f} were
1098 permitted, where @code{f} has type @code{float}. Then the following
1099 statement would try to store an integer bit-pattern where a floating
1100 point number belongs:
1106 This is quite different from what @code{(int)f = 1} would do---that
1107 would convert 1 to floating point and store it. Rather than cause this
1108 inconsistency, we think it is better to prohibit use of @samp{&} on a cast.
1110 If you really do want an @code{int *} pointer with the address of
1111 @code{f}, you can simply write @code{(int *)&f}.
1114 @section Conditionals with Omitted Operands
1115 @cindex conditional expressions, extensions
1116 @cindex omitted middle-operands
1117 @cindex middle-operands, omitted
1118 @cindex extensions, @code{?:}
1119 @cindex @code{?:} extensions
1121 The middle operand in a conditional expression may be omitted. Then
1122 if the first operand is nonzero, its value is the value of the conditional
1125 Therefore, the expression
1132 has the value of @code{x} if that is nonzero; otherwise, the value of
1135 This example is perfectly equivalent to
1141 @cindex side effect in ?:
1142 @cindex ?: side effect
1144 In this simple case, the ability to omit the middle operand is not
1145 especially useful. When it becomes useful is when the first operand does,
1146 or may (if it is a macro argument), contain a side effect. Then repeating
1147 the operand in the middle would perform the side effect twice. Omitting
1148 the middle operand uses the value already computed without the undesirable
1149 effects of recomputing it.
1152 @section Double-Word Integers
1153 @cindex @code{long long} data types
1154 @cindex double-word arithmetic
1155 @cindex multiprecision arithmetic
1156 @cindex @code{LL} integer suffix
1157 @cindex @code{ULL} integer suffix
1159 ISO C99 supports data types for integers that are at least 64 bits wide,
1160 and as an extension GCC supports them in C89 mode and in C++.
1161 Simply write @code{long long int} for a signed integer, or
1162 @code{unsigned long long int} for an unsigned integer. To make an
1163 integer constant of type @code{long long int}, add the suffix @samp{LL}
1164 to the integer. To make an integer constant of type @code{unsigned long
1165 long int}, add the suffix @samp{ULL} to the integer.
1167 You can use these types in arithmetic like any other integer types.
1168 Addition, subtraction, and bitwise boolean operations on these types
1169 are open-coded on all types of machines. Multiplication is open-coded
1170 if the machine supports fullword-to-doubleword a widening multiply
1171 instruction. Division and shifts are open-coded only on machines that
1172 provide special support. The operations that are not open-coded use
1173 special library routines that come with GCC@.
1175 There may be pitfalls when you use @code{long long} types for function
1176 arguments, unless you declare function prototypes. If a function
1177 expects type @code{int} for its argument, and you pass a value of type
1178 @code{long long int}, confusion will result because the caller and the
1179 subroutine will disagree about the number of bytes for the argument.
1180 Likewise, if the function expects @code{long long int} and you pass
1181 @code{int}. The best way to avoid such problems is to use prototypes.
1184 @section Complex Numbers
1185 @cindex complex numbers
1186 @cindex @code{_Complex} keyword
1187 @cindex @code{__complex__} keyword
1189 ISO C99 supports complex floating data types, and as an extension GCC
1190 supports them in C89 mode and in C++, and supports complex integer data
1191 types which are not part of ISO C99. You can declare complex types
1192 using the keyword @code{_Complex}. As an extension, the older GNU
1193 keyword @code{__complex__} is also supported.
1195 For example, @samp{_Complex double x;} declares @code{x} as a
1196 variable whose real part and imaginary part are both of type
1197 @code{double}. @samp{_Complex short int y;} declares @code{y} to
1198 have real and imaginary parts of type @code{short int}; this is not
1199 likely to be useful, but it shows that the set of complex types is
1202 To write a constant with a complex data type, use the suffix @samp{i} or
1203 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
1204 has type @code{_Complex float} and @code{3i} has type
1205 @code{_Complex int}. Such a constant always has a pure imaginary
1206 value, but you can form any complex value you like by adding one to a
1207 real constant. This is a GNU extension; if you have an ISO C99
1208 conforming C library (such as GNU libc), and want to construct complex
1209 constants of floating type, you should include @code{<complex.h>} and
1210 use the macros @code{I} or @code{_Complex_I} instead.
1212 @cindex @code{__real__} keyword
1213 @cindex @code{__imag__} keyword
1214 To extract the real part of a complex-valued expression @var{exp}, write
1215 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
1216 extract the imaginary part. This is a GNU extension; for values of
1217 floating type, you should use the ISO C99 functions @code{crealf},
1218 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
1219 @code{cimagl}, declared in @code{<complex.h>} and also provided as
1220 built-in functions by GCC@.
1222 @cindex complex conjugation
1223 The operator @samp{~} performs complex conjugation when used on a value
1224 with a complex type. This is a GNU extension; for values of
1225 floating type, you should use the ISO C99 functions @code{conjf},
1226 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
1227 provided as built-in functions by GCC@.
1229 GCC can allocate complex automatic variables in a noncontiguous
1230 fashion; it's even possible for the real part to be in a register while
1231 the imaginary part is on the stack (or vice-versa). None of the
1232 supported debugging info formats has a way to represent noncontiguous
1233 allocation like this, so GCC describes a noncontiguous complex
1234 variable as if it were two separate variables of noncomplex type.
1235 If the variable's actual name is @code{foo}, the two fictitious
1236 variables are named @code{foo$real} and @code{foo$imag}. You can
1237 examine and set these two fictitious variables with your debugger.
1239 A future version of GDB will know how to recognize such pairs and treat
1240 them as a single variable with a complex type.
1246 ISO C99 supports floating-point numbers written not only in the usual
1247 decimal notation, such as @code{1.55e1}, but also numbers such as
1248 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
1249 supports this in C89 mode (except in some cases when strictly
1250 conforming) and in C++. In that format the
1251 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
1252 mandatory. The exponent is a decimal number that indicates the power of
1253 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
1260 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
1261 is the same as @code{1.55e1}.
1263 Unlike for floating-point numbers in the decimal notation the exponent
1264 is always required in the hexadecimal notation. Otherwise the compiler
1265 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
1266 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
1267 extension for floating-point constants of type @code{float}.
1270 @section Arrays of Length Zero
1271 @cindex arrays of length zero
1272 @cindex zero-length arrays
1273 @cindex length-zero arrays
1274 @cindex flexible array members
1276 Zero-length arrays are allowed in GNU C@. They are very useful as the
1277 last element of a structure which is really a header for a variable-length
1286 struct line *thisline = (struct line *)
1287 malloc (sizeof (struct line) + this_length);
1288 thisline->length = this_length;
1291 In ISO C89, you would have to give @code{contents} a length of 1, which
1292 means either you waste space or complicate the argument to @code{malloc}.
1294 In ISO C99, you would use a @dfn{flexible array member}, which is
1295 slightly different in syntax and semantics:
1299 Flexible array members are written as @code{contents[]} without
1303 Flexible array members have incomplete type, and so the @code{sizeof}
1304 operator may not be applied. As a quirk of the original implementation
1305 of zero-length arrays, @code{sizeof} evaluates to zero.
1308 Flexible array members may only appear as the last member of a
1309 @code{struct} that is otherwise non-empty.
1312 GCC versions before 3.0 allowed zero-length arrays to be statically
1313 initialized, as if they were flexible arrays. In addition to those
1314 cases that were useful, it also allowed initializations in situations
1315 that would corrupt later data. Non-empty initialization of zero-length
1316 arrays is now treated like any case where there are more initializer
1317 elements than the array holds, in that a suitable warning about "excess
1318 elements in array" is given, and the excess elements (all of them, in
1319 this case) are ignored.
1321 Instead GCC allows static initialization of flexible array members.
1322 This is equivalent to defining a new structure containing the original
1323 structure followed by an array of sufficient size to contain the data.
1324 I.e.@: in the following, @code{f1} is constructed as if it were declared
1330 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
1333 struct f1 f1; int data[3];
1334 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1338 The convenience of this extension is that @code{f1} has the desired
1339 type, eliminating the need to consistently refer to @code{f2.f1}.
1341 This has symmetry with normal static arrays, in that an array of
1342 unknown size is also written with @code{[]}.
1344 Of course, this extension only makes sense if the extra data comes at
1345 the end of a top-level object, as otherwise we would be overwriting
1346 data at subsequent offsets. To avoid undue complication and confusion
1347 with initialization of deeply nested arrays, we simply disallow any
1348 non-empty initialization except when the structure is the top-level
1349 object. For example:
1352 struct foo @{ int x; int y[]; @};
1353 struct bar @{ struct foo z; @};
1355 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1356 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1357 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1358 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1361 @node Variable Length
1362 @section Arrays of Variable Length
1363 @cindex variable-length arrays
1364 @cindex arrays of variable length
1367 Variable-length automatic arrays are allowed in ISO C99, and as an
1368 extension GCC accepts them in C89 mode and in C++. (However, GCC's
1369 implementation of variable-length arrays does not yet conform in detail
1370 to the ISO C99 standard.) These arrays are
1371 declared like any other automatic arrays, but with a length that is not
1372 a constant expression. The storage is allocated at the point of
1373 declaration and deallocated when the brace-level is exited. For
1378 concat_fopen (char *s1, char *s2, char *mode)
1380 char str[strlen (s1) + strlen (s2) + 1];
1383 return fopen (str, mode);
1387 @cindex scope of a variable length array
1388 @cindex variable-length array scope
1389 @cindex deallocating variable length arrays
1390 Jumping or breaking out of the scope of the array name deallocates the
1391 storage. Jumping into the scope is not allowed; you get an error
1394 @cindex @code{alloca} vs variable-length arrays
1395 You can use the function @code{alloca} to get an effect much like
1396 variable-length arrays. The function @code{alloca} is available in
1397 many other C implementations (but not in all). On the other hand,
1398 variable-length arrays are more elegant.
1400 There are other differences between these two methods. Space allocated
1401 with @code{alloca} exists until the containing @emph{function} returns.
1402 The space for a variable-length array is deallocated as soon as the array
1403 name's scope ends. (If you use both variable-length arrays and
1404 @code{alloca} in the same function, deallocation of a variable-length array
1405 will also deallocate anything more recently allocated with @code{alloca}.)
1407 You can also use variable-length arrays as arguments to functions:
1411 tester (int len, char data[len][len])
1417 The length of an array is computed once when the storage is allocated
1418 and is remembered for the scope of the array in case you access it with
1421 If you want to pass the array first and the length afterward, you can
1422 use a forward declaration in the parameter list---another GNU extension.
1426 tester (int len; char data[len][len], int len)
1432 @cindex parameter forward declaration
1433 The @samp{int len} before the semicolon is a @dfn{parameter forward
1434 declaration}, and it serves the purpose of making the name @code{len}
1435 known when the declaration of @code{data} is parsed.
1437 You can write any number of such parameter forward declarations in the
1438 parameter list. They can be separated by commas or semicolons, but the
1439 last one must end with a semicolon, which is followed by the ``real''
1440 parameter declarations. Each forward declaration must match a ``real''
1441 declaration in parameter name and data type. ISO C99 does not support
1442 parameter forward declarations.
1444 @node Variadic Macros
1445 @section Macros with a Variable Number of Arguments.
1446 @cindex variable number of arguments
1447 @cindex macro with variable arguments
1448 @cindex rest argument (in macro)
1449 @cindex variadic macros
1451 In the ISO C standard of 1999, a macro can be declared to accept a
1452 variable number of arguments much as a function can. The syntax for
1453 defining the macro is similar to that of a function. Here is an
1457 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1460 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1461 such a macro, it represents the zero or more tokens until the closing
1462 parenthesis that ends the invocation, including any commas. This set of
1463 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1464 wherever it appears. See the CPP manual for more information.
1466 GCC has long supported variadic macros, and used a different syntax that
1467 allowed you to give a name to the variable arguments just like any other
1468 argument. Here is an example:
1471 #define debug(format, args...) fprintf (stderr, format, args)
1474 This is in all ways equivalent to the ISO C example above, but arguably
1475 more readable and descriptive.
1477 GNU CPP has two further variadic macro extensions, and permits them to
1478 be used with either of the above forms of macro definition.
1480 In standard C, you are not allowed to leave the variable argument out
1481 entirely; but you are allowed to pass an empty argument. For example,
1482 this invocation is invalid in ISO C, because there is no comma after
1489 GNU CPP permits you to completely omit the variable arguments in this
1490 way. In the above examples, the compiler would complain, though since
1491 the expansion of the macro still has the extra comma after the format
1494 To help solve this problem, CPP behaves specially for variable arguments
1495 used with the token paste operator, @samp{##}. If instead you write
1498 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1501 and if the variable arguments are omitted or empty, the @samp{##}
1502 operator causes the preprocessor to remove the comma before it. If you
1503 do provide some variable arguments in your macro invocation, GNU CPP
1504 does not complain about the paste operation and instead places the
1505 variable arguments after the comma. Just like any other pasted macro
1506 argument, these arguments are not macro expanded.
1508 @node Escaped Newlines
1509 @section Slightly Looser Rules for Escaped Newlines
1510 @cindex escaped newlines
1511 @cindex newlines (escaped)
1513 Recently, the non-traditional preprocessor has relaxed its treatment of
1514 escaped newlines. Previously, the newline had to immediately follow a
1515 backslash. The current implementation allows whitespace in the form of
1516 spaces, horizontal and vertical tabs, and form feeds between the
1517 backslash and the subsequent newline. The preprocessor issues a
1518 warning, but treats it as a valid escaped newline and combines the two
1519 lines to form a single logical line. This works within comments and
1520 tokens, including multi-line strings, as well as between tokens.
1521 Comments are @emph{not} treated as whitespace for the purposes of this
1522 relaxation, since they have not yet been replaced with spaces.
1524 @node Multi-line Strings
1525 @section String Literals with Embedded Newlines
1526 @cindex multi-line string literals
1528 As an extension, GNU CPP permits string literals to cross multiple lines
1529 without escaping the embedded newlines. Each embedded newline is
1530 replaced with a single @samp{\n} character in the resulting string
1531 literal, regardless of what form the newline took originally.
1533 CPP currently allows such strings in directives as well (other than the
1534 @samp{#include} family). This is deprecated and will eventually be
1538 @section Non-Lvalue Arrays May Have Subscripts
1539 @cindex subscripting
1540 @cindex arrays, non-lvalue
1542 @cindex subscripting and function values
1543 In ISO C99, arrays that are not lvalues still decay to pointers, and
1544 may be subscripted, although they may not be modified or used after
1545 the next sequence point and the unary @samp{&} operator may not be
1546 applied to them. As an extension, GCC allows such arrays to be
1547 subscripted in C89 mode, though otherwise they do not decay to
1548 pointers outside C99 mode. For example,
1549 this is valid in GNU C though not valid in C89:
1553 struct foo @{int a[4];@};
1559 return f().a[index];
1565 @section Arithmetic on @code{void}- and Function-Pointers
1566 @cindex void pointers, arithmetic
1567 @cindex void, size of pointer to
1568 @cindex function pointers, arithmetic
1569 @cindex function, size of pointer to
1571 In GNU C, addition and subtraction operations are supported on pointers to
1572 @code{void} and on pointers to functions. This is done by treating the
1573 size of a @code{void} or of a function as 1.
1575 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1576 and on function types, and returns 1.
1578 @opindex Wpointer-arith
1579 The option @option{-Wpointer-arith} requests a warning if these extensions
1583 @section Non-Constant Initializers
1584 @cindex initializers, non-constant
1585 @cindex non-constant initializers
1587 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1588 automatic variable are not required to be constant expressions in GNU C@.
1589 Here is an example of an initializer with run-time varying elements:
1592 foo (float f, float g)
1594 float beat_freqs[2] = @{ f-g, f+g @};
1599 @node Compound Literals
1600 @section Compound Literals
1601 @cindex constructor expressions
1602 @cindex initializations in expressions
1603 @cindex structures, constructor expression
1604 @cindex expressions, constructor
1605 @cindex compound literals
1606 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1608 ISO C99 supports compound literals. A compound literal looks like
1609 a cast containing an initializer. Its value is an object of the
1610 type specified in the cast, containing the elements specified in
1611 the initializer; it is an lvalue. As an extension, GCC supports
1612 compound literals in C89 mode and in C++.
1614 Usually, the specified type is a structure. Assume that
1615 @code{struct foo} and @code{structure} are declared as shown:
1618 struct foo @{int a; char b[2];@} structure;
1622 Here is an example of constructing a @code{struct foo} with a compound literal:
1625 structure = ((struct foo) @{x + y, 'a', 0@});
1629 This is equivalent to writing the following:
1633 struct foo temp = @{x + y, 'a', 0@};
1638 You can also construct an array. If all the elements of the compound literal
1639 are (made up of) simple constant expressions, suitable for use in
1640 initializers of objects of static storage duration, then the compound
1641 literal can be coerced to a pointer to its first element and used in
1642 such an initializer, as shown here:
1645 char **foo = (char *[]) @{ "x", "y", "z" @};
1648 Compound literals for scalar types and union types are is
1649 also allowed, but then the compound literal is equivalent
1652 @node Designated Inits
1653 @section Designated Initializers
1654 @cindex initializers with labeled elements
1655 @cindex labeled elements in initializers
1656 @cindex case labels in initializers
1657 @cindex designated initializers
1659 Standard C89 requires the elements of an initializer to appear in a fixed
1660 order, the same as the order of the elements in the array or structure
1663 In ISO C99 you can give the elements in any order, specifying the array
1664 indices or structure field names they apply to, and GNU C allows this as
1665 an extension in C89 mode as well. This extension is not
1666 implemented in GNU C++.
1668 To specify an array index, write
1669 @samp{[@var{index}] =} before the element value. For example,
1672 int a[6] = @{ [4] = 29, [2] = 15 @};
1679 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1683 The index values must be constant expressions, even if the array being
1684 initialized is automatic.
1686 An alternative syntax for this which has been obsolete since GCC 2.5 but
1687 GCC still accepts is to write @samp{[@var{index}]} before the element
1688 value, with no @samp{=}.
1690 To initialize a range of elements to the same value, write
1691 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1692 extension. For example,
1695 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1699 If the value in it has side-effects, the side-effects will happen only once,
1700 not for each initialized field by the range initializer.
1703 Note that the length of the array is the highest value specified
1706 In a structure initializer, specify the name of a field to initialize
1707 with @samp{.@var{fieldname} =} before the element value. For example,
1708 given the following structure,
1711 struct point @{ int x, y; @};
1715 the following initialization
1718 struct point p = @{ .y = yvalue, .x = xvalue @};
1725 struct point p = @{ xvalue, yvalue @};
1728 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1729 @samp{@var{fieldname}:}, as shown here:
1732 struct point p = @{ y: yvalue, x: xvalue @};
1736 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1737 @dfn{designator}. You can also use a designator (or the obsolete colon
1738 syntax) when initializing a union, to specify which element of the union
1739 should be used. For example,
1742 union foo @{ int i; double d; @};
1744 union foo f = @{ .d = 4 @};
1748 will convert 4 to a @code{double} to store it in the union using
1749 the second element. By contrast, casting 4 to type @code{union foo}
1750 would store it into the union as the integer @code{i}, since it is
1751 an integer. (@xref{Cast to Union}.)
1753 You can combine this technique of naming elements with ordinary C
1754 initialization of successive elements. Each initializer element that
1755 does not have a designator applies to the next consecutive element of the
1756 array or structure. For example,
1759 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1766 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1769 Labeling the elements of an array initializer is especially useful
1770 when the indices are characters or belong to an @code{enum} type.
1775 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1776 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1779 @cindex designator lists
1780 You can also write a series of @samp{.@var{fieldname}} and
1781 @samp{[@var{index}]} designators before an @samp{=} to specify a
1782 nested subobject to initialize; the list is taken relative to the
1783 subobject corresponding to the closest surrounding brace pair. For
1784 example, with the @samp{struct point} declaration above:
1787 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1791 If the same field is initialized multiple times, it will have value from
1792 the last initialization. If any such overridden initialization has
1793 side-effect, it is unspecified whether the side-effect happens or not.
1794 Currently, gcc will discard them and issue a warning.
1797 @section Case Ranges
1799 @cindex ranges in case statements
1801 You can specify a range of consecutive values in a single @code{case} label,
1805 case @var{low} ... @var{high}:
1809 This has the same effect as the proper number of individual @code{case}
1810 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1812 This feature is especially useful for ranges of ASCII character codes:
1818 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1819 it may be parsed wrong when you use it with integer values. For example,
1834 @section Cast to a Union Type
1835 @cindex cast to a union
1836 @cindex union, casting to a
1838 A cast to union type is similar to other casts, except that the type
1839 specified is a union type. You can specify the type either with
1840 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1841 a constructor though, not a cast, and hence does not yield an lvalue like
1842 normal casts. (@xref{Compound Literals}.)
1844 The types that may be cast to the union type are those of the members
1845 of the union. Thus, given the following union and variables:
1848 union foo @{ int i; double d; @};
1854 both @code{x} and @code{y} can be cast to type @code{union foo}.
1856 Using the cast as the right-hand side of an assignment to a variable of
1857 union type is equivalent to storing in a member of the union:
1862 u = (union foo) x @equiv{} u.i = x
1863 u = (union foo) y @equiv{} u.d = y
1866 You can also use the union cast as a function argument:
1869 void hack (union foo);
1871 hack ((union foo) x);
1874 @node Mixed Declarations
1875 @section Mixed Declarations and Code
1876 @cindex mixed declarations and code
1877 @cindex declarations, mixed with code
1878 @cindex code, mixed with declarations
1880 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1881 within compound statements. As an extension, GCC also allows this in
1882 C89 mode. For example, you could do:
1891 Each identifier is visible from where it is declared until the end of
1892 the enclosing block.
1894 @node Function Attributes
1895 @section Declaring Attributes of Functions
1896 @cindex function attributes
1897 @cindex declaring attributes of functions
1898 @cindex functions that never return
1899 @cindex functions that have no side effects
1900 @cindex functions in arbitrary sections
1901 @cindex functions that behave like malloc
1902 @cindex @code{volatile} applied to function
1903 @cindex @code{const} applied to function
1904 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1905 @cindex functions that are passed arguments in registers on the 386
1906 @cindex functions that pop the argument stack on the 386
1907 @cindex functions that do not pop the argument stack on the 386
1909 In GNU C, you declare certain things about functions called in your program
1910 which help the compiler optimize function calls and check your code more
1913 The keyword @code{__attribute__} allows you to specify special
1914 attributes when making a declaration. This keyword is followed by an
1915 attribute specification inside double parentheses. The following
1916 attributes are currently defined for functions on all targets:
1917 @code{noreturn}, @code{noinline}, @code{pure}, @code{const},
1918 @code{format}, @code{format_arg}, @code{no_instrument_function},
1919 @code{section}, @code{constructor}, @code{destructor}, @code{used},
1920 @code{unused}, @code{weak}, @code{malloc}, @code{alias} and
1921 @code{no_check_memory_usage}. Several other attributes are defined for
1922 functions on particular target systems. Other attributes, including
1923 @code{section} are supported for variables declarations (@pxref{Variable
1924 Attributes}) and for types (@pxref{Type Attributes}).
1926 You may also specify attributes with @samp{__} preceding and following
1927 each keyword. This allows you to use them in header files without
1928 being concerned about a possible macro of the same name. For example,
1929 you may use @code{__noreturn__} instead of @code{noreturn}.
1931 @xref{Attribute Syntax}, for details of the exact syntax for using
1935 @cindex @code{noreturn} function attribute
1937 A few standard library functions, such as @code{abort} and @code{exit},
1938 cannot return. GCC knows this automatically. Some programs define
1939 their own functions that never return. You can declare them
1940 @code{noreturn} to tell the compiler this fact. For example,
1944 void fatal () __attribute__ ((noreturn));
1949 @dots{} /* @r{Print error message.} */ @dots{}
1955 The @code{noreturn} keyword tells the compiler to assume that
1956 @code{fatal} cannot return. It can then optimize without regard to what
1957 would happen if @code{fatal} ever did return. This makes slightly
1958 better code. More importantly, it helps avoid spurious warnings of
1959 uninitialized variables.
1961 Do not assume that registers saved by the calling function are
1962 restored before calling the @code{noreturn} function.
1964 It does not make sense for a @code{noreturn} function to have a return
1965 type other than @code{void}.
1967 The attribute @code{noreturn} is not implemented in GCC versions
1968 earlier than 2.5. An alternative way to declare that a function does
1969 not return, which works in the current version and in some older
1970 versions, is as follows:
1973 typedef void voidfn ();
1975 volatile voidfn fatal;
1978 @cindex @code{noinline} function attribute
1980 This function attribute prevents a function from being considered for
1983 @cindex @code{pure} function attribute
1985 Many functions have no effects except the return value and their
1986 return value depends only on the parameters and/or global variables.
1987 Such a function can be subject
1988 to common subexpression elimination and loop optimization just as an
1989 arithmetic operator would be. These functions should be declared
1990 with the attribute @code{pure}. For example,
1993 int square (int) __attribute__ ((pure));
1997 says that the hypothetical function @code{square} is safe to call
1998 fewer times than the program says.
2000 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
2001 Interesting non-pure functions are functions with infinite loops or those
2002 depending on volatile memory or other system resource, that may change between
2003 two consecutive calls (such as @code{feof} in a multithreading environment).
2005 The attribute @code{pure} is not implemented in GCC versions earlier
2007 @cindex @code{const} function attribute
2009 Many functions do not examine any values except their arguments, and
2010 have no effects except the return value. Basically this is just slightly
2011 more strict class than the @code{pure} attribute above, since function is not
2012 allowed to read global memory.
2014 @cindex pointer arguments
2015 Note that a function that has pointer arguments and examines the data
2016 pointed to must @emph{not} be declared @code{const}. Likewise, a
2017 function that calls a non-@code{const} function usually must not be
2018 @code{const}. It does not make sense for a @code{const} function to
2021 The attribute @code{const} is not implemented in GCC versions earlier
2022 than 2.5. An alternative way to declare that a function has no side
2023 effects, which works in the current version and in some older versions,
2027 typedef int intfn ();
2029 extern const intfn square;
2032 This approach does not work in GNU C++ from 2.6.0 on, since the language
2033 specifies that the @samp{const} must be attached to the return value.
2036 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
2037 @cindex @code{format} function attribute
2039 The @code{format} attribute specifies that a function takes @code{printf},
2040 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
2041 should be type-checked against a format string. For example, the
2046 my_printf (void *my_object, const char *my_format, ...)
2047 __attribute__ ((format (printf, 2, 3)));
2051 causes the compiler to check the arguments in calls to @code{my_printf}
2052 for consistency with the @code{printf} style format string argument
2055 The parameter @var{archetype} determines how the format string is
2056 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime}
2057 or @code{strfmon}. (You can also use @code{__printf__},
2058 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The
2059 parameter @var{string-index} specifies which argument is the format
2060 string argument (starting from 1), while @var{first-to-check} is the
2061 number of the first argument to check against the format string. For
2062 functions where the arguments are not available to be checked (such as
2063 @code{vprintf}), specify the third parameter as zero. In this case the
2064 compiler only checks the format string for consistency. For
2065 @code{strftime} formats, the third parameter is required to be zero.
2067 In the example above, the format string (@code{my_format}) is the second
2068 argument of the function @code{my_print}, and the arguments to check
2069 start with the third argument, so the correct parameters for the format
2070 attribute are 2 and 3.
2072 @opindex ffreestanding
2073 The @code{format} attribute allows you to identify your own functions
2074 which take format strings as arguments, so that GCC can check the
2075 calls to these functions for errors. The compiler always (unless
2076 @option{-ffreestanding} is used) checks formats
2077 for the standard library functions @code{printf}, @code{fprintf},
2078 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2079 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2080 warnings are requested (using @option{-Wformat}), so there is no need to
2081 modify the header file @file{stdio.h}. In C99 mode, the functions
2082 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2083 @code{vsscanf} are also checked. Except in strictly conforming C
2084 standard modes, the X/Open function @code{strfmon} is also checked.
2085 @xref{C Dialect Options,,Options Controlling C Dialect}.
2087 @item format_arg (@var{string-index})
2088 @cindex @code{format_arg} function attribute
2089 @opindex Wformat-nonliteral
2090 The @code{format_arg} attribute specifies that a function takes a format
2091 string for a @code{printf}, @code{scanf}, @code{strftime} or
2092 @code{strfmon} style function and modifies it (for example, to translate
2093 it into another language), so the result can be passed to a
2094 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2095 function (with the remaining arguments to the format function the same
2096 as they would have been for the unmodified string). For example, the
2101 my_dgettext (char *my_domain, const char *my_format)
2102 __attribute__ ((format_arg (2)));
2106 causes the compiler to check the arguments in calls to a @code{printf},
2107 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2108 format string argument is a call to the @code{my_dgettext} function, for
2109 consistency with the format string argument @code{my_format}. If the
2110 @code{format_arg} attribute had not been specified, all the compiler
2111 could tell in such calls to format functions would be that the format
2112 string argument is not constant; this would generate a warning when
2113 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2114 without the attribute.
2116 The parameter @var{string-index} specifies which argument is the format
2117 string argument (starting from 1).
2119 The @code{format-arg} attribute allows you to identify your own
2120 functions which modify format strings, so that GCC can check the
2121 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2122 type function whose operands are a call to one of your own function.
2123 The compiler always treats @code{gettext}, @code{dgettext}, and
2124 @code{dcgettext} in this manner except when strict ISO C support is
2125 requested by @option{-ansi} or an appropriate @option{-std} option, or
2126 @option{-ffreestanding} is used. @xref{C Dialect Options,,Options
2127 Controlling C Dialect}.
2129 @item no_instrument_function
2130 @cindex @code{no_instrument_function} function attribute
2131 @opindex finstrument-functions
2132 If @option{-finstrument-functions} is given, profiling function calls will
2133 be generated at entry and exit of most user-compiled functions.
2134 Functions with this attribute will not be so instrumented.
2136 @item section ("@var{section-name}")
2137 @cindex @code{section} function attribute
2138 Normally, the compiler places the code it generates in the @code{text} section.
2139 Sometimes, however, you need additional sections, or you need certain
2140 particular functions to appear in special sections. The @code{section}
2141 attribute specifies that a function lives in a particular section.
2142 For example, the declaration:
2145 extern void foobar (void) __attribute__ ((section ("bar")));
2149 puts the function @code{foobar} in the @code{bar} section.
2151 Some file formats do not support arbitrary sections so the @code{section}
2152 attribute is not available on all platforms.
2153 If you need to map the entire contents of a module to a particular
2154 section, consider using the facilities of the linker instead.
2158 @cindex @code{constructor} function attribute
2159 @cindex @code{destructor} function attribute
2160 The @code{constructor} attribute causes the function to be called
2161 automatically before execution enters @code{main ()}. Similarly, the
2162 @code{destructor} attribute causes the function to be called
2163 automatically after @code{main ()} has completed or @code{exit ()} has
2164 been called. Functions with these attributes are useful for
2165 initializing data that will be used implicitly during the execution of
2168 These attributes are not currently implemented for Objective-C@.
2170 @cindex @code{unused} attribute.
2172 This attribute, attached to a function, means that the function is meant
2173 to be possibly unused. GCC will not produce a warning for this
2174 function. GNU C++ does not currently support this attribute as
2175 definitions without parameters are valid in C++.
2177 @cindex @code{used} attribute.
2179 This attribute, attached to a function, means that code must be emitted
2180 for the function even if it appears that the function is not referenced.
2181 This is useful, for example, when the function is referenced only in
2185 @cindex @code{weak} attribute
2186 The @code{weak} attribute causes the declaration to be emitted as a weak
2187 symbol rather than a global. This is primarily useful in defining
2188 library functions which can be overridden in user code, though it can
2189 also be used with non-function declarations. Weak symbols are supported
2190 for ELF targets, and also for a.out targets when using the GNU assembler
2194 @cindex @code{malloc} attribute
2195 The @code{malloc} attribute is used to tell the compiler that a function
2196 may be treated as if it were the malloc function. The compiler assumes
2197 that calls to malloc result in a pointers that cannot alias anything.
2198 This will often improve optimization.
2200 @item alias ("@var{target}")
2201 @cindex @code{alias} attribute
2202 The @code{alias} attribute causes the declaration to be emitted as an
2203 alias for another symbol, which must be specified. For instance,
2206 void __f () @{ /* do something */; @}
2207 void f () __attribute__ ((weak, alias ("__f")));
2210 declares @samp{f} to be a weak alias for @samp{__f}. In C++, the
2211 mangled name for the target must be used.
2213 Not all target machines support this attribute.
2215 @item no_check_memory_usage
2216 @cindex @code{no_check_memory_usage} function attribute
2217 @opindex fcheck-memory-usage
2218 The @code{no_check_memory_usage} attribute causes GCC to omit checks
2219 of memory references when it generates code for that function. Normally
2220 if you specify @option{-fcheck-memory-usage} (see @pxref{Code Gen
2221 Options}), GCC generates calls to support routines before most memory
2222 accesses to permit support code to record usage and detect uses of
2223 uninitialized or unallocated storage. Since GCC cannot handle
2224 @code{asm} statements properly they are not allowed in such functions.
2225 If you declare a function with this attribute, GCC will not generate
2226 memory checking code for that function, permitting the use of @code{asm}
2227 statements without having to compile that function with different
2228 options. This also allows you to write support routines of your own if
2229 you wish, without getting infinite recursion if they get compiled with
2230 @option{-fcheck-memory-usage}.
2232 @item regparm (@var{number})
2233 @cindex functions that are passed arguments in registers on the 386
2234 On the Intel 386, the @code{regparm} attribute causes the compiler to
2235 pass up to @var{number} integer arguments in registers EAX,
2236 EDX, and ECX instead of on the stack. Functions that take a
2237 variable number of arguments will continue to be passed all of their
2238 arguments on the stack.
2241 @cindex functions that pop the argument stack on the 386
2242 On the Intel 386, the @code{stdcall} attribute causes the compiler to
2243 assume that the called function will pop off the stack space used to
2244 pass arguments, unless it takes a variable number of arguments.
2246 The PowerPC compiler for Windows NT currently ignores the @code{stdcall}
2250 @cindex functions that do pop the argument stack on the 386
2252 On the Intel 386, the @code{cdecl} attribute causes the compiler to
2253 assume that the calling function will pop off the stack space used to
2254 pass arguments. This is
2255 useful to override the effects of the @option{-mrtd} switch.
2257 The PowerPC compiler for Windows NT currently ignores the @code{cdecl}
2261 @cindex functions called via pointer on the RS/6000 and PowerPC
2262 On the RS/6000 and PowerPC, the @code{longcall} attribute causes the
2263 compiler to always call the function via a pointer, so that functions
2264 which reside further than 64 megabytes (67,108,864 bytes) from the
2265 current location can be called.
2267 @item long_call/short_call
2268 @cindex indirect calls on ARM
2269 This attribute allows to specify how to call a particular function on
2270 ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options})
2271 command line switch and @code{#pragma long_calls} settings. The
2272 @code{long_call} attribute causes the compiler to always call the
2273 function by first loading its address into a register and then using the
2274 contents of that register. The @code{short_call} attribute always places
2275 the offset to the function from the call site into the @samp{BL}
2276 instruction directly.
2279 @cindex functions which are imported from a dll on PowerPC Windows NT
2280 On the PowerPC running Windows NT, the @code{dllimport} attribute causes
2281 the compiler to call the function via a global pointer to the function
2282 pointer that is set up by the Windows NT dll library. The pointer name
2283 is formed by combining @code{__imp_} and the function name.
2286 @cindex functions which are exported from a dll on PowerPC Windows NT
2287 On the PowerPC running Windows NT, the @code{dllexport} attribute causes
2288 the compiler to provide a global pointer to the function pointer, so
2289 that it can be called with the @code{dllimport} attribute. The pointer
2290 name is formed by combining @code{__imp_} and the function name.
2292 @item exception (@var{except-func} [, @var{except-arg}])
2293 @cindex functions which specify exception handling on PowerPC Windows NT
2294 On the PowerPC running Windows NT, the @code{exception} attribute causes
2295 the compiler to modify the structured exception table entry it emits for
2296 the declared function. The string or identifier @var{except-func} is
2297 placed in the third entry of the structured exception table. It
2298 represents a function, which is called by the exception handling
2299 mechanism if an exception occurs. If it was specified, the string or
2300 identifier @var{except-arg} is placed in the fourth entry of the
2301 structured exception table.
2303 @item function_vector
2304 @cindex calling functions through the function vector on the H8/300 processors
2305 Use this option on the H8/300 and H8/300H to indicate that the specified
2306 function should be called through the function vector. Calling a
2307 function through the function vector will reduce code size, however;
2308 the function vector has a limited size (maximum 128 entries on the H8/300
2309 and 64 entries on the H8/300H) and shares space with the interrupt vector.
2311 You must use GAS and GLD from GNU binutils version 2.7 or later for
2312 this option to work correctly.
2315 @cindex interrupt handler functions
2316 Use this option on the ARM, AVR and M32R/D ports to indicate that the
2317 specified function is an interrupt handler. The compiler will generate
2318 function entry and exit sequences suitable for use in an interrupt
2319 handler when this attribute is present.
2321 Note, interrupt handlers for the H8/300, H8/300H and SH processors can
2322 be specified via the @code{interrupt_handler} attribute.
2324 Note, on the AVR interrupts will be enabled inside the function.
2326 Note, for the ARM you can specify the kind of interrupt to be handled by
2327 adding an optional parameter to the interrupt attribute like this:
2330 void f () __attribute__ ((interrupt ("IRQ")));
2333 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
2335 @item interrupt_handler
2336 @cindex interrupt handler functions on the H8/300 and SH processors
2337 Use this option on the H8/300, H8/300H and SH to indicate that the
2338 specified function is an interrupt handler. The compiler will generate
2339 function entry and exit sequences suitable for use in an interrupt
2340 handler when this attribute is present.
2343 Use this option on the SH to indicate an @code{interrupt_handler}
2344 function should switch to an alternate stack. It expects a string
2345 argument that names a global variable holding the address of the
2350 void f () __attribute__ ((interrupt_handler,
2351 sp_switch ("alt_stack")));
2355 Use this option on the SH for an @code{interrupt_handle} to return using
2356 @code{trapa} instead of @code{rte}. This attribute expects an integer
2357 argument specifying the trap number to be used.
2360 @cindex eight bit data on the H8/300 and H8/300H
2361 Use this option on the H8/300 and H8/300H to indicate that the specified
2362 variable should be placed into the eight bit data section.
2363 The compiler will generate more efficient code for certain operations
2364 on data in the eight bit data area. Note the eight bit data area is limited to
2367 You must use GAS and GLD from GNU binutils version 2.7 or later for
2368 this option to work correctly.
2371 @cindex tiny data section on the H8/300H
2372 Use this option on the H8/300H to indicate that the specified
2373 variable should be placed into the tiny data section.
2374 The compiler will generate more efficient code for loads and stores
2375 on data in the tiny data section. Note the tiny data area is limited to
2376 slightly under 32kbytes of data.
2379 @cindex signal handler functions on the AVR processors
2380 Use this option on the AVR to indicate that the specified
2381 function is an signal handler. The compiler will generate function
2382 entry and exit sequences suitable for use in an signal handler when this
2383 attribute is present. Interrupts will be disabled inside function.
2386 @cindex function without a prologue/epilogue code
2387 Use this option on the ARM or AVR ports to indicate that the specified
2388 function do not need prologue/epilogue sequences generated by the
2389 compiler. It is up to the programmer to provide these sequences.
2391 @item model (@var{model-name})
2392 @cindex function addressability on the M32R/D
2393 Use this attribute on the M32R/D to set the addressability of an object,
2394 and the code generated for a function.
2395 The identifier @var{model-name} is one of @code{small}, @code{medium},
2396 or @code{large}, representing each of the code models.
2398 Small model objects live in the lower 16MB of memory (so that their
2399 addresses can be loaded with the @code{ld24} instruction), and are
2400 callable with the @code{bl} instruction.
2402 Medium model objects may live anywhere in the 32-bit address space (the
2403 compiler will generate @code{seth/add3} instructions to load their addresses),
2404 and are callable with the @code{bl} instruction.
2406 Large model objects may live anywhere in the 32-bit address space (the
2407 compiler will generate @code{seth/add3} instructions to load their addresses),
2408 and may not be reachable with the @code{bl} instruction (the compiler will
2409 generate the much slower @code{seth/add3/jl} instruction sequence).
2413 You can specify multiple attributes in a declaration by separating them
2414 by commas within the double parentheses or by immediately following an
2415 attribute declaration with another attribute declaration.
2417 @cindex @code{#pragma}, reason for not using
2418 @cindex pragma, reason for not using
2419 Some people object to the @code{__attribute__} feature, suggesting that
2420 ISO C's @code{#pragma} should be used instead. At the time
2421 @code{__attribute__} was designed, there were two reasons for not doing
2426 It is impossible to generate @code{#pragma} commands from a macro.
2429 There is no telling what the same @code{#pragma} might mean in another
2433 These two reasons applied to almost any application that might have been
2434 proposed for @code{#pragma}. It was basically a mistake to use
2435 @code{#pragma} for @emph{anything}.
2437 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
2438 to be generated from macros. In addition, a @code{#pragma GCC}
2439 namespace is now in use for GCC-specific pragmas. However, it has been
2440 found convenient to use @code{__attribute__} to achieve a natural
2441 attachment of attributes to their corresponding declarations, whereas
2442 @code{#pragma GCC} is of use for constructs that do not naturally form
2443 part of the grammar. @xref{Other Directives,,Miscellaneous
2444 Preprocessing Directives, cpp, The C Preprocessor}.
2446 @node Attribute Syntax
2447 @section Attribute Syntax
2448 @cindex attribute syntax
2450 This section describes the syntax with which @code{__attribute__} may be
2451 used, and the constructs to which attribute specifiers bind, for the C
2452 language. Some details may vary for C++ and Objective-C@. Because of
2453 infelicities in the grammar for attributes, some forms described here
2454 may not be successfully parsed in all cases.
2456 There are some problems with the semantics of attributes in C++. For
2457 example, there are no manglings for attributes, although they may affect
2458 code generation, so problems may arise when attributed types are used in
2459 conjunction with templates or overloading. Similarly, @code{typeid}
2460 does not distinguish between types with different attributes. Support
2461 for attributes in C++ may be restricted in future to attributes on
2462 declarations only, but not on nested declarators.
2464 @xref{Function Attributes}, for details of the semantics of attributes
2465 applying to functions. @xref{Variable Attributes}, for details of the
2466 semantics of attributes applying to variables. @xref{Type Attributes},
2467 for details of the semantics of attributes applying to structure, union
2468 and enumerated types.
2470 An @dfn{attribute specifier} is of the form
2471 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
2472 is a possibly empty comma-separated sequence of @dfn{attributes}, where
2473 each attribute is one of the following:
2477 Empty. Empty attributes are ignored.
2480 A word (which may be an identifier such as @code{unused}, or a reserved
2481 word such as @code{const}).
2484 A word, followed by, in parentheses, parameters for the attribute.
2485 These parameters take one of the following forms:
2489 An identifier. For example, @code{mode} attributes use this form.
2492 An identifier followed by a comma and a non-empty comma-separated list
2493 of expressions. For example, @code{format} attributes use this form.
2496 A possibly empty comma-separated list of expressions. For example,
2497 @code{format_arg} attributes use this form with the list being a single
2498 integer constant expression, and @code{alias} attributes use this form
2499 with the list being a single string constant.
2503 An @dfn{attribute specifier list} is a sequence of one or more attribute
2504 specifiers, not separated by any other tokens.
2506 An attribute specifier list may appear after the colon following a
2507 label, other than a @code{case} or @code{default} label. The only
2508 attribute it makes sense to use after a label is @code{unused}. This
2509 feature is intended for code generated by programs which contains labels
2510 that may be unused but which is compiled with @option{-Wall}. It would
2511 not normally be appropriate to use in it human-written code, though it
2512 could be useful in cases where the code that jumps to the label is
2513 contained within an @code{#ifdef} conditional.
2515 An attribute specifier list may appear as part of a @code{struct},
2516 @code{union} or @code{enum} specifier. It may go either immediately
2517 after the @code{struct}, @code{union} or @code{enum} keyword, or after
2518 the closing brace. It is ignored if the content of the structure, union
2519 or enumerated type is not defined in the specifier in which the
2520 attribute specifier list is used---that is, in usages such as
2521 @code{struct __attribute__((foo)) bar} with no following opening brace.
2522 Where attribute specifiers follow the closing brace, they are considered
2523 to relate to the structure, union or enumerated type defined, not to any
2524 enclosing declaration the type specifier appears in, and the type
2525 defined is not complete until after the attribute specifiers.
2526 @c Otherwise, there would be the following problems: a shift/reduce
2527 @c conflict between attributes binding the struct/union/enum and
2528 @c binding to the list of specifiers/qualifiers; and "aligned"
2529 @c attributes could use sizeof for the structure, but the size could be
2530 @c changed later by "packed" attributes.
2532 Otherwise, an attribute specifier appears as part of a declaration,
2533 counting declarations of unnamed parameters and type names, and relates
2534 to that declaration (which may be nested in another declaration, for
2535 example in the case of a parameter declaration), or to a particular declarator
2536 within a declaration. Where an
2537 attribute specifier is applied to a parameter declared as a function or
2538 an array, it should apply to the function or array rather than the
2539 pointer to which the parameter is implicitly converted, but this is not
2540 yet correctly implemented.
2542 Any list of specifiers and qualifiers at the start of a declaration may
2543 contain attribute specifiers, whether or not such a list may in that
2544 context contain storage class specifiers. (Some attributes, however,
2545 are essentially in the nature of storage class specifiers, and only make
2546 sense where storage class specifiers may be used; for example,
2547 @code{section}.) There is one necessary limitation to this syntax: the
2548 first old-style parameter declaration in a function definition cannot
2549 begin with an attribute specifier, because such an attribute applies to
2550 the function instead by syntax described below (which, however, is not
2551 yet implemented in this case). In some other cases, attribute
2552 specifiers are permitted by this grammar but not yet supported by the
2553 compiler. All attribute specifiers in this place relate to the
2554 declaration as a whole. In the obsolescent usage where a type of
2555 @code{int} is implied by the absence of type specifiers, such a list of
2556 specifiers and qualifiers may be an attribute specifier list with no
2557 other specifiers or qualifiers.
2559 An attribute specifier list may appear immediately before a declarator
2560 (other than the first) in a comma-separated list of declarators in a
2561 declaration of more than one identifier using a single list of
2562 specifiers and qualifiers. Such attribute specifiers apply
2563 only to the identifier before whose declarator they appear. For
2567 __attribute__((noreturn)) void d0 (void),
2568 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
2573 the @code{noreturn} attribute applies to all the functions
2574 declared; the @code{format} attribute only applies to @code{d1}.
2576 An attribute specifier list may appear immediately before the comma,
2577 @code{=} or semicolon terminating the declaration of an identifier other
2578 than a function definition. At present, such attribute specifiers apply
2579 to the declared object or function, but in future they may attach to the
2580 outermost adjacent declarator. In simple cases there is no difference,
2581 but, for example, in
2584 void (****f)(void) __attribute__((noreturn));
2588 at present the @code{noreturn} attribute applies to @code{f}, which
2589 causes a warning since @code{f} is not a function, but in future it may
2590 apply to the function @code{****f}. The precise semantics of what
2591 attributes in such cases will apply to are not yet specified. Where an
2592 assembler name for an object or function is specified (@pxref{Asm
2593 Labels}), at present the attribute must follow the @code{asm}
2594 specification; in future, attributes before the @code{asm} specification
2595 may apply to the adjacent declarator, and those after it to the declared
2598 An attribute specifier list may, in future, be permitted to appear after
2599 the declarator in a function definition (before any old-style parameter
2600 declarations or the function body).
2602 Attribute specifiers may be mixed with type qualifiers appearing inside
2603 the @code{[]} of a parameter array declarator, in the C99 construct by
2604 which such qualifiers are applied to the pointer to which the array is
2605 implicitly converted. Such attribute specifiers apply to the pointer,
2606 not to the array, but at present this is not implemented and they are
2609 An attribute specifier list may appear at the start of a nested
2610 declarator. At present, there are some limitations in this usage: the
2611 attributes correctly apply to the declarator, but for most individual
2612 attributes the semantics this implies are not implemented.
2613 When attribute specifiers follow the @code{*} of a pointer
2614 declarator, they may be mixed with any type qualifiers present.
2615 The following describes the formal semantics of this syntax. It will make the
2616 most sense if you are familiar with the formal specification of
2617 declarators in the ISO C standard.
2619 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
2620 D1}, where @code{T} contains declaration specifiers that specify a type
2621 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
2622 contains an identifier @var{ident}. The type specified for @var{ident}
2623 for derived declarators whose type does not include an attribute
2624 specifier is as in the ISO C standard.
2626 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
2627 and the declaration @code{T D} specifies the type
2628 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2629 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2630 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
2632 If @code{D1} has the form @code{*
2633 @var{type-qualifier-and-attribute-specifier-list} D}, and the
2634 declaration @code{T D} specifies the type
2635 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2636 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2637 @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
2643 void (__attribute__((noreturn)) ****f) (void);
2647 specifies the type ``pointer to pointer to pointer to pointer to
2648 non-returning function returning @code{void}''. As another example,
2651 char *__attribute__((aligned(8))) *f;
2655 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
2656 Note again that this does not work with most attributes; for example,
2657 the usage of @samp{aligned} and @samp{noreturn} attributes given above
2658 is not yet supported.
2660 For compatibility with existing code written for compiler versions that
2661 did not implement attributes on nested declarators, some laxity is
2662 allowed in the placing of attributes. If an attribute that only applies
2663 to types is applied to a declaration, it will be treated as applying to
2664 the type of that declaration. If an attribute that only applies to
2665 declarations is applied to the type of a declaration, it will be treated
2666 as applying to that declaration; and, for compatibility with code
2667 placing the attributes immediately before the identifier declared, such
2668 an attribute applied to a function return type will be treated as
2669 applying to the function type, and such an attribute applied to an array
2670 element type will be treated as applying to the array type. If an
2671 attribute that only applies to function types is applied to a
2672 pointer-to-function type, it will be treated as applying to the pointer
2673 target type; if such an attribute is applied to a function return type
2674 that is not a pointer-to-function type, it will be treated as applying
2675 to the function type.
2677 @node Function Prototypes
2678 @section Prototypes and Old-Style Function Definitions
2679 @cindex function prototype declarations
2680 @cindex old-style function definitions
2681 @cindex promotion of formal parameters
2683 GNU C extends ISO C to allow a function prototype to override a later
2684 old-style non-prototype definition. Consider the following example:
2687 /* @r{Use prototypes unless the compiler is old-fashioned.} */
2694 /* @r{Prototype function declaration.} */
2695 int isroot P((uid_t));
2697 /* @r{Old-style function definition.} */
2699 isroot (x) /* ??? lossage here ??? */
2706 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
2707 not allow this example, because subword arguments in old-style
2708 non-prototype definitions are promoted. Therefore in this example the
2709 function definition's argument is really an @code{int}, which does not
2710 match the prototype argument type of @code{short}.
2712 This restriction of ISO C makes it hard to write code that is portable
2713 to traditional C compilers, because the programmer does not know
2714 whether the @code{uid_t} type is @code{short}, @code{int}, or
2715 @code{long}. Therefore, in cases like these GNU C allows a prototype
2716 to override a later old-style definition. More precisely, in GNU C, a
2717 function prototype argument type overrides the argument type specified
2718 by a later old-style definition if the former type is the same as the
2719 latter type before promotion. Thus in GNU C the above example is
2720 equivalent to the following:
2733 GNU C++ does not support old-style function definitions, so this
2734 extension is irrelevant.
2737 @section C++ Style Comments
2739 @cindex C++ comments
2740 @cindex comments, C++ style
2742 In GNU C, you may use C++ style comments, which start with @samp{//} and
2743 continue until the end of the line. Many other C implementations allow
2744 such comments, and they are likely to be in a future C standard.
2745 However, C++ style comments are not recognized if you specify
2746 @w{@option{-ansi}}, a @option{-std} option specifying a version of ISO C
2747 before C99, or @w{@option{-traditional}}, since they are incompatible
2748 with traditional constructs like @code{dividend//*comment*/divisor}.
2751 @section Dollar Signs in Identifier Names
2753 @cindex dollar signs in identifier names
2754 @cindex identifier names, dollar signs in
2756 In GNU C, you may normally use dollar signs in identifier names.
2757 This is because many traditional C implementations allow such identifiers.
2758 However, dollar signs in identifiers are not supported on a few target
2759 machines, typically because the target assembler does not allow them.
2761 @node Character Escapes
2762 @section The Character @key{ESC} in Constants
2764 You can use the sequence @samp{\e} in a string or character constant to
2765 stand for the ASCII character @key{ESC}.
2768 @section Inquiring on Alignment of Types or Variables
2770 @cindex type alignment
2771 @cindex variable alignment
2773 The keyword @code{__alignof__} allows you to inquire about how an object
2774 is aligned, or the minimum alignment usually required by a type. Its
2775 syntax is just like @code{sizeof}.
2777 For example, if the target machine requires a @code{double} value to be
2778 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
2779 This is true on many RISC machines. On more traditional machine
2780 designs, @code{__alignof__ (double)} is 4 or even 2.
2782 Some machines never actually require alignment; they allow reference to any
2783 data type even at an odd addresses. For these machines, @code{__alignof__}
2784 reports the @emph{recommended} alignment of a type.
2786 When the operand of @code{__alignof__} is an lvalue rather than a type, the
2787 value is the largest alignment that the lvalue is known to have. It may
2788 have this alignment as a result of its data type, or because it is part of
2789 a structure and inherits alignment from that structure. For example, after
2793 struct foo @{ int x; char y; @} foo1;
2797 the value of @code{__alignof__ (foo1.y)} is probably 2 or 4, the same as
2798 @code{__alignof__ (int)}, even though the data type of @code{foo1.y}
2799 does not itself demand any alignment.
2801 It is an error to ask for the alignment of an incomplete type.
2803 A related feature which lets you specify the alignment of an object is
2804 @code{__attribute__ ((aligned (@var{alignment})))}; see the following
2807 @node Variable Attributes
2808 @section Specifying Attributes of Variables
2809 @cindex attribute of variables
2810 @cindex variable attributes
2812 The keyword @code{__attribute__} allows you to specify special
2813 attributes of variables or structure fields. This keyword is followed
2814 by an attribute specification inside double parentheses. Eight
2815 attributes are currently defined for variables: @code{aligned},
2816 @code{mode}, @code{nocommon}, @code{packed}, @code{section},
2817 @code{transparent_union}, @code{unused}, and @code{weak}. Some other
2818 attributes are defined for variables on particular target systems. Other
2819 attributes are available for functions (@pxref{Function Attributes}) and
2820 for types (@pxref{Type Attributes}). Other front ends might define more
2821 attributes (@pxref{C++ Extensions,,Extensions to the C++ Language}).
2823 You may also specify attributes with @samp{__} preceding and following
2824 each keyword. This allows you to use them in header files without
2825 being concerned about a possible macro of the same name. For example,
2826 you may use @code{__aligned__} instead of @code{aligned}.
2828 @xref{Attribute Syntax}, for details of the exact syntax for using
2832 @cindex @code{aligned} attribute
2833 @item aligned (@var{alignment})
2834 This attribute specifies a minimum alignment for the variable or
2835 structure field, measured in bytes. For example, the declaration:
2838 int x __attribute__ ((aligned (16))) = 0;
2842 causes the compiler to allocate the global variable @code{x} on a
2843 16-byte boundary. On a 68040, this could be used in conjunction with
2844 an @code{asm} expression to access the @code{move16} instruction which
2845 requires 16-byte aligned operands.
2847 You can also specify the alignment of structure fields. For example, to
2848 create a double-word aligned @code{int} pair, you could write:
2851 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
2855 This is an alternative to creating a union with a @code{double} member
2856 that forces the union to be double-word aligned.
2858 It is not possible to specify the alignment of functions; the alignment
2859 of functions is determined by the machine's requirements and cannot be
2860 changed. You cannot specify alignment for a typedef name because such a
2861 name is just an alias, not a distinct type.
2863 As in the preceding examples, you can explicitly specify the alignment
2864 (in bytes) that you wish the compiler to use for a given variable or
2865 structure field. Alternatively, you can leave out the alignment factor
2866 and just ask the compiler to align a variable or field to the maximum
2867 useful alignment for the target machine you are compiling for. For
2868 example, you could write:
2871 short array[3] __attribute__ ((aligned));
2874 Whenever you leave out the alignment factor in an @code{aligned} attribute
2875 specification, the compiler automatically sets the alignment for the declared
2876 variable or field to the largest alignment which is ever used for any data
2877 type on the target machine you are compiling for. Doing this can often make
2878 copy operations more efficient, because the compiler can use whatever
2879 instructions copy the biggest chunks of memory when performing copies to
2880 or from the variables or fields that you have aligned this way.
2882 The @code{aligned} attribute can only increase the alignment; but you
2883 can decrease it by specifying @code{packed} as well. See below.
2885 Note that the effectiveness of @code{aligned} attributes may be limited
2886 by inherent limitations in your linker. On many systems, the linker is
2887 only able to arrange for variables to be aligned up to a certain maximum
2888 alignment. (For some linkers, the maximum supported alignment may
2889 be very very small.) If your linker is only able to align variables
2890 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
2891 in an @code{__attribute__} will still only provide you with 8 byte
2892 alignment. See your linker documentation for further information.
2894 @item mode (@var{mode})
2895 @cindex @code{mode} attribute
2896 This attribute specifies the data type for the declaration---whichever
2897 type corresponds to the mode @var{mode}. This in effect lets you
2898 request an integer or floating point type according to its width.
2900 You may also specify a mode of @samp{byte} or @samp{__byte__} to
2901 indicate the mode corresponding to a one-byte integer, @samp{word} or
2902 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
2903 or @samp{__pointer__} for the mode used to represent pointers.
2906 @cindex @code{nocommon} attribute
2908 This attribute specifies requests GCC not to place a variable
2909 ``common'' but instead to allocate space for it directly. If you
2910 specify the @option{-fno-common} flag, GCC will do this for all
2913 Specifying the @code{nocommon} attribute for a variable provides an
2914 initialization of zeros. A variable may only be initialized in one
2918 @cindex @code{packed} attribute
2919 The @code{packed} attribute specifies that a variable or structure field
2920 should have the smallest possible alignment---one byte for a variable,
2921 and one bit for a field, unless you specify a larger value with the
2922 @code{aligned} attribute.
2924 Here is a structure in which the field @code{x} is packed, so that it
2925 immediately follows @code{a}:
2931 int x[2] __attribute__ ((packed));
2935 @item section ("@var{section-name}")
2936 @cindex @code{section} variable attribute
2937 Normally, the compiler places the objects it generates in sections like
2938 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
2939 or you need certain particular variables to appear in special sections,
2940 for example to map to special hardware. The @code{section}
2941 attribute specifies that a variable (or function) lives in a particular
2942 section. For example, this small program uses several specific section names:
2945 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
2946 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
2947 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
2948 int init_data __attribute__ ((section ("INITDATA"))) = 0;
2952 /* Initialize stack pointer */
2953 init_sp (stack + sizeof (stack));
2955 /* Initialize initialized data */
2956 memcpy (&init_data, &data, &edata - &data);
2958 /* Turn on the serial ports */
2965 Use the @code{section} attribute with an @emph{initialized} definition
2966 of a @emph{global} variable, as shown in the example. GCC issues
2967 a warning and otherwise ignores the @code{section} attribute in
2968 uninitialized variable declarations.
2970 You may only use the @code{section} attribute with a fully initialized
2971 global definition because of the way linkers work. The linker requires
2972 each object be defined once, with the exception that uninitialized
2973 variables tentatively go in the @code{common} (or @code{bss}) section
2974 and can be multiply ``defined''. You can force a variable to be
2975 initialized with the @option{-fno-common} flag or the @code{nocommon}
2978 Some file formats do not support arbitrary sections so the @code{section}
2979 attribute is not available on all platforms.
2980 If you need to map the entire contents of a module to a particular
2981 section, consider using the facilities of the linker instead.
2984 @cindex @code{shared} variable attribute
2985 On Windows NT, in addition to putting variable definitions in a named
2986 section, the section can also be shared among all running copies of an
2987 executable or DLL@. For example, this small program defines shared data
2988 by putting it in a named section @code{shared} and marking the section
2992 int foo __attribute__((section ("shared"), shared)) = 0;
2997 /* Read and write foo. All running
2998 copies see the same value. */
3004 You may only use the @code{shared} attribute along with @code{section}
3005 attribute with a fully initialized global definition because of the way
3006 linkers work. See @code{section} attribute for more information.
3008 The @code{shared} attribute is only available on Windows NT@.
3010 @item transparent_union
3011 This attribute, attached to a function parameter which is a union, means
3012 that the corresponding argument may have the type of any union member,
3013 but the argument is passed as if its type were that of the first union
3014 member. For more details see @xref{Type Attributes}. You can also use
3015 this attribute on a @code{typedef} for a union data type; then it
3016 applies to all function parameters with that type.
3019 This attribute, attached to a variable, means that the variable is meant
3020 to be possibly unused. GCC will not produce a warning for this
3024 The @code{weak} attribute is described in @xref{Function Attributes}.
3026 @item model (@var{model-name})
3027 @cindex variable addressability on the M32R/D
3028 Use this attribute on the M32R/D to set the addressability of an object.
3029 The identifier @var{model-name} is one of @code{small}, @code{medium},
3030 or @code{large}, representing each of the code models.
3032 Small model objects live in the lower 16MB of memory (so that their
3033 addresses can be loaded with the @code{ld24} instruction).
3035 Medium and large model objects may live anywhere in the 32-bit address space
3036 (the compiler will generate @code{seth/add3} instructions to load their
3041 To specify multiple attributes, separate them by commas within the
3042 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3045 @node Type Attributes
3046 @section Specifying Attributes of Types
3047 @cindex attribute of types
3048 @cindex type attributes
3050 The keyword @code{__attribute__} allows you to specify special
3051 attributes of @code{struct} and @code{union} types when you define such
3052 types. This keyword is followed by an attribute specification inside
3053 double parentheses. Four attributes are currently defined for types:
3054 @code{aligned}, @code{packed}, @code{transparent_union}, and @code{unused}.
3055 Other attributes are defined for functions (@pxref{Function Attributes}) and
3056 for variables (@pxref{Variable Attributes}).
3058 You may also specify any one of these attributes with @samp{__}
3059 preceding and following its keyword. This allows you to use these
3060 attributes in header files without being concerned about a possible
3061 macro of the same name. For example, you may use @code{__aligned__}
3062 instead of @code{aligned}.
3064 You may specify the @code{aligned} and @code{transparent_union}
3065 attributes either in a @code{typedef} declaration or just past the
3066 closing curly brace of a complete enum, struct or union type
3067 @emph{definition} and the @code{packed} attribute only past the closing
3068 brace of a definition.
3070 You may also specify attributes between the enum, struct or union
3071 tag and the name of the type rather than after the closing brace.
3073 @xref{Attribute Syntax}, for details of the exact syntax for using
3077 @cindex @code{aligned} attribute
3078 @item aligned (@var{alignment})
3079 This attribute specifies a minimum alignment (in bytes) for variables
3080 of the specified type. For example, the declarations:
3083 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
3084 typedef int more_aligned_int __attribute__ ((aligned (8)));
3088 force the compiler to insure (as far as it can) that each variable whose
3089 type is @code{struct S} or @code{more_aligned_int} will be allocated and
3090 aligned @emph{at least} on a 8-byte boundary. On a Sparc, having all
3091 variables of type @code{struct S} aligned to 8-byte boundaries allows
3092 the compiler to use the @code{ldd} and @code{std} (doubleword load and
3093 store) instructions when copying one variable of type @code{struct S} to
3094 another, thus improving run-time efficiency.
3096 Note that the alignment of any given @code{struct} or @code{union} type
3097 is required by the ISO C standard to be at least a perfect multiple of
3098 the lowest common multiple of the alignments of all of the members of
3099 the @code{struct} or @code{union} in question. This means that you @emph{can}
3100 effectively adjust the alignment of a @code{struct} or @code{union}
3101 type by attaching an @code{aligned} attribute to any one of the members
3102 of such a type, but the notation illustrated in the example above is a
3103 more obvious, intuitive, and readable way to request the compiler to
3104 adjust the alignment of an entire @code{struct} or @code{union} type.
3106 As in the preceding example, you can explicitly specify the alignment
3107 (in bytes) that you wish the compiler to use for a given @code{struct}
3108 or @code{union} type. Alternatively, you can leave out the alignment factor
3109 and just ask the compiler to align a type to the maximum
3110 useful alignment for the target machine you are compiling for. For
3111 example, you could write:
3114 struct S @{ short f[3]; @} __attribute__ ((aligned));
3117 Whenever you leave out the alignment factor in an @code{aligned}
3118 attribute specification, the compiler automatically sets the alignment
3119 for the type to the largest alignment which is ever used for any data
3120 type on the target machine you are compiling for. Doing this can often
3121 make copy operations more efficient, because the compiler can use
3122 whatever instructions copy the biggest chunks of memory when performing
3123 copies to or from the variables which have types that you have aligned
3126 In the example above, if the size of each @code{short} is 2 bytes, then
3127 the size of the entire @code{struct S} type is 6 bytes. The smallest
3128 power of two which is greater than or equal to that is 8, so the
3129 compiler sets the alignment for the entire @code{struct S} type to 8
3132 Note that although you can ask the compiler to select a time-efficient
3133 alignment for a given type and then declare only individual stand-alone
3134 objects of that type, the compiler's ability to select a time-efficient
3135 alignment is primarily useful only when you plan to create arrays of
3136 variables having the relevant (efficiently aligned) type. If you
3137 declare or use arrays of variables of an efficiently-aligned type, then
3138 it is likely that your program will also be doing pointer arithmetic (or
3139 subscripting, which amounts to the same thing) on pointers to the
3140 relevant type, and the code that the compiler generates for these
3141 pointer arithmetic operations will often be more efficient for
3142 efficiently-aligned types than for other types.
3144 The @code{aligned} attribute can only increase the alignment; but you
3145 can decrease it by specifying @code{packed} as well. See below.
3147 Note that the effectiveness of @code{aligned} attributes may be limited
3148 by inherent limitations in your linker. On many systems, the linker is
3149 only able to arrange for variables to be aligned up to a certain maximum
3150 alignment. (For some linkers, the maximum supported alignment may
3151 be very very small.) If your linker is only able to align variables
3152 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3153 in an @code{__attribute__} will still only provide you with 8 byte
3154 alignment. See your linker documentation for further information.
3157 This attribute, attached to an @code{enum}, @code{struct}, or
3158 @code{union} type definition, specified that the minimum required memory
3159 be used to represent the type.
3161 @opindex fshort-enums
3162 Specifying this attribute for @code{struct} and @code{union} types is
3163 equivalent to specifying the @code{packed} attribute on each of the
3164 structure or union members. Specifying the @option{-fshort-enums}
3165 flag on the line is equivalent to specifying the @code{packed}
3166 attribute on all @code{enum} definitions.
3168 You may only specify this attribute after a closing curly brace on an
3169 @code{enum} definition, not in a @code{typedef} declaration, unless that
3170 declaration also contains the definition of the @code{enum}.
3172 @item transparent_union
3173 This attribute, attached to a @code{union} type definition, indicates
3174 that any function parameter having that union type causes calls to that
3175 function to be treated in a special way.
3177 First, the argument corresponding to a transparent union type can be of
3178 any type in the union; no cast is required. Also, if the union contains
3179 a pointer type, the corresponding argument can be a null pointer
3180 constant or a void pointer expression; and if the union contains a void
3181 pointer type, the corresponding argument can be any pointer expression.
3182 If the union member type is a pointer, qualifiers like @code{const} on
3183 the referenced type must be respected, just as with normal pointer
3186 Second, the argument is passed to the function using the calling
3187 conventions of first member of the transparent union, not the calling
3188 conventions of the union itself. All members of the union must have the
3189 same machine representation; this is necessary for this argument passing
3192 Transparent unions are designed for library functions that have multiple
3193 interfaces for compatibility reasons. For example, suppose the
3194 @code{wait} function must accept either a value of type @code{int *} to
3195 comply with Posix, or a value of type @code{union wait *} to comply with
3196 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
3197 @code{wait} would accept both kinds of arguments, but it would also
3198 accept any other pointer type and this would make argument type checking
3199 less useful. Instead, @code{<sys/wait.h>} might define the interface
3207 @} wait_status_ptr_t __attribute__ ((__transparent_union__));
3209 pid_t wait (wait_status_ptr_t);
3212 This interface allows either @code{int *} or @code{union wait *}
3213 arguments to be passed, using the @code{int *} calling convention.
3214 The program can call @code{wait} with arguments of either type:
3217 int w1 () @{ int w; return wait (&w); @}
3218 int w2 () @{ union wait w; return wait (&w); @}
3221 With this interface, @code{wait}'s implementation might look like this:
3224 pid_t wait (wait_status_ptr_t p)
3226 return waitpid (-1, p.__ip, 0);
3231 When attached to a type (including a @code{union} or a @code{struct}),
3232 this attribute means that variables of that type are meant to appear
3233 possibly unused. GCC will not produce a warning for any variables of
3234 that type, even if the variable appears to do nothing. This is often
3235 the case with lock or thread classes, which are usually defined and then
3236 not referenced, but contain constructors and destructors that have
3237 nontrivial bookkeeping functions.
3241 To specify multiple attributes, separate them by commas within the
3242 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3246 @section An Inline Function is As Fast As a Macro
3247 @cindex inline functions
3248 @cindex integrating function code
3250 @cindex macros, inline alternative
3252 By declaring a function @code{inline}, you can direct GCC to
3253 integrate that function's code into the code for its callers. This
3254 makes execution faster by eliminating the function-call overhead; in
3255 addition, if any of the actual argument values are constant, their known
3256 values may permit simplifications at compile time so that not all of the
3257 inline function's code needs to be included. The effect on code size is
3258 less predictable; object code may be larger or smaller with function
3259 inlining, depending on the particular case. Inlining of functions is an
3260 optimization and it really ``works'' only in optimizing compilation. If
3261 you don't use @option{-O}, no function is really inline.
3263 Inline functions are included in the ISO C99 standard, but there are
3264 currently substantial differences between what GCC implements and what
3265 the ISO C99 standard requires.
3267 To declare a function inline, use the @code{inline} keyword in its
3268 declaration, like this:
3278 (If you are writing a header file to be included in ISO C programs, write
3279 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.)
3280 You can also make all ``simple enough'' functions inline with the option
3281 @option{-finline-functions}.
3284 Note that certain usages in a function definition can make it unsuitable
3285 for inline substitution. Among these usages are: use of varargs, use of
3286 alloca, use of variable sized data types (@pxref{Variable Length}),
3287 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
3288 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
3289 will warn when a function marked @code{inline} could not be substituted,
3290 and will give the reason for the failure.
3292 Note that in C and Objective-C, unlike C++, the @code{inline} keyword
3293 does not affect the linkage of the function.
3295 @cindex automatic @code{inline} for C++ member fns
3296 @cindex @code{inline} automatic for C++ member fns
3297 @cindex member fns, automatically @code{inline}
3298 @cindex C++ member fns, automatically @code{inline}
3299 @opindex fno-default-inline
3300 GCC automatically inlines member functions defined within the class
3301 body of C++ programs even if they are not explicitly declared
3302 @code{inline}. (You can override this with @option{-fno-default-inline};
3303 @pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.)
3305 @cindex inline functions, omission of
3306 @opindex fkeep-inline-functions
3307 When a function is both inline and @code{static}, if all calls to the
3308 function are integrated into the caller, and the function's address is
3309 never used, then the function's own assembler code is never referenced.
3310 In this case, GCC does not actually output assembler code for the
3311 function, unless you specify the option @option{-fkeep-inline-functions}.
3312 Some calls cannot be integrated for various reasons (in particular,
3313 calls that precede the function's definition cannot be integrated, and
3314 neither can recursive calls within the definition). If there is a
3315 nonintegrated call, then the function is compiled to assembler code as
3316 usual. The function must also be compiled as usual if the program
3317 refers to its address, because that can't be inlined.
3319 @cindex non-static inline function
3320 When an inline function is not @code{static}, then the compiler must assume
3321 that there may be calls from other source files; since a global symbol can
3322 be defined only once in any program, the function must not be defined in
3323 the other source files, so the calls therein cannot be integrated.
3324 Therefore, a non-@code{static} inline function is always compiled on its
3325 own in the usual fashion.
3327 If you specify both @code{inline} and @code{extern} in the function
3328 definition, then the definition is used only for inlining. In no case
3329 is the function compiled on its own, not even if you refer to its
3330 address explicitly. Such an address becomes an external reference, as
3331 if you had only declared the function, and had not defined it.
3333 This combination of @code{inline} and @code{extern} has almost the
3334 effect of a macro. The way to use it is to put a function definition in
3335 a header file with these keywords, and put another copy of the
3336 definition (lacking @code{inline} and @code{extern}) in a library file.
3337 The definition in the header file will cause most calls to the function
3338 to be inlined. If any uses of the function remain, they will refer to
3339 the single copy in the library.
3341 For future compatibility with when GCC implements ISO C99 semantics for
3342 inline functions, it is best to use @code{static inline} only. (The
3343 existing semantics will remain available when @option{-std=gnu89} is
3344 specified, but eventually the default will be @option{-std=gnu99} and
3345 that will implement the C99 semantics, though it does not do so yet.)
3347 GCC does not inline any functions when not optimizing. It is not
3348 clear whether it is better to inline or not, in this case, but we found
3349 that a correct implementation when not optimizing was difficult. So we
3350 did the easy thing, and turned it off.
3353 @section Assembler Instructions with C Expression Operands
3354 @cindex extended @code{asm}
3355 @cindex @code{asm} expressions
3356 @cindex assembler instructions
3359 In an assembler instruction using @code{asm}, you can specify the
3360 operands of the instruction using C expressions. This means you need not
3361 guess which registers or memory locations will contain the data you want
3364 You must specify an assembler instruction template much like what
3365 appears in a machine description, plus an operand constraint string for
3368 For example, here is how to use the 68881's @code{fsinx} instruction:
3371 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
3375 Here @code{angle} is the C expression for the input operand while
3376 @code{result} is that of the output operand. Each has @samp{"f"} as its
3377 operand constraint, saying that a floating point register is required.
3378 The @samp{=} in @samp{=f} indicates that the operand is an output; all
3379 output operands' constraints must use @samp{=}. The constraints use the
3380 same language used in the machine description (@pxref{Constraints}).
3382 Each operand is described by an operand-constraint string followed by
3383 the C expression in parentheses. A colon separates the assembler
3384 template from the first output operand and another separates the last
3385 output operand from the first input, if any. Commas separate the
3386 operands within each group. The total number of operands is currently
3387 limited to 30; this limitation may be lifted in some future version of
3390 If there are no output operands but there are input operands, you must
3391 place two consecutive colons surrounding the place where the output
3394 As of GCC version 3.1, it is also possible to specify input and output
3395 operands using symbolic names which can be referenced within the
3396 assembler code. These names are specified inside square brackets
3397 preceding the constraint string, and can be referenced inside the
3398 assembler code using @code{%[@var{name}]} instead of a percentage sign
3399 followed by the operand number. Using named operands the above example
3403 asm ("fsinx %[angle],%[output]"
3404 : [output] "=f" (result)
3405 : [angle] "f" (angle));
3409 Note that the symbolic operand names have no relation whatsoever to
3410 other C identifiers. You may use any name you like, even those of
3411 existing C symbols, but must ensure that no two operands within the same
3412 assembler construct use the same symbolic name.
3414 Output operand expressions must be lvalues; the compiler can check this.
3415 The input operands need not be lvalues. The compiler cannot check
3416 whether the operands have data types that are reasonable for the
3417 instruction being executed. It does not parse the assembler instruction
3418 template and does not know what it means or even whether it is valid
3419 assembler input. The extended @code{asm} feature is most often used for
3420 machine instructions the compiler itself does not know exist. If
3421 the output expression cannot be directly addressed (for example, it is a
3422 bit-field), your constraint must allow a register. In that case, GCC
3423 will use the register as the output of the @code{asm}, and then store
3424 that register into the output.
3426 The ordinary output operands must be write-only; GCC will assume that
3427 the values in these operands before the instruction are dead and need
3428 not be generated. Extended asm supports input-output or read-write
3429 operands. Use the constraint character @samp{+} to indicate such an
3430 operand and list it with the output operands.
3432 When the constraints for the read-write operand (or the operand in which
3433 only some of the bits are to be changed) allows a register, you may, as
3434 an alternative, logically split its function into two separate operands,
3435 one input operand and one write-only output operand. The connection
3436 between them is expressed by constraints which say they need to be in
3437 the same location when the instruction executes. You can use the same C
3438 expression for both operands, or different expressions. For example,
3439 here we write the (fictitious) @samp{combine} instruction with
3440 @code{bar} as its read-only source operand and @code{foo} as its
3441 read-write destination:
3444 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
3448 The constraint @samp{"0"} for operand 1 says that it must occupy the
3449 same location as operand 0. A number in constraint is allowed only in
3450 an input operand and it must refer to an output operand.
3452 Only a number in the constraint can guarantee that one operand will be in
3453 the same place as another. The mere fact that @code{foo} is the value
3454 of both operands is not enough to guarantee that they will be in the
3455 same place in the generated assembler code. The following would not
3459 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
3462 Various optimizations or reloading could cause operands 0 and 1 to be in
3463 different registers; GCC knows no reason not to do so. For example, the
3464 compiler might find a copy of the value of @code{foo} in one register and
3465 use it for operand 1, but generate the output operand 0 in a different
3466 register (copying it afterward to @code{foo}'s own address). Of course,
3467 since the register for operand 1 is not even mentioned in the assembler
3468 code, the result will not work, but GCC can't tell that.
3470 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
3471 the operand number for a matching constraint. For example:
3474 asm ("cmoveq %1,%2,%[result]"
3475 : [result] "=r"(result)
3476 : "r" (test), "r"(new), "[result]"(old));
3479 Some instructions clobber specific hard registers. To describe this,
3480 write a third colon after the input operands, followed by the names of
3481 the clobbered hard registers (given as strings). Here is a realistic
3482 example for the VAX:
3485 asm volatile ("movc3 %0,%1,%2"
3487 : "g" (from), "g" (to), "g" (count)
3488 : "r0", "r1", "r2", "r3", "r4", "r5");
3491 You may not write a clobber description in a way that overlaps with an
3492 input or output operand. For example, you may not have an operand
3493 describing a register class with one member if you mention that register
3494 in the clobber list. There is no way for you to specify that an input
3495 operand is modified without also specifying it as an output
3496 operand. Note that if all the output operands you specify are for this
3497 purpose (and hence unused), you will then also need to specify
3498 @code{volatile} for the @code{asm} construct, as described below, to
3499 prevent GCC from deleting the @code{asm} statement as unused.
3501 If you refer to a particular hardware register from the assembler code,
3502 you will probably have to list the register after the third colon to
3503 tell the compiler the register's value is modified. In some assemblers,
3504 the register names begin with @samp{%}; to produce one @samp{%} in the
3505 assembler code, you must write @samp{%%} in the input.
3507 If your assembler instruction can alter the condition code register, add
3508 @samp{cc} to the list of clobbered registers. GCC on some machines
3509 represents the condition codes as a specific hardware register;
3510 @samp{cc} serves to name this register. On other machines, the
3511 condition code is handled differently, and specifying @samp{cc} has no
3512 effect. But it is valid no matter what the machine.
3514 If your assembler instruction modifies memory in an unpredictable
3515 fashion, add @samp{memory} to the list of clobbered registers. This
3516 will cause GCC to not keep memory values cached in registers across
3517 the assembler instruction. You will also want to add the
3518 @code{volatile} keyword if the memory affected is not listed in the
3519 inputs or outputs of the @code{asm}, as the @samp{memory} clobber does
3520 not count as a side-effect of the @code{asm}.
3522 You can put multiple assembler instructions together in a single
3523 @code{asm} template, separated by the characters normally used in assembly
3524 code for the system. A combination that works in most places is a newline
3525 to break the line, plus a tab character to move to the instruction field
3526 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
3527 assembler allows semicolons as a line-breaking character. Note that some
3528 assembler dialects use semicolons to start a comment.
3529 The input operands are guaranteed not to use any of the clobbered
3530 registers, and neither will the output operands' addresses, so you can
3531 read and write the clobbered registers as many times as you like. Here
3532 is an example of multiple instructions in a template; it assumes the
3533 subroutine @code{_foo} accepts arguments in registers 9 and 10:
3536 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
3538 : "g" (from), "g" (to)
3542 Unless an output operand has the @samp{&} constraint modifier, GCC
3543 may allocate it in the same register as an unrelated input operand, on
3544 the assumption the inputs are consumed before the outputs are produced.
3545 This assumption may be false if the assembler code actually consists of
3546 more than one instruction. In such a case, use @samp{&} for each output
3547 operand that may not overlap an input. @xref{Modifiers}.
3549 If you want to test the condition code produced by an assembler
3550 instruction, you must include a branch and a label in the @code{asm}
3551 construct, as follows:
3554 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
3560 This assumes your assembler supports local labels, as the GNU assembler
3561 and most Unix assemblers do.
3563 Speaking of labels, jumps from one @code{asm} to another are not
3564 supported. The compiler's optimizers do not know about these jumps, and
3565 therefore they cannot take account of them when deciding how to
3568 @cindex macros containing @code{asm}
3569 Usually the most convenient way to use these @code{asm} instructions is to
3570 encapsulate them in macros that look like functions. For example,
3574 (@{ double __value, __arg = (x); \
3575 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
3580 Here the variable @code{__arg} is used to make sure that the instruction
3581 operates on a proper @code{double} value, and to accept only those
3582 arguments @code{x} which can convert automatically to a @code{double}.
3584 Another way to make sure the instruction operates on the correct data
3585 type is to use a cast in the @code{asm}. This is different from using a
3586 variable @code{__arg} in that it converts more different types. For
3587 example, if the desired type were @code{int}, casting the argument to
3588 @code{int} would accept a pointer with no complaint, while assigning the
3589 argument to an @code{int} variable named @code{__arg} would warn about
3590 using a pointer unless the caller explicitly casts it.
3592 If an @code{asm} has output operands, GCC assumes for optimization
3593 purposes the instruction has no side effects except to change the output
3594 operands. This does not mean instructions with a side effect cannot be
3595 used, but you must be careful, because the compiler may eliminate them
3596 if the output operands aren't used, or move them out of loops, or
3597 replace two with one if they constitute a common subexpression. Also,
3598 if your instruction does have a side effect on a variable that otherwise
3599 appears not to change, the old value of the variable may be reused later
3600 if it happens to be found in a register.
3602 You can prevent an @code{asm} instruction from being deleted, moved
3603 significantly, or combined, by writing the keyword @code{volatile} after
3604 the @code{asm}. For example:
3607 #define get_and_set_priority(new) \
3609 asm volatile ("get_and_set_priority %0, %1" \
3610 : "=g" (__old) : "g" (new)); \
3615 If you write an @code{asm} instruction with no outputs, GCC will know
3616 the instruction has side-effects and will not delete the instruction or
3617 move it outside of loops.
3619 The @code{volatile} keyword indicates that the instruction has
3620 important side-effects. GCC will not delete a volatile @code{asm} if
3621 it is reachable. (The instruction can still be deleted if GCC can
3622 prove that control-flow will never reach the location of the
3623 instruction.) In addition, GCC will not reschedule instructions
3624 across a volatile @code{asm} instruction. For example:
3627 *(volatile int *)addr = foo;
3628 asm volatile ("eieio" : : );
3632 Assume @code{addr} contains the address of a memory mapped device
3633 register. The PowerPC @code{eieio} instruction (Enforce In-order
3634 Execution of I/O) tells the CPU to make sure that the store to that
3635 device register happens before it issues any other I/O@.
3637 Note that even a volatile @code{asm} instruction can be moved in ways
3638 that appear insignificant to the compiler, such as across jump
3639 instructions. You can't expect a sequence of volatile @code{asm}
3640 instructions to remain perfectly consecutive. If you want consecutive
3641 output, use a single @code{asm}. Also, GCC will perform some
3642 optimizations across a volatile @code{asm} instruction; GCC does not
3643 ``forget everything'' when it encounters a volatile @code{asm}
3644 instruction the way some other compilers do.
3646 An @code{asm} instruction without any operands or clobbers (an ``old
3647 style'' @code{asm}) will be treated identically to a volatile
3648 @code{asm} instruction.
3650 It is a natural idea to look for a way to give access to the condition
3651 code left by the assembler instruction. However, when we attempted to
3652 implement this, we found no way to make it work reliably. The problem
3653 is that output operands might need reloading, which would result in
3654 additional following ``store'' instructions. On most machines, these
3655 instructions would alter the condition code before there was time to
3656 test it. This problem doesn't arise for ordinary ``test'' and
3657 ``compare'' instructions because they don't have any output operands.
3659 For reasons similar to those described above, it is not possible to give
3660 an assembler instruction access to the condition code left by previous
3663 If you are writing a header file that should be includable in ISO C
3664 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
3667 @subsection i386 floating point asm operands
3669 There are several rules on the usage of stack-like regs in
3670 asm_operands insns. These rules apply only to the operands that are
3675 Given a set of input regs that die in an asm_operands, it is
3676 necessary to know which are implicitly popped by the asm, and
3677 which must be explicitly popped by gcc.
3679 An input reg that is implicitly popped by the asm must be
3680 explicitly clobbered, unless it is constrained to match an
3684 For any input reg that is implicitly popped by an asm, it is
3685 necessary to know how to adjust the stack to compensate for the pop.
3686 If any non-popped input is closer to the top of the reg-stack than
3687 the implicitly popped reg, it would not be possible to know what the
3688 stack looked like---it's not clear how the rest of the stack ``slides
3691 All implicitly popped input regs must be closer to the top of
3692 the reg-stack than any input that is not implicitly popped.
3694 It is possible that if an input dies in an insn, reload might
3695 use the input reg for an output reload. Consider this example:
3698 asm ("foo" : "=t" (a) : "f" (b));
3701 This asm says that input B is not popped by the asm, and that
3702 the asm pushes a result onto the reg-stack, i.e., the stack is one
3703 deeper after the asm than it was before. But, it is possible that
3704 reload will think that it can use the same reg for both the input and
3705 the output, if input B dies in this insn.
3707 If any input operand uses the @code{f} constraint, all output reg
3708 constraints must use the @code{&} earlyclobber.
3710 The asm above would be written as
3713 asm ("foo" : "=&t" (a) : "f" (b));
3717 Some operands need to be in particular places on the stack. All
3718 output operands fall in this category---there is no other way to
3719 know which regs the outputs appear in unless the user indicates
3720 this in the constraints.
3722 Output operands must specifically indicate which reg an output
3723 appears in after an asm. @code{=f} is not allowed: the operand
3724 constraints must select a class with a single reg.
3727 Output operands may not be ``inserted'' between existing stack regs.
3728 Since no 387 opcode uses a read/write operand, all output operands
3729 are dead before the asm_operands, and are pushed by the asm_operands.
3730 It makes no sense to push anywhere but the top of the reg-stack.
3732 Output operands must start at the top of the reg-stack: output
3733 operands may not ``skip'' a reg.
3736 Some asm statements may need extra stack space for internal
3737 calculations. This can be guaranteed by clobbering stack registers
3738 unrelated to the inputs and outputs.
3742 Here are a couple of reasonable asms to want to write. This asm
3743 takes one input, which is internally popped, and produces two outputs.
3746 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
3749 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
3750 and replaces them with one output. The user must code the @code{st(1)}
3751 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
3754 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
3758 @c Show the details on constraints if they do not appear elsewhere in
3764 @section Controlling Names Used in Assembler Code
3765 @cindex assembler names for identifiers
3766 @cindex names used in assembler code
3767 @cindex identifiers, names in assembler code
3769 You can specify the name to be used in the assembler code for a C
3770 function or variable by writing the @code{asm} (or @code{__asm__})
3771 keyword after the declarator as follows:
3774 int foo asm ("myfoo") = 2;
3778 This specifies that the name to be used for the variable @code{foo} in
3779 the assembler code should be @samp{myfoo} rather than the usual
3782 On systems where an underscore is normally prepended to the name of a C
3783 function or variable, this feature allows you to define names for the
3784 linker that do not start with an underscore.
3786 It does not make sense to use this feature with a non-static local
3787 variable since such variables do not have assembler names. If you are
3788 trying to put the variable in a particular register, see @ref{Explicit
3789 Reg Vars}. GCC presently accepts such code with a warning, but will
3790 probably be changed to issue an error, rather than a warning, in the
3793 You cannot use @code{asm} in this way in a function @emph{definition}; but
3794 you can get the same effect by writing a declaration for the function
3795 before its definition and putting @code{asm} there, like this:
3798 extern func () asm ("FUNC");
3805 It is up to you to make sure that the assembler names you choose do not
3806 conflict with any other assembler symbols. Also, you must not use a
3807 register name; that would produce completely invalid assembler code. GCC
3808 does not as yet have the ability to store static variables in registers.
3809 Perhaps that will be added.
3811 @node Explicit Reg Vars
3812 @section Variables in Specified Registers
3813 @cindex explicit register variables
3814 @cindex variables in specified registers
3815 @cindex specified registers
3816 @cindex registers, global allocation
3818 GNU C allows you to put a few global variables into specified hardware
3819 registers. You can also specify the register in which an ordinary
3820 register variable should be allocated.
3824 Global register variables reserve registers throughout the program.
3825 This may be useful in programs such as programming language
3826 interpreters which have a couple of global variables that are accessed
3830 Local register variables in specific registers do not reserve the
3831 registers. The compiler's data flow analysis is capable of determining
3832 where the specified registers contain live values, and where they are
3833 available for other uses. Stores into local register variables may be deleted
3834 when they appear to be dead according to dataflow analysis. References
3835 to local register variables may be deleted or moved or simplified.
3837 These local variables are sometimes convenient for use with the extended
3838 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
3839 output of the assembler instruction directly into a particular register.
3840 (This will work provided the register you specify fits the constraints
3841 specified for that operand in the @code{asm}.)
3849 @node Global Reg Vars
3850 @subsection Defining Global Register Variables
3851 @cindex global register variables
3852 @cindex registers, global variables in
3854 You can define a global register variable in GNU C like this:
3857 register int *foo asm ("a5");
3861 Here @code{a5} is the name of the register which should be used. Choose a
3862 register which is normally saved and restored by function calls on your
3863 machine, so that library routines will not clobber it.
3865 Naturally the register name is cpu-dependent, so you would need to
3866 conditionalize your program according to cpu type. The register
3867 @code{a5} would be a good choice on a 68000 for a variable of pointer
3868 type. On machines with register windows, be sure to choose a ``global''
3869 register that is not affected magically by the function call mechanism.
3871 In addition, operating systems on one type of cpu may differ in how they
3872 name the registers; then you would need additional conditionals. For
3873 example, some 68000 operating systems call this register @code{%a5}.
3875 Eventually there may be a way of asking the compiler to choose a register
3876 automatically, but first we need to figure out how it should choose and
3877 how to enable you to guide the choice. No solution is evident.
3879 Defining a global register variable in a certain register reserves that
3880 register entirely for this use, at least within the current compilation.
3881 The register will not be allocated for any other purpose in the functions
3882 in the current compilation. The register will not be saved and restored by
3883 these functions. Stores into this register are never deleted even if they
3884 would appear to be dead, but references may be deleted or moved or
3887 It is not safe to access the global register variables from signal
3888 handlers, or from more than one thread of control, because the system
3889 library routines may temporarily use the register for other things (unless
3890 you recompile them specially for the task at hand).
3892 @cindex @code{qsort}, and global register variables
3893 It is not safe for one function that uses a global register variable to
3894 call another such function @code{foo} by way of a third function
3895 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
3896 different source file in which the variable wasn't declared). This is
3897 because @code{lose} might save the register and put some other value there.
3898 For example, you can't expect a global register variable to be available in
3899 the comparison-function that you pass to @code{qsort}, since @code{qsort}
3900 might have put something else in that register. (If you are prepared to
3901 recompile @code{qsort} with the same global register variable, you can
3902 solve this problem.)
3904 If you want to recompile @code{qsort} or other source files which do not
3905 actually use your global register variable, so that they will not use that
3906 register for any other purpose, then it suffices to specify the compiler
3907 option @option{-ffixed-@var{reg}}. You need not actually add a global
3908 register declaration to their source code.
3910 A function which can alter the value of a global register variable cannot
3911 safely be called from a function compiled without this variable, because it
3912 could clobber the value the caller expects to find there on return.
3913 Therefore, the function which is the entry point into the part of the
3914 program that uses the global register variable must explicitly save and
3915 restore the value which belongs to its caller.
3917 @cindex register variable after @code{longjmp}
3918 @cindex global register after @code{longjmp}
3919 @cindex value after @code{longjmp}
3922 On most machines, @code{longjmp} will restore to each global register
3923 variable the value it had at the time of the @code{setjmp}. On some
3924 machines, however, @code{longjmp} will not change the value of global
3925 register variables. To be portable, the function that called @code{setjmp}
3926 should make other arrangements to save the values of the global register
3927 variables, and to restore them in a @code{longjmp}. This way, the same
3928 thing will happen regardless of what @code{longjmp} does.
3930 All global register variable declarations must precede all function
3931 definitions. If such a declaration could appear after function
3932 definitions, the declaration would be too late to prevent the register from
3933 being used for other purposes in the preceding functions.
3935 Global register variables may not have initial values, because an
3936 executable file has no means to supply initial contents for a register.
3938 On the Sparc, there are reports that g3 @dots{} g7 are suitable
3939 registers, but certain library functions, such as @code{getwd}, as well
3940 as the subroutines for division and remainder, modify g3 and g4. g1 and
3941 g2 are local temporaries.
3943 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
3944 Of course, it will not do to use more than a few of those.
3946 @node Local Reg Vars
3947 @subsection Specifying Registers for Local Variables
3948 @cindex local variables, specifying registers
3949 @cindex specifying registers for local variables
3950 @cindex registers for local variables
3952 You can define a local register variable with a specified register
3956 register int *foo asm ("a5");
3960 Here @code{a5} is the name of the register which should be used. Note
3961 that this is the same syntax used for defining global register
3962 variables, but for a local variable it would appear within a function.
3964 Naturally the register name is cpu-dependent, but this is not a
3965 problem, since specific registers are most often useful with explicit
3966 assembler instructions (@pxref{Extended Asm}). Both of these things
3967 generally require that you conditionalize your program according to
3970 In addition, operating systems on one type of cpu may differ in how they
3971 name the registers; then you would need additional conditionals. For
3972 example, some 68000 operating systems call this register @code{%a5}.
3974 Defining such a register variable does not reserve the register; it
3975 remains available for other uses in places where flow control determines
3976 the variable's value is not live. However, these registers are made
3977 unavailable for use in the reload pass; excessive use of this feature
3978 leaves the compiler too few available registers to compile certain
3981 This option does not guarantee that GCC will generate code that has
3982 this variable in the register you specify at all times. You may not
3983 code an explicit reference to this register in an @code{asm} statement
3984 and assume it will always refer to this variable.
3986 Stores into local register variables may be deleted when they appear to be dead
3987 according to dataflow analysis. References to local register variables may
3988 be deleted or moved or simplified.
3990 @node Alternate Keywords
3991 @section Alternate Keywords
3992 @cindex alternate keywords
3993 @cindex keywords, alternate
3995 The option @option{-traditional} disables certain keywords;
3996 @option{-ansi} and the various @option{-std} options disable certain
3997 others. This causes trouble when you want to use GNU C extensions, or
3998 ISO C features, in a general-purpose header file that should be usable
3999 by all programs, including ISO C programs and traditional ones. The
4000 keywords @code{asm}, @code{typeof} and @code{inline} cannot be used
4001 since they won't work in a program compiled with @option{-ansi}
4002 (although @code{inline} can be used in a program compiled with
4003 @option{-std=c99}), while the keywords @code{const}, @code{volatile},
4004 @code{signed}, @code{typeof} and @code{inline} won't work in a program
4005 compiled with @option{-traditional}. The ISO C99 keyword
4006 @code{restrict} is only available when @option{-std=gnu99} (which will
4007 eventually be the default) or @option{-std=c99} (or the equivalent
4008 @option{-std=iso9899:1999}) is used.
4010 The way to solve these problems is to put @samp{__} at the beginning and
4011 end of each problematical keyword. For example, use @code{__asm__}
4012 instead of @code{asm}, @code{__const__} instead of @code{const}, and
4013 @code{__inline__} instead of @code{inline}.
4015 Other C compilers won't accept these alternative keywords; if you want to
4016 compile with another compiler, you can define the alternate keywords as
4017 macros to replace them with the customary keywords. It looks like this:
4025 @findex __extension__
4027 @option{-pedantic} and other options cause warnings for many GNU C extensions.
4029 prevent such warnings within one expression by writing
4030 @code{__extension__} before the expression. @code{__extension__} has no
4031 effect aside from this.
4033 @node Incomplete Enums
4034 @section Incomplete @code{enum} Types
4036 You can define an @code{enum} tag without specifying its possible values.
4037 This results in an incomplete type, much like what you get if you write
4038 @code{struct foo} without describing the elements. A later declaration
4039 which does specify the possible values completes the type.
4041 You can't allocate variables or storage using the type while it is
4042 incomplete. However, you can work with pointers to that type.
4044 This extension may not be very useful, but it makes the handling of
4045 @code{enum} more consistent with the way @code{struct} and @code{union}
4048 This extension is not supported by GNU C++.
4050 @node Function Names
4051 @section Function Names as Strings
4052 @cindex @code{__FUNCTION__} identifier
4053 @cindex @code{__PRETTY_FUNCTION__} identifier
4054 @cindex @code{__func__} identifier
4056 GCC predefines two magic identifiers to hold the name of the current
4057 function. The identifier @code{__FUNCTION__} holds the name of the function
4058 as it appears in the source. The identifier @code{__PRETTY_FUNCTION__}
4059 holds the name of the function pretty printed in a language specific
4062 These names are always the same in a C function, but in a C++ function
4063 they may be different. For example, this program:
4067 extern int printf (char *, ...);
4074 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
4075 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
4093 __PRETTY_FUNCTION__ = int a::sub (int)
4096 The compiler automagically replaces the identifiers with a string
4097 literal containing the appropriate name. Thus, they are neither
4098 preprocessor macros, like @code{__FILE__} and @code{__LINE__}, nor
4099 variables. This means that they catenate with other string literals, and
4100 that they can be used to initialize char arrays. For example
4103 char here[] = "Function " __FUNCTION__ " in " __FILE__;
4106 On the other hand, @samp{#ifdef __FUNCTION__} does not have any special
4107 meaning inside a function, since the preprocessor does not do anything
4108 special with the identifier @code{__FUNCTION__}.
4110 GCC also supports the magic word @code{__func__}, defined by the
4114 The identifier @code{__func__} is implicitly declared by the translator
4115 as if, immediately following the opening brace of each function
4116 definition, the declaration
4119 static const char __func__[] = "function-name";
4122 appeared, where function-name is the name of the lexically-enclosing
4123 function. This name is the unadorned name of the function.
4126 By this definition, @code{__func__} is a variable, not a string literal.
4127 In particular, @code{__func__} does not catenate with other string
4130 In @code{C++}, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} are
4131 variables, declared in the same way as @code{__func__}.
4133 @node Return Address
4134 @section Getting the Return or Frame Address of a Function
4136 These functions may be used to get information about the callers of a
4139 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
4140 This function returns the return address of the current function, or of
4141 one of its callers. The @var{level} argument is number of frames to
4142 scan up the call stack. A value of @code{0} yields the return address
4143 of the current function, a value of @code{1} yields the return address
4144 of the caller of the current function, and so forth.
4146 The @var{level} argument must be a constant integer.
4148 On some machines it may be impossible to determine the return address of
4149 any function other than the current one; in such cases, or when the top
4150 of the stack has been reached, this function will return @code{0} or a
4151 random value. In addition, @code{__builtin_frame_address} may be used
4152 to determine if the top of the stack has been reached.
4154 This function should only be used with a nonzero argument for debugging
4158 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
4159 This function is similar to @code{__builtin_return_address}, but it
4160 returns the address of the function frame rather than the return address
4161 of the function. Calling @code{__builtin_frame_address} with a value of
4162 @code{0} yields the frame address of the current function, a value of
4163 @code{1} yields the frame address of the caller of the current function,
4166 The frame is the area on the stack which holds local variables and saved
4167 registers. The frame address is normally the address of the first word
4168 pushed on to the stack by the function. However, the exact definition
4169 depends upon the processor and the calling convention. If the processor
4170 has a dedicated frame pointer register, and the function has a frame,
4171 then @code{__builtin_frame_address} will return the value of the frame
4174 On some machines it may be impossible to determine the frame address of
4175 any function other than the current one; in such cases, or when the top
4176 of the stack has been reached, this function will return @code{0} if
4177 the first frame pointer is properly initialized by the startup code.
4179 This function should only be used with a nonzero argument for debugging
4183 @node Vector Extensions
4184 @section Using vector instructions through built-in functions
4186 On some targets, the instruction set contains SIMD vector instructions that
4187 operate on multiple values contained in one large register at the same time.
4188 For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used
4191 The first step in using these extensions is to provide the necessary data
4192 types. This should be done using an appropriate @code{typedef}:
4195 typedef int v4si __attribute__ ((mode(V4SI)));
4198 The base type @code{int} is effectively ignored by the compiler, the
4199 actual properties of the new type @code{v4si} are defined by the
4200 @code{__attribute__}. It defines the machine mode to be used; for vector
4201 types these have the form @code{V@var{n}@var{B}}; @var{n} should be the
4202 number of elements in the vector, and @var{B} should be the base mode of the
4203 individual elements. The following can be used as base modes:
4207 An integer that is as wide as the smallest addressable unit, usually 8 bits.
4209 An integer, twice as wide as a QI mode integer, usually 16 bits.
4211 An integer, four times as wide as a QI mode integer, usually 32 bits.
4213 An integer, eight times as wide as a QI mode integer, usually 64 bits.
4215 A floating point value, as wide as a SI mode integer, usually 32 bits.
4217 A floating point value, as wide as a DI mode integer, usually 64 bits.
4220 Not all base types or combinations are always valid; which modes can be used
4221 is determined by the target machine. For example, if targetting the i386 MMX
4222 extensions, only @code{V8QI}, @code{V4HI} and @code{V2SI} are allowed modes.
4224 There are no @code{V1xx} vector modes - they would be identical to the
4225 corresponding base mode.
4227 There is no distinction between signed and unsigned vector modes. This
4228 distinction is made by the operations that perform on the vectors, not
4231 The types defined in this manner are somewhat special, they cannot be
4232 used with most normal C operations (i.e., a vector addition can @emph{not}
4233 be represented by a normal addition of two vector type variables). You
4234 can declare only variables and use them in function calls and returns, as
4235 well as in assignments and some casts. It is possible to cast from one
4236 vector type to another, provided they are of the same size (in fact, you
4237 can also cast vectors to and from other datatypes of the same size).
4239 A port that supports vector operations provides a set of built-in functions
4240 that can be used to operate on vectors. For example, a function to add two
4241 vectors and multiply the result by a third could look like this:
4244 v4si f (v4si a, v4si b, v4si c)
4246 v4si tmp = __builtin_addv4si (a, b);
4247 return __builtin_mulv4si (tmp, c);
4252 @node Other Builtins
4253 @section Other built-in functions provided by GCC
4254 @cindex built-in functions
4255 @findex __builtin_isgreater
4256 @findex __builtin_isgreaterequal
4257 @findex __builtin_isless
4258 @findex __builtin_islessequal
4259 @findex __builtin_islessgreater
4260 @findex __builtin_isunordered
4316 GCC provides a large number of built-in functions other than the ones
4317 mentioned above. Some of these are for internal use in the processing
4318 of exceptions or variable-length argument lists and will not be
4319 documented here because they may change from time to time; we do not
4320 recommend general use of these functions.
4322 The remaining functions are provided for optimization purposes.
4324 @opindex fno-builtin
4325 GCC includes built-in versions of many of the functions in the standard
4326 C library. The versions prefixed with @code{__builtin_} will always be
4327 treated as having the same meaning as the C library function even if you
4328 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
4329 Many of these functions are only optimized in certain cases; if they are
4330 not optimized in a particular case, a call to the library function will
4335 The functions @code{abort}, @code{exit}, @code{_Exit} and @code{_exit}
4336 are recognized and presumed not to return, but otherwise are not built
4337 in. @code{_exit} is not recognized in strict ISO C mode (@option{-ansi},
4338 @option{-std=c89} or @option{-std=c99}). @code{_Exit} is not recognized in
4339 strict C89 mode (@option{-ansi} or @option{-std=c89}).
4341 Outside strict ISO C mode, the functions @code{alloca}, @code{bcmp},
4342 @code{bzero}, @code{index}, @code{rindex} and @code{ffs} may be handled
4343 as built-in functions. All these functions have corresponding versions
4344 prefixed with @code{__builtin_}, which may be used even in strict C89
4347 The ISO C99 functions @code{conj}, @code{conjf}, @code{conjl},
4348 @code{creal}, @code{crealf}, @code{creall}, @code{cimag}, @code{cimagf},
4349 @code{cimagl}, @code{llabs} and @code{imaxabs} are handled as built-in
4350 functions except in strict ISO C89 mode. There are also built-in
4351 versions of the ISO C99 functions @code{cosf}, @code{cosl},
4352 @code{fabsf}, @code{fabsl}, @code{sinf}, @code{sinl}, @code{sqrtf}, and
4353 @code{sqrtl}, that are recognized in any mode since ISO C89 reserves
4354 these names for the purpose to which ISO C99 puts them. All these
4355 functions have corresponding versions prefixed with @code{__builtin_}.
4357 The ISO C89 functions @code{abs}, @code{cos}, @code{fabs},
4358 @code{fprintf}, @code{fputs}, @code{labs}, @code{memcmp}, @code{memcpy},
4359 @code{memset}, @code{printf}, @code{sin}, @code{sqrt}, @code{strcat},
4360 @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn},
4361 @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy},
4362 @code{strpbrk}, @code{strrchr}, @code{strspn}, and @code{strstr} are all
4363 recognized as built-in functions unless @option{-fno-builtin} is
4364 specified (or @option{-fno-builtin-@var{function}} is specified for an
4365 individual function). All of these functions have
4366 corresponding versions prefixed
4367 with @code{__builtin_}, except that the version for @code{sqrt} is
4368 called @code{__builtin_fsqrt}.
4370 GCC provides built-in versions of the ISO C99 floating point comparison
4371 macros that avoid raising exceptions for unordered operands. They have
4372 the same names as the standard macros ( @code{isgreater},
4373 @code{isgreaterequal}, @code{isless}, @code{islessequal},
4374 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
4375 prefixed. We intend for a library implementor to be able to simply
4376 @code{#define} each standard macro to its built-in equivalent.
4378 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
4379 You can use the built-in function @code{__builtin_constant_p} to
4380 determine if a value is known to be constant at compile-time and hence
4381 that GCC can perform constant-folding on expressions involving that
4382 value. The argument of the function is the value to test. The function
4383 returns the integer 1 if the argument is known to be a compile-time
4384 constant and 0 if it is not known to be a compile-time constant. A
4385 return of 0 does not indicate that the value is @emph{not} a constant,
4386 but merely that GCC cannot prove it is a constant with the specified
4387 value of the @option{-O} option.
4389 You would typically use this function in an embedded application where
4390 memory was a critical resource. If you have some complex calculation,
4391 you may want it to be folded if it involves constants, but need to call
4392 a function if it does not. For example:
4395 #define Scale_Value(X) \
4396 (__builtin_constant_p (X) \
4397 ? ((X) * SCALE + OFFSET) : Scale (X))
4400 You may use this built-in function in either a macro or an inline
4401 function. However, if you use it in an inlined function and pass an
4402 argument of the function as the argument to the built-in, GCC will
4403 never return 1 when you call the inline function with a string constant
4404 or compound literal (@pxref{Compound Literals}) and will not return 1
4405 when you pass a constant numeric value to the inline function unless you
4406 specify the @option{-O} option.
4408 You may also use @code{__builtin_constant_p} in initializers for static
4409 data. For instance, you can write
4412 static const int table[] = @{
4413 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
4419 This is an acceptable initializer even if @var{EXPRESSION} is not a
4420 constant expression. GCC must be more conservative about evaluating the
4421 built-in in this case, because it has no opportunity to perform
4424 Previous versions of GCC did not accept this built-in in data
4425 initializers. The earliest version where it is completely safe is
4429 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
4430 @opindex fprofile-arcs
4431 You may use @code{__builtin_expect} to provide the compiler with
4432 branch prediction information. In general, you should prefer to
4433 use actual profile feedback for this (@option{-fprofile-arcs}), as
4434 programmers are notoriously bad at predicting how their programs
4435 actually perform. However, there are applications in which this
4436 data is hard to collect.
4438 The return value is the value of @var{exp}, which should be an
4439 integral expression. The value of @var{c} must be a compile-time
4440 constant. The semantics of the built-in are that it is expected
4441 that @var{exp} == @var{c}. For example:
4444 if (__builtin_expect (x, 0))
4449 would indicate that we do not expect to call @code{foo}, since
4450 we expect @code{x} to be zero. Since you are limited to integral
4451 expressions for @var{exp}, you should use constructions such as
4454 if (__builtin_expect (ptr != NULL, 1))
4459 when testing pointer or floating-point values.
4462 @deftypefn {Built-in Function} void __builtin_prefetch (void *@var{addr}, int @var{rw}, int @var{locality})
4463 This function is used to minimize cache-miss latency by moving data into
4464 a cache before it is accessed.
4465 You can insert calls to @code{__builtin_prefetch} into code for which
4466 you know addresses of data in memory that is likely to be accessed soon.
4467 If the target supports them, data prefetch instructions will be generated.
4468 If the prefetch is done early enough before the access then the data will
4469 be in the cache by the time it is accessed.
4471 The value of @var{addr} is the address of the memory to prefetch.
4472 The value of @var{rw} is a compile-time constant one or zero; one
4473 means that the prefetch is preparing for a write to the memory address.
4474 The value @var{locality} must be a compile-time constant integer between
4475 zero and three. A value of zero means that the data has no temporal
4476 locality, so it need not be left in the cache after the access. A value
4477 of three means that the data has a high degree of temporal locality and
4478 should be left in all levels of cache possible. Values of one and two
4479 mean, respectively, a low or moderate degree of temporal locality.
4482 for (i = 0; i < n; i++)
4485 __builtin_prefetch (&a[i+j], 1, 1);
4486 __builtin_prefetch (&b[i+j], 0, 1);
4491 Data prefetch does not generate faults if @var{addr} is invalid, but
4492 the address expression itself must be valid. For example, a prefetch
4493 of @code{p->next} will not fault if @code{p->next} is not a valid
4494 address, but evaluation will fault if @code{p} is not a valid address.
4496 If the target does not support data prefetch, the address expression
4497 is evaluated if it includes side effects but no other code is generated
4498 and GCC does not issue a warning.
4502 @section Pragmas Accepted by GCC
4506 GCC supports several types of pragmas, primarily in order to compile
4507 code originally written for other compilers. Note that in general
4508 we do not recommend the use of pragmas; @xref{Function Attributes},
4509 for further explanation.
4517 @subsection ARM Pragmas
4519 The ARM target defines pragmas for controlling the default addition of
4520 @code{long_call} and @code{short_call} attributes to functions.
4521 @xref{Function Attributes}, for information about the effects of these
4526 @cindex pragma, long_calls
4527 Set all subsequent functions to have the @code{long_call} attribute.
4530 @cindex pragma, no_long_calls
4531 Set all subsequent functions to have the @code{short_call} attribute.
4533 @item long_calls_off
4534 @cindex pragma, long_calls_off
4535 Do not affect the @code{long_call} or @code{short_call} attributes of
4536 subsequent functions.
4539 @c Describe c4x pragmas here.
4540 @c Describe h8300 pragmas here.
4541 @c Describe i370 pragmas here.
4542 @c Describe i960 pragmas here.
4543 @c Describe sh pragmas here.
4544 @c Describe v850 pragmas here.
4546 @node Darwin Pragmas
4547 @subsection Darwin Pragmas
4549 The following pragmas are available for all architectures running the
4550 Darwin operating system. These are useful for compatibility with other
4554 @item mark @var{tokens}@dots{}
4555 @cindex pragma, mark
4556 This pragma is accepted, but has no effect.
4558 @item options align=@var{alignment}
4559 @cindex pragma, options align
4560 This pragma sets the alignment of fields in structures. The values of
4561 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
4562 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
4563 properly; to restore the previous setting, use @code{reset} for the
4566 @item segment @var{tokens}@dots{}
4567 @cindex pragma, segment
4568 This pragma is accepted, but has no effect.
4570 @item unused (@var{var} [, @var{var}]@dots{})
4571 @cindex pragma, unused
4572 This pragma declares variables to be possibly unused. GCC will not
4573 produce warnings for the listed variables. The effect is similar to
4574 that of the @code{unused} attribute, except that this pragma may appear
4575 anywhere within the variables' scopes.
4578 @node Unnamed Fields
4579 @section Unnamed struct/union fields within structs/unions.
4583 For compatibility with other compilers, GCC allows you to define
4584 a structure or union that contains, as fields, structures and unions
4585 without names. For example:
4598 In this example, the user would be able to access members of the unnamed
4599 union with code like @samp{foo.b}. Note that only unnamed structs and
4600 unions are allowed, you may not have, for example, an unnamed
4603 You must never create such structures that cause ambiguous field definitions.
4604 For example, this structure:
4615 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
4616 Such constructs are not supported and must be avoided. In the future,
4617 such constructs may be detected and treated as compilation errors.
4619 @node C++ Extensions
4620 @chapter Extensions to the C++ Language
4621 @cindex extensions, C++ language
4622 @cindex C++ language extensions
4624 The GNU compiler provides these extensions to the C++ language (and you
4625 can also use most of the C language extensions in your C++ programs). If you
4626 want to write code that checks whether these features are available, you can
4627 test for the GNU compiler the same way as for C programs: check for a
4628 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
4629 test specifically for GNU C++ (@pxref{Standard Predefined,,Standard
4630 Predefined Macros,cpp.info,The C Preprocessor}).
4633 * Min and Max:: C++ Minimum and maximum operators.
4634 * Volatiles:: What constitutes an access to a volatile object.
4635 * Restricted Pointers:: C99 restricted pointers and references.
4636 * Vague Linkage:: Where G++ puts inlines, vtables and such.
4637 * C++ Interface:: You can use a single C++ header file for both
4638 declarations and definitions.
4639 * Template Instantiation:: Methods for ensuring that exactly one copy of
4640 each needed template instantiation is emitted.
4641 * Bound member functions:: You can extract a function pointer to the
4642 method denoted by a @samp{->*} or @samp{.*} expression.
4643 * C++ Attributes:: Variable, function, and type attributes for C++ only.
4644 * Java Exceptions:: Tweaking exception handling to work with Java.
4645 * Deprecated Features:: Things might disappear from g++.
4646 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
4650 @section Minimum and Maximum Operators in C++
4652 It is very convenient to have operators which return the ``minimum'' or the
4653 ``maximum'' of two arguments. In GNU C++ (but not in GNU C),
4656 @item @var{a} <? @var{b}
4658 @cindex minimum operator
4659 is the @dfn{minimum}, returning the smaller of the numeric values
4660 @var{a} and @var{b};
4662 @item @var{a} >? @var{b}
4664 @cindex maximum operator
4665 is the @dfn{maximum}, returning the larger of the numeric values @var{a}
4669 These operations are not primitive in ordinary C++, since you can
4670 use a macro to return the minimum of two things in C++, as in the
4674 #define MIN(X,Y) ((X) < (Y) ? : (X) : (Y))
4678 You might then use @w{@samp{int min = MIN (i, j);}} to set @var{min} to
4679 the minimum value of variables @var{i} and @var{j}.
4681 However, side effects in @code{X} or @code{Y} may cause unintended
4682 behavior. For example, @code{MIN (i++, j++)} will fail, incrementing
4683 the smaller counter twice. A GNU C extension allows you to write safe
4684 macros that avoid this kind of problem (@pxref{Naming Types,,Naming an
4685 Expression's Type}). However, writing @code{MIN} and @code{MAX} as
4686 macros also forces you to use function-call notation for a
4687 fundamental arithmetic operation. Using GNU C++ extensions, you can
4688 write @w{@samp{int min = i <? j;}} instead.
4690 Since @code{<?} and @code{>?} are built into the compiler, they properly
4691 handle expressions with side-effects; @w{@samp{int min = i++ <? j++;}}
4695 @section When is a Volatile Object Accessed?
4696 @cindex accessing volatiles
4697 @cindex volatile read
4698 @cindex volatile write
4699 @cindex volatile access
4701 Both the C and C++ standard have the concept of volatile objects. These
4702 are normally accessed by pointers and used for accessing hardware. The
4703 standards encourage compilers to refrain from optimizations
4704 concerning accesses to volatile objects that it might perform on
4705 non-volatile objects. The C standard leaves it implementation defined
4706 as to what constitutes a volatile access. The C++ standard omits to
4707 specify this, except to say that C++ should behave in a similar manner
4708 to C with respect to volatiles, where possible. The minimum either
4709 standard specifies is that at a sequence point all previous accesses to
4710 volatile objects have stabilized and no subsequent accesses have
4711 occurred. Thus an implementation is free to reorder and combine
4712 volatile accesses which occur between sequence points, but cannot do so
4713 for accesses across a sequence point. The use of volatiles does not
4714 allow you to violate the restriction on updating objects multiple times
4715 within a sequence point.
4717 In most expressions, it is intuitively obvious what is a read and what is
4718 a write. For instance
4721 volatile int *dst = @var{somevalue};
4722 volatile int *src = @var{someothervalue};
4727 will cause a read of the volatile object pointed to by @var{src} and stores the
4728 value into the volatile object pointed to by @var{dst}. There is no
4729 guarantee that these reads and writes are atomic, especially for objects
4730 larger than @code{int}.
4732 Less obvious expressions are where something which looks like an access
4733 is used in a void context. An example would be,
4736 volatile int *src = @var{somevalue};
4740 With C, such expressions are rvalues, and as rvalues cause a read of
4741 the object, GCC interprets this as a read of the volatile being pointed
4742 to. The C++ standard specifies that such expressions do not undergo
4743 lvalue to rvalue conversion, and that the type of the dereferenced
4744 object may be incomplete. The C++ standard does not specify explicitly
4745 that it is this lvalue to rvalue conversion which is responsible for
4746 causing an access. However, there is reason to believe that it is,
4747 because otherwise certain simple expressions become undefined. However,
4748 because it would surprise most programmers, G++ treats dereferencing a
4749 pointer to volatile object of complete type in a void context as a read
4750 of the object. When the object has incomplete type, G++ issues a
4755 struct T @{int m;@};
4756 volatile S *ptr1 = @var{somevalue};
4757 volatile T *ptr2 = @var{somevalue};
4762 In this example, a warning is issued for @code{*ptr1}, and @code{*ptr2}
4763 causes a read of the object pointed to. If you wish to force an error on
4764 the first case, you must force a conversion to rvalue with, for instance
4765 a static cast, @code{static_cast<S>(*ptr1)}.
4767 When using a reference to volatile, G++ does not treat equivalent
4768 expressions as accesses to volatiles, but instead issues a warning that
4769 no volatile is accessed. The rationale for this is that otherwise it
4770 becomes difficult to determine where volatile access occur, and not
4771 possible to ignore the return value from functions returning volatile
4772 references. Again, if you wish to force a read, cast the reference to
4775 @node Restricted Pointers
4776 @section Restricting Pointer Aliasing
4777 @cindex restricted pointers
4778 @cindex restricted references
4779 @cindex restricted this pointer
4781 As with gcc, g++ understands the C99 feature of restricted pointers,
4782 specified with the @code{__restrict__}, or @code{__restrict} type
4783 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
4784 language flag, @code{restrict} is not a keyword in C++.
4786 In addition to allowing restricted pointers, you can specify restricted
4787 references, which indicate that the reference is not aliased in the local
4791 void fn (int *__restrict__ rptr, int &__restrict__ rref)
4798 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
4799 @var{rref} refers to a (different) unaliased integer.
4801 You may also specify whether a member function's @var{this} pointer is
4802 unaliased by using @code{__restrict__} as a member function qualifier.
4805 void T::fn () __restrict__
4812 Within the body of @code{T::fn}, @var{this} will have the effective
4813 definition @code{T *__restrict__ const this}. Notice that the
4814 interpretation of a @code{__restrict__} member function qualifier is
4815 different to that of @code{const} or @code{volatile} qualifier, in that it
4816 is applied to the pointer rather than the object. This is consistent with
4817 other compilers which implement restricted pointers.
4819 As with all outermost parameter qualifiers, @code{__restrict__} is
4820 ignored in function definition matching. This means you only need to
4821 specify @code{__restrict__} in a function definition, rather than
4822 in a function prototype as well.
4825 @section Vague Linkage
4826 @cindex vague linkage
4828 There are several constructs in C++ which require space in the object
4829 file but are not clearly tied to a single translation unit. We say that
4830 these constructs have ``vague linkage''. Typically such constructs are
4831 emitted wherever they are needed, though sometimes we can be more
4835 @item Inline Functions
4836 Inline functions are typically defined in a header file which can be
4837 included in many different compilations. Hopefully they can usually be
4838 inlined, but sometimes an out-of-line copy is necessary, if the address
4839 of the function is taken or if inlining fails. In general, we emit an
4840 out-of-line copy in all translation units where one is needed. As an
4841 exception, we only emit inline virtual functions with the vtable, since
4842 it will always require a copy.
4844 Local static variables and string constants used in an inline function
4845 are also considered to have vague linkage, since they must be shared
4846 between all inlined and out-of-line instances of the function.
4850 C++ virtual functions are implemented in most compilers using a lookup
4851 table, known as a vtable. The vtable contains pointers to the virtual
4852 functions provided by a class, and each object of the class contains a
4853 pointer to its vtable (or vtables, in some multiple-inheritance
4854 situations). If the class declares any non-inline, non-pure virtual
4855 functions, the first one is chosen as the ``key method'' for the class,
4856 and the vtable is only emitted in the translation unit where the key
4859 @emph{Note:} If the chosen key method is later defined as inline, the
4860 vtable will still be emitted in every translation unit which defines it.
4861 Make sure that any inline virtuals are declared inline in the class
4862 body, even if they are not defined there.
4864 @item type_info objects
4867 C++ requires information about types to be written out in order to
4868 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
4869 For polymorphic classes (classes with virtual functions), the type_info
4870 object is written out along with the vtable so that @samp{dynamic_cast}
4871 can determine the dynamic type of a class object at runtime. For all
4872 other types, we write out the type_info object when it is used: when
4873 applying @samp{typeid} to an expression, throwing an object, or
4874 referring to a type in a catch clause or exception specification.
4876 @item Template Instantiations
4877 Most everything in this section also applies to template instantiations,
4878 but there are other options as well.
4879 @xref{Template Instantiation,,Where's the Template?}.
4883 When used with GNU ld version 2.8 or later on an ELF system such as
4884 Linux/GNU or Solaris 2, or on Microsoft Windows, duplicate copies of
4885 these constructs will be discarded at link time. This is known as
4888 On targets that don't support COMDAT, but do support weak symbols, GCC
4889 will use them. This way one copy will override all the others, but
4890 the unused copies will still take up space in the executable.
4892 For targets which do not support either COMDAT or weak symbols,
4893 most entities with vague linkage will be emitted as local symbols to
4894 avoid duplicate definition errors from the linker. This will not happen
4895 for local statics in inlines, however, as having multiple copies will
4896 almost certainly break things.
4898 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
4899 another way to control placement of these constructs.
4902 @section Declarations and Definitions in One Header
4904 @cindex interface and implementation headers, C++
4905 @cindex C++ interface and implementation headers
4906 C++ object definitions can be quite complex. In principle, your source
4907 code will need two kinds of things for each object that you use across
4908 more than one source file. First, you need an @dfn{interface}
4909 specification, describing its structure with type declarations and
4910 function prototypes. Second, you need the @dfn{implementation} itself.
4911 It can be tedious to maintain a separate interface description in a
4912 header file, in parallel to the actual implementation. It is also
4913 dangerous, since separate interface and implementation definitions may
4914 not remain parallel.
4916 @cindex pragmas, interface and implementation
4917 With GNU C++, you can use a single header file for both purposes.
4920 @emph{Warning:} The mechanism to specify this is in transition. For the
4921 nonce, you must use one of two @code{#pragma} commands; in a future
4922 release of GNU C++, an alternative mechanism will make these
4923 @code{#pragma} commands unnecessary.
4926 The header file contains the full definitions, but is marked with
4927 @samp{#pragma interface} in the source code. This allows the compiler
4928 to use the header file only as an interface specification when ordinary
4929 source files incorporate it with @code{#include}. In the single source
4930 file where the full implementation belongs, you can use either a naming
4931 convention or @samp{#pragma implementation} to indicate this alternate
4932 use of the header file.
4935 @item #pragma interface
4936 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
4937 @kindex #pragma interface
4938 Use this directive in @emph{header files} that define object classes, to save
4939 space in most of the object files that use those classes. Normally,
4940 local copies of certain information (backup copies of inline member
4941 functions, debugging information, and the internal tables that implement
4942 virtual functions) must be kept in each object file that includes class
4943 definitions. You can use this pragma to avoid such duplication. When a
4944 header file containing @samp{#pragma interface} is included in a
4945 compilation, this auxiliary information will not be generated (unless
4946 the main input source file itself uses @samp{#pragma implementation}).
4947 Instead, the object files will contain references to be resolved at link
4950 The second form of this directive is useful for the case where you have
4951 multiple headers with the same name in different directories. If you
4952 use this form, you must specify the same string to @samp{#pragma
4955 @item #pragma implementation
4956 @itemx #pragma implementation "@var{objects}.h"
4957 @kindex #pragma implementation
4958 Use this pragma in a @emph{main input file}, when you want full output from
4959 included header files to be generated (and made globally visible). The
4960 included header file, in turn, should use @samp{#pragma interface}.
4961 Backup copies of inline member functions, debugging information, and the
4962 internal tables used to implement virtual functions are all generated in
4963 implementation files.
4965 @cindex implied @code{#pragma implementation}
4966 @cindex @code{#pragma implementation}, implied
4967 @cindex naming convention, implementation headers
4968 If you use @samp{#pragma implementation} with no argument, it applies to
4969 an include file with the same basename@footnote{A file's @dfn{basename}
4970 was the name stripped of all leading path information and of trailing
4971 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
4972 file. For example, in @file{allclass.cc}, giving just
4973 @samp{#pragma implementation}
4974 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
4976 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
4977 an implementation file whenever you would include it from
4978 @file{allclass.cc} even if you never specified @samp{#pragma
4979 implementation}. This was deemed to be more trouble than it was worth,
4980 however, and disabled.
4982 If you use an explicit @samp{#pragma implementation}, it must appear in
4983 your source file @emph{before} you include the affected header files.
4985 Use the string argument if you want a single implementation file to
4986 include code from multiple header files. (You must also use
4987 @samp{#include} to include the header file; @samp{#pragma
4988 implementation} only specifies how to use the file---it doesn't actually
4991 There is no way to split up the contents of a single header file into
4992 multiple implementation files.
4995 @cindex inlining and C++ pragmas
4996 @cindex C++ pragmas, effect on inlining
4997 @cindex pragmas in C++, effect on inlining
4998 @samp{#pragma implementation} and @samp{#pragma interface} also have an
4999 effect on function inlining.
5001 If you define a class in a header file marked with @samp{#pragma
5002 interface}, the effect on a function defined in that class is similar to
5003 an explicit @code{extern} declaration---the compiler emits no code at
5004 all to define an independent version of the function. Its definition
5005 is used only for inlining with its callers.
5007 @opindex fno-implement-inlines
5008 Conversely, when you include the same header file in a main source file
5009 that declares it as @samp{#pragma implementation}, the compiler emits
5010 code for the function itself; this defines a version of the function
5011 that can be found via pointers (or by callers compiled without
5012 inlining). If all calls to the function can be inlined, you can avoid
5013 emitting the function by compiling with @option{-fno-implement-inlines}.
5014 If any calls were not inlined, you will get linker errors.
5016 @node Template Instantiation
5017 @section Where's the Template?
5019 @cindex template instantiation
5021 C++ templates are the first language feature to require more
5022 intelligence from the environment than one usually finds on a UNIX
5023 system. Somehow the compiler and linker have to make sure that each
5024 template instance occurs exactly once in the executable if it is needed,
5025 and not at all otherwise. There are two basic approaches to this
5026 problem, which I will refer to as the Borland model and the Cfront model.
5030 Borland C++ solved the template instantiation problem by adding the code
5031 equivalent of common blocks to their linker; the compiler emits template
5032 instances in each translation unit that uses them, and the linker
5033 collapses them together. The advantage of this model is that the linker
5034 only has to consider the object files themselves; there is no external
5035 complexity to worry about. This disadvantage is that compilation time
5036 is increased because the template code is being compiled repeatedly.
5037 Code written for this model tends to include definitions of all
5038 templates in the header file, since they must be seen to be
5042 The AT&T C++ translator, Cfront, solved the template instantiation
5043 problem by creating the notion of a template repository, an
5044 automatically maintained place where template instances are stored. A
5045 more modern version of the repository works as follows: As individual
5046 object files are built, the compiler places any template definitions and
5047 instantiations encountered in the repository. At link time, the link
5048 wrapper adds in the objects in the repository and compiles any needed
5049 instances that were not previously emitted. The advantages of this
5050 model are more optimal compilation speed and the ability to use the
5051 system linker; to implement the Borland model a compiler vendor also
5052 needs to replace the linker. The disadvantages are vastly increased
5053 complexity, and thus potential for error; for some code this can be
5054 just as transparent, but in practice it can been very difficult to build
5055 multiple programs in one directory and one program in multiple
5056 directories. Code written for this model tends to separate definitions
5057 of non-inline member templates into a separate file, which should be
5058 compiled separately.
5061 When used with GNU ld version 2.8 or later on an ELF system such as
5062 Linux/GNU or Solaris 2, or on Microsoft Windows, g++ supports the
5063 Borland model. On other systems, g++ implements neither automatic
5066 A future version of g++ will support a hybrid model whereby the compiler
5067 will emit any instantiations for which the template definition is
5068 included in the compile, and store template definitions and
5069 instantiation context information into the object file for the rest.
5070 The link wrapper will extract that information as necessary and invoke
5071 the compiler to produce the remaining instantiations. The linker will
5072 then combine duplicate instantiations.
5074 In the mean time, you have the following options for dealing with
5075 template instantiations:
5080 Compile your template-using code with @option{-frepo}. The compiler will
5081 generate files with the extension @samp{.rpo} listing all of the
5082 template instantiations used in the corresponding object files which
5083 could be instantiated there; the link wrapper, @samp{collect2}, will
5084 then update the @samp{.rpo} files to tell the compiler where to place
5085 those instantiations and rebuild any affected object files. The
5086 link-time overhead is negligible after the first pass, as the compiler
5087 will continue to place the instantiations in the same files.
5089 This is your best option for application code written for the Borland
5090 model, as it will just work. Code written for the Cfront model will
5091 need to be modified so that the template definitions are available at
5092 one or more points of instantiation; usually this is as simple as adding
5093 @code{#include <tmethods.cc>} to the end of each template header.
5095 For library code, if you want the library to provide all of the template
5096 instantiations it needs, just try to link all of its object files
5097 together; the link will fail, but cause the instantiations to be
5098 generated as a side effect. Be warned, however, that this may cause
5099 conflicts if multiple libraries try to provide the same instantiations.
5100 For greater control, use explicit instantiation as described in the next
5104 @opindex fno-implicit-templates
5105 Compile your code with @option{-fno-implicit-templates} to disable the
5106 implicit generation of template instances, and explicitly instantiate
5107 all the ones you use. This approach requires more knowledge of exactly
5108 which instances you need than do the others, but it's less
5109 mysterious and allows greater control. You can scatter the explicit
5110 instantiations throughout your program, perhaps putting them in the
5111 translation units where the instances are used or the translation units
5112 that define the templates themselves; you can put all of the explicit
5113 instantiations you need into one big file; or you can create small files
5120 template class Foo<int>;
5121 template ostream& operator <<
5122 (ostream&, const Foo<int>&);
5125 for each of the instances you need, and create a template instantiation
5128 If you are using Cfront-model code, you can probably get away with not
5129 using @option{-fno-implicit-templates} when compiling files that don't
5130 @samp{#include} the member template definitions.
5132 If you use one big file to do the instantiations, you may want to
5133 compile it without @option{-fno-implicit-templates} so you get all of the
5134 instances required by your explicit instantiations (but not by any
5135 other files) without having to specify them as well.
5137 g++ has extended the template instantiation syntax outlined in the
5138 Working Paper to allow forward declaration of explicit instantiations
5139 (with @code{extern}), instantiation of the compiler support data for a
5140 template class (i.e.@: the vtable) without instantiating any of its
5141 members (with @code{inline}), and instantiation of only the static data
5142 members of a template class, without the support data or member
5143 functions (with (@code{static}):
5146 extern template int max (int, int);
5147 inline template class Foo<int>;
5148 static template class Foo<int>;
5152 Do nothing. Pretend g++ does implement automatic instantiation
5153 management. Code written for the Borland model will work fine, but
5154 each translation unit will contain instances of each of the templates it
5155 uses. In a large program, this can lead to an unacceptable amount of code
5159 @opindex fexternal-templates
5160 Add @samp{#pragma interface} to all files containing template
5161 definitions. For each of these files, add @samp{#pragma implementation
5162 "@var{filename}"} to the top of some @samp{.C} file which
5163 @samp{#include}s it. Then compile everything with
5164 @option{-fexternal-templates}. The templates will then only be expanded
5165 in the translation unit which implements them (i.e.@: has a @samp{#pragma
5166 implementation} line for the file where they live); all other files will
5167 use external references. If you're lucky, everything should work
5168 properly. If you get undefined symbol errors, you need to make sure
5169 that each template instance which is used in the program is used in the
5170 file which implements that template. If you don't have any use for a
5171 particular instance in that file, you can just instantiate it
5172 explicitly, using the syntax from the latest C++ working paper:
5175 template class A<int>;
5176 template ostream& operator << (ostream&, const A<int>&);
5179 This strategy will work with code written for either model. If you are
5180 using code written for the Cfront model, the file containing a class
5181 template and the file containing its member templates should be
5182 implemented in the same translation unit.
5185 @opindex falt-external-templates
5186 A slight variation on this approach is to use the flag
5187 @option{-falt-external-templates} instead. This flag causes template
5188 instances to be emitted in the translation unit that implements the
5189 header where they are first instantiated, rather than the one which
5190 implements the file where the templates are defined. This header must
5191 be the same in all translation units, or things are likely to break.
5193 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
5194 more discussion of these pragmas.
5197 @node Bound member functions
5198 @section Extracting the function pointer from a bound pointer to member function
5201 @cindex pointer to member function
5202 @cindex bound pointer to member function
5204 In C++, pointer to member functions (PMFs) are implemented using a wide
5205 pointer of sorts to handle all the possible call mechanisms; the PMF
5206 needs to store information about how to adjust the @samp{this} pointer,
5207 and if the function pointed to is virtual, where to find the vtable, and
5208 where in the vtable to look for the member function. If you are using
5209 PMFs in an inner loop, you should really reconsider that decision. If
5210 that is not an option, you can extract the pointer to the function that
5211 would be called for a given object/PMF pair and call it directly inside
5212 the inner loop, to save a bit of time.
5214 Note that you will still be paying the penalty for the call through a
5215 function pointer; on most modern architectures, such a call defeats the
5216 branch prediction features of the CPU@. This is also true of normal
5217 virtual function calls.
5219 The syntax for this extension is
5223 extern int (A::*fp)();
5224 typedef int (*fptr)(A *);
5226 fptr p = (fptr)(a.*fp);
5229 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
5230 no object is needed to obtain the address of the function. They can be
5231 converted to function pointers directly:
5234 fptr p1 = (fptr)(&A::foo);
5237 @opindex Wno-pmf-conversions
5238 You must specify @option{-Wno-pmf-conversions} to use this extension.
5240 @node C++ Attributes
5241 @section C++-Specific Variable, Function, and Type Attributes
5243 Some attributes only make sense for C++ programs.
5246 @item init_priority (@var{priority})
5247 @cindex init_priority attribute
5250 In Standard C++, objects defined at namespace scope are guaranteed to be
5251 initialized in an order in strict accordance with that of their definitions
5252 @emph{in a given translation unit}. No guarantee is made for initializations
5253 across translation units. However, GNU C++ allows users to control the
5254 order of initialization of objects defined at namespace scope with the
5255 @code{init_priority} attribute by specifying a relative @var{priority},
5256 a constant integral expression currently bounded between 101 and 65535
5257 inclusive. Lower numbers indicate a higher priority.
5259 In the following example, @code{A} would normally be created before
5260 @code{B}, but the @code{init_priority} attribute has reversed that order:
5263 Some_Class A __attribute__ ((init_priority (2000)));
5264 Some_Class B __attribute__ ((init_priority (543)));
5268 Note that the particular values of @var{priority} do not matter; only their
5271 @item java_interface
5272 @cindex java_interface attribute
5274 This type attribute informs C++ that the class is a Java interface. It may
5275 only be applied to classes declared within an @code{extern "Java"} block.
5276 Calls to methods declared in this interface will be dispatched using GCJ's
5277 interface table mechanism, instead of regular virtual table dispatch.
5281 @node Java Exceptions
5282 @section Java Exceptions
5284 The Java language uses a slightly different exception handling model
5285 from C++. Normally, GNU C++ will automatically detect when you are
5286 writing C++ code that uses Java exceptions, and handle them
5287 appropriately. However, if C++ code only needs to execute destructors
5288 when Java exceptions are thrown through it, GCC will guess incorrectly.
5289 Sample problematic code is:
5292 struct S @{ ~S(); @};
5293 extern void bar(); // is written in Java, and may throw exceptions
5302 The usual effect of an incorrect guess is a link failure, complaining of
5303 a missing routine called @samp{__gxx_personality_v0}.
5305 You can inform the compiler that Java exceptions are to be used in a
5306 translation unit, irrespective of what it might think, by writing
5307 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
5308 @samp{#pragma} must appear before any functions that throw or catch
5309 exceptions, or run destructors when exceptions are thrown through them.
5311 You cannot mix Java and C++ exceptions in the same translation unit. It
5312 is believed to be safe to throw a C++ exception from one file through
5313 another file compiled for the Java exception model, or vice versa, but
5314 there may be bugs in this area.
5316 @node Deprecated Features
5317 @section Deprecated Features
5319 In the past, the GNU C++ compiler was extended to experiment with new
5320 features, at a time when the C++ language was still evolving. Now that
5321 the C++ standard is complete, some of those features are superseded by
5322 superior alternatives. Using the old features might cause a warning in
5323 some cases that the feature will be dropped in the future. In other
5324 cases, the feature might be gone already.
5326 While the list below is not exhaustive, it documents some of the options
5327 that are now deprecated:
5330 @item -fexternal-templates
5331 @itemx -falt-external-templates
5332 These are two of the many ways for g++ to implement template
5333 instantiation. @xref{Template Instantiation}. The C++ standard clearly
5334 defines how template definitions have to be organized across
5335 implementation units. g++ has an implicit instantiation mechanism that
5336 should work just fine for standard-conforming code.
5338 @item -fstrict-prototype
5339 @itemx -fno-strict-prototype
5340 Previously it was possible to use an empty prototype parameter list to
5341 indicate an unspecified number of parameters (like C), rather than no
5342 parameters, as C++ demands. This feature has been removed, except where
5343 it is required for backwards compatibility @xref{Backwards Compatibility}.
5346 The named return value extension has been deprecated, and will be
5347 removed from g++ at some point.
5349 The use of initializer lists with new expressions has been deprecated,
5350 and will be removed from g++ at some point.
5352 @node Backwards Compatibility
5353 @section Backwards Compatibility
5354 @cindex Backwards Compatibility
5355 @cindex ARM [Annotated C++ Reference Manual]
5357 Now that there is a definitive ISO standard C++, G++ has a specification
5358 to adhere to. The C++ language evolved over time, and features that
5359 used to be acceptable in previous drafts of the standard, such as the ARM
5360 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
5361 compilation of C++ written to such drafts, G++ contains some backwards
5362 compatibilities. @emph{All such backwards compatibility features are
5363 liable to disappear in future versions of G++.} They should be considered
5364 deprecated @xref{Deprecated Features}.
5368 If a variable is declared at for scope, it used to remain in scope until
5369 the end of the scope which contained the for statement (rather than just
5370 within the for scope). G++ retains this, but issues a warning, if such a
5371 variable is accessed outside the for scope.
5373 @item implicit C language
5374 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
5375 scope to set the language. On such systems, all header files are
5376 implicitly scoped inside a C language scope. Also, an empty prototype
5377 @code{()} will be treated as an unspecified number of arguments, rather
5378 than no arguments, as C++ demands.