Commit | Line | Data |
---|---|---|
ec2bcbe7 | 1 | /* C preprocessor macro expansion for GDB. |
e2882c85 | 2 | Copyright (C) 2002-2018 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" | |
6c7a06a3 | 25 | #include "c-lang.h" |
ec2bcbe7 JB |
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 | |
b38ef47f PA |
56 | some other block, or to a string literal, and we shouldn't |
57 | reallocate it. */ | |
58 | bool shared; | |
ec2bcbe7 JB |
59 | |
60 | /* For detecting token splicing. | |
61 | ||
62 | This is the index in TEXT of the first character of the token | |
63 | that abuts the end of TEXT. If TEXT contains no tokens, then we | |
64 | set this equal to LEN. If TEXT ends in whitespace, then there is | |
65 | no token abutting the end of TEXT (it's just whitespace), and | |
66 | again, we set this equal to LEN. We set this to -1 if we don't | |
67 | know the nature of TEXT. */ | |
1739cf24 | 68 | int last_token = -1; |
ec2bcbe7 JB |
69 | |
70 | /* If this buffer is holding the result from get_token, then this | |
71 | is non-zero if it is an identifier token, zero otherwise. */ | |
1739cf24 | 72 | int is_identifier = 0; |
ec2bcbe7 JB |
73 | |
74 | ||
1739cf24 TT |
75 | macro_buffer () |
76 | : text (NULL), | |
77 | len (0), | |
78 | size (0), | |
79 | shared (false) | |
80 | { | |
81 | } | |
ec2bcbe7 | 82 | |
1739cf24 TT |
83 | /* Set the macro buffer to the empty string, guessing that its |
84 | final contents will fit in N bytes. (It'll get resized if it | |
85 | doesn't, so the guess doesn't have to be right.) Allocate the | |
86 | initial storage with xmalloc. */ | |
87 | explicit macro_buffer (int n) | |
88 | : len (0), | |
89 | size (n), | |
90 | shared (false) | |
91 | { | |
92 | if (n > 0) | |
93 | text = (char *) xmalloc (n); | |
94 | else | |
95 | text = NULL; | |
96 | } | |
abc9d0dc | 97 | |
1739cf24 TT |
98 | /* Set the macro buffer to refer to the LEN bytes at ADDR, as a |
99 | shared substring. */ | |
100 | macro_buffer (const char *addr, int len) | |
101 | { | |
102 | set_shared (addr, len); | |
103 | } | |
ec2bcbe7 | 104 | |
1739cf24 TT |
105 | /* Set the macro buffer to refer to the LEN bytes at ADDR, as a |
106 | shared substring. */ | |
107 | void set_shared (const char *addr, int len_) | |
108 | { | |
109 | text = (char *) addr; | |
110 | len = len_; | |
111 | size = 0; | |
112 | shared = true; | |
113 | } | |
ec2bcbe7 | 114 | |
1739cf24 TT |
115 | ~macro_buffer () |
116 | { | |
117 | if (! shared && size) | |
118 | xfree (text); | |
119 | } | |
ec2bcbe7 | 120 | |
1739cf24 TT |
121 | /* Release the text of the buffer to the caller, which is now |
122 | responsible for freeing it. */ | |
123 | char *release () | |
124 | { | |
125 | gdb_assert (! shared); | |
126 | gdb_assert (size); | |
127 | char *result = text; | |
128 | text = NULL; | |
129 | return result; | |
130 | } | |
ec2bcbe7 | 131 | |
1739cf24 TT |
132 | /* Resize the buffer to be at least N bytes long. Raise an error if |
133 | the buffer shouldn't be resized. */ | |
134 | void resize_buffer (int n) | |
135 | { | |
136 | /* We shouldn't be trying to resize shared strings. */ | |
137 | gdb_assert (! shared); | |
ec2bcbe7 | 138 | |
1739cf24 TT |
139 | if (size == 0) |
140 | size = n; | |
141 | else | |
142 | while (size <= n) | |
143 | size *= 2; | |
ec2bcbe7 | 144 | |
1739cf24 TT |
145 | text = (char *) xrealloc (text, size); |
146 | } | |
ec2bcbe7 | 147 | |
1739cf24 TT |
148 | /* Append the character C to the buffer. */ |
149 | void appendc (int c) | |
150 | { | |
151 | int new_len = len + 1; | |
ec2bcbe7 | 152 | |
1739cf24 TT |
153 | if (new_len > size) |
154 | resize_buffer (new_len); | |
ec2bcbe7 | 155 | |
1739cf24 TT |
156 | text[len] = c; |
157 | len = new_len; | |
158 | } | |
ec2bcbe7 | 159 | |
1739cf24 TT |
160 | /* Append the COUNT bytes at ADDR to the buffer. */ |
161 | void appendmem (const char *addr, int count) | |
162 | { | |
163 | int new_len = len + count; | |
ec2bcbe7 | 164 | |
1739cf24 TT |
165 | if (new_len > size) |
166 | resize_buffer (new_len); | |
ec2bcbe7 | 167 | |
1739cf24 TT |
168 | memcpy (text + len, addr, count); |
169 | len = new_len; | |
170 | } | |
171 | }; | |
ec2bcbe7 JB |
172 | |
173 | ||
174 | \f | |
175 | /* Recognizing preprocessor tokens. */ | |
176 | ||
177 | ||
d7d9f01e TT |
178 | int |
179 | macro_is_whitespace (int c) | |
ec2bcbe7 JB |
180 | { |
181 | return (c == ' ' | |
182 | || c == '\t' | |
183 | || c == '\n' | |
184 | || c == '\v' | |
185 | || c == '\f'); | |
186 | } | |
187 | ||
188 | ||
d7d9f01e TT |
189 | int |
190 | macro_is_digit (int c) | |
ec2bcbe7 JB |
191 | { |
192 | return ('0' <= c && c <= '9'); | |
193 | } | |
194 | ||
195 | ||
d7d9f01e TT |
196 | int |
197 | macro_is_identifier_nondigit (int c) | |
ec2bcbe7 JB |
198 | { |
199 | return (c == '_' | |
200 | || ('a' <= c && c <= 'z') | |
201 | || ('A' <= c && c <= 'Z')); | |
202 | } | |
203 | ||
204 | ||
205 | static void | |
206 | set_token (struct macro_buffer *tok, char *start, char *end) | |
207 | { | |
1739cf24 | 208 | tok->set_shared (start, end - start); |
ec2bcbe7 JB |
209 | tok->last_token = 0; |
210 | ||
025bb325 | 211 | /* Presumed; get_identifier may overwrite this. */ |
ec2bcbe7 JB |
212 | tok->is_identifier = 0; |
213 | } | |
214 | ||
215 | ||
216 | static int | |
217 | get_comment (struct macro_buffer *tok, char *p, char *end) | |
218 | { | |
219 | if (p + 2 > end) | |
220 | return 0; | |
221 | else if (p[0] == '/' | |
222 | && p[1] == '*') | |
223 | { | |
224 | char *tok_start = p; | |
225 | ||
226 | p += 2; | |
227 | ||
228 | for (; p < end; p++) | |
229 | if (p + 2 <= end | |
230 | && p[0] == '*' | |
231 | && p[1] == '/') | |
232 | { | |
233 | p += 2; | |
234 | set_token (tok, tok_start, p); | |
235 | return 1; | |
236 | } | |
237 | ||
8a3fe4f8 | 238 | error (_("Unterminated comment in macro expansion.")); |
ec2bcbe7 JB |
239 | } |
240 | else if (p[0] == '/' | |
241 | && p[1] == '/') | |
242 | { | |
243 | char *tok_start = p; | |
244 | ||
245 | p += 2; | |
246 | for (; p < end; p++) | |
247 | if (*p == '\n') | |
248 | break; | |
249 | ||
250 | set_token (tok, tok_start, p); | |
251 | return 1; | |
252 | } | |
253 | else | |
254 | return 0; | |
255 | } | |
256 | ||
257 | ||
258 | static int | |
259 | get_identifier (struct macro_buffer *tok, char *p, char *end) | |
260 | { | |
261 | if (p < end | |
d7d9f01e | 262 | && macro_is_identifier_nondigit (*p)) |
ec2bcbe7 JB |
263 | { |
264 | char *tok_start = p; | |
265 | ||
266 | while (p < end | |
d7d9f01e TT |
267 | && (macro_is_identifier_nondigit (*p) |
268 | || macro_is_digit (*p))) | |
ec2bcbe7 JB |
269 | p++; |
270 | ||
271 | set_token (tok, tok_start, p); | |
272 | tok->is_identifier = 1; | |
273 | return 1; | |
274 | } | |
275 | else | |
276 | return 0; | |
277 | } | |
278 | ||
279 | ||
280 | static int | |
281 | get_pp_number (struct macro_buffer *tok, char *p, char *end) | |
282 | { | |
283 | if (p < end | |
d7d9f01e | 284 | && (macro_is_digit (*p) |
17c8aaf5 TT |
285 | || (*p == '.' |
286 | && p + 2 <= end | |
287 | && macro_is_digit (p[1])))) | |
ec2bcbe7 JB |
288 | { |
289 | char *tok_start = p; | |
290 | ||
291 | while (p < end) | |
292 | { | |
17c8aaf5 TT |
293 | if (p + 2 <= end |
294 | && strchr ("eEpP", *p) | |
295 | && (p[1] == '+' || p[1] == '-')) | |
ec2bcbe7 | 296 | p += 2; |
17c8aaf5 TT |
297 | else if (macro_is_digit (*p) |
298 | || macro_is_identifier_nondigit (*p) | |
299 | || *p == '.') | |
300 | p++; | |
ec2bcbe7 JB |
301 | else |
302 | break; | |
303 | } | |
304 | ||
305 | set_token (tok, tok_start, p); | |
306 | return 1; | |
307 | } | |
308 | else | |
309 | return 0; | |
310 | } | |
311 | ||
312 | ||
313 | ||
314 | /* If the text starting at P going up to (but not including) END | |
315 | starts with a character constant, set *TOK to point to that | |
316 | character constant, and return 1. Otherwise, return zero. | |
317 | Signal an error if it contains a malformed or incomplete character | |
318 | constant. */ | |
319 | static int | |
320 | get_character_constant (struct macro_buffer *tok, char *p, char *end) | |
321 | { | |
322 | /* ISO/IEC 9899:1999 (E) Section 6.4.4.4 paragraph 1 | |
323 | But of course, what really matters is that we handle it the same | |
324 | way GDB's C/C++ lexer does. So we call parse_escape in utils.c | |
325 | to handle escape sequences. */ | |
326 | if ((p + 1 <= end && *p == '\'') | |
6c7a06a3 TT |
327 | || (p + 2 <= end |
328 | && (p[0] == 'L' || p[0] == 'u' || p[0] == 'U') | |
329 | && p[1] == '\'')) | |
ec2bcbe7 JB |
330 | { |
331 | char *tok_start = p; | |
6c7a06a3 | 332 | int char_count = 0; |
ec2bcbe7 JB |
333 | |
334 | if (*p == '\'') | |
335 | p++; | |
6c7a06a3 | 336 | else if (*p == 'L' || *p == 'u' || *p == 'U') |
ec2bcbe7 JB |
337 | p += 2; |
338 | else | |
f3574227 | 339 | gdb_assert_not_reached ("unexpected character constant"); |
ec2bcbe7 | 340 | |
ec2bcbe7 JB |
341 | for (;;) |
342 | { | |
343 | if (p >= end) | |
8a3fe4f8 | 344 | error (_("Unmatched single quote.")); |
ec2bcbe7 JB |
345 | else if (*p == '\'') |
346 | { | |
6c7a06a3 | 347 | if (!char_count) |
8a3fe4f8 AC |
348 | error (_("A character constant must contain at least one " |
349 | "character.")); | |
ec2bcbe7 JB |
350 | p++; |
351 | break; | |
352 | } | |
353 | else if (*p == '\\') | |
354 | { | |
d7561cbb KS |
355 | const char *s, *o; |
356 | ||
357 | s = o = ++p; | |
358 | char_count += c_parse_escape (&s, NULL); | |
359 | p += s - o; | |
ec2bcbe7 JB |
360 | } |
361 | else | |
6c7a06a3 TT |
362 | { |
363 | p++; | |
364 | char_count++; | |
365 | } | |
ec2bcbe7 JB |
366 | } |
367 | ||
368 | set_token (tok, tok_start, p); | |
369 | return 1; | |
370 | } | |
371 | else | |
372 | return 0; | |
373 | } | |
374 | ||
375 | ||
376 | /* If the text starting at P going up to (but not including) END | |
377 | starts with a string literal, set *TOK to point to that string | |
378 | literal, and return 1. Otherwise, return zero. Signal an error if | |
379 | it contains a malformed or incomplete string literal. */ | |
380 | static int | |
381 | get_string_literal (struct macro_buffer *tok, char *p, char *end) | |
382 | { | |
383 | if ((p + 1 <= end | |
6c7a06a3 | 384 | && *p == '"') |
ec2bcbe7 | 385 | || (p + 2 <= end |
6c7a06a3 TT |
386 | && (p[0] == 'L' || p[0] == 'u' || p[0] == 'U') |
387 | && p[1] == '"')) | |
ec2bcbe7 JB |
388 | { |
389 | char *tok_start = p; | |
390 | ||
6c7a06a3 | 391 | if (*p == '"') |
ec2bcbe7 | 392 | p++; |
6c7a06a3 | 393 | else if (*p == 'L' || *p == 'u' || *p == 'U') |
ec2bcbe7 JB |
394 | p += 2; |
395 | else | |
f3574227 | 396 | gdb_assert_not_reached ("unexpected string literal"); |
ec2bcbe7 JB |
397 | |
398 | for (;;) | |
399 | { | |
400 | if (p >= end) | |
8a3fe4f8 | 401 | error (_("Unterminated string in expression.")); |
6c7a06a3 | 402 | else if (*p == '"') |
ec2bcbe7 JB |
403 | { |
404 | p++; | |
405 | break; | |
406 | } | |
407 | else if (*p == '\n') | |
8a3fe4f8 AC |
408 | error (_("Newline characters may not appear in string " |
409 | "constants.")); | |
ec2bcbe7 JB |
410 | else if (*p == '\\') |
411 | { | |
d7561cbb KS |
412 | const char *s, *o; |
413 | ||
414 | s = o = ++p; | |
415 | c_parse_escape (&s, NULL); | |
416 | p += s - o; | |
ec2bcbe7 JB |
417 | } |
418 | else | |
419 | p++; | |
420 | } | |
421 | ||
422 | set_token (tok, tok_start, p); | |
423 | return 1; | |
424 | } | |
425 | else | |
426 | return 0; | |
427 | } | |
428 | ||
429 | ||
430 | static int | |
431 | get_punctuator (struct macro_buffer *tok, char *p, char *end) | |
432 | { | |
433 | /* Here, speed is much less important than correctness and clarity. */ | |
434 | ||
ccb3ac8a TT |
435 | /* ISO/IEC 9899:1999 (E) Section 6.4.6 Paragraph 1. |
436 | Note that this table is ordered in a special way. A punctuator | |
437 | which is a prefix of another punctuator must appear after its | |
438 | "extension". Otherwise, the wrong token will be returned. */ | |
ec2bcbe7 | 439 | static const char * const punctuators[] = { |
ccb3ac8a TT |
440 | "[", "]", "(", ")", "{", "}", "?", ";", ",", "~", |
441 | "...", ".", | |
442 | "->", "--", "-=", "-", | |
443 | "++", "+=", "+", | |
444 | "*=", "*", | |
445 | "!=", "!", | |
446 | "&&", "&=", "&", | |
447 | "/=", "/", | |
448 | "%>", "%:%:", "%:", "%=", "%", | |
449 | "^=", "^", | |
450 | "##", "#", | |
451 | ":>", ":", | |
452 | "||", "|=", "|", | |
453 | "<<=", "<<", "<=", "<:", "<%", "<", | |
454 | ">>=", ">>", ">=", ">", | |
455 | "==", "=", | |
ec2bcbe7 JB |
456 | 0 |
457 | }; | |
458 | ||
459 | int i; | |
460 | ||
461 | if (p + 1 <= end) | |
462 | { | |
463 | for (i = 0; punctuators[i]; i++) | |
464 | { | |
465 | const char *punctuator = punctuators[i]; | |
466 | ||
467 | if (p[0] == punctuator[0]) | |
468 | { | |
469 | int len = strlen (punctuator); | |
470 | ||
471 | if (p + len <= end | |
472 | && ! memcmp (p, punctuator, len)) | |
473 | { | |
474 | set_token (tok, p, p + len); | |
475 | return 1; | |
476 | } | |
477 | } | |
478 | } | |
479 | } | |
480 | ||
481 | return 0; | |
482 | } | |
483 | ||
484 | ||
485 | /* Peel the next preprocessor token off of SRC, and put it in TOK. | |
486 | Mutate TOK to refer to the first token in SRC, and mutate SRC to | |
487 | refer to the text after that token. SRC must be a shared buffer; | |
488 | the resulting TOK will be shared, pointing into the same string SRC | |
489 | does. Initialize TOK's last_token field. Return non-zero if we | |
490 | succeed, or 0 if we didn't find any more tokens in SRC. */ | |
491 | static int | |
492 | get_token (struct macro_buffer *tok, | |
493 | struct macro_buffer *src) | |
494 | { | |
495 | char *p = src->text; | |
496 | char *end = p + src->len; | |
497 | ||
498 | gdb_assert (src->shared); | |
499 | ||
500 | /* From the ISO C standard, ISO/IEC 9899:1999 (E), section 6.4: | |
501 | ||
502 | preprocessing-token: | |
503 | header-name | |
504 | identifier | |
505 | pp-number | |
506 | character-constant | |
507 | string-literal | |
508 | punctuator | |
509 | each non-white-space character that cannot be one of the above | |
510 | ||
511 | We don't have to deal with header-name tokens, since those can | |
512 | only occur after a #include, which we will never see. */ | |
513 | ||
514 | while (p < end) | |
d7d9f01e | 515 | if (macro_is_whitespace (*p)) |
ec2bcbe7 JB |
516 | p++; |
517 | else if (get_comment (tok, p, end)) | |
518 | p += tok->len; | |
519 | else if (get_pp_number (tok, p, end) | |
520 | || get_character_constant (tok, p, end) | |
521 | || get_string_literal (tok, p, end) | |
522 | /* Note: the grammar in the standard seems to be | |
523 | ambiguous: L'x' can be either a wide character | |
524 | constant, or an identifier followed by a normal | |
525 | character constant. By trying `get_identifier' after | |
526 | we try get_character_constant and get_string_literal, | |
527 | we give the wide character syntax precedence. Now, | |
528 | since GDB doesn't handle wide character constants | |
529 | anyway, is this the right thing to do? */ | |
530 | || get_identifier (tok, p, end) | |
531 | || get_punctuator (tok, p, end)) | |
532 | { | |
533 | /* How many characters did we consume, including whitespace? */ | |
534 | int consumed = p - src->text + tok->len; | |
b8d56208 | 535 | |
ec2bcbe7 JB |
536 | src->text += consumed; |
537 | src->len -= consumed; | |
538 | return 1; | |
539 | } | |
540 | else | |
541 | { | |
542 | /* We have found a "non-whitespace character that cannot be | |
543 | one of the above." Make a token out of it. */ | |
544 | int consumed; | |
545 | ||
546 | set_token (tok, p, p + 1); | |
547 | consumed = p - src->text + tok->len; | |
548 | src->text += consumed; | |
549 | src->len -= consumed; | |
550 | return 1; | |
551 | } | |
552 | ||
553 | return 0; | |
554 | } | |
555 | ||
556 | ||
557 | \f | |
558 | /* Appending token strings, with and without splicing */ | |
559 | ||
560 | ||
561 | /* Append the macro buffer SRC to the end of DEST, and ensure that | |
562 | doing so doesn't splice the token at the end of SRC with the token | |
563 | at the beginning of DEST. SRC and DEST must have their last_token | |
564 | fields set. Upon return, DEST's last_token field is set correctly. | |
565 | ||
566 | For example: | |
567 | ||
568 | If DEST is "(" and SRC is "y", then we can return with | |
569 | DEST set to "(y" --- we've simply appended the two buffers. | |
570 | ||
571 | However, if DEST is "x" and SRC is "y", then we must not return | |
572 | with DEST set to "xy" --- that would splice the two tokens "x" and | |
573 | "y" together to make a single token "xy". However, it would be | |
574 | fine to return with DEST set to "x y". Similarly, "<" and "<" must | |
575 | yield "< <", not "<<", etc. */ | |
576 | static void | |
577 | append_tokens_without_splicing (struct macro_buffer *dest, | |
578 | struct macro_buffer *src) | |
579 | { | |
580 | int original_dest_len = dest->len; | |
581 | struct macro_buffer dest_tail, new_token; | |
582 | ||
583 | gdb_assert (src->last_token != -1); | |
584 | gdb_assert (dest->last_token != -1); | |
585 | ||
586 | /* First, just try appending the two, and call get_token to see if | |
587 | we got a splice. */ | |
1739cf24 | 588 | dest->appendmem (src->text, src->len); |
ec2bcbe7 JB |
589 | |
590 | /* If DEST originally had no token abutting its end, then we can't | |
591 | have spliced anything, so we're done. */ | |
592 | if (dest->last_token == original_dest_len) | |
593 | { | |
594 | dest->last_token = original_dest_len + src->last_token; | |
595 | return; | |
596 | } | |
597 | ||
598 | /* Set DEST_TAIL to point to the last token in DEST, followed by | |
599 | all the stuff we just appended. */ | |
1739cf24 TT |
600 | dest_tail.set_shared (dest->text + dest->last_token, |
601 | dest->len - dest->last_token); | |
ec2bcbe7 JB |
602 | |
603 | /* Re-parse DEST's last token. We know that DEST used to contain | |
604 | at least one token, so if it doesn't contain any after the | |
605 | append, then we must have spliced "/" and "*" or "/" and "/" to | |
606 | make a comment start. (Just for the record, I got this right | |
607 | the first time. This is not a bug fix.) */ | |
608 | if (get_token (&new_token, &dest_tail) | |
609 | && (new_token.text + new_token.len | |
610 | == dest->text + original_dest_len)) | |
611 | { | |
612 | /* No splice, so we're done. */ | |
613 | dest->last_token = original_dest_len + src->last_token; | |
614 | return; | |
615 | } | |
616 | ||
617 | /* Okay, a simple append caused a splice. Let's chop dest back to | |
618 | its original length and try again, but separate the texts with a | |
619 | space. */ | |
620 | dest->len = original_dest_len; | |
1739cf24 TT |
621 | dest->appendc (' '); |
622 | dest->appendmem (src->text, src->len); | |
ec2bcbe7 | 623 | |
1739cf24 TT |
624 | dest_tail.set_shared (dest->text + dest->last_token, |
625 | dest->len - dest->last_token); | |
ec2bcbe7 JB |
626 | |
627 | /* Try to re-parse DEST's last token, as above. */ | |
628 | if (get_token (&new_token, &dest_tail) | |
629 | && (new_token.text + new_token.len | |
630 | == dest->text + original_dest_len)) | |
631 | { | |
632 | /* No splice, so we're done. */ | |
633 | dest->last_token = original_dest_len + 1 + src->last_token; | |
634 | return; | |
635 | } | |
636 | ||
637 | /* As far as I know, there's no case where inserting a space isn't | |
638 | enough to prevent a splice. */ | |
639 | internal_error (__FILE__, __LINE__, | |
e2e0b3e5 | 640 | _("unable to avoid splicing tokens during macro expansion")); |
ec2bcbe7 JB |
641 | } |
642 | ||
2fae03e8 TT |
643 | /* Stringify an argument, and insert it into DEST. ARG is the text to |
644 | stringify; it is LEN bytes long. */ | |
645 | ||
646 | static void | |
abc9d0dc | 647 | stringify (struct macro_buffer *dest, const char *arg, int len) |
2fae03e8 TT |
648 | { |
649 | /* Trim initial whitespace from ARG. */ | |
650 | while (len > 0 && macro_is_whitespace (*arg)) | |
651 | { | |
652 | ++arg; | |
653 | --len; | |
654 | } | |
655 | ||
656 | /* Trim trailing whitespace from ARG. */ | |
657 | while (len > 0 && macro_is_whitespace (arg[len - 1])) | |
658 | --len; | |
659 | ||
660 | /* Insert the string. */ | |
1739cf24 | 661 | dest->appendc ('"'); |
2fae03e8 TT |
662 | while (len > 0) |
663 | { | |
664 | /* We could try to handle strange cases here, like control | |
665 | characters, but there doesn't seem to be much point. */ | |
666 | if (macro_is_whitespace (*arg)) | |
667 | { | |
668 | /* Replace a sequence of whitespace with a single space. */ | |
1739cf24 | 669 | dest->appendc (' '); |
2fae03e8 TT |
670 | while (len > 1 && macro_is_whitespace (arg[1])) |
671 | { | |
672 | ++arg; | |
673 | --len; | |
674 | } | |
675 | } | |
676 | else if (*arg == '\\' || *arg == '"') | |
677 | { | |
1739cf24 TT |
678 | dest->appendc ('\\'); |
679 | dest->appendc (*arg); | |
2fae03e8 TT |
680 | } |
681 | else | |
1739cf24 | 682 | dest->appendc (*arg); |
2fae03e8 TT |
683 | ++arg; |
684 | --len; | |
685 | } | |
1739cf24 | 686 | dest->appendc ('"'); |
2fae03e8 TT |
687 | dest->last_token = dest->len; |
688 | } | |
ec2bcbe7 | 689 | |
abc9d0dc TT |
690 | /* See macroexp.h. */ |
691 | ||
692 | char * | |
693 | macro_stringify (const char *str) | |
694 | { | |
abc9d0dc | 695 | int len = strlen (str); |
1739cf24 | 696 | struct macro_buffer buffer (len); |
abc9d0dc | 697 | |
abc9d0dc | 698 | stringify (&buffer, str, len); |
1739cf24 | 699 | buffer.appendc ('\0'); |
abc9d0dc | 700 | |
1739cf24 | 701 | return buffer.release (); |
abc9d0dc TT |
702 | } |
703 | ||
ec2bcbe7 JB |
704 | \f |
705 | /* Expanding macros! */ | |
706 | ||
707 | ||
708 | /* A singly-linked list of the names of the macros we are currently | |
709 | expanding --- for detecting expansion loops. */ | |
710 | struct macro_name_list { | |
711 | const char *name; | |
712 | struct macro_name_list *next; | |
713 | }; | |
714 | ||
715 | ||
716 | /* Return non-zero if we are currently expanding the macro named NAME, | |
717 | according to LIST; otherwise, return zero. | |
718 | ||
719 | You know, it would be possible to get rid of all the NO_LOOP | |
720 | arguments to these functions by simply generating a new lookup | |
721 | function and baton which refuses to find the definition for a | |
722 | particular macro, and otherwise delegates the decision to another | |
723 | function/baton pair. But that makes the linked list of excluded | |
724 | macros chained through untyped baton pointers, which will make it | |
025bb325 | 725 | harder to debug. :( */ |
ec2bcbe7 JB |
726 | static int |
727 | currently_rescanning (struct macro_name_list *list, const char *name) | |
728 | { | |
729 | for (; list; list = list->next) | |
a86bc61c | 730 | if (strcmp (name, list->name) == 0) |
ec2bcbe7 JB |
731 | return 1; |
732 | ||
733 | return 0; | |
734 | } | |
735 | ||
736 | ||
737 | /* Gather the arguments to a macro expansion. | |
738 | ||
739 | NAME is the name of the macro being invoked. (It's only used for | |
740 | printing error messages.) | |
741 | ||
742 | Assume that SRC is the text of the macro invocation immediately | |
743 | following the macro name. For example, if we're processing the | |
744 | text foo(bar, baz), then NAME would be foo and SRC will be (bar, | |
745 | baz). | |
746 | ||
747 | If SRC doesn't start with an open paren ( token at all, return | |
1739cf24 | 748 | false, leave SRC unchanged, and don't set *ARGS_PTR to anything. |
ec2bcbe7 JB |
749 | |
750 | If SRC doesn't contain a properly terminated argument list, then | |
751 | raise an error. | |
1739cf24 | 752 | |
2fae03e8 TT |
753 | For a variadic macro, NARGS holds the number of formal arguments to |
754 | the macro. For a GNU-style variadic macro, this should be the | |
755 | number of named arguments. For a non-variadic macro, NARGS should | |
756 | be -1. | |
ec2bcbe7 | 757 | |
1739cf24 TT |
758 | Otherwise, return true and set *ARGS_PTR to a vector of macro |
759 | buffers referring to the argument texts. The macro buffers share | |
760 | their text with SRC, and their last_token fields are initialized. | |
ec2bcbe7 JB |
761 | |
762 | NOTE WELL: if SRC starts with a open paren ( token followed | |
763 | immediately by a close paren ) token (e.g., the invocation looks | |
764 | like "foo()"), we treat that as one argument, which happens to be | |
765 | the empty list of tokens. The caller should keep in mind that such | |
766 | a sequence of tokens is a valid way to invoke one-parameter | |
767 | function-like macros, but also a valid way to invoke zero-parameter | |
768 | function-like macros. Eeew. | |
769 | ||
770 | Consume the tokens from SRC; after this call, SRC contains the text | |
771 | following the invocation. */ | |
772 | ||
1739cf24 TT |
773 | static bool |
774 | gather_arguments (const char *name, struct macro_buffer *src, int nargs, | |
775 | std::vector<struct macro_buffer> *args_ptr) | |
ec2bcbe7 JB |
776 | { |
777 | struct macro_buffer tok; | |
1739cf24 | 778 | std::vector<struct macro_buffer> args; |
ec2bcbe7 JB |
779 | |
780 | /* Does SRC start with an opening paren token? Read from a copy of | |
781 | SRC, so SRC itself is unaffected if we don't find an opening | |
782 | paren. */ | |
783 | { | |
1739cf24 | 784 | struct macro_buffer temp (src->text, src->len); |
ec2bcbe7 JB |
785 | |
786 | if (! get_token (&tok, &temp) | |
787 | || tok.len != 1 | |
788 | || tok.text[0] != '(') | |
1739cf24 | 789 | return false; |
ec2bcbe7 JB |
790 | } |
791 | ||
792 | /* Consume SRC's opening paren. */ | |
793 | get_token (&tok, src); | |
794 | ||
ec2bcbe7 JB |
795 | for (;;) |
796 | { | |
797 | struct macro_buffer *arg; | |
798 | int depth; | |
799 | ||
ec2bcbe7 | 800 | /* Initialize the next argument. */ |
1739cf24 TT |
801 | args.emplace_back (); |
802 | arg = &args.back (); | |
ec2bcbe7 JB |
803 | set_token (arg, src->text, src->text); |
804 | ||
805 | /* Gather the argument's tokens. */ | |
806 | depth = 0; | |
807 | for (;;) | |
808 | { | |
ec2bcbe7 | 809 | if (! get_token (&tok, src)) |
8a3fe4f8 | 810 | error (_("Malformed argument list for macro `%s'."), name); |
1739cf24 | 811 | |
ec2bcbe7 JB |
812 | /* Is tok an opening paren? */ |
813 | if (tok.len == 1 && tok.text[0] == '(') | |
814 | depth++; | |
815 | ||
816 | /* Is tok is a closing paren? */ | |
817 | else if (tok.len == 1 && tok.text[0] == ')') | |
818 | { | |
819 | /* If it's a closing paren at the top level, then that's | |
820 | the end of the argument list. */ | |
821 | if (depth == 0) | |
822 | { | |
2fae03e8 TT |
823 | /* In the varargs case, the last argument may be |
824 | missing. Add an empty argument in this case. */ | |
1739cf24 | 825 | if (nargs != -1 && args.size () == nargs - 1) |
2fae03e8 | 826 | { |
1739cf24 TT |
827 | args.emplace_back (); |
828 | arg = &args.back (); | |
2fae03e8 TT |
829 | set_token (arg, src->text, src->text); |
830 | } | |
831 | ||
1739cf24 TT |
832 | *args_ptr = std::move (args); |
833 | return true; | |
ec2bcbe7 JB |
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 | |
1739cf24 | 844 | && (nargs == -1 || args.size () < 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) | |
3e43a32a MS |
897 | if (tok->len == strlen (argv[i]) |
898 | && !memcmp (tok->text, argv[i], tok->len)) | |
2fae03e8 TT |
899 | return i; |
900 | ||
901 | if (is_varargs && tok->len == va_arg_name->len | |
902 | && ! memcmp (tok->text, va_arg_name->text, tok->len)) | |
903 | return argc - 1; | |
904 | ||
905 | return -1; | |
906 | } | |
907 | ||
a9bbfbd8 TT |
908 | /* Helper function for substitute_args that gets the next token and |
909 | updates the passed-in state variables. */ | |
910 | ||
911 | static void | |
912 | get_next_token_for_substitution (struct macro_buffer *replacement_list, | |
913 | struct macro_buffer *token, | |
914 | char **start, | |
915 | struct macro_buffer *lookahead, | |
916 | char **lookahead_start, | |
917 | int *lookahead_valid, | |
918 | bool *keep_going) | |
919 | { | |
920 | if (!*lookahead_valid) | |
921 | *keep_going = false; | |
922 | else | |
923 | { | |
924 | *keep_going = true; | |
925 | *token = *lookahead; | |
926 | *start = *lookahead_start; | |
927 | *lookahead_start = replacement_list->text; | |
928 | *lookahead_valid = get_token (lookahead, replacement_list); | |
929 | } | |
930 | } | |
931 | ||
ec2bcbe7 | 932 | /* Given the macro definition DEF, being invoked with the actual |
1739cf24 | 933 | arguments given by ARGV, substitute the arguments into the |
ec2bcbe7 JB |
934 | replacement list, and store the result in DEST. |
935 | ||
2fae03e8 TT |
936 | IS_VARARGS should be true if DEF is a varargs macro. In this case, |
937 | VA_ARG_NAME should be the name of the "variable" argument -- either | |
938 | __VA_ARGS__ for c99-style varargs, or the final argument name, for | |
939 | GNU-style varargs. If IS_VARARGS is false, this parameter is | |
940 | ignored. | |
941 | ||
ec2bcbe7 JB |
942 | If it is necessary to expand macro invocations in one of the |
943 | arguments, use LOOKUP_FUNC and LOOKUP_BATON to find the macro | |
944 | definitions, and don't expand invocations of the macros listed in | |
945 | NO_LOOP. */ | |
2fae03e8 | 946 | |
ec2bcbe7 | 947 | static void |
1739cf24 | 948 | substitute_args (struct macro_buffer *dest, |
ec2bcbe7 | 949 | struct macro_definition *def, |
2fae03e8 | 950 | int is_varargs, const struct macro_buffer *va_arg_name, |
1739cf24 | 951 | const std::vector<struct macro_buffer> &argv, |
ec2bcbe7 JB |
952 | struct macro_name_list *no_loop, |
953 | macro_lookup_ftype *lookup_func, | |
954 | void *lookup_baton) | |
955 | { | |
2fae03e8 TT |
956 | /* The token we are currently considering. */ |
957 | struct macro_buffer tok; | |
958 | /* The replacement list's pointer from just before TOK was lexed. */ | |
959 | char *original_rl_start; | |
960 | /* We have a single lookahead token to handle token splicing. */ | |
961 | struct macro_buffer lookahead; | |
962 | /* The lookahead token might not be valid. */ | |
963 | int lookahead_valid; | |
964 | /* The replacement list's pointer from just before LOOKAHEAD was | |
965 | lexed. */ | |
966 | char *lookahead_rl_start; | |
ec2bcbe7 | 967 | |
1739cf24 TT |
968 | /* A macro buffer for the macro's replacement list. */ |
969 | struct macro_buffer replacement_list (def->replacement, | |
970 | strlen (def->replacement)); | |
ec2bcbe7 JB |
971 | |
972 | gdb_assert (dest->len == 0); | |
973 | dest->last_token = 0; | |
974 | ||
2fae03e8 TT |
975 | original_rl_start = replacement_list.text; |
976 | if (! get_token (&tok, &replacement_list)) | |
977 | return; | |
978 | lookahead_rl_start = replacement_list.text; | |
979 | lookahead_valid = get_token (&lookahead, &replacement_list); | |
980 | ||
a9bbfbd8 TT |
981 | /* __VA_OPT__ state variable. The states are: |
982 | 0 - nothing happening | |
983 | 1 - saw __VA_OPT__ | |
984 | >= 2 in __VA_OPT__, the value encodes the parenthesis depth. */ | |
985 | unsigned vaopt_state = 0; | |
986 | ||
987 | for (bool keep_going = true; | |
988 | keep_going; | |
989 | get_next_token_for_substitution (&replacement_list, | |
990 | &tok, | |
991 | &original_rl_start, | |
992 | &lookahead, | |
993 | &lookahead_rl_start, | |
994 | &lookahead_valid, | |
995 | &keep_going)) | |
ec2bcbe7 | 996 | { |
a9bbfbd8 TT |
997 | bool token_is_vaopt = (tok.len == 10 |
998 | && strncmp (tok.text, "__VA_OPT__", 10) == 0); | |
999 | ||
1000 | if (vaopt_state > 0) | |
1001 | { | |
1002 | if (token_is_vaopt) | |
1003 | error (_("__VA_OPT__ cannot appear inside __VA_OPT__")); | |
1004 | else if (tok.len == 1 && tok.text[0] == '(') | |
1005 | { | |
1006 | ++vaopt_state; | |
1007 | /* We just entered __VA_OPT__, so don't emit this | |
1008 | token. */ | |
1009 | continue; | |
1010 | } | |
1011 | else if (vaopt_state == 1) | |
1012 | error (_("__VA_OPT__ must be followed by an open parenthesis")); | |
1013 | else if (tok.len == 1 && tok.text[0] == ')') | |
1014 | { | |
1015 | --vaopt_state; | |
1016 | if (vaopt_state == 1) | |
1017 | { | |
1018 | /* Done with __VA_OPT__. */ | |
1019 | vaopt_state = 0; | |
1020 | /* Don't emit. */ | |
1021 | continue; | |
1022 | } | |
1023 | } | |
1024 | ||
1025 | /* If __VA_ARGS__ is empty, then drop the contents of | |
1026 | __VA_OPT__. */ | |
1739cf24 | 1027 | if (argv.back ().len == 0) |
a9bbfbd8 TT |
1028 | continue; |
1029 | } | |
1030 | else if (token_is_vaopt) | |
1031 | { | |
1032 | if (!is_varargs) | |
1033 | error (_("__VA_OPT__ is only valid in a variadic macro")); | |
1034 | vaopt_state = 1; | |
1035 | /* Don't emit this token. */ | |
1036 | continue; | |
1037 | } | |
1038 | ||
ec2bcbe7 JB |
1039 | /* Just for aesthetics. If we skipped some whitespace, copy |
1040 | that to DEST. */ | |
1041 | if (tok.text > original_rl_start) | |
1042 | { | |
1739cf24 | 1043 | dest->appendmem (original_rl_start, tok.text - original_rl_start); |
ec2bcbe7 JB |
1044 | dest->last_token = dest->len; |
1045 | } | |
1046 | ||
1047 | /* Is this token the stringification operator? */ | |
1048 | if (tok.len == 1 | |
1049 | && tok.text[0] == '#') | |
2fae03e8 TT |
1050 | { |
1051 | int arg; | |
ec2bcbe7 | 1052 | |
2fae03e8 TT |
1053 | if (!lookahead_valid) |
1054 | error (_("Stringification operator requires an argument.")); | |
ec2bcbe7 | 1055 | |
2fae03e8 TT |
1056 | arg = find_parameter (&lookahead, is_varargs, va_arg_name, |
1057 | def->argc, def->argv); | |
1058 | if (arg == -1) | |
1059 | error (_("Argument to stringification operator must name " | |
1060 | "a macro parameter.")); | |
ec2bcbe7 | 1061 | |
2fae03e8 TT |
1062 | stringify (dest, argv[arg].text, argv[arg].len); |
1063 | ||
1064 | /* Read one token and let the loop iteration code handle the | |
1065 | rest. */ | |
1066 | lookahead_rl_start = replacement_list.text; | |
1067 | lookahead_valid = get_token (&lookahead, &replacement_list); | |
1068 | } | |
1069 | /* Is this token the splicing operator? */ | |
1070 | else if (tok.len == 2 | |
1071 | && tok.text[0] == '#' | |
1072 | && tok.text[1] == '#') | |
1073 | error (_("Stray splicing operator")); | |
1074 | /* Is the next token the splicing operator? */ | |
1075 | else if (lookahead_valid | |
1076 | && lookahead.len == 2 | |
1077 | && lookahead.text[0] == '#' | |
1078 | && lookahead.text[1] == '#') | |
1079 | { | |
308d96ed | 1080 | int finished = 0; |
2fae03e8 TT |
1081 | int prev_was_comma = 0; |
1082 | ||
1083 | /* Note that GCC warns if the result of splicing is not a | |
1084 | token. In the debugger there doesn't seem to be much | |
1085 | benefit from doing this. */ | |
1086 | ||
1087 | /* Insert the first token. */ | |
1088 | if (tok.len == 1 && tok.text[0] == ',') | |
1089 | prev_was_comma = 1; | |
1090 | else | |
1091 | { | |
1092 | int arg = find_parameter (&tok, is_varargs, va_arg_name, | |
1093 | def->argc, def->argv); | |
b8d56208 | 1094 | |
2fae03e8 | 1095 | if (arg != -1) |
1739cf24 | 1096 | dest->appendmem (argv[arg].text, argv[arg].len); |
2fae03e8 | 1097 | else |
1739cf24 | 1098 | dest->appendmem (tok.text, tok.len); |
2fae03e8 TT |
1099 | } |
1100 | ||
1101 | /* Apply a possible sequence of ## operators. */ | |
1102 | for (;;) | |
1103 | { | |
1104 | if (! get_token (&tok, &replacement_list)) | |
1105 | error (_("Splicing operator at end of macro")); | |
1106 | ||
1107 | /* Handle a comma before a ##. If we are handling | |
1108 | varargs, and the token on the right hand side is the | |
1109 | varargs marker, and the final argument is empty or | |
1110 | missing, then drop the comma. This is a GNU | |
1111 | extension. There is one ambiguous case here, | |
1112 | involving pedantic behavior with an empty argument, | |
1113 | but we settle that in favor of GNU-style (GCC uses an | |
1114 | option). If we aren't dealing with varargs, we | |
1115 | simply insert the comma. */ | |
1116 | if (prev_was_comma) | |
1117 | { | |
1118 | if (! (is_varargs | |
1119 | && tok.len == va_arg_name->len | |
1120 | && !memcmp (tok.text, va_arg_name->text, tok.len) | |
1739cf24 TT |
1121 | && argv.back ().len == 0)) |
1122 | dest->appendmem (",", 1); | |
2fae03e8 TT |
1123 | prev_was_comma = 0; |
1124 | } | |
1125 | ||
1126 | /* Insert the token. If it is a parameter, insert the | |
1127 | argument. If it is a comma, treat it specially. */ | |
1128 | if (tok.len == 1 && tok.text[0] == ',') | |
1129 | prev_was_comma = 1; | |
1130 | else | |
1131 | { | |
1132 | int arg = find_parameter (&tok, is_varargs, va_arg_name, | |
1133 | def->argc, def->argv); | |
b8d56208 | 1134 | |
2fae03e8 | 1135 | if (arg != -1) |
1739cf24 | 1136 | dest->appendmem (argv[arg].text, argv[arg].len); |
2fae03e8 | 1137 | else |
1739cf24 | 1138 | dest->appendmem (tok.text, tok.len); |
2fae03e8 TT |
1139 | } |
1140 | ||
1141 | /* Now read another token. If it is another splice, we | |
1142 | loop. */ | |
1143 | original_rl_start = replacement_list.text; | |
1144 | if (! get_token (&tok, &replacement_list)) | |
1145 | { | |
1146 | finished = 1; | |
1147 | break; | |
1148 | } | |
1149 | ||
1150 | if (! (tok.len == 2 | |
1151 | && tok.text[0] == '#' | |
1152 | && tok.text[1] == '#')) | |
1153 | break; | |
1154 | } | |
1155 | ||
1156 | if (prev_was_comma) | |
1157 | { | |
1158 | /* We saw a comma. Insert it now. */ | |
1739cf24 | 1159 | dest->appendmem (",", 1); |
2fae03e8 TT |
1160 | } |
1161 | ||
1162 | dest->last_token = dest->len; | |
1163 | if (finished) | |
1164 | lookahead_valid = 0; | |
1165 | else | |
1166 | { | |
1167 | /* Set up for the loop iterator. */ | |
1168 | lookahead = tok; | |
1169 | lookahead_rl_start = original_rl_start; | |
1170 | lookahead_valid = 1; | |
1171 | } | |
1172 | } | |
1173 | else | |
1174 | { | |
1175 | /* Is this token an identifier? */ | |
1176 | int substituted = 0; | |
1177 | int arg = find_parameter (&tok, is_varargs, va_arg_name, | |
1178 | def->argc, def->argv); | |
1179 | ||
1180 | if (arg != -1) | |
1181 | { | |
2fae03e8 TT |
1182 | /* Expand any macro invocations in the argument text, |
1183 | and append the result to dest. Remember that scan | |
1184 | mutates its source, so we need to scan a new buffer | |
1185 | referring to the argument's text, not the argument | |
1186 | itself. */ | |
1739cf24 | 1187 | struct macro_buffer arg_src (argv[arg].text, argv[arg].len); |
2fae03e8 TT |
1188 | scan (dest, &arg_src, no_loop, lookup_func, lookup_baton); |
1189 | substituted = 1; | |
1190 | } | |
1191 | ||
1192 | /* If it wasn't a parameter, then just copy it across. */ | |
1193 | if (! substituted) | |
1194 | append_tokens_without_splicing (dest, &tok); | |
1195 | } | |
ec2bcbe7 | 1196 | } |
a9bbfbd8 TT |
1197 | |
1198 | if (vaopt_state > 0) | |
1199 | error (_("Unterminated __VA_OPT__")); | |
ec2bcbe7 JB |
1200 | } |
1201 | ||
1202 | ||
1203 | /* Expand a call to a macro named ID, whose definition is DEF. Append | |
1204 | its expansion to DEST. SRC is the input text following the ID | |
1205 | token. We are currently rescanning the expansions of the macros | |
1206 | named in NO_LOOP; don't re-expand them. Use LOOKUP_FUNC and | |
025bb325 | 1207 | LOOKUP_BATON to find definitions for any nested macro references. |
ec2bcbe7 JB |
1208 | |
1209 | Return 1 if we decided to expand it, zero otherwise. (If it's a | |
1210 | function-like macro name that isn't followed by an argument list, | |
1211 | we don't expand it.) If we return zero, leave SRC unchanged. */ | |
1212 | static int | |
1213 | expand (const char *id, | |
1214 | struct macro_definition *def, | |
1215 | struct macro_buffer *dest, | |
1216 | struct macro_buffer *src, | |
1217 | struct macro_name_list *no_loop, | |
1218 | macro_lookup_ftype *lookup_func, | |
1219 | void *lookup_baton) | |
1220 | { | |
1221 | struct macro_name_list new_no_loop; | |
1222 | ||
1223 | /* Create a new node to be added to the front of the no-expand list. | |
1224 | This list is appropriate for re-scanning replacement lists, but | |
1225 | it is *not* appropriate for scanning macro arguments; invocations | |
1226 | of the macro whose arguments we are gathering *do* get expanded | |
1227 | there. */ | |
1228 | new_no_loop.name = id; | |
1229 | new_no_loop.next = no_loop; | |
1230 | ||
1231 | /* What kind of macro are we expanding? */ | |
1232 | if (def->kind == macro_object_like) | |
1233 | { | |
1739cf24 TT |
1234 | struct macro_buffer replacement_list (def->replacement, |
1235 | strlen (def->replacement)); | |
ec2bcbe7 JB |
1236 | |
1237 | scan (dest, &replacement_list, &new_no_loop, lookup_func, lookup_baton); | |
1238 | return 1; | |
1239 | } | |
1240 | else if (def->kind == macro_function_like) | |
1241 | { | |
1739cf24 | 1242 | struct macro_buffer va_arg_name; |
2fae03e8 TT |
1243 | int is_varargs = 0; |
1244 | ||
1245 | if (def->argc >= 1) | |
1246 | { | |
1247 | if (strcmp (def->argv[def->argc - 1], "...") == 0) | |
1248 | { | |
1249 | /* In C99-style varargs, substitution is done using | |
1250 | __VA_ARGS__. */ | |
1739cf24 | 1251 | va_arg_name.set_shared ("__VA_ARGS__", strlen ("__VA_ARGS__")); |
2fae03e8 TT |
1252 | is_varargs = 1; |
1253 | } | |
1254 | else | |
1255 | { | |
1256 | int len = strlen (def->argv[def->argc - 1]); | |
b8d56208 | 1257 | |
2fae03e8 TT |
1258 | if (len > 3 |
1259 | && strcmp (def->argv[def->argc - 1] + len - 3, "...") == 0) | |
1260 | { | |
1261 | /* In GNU-style varargs, the name of the | |
1262 | substitution parameter is the name of the formal | |
1263 | argument without the "...". */ | |
1739cf24 | 1264 | va_arg_name.set_shared (def->argv[def->argc - 1], len - 3); |
2fae03e8 TT |
1265 | is_varargs = 1; |
1266 | } | |
1267 | } | |
1268 | } | |
ec2bcbe7 | 1269 | |
1739cf24 | 1270 | std::vector<struct macro_buffer> argv; |
ec2bcbe7 JB |
1271 | /* If we couldn't find any argument list, then we don't expand |
1272 | this macro. */ | |
1739cf24 TT |
1273 | if (!gather_arguments (id, src, is_varargs ? def->argc : -1, |
1274 | &argv)) | |
1275 | return 0; | |
ec2bcbe7 JB |
1276 | |
1277 | /* Check that we're passing an acceptable number of arguments for | |
1278 | this macro. */ | |
1739cf24 | 1279 | if (argv.size () != def->argc) |
ec2bcbe7 | 1280 | { |
1739cf24 | 1281 | if (is_varargs && argv.size () >= def->argc - 1) |
2fae03e8 TT |
1282 | { |
1283 | /* Ok. */ | |
1284 | } | |
ec2bcbe7 JB |
1285 | /* Remember that a sequence of tokens like "foo()" is a |
1286 | valid invocation of a macro expecting either zero or one | |
1287 | arguments. */ | |
1739cf24 | 1288 | else if (! (argv.size () == 1 |
2fae03e8 TT |
1289 | && argv[0].len == 0 |
1290 | && def->argc == 0)) | |
8a3fe4f8 AC |
1291 | error (_("Wrong number of arguments to macro `%s' " |
1292 | "(expected %d, got %d)."), | |
1739cf24 | 1293 | id, def->argc, int (argv.size ())); |
ec2bcbe7 JB |
1294 | } |
1295 | ||
1296 | /* Note that we don't expand macro invocations in the arguments | |
1297 | yet --- we let subst_args take care of that. Parameters that | |
1298 | appear as operands of the stringifying operator "#" or the | |
1299 | splicing operator "##" don't get macro references expanded, | |
1300 | so we can't really tell whether it's appropriate to macro- | |
1301 | expand an argument until we see how it's being used. */ | |
1739cf24 | 1302 | struct macro_buffer substituted (0); |
2fae03e8 | 1303 | substitute_args (&substituted, def, is_varargs, &va_arg_name, |
1739cf24 | 1304 | argv, no_loop, lookup_func, lookup_baton); |
ec2bcbe7 JB |
1305 | |
1306 | /* Now `substituted' is the macro's replacement list, with all | |
1307 | argument values substituted into it properly. Re-scan it for | |
1308 | macro references, but don't expand invocations of this macro. | |
1309 | ||
1310 | We create a new buffer, `substituted_src', which points into | |
1311 | `substituted', and scan that. We can't scan `substituted' | |
1312 | itself, since the tokenization process moves the buffer's | |
1313 | text pointer around, and we still need to be able to find | |
1314 | `substituted's original text buffer after scanning it so we | |
1315 | can free it. */ | |
1739cf24 | 1316 | struct macro_buffer substituted_src (substituted.text, substituted.len); |
ec2bcbe7 JB |
1317 | scan (dest, &substituted_src, &new_no_loop, lookup_func, lookup_baton); |
1318 | ||
ec2bcbe7 JB |
1319 | return 1; |
1320 | } | |
1321 | else | |
e2e0b3e5 | 1322 | internal_error (__FILE__, __LINE__, _("bad macro definition kind")); |
ec2bcbe7 JB |
1323 | } |
1324 | ||
1325 | ||
1326 | /* If the single token in SRC_FIRST followed by the tokens in SRC_REST | |
1327 | constitute a macro invokation not forbidden in NO_LOOP, append its | |
1328 | expansion to DEST and return non-zero. Otherwise, return zero, and | |
1329 | leave DEST unchanged. | |
1330 | ||
1331 | SRC_FIRST and SRC_REST must be shared buffers; DEST must not be one. | |
1332 | SRC_FIRST must be a string built by get_token. */ | |
1333 | static int | |
1334 | maybe_expand (struct macro_buffer *dest, | |
1335 | struct macro_buffer *src_first, | |
1336 | struct macro_buffer *src_rest, | |
1337 | struct macro_name_list *no_loop, | |
1338 | macro_lookup_ftype *lookup_func, | |
1339 | void *lookup_baton) | |
1340 | { | |
1341 | gdb_assert (src_first->shared); | |
1342 | gdb_assert (src_rest->shared); | |
1343 | gdb_assert (! dest->shared); | |
1344 | ||
1345 | /* Is this token an identifier? */ | |
1346 | if (src_first->is_identifier) | |
1347 | { | |
1348 | /* Make a null-terminated copy of it, since that's what our | |
1349 | lookup function expects. */ | |
0354904b | 1350 | std::string id (src_first->text, src_first->len); |
b8d56208 | 1351 | |
ec2bcbe7 JB |
1352 | /* If we're currently re-scanning the result of expanding |
1353 | this macro, don't expand it again. */ | |
0354904b | 1354 | if (! currently_rescanning (no_loop, id.c_str ())) |
ec2bcbe7 JB |
1355 | { |
1356 | /* Does this identifier have a macro definition in scope? */ | |
0354904b TT |
1357 | struct macro_definition *def = lookup_func (id.c_str (), |
1358 | lookup_baton); | |
ec2bcbe7 | 1359 | |
0354904b | 1360 | if (def && expand (id.c_str (), def, dest, src_rest, no_loop, |
ec2bcbe7 | 1361 | lookup_func, lookup_baton)) |
0354904b | 1362 | return 1; |
ec2bcbe7 | 1363 | } |
ec2bcbe7 JB |
1364 | } |
1365 | ||
1366 | return 0; | |
1367 | } | |
1368 | ||
1369 | ||
1370 | /* Expand macro references in SRC, appending the results to DEST. | |
1371 | Assume we are re-scanning the result of expanding the macros named | |
1372 | in NO_LOOP, and don't try to re-expand references to them. | |
1373 | ||
1374 | SRC must be a shared buffer; DEST must not be one. */ | |
1375 | static void | |
1376 | scan (struct macro_buffer *dest, | |
1377 | struct macro_buffer *src, | |
1378 | struct macro_name_list *no_loop, | |
1379 | macro_lookup_ftype *lookup_func, | |
1380 | void *lookup_baton) | |
1381 | { | |
1382 | gdb_assert (src->shared); | |
1383 | gdb_assert (! dest->shared); | |
1384 | ||
1385 | for (;;) | |
1386 | { | |
1387 | struct macro_buffer tok; | |
1388 | char *original_src_start = src->text; | |
1389 | ||
1390 | /* Find the next token in SRC. */ | |
1391 | if (! get_token (&tok, src)) | |
1392 | break; | |
1393 | ||
1394 | /* Just for aesthetics. If we skipped some whitespace, copy | |
1395 | that to DEST. */ | |
1396 | if (tok.text > original_src_start) | |
1397 | { | |
1739cf24 | 1398 | dest->appendmem (original_src_start, tok.text - original_src_start); |
ec2bcbe7 JB |
1399 | dest->last_token = dest->len; |
1400 | } | |
1401 | ||
1402 | if (! maybe_expand (dest, &tok, src, no_loop, lookup_func, lookup_baton)) | |
1403 | /* We didn't end up expanding tok as a macro reference, so | |
1404 | simply append it to dest. */ | |
1405 | append_tokens_without_splicing (dest, &tok); | |
1406 | } | |
1407 | ||
1408 | /* Just for aesthetics. If there was any trailing whitespace in | |
1409 | src, copy it to dest. */ | |
1410 | if (src->len) | |
1411 | { | |
1739cf24 | 1412 | dest->appendmem (src->text, src->len); |
ec2bcbe7 JB |
1413 | dest->last_token = dest->len; |
1414 | } | |
1415 | } | |
1416 | ||
1417 | ||
f6c2623e | 1418 | gdb::unique_xmalloc_ptr<char> |
ec2bcbe7 JB |
1419 | macro_expand (const char *source, |
1420 | macro_lookup_ftype *lookup_func, | |
1421 | void *lookup_func_baton) | |
1422 | { | |
1739cf24 | 1423 | struct macro_buffer src (source, strlen (source)); |
ec2bcbe7 | 1424 | |
1739cf24 | 1425 | struct macro_buffer dest (0); |
ec2bcbe7 | 1426 | dest.last_token = 0; |
ec2bcbe7 JB |
1427 | |
1428 | scan (&dest, &src, 0, lookup_func, lookup_func_baton); | |
1429 | ||
1739cf24 | 1430 | dest.appendc ('\0'); |
ec2bcbe7 | 1431 | |
1739cf24 | 1432 | return gdb::unique_xmalloc_ptr<char> (dest.release ()); |
ec2bcbe7 JB |
1433 | } |
1434 | ||
1435 | ||
f6c2623e | 1436 | gdb::unique_xmalloc_ptr<char> |
ec2bcbe7 JB |
1437 | macro_expand_once (const char *source, |
1438 | macro_lookup_ftype *lookup_func, | |
1439 | void *lookup_func_baton) | |
1440 | { | |
8a3fe4f8 | 1441 | error (_("Expand-once not implemented yet.")); |
ec2bcbe7 JB |
1442 | } |
1443 | ||
1444 | ||
1445 | char * | |
d7561cbb | 1446 | macro_expand_next (const char **lexptr, |
ec2bcbe7 JB |
1447 | macro_lookup_ftype *lookup_func, |
1448 | void *lookup_baton) | |
1449 | { | |
1739cf24 | 1450 | struct macro_buffer tok; |
ec2bcbe7 JB |
1451 | |
1452 | /* Set up SRC to refer to the input text, pointed to by *lexptr. */ | |
1739cf24 | 1453 | struct macro_buffer src (*lexptr, strlen (*lexptr)); |
ec2bcbe7 JB |
1454 | |
1455 | /* Set up DEST to receive the expansion, if there is one. */ | |
1739cf24 | 1456 | struct macro_buffer dest (0); |
ec2bcbe7 | 1457 | dest.last_token = 0; |
ec2bcbe7 JB |
1458 | |
1459 | /* Get the text's first preprocessing token. */ | |
1460 | if (! get_token (&tok, &src)) | |
1739cf24 | 1461 | return 0; |
ec2bcbe7 JB |
1462 | |
1463 | /* If it's a macro invocation, expand it. */ | |
1464 | if (maybe_expand (&dest, &tok, &src, 0, lookup_func, lookup_baton)) | |
1465 | { | |
1466 | /* It was a macro invocation! Package up the expansion as a | |
1467 | null-terminated string and return it. Set *lexptr to the | |
1468 | start of the next token in the input. */ | |
1739cf24 | 1469 | dest.appendc ('\0'); |
ec2bcbe7 | 1470 | *lexptr = src.text; |
1739cf24 | 1471 | return dest.release (); |
ec2bcbe7 JB |
1472 | } |
1473 | else | |
1474 | { | |
1475 | /* It wasn't a macro invocation. */ | |
ec2bcbe7 JB |
1476 | return 0; |
1477 | } | |
1478 | } |