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++.
383 * Statement Exprs:: Putting statements and declarations inside expressions.
384 * Local Labels:: Labels local to a statement-expression.
385 * Labels as Values:: Getting pointers to labels, and computed gotos.
386 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
387 * Constructing Calls:: Dispatching a call to another function.
388 * Naming Types:: Giving a name to the type of some expression.
389 * Typeof:: @code{typeof}: referring to the type of an expression.
390 * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues.
391 * Conditionals:: Omitting the middle operand of a @samp{?:} expression.
392 * Long Long:: Double-word integers---@code{long long int}.
393 * Complex:: Data types for complex numbers.
394 * Hex Floats:: Hexadecimal floating-point constants.
395 * Zero Length:: Zero-length arrays.
396 * Variable Length:: Arrays whose length is computed at run time.
397 * Variadic Macros:: Macros with a variable number of arguments.
398 * Escaped Newlines:: Slightly looser rules for escaped newlines.
399 * Multi-line Strings:: String literals with embedded newlines.
400 * Subscripting:: Any array can be subscripted, even if not an lvalue.
401 * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers.
402 * Initializers:: Non-constant initializers.
403 * Compound Literals:: Compound literals give structures, unions
405 * Designated Inits:: Labeling elements of initializers.
406 * Cast to Union:: Casting to union type from any member of the union.
407 * Case Ranges:: `case 1 ... 9' and such.
408 * Mixed Declarations:: Mixing declarations and code.
409 * Function Attributes:: Declaring that functions have no side effects,
410 or that they can never return.
411 * Attribute Syntax:: Formal syntax for attributes.
412 * Function Prototypes:: Prototype declarations and old-style definitions.
413 * C++ Comments:: C++ comments are recognized.
414 * Dollar Signs:: Dollar sign is allowed in identifiers.
415 * Character Escapes:: @samp{\e} stands for the character @key{ESC}.
416 * Variable Attributes:: Specifying attributes of variables.
417 * Type Attributes:: Specifying attributes of types.
418 * Alignment:: Inquiring about the alignment of a type or variable.
419 * Inline:: Defining inline functions (as fast as macros).
420 * Extended Asm:: Assembler instructions with C expressions as operands.
421 (With them you can define ``built-in'' functions.)
422 * Constraints:: Constraints for asm operands
423 * Asm Labels:: Specifying the assembler name to use for a C symbol.
424 * Explicit Reg Vars:: Defining variables residing in specified registers.
425 * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files.
426 * Incomplete Enums:: @code{enum foo;}, with details to follow.
427 * Function Names:: Printable strings which are the name of the current
429 * Return Address:: Getting the return or frame address of a function.
430 * Vector Extensions:: Using vector instructions through built-in functions.
431 * Other Builtins:: Other built-in functions.
432 * Pragmas:: Pragmas accepted by GCC.
433 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
436 @node Statement Exprs
437 @section Statements and Declarations in Expressions
438 @cindex statements inside expressions
439 @cindex declarations inside expressions
440 @cindex expressions containing statements
441 @cindex macros, statements in expressions
443 @c the above section title wrapped and causes an underfull hbox.. i
444 @c changed it from "within" to "in". --mew 4feb93
446 A compound statement enclosed in parentheses may appear as an expression
447 in GNU C@. This allows you to use loops, switches, and local variables
448 within an expression.
450 Recall that a compound statement is a sequence of statements surrounded
451 by braces; in this construct, parentheses go around the braces. For
455 (@{ int y = foo (); int z;
462 is a valid (though slightly more complex than necessary) expression
463 for the absolute value of @code{foo ()}.
465 The last thing in the compound statement should be an expression
466 followed by a semicolon; the value of this subexpression serves as the
467 value of the entire construct. (If you use some other kind of statement
468 last within the braces, the construct has type @code{void}, and thus
469 effectively no value.)
471 This feature is especially useful in making macro definitions ``safe'' (so
472 that they evaluate each operand exactly once). For example, the
473 ``maximum'' function is commonly defined as a macro in standard C as
477 #define max(a,b) ((a) > (b) ? (a) : (b))
481 @cindex side effects, macro argument
482 But this definition computes either @var{a} or @var{b} twice, with bad
483 results if the operand has side effects. In GNU C, if you know the
484 type of the operands (here let's assume @code{int}), you can define
485 the macro safely as follows:
488 #define maxint(a,b) \
489 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @})
492 Embedded statements are not allowed in constant expressions, such as
493 the value of an enumeration constant, the width of a bit-field, or
494 the initial value of a static variable.
496 If you don't know the type of the operand, you can still do this, but you
497 must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming
500 Statement expressions are not supported fully in G++, and their fate
501 there is unclear. (It is possible that they will become fully supported
502 at some point, or that they will be deprecated, or that the bugs that
503 are present will continue to exist indefinitely.) Presently, statement
504 expressions do not work well as default arguments.
506 In addition, there are semantic issues with statement-expressions in
507 C++. If you try to use statement-expressions instead of inline
508 functions in C++, you may be surprised at the way object destruction is
509 handled. For example:
512 #define foo(a) (@{int b = (a); b + 3; @})
516 does not work the same way as:
519 inline int foo(int a) @{ int b = a; return b + 3; @}
523 In particular, if the expression passed into @code{foo} involves the
524 creation of temporaries, the destructors for those temporaries will be
525 run earlier in the case of the macro than in the case of the function.
527 These considerations mean that it is probably a bad idea to use
528 statement-expressions of this form in header files that are designed to
529 work with C++. (Note that some versions of the GNU C Library contained
530 header files using statement-expression that lead to precisely this
534 @section Locally Declared Labels
536 @cindex macros, local labels
538 Each statement expression is a scope in which @dfn{local labels} can be
539 declared. A local label is simply an identifier; you can jump to it
540 with an ordinary @code{goto} statement, but only from within the
541 statement expression it belongs to.
543 A local label declaration looks like this:
546 __label__ @var{label};
553 __label__ @var{label1}, @var{label2}, @dots{};
556 Local label declarations must come at the beginning of the statement
557 expression, right after the @samp{(@{}, before any ordinary
560 The label declaration defines the label @emph{name}, but does not define
561 the label itself. You must do this in the usual way, with
562 @code{@var{label}:}, within the statements of the statement expression.
564 The local label feature is useful because statement expressions are
565 often used in macros. If the macro contains nested loops, a @code{goto}
566 can be useful for breaking out of them. However, an ordinary label
567 whose scope is the whole function cannot be used: if the macro can be
568 expanded several times in one function, the label will be multiply
569 defined in that function. A local label avoids this problem. For
573 #define SEARCH(array, target) \
576 typeof (target) _SEARCH_target = (target); \
577 typeof (*(array)) *_SEARCH_array = (array); \
580 for (i = 0; i < max; i++) \
581 for (j = 0; j < max; j++) \
582 if (_SEARCH_array[i][j] == _SEARCH_target) \
583 @{ value = i; goto found; @} \
590 @node Labels as Values
591 @section Labels as Values
592 @cindex labels as values
593 @cindex computed gotos
594 @cindex goto with computed label
595 @cindex address of a label
597 You can get the address of a label defined in the current function
598 (or a containing function) with the unary operator @samp{&&}. The
599 value has type @code{void *}. This value is a constant and can be used
600 wherever a constant of that type is valid. For example:
608 To use these values, you need to be able to jump to one. This is done
609 with the computed goto statement@footnote{The analogous feature in
610 Fortran is called an assigned goto, but that name seems inappropriate in
611 C, where one can do more than simply store label addresses in label
612 variables.}, @code{goto *@var{exp};}. For example,
619 Any expression of type @code{void *} is allowed.
621 One way of using these constants is in initializing a static array that
622 will serve as a jump table:
625 static void *array[] = @{ &&foo, &&bar, &&hack @};
628 Then you can select a label with indexing, like this:
635 Note that this does not check whether the subscript is in bounds---array
636 indexing in C never does that.
638 Such an array of label values serves a purpose much like that of the
639 @code{switch} statement. The @code{switch} statement is cleaner, so
640 use that rather than an array unless the problem does not fit a
641 @code{switch} statement very well.
643 Another use of label values is in an interpreter for threaded code.
644 The labels within the interpreter function can be stored in the
645 threaded code for super-fast dispatching.
647 You may not use this mechanism to jump to code in a different function.
648 If you do that, totally unpredictable things will happen. The best way to
649 avoid this is to store the label address only in automatic variables and
650 never pass it as an argument.
652 An alternate way to write the above example is
655 static const int array[] = @{ &&foo - &&foo, &&bar - &&foo,
657 goto *(&&foo + array[i]);
661 This is more friendly to code living in shared libraries, as it reduces
662 the number of dynamic relocations that are needed, and by consequence,
663 allows the data to be read-only.
665 @node Nested Functions
666 @section Nested Functions
667 @cindex nested functions
668 @cindex downward funargs
671 A @dfn{nested function} is a function defined inside another function.
672 (Nested functions are not supported for GNU C++.) The nested function's
673 name is local to the block where it is defined. For example, here we
674 define a nested function named @code{square}, and call it twice:
678 foo (double a, double b)
680 double square (double z) @{ return z * z; @}
682 return square (a) + square (b);
687 The nested function can access all the variables of the containing
688 function that are visible at the point of its definition. This is
689 called @dfn{lexical scoping}. For example, here we show a nested
690 function which uses an inherited variable named @code{offset}:
694 bar (int *array, int offset, int size)
696 int access (int *array, int index)
697 @{ return array[index + offset]; @}
700 for (i = 0; i < size; i++)
701 @dots{} access (array, i) @dots{}
706 Nested function definitions are permitted within functions in the places
707 where variable definitions are allowed; that is, in any block, before
708 the first statement in the block.
710 It is possible to call the nested function from outside the scope of its
711 name by storing its address or passing the address to another function:
714 hack (int *array, int size)
716 void store (int index, int value)
717 @{ array[index] = value; @}
719 intermediate (store, size);
723 Here, the function @code{intermediate} receives the address of
724 @code{store} as an argument. If @code{intermediate} calls @code{store},
725 the arguments given to @code{store} are used to store into @code{array}.
726 But this technique works only so long as the containing function
727 (@code{hack}, in this example) does not exit.
729 If you try to call the nested function through its address after the
730 containing function has exited, all hell will break loose. If you try
731 to call it after a containing scope level has exited, and if it refers
732 to some of the variables that are no longer in scope, you may be lucky,
733 but it's not wise to take the risk. If, however, the nested function
734 does not refer to anything that has gone out of scope, you should be
737 GCC implements taking the address of a nested function using a technique
738 called @dfn{trampolines}. A paper describing them is available as
741 @uref{http://people.debian.org/~karlheg/Usenix88-lexic.pdf}.
743 A nested function can jump to a label inherited from a containing
744 function, provided the label was explicitly declared in the containing
745 function (@pxref{Local Labels}). Such a jump returns instantly to the
746 containing function, exiting the nested function which did the
747 @code{goto} and any intermediate functions as well. Here is an example:
751 bar (int *array, int offset, int size)
754 int access (int *array, int index)
758 return array[index + offset];
762 for (i = 0; i < size; i++)
763 @dots{} access (array, i) @dots{}
767 /* @r{Control comes here from @code{access}
768 if it detects an error.} */
775 A nested function always has internal linkage. Declaring one with
776 @code{extern} is erroneous. If you need to declare the nested function
777 before its definition, use @code{auto} (which is otherwise meaningless
778 for function declarations).
781 bar (int *array, int offset, int size)
784 auto int access (int *, int);
786 int access (int *array, int index)
790 return array[index + offset];
796 @node Constructing Calls
797 @section Constructing Function Calls
798 @cindex constructing calls
799 @cindex forwarding calls
801 Using the built-in functions described below, you can record
802 the arguments a function received, and call another function
803 with the same arguments, without knowing the number or types
806 You can also record the return value of that function call,
807 and later return that value, without knowing what data type
808 the function tried to return (as long as your caller expects
811 @deftypefn {Built-in Function} {void *} __builtin_apply_args ()
812 This built-in function returns a pointer to data
813 describing how to perform a call with the same arguments as were passed
814 to the current function.
816 The function saves the arg pointer register, structure value address,
817 and all registers that might be used to pass arguments to a function
818 into a block of memory allocated on the stack. Then it returns the
819 address of that block.
822 @deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size})
823 This built-in function invokes @var{function}
824 with a copy of the parameters described by @var{arguments}
827 The value of @var{arguments} should be the value returned by
828 @code{__builtin_apply_args}. The argument @var{size} specifies the size
829 of the stack argument data, in bytes.
831 This function returns a pointer to data describing
832 how to return whatever value was returned by @var{function}. The data
833 is saved in a block of memory allocated on the stack.
835 It is not always simple to compute the proper value for @var{size}. The
836 value is used by @code{__builtin_apply} to compute the amount of data
837 that should be pushed on the stack and copied from the incoming argument
841 @deftypefn {Built-in Function} {void} __builtin_return (void *@var{result})
842 This built-in function returns the value described by @var{result} from
843 the containing function. You should specify, for @var{result}, a value
844 returned by @code{__builtin_apply}.
848 @section Naming an Expression's Type
851 You can give a name to the type of an expression using a @code{typedef}
852 declaration with an initializer. Here is how to define @var{name} as a
853 type name for the type of @var{exp}:
856 typedef @var{name} = @var{exp};
859 This is useful in conjunction with the statements-within-expressions
860 feature. Here is how the two together can be used to define a safe
861 ``maximum'' macro that operates on any arithmetic type:
865 (@{typedef _ta = (a), _tb = (b); \
866 _ta _a = (a); _tb _b = (b); \
867 _a > _b ? _a : _b; @})
870 @cindex underscores in variables in macros
871 @cindex @samp{_} in variables in macros
872 @cindex local variables in macros
873 @cindex variables, local, in macros
874 @cindex macros, local variables in
876 The reason for using names that start with underscores for the local
877 variables is to avoid conflicts with variable names that occur within the
878 expressions that are substituted for @code{a} and @code{b}. Eventually we
879 hope to design a new form of declaration syntax that allows you to declare
880 variables whose scopes start only after their initializers; this will be a
881 more reliable way to prevent such conflicts.
884 @section Referring to a Type with @code{typeof}
887 @cindex macros, types of arguments
889 Another way to refer to the type of an expression is with @code{typeof}.
890 The syntax of using of this keyword looks like @code{sizeof}, but the
891 construct acts semantically like a type name defined with @code{typedef}.
893 There are two ways of writing the argument to @code{typeof}: with an
894 expression or with a type. Here is an example with an expression:
901 This assumes that @code{x} is an array of pointers to functions;
902 the type described is that of the values of the functions.
904 Here is an example with a typename as the argument:
911 Here the type described is that of pointers to @code{int}.
913 If you are writing a header file that must work when included in ISO C
914 programs, write @code{__typeof__} instead of @code{typeof}.
915 @xref{Alternate Keywords}.
917 A @code{typeof}-construct can be used anywhere a typedef name could be
918 used. For example, you can use it in a declaration, in a cast, or inside
919 of @code{sizeof} or @code{typeof}.
923 This declares @code{y} with the type of what @code{x} points to.
930 This declares @code{y} as an array of such values.
937 This declares @code{y} as an array of pointers to characters:
940 typeof (typeof (char *)[4]) y;
944 It is equivalent to the following traditional C declaration:
950 To see the meaning of the declaration using @code{typeof}, and why it
951 might be a useful way to write, let's rewrite it with these macros:
954 #define pointer(T) typeof(T *)
955 #define array(T, N) typeof(T [N])
959 Now the declaration can be rewritten this way:
962 array (pointer (char), 4) y;
966 Thus, @code{array (pointer (char), 4)} is the type of arrays of 4
967 pointers to @code{char}.
971 @section Generalized Lvalues
972 @cindex compound expressions as lvalues
973 @cindex expressions, compound, as lvalues
974 @cindex conditional expressions as lvalues
975 @cindex expressions, conditional, as lvalues
976 @cindex casts as lvalues
977 @cindex generalized lvalues
978 @cindex lvalues, generalized
979 @cindex extensions, @code{?:}
980 @cindex @code{?:} extensions
981 Compound expressions, conditional expressions and casts are allowed as
982 lvalues provided their operands are lvalues. This means that you can take
983 their addresses or store values into them.
985 Standard C++ allows compound expressions and conditional expressions as
986 lvalues, and permits casts to reference type, so use of this extension
987 is deprecated for C++ code.
989 For example, a compound expression can be assigned, provided the last
990 expression in the sequence is an lvalue. These two expressions are
998 Similarly, the address of the compound expression can be taken. These two
999 expressions are equivalent:
1006 A conditional expression is a valid lvalue if its type is not void and the
1007 true and false branches are both valid lvalues. For example, these two
1008 expressions are equivalent:
1012 (a ? b = 5 : (c = 5))
1015 A cast is a valid lvalue if its operand is an lvalue. A simple
1016 assignment whose left-hand side is a cast works by converting the
1017 right-hand side first to the specified type, then to the type of the
1018 inner left-hand side expression. After this is stored, the value is
1019 converted back to the specified type to become the value of the
1020 assignment. Thus, if @code{a} has type @code{char *}, the following two
1021 expressions are equivalent:
1025 (int)(a = (char *)(int)5)
1028 An assignment-with-arithmetic operation such as @samp{+=} applied to a cast
1029 performs the arithmetic using the type resulting from the cast, and then
1030 continues as in the previous case. Therefore, these two expressions are
1035 (int)(a = (char *)(int) ((int)a + 5))
1038 You cannot take the address of an lvalue cast, because the use of its
1039 address would not work out coherently. Suppose that @code{&(int)f} were
1040 permitted, where @code{f} has type @code{float}. Then the following
1041 statement would try to store an integer bit-pattern where a floating
1042 point number belongs:
1048 This is quite different from what @code{(int)f = 1} would do---that
1049 would convert 1 to floating point and store it. Rather than cause this
1050 inconsistency, we think it is better to prohibit use of @samp{&} on a cast.
1052 If you really do want an @code{int *} pointer with the address of
1053 @code{f}, you can simply write @code{(int *)&f}.
1056 @section Conditionals with Omitted Operands
1057 @cindex conditional expressions, extensions
1058 @cindex omitted middle-operands
1059 @cindex middle-operands, omitted
1060 @cindex extensions, @code{?:}
1061 @cindex @code{?:} extensions
1063 The middle operand in a conditional expression may be omitted. Then
1064 if the first operand is nonzero, its value is the value of the conditional
1067 Therefore, the expression
1074 has the value of @code{x} if that is nonzero; otherwise, the value of
1077 This example is perfectly equivalent to
1083 @cindex side effect in ?:
1084 @cindex ?: side effect
1086 In this simple case, the ability to omit the middle operand is not
1087 especially useful. When it becomes useful is when the first operand does,
1088 or may (if it is a macro argument), contain a side effect. Then repeating
1089 the operand in the middle would perform the side effect twice. Omitting
1090 the middle operand uses the value already computed without the undesirable
1091 effects of recomputing it.
1094 @section Double-Word Integers
1095 @cindex @code{long long} data types
1096 @cindex double-word arithmetic
1097 @cindex multiprecision arithmetic
1098 @cindex @code{LL} integer suffix
1099 @cindex @code{ULL} integer suffix
1101 ISO C99 supports data types for integers that are at least 64 bits wide,
1102 and as an extension GCC supports them in C89 mode and in C++.
1103 Simply write @code{long long int} for a signed integer, or
1104 @code{unsigned long long int} for an unsigned integer. To make an
1105 integer constant of type @code{long long int}, add the suffix @samp{LL}
1106 to the integer. To make an integer constant of type @code{unsigned long
1107 long int}, add the suffix @samp{ULL} to the integer.
1109 You can use these types in arithmetic like any other integer types.
1110 Addition, subtraction, and bitwise boolean operations on these types
1111 are open-coded on all types of machines. Multiplication is open-coded
1112 if the machine supports fullword-to-doubleword a widening multiply
1113 instruction. Division and shifts are open-coded only on machines that
1114 provide special support. The operations that are not open-coded use
1115 special library routines that come with GCC@.
1117 There may be pitfalls when you use @code{long long} types for function
1118 arguments, unless you declare function prototypes. If a function
1119 expects type @code{int} for its argument, and you pass a value of type
1120 @code{long long int}, confusion will result because the caller and the
1121 subroutine will disagree about the number of bytes for the argument.
1122 Likewise, if the function expects @code{long long int} and you pass
1123 @code{int}. The best way to avoid such problems is to use prototypes.
1126 @section Complex Numbers
1127 @cindex complex numbers
1128 @cindex @code{_Complex} keyword
1129 @cindex @code{__complex__} keyword
1131 ISO C99 supports complex floating data types, and as an extension GCC
1132 supports them in C89 mode and in C++, and supports complex integer data
1133 types which are not part of ISO C99. You can declare complex types
1134 using the keyword @code{_Complex}. As an extension, the older GNU
1135 keyword @code{__complex__} is also supported.
1137 For example, @samp{_Complex double x;} declares @code{x} as a
1138 variable whose real part and imaginary part are both of type
1139 @code{double}. @samp{_Complex short int y;} declares @code{y} to
1140 have real and imaginary parts of type @code{short int}; this is not
1141 likely to be useful, but it shows that the set of complex types is
1144 To write a constant with a complex data type, use the suffix @samp{i} or
1145 @samp{j} (either one; they are equivalent). For example, @code{2.5fi}
1146 has type @code{_Complex float} and @code{3i} has type
1147 @code{_Complex int}. Such a constant always has a pure imaginary
1148 value, but you can form any complex value you like by adding one to a
1149 real constant. This is a GNU extension; if you have an ISO C99
1150 conforming C library (such as GNU libc), and want to construct complex
1151 constants of floating type, you should include @code{<complex.h>} and
1152 use the macros @code{I} or @code{_Complex_I} instead.
1154 @cindex @code{__real__} keyword
1155 @cindex @code{__imag__} keyword
1156 To extract the real part of a complex-valued expression @var{exp}, write
1157 @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to
1158 extract the imaginary part. This is a GNU extension; for values of
1159 floating type, you should use the ISO C99 functions @code{crealf},
1160 @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and
1161 @code{cimagl}, declared in @code{<complex.h>} and also provided as
1162 built-in functions by GCC@.
1164 @cindex complex conjugation
1165 The operator @samp{~} performs complex conjugation when used on a value
1166 with a complex type. This is a GNU extension; for values of
1167 floating type, you should use the ISO C99 functions @code{conjf},
1168 @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also
1169 provided as built-in functions by GCC@.
1171 GCC can allocate complex automatic variables in a noncontiguous
1172 fashion; it's even possible for the real part to be in a register while
1173 the imaginary part is on the stack (or vice-versa). None of the
1174 supported debugging info formats has a way to represent noncontiguous
1175 allocation like this, so GCC describes a noncontiguous complex
1176 variable as if it were two separate variables of noncomplex type.
1177 If the variable's actual name is @code{foo}, the two fictitious
1178 variables are named @code{foo$real} and @code{foo$imag}. You can
1179 examine and set these two fictitious variables with your debugger.
1181 A future version of GDB will know how to recognize such pairs and treat
1182 them as a single variable with a complex type.
1188 ISO C99 supports floating-point numbers written not only in the usual
1189 decimal notation, such as @code{1.55e1}, but also numbers such as
1190 @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC
1191 supports this in C89 mode (except in some cases when strictly
1192 conforming) and in C++. In that format the
1193 @samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are
1194 mandatory. The exponent is a decimal number that indicates the power of
1195 2 by which the significant part will be multiplied. Thus @samp{0x1.f} is
1202 @samp{p3} multiplies it by 8, and the value of @code{0x1.fp3}
1203 is the same as @code{1.55e1}.
1205 Unlike for floating-point numbers in the decimal notation the exponent
1206 is always required in the hexadecimal notation. Otherwise the compiler
1207 would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This
1208 could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the
1209 extension for floating-point constants of type @code{float}.
1212 @section Arrays of Length Zero
1213 @cindex arrays of length zero
1214 @cindex zero-length arrays
1215 @cindex length-zero arrays
1216 @cindex flexible array members
1218 Zero-length arrays are allowed in GNU C@. They are very useful as the
1219 last element of a structure which is really a header for a variable-length
1228 struct line *thisline = (struct line *)
1229 malloc (sizeof (struct line) + this_length);
1230 thisline->length = this_length;
1233 In ISO C89, you would have to give @code{contents} a length of 1, which
1234 means either you waste space or complicate the argument to @code{malloc}.
1236 In ISO C99, you would use a @dfn{flexible array member}, which is
1237 slightly different in syntax and semantics:
1241 Flexible array members are written as @code{contents[]} without
1245 Flexible array members have incomplete type, and so the @code{sizeof}
1246 operator may not be applied. As a quirk of the original implementation
1247 of zero-length arrays, @code{sizeof} evaluates to zero.
1250 Flexible array members may only appear as the last member of a
1251 @code{struct} that is otherwise non-empty.
1254 GCC versions before 3.0 allowed zero-length arrays to be statically
1255 initialized, as if they were flexible arrays. In addition to those
1256 cases that were useful, it also allowed initializations in situations
1257 that would corrupt later data. Non-empty initialization of zero-length
1258 arrays is now treated like any case where there are more initializer
1259 elements than the array holds, in that a suitable warning about "excess
1260 elements in array" is given, and the excess elements (all of them, in
1261 this case) are ignored.
1263 Instead GCC allows static initialization of flexible array members.
1264 This is equivalent to defining a new structure containing the original
1265 structure followed by an array of sufficient size to contain the data.
1266 I.e.@: in the following, @code{f1} is constructed as if it were declared
1272 @} f1 = @{ 1, @{ 2, 3, 4 @} @};
1275 struct f1 f1; int data[3];
1276 @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @};
1280 The convenience of this extension is that @code{f1} has the desired
1281 type, eliminating the need to consistently refer to @code{f2.f1}.
1283 This has symmetry with normal static arrays, in that an array of
1284 unknown size is also written with @code{[]}.
1286 Of course, this extension only makes sense if the extra data comes at
1287 the end of a top-level object, as otherwise we would be overwriting
1288 data at subsequent offsets. To avoid undue complication and confusion
1289 with initialization of deeply nested arrays, we simply disallow any
1290 non-empty initialization except when the structure is the top-level
1291 object. For example:
1294 struct foo @{ int x; int y[]; @};
1295 struct bar @{ struct foo z; @};
1297 struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.}
1298 struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1299 struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.}
1300 struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // @r{Invalid.}
1303 @node Variable Length
1304 @section Arrays of Variable Length
1305 @cindex variable-length arrays
1306 @cindex arrays of variable length
1309 Variable-length automatic arrays are allowed in ISO C99, and as an
1310 extension GCC accepts them in C89 mode and in C++. (However, GCC's
1311 implementation of variable-length arrays does not yet conform in detail
1312 to the ISO C99 standard.) These arrays are
1313 declared like any other automatic arrays, but with a length that is not
1314 a constant expression. The storage is allocated at the point of
1315 declaration and deallocated when the brace-level is exited. For
1320 concat_fopen (char *s1, char *s2, char *mode)
1322 char str[strlen (s1) + strlen (s2) + 1];
1325 return fopen (str, mode);
1329 @cindex scope of a variable length array
1330 @cindex variable-length array scope
1331 @cindex deallocating variable length arrays
1332 Jumping or breaking out of the scope of the array name deallocates the
1333 storage. Jumping into the scope is not allowed; you get an error
1336 @cindex @code{alloca} vs variable-length arrays
1337 You can use the function @code{alloca} to get an effect much like
1338 variable-length arrays. The function @code{alloca} is available in
1339 many other C implementations (but not in all). On the other hand,
1340 variable-length arrays are more elegant.
1342 There are other differences between these two methods. Space allocated
1343 with @code{alloca} exists until the containing @emph{function} returns.
1344 The space for a variable-length array is deallocated as soon as the array
1345 name's scope ends. (If you use both variable-length arrays and
1346 @code{alloca} in the same function, deallocation of a variable-length array
1347 will also deallocate anything more recently allocated with @code{alloca}.)
1349 You can also use variable-length arrays as arguments to functions:
1353 tester (int len, char data[len][len])
1359 The length of an array is computed once when the storage is allocated
1360 and is remembered for the scope of the array in case you access it with
1363 If you want to pass the array first and the length afterward, you can
1364 use a forward declaration in the parameter list---another GNU extension.
1368 tester (int len; char data[len][len], int len)
1374 @cindex parameter forward declaration
1375 The @samp{int len} before the semicolon is a @dfn{parameter forward
1376 declaration}, and it serves the purpose of making the name @code{len}
1377 known when the declaration of @code{data} is parsed.
1379 You can write any number of such parameter forward declarations in the
1380 parameter list. They can be separated by commas or semicolons, but the
1381 last one must end with a semicolon, which is followed by the ``real''
1382 parameter declarations. Each forward declaration must match a ``real''
1383 declaration in parameter name and data type. ISO C99 does not support
1384 parameter forward declarations.
1386 @node Variadic Macros
1387 @section Macros with a Variable Number of Arguments.
1388 @cindex variable number of arguments
1389 @cindex macro with variable arguments
1390 @cindex rest argument (in macro)
1391 @cindex variadic macros
1393 In the ISO C standard of 1999, a macro can be declared to accept a
1394 variable number of arguments much as a function can. The syntax for
1395 defining the macro is similar to that of a function. Here is an
1399 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
1402 Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of
1403 such a macro, it represents the zero or more tokens until the closing
1404 parenthesis that ends the invocation, including any commas. This set of
1405 tokens replaces the identifier @code{__VA_ARGS__} in the macro body
1406 wherever it appears. See the CPP manual for more information.
1408 GCC has long supported variadic macros, and used a different syntax that
1409 allowed you to give a name to the variable arguments just like any other
1410 argument. Here is an example:
1413 #define debug(format, args...) fprintf (stderr, format, args)
1416 This is in all ways equivalent to the ISO C example above, but arguably
1417 more readable and descriptive.
1419 GNU CPP has two further variadic macro extensions, and permits them to
1420 be used with either of the above forms of macro definition.
1422 In standard C, you are not allowed to leave the variable argument out
1423 entirely; but you are allowed to pass an empty argument. For example,
1424 this invocation is invalid in ISO C, because there is no comma after
1431 GNU CPP permits you to completely omit the variable arguments in this
1432 way. In the above examples, the compiler would complain, though since
1433 the expansion of the macro still has the extra comma after the format
1436 To help solve this problem, CPP behaves specially for variable arguments
1437 used with the token paste operator, @samp{##}. If instead you write
1440 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
1443 and if the variable arguments are omitted or empty, the @samp{##}
1444 operator causes the preprocessor to remove the comma before it. If you
1445 do provide some variable arguments in your macro invocation, GNU CPP
1446 does not complain about the paste operation and instead places the
1447 variable arguments after the comma. Just like any other pasted macro
1448 argument, these arguments are not macro expanded.
1450 @node Escaped Newlines
1451 @section Slightly Looser Rules for Escaped Newlines
1452 @cindex escaped newlines
1453 @cindex newlines (escaped)
1455 Recently, the non-traditional preprocessor has relaxed its treatment of
1456 escaped newlines. Previously, the newline had to immediately follow a
1457 backslash. The current implementation allows whitespace in the form of
1458 spaces, horizontal and vertical tabs, and form feeds between the
1459 backslash and the subsequent newline. The preprocessor issues a
1460 warning, but treats it as a valid escaped newline and combines the two
1461 lines to form a single logical line. This works within comments and
1462 tokens, including multi-line strings, as well as between tokens.
1463 Comments are @emph{not} treated as whitespace for the purposes of this
1464 relaxation, since they have not yet been replaced with spaces.
1466 @node Multi-line Strings
1467 @section String Literals with Embedded Newlines
1468 @cindex multi-line string literals
1470 As an extension, GNU CPP permits string literals to cross multiple lines
1471 without escaping the embedded newlines. Each embedded newline is
1472 replaced with a single @samp{\n} character in the resulting string
1473 literal, regardless of what form the newline took originally.
1475 CPP currently allows such strings in directives as well (other than the
1476 @samp{#include} family). This is deprecated and will eventually be
1480 @section Non-Lvalue Arrays May Have Subscripts
1481 @cindex subscripting
1482 @cindex arrays, non-lvalue
1484 @cindex subscripting and function values
1485 In ISO C99, arrays that are not lvalues still decay to pointers, and
1486 may be subscripted, although they may not be modified or used after
1487 the next sequence point and the unary @samp{&} operator may not be
1488 applied to them. As an extension, GCC allows such arrays to be
1489 subscripted in C89 mode, though otherwise they do not decay to
1490 pointers outside C99 mode. For example,
1491 this is valid in GNU C though not valid in C89:
1495 struct foo @{int a[4];@};
1501 return f().a[index];
1507 @section Arithmetic on @code{void}- and Function-Pointers
1508 @cindex void pointers, arithmetic
1509 @cindex void, size of pointer to
1510 @cindex function pointers, arithmetic
1511 @cindex function, size of pointer to
1513 In GNU C, addition and subtraction operations are supported on pointers to
1514 @code{void} and on pointers to functions. This is done by treating the
1515 size of a @code{void} or of a function as 1.
1517 A consequence of this is that @code{sizeof} is also allowed on @code{void}
1518 and on function types, and returns 1.
1520 @opindex Wpointer-arith
1521 The option @option{-Wpointer-arith} requests a warning if these extensions
1525 @section Non-Constant Initializers
1526 @cindex initializers, non-constant
1527 @cindex non-constant initializers
1529 As in standard C++ and ISO C99, the elements of an aggregate initializer for an
1530 automatic variable are not required to be constant expressions in GNU C@.
1531 Here is an example of an initializer with run-time varying elements:
1534 foo (float f, float g)
1536 float beat_freqs[2] = @{ f-g, f+g @};
1541 @node Compound Literals
1542 @section Compound Literals
1543 @cindex constructor expressions
1544 @cindex initializations in expressions
1545 @cindex structures, constructor expression
1546 @cindex expressions, constructor
1547 @cindex compound literals
1548 @c The GNU C name for what C99 calls compound literals was "constructor expressions".
1550 ISO C99 supports compound literals. A compound literal looks like
1551 a cast containing an initializer. Its value is an object of the
1552 type specified in the cast, containing the elements specified in
1553 the initializer; it is an lvalue. As an extension, GCC supports
1554 compound literals in C89 mode and in C++.
1556 Usually, the specified type is a structure. Assume that
1557 @code{struct foo} and @code{structure} are declared as shown:
1560 struct foo @{int a; char b[2];@} structure;
1564 Here is an example of constructing a @code{struct foo} with a compound literal:
1567 structure = ((struct foo) @{x + y, 'a', 0@});
1571 This is equivalent to writing the following:
1575 struct foo temp = @{x + y, 'a', 0@};
1580 You can also construct an array. If all the elements of the compound literal
1581 are (made up of) simple constant expressions, suitable for use in
1582 initializers of objects of static storage duration, then the compound
1583 literal can be coerced to a pointer to its first element and used in
1584 such an initializer, as shown here:
1587 char **foo = (char *[]) @{ "x", "y", "z" @};
1590 Compound literals for scalar types and union types are is
1591 also allowed, but then the compound literal is equivalent
1594 As a GNU extension, GCC allows initialization of objects with static storage
1595 duration by compound literals (which is not possible in ISO C99, because
1596 the initializer is not a constant).
1597 It is handled as if the object was initialized only with the bracket
1598 enclosed list if compound literal's and object types match.
1599 The initializer list of the compound literal must be constant.
1600 If the object being initialized has array type of unknown size, the size is
1601 determined by compound literal's initializer list, not by the size of the
1605 static struct foo x = (struct foo) @{1, 'a', 'b'@};
1606 static int y[] = (int []) @{1, 2, 3@};
1607 static int z[] = (int [3]) @{1@};
1611 The above lines are equivalent to the following:
1613 static struct foo x = @{1, 'a', 'b'@};
1614 static int y[] = @{1, 2, 3@};
1615 static int z[] = @{1@};
1618 @node Designated Inits
1619 @section Designated Initializers
1620 @cindex initializers with labeled elements
1621 @cindex labeled elements in initializers
1622 @cindex case labels in initializers
1623 @cindex designated initializers
1625 Standard C89 requires the elements of an initializer to appear in a fixed
1626 order, the same as the order of the elements in the array or structure
1629 In ISO C99 you can give the elements in any order, specifying the array
1630 indices or structure field names they apply to, and GNU C allows this as
1631 an extension in C89 mode as well. This extension is not
1632 implemented in GNU C++.
1634 To specify an array index, write
1635 @samp{[@var{index}] =} before the element value. For example,
1638 int a[6] = @{ [4] = 29, [2] = 15 @};
1645 int a[6] = @{ 0, 0, 15, 0, 29, 0 @};
1649 The index values must be constant expressions, even if the array being
1650 initialized is automatic.
1652 An alternative syntax for this which has been obsolete since GCC 2.5 but
1653 GCC still accepts is to write @samp{[@var{index}]} before the element
1654 value, with no @samp{=}.
1656 To initialize a range of elements to the same value, write
1657 @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU
1658 extension. For example,
1661 int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @};
1665 If the value in it has side-effects, the side-effects will happen only once,
1666 not for each initialized field by the range initializer.
1669 Note that the length of the array is the highest value specified
1672 In a structure initializer, specify the name of a field to initialize
1673 with @samp{.@var{fieldname} =} before the element value. For example,
1674 given the following structure,
1677 struct point @{ int x, y; @};
1681 the following initialization
1684 struct point p = @{ .y = yvalue, .x = xvalue @};
1691 struct point p = @{ xvalue, yvalue @};
1694 Another syntax which has the same meaning, obsolete since GCC 2.5, is
1695 @samp{@var{fieldname}:}, as shown here:
1698 struct point p = @{ y: yvalue, x: xvalue @};
1702 The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a
1703 @dfn{designator}. You can also use a designator (or the obsolete colon
1704 syntax) when initializing a union, to specify which element of the union
1705 should be used. For example,
1708 union foo @{ int i; double d; @};
1710 union foo f = @{ .d = 4 @};
1714 will convert 4 to a @code{double} to store it in the union using
1715 the second element. By contrast, casting 4 to type @code{union foo}
1716 would store it into the union as the integer @code{i}, since it is
1717 an integer. (@xref{Cast to Union}.)
1719 You can combine this technique of naming elements with ordinary C
1720 initialization of successive elements. Each initializer element that
1721 does not have a designator applies to the next consecutive element of the
1722 array or structure. For example,
1725 int a[6] = @{ [1] = v1, v2, [4] = v4 @};
1732 int a[6] = @{ 0, v1, v2, 0, v4, 0 @};
1735 Labeling the elements of an array initializer is especially useful
1736 when the indices are characters or belong to an @code{enum} type.
1741 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1,
1742 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @};
1745 @cindex designator lists
1746 You can also write a series of @samp{.@var{fieldname}} and
1747 @samp{[@var{index}]} designators before an @samp{=} to specify a
1748 nested subobject to initialize; the list is taken relative to the
1749 subobject corresponding to the closest surrounding brace pair. For
1750 example, with the @samp{struct point} declaration above:
1753 struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @};
1757 If the same field is initialized multiple times, it will have value from
1758 the last initialization. If any such overridden initialization has
1759 side-effect, it is unspecified whether the side-effect happens or not.
1760 Currently, gcc will discard them and issue a warning.
1763 @section Case Ranges
1765 @cindex ranges in case statements
1767 You can specify a range of consecutive values in a single @code{case} label,
1771 case @var{low} ... @var{high}:
1775 This has the same effect as the proper number of individual @code{case}
1776 labels, one for each integer value from @var{low} to @var{high}, inclusive.
1778 This feature is especially useful for ranges of ASCII character codes:
1784 @strong{Be careful:} Write spaces around the @code{...}, for otherwise
1785 it may be parsed wrong when you use it with integer values. For example,
1800 @section Cast to a Union Type
1801 @cindex cast to a union
1802 @cindex union, casting to a
1804 A cast to union type is similar to other casts, except that the type
1805 specified is a union type. You can specify the type either with
1806 @code{union @var{tag}} or with a typedef name. A cast to union is actually
1807 a constructor though, not a cast, and hence does not yield an lvalue like
1808 normal casts. (@xref{Compound Literals}.)
1810 The types that may be cast to the union type are those of the members
1811 of the union. Thus, given the following union and variables:
1814 union foo @{ int i; double d; @};
1820 both @code{x} and @code{y} can be cast to type @code{union foo}.
1822 Using the cast as the right-hand side of an assignment to a variable of
1823 union type is equivalent to storing in a member of the union:
1828 u = (union foo) x @equiv{} u.i = x
1829 u = (union foo) y @equiv{} u.d = y
1832 You can also use the union cast as a function argument:
1835 void hack (union foo);
1837 hack ((union foo) x);
1840 @node Mixed Declarations
1841 @section Mixed Declarations and Code
1842 @cindex mixed declarations and code
1843 @cindex declarations, mixed with code
1844 @cindex code, mixed with declarations
1846 ISO C99 and ISO C++ allow declarations and code to be freely mixed
1847 within compound statements. As an extension, GCC also allows this in
1848 C89 mode. For example, you could do:
1857 Each identifier is visible from where it is declared until the end of
1858 the enclosing block.
1860 @node Function Attributes
1861 @section Declaring Attributes of Functions
1862 @cindex function attributes
1863 @cindex declaring attributes of functions
1864 @cindex functions that never return
1865 @cindex functions that have no side effects
1866 @cindex functions in arbitrary sections
1867 @cindex functions that behave like malloc
1868 @cindex @code{volatile} applied to function
1869 @cindex @code{const} applied to function
1870 @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments
1871 @cindex functions that are passed arguments in registers on the 386
1872 @cindex functions that pop the argument stack on the 386
1873 @cindex functions that do not pop the argument stack on the 386
1875 In GNU C, you declare certain things about functions called in your program
1876 which help the compiler optimize function calls and check your code more
1879 The keyword @code{__attribute__} allows you to specify special
1880 attributes when making a declaration. This keyword is followed by an
1881 attribute specification inside double parentheses. The following
1882 attributes are currently defined for functions on all targets:
1883 @code{noreturn}, @code{noinline}, @code{pure}, @code{const},
1884 @code{format}, @code{format_arg}, @code{no_instrument_function},
1885 @code{section}, @code{constructor}, @code{destructor}, @code{used},
1886 @code{unused}, @code{weak}, @code{malloc}, and @code{alias}. Several
1887 other attributes are defined for functions on particular target systems.
1888 Other attributes, including @code{section} are supported for variables
1889 declarations (@pxref{Variable Attributes}) and for types (@pxref{Type
1892 You may also specify attributes with @samp{__} preceding and following
1893 each keyword. This allows you to use them in header files without
1894 being concerned about a possible macro of the same name. For example,
1895 you may use @code{__noreturn__} instead of @code{noreturn}.
1897 @xref{Attribute Syntax}, for details of the exact syntax for using
1901 @cindex @code{noreturn} function attribute
1903 A few standard library functions, such as @code{abort} and @code{exit},
1904 cannot return. GCC knows this automatically. Some programs define
1905 their own functions that never return. You can declare them
1906 @code{noreturn} to tell the compiler this fact. For example,
1910 void fatal () __attribute__ ((noreturn));
1915 @dots{} /* @r{Print error message.} */ @dots{}
1921 The @code{noreturn} keyword tells the compiler to assume that
1922 @code{fatal} cannot return. It can then optimize without regard to what
1923 would happen if @code{fatal} ever did return. This makes slightly
1924 better code. More importantly, it helps avoid spurious warnings of
1925 uninitialized variables.
1927 Do not assume that registers saved by the calling function are
1928 restored before calling the @code{noreturn} function.
1930 It does not make sense for a @code{noreturn} function to have a return
1931 type other than @code{void}.
1933 The attribute @code{noreturn} is not implemented in GCC versions
1934 earlier than 2.5. An alternative way to declare that a function does
1935 not return, which works in the current version and in some older
1936 versions, is as follows:
1939 typedef void voidfn ();
1941 volatile voidfn fatal;
1944 @cindex @code{noinline} function attribute
1946 This function attribute prevents a function from being considered for
1949 @cindex @code{pure} function attribute
1951 Many functions have no effects except the return value and their
1952 return value depends only on the parameters and/or global variables.
1953 Such a function can be subject
1954 to common subexpression elimination and loop optimization just as an
1955 arithmetic operator would be. These functions should be declared
1956 with the attribute @code{pure}. For example,
1959 int square (int) __attribute__ ((pure));
1963 says that the hypothetical function @code{square} is safe to call
1964 fewer times than the program says.
1966 Some of common examples of pure functions are @code{strlen} or @code{memcmp}.
1967 Interesting non-pure functions are functions with infinite loops or those
1968 depending on volatile memory or other system resource, that may change between
1969 two consecutive calls (such as @code{feof} in a multithreading environment).
1971 The attribute @code{pure} is not implemented in GCC versions earlier
1973 @cindex @code{const} function attribute
1975 Many functions do not examine any values except their arguments, and
1976 have no effects except the return value. Basically this is just slightly
1977 more strict class than the @code{pure} attribute above, since function is not
1978 allowed to read global memory.
1980 @cindex pointer arguments
1981 Note that a function that has pointer arguments and examines the data
1982 pointed to must @emph{not} be declared @code{const}. Likewise, a
1983 function that calls a non-@code{const} function usually must not be
1984 @code{const}. It does not make sense for a @code{const} function to
1987 The attribute @code{const} is not implemented in GCC versions earlier
1988 than 2.5. An alternative way to declare that a function has no side
1989 effects, which works in the current version and in some older versions,
1993 typedef int intfn ();
1995 extern const intfn square;
1998 This approach does not work in GNU C++ from 2.6.0 on, since the language
1999 specifies that the @samp{const} must be attached to the return value.
2002 @item format (@var{archetype}, @var{string-index}, @var{first-to-check})
2003 @cindex @code{format} function attribute
2005 The @code{format} attribute specifies that a function takes @code{printf},
2006 @code{scanf}, @code{strftime} or @code{strfmon} style arguments which
2007 should be type-checked against a format string. For example, the
2012 my_printf (void *my_object, const char *my_format, ...)
2013 __attribute__ ((format (printf, 2, 3)));
2017 causes the compiler to check the arguments in calls to @code{my_printf}
2018 for consistency with the @code{printf} style format string argument
2021 The parameter @var{archetype} determines how the format string is
2022 interpreted, and should be @code{printf}, @code{scanf}, @code{strftime}
2023 or @code{strfmon}. (You can also use @code{__printf__},
2024 @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The
2025 parameter @var{string-index} specifies which argument is the format
2026 string argument (starting from 1), while @var{first-to-check} is the
2027 number of the first argument to check against the format string. For
2028 functions where the arguments are not available to be checked (such as
2029 @code{vprintf}), specify the third parameter as zero. In this case the
2030 compiler only checks the format string for consistency. For
2031 @code{strftime} formats, the third parameter is required to be zero.
2033 In the example above, the format string (@code{my_format}) is the second
2034 argument of the function @code{my_print}, and the arguments to check
2035 start with the third argument, so the correct parameters for the format
2036 attribute are 2 and 3.
2038 @opindex ffreestanding
2039 The @code{format} attribute allows you to identify your own functions
2040 which take format strings as arguments, so that GCC can check the
2041 calls to these functions for errors. The compiler always (unless
2042 @option{-ffreestanding} is used) checks formats
2043 for the standard library functions @code{printf}, @code{fprintf},
2044 @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime},
2045 @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such
2046 warnings are requested (using @option{-Wformat}), so there is no need to
2047 modify the header file @file{stdio.h}. In C99 mode, the functions
2048 @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and
2049 @code{vsscanf} are also checked. Except in strictly conforming C
2050 standard modes, the X/Open function @code{strfmon} is also checked.
2051 @xref{C Dialect Options,,Options Controlling C Dialect}.
2053 @item format_arg (@var{string-index})
2054 @cindex @code{format_arg} function attribute
2055 @opindex Wformat-nonliteral
2056 The @code{format_arg} attribute specifies that a function takes a format
2057 string for a @code{printf}, @code{scanf}, @code{strftime} or
2058 @code{strfmon} style function and modifies it (for example, to translate
2059 it into another language), so the result can be passed to a
2060 @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style
2061 function (with the remaining arguments to the format function the same
2062 as they would have been for the unmodified string). For example, the
2067 my_dgettext (char *my_domain, const char *my_format)
2068 __attribute__ ((format_arg (2)));
2072 causes the compiler to check the arguments in calls to a @code{printf},
2073 @code{scanf}, @code{strftime} or @code{strfmon} type function, whose
2074 format string argument is a call to the @code{my_dgettext} function, for
2075 consistency with the format string argument @code{my_format}. If the
2076 @code{format_arg} attribute had not been specified, all the compiler
2077 could tell in such calls to format functions would be that the format
2078 string argument is not constant; this would generate a warning when
2079 @option{-Wformat-nonliteral} is used, but the calls could not be checked
2080 without the attribute.
2082 The parameter @var{string-index} specifies which argument is the format
2083 string argument (starting from 1).
2085 The @code{format-arg} attribute allows you to identify your own
2086 functions which modify format strings, so that GCC can check the
2087 calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon}
2088 type function whose operands are a call to one of your own function.
2089 The compiler always treats @code{gettext}, @code{dgettext}, and
2090 @code{dcgettext} in this manner except when strict ISO C support is
2091 requested by @option{-ansi} or an appropriate @option{-std} option, or
2092 @option{-ffreestanding} is used. @xref{C Dialect Options,,Options
2093 Controlling C Dialect}.
2095 @item no_instrument_function
2096 @cindex @code{no_instrument_function} function attribute
2097 @opindex finstrument-functions
2098 If @option{-finstrument-functions} is given, profiling function calls will
2099 be generated at entry and exit of most user-compiled functions.
2100 Functions with this attribute will not be so instrumented.
2102 @item section ("@var{section-name}")
2103 @cindex @code{section} function attribute
2104 Normally, the compiler places the code it generates in the @code{text} section.
2105 Sometimes, however, you need additional sections, or you need certain
2106 particular functions to appear in special sections. The @code{section}
2107 attribute specifies that a function lives in a particular section.
2108 For example, the declaration:
2111 extern void foobar (void) __attribute__ ((section ("bar")));
2115 puts the function @code{foobar} in the @code{bar} section.
2117 Some file formats do not support arbitrary sections so the @code{section}
2118 attribute is not available on all platforms.
2119 If you need to map the entire contents of a module to a particular
2120 section, consider using the facilities of the linker instead.
2124 @cindex @code{constructor} function attribute
2125 @cindex @code{destructor} function attribute
2126 The @code{constructor} attribute causes the function to be called
2127 automatically before execution enters @code{main ()}. Similarly, the
2128 @code{destructor} attribute causes the function to be called
2129 automatically after @code{main ()} has completed or @code{exit ()} has
2130 been called. Functions with these attributes are useful for
2131 initializing data that will be used implicitly during the execution of
2134 These attributes are not currently implemented for Objective-C@.
2136 @cindex @code{unused} attribute.
2138 This attribute, attached to a function, means that the function is meant
2139 to be possibly unused. GCC will not produce a warning for this
2140 function. GNU C++ does not currently support this attribute as
2141 definitions without parameters are valid in C++.
2143 @cindex @code{used} attribute.
2145 This attribute, attached to a function, means that code must be emitted
2146 for the function even if it appears that the function is not referenced.
2147 This is useful, for example, when the function is referenced only in
2151 @cindex @code{weak} attribute
2152 The @code{weak} attribute causes the declaration to be emitted as a weak
2153 symbol rather than a global. This is primarily useful in defining
2154 library functions which can be overridden in user code, though it can
2155 also be used with non-function declarations. Weak symbols are supported
2156 for ELF targets, and also for a.out targets when using the GNU assembler
2160 @cindex @code{malloc} attribute
2161 The @code{malloc} attribute is used to tell the compiler that a function
2162 may be treated as if it were the malloc function. The compiler assumes
2163 that calls to malloc result in a pointers that cannot alias anything.
2164 This will often improve optimization.
2166 @item alias ("@var{target}")
2167 @cindex @code{alias} attribute
2168 The @code{alias} attribute causes the declaration to be emitted as an
2169 alias for another symbol, which must be specified. For instance,
2172 void __f () @{ /* do something */; @}
2173 void f () __attribute__ ((weak, alias ("__f")));
2176 declares @samp{f} to be a weak alias for @samp{__f}. In C++, the
2177 mangled name for the target must be used.
2179 Not all target machines support this attribute.
2181 @item regparm (@var{number})
2182 @cindex functions that are passed arguments in registers on the 386
2183 On the Intel 386, the @code{regparm} attribute causes the compiler to
2184 pass up to @var{number} integer arguments in registers EAX,
2185 EDX, and ECX instead of on the stack. Functions that take a
2186 variable number of arguments will continue to be passed all of their
2187 arguments on the stack.
2190 @cindex functions that pop the argument stack on the 386
2191 On the Intel 386, the @code{stdcall} attribute causes the compiler to
2192 assume that the called function will pop off the stack space used to
2193 pass arguments, unless it takes a variable number of arguments.
2195 The PowerPC compiler for Windows NT currently ignores the @code{stdcall}
2199 @cindex functions that do pop the argument stack on the 386
2201 On the Intel 386, the @code{cdecl} attribute causes the compiler to
2202 assume that the calling function will pop off the stack space used to
2203 pass arguments. This is
2204 useful to override the effects of the @option{-mrtd} switch.
2206 The PowerPC compiler for Windows NT currently ignores the @code{cdecl}
2210 @cindex functions called via pointer on the RS/6000 and PowerPC
2211 On the RS/6000 and PowerPC, the @code{longcall} attribute causes the
2212 compiler to always call the function via a pointer, so that functions
2213 which reside further than 64 megabytes (67,108,864 bytes) from the
2214 current location can be called.
2216 @item long_call/short_call
2217 @cindex indirect calls on ARM
2218 This attribute allows to specify how to call a particular function on
2219 ARM@. Both attributes override the @option{-mlong-calls} (@pxref{ARM Options})
2220 command line switch and @code{#pragma long_calls} settings. The
2221 @code{long_call} attribute causes the compiler to always call the
2222 function by first loading its address into a register and then using the
2223 contents of that register. The @code{short_call} attribute always places
2224 the offset to the function from the call site into the @samp{BL}
2225 instruction directly.
2228 @cindex functions which are imported from a dll on PowerPC Windows NT
2229 On the PowerPC running Windows NT, the @code{dllimport} attribute causes
2230 the compiler to call the function via a global pointer to the function
2231 pointer that is set up by the Windows NT dll library. The pointer name
2232 is formed by combining @code{__imp_} and the function name.
2235 @cindex functions which are exported from a dll on PowerPC Windows NT
2236 On the PowerPC running Windows NT, the @code{dllexport} attribute causes
2237 the compiler to provide a global pointer to the function pointer, so
2238 that it can be called with the @code{dllimport} attribute. The pointer
2239 name is formed by combining @code{__imp_} and the function name.
2241 @item exception (@var{except-func} [, @var{except-arg}])
2242 @cindex functions which specify exception handling on PowerPC Windows NT
2243 On the PowerPC running Windows NT, the @code{exception} attribute causes
2244 the compiler to modify the structured exception table entry it emits for
2245 the declared function. The string or identifier @var{except-func} is
2246 placed in the third entry of the structured exception table. It
2247 represents a function, which is called by the exception handling
2248 mechanism if an exception occurs. If it was specified, the string or
2249 identifier @var{except-arg} is placed in the fourth entry of the
2250 structured exception table.
2252 @item function_vector
2253 @cindex calling functions through the function vector on the H8/300 processors
2254 Use this option on the H8/300 and H8/300H to indicate that the specified
2255 function should be called through the function vector. Calling a
2256 function through the function vector will reduce code size, however;
2257 the function vector has a limited size (maximum 128 entries on the H8/300
2258 and 64 entries on the H8/300H) and shares space with the interrupt vector.
2260 You must use GAS and GLD from GNU binutils version 2.7 or later for
2261 this option to work correctly.
2264 @cindex interrupt handler functions
2265 Use this option on the ARM, AVR and M32R/D ports to indicate that the
2266 specified function is an interrupt handler. The compiler will generate
2267 function entry and exit sequences suitable for use in an interrupt
2268 handler when this attribute is present.
2270 Note, interrupt handlers for the H8/300, H8/300H and SH processors can
2271 be specified via the @code{interrupt_handler} attribute.
2273 Note, on the AVR interrupts will be enabled inside the function.
2275 Note, for the ARM you can specify the kind of interrupt to be handled by
2276 adding an optional parameter to the interrupt attribute like this:
2279 void f () __attribute__ ((interrupt ("IRQ")));
2282 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF@.
2284 @item interrupt_handler
2285 @cindex interrupt handler functions on the H8/300 and SH processors
2286 Use this option on the H8/300, H8/300H and SH to indicate that the
2287 specified function is an interrupt handler. The compiler will generate
2288 function entry and exit sequences suitable for use in an interrupt
2289 handler when this attribute is present.
2292 Use this option on the SH to indicate an @code{interrupt_handler}
2293 function should switch to an alternate stack. It expects a string
2294 argument that names a global variable holding the address of the
2299 void f () __attribute__ ((interrupt_handler,
2300 sp_switch ("alt_stack")));
2304 Use this option on the SH for an @code{interrupt_handle} to return using
2305 @code{trapa} instead of @code{rte}. This attribute expects an integer
2306 argument specifying the trap number to be used.
2309 @cindex eight bit data on the H8/300 and H8/300H
2310 Use this option on the H8/300 and H8/300H to indicate that the specified
2311 variable should be placed into the eight bit data section.
2312 The compiler will generate more efficient code for certain operations
2313 on data in the eight bit data area. Note the eight bit data area is limited to
2316 You must use GAS and GLD from GNU binutils version 2.7 or later for
2317 this option to work correctly.
2320 @cindex tiny data section on the H8/300H
2321 Use this option on the H8/300H to indicate that the specified
2322 variable should be placed into the tiny data section.
2323 The compiler will generate more efficient code for loads and stores
2324 on data in the tiny data section. Note the tiny data area is limited to
2325 slightly under 32kbytes of data.
2328 @cindex signal handler functions on the AVR processors
2329 Use this option on the AVR to indicate that the specified
2330 function is an signal handler. The compiler will generate function
2331 entry and exit sequences suitable for use in an signal handler when this
2332 attribute is present. Interrupts will be disabled inside function.
2335 @cindex function without a prologue/epilogue code
2336 Use this option on the ARM or AVR ports to indicate that the specified
2337 function do not need prologue/epilogue sequences generated by the
2338 compiler. It is up to the programmer to provide these sequences.
2340 @item model (@var{model-name})
2341 @cindex function addressability on the M32R/D
2342 Use this attribute on the M32R/D to set the addressability of an object,
2343 and the code generated for a function.
2344 The identifier @var{model-name} is one of @code{small}, @code{medium},
2345 or @code{large}, representing each of the code models.
2347 Small model objects live in the lower 16MB of memory (so that their
2348 addresses can be loaded with the @code{ld24} instruction), and are
2349 callable with the @code{bl} instruction.
2351 Medium model objects may live anywhere in the 32-bit address space (the
2352 compiler will generate @code{seth/add3} instructions to load their addresses),
2353 and are callable with the @code{bl} instruction.
2355 Large model objects may live anywhere in the 32-bit address space (the
2356 compiler will generate @code{seth/add3} instructions to load their addresses),
2357 and may not be reachable with the @code{bl} instruction (the compiler will
2358 generate the much slower @code{seth/add3/jl} instruction sequence).
2362 You can specify multiple attributes in a declaration by separating them
2363 by commas within the double parentheses or by immediately following an
2364 attribute declaration with another attribute declaration.
2366 @cindex @code{#pragma}, reason for not using
2367 @cindex pragma, reason for not using
2368 Some people object to the @code{__attribute__} feature, suggesting that
2369 ISO C's @code{#pragma} should be used instead. At the time
2370 @code{__attribute__} was designed, there were two reasons for not doing
2375 It is impossible to generate @code{#pragma} commands from a macro.
2378 There is no telling what the same @code{#pragma} might mean in another
2382 These two reasons applied to almost any application that might have been
2383 proposed for @code{#pragma}. It was basically a mistake to use
2384 @code{#pragma} for @emph{anything}.
2386 The ISO C99 standard includes @code{_Pragma}, which now allows pragmas
2387 to be generated from macros. In addition, a @code{#pragma GCC}
2388 namespace is now in use for GCC-specific pragmas. However, it has been
2389 found convenient to use @code{__attribute__} to achieve a natural
2390 attachment of attributes to their corresponding declarations, whereas
2391 @code{#pragma GCC} is of use for constructs that do not naturally form
2392 part of the grammar. @xref{Other Directives,,Miscellaneous
2393 Preprocessing Directives, cpp, The C Preprocessor}.
2395 @node Attribute Syntax
2396 @section Attribute Syntax
2397 @cindex attribute syntax
2399 This section describes the syntax with which @code{__attribute__} may be
2400 used, and the constructs to which attribute specifiers bind, for the C
2401 language. Some details may vary for C++ and Objective-C@. Because of
2402 infelicities in the grammar for attributes, some forms described here
2403 may not be successfully parsed in all cases.
2405 There are some problems with the semantics of attributes in C++. For
2406 example, there are no manglings for attributes, although they may affect
2407 code generation, so problems may arise when attributed types are used in
2408 conjunction with templates or overloading. Similarly, @code{typeid}
2409 does not distinguish between types with different attributes. Support
2410 for attributes in C++ may be restricted in future to attributes on
2411 declarations only, but not on nested declarators.
2413 @xref{Function Attributes}, for details of the semantics of attributes
2414 applying to functions. @xref{Variable Attributes}, for details of the
2415 semantics of attributes applying to variables. @xref{Type Attributes},
2416 for details of the semantics of attributes applying to structure, union
2417 and enumerated types.
2419 An @dfn{attribute specifier} is of the form
2420 @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list}
2421 is a possibly empty comma-separated sequence of @dfn{attributes}, where
2422 each attribute is one of the following:
2426 Empty. Empty attributes are ignored.
2429 A word (which may be an identifier such as @code{unused}, or a reserved
2430 word such as @code{const}).
2433 A word, followed by, in parentheses, parameters for the attribute.
2434 These parameters take one of the following forms:
2438 An identifier. For example, @code{mode} attributes use this form.
2441 An identifier followed by a comma and a non-empty comma-separated list
2442 of expressions. For example, @code{format} attributes use this form.
2445 A possibly empty comma-separated list of expressions. For example,
2446 @code{format_arg} attributes use this form with the list being a single
2447 integer constant expression, and @code{alias} attributes use this form
2448 with the list being a single string constant.
2452 An @dfn{attribute specifier list} is a sequence of one or more attribute
2453 specifiers, not separated by any other tokens.
2455 An attribute specifier list may appear after the colon following a
2456 label, other than a @code{case} or @code{default} label. The only
2457 attribute it makes sense to use after a label is @code{unused}. This
2458 feature is intended for code generated by programs which contains labels
2459 that may be unused but which is compiled with @option{-Wall}. It would
2460 not normally be appropriate to use in it human-written code, though it
2461 could be useful in cases where the code that jumps to the label is
2462 contained within an @code{#ifdef} conditional.
2464 An attribute specifier list may appear as part of a @code{struct},
2465 @code{union} or @code{enum} specifier. It may go either immediately
2466 after the @code{struct}, @code{union} or @code{enum} keyword, or after
2467 the closing brace. It is ignored if the content of the structure, union
2468 or enumerated type is not defined in the specifier in which the
2469 attribute specifier list is used---that is, in usages such as
2470 @code{struct __attribute__((foo)) bar} with no following opening brace.
2471 Where attribute specifiers follow the closing brace, they are considered
2472 to relate to the structure, union or enumerated type defined, not to any
2473 enclosing declaration the type specifier appears in, and the type
2474 defined is not complete until after the attribute specifiers.
2475 @c Otherwise, there would be the following problems: a shift/reduce
2476 @c conflict between attributes binding the struct/union/enum and
2477 @c binding to the list of specifiers/qualifiers; and "aligned"
2478 @c attributes could use sizeof for the structure, but the size could be
2479 @c changed later by "packed" attributes.
2481 Otherwise, an attribute specifier appears as part of a declaration,
2482 counting declarations of unnamed parameters and type names, and relates
2483 to that declaration (which may be nested in another declaration, for
2484 example in the case of a parameter declaration), or to a particular declarator
2485 within a declaration. Where an
2486 attribute specifier is applied to a parameter declared as a function or
2487 an array, it should apply to the function or array rather than the
2488 pointer to which the parameter is implicitly converted, but this is not
2489 yet correctly implemented.
2491 Any list of specifiers and qualifiers at the start of a declaration may
2492 contain attribute specifiers, whether or not such a list may in that
2493 context contain storage class specifiers. (Some attributes, however,
2494 are essentially in the nature of storage class specifiers, and only make
2495 sense where storage class specifiers may be used; for example,
2496 @code{section}.) There is one necessary limitation to this syntax: the
2497 first old-style parameter declaration in a function definition cannot
2498 begin with an attribute specifier, because such an attribute applies to
2499 the function instead by syntax described below (which, however, is not
2500 yet implemented in this case). In some other cases, attribute
2501 specifiers are permitted by this grammar but not yet supported by the
2502 compiler. All attribute specifiers in this place relate to the
2503 declaration as a whole. In the obsolescent usage where a type of
2504 @code{int} is implied by the absence of type specifiers, such a list of
2505 specifiers and qualifiers may be an attribute specifier list with no
2506 other specifiers or qualifiers.
2508 An attribute specifier list may appear immediately before a declarator
2509 (other than the first) in a comma-separated list of declarators in a
2510 declaration of more than one identifier using a single list of
2511 specifiers and qualifiers. Such attribute specifiers apply
2512 only to the identifier before whose declarator they appear. For
2516 __attribute__((noreturn)) void d0 (void),
2517 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
2522 the @code{noreturn} attribute applies to all the functions
2523 declared; the @code{format} attribute only applies to @code{d1}.
2525 An attribute specifier list may appear immediately before the comma,
2526 @code{=} or semicolon terminating the declaration of an identifier other
2527 than a function definition. At present, such attribute specifiers apply
2528 to the declared object or function, but in future they may attach to the
2529 outermost adjacent declarator. In simple cases there is no difference,
2530 but, for example, in
2533 void (****f)(void) __attribute__((noreturn));
2537 at present the @code{noreturn} attribute applies to @code{f}, which
2538 causes a warning since @code{f} is not a function, but in future it may
2539 apply to the function @code{****f}. The precise semantics of what
2540 attributes in such cases will apply to are not yet specified. Where an
2541 assembler name for an object or function is specified (@pxref{Asm
2542 Labels}), at present the attribute must follow the @code{asm}
2543 specification; in future, attributes before the @code{asm} specification
2544 may apply to the adjacent declarator, and those after it to the declared
2547 An attribute specifier list may, in future, be permitted to appear after
2548 the declarator in a function definition (before any old-style parameter
2549 declarations or the function body).
2551 Attribute specifiers may be mixed with type qualifiers appearing inside
2552 the @code{[]} of a parameter array declarator, in the C99 construct by
2553 which such qualifiers are applied to the pointer to which the array is
2554 implicitly converted. Such attribute specifiers apply to the pointer,
2555 not to the array, but at present this is not implemented and they are
2558 An attribute specifier list may appear at the start of a nested
2559 declarator. At present, there are some limitations in this usage: the
2560 attributes correctly apply to the declarator, but for most individual
2561 attributes the semantics this implies are not implemented.
2562 When attribute specifiers follow the @code{*} of a pointer
2563 declarator, they may be mixed with any type qualifiers present.
2564 The following describes the formal semantics of this syntax. It will make the
2565 most sense if you are familiar with the formal specification of
2566 declarators in the ISO C standard.
2568 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T
2569 D1}, where @code{T} contains declaration specifiers that specify a type
2570 @var{Type} (such as @code{int}) and @code{D1} is a declarator that
2571 contains an identifier @var{ident}. The type specified for @var{ident}
2572 for derived declarators whose type does not include an attribute
2573 specifier is as in the ISO C standard.
2575 If @code{D1} has the form @code{( @var{attribute-specifier-list} D )},
2576 and the declaration @code{T D} specifies the type
2577 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2578 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2579 @var{attribute-specifier-list} @var{Type}'' for @var{ident}.
2581 If @code{D1} has the form @code{*
2582 @var{type-qualifier-and-attribute-specifier-list} D}, and the
2583 declaration @code{T D} specifies the type
2584 ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then
2585 @code{T D1} specifies the type ``@var{derived-declarator-type-list}
2586 @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for
2592 void (__attribute__((noreturn)) ****f) (void);
2596 specifies the type ``pointer to pointer to pointer to pointer to
2597 non-returning function returning @code{void}''. As another example,
2600 char *__attribute__((aligned(8))) *f;
2604 specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''.
2605 Note again that this does not work with most attributes; for example,
2606 the usage of @samp{aligned} and @samp{noreturn} attributes given above
2607 is not yet supported.
2609 For compatibility with existing code written for compiler versions that
2610 did not implement attributes on nested declarators, some laxity is
2611 allowed in the placing of attributes. If an attribute that only applies
2612 to types is applied to a declaration, it will be treated as applying to
2613 the type of that declaration. If an attribute that only applies to
2614 declarations is applied to the type of a declaration, it will be treated
2615 as applying to that declaration; and, for compatibility with code
2616 placing the attributes immediately before the identifier declared, such
2617 an attribute applied to a function return type will be treated as
2618 applying to the function type, and such an attribute applied to an array
2619 element type will be treated as applying to the array type. If an
2620 attribute that only applies to function types is applied to a
2621 pointer-to-function type, it will be treated as applying to the pointer
2622 target type; if such an attribute is applied to a function return type
2623 that is not a pointer-to-function type, it will be treated as applying
2624 to the function type.
2626 @node Function Prototypes
2627 @section Prototypes and Old-Style Function Definitions
2628 @cindex function prototype declarations
2629 @cindex old-style function definitions
2630 @cindex promotion of formal parameters
2632 GNU C extends ISO C to allow a function prototype to override a later
2633 old-style non-prototype definition. Consider the following example:
2636 /* @r{Use prototypes unless the compiler is old-fashioned.} */
2643 /* @r{Prototype function declaration.} */
2644 int isroot P((uid_t));
2646 /* @r{Old-style function definition.} */
2648 isroot (x) /* ??? lossage here ??? */
2655 Suppose the type @code{uid_t} happens to be @code{short}. ISO C does
2656 not allow this example, because subword arguments in old-style
2657 non-prototype definitions are promoted. Therefore in this example the
2658 function definition's argument is really an @code{int}, which does not
2659 match the prototype argument type of @code{short}.
2661 This restriction of ISO C makes it hard to write code that is portable
2662 to traditional C compilers, because the programmer does not know
2663 whether the @code{uid_t} type is @code{short}, @code{int}, or
2664 @code{long}. Therefore, in cases like these GNU C allows a prototype
2665 to override a later old-style definition. More precisely, in GNU C, a
2666 function prototype argument type overrides the argument type specified
2667 by a later old-style definition if the former type is the same as the
2668 latter type before promotion. Thus in GNU C the above example is
2669 equivalent to the following:
2682 GNU C++ does not support old-style function definitions, so this
2683 extension is irrelevant.
2686 @section C++ Style Comments
2688 @cindex C++ comments
2689 @cindex comments, C++ style
2691 In GNU C, you may use C++ style comments, which start with @samp{//} and
2692 continue until the end of the line. Many other C implementations allow
2693 such comments, and they are likely to be in a future C standard.
2694 However, C++ style comments are not recognized if you specify
2695 @w{@option{-ansi}}, a @option{-std} option specifying a version of ISO C
2696 before C99, or @w{@option{-traditional}}, since they are incompatible
2697 with traditional constructs like @code{dividend//*comment*/divisor}.
2700 @section Dollar Signs in Identifier Names
2702 @cindex dollar signs in identifier names
2703 @cindex identifier names, dollar signs in
2705 In GNU C, you may normally use dollar signs in identifier names.
2706 This is because many traditional C implementations allow such identifiers.
2707 However, dollar signs in identifiers are not supported on a few target
2708 machines, typically because the target assembler does not allow them.
2710 @node Character Escapes
2711 @section The Character @key{ESC} in Constants
2713 You can use the sequence @samp{\e} in a string or character constant to
2714 stand for the ASCII character @key{ESC}.
2717 @section Inquiring on Alignment of Types or Variables
2719 @cindex type alignment
2720 @cindex variable alignment
2722 The keyword @code{__alignof__} allows you to inquire about how an object
2723 is aligned, or the minimum alignment usually required by a type. Its
2724 syntax is just like @code{sizeof}.
2726 For example, if the target machine requires a @code{double} value to be
2727 aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8.
2728 This is true on many RISC machines. On more traditional machine
2729 designs, @code{__alignof__ (double)} is 4 or even 2.
2731 Some machines never actually require alignment; they allow reference to any
2732 data type even at an odd addresses. For these machines, @code{__alignof__}
2733 reports the @emph{recommended} alignment of a type.
2735 When the operand of @code{__alignof__} is an lvalue rather than a type, the
2736 value is the largest alignment that the lvalue is known to have. It may
2737 have this alignment as a result of its data type, or because it is part of
2738 a structure and inherits alignment from that structure. For example, after
2742 struct foo @{ int x; char y; @} foo1;
2746 the value of @code{__alignof__ (foo1.y)} is probably 2 or 4, the same as
2747 @code{__alignof__ (int)}, even though the data type of @code{foo1.y}
2748 does not itself demand any alignment.
2750 It is an error to ask for the alignment of an incomplete type.
2752 A related feature which lets you specify the alignment of an object is
2753 @code{__attribute__ ((aligned (@var{alignment})))}; see the following
2756 @node Variable Attributes
2757 @section Specifying Attributes of Variables
2758 @cindex attribute of variables
2759 @cindex variable attributes
2761 The keyword @code{__attribute__} allows you to specify special
2762 attributes of variables or structure fields. This keyword is followed
2763 by an attribute specification inside double parentheses. Nine
2764 attributes are currently defined for variables: @code{aligned},
2765 @code{mode}, @code{nocommon}, @code{packed}, @code{section},
2766 @code{transparent_union}, @code{unused}, @code{vector_size}, and
2767 @code{weak}. Some other attributes are defined for variables on
2768 particular target systems. Other attributes are available for functions
2769 (@pxref{Function Attributes}) and for types (@pxref{Type Attributes}).
2770 Other front ends might define more attributes (@pxref{C++
2771 Extensions,,Extensions to the C++ Language}).
2773 You may also specify attributes with @samp{__} preceding and following
2774 each keyword. This allows you to use them in header files without
2775 being concerned about a possible macro of the same name. For example,
2776 you may use @code{__aligned__} instead of @code{aligned}.
2778 @xref{Attribute Syntax}, for details of the exact syntax for using
2782 @cindex @code{aligned} attribute
2783 @item aligned (@var{alignment})
2784 This attribute specifies a minimum alignment for the variable or
2785 structure field, measured in bytes. For example, the declaration:
2788 int x __attribute__ ((aligned (16))) = 0;
2792 causes the compiler to allocate the global variable @code{x} on a
2793 16-byte boundary. On a 68040, this could be used in conjunction with
2794 an @code{asm} expression to access the @code{move16} instruction which
2795 requires 16-byte aligned operands.
2797 You can also specify the alignment of structure fields. For example, to
2798 create a double-word aligned @code{int} pair, you could write:
2801 struct foo @{ int x[2] __attribute__ ((aligned (8))); @};
2805 This is an alternative to creating a union with a @code{double} member
2806 that forces the union to be double-word aligned.
2808 It is not possible to specify the alignment of functions; the alignment
2809 of functions is determined by the machine's requirements and cannot be
2810 changed. You cannot specify alignment for a typedef name because such a
2811 name is just an alias, not a distinct type.
2813 As in the preceding examples, you can explicitly specify the alignment
2814 (in bytes) that you wish the compiler to use for a given variable or
2815 structure field. Alternatively, you can leave out the alignment factor
2816 and just ask the compiler to align a variable or field to the maximum
2817 useful alignment for the target machine you are compiling for. For
2818 example, you could write:
2821 short array[3] __attribute__ ((aligned));
2824 Whenever you leave out the alignment factor in an @code{aligned} attribute
2825 specification, the compiler automatically sets the alignment for the declared
2826 variable or field to the largest alignment which is ever used for any data
2827 type on the target machine you are compiling for. Doing this can often make
2828 copy operations more efficient, because the compiler can use whatever
2829 instructions copy the biggest chunks of memory when performing copies to
2830 or from the variables or fields that you have aligned this way.
2832 The @code{aligned} attribute can only increase the alignment; but you
2833 can decrease it by specifying @code{packed} as well. See below.
2835 Note that the effectiveness of @code{aligned} attributes may be limited
2836 by inherent limitations in your linker. On many systems, the linker is
2837 only able to arrange for variables to be aligned up to a certain maximum
2838 alignment. (For some linkers, the maximum supported alignment may
2839 be very very small.) If your linker is only able to align variables
2840 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
2841 in an @code{__attribute__} will still only provide you with 8 byte
2842 alignment. See your linker documentation for further information.
2844 @item mode (@var{mode})
2845 @cindex @code{mode} attribute
2846 This attribute specifies the data type for the declaration---whichever
2847 type corresponds to the mode @var{mode}. This in effect lets you
2848 request an integer or floating point type according to its width.
2850 You may also specify a mode of @samp{byte} or @samp{__byte__} to
2851 indicate the mode corresponding to a one-byte integer, @samp{word} or
2852 @samp{__word__} for the mode of a one-word integer, and @samp{pointer}
2853 or @samp{__pointer__} for the mode used to represent pointers.
2856 @cindex @code{nocommon} attribute
2858 This attribute specifies requests GCC not to place a variable
2859 ``common'' but instead to allocate space for it directly. If you
2860 specify the @option{-fno-common} flag, GCC will do this for all
2863 Specifying the @code{nocommon} attribute for a variable provides an
2864 initialization of zeros. A variable may only be initialized in one
2868 @cindex @code{packed} attribute
2869 The @code{packed} attribute specifies that a variable or structure field
2870 should have the smallest possible alignment---one byte for a variable,
2871 and one bit for a field, unless you specify a larger value with the
2872 @code{aligned} attribute.
2874 Here is a structure in which the field @code{x} is packed, so that it
2875 immediately follows @code{a}:
2881 int x[2] __attribute__ ((packed));
2885 @item section ("@var{section-name}")
2886 @cindex @code{section} variable attribute
2887 Normally, the compiler places the objects it generates in sections like
2888 @code{data} and @code{bss}. Sometimes, however, you need additional sections,
2889 or you need certain particular variables to appear in special sections,
2890 for example to map to special hardware. The @code{section}
2891 attribute specifies that a variable (or function) lives in a particular
2892 section. For example, this small program uses several specific section names:
2895 struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @};
2896 struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @};
2897 char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @};
2898 int init_data __attribute__ ((section ("INITDATA"))) = 0;
2902 /* Initialize stack pointer */
2903 init_sp (stack + sizeof (stack));
2905 /* Initialize initialized data */
2906 memcpy (&init_data, &data, &edata - &data);
2908 /* Turn on the serial ports */
2915 Use the @code{section} attribute with an @emph{initialized} definition
2916 of a @emph{global} variable, as shown in the example. GCC issues
2917 a warning and otherwise ignores the @code{section} attribute in
2918 uninitialized variable declarations.
2920 You may only use the @code{section} attribute with a fully initialized
2921 global definition because of the way linkers work. The linker requires
2922 each object be defined once, with the exception that uninitialized
2923 variables tentatively go in the @code{common} (or @code{bss}) section
2924 and can be multiply ``defined''. You can force a variable to be
2925 initialized with the @option{-fno-common} flag or the @code{nocommon}
2928 Some file formats do not support arbitrary sections so the @code{section}
2929 attribute is not available on all platforms.
2930 If you need to map the entire contents of a module to a particular
2931 section, consider using the facilities of the linker instead.
2934 @cindex @code{shared} variable attribute
2935 On Windows NT, in addition to putting variable definitions in a named
2936 section, the section can also be shared among all running copies of an
2937 executable or DLL@. For example, this small program defines shared data
2938 by putting it in a named section @code{shared} and marking the section
2942 int foo __attribute__((section ("shared"), shared)) = 0;
2947 /* Read and write foo. All running
2948 copies see the same value. */
2954 You may only use the @code{shared} attribute along with @code{section}
2955 attribute with a fully initialized global definition because of the way
2956 linkers work. See @code{section} attribute for more information.
2958 The @code{shared} attribute is only available on Windows NT@.
2960 @item transparent_union
2961 This attribute, attached to a function parameter which is a union, means
2962 that the corresponding argument may have the type of any union member,
2963 but the argument is passed as if its type were that of the first union
2964 member. For more details see @xref{Type Attributes}. You can also use
2965 this attribute on a @code{typedef} for a union data type; then it
2966 applies to all function parameters with that type.
2969 This attribute, attached to a variable, means that the variable is meant
2970 to be possibly unused. GCC will not produce a warning for this
2973 @item vector_size (@var{bytes})
2974 This attribute specifies the vector size for the variable, measured in
2975 bytes. For example, the declaration:
2978 int foo __attribute__ ((vector_size (16)));
2982 causes the compiler to set the mode for @code{foo}, to be 16 bytes,
2983 divided into @code{int} sized units. Assuming a 32-bit int (a vector of
2984 4 units of 4 bytes), the corresponding mode of @code{foo} will be V4SI@.
2986 This attribute is only applicable to integral and float scalars,
2987 although arrays, pointers, and function return values are allowed in
2988 conjunction with this construct.
2990 Aggregates with this attribute are invalid, even if they are of the same
2991 size as a corresponding scalar. For example, the declaration:
2994 struct S @{ int a; @};
2995 struct S __attribute__ ((vector_size (16))) foo;
2999 is invalid even if the size of the structure is the same as the size of
3003 The @code{weak} attribute is described in @xref{Function Attributes}.
3005 @item model (@var{model-name})
3006 @cindex variable addressability on the M32R/D
3007 Use this attribute on the M32R/D to set the addressability of an object.
3008 The identifier @var{model-name} is one of @code{small}, @code{medium},
3009 or @code{large}, representing each of the code models.
3011 Small model objects live in the lower 16MB of memory (so that their
3012 addresses can be loaded with the @code{ld24} instruction).
3014 Medium and large model objects may live anywhere in the 32-bit address space
3015 (the compiler will generate @code{seth/add3} instructions to load their
3020 To specify multiple attributes, separate them by commas within the
3021 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3024 @node Type Attributes
3025 @section Specifying Attributes of Types
3026 @cindex attribute of types
3027 @cindex type attributes
3029 The keyword @code{__attribute__} allows you to specify special
3030 attributes of @code{struct} and @code{union} types when you define such
3031 types. This keyword is followed by an attribute specification inside
3032 double parentheses. Four attributes are currently defined for types:
3033 @code{aligned}, @code{packed}, @code{transparent_union}, and @code{unused}.
3034 Other attributes are defined for functions (@pxref{Function Attributes}) and
3035 for variables (@pxref{Variable Attributes}).
3037 You may also specify any one of these attributes with @samp{__}
3038 preceding and following its keyword. This allows you to use these
3039 attributes in header files without being concerned about a possible
3040 macro of the same name. For example, you may use @code{__aligned__}
3041 instead of @code{aligned}.
3043 You may specify the @code{aligned} and @code{transparent_union}
3044 attributes either in a @code{typedef} declaration or just past the
3045 closing curly brace of a complete enum, struct or union type
3046 @emph{definition} and the @code{packed} attribute only past the closing
3047 brace of a definition.
3049 You may also specify attributes between the enum, struct or union
3050 tag and the name of the type rather than after the closing brace.
3052 @xref{Attribute Syntax}, for details of the exact syntax for using
3056 @cindex @code{aligned} attribute
3057 @item aligned (@var{alignment})
3058 This attribute specifies a minimum alignment (in bytes) for variables
3059 of the specified type. For example, the declarations:
3062 struct S @{ short f[3]; @} __attribute__ ((aligned (8)));
3063 typedef int more_aligned_int __attribute__ ((aligned (8)));
3067 force the compiler to insure (as far as it can) that each variable whose
3068 type is @code{struct S} or @code{more_aligned_int} will be allocated and
3069 aligned @emph{at least} on a 8-byte boundary. On a Sparc, having all
3070 variables of type @code{struct S} aligned to 8-byte boundaries allows
3071 the compiler to use the @code{ldd} and @code{std} (doubleword load and
3072 store) instructions when copying one variable of type @code{struct S} to
3073 another, thus improving run-time efficiency.
3075 Note that the alignment of any given @code{struct} or @code{union} type
3076 is required by the ISO C standard to be at least a perfect multiple of
3077 the lowest common multiple of the alignments of all of the members of
3078 the @code{struct} or @code{union} in question. This means that you @emph{can}
3079 effectively adjust the alignment of a @code{struct} or @code{union}
3080 type by attaching an @code{aligned} attribute to any one of the members
3081 of such a type, but the notation illustrated in the example above is a
3082 more obvious, intuitive, and readable way to request the compiler to
3083 adjust the alignment of an entire @code{struct} or @code{union} type.
3085 As in the preceding example, you can explicitly specify the alignment
3086 (in bytes) that you wish the compiler to use for a given @code{struct}
3087 or @code{union} type. Alternatively, you can leave out the alignment factor
3088 and just ask the compiler to align a type to the maximum
3089 useful alignment for the target machine you are compiling for. For
3090 example, you could write:
3093 struct S @{ short f[3]; @} __attribute__ ((aligned));
3096 Whenever you leave out the alignment factor in an @code{aligned}
3097 attribute specification, the compiler automatically sets the alignment
3098 for the type to the largest alignment which is ever used for any data
3099 type on the target machine you are compiling for. Doing this can often
3100 make copy operations more efficient, because the compiler can use
3101 whatever instructions copy the biggest chunks of memory when performing
3102 copies to or from the variables which have types that you have aligned
3105 In the example above, if the size of each @code{short} is 2 bytes, then
3106 the size of the entire @code{struct S} type is 6 bytes. The smallest
3107 power of two which is greater than or equal to that is 8, so the
3108 compiler sets the alignment for the entire @code{struct S} type to 8
3111 Note that although you can ask the compiler to select a time-efficient
3112 alignment for a given type and then declare only individual stand-alone
3113 objects of that type, the compiler's ability to select a time-efficient
3114 alignment is primarily useful only when you plan to create arrays of
3115 variables having the relevant (efficiently aligned) type. If you
3116 declare or use arrays of variables of an efficiently-aligned type, then
3117 it is likely that your program will also be doing pointer arithmetic (or
3118 subscripting, which amounts to the same thing) on pointers to the
3119 relevant type, and the code that the compiler generates for these
3120 pointer arithmetic operations will often be more efficient for
3121 efficiently-aligned types than for other types.
3123 The @code{aligned} attribute can only increase the alignment; but you
3124 can decrease it by specifying @code{packed} as well. See below.
3126 Note that the effectiveness of @code{aligned} attributes may be limited
3127 by inherent limitations in your linker. On many systems, the linker is
3128 only able to arrange for variables to be aligned up to a certain maximum
3129 alignment. (For some linkers, the maximum supported alignment may
3130 be very very small.) If your linker is only able to align variables
3131 up to a maximum of 8 byte alignment, then specifying @code{aligned(16)}
3132 in an @code{__attribute__} will still only provide you with 8 byte
3133 alignment. See your linker documentation for further information.
3136 This attribute, attached to an @code{enum}, @code{struct}, or
3137 @code{union} type definition, specified that the minimum required memory
3138 be used to represent the type.
3140 @opindex fshort-enums
3141 Specifying this attribute for @code{struct} and @code{union} types is
3142 equivalent to specifying the @code{packed} attribute on each of the
3143 structure or union members. Specifying the @option{-fshort-enums}
3144 flag on the line is equivalent to specifying the @code{packed}
3145 attribute on all @code{enum} definitions.
3147 You may only specify this attribute after a closing curly brace on an
3148 @code{enum} definition, not in a @code{typedef} declaration, unless that
3149 declaration also contains the definition of the @code{enum}.
3151 @item transparent_union
3152 This attribute, attached to a @code{union} type definition, indicates
3153 that any function parameter having that union type causes calls to that
3154 function to be treated in a special way.
3156 First, the argument corresponding to a transparent union type can be of
3157 any type in the union; no cast is required. Also, if the union contains
3158 a pointer type, the corresponding argument can be a null pointer
3159 constant or a void pointer expression; and if the union contains a void
3160 pointer type, the corresponding argument can be any pointer expression.
3161 If the union member type is a pointer, qualifiers like @code{const} on
3162 the referenced type must be respected, just as with normal pointer
3165 Second, the argument is passed to the function using the calling
3166 conventions of first member of the transparent union, not the calling
3167 conventions of the union itself. All members of the union must have the
3168 same machine representation; this is necessary for this argument passing
3171 Transparent unions are designed for library functions that have multiple
3172 interfaces for compatibility reasons. For example, suppose the
3173 @code{wait} function must accept either a value of type @code{int *} to
3174 comply with Posix, or a value of type @code{union wait *} to comply with
3175 the 4.1BSD interface. If @code{wait}'s parameter were @code{void *},
3176 @code{wait} would accept both kinds of arguments, but it would also
3177 accept any other pointer type and this would make argument type checking
3178 less useful. Instead, @code{<sys/wait.h>} might define the interface
3186 @} wait_status_ptr_t __attribute__ ((__transparent_union__));
3188 pid_t wait (wait_status_ptr_t);
3191 This interface allows either @code{int *} or @code{union wait *}
3192 arguments to be passed, using the @code{int *} calling convention.
3193 The program can call @code{wait} with arguments of either type:
3196 int w1 () @{ int w; return wait (&w); @}
3197 int w2 () @{ union wait w; return wait (&w); @}
3200 With this interface, @code{wait}'s implementation might look like this:
3203 pid_t wait (wait_status_ptr_t p)
3205 return waitpid (-1, p.__ip, 0);
3210 When attached to a type (including a @code{union} or a @code{struct}),
3211 this attribute means that variables of that type are meant to appear
3212 possibly unused. GCC will not produce a warning for any variables of
3213 that type, even if the variable appears to do nothing. This is often
3214 the case with lock or thread classes, which are usually defined and then
3215 not referenced, but contain constructors and destructors that have
3216 nontrivial bookkeeping functions.
3220 To specify multiple attributes, separate them by commas within the
3221 double parentheses: for example, @samp{__attribute__ ((aligned (16),
3225 @section An Inline Function is As Fast As a Macro
3226 @cindex inline functions
3227 @cindex integrating function code
3229 @cindex macros, inline alternative
3231 By declaring a function @code{inline}, you can direct GCC to
3232 integrate that function's code into the code for its callers. This
3233 makes execution faster by eliminating the function-call overhead; in
3234 addition, if any of the actual argument values are constant, their known
3235 values may permit simplifications at compile time so that not all of the
3236 inline function's code needs to be included. The effect on code size is
3237 less predictable; object code may be larger or smaller with function
3238 inlining, depending on the particular case. Inlining of functions is an
3239 optimization and it really ``works'' only in optimizing compilation. If
3240 you don't use @option{-O}, no function is really inline.
3242 Inline functions are included in the ISO C99 standard, but there are
3243 currently substantial differences between what GCC implements and what
3244 the ISO C99 standard requires.
3246 To declare a function inline, use the @code{inline} keyword in its
3247 declaration, like this:
3257 (If you are writing a header file to be included in ISO C programs, write
3258 @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.)
3259 You can also make all ``simple enough'' functions inline with the option
3260 @option{-finline-functions}.
3263 Note that certain usages in a function definition can make it unsuitable
3264 for inline substitution. Among these usages are: use of varargs, use of
3265 alloca, use of variable sized data types (@pxref{Variable Length}),
3266 use of computed goto (@pxref{Labels as Values}), use of nonlocal goto,
3267 and nested functions (@pxref{Nested Functions}). Using @option{-Winline}
3268 will warn when a function marked @code{inline} could not be substituted,
3269 and will give the reason for the failure.
3271 Note that in C and Objective-C, unlike C++, the @code{inline} keyword
3272 does not affect the linkage of the function.
3274 @cindex automatic @code{inline} for C++ member fns
3275 @cindex @code{inline} automatic for C++ member fns
3276 @cindex member fns, automatically @code{inline}
3277 @cindex C++ member fns, automatically @code{inline}
3278 @opindex fno-default-inline
3279 GCC automatically inlines member functions defined within the class
3280 body of C++ programs even if they are not explicitly declared
3281 @code{inline}. (You can override this with @option{-fno-default-inline};
3282 @pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.)
3284 @cindex inline functions, omission of
3285 @opindex fkeep-inline-functions
3286 When a function is both inline and @code{static}, if all calls to the
3287 function are integrated into the caller, and the function's address is
3288 never used, then the function's own assembler code is never referenced.
3289 In this case, GCC does not actually output assembler code for the
3290 function, unless you specify the option @option{-fkeep-inline-functions}.
3291 Some calls cannot be integrated for various reasons (in particular,
3292 calls that precede the function's definition cannot be integrated, and
3293 neither can recursive calls within the definition). If there is a
3294 nonintegrated call, then the function is compiled to assembler code as
3295 usual. The function must also be compiled as usual if the program
3296 refers to its address, because that can't be inlined.
3298 @cindex non-static inline function
3299 When an inline function is not @code{static}, then the compiler must assume
3300 that there may be calls from other source files; since a global symbol can
3301 be defined only once in any program, the function must not be defined in
3302 the other source files, so the calls therein cannot be integrated.
3303 Therefore, a non-@code{static} inline function is always compiled on its
3304 own in the usual fashion.
3306 If you specify both @code{inline} and @code{extern} in the function
3307 definition, then the definition is used only for inlining. In no case
3308 is the function compiled on its own, not even if you refer to its
3309 address explicitly. Such an address becomes an external reference, as
3310 if you had only declared the function, and had not defined it.
3312 This combination of @code{inline} and @code{extern} has almost the
3313 effect of a macro. The way to use it is to put a function definition in
3314 a header file with these keywords, and put another copy of the
3315 definition (lacking @code{inline} and @code{extern}) in a library file.
3316 The definition in the header file will cause most calls to the function
3317 to be inlined. If any uses of the function remain, they will refer to
3318 the single copy in the library.
3320 For future compatibility with when GCC implements ISO C99 semantics for
3321 inline functions, it is best to use @code{static inline} only. (The
3322 existing semantics will remain available when @option{-std=gnu89} is
3323 specified, but eventually the default will be @option{-std=gnu99} and
3324 that will implement the C99 semantics, though it does not do so yet.)
3326 GCC does not inline any functions when not optimizing. It is not
3327 clear whether it is better to inline or not, in this case, but we found
3328 that a correct implementation when not optimizing was difficult. So we
3329 did the easy thing, and turned it off.
3332 @section Assembler Instructions with C Expression Operands
3333 @cindex extended @code{asm}
3334 @cindex @code{asm} expressions
3335 @cindex assembler instructions
3338 In an assembler instruction using @code{asm}, you can specify the
3339 operands of the instruction using C expressions. This means you need not
3340 guess which registers or memory locations will contain the data you want
3343 You must specify an assembler instruction template much like what
3344 appears in a machine description, plus an operand constraint string for
3347 For example, here is how to use the 68881's @code{fsinx} instruction:
3350 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
3354 Here @code{angle} is the C expression for the input operand while
3355 @code{result} is that of the output operand. Each has @samp{"f"} as its
3356 operand constraint, saying that a floating point register is required.
3357 The @samp{=} in @samp{=f} indicates that the operand is an output; all
3358 output operands' constraints must use @samp{=}. The constraints use the
3359 same language used in the machine description (@pxref{Constraints}).
3361 Each operand is described by an operand-constraint string followed by
3362 the C expression in parentheses. A colon separates the assembler
3363 template from the first output operand and another separates the last
3364 output operand from the first input, if any. Commas separate the
3365 operands within each group. The total number of operands is currently
3366 limited to 30; this limitation may be lifted in some future version of
3369 If there are no output operands but there are input operands, you must
3370 place two consecutive colons surrounding the place where the output
3373 As of GCC version 3.1, it is also possible to specify input and output
3374 operands using symbolic names which can be referenced within the
3375 assembler code. These names are specified inside square brackets
3376 preceding the constraint string, and can be referenced inside the
3377 assembler code using @code{%[@var{name}]} instead of a percentage sign
3378 followed by the operand number. Using named operands the above example
3382 asm ("fsinx %[angle],%[output]"
3383 : [output] "=f" (result)
3384 : [angle] "f" (angle));
3388 Note that the symbolic operand names have no relation whatsoever to
3389 other C identifiers. You may use any name you like, even those of
3390 existing C symbols, but must ensure that no two operands within the same
3391 assembler construct use the same symbolic name.
3393 Output operand expressions must be lvalues; the compiler can check this.
3394 The input operands need not be lvalues. The compiler cannot check
3395 whether the operands have data types that are reasonable for the
3396 instruction being executed. It does not parse the assembler instruction
3397 template and does not know what it means or even whether it is valid
3398 assembler input. The extended @code{asm} feature is most often used for
3399 machine instructions the compiler itself does not know exist. If
3400 the output expression cannot be directly addressed (for example, it is a
3401 bit-field), your constraint must allow a register. In that case, GCC
3402 will use the register as the output of the @code{asm}, and then store
3403 that register into the output.
3405 The ordinary output operands must be write-only; GCC will assume that
3406 the values in these operands before the instruction are dead and need
3407 not be generated. Extended asm supports input-output or read-write
3408 operands. Use the constraint character @samp{+} to indicate such an
3409 operand and list it with the output operands.
3411 When the constraints for the read-write operand (or the operand in which
3412 only some of the bits are to be changed) allows a register, you may, as
3413 an alternative, logically split its function into two separate operands,
3414 one input operand and one write-only output operand. The connection
3415 between them is expressed by constraints which say they need to be in
3416 the same location when the instruction executes. You can use the same C
3417 expression for both operands, or different expressions. For example,
3418 here we write the (fictitious) @samp{combine} instruction with
3419 @code{bar} as its read-only source operand and @code{foo} as its
3420 read-write destination:
3423 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
3427 The constraint @samp{"0"} for operand 1 says that it must occupy the
3428 same location as operand 0. A number in constraint is allowed only in
3429 an input operand and it must refer to an output operand.
3431 Only a number in the constraint can guarantee that one operand will be in
3432 the same place as another. The mere fact that @code{foo} is the value
3433 of both operands is not enough to guarantee that they will be in the
3434 same place in the generated assembler code. The following would not
3438 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
3441 Various optimizations or reloading could cause operands 0 and 1 to be in
3442 different registers; GCC knows no reason not to do so. For example, the
3443 compiler might find a copy of the value of @code{foo} in one register and
3444 use it for operand 1, but generate the output operand 0 in a different
3445 register (copying it afterward to @code{foo}'s own address). Of course,
3446 since the register for operand 1 is not even mentioned in the assembler
3447 code, the result will not work, but GCC can't tell that.
3449 As of GCC version 3.1, one may write @code{[@var{name}]} instead of
3450 the operand number for a matching constraint. For example:
3453 asm ("cmoveq %1,%2,%[result]"
3454 : [result] "=r"(result)
3455 : "r" (test), "r"(new), "[result]"(old));
3458 Some instructions clobber specific hard registers. To describe this,
3459 write a third colon after the input operands, followed by the names of
3460 the clobbered hard registers (given as strings). Here is a realistic
3461 example for the VAX:
3464 asm volatile ("movc3 %0,%1,%2"
3466 : "g" (from), "g" (to), "g" (count)
3467 : "r0", "r1", "r2", "r3", "r4", "r5");
3470 You may not write a clobber description in a way that overlaps with an
3471 input or output operand. For example, you may not have an operand
3472 describing a register class with one member if you mention that register
3473 in the clobber list. There is no way for you to specify that an input
3474 operand is modified without also specifying it as an output
3475 operand. Note that if all the output operands you specify are for this
3476 purpose (and hence unused), you will then also need to specify
3477 @code{volatile} for the @code{asm} construct, as described below, to
3478 prevent GCC from deleting the @code{asm} statement as unused.
3480 If you refer to a particular hardware register from the assembler code,
3481 you will probably have to list the register after the third colon to
3482 tell the compiler the register's value is modified. In some assemblers,
3483 the register names begin with @samp{%}; to produce one @samp{%} in the
3484 assembler code, you must write @samp{%%} in the input.
3486 If your assembler instruction can alter the condition code register, add
3487 @samp{cc} to the list of clobbered registers. GCC on some machines
3488 represents the condition codes as a specific hardware register;
3489 @samp{cc} serves to name this register. On other machines, the
3490 condition code is handled differently, and specifying @samp{cc} has no
3491 effect. But it is valid no matter what the machine.
3493 If your assembler instruction modifies memory in an unpredictable
3494 fashion, add @samp{memory} to the list of clobbered registers. This
3495 will cause GCC to not keep memory values cached in registers across
3496 the assembler instruction. You will also want to add the
3497 @code{volatile} keyword if the memory affected is not listed in the
3498 inputs or outputs of the @code{asm}, as the @samp{memory} clobber does
3499 not count as a side-effect of the @code{asm}.
3501 You can put multiple assembler instructions together in a single
3502 @code{asm} template, separated by the characters normally used in assembly
3503 code for the system. A combination that works in most places is a newline
3504 to break the line, plus a tab character to move to the instruction field
3505 (written as @samp{\n\t}). Sometimes semicolons can be used, if the
3506 assembler allows semicolons as a line-breaking character. Note that some
3507 assembler dialects use semicolons to start a comment.
3508 The input operands are guaranteed not to use any of the clobbered
3509 registers, and neither will the output operands' addresses, so you can
3510 read and write the clobbered registers as many times as you like. Here
3511 is an example of multiple instructions in a template; it assumes the
3512 subroutine @code{_foo} accepts arguments in registers 9 and 10:
3515 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
3517 : "g" (from), "g" (to)
3521 Unless an output operand has the @samp{&} constraint modifier, GCC
3522 may allocate it in the same register as an unrelated input operand, on
3523 the assumption the inputs are consumed before the outputs are produced.
3524 This assumption may be false if the assembler code actually consists of
3525 more than one instruction. In such a case, use @samp{&} for each output
3526 operand that may not overlap an input. @xref{Modifiers}.
3528 If you want to test the condition code produced by an assembler
3529 instruction, you must include a branch and a label in the @code{asm}
3530 construct, as follows:
3533 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
3539 This assumes your assembler supports local labels, as the GNU assembler
3540 and most Unix assemblers do.
3542 Speaking of labels, jumps from one @code{asm} to another are not
3543 supported. The compiler's optimizers do not know about these jumps, and
3544 therefore they cannot take account of them when deciding how to
3547 @cindex macros containing @code{asm}
3548 Usually the most convenient way to use these @code{asm} instructions is to
3549 encapsulate them in macros that look like functions. For example,
3553 (@{ double __value, __arg = (x); \
3554 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
3559 Here the variable @code{__arg} is used to make sure that the instruction
3560 operates on a proper @code{double} value, and to accept only those
3561 arguments @code{x} which can convert automatically to a @code{double}.
3563 Another way to make sure the instruction operates on the correct data
3564 type is to use a cast in the @code{asm}. This is different from using a
3565 variable @code{__arg} in that it converts more different types. For
3566 example, if the desired type were @code{int}, casting the argument to
3567 @code{int} would accept a pointer with no complaint, while assigning the
3568 argument to an @code{int} variable named @code{__arg} would warn about
3569 using a pointer unless the caller explicitly casts it.
3571 If an @code{asm} has output operands, GCC assumes for optimization
3572 purposes the instruction has no side effects except to change the output
3573 operands. This does not mean instructions with a side effect cannot be
3574 used, but you must be careful, because the compiler may eliminate them
3575 if the output operands aren't used, or move them out of loops, or
3576 replace two with one if they constitute a common subexpression. Also,
3577 if your instruction does have a side effect on a variable that otherwise
3578 appears not to change, the old value of the variable may be reused later
3579 if it happens to be found in a register.
3581 You can prevent an @code{asm} instruction from being deleted, moved
3582 significantly, or combined, by writing the keyword @code{volatile} after
3583 the @code{asm}. For example:
3586 #define get_and_set_priority(new) \
3588 asm volatile ("get_and_set_priority %0, %1" \
3589 : "=g" (__old) : "g" (new)); \
3594 If you write an @code{asm} instruction with no outputs, GCC will know
3595 the instruction has side-effects and will not delete the instruction or
3596 move it outside of loops.
3598 The @code{volatile} keyword indicates that the instruction has
3599 important side-effects. GCC will not delete a volatile @code{asm} if
3600 it is reachable. (The instruction can still be deleted if GCC can
3601 prove that control-flow will never reach the location of the
3602 instruction.) In addition, GCC will not reschedule instructions
3603 across a volatile @code{asm} instruction. For example:
3606 *(volatile int *)addr = foo;
3607 asm volatile ("eieio" : : );
3611 Assume @code{addr} contains the address of a memory mapped device
3612 register. The PowerPC @code{eieio} instruction (Enforce In-order
3613 Execution of I/O) tells the CPU to make sure that the store to that
3614 device register happens before it issues any other I/O@.
3616 Note that even a volatile @code{asm} instruction can be moved in ways
3617 that appear insignificant to the compiler, such as across jump
3618 instructions. You can't expect a sequence of volatile @code{asm}
3619 instructions to remain perfectly consecutive. If you want consecutive
3620 output, use a single @code{asm}. Also, GCC will perform some
3621 optimizations across a volatile @code{asm} instruction; GCC does not
3622 ``forget everything'' when it encounters a volatile @code{asm}
3623 instruction the way some other compilers do.
3625 An @code{asm} instruction without any operands or clobbers (an ``old
3626 style'' @code{asm}) will be treated identically to a volatile
3627 @code{asm} instruction.
3629 It is a natural idea to look for a way to give access to the condition
3630 code left by the assembler instruction. However, when we attempted to
3631 implement this, we found no way to make it work reliably. The problem
3632 is that output operands might need reloading, which would result in
3633 additional following ``store'' instructions. On most machines, these
3634 instructions would alter the condition code before there was time to
3635 test it. This problem doesn't arise for ordinary ``test'' and
3636 ``compare'' instructions because they don't have any output operands.
3638 For reasons similar to those described above, it is not possible to give
3639 an assembler instruction access to the condition code left by previous
3642 If you are writing a header file that should be includable in ISO C
3643 programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate
3646 @subsection i386 floating point asm operands
3648 There are several rules on the usage of stack-like regs in
3649 asm_operands insns. These rules apply only to the operands that are
3654 Given a set of input regs that die in an asm_operands, it is
3655 necessary to know which are implicitly popped by the asm, and
3656 which must be explicitly popped by gcc.
3658 An input reg that is implicitly popped by the asm must be
3659 explicitly clobbered, unless it is constrained to match an
3663 For any input reg that is implicitly popped by an asm, it is
3664 necessary to know how to adjust the stack to compensate for the pop.
3665 If any non-popped input is closer to the top of the reg-stack than
3666 the implicitly popped reg, it would not be possible to know what the
3667 stack looked like---it's not clear how the rest of the stack ``slides
3670 All implicitly popped input regs must be closer to the top of
3671 the reg-stack than any input that is not implicitly popped.
3673 It is possible that if an input dies in an insn, reload might
3674 use the input reg for an output reload. Consider this example:
3677 asm ("foo" : "=t" (a) : "f" (b));
3680 This asm says that input B is not popped by the asm, and that
3681 the asm pushes a result onto the reg-stack, i.e., the stack is one
3682 deeper after the asm than it was before. But, it is possible that
3683 reload will think that it can use the same reg for both the input and
3684 the output, if input B dies in this insn.
3686 If any input operand uses the @code{f} constraint, all output reg
3687 constraints must use the @code{&} earlyclobber.
3689 The asm above would be written as
3692 asm ("foo" : "=&t" (a) : "f" (b));
3696 Some operands need to be in particular places on the stack. All
3697 output operands fall in this category---there is no other way to
3698 know which regs the outputs appear in unless the user indicates
3699 this in the constraints.
3701 Output operands must specifically indicate which reg an output
3702 appears in after an asm. @code{=f} is not allowed: the operand
3703 constraints must select a class with a single reg.
3706 Output operands may not be ``inserted'' between existing stack regs.
3707 Since no 387 opcode uses a read/write operand, all output operands
3708 are dead before the asm_operands, and are pushed by the asm_operands.
3709 It makes no sense to push anywhere but the top of the reg-stack.
3711 Output operands must start at the top of the reg-stack: output
3712 operands may not ``skip'' a reg.
3715 Some asm statements may need extra stack space for internal
3716 calculations. This can be guaranteed by clobbering stack registers
3717 unrelated to the inputs and outputs.
3721 Here are a couple of reasonable asms to want to write. This asm
3722 takes one input, which is internally popped, and produces two outputs.
3725 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
3728 This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode,
3729 and replaces them with one output. The user must code the @code{st(1)}
3730 clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs.
3733 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
3739 @section Controlling Names Used in Assembler Code
3740 @cindex assembler names for identifiers
3741 @cindex names used in assembler code
3742 @cindex identifiers, names in assembler code
3744 You can specify the name to be used in the assembler code for a C
3745 function or variable by writing the @code{asm} (or @code{__asm__})
3746 keyword after the declarator as follows:
3749 int foo asm ("myfoo") = 2;
3753 This specifies that the name to be used for the variable @code{foo} in
3754 the assembler code should be @samp{myfoo} rather than the usual
3757 On systems where an underscore is normally prepended to the name of a C
3758 function or variable, this feature allows you to define names for the
3759 linker that do not start with an underscore.
3761 It does not make sense to use this feature with a non-static local
3762 variable since such variables do not have assembler names. If you are
3763 trying to put the variable in a particular register, see @ref{Explicit
3764 Reg Vars}. GCC presently accepts such code with a warning, but will
3765 probably be changed to issue an error, rather than a warning, in the
3768 You cannot use @code{asm} in this way in a function @emph{definition}; but
3769 you can get the same effect by writing a declaration for the function
3770 before its definition and putting @code{asm} there, like this:
3773 extern func () asm ("FUNC");
3780 It is up to you to make sure that the assembler names you choose do not
3781 conflict with any other assembler symbols. Also, you must not use a
3782 register name; that would produce completely invalid assembler code. GCC
3783 does not as yet have the ability to store static variables in registers.
3784 Perhaps that will be added.
3786 @node Explicit Reg Vars
3787 @section Variables in Specified Registers
3788 @cindex explicit register variables
3789 @cindex variables in specified registers
3790 @cindex specified registers
3791 @cindex registers, global allocation
3793 GNU C allows you to put a few global variables into specified hardware
3794 registers. You can also specify the register in which an ordinary
3795 register variable should be allocated.
3799 Global register variables reserve registers throughout the program.
3800 This may be useful in programs such as programming language
3801 interpreters which have a couple of global variables that are accessed
3805 Local register variables in specific registers do not reserve the
3806 registers. The compiler's data flow analysis is capable of determining
3807 where the specified registers contain live values, and where they are
3808 available for other uses. Stores into local register variables may be deleted
3809 when they appear to be dead according to dataflow analysis. References
3810 to local register variables may be deleted or moved or simplified.
3812 These local variables are sometimes convenient for use with the extended
3813 @code{asm} feature (@pxref{Extended Asm}), if you want to write one
3814 output of the assembler instruction directly into a particular register.
3815 (This will work provided the register you specify fits the constraints
3816 specified for that operand in the @code{asm}.)
3824 @node Global Reg Vars
3825 @subsection Defining Global Register Variables
3826 @cindex global register variables
3827 @cindex registers, global variables in
3829 You can define a global register variable in GNU C like this:
3832 register int *foo asm ("a5");
3836 Here @code{a5} is the name of the register which should be used. Choose a
3837 register which is normally saved and restored by function calls on your
3838 machine, so that library routines will not clobber it.
3840 Naturally the register name is cpu-dependent, so you would need to
3841 conditionalize your program according to cpu type. The register
3842 @code{a5} would be a good choice on a 68000 for a variable of pointer
3843 type. On machines with register windows, be sure to choose a ``global''
3844 register that is not affected magically by the function call mechanism.
3846 In addition, operating systems on one type of cpu may differ in how they
3847 name the registers; then you would need additional conditionals. For
3848 example, some 68000 operating systems call this register @code{%a5}.
3850 Eventually there may be a way of asking the compiler to choose a register
3851 automatically, but first we need to figure out how it should choose and
3852 how to enable you to guide the choice. No solution is evident.
3854 Defining a global register variable in a certain register reserves that
3855 register entirely for this use, at least within the current compilation.
3856 The register will not be allocated for any other purpose in the functions
3857 in the current compilation. The register will not be saved and restored by
3858 these functions. Stores into this register are never deleted even if they
3859 would appear to be dead, but references may be deleted or moved or
3862 It is not safe to access the global register variables from signal
3863 handlers, or from more than one thread of control, because the system
3864 library routines may temporarily use the register for other things (unless
3865 you recompile them specially for the task at hand).
3867 @cindex @code{qsort}, and global register variables
3868 It is not safe for one function that uses a global register variable to
3869 call another such function @code{foo} by way of a third function
3870 @code{lose} that was compiled without knowledge of this variable (i.e.@: in a
3871 different source file in which the variable wasn't declared). This is
3872 because @code{lose} might save the register and put some other value there.
3873 For example, you can't expect a global register variable to be available in
3874 the comparison-function that you pass to @code{qsort}, since @code{qsort}
3875 might have put something else in that register. (If you are prepared to
3876 recompile @code{qsort} with the same global register variable, you can
3877 solve this problem.)
3879 If you want to recompile @code{qsort} or other source files which do not
3880 actually use your global register variable, so that they will not use that
3881 register for any other purpose, then it suffices to specify the compiler
3882 option @option{-ffixed-@var{reg}}. You need not actually add a global
3883 register declaration to their source code.
3885 A function which can alter the value of a global register variable cannot
3886 safely be called from a function compiled without this variable, because it
3887 could clobber the value the caller expects to find there on return.
3888 Therefore, the function which is the entry point into the part of the
3889 program that uses the global register variable must explicitly save and
3890 restore the value which belongs to its caller.
3892 @cindex register variable after @code{longjmp}
3893 @cindex global register after @code{longjmp}
3894 @cindex value after @code{longjmp}
3897 On most machines, @code{longjmp} will restore to each global register
3898 variable the value it had at the time of the @code{setjmp}. On some
3899 machines, however, @code{longjmp} will not change the value of global
3900 register variables. To be portable, the function that called @code{setjmp}
3901 should make other arrangements to save the values of the global register
3902 variables, and to restore them in a @code{longjmp}. This way, the same
3903 thing will happen regardless of what @code{longjmp} does.
3905 All global register variable declarations must precede all function
3906 definitions. If such a declaration could appear after function
3907 definitions, the declaration would be too late to prevent the register from
3908 being used for other purposes in the preceding functions.
3910 Global register variables may not have initial values, because an
3911 executable file has no means to supply initial contents for a register.
3913 On the Sparc, there are reports that g3 @dots{} g7 are suitable
3914 registers, but certain library functions, such as @code{getwd}, as well
3915 as the subroutines for division and remainder, modify g3 and g4. g1 and
3916 g2 are local temporaries.
3918 On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7.
3919 Of course, it will not do to use more than a few of those.
3921 @node Local Reg Vars
3922 @subsection Specifying Registers for Local Variables
3923 @cindex local variables, specifying registers
3924 @cindex specifying registers for local variables
3925 @cindex registers for local variables
3927 You can define a local register variable with a specified register
3931 register int *foo asm ("a5");
3935 Here @code{a5} is the name of the register which should be used. Note
3936 that this is the same syntax used for defining global register
3937 variables, but for a local variable it would appear within a function.
3939 Naturally the register name is cpu-dependent, but this is not a
3940 problem, since specific registers are most often useful with explicit
3941 assembler instructions (@pxref{Extended Asm}). Both of these things
3942 generally require that you conditionalize your program according to
3945 In addition, operating systems on one type of cpu may differ in how they
3946 name the registers; then you would need additional conditionals. For
3947 example, some 68000 operating systems call this register @code{%a5}.
3949 Defining such a register variable does not reserve the register; it
3950 remains available for other uses in places where flow control determines
3951 the variable's value is not live. However, these registers are made
3952 unavailable for use in the reload pass; excessive use of this feature
3953 leaves the compiler too few available registers to compile certain
3956 This option does not guarantee that GCC will generate code that has
3957 this variable in the register you specify at all times. You may not
3958 code an explicit reference to this register in an @code{asm} statement
3959 and assume it will always refer to this variable.
3961 Stores into local register variables may be deleted when they appear to be dead
3962 according to dataflow analysis. References to local register variables may
3963 be deleted or moved or simplified.
3965 @node Alternate Keywords
3966 @section Alternate Keywords
3967 @cindex alternate keywords
3968 @cindex keywords, alternate
3970 The option @option{-traditional} disables certain keywords;
3971 @option{-ansi} and the various @option{-std} options disable certain
3972 others. This causes trouble when you want to use GNU C extensions, or
3973 ISO C features, in a general-purpose header file that should be usable
3974 by all programs, including ISO C programs and traditional ones. The
3975 keywords @code{asm}, @code{typeof} and @code{inline} cannot be used
3976 since they won't work in a program compiled with @option{-ansi}
3977 (although @code{inline} can be used in a program compiled with
3978 @option{-std=c99}), while the keywords @code{const}, @code{volatile},
3979 @code{signed}, @code{typeof} and @code{inline} won't work in a program
3980 compiled with @option{-traditional}. The ISO C99 keyword
3981 @code{restrict} is only available when @option{-std=gnu99} (which will
3982 eventually be the default) or @option{-std=c99} (or the equivalent
3983 @option{-std=iso9899:1999}) is used.
3985 The way to solve these problems is to put @samp{__} at the beginning and
3986 end of each problematical keyword. For example, use @code{__asm__}
3987 instead of @code{asm}, @code{__const__} instead of @code{const}, and
3988 @code{__inline__} instead of @code{inline}.
3990 Other C compilers won't accept these alternative keywords; if you want to
3991 compile with another compiler, you can define the alternate keywords as
3992 macros to replace them with the customary keywords. It looks like this:
4000 @findex __extension__
4002 @option{-pedantic} and other options cause warnings for many GNU C extensions.
4004 prevent such warnings within one expression by writing
4005 @code{__extension__} before the expression. @code{__extension__} has no
4006 effect aside from this.
4008 @node Incomplete Enums
4009 @section Incomplete @code{enum} Types
4011 You can define an @code{enum} tag without specifying its possible values.
4012 This results in an incomplete type, much like what you get if you write
4013 @code{struct foo} without describing the elements. A later declaration
4014 which does specify the possible values completes the type.
4016 You can't allocate variables or storage using the type while it is
4017 incomplete. However, you can work with pointers to that type.
4019 This extension may not be very useful, but it makes the handling of
4020 @code{enum} more consistent with the way @code{struct} and @code{union}
4023 This extension is not supported by GNU C++.
4025 @node Function Names
4026 @section Function Names as Strings
4027 @cindex @code{__FUNCTION__} identifier
4028 @cindex @code{__PRETTY_FUNCTION__} identifier
4029 @cindex @code{__func__} identifier
4031 GCC predefines two magic identifiers to hold the name of the current
4032 function. The identifier @code{__FUNCTION__} holds the name of the function
4033 as it appears in the source. The identifier @code{__PRETTY_FUNCTION__}
4034 holds the name of the function pretty printed in a language specific
4037 These names are always the same in a C function, but in a C++ function
4038 they may be different. For example, this program:
4042 extern int printf (char *, ...);
4049 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
4050 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
4068 __PRETTY_FUNCTION__ = int a::sub (int)
4071 The compiler automagically replaces the identifiers with a string
4072 literal containing the appropriate name. Thus, they are neither
4073 preprocessor macros, like @code{__FILE__} and @code{__LINE__}, nor
4074 variables. This means that they catenate with other string literals, and
4075 that they can be used to initialize char arrays. For example
4078 char here[] = "Function " __FUNCTION__ " in " __FILE__;
4081 On the other hand, @samp{#ifdef __FUNCTION__} does not have any special
4082 meaning inside a function, since the preprocessor does not do anything
4083 special with the identifier @code{__FUNCTION__}.
4085 Note that these semantics are deprecated, and that GCC 3.2 will handle
4086 @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} the same way as
4087 @code{__func__}. @code{__func__} is defined by the ISO standard C99:
4090 The identifier @code{__func__} is implicitly declared by the translator
4091 as if, immediately following the opening brace of each function
4092 definition, the declaration
4095 static const char __func__[] = "function-name";
4098 appeared, where function-name is the name of the lexically-enclosing
4099 function. This name is the unadorned name of the function.
4102 By this definition, @code{__func__} is a variable, not a string literal.
4103 In particular, @code{__func__} does not catenate with other string
4106 In @code{C++}, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} are
4107 variables, declared in the same way as @code{__func__}.
4109 @node Return Address
4110 @section Getting the Return or Frame Address of a Function
4112 These functions may be used to get information about the callers of a
4115 @deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level})
4116 This function returns the return address of the current function, or of
4117 one of its callers. The @var{level} argument is number of frames to
4118 scan up the call stack. A value of @code{0} yields the return address
4119 of the current function, a value of @code{1} yields the return address
4120 of the caller of the current function, and so forth.
4122 The @var{level} argument must be a constant integer.
4124 On some machines it may be impossible to determine the return address of
4125 any function other than the current one; in such cases, or when the top
4126 of the stack has been reached, this function will return @code{0} or a
4127 random value. In addition, @code{__builtin_frame_address} may be used
4128 to determine if the top of the stack has been reached.
4130 This function should only be used with a nonzero argument for debugging
4134 @deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level})
4135 This function is similar to @code{__builtin_return_address}, but it
4136 returns the address of the function frame rather than the return address
4137 of the function. Calling @code{__builtin_frame_address} with a value of
4138 @code{0} yields the frame address of the current function, a value of
4139 @code{1} yields the frame address of the caller of the current function,
4142 The frame is the area on the stack which holds local variables and saved
4143 registers. The frame address is normally the address of the first word
4144 pushed on to the stack by the function. However, the exact definition
4145 depends upon the processor and the calling convention. If the processor
4146 has a dedicated frame pointer register, and the function has a frame,
4147 then @code{__builtin_frame_address} will return the value of the frame
4150 On some machines it may be impossible to determine the frame address of
4151 any function other than the current one; in such cases, or when the top
4152 of the stack has been reached, this function will return @code{0} if
4153 the first frame pointer is properly initialized by the startup code.
4155 This function should only be used with a nonzero argument for debugging
4159 @node Vector Extensions
4160 @section Using vector instructions through built-in functions
4162 On some targets, the instruction set contains SIMD vector instructions that
4163 operate on multiple values contained in one large register at the same time.
4164 For example, on the i386 the MMX, 3Dnow! and SSE extensions can be used
4167 The first step in using these extensions is to provide the necessary data
4168 types. This should be done using an appropriate @code{typedef}:
4171 typedef int v4si __attribute__ ((mode(V4SI)));
4174 The base type @code{int} is effectively ignored by the compiler, the
4175 actual properties of the new type @code{v4si} are defined by the
4176 @code{__attribute__}. It defines the machine mode to be used; for vector
4177 types these have the form @code{V@var{n}@var{B}}; @var{n} should be the
4178 number of elements in the vector, and @var{B} should be the base mode of the
4179 individual elements. The following can be used as base modes:
4183 An integer that is as wide as the smallest addressable unit, usually 8 bits.
4185 An integer, twice as wide as a QI mode integer, usually 16 bits.
4187 An integer, four times as wide as a QI mode integer, usually 32 bits.
4189 An integer, eight times as wide as a QI mode integer, usually 64 bits.
4191 A floating point value, as wide as a SI mode integer, usually 32 bits.
4193 A floating point value, as wide as a DI mode integer, usually 64 bits.
4196 Not all base types or combinations are always valid; which modes can be used
4197 is determined by the target machine. For example, if targetting the i386 MMX
4198 extensions, only @code{V8QI}, @code{V4HI} and @code{V2SI} are allowed modes.
4200 There are no @code{V1xx} vector modes - they would be identical to the
4201 corresponding base mode.
4203 There is no distinction between signed and unsigned vector modes. This
4204 distinction is made by the operations that perform on the vectors, not
4207 The types defined in this manner are somewhat special, they cannot be
4208 used with most normal C operations (i.e., a vector addition can @emph{not}
4209 be represented by a normal addition of two vector type variables). You
4210 can declare only variables and use them in function calls and returns, as
4211 well as in assignments and some casts. It is possible to cast from one
4212 vector type to another, provided they are of the same size (in fact, you
4213 can also cast vectors to and from other datatypes of the same size).
4215 A port that supports vector operations provides a set of built-in functions
4216 that can be used to operate on vectors. For example, a function to add two
4217 vectors and multiply the result by a third could look like this:
4220 v4si f (v4si a, v4si b, v4si c)
4222 v4si tmp = __builtin_addv4si (a, b);
4223 return __builtin_mulv4si (tmp, c);
4228 @node Other Builtins
4229 @section Other built-in functions provided by GCC
4230 @cindex built-in functions
4231 @findex __builtin_isgreater
4232 @findex __builtin_isgreaterequal
4233 @findex __builtin_isless
4234 @findex __builtin_islessequal
4235 @findex __builtin_islessgreater
4236 @findex __builtin_isunordered
4292 GCC provides a large number of built-in functions other than the ones
4293 mentioned above. Some of these are for internal use in the processing
4294 of exceptions or variable-length argument lists and will not be
4295 documented here because they may change from time to time; we do not
4296 recommend general use of these functions.
4298 The remaining functions are provided for optimization purposes.
4300 @opindex fno-builtin
4301 GCC includes built-in versions of many of the functions in the standard
4302 C library. The versions prefixed with @code{__builtin_} will always be
4303 treated as having the same meaning as the C library function even if you
4304 specify the @option{-fno-builtin} option. (@pxref{C Dialect Options})
4305 Many of these functions are only optimized in certain cases; if they are
4306 not optimized in a particular case, a call to the library function will
4311 The functions @code{abort}, @code{exit}, @code{_Exit} and @code{_exit}
4312 are recognized and presumed not to return, but otherwise are not built
4313 in. @code{_exit} is not recognized in strict ISO C mode (@option{-ansi},
4314 @option{-std=c89} or @option{-std=c99}). @code{_Exit} is not recognized in
4315 strict C89 mode (@option{-ansi} or @option{-std=c89}).
4317 Outside strict ISO C mode, the functions @code{alloca}, @code{bcmp},
4318 @code{bzero}, @code{index}, @code{rindex} and @code{ffs} may be handled
4319 as built-in functions. All these functions have corresponding versions
4320 prefixed with @code{__builtin_}, which may be used even in strict C89
4323 The ISO C99 functions @code{conj}, @code{conjf}, @code{conjl},
4324 @code{creal}, @code{crealf}, @code{creall}, @code{cimag}, @code{cimagf},
4325 @code{cimagl}, @code{llabs} and @code{imaxabs} are handled as built-in
4326 functions except in strict ISO C89 mode. There are also built-in
4327 versions of the ISO C99 functions @code{cosf}, @code{cosl},
4328 @code{fabsf}, @code{fabsl}, @code{sinf}, @code{sinl}, @code{sqrtf}, and
4329 @code{sqrtl}, that are recognized in any mode since ISO C89 reserves
4330 these names for the purpose to which ISO C99 puts them. All these
4331 functions have corresponding versions prefixed with @code{__builtin_}.
4333 The ISO C89 functions @code{abs}, @code{cos}, @code{fabs},
4334 @code{fprintf}, @code{fputs}, @code{labs}, @code{memcmp}, @code{memcpy},
4335 @code{memset}, @code{printf}, @code{sin}, @code{sqrt}, @code{strcat},
4336 @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn},
4337 @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy},
4338 @code{strpbrk}, @code{strrchr}, @code{strspn}, and @code{strstr} are all
4339 recognized as built-in functions unless @option{-fno-builtin} is
4340 specified (or @option{-fno-builtin-@var{function}} is specified for an
4341 individual function). All of these functions have corresponding
4342 versions prefixed with @code{__builtin_}.
4344 GCC provides built-in versions of the ISO C99 floating point comparison
4345 macros that avoid raising exceptions for unordered operands. They have
4346 the same names as the standard macros ( @code{isgreater},
4347 @code{isgreaterequal}, @code{isless}, @code{islessequal},
4348 @code{islessgreater}, and @code{isunordered}) , with @code{__builtin_}
4349 prefixed. We intend for a library implementor to be able to simply
4350 @code{#define} each standard macro to its built-in equivalent.
4352 @deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2})
4354 You can use the built-in function @code{__builtin_types_compatible_p} to
4355 determine whether two types are the same.
4357 This built-in function returns 1 if the unqualified versions of the
4358 types @var{type1} and @var{type2} (which are types, not expressions) are
4359 compatible, 0 otherwise. The result of this built-in function can be
4360 used in integer constant expressions.
4362 This built-in function ignores top level qualifiers (e.g., @code{const},
4363 @code{volatile}). For example, @code{int} is equivalent to @code{const
4366 The type @code{int[]} and @code{int[5]} are compatible. On the other
4367 hand, @code{int} and @code{char *} are not compatible, even if the size
4368 of their types, on the particular architecture are the same. Also, the
4369 amount of pointer indirection is taken into account when determining
4370 similarity. Consequently, @code{short *} is not similar to
4371 @code{short **}. Furthermore, two types that are typedefed are
4372 considered compatible if their underlying types are compatible.
4374 An @code{enum} type is considered to be compatible with another
4375 @code{enum} type. For example, @code{enum @{foo, bar@}} is similar to
4376 @code{enum @{hot, dog@}}.
4378 You would typically use this function in code whose execution varies
4379 depending on the arguments' types. For example:
4385 if (__builtin_types_compatible_p (typeof (x), long double)) \
4386 tmp = foo_long_double (tmp); \
4387 else if (__builtin_types_compatible_p (typeof (x), double)) \
4388 tmp = foo_double (tmp); \
4389 else if (__builtin_types_compatible_p (typeof (x), float)) \
4390 tmp = foo_float (tmp); \
4397 @emph{Note:} This construct is only available for C.
4401 @deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2})
4403 You can use the built-in function @code{__builtin_choose_expr} to
4404 evaluate code depending on the value of a constant expression. This
4405 built-in function returns @var{exp1} if @var{const_exp}, which is a
4406 constant expression that must be able to be determined at compile time,
4407 is nonzero. Otherwise it returns 0.
4409 This built-in function is analogous to the @samp{? :} operator in C,
4410 except that the expression returned has its type unaltered by promotion
4411 rules. Also, the built-in function does not evaluate the expression
4412 that was not chosen. For example, if @var{const_exp} evaluates to true,
4413 @var{exp2} is not evaluated even if it has side-effects.
4415 This built-in function can return an lvalue if the chosen argument is an
4418 If @var{exp1} is returned, the return type is the same as @var{exp1}'s
4419 type. Similarly, if @var{exp2} is returned, its return type is the same
4426 __builtin_choose_expr (__builtin_types_compatible_p (typeof (x), double), \
4428 __builtin_choose_expr (__builtin_types_compatible_p (typeof (x), float), \
4430 /* @r{The void expression results in a compile-time error} \
4431 @r{when assigning the result to something.} */ \
4435 @emph{Note:} This construct is only available for C. Furthermore, the
4436 unused expression (@var{exp1} or @var{exp2} depending on the value of
4437 @var{const_exp}) may still generate syntax errors. This may change in
4442 @deftypefn {Built-in Function} int __builtin_constant_p (@var{exp})
4443 You can use the built-in function @code{__builtin_constant_p} to
4444 determine if a value is known to be constant at compile-time and hence
4445 that GCC can perform constant-folding on expressions involving that
4446 value. The argument of the function is the value to test. The function
4447 returns the integer 1 if the argument is known to be a compile-time
4448 constant and 0 if it is not known to be a compile-time constant. A
4449 return of 0 does not indicate that the value is @emph{not} a constant,
4450 but merely that GCC cannot prove it is a constant with the specified
4451 value of the @option{-O} option.
4453 You would typically use this function in an embedded application where
4454 memory was a critical resource. If you have some complex calculation,
4455 you may want it to be folded if it involves constants, but need to call
4456 a function if it does not. For example:
4459 #define Scale_Value(X) \
4460 (__builtin_constant_p (X) \
4461 ? ((X) * SCALE + OFFSET) : Scale (X))
4464 You may use this built-in function in either a macro or an inline
4465 function. However, if you use it in an inlined function and pass an
4466 argument of the function as the argument to the built-in, GCC will
4467 never return 1 when you call the inline function with a string constant
4468 or compound literal (@pxref{Compound Literals}) and will not return 1
4469 when you pass a constant numeric value to the inline function unless you
4470 specify the @option{-O} option.
4472 You may also use @code{__builtin_constant_p} in initializers for static
4473 data. For instance, you can write
4476 static const int table[] = @{
4477 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
4483 This is an acceptable initializer even if @var{EXPRESSION} is not a
4484 constant expression. GCC must be more conservative about evaluating the
4485 built-in in this case, because it has no opportunity to perform
4488 Previous versions of GCC did not accept this built-in in data
4489 initializers. The earliest version where it is completely safe is
4493 @deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c})
4494 @opindex fprofile-arcs
4495 You may use @code{__builtin_expect} to provide the compiler with
4496 branch prediction information. In general, you should prefer to
4497 use actual profile feedback for this (@option{-fprofile-arcs}), as
4498 programmers are notoriously bad at predicting how their programs
4499 actually perform. However, there are applications in which this
4500 data is hard to collect.
4502 The return value is the value of @var{exp}, which should be an
4503 integral expression. The value of @var{c} must be a compile-time
4504 constant. The semantics of the built-in are that it is expected
4505 that @var{exp} == @var{c}. For example:
4508 if (__builtin_expect (x, 0))
4513 would indicate that we do not expect to call @code{foo}, since
4514 we expect @code{x} to be zero. Since you are limited to integral
4515 expressions for @var{exp}, you should use constructions such as
4518 if (__builtin_expect (ptr != NULL, 1))
4523 when testing pointer or floating-point values.
4526 @deftypefn {Built-in Function} void __builtin_prefetch (void *@var{addr}, ...)
4527 This function is used to minimize cache-miss latency by moving data into
4528 a cache before it is accessed.
4529 You can insert calls to @code{__builtin_prefetch} into code for which
4530 you know addresses of data in memory that is likely to be accessed soon.
4531 If the target supports them, data prefetch instructions will be generated.
4532 If the prefetch is done early enough before the access then the data will
4533 be in the cache by the time it is accessed.
4535 The value of @var{addr} is the address of the memory to prefetch.
4536 There are two optional arguments, @var{rw} and @var{locality}.
4537 The value of @var{rw} is a compile-time constant one or zero; one
4538 means that the prefetch is preparing for a write to the memory address
4539 and zero, the default, means that the prefetch is preparing for a read.
4540 The value @var{locality} must be a compile-time constant integer between
4541 zero and three. A value of zero means that the data has no temporal
4542 locality, so it need not be left in the cache after the access. A value
4543 of three means that the data has a high degree of temporal locality and
4544 should be left in all levels of cache possible. Values of one and two
4545 mean, respectively, a low or moderate degree of temporal locality. The
4549 for (i = 0; i < n; i++)
4552 __builtin_prefetch (&a[i+j], 1, 1);
4553 __builtin_prefetch (&b[i+j], 0, 1);
4558 Data prefetch does not generate faults if @var{addr} is invalid, but
4559 the address expression itself must be valid. For example, a prefetch
4560 of @code{p->next} will not fault if @code{p->next} is not a valid
4561 address, but evaluation will fault if @code{p} is not a valid address.
4563 If the target does not support data prefetch, the address expression
4564 is evaluated if it includes side effects but no other code is generated
4565 and GCC does not issue a warning.
4569 @section Pragmas Accepted by GCC
4573 GCC supports several types of pragmas, primarily in order to compile
4574 code originally written for other compilers. Note that in general
4575 we do not recommend the use of pragmas; @xref{Function Attributes},
4576 for further explanation.
4584 @subsection ARM Pragmas
4586 The ARM target defines pragmas for controlling the default addition of
4587 @code{long_call} and @code{short_call} attributes to functions.
4588 @xref{Function Attributes}, for information about the effects of these
4593 @cindex pragma, long_calls
4594 Set all subsequent functions to have the @code{long_call} attribute.
4597 @cindex pragma, no_long_calls
4598 Set all subsequent functions to have the @code{short_call} attribute.
4600 @item long_calls_off
4601 @cindex pragma, long_calls_off
4602 Do not affect the @code{long_call} or @code{short_call} attributes of
4603 subsequent functions.
4606 @c Describe c4x pragmas here.
4607 @c Describe h8300 pragmas here.
4608 @c Describe i370 pragmas here.
4609 @c Describe i960 pragmas here.
4610 @c Describe sh pragmas here.
4611 @c Describe v850 pragmas here.
4613 @node Darwin Pragmas
4614 @subsection Darwin Pragmas
4616 The following pragmas are available for all architectures running the
4617 Darwin operating system. These are useful for compatibility with other
4621 @item mark @var{tokens}@dots{}
4622 @cindex pragma, mark
4623 This pragma is accepted, but has no effect.
4625 @item options align=@var{alignment}
4626 @cindex pragma, options align
4627 This pragma sets the alignment of fields in structures. The values of
4628 @var{alignment} may be @code{mac68k}, to emulate m68k alignment, or
4629 @code{power}, to emulate PowerPC alignment. Uses of this pragma nest
4630 properly; to restore the previous setting, use @code{reset} for the
4633 @item segment @var{tokens}@dots{}
4634 @cindex pragma, segment
4635 This pragma is accepted, but has no effect.
4637 @item unused (@var{var} [, @var{var}]@dots{})
4638 @cindex pragma, unused
4639 This pragma declares variables to be possibly unused. GCC will not
4640 produce warnings for the listed variables. The effect is similar to
4641 that of the @code{unused} attribute, except that this pragma may appear
4642 anywhere within the variables' scopes.
4645 @node Unnamed Fields
4646 @section Unnamed struct/union fields within structs/unions.
4650 For compatibility with other compilers, GCC allows you to define
4651 a structure or union that contains, as fields, structures and unions
4652 without names. For example:
4665 In this example, the user would be able to access members of the unnamed
4666 union with code like @samp{foo.b}. Note that only unnamed structs and
4667 unions are allowed, you may not have, for example, an unnamed
4670 You must never create such structures that cause ambiguous field definitions.
4671 For example, this structure:
4682 It is ambiguous which @code{a} is being referred to with @samp{foo.a}.
4683 Such constructs are not supported and must be avoided. In the future,
4684 such constructs may be detected and treated as compilation errors.
4686 @node C++ Extensions
4687 @chapter Extensions to the C++ Language
4688 @cindex extensions, C++ language
4689 @cindex C++ language extensions
4691 The GNU compiler provides these extensions to the C++ language (and you
4692 can also use most of the C language extensions in your C++ programs). If you
4693 want to write code that checks whether these features are available, you can
4694 test for the GNU compiler the same way as for C programs: check for a
4695 predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to
4696 test specifically for GNU C++ (@pxref{Standard Predefined,,Standard
4697 Predefined Macros,cpp.info,The C Preprocessor}).
4700 * Min and Max:: C++ Minimum and maximum operators.
4701 * Volatiles:: What constitutes an access to a volatile object.
4702 * Restricted Pointers:: C99 restricted pointers and references.
4703 * Vague Linkage:: Where G++ puts inlines, vtables and such.
4704 * C++ Interface:: You can use a single C++ header file for both
4705 declarations and definitions.
4706 * Template Instantiation:: Methods for ensuring that exactly one copy of
4707 each needed template instantiation is emitted.
4708 * Bound member functions:: You can extract a function pointer to the
4709 method denoted by a @samp{->*} or @samp{.*} expression.
4710 * C++ Attributes:: Variable, function, and type attributes for C++ only.
4711 * Java Exceptions:: Tweaking exception handling to work with Java.
4712 * Deprecated Features:: Things might disappear from g++.
4713 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
4717 @section Minimum and Maximum Operators in C++
4719 It is very convenient to have operators which return the ``minimum'' or the
4720 ``maximum'' of two arguments. In GNU C++ (but not in GNU C),
4723 @item @var{a} <? @var{b}
4725 @cindex minimum operator
4726 is the @dfn{minimum}, returning the smaller of the numeric values
4727 @var{a} and @var{b};
4729 @item @var{a} >? @var{b}
4731 @cindex maximum operator
4732 is the @dfn{maximum}, returning the larger of the numeric values @var{a}
4736 These operations are not primitive in ordinary C++, since you can
4737 use a macro to return the minimum of two things in C++, as in the
4741 #define MIN(X,Y) ((X) < (Y) ? : (X) : (Y))
4745 You might then use @w{@samp{int min = MIN (i, j);}} to set @var{min} to
4746 the minimum value of variables @var{i} and @var{j}.
4748 However, side effects in @code{X} or @code{Y} may cause unintended
4749 behavior. For example, @code{MIN (i++, j++)} will fail, incrementing
4750 the smaller counter twice. A GNU C extension allows you to write safe
4751 macros that avoid this kind of problem (@pxref{Naming Types,,Naming an
4752 Expression's Type}). However, writing @code{MIN} and @code{MAX} as
4753 macros also forces you to use function-call notation for a
4754 fundamental arithmetic operation. Using GNU C++ extensions, you can
4755 write @w{@samp{int min = i <? j;}} instead.
4757 Since @code{<?} and @code{>?} are built into the compiler, they properly
4758 handle expressions with side-effects; @w{@samp{int min = i++ <? j++;}}
4762 @section When is a Volatile Object Accessed?
4763 @cindex accessing volatiles
4764 @cindex volatile read
4765 @cindex volatile write
4766 @cindex volatile access
4768 Both the C and C++ standard have the concept of volatile objects. These
4769 are normally accessed by pointers and used for accessing hardware. The
4770 standards encourage compilers to refrain from optimizations
4771 concerning accesses to volatile objects that it might perform on
4772 non-volatile objects. The C standard leaves it implementation defined
4773 as to what constitutes a volatile access. The C++ standard omits to
4774 specify this, except to say that C++ should behave in a similar manner
4775 to C with respect to volatiles, where possible. The minimum either
4776 standard specifies is that at a sequence point all previous accesses to
4777 volatile objects have stabilized and no subsequent accesses have
4778 occurred. Thus an implementation is free to reorder and combine
4779 volatile accesses which occur between sequence points, but cannot do so
4780 for accesses across a sequence point. The use of volatiles does not
4781 allow you to violate the restriction on updating objects multiple times
4782 within a sequence point.
4784 In most expressions, it is intuitively obvious what is a read and what is
4785 a write. For instance
4788 volatile int *dst = @var{somevalue};
4789 volatile int *src = @var{someothervalue};
4794 will cause a read of the volatile object pointed to by @var{src} and stores the
4795 value into the volatile object pointed to by @var{dst}. There is no
4796 guarantee that these reads and writes are atomic, especially for objects
4797 larger than @code{int}.
4799 Less obvious expressions are where something which looks like an access
4800 is used in a void context. An example would be,
4803 volatile int *src = @var{somevalue};
4807 With C, such expressions are rvalues, and as rvalues cause a read of
4808 the object, GCC interprets this as a read of the volatile being pointed
4809 to. The C++ standard specifies that such expressions do not undergo
4810 lvalue to rvalue conversion, and that the type of the dereferenced
4811 object may be incomplete. The C++ standard does not specify explicitly
4812 that it is this lvalue to rvalue conversion which is responsible for
4813 causing an access. However, there is reason to believe that it is,
4814 because otherwise certain simple expressions become undefined. However,
4815 because it would surprise most programmers, G++ treats dereferencing a
4816 pointer to volatile object of complete type in a void context as a read
4817 of the object. When the object has incomplete type, G++ issues a
4822 struct T @{int m;@};
4823 volatile S *ptr1 = @var{somevalue};
4824 volatile T *ptr2 = @var{somevalue};
4829 In this example, a warning is issued for @code{*ptr1}, and @code{*ptr2}
4830 causes a read of the object pointed to. If you wish to force an error on
4831 the first case, you must force a conversion to rvalue with, for instance
4832 a static cast, @code{static_cast<S>(*ptr1)}.
4834 When using a reference to volatile, G++ does not treat equivalent
4835 expressions as accesses to volatiles, but instead issues a warning that
4836 no volatile is accessed. The rationale for this is that otherwise it
4837 becomes difficult to determine where volatile access occur, and not
4838 possible to ignore the return value from functions returning volatile
4839 references. Again, if you wish to force a read, cast the reference to
4842 @node Restricted Pointers
4843 @section Restricting Pointer Aliasing
4844 @cindex restricted pointers
4845 @cindex restricted references
4846 @cindex restricted this pointer
4848 As with gcc, g++ understands the C99 feature of restricted pointers,
4849 specified with the @code{__restrict__}, or @code{__restrict} type
4850 qualifier. Because you cannot compile C++ by specifying the @option{-std=c99}
4851 language flag, @code{restrict} is not a keyword in C++.
4853 In addition to allowing restricted pointers, you can specify restricted
4854 references, which indicate that the reference is not aliased in the local
4858 void fn (int *__restrict__ rptr, int &__restrict__ rref)
4865 In the body of @code{fn}, @var{rptr} points to an unaliased integer and
4866 @var{rref} refers to a (different) unaliased integer.
4868 You may also specify whether a member function's @var{this} pointer is
4869 unaliased by using @code{__restrict__} as a member function qualifier.
4872 void T::fn () __restrict__
4879 Within the body of @code{T::fn}, @var{this} will have the effective
4880 definition @code{T *__restrict__ const this}. Notice that the
4881 interpretation of a @code{__restrict__} member function qualifier is
4882 different to that of @code{const} or @code{volatile} qualifier, in that it
4883 is applied to the pointer rather than the object. This is consistent with
4884 other compilers which implement restricted pointers.
4886 As with all outermost parameter qualifiers, @code{__restrict__} is
4887 ignored in function definition matching. This means you only need to
4888 specify @code{__restrict__} in a function definition, rather than
4889 in a function prototype as well.
4892 @section Vague Linkage
4893 @cindex vague linkage
4895 There are several constructs in C++ which require space in the object
4896 file but are not clearly tied to a single translation unit. We say that
4897 these constructs have ``vague linkage''. Typically such constructs are
4898 emitted wherever they are needed, though sometimes we can be more
4902 @item Inline Functions
4903 Inline functions are typically defined in a header file which can be
4904 included in many different compilations. Hopefully they can usually be
4905 inlined, but sometimes an out-of-line copy is necessary, if the address
4906 of the function is taken or if inlining fails. In general, we emit an
4907 out-of-line copy in all translation units where one is needed. As an
4908 exception, we only emit inline virtual functions with the vtable, since
4909 it will always require a copy.
4911 Local static variables and string constants used in an inline function
4912 are also considered to have vague linkage, since they must be shared
4913 between all inlined and out-of-line instances of the function.
4917 C++ virtual functions are implemented in most compilers using a lookup
4918 table, known as a vtable. The vtable contains pointers to the virtual
4919 functions provided by a class, and each object of the class contains a
4920 pointer to its vtable (or vtables, in some multiple-inheritance
4921 situations). If the class declares any non-inline, non-pure virtual
4922 functions, the first one is chosen as the ``key method'' for the class,
4923 and the vtable is only emitted in the translation unit where the key
4926 @emph{Note:} If the chosen key method is later defined as inline, the
4927 vtable will still be emitted in every translation unit which defines it.
4928 Make sure that any inline virtuals are declared inline in the class
4929 body, even if they are not defined there.
4931 @item type_info objects
4934 C++ requires information about types to be written out in order to
4935 implement @samp{dynamic_cast}, @samp{typeid} and exception handling.
4936 For polymorphic classes (classes with virtual functions), the type_info
4937 object is written out along with the vtable so that @samp{dynamic_cast}
4938 can determine the dynamic type of a class object at runtime. For all
4939 other types, we write out the type_info object when it is used: when
4940 applying @samp{typeid} to an expression, throwing an object, or
4941 referring to a type in a catch clause or exception specification.
4943 @item Template Instantiations
4944 Most everything in this section also applies to template instantiations,
4945 but there are other options as well.
4946 @xref{Template Instantiation,,Where's the Template?}.
4950 When used with GNU ld version 2.8 or later on an ELF system such as
4951 Linux/GNU or Solaris 2, or on Microsoft Windows, duplicate copies of
4952 these constructs will be discarded at link time. This is known as
4955 On targets that don't support COMDAT, but do support weak symbols, GCC
4956 will use them. This way one copy will override all the others, but
4957 the unused copies will still take up space in the executable.
4959 For targets which do not support either COMDAT or weak symbols,
4960 most entities with vague linkage will be emitted as local symbols to
4961 avoid duplicate definition errors from the linker. This will not happen
4962 for local statics in inlines, however, as having multiple copies will
4963 almost certainly break things.
4965 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
4966 another way to control placement of these constructs.
4969 @section Declarations and Definitions in One Header
4971 @cindex interface and implementation headers, C++
4972 @cindex C++ interface and implementation headers
4973 C++ object definitions can be quite complex. In principle, your source
4974 code will need two kinds of things for each object that you use across
4975 more than one source file. First, you need an @dfn{interface}
4976 specification, describing its structure with type declarations and
4977 function prototypes. Second, you need the @dfn{implementation} itself.
4978 It can be tedious to maintain a separate interface description in a
4979 header file, in parallel to the actual implementation. It is also
4980 dangerous, since separate interface and implementation definitions may
4981 not remain parallel.
4983 @cindex pragmas, interface and implementation
4984 With GNU C++, you can use a single header file for both purposes.
4987 @emph{Warning:} The mechanism to specify this is in transition. For the
4988 nonce, you must use one of two @code{#pragma} commands; in a future
4989 release of GNU C++, an alternative mechanism will make these
4990 @code{#pragma} commands unnecessary.
4993 The header file contains the full definitions, but is marked with
4994 @samp{#pragma interface} in the source code. This allows the compiler
4995 to use the header file only as an interface specification when ordinary
4996 source files incorporate it with @code{#include}. In the single source
4997 file where the full implementation belongs, you can use either a naming
4998 convention or @samp{#pragma implementation} to indicate this alternate
4999 use of the header file.
5002 @item #pragma interface
5003 @itemx #pragma interface "@var{subdir}/@var{objects}.h"
5004 @kindex #pragma interface
5005 Use this directive in @emph{header files} that define object classes, to save
5006 space in most of the object files that use those classes. Normally,
5007 local copies of certain information (backup copies of inline member
5008 functions, debugging information, and the internal tables that implement
5009 virtual functions) must be kept in each object file that includes class
5010 definitions. You can use this pragma to avoid such duplication. When a
5011 header file containing @samp{#pragma interface} is included in a
5012 compilation, this auxiliary information will not be generated (unless
5013 the main input source file itself uses @samp{#pragma implementation}).
5014 Instead, the object files will contain references to be resolved at link
5017 The second form of this directive is useful for the case where you have
5018 multiple headers with the same name in different directories. If you
5019 use this form, you must specify the same string to @samp{#pragma
5022 @item #pragma implementation
5023 @itemx #pragma implementation "@var{objects}.h"
5024 @kindex #pragma implementation
5025 Use this pragma in a @emph{main input file}, when you want full output from
5026 included header files to be generated (and made globally visible). The
5027 included header file, in turn, should use @samp{#pragma interface}.
5028 Backup copies of inline member functions, debugging information, and the
5029 internal tables used to implement virtual functions are all generated in
5030 implementation files.
5032 @cindex implied @code{#pragma implementation}
5033 @cindex @code{#pragma implementation}, implied
5034 @cindex naming convention, implementation headers
5035 If you use @samp{#pragma implementation} with no argument, it applies to
5036 an include file with the same basename@footnote{A file's @dfn{basename}
5037 was the name stripped of all leading path information and of trailing
5038 suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source
5039 file. For example, in @file{allclass.cc}, giving just
5040 @samp{#pragma implementation}
5041 by itself is equivalent to @samp{#pragma implementation "allclass.h"}.
5043 In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as
5044 an implementation file whenever you would include it from
5045 @file{allclass.cc} even if you never specified @samp{#pragma
5046 implementation}. This was deemed to be more trouble than it was worth,
5047 however, and disabled.
5049 If you use an explicit @samp{#pragma implementation}, it must appear in
5050 your source file @emph{before} you include the affected header files.
5052 Use the string argument if you want a single implementation file to
5053 include code from multiple header files. (You must also use
5054 @samp{#include} to include the header file; @samp{#pragma
5055 implementation} only specifies how to use the file---it doesn't actually
5058 There is no way to split up the contents of a single header file into
5059 multiple implementation files.
5062 @cindex inlining and C++ pragmas
5063 @cindex C++ pragmas, effect on inlining
5064 @cindex pragmas in C++, effect on inlining
5065 @samp{#pragma implementation} and @samp{#pragma interface} also have an
5066 effect on function inlining.
5068 If you define a class in a header file marked with @samp{#pragma
5069 interface}, the effect on a function defined in that class is similar to
5070 an explicit @code{extern} declaration---the compiler emits no code at
5071 all to define an independent version of the function. Its definition
5072 is used only for inlining with its callers.
5074 @opindex fno-implement-inlines
5075 Conversely, when you include the same header file in a main source file
5076 that declares it as @samp{#pragma implementation}, the compiler emits
5077 code for the function itself; this defines a version of the function
5078 that can be found via pointers (or by callers compiled without
5079 inlining). If all calls to the function can be inlined, you can avoid
5080 emitting the function by compiling with @option{-fno-implement-inlines}.
5081 If any calls were not inlined, you will get linker errors.
5083 @node Template Instantiation
5084 @section Where's the Template?
5086 @cindex template instantiation
5088 C++ templates are the first language feature to require more
5089 intelligence from the environment than one usually finds on a UNIX
5090 system. Somehow the compiler and linker have to make sure that each
5091 template instance occurs exactly once in the executable if it is needed,
5092 and not at all otherwise. There are two basic approaches to this
5093 problem, which I will refer to as the Borland model and the Cfront model.
5097 Borland C++ solved the template instantiation problem by adding the code
5098 equivalent of common blocks to their linker; the compiler emits template
5099 instances in each translation unit that uses them, and the linker
5100 collapses them together. The advantage of this model is that the linker
5101 only has to consider the object files themselves; there is no external
5102 complexity to worry about. This disadvantage is that compilation time
5103 is increased because the template code is being compiled repeatedly.
5104 Code written for this model tends to include definitions of all
5105 templates in the header file, since they must be seen to be
5109 The AT&T C++ translator, Cfront, solved the template instantiation
5110 problem by creating the notion of a template repository, an
5111 automatically maintained place where template instances are stored. A
5112 more modern version of the repository works as follows: As individual
5113 object files are built, the compiler places any template definitions and
5114 instantiations encountered in the repository. At link time, the link
5115 wrapper adds in the objects in the repository and compiles any needed
5116 instances that were not previously emitted. The advantages of this
5117 model are more optimal compilation speed and the ability to use the
5118 system linker; to implement the Borland model a compiler vendor also
5119 needs to replace the linker. The disadvantages are vastly increased
5120 complexity, and thus potential for error; for some code this can be
5121 just as transparent, but in practice it can been very difficult to build
5122 multiple programs in one directory and one program in multiple
5123 directories. Code written for this model tends to separate definitions
5124 of non-inline member templates into a separate file, which should be
5125 compiled separately.
5128 When used with GNU ld version 2.8 or later on an ELF system such as
5129 Linux/GNU or Solaris 2, or on Microsoft Windows, g++ supports the
5130 Borland model. On other systems, g++ implements neither automatic
5133 A future version of g++ will support a hybrid model whereby the compiler
5134 will emit any instantiations for which the template definition is
5135 included in the compile, and store template definitions and
5136 instantiation context information into the object file for the rest.
5137 The link wrapper will extract that information as necessary and invoke
5138 the compiler to produce the remaining instantiations. The linker will
5139 then combine duplicate instantiations.
5141 In the mean time, you have the following options for dealing with
5142 template instantiations:
5147 Compile your template-using code with @option{-frepo}. The compiler will
5148 generate files with the extension @samp{.rpo} listing all of the
5149 template instantiations used in the corresponding object files which
5150 could be instantiated there; the link wrapper, @samp{collect2}, will
5151 then update the @samp{.rpo} files to tell the compiler where to place
5152 those instantiations and rebuild any affected object files. The
5153 link-time overhead is negligible after the first pass, as the compiler
5154 will continue to place the instantiations in the same files.
5156 This is your best option for application code written for the Borland
5157 model, as it will just work. Code written for the Cfront model will
5158 need to be modified so that the template definitions are available at
5159 one or more points of instantiation; usually this is as simple as adding
5160 @code{#include <tmethods.cc>} to the end of each template header.
5162 For library code, if you want the library to provide all of the template
5163 instantiations it needs, just try to link all of its object files
5164 together; the link will fail, but cause the instantiations to be
5165 generated as a side effect. Be warned, however, that this may cause
5166 conflicts if multiple libraries try to provide the same instantiations.
5167 For greater control, use explicit instantiation as described in the next
5171 @opindex fno-implicit-templates
5172 Compile your code with @option{-fno-implicit-templates} to disable the
5173 implicit generation of template instances, and explicitly instantiate
5174 all the ones you use. This approach requires more knowledge of exactly
5175 which instances you need than do the others, but it's less
5176 mysterious and allows greater control. You can scatter the explicit
5177 instantiations throughout your program, perhaps putting them in the
5178 translation units where the instances are used or the translation units
5179 that define the templates themselves; you can put all of the explicit
5180 instantiations you need into one big file; or you can create small files
5187 template class Foo<int>;
5188 template ostream& operator <<
5189 (ostream&, const Foo<int>&);
5192 for each of the instances you need, and create a template instantiation
5195 If you are using Cfront-model code, you can probably get away with not
5196 using @option{-fno-implicit-templates} when compiling files that don't
5197 @samp{#include} the member template definitions.
5199 If you use one big file to do the instantiations, you may want to
5200 compile it without @option{-fno-implicit-templates} so you get all of the
5201 instances required by your explicit instantiations (but not by any
5202 other files) without having to specify them as well.
5204 g++ has extended the template instantiation syntax outlined in the
5205 Working Paper to allow forward declaration of explicit instantiations
5206 (with @code{extern}), instantiation of the compiler support data for a
5207 template class (i.e.@: the vtable) without instantiating any of its
5208 members (with @code{inline}), and instantiation of only the static data
5209 members of a template class, without the support data or member
5210 functions (with (@code{static}):
5213 extern template int max (int, int);
5214 inline template class Foo<int>;
5215 static template class Foo<int>;
5219 Do nothing. Pretend g++ does implement automatic instantiation
5220 management. Code written for the Borland model will work fine, but
5221 each translation unit will contain instances of each of the templates it
5222 uses. In a large program, this can lead to an unacceptable amount of code
5226 @opindex fexternal-templates
5227 Add @samp{#pragma interface} to all files containing template
5228 definitions. For each of these files, add @samp{#pragma implementation
5229 "@var{filename}"} to the top of some @samp{.C} file which
5230 @samp{#include}s it. Then compile everything with
5231 @option{-fexternal-templates}. The templates will then only be expanded
5232 in the translation unit which implements them (i.e.@: has a @samp{#pragma
5233 implementation} line for the file where they live); all other files will
5234 use external references. If you're lucky, everything should work
5235 properly. If you get undefined symbol errors, you need to make sure
5236 that each template instance which is used in the program is used in the
5237 file which implements that template. If you don't have any use for a
5238 particular instance in that file, you can just instantiate it
5239 explicitly, using the syntax from the latest C++ working paper:
5242 template class A<int>;
5243 template ostream& operator << (ostream&, const A<int>&);
5246 This strategy will work with code written for either model. If you are
5247 using code written for the Cfront model, the file containing a class
5248 template and the file containing its member templates should be
5249 implemented in the same translation unit.
5252 @opindex falt-external-templates
5253 A slight variation on this approach is to use the flag
5254 @option{-falt-external-templates} instead. This flag causes template
5255 instances to be emitted in the translation unit that implements the
5256 header where they are first instantiated, rather than the one which
5257 implements the file where the templates are defined. This header must
5258 be the same in all translation units, or things are likely to break.
5260 @xref{C++ Interface,,Declarations and Definitions in One Header}, for
5261 more discussion of these pragmas.
5264 @node Bound member functions
5265 @section Extracting the function pointer from a bound pointer to member function
5268 @cindex pointer to member function
5269 @cindex bound pointer to member function
5271 In C++, pointer to member functions (PMFs) are implemented using a wide
5272 pointer of sorts to handle all the possible call mechanisms; the PMF
5273 needs to store information about how to adjust the @samp{this} pointer,
5274 and if the function pointed to is virtual, where to find the vtable, and
5275 where in the vtable to look for the member function. If you are using
5276 PMFs in an inner loop, you should really reconsider that decision. If
5277 that is not an option, you can extract the pointer to the function that
5278 would be called for a given object/PMF pair and call it directly inside
5279 the inner loop, to save a bit of time.
5281 Note that you will still be paying the penalty for the call through a
5282 function pointer; on most modern architectures, such a call defeats the
5283 branch prediction features of the CPU@. This is also true of normal
5284 virtual function calls.
5286 The syntax for this extension is
5290 extern int (A::*fp)();
5291 typedef int (*fptr)(A *);
5293 fptr p = (fptr)(a.*fp);
5296 For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}),
5297 no object is needed to obtain the address of the function. They can be
5298 converted to function pointers directly:
5301 fptr p1 = (fptr)(&A::foo);
5304 @opindex Wno-pmf-conversions
5305 You must specify @option{-Wno-pmf-conversions} to use this extension.
5307 @node C++ Attributes
5308 @section C++-Specific Variable, Function, and Type Attributes
5310 Some attributes only make sense for C++ programs.
5313 @item init_priority (@var{priority})
5314 @cindex init_priority attribute
5317 In Standard C++, objects defined at namespace scope are guaranteed to be
5318 initialized in an order in strict accordance with that of their definitions
5319 @emph{in a given translation unit}. No guarantee is made for initializations
5320 across translation units. However, GNU C++ allows users to control the
5321 order of initialization of objects defined at namespace scope with the
5322 @code{init_priority} attribute by specifying a relative @var{priority},
5323 a constant integral expression currently bounded between 101 and 65535
5324 inclusive. Lower numbers indicate a higher priority.
5326 In the following example, @code{A} would normally be created before
5327 @code{B}, but the @code{init_priority} attribute has reversed that order:
5330 Some_Class A __attribute__ ((init_priority (2000)));
5331 Some_Class B __attribute__ ((init_priority (543)));
5335 Note that the particular values of @var{priority} do not matter; only their
5338 @item java_interface
5339 @cindex java_interface attribute
5341 This type attribute informs C++ that the class is a Java interface. It may
5342 only be applied to classes declared within an @code{extern "Java"} block.
5343 Calls to methods declared in this interface will be dispatched using GCJ's
5344 interface table mechanism, instead of regular virtual table dispatch.
5348 @node Java Exceptions
5349 @section Java Exceptions
5351 The Java language uses a slightly different exception handling model
5352 from C++. Normally, GNU C++ will automatically detect when you are
5353 writing C++ code that uses Java exceptions, and handle them
5354 appropriately. However, if C++ code only needs to execute destructors
5355 when Java exceptions are thrown through it, GCC will guess incorrectly.
5356 Sample problematic code is:
5359 struct S @{ ~S(); @};
5360 extern void bar(); // is written in Java, and may throw exceptions
5369 The usual effect of an incorrect guess is a link failure, complaining of
5370 a missing routine called @samp{__gxx_personality_v0}.
5372 You can inform the compiler that Java exceptions are to be used in a
5373 translation unit, irrespective of what it might think, by writing
5374 @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This
5375 @samp{#pragma} must appear before any functions that throw or catch
5376 exceptions, or run destructors when exceptions are thrown through them.
5378 You cannot mix Java and C++ exceptions in the same translation unit. It
5379 is believed to be safe to throw a C++ exception from one file through
5380 another file compiled for the Java exception model, or vice versa, but
5381 there may be bugs in this area.
5383 @node Deprecated Features
5384 @section Deprecated Features
5386 In the past, the GNU C++ compiler was extended to experiment with new
5387 features, at a time when the C++ language was still evolving. Now that
5388 the C++ standard is complete, some of those features are superseded by
5389 superior alternatives. Using the old features might cause a warning in
5390 some cases that the feature will be dropped in the future. In other
5391 cases, the feature might be gone already.
5393 While the list below is not exhaustive, it documents some of the options
5394 that are now deprecated:
5397 @item -fexternal-templates
5398 @itemx -falt-external-templates
5399 These are two of the many ways for g++ to implement template
5400 instantiation. @xref{Template Instantiation}. The C++ standard clearly
5401 defines how template definitions have to be organized across
5402 implementation units. g++ has an implicit instantiation mechanism that
5403 should work just fine for standard-conforming code.
5405 @item -fstrict-prototype
5406 @itemx -fno-strict-prototype
5407 Previously it was possible to use an empty prototype parameter list to
5408 indicate an unspecified number of parameters (like C), rather than no
5409 parameters, as C++ demands. This feature has been removed, except where
5410 it is required for backwards compatibility @xref{Backwards Compatibility}.
5413 The named return value extension has been deprecated, and is now
5416 The use of initializer lists with new expressions has been deprecated,
5417 and is now removed from g++.
5419 Floating and complex non-type template parameters have been deprecated,
5420 and are now removed from g++.
5422 The implicit typename extension has been deprecated and will be removed
5423 from g++ at some point. In some cases g++ determines that a dependant
5424 type such as @code{TPL<T>::X} is a type without needing a
5425 @code{typename} keyword, contrary to the standard.
5427 @node Backwards Compatibility
5428 @section Backwards Compatibility
5429 @cindex Backwards Compatibility
5430 @cindex ARM [Annotated C++ Reference Manual]
5432 Now that there is a definitive ISO standard C++, G++ has a specification
5433 to adhere to. The C++ language evolved over time, and features that
5434 used to be acceptable in previous drafts of the standard, such as the ARM
5435 [Annotated C++ Reference Manual], are no longer accepted. In order to allow
5436 compilation of C++ written to such drafts, G++ contains some backwards
5437 compatibilities. @emph{All such backwards compatibility features are
5438 liable to disappear in future versions of G++.} They should be considered
5439 deprecated @xref{Deprecated Features}.
5443 If a variable is declared at for scope, it used to remain in scope until
5444 the end of the scope which contained the for statement (rather than just
5445 within the for scope). G++ retains this, but issues a warning, if such a
5446 variable is accessed outside the for scope.
5448 @item Implicit C language
5449 Old C system header files did not contain an @code{extern "C" @{@dots{}@}}
5450 scope to set the language. On such systems, all header files are
5451 implicitly scoped inside a C language scope. Also, an empty prototype
5452 @code{()} will be treated as an unspecified number of arguments, rather
5453 than no arguments, as C++ demands.