]>
Commit | Line | Data |
---|---|---|
d60e5448 | 1 | @c Copyright (C) 1988,1989,1992,1993,1994,1996,1998,1999,2000,2001 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 | ||
4b404517 JM |
20 | Some features that are in ISO C99 but not C89 or C++ are also, as |
21 | extensions, accepted by GCC in C89 mode and in C++. | |
5490d604 | 22 | |
c1f7febf RK |
23 | @c The only difference between the two versions of this menu is that the |
24 | @c version for clear INTERNALS has an extra node, "Constraints" (which | |
25 | @c appears in a separate chapter in the other version of the manual). | |
26 | @ifset INTERNALS | |
27 | @menu | |
28 | * Statement Exprs:: Putting statements and declarations inside expressions. | |
29 | * Local Labels:: Labels local to a statement-expression. | |
30 | * Labels as Values:: Getting pointers to labels, and computed gotos. | |
31 | * Nested Functions:: As in Algol and Pascal, lexical scoping of functions. | |
32 | * Constructing Calls:: Dispatching a call to another function. | |
33 | * Naming Types:: Giving a name to the type of some expression. | |
34 | * Typeof:: @code{typeof}: referring to the type of an expression. | |
35 | * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues. | |
36 | * Conditionals:: Omitting the middle operand of a @samp{?:} expression. | |
37 | * Long Long:: Double-word integers---@code{long long int}. | |
38 | * Complex:: Data types for complex numbers. | |
6f4d7222 | 39 | * Hex Floats:: Hexadecimal floating-point constants. |
c1f7febf RK |
40 | * Zero Length:: Zero-length arrays. |
41 | * Variable Length:: Arrays whose length is computed at run time. | |
ccd96f0a NB |
42 | * Variadic Macros:: Macros with a variable number of arguments. |
43 | * Escaped Newlines:: Slightly looser rules for escaped newlines. | |
44 | * Multi-line Strings:: String literals with embedded newlines. | |
c1f7febf RK |
45 | * Subscripting:: Any array can be subscripted, even if not an lvalue. |
46 | * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers. | |
47 | * Initializers:: Non-constant initializers. | |
4b404517 | 48 | * Compound Literals:: Compound literals give structures, unions |
c1f7febf | 49 | or arrays as values. |
4b404517 | 50 | * Designated Inits:: Labeling elements of initializers. |
c1f7febf RK |
51 | * Cast to Union:: Casting to union type from any member of the union. |
52 | * Case Ranges:: `case 1 ... 9' and such. | |
4b404517 | 53 | * Mixed Declarations:: Mixing declarations and code. |
c1f7febf RK |
54 | * Function Attributes:: Declaring that functions have no side effects, |
55 | or that they can never return. | |
2c5e91d2 | 56 | * Attribute Syntax:: Formal syntax for attributes. |
c1f7febf RK |
57 | * Function Prototypes:: Prototype declarations and old-style definitions. |
58 | * C++ Comments:: C++ comments are recognized. | |
59 | * Dollar Signs:: Dollar sign is allowed in identifiers. | |
60 | * Character Escapes:: @samp{\e} stands for the character @key{ESC}. | |
61 | * Variable Attributes:: Specifying attributes of variables. | |
62 | * Type Attributes:: Specifying attributes of types. | |
63 | * Alignment:: Inquiring about the alignment of a type or variable. | |
64 | * Inline:: Defining inline functions (as fast as macros). | |
65 | * Extended Asm:: Assembler instructions with C expressions as operands. | |
66 | (With them you can define ``built-in'' functions.) | |
67 | * Asm Labels:: Specifying the assembler name to use for a C symbol. | |
68 | * Explicit Reg Vars:: Defining variables residing in specified registers. | |
69 | * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files. | |
70 | * Incomplete Enums:: @code{enum foo;}, with details to follow. | |
71 | * Function Names:: Printable strings which are the name of the current | |
72 | function. | |
73 | * Return Address:: Getting the return or frame address of a function. | |
185ebd6c | 74 | * Other Builtins:: Other built-in functions. |
c1f7febf RK |
75 | @end menu |
76 | @end ifset | |
77 | @ifclear INTERNALS | |
78 | @menu | |
79 | * Statement Exprs:: Putting statements and declarations inside expressions. | |
80 | * Local Labels:: Labels local to a statement-expression. | |
81 | * Labels as Values:: Getting pointers to labels, and computed gotos. | |
82 | * Nested Functions:: As in Algol and Pascal, lexical scoping of functions. | |
83 | * Constructing Calls:: Dispatching a call to another function. | |
84 | * Naming Types:: Giving a name to the type of some expression. | |
85 | * Typeof:: @code{typeof}: referring to the type of an expression. | |
86 | * Lvalues:: Using @samp{?:}, @samp{,} and casts in lvalues. | |
87 | * Conditionals:: Omitting the middle operand of a @samp{?:} expression. | |
88 | * Long Long:: Double-word integers---@code{long long int}. | |
89 | * Complex:: Data types for complex numbers. | |
6f4d7222 | 90 | * Hex Floats:: Hexadecimal floating-point constants. |
c1f7febf RK |
91 | * Zero Length:: Zero-length arrays. |
92 | * Variable Length:: Arrays whose length is computed at run time. | |
ccd96f0a NB |
93 | * Variadic Macros:: Macros with a variable number of arguments. |
94 | * Escaped Newlines:: Slightly looser rules for escaped newlines. | |
95 | * Multi-line Strings:: String literals with embedded newlines. | |
c1f7febf RK |
96 | * Subscripting:: Any array can be subscripted, even if not an lvalue. |
97 | * Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers. | |
98 | * Initializers:: Non-constant initializers. | |
4b404517 | 99 | * Compound Literals:: Compound literals give structures, unions |
c1f7febf | 100 | or arrays as values. |
4b404517 | 101 | * Designated Inits:: Labeling elements of initializers. |
c1f7febf RK |
102 | * Cast to Union:: Casting to union type from any member of the union. |
103 | * Case Ranges:: `case 1 ... 9' and such. | |
4b404517 | 104 | * Mixed Declarations:: Mixing declarations and code. |
c1f7febf RK |
105 | * Function Attributes:: Declaring that functions have no side effects, |
106 | or that they can never return. | |
2c5e91d2 | 107 | * Attribute Syntax:: Formal syntax for attributes. |
c1f7febf RK |
108 | * Function Prototypes:: Prototype declarations and old-style definitions. |
109 | * C++ Comments:: C++ comments are recognized. | |
110 | * Dollar Signs:: Dollar sign is allowed in identifiers. | |
111 | * Character Escapes:: @samp{\e} stands for the character @key{ESC}. | |
112 | * Variable Attributes:: Specifying attributes of variables. | |
113 | * Type Attributes:: Specifying attributes of types. | |
114 | * Alignment:: Inquiring about the alignment of a type or variable. | |
115 | * Inline:: Defining inline functions (as fast as macros). | |
116 | * Extended Asm:: Assembler instructions with C expressions as operands. | |
117 | (With them you can define ``built-in'' functions.) | |
118 | * Constraints:: Constraints for asm operands | |
119 | * Asm Labels:: Specifying the assembler name to use for a C symbol. | |
120 | * Explicit Reg Vars:: Defining variables residing in specified registers. | |
121 | * Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files. | |
122 | * Incomplete Enums:: @code{enum foo;}, with details to follow. | |
123 | * Function Names:: Printable strings which are the name of the current | |
124 | function. | |
125 | * Return Address:: Getting the return or frame address of a function. | |
c5c76735 | 126 | * Other Builtins:: Other built-in functions. |
c1f7febf RK |
127 | @end menu |
128 | @end ifclear | |
129 | ||
130 | @node Statement Exprs | |
131 | @section Statements and Declarations in Expressions | |
132 | @cindex statements inside expressions | |
133 | @cindex declarations inside expressions | |
134 | @cindex expressions containing statements | |
135 | @cindex macros, statements in expressions | |
136 | ||
137 | @c the above section title wrapped and causes an underfull hbox.. i | |
138 | @c changed it from "within" to "in". --mew 4feb93 | |
139 | ||
140 | A compound statement enclosed in parentheses may appear as an expression | |
141 | in GNU C. This allows you to use loops, switches, and local variables | |
142 | within an expression. | |
143 | ||
144 | Recall that a compound statement is a sequence of statements surrounded | |
145 | by braces; in this construct, parentheses go around the braces. For | |
146 | example: | |
147 | ||
148 | @example | |
149 | (@{ int y = foo (); int z; | |
150 | if (y > 0) z = y; | |
151 | else z = - y; | |
152 | z; @}) | |
153 | @end example | |
154 | ||
155 | @noindent | |
156 | is a valid (though slightly more complex than necessary) expression | |
157 | for the absolute value of @code{foo ()}. | |
158 | ||
159 | The last thing in the compound statement should be an expression | |
160 | followed by a semicolon; the value of this subexpression serves as the | |
161 | value of the entire construct. (If you use some other kind of statement | |
162 | last within the braces, the construct has type @code{void}, and thus | |
163 | effectively no value.) | |
164 | ||
165 | This feature is especially useful in making macro definitions ``safe'' (so | |
166 | that they evaluate each operand exactly once). For example, the | |
167 | ``maximum'' function is commonly defined as a macro in standard C as | |
168 | follows: | |
169 | ||
170 | @example | |
171 | #define max(a,b) ((a) > (b) ? (a) : (b)) | |
172 | @end example | |
173 | ||
174 | @noindent | |
175 | @cindex side effects, macro argument | |
176 | But this definition computes either @var{a} or @var{b} twice, with bad | |
177 | results if the operand has side effects. In GNU C, if you know the | |
178 | type of the operands (here let's assume @code{int}), you can define | |
179 | the macro safely as follows: | |
180 | ||
181 | @example | |
182 | #define maxint(a,b) \ | |
183 | (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @}) | |
184 | @end example | |
185 | ||
186 | Embedded statements are not allowed in constant expressions, such as | |
187 | the value of an enumeration constant, the width of a bit field, or | |
188 | the initial value of a static variable. | |
189 | ||
190 | If you don't know the type of the operand, you can still do this, but you | |
191 | must use @code{typeof} (@pxref{Typeof}) or type naming (@pxref{Naming | |
192 | Types}). | |
193 | ||
b98e139b MM |
194 | Statement expressions are not supported fully in G++, and their fate |
195 | there is unclear. (It is possible that they will become fully supported | |
196 | at some point, or that they will be deprecated, or that the bugs that | |
197 | are present will continue to exist indefinitely.) Presently, statement | |
02f52e19 | 198 | expressions do not work well as default arguments. |
b98e139b MM |
199 | |
200 | In addition, there are semantic issues with statement-expressions in | |
201 | C++. If you try to use statement-expressions instead of inline | |
202 | functions in C++, you may be surprised at the way object destruction is | |
203 | handled. For example: | |
204 | ||
205 | @example | |
206 | #define foo(a) (@{int b = (a); b + 3; @}) | |
207 | @end example | |
208 | ||
209 | @noindent | |
210 | does not work the same way as: | |
211 | ||
212 | @example | |
54e1d3a6 | 213 | inline int foo(int a) @{ int b = a; return b + 3; @} |
b98e139b MM |
214 | @end example |
215 | ||
216 | @noindent | |
217 | In particular, if the expression passed into @code{foo} involves the | |
218 | creation of temporaries, the destructors for those temporaries will be | |
219 | run earlier in the case of the macro than in the case of the function. | |
220 | ||
221 | These considerations mean that it is probably a bad idea to use | |
222 | statement-expressions of this form in header files that are designed to | |
54e1d3a6 MM |
223 | work with C++. (Note that some versions of the GNU C Library contained |
224 | header files using statement-expression that lead to precisely this | |
225 | bug.) | |
b98e139b | 226 | |
c1f7febf RK |
227 | @node Local Labels |
228 | @section Locally Declared Labels | |
229 | @cindex local labels | |
230 | @cindex macros, local labels | |
231 | ||
232 | Each statement expression is a scope in which @dfn{local labels} can be | |
233 | declared. A local label is simply an identifier; you can jump to it | |
234 | with an ordinary @code{goto} statement, but only from within the | |
235 | statement expression it belongs to. | |
236 | ||
237 | A local label declaration looks like this: | |
238 | ||
239 | @example | |
240 | __label__ @var{label}; | |
241 | @end example | |
242 | ||
243 | @noindent | |
244 | or | |
245 | ||
246 | @example | |
247 | __label__ @var{label1}, @var{label2}, @dots{}; | |
248 | @end example | |
249 | ||
250 | Local label declarations must come at the beginning of the statement | |
251 | expression, right after the @samp{(@{}, before any ordinary | |
252 | declarations. | |
253 | ||
254 | The label declaration defines the label @emph{name}, but does not define | |
255 | the label itself. You must do this in the usual way, with | |
256 | @code{@var{label}:}, within the statements of the statement expression. | |
257 | ||
258 | The local label feature is useful because statement expressions are | |
259 | often used in macros. If the macro contains nested loops, a @code{goto} | |
260 | can be useful for breaking out of them. However, an ordinary label | |
261 | whose scope is the whole function cannot be used: if the macro can be | |
262 | expanded several times in one function, the label will be multiply | |
263 | defined in that function. A local label avoids this problem. For | |
264 | example: | |
265 | ||
266 | @example | |
267 | #define SEARCH(array, target) \ | |
310668e8 | 268 | (@{ \ |
c1f7febf RK |
269 | __label__ found; \ |
270 | typeof (target) _SEARCH_target = (target); \ | |
271 | typeof (*(array)) *_SEARCH_array = (array); \ | |
272 | int i, j; \ | |
273 | int value; \ | |
274 | for (i = 0; i < max; i++) \ | |
275 | for (j = 0; j < max; j++) \ | |
276 | if (_SEARCH_array[i][j] == _SEARCH_target) \ | |
310668e8 | 277 | @{ value = i; goto found; @} \ |
c1f7febf RK |
278 | value = -1; \ |
279 | found: \ | |
280 | value; \ | |
281 | @}) | |
282 | @end example | |
283 | ||
284 | @node Labels as Values | |
285 | @section Labels as Values | |
286 | @cindex labels as values | |
287 | @cindex computed gotos | |
288 | @cindex goto with computed label | |
289 | @cindex address of a label | |
290 | ||
291 | You can get the address of a label defined in the current function | |
292 | (or a containing function) with the unary operator @samp{&&}. The | |
293 | value has type @code{void *}. This value is a constant and can be used | |
294 | wherever a constant of that type is valid. For example: | |
295 | ||
296 | @example | |
297 | void *ptr; | |
298 | @dots{} | |
299 | ptr = &&foo; | |
300 | @end example | |
301 | ||
302 | To use these values, you need to be able to jump to one. This is done | |
303 | with the computed goto statement@footnote{The analogous feature in | |
304 | Fortran is called an assigned goto, but that name seems inappropriate in | |
305 | C, where one can do more than simply store label addresses in label | |
306 | variables.}, @code{goto *@var{exp};}. For example, | |
307 | ||
308 | @example | |
309 | goto *ptr; | |
310 | @end example | |
311 | ||
312 | @noindent | |
313 | Any expression of type @code{void *} is allowed. | |
314 | ||
315 | One way of using these constants is in initializing a static array that | |
316 | will serve as a jump table: | |
317 | ||
318 | @example | |
319 | static void *array[] = @{ &&foo, &&bar, &&hack @}; | |
320 | @end example | |
321 | ||
322 | Then you can select a label with indexing, like this: | |
323 | ||
324 | @example | |
325 | goto *array[i]; | |
326 | @end example | |
327 | ||
328 | @noindent | |
329 | Note that this does not check whether the subscript is in bounds---array | |
330 | indexing in C never does that. | |
331 | ||
332 | Such an array of label values serves a purpose much like that of the | |
333 | @code{switch} statement. The @code{switch} statement is cleaner, so | |
334 | use that rather than an array unless the problem does not fit a | |
335 | @code{switch} statement very well. | |
336 | ||
337 | Another use of label values is in an interpreter for threaded code. | |
338 | The labels within the interpreter function can be stored in the | |
339 | threaded code for super-fast dispatching. | |
340 | ||
02f52e19 | 341 | You may not use this mechanism to jump to code in a different function. |
47620e09 | 342 | If you do that, totally unpredictable things will happen. The best way to |
c1f7febf RK |
343 | avoid this is to store the label address only in automatic variables and |
344 | never pass it as an argument. | |
345 | ||
47620e09 RH |
346 | An alternate way to write the above example is |
347 | ||
348 | @example | |
310668e8 JM |
349 | static const int array[] = @{ &&foo - &&foo, &&bar - &&foo, |
350 | &&hack - &&foo @}; | |
47620e09 RH |
351 | goto *(&&foo + array[i]); |
352 | @end example | |
353 | ||
354 | @noindent | |
355 | This is more friendly to code living in shared libraries, as it reduces | |
356 | the number of dynamic relocations that are needed, and by consequence, | |
357 | allows the data to be read-only. | |
358 | ||
c1f7febf RK |
359 | @node Nested Functions |
360 | @section Nested Functions | |
361 | @cindex nested functions | |
362 | @cindex downward funargs | |
363 | @cindex thunks | |
364 | ||
365 | A @dfn{nested function} is a function defined inside another function. | |
366 | (Nested functions are not supported for GNU C++.) The nested function's | |
367 | name is local to the block where it is defined. For example, here we | |
368 | define a nested function named @code{square}, and call it twice: | |
369 | ||
370 | @example | |
371 | @group | |
372 | foo (double a, double b) | |
373 | @{ | |
374 | double square (double z) @{ return z * z; @} | |
375 | ||
376 | return square (a) + square (b); | |
377 | @} | |
378 | @end group | |
379 | @end example | |
380 | ||
381 | The nested function can access all the variables of the containing | |
382 | function that are visible at the point of its definition. This is | |
383 | called @dfn{lexical scoping}. For example, here we show a nested | |
384 | function which uses an inherited variable named @code{offset}: | |
385 | ||
386 | @example | |
387 | bar (int *array, int offset, int size) | |
388 | @{ | |
389 | int access (int *array, int index) | |
390 | @{ return array[index + offset]; @} | |
391 | int i; | |
392 | @dots{} | |
393 | for (i = 0; i < size; i++) | |
394 | @dots{} access (array, i) @dots{} | |
395 | @} | |
396 | @end example | |
397 | ||
398 | Nested function definitions are permitted within functions in the places | |
399 | where variable definitions are allowed; that is, in any block, before | |
400 | the first statement in the block. | |
401 | ||
402 | It is possible to call the nested function from outside the scope of its | |
403 | name by storing its address or passing the address to another function: | |
404 | ||
405 | @example | |
406 | hack (int *array, int size) | |
407 | @{ | |
408 | void store (int index, int value) | |
409 | @{ array[index] = value; @} | |
410 | ||
411 | intermediate (store, size); | |
412 | @} | |
413 | @end example | |
414 | ||
415 | Here, the function @code{intermediate} receives the address of | |
416 | @code{store} as an argument. If @code{intermediate} calls @code{store}, | |
417 | the arguments given to @code{store} are used to store into @code{array}. | |
418 | But this technique works only so long as the containing function | |
419 | (@code{hack}, in this example) does not exit. | |
420 | ||
421 | If you try to call the nested function through its address after the | |
422 | containing function has exited, all hell will break loose. If you try | |
423 | to call it after a containing scope level has exited, and if it refers | |
424 | to some of the variables that are no longer in scope, you may be lucky, | |
425 | but it's not wise to take the risk. If, however, the nested function | |
426 | does not refer to anything that has gone out of scope, you should be | |
427 | safe. | |
428 | ||
429 | GNU CC implements taking the address of a nested function using a | |
674032e2 | 430 | technique called @dfn{trampolines}. A paper describing them is |
9734e80c | 431 | available as @uref{http://people.debian.org/~karlheg/Usenix88-lexic.pdf}. |
c1f7febf RK |
432 | |
433 | A nested function can jump to a label inherited from a containing | |
434 | function, provided the label was explicitly declared in the containing | |
435 | function (@pxref{Local Labels}). Such a jump returns instantly to the | |
436 | containing function, exiting the nested function which did the | |
437 | @code{goto} and any intermediate functions as well. Here is an example: | |
438 | ||
439 | @example | |
440 | @group | |
441 | bar (int *array, int offset, int size) | |
442 | @{ | |
443 | __label__ failure; | |
444 | int access (int *array, int index) | |
445 | @{ | |
446 | if (index > size) | |
447 | goto failure; | |
448 | return array[index + offset]; | |
449 | @} | |
450 | int i; | |
451 | @dots{} | |
452 | for (i = 0; i < size; i++) | |
453 | @dots{} access (array, i) @dots{} | |
454 | @dots{} | |
455 | return 0; | |
456 | ||
457 | /* @r{Control comes here from @code{access} | |
458 | if it detects an error.} */ | |
459 | failure: | |
460 | return -1; | |
461 | @} | |
462 | @end group | |
463 | @end example | |
464 | ||
465 | A nested function always has internal linkage. Declaring one with | |
466 | @code{extern} is erroneous. If you need to declare the nested function | |
467 | before its definition, use @code{auto} (which is otherwise meaningless | |
468 | for function declarations). | |
469 | ||
470 | @example | |
471 | bar (int *array, int offset, int size) | |
472 | @{ | |
473 | __label__ failure; | |
474 | auto int access (int *, int); | |
475 | @dots{} | |
476 | int access (int *array, int index) | |
477 | @{ | |
478 | if (index > size) | |
479 | goto failure; | |
480 | return array[index + offset]; | |
481 | @} | |
482 | @dots{} | |
483 | @} | |
484 | @end example | |
485 | ||
486 | @node Constructing Calls | |
487 | @section Constructing Function Calls | |
488 | @cindex constructing calls | |
489 | @cindex forwarding calls | |
490 | ||
491 | Using the built-in functions described below, you can record | |
492 | the arguments a function received, and call another function | |
493 | with the same arguments, without knowing the number or types | |
494 | of the arguments. | |
495 | ||
496 | You can also record the return value of that function call, | |
497 | and later return that value, without knowing what data type | |
498 | the function tried to return (as long as your caller expects | |
499 | that data type). | |
500 | ||
501 | @table @code | |
502 | @findex __builtin_apply_args | |
503 | @item __builtin_apply_args () | |
504 | This built-in function returns a pointer of type @code{void *} to data | |
505 | describing how to perform a call with the same arguments as were passed | |
506 | to the current function. | |
507 | ||
508 | The function saves the arg pointer register, structure value address, | |
509 | and all registers that might be used to pass arguments to a function | |
510 | into a block of memory allocated on the stack. Then it returns the | |
511 | address of that block. | |
512 | ||
513 | @findex __builtin_apply | |
514 | @item __builtin_apply (@var{function}, @var{arguments}, @var{size}) | |
515 | This built-in function invokes @var{function} (type @code{void (*)()}) | |
516 | with a copy of the parameters described by @var{arguments} (type | |
517 | @code{void *}) and @var{size} (type @code{int}). | |
518 | ||
519 | The value of @var{arguments} should be the value returned by | |
520 | @code{__builtin_apply_args}. The argument @var{size} specifies the size | |
521 | of the stack argument data, in bytes. | |
522 | ||
523 | This function returns a pointer of type @code{void *} to data describing | |
524 | how to return whatever value was returned by @var{function}. The data | |
525 | is saved in a block of memory allocated on the stack. | |
526 | ||
527 | It is not always simple to compute the proper value for @var{size}. The | |
528 | value is used by @code{__builtin_apply} to compute the amount of data | |
529 | that should be pushed on the stack and copied from the incoming argument | |
530 | area. | |
531 | ||
532 | @findex __builtin_return | |
533 | @item __builtin_return (@var{result}) | |
534 | This built-in function returns the value described by @var{result} from | |
535 | the containing function. You should specify, for @var{result}, a value | |
536 | returned by @code{__builtin_apply}. | |
537 | @end table | |
538 | ||
539 | @node Naming Types | |
540 | @section Naming an Expression's Type | |
541 | @cindex naming types | |
542 | ||
543 | You can give a name to the type of an expression using a @code{typedef} | |
544 | declaration with an initializer. Here is how to define @var{name} as a | |
545 | type name for the type of @var{exp}: | |
546 | ||
547 | @example | |
548 | typedef @var{name} = @var{exp}; | |
549 | @end example | |
550 | ||
551 | This is useful in conjunction with the statements-within-expressions | |
552 | feature. Here is how the two together can be used to define a safe | |
553 | ``maximum'' macro that operates on any arithmetic type: | |
554 | ||
555 | @example | |
556 | #define max(a,b) \ | |
557 | (@{typedef _ta = (a), _tb = (b); \ | |
558 | _ta _a = (a); _tb _b = (b); \ | |
559 | _a > _b ? _a : _b; @}) | |
560 | @end example | |
561 | ||
562 | @cindex underscores in variables in macros | |
563 | @cindex @samp{_} in variables in macros | |
564 | @cindex local variables in macros | |
565 | @cindex variables, local, in macros | |
566 | @cindex macros, local variables in | |
567 | ||
568 | The reason for using names that start with underscores for the local | |
569 | variables is to avoid conflicts with variable names that occur within the | |
570 | expressions that are substituted for @code{a} and @code{b}. Eventually we | |
571 | hope to design a new form of declaration syntax that allows you to declare | |
572 | variables whose scopes start only after their initializers; this will be a | |
573 | more reliable way to prevent such conflicts. | |
574 | ||
575 | @node Typeof | |
576 | @section Referring to a Type with @code{typeof} | |
577 | @findex typeof | |
578 | @findex sizeof | |
579 | @cindex macros, types of arguments | |
580 | ||
581 | Another way to refer to the type of an expression is with @code{typeof}. | |
582 | The syntax of using of this keyword looks like @code{sizeof}, but the | |
583 | construct acts semantically like a type name defined with @code{typedef}. | |
584 | ||
585 | There are two ways of writing the argument to @code{typeof}: with an | |
586 | expression or with a type. Here is an example with an expression: | |
587 | ||
588 | @example | |
589 | typeof (x[0](1)) | |
590 | @end example | |
591 | ||
592 | @noindent | |
89aed483 JM |
593 | This assumes that @code{x} is an array of pointers to functions; |
594 | the type described is that of the values of the functions. | |
c1f7febf RK |
595 | |
596 | Here is an example with a typename as the argument: | |
597 | ||
598 | @example | |
599 | typeof (int *) | |
600 | @end example | |
601 | ||
602 | @noindent | |
603 | Here the type described is that of pointers to @code{int}. | |
604 | ||
5490d604 | 605 | If you are writing a header file that must work when included in ISO C |
c1f7febf RK |
606 | programs, write @code{__typeof__} instead of @code{typeof}. |
607 | @xref{Alternate Keywords}. | |
608 | ||
609 | A @code{typeof}-construct can be used anywhere a typedef name could be | |
610 | used. For example, you can use it in a declaration, in a cast, or inside | |
611 | of @code{sizeof} or @code{typeof}. | |
612 | ||
613 | @itemize @bullet | |
614 | @item | |
615 | This declares @code{y} with the type of what @code{x} points to. | |
616 | ||
617 | @example | |
618 | typeof (*x) y; | |
619 | @end example | |
620 | ||
621 | @item | |
622 | This declares @code{y} as an array of such values. | |
623 | ||
624 | @example | |
625 | typeof (*x) y[4]; | |
626 | @end example | |
627 | ||
628 | @item | |
629 | This declares @code{y} as an array of pointers to characters: | |
630 | ||
631 | @example | |
632 | typeof (typeof (char *)[4]) y; | |
633 | @end example | |
634 | ||
635 | @noindent | |
636 | It is equivalent to the following traditional C declaration: | |
637 | ||
638 | @example | |
639 | char *y[4]; | |
640 | @end example | |
641 | ||
642 | To see the meaning of the declaration using @code{typeof}, and why it | |
643 | might be a useful way to write, let's rewrite it with these macros: | |
644 | ||
645 | @example | |
646 | #define pointer(T) typeof(T *) | |
647 | #define array(T, N) typeof(T [N]) | |
648 | @end example | |
649 | ||
650 | @noindent | |
651 | Now the declaration can be rewritten this way: | |
652 | ||
653 | @example | |
654 | array (pointer (char), 4) y; | |
655 | @end example | |
656 | ||
657 | @noindent | |
658 | Thus, @code{array (pointer (char), 4)} is the type of arrays of 4 | |
659 | pointers to @code{char}. | |
660 | @end itemize | |
661 | ||
662 | @node Lvalues | |
663 | @section Generalized Lvalues | |
664 | @cindex compound expressions as lvalues | |
665 | @cindex expressions, compound, as lvalues | |
666 | @cindex conditional expressions as lvalues | |
667 | @cindex expressions, conditional, as lvalues | |
668 | @cindex casts as lvalues | |
669 | @cindex generalized lvalues | |
670 | @cindex lvalues, generalized | |
671 | @cindex extensions, @code{?:} | |
672 | @cindex @code{?:} extensions | |
673 | Compound expressions, conditional expressions and casts are allowed as | |
674 | lvalues provided their operands are lvalues. This means that you can take | |
675 | their addresses or store values into them. | |
676 | ||
677 | Standard C++ allows compound expressions and conditional expressions as | |
678 | lvalues, and permits casts to reference type, so use of this extension | |
679 | is deprecated for C++ code. | |
680 | ||
681 | For example, a compound expression can be assigned, provided the last | |
682 | expression in the sequence is an lvalue. These two expressions are | |
683 | equivalent: | |
684 | ||
685 | @example | |
686 | (a, b) += 5 | |
687 | a, (b += 5) | |
688 | @end example | |
689 | ||
690 | Similarly, the address of the compound expression can be taken. These two | |
691 | expressions are equivalent: | |
692 | ||
693 | @example | |
694 | &(a, b) | |
695 | a, &b | |
696 | @end example | |
697 | ||
698 | A conditional expression is a valid lvalue if its type is not void and the | |
699 | true and false branches are both valid lvalues. For example, these two | |
700 | expressions are equivalent: | |
701 | ||
702 | @example | |
703 | (a ? b : c) = 5 | |
704 | (a ? b = 5 : (c = 5)) | |
705 | @end example | |
706 | ||
707 | A cast is a valid lvalue if its operand is an lvalue. A simple | |
708 | assignment whose left-hand side is a cast works by converting the | |
709 | right-hand side first to the specified type, then to the type of the | |
710 | inner left-hand side expression. After this is stored, the value is | |
711 | converted back to the specified type to become the value of the | |
712 | assignment. Thus, if @code{a} has type @code{char *}, the following two | |
713 | expressions are equivalent: | |
714 | ||
715 | @example | |
716 | (int)a = 5 | |
717 | (int)(a = (char *)(int)5) | |
718 | @end example | |
719 | ||
720 | An assignment-with-arithmetic operation such as @samp{+=} applied to a cast | |
721 | performs the arithmetic using the type resulting from the cast, and then | |
722 | continues as in the previous case. Therefore, these two expressions are | |
723 | equivalent: | |
724 | ||
725 | @example | |
726 | (int)a += 5 | |
727 | (int)(a = (char *)(int) ((int)a + 5)) | |
728 | @end example | |
729 | ||
730 | You cannot take the address of an lvalue cast, because the use of its | |
731 | address would not work out coherently. Suppose that @code{&(int)f} were | |
732 | permitted, where @code{f} has type @code{float}. Then the following | |
733 | statement would try to store an integer bit-pattern where a floating | |
734 | point number belongs: | |
735 | ||
736 | @example | |
737 | *&(int)f = 1; | |
738 | @end example | |
739 | ||
740 | This is quite different from what @code{(int)f = 1} would do---that | |
741 | would convert 1 to floating point and store it. Rather than cause this | |
742 | inconsistency, we think it is better to prohibit use of @samp{&} on a cast. | |
743 | ||
744 | If you really do want an @code{int *} pointer with the address of | |
745 | @code{f}, you can simply write @code{(int *)&f}. | |
746 | ||
747 | @node Conditionals | |
748 | @section Conditionals with Omitted Operands | |
749 | @cindex conditional expressions, extensions | |
750 | @cindex omitted middle-operands | |
751 | @cindex middle-operands, omitted | |
752 | @cindex extensions, @code{?:} | |
753 | @cindex @code{?:} extensions | |
754 | ||
755 | The middle operand in a conditional expression may be omitted. Then | |
756 | if the first operand is nonzero, its value is the value of the conditional | |
757 | expression. | |
758 | ||
759 | Therefore, the expression | |
760 | ||
761 | @example | |
762 | x ? : y | |
763 | @end example | |
764 | ||
765 | @noindent | |
766 | has the value of @code{x} if that is nonzero; otherwise, the value of | |
767 | @code{y}. | |
768 | ||
769 | This example is perfectly equivalent to | |
770 | ||
771 | @example | |
772 | x ? x : y | |
773 | @end example | |
774 | ||
775 | @cindex side effect in ?: | |
776 | @cindex ?: side effect | |
777 | @noindent | |
778 | In this simple case, the ability to omit the middle operand is not | |
779 | especially useful. When it becomes useful is when the first operand does, | |
780 | or may (if it is a macro argument), contain a side effect. Then repeating | |
781 | the operand in the middle would perform the side effect twice. Omitting | |
782 | the middle operand uses the value already computed without the undesirable | |
783 | effects of recomputing it. | |
784 | ||
785 | @node Long Long | |
786 | @section Double-Word Integers | |
787 | @cindex @code{long long} data types | |
788 | @cindex double-word arithmetic | |
789 | @cindex multiprecision arithmetic | |
4b404517 JM |
790 | @cindex @code{LL} integer suffix |
791 | @cindex @code{ULL} integer suffix | |
c1f7febf | 792 | |
4b404517 JM |
793 | ISO C99 supports data types for integers that are at least 64 bits wide, |
794 | and as an extension GCC supports them in C89 mode and in C++. | |
795 | Simply write @code{long long int} for a signed integer, or | |
c1f7febf RK |
796 | @code{unsigned long long int} for an unsigned integer. To make an |
797 | integer constant of type @code{long long int}, add the suffix @code{LL} | |
798 | to the integer. To make an integer constant of type @code{unsigned long | |
799 | long int}, add the suffix @code{ULL} to the integer. | |
800 | ||
801 | You can use these types in arithmetic like any other integer types. | |
802 | Addition, subtraction, and bitwise boolean operations on these types | |
803 | are open-coded on all types of machines. Multiplication is open-coded | |
804 | if the machine supports fullword-to-doubleword a widening multiply | |
805 | instruction. Division and shifts are open-coded only on machines that | |
806 | provide special support. The operations that are not open-coded use | |
807 | special library routines that come with GNU CC. | |
808 | ||
809 | There may be pitfalls when you use @code{long long} types for function | |
810 | arguments, unless you declare function prototypes. If a function | |
811 | expects type @code{int} for its argument, and you pass a value of type | |
812 | @code{long long int}, confusion will result because the caller and the | |
813 | subroutine will disagree about the number of bytes for the argument. | |
814 | Likewise, if the function expects @code{long long int} and you pass | |
815 | @code{int}. The best way to avoid such problems is to use prototypes. | |
816 | ||
817 | @node Complex | |
818 | @section Complex Numbers | |
819 | @cindex complex numbers | |
4b404517 JM |
820 | @cindex @code{_Complex} keyword |
821 | @cindex @code{__complex__} keyword | |
c1f7febf | 822 | |
4b404517 JM |
823 | ISO C99 supports complex floating data types, and as an extension GCC |
824 | supports them in C89 mode and in C++, and supports complex integer data | |
825 | types which are not part of ISO C99. You can declare complex types | |
826 | using the keyword @code{_Complex}. As an extension, the older GNU | |
827 | keyword @code{__complex__} is also supported. | |
c1f7febf | 828 | |
4b404517 | 829 | For example, @samp{_Complex double x;} declares @code{x} as a |
c1f7febf | 830 | variable whose real part and imaginary part are both of type |
4b404517 | 831 | @code{double}. @samp{_Complex short int y;} declares @code{y} to |
c1f7febf RK |
832 | have real and imaginary parts of type @code{short int}; this is not |
833 | likely to be useful, but it shows that the set of complex types is | |
834 | complete. | |
835 | ||
836 | To write a constant with a complex data type, use the suffix @samp{i} or | |
837 | @samp{j} (either one; they are equivalent). For example, @code{2.5fi} | |
4b404517 JM |
838 | has type @code{_Complex float} and @code{3i} has type |
839 | @code{_Complex int}. Such a constant always has a pure imaginary | |
c1f7febf | 840 | value, but you can form any complex value you like by adding one to a |
4b404517 JM |
841 | real constant. This is a GNU extension; if you have an ISO C99 |
842 | conforming C library (such as GNU libc), and want to construct complex | |
843 | constants of floating type, you should include @code{<complex.h>} and | |
844 | use the macros @code{I} or @code{_Complex_I} instead. | |
c1f7febf | 845 | |
4b404517 JM |
846 | @cindex @code{__real__} keyword |
847 | @cindex @code{__imag__} keyword | |
c1f7febf RK |
848 | To extract the real part of a complex-valued expression @var{exp}, write |
849 | @code{__real__ @var{exp}}. Likewise, use @code{__imag__} to | |
4b404517 JM |
850 | extract the imaginary part. This is a GNU extension; for values of |
851 | floating type, you should use the ISO C99 functions @code{crealf}, | |
852 | @code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and | |
853 | @code{cimagl}, declared in @code{<complex.h>} and also provided as | |
854 | builtin functions by GCC. | |
c1f7febf | 855 | |
4b404517 | 856 | @cindex complex conjugation |
c1f7febf | 857 | The operator @samp{~} performs complex conjugation when used on a value |
4b404517 JM |
858 | with a complex type. This is a GNU extension; for values of |
859 | floating type, you should use the ISO C99 functions @code{conjf}, | |
860 | @code{conj} and @code{conjl}, declared in @code{<complex.h>} and also | |
861 | provided as builtin functions by GCC. | |
c1f7febf RK |
862 | |
863 | GNU CC can allocate complex automatic variables in a noncontiguous | |
864 | fashion; it's even possible for the real part to be in a register while | |
865 | the imaginary part is on the stack (or vice-versa). None of the | |
866 | supported debugging info formats has a way to represent noncontiguous | |
867 | allocation like this, so GNU CC describes a noncontiguous complex | |
868 | variable as if it were two separate variables of noncomplex type. | |
869 | If the variable's actual name is @code{foo}, the two fictitious | |
870 | variables are named @code{foo$real} and @code{foo$imag}. You can | |
871 | examine and set these two fictitious variables with your debugger. | |
872 | ||
873 | A future version of GDB will know how to recognize such pairs and treat | |
874 | them as a single variable with a complex type. | |
875 | ||
6f4d7222 | 876 | @node Hex Floats |
6b42b9ea UD |
877 | @section Hex Floats |
878 | @cindex hex floats | |
c5c76735 | 879 | |
4b404517 | 880 | ISO C99 supports floating-point numbers written not only in the usual |
6f4d7222 | 881 | decimal notation, such as @code{1.55e1}, but also numbers such as |
4b404517 JM |
882 | @code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC |
883 | supports this in C89 mode (except in some cases when strictly | |
884 | conforming) and in C++. In that format the | |
6f4d7222 UD |
885 | @code{0x} hex introducer and the @code{p} or @code{P} exponent field are |
886 | mandatory. The exponent is a decimal number that indicates the power of | |
3b7a2e58 | 887 | 2 by which the significant part will be multiplied. Thus @code{0x1.f} is |
6f4d7222 UD |
888 | 1 15/16, @code{p3} multiplies it by 8, and the value of @code{0x1.fp3} |
889 | is the same as @code{1.55e1}. | |
890 | ||
891 | Unlike for floating-point numbers in the decimal notation the exponent | |
892 | is always required in the hexadecimal notation. Otherwise the compiler | |
893 | would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This | |
894 | could mean @code{1.0f} or @code{1.9375} since @code{f} is also the | |
895 | extension for floating-point constants of type @code{float}. | |
896 | ||
c1f7febf RK |
897 | @node Zero Length |
898 | @section Arrays of Length Zero | |
899 | @cindex arrays of length zero | |
900 | @cindex zero-length arrays | |
901 | @cindex length-zero arrays | |
ffc5c6a9 | 902 | @cindex flexible array members |
c1f7febf | 903 | |
584ef5fe RH |
904 | Zero-length arrays are allowed in GNU C. They are very useful as the |
905 | last element of a structure which is really a header for a variable-length | |
c1f7febf RK |
906 | object: |
907 | ||
908 | @example | |
909 | struct line @{ | |
910 | int length; | |
911 | char contents[0]; | |
912 | @}; | |
913 | ||
584ef5fe RH |
914 | struct line *thisline = (struct line *) |
915 | malloc (sizeof (struct line) + this_length); | |
916 | thisline->length = this_length; | |
c1f7febf RK |
917 | @end example |
918 | ||
a25f1211 | 919 | In ISO C89, you would have to give @code{contents} a length of 1, which |
c1f7febf RK |
920 | means either you waste space or complicate the argument to @code{malloc}. |
921 | ||
02f52e19 | 922 | In ISO C99, you would use a @dfn{flexible array member}, which is |
584ef5fe RH |
923 | slightly different in syntax and semantics: |
924 | ||
925 | @itemize @bullet | |
926 | @item | |
927 | Flexible array members are written as @code{contents[]} without | |
928 | the @code{0}. | |
929 | ||
930 | @item | |
931 | Flexible array members have incomplete type, and so the @code{sizeof} | |
932 | operator may not be applied. As a quirk of the original implementation | |
933 | of zero-length arrays, @code{sizeof} evaluates to zero. | |
934 | ||
935 | @item | |
936 | Flexible array members may only appear as the last member of a | |
02f52e19 | 937 | @code{struct} that is otherwise non-empty. GCC currently allows |
584ef5fe RH |
938 | zero-length arrays anywhere. You may encounter problems, however, |
939 | defining structures containing only a zero-length array. Such usage | |
940 | is deprecated, and we recommend using zero-length arrays only in | |
941 | places in which flexible array members would be allowed. | |
ffc5c6a9 | 942 | @end itemize |
a25f1211 | 943 | |
ffc5c6a9 RH |
944 | GCC versions before 3.0 allowed zero-length arrays to be statically |
945 | initialized. In addition to those cases that were useful, it also | |
946 | allowed initializations in situations that would corrupt later data. | |
947 | Non-empty initialization of zero-length arrays is now deprecated. | |
948 | ||
949 | Instead GCC allows static initialization of flexible array members. | |
950 | This is equivalent to defining a new structure containing the original | |
951 | structure followed by an array of sufficient size to contain the data. | |
952 | I.e. in the following, @code{f1} is constructed as if it were declared | |
953 | like @code{f2}. | |
a25f1211 RH |
954 | |
955 | @example | |
ffc5c6a9 RH |
956 | struct f1 @{ |
957 | int x; int y[]; | |
958 | @} f1 = @{ 1, @{ 2, 3, 4 @} @}; | |
959 | ||
960 | struct f2 @{ | |
961 | struct f1 f1; int data[3]; | |
962 | @} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @}; | |
963 | @end example | |
584ef5fe | 964 | |
ffc5c6a9 RH |
965 | @noindent |
966 | The convenience of this extension is that @code{f1} has the desired | |
967 | type, eliminating the need to consistently refer to @code{f2.f1}. | |
968 | ||
969 | This has symmetry with normal static arrays, in that an array of | |
970 | unknown size is also written with @code{[]}. | |
a25f1211 | 971 | |
ffc5c6a9 RH |
972 | Of course, this extension only makes sense if the extra data comes at |
973 | the end of a top-level object, as otherwise we would be overwriting | |
974 | data at subsequent offsets. To avoid undue complication and confusion | |
975 | with initialization of deeply nested arrays, we simply disallow any | |
976 | non-empty initialization except when the structure is the top-level | |
977 | object. For example: | |
584ef5fe | 978 | |
ffc5c6a9 RH |
979 | @example |
980 | struct foo @{ int x; int y[]; @}; | |
981 | struct bar @{ struct foo z; @}; | |
982 | ||
983 | struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // Legal. | |
984 | struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // Illegal. | |
985 | struct bar c = @{ @{ 1, @{ @} @} @}; // Legal. | |
986 | struct foo d[1] = @{ @{ 1 @{ 2, 3, 4 @} @} @}; // Illegal. | |
a25f1211 | 987 | @end example |
4b606faf | 988 | |
c1f7febf RK |
989 | @node Variable Length |
990 | @section Arrays of Variable Length | |
991 | @cindex variable-length arrays | |
992 | @cindex arrays of variable length | |
4b404517 | 993 | @cindex VLAs |
c1f7febf | 994 | |
4b404517 JM |
995 | Variable-length automatic arrays are allowed in ISO C99, and as an |
996 | extension GCC accepts them in C89 mode and in C++. (However, GCC's | |
997 | implementation of variable-length arrays does not yet conform in detail | |
998 | to the ISO C99 standard.) These arrays are | |
c1f7febf RK |
999 | declared like any other automatic arrays, but with a length that is not |
1000 | a constant expression. The storage is allocated at the point of | |
1001 | declaration and deallocated when the brace-level is exited. For | |
1002 | example: | |
1003 | ||
1004 | @example | |
1005 | FILE * | |
1006 | concat_fopen (char *s1, char *s2, char *mode) | |
1007 | @{ | |
1008 | char str[strlen (s1) + strlen (s2) + 1]; | |
1009 | strcpy (str, s1); | |
1010 | strcat (str, s2); | |
1011 | return fopen (str, mode); | |
1012 | @} | |
1013 | @end example | |
1014 | ||
1015 | @cindex scope of a variable length array | |
1016 | @cindex variable-length array scope | |
1017 | @cindex deallocating variable length arrays | |
1018 | Jumping or breaking out of the scope of the array name deallocates the | |
1019 | storage. Jumping into the scope is not allowed; you get an error | |
1020 | message for it. | |
1021 | ||
1022 | @cindex @code{alloca} vs variable-length arrays | |
1023 | You can use the function @code{alloca} to get an effect much like | |
1024 | variable-length arrays. The function @code{alloca} is available in | |
1025 | many other C implementations (but not in all). On the other hand, | |
1026 | variable-length arrays are more elegant. | |
1027 | ||
1028 | There are other differences between these two methods. Space allocated | |
1029 | with @code{alloca} exists until the containing @emph{function} returns. | |
1030 | The space for a variable-length array is deallocated as soon as the array | |
1031 | name's scope ends. (If you use both variable-length arrays and | |
1032 | @code{alloca} in the same function, deallocation of a variable-length array | |
1033 | will also deallocate anything more recently allocated with @code{alloca}.) | |
1034 | ||
1035 | You can also use variable-length arrays as arguments to functions: | |
1036 | ||
1037 | @example | |
1038 | struct entry | |
1039 | tester (int len, char data[len][len]) | |
1040 | @{ | |
1041 | @dots{} | |
1042 | @} | |
1043 | @end example | |
1044 | ||
1045 | The length of an array is computed once when the storage is allocated | |
1046 | and is remembered for the scope of the array in case you access it with | |
1047 | @code{sizeof}. | |
1048 | ||
1049 | If you want to pass the array first and the length afterward, you can | |
1050 | use a forward declaration in the parameter list---another GNU extension. | |
1051 | ||
1052 | @example | |
1053 | struct entry | |
1054 | tester (int len; char data[len][len], int len) | |
1055 | @{ | |
1056 | @dots{} | |
1057 | @} | |
1058 | @end example | |
1059 | ||
1060 | @cindex parameter forward declaration | |
1061 | The @samp{int len} before the semicolon is a @dfn{parameter forward | |
1062 | declaration}, and it serves the purpose of making the name @code{len} | |
1063 | known when the declaration of @code{data} is parsed. | |
1064 | ||
1065 | You can write any number of such parameter forward declarations in the | |
1066 | parameter list. They can be separated by commas or semicolons, but the | |
1067 | last one must end with a semicolon, which is followed by the ``real'' | |
1068 | parameter declarations. Each forward declaration must match a ``real'' | |
4b404517 JM |
1069 | declaration in parameter name and data type. ISO C99 does not support |
1070 | parameter forward declarations. | |
c1f7febf | 1071 | |
ccd96f0a NB |
1072 | @node Variadic Macros |
1073 | @section Macros with a Variable Number of Arguments. | |
c1f7febf RK |
1074 | @cindex variable number of arguments |
1075 | @cindex macro with variable arguments | |
1076 | @cindex rest argument (in macro) | |
ccd96f0a | 1077 | @cindex variadic macros |
c1f7febf | 1078 | |
ccd96f0a NB |
1079 | In the ISO C standard of 1999, a macro can be declared to accept a |
1080 | variable number of arguments much as a function can. The syntax for | |
1081 | defining the macro is similar to that of a function. Here is an | |
1082 | example: | |
c1f7febf RK |
1083 | |
1084 | @example | |
ccd96f0a | 1085 | #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__) |
c1f7febf RK |
1086 | @end example |
1087 | ||
ccd96f0a NB |
1088 | Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of |
1089 | such a macro, it represents the zero or more tokens until the closing | |
1090 | parenthesis that ends the invocation, including any commas. This set of | |
1091 | tokens replaces the identifier @code{__VA_ARGS__} in the macro body | |
1092 | wherever it appears. See the CPP manual for more information. | |
1093 | ||
1094 | GCC has long supported variadic macros, and used a different syntax that | |
1095 | allowed you to give a name to the variable arguments just like any other | |
1096 | argument. Here is an example: | |
c1f7febf RK |
1097 | |
1098 | @example | |
ccd96f0a | 1099 | #define debug(format, args...) fprintf (stderr, format, args) |
c1f7febf RK |
1100 | @end example |
1101 | ||
ccd96f0a NB |
1102 | This is in all ways equivalent to the ISO C example above, but arguably |
1103 | more readable and descriptive. | |
c1f7febf | 1104 | |
ccd96f0a NB |
1105 | GNU CPP has two further variadic macro extensions, and permits them to |
1106 | be used with either of the above forms of macro definition. | |
1107 | ||
1108 | In standard C, you are not allowed to leave the variable argument out | |
1109 | entirely; but you are allowed to pass an empty argument. For example, | |
1110 | this invocation is invalid in ISO C, because there is no comma after | |
1111 | the string: | |
c1f7febf RK |
1112 | |
1113 | @example | |
ccd96f0a | 1114 | debug ("A message") |
c1f7febf RK |
1115 | @end example |
1116 | ||
ccd96f0a NB |
1117 | GNU CPP permits you to completely omit the variable arguments in this |
1118 | way. In the above examples, the compiler would complain, though since | |
1119 | the expansion of the macro still has the extra comma after the format | |
1120 | string. | |
1121 | ||
1122 | To help solve this problem, CPP behaves specially for variable arguments | |
1123 | used with the token paste operator, @samp{##}. If instead you write | |
c1f7febf RK |
1124 | |
1125 | @example | |
ccd96f0a | 1126 | #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__) |
c1f7febf RK |
1127 | @end example |
1128 | ||
ccd96f0a NB |
1129 | and if the variable arguments are omitted or empty, the @samp{##} |
1130 | operator causes the preprocessor to remove the comma before it. If you | |
1131 | do provide some variable arguments in your macro invocation, GNU CPP | |
1132 | does not complain about the paste operation and instead places the | |
1133 | variable arguments after the comma. Just like any other pasted macro | |
1134 | argument, these arguments are not macro expanded. | |
1135 | ||
1136 | @node Escaped Newlines | |
1137 | @section Slightly Looser Rules for Escaped Newlines | |
1138 | @cindex escaped newlines | |
1139 | @cindex newlines (escaped) | |
1140 | ||
1141 | Recently, the non-traditional preprocessor has relaxed its treatment of | |
1142 | escaped newlines. Previously, the newline had to immediately follow a | |
1143 | backslash. The current implementation allows whitespace in the form of | |
1144 | spaces, horizontal and vertical tabs, and form feeds between the | |
1145 | backslash and the subsequent newline. The preprocessor issues a | |
1146 | warning, but treats it as a valid escaped newline and combines the two | |
1147 | lines to form a single logical line. This works within comments and | |
1148 | tokens, including multi-line strings, as well as between tokens. | |
1149 | Comments are @emph{not} treated as whitespace for the purposes of this | |
1150 | relaxation, since they have not yet been replaced with spaces. | |
1151 | ||
1152 | @node Multi-line Strings | |
1153 | @section String Literals with Embedded Newlines | |
1154 | @cindex multi-line string literals | |
1155 | ||
1156 | As an extension, GNU CPP permits string literals to cross multiple lines | |
1157 | without escaping the embedded newlines. Each embedded newline is | |
1158 | replaced with a single @samp{\n} character in the resulting string | |
1159 | literal, regardless of what form the newline took originally. | |
1160 | ||
1161 | CPP currently allows such strings in directives as well (other than the | |
1162 | @samp{#include} family). This is deprecated and will eventually be | |
1163 | removed. | |
c1f7febf RK |
1164 | |
1165 | @node Subscripting | |
1166 | @section Non-Lvalue Arrays May Have Subscripts | |
1167 | @cindex subscripting | |
1168 | @cindex arrays, non-lvalue | |
1169 | ||
1170 | @cindex subscripting and function values | |
1171 | Subscripting is allowed on arrays that are not lvalues, even though the | |
4b404517 JM |
1172 | unary @samp{&} operator is not. (In ISO C99, both are allowed (though |
1173 | the array may not be used after the next sequence point), but this ISO | |
1174 | C99 feature is not yet fully supported in GCC.) For example, | |
1175 | this is valid in GNU C though not valid in C89: | |
c1f7febf RK |
1176 | |
1177 | @example | |
1178 | @group | |
1179 | struct foo @{int a[4];@}; | |
1180 | ||
1181 | struct foo f(); | |
1182 | ||
1183 | bar (int index) | |
1184 | @{ | |
1185 | return f().a[index]; | |
1186 | @} | |
1187 | @end group | |
1188 | @end example | |
1189 | ||
1190 | @node Pointer Arith | |
1191 | @section Arithmetic on @code{void}- and Function-Pointers | |
1192 | @cindex void pointers, arithmetic | |
1193 | @cindex void, size of pointer to | |
1194 | @cindex function pointers, arithmetic | |
1195 | @cindex function, size of pointer to | |
1196 | ||
1197 | In GNU C, addition and subtraction operations are supported on pointers to | |
1198 | @code{void} and on pointers to functions. This is done by treating the | |
1199 | size of a @code{void} or of a function as 1. | |
1200 | ||
1201 | A consequence of this is that @code{sizeof} is also allowed on @code{void} | |
1202 | and on function types, and returns 1. | |
1203 | ||
1204 | The option @samp{-Wpointer-arith} requests a warning if these extensions | |
1205 | are used. | |
1206 | ||
1207 | @node Initializers | |
1208 | @section Non-Constant Initializers | |
1209 | @cindex initializers, non-constant | |
1210 | @cindex non-constant initializers | |
1211 | ||
4b404517 | 1212 | As in standard C++ and ISO C99, the elements of an aggregate initializer for an |
c1f7febf RK |
1213 | automatic variable are not required to be constant expressions in GNU C. |
1214 | Here is an example of an initializer with run-time varying elements: | |
1215 | ||
1216 | @example | |
1217 | foo (float f, float g) | |
1218 | @{ | |
1219 | float beat_freqs[2] = @{ f-g, f+g @}; | |
1220 | @dots{} | |
1221 | @} | |
1222 | @end example | |
1223 | ||
4b404517 JM |
1224 | @node Compound Literals |
1225 | @section Compound Literals | |
c1f7febf RK |
1226 | @cindex constructor expressions |
1227 | @cindex initializations in expressions | |
1228 | @cindex structures, constructor expression | |
1229 | @cindex expressions, constructor | |
4b404517 JM |
1230 | @cindex compound literals |
1231 | @c The GNU C name for what C99 calls compound literals was "constructor expressions". | |
c1f7febf | 1232 | |
4b404517 | 1233 | ISO C99 supports compound literals. A compound literal looks like |
c1f7febf RK |
1234 | a cast containing an initializer. Its value is an object of the |
1235 | type specified in the cast, containing the elements specified in | |
4b404517 JM |
1236 | the initializer. (GCC does not yet implement the full ISO C99 semantics |
1237 | for compound literals.) As an extension, GCC supports compound literals | |
1238 | in C89 mode and in C++. | |
c1f7febf RK |
1239 | |
1240 | Usually, the specified type is a structure. Assume that | |
1241 | @code{struct foo} and @code{structure} are declared as shown: | |
1242 | ||
1243 | @example | |
1244 | struct foo @{int a; char b[2];@} structure; | |
1245 | @end example | |
1246 | ||
1247 | @noindent | |
4b404517 | 1248 | Here is an example of constructing a @code{struct foo} with a compound literal: |
c1f7febf RK |
1249 | |
1250 | @example | |
1251 | structure = ((struct foo) @{x + y, 'a', 0@}); | |
1252 | @end example | |
1253 | ||
1254 | @noindent | |
1255 | This is equivalent to writing the following: | |
1256 | ||
1257 | @example | |
1258 | @{ | |
1259 | struct foo temp = @{x + y, 'a', 0@}; | |
1260 | structure = temp; | |
1261 | @} | |
1262 | @end example | |
1263 | ||
4b404517 | 1264 | You can also construct an array. If all the elements of the compound literal |
c1f7febf | 1265 | are (made up of) simple constant expressions, suitable for use in |
4b404517 | 1266 | initializers, then the compound literal is an lvalue and can be coerced to a |
c1f7febf RK |
1267 | pointer to its first element, as shown here: |
1268 | ||
1269 | @example | |
1270 | char **foo = (char *[]) @{ "x", "y", "z" @}; | |
1271 | @end example | |
1272 | ||
4b404517 JM |
1273 | Array compound literals whose elements are not simple constants are |
1274 | not very useful, because the compound literal is not an lvalue; ISO C99 | |
1275 | specifies that it is, being a temporary object with automatic storage | |
1276 | duration associated with the enclosing block, but GCC does not yet | |
1277 | implement this. There are currently only two valid ways to use it with | |
1278 | GCC: to subscript it, or initialize | |
c1f7febf RK |
1279 | an array variable with it. The former is probably slower than a |
1280 | @code{switch} statement, while the latter does the same thing an | |
1281 | ordinary C initializer would do. Here is an example of | |
4b404517 | 1282 | subscripting an array compound literal: |
c1f7febf RK |
1283 | |
1284 | @example | |
1285 | output = ((int[]) @{ 2, x, 28 @}) [input]; | |
1286 | @end example | |
1287 | ||
4b404517 JM |
1288 | Compound literals for scalar types and union types are is |
1289 | also allowed, but then the compound literal is equivalent | |
c1f7febf RK |
1290 | to a cast. |
1291 | ||
4b404517 JM |
1292 | @node Designated Inits |
1293 | @section Designated Initializers | |
c1f7febf RK |
1294 | @cindex initializers with labeled elements |
1295 | @cindex labeled elements in initializers | |
1296 | @cindex case labels in initializers | |
4b404517 | 1297 | @cindex designated initializers |
c1f7febf | 1298 | |
26d4fec7 | 1299 | Standard C89 requires the elements of an initializer to appear in a fixed |
c1f7febf RK |
1300 | order, the same as the order of the elements in the array or structure |
1301 | being initialized. | |
1302 | ||
26d4fec7 JM |
1303 | In ISO C99 you can give the elements in any order, specifying the array |
1304 | indices or structure field names they apply to, and GNU C allows this as | |
1305 | an extension in C89 mode as well. This extension is not | |
c1f7febf RK |
1306 | implemented in GNU C++. |
1307 | ||
26d4fec7 | 1308 | To specify an array index, write |
c1f7febf RK |
1309 | @samp{[@var{index}] =} before the element value. For example, |
1310 | ||
1311 | @example | |
26d4fec7 | 1312 | int a[6] = @{ [4] = 29, [2] = 15 @}; |
c1f7febf RK |
1313 | @end example |
1314 | ||
1315 | @noindent | |
1316 | is equivalent to | |
1317 | ||
1318 | @example | |
1319 | int a[6] = @{ 0, 0, 15, 0, 29, 0 @}; | |
1320 | @end example | |
1321 | ||
1322 | @noindent | |
1323 | The index values must be constant expressions, even if the array being | |
1324 | initialized is automatic. | |
1325 | ||
26d4fec7 JM |
1326 | An alternative syntax for this which has been obsolete since GCC 2.5 but |
1327 | GCC still accepts is to write @samp{[@var{index}]} before the element | |
1328 | value, with no @samp{=}. | |
1329 | ||
c1f7febf | 1330 | To initialize a range of elements to the same value, write |
26d4fec7 JM |
1331 | @samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU |
1332 | extension. For example, | |
c1f7febf RK |
1333 | |
1334 | @example | |
1335 | int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @}; | |
1336 | @end example | |
1337 | ||
8b6a5902 JJ |
1338 | @noindent |
1339 | If the value in it has side-effects, the side-effects will happen only once, | |
1340 | not for each initialized field by the range initializer. | |
1341 | ||
c1f7febf RK |
1342 | @noindent |
1343 | Note that the length of the array is the highest value specified | |
1344 | plus one. | |
1345 | ||
1346 | In a structure initializer, specify the name of a field to initialize | |
26d4fec7 | 1347 | with @samp{.@var{fieldname} =} before the element value. For example, |
c1f7febf RK |
1348 | given the following structure, |
1349 | ||
1350 | @example | |
1351 | struct point @{ int x, y; @}; | |
1352 | @end example | |
1353 | ||
1354 | @noindent | |
1355 | the following initialization | |
1356 | ||
1357 | @example | |
26d4fec7 | 1358 | struct point p = @{ .y = yvalue, .x = xvalue @}; |
c1f7febf RK |
1359 | @end example |
1360 | ||
1361 | @noindent | |
1362 | is equivalent to | |
1363 | ||
1364 | @example | |
1365 | struct point p = @{ xvalue, yvalue @}; | |
1366 | @end example | |
1367 | ||
26d4fec7 JM |
1368 | Another syntax which has the same meaning, obsolete since GCC 2.5, is |
1369 | @samp{@var{fieldname}:}, as shown here: | |
c1f7febf RK |
1370 | |
1371 | @example | |
26d4fec7 | 1372 | struct point p = @{ y: yvalue, x: xvalue @}; |
c1f7febf RK |
1373 | @end example |
1374 | ||
4b404517 JM |
1375 | @cindex designators |
1376 | The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a | |
1377 | @dfn{designator}. You can also use a designator (or the obsolete colon | |
1378 | syntax) when initializing a union, to specify which element of the union | |
1379 | should be used. For example, | |
c1f7febf RK |
1380 | |
1381 | @example | |
1382 | union foo @{ int i; double d; @}; | |
1383 | ||
26d4fec7 | 1384 | union foo f = @{ .d = 4 @}; |
c1f7febf RK |
1385 | @end example |
1386 | ||
1387 | @noindent | |
1388 | will convert 4 to a @code{double} to store it in the union using | |
1389 | the second element. By contrast, casting 4 to type @code{union foo} | |
1390 | would store it into the union as the integer @code{i}, since it is | |
1391 | an integer. (@xref{Cast to Union}.) | |
1392 | ||
1393 | You can combine this technique of naming elements with ordinary C | |
1394 | initialization of successive elements. Each initializer element that | |
4b404517 | 1395 | does not have a designator applies to the next consecutive element of the |
c1f7febf RK |
1396 | array or structure. For example, |
1397 | ||
1398 | @example | |
1399 | int a[6] = @{ [1] = v1, v2, [4] = v4 @}; | |
1400 | @end example | |
1401 | ||
1402 | @noindent | |
1403 | is equivalent to | |
1404 | ||
1405 | @example | |
1406 | int a[6] = @{ 0, v1, v2, 0, v4, 0 @}; | |
1407 | @end example | |
1408 | ||
1409 | Labeling the elements of an array initializer is especially useful | |
1410 | when the indices are characters or belong to an @code{enum} type. | |
1411 | For example: | |
1412 | ||
1413 | @example | |
1414 | int whitespace[256] | |
1415 | = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1, | |
1416 | ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @}; | |
1417 | @end example | |
1418 | ||
4b404517 | 1419 | @cindex designator lists |
26d4fec7 | 1420 | You can also write a series of @samp{.@var{fieldname}} and |
4b404517 | 1421 | @samp{[@var{index}]} designators before an @samp{=} to specify a |
26d4fec7 JM |
1422 | nested subobject to initialize; the list is taken relative to the |
1423 | subobject corresponding to the closest surrounding brace pair. For | |
1424 | example, with the @samp{struct point} declaration above: | |
1425 | ||
1426 | @example | |
1427 | struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @}; | |
1428 | @end example | |
1429 | ||
8b6a5902 JJ |
1430 | @noindent |
1431 | If the same field is initialized multiple times, it will have value from | |
1432 | the last initialization. If any such overridden initialization has | |
1433 | side-effect, it is unspecified whether the side-effect happens or not. | |
1434 | Currently, gcc will discard them and issue a warning. | |
1435 | ||
c1f7febf RK |
1436 | @node Case Ranges |
1437 | @section Case Ranges | |
1438 | @cindex case ranges | |
1439 | @cindex ranges in case statements | |
1440 | ||
1441 | You can specify a range of consecutive values in a single @code{case} label, | |
1442 | like this: | |
1443 | ||
1444 | @example | |
1445 | case @var{low} ... @var{high}: | |
1446 | @end example | |
1447 | ||
1448 | @noindent | |
1449 | This has the same effect as the proper number of individual @code{case} | |
1450 | labels, one for each integer value from @var{low} to @var{high}, inclusive. | |
1451 | ||
1452 | This feature is especially useful for ranges of ASCII character codes: | |
1453 | ||
1454 | @example | |
1455 | case 'A' ... 'Z': | |
1456 | @end example | |
1457 | ||
1458 | @strong{Be careful:} Write spaces around the @code{...}, for otherwise | |
1459 | it may be parsed wrong when you use it with integer values. For example, | |
1460 | write this: | |
1461 | ||
1462 | @example | |
1463 | case 1 ... 5: | |
1464 | @end example | |
1465 | ||
1466 | @noindent | |
1467 | rather than this: | |
1468 | ||
1469 | @example | |
1470 | case 1...5: | |
1471 | @end example | |
1472 | ||
1473 | @node Cast to Union | |
1474 | @section Cast to a Union Type | |
1475 | @cindex cast to a union | |
1476 | @cindex union, casting to a | |
1477 | ||
1478 | A cast to union type is similar to other casts, except that the type | |
1479 | specified is a union type. You can specify the type either with | |
1480 | @code{union @var{tag}} or with a typedef name. A cast to union is actually | |
1481 | a constructor though, not a cast, and hence does not yield an lvalue like | |
4b404517 | 1482 | normal casts. (@xref{Compound Literals}.) |
c1f7febf RK |
1483 | |
1484 | The types that may be cast to the union type are those of the members | |
1485 | of the union. Thus, given the following union and variables: | |
1486 | ||
1487 | @example | |
1488 | union foo @{ int i; double d; @}; | |
1489 | int x; | |
1490 | double y; | |
1491 | @end example | |
1492 | ||
1493 | @noindent | |
1494 | both @code{x} and @code{y} can be cast to type @code{union} foo. | |
1495 | ||
1496 | Using the cast as the right-hand side of an assignment to a variable of | |
1497 | union type is equivalent to storing in a member of the union: | |
1498 | ||
1499 | @example | |
1500 | union foo u; | |
1501 | @dots{} | |
1502 | u = (union foo) x @equiv{} u.i = x | |
1503 | u = (union foo) y @equiv{} u.d = y | |
1504 | @end example | |
1505 | ||
1506 | You can also use the union cast as a function argument: | |
1507 | ||
1508 | @example | |
1509 | void hack (union foo); | |
1510 | @dots{} | |
1511 | hack ((union foo) x); | |
1512 | @end example | |
1513 | ||
4b404517 JM |
1514 | @node Mixed Declarations |
1515 | @section Mixed Declarations and Code | |
1516 | @cindex mixed declarations and code | |
1517 | @cindex declarations, mixed with code | |
1518 | @cindex code, mixed with declarations | |
1519 | ||
1520 | ISO C99 and ISO C++ allow declarations and code to be freely mixed | |
1521 | within compound statements. As an extension, GCC also allows this in | |
1522 | C89 mode. For example, you could do: | |
1523 | ||
1524 | @example | |
1525 | int i; | |
1526 | @dots{} | |
1527 | i++; | |
1528 | int j = i + 2; | |
1529 | @end example | |
1530 | ||
1531 | Each identifier is visible from where it is declared until the end of | |
1532 | the enclosing block. | |
1533 | ||
c1f7febf RK |
1534 | @node Function Attributes |
1535 | @section Declaring Attributes of Functions | |
1536 | @cindex function attributes | |
1537 | @cindex declaring attributes of functions | |
1538 | @cindex functions that never return | |
1539 | @cindex functions that have no side effects | |
1540 | @cindex functions in arbitrary sections | |
2a59078d | 1541 | @cindex functions that behave like malloc |
c1f7febf RK |
1542 | @cindex @code{volatile} applied to function |
1543 | @cindex @code{const} applied to function | |
26f6672d | 1544 | @cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments |
c1f7febf RK |
1545 | @cindex functions that are passed arguments in registers on the 386 |
1546 | @cindex functions that pop the argument stack on the 386 | |
1547 | @cindex functions that do not pop the argument stack on the 386 | |
1548 | ||
1549 | In GNU C, you declare certain things about functions called in your program | |
1550 | which help the compiler optimize function calls and check your code more | |
1551 | carefully. | |
1552 | ||
1553 | The keyword @code{__attribute__} allows you to specify special | |
1554 | attributes when making a declaration. This keyword is followed by an | |
9f1bbeaa JM |
1555 | attribute specification inside double parentheses. Fourteen attributes, |
1556 | @code{noreturn}, @code{pure}, @code{const}, @code{format}, | |
1557 | @code{format_arg}, @code{no_instrument_function}, @code{section}, | |
1558 | @code{constructor}, @code{destructor}, @code{unused}, @code{weak}, | |
1559 | @code{malloc}, @code{alias} and @code{no_check_memory_usage} are | |
1560 | currently defined for functions. Several other attributes are defined | |
1561 | for functions on particular target systems. Other attributes, including | |
c1f7febf RK |
1562 | @code{section} are supported for variables declarations (@pxref{Variable |
1563 | Attributes}) and for types (@pxref{Type Attributes}). | |
1564 | ||
1565 | You may also specify attributes with @samp{__} preceding and following | |
1566 | each keyword. This allows you to use them in header files without | |
1567 | being concerned about a possible macro of the same name. For example, | |
1568 | you may use @code{__noreturn__} instead of @code{noreturn}. | |
1569 | ||
2c5e91d2 JM |
1570 | @xref{Attribute Syntax}, for details of the exact syntax for using |
1571 | attributes. | |
1572 | ||
c1f7febf RK |
1573 | @table @code |
1574 | @cindex @code{noreturn} function attribute | |
1575 | @item noreturn | |
1576 | A few standard library functions, such as @code{abort} and @code{exit}, | |
1577 | cannot return. GNU CC knows this automatically. Some programs define | |
1578 | their own functions that never return. You can declare them | |
1579 | @code{noreturn} to tell the compiler this fact. For example, | |
1580 | ||
1581 | @smallexample | |
1582 | void fatal () __attribute__ ((noreturn)); | |
1583 | ||
1584 | void | |
1585 | fatal (@dots{}) | |
1586 | @{ | |
1587 | @dots{} /* @r{Print error message.} */ @dots{} | |
1588 | exit (1); | |
1589 | @} | |
1590 | @end smallexample | |
1591 | ||
1592 | The @code{noreturn} keyword tells the compiler to assume that | |
1593 | @code{fatal} cannot return. It can then optimize without regard to what | |
1594 | would happen if @code{fatal} ever did return. This makes slightly | |
1595 | better code. More importantly, it helps avoid spurious warnings of | |
1596 | uninitialized variables. | |
1597 | ||
1598 | Do not assume that registers saved by the calling function are | |
1599 | restored before calling the @code{noreturn} function. | |
1600 | ||
1601 | It does not make sense for a @code{noreturn} function to have a return | |
1602 | type other than @code{void}. | |
1603 | ||
1604 | The attribute @code{noreturn} is not implemented in GNU C versions | |
1605 | earlier than 2.5. An alternative way to declare that a function does | |
1606 | not return, which works in the current version and in some older | |
1607 | versions, is as follows: | |
1608 | ||
1609 | @smallexample | |
1610 | typedef void voidfn (); | |
1611 | ||
1612 | volatile voidfn fatal; | |
1613 | @end smallexample | |
1614 | ||
2a8f6b90 JH |
1615 | @cindex @code{pure} function attribute |
1616 | @item pure | |
1617 | Many functions have no effects except the return value and their | |
d4047e24 | 1618 | return value depends only on the parameters and/or global variables. |
2a8f6b90 | 1619 | Such a function can be subject |
c1f7febf RK |
1620 | to common subexpression elimination and loop optimization just as an |
1621 | arithmetic operator would be. These functions should be declared | |
2a8f6b90 | 1622 | with the attribute @code{pure}. For example, |
c1f7febf RK |
1623 | |
1624 | @smallexample | |
2a8f6b90 | 1625 | int square (int) __attribute__ ((pure)); |
c1f7febf RK |
1626 | @end smallexample |
1627 | ||
1628 | @noindent | |
1629 | says that the hypothetical function @code{square} is safe to call | |
1630 | fewer times than the program says. | |
1631 | ||
2a8f6b90 JH |
1632 | Some of common examples of pure functions are @code{strlen} or @code{memcmp}. |
1633 | Interesting non-pure functions are functions with infinite loops or those | |
1634 | depending on volatile memory or other system resource, that may change between | |
2a59078d | 1635 | two consecutive calls (such as @code{feof} in a multithreading environment). |
2a8f6b90 JH |
1636 | |
1637 | The attribute @code{pure} is not implemented in GNU C versions earlier | |
1638 | than 2.96. | |
1639 | @cindex @code{const} function attribute | |
1640 | @item const | |
1641 | Many functions do not examine any values except their arguments, and | |
1642 | have no effects except the return value. Basically this is just slightly | |
1643 | more strict class than the "pure" attribute above, since function is not | |
2a59078d | 1644 | allowed to read global memory. |
2a8f6b90 JH |
1645 | |
1646 | @cindex pointer arguments | |
1647 | Note that a function that has pointer arguments and examines the data | |
1648 | pointed to must @emph{not} be declared @code{const}. Likewise, a | |
1649 | function that calls a non-@code{const} function usually must not be | |
1650 | @code{const}. It does not make sense for a @code{const} function to | |
1651 | return @code{void}. | |
1652 | ||
c1f7febf RK |
1653 | The attribute @code{const} is not implemented in GNU C versions earlier |
1654 | than 2.5. An alternative way to declare that a function has no side | |
1655 | effects, which works in the current version and in some older versions, | |
1656 | is as follows: | |
1657 | ||
1658 | @smallexample | |
1659 | typedef int intfn (); | |
1660 | ||
1661 | extern const intfn square; | |
1662 | @end smallexample | |
1663 | ||
1664 | This approach does not work in GNU C++ from 2.6.0 on, since the language | |
1665 | specifies that the @samp{const} must be attached to the return value. | |
1666 | ||
c1f7febf RK |
1667 | |
1668 | @item format (@var{archetype}, @var{string-index}, @var{first-to-check}) | |
1669 | @cindex @code{format} function attribute | |
bb72a084 | 1670 | The @code{format} attribute specifies that a function takes @code{printf}, |
26f6672d JM |
1671 | @code{scanf}, @code{strftime} or @code{strfmon} style arguments which |
1672 | should be type-checked against a format string. For example, the | |
1673 | declaration: | |
c1f7febf RK |
1674 | |
1675 | @smallexample | |
1676 | extern int | |
1677 | my_printf (void *my_object, const char *my_format, ...) | |
1678 | __attribute__ ((format (printf, 2, 3))); | |
1679 | @end smallexample | |
1680 | ||
1681 | @noindent | |
1682 | causes the compiler to check the arguments in calls to @code{my_printf} | |
1683 | for consistency with the @code{printf} style format string argument | |
1684 | @code{my_format}. | |
1685 | ||
1686 | The parameter @var{archetype} determines how the format string is | |
26f6672d JM |
1687 | interpreted, and should be @code{printf}, @code{scanf}, @code{strftime} |
1688 | or @code{strfmon}. (You can also use @code{__printf__}, | |
1689 | @code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) The | |
c1f7febf RK |
1690 | parameter @var{string-index} specifies which argument is the format |
1691 | string argument (starting from 1), while @var{first-to-check} is the | |
1692 | number of the first argument to check against the format string. For | |
1693 | functions where the arguments are not available to be checked (such as | |
1694 | @code{vprintf}), specify the third parameter as zero. In this case the | |
b722c82c JM |
1695 | compiler only checks the format string for consistency. For |
1696 | @code{strftime} formats, the third parameter is required to be zero. | |
c1f7febf RK |
1697 | |
1698 | In the example above, the format string (@code{my_format}) is the second | |
1699 | argument of the function @code{my_print}, and the arguments to check | |
1700 | start with the third argument, so the correct parameters for the format | |
1701 | attribute are 2 and 3. | |
1702 | ||
1703 | The @code{format} attribute allows you to identify your own functions | |
1704 | which take format strings as arguments, so that GNU CC can check the | |
b722c82c JM |
1705 | calls to these functions for errors. The compiler always (unless |
1706 | @samp{-ffreestanding} is used) checks formats | |
1707 | for the standard library functions @code{printf}, @code{fprintf}, | |
bb72a084 | 1708 | @code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime}, |
c1f7febf RK |
1709 | @code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such |
1710 | warnings are requested (using @samp{-Wformat}), so there is no need to | |
b722c82c JM |
1711 | modify the header file @file{stdio.h}. In C99 mode, the functions |
1712 | @code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and | |
26f6672d JM |
1713 | @code{vsscanf} are also checked. Except in strictly conforming C |
1714 | standard modes, the X/Open function @code{strfmon} is also checked. | |
b722c82c | 1715 | @xref{C Dialect Options,,Options Controlling C Dialect}. |
c1f7febf RK |
1716 | |
1717 | @item format_arg (@var{string-index}) | |
1718 | @cindex @code{format_arg} function attribute | |
26f6672d JM |
1719 | The @code{format_arg} attribute specifies that a function takes a format |
1720 | string for a @code{printf}, @code{scanf}, @code{strftime} or | |
1721 | @code{strfmon} style function and modifies it (for example, to translate | |
1722 | it into another language), so the result can be passed to a | |
1723 | @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style | |
1724 | function (with the remaining arguments to the format function the same | |
1725 | as they would have been for the unmodified string). For example, the | |
1726 | declaration: | |
c1f7febf RK |
1727 | |
1728 | @smallexample | |
1729 | extern char * | |
1730 | my_dgettext (char *my_domain, const char *my_format) | |
1731 | __attribute__ ((format_arg (2))); | |
1732 | @end smallexample | |
1733 | ||
1734 | @noindent | |
26f6672d JM |
1735 | causes the compiler to check the arguments in calls to a @code{printf}, |
1736 | @code{scanf}, @code{strftime} or @code{strfmon} type function, whose | |
1737 | format string argument is a call to the @code{my_dgettext} function, for | |
1738 | consistency with the format string argument @code{my_format}. If the | |
1739 | @code{format_arg} attribute had not been specified, all the compiler | |
1740 | could tell in such calls to format functions would be that the format | |
1741 | string argument is not constant; this would generate a warning when | |
1742 | @code{-Wformat-nonliteral} is used, but the calls could not be checked | |
1743 | without the attribute. | |
c1f7febf RK |
1744 | |
1745 | The parameter @var{string-index} specifies which argument is the format | |
1746 | string argument (starting from 1). | |
1747 | ||
1748 | The @code{format-arg} attribute allows you to identify your own | |
1749 | functions which modify format strings, so that GNU CC can check the | |
26f6672d JM |
1750 | calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} |
1751 | type function whose operands are a call to one of your own function. | |
1752 | The compiler always treats @code{gettext}, @code{dgettext}, and | |
1753 | @code{dcgettext} in this manner except when strict ISO C support is | |
1754 | requested by @samp{-ansi} or an appropriate @samp{-std} option, or | |
1755 | @samp{-ffreestanding} is used. @xref{C Dialect Options,,Options | |
1756 | Controlling C Dialect}. | |
c1f7febf | 1757 | |
07417085 KR |
1758 | @item no_instrument_function |
1759 | @cindex @code{no_instrument_function} function attribute | |
1760 | If @samp{-finstrument-functions} is given, profiling function calls will | |
1761 | be generated at entry and exit of most user-compiled functions. | |
1762 | Functions with this attribute will not be so instrumented. | |
1763 | ||
c1f7febf RK |
1764 | @item section ("section-name") |
1765 | @cindex @code{section} function attribute | |
1766 | Normally, the compiler places the code it generates in the @code{text} section. | |
1767 | Sometimes, however, you need additional sections, or you need certain | |
1768 | particular functions to appear in special sections. The @code{section} | |
1769 | attribute specifies that a function lives in a particular section. | |
1770 | For example, the declaration: | |
1771 | ||
1772 | @smallexample | |
1773 | extern void foobar (void) __attribute__ ((section ("bar"))); | |
1774 | @end smallexample | |
1775 | ||
1776 | @noindent | |
1777 | puts the function @code{foobar} in the @code{bar} section. | |
1778 | ||
1779 | Some file formats do not support arbitrary sections so the @code{section} | |
1780 | attribute is not available on all platforms. | |
1781 | If you need to map the entire contents of a module to a particular | |
1782 | section, consider using the facilities of the linker instead. | |
1783 | ||
1784 | @item constructor | |
1785 | @itemx destructor | |
1786 | @cindex @code{constructor} function attribute | |
1787 | @cindex @code{destructor} function attribute | |
1788 | The @code{constructor} attribute causes the function to be called | |
1789 | automatically before execution enters @code{main ()}. Similarly, the | |
1790 | @code{destructor} attribute causes the function to be called | |
1791 | automatically after @code{main ()} has completed or @code{exit ()} has | |
1792 | been called. Functions with these attributes are useful for | |
1793 | initializing data that will be used implicitly during the execution of | |
1794 | the program. | |
1795 | ||
1796 | These attributes are not currently implemented for Objective C. | |
1797 | ||
1798 | @item unused | |
1799 | This attribute, attached to a function, means that the function is meant | |
1800 | to be possibly unused. GNU CC will not produce a warning for this | |
1801 | function. GNU C++ does not currently support this attribute as | |
1802 | definitions without parameters are valid in C++. | |
1803 | ||
1804 | @item weak | |
1805 | @cindex @code{weak} attribute | |
1806 | The @code{weak} attribute causes the declaration to be emitted as a weak | |
1807 | symbol rather than a global. This is primarily useful in defining | |
1808 | library functions which can be overridden in user code, though it can | |
1809 | also be used with non-function declarations. Weak symbols are supported | |
1810 | for ELF targets, and also for a.out targets when using the GNU assembler | |
1811 | and linker. | |
1812 | ||
140592a0 AG |
1813 | @item malloc |
1814 | @cindex @code{malloc} attribute | |
1815 | The @code{malloc} attribute is used to tell the compiler that a function | |
1816 | may be treated as if it were the malloc function. The compiler assumes | |
1817 | that calls to malloc result in a pointers that cannot alias anything. | |
1818 | This will often improve optimization. | |
1819 | ||
c1f7febf RK |
1820 | @item alias ("target") |
1821 | @cindex @code{alias} attribute | |
1822 | The @code{alias} attribute causes the declaration to be emitted as an | |
1823 | alias for another symbol, which must be specified. For instance, | |
1824 | ||
1825 | @smallexample | |
1826 | void __f () @{ /* do something */; @} | |
1827 | void f () __attribute__ ((weak, alias ("__f"))); | |
1828 | @end smallexample | |
1829 | ||
1830 | declares @samp{f} to be a weak alias for @samp{__f}. In C++, the | |
1831 | mangled name for the target must be used. | |
1832 | ||
af3e86c2 RK |
1833 | Not all target machines support this attribute. |
1834 | ||
7d384cc0 KR |
1835 | @item no_check_memory_usage |
1836 | @cindex @code{no_check_memory_usage} function attribute | |
c5c76735 JL |
1837 | The @code{no_check_memory_usage} attribute causes GNU CC to omit checks |
1838 | of memory references when it generates code for that function. Normally | |
1839 | if you specify @samp{-fcheck-memory-usage} (see @pxref{Code Gen | |
1840 | Options}), GNU CC generates calls to support routines before most memory | |
1841 | accesses to permit support code to record usage and detect uses of | |
1842 | uninitialized or unallocated storage. Since GNU CC cannot handle | |
1843 | @code{asm} statements properly they are not allowed in such functions. | |
1844 | If you declare a function with this attribute, GNU CC will not generate | |
7d384cc0 | 1845 | memory checking code for that function, permitting the use of @code{asm} |
c5c76735 JL |
1846 | statements without having to compile that function with different |
1847 | options. This also allows you to write support routines of your own if | |
1848 | you wish, without getting infinite recursion if they get compiled with | |
1849 | @code{-fcheck-memory-usage}. | |
7d384cc0 | 1850 | |
c1f7febf RK |
1851 | @item regparm (@var{number}) |
1852 | @cindex functions that are passed arguments in registers on the 386 | |
1853 | On the Intel 386, the @code{regparm} attribute causes the compiler to | |
1854 | pass up to @var{number} integer arguments in registers @var{EAX}, | |
1855 | @var{EDX}, and @var{ECX} instead of on the stack. Functions that take a | |
1856 | variable number of arguments will continue to be passed all of their | |
1857 | arguments on the stack. | |
1858 | ||
1859 | @item stdcall | |
1860 | @cindex functions that pop the argument stack on the 386 | |
1861 | On the Intel 386, the @code{stdcall} attribute causes the compiler to | |
1862 | assume that the called function will pop off the stack space used to | |
1863 | pass arguments, unless it takes a variable number of arguments. | |
1864 | ||
1865 | The PowerPC compiler for Windows NT currently ignores the @code{stdcall} | |
1866 | attribute. | |
1867 | ||
1868 | @item cdecl | |
1869 | @cindex functions that do pop the argument stack on the 386 | |
1870 | On the Intel 386, the @code{cdecl} attribute causes the compiler to | |
1871 | assume that the calling function will pop off the stack space used to | |
1872 | pass arguments. This is | |
1873 | useful to override the effects of the @samp{-mrtd} switch. | |
1874 | ||
1875 | The PowerPC compiler for Windows NT currently ignores the @code{cdecl} | |
1876 | attribute. | |
1877 | ||
1878 | @item longcall | |
1879 | @cindex functions called via pointer on the RS/6000 and PowerPC | |
1880 | On the RS/6000 and PowerPC, the @code{longcall} attribute causes the | |
1881 | compiler to always call the function via a pointer, so that functions | |
1882 | which reside further than 64 megabytes (67,108,864 bytes) from the | |
1883 | current location can be called. | |
1884 | ||
c27ba912 DM |
1885 | @item long_call/short_call |
1886 | @cindex indirect calls on ARM | |
1887 | This attribute allows to specify how to call a particular function on | |
1888 | ARM. Both attributes override the @code{-mlong-calls} (@pxref{ARM Options}) | |
1889 | command line switch and @code{#pragma long_calls} settings. The | |
1890 | @code{long_call} attribute causes the compiler to always call the | |
1891 | function by first loading its address into a register and then using the | |
1892 | contents of that register. The @code{short_call} attribute always places | |
1893 | the offset to the function from the call site into the @samp{BL} | |
1894 | instruction directly. | |
1895 | ||
c1f7febf RK |
1896 | @item dllimport |
1897 | @cindex functions which are imported from a dll on PowerPC Windows NT | |
1898 | On the PowerPC running Windows NT, the @code{dllimport} attribute causes | |
1899 | the compiler to call the function via a global pointer to the function | |
1900 | pointer that is set up by the Windows NT dll library. The pointer name | |
1901 | is formed by combining @code{__imp_} and the function name. | |
1902 | ||
1903 | @item dllexport | |
1904 | @cindex functions which are exported from a dll on PowerPC Windows NT | |
1905 | On the PowerPC running Windows NT, the @code{dllexport} attribute causes | |
1906 | the compiler to provide a global pointer to the function pointer, so | |
1907 | that it can be called with the @code{dllimport} attribute. The pointer | |
1908 | name is formed by combining @code{__imp_} and the function name. | |
1909 | ||
1910 | @item exception (@var{except-func} [, @var{except-arg}]) | |
1911 | @cindex functions which specify exception handling on PowerPC Windows NT | |
1912 | On the PowerPC running Windows NT, the @code{exception} attribute causes | |
1913 | the compiler to modify the structured exception table entry it emits for | |
1914 | the declared function. The string or identifier @var{except-func} is | |
1915 | placed in the third entry of the structured exception table. It | |
1916 | represents a function, which is called by the exception handling | |
1917 | mechanism if an exception occurs. If it was specified, the string or | |
1918 | identifier @var{except-arg} is placed in the fourth entry of the | |
1919 | structured exception table. | |
1920 | ||
1921 | @item function_vector | |
1922 | @cindex calling functions through the function vector on the H8/300 processors | |
1923 | Use this option on the H8/300 and H8/300H to indicate that the specified | |
1924 | function should be called through the function vector. Calling a | |
1925 | function through the function vector will reduce code size, however; | |
1926 | the function vector has a limited size (maximum 128 entries on the H8/300 | |
1927 | and 64 entries on the H8/300H) and shares space with the interrupt vector. | |
1928 | ||
1929 | You must use GAS and GLD from GNU binutils version 2.7 or later for | |
1930 | this option to work correctly. | |
1931 | ||
1932 | @item interrupt_handler | |
1933 | @cindex interrupt handler functions on the H8/300 processors | |
1934 | Use this option on the H8/300 and H8/300H to indicate that the specified | |
1935 | function is an interrupt handler. The compiler will generate function | |
1936 | entry and exit sequences suitable for use in an interrupt handler when this | |
1937 | attribute is present. | |
1938 | ||
6d3d9133 NC |
1939 | @item interrupt |
1940 | @cindex interrupt handler functions | |
1941 | Use this option on the ARM, AVR and M32R/D ports to indicate that the | |
1942 | specified function is an interrupt handler. The compiler will generate | |
1943 | function entry and exit sequences suitable for use in an interrupt | |
1944 | handler when this attribute is present. | |
1945 | ||
a7cf60a2 | 1946 | Note, interrupt handlers for the H8/300 and H8/300H processors can be |
6d3d9133 NC |
1947 | specified via the @code{interrupt_handler} attribute. |
1948 | ||
1949 | Note, on the AVR interrupts will be enabled inside the function. | |
1950 | ||
1951 | Note, for the ARM you can specify the kind of interrupt to be handled by | |
1952 | adding an optional parameter to the interrupt attribute like this: | |
1953 | ||
1954 | @smallexample | |
1955 | void f () __attribute__ ((interrupt ("IRQ"))); | |
1956 | @end smallexample | |
1957 | ||
a7cf60a2 | 1958 | Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT and UNDEF. |
6d3d9133 | 1959 | |
c1f7febf RK |
1960 | @item eightbit_data |
1961 | @cindex eight bit data on the H8/300 and H8/300H | |
1962 | Use this option on the H8/300 and H8/300H to indicate that the specified | |
1963 | variable should be placed into the eight bit data section. | |
1964 | The compiler will generate more efficient code for certain operations | |
1965 | on data in the eight bit data area. Note the eight bit data area is limited to | |
1966 | 256 bytes of data. | |
1967 | ||
1968 | You must use GAS and GLD from GNU binutils version 2.7 or later for | |
1969 | this option to work correctly. | |
1970 | ||
1971 | @item tiny_data | |
1972 | @cindex tiny data section on the H8/300H | |
1973 | Use this option on the H8/300H to indicate that the specified | |
1974 | variable should be placed into the tiny data section. | |
1975 | The compiler will generate more efficient code for loads and stores | |
1976 | on data in the tiny data section. Note the tiny data area is limited to | |
1977 | slightly under 32kbytes of data. | |
845da534 | 1978 | |
052a4b28 DC |
1979 | @item signal |
1980 | @cindex signal handler functions on the AVR processors | |
1981 | Use this option on the AVR to indicate that the specified | |
1982 | function is an signal handler. The compiler will generate function | |
1983 | entry and exit sequences suitable for use in an signal handler when this | |
1984 | attribute is present. Interrupts will be disabled inside function. | |
1985 | ||
1986 | @item naked | |
6d3d9133 NC |
1987 | @cindex function without a prologue/epilogue code |
1988 | Use this option on the ARM or AVR ports to indicate that the specified | |
1989 | function do not need prologue/epilogue sequences generated by the | |
1990 | compiler. It is up to the programmer to provide these sequences. | |
052a4b28 | 1991 | |
845da534 DE |
1992 | @item model (@var{model-name}) |
1993 | @cindex function addressability on the M32R/D | |
1994 | Use this attribute on the M32R/D to set the addressability of an object, | |
1995 | and the code generated for a function. | |
1996 | The identifier @var{model-name} is one of @code{small}, @code{medium}, | |
1997 | or @code{large}, representing each of the code models. | |
1998 | ||
1999 | Small model objects live in the lower 16MB of memory (so that their | |
2000 | addresses can be loaded with the @code{ld24} instruction), and are | |
2001 | callable with the @code{bl} instruction. | |
2002 | ||
02f52e19 | 2003 | Medium model objects may live anywhere in the 32-bit address space (the |
845da534 DE |
2004 | compiler will generate @code{seth/add3} instructions to load their addresses), |
2005 | and are callable with the @code{bl} instruction. | |
2006 | ||
02f52e19 | 2007 | Large model objects may live anywhere in the 32-bit address space (the |
845da534 DE |
2008 | compiler will generate @code{seth/add3} instructions to load their addresses), |
2009 | and may not be reachable with the @code{bl} instruction (the compiler will | |
2010 | generate the much slower @code{seth/add3/jl} instruction sequence). | |
2011 | ||
c1f7febf RK |
2012 | @end table |
2013 | ||
2014 | You can specify multiple attributes in a declaration by separating them | |
2015 | by commas within the double parentheses or by immediately following an | |
2016 | attribute declaration with another attribute declaration. | |
2017 | ||
2018 | @cindex @code{#pragma}, reason for not using | |
2019 | @cindex pragma, reason for not using | |
9f1bbeaa JM |
2020 | Some people object to the @code{__attribute__} feature, suggesting that |
2021 | ISO C's @code{#pragma} should be used instead. At the time | |
2022 | @code{__attribute__} was designed, there were two reasons for not doing | |
2023 | this. | |
c1f7febf RK |
2024 | |
2025 | @enumerate | |
2026 | @item | |
2027 | It is impossible to generate @code{#pragma} commands from a macro. | |
2028 | ||
2029 | @item | |
2030 | There is no telling what the same @code{#pragma} might mean in another | |
2031 | compiler. | |
2032 | @end enumerate | |
2033 | ||
9f1bbeaa JM |
2034 | These two reasons applied to almost any application that might have been |
2035 | proposed for @code{#pragma}. It was basically a mistake to use | |
2036 | @code{#pragma} for @emph{anything}. | |
2037 | ||
2038 | The ISO C99 standard includes @code{_Pragma}, which now allows pragmas | |
2039 | to be generated from macros. In addition, a @code{#pragma GCC} | |
2040 | namespace is now in use for GCC-specific pragmas. However, it has been | |
2041 | found convenient to use @code{__attribute__} to achieve a natural | |
2042 | attachment of attributes to their corresponding declarations, whereas | |
2043 | @code{#pragma GCC} is of use for constructs that do not naturally form | |
2044 | part of the grammar. @xref{Other Directives,,Miscellaneous | |
2045 | Preprocessing Directives, cpp, The C Preprocessor}. | |
c1f7febf | 2046 | |
2c5e91d2 JM |
2047 | @node Attribute Syntax |
2048 | @section Attribute Syntax | |
2049 | @cindex attribute syntax | |
2050 | ||
2051 | This section describes the syntax with which @code{__attribute__} may be | |
2052 | used, and the constructs to which attribute specifiers bind, for the C | |
2053 | language. Some details may vary for C++ and Objective C. Because of | |
2054 | infelicities in the grammar for attributes, some forms described here | |
2055 | may not be successfully parsed in all cases. | |
2056 | ||
2057 | @xref{Function Attributes}, for details of the semantics of attributes | |
2058 | applying to functions. @xref{Variable Attributes}, for details of the | |
2059 | semantics of attributes applying to variables. @xref{Type Attributes}, | |
2060 | for details of the semantics of attributes applying to structure, union | |
2061 | and enumerated types. | |
2062 | ||
2063 | An @dfn{attribute specifier} is of the form | |
2064 | @code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list} | |
2065 | is a possibly empty comma-separated sequence of @dfn{attributes}, where | |
2066 | each attribute is one of the following: | |
2067 | ||
2068 | @itemize @bullet | |
2069 | @item | |
2070 | Empty. Empty attributes are ignored. | |
2071 | ||
2072 | @item | |
2073 | A word (which may be an identifier such as @code{unused}, or a reserved | |
2074 | word such as @code{const}). | |
2075 | ||
2076 | @item | |
2077 | A word, followed by, in parentheses, parameters for the attribute. | |
2078 | These parameters take one of the following forms: | |
2079 | ||
2080 | @itemize @bullet | |
2081 | @item | |
2082 | An identifier. For example, @code{mode} attributes use this form. | |
2083 | ||
2084 | @item | |
2085 | An identifier followed by a comma and a non-empty comma-separated list | |
2086 | of expressions. For example, @code{format} attributes use this form. | |
2087 | ||
2088 | @item | |
2089 | A possibly empty comma-separated list of expressions. For example, | |
2090 | @code{format_arg} attributes use this form with the list being a single | |
2091 | integer constant expression, and @code{alias} attributes use this form | |
2092 | with the list being a single string constant. | |
2093 | @end itemize | |
2094 | @end itemize | |
2095 | ||
2096 | An @dfn{attribute specifier list} is a sequence of one or more attribute | |
2097 | specifiers, not separated by any other tokens. | |
2098 | ||
2099 | An attribute specifier list may appear after the colon following a | |
2100 | label, other than a @code{case} or @code{default} label. The only | |
2101 | attribute it makes sense to use after a label is @code{unused}. This | |
2102 | feature is intended for code generated by programs which contains labels | |
2103 | that may be unused but which is compiled with @option{-Wall}. It would | |
2104 | not normally be appropriate to use in it human-written code, though it | |
2105 | could be useful in cases where the code that jumps to the label is | |
2106 | contained within an @code{#ifdef} conditional. | |
2107 | ||
2108 | An attribute specifier list may appear as part of a @code{struct}, | |
2109 | @code{union} or @code{enum} specifier. It may go either immediately | |
2110 | after the @code{struct}, @code{union} or @code{enum} keyword, or after | |
2111 | the closing brace. It is ignored if the content of the structure, union | |
2112 | or enumerated type is not defined in the specifier in which the | |
2113 | attribute specifier list is used---that is, in usages such as | |
2114 | @code{struct __attribute__((foo)) bar} with no following opening brace. | |
2115 | Where attribute specifiers follow the closing brace, they are considered | |
2116 | to relate to the structure, union or enumerated type defined, not to any | |
2117 | enclosing declaration the type specifier appears in, and the type | |
2118 | defined is not complete until after the attribute specifiers. | |
2119 | @c Otherwise, there would be the following problems: a shift/reduce | |
2120 | @c conflict between attributes binding the the struct/union/enum and | |
2121 | @c binding to the list of specifiers/qualifiers; and "aligned" | |
2122 | @c attributes could use sizeof for the structure, but the size could be | |
2123 | @c changed later by "packed" attributes. | |
2124 | ||
2125 | Otherwise, an attribute specifier appears as part of a declaration, | |
2126 | counting declarations of unnamed parameters and type names, and relates | |
2127 | to that declaration (which may be nested in another declaration, for | |
2128 | example in the case of a parameter declaration). In future, attribute | |
2129 | specifiers in some places may however apply to a particular declarator | |
ff867905 JM |
2130 | within a declaration instead; these cases are noted below. Where an |
2131 | attribute specifier is applied to a parameter declared as a function or | |
2132 | an array, it should apply to the function or array rather than the | |
2133 | pointer to which the parameter is implicitly converted, but this is not | |
2134 | yet correctly implemented. | |
2c5e91d2 JM |
2135 | |
2136 | Any list of specifiers and qualifiers at the start of a declaration may | |
2137 | contain attribute specifiers, whether or not such a list may in that | |
2138 | context contain storage class specifiers. (Some attributes, however, | |
2139 | are essentially in the nature of storage class specifiers, and only make | |
2140 | sense where storage class specifiers may be used; for example, | |
2141 | @code{section}.) There is one necessary limitation to this syntax: the | |
2142 | first old-style parameter declaration in a function definition cannot | |
2143 | begin with an attribute specifier, because such an attribute applies to | |
2144 | the function instead by syntax described below (which, however, is not | |
2145 | yet implemented in this case). In some other cases, attribute | |
2146 | specifiers are permitted by this grammar but not yet supported by the | |
2147 | compiler. All attribute specifiers in this place relate to the | |
2148 | declaration as a whole. In the obsolencent usage where a type of | |
2149 | @code{int} is implied by the absence of type specifiers, such a list of | |
2150 | specifiers and qualifiers may be an attribute specifier list with no | |
2151 | other specifiers or qualifiers. | |
2152 | ||
2153 | An attribute specifier list may appear immediately before a declarator | |
2154 | (other than the first) in a comma-separated list of declarators in a | |
2155 | declaration of more than one identifier using a single list of | |
2156 | specifiers and qualifiers. At present, such attribute specifiers apply | |
2157 | not only to the identifier before whose declarator they appear, but to | |
2158 | all subsequent identifiers declared in that declaration, but in future | |
2159 | they may apply only to that single identifier. For example, in | |
2160 | @code{__attribute__((noreturn)) void d0 (void), | |
2161 | __attribute__((format(printf, 1, 2))) d1 (const char *, ...), d2 | |
2162 | (void)}, the @code{noreturn} attribute applies to all the functions | |
2163 | declared; the @code{format} attribute should only apply to @code{d1}, | |
2164 | but at present applies to @code{d2} as well (and so causes an error). | |
2165 | ||
2166 | An attribute specifier list may appear immediately before the comma, | |
2167 | @code{=} or semicolon terminating the declaration of an identifier other | |
2168 | than a function definition. At present, such attribute specifiers apply | |
2169 | to the declared object or function, but in future they may attach to the | |
2170 | outermost adjacent declarator. In simple cases there is no difference, | |
2171 | but, for example, in @code{void (****f)(void) | |
2172 | __attribute__((noreturn));}, at present the @code{noreturn} attribute | |
2173 | applies to @code{f}, which causes a warning since @code{f} is not a | |
2174 | function, but in future it may apply to the function @code{****f}. The | |
2175 | precise semantics of what attributes in such cases will apply to are not | |
2176 | yet specified. Where an assembler name for an object or function is | |
2177 | specified (@pxref{Asm Labels}), at present the attribute must follow the | |
2178 | @code{asm} specification; in future, attributes before the @code{asm} | |
2179 | specification may apply to the adjacent declarator, and those after it | |
2180 | to the declared object or function. | |
2181 | ||
2182 | An attribute specifier list may, in future, be permitted to appear after | |
2183 | the declarator in a function definition (before any old-style parameter | |
2184 | declarations or the function body). | |
2185 | ||
0e03329a JM |
2186 | Attribute specifiers may be mixed with type qualifiers appearing inside |
2187 | the @code{[]} of a parameter array declarator, in the C99 construct by | |
2188 | which such qualifiers are applied to the pointer to which the array is | |
2189 | implicitly converted. Such attribute specifiers apply to the pointer, | |
2190 | not to the array, but at present this is not implemented and they are | |
2191 | ignored. | |
2192 | ||
2c5e91d2 JM |
2193 | An attribute specifier list may appear at the start of a nested |
2194 | declarator. At present, there are some limitations in this usage: the | |
2195 | attributes apply to the identifer declared, and to all subsequent | |
2196 | identifiers declared in that declaration (if it includes a | |
2197 | comma-separated list of declarators), rather than to a specific | |
2198 | declarator. When attribute specifiers follow the @code{*} of a pointer | |
2199 | declarator, they must presently follow any type qualifiers present, and | |
2200 | cannot be mixed with them. The following describes intended future | |
2201 | semantics which make this syntax more useful only. It will make the | |
2202 | most sense if you are familiar with the formal specification of | |
2203 | declarators in the ISO C standard. | |
2204 | ||
2205 | Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T | |
2206 | D1}, where @code{T} contains declaration specifiers that specify a type | |
2207 | @var{Type} (such as @code{int}) and @code{D1} is a declarator that | |
2208 | contains an identifier @var{ident}. The type specified for @var{ident} | |
2209 | for derived declarators whose type does not include an attribute | |
2210 | specifier is as in the ISO C standard. | |
2211 | ||
2212 | If @code{D1} has the form @code{( @var{attribute-specifier-list} D )}, | |
2213 | and the declaration @code{T D} specifies the type | |
2214 | ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then | |
2215 | @code{T D1} specifies the type ``@var{derived-declarator-type-list} | |
2216 | @var{attribute-specifier-list} @var{Type}'' for @var{ident}. | |
2217 | ||
2218 | If @code{D1} has the form @code{* | |
2219 | @var{type-qualifier-and-attribute-specifier-list} D}, and the | |
2220 | declaration @code{T D} specifies the type | |
2221 | ``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then | |
2222 | @code{T D1} specifies the type ``@var{derived-declarator-type-list} | |
2223 | @var{type-qualifier-and-attribute-specifier-list} @var{Type}'' for | |
2224 | @var{ident}. | |
2225 | ||
2226 | For example, @code{void (__attribute__((noreturn)) ****f)();} specifies | |
2227 | the type ``pointer to pointer to pointer to pointer to non-returning | |
2228 | function returning @code{void}''. As another example, @code{char | |
2229 | *__attribute__((aligned(8))) *f;} specifies the type ``pointer to | |
2230 | 8-byte-aligned pointer to @code{char}''. Note again that this describes | |
2231 | intended future semantics, not current implementation. | |
2232 | ||
c1f7febf RK |
2233 | @node Function Prototypes |
2234 | @section Prototypes and Old-Style Function Definitions | |
2235 | @cindex function prototype declarations | |
2236 | @cindex old-style function definitions | |
2237 | @cindex promotion of formal parameters | |
2238 | ||
5490d604 | 2239 | GNU C extends ISO C to allow a function prototype to override a later |
c1f7febf RK |
2240 | old-style non-prototype definition. Consider the following example: |
2241 | ||
2242 | @example | |
2243 | /* @r{Use prototypes unless the compiler is old-fashioned.} */ | |
d863830b | 2244 | #ifdef __STDC__ |
c1f7febf RK |
2245 | #define P(x) x |
2246 | #else | |
2247 | #define P(x) () | |
2248 | #endif | |
2249 | ||
2250 | /* @r{Prototype function declaration.} */ | |
2251 | int isroot P((uid_t)); | |
2252 | ||
2253 | /* @r{Old-style function definition.} */ | |
2254 | int | |
2255 | isroot (x) /* ??? lossage here ??? */ | |
2256 | uid_t x; | |
2257 | @{ | |
2258 | return x == 0; | |
2259 | @} | |
2260 | @end example | |
2261 | ||
5490d604 | 2262 | Suppose the type @code{uid_t} happens to be @code{short}. ISO C does |
c1f7febf RK |
2263 | not allow this example, because subword arguments in old-style |
2264 | non-prototype definitions are promoted. Therefore in this example the | |
2265 | function definition's argument is really an @code{int}, which does not | |
2266 | match the prototype argument type of @code{short}. | |
2267 | ||
5490d604 | 2268 | This restriction of ISO C makes it hard to write code that is portable |
c1f7febf RK |
2269 | to traditional C compilers, because the programmer does not know |
2270 | whether the @code{uid_t} type is @code{short}, @code{int}, or | |
2271 | @code{long}. Therefore, in cases like these GNU C allows a prototype | |
2272 | to override a later old-style definition. More precisely, in GNU C, a | |
2273 | function prototype argument type overrides the argument type specified | |
2274 | by a later old-style definition if the former type is the same as the | |
2275 | latter type before promotion. Thus in GNU C the above example is | |
2276 | equivalent to the following: | |
2277 | ||
2278 | @example | |
2279 | int isroot (uid_t); | |
2280 | ||
2281 | int | |
2282 | isroot (uid_t x) | |
2283 | @{ | |
2284 | return x == 0; | |
2285 | @} | |
2286 | @end example | |
2287 | ||
2288 | GNU C++ does not support old-style function definitions, so this | |
2289 | extension is irrelevant. | |
2290 | ||
2291 | @node C++ Comments | |
2292 | @section C++ Style Comments | |
2293 | @cindex // | |
2294 | @cindex C++ comments | |
2295 | @cindex comments, C++ style | |
2296 | ||
2297 | In GNU C, you may use C++ style comments, which start with @samp{//} and | |
2298 | continue until the end of the line. Many other C implementations allow | |
2299 | such comments, and they are likely to be in a future C standard. | |
2300 | However, C++ style comments are not recognized if you specify | |
5490d604 JM |
2301 | @w{@samp{-ansi}}, a @option{-std} option specifying a version of ISO C |
2302 | before C99, or @w{@samp{-traditional}}, since they are incompatible | |
c1f7febf RK |
2303 | with traditional constructs like @code{dividend//*comment*/divisor}. |
2304 | ||
2305 | @node Dollar Signs | |
2306 | @section Dollar Signs in Identifier Names | |
2307 | @cindex $ | |
2308 | @cindex dollar signs in identifier names | |
2309 | @cindex identifier names, dollar signs in | |
2310 | ||
79188db9 RK |
2311 | In GNU C, you may normally use dollar signs in identifier names. |
2312 | This is because many traditional C implementations allow such identifiers. | |
2313 | However, dollar signs in identifiers are not supported on a few target | |
2314 | machines, typically because the target assembler does not allow them. | |
c1f7febf RK |
2315 | |
2316 | @node Character Escapes | |
2317 | @section The Character @key{ESC} in Constants | |
2318 | ||
2319 | You can use the sequence @samp{\e} in a string or character constant to | |
2320 | stand for the ASCII character @key{ESC}. | |
2321 | ||
2322 | @node Alignment | |
2323 | @section Inquiring on Alignment of Types or Variables | |
2324 | @cindex alignment | |
2325 | @cindex type alignment | |
2326 | @cindex variable alignment | |
2327 | ||
2328 | The keyword @code{__alignof__} allows you to inquire about how an object | |
2329 | is aligned, or the minimum alignment usually required by a type. Its | |
2330 | syntax is just like @code{sizeof}. | |
2331 | ||
2332 | For example, if the target machine requires a @code{double} value to be | |
2333 | aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8. | |
2334 | This is true on many RISC machines. On more traditional machine | |
2335 | designs, @code{__alignof__ (double)} is 4 or even 2. | |
2336 | ||
2337 | Some machines never actually require alignment; they allow reference to any | |
2338 | data type even at an odd addresses. For these machines, @code{__alignof__} | |
2339 | reports the @emph{recommended} alignment of a type. | |
2340 | ||
2341 | When the operand of @code{__alignof__} is an lvalue rather than a type, the | |
2342 | value is the largest alignment that the lvalue is known to have. It may | |
2343 | have this alignment as a result of its data type, or because it is part of | |
2344 | a structure and inherits alignment from that structure. For example, after | |
2345 | this declaration: | |
2346 | ||
2347 | @example | |
2348 | struct foo @{ int x; char y; @} foo1; | |
2349 | @end example | |
2350 | ||
2351 | @noindent | |
2352 | the value of @code{__alignof__ (foo1.y)} is probably 2 or 4, the same as | |
2353 | @code{__alignof__ (int)}, even though the data type of @code{foo1.y} | |
2354 | does not itself demand any alignment.@refill | |
2355 | ||
9d27bffe SS |
2356 | It is an error to ask for the alignment of an incomplete type. |
2357 | ||
c1f7febf RK |
2358 | A related feature which lets you specify the alignment of an object is |
2359 | @code{__attribute__ ((aligned (@var{alignment})))}; see the following | |
2360 | section. | |
2361 | ||
2362 | @node Variable Attributes | |
2363 | @section Specifying Attributes of Variables | |
2364 | @cindex attribute of variables | |
2365 | @cindex variable attributes | |
2366 | ||
2367 | The keyword @code{__attribute__} allows you to specify special | |
2368 | attributes of variables or structure fields. This keyword is followed | |
2369 | by an attribute specification inside double parentheses. Eight | |
2370 | attributes are currently defined for variables: @code{aligned}, | |
2371 | @code{mode}, @code{nocommon}, @code{packed}, @code{section}, | |
9f1bbeaa JM |
2372 | @code{transparent_union}, @code{unused}, and @code{weak}. Some other |
2373 | attributes are defined for variables on particular target systems. Other | |
c1f7febf | 2374 | attributes are available for functions (@pxref{Function Attributes}) and |
5c25e11d PE |
2375 | for types (@pxref{Type Attributes}). Other front-ends might define more |
2376 | attributes (@pxref{C++ Extensions,,Extensions to the C++ Language}). | |
c1f7febf RK |
2377 | |
2378 | You may also specify attributes with @samp{__} preceding and following | |
2379 | each keyword. This allows you to use them in header files without | |
2380 | being concerned about a possible macro of the same name. For example, | |
2381 | you may use @code{__aligned__} instead of @code{aligned}. | |
2382 | ||
2c5e91d2 JM |
2383 | @xref{Attribute Syntax}, for details of the exact syntax for using |
2384 | attributes. | |
2385 | ||
c1f7febf RK |
2386 | @table @code |
2387 | @cindex @code{aligned} attribute | |
2388 | @item aligned (@var{alignment}) | |
2389 | This attribute specifies a minimum alignment for the variable or | |
2390 | structure field, measured in bytes. For example, the declaration: | |
2391 | ||
2392 | @smallexample | |
2393 | int x __attribute__ ((aligned (16))) = 0; | |
2394 | @end smallexample | |
2395 | ||
2396 | @noindent | |
2397 | causes the compiler to allocate the global variable @code{x} on a | |
2398 | 16-byte boundary. On a 68040, this could be used in conjunction with | |
2399 | an @code{asm} expression to access the @code{move16} instruction which | |
2400 | requires 16-byte aligned operands. | |
2401 | ||
2402 | You can also specify the alignment of structure fields. For example, to | |
2403 | create a double-word aligned @code{int} pair, you could write: | |
2404 | ||
2405 | @smallexample | |
2406 | struct foo @{ int x[2] __attribute__ ((aligned (8))); @}; | |
2407 | @end smallexample | |
2408 | ||
2409 | @noindent | |
2410 | This is an alternative to creating a union with a @code{double} member | |
2411 | that forces the union to be double-word aligned. | |
2412 | ||
2413 | It is not possible to specify the alignment of functions; the alignment | |
2414 | of functions is determined by the machine's requirements and cannot be | |
2415 | changed. You cannot specify alignment for a typedef name because such a | |
2416 | name is just an alias, not a distinct type. | |
2417 | ||
2418 | As in the preceding examples, you can explicitly specify the alignment | |
2419 | (in bytes) that you wish the compiler to use for a given variable or | |
2420 | structure field. Alternatively, you can leave out the alignment factor | |
2421 | and just ask the compiler to align a variable or field to the maximum | |
2422 | useful alignment for the target machine you are compiling for. For | |
2423 | example, you could write: | |
2424 | ||
2425 | @smallexample | |
2426 | short array[3] __attribute__ ((aligned)); | |
2427 | @end smallexample | |
2428 | ||
2429 | Whenever you leave out the alignment factor in an @code{aligned} attribute | |
2430 | specification, the compiler automatically sets the alignment for the declared | |
2431 | variable or field to the largest alignment which is ever used for any data | |
2432 | type on the target machine you are compiling for. Doing this can often make | |
2433 | copy operations more efficient, because the compiler can use whatever | |
2434 | instructions copy the biggest chunks of memory when performing copies to | |
2435 | or from the variables or fields that you have aligned this way. | |
2436 | ||
2437 | The @code{aligned} attribute can only increase the alignment; but you | |
2438 | can decrease it by specifying @code{packed} as well. See below. | |
2439 | ||
2440 | Note that the effectiveness of @code{aligned} attributes may be limited | |
2441 | by inherent limitations in your linker. On many systems, the linker is | |
2442 | only able to arrange for variables to be aligned up to a certain maximum | |
2443 | alignment. (For some linkers, the maximum supported alignment may | |
2444 | be very very small.) If your linker is only able to align variables | |
2445 | up to a maximum of 8 byte alignment, then specifying @code{aligned(16)} | |
2446 | in an @code{__attribute__} will still only provide you with 8 byte | |
2447 | alignment. See your linker documentation for further information. | |
2448 | ||
2449 | @item mode (@var{mode}) | |
2450 | @cindex @code{mode} attribute | |
2451 | This attribute specifies the data type for the declaration---whichever | |
2452 | type corresponds to the mode @var{mode}. This in effect lets you | |
2453 | request an integer or floating point type according to its width. | |
2454 | ||
2455 | You may also specify a mode of @samp{byte} or @samp{__byte__} to | |
2456 | indicate the mode corresponding to a one-byte integer, @samp{word} or | |
2457 | @samp{__word__} for the mode of a one-word integer, and @samp{pointer} | |
2458 | or @samp{__pointer__} for the mode used to represent pointers. | |
2459 | ||
2460 | @item nocommon | |
2461 | @cindex @code{nocommon} attribute | |
2462 | This attribute specifies requests GNU CC not to place a variable | |
2463 | ``common'' but instead to allocate space for it directly. If you | |
2464 | specify the @samp{-fno-common} flag, GNU CC will do this for all | |
2465 | variables. | |
2466 | ||
2467 | Specifying the @code{nocommon} attribute for a variable provides an | |
2468 | initialization of zeros. A variable may only be initialized in one | |
2469 | source file. | |
2470 | ||
2471 | @item packed | |
2472 | @cindex @code{packed} attribute | |
2473 | The @code{packed} attribute specifies that a variable or structure field | |
2474 | should have the smallest possible alignment---one byte for a variable, | |
2475 | and one bit for a field, unless you specify a larger value with the | |
2476 | @code{aligned} attribute. | |
2477 | ||
2478 | Here is a structure in which the field @code{x} is packed, so that it | |
2479 | immediately follows @code{a}: | |
2480 | ||
2481 | @example | |
2482 | struct foo | |
2483 | @{ | |
2484 | char a; | |
2485 | int x[2] __attribute__ ((packed)); | |
2486 | @}; | |
2487 | @end example | |
2488 | ||
2489 | @item section ("section-name") | |
2490 | @cindex @code{section} variable attribute | |
2491 | Normally, the compiler places the objects it generates in sections like | |
2492 | @code{data} and @code{bss}. Sometimes, however, you need additional sections, | |
2493 | or you need certain particular variables to appear in special sections, | |
2494 | for example to map to special hardware. The @code{section} | |
2495 | attribute specifies that a variable (or function) lives in a particular | |
2496 | section. For example, this small program uses several specific section names: | |
2497 | ||
2498 | @smallexample | |
2499 | struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @}; | |
2500 | struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @}; | |
2501 | char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @}; | |
2502 | int init_data __attribute__ ((section ("INITDATA"))) = 0; | |
2503 | ||
2504 | main() | |
2505 | @{ | |
2506 | /* Initialize stack pointer */ | |
2507 | init_sp (stack + sizeof (stack)); | |
2508 | ||
2509 | /* Initialize initialized data */ | |
2510 | memcpy (&init_data, &data, &edata - &data); | |
2511 | ||
2512 | /* Turn on the serial ports */ | |
2513 | init_duart (&a); | |
2514 | init_duart (&b); | |
2515 | @} | |
2516 | @end smallexample | |
2517 | ||
2518 | @noindent | |
2519 | Use the @code{section} attribute with an @emph{initialized} definition | |
2520 | of a @emph{global} variable, as shown in the example. GNU CC issues | |
2521 | a warning and otherwise ignores the @code{section} attribute in | |
2522 | uninitialized variable declarations. | |
2523 | ||
2524 | You may only use the @code{section} attribute with a fully initialized | |
2525 | global definition because of the way linkers work. The linker requires | |
2526 | each object be defined once, with the exception that uninitialized | |
2527 | variables tentatively go in the @code{common} (or @code{bss}) section | |
2528 | and can be multiply "defined". You can force a variable to be | |
2529 | initialized with the @samp{-fno-common} flag or the @code{nocommon} | |
2530 | attribute. | |
2531 | ||
2532 | Some file formats do not support arbitrary sections so the @code{section} | |
2533 | attribute is not available on all platforms. | |
2534 | If you need to map the entire contents of a module to a particular | |
2535 | section, consider using the facilities of the linker instead. | |
2536 | ||
593d3a34 MK |
2537 | @item shared |
2538 | @cindex @code{shared} variable attribute | |
02f52e19 AJ |
2539 | On Windows NT, in addition to putting variable definitions in a named |
2540 | section, the section can also be shared among all running copies of an | |
2541 | executable or DLL. For example, this small program defines shared data | |
2542 | by putting it in a named section "shared" and marking the section | |
593d3a34 MK |
2543 | shareable: |
2544 | ||
2545 | @smallexample | |
2546 | int foo __attribute__((section ("shared"), shared)) = 0; | |
2547 | ||
2548 | int | |
2549 | main() | |
2550 | @{ | |
310668e8 JM |
2551 | /* Read and write foo. All running |
2552 | copies see the same value. */ | |
593d3a34 MK |
2553 | return 0; |
2554 | @} | |
2555 | @end smallexample | |
2556 | ||
2557 | @noindent | |
2558 | You may only use the @code{shared} attribute along with @code{section} | |
02f52e19 | 2559 | attribute with a fully initialized global definition because of the way |
593d3a34 MK |
2560 | linkers work. See @code{section} attribute for more information. |
2561 | ||
2562 | The @code{shared} attribute is only available on Windows NT. | |
2563 | ||
c1f7febf RK |
2564 | @item transparent_union |
2565 | This attribute, attached to a function parameter which is a union, means | |
2566 | that the corresponding argument may have the type of any union member, | |
2567 | but the argument is passed as if its type were that of the first union | |
2568 | member. For more details see @xref{Type Attributes}. You can also use | |
2569 | this attribute on a @code{typedef} for a union data type; then it | |
2570 | applies to all function parameters with that type. | |
2571 | ||
2572 | @item unused | |
2573 | This attribute, attached to a variable, means that the variable is meant | |
2574 | to be possibly unused. GNU CC will not produce a warning for this | |
2575 | variable. | |
2576 | ||
2577 | @item weak | |
2578 | The @code{weak} attribute is described in @xref{Function Attributes}. | |
845da534 DE |
2579 | |
2580 | @item model (@var{model-name}) | |
2581 | @cindex variable addressability on the M32R/D | |
2582 | Use this attribute on the M32R/D to set the addressability of an object. | |
2583 | The identifier @var{model-name} is one of @code{small}, @code{medium}, | |
2584 | or @code{large}, representing each of the code models. | |
2585 | ||
2586 | Small model objects live in the lower 16MB of memory (so that their | |
2587 | addresses can be loaded with the @code{ld24} instruction). | |
2588 | ||
02f52e19 | 2589 | Medium and large model objects may live anywhere in the 32-bit address space |
845da534 DE |
2590 | (the compiler will generate @code{seth/add3} instructions to load their |
2591 | addresses). | |
2592 | ||
c1f7febf RK |
2593 | @end table |
2594 | ||
2595 | To specify multiple attributes, separate them by commas within the | |
2596 | double parentheses: for example, @samp{__attribute__ ((aligned (16), | |
2597 | packed))}. | |
2598 | ||
2599 | @node Type Attributes | |
2600 | @section Specifying Attributes of Types | |
2601 | @cindex attribute of types | |
2602 | @cindex type attributes | |
2603 | ||
2604 | The keyword @code{__attribute__} allows you to specify special | |
2605 | attributes of @code{struct} and @code{union} types when you define such | |
2606 | types. This keyword is followed by an attribute specification inside | |
9f1bbeaa JM |
2607 | double parentheses. Four attributes are currently defined for types: |
2608 | @code{aligned}, @code{packed}, @code{transparent_union}, and @code{unused}. | |
2609 | Other attributes are defined for functions (@pxref{Function Attributes}) and | |
c1f7febf RK |
2610 | for variables (@pxref{Variable Attributes}). |
2611 | ||
2612 | You may also specify any one of these attributes with @samp{__} | |
2613 | preceding and following its keyword. This allows you to use these | |
2614 | attributes in header files without being concerned about a possible | |
2615 | macro of the same name. For example, you may use @code{__aligned__} | |
2616 | instead of @code{aligned}. | |
2617 | ||
2618 | You may specify the @code{aligned} and @code{transparent_union} | |
2619 | attributes either in a @code{typedef} declaration or just past the | |
2620 | closing curly brace of a complete enum, struct or union type | |
2621 | @emph{definition} and the @code{packed} attribute only past the closing | |
2622 | brace of a definition. | |
2623 | ||
4051959b JM |
2624 | You may also specify attributes between the enum, struct or union |
2625 | tag and the name of the type rather than after the closing brace. | |
2626 | ||
2c5e91d2 JM |
2627 | @xref{Attribute Syntax}, for details of the exact syntax for using |
2628 | attributes. | |
2629 | ||
c1f7febf RK |
2630 | @table @code |
2631 | @cindex @code{aligned} attribute | |
2632 | @item aligned (@var{alignment}) | |
2633 | This attribute specifies a minimum alignment (in bytes) for variables | |
2634 | of the specified type. For example, the declarations: | |
2635 | ||
2636 | @smallexample | |
f69eecfb JL |
2637 | struct S @{ short f[3]; @} __attribute__ ((aligned (8))); |
2638 | typedef int more_aligned_int __attribute__ ((aligned (8))); | |
c1f7febf RK |
2639 | @end smallexample |
2640 | ||
2641 | @noindent | |
d863830b | 2642 | force the compiler to insure (as far as it can) that each variable whose |
c1f7febf RK |
2643 | type is @code{struct S} or @code{more_aligned_int} will be allocated and |
2644 | aligned @emph{at least} on a 8-byte boundary. On a Sparc, having all | |
2645 | variables of type @code{struct S} aligned to 8-byte boundaries allows | |
2646 | the compiler to use the @code{ldd} and @code{std} (doubleword load and | |
2647 | store) instructions when copying one variable of type @code{struct S} to | |
2648 | another, thus improving run-time efficiency. | |
2649 | ||
2650 | Note that the alignment of any given @code{struct} or @code{union} type | |
5490d604 | 2651 | is required by the ISO C standard to be at least a perfect multiple of |
c1f7febf RK |
2652 | the lowest common multiple of the alignments of all of the members of |
2653 | the @code{struct} or @code{union} in question. This means that you @emph{can} | |
2654 | effectively adjust the alignment of a @code{struct} or @code{union} | |
2655 | type by attaching an @code{aligned} attribute to any one of the members | |
2656 | of such a type, but the notation illustrated in the example above is a | |
2657 | more obvious, intuitive, and readable way to request the compiler to | |
2658 | adjust the alignment of an entire @code{struct} or @code{union} type. | |
2659 | ||
2660 | As in the preceding example, you can explicitly specify the alignment | |
2661 | (in bytes) that you wish the compiler to use for a given @code{struct} | |
2662 | or @code{union} type. Alternatively, you can leave out the alignment factor | |
2663 | and just ask the compiler to align a type to the maximum | |
2664 | useful alignment for the target machine you are compiling for. For | |
2665 | example, you could write: | |
2666 | ||
2667 | @smallexample | |
2668 | struct S @{ short f[3]; @} __attribute__ ((aligned)); | |
2669 | @end smallexample | |
2670 | ||
2671 | Whenever you leave out the alignment factor in an @code{aligned} | |
2672 | attribute specification, the compiler automatically sets the alignment | |
2673 | for the type to the largest alignment which is ever used for any data | |
2674 | type on the target machine you are compiling for. Doing this can often | |
2675 | make copy operations more efficient, because the compiler can use | |
2676 | whatever instructions copy the biggest chunks of memory when performing | |
2677 | copies to or from the variables which have types that you have aligned | |
2678 | this way. | |
2679 | ||
2680 | In the example above, if the size of each @code{short} is 2 bytes, then | |
2681 | the size of the entire @code{struct S} type is 6 bytes. The smallest | |
2682 | power of two which is greater than or equal to that is 8, so the | |
2683 | compiler sets the alignment for the entire @code{struct S} type to 8 | |
2684 | bytes. | |
2685 | ||
2686 | Note that although you can ask the compiler to select a time-efficient | |
2687 | alignment for a given type and then declare only individual stand-alone | |
2688 | objects of that type, the compiler's ability to select a time-efficient | |
2689 | alignment is primarily useful only when you plan to create arrays of | |
2690 | variables having the relevant (efficiently aligned) type. If you | |
2691 | declare or use arrays of variables of an efficiently-aligned type, then | |
2692 | it is likely that your program will also be doing pointer arithmetic (or | |
2693 | subscripting, which amounts to the same thing) on pointers to the | |
2694 | relevant type, and the code that the compiler generates for these | |
2695 | pointer arithmetic operations will often be more efficient for | |
2696 | efficiently-aligned types than for other types. | |
2697 | ||
2698 | The @code{aligned} attribute can only increase the alignment; but you | |
2699 | can decrease it by specifying @code{packed} as well. See below. | |
2700 | ||
2701 | Note that the effectiveness of @code{aligned} attributes may be limited | |
2702 | by inherent limitations in your linker. On many systems, the linker is | |
2703 | only able to arrange for variables to be aligned up to a certain maximum | |
2704 | alignment. (For some linkers, the maximum supported alignment may | |
2705 | be very very small.) If your linker is only able to align variables | |
2706 | up to a maximum of 8 byte alignment, then specifying @code{aligned(16)} | |
2707 | in an @code{__attribute__} will still only provide you with 8 byte | |
2708 | alignment. See your linker documentation for further information. | |
2709 | ||
2710 | @item packed | |
2711 | This attribute, attached to an @code{enum}, @code{struct}, or | |
2712 | @code{union} type definition, specified that the minimum required memory | |
2713 | be used to represent the type. | |
2714 | ||
2715 | Specifying this attribute for @code{struct} and @code{union} types is | |
2716 | equivalent to specifying the @code{packed} attribute on each of the | |
2717 | structure or union members. Specifying the @samp{-fshort-enums} | |
2718 | flag on the line is equivalent to specifying the @code{packed} | |
2719 | attribute on all @code{enum} definitions. | |
2720 | ||
2721 | You may only specify this attribute after a closing curly brace on an | |
1cd4bca9 BK |
2722 | @code{enum} definition, not in a @code{typedef} declaration, unless that |
2723 | declaration also contains the definition of the @code{enum}. | |
c1f7febf RK |
2724 | |
2725 | @item transparent_union | |
2726 | This attribute, attached to a @code{union} type definition, indicates | |
2727 | that any function parameter having that union type causes calls to that | |
2728 | function to be treated in a special way. | |
2729 | ||
2730 | First, the argument corresponding to a transparent union type can be of | |
2731 | any type in the union; no cast is required. Also, if the union contains | |
2732 | a pointer type, the corresponding argument can be a null pointer | |
2733 | constant or a void pointer expression; and if the union contains a void | |
2734 | pointer type, the corresponding argument can be any pointer expression. | |
2735 | If the union member type is a pointer, qualifiers like @code{const} on | |
2736 | the referenced type must be respected, just as with normal pointer | |
2737 | conversions. | |
2738 | ||
2739 | Second, the argument is passed to the function using the calling | |
2740 | conventions of first member of the transparent union, not the calling | |
2741 | conventions of the union itself. All members of the union must have the | |
2742 | same machine representation; this is necessary for this argument passing | |
2743 | to work properly. | |
2744 | ||
2745 | Transparent unions are designed for library functions that have multiple | |
2746 | interfaces for compatibility reasons. For example, suppose the | |
2747 | @code{wait} function must accept either a value of type @code{int *} to | |
2748 | comply with Posix, or a value of type @code{union wait *} to comply with | |
2749 | the 4.1BSD interface. If @code{wait}'s parameter were @code{void *}, | |
2750 | @code{wait} would accept both kinds of arguments, but it would also | |
2751 | accept any other pointer type and this would make argument type checking | |
2752 | less useful. Instead, @code{<sys/wait.h>} might define the interface | |
2753 | as follows: | |
2754 | ||
2755 | @smallexample | |
2756 | typedef union | |
2757 | @{ | |
2758 | int *__ip; | |
2759 | union wait *__up; | |
2760 | @} wait_status_ptr_t __attribute__ ((__transparent_union__)); | |
2761 | ||
2762 | pid_t wait (wait_status_ptr_t); | |
2763 | @end smallexample | |
2764 | ||
2765 | This interface allows either @code{int *} or @code{union wait *} | |
2766 | arguments to be passed, using the @code{int *} calling convention. | |
2767 | The program can call @code{wait} with arguments of either type: | |
2768 | ||
2769 | @example | |
2770 | int w1 () @{ int w; return wait (&w); @} | |
2771 | int w2 () @{ union wait w; return wait (&w); @} | |
2772 | @end example | |
2773 | ||
2774 | With this interface, @code{wait}'s implementation might look like this: | |
2775 | ||
2776 | @example | |
2777 | pid_t wait (wait_status_ptr_t p) | |
2778 | @{ | |
2779 | return waitpid (-1, p.__ip, 0); | |
2780 | @} | |
2781 | @end example | |
d863830b JL |
2782 | |
2783 | @item unused | |
2784 | When attached to a type (including a @code{union} or a @code{struct}), | |
2785 | this attribute means that variables of that type are meant to appear | |
2786 | possibly unused. GNU CC will not produce a warning for any variables of | |
2787 | that type, even if the variable appears to do nothing. This is often | |
2788 | the case with lock or thread classes, which are usually defined and then | |
2789 | not referenced, but contain constructors and destructors that have | |
956d6950 | 2790 | nontrivial bookkeeping functions. |
d863830b | 2791 | |
c1f7febf RK |
2792 | @end table |
2793 | ||
2794 | To specify multiple attributes, separate them by commas within the | |
2795 | double parentheses: for example, @samp{__attribute__ ((aligned (16), | |
2796 | packed))}. | |
2797 | ||
2798 | @node Inline | |
2799 | @section An Inline Function is As Fast As a Macro | |
2800 | @cindex inline functions | |
2801 | @cindex integrating function code | |
2802 | @cindex open coding | |
2803 | @cindex macros, inline alternative | |
2804 | ||
2805 | By declaring a function @code{inline}, you can direct GNU CC to | |
2806 | integrate that function's code into the code for its callers. This | |
2807 | makes execution faster by eliminating the function-call overhead; in | |
2808 | addition, if any of the actual argument values are constant, their known | |
2809 | values may permit simplifications at compile time so that not all of the | |
2810 | inline function's code needs to be included. The effect on code size is | |
2811 | less predictable; object code may be larger or smaller with function | |
2812 | inlining, depending on the particular case. Inlining of functions is an | |
2813 | optimization and it really ``works'' only in optimizing compilation. If | |
2814 | you don't use @samp{-O}, no function is really inline. | |
2815 | ||
4b404517 JM |
2816 | Inline functions are included in the ISO C99 standard, but there are |
2817 | currently substantial differences between what GCC implements and what | |
2818 | the ISO C99 standard requires. | |
2819 | ||
c1f7febf RK |
2820 | To declare a function inline, use the @code{inline} keyword in its |
2821 | declaration, like this: | |
2822 | ||
2823 | @example | |
2824 | inline int | |
2825 | inc (int *a) | |
2826 | @{ | |
2827 | (*a)++; | |
2828 | @} | |
2829 | @end example | |
2830 | ||
5490d604 | 2831 | (If you are writing a header file to be included in ISO C programs, write |
c1f7febf | 2832 | @code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}.) |
c1f7febf | 2833 | You can also make all ``simple enough'' functions inline with the option |
02f52e19 | 2834 | @samp{-finline-functions}. |
247b14bd RH |
2835 | |
2836 | Note that certain usages in a function definition can make it unsuitable | |
2837 | for inline substitution. Among these usages are: use of varargs, use of | |
2838 | alloca, use of variable sized data types (@pxref{Variable Length}), | |
2839 | use of computed goto (@pxref{Labels as Values}), use of nonlocal goto, | |
2840 | and nested functions (@pxref{Nested Functions}). Using @samp{-Winline} | |
2841 | will warn when a function marked @code{inline} could not be substituted, | |
2842 | and will give the reason for the failure. | |
c1f7febf RK |
2843 | |
2844 | Note that in C and Objective C, unlike C++, the @code{inline} keyword | |
2845 | does not affect the linkage of the function. | |
2846 | ||
2847 | @cindex automatic @code{inline} for C++ member fns | |
2848 | @cindex @code{inline} automatic for C++ member fns | |
2849 | @cindex member fns, automatically @code{inline} | |
2850 | @cindex C++ member fns, automatically @code{inline} | |
2851 | GNU CC automatically inlines member functions defined within the class | |
2852 | body of C++ programs even if they are not explicitly declared | |
2853 | @code{inline}. (You can override this with @samp{-fno-default-inline}; | |
2854 | @pxref{C++ Dialect Options,,Options Controlling C++ Dialect}.) | |
2855 | ||
2856 | @cindex inline functions, omission of | |
2857 | When a function is both inline and @code{static}, if all calls to the | |
2858 | function are integrated into the caller, and the function's address is | |
2859 | never used, then the function's own assembler code is never referenced. | |
2860 | In this case, GNU CC does not actually output assembler code for the | |
2861 | function, unless you specify the option @samp{-fkeep-inline-functions}. | |
2862 | Some calls cannot be integrated for various reasons (in particular, | |
2863 | calls that precede the function's definition cannot be integrated, and | |
2864 | neither can recursive calls within the definition). If there is a | |
2865 | nonintegrated call, then the function is compiled to assembler code as | |
2866 | usual. The function must also be compiled as usual if the program | |
2867 | refers to its address, because that can't be inlined. | |
2868 | ||
2869 | @cindex non-static inline function | |
2870 | When an inline function is not @code{static}, then the compiler must assume | |
2871 | that there may be calls from other source files; since a global symbol can | |
2872 | be defined only once in any program, the function must not be defined in | |
2873 | the other source files, so the calls therein cannot be integrated. | |
2874 | Therefore, a non-@code{static} inline function is always compiled on its | |
2875 | own in the usual fashion. | |
2876 | ||
2877 | If you specify both @code{inline} and @code{extern} in the function | |
2878 | definition, then the definition is used only for inlining. In no case | |
2879 | is the function compiled on its own, not even if you refer to its | |
2880 | address explicitly. Such an address becomes an external reference, as | |
2881 | if you had only declared the function, and had not defined it. | |
2882 | ||
2883 | This combination of @code{inline} and @code{extern} has almost the | |
2884 | effect of a macro. The way to use it is to put a function definition in | |
2885 | a header file with these keywords, and put another copy of the | |
2886 | definition (lacking @code{inline} and @code{extern}) in a library file. | |
2887 | The definition in the header file will cause most calls to the function | |
2888 | to be inlined. If any uses of the function remain, they will refer to | |
2889 | the single copy in the library. | |
2890 | ||
4b404517 JM |
2891 | For future compatibility with when GCC implements ISO C99 semantics for |
2892 | inline functions, it is best to use @code{static inline} only. (The | |
2893 | existing semantics will remain available when @option{-std=gnu89} is | |
2894 | specified, but eventually the default will be @option{-std=gnu99} and | |
2895 | that will implement the C99 semantics, though it does not do so yet.) | |
2896 | ||
c1f7febf RK |
2897 | GNU C does not inline any functions when not optimizing. It is not |
2898 | clear whether it is better to inline or not, in this case, but we found | |
2899 | that a correct implementation when not optimizing was difficult. So we | |
2900 | did the easy thing, and turned it off. | |
2901 | ||
2902 | @node Extended Asm | |
2903 | @section Assembler Instructions with C Expression Operands | |
2904 | @cindex extended @code{asm} | |
2905 | @cindex @code{asm} expressions | |
2906 | @cindex assembler instructions | |
2907 | @cindex registers | |
2908 | ||
c85f7c16 JL |
2909 | In an assembler instruction using @code{asm}, you can specify the |
2910 | operands of the instruction using C expressions. This means you need not | |
2911 | guess which registers or memory locations will contain the data you want | |
c1f7febf RK |
2912 | to use. |
2913 | ||
c85f7c16 JL |
2914 | You must specify an assembler instruction template much like what |
2915 | appears in a machine description, plus an operand constraint string for | |
2916 | each operand. | |
c1f7febf RK |
2917 | |
2918 | For example, here is how to use the 68881's @code{fsinx} instruction: | |
2919 | ||
2920 | @example | |
2921 | asm ("fsinx %1,%0" : "=f" (result) : "f" (angle)); | |
2922 | @end example | |
2923 | ||
2924 | @noindent | |
2925 | Here @code{angle} is the C expression for the input operand while | |
2926 | @code{result} is that of the output operand. Each has @samp{"f"} as its | |
c85f7c16 JL |
2927 | operand constraint, saying that a floating point register is required. |
2928 | The @samp{=} in @samp{=f} indicates that the operand is an output; all | |
2929 | output operands' constraints must use @samp{=}. The constraints use the | |
2930 | same language used in the machine description (@pxref{Constraints}). | |
2931 | ||
2932 | Each operand is described by an operand-constraint string followed by | |
2933 | the C expression in parentheses. A colon separates the assembler | |
2934 | template from the first output operand and another separates the last | |
2935 | output operand from the first input, if any. Commas separate the | |
2936 | operands within each group. The total number of operands is limited to | |
2937 | ten or to the maximum number of operands in any instruction pattern in | |
2938 | the machine description, whichever is greater. | |
2939 | ||
2940 | If there are no output operands but there are input operands, you must | |
2941 | place two consecutive colons surrounding the place where the output | |
c1f7febf RK |
2942 | operands would go. |
2943 | ||
2944 | Output operand expressions must be lvalues; the compiler can check this. | |
c85f7c16 JL |
2945 | The input operands need not be lvalues. The compiler cannot check |
2946 | whether the operands have data types that are reasonable for the | |
2947 | instruction being executed. It does not parse the assembler instruction | |
2948 | template and does not know what it means or even whether it is valid | |
2949 | assembler input. The extended @code{asm} feature is most often used for | |
2950 | machine instructions the compiler itself does not know exist. If | |
2951 | the output expression cannot be directly addressed (for example, it is a | |
2952 | bit field), your constraint must allow a register. In that case, GNU CC | |
2953 | will use the register as the output of the @code{asm}, and then store | |
2954 | that register into the output. | |
2955 | ||
2956 | The ordinary output operands must be write-only; GNU CC will assume that | |
2957 | the values in these operands before the instruction are dead and need | |
2958 | not be generated. Extended asm supports input-output or read-write | |
2959 | operands. Use the constraint character @samp{+} to indicate such an | |
2960 | operand and list it with the output operands. | |
2961 | ||
2962 | When the constraints for the read-write operand (or the operand in which | |
2963 | only some of the bits are to be changed) allows a register, you may, as | |
2964 | an alternative, logically split its function into two separate operands, | |
2965 | one input operand and one write-only output operand. The connection | |
2966 | between them is expressed by constraints which say they need to be in | |
2967 | the same location when the instruction executes. You can use the same C | |
2968 | expression for both operands, or different expressions. For example, | |
2969 | here we write the (fictitious) @samp{combine} instruction with | |
2970 | @code{bar} as its read-only source operand and @code{foo} as its | |
2971 | read-write destination: | |
c1f7febf RK |
2972 | |
2973 | @example | |
2974 | asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar)); | |
2975 | @end example | |
2976 | ||
2977 | @noindent | |
c85f7c16 JL |
2978 | The constraint @samp{"0"} for operand 1 says that it must occupy the |
2979 | same location as operand 0. A digit in constraint is allowed only in an | |
2980 | input operand and it must refer to an output operand. | |
c1f7febf RK |
2981 | |
2982 | Only a digit in the constraint can guarantee that one operand will be in | |
c85f7c16 JL |
2983 | the same place as another. The mere fact that @code{foo} is the value |
2984 | of both operands is not enough to guarantee that they will be in the | |
2985 | same place in the generated assembler code. The following would not | |
2986 | work reliably: | |
c1f7febf RK |
2987 | |
2988 | @example | |
2989 | asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar)); | |
2990 | @end example | |
2991 | ||
2992 | Various optimizations or reloading could cause operands 0 and 1 to be in | |
2993 | different registers; GNU CC knows no reason not to do so. For example, the | |
2994 | compiler might find a copy of the value of @code{foo} in one register and | |
2995 | use it for operand 1, but generate the output operand 0 in a different | |
2996 | register (copying it afterward to @code{foo}'s own address). Of course, | |
2997 | since the register for operand 1 is not even mentioned in the assembler | |
2998 | code, the result will not work, but GNU CC can't tell that. | |
2999 | ||
c85f7c16 JL |
3000 | Some instructions clobber specific hard registers. To describe this, |
3001 | write a third colon after the input operands, followed by the names of | |
3002 | the clobbered hard registers (given as strings). Here is a realistic | |
3003 | example for the VAX: | |
c1f7febf RK |
3004 | |
3005 | @example | |
3006 | asm volatile ("movc3 %0,%1,%2" | |
3007 | : /* no outputs */ | |
3008 | : "g" (from), "g" (to), "g" (count) | |
3009 | : "r0", "r1", "r2", "r3", "r4", "r5"); | |
3010 | @end example | |
3011 | ||
c5c76735 JL |
3012 | You may not write a clobber description in a way that overlaps with an |
3013 | input or output operand. For example, you may not have an operand | |
3014 | describing a register class with one member if you mention that register | |
3015 | in the clobber list. There is no way for you to specify that an input | |
3016 | operand is modified without also specifying it as an output | |
3017 | operand. Note that if all the output operands you specify are for this | |
3018 | purpose (and hence unused), you will then also need to specify | |
3019 | @code{volatile} for the @code{asm} construct, as described below, to | |
3020 | prevent GNU CC from deleting the @code{asm} statement as unused. | |
8fe1938e | 3021 | |
c1f7febf | 3022 | If you refer to a particular hardware register from the assembler code, |
c85f7c16 JL |
3023 | you will probably have to list the register after the third colon to |
3024 | tell the compiler the register's value is modified. In some assemblers, | |
3025 | the register names begin with @samp{%}; to produce one @samp{%} in the | |
3026 | assembler code, you must write @samp{%%} in the input. | |
3027 | ||
3028 | If your assembler instruction can alter the condition code register, add | |
3029 | @samp{cc} to the list of clobbered registers. GNU CC on some machines | |
3030 | represents the condition codes as a specific hardware register; | |
3031 | @samp{cc} serves to name this register. On other machines, the | |
3032 | condition code is handled differently, and specifying @samp{cc} has no | |
3033 | effect. But it is valid no matter what the machine. | |
c1f7febf RK |
3034 | |
3035 | If your assembler instruction modifies memory in an unpredictable | |
c85f7c16 JL |
3036 | fashion, add @samp{memory} to the list of clobbered registers. This |
3037 | will cause GNU CC to not keep memory values cached in registers across | |
dd40655a GK |
3038 | the assembler instruction. You will also want to add the |
3039 | @code{volatile} keyword if the memory affected is not listed in the | |
3040 | inputs or outputs of the @code{asm}, as the @samp{memory} clobber does | |
3041 | not count as a side-effect of the @code{asm}. | |
c1f7febf | 3042 | |
c85f7c16 | 3043 | You can put multiple assembler instructions together in a single |
8720914b HPN |
3044 | @code{asm} template, separated by the characters normally used in assembly |
3045 | code for the system. A combination that works in most places is a newline | |
3046 | to break the line, plus a tab character to move to the instruction field | |
3047 | (written as @samp{\n\t}). Sometimes semicolons can be used, if the | |
3048 | assembler allows semicolons as a line-breaking character. Note that some | |
3049 | assembler dialects use semicolons to start a comment. | |
3050 | The input operands are guaranteed not to use any of the clobbered | |
c85f7c16 JL |
3051 | registers, and neither will the output operands' addresses, so you can |
3052 | read and write the clobbered registers as many times as you like. Here | |
3053 | is an example of multiple instructions in a template; it assumes the | |
3054 | subroutine @code{_foo} accepts arguments in registers 9 and 10: | |
c1f7febf RK |
3055 | |
3056 | @example | |
8720914b | 3057 | asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo" |
c1f7febf RK |
3058 | : /* no outputs */ |
3059 | : "g" (from), "g" (to) | |
3060 | : "r9", "r10"); | |
3061 | @end example | |
3062 | ||
c85f7c16 JL |
3063 | Unless an output operand has the @samp{&} constraint modifier, GNU CC |
3064 | may allocate it in the same register as an unrelated input operand, on | |
3065 | the assumption the inputs are consumed before the outputs are produced. | |
c1f7febf RK |
3066 | This assumption may be false if the assembler code actually consists of |
3067 | more than one instruction. In such a case, use @samp{&} for each output | |
c85f7c16 | 3068 | operand that may not overlap an input. @xref{Modifiers}. |
c1f7febf | 3069 | |
c85f7c16 JL |
3070 | If you want to test the condition code produced by an assembler |
3071 | instruction, you must include a branch and a label in the @code{asm} | |
3072 | construct, as follows: | |
c1f7febf RK |
3073 | |
3074 | @example | |
8720914b | 3075 | asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:" |
c1f7febf RK |
3076 | : "g" (result) |
3077 | : "g" (input)); | |
3078 | @end example | |
3079 | ||
3080 | @noindent | |
3081 | This assumes your assembler supports local labels, as the GNU assembler | |
3082 | and most Unix assemblers do. | |
3083 | ||
3084 | Speaking of labels, jumps from one @code{asm} to another are not | |
c85f7c16 JL |
3085 | supported. The compiler's optimizers do not know about these jumps, and |
3086 | therefore they cannot take account of them when deciding how to | |
c1f7febf RK |
3087 | optimize. |
3088 | ||
3089 | @cindex macros containing @code{asm} | |
3090 | Usually the most convenient way to use these @code{asm} instructions is to | |
3091 | encapsulate them in macros that look like functions. For example, | |
3092 | ||
3093 | @example | |
3094 | #define sin(x) \ | |
3095 | (@{ double __value, __arg = (x); \ | |
3096 | asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \ | |
3097 | __value; @}) | |
3098 | @end example | |
3099 | ||
3100 | @noindent | |
3101 | Here the variable @code{__arg} is used to make sure that the instruction | |
3102 | operates on a proper @code{double} value, and to accept only those | |
3103 | arguments @code{x} which can convert automatically to a @code{double}. | |
3104 | ||
c85f7c16 JL |
3105 | Another way to make sure the instruction operates on the correct data |
3106 | type is to use a cast in the @code{asm}. This is different from using a | |
c1f7febf RK |
3107 | variable @code{__arg} in that it converts more different types. For |
3108 | example, if the desired type were @code{int}, casting the argument to | |
3109 | @code{int} would accept a pointer with no complaint, while assigning the | |
3110 | argument to an @code{int} variable named @code{__arg} would warn about | |
3111 | using a pointer unless the caller explicitly casts it. | |
3112 | ||
3113 | If an @code{asm} has output operands, GNU CC assumes for optimization | |
c85f7c16 JL |
3114 | purposes the instruction has no side effects except to change the output |
3115 | operands. This does not mean instructions with a side effect cannot be | |
3116 | used, but you must be careful, because the compiler may eliminate them | |
3117 | if the output operands aren't used, or move them out of loops, or | |
3118 | replace two with one if they constitute a common subexpression. Also, | |
3119 | if your instruction does have a side effect on a variable that otherwise | |
3120 | appears not to change, the old value of the variable may be reused later | |
3121 | if it happens to be found in a register. | |
c1f7febf RK |
3122 | |
3123 | You can prevent an @code{asm} instruction from being deleted, moved | |
3124 | significantly, or combined, by writing the keyword @code{volatile} after | |
3125 | the @code{asm}. For example: | |
3126 | ||
3127 | @example | |
310668e8 JM |
3128 | #define get_and_set_priority(new) \ |
3129 | (@{ int __old; \ | |
3130 | asm volatile ("get_and_set_priority %0, %1" \ | |
3131 | : "=g" (__old) : "g" (new)); \ | |
c85f7c16 | 3132 | __old; @}) |
24f98470 | 3133 | @end example |
c1f7febf RK |
3134 | |
3135 | @noindent | |
c85f7c16 JL |
3136 | If you write an @code{asm} instruction with no outputs, GNU CC will know |
3137 | the instruction has side-effects and will not delete the instruction or | |
e71b34aa | 3138 | move it outside of loops. |
c85f7c16 | 3139 | |
e71b34aa MM |
3140 | The @code{volatile} keyword indicates that the instruction has |
3141 | important side-effects. GCC will not delete a volatile @code{asm} if | |
3142 | it is reachable. (The instruction can still be deleted if GCC can | |
3143 | prove that control-flow will never reach the location of the | |
3144 | instruction.) In addition, GCC will not reschedule instructions | |
3145 | across a volatile @code{asm} instruction. For example: | |
3146 | ||
3147 | @example | |
bd78000b | 3148 | *(volatile int *)addr = foo; |
e71b34aa MM |
3149 | asm volatile ("eieio" : : ); |
3150 | @end example | |
3151 | ||
ebb48a4d | 3152 | @noindent |
e71b34aa MM |
3153 | Assume @code{addr} contains the address of a memory mapped device |
3154 | register. The PowerPC @code{eieio} instruction (Enforce In-order | |
3155 | Execution of I/O) tells the cpu to make sure that the store to that | |
3156 | device register happens before it issues any other I/O. | |
c1f7febf RK |
3157 | |
3158 | Note that even a volatile @code{asm} instruction can be moved in ways | |
3159 | that appear insignificant to the compiler, such as across jump | |
3160 | instructions. You can't expect a sequence of volatile @code{asm} | |
3161 | instructions to remain perfectly consecutive. If you want consecutive | |
e71b34aa MM |
3162 | output, use a single @code{asm}. Also, GCC will perform some |
3163 | optimizations across a volatile @code{asm} instruction; GCC does not | |
3164 | ``forget everything'' when it encounters a volatile @code{asm} | |
3165 | instruction the way some other compilers do. | |
3166 | ||
3167 | An @code{asm} instruction without any operands or clobbers (an ``old | |
3168 | style'' @code{asm}) will be treated identically to a volatile | |
3169 | @code{asm} instruction. | |
c1f7febf RK |
3170 | |
3171 | It is a natural idea to look for a way to give access to the condition | |
3172 | code left by the assembler instruction. However, when we attempted to | |
3173 | implement this, we found no way to make it work reliably. The problem | |
3174 | is that output operands might need reloading, which would result in | |
3175 | additional following ``store'' instructions. On most machines, these | |
3176 | instructions would alter the condition code before there was time to | |
3177 | test it. This problem doesn't arise for ordinary ``test'' and | |
3178 | ``compare'' instructions because they don't have any output operands. | |
3179 | ||
eda3fbbe GB |
3180 | For reasons similar to those described above, it is not possible to give |
3181 | an assembler instruction access to the condition code left by previous | |
3182 | instructions. | |
3183 | ||
5490d604 | 3184 | If you are writing a header file that should be includable in ISO C |
c1f7febf RK |
3185 | programs, write @code{__asm__} instead of @code{asm}. @xref{Alternate |
3186 | Keywords}. | |
3187 | ||
fe0ce426 JH |
3188 | @subsection i386 floating point asm operands |
3189 | ||
3190 | There are several rules on the usage of stack-like regs in | |
3191 | asm_operands insns. These rules apply only to the operands that are | |
3192 | stack-like regs: | |
3193 | ||
3194 | @enumerate | |
3195 | @item | |
3196 | Given a set of input regs that die in an asm_operands, it is | |
3197 | necessary to know which are implicitly popped by the asm, and | |
3198 | which must be explicitly popped by gcc. | |
3199 | ||
3200 | An input reg that is implicitly popped by the asm must be | |
3201 | explicitly clobbered, unless it is constrained to match an | |
3202 | output operand. | |
3203 | ||
3204 | @item | |
3205 | For any input reg that is implicitly popped by an asm, it is | |
3206 | necessary to know how to adjust the stack to compensate for the pop. | |
3207 | If any non-popped input is closer to the top of the reg-stack than | |
3208 | the implicitly popped reg, it would not be possible to know what the | |
3209 | stack looked like --- it's not clear how the rest of the stack ``slides | |
3210 | up''. | |
3211 | ||
3212 | All implicitly popped input regs must be closer to the top of | |
3213 | the reg-stack than any input that is not implicitly popped. | |
3214 | ||
3215 | It is possible that if an input dies in an insn, reload might | |
3216 | use the input reg for an output reload. Consider this example: | |
3217 | ||
3218 | @example | |
3219 | asm ("foo" : "=t" (a) : "f" (b)); | |
3220 | @end example | |
3221 | ||
3222 | This asm says that input B is not popped by the asm, and that | |
3223 | the asm pushes a result onto the reg-stack, ie, the stack is one | |
3224 | deeper after the asm than it was before. But, it is possible that | |
3225 | reload will think that it can use the same reg for both the input and | |
3226 | the output, if input B dies in this insn. | |
3227 | ||
3228 | If any input operand uses the @code{f} constraint, all output reg | |
3229 | constraints must use the @code{&} earlyclobber. | |
3230 | ||
3231 | The asm above would be written as | |
3232 | ||
3233 | @example | |
3234 | asm ("foo" : "=&t" (a) : "f" (b)); | |
3235 | @end example | |
3236 | ||
3237 | @item | |
3238 | Some operands need to be in particular places on the stack. All | |
3239 | output operands fall in this category --- there is no other way to | |
3240 | know which regs the outputs appear in unless the user indicates | |
3241 | this in the constraints. | |
3242 | ||
3243 | Output operands must specifically indicate which reg an output | |
3244 | appears in after an asm. @code{=f} is not allowed: the operand | |
3245 | constraints must select a class with a single reg. | |
3246 | ||
3247 | @item | |
3248 | Output operands may not be ``inserted'' between existing stack regs. | |
3249 | Since no 387 opcode uses a read/write operand, all output operands | |
3250 | are dead before the asm_operands, and are pushed by the asm_operands. | |
3251 | It makes no sense to push anywhere but the top of the reg-stack. | |
3252 | ||
3253 | Output operands must start at the top of the reg-stack: output | |
3254 | operands may not ``skip'' a reg. | |
3255 | ||
3256 | @item | |
3257 | Some asm statements may need extra stack space for internal | |
3258 | calculations. This can be guaranteed by clobbering stack registers | |
3259 | unrelated to the inputs and outputs. | |
3260 | ||
3261 | @end enumerate | |
3262 | ||
3263 | Here are a couple of reasonable asms to want to write. This asm | |
3264 | takes one input, which is internally popped, and produces two outputs. | |
3265 | ||
3266 | @example | |
3267 | asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp)); | |
3268 | @end example | |
3269 | ||
3270 | This asm takes two inputs, which are popped by the @code{fyl2xp1} opcode, | |
3271 | and replaces them with one output. The user must code the @code{st(1)} | |
3272 | clobber for reg-stack.c to know that @code{fyl2xp1} pops both inputs. | |
3273 | ||
3274 | @example | |
3275 | asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)"); | |
3276 | @end example | |
3277 | ||
c1f7febf RK |
3278 | @ifclear INTERNALS |
3279 | @c Show the details on constraints if they do not appear elsewhere in | |
3280 | @c the manual | |
3281 | @include md.texi | |
3282 | @end ifclear | |
3283 | ||
3284 | @node Asm Labels | |
3285 | @section Controlling Names Used in Assembler Code | |
3286 | @cindex assembler names for identifiers | |
3287 | @cindex names used in assembler code | |
3288 | @cindex identifiers, names in assembler code | |
3289 | ||
3290 | You can specify the name to be used in the assembler code for a C | |
3291 | function or variable by writing the @code{asm} (or @code{__asm__}) | |
3292 | keyword after the declarator as follows: | |
3293 | ||
3294 | @example | |
3295 | int foo asm ("myfoo") = 2; | |
3296 | @end example | |
3297 | ||
3298 | @noindent | |
3299 | This specifies that the name to be used for the variable @code{foo} in | |
3300 | the assembler code should be @samp{myfoo} rather than the usual | |
3301 | @samp{_foo}. | |
3302 | ||
3303 | On systems where an underscore is normally prepended to the name of a C | |
3304 | function or variable, this feature allows you to define names for the | |
3305 | linker that do not start with an underscore. | |
3306 | ||
0adc3c19 MM |
3307 | It does not make sense to use this feature with a non-static local |
3308 | variable since such variables do not have assembler names. If you are | |
3309 | trying to put the variable in a particular register, see @ref{Explicit | |
3310 | Reg Vars}. GCC presently accepts such code with a warning, but will | |
3311 | probably be changed to issue an error, rather than a warning, in the | |
3312 | future. | |
3313 | ||
c1f7febf RK |
3314 | You cannot use @code{asm} in this way in a function @emph{definition}; but |
3315 | you can get the same effect by writing a declaration for the function | |
3316 | before its definition and putting @code{asm} there, like this: | |
3317 | ||
3318 | @example | |
3319 | extern func () asm ("FUNC"); | |
3320 | ||
3321 | func (x, y) | |
3322 | int x, y; | |
3323 | @dots{} | |
3324 | @end example | |
3325 | ||
3326 | It is up to you to make sure that the assembler names you choose do not | |
3327 | conflict with any other assembler symbols. Also, you must not use a | |
3328 | register name; that would produce completely invalid assembler code. GNU | |
3329 | CC does not as yet have the ability to store static variables in registers. | |
3330 | Perhaps that will be added. | |
3331 | ||
3332 | @node Explicit Reg Vars | |
3333 | @section Variables in Specified Registers | |
3334 | @cindex explicit register variables | |
3335 | @cindex variables in specified registers | |
3336 | @cindex specified registers | |
3337 | @cindex registers, global allocation | |
3338 | ||
3339 | GNU C allows you to put a few global variables into specified hardware | |
3340 | registers. You can also specify the register in which an ordinary | |
3341 | register variable should be allocated. | |
3342 | ||
3343 | @itemize @bullet | |
3344 | @item | |
3345 | Global register variables reserve registers throughout the program. | |
3346 | This may be useful in programs such as programming language | |
3347 | interpreters which have a couple of global variables that are accessed | |
3348 | very often. | |
3349 | ||
3350 | @item | |
3351 | Local register variables in specific registers do not reserve the | |
3352 | registers. The compiler's data flow analysis is capable of determining | |
3353 | where the specified registers contain live values, and where they are | |
8d344fbc | 3354 | available for other uses. Stores into local register variables may be deleted |
0deaf590 JL |
3355 | when they appear to be dead according to dataflow analysis. References |
3356 | to local register variables may be deleted or moved or simplified. | |
c1f7febf RK |
3357 | |
3358 | These local variables are sometimes convenient for use with the extended | |
3359 | @code{asm} feature (@pxref{Extended Asm}), if you want to write one | |
3360 | output of the assembler instruction directly into a particular register. | |
3361 | (This will work provided the register you specify fits the constraints | |
3362 | specified for that operand in the @code{asm}.) | |
3363 | @end itemize | |
3364 | ||
3365 | @menu | |
3366 | * Global Reg Vars:: | |
3367 | * Local Reg Vars:: | |
3368 | @end menu | |
3369 | ||
3370 | @node Global Reg Vars | |
3371 | @subsection Defining Global Register Variables | |
3372 | @cindex global register variables | |
3373 | @cindex registers, global variables in | |
3374 | ||
3375 | You can define a global register variable in GNU C like this: | |
3376 | ||
3377 | @example | |
3378 | register int *foo asm ("a5"); | |
3379 | @end example | |
3380 | ||
3381 | @noindent | |
3382 | Here @code{a5} is the name of the register which should be used. Choose a | |
3383 | register which is normally saved and restored by function calls on your | |
3384 | machine, so that library routines will not clobber it. | |
3385 | ||
3386 | Naturally the register name is cpu-dependent, so you would need to | |
3387 | conditionalize your program according to cpu type. The register | |
3388 | @code{a5} would be a good choice on a 68000 for a variable of pointer | |
3389 | type. On machines with register windows, be sure to choose a ``global'' | |
3390 | register that is not affected magically by the function call mechanism. | |
3391 | ||
3392 | In addition, operating systems on one type of cpu may differ in how they | |
3393 | name the registers; then you would need additional conditionals. For | |
3394 | example, some 68000 operating systems call this register @code{%a5}. | |
3395 | ||
3396 | Eventually there may be a way of asking the compiler to choose a register | |
3397 | automatically, but first we need to figure out how it should choose and | |
3398 | how to enable you to guide the choice. No solution is evident. | |
3399 | ||
3400 | Defining a global register variable in a certain register reserves that | |
3401 | register entirely for this use, at least within the current compilation. | |
3402 | The register will not be allocated for any other purpose in the functions | |
3403 | in the current compilation. The register will not be saved and restored by | |
3404 | these functions. Stores into this register are never deleted even if they | |
3405 | would appear to be dead, but references may be deleted or moved or | |
3406 | simplified. | |
3407 | ||
3408 | It is not safe to access the global register variables from signal | |
3409 | handlers, or from more than one thread of control, because the system | |
3410 | library routines may temporarily use the register for other things (unless | |
3411 | you recompile them specially for the task at hand). | |
3412 | ||
3413 | @cindex @code{qsort}, and global register variables | |
3414 | It is not safe for one function that uses a global register variable to | |
3415 | call another such function @code{foo} by way of a third function | |
3416 | @code{lose} that was compiled without knowledge of this variable (i.e. in a | |
3417 | different source file in which the variable wasn't declared). This is | |
3418 | because @code{lose} might save the register and put some other value there. | |
3419 | For example, you can't expect a global register variable to be available in | |
3420 | the comparison-function that you pass to @code{qsort}, since @code{qsort} | |
3421 | might have put something else in that register. (If you are prepared to | |
3422 | recompile @code{qsort} with the same global register variable, you can | |
3423 | solve this problem.) | |
3424 | ||
3425 | If you want to recompile @code{qsort} or other source files which do not | |
3426 | actually use your global register variable, so that they will not use that | |
3427 | register for any other purpose, then it suffices to specify the compiler | |
3428 | option @samp{-ffixed-@var{reg}}. You need not actually add a global | |
3429 | register declaration to their source code. | |
3430 | ||
3431 | A function which can alter the value of a global register variable cannot | |
3432 | safely be called from a function compiled without this variable, because it | |
3433 | could clobber the value the caller expects to find there on return. | |
3434 | Therefore, the function which is the entry point into the part of the | |
3435 | program that uses the global register variable must explicitly save and | |
3436 | restore the value which belongs to its caller. | |
3437 | ||
3438 | @cindex register variable after @code{longjmp} | |
3439 | @cindex global register after @code{longjmp} | |
3440 | @cindex value after @code{longjmp} | |
3441 | @findex longjmp | |
3442 | @findex setjmp | |
3443 | On most machines, @code{longjmp} will restore to each global register | |
3444 | variable the value it had at the time of the @code{setjmp}. On some | |
3445 | machines, however, @code{longjmp} will not change the value of global | |
3446 | register variables. To be portable, the function that called @code{setjmp} | |
3447 | should make other arrangements to save the values of the global register | |
3448 | variables, and to restore them in a @code{longjmp}. This way, the same | |
3449 | thing will happen regardless of what @code{longjmp} does. | |
3450 | ||
3451 | All global register variable declarations must precede all function | |
3452 | definitions. If such a declaration could appear after function | |
3453 | definitions, the declaration would be too late to prevent the register from | |
3454 | being used for other purposes in the preceding functions. | |
3455 | ||
3456 | Global register variables may not have initial values, because an | |
3457 | executable file has no means to supply initial contents for a register. | |
3458 | ||
3459 | On the Sparc, there are reports that g3 @dots{} g7 are suitable | |
3460 | registers, but certain library functions, such as @code{getwd}, as well | |
3461 | as the subroutines for division and remainder, modify g3 and g4. g1 and | |
3462 | g2 are local temporaries. | |
3463 | ||
3464 | On the 68000, a2 @dots{} a5 should be suitable, as should d2 @dots{} d7. | |
3465 | Of course, it will not do to use more than a few of those. | |
3466 | ||
3467 | @node Local Reg Vars | |
3468 | @subsection Specifying Registers for Local Variables | |
3469 | @cindex local variables, specifying registers | |
3470 | @cindex specifying registers for local variables | |
3471 | @cindex registers for local variables | |
3472 | ||
3473 | You can define a local register variable with a specified register | |
3474 | like this: | |
3475 | ||
3476 | @example | |
3477 | register int *foo asm ("a5"); | |
3478 | @end example | |
3479 | ||
3480 | @noindent | |
3481 | Here @code{a5} is the name of the register which should be used. Note | |
3482 | that this is the same syntax used for defining global register | |
3483 | variables, but for a local variable it would appear within a function. | |
3484 | ||
3485 | Naturally the register name is cpu-dependent, but this is not a | |
3486 | problem, since specific registers are most often useful with explicit | |
3487 | assembler instructions (@pxref{Extended Asm}). Both of these things | |
3488 | generally require that you conditionalize your program according to | |
3489 | cpu type. | |
3490 | ||
3491 | In addition, operating systems on one type of cpu may differ in how they | |
3492 | name the registers; then you would need additional conditionals. For | |
3493 | example, some 68000 operating systems call this register @code{%a5}. | |
3494 | ||
c1f7febf RK |
3495 | Defining such a register variable does not reserve the register; it |
3496 | remains available for other uses in places where flow control determines | |
3497 | the variable's value is not live. However, these registers are made | |
e5e809f4 JL |
3498 | unavailable for use in the reload pass; excessive use of this feature |
3499 | leaves the compiler too few available registers to compile certain | |
3500 | functions. | |
3501 | ||
3502 | This option does not guarantee that GNU CC will generate code that has | |
3503 | this variable in the register you specify at all times. You may not | |
3504 | code an explicit reference to this register in an @code{asm} statement | |
3505 | and assume it will always refer to this variable. | |
c1f7febf | 3506 | |
8d344fbc | 3507 | Stores into local register variables may be deleted when they appear to be dead |
0deaf590 JL |
3508 | according to dataflow analysis. References to local register variables may |
3509 | be deleted or moved or simplified. | |
3510 | ||
c1f7febf RK |
3511 | @node Alternate Keywords |
3512 | @section Alternate Keywords | |
3513 | @cindex alternate keywords | |
3514 | @cindex keywords, alternate | |
3515 | ||
5490d604 JM |
3516 | The option @option{-traditional} disables certain keywords; |
3517 | @option{-ansi} and the various @option{-std} options disable certain | |
3518 | others. This causes trouble when you want to use GNU C extensions, or | |
3519 | ISO C features, in a general-purpose header file that should be usable | |
3520 | by all programs, including ISO C programs and traditional ones. The | |
3521 | keywords @code{asm}, @code{typeof} and @code{inline} cannot be used | |
3522 | since they won't work in a program compiled with @option{-ansi} | |
3523 | (although @code{inline} can be used in a program compiled with | |
3524 | @option{-std=c99}), while the keywords @code{const}, @code{volatile}, | |
3525 | @code{signed}, @code{typeof} and @code{inline} won't work in a program | |
3526 | compiled with @option{-traditional}. The ISO C99 keyword | |
3527 | @code{restrict} is only available when @option{-std=gnu99} (which will | |
3528 | eventually be the default) or @option{-std=c99} (or the equivalent | |
3529 | @option{-std=iso9899:1999}) is used.@refill | |
c1f7febf RK |
3530 | |
3531 | The way to solve these problems is to put @samp{__} at the beginning and | |
3532 | end of each problematical keyword. For example, use @code{__asm__} | |
3533 | instead of @code{asm}, @code{__const__} instead of @code{const}, and | |
3534 | @code{__inline__} instead of @code{inline}. | |
3535 | ||
3536 | Other C compilers won't accept these alternative keywords; if you want to | |
3537 | compile with another compiler, you can define the alternate keywords as | |
3538 | macros to replace them with the customary keywords. It looks like this: | |
3539 | ||
3540 | @example | |
3541 | #ifndef __GNUC__ | |
3542 | #define __asm__ asm | |
3543 | #endif | |
3544 | @end example | |
3545 | ||
6e6b0525 | 3546 | @findex __extension__ |
f1b5ff21 | 3547 | @samp{-pedantic} and other options cause warnings for many GNU C extensions. |
dbe519e0 | 3548 | You can |
c1f7febf RK |
3549 | prevent such warnings within one expression by writing |
3550 | @code{__extension__} before the expression. @code{__extension__} has no | |
3551 | effect aside from this. | |
3552 | ||
3553 | @node Incomplete Enums | |
3554 | @section Incomplete @code{enum} Types | |
3555 | ||
3556 | You can define an @code{enum} tag without specifying its possible values. | |
3557 | This results in an incomplete type, much like what you get if you write | |
3558 | @code{struct foo} without describing the elements. A later declaration | |
3559 | which does specify the possible values completes the type. | |
3560 | ||
3561 | You can't allocate variables or storage using the type while it is | |
3562 | incomplete. However, you can work with pointers to that type. | |
3563 | ||
3564 | This extension may not be very useful, but it makes the handling of | |
3565 | @code{enum} more consistent with the way @code{struct} and @code{union} | |
3566 | are handled. | |
3567 | ||
3568 | This extension is not supported by GNU C++. | |
3569 | ||
3570 | @node Function Names | |
3571 | @section Function Names as Strings | |
4b404517 JM |
3572 | @cindex @code{__FUNCTION__} identifier |
3573 | @cindex @code{__PRETTY_FUNCTION__} identifier | |
3574 | @cindex @code{__func__} identifier | |
c1f7febf | 3575 | |
22acfb79 NM |
3576 | GNU CC predefines two magic identifiers to hold the name of the current |
3577 | function. The identifier @code{__FUNCTION__} holds the name of the function | |
3578 | as it appears in the source. The identifier @code{__PRETTY_FUNCTION__} | |
3579 | holds the name of the function pretty printed in a language specific | |
3580 | fashion. | |
c1f7febf RK |
3581 | |
3582 | These names are always the same in a C function, but in a C++ function | |
3583 | they may be different. For example, this program: | |
3584 | ||
3585 | @smallexample | |
3586 | extern "C" @{ | |
3587 | extern int printf (char *, ...); | |
3588 | @} | |
3589 | ||
3590 | class a @{ | |
3591 | public: | |
3592 | sub (int i) | |
3593 | @{ | |
3594 | printf ("__FUNCTION__ = %s\n", __FUNCTION__); | |
3595 | printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__); | |
3596 | @} | |
3597 | @}; | |
3598 | ||
3599 | int | |
3600 | main (void) | |
3601 | @{ | |
3602 | a ax; | |
3603 | ax.sub (0); | |
3604 | return 0; | |
3605 | @} | |
3606 | @end smallexample | |
3607 | ||
3608 | @noindent | |
3609 | gives this output: | |
3610 | ||
3611 | @smallexample | |
3612 | __FUNCTION__ = sub | |
3613 | __PRETTY_FUNCTION__ = int a::sub (int) | |
3614 | @end smallexample | |
3615 | ||
22acfb79 NM |
3616 | The compiler automagically replaces the identifiers with a string |
3617 | literal containing the appropriate name. Thus, they are neither | |
3618 | preprocessor macros, like @code{__FILE__} and @code{__LINE__}, nor | |
3619 | variables. This means that they catenate with other string literals, and | |
3620 | that they can be used to initialize char arrays. For example | |
3621 | ||
3622 | @smallexample | |
3623 | char here[] = "Function " __FUNCTION__ " in " __FILE__; | |
3624 | @end smallexample | |
3625 | ||
3626 | On the other hand, @samp{#ifdef __FUNCTION__} does not have any special | |
c1f7febf RK |
3627 | meaning inside a function, since the preprocessor does not do anything |
3628 | special with the identifier @code{__FUNCTION__}. | |
3629 | ||
22acfb79 | 3630 | GNU CC also supports the magic word @code{__func__}, defined by the |
4b404517 | 3631 | ISO standard C99: |
22acfb79 NM |
3632 | |
3633 | @display | |
3634 | The identifier @code{__func__} is implicitly declared by the translator | |
3635 | as if, immediately following the opening brace of each function | |
3636 | definition, the declaration | |
3637 | ||
3638 | @smallexample | |
3639 | static const char __func__[] = "function-name"; | |
3640 | @end smallexample | |
3641 | ||
3642 | appeared, where function-name is the name of the lexically-enclosing | |
3643 | function. This name is the unadorned name of the function. | |
3644 | @end display | |
3645 | ||
3646 | By this definition, @code{__func__} is a variable, not a string literal. | |
3647 | In particular, @code{__func__} does not catenate with other string | |
3648 | literals. | |
3649 | ||
3650 | In @code{C++}, @code{__FUNCTION__} and @code{__PRETTY_FUNCTION__} are | |
3651 | variables, declared in the same way as @code{__func__}. | |
3652 | ||
c1f7febf RK |
3653 | @node Return Address |
3654 | @section Getting the Return or Frame Address of a Function | |
3655 | ||
3656 | These functions may be used to get information about the callers of a | |
3657 | function. | |
3658 | ||
3659 | @table @code | |
185ebd6c | 3660 | @findex __builtin_return_address |
c1f7febf RK |
3661 | @item __builtin_return_address (@var{level}) |
3662 | This function returns the return address of the current function, or of | |
3663 | one of its callers. The @var{level} argument is number of frames to | |
3664 | scan up the call stack. A value of @code{0} yields the return address | |
3665 | of the current function, a value of @code{1} yields the return address | |
3666 | of the caller of the current function, and so forth. | |
3667 | ||
3668 | The @var{level} argument must be a constant integer. | |
3669 | ||
3670 | On some machines it may be impossible to determine the return address of | |
3671 | any function other than the current one; in such cases, or when the top | |
3672 | of the stack has been reached, this function will return @code{0}. | |
3673 | ||
3674 | This function should only be used with a non-zero argument for debugging | |
3675 | purposes. | |
3676 | ||
185ebd6c | 3677 | @findex __builtin_frame_address |
c1f7febf RK |
3678 | @item __builtin_frame_address (@var{level}) |
3679 | This function is similar to @code{__builtin_return_address}, but it | |
3680 | returns the address of the function frame rather than the return address | |
3681 | of the function. Calling @code{__builtin_frame_address} with a value of | |
3682 | @code{0} yields the frame address of the current function, a value of | |
3683 | @code{1} yields the frame address of the caller of the current function, | |
3684 | and so forth. | |
3685 | ||
3686 | The frame is the area on the stack which holds local variables and saved | |
3687 | registers. The frame address is normally the address of the first word | |
3688 | pushed on to the stack by the function. However, the exact definition | |
3689 | depends upon the processor and the calling convention. If the processor | |
3690 | has a dedicated frame pointer register, and the function has a frame, | |
3691 | then @code{__builtin_frame_address} will return the value of the frame | |
3692 | pointer register. | |
3693 | ||
3694 | The caveats that apply to @code{__builtin_return_address} apply to this | |
3695 | function as well. | |
3696 | @end table | |
3697 | ||
185ebd6c RH |
3698 | @node Other Builtins |
3699 | @section Other built-in functions provided by GNU CC | |
01702459 JM |
3700 | @cindex builtin functions |
3701 | @findex __builtin_isgreater | |
3702 | @findex __builtin_isgreaterequal | |
3703 | @findex __builtin_isless | |
3704 | @findex __builtin_islessequal | |
3705 | @findex __builtin_islessgreater | |
3706 | @findex __builtin_isunordered | |
3707 | @findex abort | |
3708 | @findex abs | |
3709 | @findex alloca | |
3710 | @findex bcmp | |
3711 | @findex bzero | |
341e3d11 JM |
3712 | @findex cimag |
3713 | @findex cimagf | |
3714 | @findex cimagl | |
3715 | @findex conj | |
3716 | @findex conjf | |
3717 | @findex conjl | |
01702459 JM |
3718 | @findex cos |
3719 | @findex cosf | |
3720 | @findex cosl | |
341e3d11 JM |
3721 | @findex creal |
3722 | @findex crealf | |
3723 | @findex creall | |
01702459 JM |
3724 | @findex exit |
3725 | @findex _exit | |
796cdb65 | 3726 | @findex _Exit |
01702459 JM |
3727 | @findex fabs |
3728 | @findex fabsf | |
3729 | @findex fabsl | |
3730 | @findex ffs | |
18f988a0 | 3731 | @findex fprintf |
01702459 | 3732 | @findex fputs |
e78f4a97 | 3733 | @findex imaxabs |
c7b6c6cd | 3734 | @findex index |
01702459 JM |
3735 | @findex labs |
3736 | @findex llabs | |
3737 | @findex memcmp | |
3738 | @findex memcpy | |
3739 | @findex memset | |
3740 | @findex printf | |
c7b6c6cd | 3741 | @findex rindex |
01702459 JM |
3742 | @findex sin |
3743 | @findex sinf | |
3744 | @findex sinl | |
3745 | @findex sqrt | |
3746 | @findex sqrtf | |
3747 | @findex sqrtl | |
d118937d | 3748 | @findex strcat |
01702459 JM |
3749 | @findex strchr |
3750 | @findex strcmp | |
3751 | @findex strcpy | |
d118937d | 3752 | @findex strcspn |
01702459 | 3753 | @findex strlen |
d118937d | 3754 | @findex strncat |
da9e9f08 KG |
3755 | @findex strncmp |
3756 | @findex strncpy | |
01702459 JM |
3757 | @findex strpbrk |
3758 | @findex strrchr | |
d118937d | 3759 | @findex strspn |
01702459 | 3760 | @findex strstr |
185ebd6c RH |
3761 | |
3762 | GNU CC provides a large number of built-in functions other than the ones | |
3763 | mentioned above. Some of these are for internal use in the processing | |
3764 | of exceptions or variable-length argument lists and will not be | |
3765 | documented here because they may change from time to time; we do not | |
3766 | recommend general use of these functions. | |
3767 | ||
3768 | The remaining functions are provided for optimization purposes. | |
3769 | ||
3770 | GNU CC includes builtin versions of many of the functions in the | |
01702459 JM |
3771 | standard C library. The versions prefixed with @code{__builtin_} will |
3772 | always be treated as having the same meaning as the C library function | |
3773 | even if you specify the @samp{-fno-builtin} (@pxref{C Dialect Options}) | |
3774 | option. Many of these functions are only optimized in certain cases; if | |
3775 | not optimized in a particular case, a call to the library function will | |
3776 | be emitted. | |
3777 | ||
796cdb65 JM |
3778 | The functions @code{abort}, @code{exit}, @code{_Exit} and @code{_exit} |
3779 | are recognized and presumed not to return, but otherwise are not built | |
3780 | in. @code{_exit} is not recognized in strict ISO C mode (@samp{-ansi}, | |
3781 | @samp{-std=c89} or @samp{-std=c99}). @code{_Exit} is not recognized in | |
3782 | strict C89 mode (@samp{-ansi} or @samp{-std=c89}). | |
01702459 JM |
3783 | |
3784 | Outside strict ISO C mode, the functions @code{alloca}, @code{bcmp}, | |
c7b6c6cd KG |
3785 | @code{bzero}, @code{index}, @code{rindex} and @code{ffs} may be handled |
3786 | as builtins. Corresponding versions @code{__builtin_alloca}, | |
3787 | @code{__builtin_bcmp}, @code{__builtin_bzero}, @code{__builtin_index}, | |
3788 | @code{__builtin_rindex} and @code{__builtin_ffs} are also recognized in | |
01702459 JM |
3789 | strict ISO C mode. |
3790 | ||
341e3d11 JM |
3791 | The ISO C99 functions @code{conj}, @code{conjf}, @code{conjl}, |
3792 | @code{creal}, @code{crealf}, @code{creall}, @code{cimag}, @code{cimagf}, | |
3793 | @code{cimagl}, @code{llabs} and @code{imaxabs} are handled as builtins | |
3794 | except in strict ISO C89 mode. There are also builtin versions of the ISO C99 | |
01702459 JM |
3795 | functions @code{cosf}, @code{cosl}, @code{fabsf}, @code{fabsl}, |
3796 | @code{sinf}, @code{sinl}, @code{sqrtf}, and @code{sqrtl}, that are | |
3797 | recognized in any mode since ISO C89 reserves these names for the | |
3798 | purpose to which ISO C99 puts them. All these functions have | |
3799 | corresponding versions prefixed with @code{__builtin_}. | |
3800 | ||
3801 | The following ISO C89 functions are recognized as builtins unless | |
3802 | @samp{-fno-builtin} is specified: @code{abs}, @code{cos}, @code{fabs}, | |
18f988a0 KG |
3803 | @code{fprintf}, @code{fputs}, @code{labs}, @code{memcmp}, @code{memcpy}, |
3804 | @code{memset}, @code{printf}, @code{sin}, @code{sqrt}, @code{strcat}, | |
3805 | @code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn}, | |
3806 | @code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy}, | |
3807 | @code{strpbrk}, @code{strrchr}, @code{strspn}, and @code{strstr}. All | |
3808 | of these functions have corresponding versions prefixed with | |
3809 | @code{__builtin_}, except that the version for @code{sqrt} is called | |
da9e9f08 | 3810 | @code{__builtin_fsqrt}. |
01702459 JM |
3811 | |
3812 | GNU CC provides builtin versions of the ISO C99 floating point | |
3813 | comparison macros (that avoid raising exceptions for unordered | |
3814 | operands): @code{__builtin_isgreater}, @code{__builtin_isgreaterequal}, | |
3815 | @code{__builtin_isless}, @code{__builtin_islessequal}, | |
3816 | @code{__builtin_islessgreater}, and @code{__builtin_isunordered}. | |
3817 | ||
185ebd6c | 3818 | |
994a57cd | 3819 | @table @code |
185ebd6c | 3820 | @findex __builtin_constant_p |
994a57cd | 3821 | @item __builtin_constant_p (@var{exp}) |
185ebd6c RH |
3822 | You can use the builtin function @code{__builtin_constant_p} to |
3823 | determine if a value is known to be constant at compile-time and hence | |
3824 | that GNU CC can perform constant-folding on expressions involving that | |
3825 | value. The argument of the function is the value to test. The function | |
3826 | returns the integer 1 if the argument is known to be a compile-time | |
3827 | constant and 0 if it is not known to be a compile-time constant. A | |
3828 | return of 0 does not indicate that the value is @emph{not} a constant, | |
3829 | but merely that GNU CC cannot prove it is a constant with the specified | |
3830 | value of the @samp{-O} option. | |
3831 | ||
3832 | You would typically use this function in an embedded application where | |
3833 | memory was a critical resource. If you have some complex calculation, | |
3834 | you may want it to be folded if it involves constants, but need to call | |
3835 | a function if it does not. For example: | |
3836 | ||
4d390518 | 3837 | @smallexample |
310668e8 JM |
3838 | #define Scale_Value(X) \ |
3839 | (__builtin_constant_p (X) \ | |
3840 | ? ((X) * SCALE + OFFSET) : Scale (X)) | |
185ebd6c RH |
3841 | @end smallexample |
3842 | ||
3843 | You may use this builtin function in either a macro or an inline | |
3844 | function. However, if you use it in an inlined function and pass an | |
3845 | argument of the function as the argument to the builtin, GNU CC will | |
3846 | never return 1 when you call the inline function with a string constant | |
4b404517 | 3847 | or compound literal (@pxref{Compound Literals}) and will not return 1 |
185ebd6c RH |
3848 | when you pass a constant numeric value to the inline function unless you |
3849 | specify the @samp{-O} option. | |
3850 | ||
994a57cd RH |
3851 | @findex __builtin_expect |
3852 | @item __builtin_expect(@var{exp}, @var{c}) | |
02f52e19 | 3853 | You may use @code{__builtin_expect} to provide the compiler with |
994a57cd RH |
3854 | branch prediction information. In general, you should prefer to |
3855 | use actual profile feedback for this (@samp{-fprofile-arcs}), as | |
3856 | programmers are notoriously bad at predicting how their programs | |
60b6e1f5 | 3857 | actually perform. However, there are applications in which this |
994a57cd RH |
3858 | data is hard to collect. |
3859 | ||
3860 | The return value is the value of @var{exp}, which should be an | |
3861 | integral expression. The value of @var{c} must be a compile-time | |
3862 | constant. The semantics of the builtin are that it is expected | |
3863 | that @var{exp} == @var{c}. For example: | |
3864 | ||
3865 | @smallexample | |
3866 | if (__builtin_expect (x, 0)) | |
3867 | foo (); | |
3868 | @end smallexample | |
3869 | ||
3870 | @noindent | |
3871 | would indicate that we do not expect to call @code{foo}, since | |
3872 | we expect @code{x} to be zero. Since you are limited to integral | |
3873 | expressions for @var{exp}, you should use constructions such as | |
3874 | ||
3875 | @smallexample | |
3876 | if (__builtin_expect (ptr != NULL, 1)) | |
3877 | error (); | |
3878 | @end smallexample | |
3879 | ||
3880 | @noindent | |
3881 | when testing pointer or floating-point values. | |
3882 | @end table | |
3883 | ||
c1f7febf RK |
3884 | @node C++ Extensions |
3885 | @chapter Extensions to the C++ Language | |
3886 | @cindex extensions, C++ language | |
3887 | @cindex C++ language extensions | |
3888 | ||
3889 | The GNU compiler provides these extensions to the C++ language (and you | |
3890 | can also use most of the C language extensions in your C++ programs). If you | |
3891 | want to write code that checks whether these features are available, you can | |
3892 | test for the GNU compiler the same way as for C programs: check for a | |
3893 | predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to | |
3894 | test specifically for GNU C++ (@pxref{Standard Predefined,,Standard | |
3895 | Predefined Macros,cpp.info,The C Preprocessor}). | |
3896 | ||
3897 | @menu | |
c1f7febf | 3898 | * Min and Max:: C++ Minimum and maximum operators. |
02cac427 | 3899 | * Volatiles:: What constitutes an access to a volatile object. |
49419c8f | 3900 | * Restricted Pointers:: C99 restricted pointers and references. |
7a81cf7f | 3901 | * Vague Linkage:: Where G++ puts inlines, vtables and such. |
c1f7febf | 3902 | * C++ Interface:: You can use a single C++ header file for both |
e6f3b89d | 3903 | declarations and definitions. |
c1f7febf | 3904 | * Template Instantiation:: Methods for ensuring that exactly one copy of |
e6f3b89d | 3905 | each needed template instantiation is emitted. |
0ded1f18 JM |
3906 | * Bound member functions:: You can extract a function pointer to the |
3907 | method denoted by a @samp{->*} or @samp{.*} expression. | |
e6f3b89d | 3908 | * C++ Attributes:: Variable, function, and type attributes for C++ only. |
1f730ff7 | 3909 | * Java Exceptions:: Tweaking exception handling to work with Java. |
e6f3b89d PE |
3910 | * Deprecated Features:: Things might disappear from g++. |
3911 | * Backwards Compatibility:: Compatibilities with earlier definitions of C++. | |
c1f7febf RK |
3912 | @end menu |
3913 | ||
c1f7febf RK |
3914 | @node Min and Max |
3915 | @section Minimum and Maximum Operators in C++ | |
3916 | ||
3917 | It is very convenient to have operators which return the ``minimum'' or the | |
3918 | ``maximum'' of two arguments. In GNU C++ (but not in GNU C), | |
3919 | ||
3920 | @table @code | |
3921 | @item @var{a} <? @var{b} | |
3922 | @findex <? | |
3923 | @cindex minimum operator | |
3924 | is the @dfn{minimum}, returning the smaller of the numeric values | |
3925 | @var{a} and @var{b}; | |
3926 | ||
3927 | @item @var{a} >? @var{b} | |
3928 | @findex >? | |
3929 | @cindex maximum operator | |
3930 | is the @dfn{maximum}, returning the larger of the numeric values @var{a} | |
3931 | and @var{b}. | |
3932 | @end table | |
3933 | ||
3934 | These operations are not primitive in ordinary C++, since you can | |
3935 | use a macro to return the minimum of two things in C++, as in the | |
3936 | following example. | |
3937 | ||
3938 | @example | |
3939 | #define MIN(X,Y) ((X) < (Y) ? : (X) : (Y)) | |
3940 | @end example | |
3941 | ||
3942 | @noindent | |
3943 | You might then use @w{@samp{int min = MIN (i, j);}} to set @var{min} to | |
3944 | the minimum value of variables @var{i} and @var{j}. | |
3945 | ||
3946 | However, side effects in @code{X} or @code{Y} may cause unintended | |
3947 | behavior. For example, @code{MIN (i++, j++)} will fail, incrementing | |
3948 | the smaller counter twice. A GNU C extension allows you to write safe | |
3949 | macros that avoid this kind of problem (@pxref{Naming Types,,Naming an | |
3950 | Expression's Type}). However, writing @code{MIN} and @code{MAX} as | |
3951 | macros also forces you to use function-call notation for a | |
3952 | fundamental arithmetic operation. Using GNU C++ extensions, you can | |
3953 | write @w{@samp{int min = i <? j;}} instead. | |
3954 | ||
3955 | Since @code{<?} and @code{>?} are built into the compiler, they properly | |
3956 | handle expressions with side-effects; @w{@samp{int min = i++ <? j++;}} | |
3957 | works correctly. | |
3958 | ||
02cac427 NS |
3959 | @node Volatiles |
3960 | @section When is a Volatile Object Accessed? | |
3961 | @cindex accessing volatiles | |
3962 | @cindex volatile read | |
3963 | @cindex volatile write | |
3964 | @cindex volatile access | |
3965 | ||
3966 | Both the C and C++ standard have the concept of volatile objects. These | |
3967 | are normally accessed by pointers and used for accessing hardware. The | |
8117da65 | 3968 | standards encourage compilers to refrain from optimizations |
02cac427 NS |
3969 | concerning accesses to volatile objects that it might perform on |
3970 | non-volatile objects. The C standard leaves it implementation defined | |
3971 | as to what constitutes a volatile access. The C++ standard omits to | |
3972 | specify this, except to say that C++ should behave in a similar manner | |
3973 | to C with respect to volatiles, where possible. The minimum either | |
8117da65 | 3974 | standard specifies is that at a sequence point all previous accesses to |
02cac427 NS |
3975 | volatile objects have stabilized and no subsequent accesses have |
3976 | occurred. Thus an implementation is free to reorder and combine | |
3977 | volatile accesses which occur between sequence points, but cannot do so | |
3978 | for accesses across a sequence point. The use of volatiles does not | |
3979 | allow you to violate the restriction on updating objects multiple times | |
3980 | within a sequence point. | |
3981 | ||
3982 | In most expressions, it is intuitively obvious what is a read and what is | |
3983 | a write. For instance | |
3984 | ||
3985 | @example | |
3986 | volatile int *dst = <somevalue>; | |
3987 | volatile int *src = <someothervalue>; | |
3988 | *dst = *src; | |
3989 | @end example | |
3990 | ||
3991 | @noindent | |
3992 | will cause a read of the volatile object pointed to by @var{src} and stores the | |
3993 | value into the volatile object pointed to by @var{dst}. There is no | |
3994 | guarantee that these reads and writes are atomic, especially for objects | |
3995 | larger than @code{int}. | |
3996 | ||
3997 | Less obvious expressions are where something which looks like an access | |
3998 | is used in a void context. An example would be, | |
3999 | ||
4000 | @example | |
4001 | volatile int *src = <somevalue>; | |
4002 | *src; | |
4003 | @end example | |
4004 | ||
4005 | With C, such expressions are rvalues, and as rvalues cause a read of | |
4006 | the object, gcc interprets this as a read of the volatile being pointed | |
4007 | to. The C++ standard specifies that such expressions do not undergo | |
4008 | lvalue to rvalue conversion, and that the type of the dereferenced | |
4009 | object may be incomplete. The C++ standard does not specify explicitly | |
4010 | that it is this lvalue to rvalue conversion which is responsible for | |
4011 | causing an access. However, there is reason to believe that it is, | |
4012 | because otherwise certain simple expressions become undefined. However, | |
4013 | because it would surprise most programmers, g++ treats dereferencing a | |
4014 | pointer to volatile object of complete type in a void context as a read | |
4015 | of the object. When the object has incomplete type, g++ issues a | |
4016 | warning. | |
4017 | ||
4018 | @example | |
4019 | struct S; | |
4020 | struct T @{int m;@}; | |
4021 | volatile S *ptr1 = <somevalue>; | |
4022 | volatile T *ptr2 = <somevalue>; | |
4023 | *ptr1; | |
4024 | *ptr2; | |
4025 | @end example | |
4026 | ||
4027 | In this example, a warning is issued for @code{*ptr1}, and @code{*ptr2} | |
4028 | causes a read of the object pointed to. If you wish to force an error on | |
4029 | the first case, you must force a conversion to rvalue with, for instance | |
4030 | a static cast, @code{static_cast<S>(*ptr1)}. | |
4031 | ||
4032 | When using a reference to volatile, g++ does not treat equivalent | |
4033 | expressions as accesses to volatiles, but instead issues a warning that | |
4034 | no volatile is accessed. The rationale for this is that otherwise it | |
4035 | becomes difficult to determine where volatile access occur, and not | |
4036 | possible to ignore the return value from functions returning volatile | |
4037 | references. Again, if you wish to force a read, cast the reference to | |
4038 | an rvalue. | |
4039 | ||
535233a8 NS |
4040 | @node Restricted Pointers |
4041 | @section Restricting Pointer Aliasing | |
4042 | @cindex restricted pointers | |
4043 | @cindex restricted references | |
4044 | @cindex restricted this pointer | |
4045 | ||
49419c8f | 4046 | As with gcc, g++ understands the C99 feature of restricted pointers, |
535233a8 | 4047 | specified with the @code{__restrict__}, or @code{__restrict} type |
49419c8f | 4048 | qualifier. Because you cannot compile C++ by specifying the -std=c99 |
535233a8 NS |
4049 | language flag, @code{restrict} is not a keyword in C++. |
4050 | ||
4051 | In addition to allowing restricted pointers, you can specify restricted | |
4052 | references, which indicate that the reference is not aliased in the local | |
4053 | context. | |
4054 | ||
4055 | @example | |
4056 | void fn (int *__restrict__ rptr, int &__restrict__ rref) | |
4057 | @{ | |
4058 | @dots{} | |
4059 | @} | |
4060 | @end example | |
4061 | ||
4062 | @noindent | |
4063 | In the body of @code{fn}, @var{rptr} points to an unaliased integer and | |
4064 | @var{rref} refers to a (different) unaliased integer. | |
4065 | ||
4066 | You may also specify whether a member function's @var{this} pointer is | |
4067 | unaliased by using @code{__restrict__} as a member function qualifier. | |
4068 | ||
4069 | @example | |
4070 | void T::fn () __restrict__ | |
4071 | @{ | |
4072 | @dots{} | |
4073 | @} | |
4074 | @end example | |
4075 | ||
4076 | @noindent | |
4077 | Within the body of @code{T::fn}, @var{this} will have the effective | |
4078 | definition @code{T *__restrict__ const this}. Notice that the | |
4079 | interpretation of a @code{__restrict__} member function qualifier is | |
4080 | different to that of @code{const} or @code{volatile} qualifier, in that it | |
4081 | is applied to the pointer rather than the object. This is consistent with | |
4082 | other compilers which implement restricted pointers. | |
4083 | ||
4084 | As with all outermost parameter qualifiers, @code{__restrict__} is | |
4085 | ignored in function definition matching. This means you only need to | |
4086 | specify @code{__restrict__} in a function definition, rather than | |
4087 | in a function prototype as well. | |
4088 | ||
7a81cf7f JM |
4089 | @node Vague Linkage |
4090 | @section Vague Linkage | |
4091 | @cindex vague linkage | |
4092 | ||
4093 | There are several constructs in C++ which require space in the object | |
4094 | file but are not clearly tied to a single translation unit. We say that | |
4095 | these constructs have ``vague linkage''. Typically such constructs are | |
4096 | emitted wherever they are needed, though sometimes we can be more | |
4097 | clever. | |
4098 | ||
4099 | @table @asis | |
4100 | @item Inline Functions | |
4101 | Inline functions are typically defined in a header file which can be | |
4102 | included in many different compilations. Hopefully they can usually be | |
4103 | inlined, but sometimes an out-of-line copy is necessary, if the address | |
4104 | of the function is taken or if inlining fails. In general, we emit an | |
4105 | out-of-line copy in all translation units where one is needed. As an | |
4106 | exception, we only emit inline virtual functions with the vtable, since | |
4107 | it will always require a copy. | |
4108 | ||
4109 | Local static variables and string constants used in an inline function | |
4110 | are also considered to have vague linkage, since they must be shared | |
4111 | between all inlined and out-of-line instances of the function. | |
4112 | ||
4113 | @item VTables | |
4114 | @cindex vtable | |
4115 | C++ virtual functions are implemented in most compilers using a lookup | |
4116 | table, known as a vtable. The vtable contains pointers to the virtual | |
4117 | functions provided by a class, and each object of the class contains a | |
4118 | pointer to its vtable (or vtables, in some multiple-inheritance | |
4119 | situations). If the class declares any non-inline, non-pure virtual | |
4120 | functions, the first one is chosen as the ``key method'' for the class, | |
4121 | and the vtable is only emitted in the translation unit where the key | |
4122 | method is defined. | |
4123 | ||
4124 | @emph{Note:} If the chosen key method is later defined as inline, the | |
4125 | vtable will still be emitted in every translation unit which defines it. | |
4126 | Make sure that any inline virtuals are declared inline in the class | |
4127 | body, even if they are not defined there. | |
4128 | ||
4129 | @item type_info objects | |
4130 | @cindex type_info | |
4131 | @cindex RTTI | |
4132 | C++ requires information about types to be written out in order to | |
4133 | implement @samp{dynamic_cast}, @samp{typeid} and exception handling. | |
4134 | For polymorphic classes (classes with virtual functions), the type_info | |
4135 | object is written out along with the vtable so that @samp{dynamic_cast} | |
4136 | can determine the dynamic type of a class object at runtime. For all | |
4137 | other types, we write out the type_info object when it is used: when | |
4138 | applying @samp{typeid} to an expression, throwing an object, or | |
4139 | referring to a type in a catch clause or exception specification. | |
4140 | ||
4141 | @item Template Instantiations | |
4142 | Most everything in this section also applies to template instantiations, | |
4143 | but there are other options as well. | |
4144 | @xref{Template Instantiation,,Where's the Template?}. | |
4145 | ||
4146 | @end table | |
4147 | ||
4148 | When used with GNU ld version 2.8 or later on an ELF system such as | |
4149 | Linux/GNU or Solaris 2, or on Microsoft Windows, duplicate copies of | |
4150 | these constructs will be discarded at link time. This is known as | |
4151 | COMDAT support. | |
4152 | ||
4153 | On targets that don't support COMDAT, but do support weak symbols, GCC | |
4154 | will use them. This way one copy will override all the others, but | |
4155 | the unused copies will still take up space in the executable. | |
4156 | ||
4157 | For targets which do not support either COMDAT or weak symbols, | |
4158 | most entities with vague linkage will be emitted as local symbols to | |
4159 | avoid duplicate definition errors from the linker. This will not happen | |
4160 | for local statics in inlines, however, as having multiple copies will | |
4161 | almost certainly break things. | |
4162 | ||
4163 | @xref{C++ Interface,,Declarations and Definitions in One Header}, for | |
4164 | another way to control placement of these constructs. | |
4165 | ||
c1f7febf RK |
4166 | @node C++ Interface |
4167 | @section Declarations and Definitions in One Header | |
4168 | ||
4169 | @cindex interface and implementation headers, C++ | |
4170 | @cindex C++ interface and implementation headers | |
4171 | C++ object definitions can be quite complex. In principle, your source | |
4172 | code will need two kinds of things for each object that you use across | |
4173 | more than one source file. First, you need an @dfn{interface} | |
4174 | specification, describing its structure with type declarations and | |
4175 | function prototypes. Second, you need the @dfn{implementation} itself. | |
4176 | It can be tedious to maintain a separate interface description in a | |
4177 | header file, in parallel to the actual implementation. It is also | |
4178 | dangerous, since separate interface and implementation definitions may | |
4179 | not remain parallel. | |
4180 | ||
4181 | @cindex pragmas, interface and implementation | |
4182 | With GNU C++, you can use a single header file for both purposes. | |
4183 | ||
4184 | @quotation | |
4185 | @emph{Warning:} The mechanism to specify this is in transition. For the | |
4186 | nonce, you must use one of two @code{#pragma} commands; in a future | |
4187 | release of GNU C++, an alternative mechanism will make these | |
4188 | @code{#pragma} commands unnecessary. | |
4189 | @end quotation | |
4190 | ||
4191 | The header file contains the full definitions, but is marked with | |
4192 | @samp{#pragma interface} in the source code. This allows the compiler | |
4193 | to use the header file only as an interface specification when ordinary | |
4194 | source files incorporate it with @code{#include}. In the single source | |
4195 | file where the full implementation belongs, you can use either a naming | |
4196 | convention or @samp{#pragma implementation} to indicate this alternate | |
4197 | use of the header file. | |
4198 | ||
4199 | @table @code | |
4200 | @item #pragma interface | |
4201 | @itemx #pragma interface "@var{subdir}/@var{objects}.h" | |
4202 | @kindex #pragma interface | |
4203 | Use this directive in @emph{header files} that define object classes, to save | |
4204 | space in most of the object files that use those classes. Normally, | |
4205 | local copies of certain information (backup copies of inline member | |
4206 | functions, debugging information, and the internal tables that implement | |
4207 | virtual functions) must be kept in each object file that includes class | |
4208 | definitions. You can use this pragma to avoid such duplication. When a | |
4209 | header file containing @samp{#pragma interface} is included in a | |
4210 | compilation, this auxiliary information will not be generated (unless | |
4211 | the main input source file itself uses @samp{#pragma implementation}). | |
4212 | Instead, the object files will contain references to be resolved at link | |
4213 | time. | |
4214 | ||
4215 | The second form of this directive is useful for the case where you have | |
4216 | multiple headers with the same name in different directories. If you | |
4217 | use this form, you must specify the same string to @samp{#pragma | |
4218 | implementation}. | |
4219 | ||
4220 | @item #pragma implementation | |
4221 | @itemx #pragma implementation "@var{objects}.h" | |
4222 | @kindex #pragma implementation | |
4223 | Use this pragma in a @emph{main input file}, when you want full output from | |
4224 | included header files to be generated (and made globally visible). The | |
4225 | included header file, in turn, should use @samp{#pragma interface}. | |
4226 | Backup copies of inline member functions, debugging information, and the | |
4227 | internal tables used to implement virtual functions are all generated in | |
4228 | implementation files. | |
4229 | ||
4230 | @cindex implied @code{#pragma implementation} | |
4231 | @cindex @code{#pragma implementation}, implied | |
4232 | @cindex naming convention, implementation headers | |
4233 | If you use @samp{#pragma implementation} with no argument, it applies to | |
4234 | an include file with the same basename@footnote{A file's @dfn{basename} | |
4235 | was the name stripped of all leading path information and of trailing | |
4236 | suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source | |
4237 | file. For example, in @file{allclass.cc}, giving just | |
4238 | @samp{#pragma implementation} | |
4239 | by itself is equivalent to @samp{#pragma implementation "allclass.h"}. | |
4240 | ||
4241 | In versions of GNU C++ prior to 2.6.0 @file{allclass.h} was treated as | |
4242 | an implementation file whenever you would include it from | |
4243 | @file{allclass.cc} even if you never specified @samp{#pragma | |
4244 | implementation}. This was deemed to be more trouble than it was worth, | |
4245 | however, and disabled. | |
4246 | ||
4247 | If you use an explicit @samp{#pragma implementation}, it must appear in | |
4248 | your source file @emph{before} you include the affected header files. | |
4249 | ||
4250 | Use the string argument if you want a single implementation file to | |
4251 | include code from multiple header files. (You must also use | |
4252 | @samp{#include} to include the header file; @samp{#pragma | |
4253 | implementation} only specifies how to use the file---it doesn't actually | |
4254 | include it.) | |
4255 | ||
4256 | There is no way to split up the contents of a single header file into | |
4257 | multiple implementation files. | |
4258 | @end table | |
4259 | ||
4260 | @cindex inlining and C++ pragmas | |
4261 | @cindex C++ pragmas, effect on inlining | |
4262 | @cindex pragmas in C++, effect on inlining | |
4263 | @samp{#pragma implementation} and @samp{#pragma interface} also have an | |
4264 | effect on function inlining. | |
4265 | ||
4266 | If you define a class in a header file marked with @samp{#pragma | |
4267 | interface}, the effect on a function defined in that class is similar to | |
4268 | an explicit @code{extern} declaration---the compiler emits no code at | |
4269 | all to define an independent version of the function. Its definition | |
4270 | is used only for inlining with its callers. | |
4271 | ||
4272 | Conversely, when you include the same header file in a main source file | |
4273 | that declares it as @samp{#pragma implementation}, the compiler emits | |
4274 | code for the function itself; this defines a version of the function | |
4275 | that can be found via pointers (or by callers compiled without | |
4276 | inlining). If all calls to the function can be inlined, you can avoid | |
4277 | emitting the function by compiling with @samp{-fno-implement-inlines}. | |
4278 | If any calls were not inlined, you will get linker errors. | |
4279 | ||
4280 | @node Template Instantiation | |
4281 | @section Where's the Template? | |
4282 | ||
4283 | @cindex template instantiation | |
4284 | ||
4285 | C++ templates are the first language feature to require more | |
4286 | intelligence from the environment than one usually finds on a UNIX | |
4287 | system. Somehow the compiler and linker have to make sure that each | |
4288 | template instance occurs exactly once in the executable if it is needed, | |
4289 | and not at all otherwise. There are two basic approaches to this | |
4290 | problem, which I will refer to as the Borland model and the Cfront model. | |
4291 | ||
4292 | @table @asis | |
4293 | @item Borland model | |
4294 | Borland C++ solved the template instantiation problem by adding the code | |
469b759e JM |
4295 | equivalent of common blocks to their linker; the compiler emits template |
4296 | instances in each translation unit that uses them, and the linker | |
4297 | collapses them together. The advantage of this model is that the linker | |
4298 | only has to consider the object files themselves; there is no external | |
4299 | complexity to worry about. This disadvantage is that compilation time | |
4300 | is increased because the template code is being compiled repeatedly. | |
4301 | Code written for this model tends to include definitions of all | |
4302 | templates in the header file, since they must be seen to be | |
4303 | instantiated. | |
c1f7febf RK |
4304 | |
4305 | @item Cfront model | |
4306 | The AT&T C++ translator, Cfront, solved the template instantiation | |
4307 | problem by creating the notion of a template repository, an | |
469b759e JM |
4308 | automatically maintained place where template instances are stored. A |
4309 | more modern version of the repository works as follows: As individual | |
4310 | object files are built, the compiler places any template definitions and | |
4311 | instantiations encountered in the repository. At link time, the link | |
4312 | wrapper adds in the objects in the repository and compiles any needed | |
4313 | instances that were not previously emitted. The advantages of this | |
4314 | model are more optimal compilation speed and the ability to use the | |
4315 | system linker; to implement the Borland model a compiler vendor also | |
c1f7febf | 4316 | needs to replace the linker. The disadvantages are vastly increased |
469b759e JM |
4317 | complexity, and thus potential for error; for some code this can be |
4318 | just as transparent, but in practice it can been very difficult to build | |
c1f7febf | 4319 | multiple programs in one directory and one program in multiple |
469b759e JM |
4320 | directories. Code written for this model tends to separate definitions |
4321 | of non-inline member templates into a separate file, which should be | |
4322 | compiled separately. | |
c1f7febf RK |
4323 | @end table |
4324 | ||
469b759e | 4325 | When used with GNU ld version 2.8 or later on an ELF system such as |
a4b3b54a JM |
4326 | Linux/GNU or Solaris 2, or on Microsoft Windows, g++ supports the |
4327 | Borland model. On other systems, g++ implements neither automatic | |
4328 | model. | |
469b759e JM |
4329 | |
4330 | A future version of g++ will support a hybrid model whereby the compiler | |
4331 | will emit any instantiations for which the template definition is | |
4332 | included in the compile, and store template definitions and | |
4333 | instantiation context information into the object file for the rest. | |
4334 | The link wrapper will extract that information as necessary and invoke | |
4335 | the compiler to produce the remaining instantiations. The linker will | |
4336 | then combine duplicate instantiations. | |
4337 | ||
4338 | In the mean time, you have the following options for dealing with | |
4339 | template instantiations: | |
c1f7febf RK |
4340 | |
4341 | @enumerate | |
d863830b JL |
4342 | @item |
4343 | Compile your template-using code with @samp{-frepo}. The compiler will | |
4344 | generate files with the extension @samp{.rpo} listing all of the | |
4345 | template instantiations used in the corresponding object files which | |
4346 | could be instantiated there; the link wrapper, @samp{collect2}, will | |
4347 | then update the @samp{.rpo} files to tell the compiler where to place | |
4348 | those instantiations and rebuild any affected object files. The | |
4349 | link-time overhead is negligible after the first pass, as the compiler | |
4350 | will continue to place the instantiations in the same files. | |
4351 | ||
4352 | This is your best option for application code written for the Borland | |
4353 | model, as it will just work. Code written for the Cfront model will | |
4354 | need to be modified so that the template definitions are available at | |
4355 | one or more points of instantiation; usually this is as simple as adding | |
4356 | @code{#include <tmethods.cc>} to the end of each template header. | |
4357 | ||
4358 | For library code, if you want the library to provide all of the template | |
4359 | instantiations it needs, just try to link all of its object files | |
4360 | together; the link will fail, but cause the instantiations to be | |
4361 | generated as a side effect. Be warned, however, that this may cause | |
4362 | conflicts if multiple libraries try to provide the same instantiations. | |
4363 | For greater control, use explicit instantiation as described in the next | |
4364 | option. | |
4365 | ||
c1f7febf RK |
4366 | @item |
4367 | Compile your code with @samp{-fno-implicit-templates} to disable the | |
4368 | implicit generation of template instances, and explicitly instantiate | |
4369 | all the ones you use. This approach requires more knowledge of exactly | |
4370 | which instances you need than do the others, but it's less | |
4371 | mysterious and allows greater control. You can scatter the explicit | |
4372 | instantiations throughout your program, perhaps putting them in the | |
4373 | translation units where the instances are used or the translation units | |
4374 | that define the templates themselves; you can put all of the explicit | |
4375 | instantiations you need into one big file; or you can create small files | |
4376 | like | |
4377 | ||
4378 | @example | |
4379 | #include "Foo.h" | |
4380 | #include "Foo.cc" | |
4381 | ||
4382 | template class Foo<int>; | |
4383 | template ostream& operator << | |
4384 | (ostream&, const Foo<int>&); | |
4385 | @end example | |
4386 | ||
4387 | for each of the instances you need, and create a template instantiation | |
4388 | library from those. | |
4389 | ||
4390 | If you are using Cfront-model code, you can probably get away with not | |
4391 | using @samp{-fno-implicit-templates} when compiling files that don't | |
4392 | @samp{#include} the member template definitions. | |
4393 | ||
4394 | If you use one big file to do the instantiations, you may want to | |
4395 | compile it without @samp{-fno-implicit-templates} so you get all of the | |
4396 | instances required by your explicit instantiations (but not by any | |
4397 | other files) without having to specify them as well. | |
4398 | ||
4399 | g++ has extended the template instantiation syntax outlined in the | |
03d0f4af | 4400 | Working Paper to allow forward declaration of explicit instantiations |
4003d7f9 JM |
4401 | (with @code{extern}), instantiation of the compiler support data for a |
4402 | template class (i.e. the vtable) without instantiating any of its | |
4403 | members (with @code{inline}), and instantiation of only the static data | |
4404 | members of a template class, without the support data or member | |
4405 | functions (with (@code{static}): | |
c1f7febf RK |
4406 | |
4407 | @example | |
4408 | extern template int max (int, int); | |
c1f7febf | 4409 | inline template class Foo<int>; |
4003d7f9 | 4410 | static template class Foo<int>; |
c1f7febf RK |
4411 | @end example |
4412 | ||
4413 | @item | |
4414 | Do nothing. Pretend g++ does implement automatic instantiation | |
4415 | management. Code written for the Borland model will work fine, but | |
4416 | each translation unit will contain instances of each of the templates it | |
4417 | uses. In a large program, this can lead to an unacceptable amount of code | |
4418 | duplication. | |
4419 | ||
4420 | @item | |
4421 | Add @samp{#pragma interface} to all files containing template | |
4422 | definitions. For each of these files, add @samp{#pragma implementation | |
4423 | "@var{filename}"} to the top of some @samp{.C} file which | |
4424 | @samp{#include}s it. Then compile everything with | |
4425 | @samp{-fexternal-templates}. The templates will then only be expanded | |
4426 | in the translation unit which implements them (i.e. has a @samp{#pragma | |
4427 | implementation} line for the file where they live); all other files will | |
4428 | use external references. If you're lucky, everything should work | |
4429 | properly. If you get undefined symbol errors, you need to make sure | |
4430 | that each template instance which is used in the program is used in the | |
4431 | file which implements that template. If you don't have any use for a | |
4432 | particular instance in that file, you can just instantiate it | |
4433 | explicitly, using the syntax from the latest C++ working paper: | |
4434 | ||
4435 | @example | |
4436 | template class A<int>; | |
4437 | template ostream& operator << (ostream&, const A<int>&); | |
4438 | @end example | |
4439 | ||
4440 | This strategy will work with code written for either model. If you are | |
4441 | using code written for the Cfront model, the file containing a class | |
4442 | template and the file containing its member templates should be | |
4443 | implemented in the same translation unit. | |
4444 | ||
4445 | A slight variation on this approach is to instead use the flag | |
4446 | @samp{-falt-external-templates}; this flag causes template | |
4447 | instances to be emitted in the translation unit that implements the | |
4448 | header where they are first instantiated, rather than the one which | |
4449 | implements the file where the templates are defined. This header must | |
4450 | be the same in all translation units, or things are likely to break. | |
4451 | ||
4452 | @xref{C++ Interface,,Declarations and Definitions in One Header}, for | |
4453 | more discussion of these pragmas. | |
4454 | @end enumerate | |
4455 | ||
0ded1f18 JM |
4456 | @node Bound member functions |
4457 | @section Extracting the function pointer from a bound pointer to member function | |
4458 | ||
4459 | @cindex pmf | |
4460 | @cindex pointer to member function | |
4461 | @cindex bound pointer to member function | |
4462 | ||
4463 | In C++, pointer to member functions (PMFs) are implemented using a wide | |
4464 | pointer of sorts to handle all the possible call mechanisms; the PMF | |
4465 | needs to store information about how to adjust the @samp{this} pointer, | |
4466 | and if the function pointed to is virtual, where to find the vtable, and | |
4467 | where in the vtable to look for the member function. If you are using | |
4468 | PMFs in an inner loop, you should really reconsider that decision. If | |
4469 | that is not an option, you can extract the pointer to the function that | |
4470 | would be called for a given object/PMF pair and call it directly inside | |
4471 | the inner loop, to save a bit of time. | |
4472 | ||
4473 | Note that you will still be paying the penalty for the call through a | |
4474 | function pointer; on most modern architectures, such a call defeats the | |
4475 | branch prediction features of the CPU. This is also true of normal | |
4476 | virtual function calls. | |
4477 | ||
4478 | The syntax for this extension is | |
4479 | ||
4480 | @example | |
4481 | extern A a; | |
4482 | extern int (A::*fp)(); | |
4483 | typedef int (*fptr)(A *); | |
4484 | ||
4485 | fptr p = (fptr)(a.*fp); | |
4486 | @end example | |
4487 | ||
0fb6bbf5 ML |
4488 | For PMF constants (i.e. expressions of the form @samp{&Klasse::Member}), |
4489 | no object is needed to obtain the address of the function. They can be | |
4490 | converted to function pointers directly: | |
4491 | ||
4492 | @example | |
4493 | fptr p1 = (fptr)(&A::foo); | |
4494 | @end example | |
4495 | ||
0ded1f18 JM |
4496 | You must specify @samp{-Wno-pmf-conversions} to use this extension. |
4497 | ||
5c25e11d PE |
4498 | @node C++ Attributes |
4499 | @section C++-Specific Variable, Function, and Type Attributes | |
4500 | ||
4501 | Some attributes only make sense for C++ programs. | |
4502 | ||
4503 | @table @code | |
4504 | @item init_priority (@var{priority}) | |
4505 | @cindex init_priority attribute | |
4506 | ||
4507 | ||
4508 | In Standard C++, objects defined at namespace scope are guaranteed to be | |
4509 | initialized in an order in strict accordance with that of their definitions | |
4510 | @emph{in a given translation unit}. No guarantee is made for initializations | |
4511 | across translation units. However, GNU C++ allows users to control the | |
3844cd2e | 4512 | order of initialization of objects defined at namespace scope with the |
5c25e11d PE |
4513 | @code{init_priority} attribute by specifying a relative @var{priority}, |
4514 | a constant integral expression currently bounded between 101 and 65535 | |
4515 | inclusive. Lower numbers indicate a higher priority. | |
4516 | ||
4517 | In the following example, @code{A} would normally be created before | |
4518 | @code{B}, but the @code{init_priority} attribute has reversed that order: | |
4519 | ||
4520 | @example | |
4521 | Some_Class A __attribute__ ((init_priority (2000))); | |
4522 | Some_Class B __attribute__ ((init_priority (543))); | |
4523 | @end example | |
4524 | ||
4525 | @noindent | |
4526 | Note that the particular values of @var{priority} do not matter; only their | |
4527 | relative ordering. | |
4528 | ||
60c87482 BM |
4529 | @item java_interface |
4530 | @cindex java_interface attribute | |
4531 | ||
02f52e19 | 4532 | This type attribute informs C++ that the class is a Java interface. It may |
60c87482 | 4533 | only be applied to classes declared within an @code{extern "Java"} block. |
02f52e19 AJ |
4534 | Calls to methods declared in this interface will be dispatched using GCJ's |
4535 | interface table mechanism, instead of regular virtual table dispatch. | |
60c87482 | 4536 | |
5c25e11d PE |
4537 | @end table |
4538 | ||
1f730ff7 ZW |
4539 | @node Java Exceptions |
4540 | @section Java Exceptions | |
4541 | ||
4542 | The Java language uses a slightly different exception handling model | |
4543 | from C++. Normally, GNU C++ will automatically detect when you are | |
4544 | writing C++ code that uses Java exceptions, and handle them | |
4545 | appropriately. However, if C++ code only needs to execute destructors | |
4546 | when Java exceptions are thrown through it, GCC will guess incorrectly. | |
4547 | Sample problematic code: | |
4548 | ||
4549 | @example | |
4550 | struct S @{ ~S(); @}; | |
4551 | extern void bar(); // is implemented in Java and may throw exceptions | |
4552 | void foo() | |
4553 | @{ | |
4554 | S s; | |
4555 | bar(); | |
4556 | @} | |
4557 | @end example | |
4558 | ||
4559 | @noindent | |
4560 | The usual effect of an incorrect guess is a link failure, complaining of | |
4561 | a missing routine called @samp{__gxx_personality_v0}. | |
4562 | ||
4563 | You can inform the compiler that Java exceptions are to be used in a | |
4564 | translation unit, irrespective of what it might think, by writing | |
4565 | @samp{@w{#pragma GCC java_exceptions}} at the head of the file. This | |
4566 | @samp{#pragma} must appear before any functions that throw or catch | |
4567 | exceptions, or run destructors when exceptions are thrown through them. | |
4568 | ||
4569 | You cannot mix Java and C++ exceptions in the same translation unit. It | |
4570 | is believed to be safe to throw a C++ exception from one file through | |
4571 | another file compiled for the for the Java exception model, or vice | |
4572 | versa, but there may be bugs in this area. | |
4573 | ||
e6f3b89d PE |
4574 | @node Deprecated Features |
4575 | @section Deprecated Features | |
4576 | ||
4577 | In the past, the GNU C++ compiler was extended to experiment with new | |
4578 | features, at a time when the C++ language was still evolving. Now that | |
4579 | the C++ standard is complete, some of those features are superseded by | |
4580 | superior alternatives. Using the old features might cause a warning in | |
4581 | some cases that the feature will be dropped in the future. In other | |
4582 | cases, the feature might be gone already. | |
4583 | ||
4584 | While the list below is not exhaustive, it documents some of the options | |
4585 | that are now deprecated: | |
4586 | ||
4587 | @table @code | |
4588 | @item -fexternal-templates | |
4589 | @itemx -falt-external-templates | |
4590 | These are two of the many ways for g++ to implement template | |
4591 | instantiation. @xref{Template Instantiation}. The C++ standard clearly | |
4592 | defines how template definitions have to be organized across | |
4593 | implementation units. g++ has an implicit instantiation mechanism that | |
4594 | should work just fine for standard-conforming code. | |
4595 | ||
4596 | @item -fstrict-prototype | |
4597 | @itemx -fno-strict-prototype | |
4598 | Previously it was possible to use an empty prototype parameter list to | |
4599 | indicate an unspecified number of parameters (like C), rather than no | |
4600 | parameters, as C++ demands. This feature has been removed, except where | |
4601 | it is required for backwards compatibility @xref{Backwards Compatibility}. | |
4602 | @end table | |
4603 | ||
4604 | The named return value extension has been deprecated, and will be | |
4605 | removed from g++ at some point. | |
4606 | ||
82c18d5c NS |
4607 | The use of initializer lists with new expressions has been deprecated, |
4608 | and will be removed from g++ at some point. | |
4609 | ||
e6f3b89d PE |
4610 | @node Backwards Compatibility |
4611 | @section Backwards Compatibility | |
4612 | @cindex Backwards Compatibility | |
4613 | @cindex ARM [Annotated C++ Reference Manual] | |
4614 | ||
4615 | Now that there is a definitive ISO standard C++, g++ has a specification | |
4616 | to adhere to. The C++ language evolved over time, and features that | |
4617 | used to be acceptable in previous drafts of the standard, such as the ARM | |
4618 | [Annotated C++ Reference Manual], are no longer accepted. In order to allow | |
4619 | compilation of C++ written to such drafts, g++ contains some backwards | |
4620 | compatibilities. @emph{All such backwards compatibility features are | |
4621 | liable to disappear in future versions of g++.} They should be considered | |
4622 | deprecated @xref{Deprecated Features}. | |
4623 | ||
4624 | @table @code | |
4625 | @item For scope | |
4626 | If a variable is declared at for scope, it used to remain in scope until | |
4627 | the end of the scope which contained the for statement (rather than just | |
4628 | within the for scope). g++ retains this, but issues a warning, if such a | |
4629 | variable is accessed outside the for scope. | |
4630 | ||
4631 | @item implicit C language | |
4632 | Old C system header files did not contain an @code{extern "C" @{...@}} | |
4633 | scope to set the language. On such systems, all header files are | |
4634 | implicitly scoped inside a C language scope. Also, an empty prototype | |
4635 | @code{()} will be treated as an unspecified number of arguments, rather | |
4636 | than no arguments, as C++ demands. | |
4637 | @end table |