]>
Commit | Line | Data |
---|---|---|
5c25e11d | 1 | @c Copyright (C) 1988,89,92,93,94,96,98,99,2000,01 Free Software Foundation, Inc. |
c1f7febf RK |
2 | @c This is part of the GCC manual. |
3 | @c For copying conditions, see the file gcc.texi. | |
4 | ||
5 | @node C Extensions | |
6 | @chapter Extensions to the C Language Family | |
7 | @cindex extensions, C language | |
8 | @cindex C language extensions | |
9 | ||
5490d604 | 10 | GNU C provides several language features not found in ISO standard C. |
c1f7febf RK |
11 | (The @samp{-pedantic} option directs GNU CC to print a warning message if |
12 | any of these features is used.) To test for the availability of these | |
13 | features in conditional compilation, check for a predefined macro | |
14 | @code{__GNUC__}, which is always defined under GNU CC. | |
15 | ||
16 | These extensions are available in C and Objective C. Most of them are | |
17 | also available in C++. @xref{C++ Extensions,,Extensions to the | |
18 | C++ Language}, for extensions that apply @emph{only} to C++. | |
19 | ||
5490d604 JM |
20 | @c FIXME: document clearly which features are in ISO C99, but also |
21 | @c accepted as extensions for -std=gnu89 and possibly for C++. | |
22 | @c See PR other/930. | |
23 | ||
24 | @c FIXME: the documentation for preprocessor extensions here is out of | |
25 | @c date. See PR other/928. | |
26 | ||
c1f7febf RK |
27 | @c The only difference between the two versions of this menu is that the |
28 | @c version for clear INTERNALS has an extra node, "Constraints" (which | |
29 | @c appears in a separate chapter in the other version of the manual). | |
30 | @ifset INTERNALS | |
31 | @menu | |
32 | * Statement Exprs:: Putting statements and declarations inside expressions. | |
33 | * Local Labels:: Labels local to a statement-expression. | |
34 | * Labels as Values:: Getting pointers to labels, and computed gotos. | |
35 | * Nested Functions:: As in Algol and Pascal, lexical scoping of functions. | |
36 | * Constructing Calls:: Dispatching a call to another function. | |
37 | * Naming Types:: Giving a name to the type of some expression. | |
38 | * Typeof:: @code{typeof}: referring to the type of an expression. | |
39 | * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues. | |
40 | * Conditionals:: Omitting the middle operand of a @samp{?:} expression. | |
41 | * Long Long:: Double-word integers---@code{long long int}. | |
42 | * Complex:: Data types for complex numbers. | |
6f4d7222 | 43 | * Hex Floats:: Hexadecimal floating-point constants. |
c1f7febf RK |
44 | * Zero Length:: Zero-length arrays. |
45 | * Variable Length:: Arrays whose length is computed at run time. | |
ccd96f0a NB |
46 | * Variadic Macros:: Macros with a variable number of arguments. |
47 | * Escaped Newlines:: Slightly looser rules for escaped newlines. | |
48 | * Multi-line Strings:: String literals with embedded newlines. | |
c1f7febf RK |
49 | * Subscripting:: Any array can be subscripted, even if not an lvalue. |
50 | * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers. | |
51 | * Initializers:: Non-constant initializers. | |
52 | * Constructors:: Constructor expressions give structures, unions | |
53 | or arrays as values. | |
54 | * Labeled Elements:: Labeling elements of initializers. | |
55 | * Cast to Union:: Casting to union type from any member of the union. | |
56 | * Case Ranges:: `case 1 ... 9' and such. | |
57 | * Function Attributes:: Declaring that functions have no side effects, | |
58 | or that they can never return. | |
2c5e91d2 | 59 | * Attribute Syntax:: Formal syntax for attributes. |
c1f7febf RK |
60 | * Function Prototypes:: Prototype declarations and old-style definitions. |
61 | * C++ Comments:: C++ comments are recognized. | |
62 | * Dollar Signs:: Dollar sign is allowed in identifiers. | |
63 | * Character Escapes:: @samp{\e} stands for the character @key{ESC}. | |
64 | * Variable Attributes:: Specifying attributes of variables. | |
65 | * Type Attributes:: Specifying attributes of types. | |
66 | * Alignment:: Inquiring about the alignment of a type or variable. | |
67 | * Inline:: Defining inline functions (as fast as macros). | |
68 | * Extended Asm:: Assembler instructions with C expressions as operands. | |
69 | (With them you can define ``built-in'' functions.) | |
70 | * Asm Labels:: Specifying the assembler name to use for a C symbol. | |
71 | * Explicit Reg Vars:: Defining variables residing in specified registers. | |
72 | * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files. | |
73 | * Incomplete Enums:: @code{enum foo;}, with details to follow. | |
74 | * Function Names:: Printable strings which are the name of the current | |
75 | function. | |
76 | * Return Address:: Getting the return or frame address of a function. | |
185ebd6c | 77 | * Other Builtins:: Other built-in functions. |
c1f7febf RK |
78 | @end menu |
79 | @end ifset | |
80 | @ifclear INTERNALS | |
81 | @menu | |
82 | * Statement Exprs:: Putting statements and declarations inside expressions. | |
83 | * Local Labels:: Labels local to a statement-expression. | |
84 | * Labels as Values:: Getting pointers to labels, and computed gotos. | |
85 | * Nested Functions:: As in Algol and Pascal, lexical scoping of functions. | |
86 | * Constructing Calls:: Dispatching a call to another function. | |
87 | * Naming Types:: Giving a name to the type of some expression. | |
88 | * Typeof:: @code{typeof}: referring to the type of an expression. | |
89 | * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues. | |
90 | * Conditionals:: Omitting the middle operand of a @samp{?:} expression. | |
91 | * Long Long:: Double-word integers---@code{long long int}. | |
92 | * Complex:: Data types for complex numbers. | |
6f4d7222 | 93 | * Hex Floats:: Hexadecimal floating-point constants. |
c1f7febf RK |
94 | * Zero Length:: Zero-length arrays. |
95 | * Variable Length:: Arrays whose length is computed at run time. | |
ccd96f0a NB |
96 | * Variadic Macros:: Macros with a variable number of arguments. |
97 | * Escaped Newlines:: Slightly looser rules for escaped newlines. | |
98 | * Multi-line Strings:: String literals with embedded newlines. | |
c1f7febf RK |
99 | * Subscripting:: Any array can be subscripted, even if not an lvalue. |
100 | * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers. | |
101 | * Initializers:: Non-constant initializers. | |
102 | * Constructors:: Constructor expressions give structures, unions | |
103 | or arrays as values. | |
104 | * Labeled Elements:: Labeling elements of initializers. | |
105 | * Cast to Union:: Casting to union type from any member of the union. | |
106 | * Case Ranges:: `case 1 ... 9' and such. | |
107 | * Function Attributes:: Declaring that functions have no side effects, | |
108 | or that they can never return. | |
2c5e91d2 | 109 | * Attribute Syntax:: Formal syntax for attributes. |
c1f7febf RK |
110 | * Function Prototypes:: Prototype declarations and old-style definitions. |
111 | * C++ Comments:: C++ comments are recognized. | |
112 | * Dollar Signs:: Dollar sign is allowed in identifiers. | |
113 | * Character Escapes:: @samp{\e} stands for the character @key{ESC}. | |
114 | * Variable Attributes:: Specifying attributes of variables. | |
115 | * Type Attributes:: Specifying attributes of types. | |
116 | * Alignment:: Inquiring about the alignment of a type or variable. | |
117 | * Inline:: Defining inline functions (as fast as macros). | |
118 | * Extended Asm:: Assembler instructions with C expressions as operands. | |
119 | (With them you can define ``built-in'' functions.) | |
120 | * Constraints:: Constraints for asm operands | |
121 | * Asm Labels:: Specifying the assembler name to use for a C symbol. | |
122 | * Explicit Reg Vars:: Defining variables residing in specified registers. | |
123 | * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files. | |
124 | * Incomplete Enums:: @code{enum foo;}, with details to follow. | |
125 | * Function Names:: Printable strings which are the name of the current | |
126 | function. | |
127 | * Return Address:: Getting the return or frame address of a function. | |
c5c76735 | 128 | * Other Builtins:: Other built-in functions. |
c1f7febf RK |
129 | @end menu |
130 | @end ifclear | |
131 | ||
132 | @node Statement Exprs | |
133 | @section Statements and Declarations in Expressions | |
134 | @cindex statements inside expressions | |
135 | @cindex declarations inside expressions | |
136 | @cindex expressions containing statements | |
137 | @cindex macros, statements in expressions | |
138 | ||
139 | @c the above section title wrapped and causes an underfull hbox.. i | |
140 | @c changed it from "within" to "in". --mew 4feb93 | |
141 | ||
142 | A compound statement enclosed in parentheses may appear as an expression | |
143 | in GNU C. This allows you to use loops, switches, and local variables | |
144 | within an expression. | |
145 | ||
146 | Recall that a compound statement is a sequence of statements surrounded | |
147 | by braces; in this construct, parentheses go around the braces. For | |
148 | example: | |
149 | ||
150 | @example | |
151 | (@{ int y = foo (); int z; | |
152 | if (y > 0) z = y; | |
153 | else z = - y; | |
154 | z; @}) | |
155 | @end example | |
156 | ||
157 | @noindent | |
158 | is a valid (though slightly more complex than necessary) expression | |
159 | for the absolute value of @code{foo ()}. | |
160 | ||
161 | The last thing in the compound statement should be an expression | |
162 | followed by a semicolon; the value of this subexpression serves as the | |
163 | value of the entire construct. (If you use some other kind of statement | |
164 | last within the braces, the construct has type @code{void}, and thus | |
165 | effectively no value.) | |
166 | ||
167 | This feature is especially useful in making macro definitions ``safe'' (so | |
168 | that they evaluate each operand exactly once). For example, the | |
169 | ``maximum'' function is commonly defined as a macro in standard C as | |
170 | follows: | |
171 | ||
172 | @example | |
173 | #define max(a,b) ((a) > (b) ? (a) : (b)) | |
174 | @end example | |
175 | ||
176 | @noindent | |
177 | @cindex side effects, macro argument | |
178 | But this definition computes either @var{a} or @var{b} twice, with bad | |
179 | results if the operand has side effects. In GNU C, if you know the | |
180 | type of the operands (here let's assume @code{int}), you can define | |
181 | the macro safely as follows: | |
182 | ||
183 | @example | |
184 | #define maxint(a,b) \ | |
185 | (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @}) | |
186 | @end example | |
187 | ||
188 | Embedded statements are not allowed in constant expressions, such as | |
189 | the value of an enumeration constant, the width of a bit field, or | |
190 | the initial value of a static variable. | |
191 | ||
192 | If you don't know the type of the operand, you can still do this, but you | |
193 | must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming | |
194 | Types}). | |
195 | ||
b98e139b MM |
196 | Statement expressions are not supported fully in G++, and their fate |
197 | there is unclear. (It is possible that they will become fully supported | |
198 | at some point, or that they will be deprecated, or that the bugs that | |
199 | are present will continue to exist indefinitely.) Presently, statement | |
200 | expressions do not work well as default arguments. | |
201 | ||
202 | In addition, there are semantic issues with statement-expressions in | |
203 | C++. If you try to use statement-expressions instead of inline | |
204 | functions in C++, you may be surprised at the way object destruction is | |
205 | handled. For example: | |
206 | ||
207 | @example | |
208 | #define foo(a) (@{int b = (a); b + 3; @}) | |
209 | @end example | |
210 | ||
211 | @noindent | |
212 | does not work the same way as: | |
213 | ||
214 | @example | |
54e1d3a6 | 215 | inline int foo(int a) @{ int b = a; return b + 3; @} |
b98e139b MM |
216 | @end example |
217 | ||
218 | @noindent | |
219 | In particular, if the expression passed into @code{foo} involves the | |
220 | creation of temporaries, the destructors for those temporaries will be | |
221 | run earlier in the case of the macro than in the case of the function. | |
222 | ||
223 | These considerations mean that it is probably a bad idea to use | |
224 | statement-expressions of this form in header files that are designed to | |
54e1d3a6 MM |
225 | work with C++. (Note that some versions of the GNU C Library contained |
226 | header files using statement-expression that lead to precisely this | |
227 | bug.) | |
b98e139b | 228 | |
c1f7febf RK |
229 | @node Local Labels |
230 | @section Locally Declared Labels | |
231 | @cindex local labels | |
232 | @cindex macros, local labels | |
233 | ||
234 | Each statement expression is a scope in which @dfn{local labels} can be | |
235 | declared. A local label is simply an identifier; you can jump to it | |
236 | with an ordinary @code{goto} statement, but only from within the | |
237 | statement expression it belongs to. | |
238 | ||
239 | A local label declaration looks like this: | |
240 | ||
241 | @example | |
242 | __label__ @var{label}; | |
243 | @end example | |
244 | ||
245 | @noindent | |
246 | or | |
247 | ||
248 | @example | |
249 | __label__ @var{label1}, @var{label2}, @dots{}; | |
250 | @end example | |
251 | ||
252 | Local label declarations must come at the beginning of the statement | |
253 | expression, right after the @samp{(@{}, before any ordinary | |
254 | declarations. | |
255 | ||
256 | The label declaration defines the label @emph{name}, but does not define | |
257 | the label itself. You must do this in the usual way, with | |
258 | @code{@var{label}:}, within the statements of the statement expression. | |
259 | ||
260 | The local label feature is useful because statement expressions are | |
261 | often used in macros. If the macro contains nested loops, a @code{goto} | |
262 | can be useful for breaking out of them. However, an ordinary label | |
263 | whose scope is the whole function cannot be used: if the macro can be | |
264 | expanded several times in one function, the label will be multiply | |
265 | defined in that function. A local label avoids this problem. For | |
266 | example: | |
267 | ||
268 | @example | |
269 | #define SEARCH(array, target) \ | |
270 | (@{ \ | |
271 | __label__ found; \ | |
272 | typeof (target) _SEARCH_target = (target); \ | |
273 | typeof (*(array)) *_SEARCH_array = (array); \ | |
274 | int i, j; \ | |
275 | int value; \ | |
276 | for (i = 0; i < max; i++) \ | |
277 | for (j = 0; j < max; j++) \ | |
278 | if (_SEARCH_array[i][j] == _SEARCH_target) \ | |
279 | @{ value = i; goto found; @} \ | |
280 | value = -1; \ | |
281 | found: \ | |
282 | value; \ | |
283 | @}) | |
284 | @end example | |
285 | ||
286 | @node Labels as Values | |
287 | @section Labels as Values | |
288 | @cindex labels as values | |
289 | @cindex computed gotos | |
290 | @cindex goto with computed label | |
291 | @cindex address of a label | |
292 | ||
293 | You can get the address of a label defined in the current function | |
294 | (or a containing function) with the unary operator @samp{&&}. The | |
295 | value has type @code{void *}. This value is a constant and can be used | |
296 | wherever a constant of that type is valid. For example: | |
297 | ||
298 | @example | |
299 | void *ptr; | |
300 | @dots{} | |
301 | ptr = &&foo; | |
302 | @end example | |
303 | ||
304 | To use these values, you need to be able to jump to one. This is done | |
305 | with the computed goto statement@footnote{The analogous feature in | |
306 | Fortran is called an assigned goto, but that name seems inappropriate in | |
307 | C, where one can do more than simply store label addresses in label | |
308 | variables.}, @code{goto *@var{exp};}. For example, | |
309 | ||
310 | @example | |
311 | goto *ptr; | |
312 | @end example | |
313 | ||
314 | @noindent | |
315 | Any expression of type @code{void *} is allowed. | |
316 | ||
317 | One way of using these constants is in initializing a static array that | |
318 | will serve as a jump table: | |
319 | ||
320 | @example | |
321 | static void *array[] = @{ &&foo, &&bar, &&hack @}; | |
322 | @end example | |
323 | ||
324 | Then you can select a label with indexing, like this: | |
325 | ||
326 | @example | |
327 | goto *array[i]; | |
328 | @end example | |
329 | ||
330 | @noindent | |
331 | Note that this does not check whether the subscript is in bounds---array | |
332 | indexing in C never does that. | |
333 | ||
334 | Such an array of label values serves a purpose much like that of the | |
335 | @code{switch} statement. The @code{switch} statement is cleaner, so | |
336 | use that rather than an array unless the problem does not fit a | |
337 | @code{switch} statement very well. | |
338 | ||
339 | Another use of label values is in an interpreter for threaded code. | |
340 | The labels within the interpreter function can be stored in the | |
341 | threaded code for super-fast dispatching. | |
342 | ||
47620e09 RH |
343 | You may not use this mechanism to jump to code in a different function. |
344 | If you do that, totally unpredictable things will happen. The best way to | |
c1f7febf RK |
345 | avoid this is to store the label address only in automatic variables and |
346 | never pass it as an argument. | |
347 | ||
47620e09 RH |
348 | An alternate way to write the above example is |
349 | ||
350 | @example | |
351 | static const int array[] = @{ &&foo - &&foo, &&bar - &&foo, &&hack - &&foo @}; | |
352 | goto *(&&foo + array[i]); | |
353 | @end example | |
354 | ||
355 | @noindent | |
356 | This is more friendly to code living in shared libraries, as it reduces | |
357 | the number of dynamic relocations that are needed, and by consequence, | |
358 | allows the data to be read-only. | |
359 | ||
c1f7febf RK |
360 | @node Nested Functions |
361 | @section Nested Functions | |
362 | @cindex nested functions | |
363 | @cindex downward funargs | |
364 | @cindex thunks | |
365 | ||
366 | A @dfn{nested function} is a function defined inside another function. | |
367 | (Nested functions are not supported for GNU C++.) The nested function's | |
368 | name is local to the block where it is defined. For example, here we | |
369 | define a nested function named @code{square}, and call it twice: | |
370 | ||
371 | @example | |
372 | @group | |
373 | foo (double a, double b) | |
374 | @{ | |
375 | double square (double z) @{ return z * z; @} | |
376 | ||
377 | return square (a) + square (b); | |
378 | @} | |
379 | @end group | |
380 | @end example | |
381 | ||
382 | The nested function can access all the variables of the containing | |
383 | function that are visible at the point of its definition. This is | |
384 | called @dfn{lexical scoping}. For example, here we show a nested | |
385 | function which uses an inherited variable named @code{offset}: | |
386 | ||
387 | @example | |
388 | bar (int *array, int offset, int size) | |
389 | @{ | |
390 | int access (int *array, int index) | |
391 | @{ return array[index + offset]; @} | |
392 | int i; | |
393 | @dots{} | |
394 | for (i = 0; i < size; i++) | |
395 | @dots{} access (array, i) @dots{} | |
396 | @} | |
397 | @end example | |
398 | ||
399 | Nested function definitions are permitted within functions in the places | |
400 | where variable definitions are allowed; that is, in any block, before | |
401 | the first statement in the block. | |
402 | ||
403 | It is possible to call the nested function from outside the scope of its | |
404 | name by storing its address or passing the address to another function: | |
405 | ||
406 | @example | |
407 | hack (int *array, int size) | |
408 | @{ | |
409 | void store (int index, int value) | |
410 | @{ array[index] = value; @} | |
411 | ||
412 | intermediate (store, size); | |
413 | @} | |
414 | @end example | |
415 | ||
416 | Here, the function @code{intermediate} receives the address of | |
417 | @code{store} as an argument. If @code{intermediate} calls @code{store}, | |
418 | the arguments given to @code{store} are used to store into @code{array}. | |
419 | But this technique works only so long as the containing function | |
420 | (@code{hack}, in this example) does not exit. | |
421 | ||
422 | If you try to call the nested function through its address after the | |
423 | containing function has exited, all hell will break loose. If you try | |
424 | to call it after a containing scope level has exited, and if it refers | |
425 | to some of the variables that are no longer in scope, you may be lucky, | |
426 | but it's not wise to take the risk. If, however, the nested function | |
427 | does not refer to anything that has gone out of scope, you should be | |
428 | safe. | |
429 | ||
430 | GNU CC implements taking the address of a nested function using a | |
674032e2 | 431 | technique called @dfn{trampolines}. A paper describing them is |
9734e80c | 432 | available as @uref{http://people.debian.org/~karlheg/Usenix88-lexic.pdf}. |
c1f7febf RK |
433 | |
434 | A nested function can jump to a label inherited from a containing | |
435 | function, provided the label was explicitly declared in the containing | |
436 | function (@pxref{Local Labels}). Such a jump returns instantly to the | |
437 | containing function, exiting the nested function which did the | |
438 | @code{goto} and any intermediate functions as well. Here is an example: | |
439 | ||
440 | @example | |
441 | @group | |
442 | bar (int *array, int offset, int size) | |
443 | @{ | |
444 | __label__ failure; | |
445 | int access (int *array, int index) | |
446 | @{ | |
447 | if (index > size) | |
448 | goto failure; | |
449 | return array[index + offset]; | |
450 | @} | |
451 | int i; | |
452 | @dots{} | |
453 | for (i = 0; i < size; i++) | |
454 | @dots{} access (array, i) @dots{} | |
455 | @dots{} | |
456 | return 0; | |
457 | ||
458 | /* @r{Control comes here from @code{access} | |
459 | if it detects an error.} */ | |
460 | failure: | |
461 | return -1; | |
462 | @} | |
463 | @end group | |
464 | @end example | |
465 | ||
466 | A nested function always has internal linkage. Declaring one with | |
467 | @code{extern} is erroneous. If you need to declare the nested function | |
468 | before its definition, use @code{auto} (which is otherwise meaningless | |
469 | for function declarations). | |
470 | ||
471 | @example | |
472 | bar (int *array, int offset, int size) | |
473 | @{ | |
474 | __label__ failure; | |
475 | auto int access (int *, int); | |
476 | @dots{} | |
477 | int access (int *array, int index) | |
478 | @{ | |
479 | if (index > size) | |
480 | goto failure; | |
481 | return array[index + offset]; | |
482 | @} | |
483 | @dots{} | |
484 | @} | |
485 | @end example | |
486 | ||
487 | @node Constructing Calls | |
488 | @section Constructing Function Calls | |
489 | @cindex constructing calls | |
490 | @cindex forwarding calls | |
491 | ||
492 | Using the built-in functions described below, you can record | |
493 | the arguments a function received, and call another function | |
494 | with the same arguments, without knowing the number or types | |
495 | of the arguments. | |
496 | ||
497 | You can also record the return value of that function call, | |
498 | and later return that value, without knowing what data type | |
499 | the function tried to return (as long as your caller expects | |
500 | that data type). | |
501 | ||
502 | @table @code | |
503 | @findex __builtin_apply_args | |
504 | @item __builtin_apply_args () | |
505 | This built-in function returns a pointer of type @code{void *} to data | |
506 | describing how to perform a call with the same arguments as were passed | |
507 | to the current function. | |
508 | ||
509 | The function saves the arg pointer register, structure value address, | |
510 | and all registers that might be used to pass arguments to a function | |
511 | into a block of memory allocated on the stack. Then it returns the | |
512 | address of that block. | |
513 | ||
514 | @findex __builtin_apply | |
515 | @item __builtin_apply (@var{function}, @var{arguments}, @var{size}) | |
516 | This built-in function invokes @var{function} (type @code{void (*)()}) | |
517 | with a copy of the parameters described by @var{arguments} (type | |
518 | @code{void *}) and @var{size} (type @code{int}). | |
519 | ||
520 | The value of @var{arguments} should be the value returned by | |
521 | @code{__builtin_apply_args}. The argument @var{size} specifies the size | |
522 | of the stack argument data, in bytes. | |
523 | ||
524 | This function returns a pointer of type @code{void *} to data describing | |
525 | how to return whatever value was returned by @var{function}. The data | |
526 | is saved in a block of memory allocated on the stack. | |
527 | ||
528 | It is not always simple to compute the proper value for @var{size}. The | |
529 | value is used by @code{__builtin_apply} to compute the amount of data | |
530 | that should be pushed on the stack and copied from the incoming argument | |
531 | area. | |
532 | ||
533 | @findex __builtin_return | |
534 | @item __builtin_return (@var{result}) | |
535 | This built-in function returns the value described by @var{result} from | |
536 | the containing function. You should specify, for @var{result}, a value | |
537 | returned by @code{__builtin_apply}. | |
538 | @end table | |
539 | ||
540 | @node Naming Types | |
541 | @section Naming an Expression's Type | |
542 | @cindex naming types | |
543 | ||
544 | You can give a name to the type of an expression using a @code{typedef} | |
545 | declaration with an initializer. Here is how to define @var{name} as a | |
546 | type name for the type of @var{exp}: | |
547 | ||
548 | @example | |
549 | typedef @var{name} = @var{exp}; | |
550 | @end example | |
551 | ||
552 | This is useful in conjunction with the statements-within-expressions | |
553 | feature. Here is how the two together can be used to define a safe | |
554 | ``maximum'' macro that operates on any arithmetic type: | |
555 | ||
556 | @example | |
557 | #define max(a,b) \ | |
558 | (@{typedef _ta = (a), _tb = (b); \ | |
559 | _ta _a = (a); _tb _b = (b); \ | |
560 | _a > _b ? _a : _b; @}) | |
561 | @end example | |
562 | ||
563 | @cindex underscores in variables in macros | |
564 | @cindex @samp{_} in variables in macros | |
565 | @cindex local variables in macros | |
566 | @cindex variables, local, in macros | |
567 | @cindex macros, local variables in | |
568 | ||
569 | The reason for using names that start with underscores for the local | |
570 | variables is to avoid conflicts with variable names that occur within the | |
571 | expressions that are substituted for @code{a} and @code{b}. Eventually we | |
572 | hope to design a new form of declaration syntax that allows you to declare | |
573 | variables whose scopes start only after their initializers; this will be a | |
574 | more reliable way to prevent such conflicts. | |
575 | ||
576 | @node Typeof | |
577 | @section Referring to a Type with @code{typeof} | |
578 | @findex typeof | |
579 | @findex sizeof | |
580 | @cindex macros, types of arguments | |
581 | ||
582 | Another way to refer to the type of an expression is with @code{typeof}. | |
583 | The syntax of using of this keyword looks like @code{sizeof}, but the | |
584 | construct acts semantically like a type name defined with @code{typedef}. | |
585 | ||
586 | There are two ways of writing the argument to @code{typeof}: with an | |
587 | expression or with a type. Here is an example with an expression: | |
588 | ||
589 | @example | |
590 | typeof (x[0](1)) | |
591 | @end example | |
592 | ||
593 | @noindent | |
89aed483 JM |
594 | This assumes that @code{x} is an array of pointers to functions; |
595 | the type described is that of the values of the functions. | |
c1f7febf RK |
596 | |
597 | Here is an example with a typename as the argument: | |
598 | ||
599 | @example | |
600 | typeof (int *) | |
601 | @end example | |
602 | ||
603 | @noindent | |
604 | Here the type described is that of pointers to @code{int}. | |
605 | ||
5490d604 | 606 | If you are writing a header file that must work when included in ISO C |
c1f7febf RK |
607 | programs, write @code{__typeof__} instead of @code{typeof}. |
608 | @xref{Alternate Keywords}. | |
609 | ||
610 | A @code{typeof}-construct can be used anywhere a typedef name could be | |
611 | used. For example, you can use it in a declaration, in a cast, or inside | |
612 | of @code{sizeof} or @code{typeof}. | |
613 | ||
614 | @itemize @bullet | |
615 | @item | |
616 | This declares @code{y} with the type of what @code{x} points to. | |
617 | ||
618 | @example | |
619 | typeof (*x) y; | |
620 | @end example | |
621 | ||
622 | @item | |
623 | This declares @code{y} as an array of such values. | |
624 | ||
625 | @example | |
626 | typeof (*x) y[4]; | |
627 | @end example | |
628 | ||
629 | @item | |
630 | This declares @code{y} as an array of pointers to characters: | |
631 | ||
632 | @example | |
633 | typeof (typeof (char *)[4]) y; | |
634 | @end example | |
635 | ||
636 | @noindent | |
637 | It is equivalent to the following traditional C declaration: | |
638 | ||
639 | @example | |
640 | char *y[4]; | |
641 | @end example | |
642 | ||
643 | To see the meaning of the declaration using @code{typeof}, and why it | |
644 | might be a useful way to write, let's rewrite it with these macros: | |
645 | ||
646 | @example | |
647 | #define pointer(T) typeof(T *) | |
648 | #define array(T, N) typeof(T [N]) | |
649 | @end example | |
650 | ||
651 | @noindent | |
652 | Now the declaration can be rewritten this way: | |
653 | ||
654 | @example | |
655 | array (pointer (char), 4) y; | |
656 | @end example | |
657 | ||
658 | @noindent | |
659 | Thus, @code{array (pointer (char), 4)} is the type of arrays of 4 | |
660 | pointers to @code{char}. | |
661 | @end itemize | |
662 | ||
663 | @node Lvalues | |
664 | @section Generalized Lvalues | |
665 | @cindex compound expressions as lvalues | |
666 | @cindex expressions, compound, as lvalues | |
667 | @cindex conditional expressions as lvalues | |
668 | @cindex expressions, conditional, as lvalues | |
669 | @cindex casts as lvalues | |
670 | @cindex generalized lvalues | |
671 | @cindex lvalues, generalized | |
672 | @cindex extensions, @code{?:} | |
673 | @cindex @code{?:} extensions | |
674 | Compound expressions, conditional expressions and casts are allowed as | |
675 | lvalues provided their operands are lvalues. This means that you can take | |
676 | their addresses or store values into them. | |
677 | ||
678 | Standard C++ allows compound expressions and conditional expressions as | |
679 | lvalues, and permits casts to reference type, so use of this extension | |
680 | is deprecated for C++ code. | |
681 | ||
682 | For example, a compound expression can be assigned, provided the last | |
683 | expression in the sequence is an lvalue. These two expressions are | |
684 | equivalent: | |
685 | ||
686 | @example | |
687 | (a, b) += 5 | |
688 | a, (b += 5) | |
689 | @end example | |
690 | ||
691 | Similarly, the address of the compound expression can be taken. These two | |
692 | expressions are equivalent: | |
693 | ||
694 | @example | |
695 | &(a, b) | |
696 | a, &b | |
697 | @end example | |
698 | ||
699 | A conditional expression is a valid lvalue if its type is not void and the | |
700 | true and false branches are both valid lvalues. For example, these two | |
701 | expressions are equivalent: | |
702 | ||
703 | @example | |
704 | (a ? b : c) = 5 | |
705 | (a ? b = 5 : (c = 5)) | |
706 | @end example | |
707 | ||
708 | A cast is a valid lvalue if its operand is an lvalue. A simple | |
709 | assignment whose left-hand side is a cast works by converting the | |
710 | right-hand side first to the specified type, then to the type of the | |
711 | inner left-hand side expression. After this is stored, the value is | |
712 | converted back to the specified type to become the value of the | |
713 | assignment. Thus, if @code{a} has type @code{char *}, the following two | |
714 | expressions are equivalent: | |
715 | ||
716 | @example | |
717 | (int)a = 5 | |
718 | (int)(a = (char *)(int)5) | |
719 | @end example | |
720 | ||
721 | An assignment-with-arithmetic operation such as @samp{+=} applied to a cast | |
722 | performs the arithmetic using the type resulting from the cast, and then | |
723 | continues as in the previous case. Therefore, these two expressions are | |
724 | equivalent: | |
725 | ||
726 | @example | |
727 | (int)a += 5 | |
728 | (int)(a = (char *)(int) ((int)a + 5)) | |
729 | @end example | |
730 | ||
731 | You cannot take the address of an lvalue cast, because the use of its | |
732 | address would not work out coherently. Suppose that @code{&(int)f} were | |
733 | permitted, where @code{f} has type @code{float}. Then the following | |
734 | statement would try to store an integer bit-pattern where a floating | |
735 | point number belongs: | |
736 | ||
737 | @example | |
738 | *&(int)f = 1; | |
739 | @end example | |
740 | ||
741 | This is quite different from what @code{(int)f = 1} would do---that | |
742 | would convert 1 to floating point and store it. Rather than cause this | |
743 | inconsistency, we think it is better to prohibit use of @samp{&} on a cast. | |
744 | ||
745 | If you really do want an @code{int *} pointer with the address of | |
746 | @code{f}, you can simply write @code{(int *)&f}. | |
747 | ||
748 | @node Conditionals | |
749 | @section Conditionals with Omitted Operands | |
750 | @cindex conditional expressions, extensions | |
751 | @cindex omitted middle-operands | |
752 | @cindex middle-operands, omitted | |
753 | @cindex extensions, @code{?:} | |
754 | @cindex @code{?:} extensions | |
755 | ||
756 | The middle operand in a conditional expression may be omitted. Then | |
757 | if the first operand is nonzero, its value is the value of the conditional | |
758 | expression. | |
759 | ||
760 | Therefore, the expression | |
761 | ||
762 | @example | |
763 | x ? : y | |
764 | @end example | |
765 | ||
766 | @noindent | |
767 | has the value of @code{x} if that is nonzero; otherwise, the value of | |
768 | @code{y}. | |
769 | ||
770 | This example is perfectly equivalent to | |
771 | ||
772 | @example | |
773 | x ? x : y | |
774 | @end example | |
775 | ||
776 | @cindex side effect in ?: | |
777 | @cindex ?: side effect | |
778 | @noindent | |
779 | In this simple case, the ability to omit the middle operand is not | |
780 | especially useful. When it becomes useful is when the first operand does, | |
781 | or may (if it is a macro argument), contain a side effect. Then repeating | |
782 | the operand in the middle would perform the side effect twice. Omitting | |
783 | the middle operand uses the value already computed without the undesirable | |
784 | effects of recomputing it. | |
785 | ||
786 | @node Long Long | |
787 | @section Double-Word Integers | |
788 | @cindex @code{long long} data types | |
789 | @cindex double-word arithmetic | |
790 | @cindex multiprecision arithmetic | |
791 | ||
792 | GNU C supports data types for integers that are twice as long as | |
793 | @code{int}. Simply write @code{long long int} for a signed integer, or | |
794 | @code{unsigned long long int} for an unsigned integer. To make an | |
795 | integer constant of type @code{long long int}, add the suffix @code{LL} | |
796 | to the integer. To make an integer constant of type @code{unsigned long | |
797 | long int}, add the suffix @code{ULL} to the integer. | |
798 | ||
799 | You can use these types in arithmetic like any other integer types. | |
800 | Addition, subtraction, and bitwise boolean operations on these types | |
801 | are open-coded on all types of machines. Multiplication is open-coded | |
802 | if the machine supports fullword-to-doubleword a widening multiply | |
803 | instruction. Division and shifts are open-coded only on machines that | |
804 | provide special support. The operations that are not open-coded use | |
805 | special library routines that come with GNU CC. | |
806 | ||
807 | There may be pitfalls when you use @code{long long} types for function | |
808 | arguments, unless you declare function prototypes. If a function | |
809 | expects type @code{int} for its argument, and you pass a value of type | |
810 | @code{long long int}, confusion will result because the caller and the | |
811 | subroutine will disagree about the number of bytes for the argument. | |
812 | Likewise, if the function expects @code{long long int} and you pass | |
813 | @code{int}. The best way to avoid such problems is to use prototypes. | |
814 | ||
815 | @node Complex | |
816 | @section Complex Numbers | |
817 | @cindex complex numbers | |
818 | ||
819 | GNU C supports complex data types. You can declare both complex integer | |
820 | types and complex floating types, using the keyword @code{__complex__}. | |
821 | ||
822 | For example, @samp{__complex__ double x;} declares @code{x} as a | |
823 | variable whose real part and imaginary part are both of type | |
824 | @code{double}. @samp{__complex__ short int y;} declares @code{y} to | |
825 | have real and imaginary parts of type @code{short int}; this is not | |
826 | likely to be useful, but it shows that the set of complex types is | |
827 | complete. | |
828 | ||
829 | To write a constant with a complex data type, use the suffix @samp{i} or | |
830 | @samp{j} (either one; they are equivalent). For example, @code{2.5fi} | |
831 | has type @code{__complex__ float} and @code{3i} has type | |
832 | @code{__complex__ int}. Such a constant always has a pure imaginary | |
833 | value, but you can form any complex value you like by adding one to a | |
834 | real constant. | |
835 | ||
836 | To extract the real part of a complex-valued expression @var{exp}, write | |
837 | @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to | |
838 | extract the imaginary part. | |
839 | ||
840 | The operator @samp{~} performs complex conjugation when used on a value | |
841 | with a complex type. | |
842 | ||
843 | GNU CC can allocate complex automatic variables in a noncontiguous | |
844 | fashion; it's even possible for the real part to be in a register while | |
845 | the imaginary part is on the stack (or vice-versa). None of the | |
846 | supported debugging info formats has a way to represent noncontiguous | |
847 | allocation like this, so GNU CC describes a noncontiguous complex | |
848 | variable as if it were two separate variables of noncomplex type. | |
849 | If the variable's actual name is @code{foo}, the two fictitious | |
850 | variables are named @code{foo$real} and @code{foo$imag}. You can | |
851 | examine and set these two fictitious variables with your debugger. | |
852 | ||
853 | A future version of GDB will know how to recognize such pairs and treat | |
854 | them as a single variable with a complex type. | |
855 | ||
6f4d7222 | 856 | @node Hex Floats |
6b42b9ea UD |
857 | @section Hex Floats |
858 | @cindex hex floats | |
c5c76735 | 859 | |
2a59078d | 860 | GNU CC recognizes floating-point numbers written not only in the usual |
6f4d7222 UD |
861 | decimal notation, such as @code{1.55e1}, but also numbers such as |
862 | @code{0x1.fp3} written in hexadecimal format. In that format the | |
863 | @code{0x} hex introducer and the @code{p} or @code{P} exponent field are | |
864 | mandatory. The exponent is a decimal number that indicates the power of | |
3b7a2e58 | 865 | 2 by which the significant part will be multiplied. Thus @code{0x1.f} is |
6f4d7222 UD |
866 | 1 15/16, @code{p3} multiplies it by 8, and the value of @code{0x1.fp3} |
867 | is the same as @code{1.55e1}. | |
868 | ||
869 | Unlike for floating-point numbers in the decimal notation the exponent | |
870 | is always required in the hexadecimal notation. Otherwise the compiler | |
871 | would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This | |
872 | could mean @code{1.0f} or @code{1.9375} since @code{f} is also the | |
873 | extension for floating-point constants of type @code{float}. | |
874 | ||
c1f7febf RK |
875 | @node Zero Length |
876 | @section Arrays of Length Zero | |
877 | @cindex arrays of length zero | |
878 | @cindex zero-length arrays | |
879 | @cindex length-zero arrays | |
ffc5c6a9 | 880 | @cindex flexible array members |
c1f7febf | 881 | |
584ef5fe RH |
882 | Zero-length arrays are allowed in GNU C. They are very useful as the |
883 | last element of a structure which is really a header for a variable-length | |
c1f7febf RK |
884 | object: |
885 | ||
886 | @example | |
887 | struct line @{ | |
888 | int length; | |
889 | char contents[0]; | |
890 | @}; | |
891 | ||
584ef5fe RH |
892 | struct line *thisline = (struct line *) |
893 | malloc (sizeof (struct line) + this_length); | |
894 | thisline->length = this_length; | |
c1f7febf RK |
895 | @end example |
896 | ||
a25f1211 | 897 | In ISO C89, you would have to give @code{contents} a length of 1, which |
c1f7febf RK |
898 | means either you waste space or complicate the argument to @code{malloc}. |
899 | ||
584ef5fe RH |
900 | In ISO C99, you would use a @dfn{flexible array member}, which is |
901 | slightly different in syntax and semantics: | |
902 | ||
903 | @itemize @bullet | |
904 | @item | |
905 | Flexible array members are written as @code{contents[]} without | |
906 | the @code{0}. | |
907 | ||
908 | @item | |
909 | Flexible array members have incomplete type, and so the @code{sizeof} | |
910 | operator may not be applied. As a quirk of the original implementation | |
911 | of zero-length arrays, @code{sizeof} evaluates to zero. | |
912 | ||
913 | @item | |
914 | Flexible array members may only appear as the last member of a | |
915 | @code{struct} that is otherwise non-empty. GCC currently allows | |
916 | zero-length arrays anywhere. You may encounter problems, however, | |
917 | defining structures containing only a zero-length array. Such usage | |
918 | is deprecated, and we recommend using zero-length arrays only in | |
919 | places in which flexible array members would be allowed. | |
ffc5c6a9 | 920 | @end itemize |
a25f1211 | 921 | |
ffc5c6a9 RH |
922 | GCC versions before 3.0 allowed zero-length arrays to be statically |
923 | initialized. In addition to those cases that were useful, it also | |
924 | allowed initializations in situations that would corrupt later data. | |
925 | Non-empty initialization of zero-length arrays is now deprecated. | |
926 | ||
927 | Instead GCC allows static initialization of flexible array members. | |
928 | This is equivalent to defining a new structure containing the original | |
929 | structure followed by an array of sufficient size to contain the data. | |
930 | I.e. in the following, @code{f1} is constructed as if it were declared | |
931 | like @code{f2}. | |
a25f1211 RH |
932 | |
933 | @example | |
ffc5c6a9 RH |
934 | struct f1 @{ |
935 | int x; int y[]; | |
936 | @} f1 = @{ 1, @{ 2, 3, 4 @} @}; | |
937 | ||
938 | struct f2 @{ | |
939 | struct f1 f1; int data[3]; | |
940 | @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @}; | |
941 | @end example | |
584ef5fe | 942 | |
ffc5c6a9 RH |
943 | @noindent |
944 | The convenience of this extension is that @code{f1} has the desired | |
945 | type, eliminating the need to consistently refer to @code{f2.f1}. | |
946 | ||
947 | This has symmetry with normal static arrays, in that an array of | |
948 | unknown size is also written with @code{[]}. | |
a25f1211 | 949 | |
ffc5c6a9 RH |
950 | Of course, this extension only makes sense if the extra data comes at |
951 | the end of a top-level object, as otherwise we would be overwriting | |
952 | data at subsequent offsets. To avoid undue complication and confusion | |
953 | with initialization of deeply nested arrays, we simply disallow any | |
954 | non-empty initialization except when the structure is the top-level | |
955 | object. For example: | |
584ef5fe | 956 | |
ffc5c6a9 RH |
957 | @example |
958 | struct foo @{ int x; int y[]; @}; | |
959 | struct bar @{ struct foo z; @}; | |
960 | ||
961 | struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // Legal. | |
962 | struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // Illegal. | |
963 | struct bar c = @{ @{ 1, @{ @} @} @}; // Legal. | |
964 | struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // Illegal. | |
a25f1211 | 965 | @end example |
4b606faf | 966 | |
c1f7febf RK |
967 | @node Variable Length |
968 | @section Arrays of Variable Length | |
969 | @cindex variable-length arrays | |
970 | @cindex arrays of variable length | |
971 | ||
972 | Variable-length automatic arrays are allowed in GNU C. These arrays are | |
973 | declared like any other automatic arrays, but with a length that is not | |
974 | a constant expression. The storage is allocated at the point of | |
975 | declaration and deallocated when the brace-level is exited. For | |
976 | example: | |
977 | ||
978 | @example | |
979 | FILE * | |
980 | concat_fopen (char *s1, char *s2, char *mode) | |
981 | @{ | |
982 | char str[strlen (s1) + strlen (s2) + 1]; | |
983 | strcpy (str, s1); | |
984 | strcat (str, s2); | |
985 | return fopen (str, mode); | |
986 | @} | |
987 | @end example | |
988 | ||
989 | @cindex scope of a variable length array | |
990 | @cindex variable-length array scope | |
991 | @cindex deallocating variable length arrays | |
992 | Jumping or breaking out of the scope of the array name deallocates the | |
993 | storage. Jumping into the scope is not allowed; you get an error | |
994 | message for it. | |
995 | ||
996 | @cindex @code{alloca} vs variable-length arrays | |
997 | You can use the function @code{alloca} to get an effect much like | |
998 | variable-length arrays. The function @code{alloca} is available in | |
999 | many other C implementations (but not in all). On the other hand, | |
1000 | variable-length arrays are more elegant. | |
1001 | ||
1002 | There are other differences between these two methods. Space allocated | |
1003 | with @code{alloca} exists until the containing @emph{function} returns. | |
1004 | The space for a variable-length array is deallocated as soon as the array | |
1005 | name's scope ends. (If you use both variable-length arrays and | |
1006 | @code{alloca} in the same function, deallocation of a variable-length array | |
1007 | will also deallocate anything more recently allocated with @code{alloca}.) | |
1008 | ||
1009 | You can also use variable-length arrays as arguments to functions: | |
1010 | ||
1011 | @example | |
1012 | struct entry | |
1013 | tester (int len, char data[len][len]) | |
1014 | @{ | |
1015 | @dots{} | |
1016 | @} | |
1017 | @end example | |
1018 | ||
1019 | The length of an array is computed once when the storage is allocated | |
1020 | and is remembered for the scope of the array in case you access it with | |
1021 | @code{sizeof}. | |
1022 | ||
1023 | If you want to pass the array first and the length afterward, you can | |
1024 | use a forward declaration in the parameter list---another GNU extension. | |
1025 | ||
1026 | @example | |
1027 | struct entry | |
1028 | tester (int len; char data[len][len], int len) | |
1029 | @{ | |
1030 | @dots{} | |
1031 | @} | |
1032 | @end example | |
1033 | ||
1034 | @cindex parameter forward declaration | |
1035 | The @samp{int len} before the semicolon is a @dfn{parameter forward | |
1036 | declaration}, and it serves the purpose of making the name @code{len} | |
1037 | known when the declaration of @code{data} is parsed. | |
1038 | ||
1039 | You can write any number of such parameter forward declarations in the | |
1040 | parameter list. They can be separated by commas or semicolons, but the | |
1041 | last one must end with a semicolon, which is followed by the ``real'' | |
1042 | parameter declarations. Each forward declaration must match a ``real'' | |
1043 | declaration in parameter name and data type. | |
1044 | ||
ccd96f0a NB |
1045 | @node Variadic Macros |
1046 | @section Macros with a Variable Number of Arguments. | |
c1f7febf RK |
1047 | @cindex variable number of arguments |
1048 | @cindex macro with variable arguments | |
1049 | @cindex rest argument (in macro) | |
ccd96f0a | 1050 | @cindex variadic macros |
c1f7febf | 1051 | |
ccd96f0a NB |
1052 | In the ISO C standard of 1999, a macro can be declared to accept a |
1053 | variable number of arguments much as a function can. The syntax for | |
1054 | defining the macro is similar to that of a function. Here is an | |
1055 | example: | |
c1f7febf RK |
1056 | |
1057 | @example | |
ccd96f0a | 1058 | #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__) |
c1f7febf RK |
1059 | @end example |
1060 | ||
ccd96f0a NB |
1061 | Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of |
1062 | such a macro, it represents the zero or more tokens until the closing | |
1063 | parenthesis that ends the invocation, including any commas. This set of | |
1064 | tokens replaces the identifier @code{__VA_ARGS__} in the macro body | |
1065 | wherever it appears. See the CPP manual for more information. | |
1066 | ||
1067 | GCC has long supported variadic macros, and used a different syntax that | |
1068 | allowed you to give a name to the variable arguments just like any other | |
1069 | argument. Here is an example: | |
c1f7febf RK |
1070 | |
1071 | @example | |
ccd96f0a | 1072 | #define debug(format, args...) fprintf (stderr, format, args) |
c1f7febf RK |
1073 | @end example |
1074 | ||
ccd96f0a NB |
1075 | This is in all ways equivalent to the ISO C example above, but arguably |
1076 | more readable and descriptive. | |
c1f7febf | 1077 | |
ccd96f0a NB |
1078 | GNU CPP has two further variadic macro extensions, and permits them to |
1079 | be used with either of the above forms of macro definition. | |
1080 | ||
1081 | In standard C, you are not allowed to leave the variable argument out | |
1082 | entirely; but you are allowed to pass an empty argument. For example, | |
1083 | this invocation is invalid in ISO C, because there is no comma after | |
1084 | the string: | |
c1f7febf RK |
1085 | |
1086 | @example | |
ccd96f0a | 1087 | debug ("A message") |
c1f7febf RK |
1088 | @end example |
1089 | ||
ccd96f0a NB |
1090 | GNU CPP permits you to completely omit the variable arguments in this |
1091 | way. In the above examples, the compiler would complain, though since | |
1092 | the expansion of the macro still has the extra comma after the format | |
1093 | string. | |
1094 | ||
1095 | To help solve this problem, CPP behaves specially for variable arguments | |
1096 | used with the token paste operator, @samp{##}. If instead you write | |
c1f7febf RK |
1097 | |
1098 | @example | |
ccd96f0a | 1099 | #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__) |
c1f7febf RK |
1100 | @end example |
1101 | ||
ccd96f0a NB |
1102 | and if the variable arguments are omitted or empty, the @samp{##} |
1103 | operator causes the preprocessor to remove the comma before it. If you | |
1104 | do provide some variable arguments in your macro invocation, GNU CPP | |
1105 | does not complain about the paste operation and instead places the | |
1106 | variable arguments after the comma. Just like any other pasted macro | |
1107 | argument, these arguments are not macro expanded. | |
1108 | ||
1109 | @node Escaped Newlines | |
1110 | @section Slightly Looser Rules for Escaped Newlines | |
1111 | @cindex escaped newlines | |
1112 | @cindex newlines (escaped) | |
1113 | ||
1114 | Recently, the non-traditional preprocessor has relaxed its treatment of | |
1115 | escaped newlines. Previously, the newline had to immediately follow a | |
1116 | backslash. The current implementation allows whitespace in the form of | |
1117 | spaces, horizontal and vertical tabs, and form feeds between the | |
1118 | backslash and the subsequent newline. The preprocessor issues a | |
1119 | warning, but treats it as a valid escaped newline and combines the two | |
1120 | lines to form a single logical line. This works within comments and | |
1121 | tokens, including multi-line strings, as well as between tokens. | |
1122 | Comments are @emph{not} treated as whitespace for the purposes of this | |
1123 | relaxation, since they have not yet been replaced with spaces. | |
1124 | ||
1125 | @node Multi-line Strings | |
1126 | @section String Literals with Embedded Newlines | |
1127 | @cindex multi-line string literals | |
1128 | ||
1129 | As an extension, GNU CPP permits string literals to cross multiple lines | |
1130 | without escaping the embedded newlines. Each embedded newline is | |
1131 | replaced with a single @samp{\n} character in the resulting string | |
1132 | literal, regardless of what form the newline took originally. | |
1133 | ||
1134 | CPP currently allows such strings in directives as well (other than the | |
1135 | @samp{#include} family). This is deprecated and will eventually be | |
1136 | removed. | |
c1f7febf RK |
1137 | |
1138 | @node Subscripting | |
1139 | @section Non-Lvalue Arrays May Have Subscripts | |
1140 | @cindex subscripting | |
1141 | @cindex arrays, non-lvalue | |
1142 | ||
1143 | @cindex subscripting and function values | |
1144 | Subscripting is allowed on arrays that are not lvalues, even though the | |
1145 | unary @samp{&} operator is not. For example, this is valid in GNU C though | |
1146 | not valid in other C dialects: | |
1147 | ||
1148 | @example | |
1149 | @group | |
1150 | struct foo @{int a[4];@}; | |
1151 | ||
1152 | struct foo f(); | |
1153 | ||
1154 | bar (int index) | |
1155 | @{ | |
1156 | return f().a[index]; | |
1157 | @} | |
1158 | @end group | |
1159 | @end example | |
1160 | ||
1161 | @node Pointer Arith | |
1162 | @section Arithmetic on @code{void}- and Function-Pointers | |
1163 | @cindex void pointers, arithmetic | |
1164 | @cindex void, size of pointer to | |
1165 | @cindex function pointers, arithmetic | |
1166 | @cindex function, size of pointer to | |
1167 | ||
1168 | In GNU C, addition and subtraction operations are supported on pointers to | |
1169 | @code{void} and on pointers to functions. This is done by treating the | |
1170 | size of a @code{void} or of a function as 1. | |
1171 | ||
1172 | A consequence of this is that @code{sizeof} is also allowed on @code{void} | |
1173 | and on function types, and returns 1. | |
1174 | ||
1175 | The option @samp{-Wpointer-arith} requests a warning if these extensions | |
1176 | are used. | |
1177 | ||
1178 | @node Initializers | |
1179 | @section Non-Constant Initializers | |
1180 | @cindex initializers, non-constant | |
1181 | @cindex non-constant initializers | |
1182 | ||
1183 | As in standard C++, the elements of an aggregate initializer for an | |
1184 | automatic variable are not required to be constant expressions in GNU C. | |
1185 | Here is an example of an initializer with run-time varying elements: | |
1186 | ||
1187 | @example | |
1188 | foo (float f, float g) | |
1189 | @{ | |
1190 | float beat_freqs[2] = @{ f-g, f+g @}; | |
1191 | @dots{} | |
1192 | @} | |
1193 | @end example | |
1194 | ||
1195 | @node Constructors | |
1196 | @section Constructor Expressions | |
1197 | @cindex constructor expressions | |
1198 | @cindex initializations in expressions | |
1199 | @cindex structures, constructor expression | |
1200 | @cindex expressions, constructor | |
1201 | ||
1202 | GNU C supports constructor expressions. A constructor looks like | |
1203 | a cast containing an initializer. Its value is an object of the | |
1204 | type specified in the cast, containing the elements specified in | |
1205 | the initializer. | |
1206 | ||
1207 | Usually, the specified type is a structure. Assume that | |
1208 | @code{struct foo} and @code{structure} are declared as shown: | |
1209 | ||
1210 | @example | |
1211 | struct foo @{int a; char b[2];@} structure; | |
1212 | @end example | |
1213 | ||
1214 | @noindent | |
1215 | Here is an example of constructing a @code{struct foo} with a constructor: | |
1216 | ||
1217 | @example | |
1218 | structure = ((struct foo) @{x + y, 'a', 0@}); | |
1219 | @end example | |
1220 | ||
1221 | @noindent | |
1222 | This is equivalent to writing the following: | |
1223 | ||
1224 | @example | |
1225 | @{ | |
1226 | struct foo temp = @{x + y, 'a', 0@}; | |
1227 | structure = temp; | |
1228 | @} | |
1229 | @end example | |
1230 | ||
1231 | You can also construct an array. If all the elements of the constructor | |
1232 | are (made up of) simple constant expressions, suitable for use in | |
1233 | initializers, then the constructor is an lvalue and can be coerced to a | |
1234 | pointer to its first element, as shown here: | |
1235 | ||
1236 | @example | |
1237 | char **foo = (char *[]) @{ "x", "y", "z" @}; | |
1238 | @end example | |
1239 | ||
1240 | Array constructors whose elements are not simple constants are | |
1241 | not very useful, because the constructor is not an lvalue. There | |
1242 | are only two valid ways to use it: to subscript it, or initialize | |
1243 | an array variable with it. The former is probably slower than a | |
1244 | @code{switch} statement, while the latter does the same thing an | |
1245 | ordinary C initializer would do. Here is an example of | |
1246 | subscripting an array constructor: | |
1247 | ||
1248 | @example | |
1249 | output = ((int[]) @{ 2, x, 28 @}) [input]; | |
1250 | @end example | |
1251 | ||
1252 | Constructor expressions for scalar types and union types are is | |
1253 | also allowed, but then the constructor expression is equivalent | |
1254 | to a cast. | |
1255 | ||
1256 | @node Labeled Elements | |
1257 | @section Labeled Elements in Initializers | |
1258 | @cindex initializers with labeled elements | |
1259 | @cindex labeled elements in initializers | |
1260 | @cindex case labels in initializers | |
1261 | ||
26d4fec7 | 1262 | Standard C89 requires the elements of an initializer to appear in a fixed |
c1f7febf RK |
1263 | order, the same as the order of the elements in the array or structure |
1264 | being initialized. | |
1265 | ||
26d4fec7 JM |
1266 | In ISO C99 you can give the elements in any order, specifying the array |
1267 | indices or structure field names they apply to, and GNU C allows this as | |
1268 | an extension in C89 mode as well. This extension is not | |
c1f7febf RK |
1269 | implemented in GNU C++. |
1270 | ||
26d4fec7 | 1271 | To specify an array index, write |
c1f7febf RK |
1272 | @samp{[@var{index}] =} before the element value. For example, |
1273 | ||
1274 | @example | |
26d4fec7 | 1275 | int a[6] = @{ [4] = 29, [2] = 15 @}; |
c1f7febf RK |
1276 | @end example |
1277 | ||
1278 | @noindent | |
1279 | is equivalent to | |
1280 | ||
1281 | @example | |
1282 | int a[6] = @{ 0, 0, 15, 0, 29, 0 @}; | |
1283 | @end example | |
1284 | ||
1285 | @noindent | |
1286 | The index values must be constant expressions, even if the array being | |
1287 | initialized is automatic. | |
1288 | ||
26d4fec7 JM |
1289 | An alternative syntax for this which has been obsolete since GCC 2.5 but |
1290 | GCC still accepts is to write @samp{[@var{index}]} before the element | |
1291 | value, with no @samp{=}. | |
1292 | ||
c1f7febf | 1293 | To initialize a range of elements to the same value, write |
26d4fec7 JM |
1294 | @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU |
1295 | extension. For example, | |
c1f7febf RK |
1296 | |
1297 | @example | |
1298 | int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @}; | |
1299 | @end example | |
1300 | ||
8b6a5902 JJ |
1301 | @noindent |
1302 | If the value in it has side-effects, the side-effects will happen only once, | |
1303 | not for each initialized field by the range initializer. | |
1304 | ||
c1f7febf RK |
1305 | @noindent |
1306 | Note that the length of the array is the highest value specified | |
1307 | plus one. | |
1308 | ||
1309 | In a structure initializer, specify the name of a field to initialize | |
26d4fec7 | 1310 | with @samp{.@var{fieldname} =} before the element value. For example, |
c1f7febf RK |
1311 | given the following structure, |
1312 | ||
1313 | @example | |
1314 | struct point @{ int x, y; @}; | |
1315 | @end example | |
1316 | ||
1317 | @noindent | |
1318 | the following initialization | |
1319 | ||
1320 | @example | |
26d4fec7 | 1321 | struct point p = @{ .y = yvalue, .x = xvalue @}; |
c1f7febf RK |
1322 | @end example |
1323 | ||
1324 | @noindent | |
1325 | is equivalent to | |
1326 | ||
1327 | @example | |
1328 | struct point p = @{ xvalue, yvalue @}; | |
1329 | @end example | |
1330 | ||
26d4fec7 JM |
1331 | Another syntax which has the same meaning, obsolete since GCC 2.5, is |
1332 | @samp{@var{fieldname}:}, as shown here: | |
c1f7febf RK |
1333 | |
1334 | @example | |
26d4fec7 | 1335 | struct point p = @{ y: yvalue, x: xvalue @}; |
c1f7febf RK |
1336 | @end example |
1337 | ||
1338 | You can also use an element label (with either the colon syntax or the | |
1339 | period-equal syntax) when initializing a union, to specify which element | |
1340 | of the union should be used. For example, | |
1341 | ||
1342 | @example | |
1343 | union foo @{ int i; double d; @}; | |
1344 | ||
26d4fec7 | 1345 | union foo f = @{ .d = 4 @}; |
c1f7febf RK |
1346 | @end example |
1347 | ||
1348 | @noindent | |
1349 | will convert 4 to a @code{double} to store it in the union using | |
1350 | the second element. By contrast, casting 4 to type @code{union foo} | |
1351 | would store it into the union as the integer @code{i}, since it is | |
1352 | an integer. (@xref{Cast to Union}.) | |
1353 | ||
1354 | You can combine this technique of naming elements with ordinary C | |
1355 | initialization of successive elements. Each initializer element that | |
1356 | does not have a label applies to the next consecutive element of the | |
1357 | array or structure. For example, | |
1358 | ||
1359 | @example | |
1360 | int a[6] = @{ [1] = v1, v2, [4] = v4 @}; | |
1361 | @end example | |
1362 | ||
1363 | @noindent | |
1364 | is equivalent to | |
1365 | ||
1366 | @example | |
1367 | int a[6] = @{ 0, v1, v2, 0, v4, 0 @}; | |
1368 | @end example | |
1369 | ||
1370 | Labeling the elements of an array initializer is especially useful | |
1371 | when the indices are characters or belong to an @code{enum} type. | |
1372 | For example: | |
1373 | ||
1374 | @example | |
1375 | int whitespace[256] | |
1376 | = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1, | |
1377 | ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @}; | |
1378 | @end example | |
1379 | ||
26d4fec7 JM |
1380 | You can also write a series of @samp{.@var{fieldname}} and |
1381 | @samp{[@var{index}]} element labels before an @samp{=} to specify a | |
1382 | nested subobject to initialize; the list is taken relative to the | |
1383 | subobject corresponding to the closest surrounding brace pair. For | |
1384 | example, with the @samp{struct point} declaration above: | |
1385 | ||
1386 | @example | |
1387 | struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @}; | |
1388 | @end example | |
1389 | ||
8b6a5902 JJ |
1390 | @noindent |
1391 | If the same field is initialized multiple times, it will have value from | |
1392 | the last initialization. If any such overridden initialization has | |
1393 | side-effect, it is unspecified whether the side-effect happens or not. | |
1394 | Currently, gcc will discard them and issue a warning. | |
1395 | ||
c1f7febf RK |
1396 | @node Case Ranges |
1397 | @section Case Ranges | |
1398 | @cindex case ranges | |
1399 | @cindex ranges in case statements | |
1400 | ||
1401 | You can specify a range of consecutive values in a single @code{case} label, | |
1402 | like this: | |
1403 | ||
1404 | @example | |
1405 | case @var{low} ... @var{high}: | |
1406 | @end example | |
1407 | ||
1408 | @noindent | |
1409 | This has the same effect as the proper number of individual @code{case} | |
1410 | labels, one for each integer value from @var{low} to @var{high}, inclusive. | |
1411 | ||
1412 | This feature is especially useful for ranges of ASCII character codes: | |
1413 | ||
1414 | @example | |
1415 | case 'A' ... 'Z': | |
1416 | @end example | |
1417 | ||
1418 | @strong{Be careful:} Write spaces around the @code{...}, for otherwise | |
1419 | it may be parsed wrong when you use it with integer values. For example, | |
1420 | write this: | |
1421 | ||
1422 | @example | |
1423 | case 1 ... 5: | |
1424 | @end example | |
1425 | ||
1426 | @noindent | |
1427 | rather than this: | |
1428 | ||
1429 | @example | |
1430 | case 1...5: | |
1431 | @end example | |
1432 | ||
1433 | @node Cast to Union | |
1434 | @section Cast to a Union Type | |
1435 | @cindex cast to a union | |
1436 | @cindex union, casting to a | |
1437 | ||
1438 | A cast to union type is similar to other casts, except that the type | |
1439 | specified is a union type. You can specify the type either with | |
1440 | @code{union @var{tag}} or with a typedef name. A cast to union is actually | |
1441 | a constructor though, not a cast, and hence does not yield an lvalue like | |
1442 | normal casts. (@xref{Constructors}.) | |
1443 | ||
1444 | The types that may be cast to the union type are those of the members | |
1445 | of the union. Thus, given the following union and variables: | |
1446 | ||
1447 | @example | |
1448 | union foo @{ int i; double d; @}; | |
1449 | int x; | |
1450 | double y; | |
1451 | @end example | |
1452 | ||
1453 | @noindent | |
1454 | both @code{x} and @code{y} can be cast to type @code{union} foo. | |
1455 | ||
1456 | Using the cast as the right-hand side of an assignment to a variable of | |
1457 | union type is equivalent to storing in a member of the union: | |
1458 | ||
1459 | @example | |
1460 | union foo u; | |
1461 | @dots{} | |
1462 | u = (union foo) x @equiv{} u.i = x | |
1463 | u = (union foo) y @equiv{} u.d = y | |
1464 | @end example | |
1465 | ||
1466 | You can also use the union cast as a function argument: | |
1467 | ||
1468 | @example | |
1469 | void hack (union foo); | |
1470 | @dots{} | |
1471 | hack ((union foo) x); | |
1472 | @end example | |
1473 | ||
1474 | @node Function Attributes | |
1475 | @section Declaring Attributes of Functions | |
1476 | @cindex function attributes | |
1477 | @cindex declaring attributes of functions | |
1478 | @cindex functions that never return | |
1479 | @cindex functions that have no side effects | |
1480 | @cindex functions in arbitrary sections | |
2a59078d | 1481 | @cindex functions that behave like malloc |
c1f7febf RK |
1482 | @cindex @code{volatile} applied to function |
1483 | @cindex @code{const} applied to function | |
26f6672d | 1484 | @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments |
c1f7febf RK |
1485 | @cindex functions that are passed arguments in registers on the 386 |
1486 | @cindex functions that pop the argument stack on the 386 | |
1487 | @cindex functions that do not pop the argument stack on the 386 | |
1488 | ||
1489 | In GNU C, you declare certain things about functions called in your program | |
1490 | which help the compiler optimize function calls and check your code more | |
1491 | carefully. | |
1492 | ||
1493 | The keyword @code{__attribute__} allows you to specify special | |
1494 | attributes when making a declaration. This keyword is followed by an | |
9f1bbeaa JM |
1495 | attribute specification inside double parentheses. Fourteen attributes, |
1496 | @code{noreturn}, @code{pure}, @code{const}, @code{format}, | |
1497 | @code{format_arg}, @code{no_instrument_function}, @code{section}, | |
1498 | @code{constructor}, @code{destructor}, @code{unused}, @code{weak}, | |
1499 | @code{malloc}, @code{alias} and @code{no_check_memory_usage} are | |
1500 | currently defined for functions. Several other attributes are defined | |
1501 | for functions on particular target systems. Other attributes, including | |
c1f7febf RK |
1502 | @code{section} are supported for variables declarations (@pxref{Variable |
1503 | Attributes}) and for types (@pxref{Type Attributes}). | |
1504 | ||
1505 | You may also specify attributes with @samp{__} preceding and following | |
1506 | each keyword. This allows you to use them in header files without | |
1507 | being concerned about a possible macro of the same name. For example, | |
1508 | you may use @code{__noreturn__} instead of @code{noreturn}. | |
1509 | ||
2c5e91d2 JM |
1510 | @xref{Attribute Syntax}, for details of the exact syntax for using |
1511 | attributes. | |
1512 | ||
c1f7febf RK |
1513 | @table @code |
1514 | @cindex @code{noreturn} function attribute | |
1515 | @item noreturn | |
1516 | A few standard library functions, such as @code{abort} and @code{exit}, | |
1517 | cannot return. GNU CC knows this automatically. Some programs define | |
1518 | their own functions that never return. You can declare them | |
1519 | @code{noreturn} to tell the compiler this fact. For example, | |
1520 | ||
1521 | @smallexample | |
1522 | void fatal () __attribute__ ((noreturn)); | |
1523 | ||
1524 | void | |
1525 | fatal (@dots{}) | |
1526 | @{ | |
1527 | @dots{} /* @r{Print error message.} */ @dots{} | |
1528 | exit (1); | |
1529 | @} | |
1530 | @end smallexample | |
1531 | ||
1532 | The @code{noreturn} keyword tells the compiler to assume that | |
1533 | @code{fatal} cannot return. It can then optimize without regard to what | |
1534 | would happen if @code{fatal} ever did return. This makes slightly | |
1535 | better code. More importantly, it helps avoid spurious warnings of | |
1536 | uninitialized variables. | |
1537 | ||
1538 | Do not assume that registers saved by the calling function are | |
1539 | restored before calling the @code{noreturn} function. | |
1540 | ||
1541 | It does not make sense for a @code{noreturn} function to have a return | |
1542 | type other than @code{void}. | |
1543 | ||
1544 | The attribute @code{noreturn} is not implemented in GNU C versions | |
1545 | earlier than 2.5. An alternative way to declare that a function does | |
1546 | not return, which works in the current version and in some older | |
1547 | versions, is as follows: | |
1548 | ||
1549 | @smallexample | |
1550 | typedef void voidfn (); | |
1551 | ||
1552 | volatile voidfn fatal; | |
1553 | @end smallexample | |
1554 | ||
2a8f6b90 JH |
1555 | @cindex @code{pure} function attribute |
1556 | @item pure | |
1557 | Many functions have no effects except the return value and their | |
d4047e24 | 1558 | return value depends only on the parameters and/or global variables. |
2a8f6b90 | 1559 | Such a function can be subject |
c1f7febf RK |
1560 | to common subexpression elimination and loop optimization just as an |
1561 | arithmetic operator would be. These functions should be declared | |
2a8f6b90 | 1562 | with the attribute @code{pure}. For example, |
c1f7febf RK |
1563 | |
1564 | @smallexample | |
2a8f6b90 | 1565 | int square (int) __attribute__ ((pure)); |
c1f7febf RK |
1566 | @end smallexample |
1567 | ||
1568 | @noindent | |
1569 | says that the hypothetical function @code{square} is safe to call | |
1570 | fewer times than the program says. | |
1571 | ||
2a8f6b90 JH |
1572 | Some of common examples of pure functions are @code{strlen} or @code{memcmp}. |
1573 | Interesting non-pure functions are functions with infinite loops or those | |
1574 | depending on volatile memory or other system resource, that may change between | |
2a59078d | 1575 | two consecutive calls (such as @code{feof} in a multithreading environment). |
2a8f6b90 JH |
1576 | |
1577 | The attribute @code{pure} is not implemented in GNU C versions earlier | |
1578 | than 2.96. | |
1579 | @cindex @code{const} function attribute | |
1580 | @item const | |
1581 | Many functions do not examine any values except their arguments, and | |
1582 | have no effects except the return value. Basically this is just slightly | |
1583 | more strict class than the "pure" attribute above, since function is not | |
2a59078d | 1584 | allowed to read global memory. |
2a8f6b90 JH |
1585 | |
1586 | @cindex pointer arguments | |
1587 | Note that a function that has pointer arguments and examines the data | |
1588 | pointed to must @emph{not} be declared @code{const}. Likewise, a | |
1589 | function that calls a non-@code{const} function usually must not be | |
1590 | @code{const}. It does not make sense for a @code{const} function to | |
1591 | return @code{void}. | |
1592 | ||
c1f7febf RK |
1593 | The attribute @code{const} is not implemented in GNU C versions earlier |
1594 | than 2.5. An alternative way to declare that a function has no side | |
1595 | effects, which works in the current version and in some older versions, | |
1596 | is as follows: | |
1597 | ||
1598 | @smallexample | |
1599 | typedef int intfn (); | |
1600 | ||
1601 | extern const intfn square; | |
1602 | @end smallexample | |
1603 | ||
1604 | This approach does not work in GNU C++ from 2.6.0 on, since the language | |
1605 | specifies that the @samp{const} must be attached to the return value. | |
1606 | ||
c1f7febf RK |
1607 | |
1608 | @item format (@var{archetype}, @var{string-index}, @var{first-to-check}) | |
1609 | @cindex @code{format} function attribute | |
bb72a084 | 1610 | The @code{format} attribute specifies that a function takes @code{printf}, |
26f6672d JM |
1611 | @code{scanf}, @code{strftime} or @code{strfmon} style arguments which |
1612 | should be type-checked against a format string. For example, the | |
1613 | declaration: | |
c1f7febf RK |
1614 | |
1615 | @smallexample | |
1616 | extern int | |
1617 | my_printf (void *my_object, const char *my_format, ...) | |
1618 | __attribute__ ((format (printf, 2, 3))); | |
1619 | @end smallexample | |
1620 | ||
1621 | @noindent | |
1622 | causes the compiler to check the arguments in calls to @code{my_printf} | |
1623 | for consistency with the @code{printf} style format string argument | |
1624 | @code{my_format}. | |
1625 | ||
1626 | The parameter @var{archetype} determines how the format string is | |
26f6672d JM |
1627 | interpreted, and should be @code{printf}, @code{scanf}, @code{strftime} |
1628 | or @code{strfmon}. (You can also use @code{__printf__}, | |
1629 | @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The | |
c1f7febf RK |
1630 | parameter @var{string-index} specifies which argument is the format |
1631 | string argument (starting from 1), while @var{first-to-check} is the | |
1632 | number of the first argument to check against the format string. For | |
1633 | functions where the arguments are not available to be checked (such as | |
1634 | @code{vprintf}), specify the third parameter as zero. In this case the | |
b722c82c JM |
1635 | compiler only checks the format string for consistency. For |
1636 | @code{strftime} formats, the third parameter is required to be zero. | |
c1f7febf RK |
1637 | |
1638 | In the example above, the format string (@code{my_format}) is the second | |
1639 | argument of the function @code{my_print}, and the arguments to check | |
1640 | start with the third argument, so the correct parameters for the format | |
1641 | attribute are 2 and 3. | |
1642 | ||
1643 | The @code{format} attribute allows you to identify your own functions | |
1644 | which take format strings as arguments, so that GNU CC can check the | |
b722c82c JM |
1645 | calls to these functions for errors. The compiler always (unless |
1646 | @samp{-ffreestanding} is used) checks formats | |
1647 | for the standard library functions @code{printf}, @code{fprintf}, | |
bb72a084 | 1648 | @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime}, |
c1f7febf RK |
1649 | @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such |
1650 | warnings are requested (using @samp{-Wformat}), so there is no need to | |
b722c82c JM |
1651 | modify the header file @file{stdio.h}. In C99 mode, the functions |
1652 | @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and | |
26f6672d JM |
1653 | @code{vsscanf} are also checked. Except in strictly conforming C |
1654 | standard modes, the X/Open function @code{strfmon} is also checked. | |
b722c82c | 1655 | @xref{C Dialect Options,,Options Controlling C Dialect}. |
c1f7febf RK |
1656 | |
1657 | @item format_arg (@var{string-index}) | |
1658 | @cindex @code{format_arg} function attribute | |
26f6672d JM |
1659 | The @code{format_arg} attribute specifies that a function takes a format |
1660 | string for a @code{printf}, @code{scanf}, @code{strftime} or | |
1661 | @code{strfmon} style function and modifies it (for example, to translate | |
1662 | it into another language), so the result can be passed to a | |
1663 | @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style | |
1664 | function (with the remaining arguments to the format function the same | |
1665 | as they would have been for the unmodified string). For example, the | |
1666 | declaration: | |
c1f7febf RK |
1667 | |
1668 | @smallexample | |
1669 | extern char * | |
1670 | my_dgettext (char *my_domain, const char *my_format) | |
1671 | __attribute__ ((format_arg (2))); | |
1672 | @end smallexample | |
1673 | ||
1674 | @noindent | |
26f6672d JM |
1675 | causes the compiler to check the arguments in calls to a @code{printf}, |
1676 | @code{scanf}, @code{strftime} or @code{strfmon} type function, whose | |
1677 | format string argument is a call to the @code{my_dgettext} function, for | |
1678 | consistency with the format string argument @code{my_format}. If the | |
1679 | @code{format_arg} attribute had not been specified, all the compiler | |
1680 | could tell in such calls to format functions would be that the format | |
1681 | string argument is not constant; this would generate a warning when | |
1682 | @code{-Wformat-nonliteral} is used, but the calls could not be checked | |
1683 | without the attribute. | |
c1f7febf RK |
1684 | |
1685 | The parameter @var{string-index} specifies which argument is the format | |
1686 | string argument (starting from 1). | |
1687 | ||
1688 | The @code{format-arg} attribute allows you to identify your own | |
1689 | functions which modify format strings, so that GNU CC can check the | |
26f6672d JM |
1690 | calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} |
1691 | type function whose operands are a call to one of your own function. | |
1692 | The compiler always treats @code{gettext}, @code{dgettext}, and | |
1693 | @code{dcgettext} in this manner except when strict ISO C support is | |
1694 | requested by @samp{-ansi} or an appropriate @samp{-std} option, or | |
1695 | @samp{-ffreestanding} is used. @xref{C Dialect Options,,Options | |
1696 | Controlling C Dialect}. | |
c1f7febf | 1697 | |
07417085 KR |
1698 | @item no_instrument_function |
1699 | @cindex @code{no_instrument_function} function attribute | |
1700 | If @samp{-finstrument-functions} is given, profiling function calls will | |
1701 | be generated at entry and exit of most user-compiled functions. | |
1702 | Functions with this attribute will not be so instrumented. | |
1703 | ||
c1f7febf RK |
1704 | @item section ("section-name") |
1705 | @cindex @code{section} function attribute | |
1706 | Normally, the compiler places the code it generates in the @code{text} section. | |
1707 | Sometimes, however, you need additional sections, or you need certain | |
1708 | particular functions to appear in special sections. The @code{section} | |
1709 | attribute specifies that a function lives in a particular section. | |
1710 | For example, the declaration: | |
1711 | ||
1712 | @smallexample | |
1713 | extern void foobar (void) __attribute__ ((section ("bar"))); | |
1714 | @end smallexample | |
1715 | ||
1716 | @noindent | |
1717 | puts the function @code{foobar} in the @code{bar} section. | |
1718 | ||
1719 | Some file formats do not support arbitrary sections so the @code{section} | |
1720 | attribute is not available on all platforms. | |
1721 | If you need to map the entire contents of a module to a particular | |
1722 | section, consider using the facilities of the linker instead. | |
1723 | ||
1724 | @item constructor | |
1725 | @itemx destructor | |
1726 | @cindex @code{constructor} function attribute | |
1727 | @cindex @code{destructor} function attribute | |
1728 | The @code{constructor} attribute causes the function to be called | |
1729 | automatically before execution enters @code{main ()}. Similarly, the | |
1730 | @code{destructor} attribute causes the function to be called | |
1731 | automatically after @code{main ()} has completed or @code{exit ()} has | |
1732 | been called. Functions with these attributes are useful for | |
1733 | initializing data that will be used implicitly during the execution of | |
1734 | the program. | |
1735 | ||
1736 | These attributes are not currently implemented for Objective C. | |
1737 | ||
1738 | @item unused | |
1739 | This attribute, attached to a function, means that the function is meant | |
1740 | to be possibly unused. GNU CC will not produce a warning for this | |
1741 | function. GNU C++ does not currently support this attribute as | |
1742 | definitions without parameters are valid in C++. | |
1743 | ||
1744 | @item weak | |
1745 | @cindex @code{weak} attribute | |
1746 | The @code{weak} attribute causes the declaration to be emitted as a weak | |
1747 | symbol rather than a global. This is primarily useful in defining | |
1748 | library functions which can be overridden in user code, though it can | |
1749 | also be used with non-function declarations. Weak symbols are supported | |
1750 | for ELF targets, and also for a.out targets when using the GNU assembler | |
1751 | and linker. | |
1752 | ||
140592a0 AG |
1753 | @item malloc |
1754 | @cindex @code{malloc} attribute | |
1755 | The @code{malloc} attribute is used to tell the compiler that a function | |
1756 | may be treated as if it were the malloc function. The compiler assumes | |
1757 | that calls to malloc result in a pointers that cannot alias anything. | |
1758 | This will often improve optimization. | |
1759 | ||
c1f7febf RK |
1760 | @item alias ("target") |
1761 | @cindex @code{alias} attribute | |
1762 | The @code{alias} attribute causes the declaration to be emitted as an | |
1763 | alias for another symbol, which must be specified. For instance, | |
1764 | ||
1765 | @smallexample | |
1766 | void __f () @{ /* do something */; @} | |
1767 | void f () __attribute__ ((weak, alias ("__f"))); | |
1768 | @end smallexample | |
1769 | ||
1770 | declares @samp{f} to be a weak alias for @samp{__f}. In C++, the | |
1771 | mangled name for the target must be used. | |
1772 | ||
af3e86c2 RK |
1773 | Not all target machines support this attribute. |
1774 | ||
7d384cc0 KR |
1775 | @item no_check_memory_usage |
1776 | @cindex @code{no_check_memory_usage} function attribute | |
c5c76735 JL |
1777 | The @code{no_check_memory_usage} attribute causes GNU CC to omit checks |
1778 | of memory references when it generates code for that function. Normally | |
1779 | if you specify @samp{-fcheck-memory-usage} (see @pxref{Code Gen | |
1780 | Options}), GNU CC generates calls to support routines before most memory | |
1781 | accesses to permit support code to record usage and detect uses of | |
1782 | uninitialized or unallocated storage. Since GNU CC cannot handle | |
1783 | @code{asm} statements properly they are not allowed in such functions. | |
1784 | If you declare a function with this attribute, GNU CC will not generate | |
7d384cc0 | 1785 | memory checking code for that function, permitting the use of @code{asm} |
c5c76735 JL |
1786 | statements without having to compile that function with different |
1787 | options. This also allows you to write support routines of your own if | |
1788 | you wish, without getting infinite recursion if they get compiled with | |
1789 | @code{-fcheck-memory-usage}. | |
7d384cc0 | 1790 | |
c1f7febf RK |
1791 | @item regparm (@var{number}) |
1792 | @cindex functions that are passed arguments in registers on the 386 | |
1793 | On the Intel 386, the @code{regparm} attribute causes the compiler to | |
1794 | pass up to @var{number} integer arguments in registers @var{EAX}, | |
1795 | @var{EDX}, and @var{ECX} instead of on the stack. Functions that take a | |
1796 | variable number of arguments will continue to be passed all of their | |
1797 | arguments on the stack. | |
1798 | ||
1799 | @item stdcall | |
1800 | @cindex functions that pop the argument stack on the 386 | |
1801 | On the Intel 386, the @code{stdcall} attribute causes the compiler to | |
1802 | assume that the called function will pop off the stack space used to | |
1803 | pass arguments, unless it takes a variable number of arguments. | |
1804 | ||
1805 | The PowerPC compiler for Windows NT currently ignores the @code{stdcall} | |
1806 | attribute. | |
1807 | ||
1808 | @item cdecl | |
1809 | @cindex functions that do pop the argument stack on the 386 | |
1810 | On the Intel 386, the @code{cdecl} attribute causes the compiler to | |
1811 | assume that the calling function will pop off the stack space used to | |
1812 | pass arguments. This is | |
1813 | useful to override the effects of the @samp{-mrtd} switch. | |
1814 | ||
1815 | The PowerPC compiler for Windows NT currently ignores the @code{cdecl} | |
1816 | attribute. | |
1817 | ||
1818 | @item longcall | |
1819 | @cindex functions called via pointer on the RS/6000 and PowerPC | |
1820 | On the RS/6000 and PowerPC, the @code{longcall} attribute causes the | |
1821 | compiler to always call the function via a pointer, so that functions | |
1822 | which reside further than 64 megabytes (67,108,864 bytes) from the | |
1823 | current location can be called. | |
1824 | ||
c27ba912 DM |
1825 | @item long_call/short_call |
1826 | @cindex indirect calls on ARM | |
1827 | This attribute allows to specify how to call a particular function on | |
1828 | ARM. Both attributes override the @code{-mlong-calls} (@pxref{ARM Options}) | |
1829 | command line switch and @code{#pragma long_calls} settings. The | |
1830 | @code{long_call} attribute causes the compiler to always call the | |
1831 | function by first loading its address into a register and then using the | |
1832 | contents of that register. The @code{short_call} attribute always places | |
1833 | the offset to the function from the call site into the @samp{BL} | |
1834 | instruction directly. | |
1835 | ||
c1f7febf RK |
1836 | @item dllimport |
1837 | @cindex functions which are imported from a dll on PowerPC Windows NT | |
1838 | On the PowerPC running Windows NT, the @code{dllimport} attribute causes | |
1839 | the compiler to call the function via a global pointer to the function | |
1840 | pointer that is set up by the Windows NT dll library. The pointer name | |
1841 | is formed by combining @code{__imp_} and the function name. | |
1842 | ||
1843 | @item dllexport | |
1844 | @cindex functions which are exported from a dll on PowerPC Windows NT | |
1845 | On the PowerPC running Windows NT, the @code{dllexport} attribute causes | |
1846 | the compiler to provide a global pointer to the function pointer, so | |
1847 | that it can be called with the @code{dllimport} attribute. The pointer | |
1848 | name is formed by combining @code{__imp_} and the function name. | |
1849 | ||
1850 | @item exception (@var{except-func} [, @var{except-arg}]) | |
1851 | @cindex functions which specify exception handling on PowerPC Windows NT | |
1852 | On the PowerPC running Windows NT, the @code{exception} attribute causes | |
1853 | the compiler to modify the structured exception table entry it emits for | |
1854 | the declared function. The string or identifier @var{except-func} is | |
1855 | placed in the third entry of the structured exception table. It | |
1856 | represents a function, which is called by the exception handling | |
1857 | mechanism if an exception occurs. If it was specified, the string or | |
1858 | identifier @var{except-arg} is placed in the fourth entry of the | |
1859 | structured exception table. | |
1860 | ||
1861 | @item function_vector | |
1862 | @cindex calling functions through the function vector on the H8/300 processors | |
1863 | Use this option on the H8/300 and H8/300H to indicate that the specified | |
1864 | function should be called through the function vector. Calling a | |
1865 | function through the function vector will reduce code size, however; | |
1866 | the function vector has a limited size (maximum 128 entries on the H8/300 | |
1867 | and 64 entries on the H8/300H) and shares space with the interrupt vector. | |
1868 | ||
1869 | You must use GAS and GLD from GNU binutils version 2.7 or later for | |
1870 | this option to work correctly. | |
1871 | ||
1872 | @item interrupt_handler | |
1873 | @cindex interrupt handler functions on the H8/300 processors | |
1874 | Use this option on the H8/300 and H8/300H to indicate that the specified | |
1875 | function is an interrupt handler. The compiler will generate function | |
1876 | entry and exit sequences suitable for use in an interrupt handler when this | |
1877 | attribute is present. | |
1878 | ||
6d3d9133 NC |
1879 | @item interrupt |
1880 | @cindex interrupt handler functions | |
1881 | Use this option on the ARM, AVR and M32R/D ports to indicate that the | |
1882 | specified function is an interrupt handler. The compiler will generate | |
1883 | function entry and exit sequences suitable for use in an interrupt | |
1884 | handler when this attribute is present. | |
1885 | ||
a7cf60a2 | 1886 | Note, interrupt handlers for the H8/300 and H8/300H processors can be |
6d3d9133 NC |
1887 | specified via the @code{interrupt_handler} attribute. |
1888 | ||
1889 | Note, on the AVR interrupts will be enabled inside the function. | |
1890 | ||
1891 | Note, for the ARM you can specify the kind of interrupt to be handled by | |
1892 | adding an optional parameter to the interrupt attribute like this: | |
1893 | ||
1894 | @smallexample | |
1895 | void f () __attribute__ ((interrupt ("IRQ"))); | |
1896 | @end smallexample | |
1897 | ||
a7cf60a2 | 1898 | Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF. |
6d3d9133 | 1899 | |
c1f7febf RK |
1900 | @item eightbit_data |
1901 | @cindex eight bit data on the H8/300 and H8/300H | |
1902 | Use this option on the H8/300 and H8/300H to indicate that the specified | |
1903 | variable should be placed into the eight bit data section. | |
1904 | The compiler will generate more efficient code for certain operations | |
1905 | on data in the eight bit data area. Note the eight bit data area is limited to | |
1906 | 256 bytes of data. | |
1907 | ||
1908 | You must use GAS and GLD from GNU binutils version 2.7 or later for | |
1909 | this option to work correctly. | |
1910 | ||
1911 | @item tiny_data | |
1912 | @cindex tiny data section on the H8/300H | |
1913 | Use this option on the H8/300H to indicate that the specified | |
1914 | variable should be placed into the tiny data section. | |
1915 | The compiler will generate more efficient code for loads and stores | |
1916 | on data in the tiny data section. Note the tiny data area is limited to | |
1917 | slightly under 32kbytes of data. | |
845da534 | 1918 | |
052a4b28 DC |
1919 | @item signal |
1920 | @cindex signal handler functions on the AVR processors | |
1921 | Use this option on the AVR to indicate that the specified | |
1922 | function is an signal handler. The compiler will generate function | |
1923 | entry and exit sequences suitable for use in an signal handler when this | |
1924 | attribute is present. Interrupts will be disabled inside function. | |
1925 | ||
1926 | @item naked | |
6d3d9133 NC |
1927 | @cindex function without a prologue/epilogue code |
1928 | Use this option on the ARM or AVR ports to indicate that the specified | |
1929 | function do not need prologue/epilogue sequences generated by the | |
1930 | compiler. It is up to the programmer to provide these sequences. | |
052a4b28 | 1931 | |
845da534 DE |
1932 | @item model (@var{model-name}) |
1933 | @cindex function addressability on the M32R/D | |
1934 | Use this attribute on the M32R/D to set the addressability of an object, | |
1935 | and the code generated for a function. | |
1936 | The identifier @var{model-name} is one of @code{small}, @code{medium}, | |
1937 | or @code{large}, representing each of the code models. | |
1938 | ||
1939 | Small model objects live in the lower 16MB of memory (so that their | |
1940 | addresses can be loaded with the @code{ld24} instruction), and are | |
1941 | callable with the @code{bl} instruction. | |
1942 | ||
1943 | Medium model objects may live anywhere in the 32 bit address space (the | |
1944 | compiler will generate @code{seth/add3} instructions to load their addresses), | |
1945 | and are callable with the @code{bl} instruction. | |
1946 | ||
1947 | Large model objects may live anywhere in the 32 bit address space (the | |
1948 | compiler will generate @code{seth/add3} instructions to load their addresses), | |
1949 | and may not be reachable with the @code{bl} instruction (the compiler will | |
1950 | generate the much slower @code{seth/add3/jl} instruction sequence). | |
1951 | ||
c1f7febf RK |
1952 | @end table |
1953 | ||
1954 | You can specify multiple attributes in a declaration by separating them | |
1955 | by commas within the double parentheses or by immediately following an | |
1956 | attribute declaration with another attribute declaration. | |
1957 | ||
1958 | @cindex @code{#pragma}, reason for not using | |
1959 | @cindex pragma, reason for not using | |
9f1bbeaa JM |
1960 | Some people object to the @code{__attribute__} feature, suggesting that |
1961 | ISO C's @code{#pragma} should be used instead. At the time | |
1962 | @code{__attribute__} was designed, there were two reasons for not doing | |
1963 | this. | |
c1f7febf RK |
1964 | |
1965 | @enumerate | |
1966 | @item | |
1967 | It is impossible to generate @code{#pragma} commands from a macro. | |
1968 | ||
1969 | @item | |
1970 | There is no telling what the same @code{#pragma} might mean in another | |
1971 | compiler. | |
1972 | @end enumerate | |
1973 | ||
9f1bbeaa JM |
1974 | These two reasons applied to almost any application that might have been |
1975 | proposed for @code{#pragma}. It was basically a mistake to use | |
1976 | @code{#pragma} for @emph{anything}. | |
1977 | ||
1978 | The ISO C99 standard includes @code{_Pragma}, which now allows pragmas | |
1979 | to be generated from macros. In addition, a @code{#pragma GCC} | |
1980 | namespace is now in use for GCC-specific pragmas. However, it has been | |
1981 | found convenient to use @code{__attribute__} to achieve a natural | |
1982 | attachment of attributes to their corresponding declarations, whereas | |
1983 | @code{#pragma GCC} is of use for constructs that do not naturally form | |
1984 | part of the grammar. @xref{Other Directives,,Miscellaneous | |
1985 | Preprocessing Directives, cpp, The C Preprocessor}. | |
c1f7febf | 1986 | |
2c5e91d2 JM |
1987 | @node Attribute Syntax |
1988 | @section Attribute Syntax | |
1989 | @cindex attribute syntax | |
1990 | ||
1991 | This section describes the syntax with which @code{__attribute__} may be | |
1992 | used, and the constructs to which attribute specifiers bind, for the C | |
1993 | language. Some details may vary for C++ and Objective C. Because of | |
1994 | infelicities in the grammar for attributes, some forms described here | |
1995 | may not be successfully parsed in all cases. | |
1996 | ||
1997 | @xref{Function Attributes}, for details of the semantics of attributes | |
1998 | applying to functions. @xref{Variable Attributes}, for details of the | |
1999 | semantics of attributes applying to variables. @xref{Type Attributes}, | |
2000 | for details of the semantics of attributes applying to structure, union | |
2001 | and enumerated types. | |
2002 | ||
2003 | An @dfn{attribute specifier} is of the form | |
2004 | @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list} | |
2005 | is a possibly empty comma-separated sequence of @dfn{attributes}, where | |
2006 | each attribute is one of the following: | |
2007 | ||
2008 | @itemize @bullet | |
2009 | @item | |
2010 | Empty. Empty attributes are ignored. | |
2011 | ||
2012 | @item | |
2013 | A word (which may be an identifier such as @code{unused}, or a reserved | |
2014 | word such as @code{const}). | |
2015 | ||
2016 | @item | |
2017 | A word, followed by, in parentheses, parameters for the attribute. | |
2018 | These parameters take one of the following forms: | |
2019 | ||
2020 | @itemize @bullet | |
2021 | @item | |
2022 | An identifier. For example, @code{mode} attributes use this form. | |
2023 | ||
2024 | @item | |
2025 | An identifier followed by a comma and a non-empty comma-separated list | |
2026 | of expressions. For example, @code{format} attributes use this form. | |
2027 | ||
2028 | @item | |
2029 | A possibly empty comma-separated list of expressions. For example, | |
2030 | @code{format_arg} attributes use this form with the list being a single | |
2031 | integer constant expression, and @code{alias} attributes use this form | |
2032 | with the list being a single string constant. | |
2033 | @end itemize | |
2034 | @end itemize | |
2035 | ||
2036 | An @dfn{attribute specifier list} is a sequence of one or more attribute | |
2037 | specifiers, not separated by any other tokens. | |
2038 | ||
2039 | An attribute specifier list may appear after the colon following a | |
2040 | label, other than a @code{case} or @code{default} label. The only | |
2041 | attribute it makes sense to use after a label is @code{unused}. This | |
2042 | feature is intended for code generated by programs which contains labels | |
2043 | that may be unused but which is compiled with @option{-Wall}. It would | |
2044 | not normally be appropriate to use in it human-written code, though it | |
2045 | could be useful in cases where the code that jumps to the label is | |
2046 | contained within an @code{#ifdef} conditional. | |
2047 | ||
2048 | An attribute specifier list may appear as part of a @code{struct}, | |
2049 | @code{union} or @code{enum} specifier. It may go either immediately | |
2050 | after the @code{struct}, @code{union} or @code{enum} keyword, or after | |
2051 | the closing brace. It is ignored if the content of the structure, union | |
2052 | or enumerated type is not defined in the specifier in which the | |
2053 | attribute specifier list is used---that is, in usages such as | |
2054 | @code{struct __attribute__((foo)) bar} with no following opening brace. | |
2055 | Where attribute specifiers follow the closing brace, they are considered | |
2056 | to relate to the structure, union or enumerated type defined, not to any | |
2057 | enclosing declaration the type specifier appears in, and the type | |
2058 | defined is not complete until after the attribute specifiers. | |
2059 | @c Otherwise, there would be the following problems: a shift/reduce | |
2060 | @c conflict between attributes binding the the struct/union/enum and | |
2061 | @c binding to the list of specifiers/qualifiers; and "aligned" | |
2062 | @c attributes could use sizeof for the structure, but the size could be | |
2063 | @c changed later by "packed" attributes. | |
2064 | ||
2065 | Otherwise, an attribute specifier appears as part of a declaration, | |
2066 | counting declarations of unnamed parameters and type names, and relates | |
2067 | to that declaration (which may be nested in another declaration, for | |
2068 | example in the case of a parameter declaration). In future, attribute | |
2069 | specifiers in some places may however apply to a particular declarator | |
2070 | within a declaration instead; these cases are noted below. | |
2071 | ||
2072 | Any list of specifiers and qualifiers at the start of a declaration may | |
2073 | contain attribute specifiers, whether or not such a list may in that | |
2074 | context contain storage class specifiers. (Some attributes, however, | |
2075 | are essentially in the nature of storage class specifiers, and only make | |
2076 | sense where storage class specifiers may be used; for example, | |
2077 | @code{section}.) There is one necessary limitation to this syntax: the | |
2078 | first old-style parameter declaration in a function definition cannot | |
2079 | begin with an attribute specifier, because such an attribute applies to | |
2080 | the function instead by syntax described below (which, however, is not | |
2081 | yet implemented in this case). In some other cases, attribute | |
2082 | specifiers are permitted by this grammar but not yet supported by the | |
2083 | compiler. All attribute specifiers in this place relate to the | |
2084 | declaration as a whole. In the obsolencent usage where a type of | |
2085 | @code{int} is implied by the absence of type specifiers, such a list of | |
2086 | specifiers and qualifiers may be an attribute specifier list with no | |
2087 | other specifiers or qualifiers. | |
2088 | ||
2089 | An attribute specifier list may appear immediately before a declarator | |
2090 | (other than the first) in a comma-separated list of declarators in a | |
2091 | declaration of more than one identifier using a single list of | |
2092 | specifiers and qualifiers. At present, such attribute specifiers apply | |
2093 | not only to the identifier before whose declarator they appear, but to | |
2094 | all subsequent identifiers declared in that declaration, but in future | |
2095 | they may apply only to that single identifier. For example, in | |
2096 | @code{__attribute__((noreturn)) void d0 (void), | |
2097 | __attribute__((format(printf, 1, 2))) d1 (const char *, ...), d2 | |
2098 | (void)}, the @code{noreturn} attribute applies to all the functions | |
2099 | declared; the @code{format} attribute should only apply to @code{d1}, | |
2100 | but at present applies to @code{d2} as well (and so causes an error). | |
2101 | ||
2102 | An attribute specifier list may appear immediately before the comma, | |
2103 | @code{=} or semicolon terminating the declaration of an identifier other | |
2104 | than a function definition. At present, such attribute specifiers apply | |
2105 | to the declared object or function, but in future they may attach to the | |
2106 | outermost adjacent declarator. In simple cases there is no difference, | |
2107 | but, for example, in @code{void (****f)(void) | |
2108 | __attribute__((noreturn));}, at present the @code{noreturn} attribute | |
2109 | applies to @code{f}, which causes a warning since @code{f} is not a | |
2110 | function, but in future it may apply to the function @code{****f}. The | |
2111 | precise semantics of what attributes in such cases will apply to are not | |
2112 | yet specified. Where an assembler name for an object or function is | |
2113 | specified (@pxref{Asm Labels}), at present the attribute must follow the | |
2114 | @code{asm} specification; in future, attributes before the @code{asm} | |
2115 | specification may apply to the adjacent declarator, and those after it | |
2116 | to the declared object or function. | |
2117 | ||
2118 | An attribute specifier list may, in future, be permitted to appear after | |
2119 | the declarator in a function definition (before any old-style parameter | |
2120 | declarations or the function body). | |
2121 | ||
2122 | An attribute specifier list may appear at the start of a nested | |
2123 | declarator. At present, there are some limitations in this usage: the | |
2124 | attributes apply to the identifer declared, and to all subsequent | |
2125 | identifiers declared in that declaration (if it includes a | |
2126 | comma-separated list of declarators), rather than to a specific | |
2127 | declarator. When attribute specifiers follow the @code{*} of a pointer | |
2128 | declarator, they must presently follow any type qualifiers present, and | |
2129 | cannot be mixed with them. The following describes intended future | |
2130 | semantics which make this syntax more useful only. It will make the | |
2131 | most sense if you are familiar with the formal specification of | |
2132 | declarators in the ISO C standard. | |
2133 | ||
2134 | Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T | |
2135 | D1}, where @code{T} contains declaration specifiers that specify a type | |
2136 | @var{Type} (such as @code{int}) and @code{D1} is a declarator that | |
2137 | contains an identifier @var{ident}. The type specified for @var{ident} | |
2138 | for derived declarators whose type does not include an attribute | |
2139 | specifier is as in the ISO C standard. | |
2140 | ||
2141 | If @code{D1} has the form @code{( @var{attribute-specifier-list} D )}, | |
2142 | and the declaration @code{T D} specifies the type | |
2143 | ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then | |
2144 | @code{T D1} specifies the type ``@var{derived-declarator-type-list} | |
2145 | @var{attribute-specifier-list} @var{Type}'' for @var{ident}. | |
2146 | ||
2147 | If @code{D1} has the form @code{* | |
2148 | @var{type-qualifier-and-attribute-specifier-list} D}, and the | |
2149 | declaration @code{T D} specifies the type | |
2150 | ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then | |
2151 | @code{T D1} specifies the type ``@var{derived-declarator-type-list} | |
2152 | @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for | |
2153 | @var{ident}. | |
2154 | ||
2155 | For example, @code{void (__attribute__((noreturn)) ****f)();} specifies | |
2156 | the type ``pointer to pointer to pointer to pointer to non-returning | |
2157 | function returning @code{void}''. As another example, @code{char | |
2158 | *__attribute__((aligned(8))) *f;} specifies the type ``pointer to | |
2159 | 8-byte-aligned pointer to @code{char}''. Note again that this describes | |
2160 | intended future semantics, not current implementation. | |
2161 | ||
c1f7febf RK |
2162 | @node Function Prototypes |
2163 | @section Prototypes and Old-Style Function Definitions | |
2164 | @cindex function prototype declarations | |
2165 | @cindex old-style function definitions | |
2166 | @cindex promotion of formal parameters | |
2167 | ||
5490d604 | 2168 | GNU C extends ISO C to allow a function prototype to override a later |
c1f7febf RK |
2169 | old-style non-prototype definition. Consider the following example: |
2170 | ||
2171 | @example | |
2172 | /* @r{Use prototypes unless the compiler is old-fashioned.} */ | |
d863830b | 2173 | #ifdef __STDC__ |
c1f7febf RK |
2174 | #define P(x) x |
2175 | #else | |
2176 | #define P(x) () | |
2177 | #endif | |
2178 | ||
2179 | /* @r{Prototype function declaration.} */ | |
2180 | int isroot P((uid_t)); | |
2181 | ||
2182 | /* @r{Old-style function definition.} */ | |
2183 | int | |
2184 | isroot (x) /* ??? lossage here ??? */ | |
2185 | uid_t x; | |
2186 | @{ | |
2187 | return x == 0; | |
2188 | @} | |
2189 | @end example | |
2190 | ||
5490d604 | 2191 | Suppose the type @code{uid_t} happens to be @code{short}. ISO C does |
c1f7febf RK |
2192 | not allow this example, because subword arguments in old-style |
2193 | non-prototype definitions are promoted. Therefore in this example the | |
2194 | function definition's argument is really an @code{int}, which does not | |
2195 | match the prototype argument type of @code{short}. | |
2196 | ||
5490d604 | 2197 | This restriction of ISO C makes it hard to write code that is portable |
c1f7febf RK |
2198 | to traditional C compilers, because the programmer does not know |
2199 | whether the @code{uid_t} type is @code{short}, @code{int}, or | |
2200 | @code{long}. Therefore, in cases like these GNU C allows a prototype | |
2201 | to override a later old-style definition. More precisely, in GNU C, a | |
2202 | function prototype argument type overrides the argument type specified | |
2203 | by a later old-style definition if the former type is the same as the | |
2204 | latter type before promotion. Thus in GNU C the above example is | |
2205 | equivalent to the following: | |
2206 | ||
2207 | @example | |
2208 | int isroot (uid_t); | |
2209 | ||
2210 | int | |
2211 | isroot (uid_t x) | |
2212 | @{ | |
2213 | return x == 0; | |
2214 | @} | |
2215 | @end example | |
2216 | ||
2217 | GNU C++ does not support old-style function definitions, so this | |
2218 | extension is irrelevant. | |
2219 | ||
2220 | @node C++ Comments | |
2221 | @section C++ Style Comments | |
2222 | @cindex // | |
2223 | @cindex C++ comments | |
2224 | @cindex comments, C++ style | |
2225 | ||
2226 | In GNU C, you may use C++ style comments, which start with @samp{//} and | |
2227 | continue until the end of the line. Many other C implementations allow | |
2228 | such comments, and they are likely to be in a future C standard. | |
2229 | However, C++ style comments are not recognized if you specify | |
5490d604 JM |
2230 | @w{@samp{-ansi}}, a @option{-std} option specifying a version of ISO C |
2231 | before C99, or @w{@samp{-traditional}}, since they are incompatible | |
c1f7febf RK |
2232 | with traditional constructs like @code{dividend//*comment*/divisor}. |
2233 | ||
2234 | @node Dollar Signs | |
2235 | @section Dollar Signs in Identifier Names | |
2236 | @cindex $ | |
2237 | @cindex dollar signs in identifier names | |
2238 | @cindex identifier names, dollar signs in | |
2239 | ||
79188db9 RK |
2240 | In GNU C, you may normally use dollar signs in identifier names. |
2241 | This is because many traditional C implementations allow such identifiers. | |
2242 | However, dollar signs in identifiers are not supported on a few target | |
2243 | machines, typically because the target assembler does not allow them. | |
c1f7febf RK |
2244 | |
2245 | @node Character Escapes | |
2246 | @section The Character @key{ESC} in Constants | |
2247 | ||
2248 | You can use the sequence @samp{\e} in a string or character constant to | |
2249 | stand for the ASCII character @key{ESC}. | |
2250 | ||
2251 | @node Alignment | |
2252 | @section Inquiring on Alignment of Types or Variables | |
2253 | @cindex alignment | |
2254 | @cindex type alignment | |
2255 | @cindex variable alignment | |
2256 | ||
2257 | The keyword @code{__alignof__} allows you to inquire about how an object | |
2258 | is aligned, or the minimum alignment usually required by a type. Its | |
2259 | syntax is just like @code{sizeof}. | |
2260 | ||
2261 | For example, if the target machine requires a @code{double} value to be | |
2262 | aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8. | |
2263 | This is true on many RISC machines. On more traditional machine | |
2264 | designs, @code{__alignof__ (double)} is 4 or even 2. | |
2265 | ||
2266 | Some machines never actually require alignment; they allow reference to any | |
2267 | data type even at an odd addresses. For these machines, @code{__alignof__} | |
2268 | reports the @emph{recommended} alignment of a type. | |
2269 | ||
2270 | When the operand of @code{__alignof__} is an lvalue rather than a type, the | |
2271 | value is the largest alignment that the lvalue is known to have. It may | |
2272 | have this alignment as a result of its data type, or because it is part of | |
2273 | a structure and inherits alignment from that structure. For example, after | |
2274 | this declaration: | |
2275 | ||
2276 | @example | |
2277 | struct foo @{ int x; char y; @} foo1; | |
2278 | @end example | |
2279 | ||
2280 | @noindent | |
2281 | the value of @code{__alignof__ (foo1.y)} is probably 2 or 4, the same as | |
2282 | @code{__alignof__ (int)}, even though the data type of @code{foo1.y} | |
2283 | does not itself demand any alignment.@refill | |
2284 | ||
9d27bffe SS |
2285 | It is an error to ask for the alignment of an incomplete type. |
2286 | ||
c1f7febf RK |
2287 | A related feature which lets you specify the alignment of an object is |
2288 | @code{__attribute__ ((aligned (@var{alignment})))}; see the following | |
2289 | section. | |
2290 | ||
2291 | @node Variable Attributes | |
2292 | @section Specifying Attributes of Variables | |
2293 | @cindex attribute of variables | |
2294 | @cindex variable attributes | |
2295 | ||
2296 | The keyword @code{__attribute__} allows you to specify special | |
2297 | attributes of variables or structure fields. This keyword is followed | |
2298 | by an attribute specification inside double parentheses. Eight | |
2299 | attributes are currently defined for variables: @code{aligned}, | |
2300 | @code{mode}, @code{nocommon}, @code{packed}, @code{section}, | |
9f1bbeaa JM |
2301 | @code{transparent_union}, @code{unused}, and @code{weak}. Some other |
2302 | attributes are defined for variables on particular target systems. Other | |
c1f7febf | 2303 | attributes are available for functions (@pxref{Function Attributes}) and |
5c25e11d PE |
2304 | for types (@pxref{Type Attributes}). Other front-ends might define more |
2305 | attributes (@pxref{C++ Extensions,,Extensions to the C++ Language}). | |
c1f7febf RK |
2306 | |
2307 | You may also specify attributes with @samp{__} preceding and following | |
2308 | each keyword. This allows you to use them in header files without | |
2309 | being concerned about a possible macro of the same name. For example, | |
2310 | you may use @code{__aligned__} instead of @code{aligned}. | |
2311 | ||
2c5e91d2 JM |
2312 | @xref{Attribute Syntax}, for details of the exact syntax for using |
2313 | attributes. | |
2314 | ||
c1f7febf RK |
2315 | @table @code |
2316 | @cindex @code{aligned} attribute | |
2317 | @item aligned (@var{alignment}) | |
2318 | This attribute specifies a minimum alignment for the variable or | |
2319 | structure field, measured in bytes. For example, the declaration: | |
2320 | ||
2321 | @smallexample | |
2322 | int x __attribute__ ((aligned (16))) = 0; | |
2323 | @end smallexample | |
2324 | ||
2325 | @noindent | |
2326 | causes the compiler to allocate the global variable @code{x} on a | |
2327 | 16-byte boundary. On a 68040, this could be used in conjunction with | |
2328 | an @code{asm} expression to access the @code{move16} instruction which | |
2329 | requires 16-byte aligned operands. | |
2330 | ||
2331 | You can also specify the alignment of structure fields. For example, to | |
2332 | create a double-word aligned @code{int} pair, you could write: | |
2333 | ||
2334 | @smallexample | |
2335 | struct foo @{ int x[2] __attribute__ ((aligned (8))); @}; | |
2336 | @end smallexample | |
2337 | ||
2338 | @noindent | |
2339 | This is an alternative to creating a union with a @code{double} member | |
2340 | that forces the union to be double-word aligned. | |
2341 | ||
2342 | It is not possible to specify the alignment of functions; the alignment | |
2343 | of functions is determined by the machine's requirements and cannot be | |
2344 | changed. You cannot specify alignment for a typedef name because such a | |
2345 | name is just an alias, not a distinct type. | |
2346 | ||
2347 | As in the preceding examples, you can explicitly specify the alignment | |
2348 | (in bytes) that you wish the compiler to use for a given variable or | |
2349 | structure field. Alternatively, you can leave out the alignment factor | |
2350 | and just ask the compiler to align a variable or field to the maximum | |
2351 | useful alignment for the target machine you are compiling for. For | |
2352 | example, you could write: | |
2353 | ||
2354 | @smallexample | |
2355 | short array[3] __attribute__ ((aligned)); | |
2356 | @end smallexample | |
2357 | ||
2358 | Whenever you leave out the alignment factor in an @code{aligned} attribute | |
2359 | specification, the compiler automatically sets the alignment for the declared | |
2360 | variable or field to the largest alignment which is ever used for any data | |
2361 | type on the target machine you are compiling for. Doing this can often make | |
2362 | copy operations more efficient, because the compiler can use whatever | |
2363 | instructions copy the biggest chunks of memory when performing copies to | |
2364 | or from the variables or fields that you have aligned this way. | |
2365 | ||
2366 | The @code{aligned} attribute can only increase the alignment; but you | |
2367 | can decrease it by specifying @code{packed} as well. See below. | |
2368 | ||
2369 | Note that the effectiveness of @code{aligned} attributes may be limited | |
2370 | by inherent limitations in your linker. On many systems, the linker is | |
2371 | only able to arrange for variables to be aligned up to a certain maximum | |
2372 | alignment. (For some linkers, the maximum supported alignment may | |
2373 | be very very small.) If your linker is only able to align variables | |
2374 | up to a maximum of 8 byte alignment, then specifying @code{aligned(16)} | |
2375 | in an @code{__attribute__} will still only provide you with 8 byte | |
2376 | alignment. See your linker documentation for further information. | |
2377 | ||
2378 | @item mode (@var{mode}) | |
2379 | @cindex @code{mode} attribute | |
2380 | This attribute specifies the data type for the declaration---whichever | |
2381 | type corresponds to the mode @var{mode}. This in effect lets you | |
2382 | request an integer or floating point type according to its width. | |
2383 | ||
2384 | You may also specify a mode of @samp{byte} or @samp{__byte__} to | |
2385 | indicate the mode corresponding to a one-byte integer, @samp{word} or | |
2386 | @samp{__word__} for the mode of a one-word integer, and @samp{pointer} | |
2387 | or @samp{__pointer__} for the mode used to represent pointers. | |
2388 | ||
2389 | @item nocommon | |
2390 | @cindex @code{nocommon} attribute | |
2391 | This attribute specifies requests GNU CC not to place a variable | |
2392 | ``common'' but instead to allocate space for it directly. If you | |
2393 | specify the @samp{-fno-common} flag, GNU CC will do this for all | |
2394 | variables. | |
2395 | ||
2396 | Specifying the @code{nocommon} attribute for a variable provides an | |
2397 | initialization of zeros. A variable may only be initialized in one | |
2398 | source file. | |
2399 | ||
2400 | @item packed | |
2401 | @cindex @code{packed} attribute | |
2402 | The @code{packed} attribute specifies that a variable or structure field | |
2403 | should have the smallest possible alignment---one byte for a variable, | |
2404 | and one bit for a field, unless you specify a larger value with the | |
2405 | @code{aligned} attribute. | |
2406 | ||
2407 | Here is a structure in which the field @code{x} is packed, so that it | |
2408 | immediately follows @code{a}: | |
2409 | ||
2410 | @example | |
2411 | struct foo | |
2412 | @{ | |
2413 | char a; | |
2414 | int x[2] __attribute__ ((packed)); | |
2415 | @}; | |
2416 | @end example | |
2417 | ||
2418 | @item section ("section-name") | |
2419 | @cindex @code{section} variable attribute | |
2420 | Normally, the compiler places the objects it generates in sections like | |
2421 | @code{data} and @code{bss}. Sometimes, however, you need additional sections, | |
2422 | or you need certain particular variables to appear in special sections, | |
2423 | for example to map to special hardware. The @code{section} | |
2424 | attribute specifies that a variable (or function) lives in a particular | |
2425 | section. For example, this small program uses several specific section names: | |
2426 | ||
2427 | @smallexample | |
2428 | struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @}; | |
2429 | struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @}; | |
2430 | char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @}; | |
2431 | int init_data __attribute__ ((section ("INITDATA"))) = 0; | |
2432 | ||
2433 | main() | |
2434 | @{ | |
2435 | /* Initialize stack pointer */ | |
2436 | init_sp (stack + sizeof (stack)); | |
2437 | ||
2438 | /* Initialize initialized data */ | |
2439 | memcpy (&init_data, &data, &edata - &data); | |
2440 | ||
2441 | /* Turn on the serial ports */ | |
2442 | init_duart (&a); | |
2443 | init_duart (&b); | |
2444 | @} | |
2445 | @end smallexample | |
2446 | ||
2447 | @noindent | |
2448 | Use the @code{section} attribute with an @emph{initialized} definition | |
2449 | of a @emph{global} variable, as shown in the example. GNU CC issues | |
2450 | a warning and otherwise ignores the @code{section} attribute in | |
2451 | uninitialized variable declarations. | |
2452 | ||
2453 | You may only use the @code{section} attribute with a fully initialized | |
2454 | global definition because of the way linkers work. The linker requires | |
2455 | each object be defined once, with the exception that uninitialized | |
2456 | variables tentatively go in the @code{common} (or @code{bss}) section | |
2457 | and can be multiply "defined". You can force a variable to be | |
2458 | initialized with the @samp{-fno-common} flag or the @code{nocommon} | |
2459 | attribute. | |
2460 | ||
2461 | Some file formats do not support arbitrary sections so the @code{section} | |
2462 | attribute is not available on all platforms. | |
2463 | If you need to map the entire contents of a module to a particular | |
2464 | section, consider using the facilities of the linker instead. | |
2465 | ||
593d3a34 MK |
2466 | @item shared |
2467 | @cindex @code{shared} variable attribute | |
2a59078d | 2468 | On Windows NT, in addition to putting variable definitions in a named |
593d3a34 MK |
2469 | section, the section can also be shared among all running copies of an |
2470 | executable or DLL. For example, this small program defines shared data | |
2471 | by putting it in a named section "shared" and marking the section | |
2472 | shareable: | |
2473 | ||
2474 | @smallexample | |
2475 | int foo __attribute__((section ("shared"), shared)) = 0; | |
2476 | ||
2477 | int | |
2478 | main() | |
2479 | @{ | |
2480 | /* Read and write foo. All running copies see the same value. */ | |
2481 | return 0; | |
2482 | @} | |
2483 | @end smallexample | |
2484 | ||
2485 | @noindent | |
2486 | You may only use the @code{shared} attribute along with @code{section} | |
2487 | attribute with a fully initialized global definition because of the way | |
2488 | linkers work. See @code{section} attribute for more information. | |
2489 | ||
2490 | The @code{shared} attribute is only available on Windows NT. | |
2491 | ||
c1f7febf RK |
2492 | @item transparent_union |
2493 | This attribute, attached to a function parameter which is a union, means | |
2494 | that the corresponding argument may have the type of any union member, | |
2495 | but the argument is passed as if its type were that of the first union | |
2496 | member. For more details see @xref{Type Attributes}. You can also use | |
2497 | this attribute on a @code{typedef} for a union data type; then it | |
2498 | applies to all function parameters with that type. | |
2499 | ||
2500 | @item unused | |
2501 | This attribute, attached to a variable, means that the variable is meant | |
2502 | to be possibly unused. GNU CC will not produce a warning for this | |
2503 | variable. | |
2504 | ||
2505 | @item weak | |
2506 | The @code{weak} attribute is described in @xref{Function Attributes}. | |
845da534 DE |
2507 | |
2508 | @item model (@var{model-name}) | |
2509 | @cindex variable addressability on the M32R/D | |
2510 | Use this attribute on the M32R/D to set the addressability of an object. | |
2511 | The identifier @var{model-name} is one of @code{small}, @code{medium}, | |
2512 | or @code{large}, representing each of the code models. | |
2513 | ||
2514 | Small model objects live in the lower 16MB of memory (so that their | |
2515 | addresses can be loaded with the @code{ld24} instruction). | |
2516 | ||
2517 | Medium and large model objects may live anywhere in the 32 bit address space | |
2518 | (the compiler will generate @code{seth/add3} instructions to load their | |
2519 | addresses). | |
2520 | ||
c1f7febf RK |
2521 | @end table |
2522 | ||
2523 | To specify multiple attributes, separate them by commas within the | |
2524 | double parentheses: for example, @samp{__attribute__ ((aligned (16), | |
2525 | packed))}. | |
2526 | ||
2527 | @node Type Attributes | |
2528 | @section Specifying Attributes of Types | |
2529 | @cindex attribute of types | |
2530 | @cindex type attributes | |
2531 | ||
2532 | The keyword @code{__attribute__} allows you to specify special | |
2533 | attributes of @code{struct} and @code{union} types when you define such | |
2534 | types. This keyword is followed by an attribute specification inside | |
9f1bbeaa JM |
2535 | double parentheses. Four attributes are currently defined for types: |
2536 | @code{aligned}, @code{packed}, @code{transparent_union}, and @code{unused}. | |
2537 | Other attributes are defined for functions (@pxref{Function Attributes}) and | |
c1f7febf RK |
2538 | for variables (@pxref{Variable Attributes}). |
2539 | ||
2540 | You may also specify any one of these attributes with @samp{__} | |
2541 | preceding and following its keyword. This allows you to use these | |
2542 | attributes in header files without being concerned about a possible | |
2543 | macro of the same name. For example, you may use @code{__aligned__} | |
2544 | instead of @code{aligned}. | |
2545 | ||
2546 | You may specify the @code{aligned} and @code{transparent_union} | |
2547 | attributes either in a @code{typedef} declaration or just past the | |
2548 | closing curly brace of a complete enum, struct or union type | |
2549 | @emph{definition} and the @code{packed} attribute only past the closing | |
2550 | brace of a definition. | |
2551 | ||
4051959b JM |
2552 | You may also specify attributes between the enum, struct or union |
2553 | tag and the name of the type rather than after the closing brace. | |
2554 | ||
2c5e91d2 JM |
2555 | @xref{Attribute Syntax}, for details of the exact syntax for using |
2556 | attributes. | |
2557 | ||
c1f7febf RK |
2558 | @table @code |
2559 | @cindex @code{aligned} attribute | |
2560 | @item aligned (@var{alignment}) | |
2561 | This attribute specifies a minimum alignment (in bytes) for variables | |
2562 | of the specified type. For example, the declarations: | |
2563 | ||
2564 | @smallexample | |
f69eecfb JL |
2565 | struct S @{ short f[3]; @} __attribute__ ((aligned (8))); |
2566 | typedef int more_aligned_int __attribute__ ((aligned (8))); | |
c1f7febf RK |
2567 | @end smallexample |
2568 | ||
2569 | @noindent | |
d863830b | 2570 | force the compiler to insure (as far as it can) that each variable whose |
c1f7febf RK |
2571 | type is @code{struct S} or @code{more_aligned_int} will be allocated and |
2572 | aligned @emph{at least} on a 8-byte boundary. On a Sparc, having all | |
2573 | variables of type @code{struct S} aligned to 8-byte boundaries allows | |
2574 | the compiler to use the @code{ldd} and @code{std} (doubleword load and | |
2575 | store) instructions when copying one variable of type @code{struct S} to | |
2576 | another, thus improving run-time efficiency. | |
2577 | ||
2578 | Note that the alignment of any given @code{struct} or @code{union} type | |
5490d604 | 2579 | is required by the ISO C standard to be at least a perfect multiple of |
c1f7febf RK |
2580 | the lowest common multiple of the alignments of all of the members of |
2581 | the @code{struct} or @code{union} in question. This means that you @emph{can} | |
2582 | effectively adjust the alignment of a @code{struct} or @code{union} | |
2583 | type by attaching an @code{aligned} attribute to any one of the members | |
2584 | of such a type, but the notation illustrated in the example above is a | |
2585 | more obvious, intuitive, and readable way to request the compiler to | |
2586 | adjust the alignment of an entire @code{struct} or @code{union} type. | |
2587 | ||
2588 | As in the preceding example, you can explicitly specify the alignment | |
2589 | (in bytes) that you wish the compiler to use for a given @code{struct} | |
2590 | or @code{union} type. Alternatively, you can leave out the alignment factor | |
2591 | and just ask the compiler to align a type to the maximum | |
2592 | useful alignment for the target machine you are compiling for. For | |
2593 | example, you could write: | |
2594 | ||
2595 | @smallexample | |
2596 | struct S @{ short f[3]; @} __attribute__ ((aligned)); | |
2597 | @end smallexample | |
2598 | ||
2599 | Whenever you leave out the alignment factor in an @code{aligned} | |
2600 | attribute specification, the compiler automatically sets the alignment | |
2601 | for the type to the largest alignment which is ever used for any data | |
2602 | type on the target machine you are compiling for. Doing this can often | |
2603 | make copy operations more efficient, because the compiler can use | |
2604 | whatever instructions copy the biggest chunks of memory when performing | |
2605 | copies to or from the variables which have types that you have aligned | |
2606 | this way. | |
2607 | ||
2608 | In the example above, if the size of each @code{short} is 2 bytes, then | |
2609 | the size of the entire @code{struct S} type is 6 bytes. The smallest | |
2610 | power of two which is greater than or equal to that is 8, so the | |
2611 | compiler sets the alignment for the entire @code{struct S} type to 8 | |
2612 | bytes. | |
2613 | ||
2614 | Note that although you can ask the compiler to select a time-efficient | |
2615 | alignment for a given type and then declare only individual stand-alone | |
2616 | objects of that type, the compiler's ability to select a time-efficient | |
2617 | alignment is primarily useful only when you plan to create arrays of | |
2618 | variables having the relevant (efficiently aligned) type. If you | |
2619 | declare or use arrays of variables of an efficiently-aligned type, then | |
2620 | it is likely that your program will also be doing pointer arithmetic (or | |
2621 | subscripting, which amounts to the same thing) on pointers to the | |
2622 | relevant type, and the code that the compiler generates for these | |
2623 | pointer arithmetic operations will often be more efficient for | |
2624 | efficiently-aligned types than for other types. | |
2625 | ||
2626 | The @code{aligned} attribute can only increase the alignment; but you | |
2627 | can decrease it by specifying @code{packed} as well. See below. | |
2628 | ||
2629 | Note that the effectiveness of @code{aligned} attributes may be limited | |
2630 | by inherent limitations in your linker. On many systems, the linker is | |
2631 | only able to arrange for variables to be aligned up to a certain maximum | |
2632 | alignment. (For some linkers, the maximum supported alignment may | |
2633 | be very very small.) If your linker is only able to align variables | |
2634 | up to a maximum of 8 byte alignment, then specifying @code{aligned(16)} | |
2635 | in an @code{__attribute__} will still only provide you with 8 byte | |
2636 | alignment. See your linker documentation for further information. | |
2637 | ||
2638 | @item packed | |
2639 | This attribute, attached to an @code{enum}, @code{struct}, or | |
2640 | @code{union} type definition, specified that the minimum required memory | |
2641 | be used to represent the type. | |
2642 | ||
2643 | Specifying this attribute for @code{struct} and @code{union} types is | |
2644 | equivalent to specifying the @code{packed} attribute on each of the | |
2645 | structure or union members. Specifying the @samp{-fshort-enums} | |
2646 | flag on the line is equivalent to specifying the @code{packed} | |
2647 | attribute on all @code{enum} definitions. | |
2648 | ||
2649 | You may only specify this attribute after a closing curly brace on an | |
1cd4bca9 BK |
2650 | @code{enum} definition, not in a @code{typedef} declaration, unless that |
2651 | declaration also contains the definition of the @code{enum}. | |
c1f7febf RK |
2652 | |
2653 | @item transparent_union | |
2654 | This attribute, attached to a @code{union} type definition, indicates | |
2655 | that any function parameter having that union type causes calls to that | |
2656 | function to be treated in a special way. | |
2657 | ||
2658 | First, the argument corresponding to a transparent union type can be of | |
2659 | any type in the union; no cast is required. Also, if the union contains | |
2660 | a pointer type, the corresponding argument can be a null pointer | |
2661 | constant or a void pointer expression; and if the union contains a void | |
2662 | pointer type, the corresponding argument can be any pointer expression. | |
2663 | If the union member type is a pointer, qualifiers like @code{const} on | |
2664 | the referenced type must be respected, just as with normal pointer | |
2665 | conversions. | |
2666 | ||
2667 | Second, the argument is passed to the function using the calling | |
2668 | conventions of first member of the transparent union, not the calling | |
2669 | conventions of the union itself. All members of the union must have the | |
2670 | same machine representation; this is necessary for this argument passing | |
2671 | to work properly. | |
2672 | ||
2673 | Transparent unions are designed for library functions that have multiple | |
2674 | interfaces for compatibility reasons. For example, suppose the | |
2675 | @code{wait} function must accept either a value of type @code{int *} to | |
2676 | comply with Posix, or a value of type @code{union wait *} to comply with | |
2677 | the 4.1BSD interface. If @code{wait}'s parameter were @code{void *}, | |
2678 | @code{wait} would accept both kinds of arguments, but it would also | |
2679 | accept any other pointer type and this would make argument type checking | |
2680 | less useful. Instead, @code{<sys/wait.h>} might define the interface | |
2681 | as follows: | |
2682 | ||
2683 | @smallexample | |
2684 | typedef union | |
2685 | @{ | |
2686 | int *__ip; | |
2687 | union wait *__up; | |
2688 | @} wait_status_ptr_t __attribute__ ((__transparent_union__)); | |
2689 | ||
2690 | pid_t wait (wait_status_ptr_t); | |
2691 | @end smallexample | |
2692 | ||
2693 | This interface allows either @code{int *} or @code{union wait *} | |
2694 | arguments to be passed, using the @code{int *} calling convention. | |
2695 | The program can call @code{wait} with arguments of either type: | |
2696 | ||
2697 | @example | |
2698 | int w1 () @{ int w; return wait (&w); @} | |
2699 | int w2 () @{ union wait w; return wait (&w); @} | |
2700 | @end example | |
2701 | ||
2702 | With this interface, @code{wait}'s implementation might look like this: | |
2703 | ||
2704 | @example | |
2705 | pid_t wait (wait_status_ptr_t p) | |
2706 | @{ | |
2707 | return waitpid (-1, p.__ip, 0); | |
2708 | @} | |
2709 | @end example | |
d863830b JL |
2710 | |
2711 | @item unused | |
2712 | When attached to a type (including a @code{union} or a @code{struct}), | |
2713 | this attribute means that variables of that type are meant to appear | |
2714 | possibly unused. GNU CC will not produce a warning for any variables of | |
2715 | that type, even if the variable appears to do nothing. This is often | |
2716 | the case with lock or thread classes, which are usually defined and then | |
2717 | not referenced, but contain constructors and destructors that have | |
956d6950 | 2718 | nontrivial bookkeeping functions. |
d863830b | 2719 | |
c1f7febf RK |
2720 | @end table |
2721 | ||
2722 | To specify multiple attributes, separate them by commas within the | |
2723 | double parentheses: for example, @samp{__attribute__ ((aligned (16), | |
2724 | packed))}. | |
2725 | ||
2726 | @node Inline | |
2727 | @section An Inline Function is As Fast As a Macro | |
2728 | @cindex inline functions | |
2729 | @cindex integrating function code | |
2730 | @cindex open coding | |
2731 | @cindex macros, inline alternative | |
2732 | ||
2733 | By declaring a function @code{inline}, you can direct GNU CC to | |
2734 | integrate that function's code into the code for its callers. This | |
2735 | makes execution faster by eliminating the function-call overhead; in | |
2736 | addition, if any of the actual argument values are constant, their known | |
2737 | values may permit simplifications at compile time so that not all of the | |
2738 | inline function's code needs to be included. The effect on code size is | |
2739 | less predictable; object code may be larger or smaller with function | |
2740 | inlining, depending on the particular case. Inlining of functions is an | |
2741 | optimization and it really ``works'' only in optimizing compilation. If | |
2742 | you don't use @samp{-O}, no function is really inline. | |
2743 | ||
2744 | To declare a function inline, use the @code{inline} keyword in its | |
2745 | declaration, like this: | |
2746 | ||
2747 | @example | |
2748 | inline int | |
2749 | inc (int *a) | |
2750 | @{ | |
2751 | (*a)++; | |
2752 | @} | |
2753 | @end example | |
2754 | ||
5490d604 | 2755 | (If you are writing a header file to be included in ISO C programs, write |
c1f7febf | 2756 | @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.) |
c1f7febf | 2757 | You can also make all ``simple enough'' functions inline with the option |
247b14bd RH |
2758 | @samp{-finline-functions}. |
2759 | ||
2760 | Note that certain usages in a function definition can make it unsuitable | |
2761 | for inline substitution. Among these usages are: use of varargs, use of | |
2762 | alloca, use of variable sized data types (@pxref{Variable Length}), | |
2763 | use of computed goto (@pxref{Labels as Values}), use of nonlocal goto, | |
2764 | and nested functions (@pxref{Nested Functions}). Using @samp{-Winline} | |
2765 | will warn when a function marked @code{inline} could not be substituted, | |
2766 | and will give the reason for the failure. | |
c1f7febf RK |
2767 | |
2768 | Note that in C and Objective C, unlike C++, the @code{inline} keyword | |
2769 | does not affect the linkage of the function. | |
2770 | ||
2771 | @cindex automatic @code{inline} for C++ member fns | |
2772 | @cindex @code{inline} automatic for C++ member fns | |
2773 | @cindex member fns, automatically @code{inline} | |
2774 | @cindex C++ member fns, automatically @code{inline} | |
2775 | GNU CC automatically inlines member functions defined within the class | |
2776 | body of C++ programs even if they are not explicitly declared | |
2777 | @code{inline}. (You can override this with @samp{-fno-default-inline}; | |
2778 | @pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.) | |
2779 | ||
2780 | @cindex inline functions, omission of | |
2781 | When a function is both inline and @code{static}, if all calls to the | |
2782 | function are integrated into the caller, and the function's address is | |
2783 | never used, then the function's own assembler code is never referenced. | |
2784 | In this case, GNU CC does not actually output assembler code for the | |
2785 | function, unless you specify the option @samp{-fkeep-inline-functions}. | |
2786 | Some calls cannot be integrated for various reasons (in particular, | |
2787 | calls that precede the function's definition cannot be integrated, and | |
2788 | neither can recursive calls within the definition). If there is a | |
2789 | nonintegrated call, then the function is compiled to assembler code as | |
2790 | usual. The function must also be compiled as usual if the program | |
2791 | refers to its address, because that can't be inlined. | |
2792 | ||
2793 | @cindex non-static inline function | |
2794 | When an inline function is not @code{static}, then the compiler must assume | |
2795 | that there may be calls from other source files; since a global symbol can | |
2796 | be defined only once in any program, the function must not be defined in | |
2797 | the other source files, so the calls therein cannot be integrated. | |
2798 | Therefore, a non-@code{static} inline function is always compiled on its | |
2799 | own in the usual fashion. | |
2800 | ||
2801 | If you specify both @code{inline} and @code{extern} in the function | |
2802 | definition, then the definition is used only for inlining. In no case | |
2803 | is the function compiled on its own, not even if you refer to its | |
2804 | address explicitly. Such an address becomes an external reference, as | |
2805 | if you had only declared the function, and had not defined it. | |
2806 | ||
2807 | This combination of @code{inline} and @code{extern} has almost the | |
2808 | effect of a macro. The way to use it is to put a function definition in | |
2809 | a header file with these keywords, and put another copy of the | |
2810 | definition (lacking @code{inline} and @code{extern}) in a library file. | |
2811 | The definition in the header file will cause most calls to the function | |
2812 | to be inlined. If any uses of the function remain, they will refer to | |
2813 | the single copy in the library. | |
2814 | ||
2815 | GNU C does not inline any functions when not optimizing. It is not | |
2816 | clear whether it is better to inline or not, in this case, but we found | |
2817 | that a correct implementation when not optimizing was difficult. So we | |
2818 | did the easy thing, and turned it off. | |
2819 | ||
2820 | @node Extended Asm | |
2821 | @section Assembler Instructions with C Expression Operands | |
2822 | @cindex extended @code{asm} | |
2823 | @cindex @code{asm} expressions | |
2824 | @cindex assembler instructions | |
2825 | @cindex registers | |
2826 | ||
c85f7c16 JL |
2827 | In an assembler instruction using @code{asm}, you can specify the |
2828 | operands of the instruction using C expressions. This means you need not | |
2829 | guess which registers or memory locations will contain the data you want | |
c1f7febf RK |
2830 | to use. |
2831 | ||
c85f7c16 JL |
2832 | You must specify an assembler instruction template much like what |
2833 | appears in a machine description, plus an operand constraint string for | |
2834 | each operand. | |
c1f7febf RK |
2835 | |
2836 | For example, here is how to use the 68881's @code{fsinx} instruction: | |
2837 | ||
2838 | @example | |
2839 | asm ("fsinx %1,%0" : "=f" (result) : "f" (angle)); | |
2840 | @end example | |
2841 | ||
2842 | @noindent | |
2843 | Here @code{angle} is the C expression for the input operand while | |
2844 | @code{result} is that of the output operand. Each has @samp{"f"} as its | |
c85f7c16 JL |
2845 | operand constraint, saying that a floating point register is required. |
2846 | The @samp{=} in @samp{=f} indicates that the operand is an output; all | |
2847 | output operands' constraints must use @samp{=}. The constraints use the | |
2848 | same language used in the machine description (@pxref{Constraints}). | |
2849 | ||
2850 | Each operand is described by an operand-constraint string followed by | |
2851 | the C expression in parentheses. A colon separates the assembler | |
2852 | template from the first output operand and another separates the last | |
2853 | output operand from the first input, if any. Commas separate the | |
2854 | operands within each group. The total number of operands is limited to | |
2855 | ten or to the maximum number of operands in any instruction pattern in | |
2856 | the machine description, whichever is greater. | |
2857 | ||
2858 | If there are no output operands but there are input operands, you must | |
2859 | place two consecutive colons surrounding the place where the output | |
c1f7febf RK |
2860 | operands would go. |
2861 | ||
2862 | Output operand expressions must be lvalues; the compiler can check this. | |
c85f7c16 JL |
2863 | The input operands need not be lvalues. The compiler cannot check |
2864 | whether the operands have data types that are reasonable for the | |
2865 | instruction being executed. It does not parse the assembler instruction | |
2866 | template and does not know what it means or even whether it is valid | |
2867 | assembler input. The extended @code{asm} feature is most often used for | |
2868 | machine instructions the compiler itself does not know exist. If | |
2869 | the output expression cannot be directly addressed (for example, it is a | |
2870 | bit field), your constraint must allow a register. In that case, GNU CC | |
2871 | will use the register as the output of the @code{asm}, and then store | |
2872 | that register into the output. | |
2873 | ||
2874 | The ordinary output operands must be write-only; GNU CC will assume that | |
2875 | the values in these operands before the instruction are dead and need | |
2876 | not be generated. Extended asm supports input-output or read-write | |
2877 | operands. Use the constraint character @samp{+} to indicate such an | |
2878 | operand and list it with the output operands. | |
2879 | ||
2880 | When the constraints for the read-write operand (or the operand in which | |
2881 | only some of the bits are to be changed) allows a register, you may, as | |
2882 | an alternative, logically split its function into two separate operands, | |
2883 | one input operand and one write-only output operand. The connection | |
2884 | between them is expressed by constraints which say they need to be in | |
2885 | the same location when the instruction executes. You can use the same C | |
2886 | expression for both operands, or different expressions. For example, | |
2887 | here we write the (fictitious) @samp{combine} instruction with | |
2888 | @code{bar} as its read-only source operand and @code{foo} as its | |
2889 | read-write destination: | |
c1f7febf RK |
2890 | |
2891 | @example | |
2892 | asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar)); | |
2893 | @end example | |
2894 | ||
2895 | @noindent | |
c85f7c16 JL |
2896 | The constraint @samp{"0"} for operand 1 says that it must occupy the |
2897 | same location as operand 0. A digit in constraint is allowed only in an | |
2898 | input operand and it must refer to an output operand. | |
c1f7febf RK |
2899 | |
2900 | Only a digit in the constraint can guarantee that one operand will be in | |
c85f7c16 JL |
2901 | the same place as another. The mere fact that @code{foo} is the value |
2902 | of both operands is not enough to guarantee that they will be in the | |
2903 | same place in the generated assembler code. The following would not | |
2904 | work reliably: | |
c1f7febf RK |
2905 | |
2906 | @example | |
2907 | asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar)); | |
2908 | @end example | |
2909 | ||
2910 | Various optimizations or reloading could cause operands 0 and 1 to be in | |
2911 | different registers; GNU CC knows no reason not to do so. For example, the | |
2912 | compiler might find a copy of the value of @code{foo} in one register and | |
2913 | use it for operand 1, but generate the output operand 0 in a different | |
2914 | register (copying it afterward to @code{foo}'s own address). Of course, | |
2915 | since the register for operand 1 is not even mentioned in the assembler | |
2916 | code, the result will not work, but GNU CC can't tell that. | |
2917 | ||
c85f7c16 JL |
2918 | Some instructions clobber specific hard registers. To describe this, |
2919 | write a third colon after the input operands, followed by the names of | |
2920 | the clobbered hard registers (given as strings). Here is a realistic | |
2921 | example for the VAX: | |
c1f7febf RK |
2922 | |
2923 | @example | |
2924 | asm volatile ("movc3 %0,%1,%2" | |
2925 | : /* no outputs */ | |
2926 | : "g" (from), "g" (to), "g" (count) | |
2927 | : "r0", "r1", "r2", "r3", "r4", "r5"); | |
2928 | @end example | |
2929 | ||
c5c76735 JL |
2930 | You may not write a clobber description in a way that overlaps with an |
2931 | input or output operand. For example, you may not have an operand | |
2932 | describing a register class with one member if you mention that register | |
2933 | in the clobber list. There is no way for you to specify that an input | |
2934 | operand is modified without also specifying it as an output | |
2935 | operand. Note that if all the output operands you specify are for this | |
2936 | purpose (and hence unused), you will then also need to specify | |
2937 | @code{volatile} for the @code{asm} construct, as described below, to | |
2938 | prevent GNU CC from deleting the @code{asm} statement as unused. | |
8fe1938e | 2939 | |
c1f7febf | 2940 | If you refer to a particular hardware register from the assembler code, |
c85f7c16 JL |
2941 | you will probably have to list the register after the third colon to |
2942 | tell the compiler the register's value is modified. In some assemblers, | |
2943 | the register names begin with @samp{%}; to produce one @samp{%} in the | |
2944 | assembler code, you must write @samp{%%} in the input. | |
2945 | ||
2946 | If your assembler instruction can alter the condition code register, add | |
2947 | @samp{cc} to the list of clobbered registers. GNU CC on some machines | |
2948 | represents the condition codes as a specific hardware register; | |
2949 | @samp{cc} serves to name this register. On other machines, the | |
2950 | condition code is handled differently, and specifying @samp{cc} has no | |
2951 | effect. But it is valid no matter what the machine. | |
c1f7febf RK |
2952 | |
2953 | If your assembler instruction modifies memory in an unpredictable | |
c85f7c16 JL |
2954 | fashion, add @samp{memory} to the list of clobbered registers. This |
2955 | will cause GNU CC to not keep memory values cached in registers across | |
dd40655a GK |
2956 | the assembler instruction. You will also want to add the |
2957 | @code{volatile} keyword if the memory affected is not listed in the | |
2958 | inputs or outputs of the @code{asm}, as the @samp{memory} clobber does | |
2959 | not count as a side-effect of the @code{asm}. | |
c1f7febf | 2960 | |
c85f7c16 | 2961 | You can put multiple assembler instructions together in a single |
8720914b HPN |
2962 | @code{asm} template, separated by the characters normally used in assembly |
2963 | code for the system. A combination that works in most places is a newline | |
2964 | to break the line, plus a tab character to move to the instruction field | |
2965 | (written as @samp{\n\t}). Sometimes semicolons can be used, if the | |
2966 | assembler allows semicolons as a line-breaking character. Note that some | |
2967 | assembler dialects use semicolons to start a comment. | |
2968 | The input operands are guaranteed not to use any of the clobbered | |
c85f7c16 JL |
2969 | registers, and neither will the output operands' addresses, so you can |
2970 | read and write the clobbered registers as many times as you like. Here | |
2971 | is an example of multiple instructions in a template; it assumes the | |
2972 | subroutine @code{_foo} accepts arguments in registers 9 and 10: | |
c1f7febf RK |
2973 | |
2974 | @example | |
8720914b | 2975 | asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo" |
c1f7febf RK |
2976 | : /* no outputs */ |
2977 | : "g" (from), "g" (to) | |
2978 | : "r9", "r10"); | |
2979 | @end example | |
2980 | ||
c85f7c16 JL |
2981 | Unless an output operand has the @samp{&} constraint modifier, GNU CC |
2982 | may allocate it in the same register as an unrelated input operand, on | |
2983 | the assumption the inputs are consumed before the outputs are produced. | |
c1f7febf RK |
2984 | This assumption may be false if the assembler code actually consists of |
2985 | more than one instruction. In such a case, use @samp{&} for each output | |
c85f7c16 | 2986 | operand that may not overlap an input. @xref{Modifiers}. |
c1f7febf | 2987 | |
c85f7c16 JL |
2988 | If you want to test the condition code produced by an assembler |
2989 | instruction, you must include a branch and a label in the @code{asm} | |
2990 | construct, as follows: | |
c1f7febf RK |
2991 | |
2992 | @example | |
8720914b | 2993 | asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:" |
c1f7febf RK |
2994 | : "g" (result) |
2995 | : "g" (input)); | |
2996 | @end example | |
2997 | ||
2998 | @noindent | |
2999 | This assumes your assembler supports local labels, as the GNU assembler | |
3000 | and most Unix assemblers do. | |
3001 | ||
3002 | Speaking of labels, jumps from one @code{asm} to another are not | |
c85f7c16 JL |
3003 | supported. The compiler's optimizers do not know about these jumps, and |
3004 | therefore they cannot take account of them when deciding how to | |
c1f7febf RK |
3005 | optimize. |
3006 | ||
3007 | @cindex macros containing @code{asm} | |
3008 | Usually the most convenient way to use these @code{asm} instructions is to | |
3009 | encapsulate them in macros that look like functions. For example, | |
3010 | ||
3011 | @example | |
3012 | #define sin(x) \ | |
3013 | (@{ double __value, __arg = (x); \ | |
3014 | asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \ | |
3015 | __value; @}) | |
3016 | @end example | |
3017 | ||
3018 | @noindent | |
3019 | Here the variable @code{__arg} is used to make sure that the instruction | |
3020 | operates on a proper @code{double} value, and to accept only those | |
3021 | arguments @code{x} which can convert automatically to a @code{double}. | |
3022 | ||
c85f7c16 JL |
3023 | Another way to make sure the instruction operates on the correct data |
3024 | type is to use a cast in the @code{asm}. This is different from using a | |
c1f7febf RK |
3025 | variable @code{__arg} in that it converts more different types. For |
3026 | example, if the desired type were @code{int}, casting the argument to | |
3027 | @code{int} would accept a pointer with no complaint, while assigning the | |
3028 | argument to an @code{int} variable named @code{__arg} would warn about | |
3029 | using a pointer unless the caller explicitly casts it. | |
3030 | ||
3031 | If an @code{asm} has output operands, GNU CC assumes for optimization | |
c85f7c16 JL |
3032 | purposes the instruction has no side effects except to change the output |
3033 | operands. This does not mean instructions with a side effect cannot be | |
3034 | used, but you must be careful, because the compiler may eliminate them | |
3035 | if the output operands aren't used, or move them out of loops, or | |
3036 | replace two with one if they constitute a common subexpression. Also, | |
3037 | if your instruction does have a side effect on a variable that otherwise | |
3038 | appears not to change, the old value of the variable may be reused later | |
3039 | if it happens to be found in a register. | |
c1f7febf RK |
3040 | |
3041 | You can prevent an @code{asm} instruction from being deleted, moved | |
3042 | significantly, or combined, by writing the keyword @code{volatile} after | |
3043 | the @code{asm}. For example: | |
3044 | ||
3045 | @example | |
c85f7c16 JL |
3046 | #define get_and_set_priority(new) \ |
3047 | (@{ int __old; \ | |
3048 | asm volatile ("get_and_set_priority %0, %1": "=g" (__old) : "g" (new)); \ | |
3049 | __old; @}) | |
24f98470 | 3050 | @end example |
c1f7febf RK |
3051 | |
3052 | @noindent | |
c85f7c16 JL |
3053 | If you write an @code{asm} instruction with no outputs, GNU CC will know |
3054 | the instruction has side-effects and will not delete the instruction or | |
3055 | move it outside of loops. If the side-effects of your instruction are | |
3056 | not purely external, but will affect variables in your program in ways | |
3057 | other than reading the inputs and clobbering the specified registers or | |
3058 | memory, you should write the @code{volatile} keyword to prevent future | |
3059 | versions of GNU CC from moving the instruction around within a core | |
3060 | region. | |
3061 | ||
3062 | An @code{asm} instruction without any operands or clobbers (and ``old | |
3063 | style'' @code{asm}) will not be deleted or moved significantly, | |
3b7a2e58 | 3064 | regardless, unless it is unreachable, the same way as if you had |
c85f7c16 | 3065 | written a @code{volatile} keyword. |
c1f7febf RK |
3066 | |
3067 | Note that even a volatile @code{asm} instruction can be moved in ways | |
3068 | that appear insignificant to the compiler, such as across jump | |
3069 | instructions. You can't expect a sequence of volatile @code{asm} | |
3070 | instructions to remain perfectly consecutive. If you want consecutive | |
3071 | output, use a single @code{asm}. | |
3072 | ||
3073 | It is a natural idea to look for a way to give access to the condition | |
3074 | code left by the assembler instruction. However, when we attempted to | |
3075 | implement this, we found no way to make it work reliably. The problem | |
3076 | is that output operands might need reloading, which would result in | |
3077 | additional following ``store'' instructions. On most machines, these | |
3078 | instructions would alter the condition code before there was time to | |
3079 | test it. This problem doesn't arise for ordinary ``test'' and | |
3080 | ``compare'' instructions because they don't have any output operands. | |
3081 | ||
eda3fbbe GB |
3082 | For reasons similar to those described above, it is not possible to give |
3083 | an assembler instruction access to the condition code left by previous | |
3084 | instructions. | |
3085 | ||
5490d604 | 3086 | If you are writing a header file that should be includable in ISO C |
c1f7febf RK |
3087 | programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate |
3088 | Keywords}. | |
3089 | ||
fe0ce426 JH |
3090 | @subsection i386 floating point asm operands |
3091 | ||
3092 | There are several rules on the usage of stack-like regs in | |
3093 | asm_operands insns. These rules apply only to the operands that are | |
3094 | stack-like regs: | |
3095 | ||
3096 | @enumerate | |
3097 | @item | |
3098 | Given a set of input regs that die in an asm_operands, it is | |
3099 | necessary to know which are implicitly popped by the asm, and | |
3100 | which must be explicitly popped by gcc. | |
3101 | ||
3102 | An input reg that is implicitly popped by the asm must be | |
3103 | explicitly clobbered, unless it is constrained to match an | |
3104 | output operand. | |
3105 | ||
3106 | @item | |
3107 | For any input reg that is implicitly popped by an asm, it is | |
3108 | necessary to know how to adjust the stack to compensate for the pop. | |
3109 | If any non-popped input is closer to the top of the reg-stack than | |
3110 | the implicitly popped reg, it would not be possible to know what the | |
3111 | stack looked like --- it's not clear how the rest of the stack ``slides | |
3112 | up''. | |
3113 | ||
3114 | All implicitly popped input regs must be closer to the top of | |
3115 | the reg-stack than any input that is not implicitly popped. | |
3116 | ||
3117 | It is possible that if an input dies in an insn, reload might | |
3118 | use the input reg for an output reload. Consider this example: | |
3119 | ||
3120 | @example | |
3121 | asm ("foo" : "=t" (a) : "f" (b)); | |
3122 | @end example | |
3123 | ||
3124 | This asm says that input B is not popped by the asm, and that | |
3125 | the asm pushes a result onto the reg-stack, ie, the stack is one | |
3126 | deeper after the asm than it was before. But, it is possible that | |
3127 | reload will think that it can use the same reg for both the input and | |
3128 | the output, if input B dies in this insn. | |
3129 | ||
3130 | If any input operand uses the @code{f} constraint, all output reg | |
3131 | constraints must use the @code{&} earlyclobber. | |
3132 | ||
3133 | The asm above would be written as | |
3134 | ||
3135 | @example | |
3136 | asm ("foo" : "=&t" (a) : "f" (b)); | |
3137 | @end example | |
3138 | ||
3139 | @item | |
3140 | Some operands need to be in particular places on the stack. All | |
3141 | output operands fall in this category --- there is no other way to | |
3142 | know which regs the outputs appear in unless the user indicates | |
3143 | this in the constraints. | |
3144 | ||
3145 | Output operands must specifically indicate which reg an output | |
3146 | appears in after an asm. @code{=f} is not allowed: the operand | |
3147 | constraints must select a class with a single reg. | |
3148 | ||
3149 | @item | |
3150 | Output operands may not be ``inserted'' between existing stack regs. | |
3151 | Since no 387 opcode uses a read/write operand, all output operands | |
3152 | are dead before the asm_operands, and are pushed by the asm_operands. | |
3153 | It makes no sense to push anywhere but the top of the reg-stack. | |
3154 | ||
3155 | Output operands must start at the top of the reg-stack: output | |
3156 | operands may not ``skip'' a reg. | |
3157 | ||
3158 | @item | |
3159 | Some asm statements may need extra stack space for internal | |
3160 | calculations. This can be guaranteed by clobbering stack registers | |
3161 | unrelated to the inputs and outputs. | |
3162 | ||
3163 | @end enumerate | |
3164 | ||
3165 | Here are a couple of reasonable asms to want to write. This asm | |
3166 | takes one input, which is internally popped, and produces two outputs. | |
3167 | ||
3168 | @example | |
3169 | asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp)); | |
3170 | @end example | |
3171 | ||
3172 | This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode, | |
3173 | and replaces them with one output. The user must code the @code{st(1)} | |
3174 | clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs. | |
3175 | ||
3176 | @example | |
3177 | asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)"); | |
3178 | @end example | |
3179 | ||
c1f7febf RK |
3180 | @ifclear INTERNALS |
3181 | @c Show the details on constraints if they do not appear elsewhere in | |
3182 | @c the manual | |
3183 | @include md.texi | |
3184 | @end ifclear | |
3185 | ||
3186 | @node Asm Labels | |
3187 | @section Controlling Names Used in Assembler Code | |
3188 | @cindex assembler names for identifiers | |
3189 | @cindex names used in assembler code | |
3190 | @cindex identifiers, names in assembler code | |
3191 | ||
3192 | You can specify the name to be used in the assembler code for a C | |
3193 | function or variable by writing the @code{asm} (or @code{__asm__}) | |
3194 | keyword after the declarator as follows: | |
3195 | ||
3196 | @example | |
3197 | int foo asm ("myfoo") = 2; | |
3198 | @end example | |
3199 | ||
3200 | @noindent | |
3201 | This specifies that the name to be used for the variable @code{foo} in | |
3202 | the assembler code should be @samp{myfoo} rather than the usual | |
3203 | @samp{_foo}. | |
3204 | ||
3205 | On systems where an underscore is normally prepended to the name of a C | |
3206 | function or variable, this feature allows you to define names for the | |
3207 | linker that do not start with an underscore. | |
3208 | ||
3209 | You cannot use @code{asm} in this way in a function @emph{definition}; but | |
3210 | you can get the same effect by writing a declaration for the function | |
3211 | before its definition and putting @code{asm} there, like this: | |
3212 | ||
3213 | @example | |
3214 | extern func () asm ("FUNC"); | |
3215 | ||
3216 | func (x, y) | |
3217 | int x, y; | |
3218 | @dots{} | |
3219 | @end example | |
3220 | ||
3221 | It is up to you to make sure that the assembler names you choose do not | |
3222 | conflict with any other assembler symbols. Also, you must not use a | |
3223 | register name; that would produce completely invalid assembler code. GNU | |
3224 | CC does not as yet have the ability to store static variables in registers. | |
3225 | Perhaps that will be added. | |
3226 | ||
3227 | @node Explicit Reg Vars | |
3228 | @section Variables in Specified Registers | |
3229 | @cindex explicit register variables | |
3230 | @cindex variables in specified registers | |
3231 | @cindex specified registers | |
3232 | @cindex registers, global allocation | |
3233 | ||
3234 | GNU C allows you to put a few global variables into specified hardware | |
3235 | registers. You can also specify the register in which an ordinary | |
3236 | register variable should be allocated. | |
3237 | ||
3238 | @itemize @bullet | |
3239 | @item | |
3240 | Global register variables reserve registers throughout the program. | |
3241 | This may be useful in programs such as programming language | |
3242 | interpreters which have a couple of global variables that are accessed | |
3243 | very often. | |
3244 | ||
3245 | @item | |
3246 | Local register variables in specific registers do not reserve the | |
3247 | registers. The compiler's data flow analysis is capable of determining | |
3248 | where the specified registers contain live values, and where they are | |
8d344fbc | 3249 | available for other uses. Stores into local register variables may be deleted |
0deaf590 JL |
3250 | when they appear to be dead according to dataflow analysis. References |
3251 | to local register variables may be deleted or moved or simplified. | |
c1f7febf RK |
3252 | |
3253 | These local variables are sometimes convenient for use with the extended | |
3254 | @code{asm} feature (@pxref{Extended Asm}), if you want to write one | |
3255 | output of the assembler instruction directly into a particular register. | |
3256 | (This will work provided the register you specify fits the constraints | |
3257 | specified for that operand in the @code{asm}.) | |
3258 | @end itemize | |
3259 | ||
3260 | @menu | |
3261 | * Global Reg Vars:: | |
3262 | * Local Reg Vars:: | |
3263 | @end menu | |
3264 | ||
3265 | @node Global Reg Vars | |
3266 | @subsection Defining Global Register Variables | |
3267 | @cindex global register variables | |
3268 | @cindex registers, global variables in | |
3269 | ||
3270 | You can define a global register variable in GNU C like this: | |
3271 | ||
3272 | @example | |
3273 | register int *foo asm ("a5"); | |
3274 | @end example | |
3275 | ||
3276 | @noindent | |
3277 | Here @code{a5} is the name of the register which should be used. Choose a | |
3278 | register which is normally saved and restored by function calls on your | |
3279 | machine, so that library routines will not clobber it. | |
3280 | ||
3281 | Naturally the register name is cpu-dependent, so you would need to | |
3282 | conditionalize your program according to cpu type. The register | |
3283 | @code{a5} would be a good choice on a 68000 for a variable of pointer | |
3284 | type. On machines with register windows, be sure to choose a ``global'' | |
3285 | register that is not affected magically by the function call mechanism. | |
3286 | ||
3287 | In addition, operating systems on one type of cpu may differ in how they | |
3288 | name the registers; then you would need additional conditionals. For | |
3289 | example, some 68000 operating systems call this register @code{%a5}. | |
3290 | ||
3291 | Eventually there may be a way of asking the compiler to choose a register | |
3292 | automatically, but first we need to figure out how it should choose and | |
3293 | how to enable you to guide the choice. No solution is evident. | |
3294 | ||
3295 | Defining a global register variable in a certain register reserves that | |
3296 | register entirely for this use, at least within the current compilation. | |
3297 | The register will not be allocated for any other purpose in the functions | |
3298 | in the current compilation. The register will not be saved and restored by | |
3299 | these functions. Stores into this register are never deleted even if they | |
3300 | would appear to be dead, but references may be deleted or moved or | |
3301 | simplified. | |
3302 | ||
3303 | It is not safe to access the global register variables from signal | |
3304 | handlers, or from more than one thread of control, because the system | |
3305 | library routines may temporarily use the register for other things (unless | |
3306 | you recompile them specially for the task at hand). | |
3307 | ||
3308 | @cindex @code{qsort}, and global register variables | |
3309 | It is not safe for one function that uses a global register variable to | |
3310 | call another such function @code{foo} by way of a third function | |
3311 | @code{lose} that was compiled without knowledge of this variable (i.e. in a | |
3312 | different source file in which the variable wasn't declared). This is | |
3313 | because @code{lose} might save the register and put some other value there. | |
3314 | For example, you can't expect a global register variable to be available in | |
3315 | the comparison-function that you pass to @code{qsort}, since @code{qsort} | |
3316 | might have put something else in that register. (If you are prepared to | |
3317 | recompile @code{qsort} with the same global register variable, you can | |
3318 | solve this problem.) | |
3319 | ||
3320 | If you want to recompile @code{qsort} or other source files which do not | |
3321 | actually use your global register variable, so that they will not use that | |
3322 | register for any other purpose, then it suffices to specify the compiler | |
3323 | option @samp{-ffixed-@var{reg}}. You need not actually add a global | |
3324 | register declaration to their source code. | |
3325 | ||
3326 | A function which can alter the value of a global register variable cannot | |
3327 | safely be called from a function compiled without this variable, because it | |
3328 | could clobber the value the caller expects to find there on return. | |
3329 | Therefore, the function which is the entry point into the part of the | |
3330 | program that uses the global register variable must explicitly save and | |
3331 | restore the value which belongs to its caller. | |
3332 | ||
3333 | @cindex register variable after @code{longjmp} | |
3334 | @cindex global register after @code{longjmp} | |
3335 | @cindex value after @code{longjmp} | |
3336 | @findex longjmp | |
3337 | @findex setjmp | |
3338 | On most machines, @code{longjmp} will restore to each global register | |
3339 | variable the value it had at the time of the @code{setjmp}. On some | |
3340 | machines, however, @code{longjmp} will not change the value of global | |
3341 | register variables. To be portable, the function that called @code{setjmp} | |
3342 | should make other arrangements to save the values of the global register | |
3343 | variables, and to restore them in a @code{longjmp}. This way, the same | |
3344 | thing will happen regardless of what @code{longjmp} does. | |
3345 | ||
3346 | All global register variable declarations must precede all function | |
3347 | definitions. If such a declaration could appear after function | |
3348 | definitions, the declaration would be too late to prevent the register from | |
3349 | being used for other purposes in the preceding functions. | |
3350 | ||
3351 | Global register variables may not have initial values, because an | |
3352 | executable file has no means to supply initial contents for a register. | |
3353 | ||
3354 | On the Sparc, there are reports that g3 @dots{} g7 are suitable | |
3355 | registers, but certain library functions, such as @code{getwd}, as well | |
3356 | as the subroutines for division and remainder, modify g3 and g4. g1 and | |
3357 | g2 are local temporaries. | |
3358 | ||
3359 | On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7. | |
3360 | Of course, it will not do to use more than a few of those. | |
3361 | ||
3362 | @node Local Reg Vars | |
3363 | @subsection Specifying Registers for Local Variables | |
3364 | @cindex local variables, specifying registers | |
3365 | @cindex specifying registers for local variables | |
3366 | @cindex registers for local variables | |
3367 | ||
3368 | You can define a local register variable with a specified register | |
3369 | like this: | |
3370 | ||
3371 | @example | |
3372 | register int *foo asm ("a5"); | |
3373 | @end example | |
3374 | ||
3375 | @noindent | |
3376 | Here @code{a5} is the name of the register which should be used. Note | |
3377 | that this is the same syntax used for defining global register | |
3378 | variables, but for a local variable it would appear within a function. | |
3379 | ||
3380 | Naturally the register name is cpu-dependent, but this is not a | |
3381 | problem, since specific registers are most often useful with explicit | |
3382 | assembler instructions (@pxref{Extended Asm}). Both of these things | |
3383 | generally require that you conditionalize your program according to | |
3384 | cpu type. | |
3385 | ||
3386 | In addition, operating systems on one type of cpu may differ in how they | |
3387 | name the registers; then you would need additional conditionals. For | |
3388 | example, some 68000 operating systems call this register @code{%a5}. | |
3389 | ||
c1f7febf RK |
3390 | Defining such a register variable does not reserve the register; it |
3391 | remains available for other uses in places where flow control determines | |
3392 | the variable's value is not live. However, these registers are made | |
e5e809f4 JL |
3393 | unavailable for use in the reload pass; excessive use of this feature |
3394 | leaves the compiler too few available registers to compile certain | |
3395 | functions. | |
3396 | ||
3397 | This option does not guarantee that GNU CC will generate code that has | |
3398 | this variable in the register you specify at all times. You may not | |
3399 | code an explicit reference to this register in an @code{asm} statement | |
3400 | and assume it will always refer to this variable. | |
c1f7febf | 3401 | |
8d344fbc | 3402 | Stores into local register variables may be deleted when they appear to be dead |
0deaf590 JL |
3403 | according to dataflow analysis. References to local register variables may |
3404 | be deleted or moved or simplified. | |
3405 | ||
c1f7febf RK |
3406 | @node Alternate Keywords |
3407 | @section Alternate Keywords | |
3408 | @cindex alternate keywords | |
3409 | @cindex keywords, alternate | |
3410 | ||
5490d604 JM |
3411 | The option @option{-traditional} disables certain keywords; |
3412 | @option{-ansi} and the various @option{-std} options disable certain | |
3413 | others. This causes trouble when you want to use GNU C extensions, or | |
3414 | ISO C features, in a general-purpose header file that should be usable | |
3415 | by all programs, including ISO C programs and traditional ones. The | |
3416 | keywords @code{asm}, @code{typeof} and @code{inline} cannot be used | |
3417 | since they won't work in a program compiled with @option{-ansi} | |
3418 | (although @code{inline} can be used in a program compiled with | |
3419 | @option{-std=c99}), while the keywords @code{const}, @code{volatile}, | |
3420 | @code{signed}, @code{typeof} and @code{inline} won't work in a program | |
3421 | compiled with @option{-traditional}. The ISO C99 keyword | |
3422 | @code{restrict} is only available when @option{-std=gnu99} (which will | |
3423 | eventually be the default) or @option{-std=c99} (or the equivalent | |
3424 | @option{-std=iso9899:1999}) is used.@refill | |
c1f7febf RK |
3425 | |
3426 | The way to solve these problems is to put @samp{__} at the beginning and | |
3427 | end of each problematical keyword. For example, use @code{__asm__} | |
3428 | instead of @code{asm}, @code{__const__} instead of @code{const}, and | |
3429 | @code{__inline__} instead of @code{inline}. | |
3430 | ||
3431 | Other C compilers won't accept these alternative keywords; if you want to | |
3432 | compile with another compiler, you can define the alternate keywords as | |
3433 | macros to replace them with the customary keywords. It looks like this: | |
3434 | ||
3435 | @example | |
3436 | #ifndef __GNUC__ | |
3437 | #define __asm__ asm | |
3438 | #endif | |
3439 | @end example | |
3440 | ||
6e6b0525 | 3441 | @findex __extension__ |
f1b5ff21 | 3442 | @samp{-pedantic} and other options cause warnings for many GNU C extensions. |
dbe519e0 | 3443 | You can |
c1f7febf RK |
3444 | prevent such warnings within one expression by writing |
3445 | @code{__extension__} before the expression. @code{__extension__} has no | |
3446 | effect aside from this. | |
3447 | ||
3448 | @node Incomplete Enums | |
3449 | @section Incomplete @code{enum} Types | |
3450 | ||
3451 | You can define an @code{enum} tag without specifying its possible values. | |
3452 | This results in an incomplete type, much like what you get if you write | |
3453 | @code{struct foo} without describing the elements. A later declaration | |
3454 | which does specify the possible values completes the type. | |
3455 | ||
3456 | You can't allocate variables or storage using the type while it is | |
3457 | incomplete. However, you can work with pointers to that type. | |
3458 | ||
3459 | This extension may not be very useful, but it makes the handling of | |
3460 | @code{enum} more consistent with the way @code{struct} and @code{union} | |
3461 | are handled. | |
3462 | ||
3463 | This extension is not supported by GNU C++. | |
3464 | ||
3465 | @node Function Names | |
3466 | @section Function Names as Strings | |
3467 | ||
22acfb79 NM |
3468 | GNU CC predefines two magic identifiers to hold the name of the current |
3469 | function. The identifier @code{__FUNCTION__} holds the name of the function | |
3470 | as it appears in the source. The identifier @code{__PRETTY_FUNCTION__} | |
3471 | holds the name of the function pretty printed in a language specific | |
3472 | fashion. | |
c1f7febf RK |
3473 | |
3474 | These names are always the same in a C function, but in a C++ function | |
3475 | they may be different. For example, this program: | |
3476 | ||
3477 | @smallexample | |
3478 | extern "C" @{ | |
3479 | extern int printf (char *, ...); | |
3480 | @} | |
3481 | ||
3482 | class a @{ | |
3483 | public: | |
3484 | sub (int i) | |
3485 | @{ | |
3486 | printf ("__FUNCTION__ = %s\n", __FUNCTION__); | |
3487 | printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__); | |
3488 | @} | |
3489 | @}; | |
3490 | ||
3491 | int | |
3492 | main (void) | |
3493 | @{ | |
3494 | a ax; | |
3495 | ax.sub (0); | |
3496 | return 0; | |
3497 | @} | |
3498 | @end smallexample | |
3499 | ||
3500 | @noindent | |
3501 | gives this output: | |
3502 | ||
3503 | @smallexample | |
3504 | __FUNCTION__ = sub | |
3505 | __PRETTY_FUNCTION__ = int a::sub (int) | |
3506 | @end smallexample | |
3507 | ||
22acfb79 NM |
3508 | The compiler automagically replaces the identifiers with a string |
3509 | literal containing the appropriate name. Thus, they are neither | |
3510 | preprocessor macros, like @code{__FILE__} and @code{__LINE__}, nor | |
3511 | variables. This means that they catenate with other string literals, and | |
3512 | that they can be used to initialize char arrays. For example | |
3513 | ||
3514 | @smallexample | |
3515 | char here[] = "Function " __FUNCTION__ " in " __FILE__; | |
3516 | @end smallexample | |
3517 | ||
3518 | On the other hand, @samp{#ifdef __FUNCTION__} does not have any special | |
c1f7febf RK |
3519 | meaning inside a function, since the preprocessor does not do anything |
3520 | special with the identifier @code{__FUNCTION__}. | |
3521 | ||
22acfb79 | 3522 | GNU CC also supports the magic word @code{__func__}, defined by the |
34527c47 | 3523 | ISO standard C-99: |
22acfb79 NM |
3524 | |
3525 | @display | |
3526 | The identifier @code{__func__} is implicitly declared by the translator | |
3527 | as if, immediately following the opening brace of each function | |
3528 | definition, the declaration | |
3529 | ||
3530 | @smallexample | |
3531 | static const char __func__[] = "function-name"; | |
3532 | @end smallexample | |
3533 | ||
3534 | appeared, where function-name is the name of the lexically-enclosing | |
3535 | function. This name is the unadorned name of the function. | |
3536 | @end display | |
3537 | ||
3538 | By this definition, @code{__func__} is a variable, not a string literal. | |
3539 | In particular, @code{__func__} does not catenate with other string | |
3540 | literals. | |
3541 | ||
3542 | In @code{C++}, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} are | |
3543 | variables, declared in the same way as @code{__func__}. | |
3544 | ||
c1f7febf RK |
3545 | @node Return Address |
3546 | @section Getting the Return or Frame Address of a Function | |
3547 | ||
3548 | These functions may be used to get information about the callers of a | |
3549 | function. | |
3550 | ||
3551 | @table @code | |
185ebd6c | 3552 | @findex __builtin_return_address |
c1f7febf RK |
3553 | @item __builtin_return_address (@var{level}) |
3554 | This function returns the return address of the current function, or of | |
3555 | one of its callers. The @var{level} argument is number of frames to | |
3556 | scan up the call stack. A value of @code{0} yields the return address | |
3557 | of the current function, a value of @code{1} yields the return address | |
3558 | of the caller of the current function, and so forth. | |
3559 | ||
3560 | The @var{level} argument must be a constant integer. | |
3561 | ||
3562 | On some machines it may be impossible to determine the return address of | |
3563 | any function other than the current one; in such cases, or when the top | |
3564 | of the stack has been reached, this function will return @code{0}. | |
3565 | ||
3566 | This function should only be used with a non-zero argument for debugging | |
3567 | purposes. | |
3568 | ||
185ebd6c | 3569 | @findex __builtin_frame_address |
c1f7febf RK |
3570 | @item __builtin_frame_address (@var{level}) |
3571 | This function is similar to @code{__builtin_return_address}, but it | |
3572 | returns the address of the function frame rather than the return address | |
3573 | of the function. Calling @code{__builtin_frame_address} with a value of | |
3574 | @code{0} yields the frame address of the current function, a value of | |
3575 | @code{1} yields the frame address of the caller of the current function, | |
3576 | and so forth. | |
3577 | ||
3578 | The frame is the area on the stack which holds local variables and saved | |
3579 | registers. The frame address is normally the address of the first word | |
3580 | pushed on to the stack by the function. However, the exact definition | |
3581 | depends upon the processor and the calling convention. If the processor | |
3582 | has a dedicated frame pointer register, and the function has a frame, | |
3583 | then @code{__builtin_frame_address} will return the value of the frame | |
3584 | pointer register. | |
3585 | ||
3586 | The caveats that apply to @code{__builtin_return_address} apply to this | |
3587 | function as well. | |
3588 | @end table | |
3589 | ||
185ebd6c RH |
3590 | @node Other Builtins |
3591 | @section Other built-in functions provided by GNU CC | |
01702459 JM |
3592 | @cindex builtin functions |
3593 | @findex __builtin_isgreater | |
3594 | @findex __builtin_isgreaterequal | |
3595 | @findex __builtin_isless | |
3596 | @findex __builtin_islessequal | |
3597 | @findex __builtin_islessgreater | |
3598 | @findex __builtin_isunordered | |
3599 | @findex abort | |
3600 | @findex abs | |
3601 | @findex alloca | |
3602 | @findex bcmp | |
3603 | @findex bzero | |
341e3d11 JM |
3604 | @findex cimag |
3605 | @findex cimagf | |
3606 | @findex cimagl | |
3607 | @findex conj | |
3608 | @findex conjf | |
3609 | @findex conjl | |
01702459 JM |
3610 | @findex cos |
3611 | @findex cosf | |
3612 | @findex cosl | |
341e3d11 JM |
3613 | @findex creal |
3614 | @findex crealf | |
3615 | @findex creall | |
01702459 JM |
3616 | @findex exit |
3617 | @findex _exit | |
796cdb65 | 3618 | @findex _Exit |
01702459 JM |
3619 | @findex fabs |
3620 | @findex fabsf | |
3621 | @findex fabsl | |
3622 | @findex ffs | |
18f988a0 | 3623 | @findex fprintf |
01702459 | 3624 | @findex fputs |
e78f4a97 | 3625 | @findex imaxabs |
c7b6c6cd | 3626 | @findex index |
01702459 JM |
3627 | @findex labs |
3628 | @findex llabs | |
3629 | @findex memcmp | |
3630 | @findex memcpy | |
3631 | @findex memset | |
3632 | @findex printf | |
c7b6c6cd | 3633 | @findex rindex |
01702459 JM |
3634 | @findex sin |
3635 | @findex sinf | |
3636 | @findex sinl | |
3637 | @findex sqrt | |
3638 | @findex sqrtf | |
3639 | @findex sqrtl | |
d118937d | 3640 | @findex strcat |
01702459 JM |
3641 | @findex strchr |
3642 | @findex strcmp | |
3643 | @findex strcpy | |
d118937d | 3644 | @findex strcspn |
01702459 | 3645 | @findex strlen |
d118937d | 3646 | @findex strncat |
da9e9f08 KG |
3647 | @findex strncmp |
3648 | @findex strncpy | |
01702459 JM |
3649 | @findex strpbrk |
3650 | @findex strrchr | |
d118937d | 3651 | @findex strspn |
01702459 | 3652 | @findex strstr |
185ebd6c RH |
3653 | |
3654 | GNU CC provides a large number of built-in functions other than the ones | |
3655 | mentioned above. Some of these are for internal use in the processing | |
3656 | of exceptions or variable-length argument lists and will not be | |
3657 | documented here because they may change from time to time; we do not | |
3658 | recommend general use of these functions. | |
3659 | ||
3660 | The remaining functions are provided for optimization purposes. | |
3661 | ||
3662 | GNU CC includes builtin versions of many of the functions in the | |
01702459 JM |
3663 | standard C library. The versions prefixed with @code{__builtin_} will |
3664 | always be treated as having the same meaning as the C library function | |
3665 | even if you specify the @samp{-fno-builtin} (@pxref{C Dialect Options}) | |
3666 | option. Many of these functions are only optimized in certain cases; if | |
3667 | not optimized in a particular case, a call to the library function will | |
3668 | be emitted. | |
3669 | ||
796cdb65 JM |
3670 | The functions @code{abort}, @code{exit}, @code{_Exit} and @code{_exit} |
3671 | are recognized and presumed not to return, but otherwise are not built | |
3672 | in. @code{_exit} is not recognized in strict ISO C mode (@samp{-ansi}, | |
3673 | @samp{-std=c89} or @samp{-std=c99}). @code{_Exit} is not recognized in | |
3674 | strict C89 mode (@samp{-ansi} or @samp{-std=c89}). | |
01702459 JM |
3675 | |
3676 | Outside strict ISO C mode, the functions @code{alloca}, @code{bcmp}, | |
c7b6c6cd KG |
3677 | @code{bzero}, @code{index}, @code{rindex} and @code{ffs} may be handled |
3678 | as builtins. Corresponding versions @code{__builtin_alloca}, | |
3679 | @code{__builtin_bcmp}, @code{__builtin_bzero}, @code{__builtin_index}, | |
3680 | @code{__builtin_rindex} and @code{__builtin_ffs} are also recognized in | |
01702459 JM |
3681 | strict ISO C mode. |
3682 | ||
341e3d11 JM |
3683 | The ISO C99 functions @code{conj}, @code{conjf}, @code{conjl}, |
3684 | @code{creal}, @code{crealf}, @code{creall}, @code{cimag}, @code{cimagf}, | |
3685 | @code{cimagl}, @code{llabs} and @code{imaxabs} are handled as builtins | |
3686 | except in strict ISO C89 mode. There are also builtin versions of the ISO C99 | |
01702459 JM |
3687 | functions @code{cosf}, @code{cosl}, @code{fabsf}, @code{fabsl}, |
3688 | @code{sinf}, @code{sinl}, @code{sqrtf}, and @code{sqrtl}, that are | |
3689 | recognized in any mode since ISO C89 reserves these names for the | |
3690 | purpose to which ISO C99 puts them. All these functions have | |
3691 | corresponding versions prefixed with @code{__builtin_}. | |
3692 | ||
3693 | The following ISO C89 functions are recognized as builtins unless | |
3694 | @samp{-fno-builtin} is specified: @code{abs}, @code{cos}, @code{fabs}, | |
18f988a0 KG |
3695 | @code{fprintf}, @code{fputs}, @code{labs}, @code{memcmp}, @code{memcpy}, |
3696 | @code{memset}, @code{printf}, @code{sin}, @code{sqrt}, @code{strcat}, | |
3697 | @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn}, | |
3698 | @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy}, | |
3699 | @code{strpbrk}, @code{strrchr}, @code{strspn}, and @code{strstr}. All | |
3700 | of these functions have corresponding versions prefixed with | |
3701 | @code{__builtin_}, except that the version for @code{sqrt} is called | |
da9e9f08 | 3702 | @code{__builtin_fsqrt}. |
01702459 JM |
3703 | |
3704 | GNU CC provides builtin versions of the ISO C99 floating point | |
3705 | comparison macros (that avoid raising exceptions for unordered | |
3706 | operands): @code{__builtin_isgreater}, @code{__builtin_isgreaterequal}, | |
3707 | @code{__builtin_isless}, @code{__builtin_islessequal}, | |
3708 | @code{__builtin_islessgreater}, and @code{__builtin_isunordered}. | |
3709 | ||
185ebd6c | 3710 | |
994a57cd | 3711 | @table @code |
185ebd6c | 3712 | @findex __builtin_constant_p |
994a57cd | 3713 | @item __builtin_constant_p (@var{exp}) |
185ebd6c RH |
3714 | You can use the builtin function @code{__builtin_constant_p} to |
3715 | determine if a value is known to be constant at compile-time and hence | |
3716 | that GNU CC can perform constant-folding on expressions involving that | |
3717 | value. The argument of the function is the value to test. The function | |
3718 | returns the integer 1 if the argument is known to be a compile-time | |
3719 | constant and 0 if it is not known to be a compile-time constant. A | |
3720 | return of 0 does not indicate that the value is @emph{not} a constant, | |
3721 | but merely that GNU CC cannot prove it is a constant with the specified | |
3722 | value of the @samp{-O} option. | |
3723 | ||
3724 | You would typically use this function in an embedded application where | |
3725 | memory was a critical resource. If you have some complex calculation, | |
3726 | you may want it to be folded if it involves constants, but need to call | |
3727 | a function if it does not. For example: | |
3728 | ||
4d390518 | 3729 | @smallexample |
185ebd6c RH |
3730 | #define Scale_Value(X) \ |
3731 | (__builtin_constant_p (X) ? ((X) * SCALE + OFFSET) : Scale (X)) | |
3732 | @end smallexample | |
3733 | ||
3734 | You may use this builtin function in either a macro or an inline | |
3735 | function. However, if you use it in an inlined function and pass an | |
3736 | argument of the function as the argument to the builtin, GNU CC will | |
3737 | never return 1 when you call the inline function with a string constant | |
3738 | or constructor expression (@pxref{Constructors}) and will not return 1 | |
3739 | when you pass a constant numeric value to the inline function unless you | |
3740 | specify the @samp{-O} option. | |
3741 | ||
994a57cd RH |
3742 | @findex __builtin_expect |
3743 | @item __builtin_expect(@var{exp}, @var{c}) | |
3744 | You may use @code{__builtin_expect} to provide the compiler with | |
3745 | branch prediction information. In general, you should prefer to | |
3746 | use actual profile feedback for this (@samp{-fprofile-arcs}), as | |
3747 | programmers are notoriously bad at predicting how their programs | |
60b6e1f5 | 3748 | actually perform. However, there are applications in which this |
994a57cd RH |
3749 | data is hard to collect. |
3750 | ||
3751 | The return value is the value of @var{exp}, which should be an | |
3752 | integral expression. The value of @var{c} must be a compile-time | |
3753 | constant. The semantics of the builtin are that it is expected | |
3754 | that @var{exp} == @var{c}. For example: | |
3755 | ||
3756 | @smallexample | |
3757 | if (__builtin_expect (x, 0)) | |
3758 | foo (); | |
3759 | @end smallexample | |
3760 | ||
3761 | @noindent | |
3762 | would indicate that we do not expect to call @code{foo}, since | |
3763 | we expect @code{x} to be zero. Since you are limited to integral | |
3764 | expressions for @var{exp}, you should use constructions such as | |
3765 | ||
3766 | @smallexample | |
3767 | if (__builtin_expect (ptr != NULL, 1)) | |
3768 | error (); | |
3769 | @end smallexample | |
3770 | ||
3771 | @noindent | |
3772 | when testing pointer or floating-point values. | |
3773 | @end table | |
3774 | ||
c1f7febf RK |
3775 | @node C++ Extensions |
3776 | @chapter Extensions to the C++ Language | |
3777 | @cindex extensions, C++ language | |
3778 | @cindex C++ language extensions | |
3779 | ||
3780 | The GNU compiler provides these extensions to the C++ language (and you | |
3781 | can also use most of the C language extensions in your C++ programs). If you | |
3782 | want to write code that checks whether these features are available, you can | |
3783 | test for the GNU compiler the same way as for C programs: check for a | |
3784 | predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to | |
3785 | test specifically for GNU C++ (@pxref{Standard Predefined,,Standard | |
3786 | Predefined Macros,cpp.info,The C Preprocessor}). | |
3787 | ||
3788 | @menu | |
c1f7febf | 3789 | * Min and Max:: C++ Minimum and maximum operators. |
02cac427 | 3790 | * Volatiles:: What constitutes an access to a volatile object. |
49419c8f | 3791 | * Restricted Pointers:: C99 restricted pointers and references. |
c1f7febf | 3792 | * C++ Interface:: You can use a single C++ header file for both |
e6f3b89d | 3793 | declarations and definitions. |
c1f7febf | 3794 | * Template Instantiation:: Methods for ensuring that exactly one copy of |
e6f3b89d | 3795 | each needed template instantiation is emitted. |
0ded1f18 JM |
3796 | * Bound member functions:: You can extract a function pointer to the |
3797 | method denoted by a @samp{->*} or @samp{.*} expression. | |
e6f3b89d PE |
3798 | * C++ Attributes:: Variable, function, and type attributes for C++ only. |
3799 | * Deprecated Features:: Things might disappear from g++. | |
3800 | * Backwards Compatibility:: Compatibilities with earlier definitions of C++. | |
c1f7febf RK |
3801 | @end menu |
3802 | ||
c1f7febf RK |
3803 | @node Min and Max |
3804 | @section Minimum and Maximum Operators in C++ | |
3805 | ||
3806 | It is very convenient to have operators which return the ``minimum'' or the | |
3807 | ``maximum'' of two arguments. In GNU C++ (but not in GNU C), | |
3808 | ||
3809 | @table @code | |
3810 | @item @var{a} <? @var{b} | |
3811 | @findex <? | |
3812 | @cindex minimum operator | |
3813 | is the @dfn{minimum}, returning the smaller of the numeric values | |
3814 | @var{a} and @var{b}; | |
3815 | ||
3816 | @item @var{a} >? @var{b} | |
3817 | @findex >? | |
3818 | @cindex maximum operator | |
3819 | is the @dfn{maximum}, returning the larger of the numeric values @var{a} | |
3820 | and @var{b}. | |
3821 | @end table | |
3822 | ||
3823 | These operations are not primitive in ordinary C++, since you can | |
3824 | use a macro to return the minimum of two things in C++, as in the | |
3825 | following example. | |
3826 | ||
3827 | @example | |
3828 | #define MIN(X,Y) ((X) < (Y) ? : (X) : (Y)) | |
3829 | @end example | |
3830 | ||
3831 | @noindent | |
3832 | You might then use @w{@samp{int min = MIN (i, j);}} to set @var{min} to | |
3833 | the minimum value of variables @var{i} and @var{j}. | |
3834 | ||
3835 | However, side effects in @code{X} or @code{Y} may cause unintended | |
3836 | behavior. For example, @code{MIN (i++, j++)} will fail, incrementing | |
3837 | the smaller counter twice. A GNU C extension allows you to write safe | |
3838 | macros that avoid this kind of problem (@pxref{Naming Types,,Naming an | |
3839 | Expression's Type}). However, writing @code{MIN} and @code{MAX} as | |
3840 | macros also forces you to use function-call notation for a | |
3841 | fundamental arithmetic operation. Using GNU C++ extensions, you can | |
3842 | write @w{@samp{int min = i <? j;}} instead. | |
3843 | ||
3844 | Since @code{<?} and @code{>?} are built into the compiler, they properly | |
3845 | handle expressions with side-effects; @w{@samp{int min = i++ <? j++;}} | |
3846 | works correctly. | |
3847 | ||
02cac427 NS |
3848 | @node Volatiles |
3849 | @section When is a Volatile Object Accessed? | |
3850 | @cindex accessing volatiles | |
3851 | @cindex volatile read | |
3852 | @cindex volatile write | |
3853 | @cindex volatile access | |
3854 | ||
3855 | Both the C and C++ standard have the concept of volatile objects. These | |
3856 | are normally accessed by pointers and used for accessing hardware. The | |
8117da65 | 3857 | standards encourage compilers to refrain from optimizations |
02cac427 NS |
3858 | concerning accesses to volatile objects that it might perform on |
3859 | non-volatile objects. The C standard leaves it implementation defined | |
3860 | as to what constitutes a volatile access. The C++ standard omits to | |
3861 | specify this, except to say that C++ should behave in a similar manner | |
3862 | to C with respect to volatiles, where possible. The minimum either | |
8117da65 | 3863 | standard specifies is that at a sequence point all previous accesses to |
02cac427 NS |
3864 | volatile objects have stabilized and no subsequent accesses have |
3865 | occurred. Thus an implementation is free to reorder and combine | |
3866 | volatile accesses which occur between sequence points, but cannot do so | |
3867 | for accesses across a sequence point. The use of volatiles does not | |
3868 | allow you to violate the restriction on updating objects multiple times | |
3869 | within a sequence point. | |
3870 | ||
3871 | In most expressions, it is intuitively obvious what is a read and what is | |
3872 | a write. For instance | |
3873 | ||
3874 | @example | |
3875 | volatile int *dst = <somevalue>; | |
3876 | volatile int *src = <someothervalue>; | |
3877 | *dst = *src; | |
3878 | @end example | |
3879 | ||
3880 | @noindent | |
3881 | will cause a read of the volatile object pointed to by @var{src} and stores the | |
3882 | value into the volatile object pointed to by @var{dst}. There is no | |
3883 | guarantee that these reads and writes are atomic, especially for objects | |
3884 | larger than @code{int}. | |
3885 | ||
3886 | Less obvious expressions are where something which looks like an access | |
3887 | is used in a void context. An example would be, | |
3888 | ||
3889 | @example | |
3890 | volatile int *src = <somevalue>; | |
3891 | *src; | |
3892 | @end example | |
3893 | ||
3894 | With C, such expressions are rvalues, and as rvalues cause a read of | |
3895 | the object, gcc interprets this as a read of the volatile being pointed | |
3896 | to. The C++ standard specifies that such expressions do not undergo | |
3897 | lvalue to rvalue conversion, and that the type of the dereferenced | |
3898 | object may be incomplete. The C++ standard does not specify explicitly | |
3899 | that it is this lvalue to rvalue conversion which is responsible for | |
3900 | causing an access. However, there is reason to believe that it is, | |
3901 | because otherwise certain simple expressions become undefined. However, | |
3902 | because it would surprise most programmers, g++ treats dereferencing a | |
3903 | pointer to volatile object of complete type in a void context as a read | |
3904 | of the object. When the object has incomplete type, g++ issues a | |
3905 | warning. | |
3906 | ||
3907 | @example | |
3908 | struct S; | |
3909 | struct T @{int m;@}; | |
3910 | volatile S *ptr1 = <somevalue>; | |
3911 | volatile T *ptr2 = <somevalue>; | |
3912 | *ptr1; | |
3913 | *ptr2; | |
3914 | @end example | |
3915 | ||
3916 | In this example, a warning is issued for @code{*ptr1}, and @code{*ptr2} | |
3917 | causes a read of the object pointed to. If you wish to force an error on | |
3918 | the first case, you must force a conversion to rvalue with, for instance | |
3919 | a static cast, @code{static_cast<S>(*ptr1)}. | |
3920 | ||
3921 | When using a reference to volatile, g++ does not treat equivalent | |
3922 | expressions as accesses to volatiles, but instead issues a warning that | |
3923 | no volatile is accessed. The rationale for this is that otherwise it | |
3924 | becomes difficult to determine where volatile access occur, and not | |
3925 | possible to ignore the return value from functions returning volatile | |
3926 | references. Again, if you wish to force a read, cast the reference to | |
3927 | an rvalue. | |
3928 | ||
535233a8 NS |
3929 | @node Restricted Pointers |
3930 | @section Restricting Pointer Aliasing | |
3931 | @cindex restricted pointers | |
3932 | @cindex restricted references | |
3933 | @cindex restricted this pointer | |
3934 | ||
49419c8f | 3935 | As with gcc, g++ understands the C99 feature of restricted pointers, |
535233a8 | 3936 | specified with the @code{__restrict__}, or @code{__restrict} type |
49419c8f | 3937 | qualifier. Because you cannot compile C++ by specifying the -std=c99 |
535233a8 NS |
3938 | language flag, @code{restrict} is not a keyword in C++. |
3939 | ||
3940 | In addition to allowing restricted pointers, you can specify restricted | |
3941 | references, which indicate that the reference is not aliased in the local | |
3942 | context. | |
3943 | ||
3944 | @example | |
3945 | void fn (int *__restrict__ rptr, int &__restrict__ rref) | |
3946 | @{ | |
3947 | @dots{} | |
3948 | @} | |
3949 | @end example | |
3950 | ||
3951 | @noindent | |
3952 | In the body of @code{fn}, @var{rptr} points to an unaliased integer and | |
3953 | @var{rref} refers to a (different) unaliased integer. | |
3954 | ||
3955 | You may also specify whether a member function's @var{this} pointer is | |
3956 | unaliased by using @code{__restrict__} as a member function qualifier. | |
3957 | ||
3958 | @example | |
3959 | void T::fn () __restrict__ | |
3960 | @{ | |
3961 | @dots{} | |
3962 | @} | |
3963 | @end example | |
3964 | ||
3965 | @noindent | |
3966 | Within the body of @code{T::fn}, @var{this} will have the effective | |
3967 | definition @code{T *__restrict__ const this}. Notice that the | |
3968 | interpretation of a @code{__restrict__} member function qualifier is | |
3969 | different to that of @code{const} or @code{volatile} qualifier, in that it | |
3970 | is applied to the pointer rather than the object. This is consistent with | |
3971 | other compilers which implement restricted pointers. | |
3972 | ||
3973 | As with all outermost parameter qualifiers, @code{__restrict__} is | |
3974 | ignored in function definition matching. This means you only need to | |
3975 | specify @code{__restrict__} in a function definition, rather than | |
3976 | in a function prototype as well. | |
3977 | ||
c1f7febf RK |
3978 | @node C++ Interface |
3979 | @section Declarations and Definitions in One Header | |
3980 | ||
3981 | @cindex interface and implementation headers, C++ | |
3982 | @cindex C++ interface and implementation headers | |
3983 | C++ object definitions can be quite complex. In principle, your source | |
3984 | code will need two kinds of things for each object that you use across | |
3985 | more than one source file. First, you need an @dfn{interface} | |
3986 | specification, describing its structure with type declarations and | |
3987 | function prototypes. Second, you need the @dfn{implementation} itself. | |
3988 | It can be tedious to maintain a separate interface description in a | |
3989 | header file, in parallel to the actual implementation. It is also | |
3990 | dangerous, since separate interface and implementation definitions may | |
3991 | not remain parallel. | |
3992 | ||
3993 | @cindex pragmas, interface and implementation | |
3994 | With GNU C++, you can use a single header file for both purposes. | |
3995 | ||
3996 | @quotation | |
3997 | @emph{Warning:} The mechanism to specify this is in transition. For the | |
3998 | nonce, you must use one of two @code{#pragma} commands; in a future | |
3999 | release of GNU C++, an alternative mechanism will make these | |
4000 | @code{#pragma} commands unnecessary. | |
4001 | @end quotation | |
4002 | ||
4003 | The header file contains the full definitions, but is marked with | |
4004 | @samp{#pragma interface} in the source code. This allows the compiler | |
4005 | to use the header file only as an interface specification when ordinary | |
4006 | source files incorporate it with @code{#include}. In the single source | |
4007 | file where the full implementation belongs, you can use either a naming | |
4008 | convention or @samp{#pragma implementation} to indicate this alternate | |
4009 | use of the header file. | |
4010 | ||
4011 | @table @code | |
4012 | @item #pragma interface | |
4013 | @itemx #pragma interface "@var{subdir}/@var{objects}.h" | |
4014 | @kindex #pragma interface | |
4015 | Use this directive in @emph{header files} that define object classes, to save | |
4016 | space in most of the object files that use those classes. Normally, | |
4017 | local copies of certain information (backup copies of inline member | |
4018 | functions, debugging information, and the internal tables that implement | |
4019 | virtual functions) must be kept in each object file that includes class | |
4020 | definitions. You can use this pragma to avoid such duplication. When a | |
4021 | header file containing @samp{#pragma interface} is included in a | |
4022 | compilation, this auxiliary information will not be generated (unless | |
4023 | the main input source file itself uses @samp{#pragma implementation}). | |
4024 | Instead, the object files will contain references to be resolved at link | |
4025 | time. | |
4026 | ||
4027 | The second form of this directive is useful for the case where you have | |
4028 | multiple headers with the same name in different directories. If you | |
4029 | use this form, you must specify the same string to @samp{#pragma | |
4030 | implementation}. | |
4031 | ||
4032 | @item #pragma implementation | |
4033 | @itemx #pragma implementation "@var{objects}.h" | |
4034 | @kindex #pragma implementation | |
4035 | Use this pragma in a @emph{main input file}, when you want full output from | |
4036 | included header files to be generated (and made globally visible). The | |
4037 | included header file, in turn, should use @samp{#pragma interface}. | |
4038 | Backup copies of inline member functions, debugging information, and the | |
4039 | internal tables used to implement virtual functions are all generated in | |
4040 | implementation files. | |
4041 | ||
4042 | @cindex implied @code{#pragma implementation} | |
4043 | @cindex @code{#pragma implementation}, implied | |
4044 | @cindex naming convention, implementation headers | |
4045 | If you use @samp{#pragma implementation} with no argument, it applies to | |
4046 | an include file with the same basename@footnote{A file's @dfn{basename} | |
4047 | was the name stripped of all leading path information and of trailing | |
4048 | suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source | |
4049 | file. For example, in @file{allclass.cc}, giving just | |
4050 | @samp{#pragma implementation} | |
4051 | by itself is equivalent to @samp{#pragma implementation "allclass.h"}. | |
4052 | ||
4053 | In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as | |
4054 | an implementation file whenever you would include it from | |
4055 | @file{allclass.cc} even if you never specified @samp{#pragma | |
4056 | implementation}. This was deemed to be more trouble than it was worth, | |
4057 | however, and disabled. | |
4058 | ||
4059 | If you use an explicit @samp{#pragma implementation}, it must appear in | |
4060 | your source file @emph{before} you include the affected header files. | |
4061 | ||
4062 | Use the string argument if you want a single implementation file to | |
4063 | include code from multiple header files. (You must also use | |
4064 | @samp{#include} to include the header file; @samp{#pragma | |
4065 | implementation} only specifies how to use the file---it doesn't actually | |
4066 | include it.) | |
4067 | ||
4068 | There is no way to split up the contents of a single header file into | |
4069 | multiple implementation files. | |
4070 | @end table | |
4071 | ||
4072 | @cindex inlining and C++ pragmas | |
4073 | @cindex C++ pragmas, effect on inlining | |
4074 | @cindex pragmas in C++, effect on inlining | |
4075 | @samp{#pragma implementation} and @samp{#pragma interface} also have an | |
4076 | effect on function inlining. | |
4077 | ||
4078 | If you define a class in a header file marked with @samp{#pragma | |
4079 | interface}, the effect on a function defined in that class is similar to | |
4080 | an explicit @code{extern} declaration---the compiler emits no code at | |
4081 | all to define an independent version of the function. Its definition | |
4082 | is used only for inlining with its callers. | |
4083 | ||
4084 | Conversely, when you include the same header file in a main source file | |
4085 | that declares it as @samp{#pragma implementation}, the compiler emits | |
4086 | code for the function itself; this defines a version of the function | |
4087 | that can be found via pointers (or by callers compiled without | |
4088 | inlining). If all calls to the function can be inlined, you can avoid | |
4089 | emitting the function by compiling with @samp{-fno-implement-inlines}. | |
4090 | If any calls were not inlined, you will get linker errors. | |
4091 | ||
4092 | @node Template Instantiation | |
4093 | @section Where's the Template? | |
4094 | ||
4095 | @cindex template instantiation | |
4096 | ||
4097 | C++ templates are the first language feature to require more | |
4098 | intelligence from the environment than one usually finds on a UNIX | |
4099 | system. Somehow the compiler and linker have to make sure that each | |
4100 | template instance occurs exactly once in the executable if it is needed, | |
4101 | and not at all otherwise. There are two basic approaches to this | |
4102 | problem, which I will refer to as the Borland model and the Cfront model. | |
4103 | ||
4104 | @table @asis | |
4105 | @item Borland model | |
4106 | Borland C++ solved the template instantiation problem by adding the code | |
469b759e JM |
4107 | equivalent of common blocks to their linker; the compiler emits template |
4108 | instances in each translation unit that uses them, and the linker | |
4109 | collapses them together. The advantage of this model is that the linker | |
4110 | only has to consider the object files themselves; there is no external | |
4111 | complexity to worry about. This disadvantage is that compilation time | |
4112 | is increased because the template code is being compiled repeatedly. | |
4113 | Code written for this model tends to include definitions of all | |
4114 | templates in the header file, since they must be seen to be | |
4115 | instantiated. | |
c1f7febf RK |
4116 | |
4117 | @item Cfront model | |
4118 | The AT&T C++ translator, Cfront, solved the template instantiation | |
4119 | problem by creating the notion of a template repository, an | |
469b759e JM |
4120 | automatically maintained place where template instances are stored. A |
4121 | more modern version of the repository works as follows: As individual | |
4122 | object files are built, the compiler places any template definitions and | |
4123 | instantiations encountered in the repository. At link time, the link | |
4124 | wrapper adds in the objects in the repository and compiles any needed | |
4125 | instances that were not previously emitted. The advantages of this | |
4126 | model are more optimal compilation speed and the ability to use the | |
4127 | system linker; to implement the Borland model a compiler vendor also | |
c1f7febf | 4128 | needs to replace the linker. The disadvantages are vastly increased |
469b759e JM |
4129 | complexity, and thus potential for error; for some code this can be |
4130 | just as transparent, but in practice it can been very difficult to build | |
c1f7febf | 4131 | multiple programs in one directory and one program in multiple |
469b759e JM |
4132 | directories. Code written for this model tends to separate definitions |
4133 | of non-inline member templates into a separate file, which should be | |
4134 | compiled separately. | |
c1f7febf RK |
4135 | @end table |
4136 | ||
469b759e | 4137 | When used with GNU ld version 2.8 or later on an ELF system such as |
a4b3b54a JM |
4138 | Linux/GNU or Solaris 2, or on Microsoft Windows, g++ supports the |
4139 | Borland model. On other systems, g++ implements neither automatic | |
4140 | model. | |
469b759e JM |
4141 | |
4142 | A future version of g++ will support a hybrid model whereby the compiler | |
4143 | will emit any instantiations for which the template definition is | |
4144 | included in the compile, and store template definitions and | |
4145 | instantiation context information into the object file for the rest. | |
4146 | The link wrapper will extract that information as necessary and invoke | |
4147 | the compiler to produce the remaining instantiations. The linker will | |
4148 | then combine duplicate instantiations. | |
4149 | ||
4150 | In the mean time, you have the following options for dealing with | |
4151 | template instantiations: | |
c1f7febf RK |
4152 | |
4153 | @enumerate | |
d863830b JL |
4154 | @item |
4155 | Compile your template-using code with @samp{-frepo}. The compiler will | |
4156 | generate files with the extension @samp{.rpo} listing all of the | |
4157 | template instantiations used in the corresponding object files which | |
4158 | could be instantiated there; the link wrapper, @samp{collect2}, will | |
4159 | then update the @samp{.rpo} files to tell the compiler where to place | |
4160 | those instantiations and rebuild any affected object files. The | |
4161 | link-time overhead is negligible after the first pass, as the compiler | |
4162 | will continue to place the instantiations in the same files. | |
4163 | ||
4164 | This is your best option for application code written for the Borland | |
4165 | model, as it will just work. Code written for the Cfront model will | |
4166 | need to be modified so that the template definitions are available at | |
4167 | one or more points of instantiation; usually this is as simple as adding | |
4168 | @code{#include <tmethods.cc>} to the end of each template header. | |
4169 | ||
4170 | For library code, if you want the library to provide all of the template | |
4171 | instantiations it needs, just try to link all of its object files | |
4172 | together; the link will fail, but cause the instantiations to be | |
4173 | generated as a side effect. Be warned, however, that this may cause | |
4174 | conflicts if multiple libraries try to provide the same instantiations. | |
4175 | For greater control, use explicit instantiation as described in the next | |
4176 | option. | |
4177 | ||
c1f7febf RK |
4178 | @item |
4179 | Compile your code with @samp{-fno-implicit-templates} to disable the | |
4180 | implicit generation of template instances, and explicitly instantiate | |
4181 | all the ones you use. This approach requires more knowledge of exactly | |
4182 | which instances you need than do the others, but it's less | |
4183 | mysterious and allows greater control. You can scatter the explicit | |
4184 | instantiations throughout your program, perhaps putting them in the | |
4185 | translation units where the instances are used or the translation units | |
4186 | that define the templates themselves; you can put all of the explicit | |
4187 | instantiations you need into one big file; or you can create small files | |
4188 | like | |
4189 | ||
4190 | @example | |
4191 | #include "Foo.h" | |
4192 | #include "Foo.cc" | |
4193 | ||
4194 | template class Foo<int>; | |
4195 | template ostream& operator << | |
4196 | (ostream&, const Foo<int>&); | |
4197 | @end example | |
4198 | ||
4199 | for each of the instances you need, and create a template instantiation | |
4200 | library from those. | |
4201 | ||
4202 | If you are using Cfront-model code, you can probably get away with not | |
4203 | using @samp{-fno-implicit-templates} when compiling files that don't | |
4204 | @samp{#include} the member template definitions. | |
4205 | ||
4206 | If you use one big file to do the instantiations, you may want to | |
4207 | compile it without @samp{-fno-implicit-templates} so you get all of the | |
4208 | instances required by your explicit instantiations (but not by any | |
4209 | other files) without having to specify them as well. | |
4210 | ||
4211 | g++ has extended the template instantiation syntax outlined in the | |
03d0f4af | 4212 | Working Paper to allow forward declaration of explicit instantiations |
4003d7f9 JM |
4213 | (with @code{extern}), instantiation of the compiler support data for a |
4214 | template class (i.e. the vtable) without instantiating any of its | |
4215 | members (with @code{inline}), and instantiation of only the static data | |
4216 | members of a template class, without the support data or member | |
4217 | functions (with (@code{static}): | |
c1f7febf RK |
4218 | |
4219 | @example | |
4220 | extern template int max (int, int); | |
c1f7febf | 4221 | inline template class Foo<int>; |
4003d7f9 | 4222 | static template class Foo<int>; |
c1f7febf RK |
4223 | @end example |
4224 | ||
4225 | @item | |
4226 | Do nothing. Pretend g++ does implement automatic instantiation | |
4227 | management. Code written for the Borland model will work fine, but | |
4228 | each translation unit will contain instances of each of the templates it | |
4229 | uses. In a large program, this can lead to an unacceptable amount of code | |
4230 | duplication. | |
4231 | ||
4232 | @item | |
4233 | Add @samp{#pragma interface} to all files containing template | |
4234 | definitions. For each of these files, add @samp{#pragma implementation | |
4235 | "@var{filename}"} to the top of some @samp{.C} file which | |
4236 | @samp{#include}s it. Then compile everything with | |
4237 | @samp{-fexternal-templates}. The templates will then only be expanded | |
4238 | in the translation unit which implements them (i.e. has a @samp{#pragma | |
4239 | implementation} line for the file where they live); all other files will | |
4240 | use external references. If you're lucky, everything should work | |
4241 | properly. If you get undefined symbol errors, you need to make sure | |
4242 | that each template instance which is used in the program is used in the | |
4243 | file which implements that template. If you don't have any use for a | |
4244 | particular instance in that file, you can just instantiate it | |
4245 | explicitly, using the syntax from the latest C++ working paper: | |
4246 | ||
4247 | @example | |
4248 | template class A<int>; | |
4249 | template ostream& operator << (ostream&, const A<int>&); | |
4250 | @end example | |
4251 | ||
4252 | This strategy will work with code written for either model. If you are | |
4253 | using code written for the Cfront model, the file containing a class | |
4254 | template and the file containing its member templates should be | |
4255 | implemented in the same translation unit. | |
4256 | ||
4257 | A slight variation on this approach is to instead use the flag | |
4258 | @samp{-falt-external-templates}; this flag causes template | |
4259 | instances to be emitted in the translation unit that implements the | |
4260 | header where they are first instantiated, rather than the one which | |
4261 | implements the file where the templates are defined. This header must | |
4262 | be the same in all translation units, or things are likely to break. | |
4263 | ||
4264 | @xref{C++ Interface,,Declarations and Definitions in One Header}, for | |
4265 | more discussion of these pragmas. | |
4266 | @end enumerate | |
4267 | ||
0ded1f18 JM |
4268 | @node Bound member functions |
4269 | @section Extracting the function pointer from a bound pointer to member function | |
4270 | ||
4271 | @cindex pmf | |
4272 | @cindex pointer to member function | |
4273 | @cindex bound pointer to member function | |
4274 | ||
4275 | In C++, pointer to member functions (PMFs) are implemented using a wide | |
4276 | pointer of sorts to handle all the possible call mechanisms; the PMF | |
4277 | needs to store information about how to adjust the @samp{this} pointer, | |
4278 | and if the function pointed to is virtual, where to find the vtable, and | |
4279 | where in the vtable to look for the member function. If you are using | |
4280 | PMFs in an inner loop, you should really reconsider that decision. If | |
4281 | that is not an option, you can extract the pointer to the function that | |
4282 | would be called for a given object/PMF pair and call it directly inside | |
4283 | the inner loop, to save a bit of time. | |
4284 | ||
4285 | Note that you will still be paying the penalty for the call through a | |
4286 | function pointer; on most modern architectures, such a call defeats the | |
4287 | branch prediction features of the CPU. This is also true of normal | |
4288 | virtual function calls. | |
4289 | ||
4290 | The syntax for this extension is | |
4291 | ||
4292 | @example | |
4293 | extern A a; | |
4294 | extern int (A::*fp)(); | |
4295 | typedef int (*fptr)(A *); | |
4296 | ||
4297 | fptr p = (fptr)(a.*fp); | |
4298 | @end example | |
4299 | ||
0fb6bbf5 ML |
4300 | For PMF constants (i.e. expressions of the form @samp{&Klasse::Member}), |
4301 | no object is needed to obtain the address of the function. They can be | |
4302 | converted to function pointers directly: | |
4303 | ||
4304 | @example | |
4305 | fptr p1 = (fptr)(&A::foo); | |
4306 | @end example | |
4307 | ||
0ded1f18 JM |
4308 | You must specify @samp{-Wno-pmf-conversions} to use this extension. |
4309 | ||
5c25e11d PE |
4310 | @node C++ Attributes |
4311 | @section C++-Specific Variable, Function, and Type Attributes | |
4312 | ||
4313 | Some attributes only make sense for C++ programs. | |
4314 | ||
4315 | @table @code | |
4316 | @item init_priority (@var{priority}) | |
4317 | @cindex init_priority attribute | |
4318 | ||
4319 | ||
4320 | In Standard C++, objects defined at namespace scope are guaranteed to be | |
4321 | initialized in an order in strict accordance with that of their definitions | |
4322 | @emph{in a given translation unit}. No guarantee is made for initializations | |
4323 | across translation units. However, GNU C++ allows users to control the | |
3844cd2e | 4324 | order of initialization of objects defined at namespace scope with the |
5c25e11d PE |
4325 | @code{init_priority} attribute by specifying a relative @var{priority}, |
4326 | a constant integral expression currently bounded between 101 and 65535 | |
4327 | inclusive. Lower numbers indicate a higher priority. | |
4328 | ||
4329 | In the following example, @code{A} would normally be created before | |
4330 | @code{B}, but the @code{init_priority} attribute has reversed that order: | |
4331 | ||
4332 | @example | |
4333 | Some_Class A __attribute__ ((init_priority (2000))); | |
4334 | Some_Class B __attribute__ ((init_priority (543))); | |
4335 | @end example | |
4336 | ||
4337 | @noindent | |
4338 | Note that the particular values of @var{priority} do not matter; only their | |
4339 | relative ordering. | |
4340 | ||
4341 | ||
4342 | @item com_interface | |
4343 | @cindex com_interface attribute | |
4344 | ||
4345 | @c This is based on: 1) grepping the code, | |
4346 | @c 2) http://gcc.gnu.org/ml/gcc-bugs/1999-08n/msg01212.html | |
4347 | @c 3) http://gcc.gnu.org/ml/gcc-bugs/1999-08n/msg01215.html | |
4348 | @c and 4) a lot of guesswork. You can tell I don't use COM. -pme 21Dec00 | |
4349 | ||
4350 | This type attribute takes no parameters, and marks a class or struct as an | |
4351 | interface for communication via COM; the class will support the COM ABI | |
4352 | rather than the full C++ ABI. Currently this means that RTTI is not possible | |
4353 | with the resulting class heirarchy. The virtual pointer table will be | |
4354 | changed to be COM-compliant. Also, all classes and structs derived from one | |
4355 | marked with this attribute are implicitly marked with the same attribute; | |
4356 | thus, only the base class in a COM hierarchy needs @code{com_interface}. | |
4357 | ||
60c87482 BM |
4358 | @item java_interface |
4359 | @cindex java_interface attribute | |
4360 | ||
4361 | This type attribute informs C++ that the class is a Java interface. It may | |
4362 | only be applied to classes declared within an @code{extern "Java"} block. | |
4363 | Calls to methods declared in this interface will be dispatched using GCJ's | |
4364 | interface table mechanism, instead of regular virtual table dispatch. | |
4365 | ||
5c25e11d PE |
4366 | @end table |
4367 | ||
e6f3b89d PE |
4368 | @node Deprecated Features |
4369 | @section Deprecated Features | |
4370 | ||
4371 | In the past, the GNU C++ compiler was extended to experiment with new | |
4372 | features, at a time when the C++ language was still evolving. Now that | |
4373 | the C++ standard is complete, some of those features are superseded by | |
4374 | superior alternatives. Using the old features might cause a warning in | |
4375 | some cases that the feature will be dropped in the future. In other | |
4376 | cases, the feature might be gone already. | |
4377 | ||
4378 | While the list below is not exhaustive, it documents some of the options | |
4379 | that are now deprecated: | |
4380 | ||
4381 | @table @code | |
4382 | @item -fexternal-templates | |
4383 | @itemx -falt-external-templates | |
4384 | These are two of the many ways for g++ to implement template | |
4385 | instantiation. @xref{Template Instantiation}. The C++ standard clearly | |
4386 | defines how template definitions have to be organized across | |
4387 | implementation units. g++ has an implicit instantiation mechanism that | |
4388 | should work just fine for standard-conforming code. | |
4389 | ||
4390 | @item -fstrict-prototype | |
4391 | @itemx -fno-strict-prototype | |
4392 | Previously it was possible to use an empty prototype parameter list to | |
4393 | indicate an unspecified number of parameters (like C), rather than no | |
4394 | parameters, as C++ demands. This feature has been removed, except where | |
4395 | it is required for backwards compatibility @xref{Backwards Compatibility}. | |
4396 | @end table | |
4397 | ||
4398 | The named return value extension has been deprecated, and will be | |
4399 | removed from g++ at some point. | |
4400 | ||
82c18d5c NS |
4401 | The use of initializer lists with new expressions has been deprecated, |
4402 | and will be removed from g++ at some point. | |
4403 | ||
e6f3b89d PE |
4404 | @node Backwards Compatibility |
4405 | @section Backwards Compatibility | |
4406 | @cindex Backwards Compatibility | |
4407 | @cindex ARM [Annotated C++ Reference Manual] | |
4408 | ||
4409 | Now that there is a definitive ISO standard C++, g++ has a specification | |
4410 | to adhere to. The C++ language evolved over time, and features that | |
4411 | used to be acceptable in previous drafts of the standard, such as the ARM | |
4412 | [Annotated C++ Reference Manual], are no longer accepted. In order to allow | |
4413 | compilation of C++ written to such drafts, g++ contains some backwards | |
4414 | compatibilities. @emph{All such backwards compatibility features are | |
4415 | liable to disappear in future versions of g++.} They should be considered | |
4416 | deprecated @xref{Deprecated Features}. | |
4417 | ||
4418 | @table @code | |
4419 | @item For scope | |
4420 | If a variable is declared at for scope, it used to remain in scope until | |
4421 | the end of the scope which contained the for statement (rather than just | |
4422 | within the for scope). g++ retains this, but issues a warning, if such a | |
4423 | variable is accessed outside the for scope. | |
4424 | ||
4425 | @item implicit C language | |
4426 | Old C system header files did not contain an @code{extern "C" @{...@}} | |
4427 | scope to set the language. On such systems, all header files are | |
4428 | implicitly scoped inside a C language scope. Also, an empty prototype | |
4429 | @code{()} will be treated as an unspecified number of arguments, rather | |
4430 | than no arguments, as C++ demands. | |
4431 | @end table | |
4432 |