Commit | Line | Data |
---|---|---|
ec2bcbe7 | 1 | /* C preprocessor macro expansion for GDB. |
0fb0cc75 | 2 | Copyright (C) 2002, 2007, 2008, 2009 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 | ||
d7d9f01e TT |
174 | int |
175 | macro_is_whitespace (int c) | |
ec2bcbe7 JB |
176 | { |
177 | return (c == ' ' | |
178 | || c == '\t' | |
179 | || c == '\n' | |
180 | || c == '\v' | |
181 | || c == '\f'); | |
182 | } | |
183 | ||
184 | ||
d7d9f01e TT |
185 | int |
186 | macro_is_digit (int c) | |
ec2bcbe7 JB |
187 | { |
188 | return ('0' <= c && c <= '9'); | |
189 | } | |
190 | ||
191 | ||
d7d9f01e TT |
192 | int |
193 | macro_is_identifier_nondigit (int c) | |
ec2bcbe7 JB |
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 | |
d7d9f01e | 258 | && macro_is_identifier_nondigit (*p)) |
ec2bcbe7 JB |
259 | { |
260 | char *tok_start = p; | |
261 | ||
262 | while (p < end | |
d7d9f01e TT |
263 | && (macro_is_identifier_nondigit (*p) |
264 | || macro_is_digit (*p))) | |
ec2bcbe7 JB |
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 | |
d7d9f01e | 280 | && (macro_is_digit (*p) |
17c8aaf5 TT |
281 | || (*p == '.' |
282 | && p + 2 <= end | |
283 | && macro_is_digit (p[1])))) | |
ec2bcbe7 JB |
284 | { |
285 | char *tok_start = p; | |
286 | ||
287 | while (p < end) | |
288 | { | |
17c8aaf5 TT |
289 | if (p + 2 <= end |
290 | && strchr ("eEpP", *p) | |
291 | && (p[1] == '+' || p[1] == '-')) | |
ec2bcbe7 | 292 | p += 2; |
17c8aaf5 TT |
293 | else if (macro_is_digit (*p) |
294 | || macro_is_identifier_nondigit (*p) | |
295 | || *p == '.') | |
296 | p++; | |
ec2bcbe7 JB |
297 | else |
298 | break; | |
299 | } | |
300 | ||
301 | set_token (tok, tok_start, p); | |
302 | return 1; | |
303 | } | |
304 | else | |
305 | return 0; | |
306 | } | |
307 | ||
308 | ||
309 | ||
310 | /* If the text starting at P going up to (but not including) END | |
311 | starts with a character constant, set *TOK to point to that | |
312 | character constant, and return 1. Otherwise, return zero. | |
313 | Signal an error if it contains a malformed or incomplete character | |
314 | constant. */ | |
315 | static int | |
316 | get_character_constant (struct macro_buffer *tok, char *p, char *end) | |
317 | { | |
318 | /* ISO/IEC 9899:1999 (E) Section 6.4.4.4 paragraph 1 | |
319 | But of course, what really matters is that we handle it the same | |
320 | way GDB's C/C++ lexer does. So we call parse_escape in utils.c | |
321 | to handle escape sequences. */ | |
322 | if ((p + 1 <= end && *p == '\'') | |
323 | || (p + 2 <= end && p[0] == 'L' && p[1] == '\'')) | |
324 | { | |
325 | char *tok_start = p; | |
326 | char *body_start; | |
327 | ||
328 | if (*p == '\'') | |
329 | p++; | |
330 | else if (*p == 'L') | |
331 | p += 2; | |
332 | else | |
333 | gdb_assert (0); | |
334 | ||
335 | body_start = p; | |
336 | for (;;) | |
337 | { | |
338 | if (p >= end) | |
8a3fe4f8 | 339 | error (_("Unmatched single quote.")); |
ec2bcbe7 JB |
340 | else if (*p == '\'') |
341 | { | |
342 | if (p == body_start) | |
8a3fe4f8 AC |
343 | error (_("A character constant must contain at least one " |
344 | "character.")); | |
ec2bcbe7 JB |
345 | p++; |
346 | break; | |
347 | } | |
348 | else if (*p == '\\') | |
349 | { | |
350 | p++; | |
351 | parse_escape (&p); | |
352 | } | |
353 | else | |
354 | p++; | |
355 | } | |
356 | ||
357 | set_token (tok, tok_start, p); | |
358 | return 1; | |
359 | } | |
360 | else | |
361 | return 0; | |
362 | } | |
363 | ||
364 | ||
365 | /* If the text starting at P going up to (but not including) END | |
366 | starts with a string literal, set *TOK to point to that string | |
367 | literal, and return 1. Otherwise, return zero. Signal an error if | |
368 | it contains a malformed or incomplete string literal. */ | |
369 | static int | |
370 | get_string_literal (struct macro_buffer *tok, char *p, char *end) | |
371 | { | |
372 | if ((p + 1 <= end | |
373 | && *p == '\"') | |
374 | || (p + 2 <= end | |
375 | && p[0] == 'L' | |
376 | && p[1] == '\"')) | |
377 | { | |
378 | char *tok_start = p; | |
379 | ||
380 | if (*p == '\"') | |
381 | p++; | |
382 | else if (*p == 'L') | |
383 | p += 2; | |
384 | else | |
385 | gdb_assert (0); | |
386 | ||
387 | for (;;) | |
388 | { | |
389 | if (p >= end) | |
8a3fe4f8 | 390 | error (_("Unterminated string in expression.")); |
ec2bcbe7 JB |
391 | else if (*p == '\"') |
392 | { | |
393 | p++; | |
394 | break; | |
395 | } | |
396 | else if (*p == '\n') | |
8a3fe4f8 AC |
397 | error (_("Newline characters may not appear in string " |
398 | "constants.")); | |
ec2bcbe7 JB |
399 | else if (*p == '\\') |
400 | { | |
401 | p++; | |
402 | parse_escape (&p); | |
403 | } | |
404 | else | |
405 | p++; | |
406 | } | |
407 | ||
408 | set_token (tok, tok_start, p); | |
409 | return 1; | |
410 | } | |
411 | else | |
412 | return 0; | |
413 | } | |
414 | ||
415 | ||
416 | static int | |
417 | get_punctuator (struct macro_buffer *tok, char *p, char *end) | |
418 | { | |
419 | /* Here, speed is much less important than correctness and clarity. */ | |
420 | ||
ccb3ac8a TT |
421 | /* ISO/IEC 9899:1999 (E) Section 6.4.6 Paragraph 1. |
422 | Note that this table is ordered in a special way. A punctuator | |
423 | which is a prefix of another punctuator must appear after its | |
424 | "extension". Otherwise, the wrong token will be returned. */ | |
ec2bcbe7 | 425 | static const char * const punctuators[] = { |
ccb3ac8a TT |
426 | "[", "]", "(", ")", "{", "}", "?", ";", ",", "~", |
427 | "...", ".", | |
428 | "->", "--", "-=", "-", | |
429 | "++", "+=", "+", | |
430 | "*=", "*", | |
431 | "!=", "!", | |
432 | "&&", "&=", "&", | |
433 | "/=", "/", | |
434 | "%>", "%:%:", "%:", "%=", "%", | |
435 | "^=", "^", | |
436 | "##", "#", | |
437 | ":>", ":", | |
438 | "||", "|=", "|", | |
439 | "<<=", "<<", "<=", "<:", "<%", "<", | |
440 | ">>=", ">>", ">=", ">", | |
441 | "==", "=", | |
ec2bcbe7 JB |
442 | 0 |
443 | }; | |
444 | ||
445 | int i; | |
446 | ||
447 | if (p + 1 <= end) | |
448 | { | |
449 | for (i = 0; punctuators[i]; i++) | |
450 | { | |
451 | const char *punctuator = punctuators[i]; | |
452 | ||
453 | if (p[0] == punctuator[0]) | |
454 | { | |
455 | int len = strlen (punctuator); | |
456 | ||
457 | if (p + len <= end | |
458 | && ! memcmp (p, punctuator, len)) | |
459 | { | |
460 | set_token (tok, p, p + len); | |
461 | return 1; | |
462 | } | |
463 | } | |
464 | } | |
465 | } | |
466 | ||
467 | return 0; | |
468 | } | |
469 | ||
470 | ||
471 | /* Peel the next preprocessor token off of SRC, and put it in TOK. | |
472 | Mutate TOK to refer to the first token in SRC, and mutate SRC to | |
473 | refer to the text after that token. SRC must be a shared buffer; | |
474 | the resulting TOK will be shared, pointing into the same string SRC | |
475 | does. Initialize TOK's last_token field. Return non-zero if we | |
476 | succeed, or 0 if we didn't find any more tokens in SRC. */ | |
477 | static int | |
478 | get_token (struct macro_buffer *tok, | |
479 | struct macro_buffer *src) | |
480 | { | |
481 | char *p = src->text; | |
482 | char *end = p + src->len; | |
483 | ||
484 | gdb_assert (src->shared); | |
485 | ||
486 | /* From the ISO C standard, ISO/IEC 9899:1999 (E), section 6.4: | |
487 | ||
488 | preprocessing-token: | |
489 | header-name | |
490 | identifier | |
491 | pp-number | |
492 | character-constant | |
493 | string-literal | |
494 | punctuator | |
495 | each non-white-space character that cannot be one of the above | |
496 | ||
497 | We don't have to deal with header-name tokens, since those can | |
498 | only occur after a #include, which we will never see. */ | |
499 | ||
500 | while (p < end) | |
d7d9f01e | 501 | if (macro_is_whitespace (*p)) |
ec2bcbe7 JB |
502 | p++; |
503 | else if (get_comment (tok, p, end)) | |
504 | p += tok->len; | |
505 | else if (get_pp_number (tok, p, end) | |
506 | || get_character_constant (tok, p, end) | |
507 | || get_string_literal (tok, p, end) | |
508 | /* Note: the grammar in the standard seems to be | |
509 | ambiguous: L'x' can be either a wide character | |
510 | constant, or an identifier followed by a normal | |
511 | character constant. By trying `get_identifier' after | |
512 | we try get_character_constant and get_string_literal, | |
513 | we give the wide character syntax precedence. Now, | |
514 | since GDB doesn't handle wide character constants | |
515 | anyway, is this the right thing to do? */ | |
516 | || get_identifier (tok, p, end) | |
517 | || get_punctuator (tok, p, end)) | |
518 | { | |
519 | /* How many characters did we consume, including whitespace? */ | |
520 | int consumed = p - src->text + tok->len; | |
521 | src->text += consumed; | |
522 | src->len -= consumed; | |
523 | return 1; | |
524 | } | |
525 | else | |
526 | { | |
527 | /* We have found a "non-whitespace character that cannot be | |
528 | one of the above." Make a token out of it. */ | |
529 | int consumed; | |
530 | ||
531 | set_token (tok, p, p + 1); | |
532 | consumed = p - src->text + tok->len; | |
533 | src->text += consumed; | |
534 | src->len -= consumed; | |
535 | return 1; | |
536 | } | |
537 | ||
538 | return 0; | |
539 | } | |
540 | ||
541 | ||
542 | \f | |
543 | /* Appending token strings, with and without splicing */ | |
544 | ||
545 | ||
546 | /* Append the macro buffer SRC to the end of DEST, and ensure that | |
547 | doing so doesn't splice the token at the end of SRC with the token | |
548 | at the beginning of DEST. SRC and DEST must have their last_token | |
549 | fields set. Upon return, DEST's last_token field is set correctly. | |
550 | ||
551 | For example: | |
552 | ||
553 | If DEST is "(" and SRC is "y", then we can return with | |
554 | DEST set to "(y" --- we've simply appended the two buffers. | |
555 | ||
556 | However, if DEST is "x" and SRC is "y", then we must not return | |
557 | with DEST set to "xy" --- that would splice the two tokens "x" and | |
558 | "y" together to make a single token "xy". However, it would be | |
559 | fine to return with DEST set to "x y". Similarly, "<" and "<" must | |
560 | yield "< <", not "<<", etc. */ | |
561 | static void | |
562 | append_tokens_without_splicing (struct macro_buffer *dest, | |
563 | struct macro_buffer *src) | |
564 | { | |
565 | int original_dest_len = dest->len; | |
566 | struct macro_buffer dest_tail, new_token; | |
567 | ||
568 | gdb_assert (src->last_token != -1); | |
569 | gdb_assert (dest->last_token != -1); | |
570 | ||
571 | /* First, just try appending the two, and call get_token to see if | |
572 | we got a splice. */ | |
573 | appendmem (dest, src->text, src->len); | |
574 | ||
575 | /* If DEST originally had no token abutting its end, then we can't | |
576 | have spliced anything, so we're done. */ | |
577 | if (dest->last_token == original_dest_len) | |
578 | { | |
579 | dest->last_token = original_dest_len + src->last_token; | |
580 | return; | |
581 | } | |
582 | ||
583 | /* Set DEST_TAIL to point to the last token in DEST, followed by | |
584 | all the stuff we just appended. */ | |
585 | init_shared_buffer (&dest_tail, | |
586 | dest->text + dest->last_token, | |
587 | dest->len - dest->last_token); | |
588 | ||
589 | /* Re-parse DEST's last token. We know that DEST used to contain | |
590 | at least one token, so if it doesn't contain any after the | |
591 | append, then we must have spliced "/" and "*" or "/" and "/" to | |
592 | make a comment start. (Just for the record, I got this right | |
593 | the first time. This is not a bug fix.) */ | |
594 | if (get_token (&new_token, &dest_tail) | |
595 | && (new_token.text + new_token.len | |
596 | == dest->text + original_dest_len)) | |
597 | { | |
598 | /* No splice, so we're done. */ | |
599 | dest->last_token = original_dest_len + src->last_token; | |
600 | return; | |
601 | } | |
602 | ||
603 | /* Okay, a simple append caused a splice. Let's chop dest back to | |
604 | its original length and try again, but separate the texts with a | |
605 | space. */ | |
606 | dest->len = original_dest_len; | |
607 | appendc (dest, ' '); | |
608 | appendmem (dest, src->text, src->len); | |
609 | ||
610 | init_shared_buffer (&dest_tail, | |
611 | dest->text + dest->last_token, | |
612 | dest->len - dest->last_token); | |
613 | ||
614 | /* Try to re-parse DEST's last token, as above. */ | |
615 | if (get_token (&new_token, &dest_tail) | |
616 | && (new_token.text + new_token.len | |
617 | == dest->text + original_dest_len)) | |
618 | { | |
619 | /* No splice, so we're done. */ | |
620 | dest->last_token = original_dest_len + 1 + src->last_token; | |
621 | return; | |
622 | } | |
623 | ||
624 | /* As far as I know, there's no case where inserting a space isn't | |
625 | enough to prevent a splice. */ | |
626 | internal_error (__FILE__, __LINE__, | |
e2e0b3e5 | 627 | _("unable to avoid splicing tokens during macro expansion")); |
ec2bcbe7 JB |
628 | } |
629 | ||
2fae03e8 TT |
630 | /* Stringify an argument, and insert it into DEST. ARG is the text to |
631 | stringify; it is LEN bytes long. */ | |
632 | ||
633 | static void | |
634 | stringify (struct macro_buffer *dest, char *arg, int len) | |
635 | { | |
636 | /* Trim initial whitespace from ARG. */ | |
637 | while (len > 0 && macro_is_whitespace (*arg)) | |
638 | { | |
639 | ++arg; | |
640 | --len; | |
641 | } | |
642 | ||
643 | /* Trim trailing whitespace from ARG. */ | |
644 | while (len > 0 && macro_is_whitespace (arg[len - 1])) | |
645 | --len; | |
646 | ||
647 | /* Insert the string. */ | |
648 | appendc (dest, '"'); | |
649 | while (len > 0) | |
650 | { | |
651 | /* We could try to handle strange cases here, like control | |
652 | characters, but there doesn't seem to be much point. */ | |
653 | if (macro_is_whitespace (*arg)) | |
654 | { | |
655 | /* Replace a sequence of whitespace with a single space. */ | |
656 | appendc (dest, ' '); | |
657 | while (len > 1 && macro_is_whitespace (arg[1])) | |
658 | { | |
659 | ++arg; | |
660 | --len; | |
661 | } | |
662 | } | |
663 | else if (*arg == '\\' || *arg == '"') | |
664 | { | |
665 | appendc (dest, '\\'); | |
666 | appendc (dest, *arg); | |
667 | } | |
668 | else | |
669 | appendc (dest, *arg); | |
670 | ++arg; | |
671 | --len; | |
672 | } | |
673 | appendc (dest, '"'); | |
674 | dest->last_token = dest->len; | |
675 | } | |
ec2bcbe7 JB |
676 | |
677 | \f | |
678 | /* Expanding macros! */ | |
679 | ||
680 | ||
681 | /* A singly-linked list of the names of the macros we are currently | |
682 | expanding --- for detecting expansion loops. */ | |
683 | struct macro_name_list { | |
684 | const char *name; | |
685 | struct macro_name_list *next; | |
686 | }; | |
687 | ||
688 | ||
689 | /* Return non-zero if we are currently expanding the macro named NAME, | |
690 | according to LIST; otherwise, return zero. | |
691 | ||
692 | You know, it would be possible to get rid of all the NO_LOOP | |
693 | arguments to these functions by simply generating a new lookup | |
694 | function and baton which refuses to find the definition for a | |
695 | particular macro, and otherwise delegates the decision to another | |
696 | function/baton pair. But that makes the linked list of excluded | |
697 | macros chained through untyped baton pointers, which will make it | |
698 | harder to debug. :( */ | |
699 | static int | |
700 | currently_rescanning (struct macro_name_list *list, const char *name) | |
701 | { | |
702 | for (; list; list = list->next) | |
a86bc61c | 703 | if (strcmp (name, list->name) == 0) |
ec2bcbe7 JB |
704 | return 1; |
705 | ||
706 | return 0; | |
707 | } | |
708 | ||
709 | ||
710 | /* Gather the arguments to a macro expansion. | |
711 | ||
712 | NAME is the name of the macro being invoked. (It's only used for | |
713 | printing error messages.) | |
714 | ||
715 | Assume that SRC is the text of the macro invocation immediately | |
716 | following the macro name. For example, if we're processing the | |
717 | text foo(bar, baz), then NAME would be foo and SRC will be (bar, | |
718 | baz). | |
719 | ||
720 | If SRC doesn't start with an open paren ( token at all, return | |
721 | zero, leave SRC unchanged, and don't set *ARGC_P to anything. | |
722 | ||
723 | If SRC doesn't contain a properly terminated argument list, then | |
724 | raise an error. | |
2fae03e8 TT |
725 | |
726 | For a variadic macro, NARGS holds the number of formal arguments to | |
727 | the macro. For a GNU-style variadic macro, this should be the | |
728 | number of named arguments. For a non-variadic macro, NARGS should | |
729 | be -1. | |
ec2bcbe7 JB |
730 | |
731 | Otherwise, return a pointer to the first element of an array of | |
732 | macro buffers referring to the argument texts, and set *ARGC_P to | |
733 | the number of arguments we found --- the number of elements in the | |
734 | array. The macro buffers share their text with SRC, and their | |
735 | last_token fields are initialized. The array is allocated with | |
736 | xmalloc, and the caller is responsible for freeing it. | |
737 | ||
738 | NOTE WELL: if SRC starts with a open paren ( token followed | |
739 | immediately by a close paren ) token (e.g., the invocation looks | |
740 | like "foo()"), we treat that as one argument, which happens to be | |
741 | the empty list of tokens. The caller should keep in mind that such | |
742 | a sequence of tokens is a valid way to invoke one-parameter | |
743 | function-like macros, but also a valid way to invoke zero-parameter | |
744 | function-like macros. Eeew. | |
745 | ||
746 | Consume the tokens from SRC; after this call, SRC contains the text | |
747 | following the invocation. */ | |
748 | ||
749 | static struct macro_buffer * | |
2fae03e8 TT |
750 | gather_arguments (const char *name, struct macro_buffer *src, |
751 | int nargs, int *argc_p) | |
ec2bcbe7 JB |
752 | { |
753 | struct macro_buffer tok; | |
754 | int args_len, args_size; | |
a86bc61c | 755 | struct macro_buffer *args = NULL; |
ec2bcbe7 JB |
756 | struct cleanup *back_to = make_cleanup (free_current_contents, &args); |
757 | ||
758 | /* Does SRC start with an opening paren token? Read from a copy of | |
759 | SRC, so SRC itself is unaffected if we don't find an opening | |
760 | paren. */ | |
761 | { | |
762 | struct macro_buffer temp; | |
763 | init_shared_buffer (&temp, src->text, src->len); | |
764 | ||
765 | if (! get_token (&tok, &temp) | |
766 | || tok.len != 1 | |
767 | || tok.text[0] != '(') | |
768 | { | |
769 | discard_cleanups (back_to); | |
770 | return 0; | |
771 | } | |
772 | } | |
773 | ||
774 | /* Consume SRC's opening paren. */ | |
775 | get_token (&tok, src); | |
776 | ||
777 | args_len = 0; | |
b1ddacc7 | 778 | args_size = 6; |
ec2bcbe7 JB |
779 | args = (struct macro_buffer *) xmalloc (sizeof (*args) * args_size); |
780 | ||
781 | for (;;) | |
782 | { | |
783 | struct macro_buffer *arg; | |
784 | int depth; | |
785 | ||
786 | /* Make sure we have room for the next argument. */ | |
787 | if (args_len >= args_size) | |
788 | { | |
789 | args_size *= 2; | |
790 | args = xrealloc (args, sizeof (*args) * args_size); | |
791 | } | |
792 | ||
793 | /* Initialize the next argument. */ | |
794 | arg = &args[args_len++]; | |
795 | set_token (arg, src->text, src->text); | |
796 | ||
797 | /* Gather the argument's tokens. */ | |
798 | depth = 0; | |
799 | for (;;) | |
800 | { | |
801 | char *start = src->text; | |
802 | ||
803 | if (! get_token (&tok, src)) | |
8a3fe4f8 | 804 | error (_("Malformed argument list for macro `%s'."), name); |
ec2bcbe7 JB |
805 | |
806 | /* Is tok an opening paren? */ | |
807 | if (tok.len == 1 && tok.text[0] == '(') | |
808 | depth++; | |
809 | ||
810 | /* Is tok is a closing paren? */ | |
811 | else if (tok.len == 1 && tok.text[0] == ')') | |
812 | { | |
813 | /* If it's a closing paren at the top level, then that's | |
814 | the end of the argument list. */ | |
815 | if (depth == 0) | |
816 | { | |
2fae03e8 TT |
817 | /* In the varargs case, the last argument may be |
818 | missing. Add an empty argument in this case. */ | |
819 | if (nargs != -1 && args_len == nargs - 1) | |
820 | { | |
821 | /* Make sure we have room for the argument. */ | |
822 | if (args_len >= args_size) | |
823 | { | |
824 | args_size++; | |
825 | args = xrealloc (args, sizeof (*args) * args_size); | |
826 | } | |
827 | arg = &args[args_len++]; | |
828 | set_token (arg, src->text, src->text); | |
829 | } | |
830 | ||
ec2bcbe7 JB |
831 | discard_cleanups (back_to); |
832 | *argc_p = args_len; | |
833 | return args; | |
834 | } | |
835 | ||
836 | depth--; | |
837 | } | |
838 | ||
839 | /* If tok is a comma at top level, then that's the end of | |
2fae03e8 TT |
840 | the current argument. However, if we are handling a |
841 | variadic macro and we are computing the last argument, we | |
842 | want to include the comma and remaining tokens. */ | |
843 | else if (tok.len == 1 && tok.text[0] == ',' && depth == 0 | |
844 | && (nargs == -1 || args_len < nargs)) | |
ec2bcbe7 JB |
845 | break; |
846 | ||
847 | /* Extend the current argument to enclose this token. If | |
848 | this is the current argument's first token, leave out any | |
849 | leading whitespace, just for aesthetics. */ | |
850 | if (arg->len == 0) | |
851 | { | |
852 | arg->text = tok.text; | |
853 | arg->len = tok.len; | |
854 | arg->last_token = 0; | |
855 | } | |
856 | else | |
857 | { | |
858 | arg->len = (tok.text + tok.len) - arg->text; | |
859 | arg->last_token = tok.text - arg->text; | |
860 | } | |
861 | } | |
862 | } | |
863 | } | |
864 | ||
865 | ||
866 | /* The `expand' and `substitute_args' functions both invoke `scan' | |
867 | recursively, so we need a forward declaration somewhere. */ | |
868 | static void scan (struct macro_buffer *dest, | |
869 | struct macro_buffer *src, | |
870 | struct macro_name_list *no_loop, | |
871 | macro_lookup_ftype *lookup_func, | |
872 | void *lookup_baton); | |
873 | ||
874 | ||
2fae03e8 TT |
875 | /* A helper function for substitute_args. |
876 | ||
877 | ARGV is a vector of all the arguments; ARGC is the number of | |
878 | arguments. IS_VARARGS is true if the macro being substituted is a | |
879 | varargs macro; in this case VA_ARG_NAME is the name of the | |
880 | "variable" argument. VA_ARG_NAME is ignored if IS_VARARGS is | |
881 | false. | |
882 | ||
883 | If the token TOK is the name of a parameter, return the parameter's | |
884 | index. If TOK is not an argument, return -1. */ | |
885 | ||
886 | static int | |
887 | find_parameter (const struct macro_buffer *tok, | |
888 | int is_varargs, const struct macro_buffer *va_arg_name, | |
889 | int argc, const char * const *argv) | |
890 | { | |
891 | int i; | |
892 | ||
893 | if (! tok->is_identifier) | |
894 | return -1; | |
895 | ||
896 | for (i = 0; i < argc; ++i) | |
897 | if (tok->len == strlen (argv[i]) && ! memcmp (tok->text, argv[i], tok->len)) | |
898 | return i; | |
899 | ||
900 | if (is_varargs && tok->len == va_arg_name->len | |
901 | && ! memcmp (tok->text, va_arg_name->text, tok->len)) | |
902 | return argc - 1; | |
903 | ||
904 | return -1; | |
905 | } | |
906 | ||
ec2bcbe7 JB |
907 | /* Given the macro definition DEF, being invoked with the actual |
908 | arguments given by ARGC and ARGV, substitute the arguments into the | |
909 | replacement list, and store the result in DEST. | |
910 | ||
2fae03e8 TT |
911 | IS_VARARGS should be true if DEF is a varargs macro. In this case, |
912 | VA_ARG_NAME should be the name of the "variable" argument -- either | |
913 | __VA_ARGS__ for c99-style varargs, or the final argument name, for | |
914 | GNU-style varargs. If IS_VARARGS is false, this parameter is | |
915 | ignored. | |
916 | ||
ec2bcbe7 JB |
917 | If it is necessary to expand macro invocations in one of the |
918 | arguments, use LOOKUP_FUNC and LOOKUP_BATON to find the macro | |
919 | definitions, and don't expand invocations of the macros listed in | |
920 | NO_LOOP. */ | |
2fae03e8 | 921 | |
ec2bcbe7 JB |
922 | static void |
923 | substitute_args (struct macro_buffer *dest, | |
924 | struct macro_definition *def, | |
2fae03e8 | 925 | int is_varargs, const struct macro_buffer *va_arg_name, |
ec2bcbe7 JB |
926 | int argc, struct macro_buffer *argv, |
927 | struct macro_name_list *no_loop, | |
928 | macro_lookup_ftype *lookup_func, | |
929 | void *lookup_baton) | |
930 | { | |
931 | /* A macro buffer for the macro's replacement list. */ | |
932 | struct macro_buffer replacement_list; | |
2fae03e8 TT |
933 | /* The token we are currently considering. */ |
934 | struct macro_buffer tok; | |
935 | /* The replacement list's pointer from just before TOK was lexed. */ | |
936 | char *original_rl_start; | |
937 | /* We have a single lookahead token to handle token splicing. */ | |
938 | struct macro_buffer lookahead; | |
939 | /* The lookahead token might not be valid. */ | |
940 | int lookahead_valid; | |
941 | /* The replacement list's pointer from just before LOOKAHEAD was | |
942 | lexed. */ | |
943 | char *lookahead_rl_start; | |
ec2bcbe7 JB |
944 | |
945 | init_shared_buffer (&replacement_list, (char *) def->replacement, | |
946 | strlen (def->replacement)); | |
947 | ||
948 | gdb_assert (dest->len == 0); | |
949 | dest->last_token = 0; | |
950 | ||
2fae03e8 TT |
951 | original_rl_start = replacement_list.text; |
952 | if (! get_token (&tok, &replacement_list)) | |
953 | return; | |
954 | lookahead_rl_start = replacement_list.text; | |
955 | lookahead_valid = get_token (&lookahead, &replacement_list); | |
956 | ||
ec2bcbe7 JB |
957 | for (;;) |
958 | { | |
ec2bcbe7 JB |
959 | /* Just for aesthetics. If we skipped some whitespace, copy |
960 | that to DEST. */ | |
961 | if (tok.text > original_rl_start) | |
962 | { | |
963 | appendmem (dest, original_rl_start, tok.text - original_rl_start); | |
964 | dest->last_token = dest->len; | |
965 | } | |
966 | ||
967 | /* Is this token the stringification operator? */ | |
968 | if (tok.len == 1 | |
969 | && tok.text[0] == '#') | |
2fae03e8 TT |
970 | { |
971 | int arg; | |
ec2bcbe7 | 972 | |
2fae03e8 TT |
973 | if (!lookahead_valid) |
974 | error (_("Stringification operator requires an argument.")); | |
ec2bcbe7 | 975 | |
2fae03e8 TT |
976 | arg = find_parameter (&lookahead, is_varargs, va_arg_name, |
977 | def->argc, def->argv); | |
978 | if (arg == -1) | |
979 | error (_("Argument to stringification operator must name " | |
980 | "a macro parameter.")); | |
ec2bcbe7 | 981 | |
2fae03e8 TT |
982 | stringify (dest, argv[arg].text, argv[arg].len); |
983 | ||
984 | /* Read one token and let the loop iteration code handle the | |
985 | rest. */ | |
986 | lookahead_rl_start = replacement_list.text; | |
987 | lookahead_valid = get_token (&lookahead, &replacement_list); | |
988 | } | |
989 | /* Is this token the splicing operator? */ | |
990 | else if (tok.len == 2 | |
991 | && tok.text[0] == '#' | |
992 | && tok.text[1] == '#') | |
993 | error (_("Stray splicing operator")); | |
994 | /* Is the next token the splicing operator? */ | |
995 | else if (lookahead_valid | |
996 | && lookahead.len == 2 | |
997 | && lookahead.text[0] == '#' | |
998 | && lookahead.text[1] == '#') | |
999 | { | |
1000 | int arg, finished = 0; | |
1001 | int prev_was_comma = 0; | |
1002 | ||
1003 | /* Note that GCC warns if the result of splicing is not a | |
1004 | token. In the debugger there doesn't seem to be much | |
1005 | benefit from doing this. */ | |
1006 | ||
1007 | /* Insert the first token. */ | |
1008 | if (tok.len == 1 && tok.text[0] == ',') | |
1009 | prev_was_comma = 1; | |
1010 | else | |
1011 | { | |
1012 | int arg = find_parameter (&tok, is_varargs, va_arg_name, | |
1013 | def->argc, def->argv); | |
1014 | if (arg != -1) | |
1015 | appendmem (dest, argv[arg].text, argv[arg].len); | |
1016 | else | |
1017 | appendmem (dest, tok.text, tok.len); | |
1018 | } | |
1019 | ||
1020 | /* Apply a possible sequence of ## operators. */ | |
1021 | for (;;) | |
1022 | { | |
1023 | if (! get_token (&tok, &replacement_list)) | |
1024 | error (_("Splicing operator at end of macro")); | |
1025 | ||
1026 | /* Handle a comma before a ##. If we are handling | |
1027 | varargs, and the token on the right hand side is the | |
1028 | varargs marker, and the final argument is empty or | |
1029 | missing, then drop the comma. This is a GNU | |
1030 | extension. There is one ambiguous case here, | |
1031 | involving pedantic behavior with an empty argument, | |
1032 | but we settle that in favor of GNU-style (GCC uses an | |
1033 | option). If we aren't dealing with varargs, we | |
1034 | simply insert the comma. */ | |
1035 | if (prev_was_comma) | |
1036 | { | |
1037 | if (! (is_varargs | |
1038 | && tok.len == va_arg_name->len | |
1039 | && !memcmp (tok.text, va_arg_name->text, tok.len) | |
1040 | && argv[argc - 1].len == 0)) | |
1041 | appendmem (dest, ",", 1); | |
1042 | prev_was_comma = 0; | |
1043 | } | |
1044 | ||
1045 | /* Insert the token. If it is a parameter, insert the | |
1046 | argument. If it is a comma, treat it specially. */ | |
1047 | if (tok.len == 1 && tok.text[0] == ',') | |
1048 | prev_was_comma = 1; | |
1049 | else | |
1050 | { | |
1051 | int arg = find_parameter (&tok, is_varargs, va_arg_name, | |
1052 | def->argc, def->argv); | |
1053 | if (arg != -1) | |
1054 | appendmem (dest, argv[arg].text, argv[arg].len); | |
1055 | else | |
1056 | appendmem (dest, tok.text, tok.len); | |
1057 | } | |
1058 | ||
1059 | /* Now read another token. If it is another splice, we | |
1060 | loop. */ | |
1061 | original_rl_start = replacement_list.text; | |
1062 | if (! get_token (&tok, &replacement_list)) | |
1063 | { | |
1064 | finished = 1; | |
1065 | break; | |
1066 | } | |
1067 | ||
1068 | if (! (tok.len == 2 | |
1069 | && tok.text[0] == '#' | |
1070 | && tok.text[1] == '#')) | |
1071 | break; | |
1072 | } | |
1073 | ||
1074 | if (prev_was_comma) | |
1075 | { | |
1076 | /* We saw a comma. Insert it now. */ | |
1077 | appendmem (dest, ",", 1); | |
1078 | } | |
1079 | ||
1080 | dest->last_token = dest->len; | |
1081 | if (finished) | |
1082 | lookahead_valid = 0; | |
1083 | else | |
1084 | { | |
1085 | /* Set up for the loop iterator. */ | |
1086 | lookahead = tok; | |
1087 | lookahead_rl_start = original_rl_start; | |
1088 | lookahead_valid = 1; | |
1089 | } | |
1090 | } | |
1091 | else | |
1092 | { | |
1093 | /* Is this token an identifier? */ | |
1094 | int substituted = 0; | |
1095 | int arg = find_parameter (&tok, is_varargs, va_arg_name, | |
1096 | def->argc, def->argv); | |
1097 | ||
1098 | if (arg != -1) | |
1099 | { | |
1100 | struct macro_buffer arg_src; | |
1101 | ||
1102 | /* Expand any macro invocations in the argument text, | |
1103 | and append the result to dest. Remember that scan | |
1104 | mutates its source, so we need to scan a new buffer | |
1105 | referring to the argument's text, not the argument | |
1106 | itself. */ | |
1107 | init_shared_buffer (&arg_src, argv[arg].text, argv[arg].len); | |
1108 | scan (dest, &arg_src, no_loop, lookup_func, lookup_baton); | |
1109 | substituted = 1; | |
1110 | } | |
1111 | ||
1112 | /* If it wasn't a parameter, then just copy it across. */ | |
1113 | if (! substituted) | |
1114 | append_tokens_without_splicing (dest, &tok); | |
1115 | } | |
1116 | ||
1117 | if (! lookahead_valid) | |
1118 | break; | |
1119 | ||
1120 | tok = lookahead; | |
1121 | original_rl_start = lookahead_rl_start; | |
1122 | ||
1123 | lookahead_rl_start = replacement_list.text; | |
1124 | lookahead_valid = get_token (&lookahead, &replacement_list); | |
ec2bcbe7 JB |
1125 | } |
1126 | } | |
1127 | ||
1128 | ||
1129 | /* Expand a call to a macro named ID, whose definition is DEF. Append | |
1130 | its expansion to DEST. SRC is the input text following the ID | |
1131 | token. We are currently rescanning the expansions of the macros | |
1132 | named in NO_LOOP; don't re-expand them. Use LOOKUP_FUNC and | |
1133 | LOOKUP_BATON to find definitions for any nested macro references. | |
1134 | ||
1135 | Return 1 if we decided to expand it, zero otherwise. (If it's a | |
1136 | function-like macro name that isn't followed by an argument list, | |
1137 | we don't expand it.) If we return zero, leave SRC unchanged. */ | |
1138 | static int | |
1139 | expand (const char *id, | |
1140 | struct macro_definition *def, | |
1141 | struct macro_buffer *dest, | |
1142 | struct macro_buffer *src, | |
1143 | struct macro_name_list *no_loop, | |
1144 | macro_lookup_ftype *lookup_func, | |
1145 | void *lookup_baton) | |
1146 | { | |
1147 | struct macro_name_list new_no_loop; | |
1148 | ||
1149 | /* Create a new node to be added to the front of the no-expand list. | |
1150 | This list is appropriate for re-scanning replacement lists, but | |
1151 | it is *not* appropriate for scanning macro arguments; invocations | |
1152 | of the macro whose arguments we are gathering *do* get expanded | |
1153 | there. */ | |
1154 | new_no_loop.name = id; | |
1155 | new_no_loop.next = no_loop; | |
1156 | ||
1157 | /* What kind of macro are we expanding? */ | |
1158 | if (def->kind == macro_object_like) | |
1159 | { | |
1160 | struct macro_buffer replacement_list; | |
1161 | ||
1162 | init_shared_buffer (&replacement_list, (char *) def->replacement, | |
1163 | strlen (def->replacement)); | |
1164 | ||
1165 | scan (dest, &replacement_list, &new_no_loop, lookup_func, lookup_baton); | |
1166 | return 1; | |
1167 | } | |
1168 | else if (def->kind == macro_function_like) | |
1169 | { | |
1170 | struct cleanup *back_to = make_cleanup (null_cleanup, 0); | |
0a029df5 | 1171 | int argc = 0; |
a86bc61c | 1172 | struct macro_buffer *argv = NULL; |
ec2bcbe7 JB |
1173 | struct macro_buffer substituted; |
1174 | struct macro_buffer substituted_src; | |
2fae03e8 TT |
1175 | struct macro_buffer va_arg_name; |
1176 | int is_varargs = 0; | |
1177 | ||
1178 | if (def->argc >= 1) | |
1179 | { | |
1180 | if (strcmp (def->argv[def->argc - 1], "...") == 0) | |
1181 | { | |
1182 | /* In C99-style varargs, substitution is done using | |
1183 | __VA_ARGS__. */ | |
1184 | init_shared_buffer (&va_arg_name, "__VA_ARGS__", | |
1185 | strlen ("__VA_ARGS__")); | |
1186 | is_varargs = 1; | |
1187 | } | |
1188 | else | |
1189 | { | |
1190 | int len = strlen (def->argv[def->argc - 1]); | |
1191 | if (len > 3 | |
1192 | && strcmp (def->argv[def->argc - 1] + len - 3, "...") == 0) | |
1193 | { | |
1194 | /* In GNU-style varargs, the name of the | |
1195 | substitution parameter is the name of the formal | |
1196 | argument without the "...". */ | |
1197 | init_shared_buffer (&va_arg_name, | |
1198 | (char *) def->argv[def->argc - 1], | |
1199 | len - 3); | |
1200 | is_varargs = 1; | |
1201 | } | |
1202 | } | |
1203 | } | |
ec2bcbe7 JB |
1204 | |
1205 | make_cleanup (free_current_contents, &argv); | |
2fae03e8 TT |
1206 | argv = gather_arguments (id, src, is_varargs ? def->argc : -1, |
1207 | &argc); | |
ec2bcbe7 JB |
1208 | |
1209 | /* If we couldn't find any argument list, then we don't expand | |
1210 | this macro. */ | |
1211 | if (! argv) | |
1212 | { | |
1213 | do_cleanups (back_to); | |
1214 | return 0; | |
1215 | } | |
1216 | ||
1217 | /* Check that we're passing an acceptable number of arguments for | |
1218 | this macro. */ | |
1219 | if (argc != def->argc) | |
1220 | { | |
2fae03e8 TT |
1221 | if (is_varargs && argc >= def->argc - 1) |
1222 | { | |
1223 | /* Ok. */ | |
1224 | } | |
ec2bcbe7 JB |
1225 | /* Remember that a sequence of tokens like "foo()" is a |
1226 | valid invocation of a macro expecting either zero or one | |
1227 | arguments. */ | |
2fae03e8 TT |
1228 | else if (! (argc == 1 |
1229 | && argv[0].len == 0 | |
1230 | && def->argc == 0)) | |
8a3fe4f8 AC |
1231 | error (_("Wrong number of arguments to macro `%s' " |
1232 | "(expected %d, got %d)."), | |
ec2bcbe7 JB |
1233 | id, def->argc, argc); |
1234 | } | |
1235 | ||
1236 | /* Note that we don't expand macro invocations in the arguments | |
1237 | yet --- we let subst_args take care of that. Parameters that | |
1238 | appear as operands of the stringifying operator "#" or the | |
1239 | splicing operator "##" don't get macro references expanded, | |
1240 | so we can't really tell whether it's appropriate to macro- | |
1241 | expand an argument until we see how it's being used. */ | |
1242 | init_buffer (&substituted, 0); | |
1243 | make_cleanup (cleanup_macro_buffer, &substituted); | |
2fae03e8 TT |
1244 | substitute_args (&substituted, def, is_varargs, &va_arg_name, |
1245 | argc, argv, no_loop, lookup_func, lookup_baton); | |
ec2bcbe7 JB |
1246 | |
1247 | /* Now `substituted' is the macro's replacement list, with all | |
1248 | argument values substituted into it properly. Re-scan it for | |
1249 | macro references, but don't expand invocations of this macro. | |
1250 | ||
1251 | We create a new buffer, `substituted_src', which points into | |
1252 | `substituted', and scan that. We can't scan `substituted' | |
1253 | itself, since the tokenization process moves the buffer's | |
1254 | text pointer around, and we still need to be able to find | |
1255 | `substituted's original text buffer after scanning it so we | |
1256 | can free it. */ | |
1257 | init_shared_buffer (&substituted_src, substituted.text, substituted.len); | |
1258 | scan (dest, &substituted_src, &new_no_loop, lookup_func, lookup_baton); | |
1259 | ||
1260 | do_cleanups (back_to); | |
1261 | ||
1262 | return 1; | |
1263 | } | |
1264 | else | |
e2e0b3e5 | 1265 | internal_error (__FILE__, __LINE__, _("bad macro definition kind")); |
ec2bcbe7 JB |
1266 | } |
1267 | ||
1268 | ||
1269 | /* If the single token in SRC_FIRST followed by the tokens in SRC_REST | |
1270 | constitute a macro invokation not forbidden in NO_LOOP, append its | |
1271 | expansion to DEST and return non-zero. Otherwise, return zero, and | |
1272 | leave DEST unchanged. | |
1273 | ||
1274 | SRC_FIRST and SRC_REST must be shared buffers; DEST must not be one. | |
1275 | SRC_FIRST must be a string built by get_token. */ | |
1276 | static int | |
1277 | maybe_expand (struct macro_buffer *dest, | |
1278 | struct macro_buffer *src_first, | |
1279 | struct macro_buffer *src_rest, | |
1280 | struct macro_name_list *no_loop, | |
1281 | macro_lookup_ftype *lookup_func, | |
1282 | void *lookup_baton) | |
1283 | { | |
1284 | gdb_assert (src_first->shared); | |
1285 | gdb_assert (src_rest->shared); | |
1286 | gdb_assert (! dest->shared); | |
1287 | ||
1288 | /* Is this token an identifier? */ | |
1289 | if (src_first->is_identifier) | |
1290 | { | |
1291 | /* Make a null-terminated copy of it, since that's what our | |
1292 | lookup function expects. */ | |
1293 | char *id = xmalloc (src_first->len + 1); | |
1294 | struct cleanup *back_to = make_cleanup (xfree, id); | |
1295 | memcpy (id, src_first->text, src_first->len); | |
1296 | id[src_first->len] = 0; | |
1297 | ||
1298 | /* If we're currently re-scanning the result of expanding | |
1299 | this macro, don't expand it again. */ | |
1300 | if (! currently_rescanning (no_loop, id)) | |
1301 | { | |
1302 | /* Does this identifier have a macro definition in scope? */ | |
1303 | struct macro_definition *def = lookup_func (id, lookup_baton); | |
1304 | ||
1305 | if (def && expand (id, def, dest, src_rest, no_loop, | |
1306 | lookup_func, lookup_baton)) | |
1307 | { | |
1308 | do_cleanups (back_to); | |
1309 | return 1; | |
1310 | } | |
1311 | } | |
1312 | ||
1313 | do_cleanups (back_to); | |
1314 | } | |
1315 | ||
1316 | return 0; | |
1317 | } | |
1318 | ||
1319 | ||
1320 | /* Expand macro references in SRC, appending the results to DEST. | |
1321 | Assume we are re-scanning the result of expanding the macros named | |
1322 | in NO_LOOP, and don't try to re-expand references to them. | |
1323 | ||
1324 | SRC must be a shared buffer; DEST must not be one. */ | |
1325 | static void | |
1326 | scan (struct macro_buffer *dest, | |
1327 | struct macro_buffer *src, | |
1328 | struct macro_name_list *no_loop, | |
1329 | macro_lookup_ftype *lookup_func, | |
1330 | void *lookup_baton) | |
1331 | { | |
1332 | gdb_assert (src->shared); | |
1333 | gdb_assert (! dest->shared); | |
1334 | ||
1335 | for (;;) | |
1336 | { | |
1337 | struct macro_buffer tok; | |
1338 | char *original_src_start = src->text; | |
1339 | ||
1340 | /* Find the next token in SRC. */ | |
1341 | if (! get_token (&tok, src)) | |
1342 | break; | |
1343 | ||
1344 | /* Just for aesthetics. If we skipped some whitespace, copy | |
1345 | that to DEST. */ | |
1346 | if (tok.text > original_src_start) | |
1347 | { | |
1348 | appendmem (dest, original_src_start, tok.text - original_src_start); | |
1349 | dest->last_token = dest->len; | |
1350 | } | |
1351 | ||
1352 | if (! maybe_expand (dest, &tok, src, no_loop, lookup_func, lookup_baton)) | |
1353 | /* We didn't end up expanding tok as a macro reference, so | |
1354 | simply append it to dest. */ | |
1355 | append_tokens_without_splicing (dest, &tok); | |
1356 | } | |
1357 | ||
1358 | /* Just for aesthetics. If there was any trailing whitespace in | |
1359 | src, copy it to dest. */ | |
1360 | if (src->len) | |
1361 | { | |
1362 | appendmem (dest, src->text, src->len); | |
1363 | dest->last_token = dest->len; | |
1364 | } | |
1365 | } | |
1366 | ||
1367 | ||
1368 | char * | |
1369 | macro_expand (const char *source, | |
1370 | macro_lookup_ftype *lookup_func, | |
1371 | void *lookup_func_baton) | |
1372 | { | |
1373 | struct macro_buffer src, dest; | |
1374 | struct cleanup *back_to; | |
1375 | ||
1376 | init_shared_buffer (&src, (char *) source, strlen (source)); | |
1377 | ||
1378 | init_buffer (&dest, 0); | |
1379 | dest.last_token = 0; | |
1380 | back_to = make_cleanup (cleanup_macro_buffer, &dest); | |
1381 | ||
1382 | scan (&dest, &src, 0, lookup_func, lookup_func_baton); | |
1383 | ||
1384 | appendc (&dest, '\0'); | |
1385 | ||
1386 | discard_cleanups (back_to); | |
1387 | return dest.text; | |
1388 | } | |
1389 | ||
1390 | ||
1391 | char * | |
1392 | macro_expand_once (const char *source, | |
1393 | macro_lookup_ftype *lookup_func, | |
1394 | void *lookup_func_baton) | |
1395 | { | |
8a3fe4f8 | 1396 | error (_("Expand-once not implemented yet.")); |
ec2bcbe7 JB |
1397 | } |
1398 | ||
1399 | ||
1400 | char * | |
1401 | macro_expand_next (char **lexptr, | |
1402 | macro_lookup_ftype *lookup_func, | |
1403 | void *lookup_baton) | |
1404 | { | |
1405 | struct macro_buffer src, dest, tok; | |
1406 | struct cleanup *back_to; | |
1407 | ||
1408 | /* Set up SRC to refer to the input text, pointed to by *lexptr. */ | |
1409 | init_shared_buffer (&src, *lexptr, strlen (*lexptr)); | |
1410 | ||
1411 | /* Set up DEST to receive the expansion, if there is one. */ | |
1412 | init_buffer (&dest, 0); | |
1413 | dest.last_token = 0; | |
1414 | back_to = make_cleanup (cleanup_macro_buffer, &dest); | |
1415 | ||
1416 | /* Get the text's first preprocessing token. */ | |
1417 | if (! get_token (&tok, &src)) | |
1418 | { | |
1419 | do_cleanups (back_to); | |
1420 | return 0; | |
1421 | } | |
1422 | ||
1423 | /* If it's a macro invocation, expand it. */ | |
1424 | if (maybe_expand (&dest, &tok, &src, 0, lookup_func, lookup_baton)) | |
1425 | { | |
1426 | /* It was a macro invocation! Package up the expansion as a | |
1427 | null-terminated string and return it. Set *lexptr to the | |
1428 | start of the next token in the input. */ | |
1429 | appendc (&dest, '\0'); | |
1430 | discard_cleanups (back_to); | |
1431 | *lexptr = src.text; | |
1432 | return dest.text; | |
1433 | } | |
1434 | else | |
1435 | { | |
1436 | /* It wasn't a macro invocation. */ | |
1437 | do_cleanups (back_to); | |
1438 | return 0; | |
1439 | } | |
1440 | } |