2 * Parse and validate C declarations.
3 * Copyright © 2011-2012, 2020 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 <http://www.gnu.org/licenses/>.
28 #include "normalize.h"
31 * Verify the declaration specifiers of a declaration. If top is true, treat
32 * this as a top-level declaration. Otherwise, treat this as a function
33 * parameter (which carries additional constraints).
35 static bool valid_declspecs(struct cdecl *decl, bool top)
37 struct cdecl_declspec *specs = decl->specifiers;
38 struct cdecl_declarator *d = decl->declarators;
39 bool abstract = cdecl_is_abstract(d);
40 unsigned num_storage = 0;
41 unsigned long typemap;
43 typemap = cdecl__build_typemap(specs);
47 for (struct cdecl_declspec *c = specs; c; c = c->next) {
48 switch (cdecl_spec_kind(c)) {
50 if (c->type == CDECL_TYPE_VOID &&
51 (d->type == CDECL_DECL_IDENT
52 || d->type == CDECL_DECL_ARRAY)) {
53 fprintf(stderr, "invalid declaration of type void\n");
58 if (top && abstract) {
59 fprintf(stderr, "type names cannot have storage-class specifiers\n");
63 if (!top && c->type != CDECL_STOR_REGISTER) {
64 fprintf(stderr, "function parameters may only have register storage\n");
68 if (++num_storage > 1) {
69 fprintf(stderr, "too many storage-class specifiers\n");
75 * Restrict qualifiers are only valid in the
76 * pointer qualifier list, which isn't checked here.
78 if (c->type == CDECL_QUAL_RESTRICT) {
79 fprintf(stderr, "only pointer types can be restrict-qualified.\n");
85 fprintf(stderr, "type names cannot have function specifiers\n");
89 if (!top || d->type != CDECL_DECL_FUNCTION) {
90 fprintf(stderr, "only function declarations may have function specifiers.\n");
103 * The C grammar leaves ambiguous some cases where parentheses represent a
104 * function declarator or just parentheses. The language uses additional
105 * context (whether or not a typedef is in scope, etc.) to resolve these
106 * ambiguities, but we don't have access to that kind of information.
108 * The cdecl99 parser uses an unambiguous grammar which treats almost
109 * everything as a function, and thus considers things like 'int (x)' to
110 * be a function type with a single parameter of type 'x' (a typedef name),
111 * returning int. This can result in very complicated types for simple
112 * declarations. Ideally, cdecl99 should try and find the "simplest"
113 * explanation for a given declaration.
115 * Whether or not it achieves the simplest explanation, we apply a simple rule:
116 * if a declarator could be interpreted as something other than a function,
119 * Since cdecl99 supports things like [*] in any context (in C, such constructs
120 * are only valid in function parameter lists), we don't treat them specially
124 static struct cdecl_declarator *reduce_function(struct cdecl *param)
126 struct cdecl_declspec *spec = param->specifiers;
127 struct cdecl_declarator *decl = param->declarators;
128 struct cdecl_declarator *last;
130 for (last = decl; last && last->type != CDECL_DECL_NULL;)
136 last->type = CDECL_DECL_IDENT;
137 last->u.ident = spec->ident;
144 static bool function_is_reducible(struct cdecl_declarator *d)
146 if (d->type != CDECL_DECL_FUNCTION)
148 if (d->child->type != CDECL_DECL_NULL)
149 return false; /* e.g., int (*)(x) */
151 if (!d->u.function.parameters)
152 return false; /* e.g., int f() */
153 if (d->u.function.parameters->next)
154 return false; /* e.g., int (x, y) */
155 if (d->u.function.variadic)
156 return false; /* e.g., int (x, ...) */
158 if (d->u.function.parameters->specifiers->type != CDECL_TYPE_IDENT)
159 return false; /* e.g. int (int) */
160 if (d->u.function.parameters->specifiers->next)
161 return false; /* e.g. int (size_t const) */
162 if (d->u.function.parameters->declarators->type == CDECL_DECL_POINTER)
163 return false; /* e.g. int (x *) */
169 simplify_functions(struct cdecl_declarator **p, struct cdecl_declarator *d)
171 struct cdecl_declarator *new;
173 if (!function_is_reducible(d))
176 new = reduce_function(d->u.function.parameters);
178 return 0; /* e.g. int (foo bar) */
187 * The parser's bias towards considering things as functions whenever possible
188 * makes nested parentheses tricky. (x) is considered to be part of a function
189 * declarator until simplify_functions converts it. The problem is that
190 * (((x))) is not valid as part of a function declarator, but it *is* valid
191 * as an identifier enclosed 3 times in parentheses. This is complicated by
192 * the fact that things like (((int))) are not valid anywhere.
194 * To avoid ambiguities, the parser actually emits a "function" declarator for
195 * every pair of parentheses. The ones that can't reasonably be functions
196 * consist of a single "parameter" with no declaration specifiers (note that
197 * every valid function parameter will have at least one type specifier).
199 * This pass is to remove these fake functions from the parse tree. We take
200 * care to avoid turning invalid things like ((int)) into valid things like
201 * (int) by observing that the only valid function declarators that appear
202 * in these "fake" parentheses are those that have a non-null child declarator
203 * (for instance, int ((*)(int)) *or* those that will be eliminated by the
204 * simplify_functions pass.
208 reduce_parentheses(struct cdecl_declarator **p, struct cdecl_declarator *d)
212 if (d->type != CDECL_DECL_FUNCTION)
215 param = d->u.function.parameters;
216 if (param && param->specifiers == NULL) {
217 struct cdecl_declarator *decl;
219 assert(!param->next);
221 decl = param->declarators;
222 if (decl->type == CDECL_DECL_NULL) {
225 d->u.function.parameters = NULL;
229 if (d->child->type != CDECL_DECL_NULL) {
230 fprintf(stderr, "invalid function parameter\n");
240 * We may have replaced d with another fake function which
241 * also needs to be eliminated.
243 if (reduce_parentheses(p, decl) < 0)
247 * If the remaining declarator is a function, make sure it's
248 * valid by checking its reducibility.
251 if (decl->type == CDECL_DECL_FUNCTION
252 && decl->child->type == CDECL_DECL_NULL
253 && !function_is_reducible(decl)) {
254 fprintf(stderr, "too many parentheses in function\n");
265 * Function parameters and return types have a few restrictions that are
266 * really easy to check in comparison to the above absurdity.
269 check_parameters(struct cdecl_declarator **p, struct cdecl_declarator *d)
271 struct cdecl_declspec *spec;
273 bool has_void = false;
275 if (d->type != CDECL_DECL_FUNCTION)
278 for (param = d->u.function.parameters; param; param = param->next) {
279 if (!valid_declspecs(param, false))
282 /* Check for "void" function parameters as a special case. */
283 for (spec = param->specifiers; spec; spec = spec->next) {
284 if (param->declarators->type != CDECL_DECL_NULL)
286 if (spec->type != CDECL_TYPE_VOID)
289 if (spec != param->specifiers || spec->next != NULL) {
290 fprintf(stderr, "void parameter must not have extra specifiers\n");
292 } else if (d->u.function.parameters->next) {
293 fprintf(stderr, "a void parameter must stand alone\n");
295 } else if (d->u.function.variadic) {
296 fprintf(stderr, "variadic functions cannot have a void parameter\n");
306 * Functions cannot return arrays or functions. Since the parse tree is
307 * "inside-out", we need to look for functions as the child declarator.
310 check_rettypes(struct cdecl_declarator **p, struct cdecl_declarator *d)
312 if (!d->child || d->child->type != CDECL_DECL_FUNCTION)
316 case CDECL_DECL_FUNCTION:
317 fprintf(stderr, "functions cannot return functions\n");
319 case CDECL_DECL_ARRAY:
320 fprintf(stderr, "functions cannot return arrays\n");
328 check_arrays(struct cdecl_declarator **p, struct cdecl_declarator *d)
330 if (!d->child || d->child->type != CDECL_DECL_ARRAY)
334 case CDECL_DECL_FUNCTION:
335 fprintf(stderr, "array members cannot be functions\n");
343 normalize_specs(struct cdecl_declarator **p, struct cdecl_declarator *d)
345 struct cdecl_function *func;
346 struct cdecl_pointer *ptr;
349 case CDECL_DECL_POINTER:
351 ptr->qualifiers = cdecl__normalize_specs(ptr->qualifiers);
353 case CDECL_DECL_FUNCTION:
354 func = &d->u.function;
355 for (struct cdecl *i = func->parameters; i; i = i->next)
356 i->specifiers = cdecl__normalize_specs(i->specifiers);
364 check_qualifiers(struct cdecl_declarator **p, struct cdecl_declarator *d)
366 struct cdecl_declspec *spec;
367 struct cdecl_pointer *ptr;
369 if (!d->child || d->child->type != CDECL_DECL_POINTER)
372 ptr = &d->child->u.pointer;
373 for (spec = ptr->qualifiers; spec; spec = spec->next) {
374 if (spec->type == CDECL_QUAL_RESTRICT
375 && d->type == CDECL_DECL_FUNCTION) {
376 fprintf(stderr, "function pointers cannot be restrict-qualified\n");
385 * Traverse the parse tree, calling a function on every declarator in a
386 * depth-first preorder traversal. The function is given a pointer to the
387 * declarator as well as to the pointer which was used to reach that
388 * declarator: this can be used to rewrite entire subtrees.
390 static bool forall_declarators(struct cdecl *decl,
391 int f(struct cdecl_declarator **, struct cdecl_declarator *))
393 struct cdecl_declarator *d, **p;
395 for (p = &decl->declarators, d = *p; d; p = &d->child, d = *p) {
408 if (d->type == CDECL_DECL_FUNCTION) {
411 for (i = d->u.function.parameters; i; i = i->next) {
412 if (!forall_declarators(i, f))
421 struct cdecl *cdecl_parse_decl(const char *declstr)
423 struct cdecl_declspec *norm_specs;
424 YY_BUFFER_STATE state;
431 rc = cdecl__yylex_init(&scanner);
435 state = cdecl__yy_scan_string(declstr, scanner);
436 rc = cdecl__yyparse(scanner, &decl);
437 cdecl__yy_delete_buffer(state, scanner);
438 cdecl__yylex_destroy(scanner);
444 * Since the top-level specifiers are shared between each top-level
445 * declarator, we need to normalize them once and then propagate the
446 * new specifier list.
448 norm_specs = cdecl__normalize_specs(decl->specifiers);
450 for (struct cdecl *i = decl; i; i = i->next) {
451 i->specifiers = norm_specs;
453 if (!forall_declarators(i, reduce_parentheses))
455 if (!forall_declarators(i, simplify_functions))
457 if (!forall_declarators(i, check_parameters))
459 if (!forall_declarators(i, check_rettypes))
461 if (!forall_declarators(i, check_arrays))
463 if (!forall_declarators(i, normalize_specs))
465 if (!forall_declarators(i, check_qualifiers))
468 if (!valid_declspecs(i, true))
471 if (cdecl_is_abstract(i->declarators)
472 && (i != decl || i->next)) {
473 fprintf(stderr, "mixing type names and declarations is not allowed\n");
484 struct cdecl *cdecl_parse_english(const char *english)
486 YY_BUFFER_STATE state;
493 rc = cdecl__yylex_init_extra(true, &scanner);
497 state = cdecl__yy_scan_string(english, scanner);
498 rc = cdecl__yyparse(scanner, &decl);
499 cdecl__yy_delete_buffer(state, scanner);
500 cdecl__yylex_destroy(scanner);
505 for (struct cdecl *i = decl; i; i = i->next) {
506 i->specifiers = cdecl__normalize_specs(i->specifiers);
508 if (!forall_declarators(i, check_parameters))
510 if (!forall_declarators(i, check_rettypes))
512 if (!forall_declarators(i, check_arrays))
514 if (!forall_declarators(i, normalize_specs))
516 if (!forall_declarators(i, check_qualifiers))
519 if (!valid_declspecs(i, true))
529 void cdecl_free(struct cdecl *decl)