2 * Parse and validate C declarations.
3 * Copyright © 2011-2012, 2020-2021 Nick Bowler
5 * This program is free software: you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation, either version 3 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program. If not, see <https://www.gnu.org/licenses/>.
25 #include "cdecl-internal.h"
30 * Verify the declaration specifiers of a declaration. If top is true, treat
31 * this as a top-level declaration. Otherwise, treat this as a function
32 * parameter (which carries additional constraints).
34 static bool valid_declspecs(struct cdecl *decl, bool top)
36 struct cdecl_declspec *specs = decl->specifiers;
37 struct cdecl_declarator *d = decl->declarators;
38 bool abstract = cdecl_is_abstract(d);
39 unsigned num_storage = 0;
40 unsigned long typemap;
42 typemap = cdecl__build_typemap(specs);
46 for (struct cdecl_declspec *c = specs; c; c = c->next) {
47 switch (cdecl_spec_kind(c)) {
49 if (c->type == CDECL_TYPE_VOID &&
50 (d->type == CDECL_DECL_IDENT
51 || d->type == CDECL_DECL_ARRAY)) {
52 fprintf(stderr, "invalid declaration of type void\n");
57 if (top && abstract) {
58 fprintf(stderr, "type names cannot have storage-class specifiers\n");
62 if (!top && c->type != CDECL_STOR_REGISTER) {
63 fprintf(stderr, "function parameters may only have register storage\n");
67 if (++num_storage > 1) {
68 fprintf(stderr, "too many storage-class specifiers\n");
74 * Restrict qualifiers are only valid in the
75 * pointer qualifier list, which isn't checked here.
77 if (c->type == CDECL_QUAL_RESTRICT) {
78 fprintf(stderr, "only pointer types can be restrict-qualified.\n");
84 fprintf(stderr, "type names cannot have function specifiers\n");
88 if (!top || d->type != CDECL_DECL_FUNCTION) {
89 fprintf(stderr, "only function declarations may have function specifiers.\n");
102 * The C grammar leaves ambiguous some cases where parentheses represent a
103 * function declarator or just parentheses. The language uses additional
104 * context (whether or not a typedef is in scope, etc.) to resolve these
105 * ambiguities, but we don't have access to that kind of information.
107 * The cdecl99 parser uses an unambiguous grammar which treats almost
108 * everything as a function, and thus considers things like 'int (x)' to
109 * be a function type with a single parameter of type 'x' (a typedef name),
110 * returning int. This can result in very complicated types for simple
111 * declarations. Ideally, cdecl99 should try and find the "simplest"
112 * explanation for a given declaration.
114 * Whether or not it achieves the simplest explanation, we apply a simple rule:
115 * if a declarator could be interpreted as something other than a function,
118 * Since cdecl99 supports things like [*] in any context (in C, such constructs
119 * are only valid in function parameter lists), we don't treat them specially
123 static struct cdecl_declarator *reduce_function(struct cdecl *param)
125 struct cdecl_declspec *spec = param->specifiers;
126 struct cdecl_declarator *decl = param->declarators;
127 struct cdecl_declarator *last;
129 for (last = decl; last && last->type != CDECL_DECL_NULL;)
135 last->type = CDECL_DECL_IDENT;
136 last->u.ident = spec->ident;
143 static bool function_is_reducible(struct cdecl_declarator *d)
145 if (d->type != CDECL_DECL_FUNCTION)
147 if (d->child->type != CDECL_DECL_NULL)
148 return false; /* e.g., int (*)(x) */
150 if (!d->u.function.parameters)
151 return false; /* e.g., int f() */
152 if (d->u.function.parameters->next)
153 return false; /* e.g., int (x, y) */
154 if (d->u.function.variadic)
155 return false; /* e.g., int (x, ...) */
157 if (d->u.function.parameters->specifiers->type != CDECL_TYPE_IDENT)
158 return false; /* e.g. int (int) */
159 if (d->u.function.parameters->specifiers->next)
160 return false; /* e.g. int (size_t const) */
161 if (d->u.function.parameters->declarators->type == CDECL_DECL_POINTER)
162 return false; /* e.g. int (x *) */
168 simplify_functions(struct cdecl_declarator **p, struct cdecl_declarator *d)
170 struct cdecl_declarator *new;
172 if (!function_is_reducible(d))
175 new = reduce_function(d->u.function.parameters);
177 return 0; /* e.g. int (foo bar) */
186 * The parser's bias towards considering things as functions whenever possible
187 * makes nested parentheses tricky. (x) is considered to be part of a function
188 * declarator until simplify_functions converts it. The problem is that
189 * (((x))) is not valid as part of a function declarator, but it *is* valid
190 * as an identifier enclosed 3 times in parentheses. This is complicated by
191 * the fact that things like (((int))) are not valid anywhere.
193 * To avoid ambiguities, the parser actually emits a "function" declarator for
194 * every pair of parentheses. The ones that can't reasonably be functions
195 * consist of a single "parameter" with no declaration specifiers (note that
196 * every valid function parameter will have at least one type specifier).
198 * This pass is to remove these fake functions from the parse tree. We take
199 * care to avoid turning invalid things like ((int)) into valid things like
200 * (int) by observing that the only valid function declarators that appear
201 * in these "fake" parentheses are those that have a non-null child declarator
202 * (for instance, int ((*)(int)) *or* those that will be eliminated by the
203 * simplify_functions pass.
207 reduce_parentheses(struct cdecl_declarator **p, struct cdecl_declarator *d)
211 if (d->type != CDECL_DECL_FUNCTION)
214 param = d->u.function.parameters;
215 if (param && param->specifiers == NULL) {
216 struct cdecl_declarator *decl;
218 assert(!param->next);
220 decl = param->declarators;
221 if (decl->type == CDECL_DECL_NULL) {
224 d->u.function.parameters = NULL;
228 if (d->child->type != CDECL_DECL_NULL) {
229 fprintf(stderr, "invalid function parameter\n");
239 * We may have replaced d with another fake function which
240 * also needs to be eliminated.
242 if (reduce_parentheses(p, decl) < 0)
246 * If the remaining declarator is a function, make sure it's
247 * valid by checking its reducibility.
250 if (decl->type == CDECL_DECL_FUNCTION
251 && decl->child->type == CDECL_DECL_NULL
252 && !function_is_reducible(decl)) {
253 fprintf(stderr, "too many parentheses in function\n");
264 * Function parameters and return types have a few restrictions that are
265 * really easy to check in comparison to the above absurdity.
268 check_parameters(struct cdecl_declarator **p, struct cdecl_declarator *d)
270 struct cdecl_declspec *spec;
272 bool has_void = false;
274 if (d->type != CDECL_DECL_FUNCTION)
277 for (param = d->u.function.parameters; param; param = param->next) {
278 if (!valid_declspecs(param, false))
281 /* Check for "void" function parameters as a special case. */
282 for (spec = param->specifiers; spec; spec = spec->next) {
283 if (param->declarators->type != CDECL_DECL_NULL)
285 if (spec->type != CDECL_TYPE_VOID)
288 if (spec != param->specifiers || spec->next != NULL) {
289 fprintf(stderr, "void parameter must not have extra specifiers\n");
291 } else if (d->u.function.parameters->next) {
292 fprintf(stderr, "a void parameter must stand alone\n");
294 } else if (d->u.function.variadic) {
295 fprintf(stderr, "variadic functions cannot have a void parameter\n");
305 * Functions cannot return arrays or functions. Since the parse tree is
306 * "inside-out", we need to look for functions as the child declarator.
309 check_rettypes(struct cdecl_declarator **p, struct cdecl_declarator *d)
311 if (!d->child || d->child->type != CDECL_DECL_FUNCTION)
315 case CDECL_DECL_FUNCTION:
316 fprintf(stderr, "functions cannot return functions\n");
318 case CDECL_DECL_ARRAY:
319 fprintf(stderr, "functions cannot return arrays\n");
327 check_arrays(struct cdecl_declarator **p, struct cdecl_declarator *d)
329 if (!d->child || d->child->type != CDECL_DECL_ARRAY)
333 case CDECL_DECL_FUNCTION:
334 fprintf(stderr, "array members cannot be functions\n");
342 normalize_specs(struct cdecl_declarator **p, struct cdecl_declarator *d)
344 struct cdecl_function *func;
345 struct cdecl_pointer *ptr;
348 case CDECL_DECL_POINTER:
350 ptr->qualifiers = cdecl__normalize_specs(ptr->qualifiers);
352 case CDECL_DECL_FUNCTION:
353 func = &d->u.function;
354 for (struct cdecl *i = func->parameters; i; i = i->next)
355 i->specifiers = cdecl__normalize_specs(i->specifiers);
363 check_qualifiers(struct cdecl_declarator **p, struct cdecl_declarator *d)
365 struct cdecl_declspec *spec;
366 struct cdecl_pointer *ptr;
368 if (!d->child || d->child->type != CDECL_DECL_POINTER)
371 ptr = &d->child->u.pointer;
372 for (spec = ptr->qualifiers; spec; spec = spec->next) {
373 if (spec->type == CDECL_QUAL_RESTRICT
374 && d->type == CDECL_DECL_FUNCTION) {
375 fprintf(stderr, "function pointers cannot be restrict-qualified\n");
384 * Traverse the parse tree, calling a function on every declarator in a
385 * depth-first preorder traversal. The function is given a pointer to the
386 * declarator as well as to the pointer which was used to reach that
387 * declarator: this can be used to rewrite entire subtrees.
389 static bool forall_declarators(struct cdecl *decl,
390 int f(struct cdecl_declarator **, struct cdecl_declarator *))
392 struct cdecl_declarator *d, **p;
394 for (p = &decl->declarators, d = *p; d; p = &d->child, d = *p) {
407 if (d->type == CDECL_DECL_FUNCTION) {
410 for (i = d->u.function.parameters; i; i = i->next) {
411 if (!forall_declarators(i, f))
420 struct cdecl *cdecl_parse_decl(const char *declstr)
422 struct cdecl_declspec *norm_specs;
423 YY_BUFFER_STATE state;
430 rc = cdecl__yylex_init(&scanner);
434 state = cdecl__yy_scan_string(declstr, scanner);
435 rc = cdecl__yyparse(scanner, &decl);
436 cdecl__yy_delete_buffer(state, scanner);
437 cdecl__yylex_destroy(scanner);
443 * Since the top-level specifiers are shared between each top-level
444 * declarator, we need to normalize them once and then propagate the
445 * new specifier list.
447 norm_specs = cdecl__normalize_specs(decl->specifiers);
448 for (struct cdecl *i = decl; i; i = i->next) {
449 i->specifiers = norm_specs;
452 /* Now perform checks and simplifications on each declarator. */
453 for (struct cdecl *i = decl; i; i = i->next) {
454 if (!forall_declarators(i, reduce_parentheses))
456 if (!forall_declarators(i, simplify_functions))
458 if (!forall_declarators(i, check_parameters))
460 if (!forall_declarators(i, check_rettypes))
462 if (!forall_declarators(i, check_arrays))
464 if (!forall_declarators(i, normalize_specs))
466 if (!forall_declarators(i, check_qualifiers))
469 if (!valid_declspecs(i, true))
472 if (cdecl_is_abstract(i->declarators)
473 && (i != decl || i->next)) {
474 fprintf(stderr, "mixing type names and declarations is not allowed\n");
485 struct cdecl *cdecl_parse_english(const char *english)
487 YY_BUFFER_STATE state;
494 rc = cdecl__yylex_init_extra(true, &scanner);
498 state = cdecl__yy_scan_string(english, scanner);
499 rc = cdecl__yyparse(scanner, &decl);
500 cdecl__yy_delete_buffer(state, scanner);
501 cdecl__yylex_destroy(scanner);
506 for (struct cdecl *i = decl; i; i = i->next) {
507 i->specifiers = cdecl__normalize_specs(i->specifiers);
509 if (!forall_declarators(i, check_parameters))
511 if (!forall_declarators(i, check_rettypes))
513 if (!forall_declarators(i, check_arrays))
515 if (!forall_declarators(i, normalize_specs))
517 if (!forall_declarators(i, check_qualifiers))
520 if (!valid_declspecs(i, true))
530 void cdecl_free(struct cdecl *decl)