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 * Determine if a declarator declares an identifier (other than a function
31 static bool is_abstract(struct cdecl_declarator *d)
36 return d->type != CDECL_DECL_IDENT;
40 * Verify the declaration specifiers of a declaration. If top is true, treat
41 * this as a top-level declaration. Otherwise, treat this as a function
42 * parameter (which carries additional constraints).
44 static bool valid_declspecs(struct cdecl *decl, bool top)
46 struct cdecl_declspec *specs = decl->specifiers;
47 struct cdecl_declarator *d = decl->declarators;
48 bool abstract = is_abstract(d);
49 unsigned num_storage = 0;
50 unsigned long typemap;
52 typemap = cdecl__build_typemap(specs);
57 for (struct cdecl_declspec *c = specs; c; c = c->next) {
58 switch (cdecl_spec_kind(c)) {
60 if (c->type == CDECL_TYPE_VOID &&
61 (d->type == CDECL_DECL_IDENT
62 || d->type == CDECL_DECL_ARRAY)) {
63 fprintf(stderr, "invalid declaration of type void\n");
68 if (top && abstract) {
69 fprintf(stderr, "type names cannot have storage-class specifiers\n");
73 if (!top && c->type != CDECL_STOR_REGISTER) {
74 fprintf(stderr, "function parameters may only have register storage\n");
78 if (++num_storage > 1) {
79 fprintf(stderr, "too many storage-class specifiers\n");
85 * Restrict qualifiers are only valid in the
86 * pointer qualifier list, which isn't checked here.
88 if (c->type == CDECL_QUAL_RESTRICT) {
89 fprintf(stderr, "only pointer types can be restrict-qualified.\n");
95 fprintf(stderr, "type names cannot have function specifiers\n");
99 if (!top || d->type != CDECL_DECL_FUNCTION) {
100 fprintf(stderr, "only function declarations may have function specifiers.\n");
113 * The C grammar leaves ambiguous some cases where parentheses represent a
114 * function declarator or just parentheses. The language uses additional
115 * context (whether or not a typedef is in scope, etc.) to resolve these
116 * ambiguities, but we don't have access to that kind of information.
118 * The cdecl99 parser uses an unambiguous grammar which treats almost
119 * everything as a function, and thus considers things like 'int (x)' to
120 * be a function type with a single parameter of type 'x' (a typedef name),
121 * returning int. This can result in very complicated types for simple
122 * declarations. Ideally, cdecl99 should try and find the "simplest"
123 * explanation for a given declaration.
125 * Whether or not it achieves the simplest explanation, we apply a simple rule:
126 * if a declarator could be interpreted as something other than a function,
129 * - The function declarator has a null child declarator.
130 * - The function declarator has exactly one parameter, and is not variadic.
131 * - The function parameter has a type specifier, and it is a typedef name.
132 * - The function parameter has no other declaration specifiers.
133 * - The function parameter does not declare an identifier.
135 * Since cdecl99 supports things like [*] in any context (in C, such constructs
136 * are only valid in function parameter lists), we don't treat them specially
140 static struct cdecl_declarator *reduce_function(struct cdecl *param)
142 struct cdecl_declspec *spec = param->specifiers;
143 struct cdecl_declarator *decl = param->declarators;
144 struct cdecl_declarator *last;
146 for (last = decl; last && last->type != CDECL_DECL_NULL;)
152 last->type = CDECL_DECL_IDENT;
153 last->u.ident = spec->ident;
160 static bool function_is_reducible(struct cdecl_declarator *d)
162 if (d->type != CDECL_DECL_FUNCTION)
164 if (d->child->type != CDECL_DECL_NULL)
165 return false; /* e.g., int (*)(x) */
167 if (!d->u.function.parameters)
168 return false; /* e.g., int f() */
169 if (d->u.function.parameters->next)
170 return false; /* e.g., int (x, y) */
171 if (d->u.function.variadic)
172 return false; /* e.g., int (x, ...) */
174 if (d->u.function.parameters->specifiers->type != CDECL_TYPE_IDENT)
175 return false; /* e.g. int (int) */
176 if (d->u.function.parameters->specifiers->next)
177 return false; /* e.g. int (size_t const) */
183 simplify_functions(struct cdecl_declarator **p, struct cdecl_declarator *d)
185 struct cdecl_declarator *new;
187 if (!function_is_reducible(d))
190 new = reduce_function(d->u.function.parameters);
192 return 0; /* e.g. int (foo bar) */
201 * The parser's bias towards considering things as functions whenever possible
202 * makes nested parentheses tricky. (x) is considered to be part of a function
203 * declarator until simplify_functions converts it. The problem is that
204 * (((x))) is not valid as part of a function declarator, but it *is* valid
205 * as an identifier enclosed 3 times in parentheses. This is complicated by
206 * the fact that things like (((int))) are not valid anywhere.
208 * To avoid ambiguities, the parser actually emits a "function" declarator for
209 * every pair of parentheses. The ones that can't reasonably be functions
210 * consist of a single "parameter" with no declaration specifiers (note that
211 * every valid function parameter will have at least one type specifier).
213 * This pass is to remove these fake functions from the parse tree. We take
214 * care to avoid turning invalid things like ((int)) into valid things like
215 * (int) by observing that the only valid function declarators that appear
216 * in these "fake" parentheses are those that have a non-null child declarator
217 * (for instance, int ((*)(int)) *or* those that will be eliminated by the
218 * simplify_functions pass.
222 reduce_parentheses(struct cdecl_declarator **p, struct cdecl_declarator *d)
226 if (d->type != CDECL_DECL_FUNCTION)
229 param = d->u.function.parameters;
230 if (param && param->specifiers == NULL) {
231 struct cdecl_declarator *decl;
233 assert(!param->next);
235 decl = param->declarators;
236 if (decl->type == CDECL_DECL_NULL) {
239 d->u.function.parameters = NULL;
243 if (d->child->type != CDECL_DECL_NULL) {
244 fprintf(stderr, "invalid function parameter\n");
254 * We may have replaced d with another fake function which
255 * also needs to be eliminated.
257 if (reduce_parentheses(p, decl) < 0)
261 * If the remaining declarator is a function, make sure it's
262 * valid by checking its reducibility.
265 if (decl->type == CDECL_DECL_FUNCTION
266 && decl->child->type == CDECL_DECL_NULL
267 && !function_is_reducible(decl)) {
268 fprintf(stderr, "too many parentheses in function\n");
279 * Function parameters and return types have a few restrictions that are
280 * really easy to check in comparison to the above absurdity.
283 check_parameters(struct cdecl_declarator **p, struct cdecl_declarator *d)
285 struct cdecl_declspec *spec;
287 bool has_void = false;
289 if (d->type != CDECL_DECL_FUNCTION)
292 for (param = d->u.function.parameters; param; param = param->next) {
293 if (!valid_declspecs(param, false))
296 for (spec = param->specifiers; spec; spec = spec->next) {
297 if (spec->type == CDECL_TYPE_VOID
298 && param->declarators->type == CDECL_DECL_NULL)
303 if (has_void && d->u.function.parameters->next) {
304 fprintf(stderr, "a void parameter must stand alone\n");
306 } else if (has_void && d->u.function.variadic) {
307 fprintf(stderr, "variadic functions cannot have a void parameter\n");
315 * Functions cannot return arrays or functions. Since the parse tree is
316 * "inside-out", we need to look for functions as the child declarator.
319 check_rettypes(struct cdecl_declarator **p, struct cdecl_declarator *d)
321 if (!d->child || d->child->type != CDECL_DECL_FUNCTION)
325 case CDECL_DECL_FUNCTION:
326 fprintf(stderr, "functions cannot return functions\n");
328 case CDECL_DECL_ARRAY:
329 fprintf(stderr, "functions cannot return arrays\n");
337 * Traverse the parse tree, calling a function on every declarator in a
338 * depth-first preorder traversal. The function is given a pointer to the
339 * declarator as well as to the pointer which was used to reach that
340 * declarator: this can be used to rewrite entire subtrees.
342 static bool forall_declarators(struct cdecl *decl,
343 int f(struct cdecl_declarator **, struct cdecl_declarator *))
345 struct cdecl_declarator *d, **p;
347 for (p = &decl->declarators, d = *p; d; p = &d->child, d = *p) {
360 if (d->type == CDECL_DECL_FUNCTION) {
363 for (i = d->u.function.parameters; i; i = i->next) {
364 if (!forall_declarators(i, f))
373 struct cdecl *cdecl_parse_decl(const char *declstr)
375 YY_BUFFER_STATE state;
379 state = yy_scan_string(declstr);
381 yy_delete_buffer(state);
386 for (struct cdecl *i = decl; i; i = i->next) {
387 if (!forall_declarators(i, reduce_parentheses))
389 if (!forall_declarators(i, simplify_functions))
391 if (!forall_declarators(i, check_parameters))
393 if (!forall_declarators(i, check_rettypes))
396 if (!valid_declspecs(i, true))