3 * Parser for C declarations.
4 * Copyright © 2011-2012, 2021, 2023-2024 Nick Bowler
6 * This program is free software: you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation, either version 3 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program. If not, see <http://www.gnu.org/licenses/>.
21 %name-prefix "cdecl__yy"
22 %parse-param {void *scanner}
23 %parse-param {struct cdecl **out}
24 %lex-param {yyscan_t scanner}
36 #include "cdecl-internal.h"
40 * Allocate a parse tree node via cdecl__alloc_item.
42 * - m1 specifies the item's union member to assign to ptr, which selects the
43 * type of node to allocate.
45 * - m2 specifies the "next" or "child" member, which is initialized to a null
46 * pointer. The cdecl__alloc_item function sets the declarator.child member
47 * to null; we explicitly copy this null pointer to the returned union member
48 * to avoid type punning. It is hoped that compilers will notice that these
49 * pointers are at the same offset therefore the assignment can be elided.
51 * The resulting pointer can be passed directly to free, as the union is the
52 * first member of the parse_item structure.
54 * Use the wrapper macros below instead of this one.
56 #define ALLOC_ITEM(ptr, m1, m2) do { \
57 struct parse_item *item; \
58 if (!(item = cdecl__alloc_item(0))) \
60 item->u.m1.m2 = (void *)item->u.declarator.child; \
61 (ptr) = &item->u.m1; \
64 /* Wrappers for ALLOC_ITEM to allocate various kinds of parser structures. */
65 #define ALLOC_ITEM_DECLARATOR(ptr) ALLOC_ITEM(ptr, declarator, child)
66 #define ALLOC_ITEM_DECLSPEC(ptr) ALLOC_ITEM(ptr, declspec, next)
67 #define ALLOC_ITEM_DECL(ptr) ALLOC_ITEM(ptr, decl, next)
69 #define ALLOC_FUNCTION(ptr, parameters_, variadic_) do { \
70 ALLOC_ITEM_DECLARATOR(ptr); \
71 (ptr)->type = CDECL_DECL_FUNCTION; \
72 (ptr)->u.function.parameters = parameters_; \
73 (ptr)->u.function.variadic = variadic_; \
76 #define ALLOC_ARRAY(ptr, length_) do { \
77 ALLOC_ITEM_DECLARATOR(ptr); \
78 (ptr)->type = CDECL_DECL_ARRAY; \
79 (ptr)->u.array.vla = NULL; \
80 (ptr)->u.array.length = length_; \
83 #define ALLOC_POINTER(ptr, qualifiers_, child_) do { \
84 ALLOC_ITEM_DECLARATOR(ptr); \
85 (ptr)->child = child_; \
86 (ptr)->type = CDECL_DECL_POINTER; \
87 (ptr)->u.pointer.qualifiers = qualifiers_; \
90 #define ALLOC_DECLSPEC(ptr, type_) do { \
91 ALLOC_ITEM_DECLSPEC(ptr); \
92 (ptr)->type = type_; \
93 (ptr)->ident = NULL; \
96 #define ALLOC_DECL(ptr, specifiers_, declarators_) do { \
97 ALLOC_ITEM_DECL(ptr); \
98 (ptr)->specifiers = specifiers_; \
99 (ptr)->declarators = declarators_; \
103 * With the postprocessing performed by fix-yytname.awk, all the symbol
104 * name strings can be used directly in error messages and there is no
105 * need for any string processing.
107 #define yytnamerr(a, b) ( (a) ? yytnamerr_copy(a, b) \
110 static size_t yytnamerr_copy(char *dst, const char *src)
112 return cdecl__strlcpy(dst, src, strlen(src)+1);
117 #include <inttypes.h>
122 void cdecl__free(struct cdecl *);
123 int cdecl__yyparse(void *scanner, struct cdecl **out);
124 const char *cdecl__token_name(unsigned token);
131 struct cdecl_declspec *declspec;
132 struct cdecl_declarator *declarator;
134 struct parse_item *item;
138 static void yyerror(YYLTYPE *, yyscan_t, struct cdecl **, const char *);
139 static void free_decl(struct cdecl *);
141 static void free_declspec(struct cdecl_declspec *x)
143 struct cdecl_declspec *p;
151 static void free_declarator(struct cdecl_declarator *x)
153 struct cdecl_declarator *p;
159 case CDECL_DECL_NULL:
161 case CDECL_DECL_IDENT:
162 case CDECL_DECL_ARRAY:
164 case CDECL_DECL_POINTER:
165 free_declspec(x->u.pointer.qualifiers);
167 case CDECL_DECL_FUNCTION:
168 free_decl(x->u.function.parameters);
179 static void free_decl(struct cdecl *x)
186 /* The specifiers may be shared by an entire chain. */
187 if (!p || p->specifiers != x->specifiers)
188 free_declspec(x->specifiers);
190 free_declarator(x->declarators);
196 void cdecl__free(struct cdecl *decl)
202 * Join two declaration specifier lists into a single list, with "a" being the
203 * head of the new list.
205 * The list "a" is assumed to be nonempty.
207 static void join_specs(struct cdecl_declspec *a, struct cdecl_declspec *b)
215 * Join three specifier lists into a single list, and returns the head of
218 * The list "b" is assumed to be a singleton list.
220 static struct cdecl_declspec *join_specs3(struct cdecl_declspec *a,
221 struct cdecl_declspec *b,
222 struct cdecl_declspec *c)
230 * Reverse the order of a "struct cdecl" list, and return the new first
231 * element of the list (i.e., the last element of the original list).
233 static struct cdecl *reverse_decls(struct cdecl *decl)
235 struct cdecl *prev, *next;
237 for (prev = NULL; decl; decl = next) {
247 * Alter an abstract declarator (type name) to declare an identifier instead,
248 * used by the English parser rules to reduce "identifier as type" sequences.
250 static struct cdecl *
251 insert_identifier(struct cdecl *decl, struct parse_item *ident)
253 struct cdecl_declarator *d, **p = &decl->declarators;
255 while ((d = *p)->child)
258 *p = &ident->u.declarator;
262 static struct cdecl_declarator *nulldecl(void)
264 static const struct cdecl_declarator nulldecl = {0};
265 return (void *)&nulldecl;
267 #define NULLDECL (nulldecl())
271 %destructor { free($$); } <item>
272 %destructor { free_declspec($$); } <declspec>
273 %destructor { free_declarator($$); } <declarator>
274 %destructor { free_decl($$); } <decl>
279 %token <item> T_IDENT "identifier"
280 %token <uintval> T_UINT "integer constant"
282 %token T_SEMICOLON ";"
283 %token T_ASTERISK "*"
286 %token T_LBRACKET "["
287 %token T_RBRACKET "]"
289 %token T_ELLIPSIS "..."
291 %token <spectype> T_TYPEDEF "typedef"
292 %token <spectype> T_EXTERN "extern"
293 %token <spectype> T_STATIC "static"
294 %token <spectype> T_AUTO "auto"
295 %token <spectype> T_REGISTER "register"
297 %token <spectype> T_INLINE "inline"
299 %token <spectype> T_RESTRICT "restrict"
300 %token <spectype> T_VOLATILE "volatile"
301 %token <spectype> T_CONST "const"
303 %token <spectype> T_VOID "void"
304 %token <spectype> T_CHAR "char"
305 %token <spectype> T_SHORT "short"
306 %token <spectype> T_INT "int"
307 %token <spectype> T_LONG "long"
308 %token <spectype> T_FLOAT "float"
309 %token <spectype> T_DOUBLE "double"
310 %token <spectype> T_SIGNED "signed"
311 %token <spectype> T_UNSIGNED "unsigned"
312 %token <spectype> T_BOOL "_Bool"
313 %token <spectype> T_COMPLEX "_Complex"
314 %token <spectype> T_IMAGINARY "_Imaginary"
316 %token <spectype> T_STRUCT "struct"
317 %token <spectype> T_UNION "union"
318 %token <spectype> T_ENUM "enum"
324 %token T_DECLARE "declare"
325 %token T_POINTER "pointer"
326 %token T_FUNCTION "function"
327 %token T_RETURNING "returning"
328 %token T_ARRAY "array"
332 %token T_VLA "variable-length"
334 %type <item> vla_ident
335 %type <uintval> array_length
336 %type <boolval> varargs
337 %type <spectype> declspec_simple qualifier_simple
338 %type <spectype> typespec_simple typespec_tagged
339 %type <declspec> declspec_notype declspec_noid typespec_noid typespec
340 %type <declspec> qualifier qualifiers
341 %type <declspec> declspecs declspecs_notype declspecs_noid
342 %type <declarator> direct_declarator declarator pointer array parens postfix
343 %type <declarator> direct_declarator_ish declarator_ish parameter_type_list
344 %type <decl> cdecl declaration declarators declarator_wrap parameter
346 %type <item> english_vla
347 %type <declspec> storage_func_specs post_specs
348 %type <declspec> type_qual_spec type_qual_specs typedef_name_qual
349 %type <declarator> english_declarator english_array english_function
350 %type <declarator> english_parameter_list null_decl
351 %type <decl> english english_declaration english_parameter
353 /* Precedence declaration to avoid conflict in english_parameter; see below. */
359 input: cdecl { *out = $1; }
360 cdecl: english | declaration
363 declaration: declspecs declarators semi {
364 $$ = reverse_decls($2);
369 * We support parsing declarations using arbitrary identifiers as type
370 * specifiers (a la C typedef). To avoid confusion with identifiers that
371 * may also be used as declarators, note the following:
373 * (a) Every valid C declaration must have at least one type specifier, and
374 * (b) Valid declarations with typedef names have exactly one type specifier.
376 * So the rule applied when parsing specifiers is: an identifier is a type
377 * specifier only if we have not yet seen any type specifiers whatsoever
378 * (within one declaration specifier list).
380 * Treating identifiers as type specifiers by default can lead to strange and
381 * unexpected parses; libcdecl applies a simplification step to the resulting
382 * parse tree afterwards.
384 declspecs: declspecs_notype typespec declspecs_noid {
385 $$ = join_specs3($1, $2, $3);
388 declspecs_notype: { $$ = NULL; } | declspecs_notype declspec_notype {
393 declspecs_noid: { $$ = NULL; } | declspecs_noid declspec_noid {
398 qualifiers: { $$ = NULL; } | qualifiers qualifier {
403 declarators: declarator_wrap | declarators T_COMMA declarator_wrap {
408 declarator_wrap: declarator {
409 ALLOC_DECL($$, NULL, $1);
412 declspec_simple: T_AUTO
419 typespec_simple: T_VOID
432 typespec_tagged: T_STRUCT | T_UNION | T_ENUM | { $$ = CDECL_TYPE_IDENT; }
434 qualifier_simple: T_CONST
438 declspec_notype: qualifier | declspec_simple { ALLOC_DECLSPEC($$, $1); }
439 typespec_noid: typespec_simple { ALLOC_DECLSPEC($$, $1); }
440 qualifier: qualifier_simple { ALLOC_DECLSPEC($$, $1); }
442 typespec: typespec_noid | typespec_tagged T_IDENT {
443 /* Compiler should be able to elide this assignment. */
444 $2->u.declspec.ident = $2->u.declarator.u.ident;
446 $$ = &$2->u.declspec;
450 declspec_noid: declspec_notype | typespec_noid
452 vla_ident: T_IDENT | T_ASTERISK {
453 if (!($$ = cdecl__alloc_item(1)))
458 array: T_LBRACKET array_length T_RBRACKET {
460 } | T_LBRACKET vla_ident T_RBRACKET {
461 $$ = &$2->u.declarator;
462 $$->type = CDECL_DECL_ARRAY;
463 $$->u.array.vla = $$->u.ident;
464 $$->u.array.length = 0;
467 parameter: declspecs declarator {
468 ALLOC_DECL($$, $1, $2);
471 varargs: { $$ = false; } | T_COMMA T_ELLIPSIS { $$ = true; }
473 parameter_type_list: parameter varargs {
474 ALLOC_FUNCTION($$, $1, $2);
475 } | parameter T_COMMA parameter_type_list {
477 $1->next = $$->u.function.parameters;
478 $$->u.function.parameters = $1;
481 parens: T_LPAREN parameter_type_list T_RPAREN {
483 } | T_LPAREN declarator_ish T_RPAREN {
484 struct cdecl *fake_params;
486 ALLOC_DECL(fake_params, NULL, $2);
487 ALLOC_FUNCTION($$, fake_params, false);
490 pointer: T_ASTERISK qualifiers direct_declarator {
491 ALLOC_POINTER($$, $2, $3);
492 } | T_ASTERISK qualifiers pointer {
493 ALLOC_POINTER($$, $2, $3);
496 declarator: direct_declarator | pointer
497 declarator_ish: direct_declarator_ish | pointer
498 postfix: array | parens
500 direct_declarator_ish: {
502 } | direct_declarator_ish postfix {
510 $$ = &$1->u.declarator;
511 } | direct_declarator postfix {
516 english: T_DECLARE T_IDENT T_AS english_declaration {
517 $$ = insert_identifier($4, $2);
518 } | T_TYPE english_declaration {
523 * We use a precedence declaration to prefer shifting an identifier
524 * over reducing this empty rule; see below.
526 storage_func_specs: %prec T_TYPE { $$ = NULL; }
527 storage_func_specs: storage_func_specs declspec_simple {
528 ALLOC_DECLSPEC($$, $2);
532 type_qual_spec: typespec_noid | qualifier
534 type_qual_specs: { $$ = NULL; } | type_qual_specs type_qual_spec {
540 * The "qualifiers" nonterminal needs to be used here to avoid shift/reduce
541 * conflicts with pointer declarators. So we end up needing to stitch
542 * together three different specifiers lists.
544 post_specs: qualifiers typespec type_qual_specs {
545 $$ = join_specs3($1, $2, $3);
548 english_declaration: storage_func_specs english_declarator post_specs {
550 ALLOC_DECL($$, $3, $2);
553 english_declarator: {
555 } | english_declarator qualifiers T_POINTER T_TO {
556 ALLOC_POINTER($$, $2, $1);
557 } | english_declarator english_array {
560 } | english_declarator english_function {
565 english_function: T_FUNCTION T_RETURNING {
566 ALLOC_FUNCTION($$, NULL, false);
567 } | T_FUNCTION T_LPAREN english_parameter_list T_RPAREN T_RETURNING {
571 english_parameter_list: english_parameter varargs {
572 ALLOC_FUNCTION($$, $1, $2);
573 } | english_parameter T_COMMA english_parameter_list {
575 $1->next = $$->u.function.parameters;
576 $$->u.function.parameters = $1;
579 typedef_name_qual: T_IDENT qualifiers {
580 /* Compiler should be able to elide this assignment. */
581 $1->u.declspec.ident = $1->u.declarator.u.ident;
583 $$ = &$1->u.declspec;
584 $$->type = CDECL_TYPE_IDENT;
593 * There is a shift/reduce conflict here when an identifier appears as the
594 * first token. The conflict is between shifting T_IDENT, or reducing the
595 * empty production for storage_func_specs (cf. english_declaration).
597 * - In either case, if we reduce, we won't match T_IDENT T_AS since the
598 * stack now has the extra storage_func_specs nonterminal symbol.
599 * - And if we shift, we won't match english_declaration since it is
600 * too late to add storage_func_specs to the stack.
602 * The only valid input affected by the conflict is a simple type names,
603 * possibly followed by qualifiers. So the conflict is adequately resolved
604 * by shifting, so long as we have a special-case reduction to handle this.
606 english_parameter: english_declaration | typedef_name_qual null_decl {
607 ALLOC_DECL($$, $1, $2);
608 } | T_IDENT T_AS english_declaration {
609 $$ = insert_identifier($3, $1);
612 english_array: T_VLA T_ARRAY english_vla T_OF {
613 $$ = &$3->u.declarator;
614 $$->type = CDECL_DECL_ARRAY;
615 $$->u.array.vla = $$->u.ident;
616 $$->u.array.length = 0;
617 } | T_ARRAY array_length T_OF {
621 array_length: { $$ = 0; }
622 array_length: T_UINT {
624 cdecl__errmsg(CDECL__EZEROARRAY);
629 english_vla: T_IDENT | {
630 if (!($$ = cdecl__alloc_item(1)))
638 * Expose the token string table to the rest of the library, in order to
639 * produce strings that match parser keywords.
641 * In order for this to work properly, the Bison output must be postprocessed
642 * by fix-yytname.awk to remove pointless quotation marks from the keyword
645 const char *cdecl__token_name(unsigned token)
647 return yytname[YYTRANSLATE(token)];
651 * Current versions of GCC (up to 13) want to inline this function into the
652 * parser even when optimizing for size and the results are not great, so
653 * try to prevent such inlining.
655 CDECL__NOINLINE static void
656 yyerror(YYLTYPE *loc, yyscan_t scanner, struct cdecl **out, const char *err)
658 if (strstr(err, yytname[YYTRANSLATE(T_LEX_ERROR)]))
661 cdecl__err("%s", err);