]>
Commit | Line | Data |
---|---|---|
1 | /* Process source files and output type information. | |
2 | Copyright (C) 2006-2020 Free Software Foundation, Inc. | |
3 | ||
4 | This file is part of GCC. | |
5 | ||
6 | GCC is free software; you can redistribute it and/or modify it under | |
7 | the terms of the GNU General Public License as published by the Free | |
8 | Software Foundation; either version 3, or (at your option) any later | |
9 | version. | |
10 | ||
11 | GCC is distributed in the hope that it will be useful, but WITHOUT ANY | |
12 | WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
13 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
14 | for more details. | |
15 | ||
16 | You should have received a copy of the GNU General Public License | |
17 | along with GCC; see the file COPYING3. If not see | |
18 | <http://www.gnu.org/licenses/>. */ | |
19 | ||
20 | #ifdef HOST_GENERATOR_FILE | |
21 | #include "config.h" | |
22 | #define GENERATOR_FILE 1 | |
23 | #else | |
24 | #include "bconfig.h" | |
25 | #endif | |
26 | #include "system.h" | |
27 | #include "gengtype.h" | |
28 | ||
29 | /* This is a simple recursive-descent parser which understands a subset of | |
30 | the C type grammar. | |
31 | ||
32 | Rule functions are suffixed _seq if they scan a sequence of items; | |
33 | _opt if they may consume zero tokens; _seqopt if both are true. The | |
34 | "consume_" prefix indicates that a sequence of tokens is parsed for | |
35 | syntactic correctness and then thrown away. */ | |
36 | ||
37 | /* Simple one-token lookahead mechanism. */ | |
38 | ||
39 | struct token | |
40 | { | |
41 | const char *value; | |
42 | int code; | |
43 | bool valid; | |
44 | }; | |
45 | static struct token T; | |
46 | ||
47 | /* Retrieve the code of the current token; if there is no current token, | |
48 | get the next one from the lexer. */ | |
49 | static inline int | |
50 | token (void) | |
51 | { | |
52 | if (!T.valid) | |
53 | { | |
54 | T.code = yylex (&T.value); | |
55 | T.valid = true; | |
56 | } | |
57 | return T.code; | |
58 | } | |
59 | ||
60 | /* Retrieve the value of the current token (if any) and mark it consumed. | |
61 | The next call to token() will get another token from the lexer. */ | |
62 | static inline const char * | |
63 | advance (void) | |
64 | { | |
65 | T.valid = false; | |
66 | return T.value; | |
67 | } | |
68 | ||
69 | /* Diagnostics. */ | |
70 | ||
71 | /* This array is indexed by the token code minus CHAR_TOKEN_OFFSET. */ | |
72 | static const char *const token_names[] = { | |
73 | "GTY", | |
74 | "typedef", | |
75 | "extern", | |
76 | "static", | |
77 | "union", | |
78 | "struct", | |
79 | "enum", | |
80 | "...", | |
81 | "ptr_alias", | |
82 | "nested_ptr", | |
83 | "a param<N>_is option", | |
84 | "a number", | |
85 | "a scalar type", | |
86 | "an identifier", | |
87 | "a string constant", | |
88 | "a character constant", | |
89 | "an array declarator", | |
90 | "a C++ keyword to ignore" | |
91 | }; | |
92 | ||
93 | /* This array is indexed by token code minus FIRST_TOKEN_WITH_VALUE. */ | |
94 | static const char *const token_value_format[] = { | |
95 | "%s", | |
96 | "'%s'", | |
97 | "'%s'", | |
98 | "'%s'", | |
99 | "'\"%s\"'", | |
100 | "\"'%s'\"", | |
101 | "'[%s]'", | |
102 | "'%s'", | |
103 | }; | |
104 | ||
105 | /* Produce a printable representation for a token defined by CODE and | |
106 | VALUE. This sometimes returns pointers into malloc memory and | |
107 | sometimes not, therefore it is unsafe to free the pointer it | |
108 | returns, so that memory is leaked. This does not matter, as this | |
109 | function is only used for diagnostics, and in a successful run of | |
110 | the program there will be none. */ | |
111 | static const char * | |
112 | print_token (int code, const char *value) | |
113 | { | |
114 | if (code < CHAR_TOKEN_OFFSET) | |
115 | return xasprintf ("'%c'", code); | |
116 | else if (code < FIRST_TOKEN_WITH_VALUE) | |
117 | return xasprintf ("'%s'", token_names[code - CHAR_TOKEN_OFFSET]); | |
118 | else if (!value) | |
119 | return token_names[code - CHAR_TOKEN_OFFSET]; /* don't quote these */ | |
120 | else | |
121 | return xasprintf (token_value_format[code - FIRST_TOKEN_WITH_VALUE], | |
122 | value); | |
123 | } | |
124 | ||
125 | /* Convenience wrapper around print_token which produces the printable | |
126 | representation of the current token. */ | |
127 | static inline const char * | |
128 | print_cur_token (void) | |
129 | { | |
130 | return print_token (T.code, T.value); | |
131 | } | |
132 | ||
133 | /* Report a parse error on the current line, with diagnostic MSG. | |
134 | Behaves as standard printf with respect to additional arguments and | |
135 | format escapes. */ | |
136 | static void ATTRIBUTE_PRINTF_1 | |
137 | parse_error (const char *msg, ...) | |
138 | { | |
139 | va_list ap; | |
140 | ||
141 | fprintf (stderr, "%s:%d: parse error: ", | |
142 | get_input_file_name (lexer_line.file), lexer_line.line); | |
143 | ||
144 | va_start (ap, msg); | |
145 | vfprintf (stderr, msg, ap); | |
146 | va_end (ap); | |
147 | ||
148 | fputc ('\n', stderr); | |
149 | ||
150 | hit_error = true; | |
151 | } | |
152 | ||
153 | /* If the next token does not have code T, report a parse error; otherwise | |
154 | return the token's value. */ | |
155 | static const char * | |
156 | require (int t) | |
157 | { | |
158 | int u = token (); | |
159 | const char *v = advance (); | |
160 | if (u != t) | |
161 | { | |
162 | parse_error ("expected %s, have %s", | |
163 | print_token (t, 0), print_token (u, v)); | |
164 | return 0; | |
165 | } | |
166 | return v; | |
167 | } | |
168 | ||
169 | /* As per require, but do not advance. */ | |
170 | static const char * | |
171 | require_without_advance (int t) | |
172 | { | |
173 | int u = token (); | |
174 | const char *v = T.value; | |
175 | if (u != t) | |
176 | { | |
177 | parse_error ("expected %s, have %s", | |
178 | print_token (t, 0), print_token (u, v)); | |
179 | return 0; | |
180 | } | |
181 | return v; | |
182 | } | |
183 | ||
184 | /* If the next token does not have one of the codes T1 or T2, report a | |
185 | parse error; otherwise return the token's value. */ | |
186 | static const char * | |
187 | require2 (int t1, int t2) | |
188 | { | |
189 | int u = token (); | |
190 | const char *v = advance (); | |
191 | if (u != t1 && u != t2) | |
192 | { | |
193 | parse_error ("expected %s or %s, have %s", | |
194 | print_token (t1, 0), print_token (t2, 0), | |
195 | print_token (u, v)); | |
196 | return 0; | |
197 | } | |
198 | return v; | |
199 | } | |
200 | ||
201 | /* If the next token does not have one of the codes T1, T2, T3 or T4, report a | |
202 | parse error; otherwise return the token's value. */ | |
203 | static const char * | |
204 | require4 (int t1, int t2, int t3, int t4) | |
205 | { | |
206 | int u = token (); | |
207 | const char *v = advance (); | |
208 | if (u != t1 && u != t2 && u != t3 && u != t4) | |
209 | { | |
210 | parse_error ("expected %s, %s, %s or %s, have %s", | |
211 | print_token (t1, 0), print_token (t2, 0), | |
212 | print_token (t3, 0), print_token (t4, 0), | |
213 | print_token (u, v)); | |
214 | return 0; | |
215 | } | |
216 | return v; | |
217 | } | |
218 | ||
219 | /* Near-terminals. */ | |
220 | ||
221 | /* C-style string constant concatenation: STRING+ | |
222 | Bare STRING should appear nowhere else in this file. */ | |
223 | static const char * | |
224 | string_seq (void) | |
225 | { | |
226 | const char *s1, *s2; | |
227 | size_t l1, l2; | |
228 | char *buf; | |
229 | ||
230 | s1 = require (STRING); | |
231 | if (s1 == 0) | |
232 | return ""; | |
233 | while (token () == STRING) | |
234 | { | |
235 | s2 = advance (); | |
236 | ||
237 | l1 = strlen (s1); | |
238 | l2 = strlen (s2); | |
239 | buf = XRESIZEVEC (char, CONST_CAST (char *, s1), l1 + l2 + 1); | |
240 | memcpy (buf + l1, s2, l2 + 1); | |
241 | XDELETE (CONST_CAST (char *, s2)); | |
242 | s1 = buf; | |
243 | } | |
244 | return s1; | |
245 | } | |
246 | ||
247 | ||
248 | /* The caller has detected a template declaration that starts | |
249 | with TMPL_NAME. Parse up to the closing '>'. This recognizes | |
250 | simple template declarations of the form ID<ID1,ID2,...,IDn>, | |
251 | potentially with a single level of indirection e.g. | |
252 | ID<ID1 *, ID2, ID3 *, ..., IDn>. | |
253 | It does not try to parse anything more sophisticated than that. | |
254 | ||
255 | Returns the template declaration string "ID<ID1,ID2,...,IDn>". */ | |
256 | ||
257 | static const char * | |
258 | require_template_declaration (const char *tmpl_name) | |
259 | { | |
260 | char *str; | |
261 | int num_indirections = 0; | |
262 | ||
263 | /* Recognize the opening '<'. */ | |
264 | require ('<'); | |
265 | str = concat (tmpl_name, "<", (char *) 0); | |
266 | ||
267 | /* Read the comma-separated list of identifiers. */ | |
268 | int depth = 1; | |
269 | while (depth > 0) | |
270 | { | |
271 | if (token () == ENUM) | |
272 | { | |
273 | advance (); | |
274 | str = concat (str, "enum ", (char *) 0); | |
275 | continue; | |
276 | } | |
277 | if (token () == NUM | |
278 | || token () == ':' | |
279 | || token () == '+') | |
280 | { | |
281 | str = concat (str, advance (), (char *) 0); | |
282 | continue; | |
283 | } | |
284 | if (token () == '<') | |
285 | { | |
286 | advance (); | |
287 | str = concat (str, "<", (char *) 0); | |
288 | depth += 1; | |
289 | continue; | |
290 | } | |
291 | if (token () == '>') | |
292 | { | |
293 | advance (); | |
294 | str = concat (str, ">", (char *) 0); | |
295 | depth -= 1; | |
296 | continue; | |
297 | } | |
298 | const char *id = require4 (SCALAR, ID, '*', ','); | |
299 | if (id == NULL) | |
300 | { | |
301 | if (T.code == '*') | |
302 | { | |
303 | id = "*"; | |
304 | if (num_indirections++) | |
305 | parse_error ("only one level of indirection is supported" | |
306 | " in template arguments"); | |
307 | } | |
308 | else | |
309 | id = ","; | |
310 | } | |
311 | else | |
312 | num_indirections = 0; | |
313 | str = concat (str, id, (char *) 0); | |
314 | } | |
315 | return str; | |
316 | } | |
317 | ||
318 | ||
319 | /* typedef_name: either an ID, or a template type | |
320 | specification of the form ID<t1,t2,...,tn>. */ | |
321 | ||
322 | static const char * | |
323 | typedef_name (void) | |
324 | { | |
325 | const char *id = require (ID); | |
326 | if (token () == '<') | |
327 | return require_template_declaration (id); | |
328 | else | |
329 | return id; | |
330 | } | |
331 | ||
332 | /* Absorb a sequence of tokens delimited by balanced ()[]{}. */ | |
333 | static void | |
334 | consume_balanced (int opener, int closer) | |
335 | { | |
336 | require (opener); | |
337 | for (;;) | |
338 | switch (token ()) | |
339 | { | |
340 | default: | |
341 | advance (); | |
342 | break; | |
343 | case '(': | |
344 | consume_balanced ('(', ')'); | |
345 | break; | |
346 | case '[': | |
347 | consume_balanced ('[', ']'); | |
348 | break; | |
349 | case '{': | |
350 | consume_balanced ('{', '}'); | |
351 | break; | |
352 | ||
353 | case '}': | |
354 | case ']': | |
355 | case ')': | |
356 | if (token () != closer) | |
357 | parse_error ("unbalanced delimiters - expected '%c', have '%c'", | |
358 | closer, token ()); | |
359 | advance (); | |
360 | return; | |
361 | ||
362 | case EOF_TOKEN: | |
363 | parse_error ("unexpected end of file within %c%c-delimited construct", | |
364 | opener, closer); | |
365 | return; | |
366 | } | |
367 | } | |
368 | ||
369 | /* Absorb a sequence of tokens, possibly including ()[]{}-delimited | |
370 | expressions, until we encounter an end-of-statement marker (a ';' or | |
371 | a '}') outside any such delimiters; absorb that too. */ | |
372 | ||
373 | static void | |
374 | consume_until_eos (void) | |
375 | { | |
376 | for (;;) | |
377 | switch (token ()) | |
378 | { | |
379 | case ';': | |
380 | advance (); | |
381 | return; | |
382 | ||
383 | case '{': | |
384 | consume_balanced ('{', '}'); | |
385 | return; | |
386 | ||
387 | case '(': | |
388 | consume_balanced ('(', ')'); | |
389 | break; | |
390 | ||
391 | case '[': | |
392 | consume_balanced ('[', ']'); | |
393 | break; | |
394 | ||
395 | case '}': | |
396 | case ']': | |
397 | case ')': | |
398 | parse_error ("unmatched '%c' while scanning for ';'", token ()); | |
399 | return; | |
400 | ||
401 | case EOF_TOKEN: | |
402 | parse_error ("unexpected end of file while scanning for ';'"); | |
403 | return; | |
404 | ||
405 | default: | |
406 | advance (); | |
407 | break; | |
408 | } | |
409 | } | |
410 | ||
411 | /* Absorb a sequence of tokens, possibly including ()[]{}-delimited | |
412 | expressions, until we encounter a comma or semicolon outside any | |
413 | such delimiters; absorb that too. Returns true if the loop ended | |
414 | with a comma. */ | |
415 | ||
416 | static bool | |
417 | consume_until_comma_or_eos () | |
418 | { | |
419 | for (;;) | |
420 | switch (token ()) | |
421 | { | |
422 | case ',': | |
423 | advance (); | |
424 | return true; | |
425 | ||
426 | case ';': | |
427 | advance (); | |
428 | return false; | |
429 | ||
430 | case '{': | |
431 | consume_balanced ('{', '}'); | |
432 | return false; | |
433 | ||
434 | case '(': | |
435 | consume_balanced ('(', ')'); | |
436 | break; | |
437 | ||
438 | case '[': | |
439 | consume_balanced ('[', ']'); | |
440 | break; | |
441 | ||
442 | case '}': | |
443 | case ']': | |
444 | case ')': | |
445 | parse_error ("unmatched '%s' while scanning for ',' or ';'", | |
446 | print_cur_token ()); | |
447 | return false; | |
448 | ||
449 | case EOF_TOKEN: | |
450 | parse_error ("unexpected end of file while scanning for ',' or ';'"); | |
451 | return false; | |
452 | ||
453 | default: | |
454 | advance (); | |
455 | break; | |
456 | } | |
457 | } | |
458 | \f | |
459 | ||
460 | /* GTY(()) option handling. */ | |
461 | static type_p type (options_p *optsp, bool nested); | |
462 | ||
463 | /* Optional parenthesized string: ('(' string_seq ')')? */ | |
464 | static options_p | |
465 | str_optvalue_opt (options_p prev) | |
466 | { | |
467 | const char *name = advance (); | |
468 | const char *value = ""; | |
469 | if (token () == '(') | |
470 | { | |
471 | advance (); | |
472 | value = string_seq (); | |
473 | require (')'); | |
474 | } | |
475 | return create_string_option (prev, name, value); | |
476 | } | |
477 | ||
478 | /* absdecl: type '*'* | |
479 | -- a vague approximation to what the C standard calls an abstract | |
480 | declarator. The only kinds that are actually used are those that | |
481 | are just a bare type and those that have trailing pointer-stars. | |
482 | Further kinds should be implemented if and when they become | |
483 | necessary. Used only within GTY(()) option values, therefore | |
484 | further GTY(()) tags within the type are invalid. Note that the | |
485 | return value has already been run through adjust_field_type. */ | |
486 | static type_p | |
487 | absdecl (void) | |
488 | { | |
489 | type_p ty; | |
490 | options_p opts; | |
491 | ||
492 | ty = type (&opts, true); | |
493 | while (token () == '*') | |
494 | { | |
495 | ty = create_pointer (ty); | |
496 | advance (); | |
497 | } | |
498 | ||
499 | if (opts) | |
500 | parse_error ("nested GTY(()) options are invalid"); | |
501 | ||
502 | return adjust_field_type (ty, 0); | |
503 | } | |
504 | ||
505 | /* Type-option: '(' absdecl ')' */ | |
506 | static options_p | |
507 | type_optvalue (options_p prev, const char *name) | |
508 | { | |
509 | type_p ty; | |
510 | require ('('); | |
511 | ty = absdecl (); | |
512 | require (')'); | |
513 | return create_type_option (prev, name, ty); | |
514 | } | |
515 | ||
516 | /* Nested pointer data: '(' type '*'* ',' string_seq ',' string_seq ')' */ | |
517 | static options_p | |
518 | nestedptr_optvalue (options_p prev) | |
519 | { | |
520 | type_p ty; | |
521 | const char *from, *to; | |
522 | ||
523 | require ('('); | |
524 | ty = absdecl (); | |
525 | require (','); | |
526 | to = string_seq (); | |
527 | require (','); | |
528 | from = string_seq (); | |
529 | require (')'); | |
530 | ||
531 | return create_nested_ptr_option (prev, ty, to, from); | |
532 | } | |
533 | ||
534 | /* One GTY(()) option: | |
535 | ID str_optvalue_opt | |
536 | | PTR_ALIAS type_optvalue | |
537 | | NESTED_PTR nestedptr_optvalue | |
538 | */ | |
539 | static options_p | |
540 | option (options_p prev) | |
541 | { | |
542 | switch (token ()) | |
543 | { | |
544 | case ID: | |
545 | return str_optvalue_opt (prev); | |
546 | ||
547 | case PTR_ALIAS: | |
548 | advance (); | |
549 | return type_optvalue (prev, "ptr_alias"); | |
550 | ||
551 | case NESTED_PTR: | |
552 | advance (); | |
553 | return nestedptr_optvalue (prev); | |
554 | ||
555 | case USER_GTY: | |
556 | advance (); | |
557 | return create_string_option (prev, "user", ""); | |
558 | ||
559 | default: | |
560 | parse_error ("expected an option keyword, have %s", print_cur_token ()); | |
561 | advance (); | |
562 | return create_string_option (prev, "", ""); | |
563 | } | |
564 | } | |
565 | ||
566 | /* One comma-separated list of options. */ | |
567 | static options_p | |
568 | option_seq (void) | |
569 | { | |
570 | options_p o; | |
571 | ||
572 | o = option (0); | |
573 | while (token () == ',') | |
574 | { | |
575 | advance (); | |
576 | o = option (o); | |
577 | } | |
578 | return o; | |
579 | } | |
580 | ||
581 | /* GTY marker: 'GTY' '(' '(' option_seq? ')' ')' */ | |
582 | static options_p | |
583 | gtymarker (void) | |
584 | { | |
585 | options_p result = 0; | |
586 | require (GTY_TOKEN); | |
587 | require ('('); | |
588 | require ('('); | |
589 | if (token () != ')') | |
590 | result = option_seq (); | |
591 | require (')'); | |
592 | require (')'); | |
593 | return result; | |
594 | } | |
595 | ||
596 | /* Optional GTY marker. */ | |
597 | static options_p | |
598 | gtymarker_opt (void) | |
599 | { | |
600 | if (token () != GTY_TOKEN) | |
601 | return 0; | |
602 | return gtymarker (); | |
603 | } | |
604 | ||
605 | ||
606 | \f | |
607 | /* Declarators. The logic here is largely lifted from c-parser.c. | |
608 | Note that we do not have to process abstract declarators, which can | |
609 | appear only in parameter type lists or casts (but see absdecl, | |
610 | above). Also, type qualifiers are thrown out in gengtype-lex.l so | |
611 | we don't have to do it. */ | |
612 | ||
613 | /* array_and_function_declarators_opt: | |
614 | \epsilon | |
615 | array_and_function_declarators_opt ARRAY | |
616 | array_and_function_declarators_opt '(' ... ')' | |
617 | ||
618 | where '...' indicates stuff we ignore except insofar as grouping | |
619 | symbols ()[]{} must balance. | |
620 | ||
621 | Subroutine of direct_declarator - do not use elsewhere. */ | |
622 | ||
623 | static type_p | |
624 | array_and_function_declarators_opt (type_p ty) | |
625 | { | |
626 | if (token () == ARRAY) | |
627 | { | |
628 | const char *array = advance (); | |
629 | return create_array (array_and_function_declarators_opt (ty), array); | |
630 | } | |
631 | else if (token () == '(') | |
632 | { | |
633 | /* We don't need exact types for functions. */ | |
634 | consume_balanced ('(', ')'); | |
635 | array_and_function_declarators_opt (ty); | |
636 | return create_scalar_type ("function type"); | |
637 | } | |
638 | else | |
639 | return ty; | |
640 | } | |
641 | ||
642 | static type_p inner_declarator (type_p, const char **, options_p *, bool); | |
643 | ||
644 | /* direct_declarator: | |
645 | '(' inner_declarator ')' | |
646 | '(' \epsilon ')' <-- C++ ctors/dtors | |
647 | gtymarker_opt ID array_and_function_declarators_opt | |
648 | ||
649 | Subroutine of declarator, mutually recursive with inner_declarator; | |
650 | do not use elsewhere. | |
651 | ||
652 | IN_STRUCT is true if we are called while parsing structures or classes. */ | |
653 | ||
654 | static type_p | |
655 | direct_declarator (type_p ty, const char **namep, options_p *optsp, | |
656 | bool in_struct) | |
657 | { | |
658 | /* The first token in a direct-declarator must be an ID, a | |
659 | GTY marker, or an open parenthesis. */ | |
660 | switch (token ()) | |
661 | { | |
662 | case GTY_TOKEN: | |
663 | *optsp = gtymarker (); | |
664 | /* fall through */ | |
665 | ||
666 | case ID: | |
667 | *namep = require (ID); | |
668 | /* If the next token is '(', we are parsing a function declaration. | |
669 | Functions are ignored by gengtype, so we return NULL. */ | |
670 | if (token () == '(') | |
671 | return NULL; | |
672 | break; | |
673 | ||
674 | case '(': | |
675 | /* If the declarator starts with a '(', we have three options. We | |
676 | are either parsing 'TYPE (*ID)' (i.e., a function pointer) | |
677 | or 'TYPE(...)'. | |
678 | ||
679 | The latter will be a constructor iff we are inside a | |
680 | structure or class. Otherwise, it could be a typedef, but | |
681 | since we explicitly reject typedefs inside structures, we can | |
682 | assume that we found a ctor and return NULL. */ | |
683 | advance (); | |
684 | if (in_struct && token () != '*') | |
685 | { | |
686 | /* Found a constructor. Find and consume the closing ')'. */ | |
687 | while (token () != ')') | |
688 | advance (); | |
689 | advance (); | |
690 | /* Tell the caller to ignore this. */ | |
691 | return NULL; | |
692 | } | |
693 | ty = inner_declarator (ty, namep, optsp, in_struct); | |
694 | require (')'); | |
695 | break; | |
696 | ||
697 | case IGNORABLE_CXX_KEYWORD: | |
698 | /* Any C++ keyword like 'operator' means that we are not looking | |
699 | at a regular data declarator. */ | |
700 | return NULL; | |
701 | ||
702 | default: | |
703 | parse_error ("expected '(', ')', 'GTY', or an identifier, have %s", | |
704 | print_cur_token ()); | |
705 | /* Do _not_ advance if what we have is a close squiggle brace, as | |
706 | we will get much better error recovery that way. */ | |
707 | if (token () != '}') | |
708 | advance (); | |
709 | return 0; | |
710 | } | |
711 | return array_and_function_declarators_opt (ty); | |
712 | } | |
713 | ||
714 | /* The difference between inner_declarator and declarator is in the | |
715 | handling of stars. Consider this declaration: | |
716 | ||
717 | char * (*pfc) (void) | |
718 | ||
719 | It declares a pointer to a function that takes no arguments and | |
720 | returns a char*. To construct the correct type for this | |
721 | declaration, the star outside the parentheses must be processed | |
722 | _before_ the function type, the star inside the parentheses must | |
723 | be processed _after_ the function type. To accomplish this, | |
724 | declarator() creates pointers before recursing (it is actually | |
725 | coded as a while loop), whereas inner_declarator() recurses before | |
726 | creating pointers. */ | |
727 | ||
728 | /* inner_declarator: | |
729 | '*' inner_declarator | |
730 | direct_declarator | |
731 | ||
732 | Mutually recursive subroutine of direct_declarator; do not use | |
733 | elsewhere. | |
734 | ||
735 | IN_STRUCT is true if we are called while parsing structures or classes. */ | |
736 | ||
737 | static type_p | |
738 | inner_declarator (type_p ty, const char **namep, options_p *optsp, | |
739 | bool in_struct) | |
740 | { | |
741 | if (token () == '*') | |
742 | { | |
743 | type_p inner; | |
744 | advance (); | |
745 | inner = inner_declarator (ty, namep, optsp, in_struct); | |
746 | if (inner == 0) | |
747 | return 0; | |
748 | else | |
749 | return create_pointer (ty); | |
750 | } | |
751 | else | |
752 | return direct_declarator (ty, namep, optsp, in_struct); | |
753 | } | |
754 | ||
755 | /* declarator: '*'+ direct_declarator | |
756 | ||
757 | This is the sole public interface to this part of the grammar. | |
758 | Arguments are the type known so far, a pointer to where the name | |
759 | may be stored, and a pointer to where GTY options may be stored. | |
760 | ||
761 | IN_STRUCT is true when we are called to parse declarators inside | |
762 | a structure or class. | |
763 | ||
764 | Returns the final type. */ | |
765 | ||
766 | static type_p | |
767 | declarator (type_p ty, const char **namep, options_p *optsp, | |
768 | bool in_struct = false) | |
769 | { | |
770 | *namep = 0; | |
771 | *optsp = 0; | |
772 | while (token () == '*') | |
773 | { | |
774 | advance (); | |
775 | ty = create_pointer (ty); | |
776 | } | |
777 | return direct_declarator (ty, namep, optsp, in_struct); | |
778 | } | |
779 | \f | |
780 | /* Types and declarations. */ | |
781 | ||
782 | /* Structure field(s) declaration: | |
783 | ( | |
784 | type bitfield ';' | |
785 | | type declarator bitfield? ( ',' declarator bitfield? )+ ';' | |
786 | )* | |
787 | ||
788 | Knows that such declarations must end with a close brace (or, | |
789 | erroneously, at EOF). | |
790 | */ | |
791 | static pair_p | |
792 | struct_field_seq (void) | |
793 | { | |
794 | pair_p f = 0; | |
795 | type_p ty, dty; | |
796 | options_p opts, dopts; | |
797 | const char *name; | |
798 | bool another; | |
799 | ||
800 | while (token () != '}' && token () != EOF_TOKEN) | |
801 | { | |
802 | ty = type (&opts, true); | |
803 | ||
804 | /* Ignore access-control keywords ("public:" etc). */ | |
805 | while (!ty && token () == IGNORABLE_CXX_KEYWORD) | |
806 | { | |
807 | const char *keyword = advance (); | |
808 | if (strcmp (keyword, "public:") != 0 | |
809 | && strcmp (keyword, "private:") != 0 | |
810 | && strcmp (keyword, "protected:") != 0) | |
811 | break; | |
812 | ty = type (&opts, true); | |
813 | } | |
814 | ||
815 | if (!ty || token () == ':') | |
816 | { | |
817 | consume_until_eos (); | |
818 | continue; | |
819 | } | |
820 | ||
821 | do | |
822 | { | |
823 | dty = declarator (ty, &name, &dopts, true); | |
824 | ||
825 | /* There could be any number of weird things after the declarator, | |
826 | notably bitfield declarations and __attribute__s. If this | |
827 | function returns true, the last thing was a comma, so we have | |
828 | more than one declarator paired with the current type. */ | |
829 | another = consume_until_comma_or_eos (); | |
830 | ||
831 | if (!dty) | |
832 | continue; | |
833 | ||
834 | if (opts && dopts) | |
835 | parse_error ("two GTY(()) options for field %s", name); | |
836 | if (opts && !dopts) | |
837 | dopts = opts; | |
838 | ||
839 | f = create_field_at (f, dty, name, dopts, &lexer_line); | |
840 | } | |
841 | while (another); | |
842 | } | |
843 | return nreverse_pairs (f); | |
844 | } | |
845 | ||
846 | /* Return true if OPTS contain the option named STR. */ | |
847 | ||
848 | bool | |
849 | opts_have (options_p opts, const char *str) | |
850 | { | |
851 | for (options_p opt = opts; opt; opt = opt->next) | |
852 | if (strcmp (opt->name, str) == 0) | |
853 | return true; | |
854 | return false; | |
855 | } | |
856 | ||
857 | ||
858 | /* This is called type(), but what it parses (sort of) is what C calls | |
859 | declaration-specifiers and specifier-qualifier-list: | |
860 | ||
861 | SCALAR | |
862 | | ID // typedef | |
863 | | (STRUCT|UNION) ID? gtymarker? ( '{' gtymarker? struct_field_seq '}' )? | |
864 | | ENUM ID ( '{' ... '}' )? | |
865 | ||
866 | Returns a partial type; under some conditions (notably | |
867 | "struct foo GTY((...)) thing;") it may write an options | |
868 | structure to *OPTSP. | |
869 | ||
870 | NESTED is true when parsing a declaration already known to have a | |
871 | GTY marker. In these cases, typedef and enum declarations are not | |
872 | allowed because gengtype only understands types at the global | |
873 | scope. */ | |
874 | ||
875 | static type_p | |
876 | type (options_p *optsp, bool nested) | |
877 | { | |
878 | const char *s; | |
879 | *optsp = 0; | |
880 | switch (token ()) | |
881 | { | |
882 | case SCALAR: | |
883 | s = advance (); | |
884 | return create_scalar_type (s); | |
885 | ||
886 | case ID: | |
887 | s = typedef_name (); | |
888 | return resolve_typedef (s, &lexer_line); | |
889 | ||
890 | case IGNORABLE_CXX_KEYWORD: | |
891 | /* By returning NULL here, we indicate to the caller that they | |
892 | should ignore everything following this keyword up to the | |
893 | next ';' or '}'. */ | |
894 | return NULL; | |
895 | ||
896 | case STRUCT: | |
897 | case UNION: | |
898 | { | |
899 | type_p base_class = NULL; | |
900 | options_p opts = 0; | |
901 | /* GTY annotations follow attribute syntax | |
902 | GTY_BEFORE_ID is for union/struct declarations | |
903 | GTY_AFTER_ID is for variable declarations. */ | |
904 | enum | |
905 | { | |
906 | NO_GTY, | |
907 | GTY_BEFORE_ID, | |
908 | GTY_AFTER_ID | |
909 | } is_gty = NO_GTY; | |
910 | enum typekind kind = (token () == UNION) ? TYPE_UNION : TYPE_STRUCT; | |
911 | advance (); | |
912 | ||
913 | /* Top-level structures that are not explicitly tagged GTY(()) | |
914 | are treated as mere forward declarations. This is because | |
915 | there are a lot of structures that we don't need to know | |
916 | about, and some of those have C++ and macro constructs that | |
917 | we cannot handle. */ | |
918 | if (nested || token () == GTY_TOKEN) | |
919 | { | |
920 | is_gty = GTY_BEFORE_ID; | |
921 | opts = gtymarker_opt (); | |
922 | } | |
923 | ||
924 | if (token () == ID) | |
925 | s = advance (); | |
926 | else | |
927 | s = xasprintf ("anonymous:%s:%d", | |
928 | get_input_file_name (lexer_line.file), | |
929 | lexer_line.line); | |
930 | ||
931 | /* Unfortunately above GTY_TOKEN check does not capture the | |
932 | typedef struct_type GTY case. */ | |
933 | if (token () == GTY_TOKEN) | |
934 | { | |
935 | is_gty = GTY_AFTER_ID; | |
936 | opts = gtymarker_opt (); | |
937 | } | |
938 | ||
939 | bool is_user_gty = opts_have (opts, "user"); | |
940 | ||
941 | if (token () == ':') | |
942 | { | |
943 | if (is_gty && !is_user_gty) | |
944 | { | |
945 | /* For GTY-marked types that are not "user", parse some C++ | |
946 | inheritance specifications. | |
947 | We require single-inheritance from a non-template type. */ | |
948 | advance (); | |
949 | const char *basename = require (ID); | |
950 | /* This may be either an access specifier, or the base name. */ | |
951 | if (strcmp (basename, "public") == 0 | |
952 | || strcmp (basename, "protected") == 0 | |
953 | || strcmp (basename, "private") == 0) | |
954 | basename = require (ID); | |
955 | base_class = find_structure (basename, TYPE_STRUCT); | |
956 | if (!base_class) | |
957 | parse_error ("unrecognized base class: %s", basename); | |
958 | require_without_advance ('{'); | |
959 | } | |
960 | else | |
961 | { | |
962 | /* For types lacking GTY-markings, skip over C++ inheritance | |
963 | specification (and thus avoid having to parse e.g. template | |
964 | types). */ | |
965 | while (token () != '{') | |
966 | advance (); | |
967 | } | |
968 | } | |
969 | ||
970 | if (is_gty) | |
971 | { | |
972 | if (token () == '{') | |
973 | { | |
974 | pair_p fields; | |
975 | ||
976 | if (is_gty == GTY_AFTER_ID) | |
977 | parse_error ("GTY must be specified before identifier"); | |
978 | ||
979 | if (!is_user_gty) | |
980 | { | |
981 | advance (); | |
982 | fields = struct_field_seq (); | |
983 | require ('}'); | |
984 | } | |
985 | else | |
986 | { | |
987 | /* Do not look inside user defined structures. */ | |
988 | fields = NULL; | |
989 | kind = TYPE_USER_STRUCT; | |
990 | consume_balanced ('{', '}'); | |
991 | return create_user_defined_type (s, &lexer_line); | |
992 | } | |
993 | ||
994 | return new_structure (s, kind, &lexer_line, fields, opts, | |
995 | base_class); | |
996 | } | |
997 | } | |
998 | else if (token () == '{') | |
999 | consume_balanced ('{', '}'); | |
1000 | if (opts) | |
1001 | *optsp = opts; | |
1002 | return find_structure (s, kind); | |
1003 | } | |
1004 | ||
1005 | case TYPEDEF: | |
1006 | /* In C++, a typedef inside a struct/class/union defines a new | |
1007 | type for that inner scope. We cannot support this in | |
1008 | gengtype because we have no concept of scoping. | |
1009 | ||
1010 | We handle typedefs in the global scope separately (see | |
1011 | parse_file), so if we find a 'typedef', we must be inside | |
1012 | a struct. */ | |
1013 | gcc_assert (nested); | |
1014 | parse_error ("typedefs not supported in structures marked with " | |
1015 | "automatic GTY markers. Use GTY((user)) to mark " | |
1016 | "this structure."); | |
1017 | advance (); | |
1018 | return NULL; | |
1019 | ||
1020 | case ENUM: | |
1021 | advance (); | |
1022 | if (token () == ID) | |
1023 | s = advance (); | |
1024 | else | |
1025 | s = xasprintf ("anonymous:%s:%d", | |
1026 | get_input_file_name (lexer_line.file), | |
1027 | lexer_line.line); | |
1028 | ||
1029 | if (token () == '{') | |
1030 | consume_balanced ('{', '}'); | |
1031 | ||
1032 | /* If after parsing the enum we are at the end of the statement, | |
1033 | and we are currently inside a structure, then this was an | |
1034 | enum declaration inside this scope. | |
1035 | ||
1036 | We cannot support this for the same reason we cannot support | |
1037 | 'typedef' inside structures (see the TYPEDEF handler above). | |
1038 | If this happens, emit an error and return NULL. */ | |
1039 | if (nested && token () == ';') | |
1040 | { | |
1041 | parse_error ("enum definitions not supported in structures marked " | |
1042 | "with automatic GTY markers. Use GTY((user)) to mark " | |
1043 | "this structure."); | |
1044 | advance (); | |
1045 | return NULL; | |
1046 | } | |
1047 | ||
1048 | return create_scalar_type (s); | |
1049 | ||
1050 | default: | |
1051 | parse_error ("expected a type specifier, have %s", print_cur_token ()); | |
1052 | advance (); | |
1053 | return create_scalar_type ("erroneous type"); | |
1054 | } | |
1055 | } | |
1056 | \f | |
1057 | /* Top level constructs. */ | |
1058 | ||
1059 | /* Dispatch declarations beginning with 'typedef'. */ | |
1060 | ||
1061 | static void | |
1062 | typedef_decl (void) | |
1063 | { | |
1064 | type_p ty, dty; | |
1065 | const char *name; | |
1066 | options_p opts; | |
1067 | bool another; | |
1068 | ||
1069 | gcc_assert (token () == TYPEDEF); | |
1070 | advance (); | |
1071 | ||
1072 | ty = type (&opts, false); | |
1073 | if (!ty) | |
1074 | return; | |
1075 | if (opts) | |
1076 | parse_error ("GTY((...)) cannot be applied to a typedef"); | |
1077 | do | |
1078 | { | |
1079 | dty = declarator (ty, &name, &opts); | |
1080 | if (opts) | |
1081 | parse_error ("GTY((...)) cannot be applied to a typedef"); | |
1082 | ||
1083 | /* Yet another place where we could have junk (notably attributes) | |
1084 | after the declarator. */ | |
1085 | another = consume_until_comma_or_eos (); | |
1086 | if (dty) | |
1087 | do_typedef (name, dty, &lexer_line); | |
1088 | } | |
1089 | while (another); | |
1090 | } | |
1091 | ||
1092 | /* Structure definition: type() does all the work. */ | |
1093 | ||
1094 | static void | |
1095 | struct_or_union (void) | |
1096 | { | |
1097 | options_p dummy; | |
1098 | type (&dummy, false); | |
1099 | /* There may be junk after the type: notably, we cannot currently | |
1100 | distinguish 'struct foo *function(prototype);' from 'struct foo;' | |
1101 | ... we could call declarator(), but it's a waste of time at | |
1102 | present. Instead, just eat whatever token is currently lookahead | |
1103 | and go back to lexical skipping mode. */ | |
1104 | advance (); | |
1105 | } | |
1106 | ||
1107 | /* GC root declaration: | |
1108 | (extern|static) gtymarker? type ID array_declarators_opt (';'|'=') | |
1109 | If the gtymarker is not present, we ignore the rest of the declaration. */ | |
1110 | static void | |
1111 | extern_or_static (void) | |
1112 | { | |
1113 | options_p opts, opts2, dopts; | |
1114 | type_p ty, dty; | |
1115 | const char *name; | |
1116 | require2 (EXTERN, STATIC); | |
1117 | ||
1118 | if (token () != GTY_TOKEN) | |
1119 | { | |
1120 | advance (); | |
1121 | return; | |
1122 | } | |
1123 | ||
1124 | opts = gtymarker (); | |
1125 | ty = type (&opts2, true); /* if we get here, it's got a GTY(()) */ | |
1126 | dty = declarator (ty, &name, &dopts); | |
1127 | ||
1128 | if ((opts && dopts) || (opts && opts2) || (opts2 && dopts)) | |
1129 | parse_error ("GTY((...)) specified more than once for %s", name); | |
1130 | else if (opts2) | |
1131 | opts = opts2; | |
1132 | else if (dopts) | |
1133 | opts = dopts; | |
1134 | ||
1135 | if (dty) | |
1136 | { | |
1137 | note_variable (name, adjust_field_type (dty, opts), opts, &lexer_line); | |
1138 | require2 (';', '='); | |
1139 | } | |
1140 | } | |
1141 | ||
1142 | /* Parse the file FNAME for GC-relevant declarations and definitions. | |
1143 | This is the only entry point to this file. */ | |
1144 | void | |
1145 | parse_file (const char *fname) | |
1146 | { | |
1147 | yybegin (fname); | |
1148 | for (;;) | |
1149 | { | |
1150 | switch (token ()) | |
1151 | { | |
1152 | case EXTERN: | |
1153 | case STATIC: | |
1154 | extern_or_static (); | |
1155 | break; | |
1156 | ||
1157 | case STRUCT: | |
1158 | case UNION: | |
1159 | struct_or_union (); | |
1160 | break; | |
1161 | ||
1162 | case TYPEDEF: | |
1163 | typedef_decl (); | |
1164 | break; | |
1165 | ||
1166 | case EOF_TOKEN: | |
1167 | goto eof; | |
1168 | ||
1169 | default: | |
1170 | parse_error ("unexpected top level token, %s", print_cur_token ()); | |
1171 | goto eof; | |
1172 | } | |
1173 | lexer_toplevel_done = 1; | |
1174 | } | |
1175 | ||
1176 | eof: | |
1177 | advance (); | |
1178 | yyend (); | |
1179 | } |