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/>.
30 * Determine if a declarator declares an identifier (other than a function
33 static bool is_abstract(struct cdecl_declarator *d)
38 return d->type != CDECL_DECL_IDENT;
42 * Verify the declaration specifiers of a declaration. If top is true, treat
43 * this as a top-level declaration. Otherwise, treat this as a function
44 * parameter (which carries additional constraints).
46 static bool valid_declspecs(struct cdecl *decl, bool top)
48 struct cdecl_declspec *specs = decl->specifiers;
49 struct cdecl_declarator *d = decl->declarators;
50 bool abstract = is_abstract(d);
51 unsigned num_storage = 0;
52 unsigned long typemap;
54 typemap = cdecl__build_typemap(specs);
58 for (struct cdecl_declspec *c = specs; c; c = c->next) {
59 switch (cdecl_spec_kind(c)) {
61 if (c->type == CDECL_TYPE_VOID &&
62 (d->type == CDECL_DECL_IDENT
63 || d->type == CDECL_DECL_ARRAY)) {
64 fprintf(stderr, "invalid declaration of type void\n");
69 if (top && abstract) {
70 fprintf(stderr, "type names cannot have storage-class specifiers\n");
74 if (!top && c->type != CDECL_STOR_REGISTER) {
75 fprintf(stderr, "function parameters may only have register storage\n");
79 if (++num_storage > 1) {
80 fprintf(stderr, "too many storage-class specifiers\n");
86 * Restrict qualifiers are only valid in the
87 * pointer qualifier list, which isn't checked here.
89 if (c->type == CDECL_QUAL_RESTRICT) {
90 fprintf(stderr, "only pointer types can be restrict-qualified.\n");
96 fprintf(stderr, "type names cannot have function specifiers\n");
100 if (!top || d->type != CDECL_DECL_FUNCTION) {
101 fprintf(stderr, "only function declarations may have function specifiers.\n");
114 * The C grammar leaves ambiguous some cases where parentheses represent a
115 * function declarator or just parentheses. The language uses additional
116 * context (whether or not a typedef is in scope, etc.) to resolve these
117 * ambiguities, but we don't have access to that kind of information.
119 * The cdecl99 parser uses an unambiguous grammar which treats almost
120 * everything as a function, and thus considers things like 'int (x)' to
121 * be a function type with a single parameter of type 'x' (a typedef name),
122 * returning int. This can result in very complicated types for simple
123 * declarations. Ideally, cdecl99 should try and find the "simplest"
124 * explanation for a given declaration.
126 * Whether or not it achieves the simplest explanation, we apply a simple rule:
127 * if a declarator could be interpreted as something other than a function,
130 * - The function declarator has a null child declarator.
131 * - The function declarator has exactly one parameter, and is not variadic.
132 * - The function parameter has a type specifier, and it is a typedef name.
133 * - The function parameter has no other declaration specifiers.
134 * - The function parameter does not declare an identifier.
136 * Since cdecl99 supports things like [*] in any context (in C, such constructs
137 * are only valid in function parameter lists), we don't treat them specially
141 static struct cdecl_declarator *reduce_function(struct cdecl *param)
143 struct cdecl_declspec *spec = param->specifiers;
144 struct cdecl_declarator *decl = param->declarators;
145 struct cdecl_declarator *last;
147 for (last = decl; last && last->type != CDECL_DECL_NULL;)
153 last->type = CDECL_DECL_IDENT;
154 last->u.ident = spec->ident;
161 static bool function_is_reducible(struct cdecl_declarator *d)
163 if (d->type != CDECL_DECL_FUNCTION)
165 if (d->child->type != CDECL_DECL_NULL)
166 return false; /* e.g., int (*)(x) */
168 if (!d->u.function.parameters)
169 return false; /* e.g., int f() */
170 if (d->u.function.parameters->next)
171 return false; /* e.g., int (x, y) */
172 if (d->u.function.variadic)
173 return false; /* e.g., int (x, ...) */
175 if (d->u.function.parameters->specifiers->type != CDECL_TYPE_IDENT)
176 return false; /* e.g. int (int) */
177 if (d->u.function.parameters->specifiers->next)
178 return false; /* e.g. int (size_t const) */
184 simplify_functions(struct cdecl_declarator **p, struct cdecl_declarator *d)
186 struct cdecl_declarator *new;
188 if (!function_is_reducible(d))
191 new = reduce_function(d->u.function.parameters);
193 return 0; /* e.g. int (foo bar) */
202 * The parser's bias towards considering things as functions whenever possible
203 * makes nested parentheses tricky. (x) is considered to be part of a function
204 * declarator until simplify_functions converts it. The problem is that
205 * (((x))) is not valid as part of a function declarator, but it *is* valid
206 * as an identifier enclosed 3 times in parentheses. This is complicated by
207 * the fact that things like (((int))) are not valid anywhere.
209 * To avoid ambiguities, the parser actually emits a "function" declarator for
210 * every pair of parentheses. The ones that can't reasonably be functions
211 * consist of a single "parameter" with no declaration specifiers (note that
212 * every valid function parameter will have at least one type specifier).
214 * This pass is to remove these fake functions from the parse tree. We take
215 * care to avoid turning invalid things like ((int)) into valid things like
216 * (int) by observing that the only valid function declarators that appear
217 * in these "fake" parentheses are those that have a non-null child declarator
218 * (for instance, int ((*)(int)) *or* those that will be eliminated by the
219 * simplify_functions pass.
223 reduce_parentheses(struct cdecl_declarator **p, struct cdecl_declarator *d)
227 if (d->type != CDECL_DECL_FUNCTION)
230 param = d->u.function.parameters;
231 if (param && param->specifiers == NULL) {
232 struct cdecl_declarator *decl;
234 assert(!param->next);
236 decl = param->declarators;
237 if (decl->type == CDECL_DECL_NULL) {
240 d->u.function.parameters = NULL;
244 if (d->child->type != CDECL_DECL_NULL) {
245 fprintf(stderr, "invalid function parameter\n");
255 * We may have replaced d with another fake function which
256 * also needs to be eliminated.
258 if (reduce_parentheses(p, decl) < 0)
262 * If the remaining declarator is a function, make sure it's
263 * valid by checking its reducibility.
266 if (decl->type == CDECL_DECL_FUNCTION
267 && decl->child->type == CDECL_DECL_NULL
268 && !function_is_reducible(decl)) {
269 fprintf(stderr, "too many parentheses in function\n");
280 * Function parameters and return types have a few restrictions that are
281 * really easy to check in comparison to the above absurdity.
284 check_parameters(struct cdecl_declarator **p, struct cdecl_declarator *d)
286 struct cdecl_declspec *spec;
288 bool has_void = false;
290 if (d->type != CDECL_DECL_FUNCTION)
293 for (param = d->u.function.parameters; param; param = param->next) {
294 if (!valid_declspecs(param, false))
297 for (spec = param->specifiers; spec; spec = spec->next) {
298 if (spec->type == CDECL_TYPE_VOID
299 && param->declarators->type == CDECL_DECL_NULL)
304 if (has_void && d->u.function.parameters->next) {
305 fprintf(stderr, "a void parameter must stand alone\n");
307 } else if (has_void && d->u.function.variadic) {
308 fprintf(stderr, "variadic functions cannot have a void parameter\n");
316 * Functions cannot return arrays or functions. Since the parse tree is
317 * "inside-out", we need to look for functions as the child declarator.
320 check_rettypes(struct cdecl_declarator **p, struct cdecl_declarator *d)
322 if (!d->child || d->child->type != CDECL_DECL_FUNCTION)
326 case CDECL_DECL_FUNCTION:
327 fprintf(stderr, "functions cannot return functions\n");
329 case CDECL_DECL_ARRAY:
330 fprintf(stderr, "functions cannot return arrays\n");
338 check_arrays(struct cdecl_declarator **p, struct cdecl_declarator *d)
340 if (!d->child || d->child->type != CDECL_DECL_ARRAY)
344 case CDECL_DECL_FUNCTION:
345 fprintf(stderr, "array members cannot be functions\n");
353 * Traverse the parse tree, calling a function on every declarator in a
354 * depth-first preorder traversal. The function is given a pointer to the
355 * declarator as well as to the pointer which was used to reach that
356 * declarator: this can be used to rewrite entire subtrees.
358 static bool forall_declarators(struct cdecl *decl,
359 int f(struct cdecl_declarator **, struct cdecl_declarator *))
361 struct cdecl_declarator *d, **p;
363 for (p = &decl->declarators, d = *p; d; p = &d->child, d = *p) {
376 if (d->type == CDECL_DECL_FUNCTION) {
379 for (i = d->u.function.parameters; i; i = i->next) {
380 if (!forall_declarators(i, f))
389 struct cdecl *cdecl_parse_decl(const char *declstr)
391 YY_BUFFER_STATE state;
398 rc = cdecl__yylex_init(&scanner);
402 state = cdecl__yy_scan_string(declstr, scanner);
403 rc = cdecl__yyparse(scanner, &decl);
404 cdecl__yy_delete_buffer(state, scanner);
405 cdecl__yylex_destroy(scanner);
410 for (struct cdecl *i = decl; i; i = i->next) {
411 if (!forall_declarators(i, reduce_parentheses))
413 if (!forall_declarators(i, simplify_functions))
415 if (!forall_declarators(i, check_parameters))
417 if (!forall_declarators(i, check_rettypes))
419 if (!forall_declarators(i, check_arrays))
422 if (!valid_declspecs(i, true))
425 if (is_abstract(i->declarators) && (i != decl || i->next)) {
426 fprintf(stderr, "mixing type names and declarations is not allowed\n");
437 struct cdecl *cdecl_parse_english(const char *english)
439 YY_BUFFER_STATE state;
446 rc = cdecl__yylex_init_extra(true, &scanner);
450 state = cdecl__yy_scan_string(english, scanner);
451 rc = cdecl__yyparse(scanner, &decl);
452 cdecl__yy_delete_buffer(state, scanner);
453 cdecl__yylex_destroy(scanner);
458 for (struct cdecl *i = decl; i; i = i->next) {
459 if (!forall_declarators(i, check_parameters))
461 if (!forall_declarators(i, check_rettypes))
463 if (!forall_declarators(i, check_arrays))
466 if (!valid_declspecs(i, true))
476 void cdecl_free(struct cdecl *decl)