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 * Allocate a "parse item", which is a union of several parse tree
32 * structure types, together with a string buffer. The s_sz argument
33 * specifies the size of the string (including its terminator), which
36 * The union's declarator member is pre-initialized to a valid "identifier"
37 * declarator, which shares several interesting offsets with the "declspec"
38 * structure for an "identifier" type specifier.
40 struct parse_item *cdecl__alloc_item(size_t s_sz)
42 struct parse_item *ret;
44 ret = malloc(offsetof(struct parse_item, s) + s_sz);
46 cdecl__errmsg(CDECL__ENOMEM);
50 ret->u.declarator.child = NULL;
51 ret->u.declarator.type = CDECL_DECL_IDENT;
52 ret->u.declarator.u.ident = ret->s;
58 * We can represent type specifiers as a bitmap, which gives us a finite
59 * list of acceptable bitmap values according to the C standard. However,
60 * the "long" specifier is allowed to occur more than once, but only at most
61 * 2 times. Treat it as a special case, assigning an unused bit to represent
64 #define MAP_LLONG_BIT 31
65 #define MAP_LONG_BIT (CDECL_TYPE_LONG-CDECL_SPEC_TYPE)
66 #define CDECL_TYPE_LLONG (CDECL_SPEC_TYPE+MAP_LLONG_BIT)
71 * Convert the declaration specifiers to a bitmap with each bit
72 * corresponding to one specific type specifier.
74 static int valid_typespec(struct cdecl_declspec *s)
76 struct cdecl_declspec *c;
77 unsigned long map = 0;
79 for (c = s; c; c = c->next) {
82 if (cdecl_spec_kind(c) != CDECL_SPEC_TYPE)
85 bit = c->type - CDECL_SPEC_TYPE;
86 assert(bit < MAP_LLONG_BIT);
89 /* "long" special case */
90 if ((map & bit) == 1ul << MAP_LONG_BIT)
91 bit = 1ul << MAP_LLONG_BIT;
94 if (bit == 1ul << MAP_LLONG_BIT)
95 cdecl__errmsg(CDECL__ETOOLONG);
97 cdecl__errmsg(CDECL__EDUPTYPE);
103 if (typemap_is_valid(map))
107 cdecl__errmsg(CDECL__ENOTYPE);
109 cdecl__errmsg(CDECL__EBADTYPE);
115 * Verify the declaration specifiers of a declaration. If top is true, treat
116 * this as a top-level declaration. Otherwise, treat this as a function
117 * parameter (which carries additional constraints).
119 static bool valid_declspecs(struct cdecl *decl, bool top)
121 struct cdecl_declspec *c, *specs = decl->specifiers;
122 struct cdecl_declarator *d = decl->declarators;
123 bool abstract = cdecl_is_abstract(d);
124 unsigned num_storage = 0;
126 if (!valid_typespec(specs))
129 for (c = specs; c; c = c->next) {
130 switch (cdecl_spec_kind(c)) {
131 case CDECL_SPEC_TYPE:
132 if (c->type == CDECL_TYPE_VOID &&
133 (d->type == CDECL_DECL_IDENT
134 || d->type == CDECL_DECL_ARRAY)) {
135 cdecl__errmsg(CDECL__EBADVOID);
139 case CDECL_SPEC_STOR:
140 if (top && abstract) {
141 cdecl__errmsg(CDECL__ETYPESTOR);
145 if (!top && c->type != CDECL_STOR_REGISTER) {
146 cdecl__errmsg(CDECL__EFUNCSTOR);
150 if (++num_storage > 1) {
151 cdecl__errmsg(CDECL__EMANYSTOR);
155 case CDECL_SPEC_QUAL:
157 * Restrict qualifiers are only valid in the
158 * pointer qualifier list, which isn't checked here.
160 if (c->type == CDECL_QUAL_RESTRICT) {
161 cdecl__errmsg(CDECL__EBADQUAL);
165 case CDECL_SPEC_FUNC:
166 if (abstract || !top || d->type != CDECL_DECL_FUNCTION) {
167 cdecl__errmsg(CDECL__ENOTFUNC);
181 * The C grammar leaves ambiguous some cases where parentheses represent a
182 * function declarator or just parentheses. The language uses additional
183 * context (whether or not a typedef is in scope, etc.) to resolve these
184 * ambiguities, but we don't have access to that kind of information.
186 * The cdecl99 parser uses an unambiguous grammar which treats almost
187 * everything as a function, and thus considers things like 'int (x)' to
188 * be a function type with a single parameter of type 'x' (a typedef name),
189 * returning int. This can result in very complicated types for simple
190 * declarations. Ideally, cdecl99 should try and find the "simplest"
191 * explanation for a given declaration.
193 * Whether or not it achieves the simplest explanation, we apply a simple rule:
194 * if a declarator could be interpreted as something other than a function,
197 * Since cdecl99 supports things like [*] in any context (in C, such constructs
198 * are only valid in function parameter lists), we don't treat them specially
202 static struct cdecl_declarator *reduce_function(struct cdecl *param)
204 struct cdecl_declarator *d, **p = ¶m->declarators;
205 struct parse_item *spec = (void *)param->specifiers;
207 while ((d = *p)->child)
210 if (d->type != CDECL_DECL_NULL)
214 * The child and u.ident members of cdecl_declarator are expected
215 * to be located at identical offsets as, respectively, the next
216 * and ident members within cdecl_declspec, so the expectation is
217 * that the compiler can elide both assignments.
219 spec->u.declarator.child = (void *)spec->u.declspec.next;
220 spec->u.declarator.u.ident = spec->u.declspec.ident;
221 spec->u.declarator.type = CDECL_DECL_IDENT;
222 *p = &spec->u.declarator;
224 d = param->declarators;
229 static bool function_is_reducible(struct cdecl_declarator *d)
231 if (d->type != CDECL_DECL_FUNCTION)
233 if (d->child->type != CDECL_DECL_NULL)
234 return false; /* e.g., int (*)(x) */
236 if (!d->u.function.parameters)
237 return false; /* e.g., int f() */
238 if (d->u.function.parameters->next)
239 return false; /* e.g., int (x, y) */
240 if (d->u.function.variadic)
241 return false; /* e.g., int (x, ...) */
243 if (d->u.function.parameters->specifiers->type != CDECL_TYPE_IDENT)
244 return false; /* e.g. int (int) */
245 if (d->u.function.parameters->specifiers->next)
246 return false; /* e.g. int (size_t const) */
247 if (d->u.function.parameters->declarators->type == CDECL_DECL_POINTER)
248 return false; /* e.g. int (x *) */
254 simplify_functions(struct cdecl_declarator **p, struct cdecl_declarator *d)
256 struct cdecl_declarator *new;
258 if (!function_is_reducible(d))
261 new = reduce_function(d->u.function.parameters);
263 return 0; /* e.g. int (foo bar) */
271 * The main parser's bias towards considering things as functions whenever
272 * possible makes nested parentheses tricky. "(x)" is considered to be part
273 * of a function declarator until simplify_functions converts it. The problem
274 * is that "(((x)))" is not valid as part of a function declarator, but it _is_
275 * valid as either an identifier enclosed thrice in parentheses, or an abstract
276 * function declarator enclosed twice in parentheses.
278 * To avoid ambiguities, the main parser actually returns a function declarator
279 * for every pair of parentheses. The ones we need to look at consist of a
280 * single parameter with an empty specifier list (noting that every real
281 * function parameter will have at least one type specifier).
283 * There are two cases:
285 * - For (), the parser emits a parameter with a lone null declarator.
286 * This fake parameter simply gets deleted, leaving us with a normal
287 * function declarator with an empty identifier list.
289 * - Otherwise, the parameter's outermost declarator is not null. The
290 * function itself is deleted, replaced in the parse tree with the
291 * fake parameter's declarator.
293 * Repeating until there no fake parameters, this reduction transforms, for
294 * example, "(((x)))" into "(x)", an abstract function declarator. The result
295 * is then subject to the function simplification step, which will turn "(x)"
296 * into x (declaring an identifier).
298 * The whole process is repeated until no more changes are made to the parse
299 * tree, or a syntax error is detected.
301 static struct cdecl *fake_function_param(struct cdecl_declarator *d)
305 if (d->type != CDECL_DECL_FUNCTION)
308 param = d->u.function.parameters;
309 if (!param || param->specifiers)
312 assert(!param->next);
317 reduce_parentheses(struct cdecl_declarator **p, struct cdecl_declarator *d)
323 while ((param = fake_function_param(d))) {
324 struct cdecl_declarator *decl = param->declarators;
325 d->u.function.parameters = NULL;
327 if (decl->type != CDECL_DECL_NULL) {
328 if (d->child->type != CDECL_DECL_NULL) {
329 /* Fake parameter on real function. */
330 d->u.function.parameters = param;
331 cdecl__errmsg(CDECL__EBADPARAM);
335 param->declarators = d;
341 } while (simplify_functions(p, d));
347 * Returns nonzero iff the given specifier list contains a specifier
348 * of the indicated type.
350 static int have_specifier(struct cdecl_declspec *s, unsigned type)
352 for (; s; s = s->next)
359 * Check syntax restrictions on a function declarator's child declarator.
360 * That is, "pointer to function", "array of function" and "function
361 * returning function".
363 * Returns -1 if the declaration is invalid, or 0 otherwise.
365 static int check_function_child(struct cdecl_declarator *d)
367 struct cdecl_pointer *ptr;
370 case CDECL_DECL_POINTER:
372 if (have_specifier(ptr->qualifiers, CDECL_QUAL_RESTRICT)) {
373 /* pointer to function cannot be restrict qualified. */
374 cdecl__errmsg(CDECL__ERESTRICTFUNC);
378 case CDECL_DECL_FUNCTION:
379 /* function returning function is never allowed. */
380 cdecl__errmsg(CDECL__ERETFUNC);
382 case CDECL_DECL_ARRAY:
383 /* array of function is never allowed. */
384 cdecl__errmsg(CDECL__EFUNCARRAY);
392 * Check a function parameter declaration for validity, which means it has a
393 * valid combination of declaration specifiers and, if it is a void parameter,
394 * that it is the one special case where this is allowed.
396 * Returns -1 if the declaration is invalid, or 0 otherwise.
398 static int check_function_param(struct cdecl_function *f, struct cdecl *param)
400 if (!valid_declspecs(param, false))
403 /* Check for "void" function parameters as a special case. */
404 if (param->declarators->type == CDECL_DECL_NULL
405 && have_specifier(param->specifiers, CDECL_TYPE_VOID))
407 struct cdecl *fp = f->parameters;
409 if (f->variadic || fp->next || fp->specifiers->next) {
410 cdecl__errmsg(CDECL__EVOIDPARAM);
419 * Normalize the specifier lists for function parameters, and then check the
420 * function declarator for validity.
422 * Returns -1 if the declaration is invalid, or 0 otherwise.
424 static int postproc_function(struct cdecl_declarator *d)
426 struct cdecl_function *func = &d->u.function;
430 for (param = func->parameters; param; param = param->next) {
431 param->specifiers = cdecl__normalize_specs(param->specifiers);
433 if ((rc = check_function_param(func, param)) < 0)
437 return check_function_child(d->child);
441 postproc_common(struct cdecl_declarator **p, struct cdecl_declarator *d)
443 struct cdecl_pointer *ptr;
446 case CDECL_DECL_POINTER:
448 ptr->qualifiers = cdecl__normalize_specs(ptr->qualifiers);
450 case CDECL_DECL_FUNCTION:
451 return postproc_function(d);
452 case CDECL_DECL_ARRAY:
453 if (d->child && d->child->type == CDECL_DECL_FUNCTION) {
454 /* function returning array is never allowed. */
455 cdecl__errmsg(CDECL__ERETARRAY);
465 * Traverse the parse tree, calling a function on every declarator in a
466 * depth-first preorder traversal. The function is given a pointer to the
467 * declarator as well as to the pointer which was used to reach that
468 * declarator: this can be used to rewrite entire subtrees.
470 * The called function may return a negative value to indicate an error
471 * which terminates traversal.
473 * Returns 0 on success, or a negative value on failure.
475 static int forall_declarators(struct cdecl *decl,
476 int f(struct cdecl_declarator **, struct cdecl_declarator *))
478 struct cdecl_declarator *d, **p;
480 for (p = &decl->declarators; *p; p = &d->child) {
488 if (d->type == CDECL_DECL_FUNCTION) {
491 for (i = d->u.function.parameters; i; i = i->next) {
492 rc = forall_declarators(i, f);
502 static struct cdecl *do_parse(const char *str, int english_mode)
504 struct cdecl *decl = NULL;
505 YY_BUFFER_STATE state;
509 extern int cdecl__yydebug;
514 if (cdecl__yylex_init_extra(english_mode, &scanner) != 0)
517 state = cdecl__yy_scan_string(str, scanner);
518 if (cdecl__yyparse(scanner, &decl) != 0) {
520 * If the input consists of a complete, valid declaration
521 * followed by some garbage, that parsed declaration will
522 * be output by the parser and we need to free it here.
527 cdecl__yy_delete_buffer(state, scanner);
528 cdecl__yylex_destroy(scanner);
533 static int do_postprocess(struct cdecl *decl, int english_mode)
535 struct cdecl_declspec *norm_specs;
539 * For a C declaration with more than one full declarator, the
540 * specifier list is common to all of them. Normalize it once,
541 * then propagate that to all the linked cdecl structures.
543 * In english mode, the cdecl structure list always has exactly
544 * one entry so we don't need to do anything differently.
546 norm_specs = cdecl__normalize_specs(decl->specifiers);
547 for (i = decl; i; i = i->next)
548 i->specifiers = norm_specs;
550 for (i = decl; i; i = i->next) {
552 if (forall_declarators(i, reduce_parentheses) < 0)
556 if (forall_declarators(i, postproc_common) < 0)
559 if (!valid_declspecs(i, true))
562 if (decl->next && cdecl_is_abstract(i->declarators)) {
563 /* Abstract full declarators: there can only be one. */
564 cdecl__errmsg(CDECL__EDECLTYPE);
572 static struct cdecl *parse_common(const char *str, int english_mode)
576 if (!(decl = do_parse(str, english_mode)))
579 if (!do_postprocess(decl, english_mode)) {
587 struct cdecl *cdecl_parse_decl(const char *declstr)
589 return parse_common(declstr, false);
592 struct cdecl *cdecl_parse_english(const char *english)
594 return parse_common(english, true);
597 void cdecl_free(struct cdecl *decl)