2 * Parse and validate C declarations.
3 * Copyright © 2011-2012, 2020-2021, 2023 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"
31 * We can represent type specifiers as a bitmap, which gives us a finite
32 * list of acceptable bitmap values according to the C standard. However,
33 * the "long" specifier is allowed to occur more than once, but only at most
34 * 2 times. Treat it as a special case, assigning an unused bit to represent
37 #define MAP_LLONG_BIT 31
38 #define MAP_LONG_BIT (CDECL_TYPE_LONG-CDECL_SPEC_TYPE)
39 #define CDECL_TYPE_LLONG (CDECL_SPEC_TYPE+MAP_LLONG_BIT)
44 * Convert the declaration specifiers to a bitmap with each bit
45 * corresponding to one specific type specifier.
47 static int valid_typespec(struct cdecl_declspec *s)
49 unsigned long map = 0;
51 for (struct cdecl_declspec *c = s; c; c = c->next) {
54 if (cdecl_spec_kind(c) != CDECL_SPEC_TYPE)
57 bit = c->type - CDECL_SPEC_TYPE;
58 assert(bit < MAP_LLONG_BIT);
61 /* "long" special case */
62 if ((map & bit) == 1ul << MAP_LONG_BIT)
63 bit = 1ul << MAP_LLONG_BIT;
66 if (bit == 1ul << MAP_LLONG_BIT)
67 cdecl__errmsg(CDECL__ETOOLONG);
69 cdecl__errmsg(CDECL__EDUPTYPE);
75 if (typemap_is_valid(map))
79 cdecl__errmsg(CDECL__ENOTYPE);
81 cdecl__errmsg(CDECL__EBADTYPE);
87 * Verify the declaration specifiers of a declaration. If top is true, treat
88 * this as a top-level declaration. Otherwise, treat this as a function
89 * parameter (which carries additional constraints).
91 static bool valid_declspecs(struct cdecl *decl, bool top)
93 struct cdecl_declspec *specs = decl->specifiers;
94 struct cdecl_declarator *d = decl->declarators;
95 bool abstract = cdecl_is_abstract(d);
96 unsigned num_storage = 0;
98 if (!valid_typespec(specs))
101 for (struct cdecl_declspec *c = specs; c; c = c->next) {
102 switch (cdecl_spec_kind(c)) {
103 case CDECL_SPEC_TYPE:
104 if (c->type == CDECL_TYPE_VOID &&
105 (d->type == CDECL_DECL_IDENT
106 || d->type == CDECL_DECL_ARRAY)) {
107 cdecl__errmsg(CDECL__EBADVOID);
111 case CDECL_SPEC_STOR:
112 if (top && abstract) {
113 cdecl__errmsg(CDECL__ETYPESTOR);
117 if (!top && c->type != CDECL_STOR_REGISTER) {
118 cdecl__errmsg(CDECL__EFUNCSTOR);
122 if (++num_storage > 1) {
123 cdecl__errmsg(CDECL__EMANYSTOR);
127 case CDECL_SPEC_QUAL:
129 * Restrict qualifiers are only valid in the
130 * pointer qualifier list, which isn't checked here.
132 if (c->type == CDECL_QUAL_RESTRICT) {
133 cdecl__errmsg(CDECL__EBADQUAL);
137 case CDECL_SPEC_FUNC:
138 if (abstract || !top || d->type != CDECL_DECL_FUNCTION) {
139 cdecl__errmsg(CDECL__ENOTFUNC);
153 * The C grammar leaves ambiguous some cases where parentheses represent a
154 * function declarator or just parentheses. The language uses additional
155 * context (whether or not a typedef is in scope, etc.) to resolve these
156 * ambiguities, but we don't have access to that kind of information.
158 * The cdecl99 parser uses an unambiguous grammar which treats almost
159 * everything as a function, and thus considers things like 'int (x)' to
160 * be a function type with a single parameter of type 'x' (a typedef name),
161 * returning int. This can result in very complicated types for simple
162 * declarations. Ideally, cdecl99 should try and find the "simplest"
163 * explanation for a given declaration.
165 * Whether or not it achieves the simplest explanation, we apply a simple rule:
166 * if a declarator could be interpreted as something other than a function,
169 * Since cdecl99 supports things like [*] in any context (in C, such constructs
170 * are only valid in function parameter lists), we don't treat them specially
174 static struct cdecl_declarator *reduce_function(struct cdecl *param)
176 struct cdecl_declspec *spec = param->specifiers;
177 struct cdecl_declarator *decl = param->declarators;
178 struct cdecl_declarator *last;
180 for (last = decl; last && last->type != CDECL_DECL_NULL;)
186 last->type = CDECL_DECL_IDENT;
187 last->u.ident = spec->ident;
194 static bool function_is_reducible(struct cdecl_declarator *d)
196 if (d->type != CDECL_DECL_FUNCTION)
198 if (d->child->type != CDECL_DECL_NULL)
199 return false; /* e.g., int (*)(x) */
201 if (!d->u.function.parameters)
202 return false; /* e.g., int f() */
203 if (d->u.function.parameters->next)
204 return false; /* e.g., int (x, y) */
205 if (d->u.function.variadic)
206 return false; /* e.g., int (x, ...) */
208 if (d->u.function.parameters->specifiers->type != CDECL_TYPE_IDENT)
209 return false; /* e.g. int (int) */
210 if (d->u.function.parameters->specifiers->next)
211 return false; /* e.g. int (size_t const) */
212 if (d->u.function.parameters->declarators->type == CDECL_DECL_POINTER)
213 return false; /* e.g. int (x *) */
219 simplify_functions(struct cdecl_declarator **p, struct cdecl_declarator *d)
221 struct cdecl_declarator *new;
223 if (!function_is_reducible(d))
226 new = reduce_function(d->u.function.parameters);
228 return 0; /* e.g. int (foo bar) */
237 * The parser's bias towards considering things as functions whenever possible
238 * makes nested parentheses tricky. (x) is considered to be part of a function
239 * declarator until simplify_functions converts it. The problem is that
240 * (((x))) is not valid as part of a function declarator, but it *is* valid
241 * as an identifier enclosed 3 times in parentheses. This is complicated by
242 * the fact that things like (((int))) are not valid anywhere.
244 * To avoid ambiguities, the parser actually emits a "function" declarator for
245 * every pair of parentheses. The ones that can't reasonably be functions
246 * consist of a single "parameter" with no declaration specifiers (note that
247 * every valid function parameter will have at least one type specifier).
249 * This pass is to remove these fake functions from the parse tree. We take
250 * care to avoid turning invalid things like ((int)) into valid things like
251 * (int) by observing that the only valid function declarators that appear
252 * in these "fake" parentheses are those that have a non-null child declarator
253 * (for instance, int ((*)(int)) *or* those that will be eliminated by the
254 * simplify_functions pass.
258 reduce_parentheses(struct cdecl_declarator **p, struct cdecl_declarator *d)
262 if (d->type != CDECL_DECL_FUNCTION)
265 param = d->u.function.parameters;
266 if (param && param->specifiers == NULL) {
267 struct cdecl_declarator *decl;
269 assert(!param->next);
271 decl = param->declarators;
272 if (decl->type == CDECL_DECL_NULL) {
275 d->u.function.parameters = NULL;
279 if (d->child->type != CDECL_DECL_NULL) {
280 cdecl__errmsg(CDECL__EBADPARAM);
290 * We may have replaced d with another fake function which
291 * also needs to be eliminated.
293 if (reduce_parentheses(p, decl) < 0)
297 * If the remaining declarator is a function, make sure it's
298 * valid by checking its reducibility.
301 if (decl->type == CDECL_DECL_FUNCTION
302 && decl->child->type == CDECL_DECL_NULL
303 && !function_is_reducible(decl)) {
304 cdecl__errmsg(CDECL__EMANYPAREN);
315 * Function parameters and return types have a few restrictions that are
316 * really easy to check in comparison to the above absurdity.
319 check_parameters(struct cdecl_declarator **p, struct cdecl_declarator *d)
321 struct cdecl_declspec *spec;
324 if (d->type != CDECL_DECL_FUNCTION)
327 for (param = d->u.function.parameters; param; param = param->next) {
328 if (!valid_declspecs(param, false))
331 /* Check for "void" function parameters as a special case. */
332 for (spec = param->specifiers; spec; spec = spec->next) {
333 if (param->declarators->type != CDECL_DECL_NULL)
335 if (spec->type != CDECL_TYPE_VOID)
338 if (spec != param->specifiers || spec->next != NULL) {
339 cdecl__errmsg(CDECL__EVOIDPARAM);
341 } else if (d->u.function.parameters->next) {
342 cdecl__errmsg(CDECL__EVOIDPARAM);
344 } else if (d->u.function.variadic) {
345 cdecl__errmsg(CDECL__EVOIDPARAM);
355 * Functions cannot return arrays or functions. Since the parse tree is
356 * "inside-out", we need to look for functions as the child declarator.
359 check_rettypes(struct cdecl_declarator **p, struct cdecl_declarator *d)
361 if (!d->child || d->child->type != CDECL_DECL_FUNCTION)
365 case CDECL_DECL_FUNCTION:
366 cdecl__errmsg(CDECL__ERETFUNC);
368 case CDECL_DECL_ARRAY:
369 cdecl__errmsg(CDECL__ERETARRAY);
377 check_arrays(struct cdecl_declarator **p, struct cdecl_declarator *d)
379 if (!d->child || d->child->type != CDECL_DECL_ARRAY)
383 case CDECL_DECL_FUNCTION:
384 cdecl__errmsg(CDECL__EFUNCARRAY);
392 normalize_specs(struct cdecl_declarator **p, struct cdecl_declarator *d)
394 struct cdecl_function *func;
395 struct cdecl_pointer *ptr;
398 case CDECL_DECL_POINTER:
400 ptr->qualifiers = cdecl__normalize_specs(ptr->qualifiers);
402 case CDECL_DECL_FUNCTION:
403 func = &d->u.function;
404 for (struct cdecl *i = func->parameters; i; i = i->next)
405 i->specifiers = cdecl__normalize_specs(i->specifiers);
413 check_qualifiers(struct cdecl_declarator **p, struct cdecl_declarator *d)
415 struct cdecl_declspec *spec;
416 struct cdecl_pointer *ptr;
418 if (!d->child || d->child->type != CDECL_DECL_POINTER)
421 ptr = &d->child->u.pointer;
422 for (spec = ptr->qualifiers; spec; spec = spec->next) {
423 if (spec->type == CDECL_QUAL_RESTRICT
424 && d->type == CDECL_DECL_FUNCTION) {
425 cdecl__errmsg(CDECL__ERESTRICTFUNC);
434 * Traverse the parse tree, calling a function on every declarator in a
435 * depth-first preorder traversal. The function is given a pointer to the
436 * declarator as well as to the pointer which was used to reach that
437 * declarator: this can be used to rewrite entire subtrees.
439 * The called function may return a negative value to indicate an error
440 * which terminates traversal.
442 * Returns 0 on success, or a negative value on failure.
444 static int forall_declarators(struct cdecl *decl,
445 int f(struct cdecl_declarator **, struct cdecl_declarator *))
447 struct cdecl_declarator *d, **p;
449 for (p = &decl->declarators; *p; p = &d->child) {
457 if (d->type == CDECL_DECL_FUNCTION) {
460 for (i = d->u.function.parameters; i; i = i->next) {
461 rc = forall_declarators(i, f);
471 static struct cdecl *do_parse(const char *str, int english_mode)
473 YY_BUFFER_STATE state;
478 extern int cdecl__yydebug;
483 if (cdecl__yylex_init_extra(english_mode, &scanner) != 0)
486 state = cdecl__yy_scan_string(str, scanner);
487 if (cdecl__yyparse(scanner, &decl) != 0)
489 cdecl__yy_delete_buffer(state, scanner);
490 cdecl__yylex_destroy(scanner);
495 static int do_postprocess(struct cdecl *decl, int english_mode)
497 struct cdecl_declspec *norm_specs;
501 * For a C declaration with more than one full declarator, the
502 * specifier list is common to all of them. Normalize it once,
503 * then propagate that to all the linked cdecl structures.
505 * In english mode, the cdecl structure list always has exactly
506 * one entry so we don't need to do anything differently.
508 norm_specs = cdecl__normalize_specs(decl->specifiers);
509 for (i = decl; i; i = i->next)
510 i->specifiers = norm_specs;
512 for (i = decl; i; i = i->next) {
514 if (forall_declarators(i, reduce_parentheses) < 0)
516 if (forall_declarators(i, simplify_functions) < 0)
520 if (forall_declarators(i, check_parameters) < 0)
522 if (forall_declarators(i, check_rettypes) < 0)
524 if (forall_declarators(i, check_arrays) < 0)
526 if (forall_declarators(i, normalize_specs) < 0)
528 if (forall_declarators(i, check_qualifiers) < 0)
531 if (!valid_declspecs(i, true))
534 if (cdecl_is_abstract(i->declarators)
535 && (i != decl || i->next)) {
536 cdecl__errmsg(CDECL__EDECLTYPE);
544 static struct cdecl *parse_common(const char *str, int english_mode)
548 if (!(decl = do_parse(str, english_mode)))
551 if (!do_postprocess(decl, english_mode)) {
559 struct cdecl *cdecl_parse_decl(const char *declstr)
561 return parse_common(declstr, false);
564 struct cdecl *cdecl_parse_english(const char *english)
566 return parse_common(english, true);
569 void cdecl_free(struct cdecl *decl)