2 * Parse and validate C declarations.
3 * Copyright © 2011 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 * Determine if a declarator declares an identifier (other than a function
34 static bool is_abstract(struct cdecl_declarator *d)
39 return d->type != CDECL_DECL_IDENT;
43 * Verify the declaration specifiers of a declaration. If top is true, treat
44 * this as a top-level declaration. Otherwise, treat this as a function
45 * parameter (which carries additional constraints).
47 static bool valid_declspecs(struct cdecl *decl, bool top)
49 struct cdecl_declspec *specs = decl->specifiers;
50 struct cdecl_declarator *d = decl->declarators;
51 bool abstract = is_abstract(d);
52 unsigned num_storage = 0;
53 unsigned long typemap;
55 typemap = cdecl__build_typemap(specs);
59 for (struct cdecl_declspec *c = specs; c; c = c->next) {
60 switch (cdecl_spec_kind(c)) {
62 if (c->type == CDECL_TYPE_VOID &&
63 (d->type == CDECL_DECL_IDENT
64 || d->type == CDECL_DECL_ARRAY)) {
65 fprintf(stderr, "invalid declaration of type void\n");
70 if (top && abstract) {
71 fprintf(stderr, "type names cannot have storage-class specifiers\n");
75 if (!top && c->type != CDECL_STOR_REGISTER) {
76 fprintf(stderr, "function parameters may only have register storage\n");
80 if (++num_storage > 1) {
81 fprintf(stderr, "too many storage-class specifiers\n");
87 * Restrict qualifiers are only valid in the
88 * pointer qualifier list, which isn't checked here.
90 if (c->type == CDECL_QUAL_RESTRICT) {
91 fprintf(stderr, "only pointer types can be restrict-qualified.\n");
97 fprintf(stderr, "type names cannot have function specifiers\n");
101 if (!top || d->type != CDECL_DECL_FUNCTION) {
102 fprintf(stderr, "only function declarations may have function specifiers.\n");
115 * The C grammar leaves ambiguous some cases where parentheses represent a
116 * function declarator or just parentheses. The language uses additional
117 * context (whether or not a typedef is in scope, etc.) to resolve these
118 * ambiguities, but we don't have access to that kind of information.
120 * The cdecl99 parser uses an unambiguous grammar which treats almost
121 * everything as a function, and thus considers things like 'int (x)' to
122 * be a function type with a single parameter of type 'x' (a typedef name),
123 * returning int. This can result in very complicated types for simple
124 * declarations. Ideally, cdecl99 should try and find the "simplest"
125 * explanation for a given declaration.
127 * Whether or not it achieves the simplest explanation, we apply a simple rule:
128 * if a declarator could be interpreted as something other than a function,
131 * - The function declarator has a null child declarator.
132 * - The function declarator has exactly one parameter, and is not variadic.
133 * - The function parameter has a type specifier, and it is a typedef name.
134 * - The function parameter has no other declaration specifiers.
135 * - The function parameter does not declare an identifier.
137 * Since cdecl99 supports things like [*] in any context (in C, such constructs
138 * are only valid in function parameter lists), we don't treat them specially
142 static struct cdecl_declarator *reduce_function(struct cdecl *param)
144 struct cdecl_declspec *spec = param->specifiers;
145 struct cdecl_declarator *decl = param->declarators;
146 struct cdecl_declarator *last;
148 for (last = decl; last && last->type != CDECL_DECL_NULL;)
154 last->type = CDECL_DECL_IDENT;
155 last->u.ident = spec->ident;
162 static bool function_is_reducible(struct cdecl_declarator *d)
164 if (d->type != CDECL_DECL_FUNCTION)
166 if (d->child->type != CDECL_DECL_NULL)
167 return false; /* e.g., int (*)(x) */
169 if (!d->u.function.parameters)
170 return false; /* e.g., int f() */
171 if (d->u.function.parameters->next)
172 return false; /* e.g., int (x, y) */
173 if (d->u.function.variadic)
174 return false; /* e.g., int (x, ...) */
176 if (d->u.function.parameters->specifiers->type != CDECL_TYPE_IDENT)
177 return false; /* e.g. int (int) */
178 if (d->u.function.parameters->specifiers->next)
179 return false; /* e.g. int (size_t const) */
185 simplify_functions(struct cdecl_declarator **p, struct cdecl_declarator *d)
187 struct cdecl_declarator *new;
189 if (!function_is_reducible(d))
192 new = reduce_function(d->u.function.parameters);
194 return 0; /* e.g. int (foo bar) */
203 * The parser's bias towards considering things as functions whenever possible
204 * makes nested parentheses tricky. (x) is considered to be part of a function
205 * declarator until simplify_functions converts it. The problem is that
206 * (((x))) is not valid as part of a function declarator, but it *is* valid
207 * as an identifier enclosed 3 times in parentheses. This is complicated by
208 * the fact that things like (((int))) are not valid anywhere.
210 * To avoid ambiguities, the parser actually emits a "function" declarator for
211 * every pair of parentheses. The ones that can't reasonably be functions
212 * consist of a single "parameter" with no declaration specifiers (note that
213 * every valid function parameter will have at least one type specifier).
215 * This pass is to remove these fake functions from the parse tree. We take
216 * care to avoid turning invalid things like ((int)) into valid things like
217 * (int) by observing that the only valid function declarators that appear
218 * in these "fake" parentheses are those that have a non-null child declarator
219 * (for instance, int ((*)(int)) *or* those that will be eliminated by the
220 * simplify_functions pass.
224 reduce_parentheses(struct cdecl_declarator **p, struct cdecl_declarator *d)
228 if (d->type != CDECL_DECL_FUNCTION)
231 param = d->u.function.parameters;
232 if (param && param->specifiers == NULL) {
233 struct cdecl_declarator *decl;
235 assert(!param->next);
237 decl = param->declarators;
238 if (decl->type == CDECL_DECL_NULL) {
241 d->u.function.parameters = NULL;
245 if (d->child->type != CDECL_DECL_NULL) {
246 fprintf(stderr, "invalid function parameter\n");
256 * We may have replaced d with another fake function which
257 * also needs to be eliminated.
259 if (reduce_parentheses(p, decl) < 0)
263 * If the remaining declarator is a function, make sure it's
264 * valid by checking its reducibility.
267 if (decl->type == CDECL_DECL_FUNCTION
268 && decl->child->type == CDECL_DECL_NULL
269 && !function_is_reducible(decl)) {
270 fprintf(stderr, "too many parentheses in function\n");
281 * Function parameters and return types have a few restrictions that are
282 * really easy to check in comparison to the above absurdity.
285 check_parameters(struct cdecl_declarator **p, struct cdecl_declarator *d)
287 struct cdecl_declspec *spec;
289 bool has_void = false;
291 if (d->type != CDECL_DECL_FUNCTION)
294 for (param = d->u.function.parameters; param; param = param->next) {
295 if (!valid_declspecs(param, false))
298 for (spec = param->specifiers; spec; spec = spec->next) {
299 if (spec->type == CDECL_TYPE_VOID
300 && param->declarators->type == CDECL_DECL_NULL)
305 if (has_void && d->u.function.parameters->next) {
306 fprintf(stderr, "a void parameter must stand alone\n");
308 } else if (has_void && d->u.function.variadic) {
309 fprintf(stderr, "variadic functions cannot have a void parameter\n");
317 * Functions cannot return arrays or functions. Since the parse tree is
318 * "inside-out", we need to look for functions as the child declarator.
321 check_rettypes(struct cdecl_declarator **p, struct cdecl_declarator *d)
323 if (!d->child || d->child->type != CDECL_DECL_FUNCTION)
327 case CDECL_DECL_FUNCTION:
328 fprintf(stderr, "functions cannot return functions\n");
330 case CDECL_DECL_ARRAY:
331 fprintf(stderr, "functions cannot return arrays\n");
339 check_arrays(struct cdecl_declarator **p, struct cdecl_declarator *d)
341 if (!d->child || d->child->type != CDECL_DECL_ARRAY)
345 case CDECL_DECL_FUNCTION:
346 fprintf(stderr, "array members cannot be functions\n");
354 normalize_specs(struct cdecl_declarator **p, struct cdecl_declarator *d)
356 struct cdecl_function *func;
357 struct cdecl_pointer *ptr;
360 case CDECL_DECL_POINTER:
362 ptr->qualifiers = cdecl__normalize_specs(ptr->qualifiers);
364 case CDECL_DECL_FUNCTION:
365 func = &d->u.function;
366 for (struct cdecl *i = func->parameters; i; i = i->next)
367 i->specifiers = cdecl__normalize_specs(i->specifiers);
375 * Traverse the parse tree, calling a function on every declarator in a
376 * depth-first preorder traversal. The function is given a pointer to the
377 * declarator as well as to the pointer which was used to reach that
378 * declarator: this can be used to rewrite entire subtrees.
380 static bool forall_declarators(struct cdecl *decl,
381 int f(struct cdecl_declarator **, struct cdecl_declarator *))
383 struct cdecl_declarator *d, **p;
385 for (p = &decl->declarators, d = *p; d; p = &d->child, d = *p) {
398 if (d->type == CDECL_DECL_FUNCTION) {
401 for (i = d->u.function.parameters; i; i = i->next) {
402 if (!forall_declarators(i, f))
411 struct cdecl *cdecl_parse_decl(const char *declstr)
413 struct cdecl_declspec *norm_specs;
414 YY_BUFFER_STATE state;
421 rc = cdecl__yylex_init(&scanner);
425 state = cdecl__yy_scan_string(declstr, scanner);
426 rc = cdecl__yyparse(scanner, &decl);
427 cdecl__yy_delete_buffer(state, scanner);
428 cdecl__yylex_destroy(scanner);
434 * Since the top-level specifiers are shared between each top-level
435 * declarator, we need to normalize them once and then propagate the
436 * new specifier list.
438 norm_specs = cdecl__normalize_specs(decl->specifiers);
440 for (struct cdecl *i = decl; i; i = i->next) {
441 i->specifiers = norm_specs;
443 if (!forall_declarators(i, reduce_parentheses))
445 if (!forall_declarators(i, simplify_functions))
447 if (!forall_declarators(i, check_parameters))
449 if (!forall_declarators(i, check_rettypes))
451 if (!forall_declarators(i, check_arrays))
453 if (!forall_declarators(i, normalize_specs))
456 if (!valid_declspecs(i, true))
459 if (is_abstract(i->declarators) && (i != decl || i->next)) {
460 fprintf(stderr, "mixing type names and declarations is not allowed\n");
471 struct cdecl *cdecl_parse_english(const char *english)
473 YY_BUFFER_STATE state;
480 rc = cdecl__yylex_init_extra(true, &scanner);
484 state = cdecl__yy_scan_string(english, scanner);
485 rc = cdecl__yyparse(scanner, &decl);
486 cdecl__yy_delete_buffer(state, scanner);
487 cdecl__yylex_destroy(scanner);
492 for (struct cdecl *i = decl; i; i = i->next) {
493 i->specifiers = cdecl__normalize_specs(i->specifiers);
495 if (!forall_declarators(i, check_parameters))
497 if (!forall_declarators(i, check_rettypes))
499 if (!forall_declarators(i, check_arrays))
501 if (!forall_declarators(i, normalize_specs))
504 if (!valid_declspecs(i, true))
514 void cdecl_free(struct cdecl *decl)