Commit | Line | Data |
---|---|---|
ec2bcbe7 | 1 | /* C preprocessor macro expansion for GDB. |
9b254dd1 | 2 | Copyright (C) 2002, 2007, 2008 Free Software Foundation, Inc. |
ec2bcbe7 JB |
3 | Contributed by Red Hat, Inc. |
4 | ||
5 | This file is part of GDB. | |
6 | ||
7 | This program is free software; you can redistribute it and/or modify | |
8 | it under the terms of the GNU General Public License as published by | |
a9762ec7 | 9 | the Free Software Foundation; either version 3 of the License, or |
ec2bcbe7 JB |
10 | (at your option) any later version. |
11 | ||
12 | This program is distributed in the hope that it will be useful, | |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | GNU General Public License for more details. | |
16 | ||
17 | You should have received a copy of the GNU General Public License | |
a9762ec7 | 18 | along with this program. If not, see <http://www.gnu.org/licenses/>. */ |
ec2bcbe7 JB |
19 | |
20 | #include "defs.h" | |
04ea0df1 | 21 | #include "gdb_obstack.h" |
ec2bcbe7 JB |
22 | #include "bcache.h" |
23 | #include "macrotab.h" | |
24 | #include "macroexp.h" | |
25 | #include "gdb_assert.h" | |
26 | ||
27 | ||
28 | \f | |
29 | /* A resizeable, substringable string type. */ | |
30 | ||
31 | ||
32 | /* A string type that we can resize, quickly append to, and use to | |
33 | refer to substrings of other strings. */ | |
34 | struct macro_buffer | |
35 | { | |
36 | /* An array of characters. The first LEN bytes are the real text, | |
37 | but there are SIZE bytes allocated to the array. If SIZE is | |
38 | zero, then this doesn't point to a malloc'ed block. If SHARED is | |
39 | non-zero, then this buffer is actually a pointer into some larger | |
40 | string, and we shouldn't append characters to it, etc. Because | |
41 | of sharing, we can't assume in general that the text is | |
42 | null-terminated. */ | |
43 | char *text; | |
44 | ||
45 | /* The number of characters in the string. */ | |
46 | int len; | |
47 | ||
48 | /* The number of characters allocated to the string. If SHARED is | |
49 | non-zero, this is meaningless; in this case, we set it to zero so | |
50 | that any "do we have room to append something?" tests will fail, | |
51 | so we don't always have to check SHARED before using this field. */ | |
52 | int size; | |
53 | ||
54 | /* Zero if TEXT can be safely realloc'ed (i.e., it's its own malloc | |
55 | block). Non-zero if TEXT is actually pointing into the middle of | |
56 | some other block, and we shouldn't reallocate it. */ | |
57 | int shared; | |
58 | ||
59 | /* For detecting token splicing. | |
60 | ||
61 | This is the index in TEXT of the first character of the token | |
62 | that abuts the end of TEXT. If TEXT contains no tokens, then we | |
63 | set this equal to LEN. If TEXT ends in whitespace, then there is | |
64 | no token abutting the end of TEXT (it's just whitespace), and | |
65 | again, we set this equal to LEN. We set this to -1 if we don't | |
66 | know the nature of TEXT. */ | |
67 | int last_token; | |
68 | ||
69 | /* If this buffer is holding the result from get_token, then this | |
70 | is non-zero if it is an identifier token, zero otherwise. */ | |
71 | int is_identifier; | |
72 | }; | |
73 | ||
74 | ||
75 | /* Set the macro buffer *B to the empty string, guessing that its | |
76 | final contents will fit in N bytes. (It'll get resized if it | |
77 | doesn't, so the guess doesn't have to be right.) Allocate the | |
78 | initial storage with xmalloc. */ | |
79 | static void | |
80 | init_buffer (struct macro_buffer *b, int n) | |
81 | { | |
ec2bcbe7 JB |
82 | b->size = n; |
83 | if (n > 0) | |
84 | b->text = (char *) xmalloc (n); | |
85 | else | |
a86bc61c | 86 | b->text = NULL; |
ec2bcbe7 JB |
87 | b->len = 0; |
88 | b->shared = 0; | |
89 | b->last_token = -1; | |
90 | } | |
91 | ||
92 | ||
93 | /* Set the macro buffer *BUF to refer to the LEN bytes at ADDR, as a | |
94 | shared substring. */ | |
95 | static void | |
96 | init_shared_buffer (struct macro_buffer *buf, char *addr, int len) | |
97 | { | |
98 | buf->text = addr; | |
99 | buf->len = len; | |
100 | buf->shared = 1; | |
101 | buf->size = 0; | |
102 | buf->last_token = -1; | |
103 | } | |
104 | ||
105 | ||
106 | /* Free the text of the buffer B. Raise an error if B is shared. */ | |
107 | static void | |
108 | free_buffer (struct macro_buffer *b) | |
109 | { | |
110 | gdb_assert (! b->shared); | |
111 | if (b->size) | |
112 | xfree (b->text); | |
113 | } | |
114 | ||
115 | ||
116 | /* A cleanup function for macro buffers. */ | |
117 | static void | |
118 | cleanup_macro_buffer (void *untyped_buf) | |
119 | { | |
120 | free_buffer ((struct macro_buffer *) untyped_buf); | |
121 | } | |
122 | ||
123 | ||
124 | /* Resize the buffer B to be at least N bytes long. Raise an error if | |
125 | B shouldn't be resized. */ | |
126 | static void | |
127 | resize_buffer (struct macro_buffer *b, int n) | |
128 | { | |
129 | /* We shouldn't be trying to resize shared strings. */ | |
130 | gdb_assert (! b->shared); | |
131 | ||
132 | if (b->size == 0) | |
133 | b->size = n; | |
134 | else | |
135 | while (b->size <= n) | |
136 | b->size *= 2; | |
137 | ||
138 | b->text = xrealloc (b->text, b->size); | |
139 | } | |
140 | ||
141 | ||
142 | /* Append the character C to the buffer B. */ | |
39efb398 | 143 | static void |
ec2bcbe7 JB |
144 | appendc (struct macro_buffer *b, int c) |
145 | { | |
146 | int new_len = b->len + 1; | |
147 | ||
148 | if (new_len > b->size) | |
149 | resize_buffer (b, new_len); | |
150 | ||
151 | b->text[b->len] = c; | |
152 | b->len = new_len; | |
153 | } | |
154 | ||
155 | ||
156 | /* Append the LEN bytes at ADDR to the buffer B. */ | |
39efb398 | 157 | static void |
ec2bcbe7 JB |
158 | appendmem (struct macro_buffer *b, char *addr, int len) |
159 | { | |
160 | int new_len = b->len + len; | |
161 | ||
162 | if (new_len > b->size) | |
163 | resize_buffer (b, new_len); | |
164 | ||
165 | memcpy (b->text + b->len, addr, len); | |
166 | b->len = new_len; | |
167 | } | |
168 | ||
169 | ||
170 | \f | |
171 | /* Recognizing preprocessor tokens. */ | |
172 | ||
173 | ||
174 | static int | |
175 | is_whitespace (int c) | |
176 | { | |
177 | return (c == ' ' | |
178 | || c == '\t' | |
179 | || c == '\n' | |
180 | || c == '\v' | |
181 | || c == '\f'); | |
182 | } | |
183 | ||
184 | ||
185 | static int | |
186 | is_digit (int c) | |
187 | { | |
188 | return ('0' <= c && c <= '9'); | |
189 | } | |
190 | ||
191 | ||
192 | static int | |
193 | is_identifier_nondigit (int c) | |
194 | { | |
195 | return (c == '_' | |
196 | || ('a' <= c && c <= 'z') | |
197 | || ('A' <= c && c <= 'Z')); | |
198 | } | |
199 | ||
200 | ||
201 | static void | |
202 | set_token (struct macro_buffer *tok, char *start, char *end) | |
203 | { | |
204 | init_shared_buffer (tok, start, end - start); | |
205 | tok->last_token = 0; | |
206 | ||
207 | /* Presumed; get_identifier may overwrite this. */ | |
208 | tok->is_identifier = 0; | |
209 | } | |
210 | ||
211 | ||
212 | static int | |
213 | get_comment (struct macro_buffer *tok, char *p, char *end) | |
214 | { | |
215 | if (p + 2 > end) | |
216 | return 0; | |
217 | else if (p[0] == '/' | |
218 | && p[1] == '*') | |
219 | { | |
220 | char *tok_start = p; | |
221 | ||
222 | p += 2; | |
223 | ||
224 | for (; p < end; p++) | |
225 | if (p + 2 <= end | |
226 | && p[0] == '*' | |
227 | && p[1] == '/') | |
228 | { | |
229 | p += 2; | |
230 | set_token (tok, tok_start, p); | |
231 | return 1; | |
232 | } | |
233 | ||
8a3fe4f8 | 234 | error (_("Unterminated comment in macro expansion.")); |
ec2bcbe7 JB |
235 | } |
236 | else if (p[0] == '/' | |
237 | && p[1] == '/') | |
238 | { | |
239 | char *tok_start = p; | |
240 | ||
241 | p += 2; | |
242 | for (; p < end; p++) | |
243 | if (*p == '\n') | |
244 | break; | |
245 | ||
246 | set_token (tok, tok_start, p); | |
247 | return 1; | |
248 | } | |
249 | else | |
250 | return 0; | |
251 | } | |
252 | ||
253 | ||
254 | static int | |
255 | get_identifier (struct macro_buffer *tok, char *p, char *end) | |
256 | { | |
257 | if (p < end | |
258 | && is_identifier_nondigit (*p)) | |
259 | { | |
260 | char *tok_start = p; | |
261 | ||
262 | while (p < end | |
263 | && (is_identifier_nondigit (*p) | |
264 | || is_digit (*p))) | |
265 | p++; | |
266 | ||
267 | set_token (tok, tok_start, p); | |
268 | tok->is_identifier = 1; | |
269 | return 1; | |
270 | } | |
271 | else | |
272 | return 0; | |
273 | } | |
274 | ||
275 | ||
276 | static int | |
277 | get_pp_number (struct macro_buffer *tok, char *p, char *end) | |
278 | { | |
279 | if (p < end | |
280 | && (is_digit (*p) | |
281 | || *p == '.')) | |
282 | { | |
283 | char *tok_start = p; | |
284 | ||
285 | while (p < end) | |
286 | { | |
287 | if (is_digit (*p) | |
288 | || is_identifier_nondigit (*p) | |
289 | || *p == '.') | |
290 | p++; | |
291 | else if (p + 2 <= end | |
292 | && strchr ("eEpP.", *p) | |
293 | && (p[1] == '+' || p[1] == '-')) | |
294 | p += 2; | |
295 | else | |
296 | break; | |
297 | } | |
298 | ||
299 | set_token (tok, tok_start, p); | |
300 | return 1; | |
301 | } | |
302 | else | |
303 | return 0; | |
304 | } | |
305 | ||
306 | ||
307 | ||
308 | /* If the text starting at P going up to (but not including) END | |
309 | starts with a character constant, set *TOK to point to that | |
310 | character constant, and return 1. Otherwise, return zero. | |
311 | Signal an error if it contains a malformed or incomplete character | |
312 | constant. */ | |
313 | static int | |
314 | get_character_constant (struct macro_buffer *tok, char *p, char *end) | |
315 | { | |
316 | /* ISO/IEC 9899:1999 (E) Section 6.4.4.4 paragraph 1 | |
317 | But of course, what really matters is that we handle it the same | |
318 | way GDB's C/C++ lexer does. So we call parse_escape in utils.c | |
319 | to handle escape sequences. */ | |
320 | if ((p + 1 <= end && *p == '\'') | |
321 | || (p + 2 <= end && p[0] == 'L' && p[1] == '\'')) | |
322 | { | |
323 | char *tok_start = p; | |
324 | char *body_start; | |
325 | ||
326 | if (*p == '\'') | |
327 | p++; | |
328 | else if (*p == 'L') | |
329 | p += 2; | |
330 | else | |
331 | gdb_assert (0); | |
332 | ||
333 | body_start = p; | |
334 | for (;;) | |
335 | { | |
336 | if (p >= end) | |
8a3fe4f8 | 337 | error (_("Unmatched single quote.")); |
ec2bcbe7 JB |
338 | else if (*p == '\'') |
339 | { | |
340 | if (p == body_start) | |
8a3fe4f8 AC |
341 | error (_("A character constant must contain at least one " |
342 | "character.")); | |
ec2bcbe7 JB |
343 | p++; |
344 | break; | |
345 | } | |
346 | else if (*p == '\\') | |
347 | { | |
348 | p++; | |
349 | parse_escape (&p); | |
350 | } | |
351 | else | |
352 | p++; | |
353 | } | |
354 | ||
355 | set_token (tok, tok_start, p); | |
356 | return 1; | |
357 | } | |
358 | else | |
359 | return 0; | |
360 | } | |
361 | ||
362 | ||
363 | /* If the text starting at P going up to (but not including) END | |
364 | starts with a string literal, set *TOK to point to that string | |
365 | literal, and return 1. Otherwise, return zero. Signal an error if | |
366 | it contains a malformed or incomplete string literal. */ | |
367 | static int | |
368 | get_string_literal (struct macro_buffer *tok, char *p, char *end) | |
369 | { | |
370 | if ((p + 1 <= end | |
371 | && *p == '\"') | |
372 | || (p + 2 <= end | |
373 | && p[0] == 'L' | |
374 | && p[1] == '\"')) | |
375 | { | |
376 | char *tok_start = p; | |
377 | ||
378 | if (*p == '\"') | |
379 | p++; | |
380 | else if (*p == 'L') | |
381 | p += 2; | |
382 | else | |
383 | gdb_assert (0); | |
384 | ||
385 | for (;;) | |
386 | { | |
387 | if (p >= end) | |
8a3fe4f8 | 388 | error (_("Unterminated string in expression.")); |
ec2bcbe7 JB |
389 | else if (*p == '\"') |
390 | { | |
391 | p++; | |
392 | break; | |
393 | } | |
394 | else if (*p == '\n') | |
8a3fe4f8 AC |
395 | error (_("Newline characters may not appear in string " |
396 | "constants.")); | |
ec2bcbe7 JB |
397 | else if (*p == '\\') |
398 | { | |
399 | p++; | |
400 | parse_escape (&p); | |
401 | } | |
402 | else | |
403 | p++; | |
404 | } | |
405 | ||
406 | set_token (tok, tok_start, p); | |
407 | return 1; | |
408 | } | |
409 | else | |
410 | return 0; | |
411 | } | |
412 | ||
413 | ||
414 | static int | |
415 | get_punctuator (struct macro_buffer *tok, char *p, char *end) | |
416 | { | |
417 | /* Here, speed is much less important than correctness and clarity. */ | |
418 | ||
419 | /* ISO/IEC 9899:1999 (E) Section 6.4.6 Paragraph 1 */ | |
420 | static const char * const punctuators[] = { | |
421 | "[", "]", "(", ")", "{", "}", ".", "->", | |
422 | "++", "--", "&", "*", "+", "-", "~", "!", | |
423 | "/", "%", "<<", ">>", "<", ">", "<=", ">=", "==", "!=", | |
424 | "^", "|", "&&", "||", | |
425 | "?", ":", ";", "...", | |
426 | "=", "*=", "/=", "%=", "+=", "-=", "<<=", ">>=", "&=", "^=", "|=", | |
427 | ",", "#", "##", | |
428 | "<:", ":>", "<%", "%>", "%:", "%:%:", | |
429 | 0 | |
430 | }; | |
431 | ||
432 | int i; | |
433 | ||
434 | if (p + 1 <= end) | |
435 | { | |
436 | for (i = 0; punctuators[i]; i++) | |
437 | { | |
438 | const char *punctuator = punctuators[i]; | |
439 | ||
440 | if (p[0] == punctuator[0]) | |
441 | { | |
442 | int len = strlen (punctuator); | |
443 | ||
444 | if (p + len <= end | |
445 | && ! memcmp (p, punctuator, len)) | |
446 | { | |
447 | set_token (tok, p, p + len); | |
448 | return 1; | |
449 | } | |
450 | } | |
451 | } | |
452 | } | |
453 | ||
454 | return 0; | |
455 | } | |
456 | ||
457 | ||
458 | /* Peel the next preprocessor token off of SRC, and put it in TOK. | |
459 | Mutate TOK to refer to the first token in SRC, and mutate SRC to | |
460 | refer to the text after that token. SRC must be a shared buffer; | |
461 | the resulting TOK will be shared, pointing into the same string SRC | |
462 | does. Initialize TOK's last_token field. Return non-zero if we | |
463 | succeed, or 0 if we didn't find any more tokens in SRC. */ | |
464 | static int | |
465 | get_token (struct macro_buffer *tok, | |
466 | struct macro_buffer *src) | |
467 | { | |
468 | char *p = src->text; | |
469 | char *end = p + src->len; | |
470 | ||
471 | gdb_assert (src->shared); | |
472 | ||
473 | /* From the ISO C standard, ISO/IEC 9899:1999 (E), section 6.4: | |
474 | ||
475 | preprocessing-token: | |
476 | header-name | |
477 | identifier | |
478 | pp-number | |
479 | character-constant | |
480 | string-literal | |
481 | punctuator | |
482 | each non-white-space character that cannot be one of the above | |
483 | ||
484 | We don't have to deal with header-name tokens, since those can | |
485 | only occur after a #include, which we will never see. */ | |
486 | ||
487 | while (p < end) | |
488 | if (is_whitespace (*p)) | |
489 | p++; | |
490 | else if (get_comment (tok, p, end)) | |
491 | p += tok->len; | |
492 | else if (get_pp_number (tok, p, end) | |
493 | || get_character_constant (tok, p, end) | |
494 | || get_string_literal (tok, p, end) | |
495 | /* Note: the grammar in the standard seems to be | |
496 | ambiguous: L'x' can be either a wide character | |
497 | constant, or an identifier followed by a normal | |
498 | character constant. By trying `get_identifier' after | |
499 | we try get_character_constant and get_string_literal, | |
500 | we give the wide character syntax precedence. Now, | |
501 | since GDB doesn't handle wide character constants | |
502 | anyway, is this the right thing to do? */ | |
503 | || get_identifier (tok, p, end) | |
504 | || get_punctuator (tok, p, end)) | |
505 | { | |
506 | /* How many characters did we consume, including whitespace? */ | |
507 | int consumed = p - src->text + tok->len; | |
508 | src->text += consumed; | |
509 | src->len -= consumed; | |
510 | return 1; | |
511 | } | |
512 | else | |
513 | { | |
514 | /* We have found a "non-whitespace character that cannot be | |
515 | one of the above." Make a token out of it. */ | |
516 | int consumed; | |
517 | ||
518 | set_token (tok, p, p + 1); | |
519 | consumed = p - src->text + tok->len; | |
520 | src->text += consumed; | |
521 | src->len -= consumed; | |
522 | return 1; | |
523 | } | |
524 | ||
525 | return 0; | |
526 | } | |
527 | ||
528 | ||
529 | \f | |
530 | /* Appending token strings, with and without splicing */ | |
531 | ||
532 | ||
533 | /* Append the macro buffer SRC to the end of DEST, and ensure that | |
534 | doing so doesn't splice the token at the end of SRC with the token | |
535 | at the beginning of DEST. SRC and DEST must have their last_token | |
536 | fields set. Upon return, DEST's last_token field is set correctly. | |
537 | ||
538 | For example: | |
539 | ||
540 | If DEST is "(" and SRC is "y", then we can return with | |
541 | DEST set to "(y" --- we've simply appended the two buffers. | |
542 | ||
543 | However, if DEST is "x" and SRC is "y", then we must not return | |
544 | with DEST set to "xy" --- that would splice the two tokens "x" and | |
545 | "y" together to make a single token "xy". However, it would be | |
546 | fine to return with DEST set to "x y". Similarly, "<" and "<" must | |
547 | yield "< <", not "<<", etc. */ | |
548 | static void | |
549 | append_tokens_without_splicing (struct macro_buffer *dest, | |
550 | struct macro_buffer *src) | |
551 | { | |
552 | int original_dest_len = dest->len; | |
553 | struct macro_buffer dest_tail, new_token; | |
554 | ||
555 | gdb_assert (src->last_token != -1); | |
556 | gdb_assert (dest->last_token != -1); | |
557 | ||
558 | /* First, just try appending the two, and call get_token to see if | |
559 | we got a splice. */ | |
560 | appendmem (dest, src->text, src->len); | |
561 | ||
562 | /* If DEST originally had no token abutting its end, then we can't | |
563 | have spliced anything, so we're done. */ | |
564 | if (dest->last_token == original_dest_len) | |
565 | { | |
566 | dest->last_token = original_dest_len + src->last_token; | |
567 | return; | |
568 | } | |
569 | ||
570 | /* Set DEST_TAIL to point to the last token in DEST, followed by | |
571 | all the stuff we just appended. */ | |
572 | init_shared_buffer (&dest_tail, | |
573 | dest->text + dest->last_token, | |
574 | dest->len - dest->last_token); | |
575 | ||
576 | /* Re-parse DEST's last token. We know that DEST used to contain | |
577 | at least one token, so if it doesn't contain any after the | |
578 | append, then we must have spliced "/" and "*" or "/" and "/" to | |
579 | make a comment start. (Just for the record, I got this right | |
580 | the first time. This is not a bug fix.) */ | |
581 | if (get_token (&new_token, &dest_tail) | |
582 | && (new_token.text + new_token.len | |
583 | == dest->text + original_dest_len)) | |
584 | { | |
585 | /* No splice, so we're done. */ | |
586 | dest->last_token = original_dest_len + src->last_token; | |
587 | return; | |
588 | } | |
589 | ||
590 | /* Okay, a simple append caused a splice. Let's chop dest back to | |
591 | its original length and try again, but separate the texts with a | |
592 | space. */ | |
593 | dest->len = original_dest_len; | |
594 | appendc (dest, ' '); | |
595 | appendmem (dest, src->text, src->len); | |
596 | ||
597 | init_shared_buffer (&dest_tail, | |
598 | dest->text + dest->last_token, | |
599 | dest->len - dest->last_token); | |
600 | ||
601 | /* Try to re-parse DEST's last token, as above. */ | |
602 | if (get_token (&new_token, &dest_tail) | |
603 | && (new_token.text + new_token.len | |
604 | == dest->text + original_dest_len)) | |
605 | { | |
606 | /* No splice, so we're done. */ | |
607 | dest->last_token = original_dest_len + 1 + src->last_token; | |
608 | return; | |
609 | } | |
610 | ||
611 | /* As far as I know, there's no case where inserting a space isn't | |
612 | enough to prevent a splice. */ | |
613 | internal_error (__FILE__, __LINE__, | |
e2e0b3e5 | 614 | _("unable to avoid splicing tokens during macro expansion")); |
ec2bcbe7 JB |
615 | } |
616 | ||
617 | ||
618 | \f | |
619 | /* Expanding macros! */ | |
620 | ||
621 | ||
622 | /* A singly-linked list of the names of the macros we are currently | |
623 | expanding --- for detecting expansion loops. */ | |
624 | struct macro_name_list { | |
625 | const char *name; | |
626 | struct macro_name_list *next; | |
627 | }; | |
628 | ||
629 | ||
630 | /* Return non-zero if we are currently expanding the macro named NAME, | |
631 | according to LIST; otherwise, return zero. | |
632 | ||
633 | You know, it would be possible to get rid of all the NO_LOOP | |
634 | arguments to these functions by simply generating a new lookup | |
635 | function and baton which refuses to find the definition for a | |
636 | particular macro, and otherwise delegates the decision to another | |
637 | function/baton pair. But that makes the linked list of excluded | |
638 | macros chained through untyped baton pointers, which will make it | |
639 | harder to debug. :( */ | |
640 | static int | |
641 | currently_rescanning (struct macro_name_list *list, const char *name) | |
642 | { | |
643 | for (; list; list = list->next) | |
a86bc61c | 644 | if (strcmp (name, list->name) == 0) |
ec2bcbe7 JB |
645 | return 1; |
646 | ||
647 | return 0; | |
648 | } | |
649 | ||
650 | ||
651 | /* Gather the arguments to a macro expansion. | |
652 | ||
653 | NAME is the name of the macro being invoked. (It's only used for | |
654 | printing error messages.) | |
655 | ||
656 | Assume that SRC is the text of the macro invocation immediately | |
657 | following the macro name. For example, if we're processing the | |
658 | text foo(bar, baz), then NAME would be foo and SRC will be (bar, | |
659 | baz). | |
660 | ||
661 | If SRC doesn't start with an open paren ( token at all, return | |
662 | zero, leave SRC unchanged, and don't set *ARGC_P to anything. | |
663 | ||
664 | If SRC doesn't contain a properly terminated argument list, then | |
665 | raise an error. | |
666 | ||
667 | Otherwise, return a pointer to the first element of an array of | |
668 | macro buffers referring to the argument texts, and set *ARGC_P to | |
669 | the number of arguments we found --- the number of elements in the | |
670 | array. The macro buffers share their text with SRC, and their | |
671 | last_token fields are initialized. The array is allocated with | |
672 | xmalloc, and the caller is responsible for freeing it. | |
673 | ||
674 | NOTE WELL: if SRC starts with a open paren ( token followed | |
675 | immediately by a close paren ) token (e.g., the invocation looks | |
676 | like "foo()"), we treat that as one argument, which happens to be | |
677 | the empty list of tokens. The caller should keep in mind that such | |
678 | a sequence of tokens is a valid way to invoke one-parameter | |
679 | function-like macros, but also a valid way to invoke zero-parameter | |
680 | function-like macros. Eeew. | |
681 | ||
682 | Consume the tokens from SRC; after this call, SRC contains the text | |
683 | following the invocation. */ | |
684 | ||
685 | static struct macro_buffer * | |
686 | gather_arguments (const char *name, struct macro_buffer *src, int *argc_p) | |
687 | { | |
688 | struct macro_buffer tok; | |
689 | int args_len, args_size; | |
a86bc61c | 690 | struct macro_buffer *args = NULL; |
ec2bcbe7 JB |
691 | struct cleanup *back_to = make_cleanup (free_current_contents, &args); |
692 | ||
693 | /* Does SRC start with an opening paren token? Read from a copy of | |
694 | SRC, so SRC itself is unaffected if we don't find an opening | |
695 | paren. */ | |
696 | { | |
697 | struct macro_buffer temp; | |
698 | init_shared_buffer (&temp, src->text, src->len); | |
699 | ||
700 | if (! get_token (&tok, &temp) | |
701 | || tok.len != 1 | |
702 | || tok.text[0] != '(') | |
703 | { | |
704 | discard_cleanups (back_to); | |
705 | return 0; | |
706 | } | |
707 | } | |
708 | ||
709 | /* Consume SRC's opening paren. */ | |
710 | get_token (&tok, src); | |
711 | ||
712 | args_len = 0; | |
b1ddacc7 | 713 | args_size = 6; |
ec2bcbe7 JB |
714 | args = (struct macro_buffer *) xmalloc (sizeof (*args) * args_size); |
715 | ||
716 | for (;;) | |
717 | { | |
718 | struct macro_buffer *arg; | |
719 | int depth; | |
720 | ||
721 | /* Make sure we have room for the next argument. */ | |
722 | if (args_len >= args_size) | |
723 | { | |
724 | args_size *= 2; | |
725 | args = xrealloc (args, sizeof (*args) * args_size); | |
726 | } | |
727 | ||
728 | /* Initialize the next argument. */ | |
729 | arg = &args[args_len++]; | |
730 | set_token (arg, src->text, src->text); | |
731 | ||
732 | /* Gather the argument's tokens. */ | |
733 | depth = 0; | |
734 | for (;;) | |
735 | { | |
736 | char *start = src->text; | |
737 | ||
738 | if (! get_token (&tok, src)) | |
8a3fe4f8 | 739 | error (_("Malformed argument list for macro `%s'."), name); |
ec2bcbe7 JB |
740 | |
741 | /* Is tok an opening paren? */ | |
742 | if (tok.len == 1 && tok.text[0] == '(') | |
743 | depth++; | |
744 | ||
745 | /* Is tok is a closing paren? */ | |
746 | else if (tok.len == 1 && tok.text[0] == ')') | |
747 | { | |
748 | /* If it's a closing paren at the top level, then that's | |
749 | the end of the argument list. */ | |
750 | if (depth == 0) | |
751 | { | |
752 | discard_cleanups (back_to); | |
753 | *argc_p = args_len; | |
754 | return args; | |
755 | } | |
756 | ||
757 | depth--; | |
758 | } | |
759 | ||
760 | /* If tok is a comma at top level, then that's the end of | |
761 | the current argument. */ | |
762 | else if (tok.len == 1 && tok.text[0] == ',' && depth == 0) | |
763 | break; | |
764 | ||
765 | /* Extend the current argument to enclose this token. If | |
766 | this is the current argument's first token, leave out any | |
767 | leading whitespace, just for aesthetics. */ | |
768 | if (arg->len == 0) | |
769 | { | |
770 | arg->text = tok.text; | |
771 | arg->len = tok.len; | |
772 | arg->last_token = 0; | |
773 | } | |
774 | else | |
775 | { | |
776 | arg->len = (tok.text + tok.len) - arg->text; | |
777 | arg->last_token = tok.text - arg->text; | |
778 | } | |
779 | } | |
780 | } | |
781 | } | |
782 | ||
783 | ||
784 | /* The `expand' and `substitute_args' functions both invoke `scan' | |
785 | recursively, so we need a forward declaration somewhere. */ | |
786 | static void scan (struct macro_buffer *dest, | |
787 | struct macro_buffer *src, | |
788 | struct macro_name_list *no_loop, | |
789 | macro_lookup_ftype *lookup_func, | |
790 | void *lookup_baton); | |
791 | ||
792 | ||
793 | /* Given the macro definition DEF, being invoked with the actual | |
794 | arguments given by ARGC and ARGV, substitute the arguments into the | |
795 | replacement list, and store the result in DEST. | |
796 | ||
797 | If it is necessary to expand macro invocations in one of the | |
798 | arguments, use LOOKUP_FUNC and LOOKUP_BATON to find the macro | |
799 | definitions, and don't expand invocations of the macros listed in | |
800 | NO_LOOP. */ | |
801 | static void | |
802 | substitute_args (struct macro_buffer *dest, | |
803 | struct macro_definition *def, | |
804 | int argc, struct macro_buffer *argv, | |
805 | struct macro_name_list *no_loop, | |
806 | macro_lookup_ftype *lookup_func, | |
807 | void *lookup_baton) | |
808 | { | |
809 | /* A macro buffer for the macro's replacement list. */ | |
810 | struct macro_buffer replacement_list; | |
811 | ||
812 | init_shared_buffer (&replacement_list, (char *) def->replacement, | |
813 | strlen (def->replacement)); | |
814 | ||
815 | gdb_assert (dest->len == 0); | |
816 | dest->last_token = 0; | |
817 | ||
818 | for (;;) | |
819 | { | |
820 | struct macro_buffer tok; | |
821 | char *original_rl_start = replacement_list.text; | |
822 | int substituted = 0; | |
823 | ||
824 | /* Find the next token in the replacement list. */ | |
825 | if (! get_token (&tok, &replacement_list)) | |
826 | break; | |
827 | ||
828 | /* Just for aesthetics. If we skipped some whitespace, copy | |
829 | that to DEST. */ | |
830 | if (tok.text > original_rl_start) | |
831 | { | |
832 | appendmem (dest, original_rl_start, tok.text - original_rl_start); | |
833 | dest->last_token = dest->len; | |
834 | } | |
835 | ||
836 | /* Is this token the stringification operator? */ | |
837 | if (tok.len == 1 | |
838 | && tok.text[0] == '#') | |
8a3fe4f8 | 839 | error (_("Stringification is not implemented yet.")); |
ec2bcbe7 JB |
840 | |
841 | /* Is this token the splicing operator? */ | |
842 | if (tok.len == 2 | |
843 | && tok.text[0] == '#' | |
844 | && tok.text[1] == '#') | |
8a3fe4f8 | 845 | error (_("Token splicing is not implemented yet.")); |
ec2bcbe7 JB |
846 | |
847 | /* Is this token an identifier? */ | |
848 | if (tok.is_identifier) | |
849 | { | |
850 | int i; | |
851 | ||
852 | /* Is it the magic varargs parameter? */ | |
853 | if (tok.len == 11 | |
854 | && ! memcmp (tok.text, "__VA_ARGS__", 11)) | |
8a3fe4f8 | 855 | error (_("Variable-arity macros not implemented yet.")); |
ec2bcbe7 JB |
856 | |
857 | /* Is it one of the parameters? */ | |
858 | for (i = 0; i < def->argc; i++) | |
859 | if (tok.len == strlen (def->argv[i]) | |
860 | && ! memcmp (tok.text, def->argv[i], tok.len)) | |
861 | { | |
862 | struct macro_buffer arg_src; | |
863 | ||
864 | /* Expand any macro invocations in the argument text, | |
865 | and append the result to dest. Remember that scan | |
866 | mutates its source, so we need to scan a new buffer | |
867 | referring to the argument's text, not the argument | |
868 | itself. */ | |
869 | init_shared_buffer (&arg_src, argv[i].text, argv[i].len); | |
870 | scan (dest, &arg_src, no_loop, lookup_func, lookup_baton); | |
871 | substituted = 1; | |
872 | break; | |
873 | } | |
874 | } | |
875 | ||
876 | /* If it wasn't a parameter, then just copy it across. */ | |
877 | if (! substituted) | |
878 | append_tokens_without_splicing (dest, &tok); | |
879 | } | |
880 | } | |
881 | ||
882 | ||
883 | /* Expand a call to a macro named ID, whose definition is DEF. Append | |
884 | its expansion to DEST. SRC is the input text following the ID | |
885 | token. We are currently rescanning the expansions of the macros | |
886 | named in NO_LOOP; don't re-expand them. Use LOOKUP_FUNC and | |
887 | LOOKUP_BATON to find definitions for any nested macro references. | |
888 | ||
889 | Return 1 if we decided to expand it, zero otherwise. (If it's a | |
890 | function-like macro name that isn't followed by an argument list, | |
891 | we don't expand it.) If we return zero, leave SRC unchanged. */ | |
892 | static int | |
893 | expand (const char *id, | |
894 | struct macro_definition *def, | |
895 | struct macro_buffer *dest, | |
896 | struct macro_buffer *src, | |
897 | struct macro_name_list *no_loop, | |
898 | macro_lookup_ftype *lookup_func, | |
899 | void *lookup_baton) | |
900 | { | |
901 | struct macro_name_list new_no_loop; | |
902 | ||
903 | /* Create a new node to be added to the front of the no-expand list. | |
904 | This list is appropriate for re-scanning replacement lists, but | |
905 | it is *not* appropriate for scanning macro arguments; invocations | |
906 | of the macro whose arguments we are gathering *do* get expanded | |
907 | there. */ | |
908 | new_no_loop.name = id; | |
909 | new_no_loop.next = no_loop; | |
910 | ||
911 | /* What kind of macro are we expanding? */ | |
912 | if (def->kind == macro_object_like) | |
913 | { | |
914 | struct macro_buffer replacement_list; | |
915 | ||
916 | init_shared_buffer (&replacement_list, (char *) def->replacement, | |
917 | strlen (def->replacement)); | |
918 | ||
919 | scan (dest, &replacement_list, &new_no_loop, lookup_func, lookup_baton); | |
920 | return 1; | |
921 | } | |
922 | else if (def->kind == macro_function_like) | |
923 | { | |
924 | struct cleanup *back_to = make_cleanup (null_cleanup, 0); | |
0a029df5 | 925 | int argc = 0; |
a86bc61c | 926 | struct macro_buffer *argv = NULL; |
ec2bcbe7 JB |
927 | struct macro_buffer substituted; |
928 | struct macro_buffer substituted_src; | |
929 | ||
930 | if (def->argc >= 1 | |
a86bc61c | 931 | && strcmp (def->argv[def->argc - 1], "...") == 0) |
8a3fe4f8 | 932 | error (_("Varargs macros not implemented yet.")); |
ec2bcbe7 JB |
933 | |
934 | make_cleanup (free_current_contents, &argv); | |
935 | argv = gather_arguments (id, src, &argc); | |
936 | ||
937 | /* If we couldn't find any argument list, then we don't expand | |
938 | this macro. */ | |
939 | if (! argv) | |
940 | { | |
941 | do_cleanups (back_to); | |
942 | return 0; | |
943 | } | |
944 | ||
945 | /* Check that we're passing an acceptable number of arguments for | |
946 | this macro. */ | |
947 | if (argc != def->argc) | |
948 | { | |
949 | /* Remember that a sequence of tokens like "foo()" is a | |
950 | valid invocation of a macro expecting either zero or one | |
951 | arguments. */ | |
952 | if (! (argc == 1 | |
953 | && argv[0].len == 0 | |
954 | && def->argc == 0)) | |
8a3fe4f8 AC |
955 | error (_("Wrong number of arguments to macro `%s' " |
956 | "(expected %d, got %d)."), | |
ec2bcbe7 JB |
957 | id, def->argc, argc); |
958 | } | |
959 | ||
960 | /* Note that we don't expand macro invocations in the arguments | |
961 | yet --- we let subst_args take care of that. Parameters that | |
962 | appear as operands of the stringifying operator "#" or the | |
963 | splicing operator "##" don't get macro references expanded, | |
964 | so we can't really tell whether it's appropriate to macro- | |
965 | expand an argument until we see how it's being used. */ | |
966 | init_buffer (&substituted, 0); | |
967 | make_cleanup (cleanup_macro_buffer, &substituted); | |
968 | substitute_args (&substituted, def, argc, argv, no_loop, | |
969 | lookup_func, lookup_baton); | |
970 | ||
971 | /* Now `substituted' is the macro's replacement list, with all | |
972 | argument values substituted into it properly. Re-scan it for | |
973 | macro references, but don't expand invocations of this macro. | |
974 | ||
975 | We create a new buffer, `substituted_src', which points into | |
976 | `substituted', and scan that. We can't scan `substituted' | |
977 | itself, since the tokenization process moves the buffer's | |
978 | text pointer around, and we still need to be able to find | |
979 | `substituted's original text buffer after scanning it so we | |
980 | can free it. */ | |
981 | init_shared_buffer (&substituted_src, substituted.text, substituted.len); | |
982 | scan (dest, &substituted_src, &new_no_loop, lookup_func, lookup_baton); | |
983 | ||
984 | do_cleanups (back_to); | |
985 | ||
986 | return 1; | |
987 | } | |
988 | else | |
e2e0b3e5 | 989 | internal_error (__FILE__, __LINE__, _("bad macro definition kind")); |
ec2bcbe7 JB |
990 | } |
991 | ||
992 | ||
993 | /* If the single token in SRC_FIRST followed by the tokens in SRC_REST | |
994 | constitute a macro invokation not forbidden in NO_LOOP, append its | |
995 | expansion to DEST and return non-zero. Otherwise, return zero, and | |
996 | leave DEST unchanged. | |
997 | ||
998 | SRC_FIRST and SRC_REST must be shared buffers; DEST must not be one. | |
999 | SRC_FIRST must be a string built by get_token. */ | |
1000 | static int | |
1001 | maybe_expand (struct macro_buffer *dest, | |
1002 | struct macro_buffer *src_first, | |
1003 | struct macro_buffer *src_rest, | |
1004 | struct macro_name_list *no_loop, | |
1005 | macro_lookup_ftype *lookup_func, | |
1006 | void *lookup_baton) | |
1007 | { | |
1008 | gdb_assert (src_first->shared); | |
1009 | gdb_assert (src_rest->shared); | |
1010 | gdb_assert (! dest->shared); | |
1011 | ||
1012 | /* Is this token an identifier? */ | |
1013 | if (src_first->is_identifier) | |
1014 | { | |
1015 | /* Make a null-terminated copy of it, since that's what our | |
1016 | lookup function expects. */ | |
1017 | char *id = xmalloc (src_first->len + 1); | |
1018 | struct cleanup *back_to = make_cleanup (xfree, id); | |
1019 | memcpy (id, src_first->text, src_first->len); | |
1020 | id[src_first->len] = 0; | |
1021 | ||
1022 | /* If we're currently re-scanning the result of expanding | |
1023 | this macro, don't expand it again. */ | |
1024 | if (! currently_rescanning (no_loop, id)) | |
1025 | { | |
1026 | /* Does this identifier have a macro definition in scope? */ | |
1027 | struct macro_definition *def = lookup_func (id, lookup_baton); | |
1028 | ||
1029 | if (def && expand (id, def, dest, src_rest, no_loop, | |
1030 | lookup_func, lookup_baton)) | |
1031 | { | |
1032 | do_cleanups (back_to); | |
1033 | return 1; | |
1034 | } | |
1035 | } | |
1036 | ||
1037 | do_cleanups (back_to); | |
1038 | } | |
1039 | ||
1040 | return 0; | |
1041 | } | |
1042 | ||
1043 | ||
1044 | /* Expand macro references in SRC, appending the results to DEST. | |
1045 | Assume we are re-scanning the result of expanding the macros named | |
1046 | in NO_LOOP, and don't try to re-expand references to them. | |
1047 | ||
1048 | SRC must be a shared buffer; DEST must not be one. */ | |
1049 | static void | |
1050 | scan (struct macro_buffer *dest, | |
1051 | struct macro_buffer *src, | |
1052 | struct macro_name_list *no_loop, | |
1053 | macro_lookup_ftype *lookup_func, | |
1054 | void *lookup_baton) | |
1055 | { | |
1056 | gdb_assert (src->shared); | |
1057 | gdb_assert (! dest->shared); | |
1058 | ||
1059 | for (;;) | |
1060 | { | |
1061 | struct macro_buffer tok; | |
1062 | char *original_src_start = src->text; | |
1063 | ||
1064 | /* Find the next token in SRC. */ | |
1065 | if (! get_token (&tok, src)) | |
1066 | break; | |
1067 | ||
1068 | /* Just for aesthetics. If we skipped some whitespace, copy | |
1069 | that to DEST. */ | |
1070 | if (tok.text > original_src_start) | |
1071 | { | |
1072 | appendmem (dest, original_src_start, tok.text - original_src_start); | |
1073 | dest->last_token = dest->len; | |
1074 | } | |
1075 | ||
1076 | if (! maybe_expand (dest, &tok, src, no_loop, lookup_func, lookup_baton)) | |
1077 | /* We didn't end up expanding tok as a macro reference, so | |
1078 | simply append it to dest. */ | |
1079 | append_tokens_without_splicing (dest, &tok); | |
1080 | } | |
1081 | ||
1082 | /* Just for aesthetics. If there was any trailing whitespace in | |
1083 | src, copy it to dest. */ | |
1084 | if (src->len) | |
1085 | { | |
1086 | appendmem (dest, src->text, src->len); | |
1087 | dest->last_token = dest->len; | |
1088 | } | |
1089 | } | |
1090 | ||
1091 | ||
1092 | char * | |
1093 | macro_expand (const char *source, | |
1094 | macro_lookup_ftype *lookup_func, | |
1095 | void *lookup_func_baton) | |
1096 | { | |
1097 | struct macro_buffer src, dest; | |
1098 | struct cleanup *back_to; | |
1099 | ||
1100 | init_shared_buffer (&src, (char *) source, strlen (source)); | |
1101 | ||
1102 | init_buffer (&dest, 0); | |
1103 | dest.last_token = 0; | |
1104 | back_to = make_cleanup (cleanup_macro_buffer, &dest); | |
1105 | ||
1106 | scan (&dest, &src, 0, lookup_func, lookup_func_baton); | |
1107 | ||
1108 | appendc (&dest, '\0'); | |
1109 | ||
1110 | discard_cleanups (back_to); | |
1111 | return dest.text; | |
1112 | } | |
1113 | ||
1114 | ||
1115 | char * | |
1116 | macro_expand_once (const char *source, | |
1117 | macro_lookup_ftype *lookup_func, | |
1118 | void *lookup_func_baton) | |
1119 | { | |
8a3fe4f8 | 1120 | error (_("Expand-once not implemented yet.")); |
ec2bcbe7 JB |
1121 | } |
1122 | ||
1123 | ||
1124 | char * | |
1125 | macro_expand_next (char **lexptr, | |
1126 | macro_lookup_ftype *lookup_func, | |
1127 | void *lookup_baton) | |
1128 | { | |
1129 | struct macro_buffer src, dest, tok; | |
1130 | struct cleanup *back_to; | |
1131 | ||
1132 | /* Set up SRC to refer to the input text, pointed to by *lexptr. */ | |
1133 | init_shared_buffer (&src, *lexptr, strlen (*lexptr)); | |
1134 | ||
1135 | /* Set up DEST to receive the expansion, if there is one. */ | |
1136 | init_buffer (&dest, 0); | |
1137 | dest.last_token = 0; | |
1138 | back_to = make_cleanup (cleanup_macro_buffer, &dest); | |
1139 | ||
1140 | /* Get the text's first preprocessing token. */ | |
1141 | if (! get_token (&tok, &src)) | |
1142 | { | |
1143 | do_cleanups (back_to); | |
1144 | return 0; | |
1145 | } | |
1146 | ||
1147 | /* If it's a macro invocation, expand it. */ | |
1148 | if (maybe_expand (&dest, &tok, &src, 0, lookup_func, lookup_baton)) | |
1149 | { | |
1150 | /* It was a macro invocation! Package up the expansion as a | |
1151 | null-terminated string and return it. Set *lexptr to the | |
1152 | start of the next token in the input. */ | |
1153 | appendc (&dest, '\0'); | |
1154 | discard_cleanups (back_to); | |
1155 | *lexptr = src.text; | |
1156 | return dest.text; | |
1157 | } | |
1158 | else | |
1159 | { | |
1160 | /* It wasn't a macro invocation. */ | |
1161 | do_cleanups (back_to); | |
1162 | return 0; | |
1163 | } | |
1164 | } |