Commit | Line | Data |
---|---|---|
bc78d348 | 1 | /* Extended regular expression matching and search library, |
8b20806d | 2 | version 0.12. |
bc78d348 KB |
3 | (Implements POSIX draft P10003.2/D11.2, except for |
4 | internationalization features.) | |
5 | ||
2860be63 | 6 | Copyright (C) 1993, 1994, 1995 Free Software Foundation, Inc. |
bc78d348 | 7 | |
fa9a63c5 RM |
8 | This program is free software; you can redistribute it and/or modify |
9 | it under the terms of the GNU General Public License as published by | |
10 | the Free Software Foundation; either version 2, or (at your option) | |
11 | any later version. | |
12 | ||
13 | This program is distributed in the hope that it will be useful, | |
14 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | GNU General Public License for more details. | |
17 | ||
18 | You should have received a copy of the GNU General Public License | |
19 | along with this program; if not, write to the Free Software | |
20 | Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ | |
21 | ||
22 | /* AIX requires this to be the first thing in the file. */ | |
23 | #if defined (_AIX) && !defined (REGEX_MALLOC) | |
24 | #pragma alloca | |
25 | #endif | |
26 | ||
27 | #define _GNU_SOURCE | |
28 | ||
29 | #ifdef HAVE_CONFIG_H | |
30 | #include <config.h> | |
31 | #endif | |
32 | ||
33 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ | |
34 | #include <sys/types.h> | |
35 | ||
36 | /* This is for other GNU distributions with internationalized messages. */ | |
37 | #if HAVE_LIBINTL_H || defined (_LIBC) | |
38 | # include <libintl.h> | |
39 | #else | |
40 | # define gettext(msgid) (msgid) | |
41 | #endif | |
42 | ||
43 | /* The `emacs' switch turns on certain matching commands | |
44 | that make sense only in Emacs. */ | |
45 | #ifdef emacs | |
46 | ||
47 | #include "lisp.h" | |
48 | #include "buffer.h" | |
49 | #include "syntax.h" | |
50 | ||
51 | #else /* not emacs */ | |
52 | ||
53 | /* If we are not linking with Emacs proper, | |
54 | we can't use the relocating allocator | |
55 | even if config.h says that we can. */ | |
56 | #undef REL_ALLOC | |
57 | ||
58 | #if defined (STDC_HEADERS) || defined (_LIBC) | |
59 | #include <stdlib.h> | |
60 | #else | |
61 | char *malloc (); | |
62 | char *realloc (); | |
63 | #endif | |
64 | ||
9e4ecb26 KH |
65 | /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow. |
66 | If nothing else has been done, use the method below. */ | |
67 | #ifdef INHIBIT_STRING_HEADER | |
68 | #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY)) | |
69 | #if !defined (bzero) && !defined (bcopy) | |
70 | #undef INHIBIT_STRING_HEADER | |
71 | #endif | |
72 | #endif | |
73 | #endif | |
74 | ||
75 | /* This is the normal way of making sure we have a bcopy and a bzero. | |
76 | This is used in most programs--a few other programs avoid this | |
77 | by defining INHIBIT_STRING_HEADER. */ | |
fa9a63c5 | 78 | #ifndef INHIBIT_STRING_HEADER |
7f998252 | 79 | #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC) |
fa9a63c5 RM |
80 | #include <string.h> |
81 | #ifndef bcmp | |
82 | #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n)) | |
83 | #endif | |
84 | #ifndef bcopy | |
85 | #define bcopy(s, d, n) memcpy ((d), (s), (n)) | |
86 | #endif | |
87 | #ifndef bzero | |
88 | #define bzero(s, n) memset ((s), 0, (n)) | |
89 | #endif | |
90 | #else | |
91 | #include <strings.h> | |
92 | #endif | |
93 | #endif | |
94 | ||
95 | /* Define the syntax stuff for \<, \>, etc. */ | |
96 | ||
97 | /* This must be nonzero for the wordchar and notwordchar pattern | |
98 | commands in re_match_2. */ | |
99 | #ifndef Sword | |
100 | #define Sword 1 | |
101 | #endif | |
102 | ||
103 | #ifdef SWITCH_ENUM_BUG | |
104 | #define SWITCH_ENUM_CAST(x) ((int)(x)) | |
105 | #else | |
106 | #define SWITCH_ENUM_CAST(x) (x) | |
107 | #endif | |
108 | ||
109 | #ifdef SYNTAX_TABLE | |
110 | ||
111 | extern char *re_syntax_table; | |
112 | ||
113 | #else /* not SYNTAX_TABLE */ | |
114 | ||
115 | /* How many characters in the character set. */ | |
116 | #define CHAR_SET_SIZE 256 | |
117 | ||
118 | static char re_syntax_table[CHAR_SET_SIZE]; | |
119 | ||
120 | static void | |
121 | init_syntax_once () | |
122 | { | |
123 | register int c; | |
124 | static int done = 0; | |
125 | ||
126 | if (done) | |
127 | return; | |
128 | ||
129 | bzero (re_syntax_table, sizeof re_syntax_table); | |
130 | ||
131 | for (c = 'a'; c <= 'z'; c++) | |
132 | re_syntax_table[c] = Sword; | |
133 | ||
134 | for (c = 'A'; c <= 'Z'; c++) | |
135 | re_syntax_table[c] = Sword; | |
136 | ||
137 | for (c = '0'; c <= '9'; c++) | |
138 | re_syntax_table[c] = Sword; | |
139 | ||
140 | re_syntax_table['_'] = Sword; | |
141 | ||
142 | done = 1; | |
143 | } | |
144 | ||
145 | #endif /* not SYNTAX_TABLE */ | |
146 | ||
147 | #define SYNTAX(c) re_syntax_table[c] | |
148 | ||
149 | #endif /* not emacs */ | |
150 | \f | |
151 | /* Get the interface, including the syntax bits. */ | |
152 | #include "regex.h" | |
153 | ||
154 | /* isalpha etc. are used for the character classes. */ | |
155 | #include <ctype.h> | |
156 | ||
157 | /* Jim Meyering writes: | |
158 | ||
159 | "... Some ctype macros are valid only for character codes that | |
160 | isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when | |
161 | using /bin/cc or gcc but without giving an ansi option). So, all | |
162 | ctype uses should be through macros like ISPRINT... If | |
163 | STDC_HEADERS is defined, then autoconf has verified that the ctype | |
164 | macros don't need to be guarded with references to isascii. ... | |
165 | Defining isascii to 1 should let any compiler worth its salt | |
166 | eliminate the && through constant folding." */ | |
167 | ||
168 | #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII)) | |
169 | #define ISASCII(c) 1 | |
170 | #else | |
171 | #define ISASCII(c) isascii(c) | |
172 | #endif | |
173 | ||
174 | #ifdef isblank | |
175 | #define ISBLANK(c) (ISASCII (c) && isblank (c)) | |
176 | #else | |
177 | #define ISBLANK(c) ((c) == ' ' || (c) == '\t') | |
178 | #endif | |
179 | #ifdef isgraph | |
180 | #define ISGRAPH(c) (ISASCII (c) && isgraph (c)) | |
181 | #else | |
182 | #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) | |
183 | #endif | |
184 | ||
185 | #define ISPRINT(c) (ISASCII (c) && isprint (c)) | |
186 | #define ISDIGIT(c) (ISASCII (c) && isdigit (c)) | |
187 | #define ISALNUM(c) (ISASCII (c) && isalnum (c)) | |
188 | #define ISALPHA(c) (ISASCII (c) && isalpha (c)) | |
189 | #define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) | |
190 | #define ISLOWER(c) (ISASCII (c) && islower (c)) | |
191 | #define ISPUNCT(c) (ISASCII (c) && ispunct (c)) | |
192 | #define ISSPACE(c) (ISASCII (c) && isspace (c)) | |
193 | #define ISUPPER(c) (ISASCII (c) && isupper (c)) | |
194 | #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) | |
195 | ||
196 | #ifndef NULL | |
075f06ec | 197 | #define NULL (void *)0 |
fa9a63c5 RM |
198 | #endif |
199 | ||
200 | /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
201 | since ours (we hope) works properly with all combinations of | |
202 | machines, compilers, `char' and `unsigned char' argument types. | |
203 | (Per Bothner suggested the basic approach.) */ | |
204 | #undef SIGN_EXTEND_CHAR | |
205 | #if __STDC__ | |
206 | #define SIGN_EXTEND_CHAR(c) ((signed char) (c)) | |
207 | #else /* not __STDC__ */ | |
208 | /* As in Harbison and Steele. */ | |
209 | #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) | |
210 | #endif | |
211 | \f | |
212 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we | |
213 | use `alloca' instead of `malloc'. This is because using malloc in | |
214 | re_search* or re_match* could cause memory leaks when C-g is used in | |
215 | Emacs; also, malloc is slower and causes storage fragmentation. On | |
216 | the other hand, malloc is more portable, and easier to debug. | |
217 | ||
218 | Because we sometimes use alloca, some routines have to be macros, | |
219 | not functions -- `alloca'-allocated space disappears at the end of the | |
220 | function it is called in. */ | |
221 | ||
222 | #ifdef REGEX_MALLOC | |
223 | ||
224 | #define REGEX_ALLOCATE malloc | |
225 | #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) | |
226 | #define REGEX_FREE free | |
227 | ||
228 | #else /* not REGEX_MALLOC */ | |
229 | ||
230 | /* Emacs already defines alloca, sometimes. */ | |
231 | #ifndef alloca | |
232 | ||
233 | /* Make alloca work the best possible way. */ | |
234 | #ifdef __GNUC__ | |
235 | #define alloca __builtin_alloca | |
236 | #else /* not __GNUC__ */ | |
237 | #if HAVE_ALLOCA_H | |
238 | #include <alloca.h> | |
239 | #else /* not __GNUC__ or HAVE_ALLOCA_H */ | |
240 | #ifndef _AIX /* Already did AIX, up at the top. */ | |
241 | char *alloca (); | |
242 | #endif /* not _AIX */ | |
243 | #endif /* not HAVE_ALLOCA_H */ | |
244 | #endif /* not __GNUC__ */ | |
245 | ||
246 | #endif /* not alloca */ | |
247 | ||
248 | #define REGEX_ALLOCATE alloca | |
249 | ||
250 | /* Assumes a `char *destination' variable. */ | |
251 | #define REGEX_REALLOCATE(source, osize, nsize) \ | |
252 | (destination = (char *) alloca (nsize), \ | |
253 | bcopy (source, destination, osize), \ | |
254 | destination) | |
255 | ||
256 | /* No need to do anything to free, after alloca. */ | |
c2e1680a | 257 | #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ |
fa9a63c5 RM |
258 | |
259 | #endif /* not REGEX_MALLOC */ | |
260 | ||
261 | /* Define how to allocate the failure stack. */ | |
262 | ||
33487cc8 | 263 | #if defined (REL_ALLOC) && defined (REGEX_MALLOC) |
4297555e | 264 | |
fa9a63c5 RM |
265 | #define REGEX_ALLOCATE_STACK(size) \ |
266 | r_alloc (&failure_stack_ptr, (size)) | |
267 | #define REGEX_REALLOCATE_STACK(source, osize, nsize) \ | |
268 | r_re_alloc (&failure_stack_ptr, (nsize)) | |
269 | #define REGEX_FREE_STACK(ptr) \ | |
270 | r_alloc_free (&failure_stack_ptr) | |
271 | ||
4297555e | 272 | #else /* not using relocating allocator */ |
fa9a63c5 RM |
273 | |
274 | #ifdef REGEX_MALLOC | |
275 | ||
276 | #define REGEX_ALLOCATE_STACK malloc | |
277 | #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) | |
278 | #define REGEX_FREE_STACK free | |
279 | ||
280 | #else /* not REGEX_MALLOC */ | |
281 | ||
282 | #define REGEX_ALLOCATE_STACK alloca | |
283 | ||
284 | #define REGEX_REALLOCATE_STACK(source, osize, nsize) \ | |
285 | REGEX_REALLOCATE (source, osize, nsize) | |
286 | /* No need to explicitly free anything. */ | |
287 | #define REGEX_FREE_STACK(arg) | |
288 | ||
289 | #endif /* not REGEX_MALLOC */ | |
4297555e | 290 | #endif /* not using relocating allocator */ |
fa9a63c5 RM |
291 | |
292 | ||
293 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
294 | `string1' or just past its end. This works if PTR is NULL, which is | |
295 | a good thing. */ | |
296 | #define FIRST_STRING_P(ptr) \ | |
297 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) | |
298 | ||
299 | /* (Re)Allocate N items of type T using malloc, or fail. */ | |
300 | #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) | |
301 | #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) | |
302 | #define RETALLOC_IF(addr, n, t) \ | |
303 | if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) | |
304 | #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) | |
305 | ||
306 | #define BYTEWIDTH 8 /* In bits. */ | |
307 | ||
308 | #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) | |
309 | ||
310 | #undef MAX | |
311 | #undef MIN | |
312 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
313 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
314 | ||
315 | typedef char boolean; | |
316 | #define false 0 | |
317 | #define true 1 | |
318 | ||
319 | static int re_match_2_internal (); | |
320 | \f | |
321 | /* These are the command codes that appear in compiled regular | |
322 | expressions. Some opcodes are followed by argument bytes. A | |
323 | command code can specify any interpretation whatsoever for its | |
324 | arguments. Zero bytes may appear in the compiled regular expression. */ | |
325 | ||
326 | typedef enum | |
327 | { | |
328 | no_op = 0, | |
329 | ||
330 | /* Succeed right away--no more backtracking. */ | |
331 | succeed, | |
332 | ||
333 | /* Followed by one byte giving n, then by n literal bytes. */ | |
334 | exactn, | |
335 | ||
336 | /* Matches any (more or less) character. */ | |
337 | anychar, | |
338 | ||
339 | /* Matches any one char belonging to specified set. First | |
340 | following byte is number of bitmap bytes. Then come bytes | |
341 | for a bitmap saying which chars are in. Bits in each byte | |
342 | are ordered low-bit-first. A character is in the set if its | |
343 | bit is 1. A character too large to have a bit in the map is | |
344 | automatically not in the set. */ | |
345 | charset, | |
346 | ||
347 | /* Same parameters as charset, but match any character that is | |
348 | not one of those specified. */ | |
349 | charset_not, | |
350 | ||
351 | /* Start remembering the text that is matched, for storing in a | |
352 | register. Followed by one byte with the register number, in | |
353 | the range 0 to one less than the pattern buffer's re_nsub | |
354 | field. Then followed by one byte with the number of groups | |
355 | inner to this one. (This last has to be part of the | |
356 | start_memory only because we need it in the on_failure_jump | |
357 | of re_match_2.) */ | |
358 | start_memory, | |
359 | ||
360 | /* Stop remembering the text that is matched and store it in a | |
361 | memory register. Followed by one byte with the register | |
362 | number, in the range 0 to one less than `re_nsub' in the | |
363 | pattern buffer, and one byte with the number of inner groups, | |
364 | just like `start_memory'. (We need the number of inner | |
365 | groups here because we don't have any easy way of finding the | |
366 | corresponding start_memory when we're at a stop_memory.) */ | |
367 | stop_memory, | |
368 | ||
369 | /* Match a duplicate of something remembered. Followed by one | |
370 | byte containing the register number. */ | |
371 | duplicate, | |
372 | ||
373 | /* Fail unless at beginning of line. */ | |
374 | begline, | |
375 | ||
376 | /* Fail unless at end of line. */ | |
377 | endline, | |
378 | ||
379 | /* Succeeds if at beginning of buffer (if emacs) or at beginning | |
380 | of string to be matched (if not). */ | |
381 | begbuf, | |
382 | ||
383 | /* Analogously, for end of buffer/string. */ | |
384 | endbuf, | |
385 | ||
386 | /* Followed by two byte relative address to which to jump. */ | |
387 | jump, | |
388 | ||
389 | /* Same as jump, but marks the end of an alternative. */ | |
390 | jump_past_alt, | |
391 | ||
392 | /* Followed by two-byte relative address of place to resume at | |
393 | in case of failure. */ | |
394 | on_failure_jump, | |
395 | ||
396 | /* Like on_failure_jump, but pushes a placeholder instead of the | |
397 | current string position when executed. */ | |
398 | on_failure_keep_string_jump, | |
399 | ||
400 | /* Throw away latest failure point and then jump to following | |
401 | two-byte relative address. */ | |
402 | pop_failure_jump, | |
403 | ||
404 | /* Change to pop_failure_jump if know won't have to backtrack to | |
405 | match; otherwise change to jump. This is used to jump | |
406 | back to the beginning of a repeat. If what follows this jump | |
407 | clearly won't match what the repeat does, such that we can be | |
408 | sure that there is no use backtracking out of repetitions | |
409 | already matched, then we change it to a pop_failure_jump. | |
410 | Followed by two-byte address. */ | |
411 | maybe_pop_jump, | |
412 | ||
413 | /* Jump to following two-byte address, and push a dummy failure | |
414 | point. This failure point will be thrown away if an attempt | |
415 | is made to use it for a failure. A `+' construct makes this | |
416 | before the first repeat. Also used as an intermediary kind | |
417 | of jump when compiling an alternative. */ | |
418 | dummy_failure_jump, | |
419 | ||
420 | /* Push a dummy failure point and continue. Used at the end of | |
421 | alternatives. */ | |
422 | push_dummy_failure, | |
423 | ||
424 | /* Followed by two-byte relative address and two-byte number n. | |
425 | After matching N times, jump to the address upon failure. */ | |
426 | succeed_n, | |
427 | ||
428 | /* Followed by two-byte relative address, and two-byte number n. | |
429 | Jump to the address N times, then fail. */ | |
430 | jump_n, | |
431 | ||
432 | /* Set the following two-byte relative address to the | |
433 | subsequent two-byte number. The address *includes* the two | |
434 | bytes of number. */ | |
435 | set_number_at, | |
436 | ||
437 | wordchar, /* Matches any word-constituent character. */ | |
438 | notwordchar, /* Matches any char that is not a word-constituent. */ | |
439 | ||
440 | wordbeg, /* Succeeds if at word beginning. */ | |
441 | wordend, /* Succeeds if at word end. */ | |
442 | ||
443 | wordbound, /* Succeeds if at a word boundary. */ | |
444 | notwordbound /* Succeeds if not at a word boundary. */ | |
445 | ||
446 | #ifdef emacs | |
447 | ,before_dot, /* Succeeds if before point. */ | |
448 | at_dot, /* Succeeds if at point. */ | |
449 | after_dot, /* Succeeds if after point. */ | |
450 | ||
451 | /* Matches any character whose syntax is specified. Followed by | |
452 | a byte which contains a syntax code, e.g., Sword. */ | |
453 | syntaxspec, | |
454 | ||
455 | /* Matches any character whose syntax is not that specified. */ | |
456 | notsyntaxspec | |
457 | #endif /* emacs */ | |
458 | } re_opcode_t; | |
459 | \f | |
460 | /* Common operations on the compiled pattern. */ | |
461 | ||
462 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
463 | ||
464 | #define STORE_NUMBER(destination, number) \ | |
465 | do { \ | |
466 | (destination)[0] = (number) & 0377; \ | |
467 | (destination)[1] = (number) >> 8; \ | |
468 | } while (0) | |
469 | ||
470 | /* Same as STORE_NUMBER, except increment DESTINATION to | |
471 | the byte after where the number is stored. Therefore, DESTINATION | |
472 | must be an lvalue. */ | |
473 | ||
474 | #define STORE_NUMBER_AND_INCR(destination, number) \ | |
475 | do { \ | |
476 | STORE_NUMBER (destination, number); \ | |
477 | (destination) += 2; \ | |
478 | } while (0) | |
479 | ||
480 | /* Put into DESTINATION a number stored in two contiguous bytes starting | |
481 | at SOURCE. */ | |
482 | ||
483 | #define EXTRACT_NUMBER(destination, source) \ | |
484 | do { \ | |
485 | (destination) = *(source) & 0377; \ | |
486 | (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
487 | } while (0) | |
488 | ||
489 | #ifdef DEBUG | |
490 | static void | |
491 | extract_number (dest, source) | |
492 | int *dest; | |
493 | unsigned char *source; | |
494 | { | |
495 | int temp = SIGN_EXTEND_CHAR (*(source + 1)); | |
496 | *dest = *source & 0377; | |
497 | *dest += temp << 8; | |
498 | } | |
499 | ||
500 | #ifndef EXTRACT_MACROS /* To debug the macros. */ | |
501 | #undef EXTRACT_NUMBER | |
502 | #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
503 | #endif /* not EXTRACT_MACROS */ | |
504 | ||
505 | #endif /* DEBUG */ | |
506 | ||
507 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
508 | SOURCE must be an lvalue. */ | |
509 | ||
510 | #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
511 | do { \ | |
512 | EXTRACT_NUMBER (destination, source); \ | |
513 | (source) += 2; \ | |
514 | } while (0) | |
515 | ||
516 | #ifdef DEBUG | |
517 | static void | |
518 | extract_number_and_incr (destination, source) | |
519 | int *destination; | |
520 | unsigned char **source; | |
521 | { | |
522 | extract_number (destination, *source); | |
523 | *source += 2; | |
524 | } | |
525 | ||
526 | #ifndef EXTRACT_MACROS | |
527 | #undef EXTRACT_NUMBER_AND_INCR | |
528 | #define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
529 | extract_number_and_incr (&dest, &src) | |
530 | #endif /* not EXTRACT_MACROS */ | |
531 | ||
532 | #endif /* DEBUG */ | |
533 | \f | |
534 | /* If DEBUG is defined, Regex prints many voluminous messages about what | |
535 | it is doing (if the variable `debug' is nonzero). If linked with the | |
536 | main program in `iregex.c', you can enter patterns and strings | |
537 | interactively. And if linked with the main program in `main.c' and | |
538 | the other test files, you can run the already-written tests. */ | |
539 | ||
540 | #ifdef DEBUG | |
541 | ||
542 | /* We use standard I/O for debugging. */ | |
543 | #include <stdio.h> | |
544 | ||
545 | /* It is useful to test things that ``must'' be true when debugging. */ | |
546 | #include <assert.h> | |
547 | ||
548 | static int debug = 0; | |
549 | ||
550 | #define DEBUG_STATEMENT(e) e | |
551 | #define DEBUG_PRINT1(x) if (debug) printf (x) | |
552 | #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) | |
553 | #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) | |
554 | #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) | |
555 | #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ | |
556 | if (debug) print_partial_compiled_pattern (s, e) | |
557 | #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ | |
558 | if (debug) print_double_string (w, s1, sz1, s2, sz2) | |
559 | ||
560 | ||
561 | /* Print the fastmap in human-readable form. */ | |
562 | ||
563 | void | |
564 | print_fastmap (fastmap) | |
565 | char *fastmap; | |
566 | { | |
567 | unsigned was_a_range = 0; | |
568 | unsigned i = 0; | |
569 | ||
570 | while (i < (1 << BYTEWIDTH)) | |
571 | { | |
572 | if (fastmap[i++]) | |
573 | { | |
574 | was_a_range = 0; | |
575 | putchar (i - 1); | |
576 | while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
577 | { | |
578 | was_a_range = 1; | |
579 | i++; | |
580 | } | |
581 | if (was_a_range) | |
582 | { | |
583 | printf ("-"); | |
584 | putchar (i - 1); | |
585 | } | |
586 | } | |
587 | } | |
588 | putchar ('\n'); | |
589 | } | |
590 | ||
591 | ||
592 | /* Print a compiled pattern string in human-readable form, starting at | |
593 | the START pointer into it and ending just before the pointer END. */ | |
594 | ||
595 | void | |
596 | print_partial_compiled_pattern (start, end) | |
597 | unsigned char *start; | |
598 | unsigned char *end; | |
599 | { | |
600 | int mcnt, mcnt2; | |
601 | unsigned char *p = start; | |
602 | unsigned char *pend = end; | |
603 | ||
604 | if (start == NULL) | |
605 | { | |
606 | printf ("(null)\n"); | |
607 | return; | |
608 | } | |
609 | ||
610 | /* Loop over pattern commands. */ | |
611 | while (p < pend) | |
612 | { | |
613 | printf ("%d:\t", p - start); | |
614 | ||
615 | switch ((re_opcode_t) *p++) | |
616 | { | |
617 | case no_op: | |
618 | printf ("/no_op"); | |
619 | break; | |
620 | ||
621 | case exactn: | |
622 | mcnt = *p++; | |
623 | printf ("/exactn/%d", mcnt); | |
624 | do | |
625 | { | |
626 | putchar ('/'); | |
627 | putchar (*p++); | |
628 | } | |
629 | while (--mcnt); | |
630 | break; | |
631 | ||
632 | case start_memory: | |
633 | mcnt = *p++; | |
634 | printf ("/start_memory/%d/%d", mcnt, *p++); | |
635 | break; | |
636 | ||
637 | case stop_memory: | |
638 | mcnt = *p++; | |
639 | printf ("/stop_memory/%d/%d", mcnt, *p++); | |
640 | break; | |
641 | ||
642 | case duplicate: | |
643 | printf ("/duplicate/%d", *p++); | |
644 | break; | |
645 | ||
646 | case anychar: | |
647 | printf ("/anychar"); | |
648 | break; | |
649 | ||
650 | case charset: | |
651 | case charset_not: | |
652 | { | |
653 | register int c, last = -100; | |
654 | register int in_range = 0; | |
655 | ||
656 | printf ("/charset [%s", | |
657 | (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); | |
658 | ||
659 | assert (p + *p < pend); | |
660 | ||
661 | for (c = 0; c < 256; c++) | |
662 | if (c / 8 < *p | |
663 | && (p[1 + (c/8)] & (1 << (c % 8)))) | |
664 | { | |
665 | /* Are we starting a range? */ | |
666 | if (last + 1 == c && ! in_range) | |
667 | { | |
668 | putchar ('-'); | |
669 | in_range = 1; | |
670 | } | |
671 | /* Have we broken a range? */ | |
672 | else if (last + 1 != c && in_range) | |
673 | { | |
674 | putchar (last); | |
675 | in_range = 0; | |
676 | } | |
677 | ||
678 | if (! in_range) | |
679 | putchar (c); | |
680 | ||
681 | last = c; | |
682 | } | |
683 | ||
684 | if (in_range) | |
685 | putchar (last); | |
686 | ||
687 | putchar (']'); | |
688 | ||
689 | p += 1 + *p; | |
690 | } | |
691 | break; | |
692 | ||
693 | case begline: | |
694 | printf ("/begline"); | |
695 | break; | |
696 | ||
697 | case endline: | |
698 | printf ("/endline"); | |
699 | break; | |
700 | ||
701 | case on_failure_jump: | |
702 | extract_number_and_incr (&mcnt, &p); | |
703 | printf ("/on_failure_jump to %d", p + mcnt - start); | |
704 | break; | |
705 | ||
706 | case on_failure_keep_string_jump: | |
707 | extract_number_and_incr (&mcnt, &p); | |
708 | printf ("/on_failure_keep_string_jump to %d", p + mcnt - start); | |
709 | break; | |
710 | ||
711 | case dummy_failure_jump: | |
712 | extract_number_and_incr (&mcnt, &p); | |
713 | printf ("/dummy_failure_jump to %d", p + mcnt - start); | |
714 | break; | |
715 | ||
716 | case push_dummy_failure: | |
717 | printf ("/push_dummy_failure"); | |
718 | break; | |
719 | ||
720 | case maybe_pop_jump: | |
721 | extract_number_and_incr (&mcnt, &p); | |
722 | printf ("/maybe_pop_jump to %d", p + mcnt - start); | |
723 | break; | |
724 | ||
725 | case pop_failure_jump: | |
726 | extract_number_and_incr (&mcnt, &p); | |
727 | printf ("/pop_failure_jump to %d", p + mcnt - start); | |
728 | break; | |
729 | ||
730 | case jump_past_alt: | |
731 | extract_number_and_incr (&mcnt, &p); | |
732 | printf ("/jump_past_alt to %d", p + mcnt - start); | |
733 | break; | |
734 | ||
735 | case jump: | |
736 | extract_number_and_incr (&mcnt, &p); | |
737 | printf ("/jump to %d", p + mcnt - start); | |
738 | break; | |
739 | ||
740 | case succeed_n: | |
741 | extract_number_and_incr (&mcnt, &p); | |
742 | extract_number_and_incr (&mcnt2, &p); | |
743 | printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2); | |
744 | break; | |
745 | ||
746 | case jump_n: | |
747 | extract_number_and_incr (&mcnt, &p); | |
748 | extract_number_and_incr (&mcnt2, &p); | |
749 | printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2); | |
750 | break; | |
751 | ||
752 | case set_number_at: | |
753 | extract_number_and_incr (&mcnt, &p); | |
754 | extract_number_and_incr (&mcnt2, &p); | |
755 | printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2); | |
756 | break; | |
757 | ||
758 | case wordbound: | |
759 | printf ("/wordbound"); | |
760 | break; | |
761 | ||
762 | case notwordbound: | |
763 | printf ("/notwordbound"); | |
764 | break; | |
765 | ||
766 | case wordbeg: | |
767 | printf ("/wordbeg"); | |
768 | break; | |
769 | ||
770 | case wordend: | |
771 | printf ("/wordend"); | |
772 | ||
773 | #ifdef emacs | |
774 | case before_dot: | |
775 | printf ("/before_dot"); | |
776 | break; | |
777 | ||
778 | case at_dot: | |
779 | printf ("/at_dot"); | |
780 | break; | |
781 | ||
782 | case after_dot: | |
783 | printf ("/after_dot"); | |
784 | break; | |
785 | ||
786 | case syntaxspec: | |
787 | printf ("/syntaxspec"); | |
788 | mcnt = *p++; | |
789 | printf ("/%d", mcnt); | |
790 | break; | |
791 | ||
792 | case notsyntaxspec: | |
793 | printf ("/notsyntaxspec"); | |
794 | mcnt = *p++; | |
795 | printf ("/%d", mcnt); | |
796 | break; | |
797 | #endif /* emacs */ | |
798 | ||
799 | case wordchar: | |
800 | printf ("/wordchar"); | |
801 | break; | |
802 | ||
803 | case notwordchar: | |
804 | printf ("/notwordchar"); | |
805 | break; | |
806 | ||
807 | case begbuf: | |
808 | printf ("/begbuf"); | |
809 | break; | |
810 | ||
811 | case endbuf: | |
812 | printf ("/endbuf"); | |
813 | break; | |
814 | ||
815 | default: | |
816 | printf ("?%d", *(p-1)); | |
817 | } | |
818 | ||
819 | putchar ('\n'); | |
820 | } | |
821 | ||
822 | printf ("%d:\tend of pattern.\n", p - start); | |
823 | } | |
824 | ||
825 | ||
826 | void | |
827 | print_compiled_pattern (bufp) | |
828 | struct re_pattern_buffer *bufp; | |
829 | { | |
830 | unsigned char *buffer = bufp->buffer; | |
831 | ||
832 | print_partial_compiled_pattern (buffer, buffer + bufp->used); | |
833 | printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated); | |
834 | ||
835 | if (bufp->fastmap_accurate && bufp->fastmap) | |
836 | { | |
837 | printf ("fastmap: "); | |
838 | print_fastmap (bufp->fastmap); | |
839 | } | |
840 | ||
841 | printf ("re_nsub: %d\t", bufp->re_nsub); | |
842 | printf ("regs_alloc: %d\t", bufp->regs_allocated); | |
843 | printf ("can_be_null: %d\t", bufp->can_be_null); | |
844 | printf ("newline_anchor: %d\n", bufp->newline_anchor); | |
845 | printf ("no_sub: %d\t", bufp->no_sub); | |
846 | printf ("not_bol: %d\t", bufp->not_bol); | |
847 | printf ("not_eol: %d\t", bufp->not_eol); | |
848 | printf ("syntax: %d\n", bufp->syntax); | |
849 | /* Perhaps we should print the translate table? */ | |
850 | } | |
851 | ||
852 | ||
853 | void | |
854 | print_double_string (where, string1, size1, string2, size2) | |
855 | const char *where; | |
856 | const char *string1; | |
857 | const char *string2; | |
858 | int size1; | |
859 | int size2; | |
860 | { | |
861 | unsigned this_char; | |
862 | ||
863 | if (where == NULL) | |
864 | printf ("(null)"); | |
865 | else | |
866 | { | |
867 | if (FIRST_STRING_P (where)) | |
868 | { | |
869 | for (this_char = where - string1; this_char < size1; this_char++) | |
870 | putchar (string1[this_char]); | |
871 | ||
872 | where = string2; | |
873 | } | |
874 | ||
875 | for (this_char = where - string2; this_char < size2; this_char++) | |
876 | putchar (string2[this_char]); | |
877 | } | |
878 | } | |
879 | ||
880 | #else /* not DEBUG */ | |
881 | ||
882 | #undef assert | |
883 | #define assert(e) | |
884 | ||
885 | #define DEBUG_STATEMENT(e) | |
886 | #define DEBUG_PRINT1(x) | |
887 | #define DEBUG_PRINT2(x1, x2) | |
888 | #define DEBUG_PRINT3(x1, x2, x3) | |
889 | #define DEBUG_PRINT4(x1, x2, x3, x4) | |
890 | #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) | |
891 | #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
892 | ||
893 | #endif /* not DEBUG */ | |
894 | \f | |
895 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
896 | also be assigned to arbitrarily: each pattern buffer stores its own | |
897 | syntax, so it can be changed between regex compilations. */ | |
898 | /* This has no initializer because initialized variables in Emacs | |
899 | become read-only after dumping. */ | |
900 | reg_syntax_t re_syntax_options; | |
901 | ||
902 | ||
903 | /* Specify the precise syntax of regexps for compilation. This provides | |
904 | for compatibility for various utilities which historically have | |
905 | different, incompatible syntaxes. | |
906 | ||
907 | The argument SYNTAX is a bit mask comprised of the various bits | |
908 | defined in regex.h. We return the old syntax. */ | |
909 | ||
910 | reg_syntax_t | |
911 | re_set_syntax (syntax) | |
912 | reg_syntax_t syntax; | |
913 | { | |
914 | reg_syntax_t ret = re_syntax_options; | |
915 | ||
916 | re_syntax_options = syntax; | |
917 | return ret; | |
918 | } | |
919 | \f | |
920 | /* This table gives an error message for each of the error codes listed | |
921 | in regex.h. Obviously the order here has to be same as there. | |
922 | POSIX doesn't require that we do anything for REG_NOERROR, | |
923 | but why not be nice? */ | |
924 | ||
925 | static const char *re_error_msgid[] = | |
926 | { "Success", /* REG_NOERROR */ | |
927 | "No match", /* REG_NOMATCH */ | |
928 | "Invalid regular expression", /* REG_BADPAT */ | |
929 | "Invalid collation character", /* REG_ECOLLATE */ | |
930 | "Invalid character class name", /* REG_ECTYPE */ | |
931 | "Trailing backslash", /* REG_EESCAPE */ | |
932 | "Invalid back reference", /* REG_ESUBREG */ | |
933 | "Unmatched [ or [^", /* REG_EBRACK */ | |
934 | "Unmatched ( or \\(", /* REG_EPAREN */ | |
935 | "Unmatched \\{", /* REG_EBRACE */ | |
936 | "Invalid content of \\{\\}", /* REG_BADBR */ | |
937 | "Invalid range end", /* REG_ERANGE */ | |
938 | "Memory exhausted", /* REG_ESPACE */ | |
939 | "Invalid preceding regular expression", /* REG_BADRPT */ | |
940 | "Premature end of regular expression", /* REG_EEND */ | |
941 | "Regular expression too big", /* REG_ESIZE */ | |
942 | "Unmatched ) or \\)", /* REG_ERPAREN */ | |
943 | }; | |
944 | \f | |
945 | /* Avoiding alloca during matching, to placate r_alloc. */ | |
946 | ||
947 | /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the | |
948 | searching and matching functions should not call alloca. On some | |
949 | systems, alloca is implemented in terms of malloc, and if we're | |
950 | using the relocating allocator routines, then malloc could cause a | |
951 | relocation, which might (if the strings being searched are in the | |
952 | ralloc heap) shift the data out from underneath the regexp | |
953 | routines. | |
954 | ||
955 | Here's another reason to avoid allocation: Emacs | |
956 | processes input from X in a signal handler; processing X input may | |
957 | call malloc; if input arrives while a matching routine is calling | |
958 | malloc, then we're scrod. But Emacs can't just block input while | |
959 | calling matching routines; then we don't notice interrupts when | |
960 | they come in. So, Emacs blocks input around all regexp calls | |
961 | except the matching calls, which it leaves unprotected, in the | |
962 | faith that they will not malloc. */ | |
963 | ||
964 | /* Normally, this is fine. */ | |
965 | #define MATCH_MAY_ALLOCATE | |
966 | ||
967 | /* When using GNU C, we are not REALLY using the C alloca, no matter | |
968 | what config.h may say. So don't take precautions for it. */ | |
969 | #ifdef __GNUC__ | |
970 | #undef C_ALLOCA | |
971 | #endif | |
972 | ||
973 | /* The match routines may not allocate if (1) they would do it with malloc | |
974 | and (2) it's not safe for them to use malloc. | |
975 | Note that if REL_ALLOC is defined, matching would not use malloc for the | |
976 | failure stack, but we would still use it for the register vectors; | |
977 | so REL_ALLOC should not affect this. */ | |
978 | #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs) | |
979 | #undef MATCH_MAY_ALLOCATE | |
980 | #endif | |
981 | ||
982 | \f | |
983 | /* Failure stack declarations and macros; both re_compile_fastmap and | |
984 | re_match_2 use a failure stack. These have to be macros because of | |
985 | REGEX_ALLOCATE_STACK. */ | |
986 | ||
987 | ||
988 | /* Number of failure points for which to initially allocate space | |
989 | when matching. If this number is exceeded, we allocate more | |
990 | space, so it is not a hard limit. */ | |
991 | #ifndef INIT_FAILURE_ALLOC | |
992 | #define INIT_FAILURE_ALLOC 5 | |
993 | #endif | |
994 | ||
995 | /* Roughly the maximum number of failure points on the stack. Would be | |
996 | exactly that if always used MAX_FAILURE_SPACE each time we failed. | |
997 | This is a variable only so users of regex can assign to it; we never | |
998 | change it ourselves. */ | |
999 | #if defined (MATCH_MAY_ALLOCATE) | |
1000 | int re_max_failures = 200000; | |
1001 | #else | |
1002 | int re_max_failures = 2000; | |
1003 | #endif | |
1004 | ||
1005 | union fail_stack_elt | |
1006 | { | |
1007 | unsigned char *pointer; | |
1008 | int integer; | |
1009 | }; | |
1010 | ||
1011 | typedef union fail_stack_elt fail_stack_elt_t; | |
1012 | ||
1013 | typedef struct | |
1014 | { | |
1015 | fail_stack_elt_t *stack; | |
1016 | unsigned size; | |
1017 | unsigned avail; /* Offset of next open position. */ | |
1018 | } fail_stack_type; | |
1019 | ||
1020 | #define FAIL_STACK_EMPTY() (fail_stack.avail == 0) | |
1021 | #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) | |
1022 | #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) | |
1023 | ||
1024 | ||
1025 | /* Define macros to initialize and free the failure stack. | |
1026 | Do `return -2' if the alloc fails. */ | |
1027 | ||
1028 | #ifdef MATCH_MAY_ALLOCATE | |
1029 | #define INIT_FAIL_STACK() \ | |
1030 | do { \ | |
1031 | fail_stack.stack = (fail_stack_elt_t *) \ | |
1032 | REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \ | |
1033 | \ | |
1034 | if (fail_stack.stack == NULL) \ | |
1035 | return -2; \ | |
1036 | \ | |
1037 | fail_stack.size = INIT_FAILURE_ALLOC; \ | |
1038 | fail_stack.avail = 0; \ | |
1039 | } while (0) | |
1040 | ||
1041 | #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) | |
1042 | #else | |
1043 | #define INIT_FAIL_STACK() \ | |
1044 | do { \ | |
1045 | fail_stack.avail = 0; \ | |
1046 | } while (0) | |
1047 | ||
1048 | #define RESET_FAIL_STACK() | |
1049 | #endif | |
1050 | ||
1051 | ||
1052 | /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. | |
1053 | ||
1054 | Return 1 if succeeds, and 0 if either ran out of memory | |
1055 | allocating space for it or it was already too large. | |
1056 | ||
1057 | REGEX_REALLOCATE_STACK requires `destination' be declared. */ | |
1058 | ||
1059 | #define DOUBLE_FAIL_STACK(fail_stack) \ | |
1060 | ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \ | |
1061 | ? 0 \ | |
1062 | : ((fail_stack).stack = (fail_stack_elt_t *) \ | |
1063 | REGEX_REALLOCATE_STACK ((fail_stack).stack, \ | |
1064 | (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
1065 | ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \ | |
1066 | \ | |
1067 | (fail_stack).stack == NULL \ | |
1068 | ? 0 \ | |
1069 | : ((fail_stack).size <<= 1, \ | |
1070 | 1))) | |
1071 | ||
1072 | ||
1073 | /* Push pointer POINTER on FAIL_STACK. | |
1074 | Return 1 if was able to do so and 0 if ran out of memory allocating | |
1075 | space to do so. */ | |
1076 | #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \ | |
1077 | ((FAIL_STACK_FULL () \ | |
1078 | && !DOUBLE_FAIL_STACK (FAIL_STACK)) \ | |
1079 | ? 0 \ | |
1080 | : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \ | |
1081 | 1)) | |
1082 | ||
1083 | /* Push a pointer value onto the failure stack. | |
1084 | Assumes the variable `fail_stack'. Probably should only | |
1085 | be called from within `PUSH_FAILURE_POINT'. */ | |
1086 | #define PUSH_FAILURE_POINTER(item) \ | |
1087 | fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item) | |
1088 | ||
1089 | /* This pushes an integer-valued item onto the failure stack. | |
1090 | Assumes the variable `fail_stack'. Probably should only | |
1091 | be called from within `PUSH_FAILURE_POINT'. */ | |
1092 | #define PUSH_FAILURE_INT(item) \ | |
1093 | fail_stack.stack[fail_stack.avail++].integer = (item) | |
1094 | ||
1095 | /* Push a fail_stack_elt_t value onto the failure stack. | |
1096 | Assumes the variable `fail_stack'. Probably should only | |
1097 | be called from within `PUSH_FAILURE_POINT'. */ | |
1098 | #define PUSH_FAILURE_ELT(item) \ | |
1099 | fail_stack.stack[fail_stack.avail++] = (item) | |
1100 | ||
1101 | /* These three POP... operations complement the three PUSH... operations. | |
1102 | All assume that `fail_stack' is nonempty. */ | |
1103 | #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer | |
1104 | #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer | |
1105 | #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] | |
1106 | ||
1107 | /* Used to omit pushing failure point id's when we're not debugging. */ | |
1108 | #ifdef DEBUG | |
1109 | #define DEBUG_PUSH PUSH_FAILURE_INT | |
1110 | #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT () | |
1111 | #else | |
1112 | #define DEBUG_PUSH(item) | |
1113 | #define DEBUG_POP(item_addr) | |
1114 | #endif | |
1115 | ||
1116 | ||
1117 | /* Push the information about the state we will need | |
1118 | if we ever fail back to it. | |
1119 | ||
1120 | Requires variables fail_stack, regstart, regend, reg_info, and | |
1121 | num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be | |
1122 | declared. | |
1123 | ||
1124 | Does `return FAILURE_CODE' if runs out of memory. */ | |
1125 | ||
1126 | #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ | |
1127 | do { \ | |
1128 | char *destination; \ | |
1129 | /* Must be int, so when we don't save any registers, the arithmetic \ | |
1130 | of 0 + -1 isn't done as unsigned. */ \ | |
1131 | int this_reg; \ | |
1132 | \ | |
1133 | DEBUG_STATEMENT (failure_id++); \ | |
1134 | DEBUG_STATEMENT (nfailure_points_pushed++); \ | |
1135 | DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ | |
1136 | DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ | |
1137 | DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ | |
1138 | \ | |
1139 | DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \ | |
1140 | DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ | |
1141 | \ | |
1142 | /* Ensure we have enough space allocated for what we will push. */ \ | |
1143 | while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ | |
1144 | { \ | |
1145 | if (!DOUBLE_FAIL_STACK (fail_stack)) \ | |
1146 | return failure_code; \ | |
1147 | \ | |
1148 | DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ | |
1149 | (fail_stack).size); \ | |
1150 | DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ | |
1151 | } \ | |
1152 | \ | |
1153 | /* Push the info, starting with the registers. */ \ | |
1154 | DEBUG_PRINT1 ("\n"); \ | |
1155 | \ | |
faec11db RS |
1156 | if (!RE_NO_POSIX_BACKTRACKING & bufp->syntax) \ |
1157 | for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ | |
1158 | this_reg++) \ | |
1159 | { \ | |
1160 | DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \ | |
1161 | DEBUG_STATEMENT (num_regs_pushed++); \ | |
fa9a63c5 | 1162 | \ |
faec11db RS |
1163 | DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ |
1164 | PUSH_FAILURE_POINTER (regstart[this_reg]); \ | |
fa9a63c5 | 1165 | \ |
faec11db RS |
1166 | DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ |
1167 | PUSH_FAILURE_POINTER (regend[this_reg]); \ | |
1168 | \ | |
1169 | DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \ | |
1170 | DEBUG_PRINT2 (" match_null=%d", \ | |
1171 | REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ | |
1172 | DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ | |
1173 | DEBUG_PRINT2 (" matched_something=%d", \ | |
1174 | MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1175 | DEBUG_PRINT2 (" ever_matched=%d", \ | |
1176 | EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1177 | DEBUG_PRINT1 ("\n"); \ | |
1178 | PUSH_FAILURE_ELT (reg_info[this_reg].word); \ | |
1179 | } \ | |
fa9a63c5 RM |
1180 | \ |
1181 | DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\ | |
1182 | PUSH_FAILURE_INT (lowest_active_reg); \ | |
1183 | \ | |
1184 | DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\ | |
1185 | PUSH_FAILURE_INT (highest_active_reg); \ | |
1186 | \ | |
1187 | DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \ | |
1188 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ | |
1189 | PUSH_FAILURE_POINTER (pattern_place); \ | |
1190 | \ | |
1191 | DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \ | |
1192 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ | |
1193 | size2); \ | |
1194 | DEBUG_PRINT1 ("'\n"); \ | |
1195 | PUSH_FAILURE_POINTER (string_place); \ | |
1196 | \ | |
1197 | DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ | |
1198 | DEBUG_PUSH (failure_id); \ | |
1199 | } while (0) | |
1200 | ||
1201 | /* This is the number of items that are pushed and popped on the stack | |
1202 | for each register. */ | |
1203 | #define NUM_REG_ITEMS 3 | |
1204 | ||
1205 | /* Individual items aside from the registers. */ | |
1206 | #ifdef DEBUG | |
1207 | #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ | |
1208 | #else | |
1209 | #define NUM_NONREG_ITEMS 4 | |
1210 | #endif | |
1211 | ||
1212 | /* We push at most this many items on the stack. */ | |
1213 | #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS) | |
1214 | ||
1215 | /* We actually push this many items. */ | |
faec11db RS |
1216 | #define NUM_FAILURE_ITEMS \ |
1217 | (((RE_NO_POSIX_BACKTRACKING & bufp->syntax \ | |
1218 | ? 0 : highest_active_reg - lowest_active_reg + 1) \ | |
1219 | * NUM_REG_ITEMS) \ | |
1220 | + NUM_NONREG_ITEMS) | |
fa9a63c5 RM |
1221 | |
1222 | /* How many items can still be added to the stack without overflowing it. */ | |
1223 | #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
1224 | ||
1225 | ||
1226 | /* Pops what PUSH_FAIL_STACK pushes. | |
1227 | ||
1228 | We restore into the parameters, all of which should be lvalues: | |
1229 | STR -- the saved data position. | |
1230 | PAT -- the saved pattern position. | |
1231 | LOW_REG, HIGH_REG -- the highest and lowest active registers. | |
1232 | REGSTART, REGEND -- arrays of string positions. | |
1233 | REG_INFO -- array of information about each subexpression. | |
1234 | ||
1235 | Also assumes the variables `fail_stack' and (if debugging), `bufp', | |
1236 | `pend', `string1', `size1', `string2', and `size2'. */ | |
1237 | ||
1238 | #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ | |
1239 | { \ | |
1240 | DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \ | |
1241 | int this_reg; \ | |
1242 | const unsigned char *string_temp; \ | |
1243 | \ | |
1244 | assert (!FAIL_STACK_EMPTY ()); \ | |
1245 | \ | |
1246 | /* Remove failure points and point to how many regs pushed. */ \ | |
1247 | DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
1248 | DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
1249 | DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ | |
1250 | \ | |
1251 | assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ | |
1252 | \ | |
1253 | DEBUG_POP (&failure_id); \ | |
1254 | DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ | |
1255 | \ | |
1256 | /* If the saved string location is NULL, it came from an \ | |
1257 | on_failure_keep_string_jump opcode, and we want to throw away the \ | |
1258 | saved NULL, thus retaining our current position in the string. */ \ | |
1259 | string_temp = POP_FAILURE_POINTER (); \ | |
1260 | if (string_temp != NULL) \ | |
1261 | str = (const char *) string_temp; \ | |
1262 | \ | |
1263 | DEBUG_PRINT2 (" Popping string 0x%x: `", str); \ | |
1264 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ | |
1265 | DEBUG_PRINT1 ("'\n"); \ | |
1266 | \ | |
1267 | pat = (unsigned char *) POP_FAILURE_POINTER (); \ | |
1268 | DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \ | |
1269 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
1270 | \ | |
1271 | /* Restore register info. */ \ | |
1272 | high_reg = (unsigned) POP_FAILURE_INT (); \ | |
1273 | DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \ | |
1274 | \ | |
1275 | low_reg = (unsigned) POP_FAILURE_INT (); \ | |
1276 | DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \ | |
1277 | \ | |
faec11db RS |
1278 | if (!RE_NO_POSIX_BACKTRACKING & bufp->syntax) \ |
1279 | for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ | |
1280 | { \ | |
1281 | DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \ | |
fa9a63c5 | 1282 | \ |
faec11db RS |
1283 | reg_info[this_reg].word = POP_FAILURE_ELT (); \ |
1284 | DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \ | |
fa9a63c5 | 1285 | \ |
faec11db RS |
1286 | regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \ |
1287 | DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ | |
fa9a63c5 | 1288 | \ |
faec11db RS |
1289 | regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \ |
1290 | DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ | |
1291 | } \ | |
fa9a63c5 RM |
1292 | \ |
1293 | set_regs_matched_done = 0; \ | |
1294 | DEBUG_STATEMENT (nfailure_points_popped++); \ | |
1295 | } /* POP_FAILURE_POINT */ | |
1296 | ||
1297 | ||
1298 | \f | |
1299 | /* Structure for per-register (a.k.a. per-group) information. | |
1300 | Other register information, such as the | |
1301 | starting and ending positions (which are addresses), and the list of | |
1302 | inner groups (which is a bits list) are maintained in separate | |
1303 | variables. | |
1304 | ||
1305 | We are making a (strictly speaking) nonportable assumption here: that | |
1306 | the compiler will pack our bit fields into something that fits into | |
1307 | the type of `word', i.e., is something that fits into one item on the | |
1308 | failure stack. */ | |
1309 | ||
1310 | typedef union | |
1311 | { | |
1312 | fail_stack_elt_t word; | |
1313 | struct | |
1314 | { | |
1315 | /* This field is one if this group can match the empty string, | |
1316 | zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ | |
1317 | #define MATCH_NULL_UNSET_VALUE 3 | |
1318 | unsigned match_null_string_p : 2; | |
1319 | unsigned is_active : 1; | |
1320 | unsigned matched_something : 1; | |
1321 | unsigned ever_matched_something : 1; | |
1322 | } bits; | |
1323 | } register_info_type; | |
1324 | ||
1325 | #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) | |
1326 | #define IS_ACTIVE(R) ((R).bits.is_active) | |
1327 | #define MATCHED_SOMETHING(R) ((R).bits.matched_something) | |
1328 | #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) | |
1329 | ||
1330 | ||
1331 | /* Call this when have matched a real character; it sets `matched' flags | |
1332 | for the subexpressions which we are currently inside. Also records | |
1333 | that those subexprs have matched. */ | |
1334 | #define SET_REGS_MATCHED() \ | |
1335 | do \ | |
1336 | { \ | |
1337 | if (!set_regs_matched_done) \ | |
1338 | { \ | |
1339 | unsigned r; \ | |
1340 | set_regs_matched_done = 1; \ | |
1341 | for (r = lowest_active_reg; r <= highest_active_reg; r++) \ | |
1342 | { \ | |
1343 | MATCHED_SOMETHING (reg_info[r]) \ | |
1344 | = EVER_MATCHED_SOMETHING (reg_info[r]) \ | |
1345 | = 1; \ | |
1346 | } \ | |
1347 | } \ | |
1348 | } \ | |
1349 | while (0) | |
1350 | ||
1351 | /* Registers are set to a sentinel when they haven't yet matched. */ | |
1352 | static char reg_unset_dummy; | |
1353 | #define REG_UNSET_VALUE (®_unset_dummy) | |
1354 | #define REG_UNSET(e) ((e) == REG_UNSET_VALUE) | |
1355 | \f | |
1356 | /* Subroutine declarations and macros for regex_compile. */ | |
1357 | ||
1358 | static void store_op1 (), store_op2 (); | |
1359 | static void insert_op1 (), insert_op2 (); | |
1360 | static boolean at_begline_loc_p (), at_endline_loc_p (); | |
1361 | static boolean group_in_compile_stack (); | |
1362 | static reg_errcode_t compile_range (); | |
1363 | ||
1364 | /* Fetch the next character in the uncompiled pattern---translating it | |
1365 | if necessary. Also cast from a signed character in the constant | |
1366 | string passed to us by the user to an unsigned char that we can use | |
1367 | as an array index (in, e.g., `translate'). */ | |
1368 | #define PATFETCH(c) \ | |
1369 | do {if (p == pend) return REG_EEND; \ | |
1370 | c = (unsigned char) *p++; \ | |
1371 | if (translate) c = translate[c]; \ | |
1372 | } while (0) | |
1373 | ||
1374 | /* Fetch the next character in the uncompiled pattern, with no | |
1375 | translation. */ | |
1376 | #define PATFETCH_RAW(c) \ | |
1377 | do {if (p == pend) return REG_EEND; \ | |
1378 | c = (unsigned char) *p++; \ | |
1379 | } while (0) | |
1380 | ||
1381 | /* Go backwards one character in the pattern. */ | |
1382 | #define PATUNFETCH p-- | |
1383 | ||
1384 | ||
1385 | /* If `translate' is non-null, return translate[D], else just D. We | |
1386 | cast the subscript to translate because some data is declared as | |
1387 | `char *', to avoid warnings when a string constant is passed. But | |
1388 | when we use a character as a subscript we must make it unsigned. */ | |
1389 | #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d)) | |
1390 | ||
1391 | ||
1392 | /* Macros for outputting the compiled pattern into `buffer'. */ | |
1393 | ||
1394 | /* If the buffer isn't allocated when it comes in, use this. */ | |
1395 | #define INIT_BUF_SIZE 32 | |
1396 | ||
1397 | /* Make sure we have at least N more bytes of space in buffer. */ | |
1398 | #define GET_BUFFER_SPACE(n) \ | |
1399 | while (b - bufp->buffer + (n) > bufp->allocated) \ | |
1400 | EXTEND_BUFFER () | |
1401 | ||
1402 | /* Make sure we have one more byte of buffer space and then add C to it. */ | |
1403 | #define BUF_PUSH(c) \ | |
1404 | do { \ | |
1405 | GET_BUFFER_SPACE (1); \ | |
1406 | *b++ = (unsigned char) (c); \ | |
1407 | } while (0) | |
1408 | ||
1409 | ||
1410 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
1411 | #define BUF_PUSH_2(c1, c2) \ | |
1412 | do { \ | |
1413 | GET_BUFFER_SPACE (2); \ | |
1414 | *b++ = (unsigned char) (c1); \ | |
1415 | *b++ = (unsigned char) (c2); \ | |
1416 | } while (0) | |
1417 | ||
1418 | ||
1419 | /* As with BUF_PUSH_2, except for three bytes. */ | |
1420 | #define BUF_PUSH_3(c1, c2, c3) \ | |
1421 | do { \ | |
1422 | GET_BUFFER_SPACE (3); \ | |
1423 | *b++ = (unsigned char) (c1); \ | |
1424 | *b++ = (unsigned char) (c2); \ | |
1425 | *b++ = (unsigned char) (c3); \ | |
1426 | } while (0) | |
1427 | ||
1428 | ||
1429 | /* Store a jump with opcode OP at LOC to location TO. We store a | |
1430 | relative address offset by the three bytes the jump itself occupies. */ | |
1431 | #define STORE_JUMP(op, loc, to) \ | |
1432 | store_op1 (op, loc, (to) - (loc) - 3) | |
1433 | ||
1434 | /* Likewise, for a two-argument jump. */ | |
1435 | #define STORE_JUMP2(op, loc, to, arg) \ | |
1436 | store_op2 (op, loc, (to) - (loc) - 3, arg) | |
1437 | ||
1438 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ | |
1439 | #define INSERT_JUMP(op, loc, to) \ | |
1440 | insert_op1 (op, loc, (to) - (loc) - 3, b) | |
1441 | ||
1442 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
1443 | #define INSERT_JUMP2(op, loc, to, arg) \ | |
1444 | insert_op2 (op, loc, (to) - (loc) - 3, arg, b) | |
1445 | ||
1446 | ||
1447 | /* This is not an arbitrary limit: the arguments which represent offsets | |
1448 | into the pattern are two bytes long. So if 2^16 bytes turns out to | |
1449 | be too small, many things would have to change. */ | |
1450 | #define MAX_BUF_SIZE (1L << 16) | |
1451 | ||
1452 | ||
1453 | /* Extend the buffer by twice its current size via realloc and | |
1454 | reset the pointers that pointed into the old block to point to the | |
1455 | correct places in the new one. If extending the buffer results in it | |
1456 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ | |
1457 | #define EXTEND_BUFFER() \ | |
1458 | do { \ | |
1459 | unsigned char *old_buffer = bufp->buffer; \ | |
1460 | if (bufp->allocated == MAX_BUF_SIZE) \ | |
1461 | return REG_ESIZE; \ | |
1462 | bufp->allocated <<= 1; \ | |
1463 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
1464 | bufp->allocated = MAX_BUF_SIZE; \ | |
1465 | bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\ | |
1466 | if (bufp->buffer == NULL) \ | |
1467 | return REG_ESPACE; \ | |
1468 | /* If the buffer moved, move all the pointers into it. */ \ | |
1469 | if (old_buffer != bufp->buffer) \ | |
1470 | { \ | |
1471 | b = (b - old_buffer) + bufp->buffer; \ | |
1472 | begalt = (begalt - old_buffer) + bufp->buffer; \ | |
1473 | if (fixup_alt_jump) \ | |
1474 | fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\ | |
1475 | if (laststart) \ | |
1476 | laststart = (laststart - old_buffer) + bufp->buffer; \ | |
1477 | if (pending_exact) \ | |
1478 | pending_exact = (pending_exact - old_buffer) + bufp->buffer; \ | |
1479 | } \ | |
1480 | } while (0) | |
1481 | ||
1482 | ||
1483 | /* Since we have one byte reserved for the register number argument to | |
1484 | {start,stop}_memory, the maximum number of groups we can report | |
1485 | things about is what fits in that byte. */ | |
1486 | #define MAX_REGNUM 255 | |
1487 | ||
1488 | /* But patterns can have more than `MAX_REGNUM' registers. We just | |
1489 | ignore the excess. */ | |
1490 | typedef unsigned regnum_t; | |
1491 | ||
1492 | ||
1493 | /* Macros for the compile stack. */ | |
1494 | ||
1495 | /* Since offsets can go either forwards or backwards, this type needs to | |
1496 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ | |
1497 | typedef int pattern_offset_t; | |
1498 | ||
1499 | typedef struct | |
1500 | { | |
1501 | pattern_offset_t begalt_offset; | |
1502 | pattern_offset_t fixup_alt_jump; | |
1503 | pattern_offset_t inner_group_offset; | |
1504 | pattern_offset_t laststart_offset; | |
1505 | regnum_t regnum; | |
1506 | } compile_stack_elt_t; | |
1507 | ||
1508 | ||
1509 | typedef struct | |
1510 | { | |
1511 | compile_stack_elt_t *stack; | |
1512 | unsigned size; | |
1513 | unsigned avail; /* Offset of next open position. */ | |
1514 | } compile_stack_type; | |
1515 | ||
1516 | ||
1517 | #define INIT_COMPILE_STACK_SIZE 32 | |
1518 | ||
1519 | #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
1520 | #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
1521 | ||
1522 | /* The next available element. */ | |
1523 | #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) | |
1524 | ||
1525 | ||
1526 | /* Set the bit for character C in a list. */ | |
1527 | #define SET_LIST_BIT(c) \ | |
1528 | (b[((unsigned char) (c)) / BYTEWIDTH] \ | |
1529 | |= 1 << (((unsigned char) c) % BYTEWIDTH)) | |
1530 | ||
1531 | ||
1532 | /* Get the next unsigned number in the uncompiled pattern. */ | |
1533 | #define GET_UNSIGNED_NUMBER(num) \ | |
1534 | { if (p != pend) \ | |
1535 | { \ | |
1536 | PATFETCH (c); \ | |
1537 | while (ISDIGIT (c)) \ | |
1538 | { \ | |
1539 | if (num < 0) \ | |
1540 | num = 0; \ | |
1541 | num = num * 10 + c - '0'; \ | |
1542 | if (p == pend) \ | |
1543 | break; \ | |
1544 | PATFETCH (c); \ | |
1545 | } \ | |
1546 | } \ | |
1547 | } | |
1548 | ||
1549 | #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ | |
1550 | ||
1551 | #define IS_CHAR_CLASS(string) \ | |
1552 | (STREQ (string, "alpha") || STREQ (string, "upper") \ | |
1553 | || STREQ (string, "lower") || STREQ (string, "digit") \ | |
1554 | || STREQ (string, "alnum") || STREQ (string, "xdigit") \ | |
1555 | || STREQ (string, "space") || STREQ (string, "print") \ | |
1556 | || STREQ (string, "punct") || STREQ (string, "graph") \ | |
1557 | || STREQ (string, "cntrl") || STREQ (string, "blank")) | |
1558 | \f | |
1559 | #ifndef MATCH_MAY_ALLOCATE | |
1560 | ||
1561 | /* If we cannot allocate large objects within re_match_2_internal, | |
1562 | we make the fail stack and register vectors global. | |
1563 | The fail stack, we grow to the maximum size when a regexp | |
1564 | is compiled. | |
1565 | The register vectors, we adjust in size each time we | |
1566 | compile a regexp, according to the number of registers it needs. */ | |
1567 | ||
1568 | static fail_stack_type fail_stack; | |
1569 | ||
1570 | /* Size with which the following vectors are currently allocated. | |
1571 | That is so we can make them bigger as needed, | |
1572 | but never make them smaller. */ | |
1573 | static int regs_allocated_size; | |
1574 | ||
1575 | static const char ** regstart, ** regend; | |
1576 | static const char ** old_regstart, ** old_regend; | |
1577 | static const char **best_regstart, **best_regend; | |
1578 | static register_info_type *reg_info; | |
1579 | static const char **reg_dummy; | |
1580 | static register_info_type *reg_info_dummy; | |
1581 | ||
1582 | /* Make the register vectors big enough for NUM_REGS registers, | |
1583 | but don't make them smaller. */ | |
1584 | ||
1585 | static | |
1586 | regex_grow_registers (num_regs) | |
1587 | int num_regs; | |
1588 | { | |
1589 | if (num_regs > regs_allocated_size) | |
1590 | { | |
1591 | RETALLOC_IF (regstart, num_regs, const char *); | |
1592 | RETALLOC_IF (regend, num_regs, const char *); | |
1593 | RETALLOC_IF (old_regstart, num_regs, const char *); | |
1594 | RETALLOC_IF (old_regend, num_regs, const char *); | |
1595 | RETALLOC_IF (best_regstart, num_regs, const char *); | |
1596 | RETALLOC_IF (best_regend, num_regs, const char *); | |
1597 | RETALLOC_IF (reg_info, num_regs, register_info_type); | |
1598 | RETALLOC_IF (reg_dummy, num_regs, const char *); | |
1599 | RETALLOC_IF (reg_info_dummy, num_regs, register_info_type); | |
1600 | ||
1601 | regs_allocated_size = num_regs; | |
1602 | } | |
1603 | } | |
1604 | ||
1605 | #endif /* not MATCH_MAY_ALLOCATE */ | |
1606 | \f | |
1607 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. | |
1608 | Returns one of error codes defined in `regex.h', or zero for success. | |
1609 | ||
1610 | Assumes the `allocated' (and perhaps `buffer') and `translate' | |
1611 | fields are set in BUFP on entry. | |
1612 | ||
1613 | If it succeeds, results are put in BUFP (if it returns an error, the | |
1614 | contents of BUFP are undefined): | |
1615 | `buffer' is the compiled pattern; | |
1616 | `syntax' is set to SYNTAX; | |
1617 | `used' is set to the length of the compiled pattern; | |
1618 | `fastmap_accurate' is zero; | |
1619 | `re_nsub' is the number of subexpressions in PATTERN; | |
1620 | `not_bol' and `not_eol' are zero; | |
1621 | ||
1622 | The `fastmap' and `newline_anchor' fields are neither | |
1623 | examined nor set. */ | |
1624 | ||
1625 | /* Return, freeing storage we allocated. */ | |
1626 | #define FREE_STACK_RETURN(value) \ | |
1627 | return (free (compile_stack.stack), value) | |
1628 | ||
1629 | static reg_errcode_t | |
1630 | regex_compile (pattern, size, syntax, bufp) | |
1631 | const char *pattern; | |
1632 | int size; | |
1633 | reg_syntax_t syntax; | |
1634 | struct re_pattern_buffer *bufp; | |
1635 | { | |
1636 | /* We fetch characters from PATTERN here. Even though PATTERN is | |
1637 | `char *' (i.e., signed), we declare these variables as unsigned, so | |
1638 | they can be reliably used as array indices. */ | |
1639 | register unsigned char c, c1; | |
1640 | ||
1641 | /* A random temporary spot in PATTERN. */ | |
1642 | const char *p1; | |
1643 | ||
1644 | /* Points to the end of the buffer, where we should append. */ | |
1645 | register unsigned char *b; | |
1646 | ||
1647 | /* Keeps track of unclosed groups. */ | |
1648 | compile_stack_type compile_stack; | |
1649 | ||
1650 | /* Points to the current (ending) position in the pattern. */ | |
1651 | const char *p = pattern; | |
1652 | const char *pend = pattern + size; | |
1653 | ||
1654 | /* How to translate the characters in the pattern. */ | |
1655 | char *translate = bufp->translate; | |
1656 | ||
1657 | /* Address of the count-byte of the most recently inserted `exactn' | |
1658 | command. This makes it possible to tell if a new exact-match | |
1659 | character can be added to that command or if the character requires | |
1660 | a new `exactn' command. */ | |
1661 | unsigned char *pending_exact = 0; | |
1662 | ||
1663 | /* Address of start of the most recently finished expression. | |
1664 | This tells, e.g., postfix * where to find the start of its | |
1665 | operand. Reset at the beginning of groups and alternatives. */ | |
1666 | unsigned char *laststart = 0; | |
1667 | ||
1668 | /* Address of beginning of regexp, or inside of last group. */ | |
1669 | unsigned char *begalt; | |
1670 | ||
1671 | /* Place in the uncompiled pattern (i.e., the {) to | |
1672 | which to go back if the interval is invalid. */ | |
1673 | const char *beg_interval; | |
1674 | ||
1675 | /* Address of the place where a forward jump should go to the end of | |
1676 | the containing expression. Each alternative of an `or' -- except the | |
1677 | last -- ends with a forward jump of this sort. */ | |
1678 | unsigned char *fixup_alt_jump = 0; | |
1679 | ||
1680 | /* Counts open-groups as they are encountered. Remembered for the | |
1681 | matching close-group on the compile stack, so the same register | |
1682 | number is put in the stop_memory as the start_memory. */ | |
1683 | regnum_t regnum = 0; | |
1684 | ||
1685 | #ifdef DEBUG | |
1686 | DEBUG_PRINT1 ("\nCompiling pattern: "); | |
1687 | if (debug) | |
1688 | { | |
1689 | unsigned debug_count; | |
1690 | ||
1691 | for (debug_count = 0; debug_count < size; debug_count++) | |
1692 | putchar (pattern[debug_count]); | |
1693 | putchar ('\n'); | |
1694 | } | |
1695 | #endif /* DEBUG */ | |
1696 | ||
1697 | /* Initialize the compile stack. */ | |
1698 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
1699 | if (compile_stack.stack == NULL) | |
1700 | return REG_ESPACE; | |
1701 | ||
1702 | compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
1703 | compile_stack.avail = 0; | |
1704 | ||
1705 | /* Initialize the pattern buffer. */ | |
1706 | bufp->syntax = syntax; | |
1707 | bufp->fastmap_accurate = 0; | |
1708 | bufp->not_bol = bufp->not_eol = 0; | |
1709 | ||
1710 | /* Set `used' to zero, so that if we return an error, the pattern | |
1711 | printer (for debugging) will think there's no pattern. We reset it | |
1712 | at the end. */ | |
1713 | bufp->used = 0; | |
1714 | ||
1715 | /* Always count groups, whether or not bufp->no_sub is set. */ | |
1716 | bufp->re_nsub = 0; | |
1717 | ||
1718 | #if !defined (emacs) && !defined (SYNTAX_TABLE) | |
1719 | /* Initialize the syntax table. */ | |
1720 | init_syntax_once (); | |
1721 | #endif | |
1722 | ||
1723 | if (bufp->allocated == 0) | |
1724 | { | |
1725 | if (bufp->buffer) | |
1726 | { /* If zero allocated, but buffer is non-null, try to realloc | |
1727 | enough space. This loses if buffer's address is bogus, but | |
1728 | that is the user's responsibility. */ | |
1729 | RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); | |
1730 | } | |
1731 | else | |
1732 | { /* Caller did not allocate a buffer. Do it for them. */ | |
1733 | bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); | |
1734 | } | |
1735 | if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE); | |
1736 | ||
1737 | bufp->allocated = INIT_BUF_SIZE; | |
1738 | } | |
1739 | ||
1740 | begalt = b = bufp->buffer; | |
1741 | ||
1742 | /* Loop through the uncompiled pattern until we're at the end. */ | |
1743 | while (p != pend) | |
1744 | { | |
1745 | PATFETCH (c); | |
1746 | ||
1747 | switch (c) | |
1748 | { | |
1749 | case '^': | |
1750 | { | |
1751 | if ( /* If at start of pattern, it's an operator. */ | |
1752 | p == pattern + 1 | |
1753 | /* If context independent, it's an operator. */ | |
1754 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
1755 | /* Otherwise, depends on what's come before. */ | |
1756 | || at_begline_loc_p (pattern, p, syntax)) | |
1757 | BUF_PUSH (begline); | |
1758 | else | |
1759 | goto normal_char; | |
1760 | } | |
1761 | break; | |
1762 | ||
1763 | ||
1764 | case '$': | |
1765 | { | |
1766 | if ( /* If at end of pattern, it's an operator. */ | |
1767 | p == pend | |
1768 | /* If context independent, it's an operator. */ | |
1769 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
1770 | /* Otherwise, depends on what's next. */ | |
1771 | || at_endline_loc_p (p, pend, syntax)) | |
1772 | BUF_PUSH (endline); | |
1773 | else | |
1774 | goto normal_char; | |
1775 | } | |
1776 | break; | |
1777 | ||
1778 | ||
1779 | case '+': | |
1780 | case '?': | |
1781 | if ((syntax & RE_BK_PLUS_QM) | |
1782 | || (syntax & RE_LIMITED_OPS)) | |
1783 | goto normal_char; | |
1784 | handle_plus: | |
1785 | case '*': | |
1786 | /* If there is no previous pattern... */ | |
1787 | if (!laststart) | |
1788 | { | |
1789 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
1790 | FREE_STACK_RETURN (REG_BADRPT); | |
1791 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
1792 | goto normal_char; | |
1793 | } | |
1794 | ||
1795 | { | |
1796 | /* Are we optimizing this jump? */ | |
1797 | boolean keep_string_p = false; | |
1798 | ||
1799 | /* 1 means zero (many) matches is allowed. */ | |
1800 | char zero_times_ok = 0, many_times_ok = 0; | |
1801 | ||
1802 | /* If there is a sequence of repetition chars, collapse it | |
1803 | down to just one (the right one). We can't combine | |
1804 | interval operators with these because of, e.g., `a{2}*', | |
1805 | which should only match an even number of `a's. */ | |
1806 | ||
1807 | for (;;) | |
1808 | { | |
1809 | zero_times_ok |= c != '+'; | |
1810 | many_times_ok |= c != '?'; | |
1811 | ||
1812 | if (p == pend) | |
1813 | break; | |
1814 | ||
1815 | PATFETCH (c); | |
1816 | ||
1817 | if (c == '*' | |
1818 | || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) | |
1819 | ; | |
1820 | ||
1821 | else if (syntax & RE_BK_PLUS_QM && c == '\\') | |
1822 | { | |
1823 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
1824 | ||
1825 | PATFETCH (c1); | |
1826 | if (!(c1 == '+' || c1 == '?')) | |
1827 | { | |
1828 | PATUNFETCH; | |
1829 | PATUNFETCH; | |
1830 | break; | |
1831 | } | |
1832 | ||
1833 | c = c1; | |
1834 | } | |
1835 | else | |
1836 | { | |
1837 | PATUNFETCH; | |
1838 | break; | |
1839 | } | |
1840 | ||
1841 | /* If we get here, we found another repeat character. */ | |
1842 | } | |
1843 | ||
1844 | /* Star, etc. applied to an empty pattern is equivalent | |
1845 | to an empty pattern. */ | |
1846 | if (!laststart) | |
1847 | break; | |
1848 | ||
1849 | /* Now we know whether or not zero matches is allowed | |
1850 | and also whether or not two or more matches is allowed. */ | |
1851 | if (many_times_ok) | |
1852 | { /* More than one repetition is allowed, so put in at the | |
1853 | end a backward relative jump from `b' to before the next | |
1854 | jump we're going to put in below (which jumps from | |
1855 | laststart to after this jump). | |
1856 | ||
1857 | But if we are at the `*' in the exact sequence `.*\n', | |
1858 | insert an unconditional jump backwards to the ., | |
1859 | instead of the beginning of the loop. This way we only | |
1860 | push a failure point once, instead of every time | |
1861 | through the loop. */ | |
1862 | assert (p - 1 > pattern); | |
1863 | ||
1864 | /* Allocate the space for the jump. */ | |
1865 | GET_BUFFER_SPACE (3); | |
1866 | ||
1867 | /* We know we are not at the first character of the pattern, | |
1868 | because laststart was nonzero. And we've already | |
1869 | incremented `p', by the way, to be the character after | |
1870 | the `*'. Do we have to do something analogous here | |
1871 | for null bytes, because of RE_DOT_NOT_NULL? */ | |
1872 | if (TRANSLATE (*(p - 2)) == TRANSLATE ('.') | |
1873 | && zero_times_ok | |
1874 | && p < pend && TRANSLATE (*p) == TRANSLATE ('\n') | |
1875 | && !(syntax & RE_DOT_NEWLINE)) | |
1876 | { /* We have .*\n. */ | |
1877 | STORE_JUMP (jump, b, laststart); | |
1878 | keep_string_p = true; | |
1879 | } | |
1880 | else | |
1881 | /* Anything else. */ | |
1882 | STORE_JUMP (maybe_pop_jump, b, laststart - 3); | |
1883 | ||
1884 | /* We've added more stuff to the buffer. */ | |
1885 | b += 3; | |
1886 | } | |
1887 | ||
1888 | /* On failure, jump from laststart to b + 3, which will be the | |
1889 | end of the buffer after this jump is inserted. */ | |
1890 | GET_BUFFER_SPACE (3); | |
1891 | INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump | |
1892 | : on_failure_jump, | |
1893 | laststart, b + 3); | |
1894 | pending_exact = 0; | |
1895 | b += 3; | |
1896 | ||
1897 | if (!zero_times_ok) | |
1898 | { | |
1899 | /* At least one repetition is required, so insert a | |
1900 | `dummy_failure_jump' before the initial | |
1901 | `on_failure_jump' instruction of the loop. This | |
1902 | effects a skip over that instruction the first time | |
1903 | we hit that loop. */ | |
1904 | GET_BUFFER_SPACE (3); | |
1905 | INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6); | |
1906 | b += 3; | |
1907 | } | |
1908 | } | |
1909 | break; | |
1910 | ||
1911 | ||
1912 | case '.': | |
1913 | laststart = b; | |
1914 | BUF_PUSH (anychar); | |
1915 | break; | |
1916 | ||
1917 | ||
1918 | case '[': | |
1919 | { | |
1920 | boolean had_char_class = false; | |
1921 | ||
1922 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
1923 | ||
1924 | /* Ensure that we have enough space to push a charset: the | |
1925 | opcode, the length count, and the bitset; 34 bytes in all. */ | |
1926 | GET_BUFFER_SPACE (34); | |
1927 | ||
1928 | laststart = b; | |
1929 | ||
1930 | /* We test `*p == '^' twice, instead of using an if | |
1931 | statement, so we only need one BUF_PUSH. */ | |
1932 | BUF_PUSH (*p == '^' ? charset_not : charset); | |
1933 | if (*p == '^') | |
1934 | p++; | |
1935 | ||
1936 | /* Remember the first position in the bracket expression. */ | |
1937 | p1 = p; | |
1938 | ||
1939 | /* Push the number of bytes in the bitmap. */ | |
1940 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
1941 | ||
1942 | /* Clear the whole map. */ | |
1943 | bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); | |
1944 | ||
1945 | /* charset_not matches newline according to a syntax bit. */ | |
1946 | if ((re_opcode_t) b[-2] == charset_not | |
1947 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
1948 | SET_LIST_BIT ('\n'); | |
1949 | ||
1950 | /* Read in characters and ranges, setting map bits. */ | |
1951 | for (;;) | |
1952 | { | |
1953 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
1954 | ||
1955 | PATFETCH (c); | |
1956 | ||
1957 | /* \ might escape characters inside [...] and [^...]. */ | |
1958 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
1959 | { | |
1960 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
1961 | ||
1962 | PATFETCH (c1); | |
1963 | SET_LIST_BIT (c1); | |
1964 | continue; | |
1965 | } | |
1966 | ||
1967 | /* Could be the end of the bracket expression. If it's | |
1968 | not (i.e., when the bracket expression is `[]' so | |
1969 | far), the ']' character bit gets set way below. */ | |
1970 | if (c == ']' && p != p1 + 1) | |
1971 | break; | |
1972 | ||
1973 | /* Look ahead to see if it's a range when the last thing | |
1974 | was a character class. */ | |
1975 | if (had_char_class && c == '-' && *p != ']') | |
1976 | FREE_STACK_RETURN (REG_ERANGE); | |
1977 | ||
1978 | /* Look ahead to see if it's a range when the last thing | |
1979 | was a character: if this is a hyphen not at the | |
1980 | beginning or the end of a list, then it's the range | |
1981 | operator. */ | |
1982 | if (c == '-' | |
1983 | && !(p - 2 >= pattern && p[-2] == '[') | |
1984 | && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') | |
1985 | && *p != ']') | |
1986 | { | |
1987 | reg_errcode_t ret | |
1988 | = compile_range (&p, pend, translate, syntax, b); | |
1989 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); | |
1990 | } | |
1991 | ||
1992 | else if (p[0] == '-' && p[1] != ']') | |
1993 | { /* This handles ranges made up of characters only. */ | |
1994 | reg_errcode_t ret; | |
1995 | ||
1996 | /* Move past the `-'. */ | |
1997 | PATFETCH (c1); | |
1998 | ||
1999 | ret = compile_range (&p, pend, translate, syntax, b); | |
2000 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); | |
2001 | } | |
2002 | ||
2003 | /* See if we're at the beginning of a possible character | |
2004 | class. */ | |
2005 | ||
2006 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
2007 | { /* Leave room for the null. */ | |
2008 | char str[CHAR_CLASS_MAX_LENGTH + 1]; | |
2009 | ||
2010 | PATFETCH (c); | |
2011 | c1 = 0; | |
2012 | ||
2013 | /* If pattern is `[[:'. */ | |
2014 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
2015 | ||
2016 | for (;;) | |
2017 | { | |
2018 | PATFETCH (c); | |
2019 | if (c == ':' || c == ']' || p == pend | |
2020 | || c1 == CHAR_CLASS_MAX_LENGTH) | |
2021 | break; | |
2022 | str[c1++] = c; | |
2023 | } | |
2024 | str[c1] = '\0'; | |
2025 | ||
2026 | /* If isn't a word bracketed by `[:' and:`]': | |
2027 | undo the ending character, the letters, and leave | |
2028 | the leading `:' and `[' (but set bits for them). */ | |
2029 | if (c == ':' && *p == ']') | |
2030 | { | |
2031 | int ch; | |
2032 | boolean is_alnum = STREQ (str, "alnum"); | |
2033 | boolean is_alpha = STREQ (str, "alpha"); | |
2034 | boolean is_blank = STREQ (str, "blank"); | |
2035 | boolean is_cntrl = STREQ (str, "cntrl"); | |
2036 | boolean is_digit = STREQ (str, "digit"); | |
2037 | boolean is_graph = STREQ (str, "graph"); | |
2038 | boolean is_lower = STREQ (str, "lower"); | |
2039 | boolean is_print = STREQ (str, "print"); | |
2040 | boolean is_punct = STREQ (str, "punct"); | |
2041 | boolean is_space = STREQ (str, "space"); | |
2042 | boolean is_upper = STREQ (str, "upper"); | |
2043 | boolean is_xdigit = STREQ (str, "xdigit"); | |
2044 | ||
2045 | if (!IS_CHAR_CLASS (str)) | |
2046 | FREE_STACK_RETURN (REG_ECTYPE); | |
2047 | ||
2048 | /* Throw away the ] at the end of the character | |
2049 | class. */ | |
2050 | PATFETCH (c); | |
2051 | ||
2052 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
2053 | ||
2054 | for (ch = 0; ch < 1 << BYTEWIDTH; ch++) | |
2055 | { | |
2056 | /* This was split into 3 if's to | |
2057 | avoid an arbitrary limit in some compiler. */ | |
2058 | if ( (is_alnum && ISALNUM (ch)) | |
2059 | || (is_alpha && ISALPHA (ch)) | |
2060 | || (is_blank && ISBLANK (ch)) | |
2061 | || (is_cntrl && ISCNTRL (ch))) | |
2062 | SET_LIST_BIT (ch); | |
2063 | if ( (is_digit && ISDIGIT (ch)) | |
2064 | || (is_graph && ISGRAPH (ch)) | |
2065 | || (is_lower && ISLOWER (ch)) | |
2066 | || (is_print && ISPRINT (ch))) | |
2067 | SET_LIST_BIT (ch); | |
2068 | if ( (is_punct && ISPUNCT (ch)) | |
2069 | || (is_space && ISSPACE (ch)) | |
2070 | || (is_upper && ISUPPER (ch)) | |
2071 | || (is_xdigit && ISXDIGIT (ch))) | |
2072 | SET_LIST_BIT (ch); | |
2073 | } | |
2074 | had_char_class = true; | |
2075 | } | |
2076 | else | |
2077 | { | |
2078 | c1++; | |
2079 | while (c1--) | |
2080 | PATUNFETCH; | |
2081 | SET_LIST_BIT ('['); | |
2082 | SET_LIST_BIT (':'); | |
2083 | had_char_class = false; | |
2084 | } | |
2085 | } | |
2086 | else | |
2087 | { | |
2088 | had_char_class = false; | |
2089 | SET_LIST_BIT (c); | |
2090 | } | |
2091 | } | |
2092 | ||
2093 | /* Discard any (non)matching list bytes that are all 0 at the | |
2094 | end of the map. Decrease the map-length byte too. */ | |
2095 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) | |
2096 | b[-1]--; | |
2097 | b += b[-1]; | |
2098 | } | |
2099 | break; | |
2100 | ||
2101 | ||
2102 | case '(': | |
2103 | if (syntax & RE_NO_BK_PARENS) | |
2104 | goto handle_open; | |
2105 | else | |
2106 | goto normal_char; | |
2107 | ||
2108 | ||
2109 | case ')': | |
2110 | if (syntax & RE_NO_BK_PARENS) | |
2111 | goto handle_close; | |
2112 | else | |
2113 | goto normal_char; | |
2114 | ||
2115 | ||
2116 | case '\n': | |
2117 | if (syntax & RE_NEWLINE_ALT) | |
2118 | goto handle_alt; | |
2119 | else | |
2120 | goto normal_char; | |
2121 | ||
2122 | ||
2123 | case '|': | |
2124 | if (syntax & RE_NO_BK_VBAR) | |
2125 | goto handle_alt; | |
2126 | else | |
2127 | goto normal_char; | |
2128 | ||
2129 | ||
2130 | case '{': | |
2131 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
2132 | goto handle_interval; | |
2133 | else | |
2134 | goto normal_char; | |
2135 | ||
2136 | ||
2137 | case '\\': | |
2138 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
2139 | ||
2140 | /* Do not translate the character after the \, so that we can | |
2141 | distinguish, e.g., \B from \b, even if we normally would | |
2142 | translate, e.g., B to b. */ | |
2143 | PATFETCH_RAW (c); | |
2144 | ||
2145 | switch (c) | |
2146 | { | |
2147 | case '(': | |
2148 | if (syntax & RE_NO_BK_PARENS) | |
2149 | goto normal_backslash; | |
2150 | ||
2151 | handle_open: | |
2152 | bufp->re_nsub++; | |
2153 | regnum++; | |
2154 | ||
2155 | if (COMPILE_STACK_FULL) | |
2156 | { | |
2157 | RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
2158 | compile_stack_elt_t); | |
2159 | if (compile_stack.stack == NULL) return REG_ESPACE; | |
2160 | ||
2161 | compile_stack.size <<= 1; | |
2162 | } | |
2163 | ||
2164 | /* These are the values to restore when we hit end of this | |
2165 | group. They are all relative offsets, so that if the | |
2166 | whole pattern moves because of realloc, they will still | |
2167 | be valid. */ | |
2168 | COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; | |
2169 | COMPILE_STACK_TOP.fixup_alt_jump | |
2170 | = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; | |
2171 | COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; | |
2172 | COMPILE_STACK_TOP.regnum = regnum; | |
2173 | ||
2174 | /* We will eventually replace the 0 with the number of | |
2175 | groups inner to this one. But do not push a | |
2176 | start_memory for groups beyond the last one we can | |
2177 | represent in the compiled pattern. */ | |
2178 | if (regnum <= MAX_REGNUM) | |
2179 | { | |
2180 | COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2; | |
2181 | BUF_PUSH_3 (start_memory, regnum, 0); | |
2182 | } | |
2183 | ||
2184 | compile_stack.avail++; | |
2185 | ||
2186 | fixup_alt_jump = 0; | |
2187 | laststart = 0; | |
2188 | begalt = b; | |
2189 | /* If we've reached MAX_REGNUM groups, then this open | |
2190 | won't actually generate any code, so we'll have to | |
2191 | clear pending_exact explicitly. */ | |
2192 | pending_exact = 0; | |
2193 | break; | |
2194 | ||
2195 | ||
2196 | case ')': | |
2197 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
2198 | ||
2199 | if (COMPILE_STACK_EMPTY) | |
2200 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
2201 | goto normal_backslash; | |
2202 | else | |
2203 | FREE_STACK_RETURN (REG_ERPAREN); | |
2204 | ||
2205 | handle_close: | |
2206 | if (fixup_alt_jump) | |
2207 | { /* Push a dummy failure point at the end of the | |
2208 | alternative for a possible future | |
2209 | `pop_failure_jump' to pop. See comments at | |
2210 | `push_dummy_failure' in `re_match_2'. */ | |
2211 | BUF_PUSH (push_dummy_failure); | |
2212 | ||
2213 | /* We allocated space for this jump when we assigned | |
2214 | to `fixup_alt_jump', in the `handle_alt' case below. */ | |
2215 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); | |
2216 | } | |
2217 | ||
2218 | /* See similar code for backslashed left paren above. */ | |
2219 | if (COMPILE_STACK_EMPTY) | |
2220 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
2221 | goto normal_char; | |
2222 | else | |
2223 | FREE_STACK_RETURN (REG_ERPAREN); | |
2224 | ||
2225 | /* Since we just checked for an empty stack above, this | |
2226 | ``can't happen''. */ | |
2227 | assert (compile_stack.avail != 0); | |
2228 | { | |
2229 | /* We don't just want to restore into `regnum', because | |
2230 | later groups should continue to be numbered higher, | |
2231 | as in `(ab)c(de)' -- the second group is #2. */ | |
2232 | regnum_t this_group_regnum; | |
2233 | ||
2234 | compile_stack.avail--; | |
2235 | begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; | |
2236 | fixup_alt_jump | |
2237 | = COMPILE_STACK_TOP.fixup_alt_jump | |
2238 | ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
2239 | : 0; | |
2240 | laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; | |
2241 | this_group_regnum = COMPILE_STACK_TOP.regnum; | |
2242 | /* If we've reached MAX_REGNUM groups, then this open | |
2243 | won't actually generate any code, so we'll have to | |
2244 | clear pending_exact explicitly. */ | |
2245 | pending_exact = 0; | |
2246 | ||
2247 | /* We're at the end of the group, so now we know how many | |
2248 | groups were inside this one. */ | |
2249 | if (this_group_regnum <= MAX_REGNUM) | |
2250 | { | |
2251 | unsigned char *inner_group_loc | |
2252 | = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset; | |
2253 | ||
2254 | *inner_group_loc = regnum - this_group_regnum; | |
2255 | BUF_PUSH_3 (stop_memory, this_group_regnum, | |
2256 | regnum - this_group_regnum); | |
2257 | } | |
2258 | } | |
2259 | break; | |
2260 | ||
2261 | ||
2262 | case '|': /* `\|'. */ | |
2263 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
2264 | goto normal_backslash; | |
2265 | handle_alt: | |
2266 | if (syntax & RE_LIMITED_OPS) | |
2267 | goto normal_char; | |
2268 | ||
2269 | /* Insert before the previous alternative a jump which | |
2270 | jumps to this alternative if the former fails. */ | |
2271 | GET_BUFFER_SPACE (3); | |
2272 | INSERT_JUMP (on_failure_jump, begalt, b + 6); | |
2273 | pending_exact = 0; | |
2274 | b += 3; | |
2275 | ||
2276 | /* The alternative before this one has a jump after it | |
2277 | which gets executed if it gets matched. Adjust that | |
2278 | jump so it will jump to this alternative's analogous | |
2279 | jump (put in below, which in turn will jump to the next | |
2280 | (if any) alternative's such jump, etc.). The last such | |
2281 | jump jumps to the correct final destination. A picture: | |
2282 | _____ _____ | |
2283 | | | | | | |
2284 | | v | v | |
2285 | a | b | c | |
2286 | ||
2287 | If we are at `b', then fixup_alt_jump right now points to a | |
2288 | three-byte space after `a'. We'll put in the jump, set | |
2289 | fixup_alt_jump to right after `b', and leave behind three | |
2290 | bytes which we'll fill in when we get to after `c'. */ | |
2291 | ||
2292 | if (fixup_alt_jump) | |
2293 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
2294 | ||
2295 | /* Mark and leave space for a jump after this alternative, | |
2296 | to be filled in later either by next alternative or | |
2297 | when know we're at the end of a series of alternatives. */ | |
2298 | fixup_alt_jump = b; | |
2299 | GET_BUFFER_SPACE (3); | |
2300 | b += 3; | |
2301 | ||
2302 | laststart = 0; | |
2303 | begalt = b; | |
2304 | break; | |
2305 | ||
2306 | ||
2307 | case '{': | |
2308 | /* If \{ is a literal. */ | |
2309 | if (!(syntax & RE_INTERVALS) | |
2310 | /* If we're at `\{' and it's not the open-interval | |
2311 | operator. */ | |
2312 | || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES)) | |
2313 | || (p - 2 == pattern && p == pend)) | |
2314 | goto normal_backslash; | |
2315 | ||
2316 | handle_interval: | |
2317 | { | |
2318 | /* If got here, then the syntax allows intervals. */ | |
2319 | ||
2320 | /* At least (most) this many matches must be made. */ | |
2321 | int lower_bound = -1, upper_bound = -1; | |
2322 | ||
2323 | beg_interval = p - 1; | |
2324 | ||
2325 | if (p == pend) | |
2326 | { | |
2327 | if (syntax & RE_NO_BK_BRACES) | |
2328 | goto unfetch_interval; | |
2329 | else | |
2330 | FREE_STACK_RETURN (REG_EBRACE); | |
2331 | } | |
2332 | ||
2333 | GET_UNSIGNED_NUMBER (lower_bound); | |
2334 | ||
2335 | if (c == ',') | |
2336 | { | |
2337 | GET_UNSIGNED_NUMBER (upper_bound); | |
2338 | if (upper_bound < 0) upper_bound = RE_DUP_MAX; | |
2339 | } | |
2340 | else | |
2341 | /* Interval such as `{1}' => match exactly once. */ | |
2342 | upper_bound = lower_bound; | |
2343 | ||
2344 | if (lower_bound < 0 || upper_bound > RE_DUP_MAX | |
2345 | || lower_bound > upper_bound) | |
2346 | { | |
2347 | if (syntax & RE_NO_BK_BRACES) | |
2348 | goto unfetch_interval; | |
2349 | else | |
2350 | FREE_STACK_RETURN (REG_BADBR); | |
2351 | } | |
2352 | ||
2353 | if (!(syntax & RE_NO_BK_BRACES)) | |
2354 | { | |
2355 | if (c != '\\') FREE_STACK_RETURN (REG_EBRACE); | |
2356 | ||
2357 | PATFETCH (c); | |
2358 | } | |
2359 | ||
2360 | if (c != '}') | |
2361 | { | |
2362 | if (syntax & RE_NO_BK_BRACES) | |
2363 | goto unfetch_interval; | |
2364 | else | |
2365 | FREE_STACK_RETURN (REG_BADBR); | |
2366 | } | |
2367 | ||
2368 | /* We just parsed a valid interval. */ | |
2369 | ||
2370 | /* If it's invalid to have no preceding re. */ | |
2371 | if (!laststart) | |
2372 | { | |
2373 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
2374 | FREE_STACK_RETURN (REG_BADRPT); | |
2375 | else if (syntax & RE_CONTEXT_INDEP_OPS) | |
2376 | laststart = b; | |
2377 | else | |
2378 | goto unfetch_interval; | |
2379 | } | |
2380 | ||
2381 | /* If the upper bound is zero, don't want to succeed at | |
2382 | all; jump from `laststart' to `b + 3', which will be | |
2383 | the end of the buffer after we insert the jump. */ | |
2384 | if (upper_bound == 0) | |
2385 | { | |
2386 | GET_BUFFER_SPACE (3); | |
2387 | INSERT_JUMP (jump, laststart, b + 3); | |
2388 | b += 3; | |
2389 | } | |
2390 | ||
2391 | /* Otherwise, we have a nontrivial interval. When | |
2392 | we're all done, the pattern will look like: | |
2393 | set_number_at <jump count> <upper bound> | |
2394 | set_number_at <succeed_n count> <lower bound> | |
2395 | succeed_n <after jump addr> <succeed_n count> | |
2396 | <body of loop> | |
2397 | jump_n <succeed_n addr> <jump count> | |
2398 | (The upper bound and `jump_n' are omitted if | |
2399 | `upper_bound' is 1, though.) */ | |
2400 | else | |
2401 | { /* If the upper bound is > 1, we need to insert | |
2402 | more at the end of the loop. */ | |
2403 | unsigned nbytes = 10 + (upper_bound > 1) * 10; | |
2404 | ||
2405 | GET_BUFFER_SPACE (nbytes); | |
2406 | ||
2407 | /* Initialize lower bound of the `succeed_n', even | |
2408 | though it will be set during matching by its | |
2409 | attendant `set_number_at' (inserted next), | |
2410 | because `re_compile_fastmap' needs to know. | |
2411 | Jump to the `jump_n' we might insert below. */ | |
2412 | INSERT_JUMP2 (succeed_n, laststart, | |
2413 | b + 5 + (upper_bound > 1) * 5, | |
2414 | lower_bound); | |
2415 | b += 5; | |
2416 | ||
2417 | /* Code to initialize the lower bound. Insert | |
2418 | before the `succeed_n'. The `5' is the last two | |
2419 | bytes of this `set_number_at', plus 3 bytes of | |
2420 | the following `succeed_n'. */ | |
2421 | insert_op2 (set_number_at, laststart, 5, lower_bound, b); | |
2422 | b += 5; | |
2423 | ||
2424 | if (upper_bound > 1) | |
2425 | { /* More than one repetition is allowed, so | |
2426 | append a backward jump to the `succeed_n' | |
2427 | that starts this interval. | |
2428 | ||
2429 | When we've reached this during matching, | |
2430 | we'll have matched the interval once, so | |
2431 | jump back only `upper_bound - 1' times. */ | |
2432 | STORE_JUMP2 (jump_n, b, laststart + 5, | |
2433 | upper_bound - 1); | |
2434 | b += 5; | |
2435 | ||
2436 | /* The location we want to set is the second | |
2437 | parameter of the `jump_n'; that is `b-2' as | |
2438 | an absolute address. `laststart' will be | |
2439 | the `set_number_at' we're about to insert; | |
2440 | `laststart+3' the number to set, the source | |
2441 | for the relative address. But we are | |
2442 | inserting into the middle of the pattern -- | |
2443 | so everything is getting moved up by 5. | |
2444 | Conclusion: (b - 2) - (laststart + 3) + 5, | |
2445 | i.e., b - laststart. | |
2446 | ||
2447 | We insert this at the beginning of the loop | |
2448 | so that if we fail during matching, we'll | |
2449 | reinitialize the bounds. */ | |
2450 | insert_op2 (set_number_at, laststart, b - laststart, | |
2451 | upper_bound - 1, b); | |
2452 | b += 5; | |
2453 | } | |
2454 | } | |
2455 | pending_exact = 0; | |
2456 | beg_interval = NULL; | |
2457 | } | |
2458 | break; | |
2459 | ||
2460 | unfetch_interval: | |
2461 | /* If an invalid interval, match the characters as literals. */ | |
2462 | assert (beg_interval); | |
2463 | p = beg_interval; | |
2464 | beg_interval = NULL; | |
2465 | ||
2466 | /* normal_char and normal_backslash need `c'. */ | |
2467 | PATFETCH (c); | |
2468 | ||
2469 | if (!(syntax & RE_NO_BK_BRACES)) | |
2470 | { | |
2471 | if (p > pattern && p[-1] == '\\') | |
2472 | goto normal_backslash; | |
2473 | } | |
2474 | goto normal_char; | |
2475 | ||
2476 | #ifdef emacs | |
2477 | /* There is no way to specify the before_dot and after_dot | |
2478 | operators. rms says this is ok. --karl */ | |
2479 | case '=': | |
2480 | BUF_PUSH (at_dot); | |
2481 | break; | |
2482 | ||
2483 | case 's': | |
2484 | laststart = b; | |
2485 | PATFETCH (c); | |
2486 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
2487 | break; | |
2488 | ||
2489 | case 'S': | |
2490 | laststart = b; | |
2491 | PATFETCH (c); | |
2492 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
2493 | break; | |
2494 | #endif /* emacs */ | |
2495 | ||
2496 | ||
2497 | case 'w': | |
2498 | laststart = b; | |
2499 | BUF_PUSH (wordchar); | |
2500 | break; | |
2501 | ||
2502 | ||
2503 | case 'W': | |
2504 | laststart = b; | |
2505 | BUF_PUSH (notwordchar); | |
2506 | break; | |
2507 | ||
2508 | ||
2509 | case '<': | |
2510 | BUF_PUSH (wordbeg); | |
2511 | break; | |
2512 | ||
2513 | case '>': | |
2514 | BUF_PUSH (wordend); | |
2515 | break; | |
2516 | ||
2517 | case 'b': | |
2518 | BUF_PUSH (wordbound); | |
2519 | break; | |
2520 | ||
2521 | case 'B': | |
2522 | BUF_PUSH (notwordbound); | |
2523 | break; | |
2524 | ||
2525 | case '`': | |
2526 | BUF_PUSH (begbuf); | |
2527 | break; | |
2528 | ||
2529 | case '\'': | |
2530 | BUF_PUSH (endbuf); | |
2531 | break; | |
2532 | ||
2533 | case '1': case '2': case '3': case '4': case '5': | |
2534 | case '6': case '7': case '8': case '9': | |
2535 | if (syntax & RE_NO_BK_REFS) | |
2536 | goto normal_char; | |
2537 | ||
2538 | c1 = c - '0'; | |
2539 | ||
2540 | if (c1 > regnum) | |
2541 | FREE_STACK_RETURN (REG_ESUBREG); | |
2542 | ||
2543 | /* Can't back reference to a subexpression if inside of it. */ | |
2544 | if (group_in_compile_stack (compile_stack, c1)) | |
2545 | goto normal_char; | |
2546 | ||
2547 | laststart = b; | |
2548 | BUF_PUSH_2 (duplicate, c1); | |
2549 | break; | |
2550 | ||
2551 | ||
2552 | case '+': | |
2553 | case '?': | |
2554 | if (syntax & RE_BK_PLUS_QM) | |
2555 | goto handle_plus; | |
2556 | else | |
2557 | goto normal_backslash; | |
2558 | ||
2559 | default: | |
2560 | normal_backslash: | |
2561 | /* You might think it would be useful for \ to mean | |
2562 | not to translate; but if we don't translate it | |
2563 | it will never match anything. */ | |
2564 | c = TRANSLATE (c); | |
2565 | goto normal_char; | |
2566 | } | |
2567 | break; | |
2568 | ||
2569 | ||
2570 | default: | |
2571 | /* Expects the character in `c'. */ | |
2572 | normal_char: | |
2573 | /* If no exactn currently being built. */ | |
2574 | if (!pending_exact | |
2575 | ||
2576 | /* If last exactn not at current position. */ | |
2577 | || pending_exact + *pending_exact + 1 != b | |
2578 | ||
2579 | /* We have only one byte following the exactn for the count. */ | |
2580 | || *pending_exact == (1 << BYTEWIDTH) - 1 | |
2581 | ||
2582 | /* If followed by a repetition operator. */ | |
2583 | || *p == '*' || *p == '^' | |
2584 | || ((syntax & RE_BK_PLUS_QM) | |
2585 | ? *p == '\\' && (p[1] == '+' || p[1] == '?') | |
2586 | : (*p == '+' || *p == '?')) | |
2587 | || ((syntax & RE_INTERVALS) | |
2588 | && ((syntax & RE_NO_BK_BRACES) | |
2589 | ? *p == '{' | |
2590 | : (p[0] == '\\' && p[1] == '{')))) | |
2591 | { | |
2592 | /* Start building a new exactn. */ | |
2593 | ||
2594 | laststart = b; | |
2595 | ||
2596 | BUF_PUSH_2 (exactn, 0); | |
2597 | pending_exact = b - 1; | |
2598 | } | |
2599 | ||
2600 | BUF_PUSH (c); | |
2601 | (*pending_exact)++; | |
2602 | break; | |
2603 | } /* switch (c) */ | |
2604 | } /* while p != pend */ | |
2605 | ||
2606 | ||
2607 | /* Through the pattern now. */ | |
2608 | ||
2609 | if (fixup_alt_jump) | |
2610 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
2611 | ||
2612 | if (!COMPILE_STACK_EMPTY) | |
2613 | FREE_STACK_RETURN (REG_EPAREN); | |
2614 | ||
2615 | /* If we don't want backtracking, force success | |
2616 | the first time we reach the end of the compiled pattern. */ | |
2617 | if (syntax & RE_NO_POSIX_BACKTRACKING) | |
2618 | BUF_PUSH (succeed); | |
2619 | ||
2620 | free (compile_stack.stack); | |
2621 | ||
2622 | /* We have succeeded; set the length of the buffer. */ | |
2623 | bufp->used = b - bufp->buffer; | |
2624 | ||
2625 | #ifdef DEBUG | |
2626 | if (debug) | |
2627 | { | |
2628 | DEBUG_PRINT1 ("\nCompiled pattern: \n"); | |
2629 | print_compiled_pattern (bufp); | |
2630 | } | |
2631 | #endif /* DEBUG */ | |
2632 | ||
2633 | #ifndef MATCH_MAY_ALLOCATE | |
2634 | /* Initialize the failure stack to the largest possible stack. This | |
2635 | isn't necessary unless we're trying to avoid calling alloca in | |
2636 | the search and match routines. */ | |
2637 | { | |
2638 | int num_regs = bufp->re_nsub + 1; | |
2639 | ||
2640 | /* Since DOUBLE_FAIL_STACK refuses to double only if the current size | |
2641 | is strictly greater than re_max_failures, the largest possible stack | |
2642 | is 2 * re_max_failures failure points. */ | |
2643 | if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS)) | |
2644 | { | |
2645 | fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); | |
2646 | ||
2647 | #ifdef emacs | |
2648 | if (! fail_stack.stack) | |
2649 | fail_stack.stack | |
2650 | = (fail_stack_elt_t *) xmalloc (fail_stack.size | |
2651 | * sizeof (fail_stack_elt_t)); | |
2652 | else | |
2653 | fail_stack.stack | |
2654 | = (fail_stack_elt_t *) xrealloc (fail_stack.stack, | |
2655 | (fail_stack.size | |
2656 | * sizeof (fail_stack_elt_t))); | |
2657 | #else /* not emacs */ | |
2658 | if (! fail_stack.stack) | |
2659 | fail_stack.stack | |
2660 | = (fail_stack_elt_t *) malloc (fail_stack.size | |
2661 | * sizeof (fail_stack_elt_t)); | |
2662 | else | |
2663 | fail_stack.stack | |
2664 | = (fail_stack_elt_t *) realloc (fail_stack.stack, | |
2665 | (fail_stack.size | |
2666 | * sizeof (fail_stack_elt_t))); | |
2667 | #endif /* not emacs */ | |
2668 | } | |
2669 | ||
2670 | regex_grow_registers (num_regs); | |
2671 | } | |
2672 | #endif /* not MATCH_MAY_ALLOCATE */ | |
2673 | ||
2674 | return REG_NOERROR; | |
2675 | } /* regex_compile */ | |
2676 | \f | |
2677 | /* Subroutines for `regex_compile'. */ | |
2678 | ||
2679 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ | |
2680 | ||
2681 | static void | |
2682 | store_op1 (op, loc, arg) | |
2683 | re_opcode_t op; | |
2684 | unsigned char *loc; | |
2685 | int arg; | |
2686 | { | |
2687 | *loc = (unsigned char) op; | |
2688 | STORE_NUMBER (loc + 1, arg); | |
2689 | } | |
2690 | ||
2691 | ||
2692 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
2693 | ||
2694 | static void | |
2695 | store_op2 (op, loc, arg1, arg2) | |
2696 | re_opcode_t op; | |
2697 | unsigned char *loc; | |
2698 | int arg1, arg2; | |
2699 | { | |
2700 | *loc = (unsigned char) op; | |
2701 | STORE_NUMBER (loc + 1, arg1); | |
2702 | STORE_NUMBER (loc + 3, arg2); | |
2703 | } | |
2704 | ||
2705 | ||
2706 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
2707 | for OP followed by two-byte integer parameter ARG. */ | |
2708 | ||
2709 | static void | |
2710 | insert_op1 (op, loc, arg, end) | |
2711 | re_opcode_t op; | |
2712 | unsigned char *loc; | |
2713 | int arg; | |
2714 | unsigned char *end; | |
2715 | { | |
2716 | register unsigned char *pfrom = end; | |
2717 | register unsigned char *pto = end + 3; | |
2718 | ||
2719 | while (pfrom != loc) | |
2720 | *--pto = *--pfrom; | |
2721 | ||
2722 | store_op1 (op, loc, arg); | |
2723 | } | |
2724 | ||
2725 | ||
2726 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
2727 | ||
2728 | static void | |
2729 | insert_op2 (op, loc, arg1, arg2, end) | |
2730 | re_opcode_t op; | |
2731 | unsigned char *loc; | |
2732 | int arg1, arg2; | |
2733 | unsigned char *end; | |
2734 | { | |
2735 | register unsigned char *pfrom = end; | |
2736 | register unsigned char *pto = end + 5; | |
2737 | ||
2738 | while (pfrom != loc) | |
2739 | *--pto = *--pfrom; | |
2740 | ||
2741 | store_op2 (op, loc, arg1, arg2); | |
2742 | } | |
2743 | ||
2744 | ||
2745 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
2746 | after an alternative or a begin-subexpression. We assume there is at | |
2747 | least one character before the ^. */ | |
2748 | ||
2749 | static boolean | |
2750 | at_begline_loc_p (pattern, p, syntax) | |
2751 | const char *pattern, *p; | |
2752 | reg_syntax_t syntax; | |
2753 | { | |
2754 | const char *prev = p - 2; | |
2755 | boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; | |
2756 | ||
2757 | return | |
2758 | /* After a subexpression? */ | |
2759 | (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
2760 | /* After an alternative? */ | |
2761 | || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); | |
2762 | } | |
2763 | ||
2764 | ||
2765 | /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
2766 | at least one character after the $, i.e., `P < PEND'. */ | |
2767 | ||
2768 | static boolean | |
2769 | at_endline_loc_p (p, pend, syntax) | |
2770 | const char *p, *pend; | |
2771 | int syntax; | |
2772 | { | |
2773 | const char *next = p; | |
2774 | boolean next_backslash = *next == '\\'; | |
5bb52971 | 2775 | const char *next_next = p + 1 < pend ? p + 1 : 0; |
fa9a63c5 RM |
2776 | |
2777 | return | |
2778 | /* Before a subexpression? */ | |
2779 | (syntax & RE_NO_BK_PARENS ? *next == ')' | |
2780 | : next_backslash && next_next && *next_next == ')') | |
2781 | /* Before an alternative? */ | |
2782 | || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
2783 | : next_backslash && next_next && *next_next == '|'); | |
2784 | } | |
2785 | ||
2786 | ||
2787 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and | |
2788 | false if it's not. */ | |
2789 | ||
2790 | static boolean | |
2791 | group_in_compile_stack (compile_stack, regnum) | |
2792 | compile_stack_type compile_stack; | |
2793 | regnum_t regnum; | |
2794 | { | |
2795 | int this_element; | |
2796 | ||
2797 | for (this_element = compile_stack.avail - 1; | |
2798 | this_element >= 0; | |
2799 | this_element--) | |
2800 | if (compile_stack.stack[this_element].regnum == regnum) | |
2801 | return true; | |
2802 | ||
2803 | return false; | |
2804 | } | |
2805 | ||
2806 | ||
2807 | /* Read the ending character of a range (in a bracket expression) from the | |
2808 | uncompiled pattern *P_PTR (which ends at PEND). We assume the | |
2809 | starting character is in `P[-2]'. (`P[-1]' is the character `-'.) | |
2810 | Then we set the translation of all bits between the starting and | |
2811 | ending characters (inclusive) in the compiled pattern B. | |
2812 | ||
2813 | Return an error code. | |
2814 | ||
2815 | We use these short variable names so we can use the same macros as | |
2816 | `regex_compile' itself. */ | |
2817 | ||
2818 | static reg_errcode_t | |
2819 | compile_range (p_ptr, pend, translate, syntax, b) | |
2820 | const char **p_ptr, *pend; | |
2821 | char *translate; | |
2822 | reg_syntax_t syntax; | |
2823 | unsigned char *b; | |
2824 | { | |
2825 | unsigned this_char; | |
2826 | ||
2827 | const char *p = *p_ptr; | |
2828 | int range_start, range_end; | |
2829 | ||
2830 | if (p == pend) | |
2831 | return REG_ERANGE; | |
2832 | ||
2833 | /* Even though the pattern is a signed `char *', we need to fetch | |
2834 | with unsigned char *'s; if the high bit of the pattern character | |
2835 | is set, the range endpoints will be negative if we fetch using a | |
2836 | signed char *. | |
2837 | ||
2838 | We also want to fetch the endpoints without translating them; the | |
2839 | appropriate translation is done in the bit-setting loop below. */ | |
2840 | /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */ | |
2841 | range_start = ((const unsigned char *) p)[-2]; | |
2842 | range_end = ((const unsigned char *) p)[0]; | |
2843 | ||
2844 | /* Have to increment the pointer into the pattern string, so the | |
2845 | caller isn't still at the ending character. */ | |
2846 | (*p_ptr)++; | |
2847 | ||
2848 | /* If the start is after the end, the range is empty. */ | |
2849 | if (range_start > range_end) | |
2850 | return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; | |
2851 | ||
2852 | /* Here we see why `this_char' has to be larger than an `unsigned | |
2853 | char' -- the range is inclusive, so if `range_end' == 0xff | |
2854 | (assuming 8-bit characters), we would otherwise go into an infinite | |
2855 | loop, since all characters <= 0xff. */ | |
2856 | for (this_char = range_start; this_char <= range_end; this_char++) | |
2857 | { | |
2858 | SET_LIST_BIT (TRANSLATE (this_char)); | |
2859 | } | |
2860 | ||
2861 | return REG_NOERROR; | |
2862 | } | |
2863 | \f | |
2864 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
2865 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
2866 | characters can start a string that matches the pattern. This fastmap | |
2867 | is used by re_search to skip quickly over impossible starting points. | |
2868 | ||
2869 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
2870 | area as BUFP->fastmap. | |
2871 | ||
2872 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in | |
2873 | the pattern buffer. | |
2874 | ||
2875 | Returns 0 if we succeed, -2 if an internal error. */ | |
2876 | ||
2877 | int | |
2878 | re_compile_fastmap (bufp) | |
2879 | struct re_pattern_buffer *bufp; | |
2880 | { | |
2881 | int j, k; | |
2882 | #ifdef MATCH_MAY_ALLOCATE | |
2883 | fail_stack_type fail_stack; | |
2884 | #endif | |
2885 | #ifndef REGEX_MALLOC | |
2886 | char *destination; | |
2887 | #endif | |
2888 | /* We don't push any register information onto the failure stack. */ | |
2889 | unsigned num_regs = 0; | |
2890 | ||
2891 | register char *fastmap = bufp->fastmap; | |
2892 | unsigned char *pattern = bufp->buffer; | |
2893 | unsigned long size = bufp->used; | |
2894 | unsigned char *p = pattern; | |
2895 | register unsigned char *pend = pattern + size; | |
2896 | ||
2897 | /* This holds the pointer to the failure stack, when | |
2898 | it is allocated relocatably. */ | |
2899 | fail_stack_elt_t *failure_stack_ptr; | |
2900 | ||
2901 | /* Assume that each path through the pattern can be null until | |
2902 | proven otherwise. We set this false at the bottom of switch | |
2903 | statement, to which we get only if a particular path doesn't | |
2904 | match the empty string. */ | |
2905 | boolean path_can_be_null = true; | |
2906 | ||
2907 | /* We aren't doing a `succeed_n' to begin with. */ | |
2908 | boolean succeed_n_p = false; | |
2909 | ||
2910 | assert (fastmap != NULL && p != NULL); | |
2911 | ||
2912 | INIT_FAIL_STACK (); | |
2913 | bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ | |
2914 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ | |
2915 | bufp->can_be_null = 0; | |
2916 | ||
2917 | while (1) | |
2918 | { | |
2919 | if (p == pend || *p == succeed) | |
2920 | { | |
2921 | /* We have reached the (effective) end of pattern. */ | |
2922 | if (!FAIL_STACK_EMPTY ()) | |
2923 | { | |
2924 | bufp->can_be_null |= path_can_be_null; | |
2925 | ||
2926 | /* Reset for next path. */ | |
2927 | path_can_be_null = true; | |
2928 | ||
2929 | p = fail_stack.stack[--fail_stack.avail].pointer; | |
2930 | ||
2931 | continue; | |
2932 | } | |
2933 | else | |
2934 | break; | |
2935 | } | |
2936 | ||
2937 | /* We should never be about to go beyond the end of the pattern. */ | |
2938 | assert (p < pend); | |
2939 | ||
2940 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) | |
2941 | { | |
2942 | ||
2943 | /* I guess the idea here is to simply not bother with a fastmap | |
2944 | if a backreference is used, since it's too hard to figure out | |
2945 | the fastmap for the corresponding group. Setting | |
2946 | `can_be_null' stops `re_search_2' from using the fastmap, so | |
2947 | that is all we do. */ | |
2948 | case duplicate: | |
2949 | bufp->can_be_null = 1; | |
2950 | goto done; | |
2951 | ||
2952 | ||
2953 | /* Following are the cases which match a character. These end | |
2954 | with `break'. */ | |
2955 | ||
2956 | case exactn: | |
2957 | fastmap[p[1]] = 1; | |
2958 | break; | |
2959 | ||
2960 | ||
2961 | case charset: | |
2962 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
2963 | if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) | |
2964 | fastmap[j] = 1; | |
2965 | break; | |
2966 | ||
2967 | ||
2968 | case charset_not: | |
2969 | /* Chars beyond end of map must be allowed. */ | |
2970 | for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) | |
2971 | fastmap[j] = 1; | |
2972 | ||
2973 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
2974 | if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) | |
2975 | fastmap[j] = 1; | |
2976 | break; | |
2977 | ||
2978 | ||
2979 | case wordchar: | |
2980 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2981 | if (SYNTAX (j) == Sword) | |
2982 | fastmap[j] = 1; | |
2983 | break; | |
2984 | ||
2985 | ||
2986 | case notwordchar: | |
2987 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2988 | if (SYNTAX (j) != Sword) | |
2989 | fastmap[j] = 1; | |
2990 | break; | |
2991 | ||
2992 | ||
2993 | case anychar: | |
2994 | { | |
2995 | int fastmap_newline = fastmap['\n']; | |
2996 | ||
2997 | /* `.' matches anything ... */ | |
2998 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2999 | fastmap[j] = 1; | |
3000 | ||
3001 | /* ... except perhaps newline. */ | |
3002 | if (!(bufp->syntax & RE_DOT_NEWLINE)) | |
3003 | fastmap['\n'] = fastmap_newline; | |
3004 | ||
3005 | /* Return if we have already set `can_be_null'; if we have, | |
3006 | then the fastmap is irrelevant. Something's wrong here. */ | |
3007 | else if (bufp->can_be_null) | |
3008 | goto done; | |
3009 | ||
3010 | /* Otherwise, have to check alternative paths. */ | |
3011 | break; | |
3012 | } | |
3013 | ||
3014 | #ifdef emacs | |
3015 | case syntaxspec: | |
3016 | k = *p++; | |
3017 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
3018 | if (SYNTAX (j) == (enum syntaxcode) k) | |
3019 | fastmap[j] = 1; | |
3020 | break; | |
3021 | ||
3022 | ||
3023 | case notsyntaxspec: | |
3024 | k = *p++; | |
3025 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
3026 | if (SYNTAX (j) != (enum syntaxcode) k) | |
3027 | fastmap[j] = 1; | |
3028 | break; | |
3029 | ||
3030 | ||
3031 | /* All cases after this match the empty string. These end with | |
3032 | `continue'. */ | |
3033 | ||
3034 | ||
3035 | case before_dot: | |
3036 | case at_dot: | |
3037 | case after_dot: | |
3038 | continue; | |
ae4788a8 | 3039 | #endif /* emacs */ |
fa9a63c5 RM |
3040 | |
3041 | ||
3042 | case no_op: | |
3043 | case begline: | |
3044 | case endline: | |
3045 | case begbuf: | |
3046 | case endbuf: | |
3047 | case wordbound: | |
3048 | case notwordbound: | |
3049 | case wordbeg: | |
3050 | case wordend: | |
3051 | case push_dummy_failure: | |
3052 | continue; | |
3053 | ||
3054 | ||
3055 | case jump_n: | |
3056 | case pop_failure_jump: | |
3057 | case maybe_pop_jump: | |
3058 | case jump: | |
3059 | case jump_past_alt: | |
3060 | case dummy_failure_jump: | |
3061 | EXTRACT_NUMBER_AND_INCR (j, p); | |
3062 | p += j; | |
3063 | if (j > 0) | |
3064 | continue; | |
3065 | ||
3066 | /* Jump backward implies we just went through the body of a | |
3067 | loop and matched nothing. Opcode jumped to should be | |
3068 | `on_failure_jump' or `succeed_n'. Just treat it like an | |
3069 | ordinary jump. For a * loop, it has pushed its failure | |
3070 | point already; if so, discard that as redundant. */ | |
3071 | if ((re_opcode_t) *p != on_failure_jump | |
3072 | && (re_opcode_t) *p != succeed_n) | |
3073 | continue; | |
3074 | ||
3075 | p++; | |
3076 | EXTRACT_NUMBER_AND_INCR (j, p); | |
3077 | p += j; | |
3078 | ||
3079 | /* If what's on the stack is where we are now, pop it. */ | |
3080 | if (!FAIL_STACK_EMPTY () | |
3081 | && fail_stack.stack[fail_stack.avail - 1].pointer == p) | |
3082 | fail_stack.avail--; | |
3083 | ||
3084 | continue; | |
3085 | ||
3086 | ||
3087 | case on_failure_jump: | |
3088 | case on_failure_keep_string_jump: | |
3089 | handle_on_failure_jump: | |
3090 | EXTRACT_NUMBER_AND_INCR (j, p); | |
3091 | ||
3092 | /* For some patterns, e.g., `(a?)?', `p+j' here points to the | |
3093 | end of the pattern. We don't want to push such a point, | |
3094 | since when we restore it above, entering the switch will | |
3095 | increment `p' past the end of the pattern. We don't need | |
3096 | to push such a point since we obviously won't find any more | |
3097 | fastmap entries beyond `pend'. Such a pattern can match | |
3098 | the null string, though. */ | |
3099 | if (p + j < pend) | |
3100 | { | |
3101 | if (!PUSH_PATTERN_OP (p + j, fail_stack)) | |
3102 | { | |
3103 | RESET_FAIL_STACK (); | |
3104 | return -2; | |
3105 | } | |
3106 | } | |
3107 | else | |
3108 | bufp->can_be_null = 1; | |
3109 | ||
3110 | if (succeed_n_p) | |
3111 | { | |
3112 | EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ | |
3113 | succeed_n_p = false; | |
3114 | } | |
3115 | ||
3116 | continue; | |
3117 | ||
3118 | ||
3119 | case succeed_n: | |
3120 | /* Get to the number of times to succeed. */ | |
3121 | p += 2; | |
3122 | ||
3123 | /* Increment p past the n for when k != 0. */ | |
3124 | EXTRACT_NUMBER_AND_INCR (k, p); | |
3125 | if (k == 0) | |
3126 | { | |
3127 | p -= 4; | |
3128 | succeed_n_p = true; /* Spaghetti code alert. */ | |
3129 | goto handle_on_failure_jump; | |
3130 | } | |
3131 | continue; | |
3132 | ||
3133 | ||
3134 | case set_number_at: | |
3135 | p += 4; | |
3136 | continue; | |
3137 | ||
3138 | ||
3139 | case start_memory: | |
3140 | case stop_memory: | |
3141 | p += 2; | |
3142 | continue; | |
3143 | ||
3144 | ||
3145 | default: | |
3146 | abort (); /* We have listed all the cases. */ | |
3147 | } /* switch *p++ */ | |
3148 | ||
3149 | /* Getting here means we have found the possible starting | |
3150 | characters for one path of the pattern -- and that the empty | |
3151 | string does not match. We need not follow this path further. | |
3152 | Instead, look at the next alternative (remembered on the | |
3153 | stack), or quit if no more. The test at the top of the loop | |
3154 | does these things. */ | |
3155 | path_can_be_null = false; | |
3156 | p = pend; | |
3157 | } /* while p */ | |
3158 | ||
3159 | /* Set `can_be_null' for the last path (also the first path, if the | |
3160 | pattern is empty). */ | |
3161 | bufp->can_be_null |= path_can_be_null; | |
3162 | ||
3163 | done: | |
3164 | RESET_FAIL_STACK (); | |
3165 | return 0; | |
3166 | } /* re_compile_fastmap */ | |
3167 | \f | |
3168 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
3169 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
3170 | this memory for recording register information. STARTS and ENDS | |
3171 | must be allocated using the malloc library routine, and must each | |
3172 | be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
3173 | ||
3174 | If NUM_REGS == 0, then subsequent matches should allocate their own | |
3175 | register data. | |
3176 | ||
3177 | Unless this function is called, the first search or match using | |
3178 | PATTERN_BUFFER will allocate its own register data, without | |
3179 | freeing the old data. */ | |
3180 | ||
3181 | void | |
3182 | re_set_registers (bufp, regs, num_regs, starts, ends) | |
3183 | struct re_pattern_buffer *bufp; | |
3184 | struct re_registers *regs; | |
3185 | unsigned num_regs; | |
3186 | regoff_t *starts, *ends; | |
3187 | { | |
3188 | if (num_regs) | |
3189 | { | |
3190 | bufp->regs_allocated = REGS_REALLOCATE; | |
3191 | regs->num_regs = num_regs; | |
3192 | regs->start = starts; | |
3193 | regs->end = ends; | |
3194 | } | |
3195 | else | |
3196 | { | |
3197 | bufp->regs_allocated = REGS_UNALLOCATED; | |
3198 | regs->num_regs = 0; | |
3199 | regs->start = regs->end = (regoff_t *) 0; | |
3200 | } | |
3201 | } | |
3202 | \f | |
3203 | /* Searching routines. */ | |
3204 | ||
3205 | /* Like re_search_2, below, but only one string is specified, and | |
3206 | doesn't let you say where to stop matching. */ | |
3207 | ||
3208 | int | |
3209 | re_search (bufp, string, size, startpos, range, regs) | |
3210 | struct re_pattern_buffer *bufp; | |
3211 | const char *string; | |
3212 | int size, startpos, range; | |
3213 | struct re_registers *regs; | |
3214 | { | |
3215 | return re_search_2 (bufp, NULL, 0, string, size, startpos, range, | |
3216 | regs, size); | |
3217 | } | |
3218 | ||
3219 | ||
3220 | /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
3221 | virtual concatenation of STRING1 and STRING2, starting first at index | |
3222 | STARTPOS, then at STARTPOS + 1, and so on. | |
3223 | ||
3224 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. | |
3225 | ||
3226 | RANGE is how far to scan while trying to match. RANGE = 0 means try | |
3227 | only at STARTPOS; in general, the last start tried is STARTPOS + | |
3228 | RANGE. | |
3229 | ||
3230 | In REGS, return the indices of the virtual concatenation of STRING1 | |
3231 | and STRING2 that matched the entire BUFP->buffer and its contained | |
3232 | subexpressions. | |
3233 | ||
3234 | Do not consider matching one past the index STOP in the virtual | |
3235 | concatenation of STRING1 and STRING2. | |
3236 | ||
3237 | We return either the position in the strings at which the match was | |
3238 | found, -1 if no match, or -2 if error (such as failure | |
3239 | stack overflow). */ | |
3240 | ||
3241 | int | |
3242 | re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop) | |
3243 | struct re_pattern_buffer *bufp; | |
3244 | const char *string1, *string2; | |
3245 | int size1, size2; | |
3246 | int startpos; | |
3247 | int range; | |
3248 | struct re_registers *regs; | |
3249 | int stop; | |
3250 | { | |
3251 | int val; | |
3252 | register char *fastmap = bufp->fastmap; | |
3253 | register char *translate = bufp->translate; | |
3254 | int total_size = size1 + size2; | |
3255 | int endpos = startpos + range; | |
3256 | ||
3257 | /* Check for out-of-range STARTPOS. */ | |
3258 | if (startpos < 0 || startpos > total_size) | |
3259 | return -1; | |
3260 | ||
3261 | /* Fix up RANGE if it might eventually take us outside | |
34597fa9 RS |
3262 | the virtual concatenation of STRING1 and STRING2. |
3263 | Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ | |
3264 | if (endpos < 0) | |
3265 | range = 0 - startpos; | |
fa9a63c5 RM |
3266 | else if (endpos > total_size) |
3267 | range = total_size - startpos; | |
3268 | ||
3269 | /* If the search isn't to be a backwards one, don't waste time in a | |
3270 | search for a pattern that must be anchored. */ | |
3271 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) | |
3272 | { | |
3273 | if (startpos > 0) | |
3274 | return -1; | |
3275 | else | |
3276 | range = 1; | |
3277 | } | |
3278 | ||
ae4788a8 RS |
3279 | #ifdef emacs |
3280 | /* In a forward search for something that starts with \=. | |
3281 | don't keep searching past point. */ | |
3282 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) | |
3283 | { | |
3284 | range = PT - startpos; | |
3285 | if (range <= 0) | |
3286 | return -1; | |
3287 | } | |
3288 | #endif /* emacs */ | |
3289 | ||
fa9a63c5 RM |
3290 | /* Update the fastmap now if not correct already. */ |
3291 | if (fastmap && !bufp->fastmap_accurate) | |
3292 | if (re_compile_fastmap (bufp) == -2) | |
3293 | return -2; | |
3294 | ||
3295 | /* Loop through the string, looking for a place to start matching. */ | |
3296 | for (;;) | |
3297 | { | |
3298 | /* If a fastmap is supplied, skip quickly over characters that | |
3299 | cannot be the start of a match. If the pattern can match the | |
3300 | null string, however, we don't need to skip characters; we want | |
3301 | the first null string. */ | |
3302 | if (fastmap && startpos < total_size && !bufp->can_be_null) | |
3303 | { | |
3304 | if (range > 0) /* Searching forwards. */ | |
3305 | { | |
3306 | register const char *d; | |
3307 | register int lim = 0; | |
3308 | int irange = range; | |
3309 | ||
3310 | if (startpos < size1 && startpos + range >= size1) | |
3311 | lim = range - (size1 - startpos); | |
3312 | ||
3313 | d = (startpos >= size1 ? string2 - size1 : string1) + startpos; | |
3314 | ||
3315 | /* Written out as an if-else to avoid testing `translate' | |
3316 | inside the loop. */ | |
3317 | if (translate) | |
3318 | while (range > lim | |
3319 | && !fastmap[(unsigned char) | |
3320 | translate[(unsigned char) *d++]]) | |
3321 | range--; | |
3322 | else | |
3323 | while (range > lim && !fastmap[(unsigned char) *d++]) | |
3324 | range--; | |
3325 | ||
3326 | startpos += irange - range; | |
3327 | } | |
3328 | else /* Searching backwards. */ | |
3329 | { | |
3330 | register char c = (size1 == 0 || startpos >= size1 | |
3331 | ? string2[startpos - size1] | |
3332 | : string1[startpos]); | |
3333 | ||
3334 | if (!fastmap[(unsigned char) TRANSLATE (c)]) | |
3335 | goto advance; | |
3336 | } | |
3337 | } | |
3338 | ||
3339 | /* If can't match the null string, and that's all we have left, fail. */ | |
3340 | if (range >= 0 && startpos == total_size && fastmap | |
3341 | && !bufp->can_be_null) | |
3342 | return -1; | |
3343 | ||
3344 | val = re_match_2_internal (bufp, string1, size1, string2, size2, | |
3345 | startpos, regs, stop); | |
3346 | #ifndef REGEX_MALLOC | |
3347 | #ifdef C_ALLOCA | |
3348 | alloca (0); | |
3349 | #endif | |
3350 | #endif | |
3351 | ||
3352 | if (val >= 0) | |
3353 | return startpos; | |
3354 | ||
3355 | if (val == -2) | |
3356 | return -2; | |
3357 | ||
3358 | advance: | |
3359 | if (!range) | |
3360 | break; | |
3361 | else if (range > 0) | |
3362 | { | |
3363 | range--; | |
3364 | startpos++; | |
3365 | } | |
3366 | else | |
3367 | { | |
3368 | range++; | |
3369 | startpos--; | |
3370 | } | |
3371 | } | |
3372 | return -1; | |
3373 | } /* re_search_2 */ | |
3374 | \f | |
3375 | /* Declarations and macros for re_match_2. */ | |
3376 | ||
3377 | static int bcmp_translate (); | |
3378 | static boolean alt_match_null_string_p (), | |
3379 | common_op_match_null_string_p (), | |
3380 | group_match_null_string_p (); | |
3381 | ||
3382 | /* This converts PTR, a pointer into one of the search strings `string1' | |
3383 | and `string2' into an offset from the beginning of that string. */ | |
3384 | #define POINTER_TO_OFFSET(ptr) \ | |
3385 | (FIRST_STRING_P (ptr) \ | |
3386 | ? ((regoff_t) ((ptr) - string1)) \ | |
3387 | : ((regoff_t) ((ptr) - string2 + size1))) | |
3388 | ||
3389 | /* Macros for dealing with the split strings in re_match_2. */ | |
3390 | ||
3391 | #define MATCHING_IN_FIRST_STRING (dend == end_match_1) | |
3392 | ||
3393 | /* Call before fetching a character with *d. This switches over to | |
3394 | string2 if necessary. */ | |
3395 | #define PREFETCH() \ | |
3396 | while (d == dend) \ | |
3397 | { \ | |
3398 | /* End of string2 => fail. */ \ | |
3399 | if (dend == end_match_2) \ | |
3400 | goto fail; \ | |
3401 | /* End of string1 => advance to string2. */ \ | |
3402 | d = string2; \ | |
3403 | dend = end_match_2; \ | |
3404 | } | |
3405 | ||
3406 | ||
3407 | /* Test if at very beginning or at very end of the virtual concatenation | |
3408 | of `string1' and `string2'. If only one string, it's `string2'. */ | |
3409 | #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) | |
3410 | #define AT_STRINGS_END(d) ((d) == end2) | |
3411 | ||
3412 | ||
3413 | /* Test if D points to a character which is word-constituent. We have | |
3414 | two special cases to check for: if past the end of string1, look at | |
3415 | the first character in string2; and if before the beginning of | |
3416 | string2, look at the last character in string1. */ | |
3417 | #define WORDCHAR_P(d) \ | |
3418 | (SYNTAX ((d) == end1 ? *string2 \ | |
3419 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ | |
3420 | == Sword) | |
3421 | ||
3422 | /* Test if the character before D and the one at D differ with respect | |
3423 | to being word-constituent. */ | |
3424 | #define AT_WORD_BOUNDARY(d) \ | |
3425 | (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ | |
3426 | || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) | |
3427 | ||
3428 | ||
3429 | /* Free everything we malloc. */ | |
3430 | #ifdef MATCH_MAY_ALLOCATE | |
3431 | #define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL | |
3432 | #define FREE_VARIABLES() \ | |
3433 | do { \ | |
3434 | REGEX_FREE_STACK (fail_stack.stack); \ | |
3435 | FREE_VAR (regstart); \ | |
3436 | FREE_VAR (regend); \ | |
3437 | FREE_VAR (old_regstart); \ | |
3438 | FREE_VAR (old_regend); \ | |
3439 | FREE_VAR (best_regstart); \ | |
3440 | FREE_VAR (best_regend); \ | |
3441 | FREE_VAR (reg_info); \ | |
3442 | FREE_VAR (reg_dummy); \ | |
3443 | FREE_VAR (reg_info_dummy); \ | |
3444 | } while (0) | |
3445 | #else | |
3446 | #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ | |
3447 | #endif /* not MATCH_MAY_ALLOCATE */ | |
3448 | ||
3449 | /* These values must meet several constraints. They must not be valid | |
3450 | register values; since we have a limit of 255 registers (because | |
3451 | we use only one byte in the pattern for the register number), we can | |
3452 | use numbers larger than 255. They must differ by 1, because of | |
3453 | NUM_FAILURE_ITEMS above. And the value for the lowest register must | |
3454 | be larger than the value for the highest register, so we do not try | |
3455 | to actually save any registers when none are active. */ | |
3456 | #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) | |
3457 | #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) | |
3458 | \f | |
3459 | /* Matching routines. */ | |
3460 | ||
3461 | #ifndef emacs /* Emacs never uses this. */ | |
3462 | /* re_match is like re_match_2 except it takes only a single string. */ | |
3463 | ||
3464 | int | |
3465 | re_match (bufp, string, size, pos, regs) | |
3466 | struct re_pattern_buffer *bufp; | |
3467 | const char *string; | |
3468 | int size, pos; | |
3469 | struct re_registers *regs; | |
3470 | { | |
3471 | int result = re_match_2_internal (bufp, NULL, 0, string, size, | |
3472 | pos, regs, size); | |
3473 | alloca (0); | |
3474 | return result; | |
3475 | } | |
3476 | #endif /* not emacs */ | |
3477 | ||
3478 | ||
3479 | /* re_match_2 matches the compiled pattern in BUFP against the | |
3480 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
3481 | and SIZE2, respectively). We start matching at POS, and stop | |
3482 | matching at STOP. | |
3483 | ||
3484 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we | |
3485 | store offsets for the substring each group matched in REGS. See the | |
3486 | documentation for exactly how many groups we fill. | |
3487 | ||
3488 | We return -1 if no match, -2 if an internal error (such as the | |
3489 | failure stack overflowing). Otherwise, we return the length of the | |
3490 | matched substring. */ | |
3491 | ||
3492 | int | |
3493 | re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
3494 | struct re_pattern_buffer *bufp; | |
3495 | const char *string1, *string2; | |
3496 | int size1, size2; | |
3497 | int pos; | |
3498 | struct re_registers *regs; | |
3499 | int stop; | |
3500 | { | |
3501 | int result = re_match_2_internal (bufp, string1, size1, string2, size2, | |
3502 | pos, regs, stop); | |
3503 | alloca (0); | |
3504 | return result; | |
3505 | } | |
3506 | ||
3507 | /* This is a separate function so that we can force an alloca cleanup | |
3508 | afterwards. */ | |
3509 | static int | |
3510 | re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop) | |
3511 | struct re_pattern_buffer *bufp; | |
3512 | const char *string1, *string2; | |
3513 | int size1, size2; | |
3514 | int pos; | |
3515 | struct re_registers *regs; | |
3516 | int stop; | |
3517 | { | |
3518 | /* General temporaries. */ | |
3519 | int mcnt; | |
3520 | unsigned char *p1; | |
3521 | ||
3522 | /* Just past the end of the corresponding string. */ | |
3523 | const char *end1, *end2; | |
3524 | ||
3525 | /* Pointers into string1 and string2, just past the last characters in | |
3526 | each to consider matching. */ | |
3527 | const char *end_match_1, *end_match_2; | |
3528 | ||
3529 | /* Where we are in the data, and the end of the current string. */ | |
3530 | const char *d, *dend; | |
3531 | ||
3532 | /* Where we are in the pattern, and the end of the pattern. */ | |
3533 | unsigned char *p = bufp->buffer; | |
3534 | register unsigned char *pend = p + bufp->used; | |
3535 | ||
3536 | /* Mark the opcode just after a start_memory, so we can test for an | |
3537 | empty subpattern when we get to the stop_memory. */ | |
3538 | unsigned char *just_past_start_mem = 0; | |
3539 | ||
3540 | /* We use this to map every character in the string. */ | |
3541 | char *translate = bufp->translate; | |
3542 | ||
3543 | /* Failure point stack. Each place that can handle a failure further | |
3544 | down the line pushes a failure point on this stack. It consists of | |
3545 | restart, regend, and reg_info for all registers corresponding to | |
3546 | the subexpressions we're currently inside, plus the number of such | |
3547 | registers, and, finally, two char *'s. The first char * is where | |
3548 | to resume scanning the pattern; the second one is where to resume | |
3549 | scanning the strings. If the latter is zero, the failure point is | |
3550 | a ``dummy''; if a failure happens and the failure point is a dummy, | |
3551 | it gets discarded and the next next one is tried. */ | |
3552 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ | |
3553 | fail_stack_type fail_stack; | |
3554 | #endif | |
3555 | #ifdef DEBUG | |
3556 | static unsigned failure_id = 0; | |
3557 | unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; | |
3558 | #endif | |
3559 | ||
3560 | /* This holds the pointer to the failure stack, when | |
3561 | it is allocated relocatably. */ | |
3562 | fail_stack_elt_t *failure_stack_ptr; | |
3563 | ||
3564 | /* We fill all the registers internally, independent of what we | |
3565 | return, for use in backreferences. The number here includes | |
3566 | an element for register zero. */ | |
3567 | unsigned num_regs = bufp->re_nsub + 1; | |
3568 | ||
3569 | /* The currently active registers. */ | |
3570 | unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3571 | unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3572 | ||
3573 | /* Information on the contents of registers. These are pointers into | |
3574 | the input strings; they record just what was matched (on this | |
3575 | attempt) by a subexpression part of the pattern, that is, the | |
3576 | regnum-th regstart pointer points to where in the pattern we began | |
3577 | matching and the regnum-th regend points to right after where we | |
3578 | stopped matching the regnum-th subexpression. (The zeroth register | |
3579 | keeps track of what the whole pattern matches.) */ | |
3580 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
3581 | const char **regstart, **regend; | |
3582 | #endif | |
3583 | ||
3584 | /* If a group that's operated upon by a repetition operator fails to | |
3585 | match anything, then the register for its start will need to be | |
3586 | restored because it will have been set to wherever in the string we | |
3587 | are when we last see its open-group operator. Similarly for a | |
3588 | register's end. */ | |
3589 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
3590 | const char **old_regstart, **old_regend; | |
3591 | #endif | |
3592 | ||
3593 | /* The is_active field of reg_info helps us keep track of which (possibly | |
3594 | nested) subexpressions we are currently in. The matched_something | |
3595 | field of reg_info[reg_num] helps us tell whether or not we have | |
3596 | matched any of the pattern so far this time through the reg_num-th | |
3597 | subexpression. These two fields get reset each time through any | |
3598 | loop their register is in. */ | |
3599 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ | |
3600 | register_info_type *reg_info; | |
3601 | #endif | |
3602 | ||
3603 | /* The following record the register info as found in the above | |
3604 | variables when we find a match better than any we've seen before. | |
3605 | This happens as we backtrack through the failure points, which in | |
3606 | turn happens only if we have not yet matched the entire string. */ | |
3607 | unsigned best_regs_set = false; | |
3608 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
3609 | const char **best_regstart, **best_regend; | |
3610 | #endif | |
3611 | ||
3612 | /* Logically, this is `best_regend[0]'. But we don't want to have to | |
3613 | allocate space for that if we're not allocating space for anything | |
3614 | else (see below). Also, we never need info about register 0 for | |
3615 | any of the other register vectors, and it seems rather a kludge to | |
3616 | treat `best_regend' differently than the rest. So we keep track of | |
3617 | the end of the best match so far in a separate variable. We | |
3618 | initialize this to NULL so that when we backtrack the first time | |
3619 | and need to test it, it's not garbage. */ | |
3620 | const char *match_end = NULL; | |
3621 | ||
3622 | /* This helps SET_REGS_MATCHED avoid doing redundant work. */ | |
3623 | int set_regs_matched_done = 0; | |
3624 | ||
3625 | /* Used when we pop values we don't care about. */ | |
3626 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
3627 | const char **reg_dummy; | |
3628 | register_info_type *reg_info_dummy; | |
3629 | #endif | |
3630 | ||
3631 | #ifdef DEBUG | |
3632 | /* Counts the total number of registers pushed. */ | |
3633 | unsigned num_regs_pushed = 0; | |
3634 | #endif | |
3635 | ||
3636 | DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
3637 | ||
3638 | INIT_FAIL_STACK (); | |
3639 | ||
3640 | #ifdef MATCH_MAY_ALLOCATE | |
3641 | /* Do not bother to initialize all the register variables if there are | |
3642 | no groups in the pattern, as it takes a fair amount of time. If | |
3643 | there are groups, we include space for register 0 (the whole | |
3644 | pattern), even though we never use it, since it simplifies the | |
3645 | array indexing. We should fix this. */ | |
3646 | if (bufp->re_nsub) | |
3647 | { | |
3648 | regstart = REGEX_TALLOC (num_regs, const char *); | |
3649 | regend = REGEX_TALLOC (num_regs, const char *); | |
3650 | old_regstart = REGEX_TALLOC (num_regs, const char *); | |
3651 | old_regend = REGEX_TALLOC (num_regs, const char *); | |
3652 | best_regstart = REGEX_TALLOC (num_regs, const char *); | |
3653 | best_regend = REGEX_TALLOC (num_regs, const char *); | |
3654 | reg_info = REGEX_TALLOC (num_regs, register_info_type); | |
3655 | reg_dummy = REGEX_TALLOC (num_regs, const char *); | |
3656 | reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type); | |
3657 | ||
3658 | if (!(regstart && regend && old_regstart && old_regend && reg_info | |
3659 | && best_regstart && best_regend && reg_dummy && reg_info_dummy)) | |
3660 | { | |
3661 | FREE_VARIABLES (); | |
3662 | return -2; | |
3663 | } | |
3664 | } | |
3665 | else | |
3666 | { | |
3667 | /* We must initialize all our variables to NULL, so that | |
3668 | `FREE_VARIABLES' doesn't try to free them. */ | |
3669 | regstart = regend = old_regstart = old_regend = best_regstart | |
3670 | = best_regend = reg_dummy = NULL; | |
3671 | reg_info = reg_info_dummy = (register_info_type *) NULL; | |
3672 | } | |
3673 | #endif /* MATCH_MAY_ALLOCATE */ | |
3674 | ||
3675 | /* The starting position is bogus. */ | |
3676 | if (pos < 0 || pos > size1 + size2) | |
3677 | { | |
3678 | FREE_VARIABLES (); | |
3679 | return -1; | |
3680 | } | |
3681 | ||
3682 | /* Initialize subexpression text positions to -1 to mark ones that no | |
3683 | start_memory/stop_memory has been seen for. Also initialize the | |
3684 | register information struct. */ | |
3685 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3686 | { | |
3687 | regstart[mcnt] = regend[mcnt] | |
3688 | = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; | |
3689 | ||
3690 | REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; | |
3691 | IS_ACTIVE (reg_info[mcnt]) = 0; | |
3692 | MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
3693 | EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
3694 | } | |
3695 | ||
3696 | /* We move `string1' into `string2' if the latter's empty -- but not if | |
3697 | `string1' is null. */ | |
3698 | if (size2 == 0 && string1 != NULL) | |
3699 | { | |
3700 | string2 = string1; | |
3701 | size2 = size1; | |
3702 | string1 = 0; | |
3703 | size1 = 0; | |
3704 | } | |
3705 | end1 = string1 + size1; | |
3706 | end2 = string2 + size2; | |
3707 | ||
3708 | /* Compute where to stop matching, within the two strings. */ | |
3709 | if (stop <= size1) | |
3710 | { | |
3711 | end_match_1 = string1 + stop; | |
3712 | end_match_2 = string2; | |
3713 | } | |
3714 | else | |
3715 | { | |
3716 | end_match_1 = end1; | |
3717 | end_match_2 = string2 + stop - size1; | |
3718 | } | |
3719 | ||
3720 | /* `p' scans through the pattern as `d' scans through the data. | |
3721 | `dend' is the end of the input string that `d' points within. `d' | |
3722 | is advanced into the following input string whenever necessary, but | |
3723 | this happens before fetching; therefore, at the beginning of the | |
3724 | loop, `d' can be pointing at the end of a string, but it cannot | |
3725 | equal `string2'. */ | |
3726 | if (size1 > 0 && pos <= size1) | |
3727 | { | |
3728 | d = string1 + pos; | |
3729 | dend = end_match_1; | |
3730 | } | |
3731 | else | |
3732 | { | |
3733 | d = string2 + pos - size1; | |
3734 | dend = end_match_2; | |
3735 | } | |
3736 | ||
3737 | DEBUG_PRINT1 ("The compiled pattern is: "); | |
3738 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
3739 | DEBUG_PRINT1 ("The string to match is: `"); | |
3740 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
3741 | DEBUG_PRINT1 ("'\n"); | |
3742 | ||
3743 | /* This loops over pattern commands. It exits by returning from the | |
3744 | function if the match is complete, or it drops through if the match | |
3745 | fails at this starting point in the input data. */ | |
3746 | for (;;) | |
3747 | { | |
3748 | DEBUG_PRINT2 ("\n0x%x: ", p); | |
3749 | ||
3750 | if (p == pend) | |
3751 | { /* End of pattern means we might have succeeded. */ | |
3752 | DEBUG_PRINT1 ("end of pattern ... "); | |
3753 | ||
3754 | /* If we haven't matched the entire string, and we want the | |
3755 | longest match, try backtracking. */ | |
3756 | if (d != end_match_2) | |
3757 | { | |
3758 | /* 1 if this match ends in the same string (string1 or string2) | |
3759 | as the best previous match. */ | |
3760 | boolean same_str_p = (FIRST_STRING_P (match_end) | |
3761 | == MATCHING_IN_FIRST_STRING); | |
3762 | /* 1 if this match is the best seen so far. */ | |
3763 | boolean best_match_p; | |
3764 | ||
3765 | /* AIX compiler got confused when this was combined | |
3766 | with the previous declaration. */ | |
3767 | if (same_str_p) | |
3768 | best_match_p = d > match_end; | |
3769 | else | |
3770 | best_match_p = !MATCHING_IN_FIRST_STRING; | |
3771 | ||
3772 | DEBUG_PRINT1 ("backtracking.\n"); | |
3773 | ||
3774 | if (!FAIL_STACK_EMPTY ()) | |
3775 | { /* More failure points to try. */ | |
3776 | ||
3777 | /* If exceeds best match so far, save it. */ | |
3778 | if (!best_regs_set || best_match_p) | |
3779 | { | |
3780 | best_regs_set = true; | |
3781 | match_end = d; | |
3782 | ||
3783 | DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); | |
3784 | ||
3785 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3786 | { | |
3787 | best_regstart[mcnt] = regstart[mcnt]; | |
3788 | best_regend[mcnt] = regend[mcnt]; | |
3789 | } | |
3790 | } | |
3791 | goto fail; | |
3792 | } | |
3793 | ||
3794 | /* If no failure points, don't restore garbage. And if | |
3795 | last match is real best match, don't restore second | |
3796 | best one. */ | |
3797 | else if (best_regs_set && !best_match_p) | |
3798 | { | |
3799 | restore_best_regs: | |
3800 | /* Restore best match. It may happen that `dend == | |
3801 | end_match_1' while the restored d is in string2. | |
3802 | For example, the pattern `x.*y.*z' against the | |
3803 | strings `x-' and `y-z-', if the two strings are | |
3804 | not consecutive in memory. */ | |
3805 | DEBUG_PRINT1 ("Restoring best registers.\n"); | |
3806 | ||
3807 | d = match_end; | |
3808 | dend = ((d >= string1 && d <= end1) | |
3809 | ? end_match_1 : end_match_2); | |
3810 | ||
3811 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3812 | { | |
3813 | regstart[mcnt] = best_regstart[mcnt]; | |
3814 | regend[mcnt] = best_regend[mcnt]; | |
3815 | } | |
3816 | } | |
3817 | } /* d != end_match_2 */ | |
3818 | ||
3819 | succeed_label: | |
3820 | DEBUG_PRINT1 ("Accepting match.\n"); | |
3821 | ||
3822 | /* If caller wants register contents data back, do it. */ | |
3823 | if (regs && !bufp->no_sub) | |
3824 | { | |
3825 | /* Have the register data arrays been allocated? */ | |
3826 | if (bufp->regs_allocated == REGS_UNALLOCATED) | |
3827 | { /* No. So allocate them with malloc. We need one | |
3828 | extra element beyond `num_regs' for the `-1' marker | |
3829 | GNU code uses. */ | |
3830 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
3831 | regs->start = TALLOC (regs->num_regs, regoff_t); | |
3832 | regs->end = TALLOC (regs->num_regs, regoff_t); | |
3833 | if (regs->start == NULL || regs->end == NULL) | |
3834 | { | |
3835 | FREE_VARIABLES (); | |
3836 | return -2; | |
3837 | } | |
3838 | bufp->regs_allocated = REGS_REALLOCATE; | |
3839 | } | |
3840 | else if (bufp->regs_allocated == REGS_REALLOCATE) | |
3841 | { /* Yes. If we need more elements than were already | |
3842 | allocated, reallocate them. If we need fewer, just | |
3843 | leave it alone. */ | |
3844 | if (regs->num_regs < num_regs + 1) | |
3845 | { | |
3846 | regs->num_regs = num_regs + 1; | |
3847 | RETALLOC (regs->start, regs->num_regs, regoff_t); | |
3848 | RETALLOC (regs->end, regs->num_regs, regoff_t); | |
3849 | if (regs->start == NULL || regs->end == NULL) | |
3850 | { | |
3851 | FREE_VARIABLES (); | |
3852 | return -2; | |
3853 | } | |
3854 | } | |
3855 | } | |
3856 | else | |
3857 | { | |
3858 | /* These braces fend off a "empty body in an else-statement" | |
3859 | warning under GCC when assert expands to nothing. */ | |
3860 | assert (bufp->regs_allocated == REGS_FIXED); | |
3861 | } | |
3862 | ||
3863 | /* Convert the pointer data in `regstart' and `regend' to | |
3864 | indices. Register zero has to be set differently, | |
3865 | since we haven't kept track of any info for it. */ | |
3866 | if (regs->num_regs > 0) | |
3867 | { | |
3868 | regs->start[0] = pos; | |
3869 | regs->end[0] = (MATCHING_IN_FIRST_STRING | |
3870 | ? ((regoff_t) (d - string1)) | |
3871 | : ((regoff_t) (d - string2 + size1))); | |
3872 | } | |
3873 | ||
3874 | /* Go through the first `min (num_regs, regs->num_regs)' | |
3875 | registers, since that is all we initialized. */ | |
3876 | for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++) | |
3877 | { | |
3878 | if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) | |
3879 | regs->start[mcnt] = regs->end[mcnt] = -1; | |
3880 | else | |
3881 | { | |
3882 | regs->start[mcnt] | |
3883 | = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]); | |
3884 | regs->end[mcnt] | |
3885 | = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]); | |
3886 | } | |
3887 | } | |
3888 | ||
3889 | /* If the regs structure we return has more elements than | |
3890 | were in the pattern, set the extra elements to -1. If | |
3891 | we (re)allocated the registers, this is the case, | |
3892 | because we always allocate enough to have at least one | |
3893 | -1 at the end. */ | |
3894 | for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++) | |
3895 | regs->start[mcnt] = regs->end[mcnt] = -1; | |
3896 | } /* regs && !bufp->no_sub */ | |
3897 | ||
3898 | DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", | |
3899 | nfailure_points_pushed, nfailure_points_popped, | |
3900 | nfailure_points_pushed - nfailure_points_popped); | |
3901 | DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); | |
3902 | ||
3903 | mcnt = d - pos - (MATCHING_IN_FIRST_STRING | |
3904 | ? string1 | |
3905 | : string2 - size1); | |
3906 | ||
3907 | DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); | |
3908 | ||
3909 | FREE_VARIABLES (); | |
3910 | return mcnt; | |
3911 | } | |
3912 | ||
3913 | /* Otherwise match next pattern command. */ | |
3914 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) | |
3915 | { | |
3916 | /* Ignore these. Used to ignore the n of succeed_n's which | |
3917 | currently have n == 0. */ | |
3918 | case no_op: | |
3919 | DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
3920 | break; | |
3921 | ||
3922 | case succeed: | |
3923 | DEBUG_PRINT1 ("EXECUTING succeed.\n"); | |
3924 | goto succeed_label; | |
3925 | ||
3926 | /* Match the next n pattern characters exactly. The following | |
3927 | byte in the pattern defines n, and the n bytes after that | |
3928 | are the characters to match. */ | |
3929 | case exactn: | |
3930 | mcnt = *p++; | |
3931 | DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); | |
3932 | ||
3933 | /* This is written out as an if-else so we don't waste time | |
3934 | testing `translate' inside the loop. */ | |
3935 | if (translate) | |
3936 | { | |
3937 | do | |
3938 | { | |
3939 | PREFETCH (); | |
3940 | if (translate[(unsigned char) *d++] != (char) *p++) | |
3941 | goto fail; | |
3942 | } | |
3943 | while (--mcnt); | |
3944 | } | |
3945 | else | |
3946 | { | |
3947 | do | |
3948 | { | |
3949 | PREFETCH (); | |
3950 | if (*d++ != (char) *p++) goto fail; | |
3951 | } | |
3952 | while (--mcnt); | |
3953 | } | |
3954 | SET_REGS_MATCHED (); | |
3955 | break; | |
3956 | ||
3957 | ||
3958 | /* Match any character except possibly a newline or a null. */ | |
3959 | case anychar: | |
3960 | DEBUG_PRINT1 ("EXECUTING anychar.\n"); | |
3961 | ||
3962 | PREFETCH (); | |
3963 | ||
3964 | if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n') | |
3965 | || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000')) | |
3966 | goto fail; | |
3967 | ||
3968 | SET_REGS_MATCHED (); | |
3969 | DEBUG_PRINT2 (" Matched `%d'.\n", *d); | |
3970 | d++; | |
3971 | break; | |
3972 | ||
3973 | ||
3974 | case charset: | |
3975 | case charset_not: | |
3976 | { | |
3977 | register unsigned char c; | |
3978 | boolean not = (re_opcode_t) *(p - 1) == charset_not; | |
3979 | ||
3980 | DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); | |
3981 | ||
3982 | PREFETCH (); | |
3983 | c = TRANSLATE (*d); /* The character to match. */ | |
3984 | ||
3985 | /* Cast to `unsigned' instead of `unsigned char' in case the | |
3986 | bit list is a full 32 bytes long. */ | |
3987 | if (c < (unsigned) (*p * BYTEWIDTH) | |
3988 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
3989 | not = !not; | |
3990 | ||
3991 | p += 1 + *p; | |
3992 | ||
3993 | if (!not) goto fail; | |
3994 | ||
3995 | SET_REGS_MATCHED (); | |
3996 | d++; | |
3997 | break; | |
3998 | } | |
3999 | ||
4000 | ||
4001 | /* The beginning of a group is represented by start_memory. | |
4002 | The arguments are the register number in the next byte, and the | |
4003 | number of groups inner to this one in the next. The text | |
4004 | matched within the group is recorded (in the internal | |
4005 | registers data structure) under the register number. */ | |
4006 | case start_memory: | |
4007 | DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]); | |
4008 | ||
4009 | /* Find out if this group can match the empty string. */ | |
4010 | p1 = p; /* To send to group_match_null_string_p. */ | |
4011 | ||
4012 | if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) | |
4013 | REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
4014 | = group_match_null_string_p (&p1, pend, reg_info); | |
4015 | ||
4016 | /* Save the position in the string where we were the last time | |
4017 | we were at this open-group operator in case the group is | |
4018 | operated upon by a repetition operator, e.g., with `(a*)*b' | |
4019 | against `ab'; then we want to ignore where we are now in | |
4020 | the string in case this attempt to match fails. */ | |
4021 | old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
4022 | ? REG_UNSET (regstart[*p]) ? d : regstart[*p] | |
4023 | : regstart[*p]; | |
4024 | DEBUG_PRINT2 (" old_regstart: %d\n", | |
4025 | POINTER_TO_OFFSET (old_regstart[*p])); | |
4026 | ||
4027 | regstart[*p] = d; | |
4028 | DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); | |
4029 | ||
4030 | IS_ACTIVE (reg_info[*p]) = 1; | |
4031 | MATCHED_SOMETHING (reg_info[*p]) = 0; | |
4032 | ||
4033 | /* Clear this whenever we change the register activity status. */ | |
4034 | set_regs_matched_done = 0; | |
4035 | ||
4036 | /* This is the new highest active register. */ | |
4037 | highest_active_reg = *p; | |
4038 | ||
4039 | /* If nothing was active before, this is the new lowest active | |
4040 | register. */ | |
4041 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
4042 | lowest_active_reg = *p; | |
4043 | ||
4044 | /* Move past the register number and inner group count. */ | |
4045 | p += 2; | |
4046 | just_past_start_mem = p; | |
4047 | ||
4048 | break; | |
4049 | ||
4050 | ||
4051 | /* The stop_memory opcode represents the end of a group. Its | |
4052 | arguments are the same as start_memory's: the register | |
4053 | number, and the number of inner groups. */ | |
4054 | case stop_memory: | |
4055 | DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]); | |
4056 | ||
4057 | /* We need to save the string position the last time we were at | |
4058 | this close-group operator in case the group is operated | |
4059 | upon by a repetition operator, e.g., with `((a*)*(b*)*)*' | |
4060 | against `aba'; then we want to ignore where we are now in | |
4061 | the string in case this attempt to match fails. */ | |
4062 | old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
4063 | ? REG_UNSET (regend[*p]) ? d : regend[*p] | |
4064 | : regend[*p]; | |
4065 | DEBUG_PRINT2 (" old_regend: %d\n", | |
4066 | POINTER_TO_OFFSET (old_regend[*p])); | |
4067 | ||
4068 | regend[*p] = d; | |
4069 | DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); | |
4070 | ||
4071 | /* This register isn't active anymore. */ | |
4072 | IS_ACTIVE (reg_info[*p]) = 0; | |
4073 | ||
4074 | /* Clear this whenever we change the register activity status. */ | |
4075 | set_regs_matched_done = 0; | |
4076 | ||
4077 | /* If this was the only register active, nothing is active | |
4078 | anymore. */ | |
4079 | if (lowest_active_reg == highest_active_reg) | |
4080 | { | |
4081 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
4082 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
4083 | } | |
4084 | else | |
4085 | { /* We must scan for the new highest active register, since | |
4086 | it isn't necessarily one less than now: consider | |
4087 | (a(b)c(d(e)f)g). When group 3 ends, after the f), the | |
4088 | new highest active register is 1. */ | |
4089 | unsigned char r = *p - 1; | |
4090 | while (r > 0 && !IS_ACTIVE (reg_info[r])) | |
4091 | r--; | |
4092 | ||
4093 | /* If we end up at register zero, that means that we saved | |
4094 | the registers as the result of an `on_failure_jump', not | |
4095 | a `start_memory', and we jumped to past the innermost | |
4096 | `stop_memory'. For example, in ((.)*) we save | |
4097 | registers 1 and 2 as a result of the *, but when we pop | |
4098 | back to the second ), we are at the stop_memory 1. | |
4099 | Thus, nothing is active. */ | |
4100 | if (r == 0) | |
4101 | { | |
4102 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
4103 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
4104 | } | |
4105 | else | |
4106 | highest_active_reg = r; | |
4107 | } | |
4108 | ||
4109 | /* If just failed to match something this time around with a | |
4110 | group that's operated on by a repetition operator, try to | |
4111 | force exit from the ``loop'', and restore the register | |
4112 | information for this group that we had before trying this | |
4113 | last match. */ | |
4114 | if ((!MATCHED_SOMETHING (reg_info[*p]) | |
4115 | || just_past_start_mem == p - 1) | |
4116 | && (p + 2) < pend) | |
4117 | { | |
4118 | boolean is_a_jump_n = false; | |
4119 | ||
4120 | p1 = p + 2; | |
4121 | mcnt = 0; | |
4122 | switch ((re_opcode_t) *p1++) | |
4123 | { | |
4124 | case jump_n: | |
4125 | is_a_jump_n = true; | |
4126 | case pop_failure_jump: | |
4127 | case maybe_pop_jump: | |
4128 | case jump: | |
4129 | case dummy_failure_jump: | |
4130 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4131 | if (is_a_jump_n) | |
4132 | p1 += 2; | |
4133 | break; | |
4134 | ||
4135 | default: | |
4136 | /* do nothing */ ; | |
4137 | } | |
4138 | p1 += mcnt; | |
4139 | ||
4140 | /* If the next operation is a jump backwards in the pattern | |
4141 | to an on_failure_jump right before the start_memory | |
4142 | corresponding to this stop_memory, exit from the loop | |
4143 | by forcing a failure after pushing on the stack the | |
4144 | on_failure_jump's jump in the pattern, and d. */ | |
4145 | if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump | |
4146 | && (re_opcode_t) p1[3] == start_memory && p1[4] == *p) | |
4147 | { | |
4148 | /* If this group ever matched anything, then restore | |
4149 | what its registers were before trying this last | |
4150 | failed match, e.g., with `(a*)*b' against `ab' for | |
4151 | regstart[1], and, e.g., with `((a*)*(b*)*)*' | |
4152 | against `aba' for regend[3]. | |
4153 | ||
4154 | Also restore the registers for inner groups for, | |
4155 | e.g., `((a*)(b*))*' against `aba' (register 3 would | |
4156 | otherwise get trashed). */ | |
4157 | ||
4158 | if (EVER_MATCHED_SOMETHING (reg_info[*p])) | |
4159 | { | |
4160 | unsigned r; | |
4161 | ||
4162 | EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; | |
4163 | ||
4164 | /* Restore this and inner groups' (if any) registers. */ | |
4165 | for (r = *p; r < *p + *(p + 1); r++) | |
4166 | { | |
4167 | regstart[r] = old_regstart[r]; | |
4168 | ||
4169 | /* xx why this test? */ | |
4170 | if (old_regend[r] >= regstart[r]) | |
4171 | regend[r] = old_regend[r]; | |
4172 | } | |
4173 | } | |
4174 | p1++; | |
4175 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4176 | PUSH_FAILURE_POINT (p1 + mcnt, d, -2); | |
4177 | ||
4178 | goto fail; | |
4179 | } | |
4180 | } | |
4181 | ||
4182 | /* Move past the register number and the inner group count. */ | |
4183 | p += 2; | |
4184 | break; | |
4185 | ||
4186 | ||
4187 | /* \<digit> has been turned into a `duplicate' command which is | |
4188 | followed by the numeric value of <digit> as the register number. */ | |
4189 | case duplicate: | |
4190 | { | |
4191 | register const char *d2, *dend2; | |
4192 | int regno = *p++; /* Get which register to match against. */ | |
4193 | DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); | |
4194 | ||
4195 | /* Can't back reference a group which we've never matched. */ | |
4196 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
4197 | goto fail; | |
4198 | ||
4199 | /* Where in input to try to start matching. */ | |
4200 | d2 = regstart[regno]; | |
4201 | ||
4202 | /* Where to stop matching; if both the place to start and | |
4203 | the place to stop matching are in the same string, then | |
4204 | set to the place to stop, otherwise, for now have to use | |
4205 | the end of the first string. */ | |
4206 | ||
4207 | dend2 = ((FIRST_STRING_P (regstart[regno]) | |
4208 | == FIRST_STRING_P (regend[regno])) | |
4209 | ? regend[regno] : end_match_1); | |
4210 | for (;;) | |
4211 | { | |
4212 | /* If necessary, advance to next segment in register | |
4213 | contents. */ | |
4214 | while (d2 == dend2) | |
4215 | { | |
4216 | if (dend2 == end_match_2) break; | |
4217 | if (dend2 == regend[regno]) break; | |
4218 | ||
4219 | /* End of string1 => advance to string2. */ | |
4220 | d2 = string2; | |
4221 | dend2 = regend[regno]; | |
4222 | } | |
4223 | /* At end of register contents => success */ | |
4224 | if (d2 == dend2) break; | |
4225 | ||
4226 | /* If necessary, advance to next segment in data. */ | |
4227 | PREFETCH (); | |
4228 | ||
4229 | /* How many characters left in this segment to match. */ | |
4230 | mcnt = dend - d; | |
4231 | ||
4232 | /* Want how many consecutive characters we can match in | |
4233 | one shot, so, if necessary, adjust the count. */ | |
4234 | if (mcnt > dend2 - d2) | |
4235 | mcnt = dend2 - d2; | |
4236 | ||
4237 | /* Compare that many; failure if mismatch, else move | |
4238 | past them. */ | |
4239 | if (translate | |
4240 | ? bcmp_translate (d, d2, mcnt, translate) | |
4241 | : bcmp (d, d2, mcnt)) | |
4242 | goto fail; | |
4243 | d += mcnt, d2 += mcnt; | |
4244 | ||
4245 | /* Do this because we've match some characters. */ | |
4246 | SET_REGS_MATCHED (); | |
4247 | } | |
4248 | } | |
4249 | break; | |
4250 | ||
4251 | ||
4252 | /* begline matches the empty string at the beginning of the string | |
4253 | (unless `not_bol' is set in `bufp'), and, if | |
4254 | `newline_anchor' is set, after newlines. */ | |
4255 | case begline: | |
4256 | DEBUG_PRINT1 ("EXECUTING begline.\n"); | |
4257 | ||
4258 | if (AT_STRINGS_BEG (d)) | |
4259 | { | |
4260 | if (!bufp->not_bol) break; | |
4261 | } | |
4262 | else if (d[-1] == '\n' && bufp->newline_anchor) | |
4263 | { | |
4264 | break; | |
4265 | } | |
4266 | /* In all other cases, we fail. */ | |
4267 | goto fail; | |
4268 | ||
4269 | ||
4270 | /* endline is the dual of begline. */ | |
4271 | case endline: | |
4272 | DEBUG_PRINT1 ("EXECUTING endline.\n"); | |
4273 | ||
4274 | if (AT_STRINGS_END (d)) | |
4275 | { | |
4276 | if (!bufp->not_eol) break; | |
4277 | } | |
4278 | ||
4279 | /* We have to ``prefetch'' the next character. */ | |
4280 | else if ((d == end1 ? *string2 : *d) == '\n' | |
4281 | && bufp->newline_anchor) | |
4282 | { | |
4283 | break; | |
4284 | } | |
4285 | goto fail; | |
4286 | ||
4287 | ||
4288 | /* Match at the very beginning of the data. */ | |
4289 | case begbuf: | |
4290 | DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
4291 | if (AT_STRINGS_BEG (d)) | |
4292 | break; | |
4293 | goto fail; | |
4294 | ||
4295 | ||
4296 | /* Match at the very end of the data. */ | |
4297 | case endbuf: | |
4298 | DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
4299 | if (AT_STRINGS_END (d)) | |
4300 | break; | |
4301 | goto fail; | |
4302 | ||
4303 | ||
4304 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It | |
4305 | pushes NULL as the value for the string on the stack. Then | |
4306 | `pop_failure_point' will keep the current value for the | |
4307 | string, instead of restoring it. To see why, consider | |
4308 | matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
4309 | then the . fails against the \n. But the next thing we want | |
4310 | to do is match the \n against the \n; if we restored the | |
4311 | string value, we would be back at the foo. | |
4312 | ||
4313 | Because this is used only in specific cases, we don't need to | |
4314 | check all the things that `on_failure_jump' does, to make | |
4315 | sure the right things get saved on the stack. Hence we don't | |
4316 | share its code. The only reason to push anything on the | |
4317 | stack at all is that otherwise we would have to change | |
4318 | `anychar's code to do something besides goto fail in this | |
4319 | case; that seems worse than this. */ | |
4320 | case on_failure_keep_string_jump: | |
4321 | DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); | |
4322 | ||
4323 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4324 | DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); | |
4325 | ||
4326 | PUSH_FAILURE_POINT (p + mcnt, NULL, -2); | |
4327 | break; | |
4328 | ||
4329 | ||
4330 | /* Uses of on_failure_jump: | |
4331 | ||
4332 | Each alternative starts with an on_failure_jump that points | |
4333 | to the beginning of the next alternative. Each alternative | |
4334 | except the last ends with a jump that in effect jumps past | |
4335 | the rest of the alternatives. (They really jump to the | |
4336 | ending jump of the following alternative, because tensioning | |
4337 | these jumps is a hassle.) | |
4338 | ||
4339 | Repeats start with an on_failure_jump that points past both | |
4340 | the repetition text and either the following jump or | |
4341 | pop_failure_jump back to this on_failure_jump. */ | |
4342 | case on_failure_jump: | |
4343 | on_failure: | |
4344 | DEBUG_PRINT1 ("EXECUTING on_failure_jump"); | |
4345 | ||
4346 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4347 | DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); | |
4348 | ||
4349 | /* If this on_failure_jump comes right before a group (i.e., | |
4350 | the original * applied to a group), save the information | |
4351 | for that group and all inner ones, so that if we fail back | |
4352 | to this point, the group's information will be correct. | |
4353 | For example, in \(a*\)*\1, we need the preceding group, | |
4354 | and in \(\(a*\)b*\)\2, we need the inner group. */ | |
4355 | ||
4356 | /* We can't use `p' to check ahead because we push | |
4357 | a failure point to `p + mcnt' after we do this. */ | |
4358 | p1 = p; | |
4359 | ||
4360 | /* We need to skip no_op's before we look for the | |
4361 | start_memory in case this on_failure_jump is happening as | |
4362 | the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 | |
4363 | against aba. */ | |
4364 | while (p1 < pend && (re_opcode_t) *p1 == no_op) | |
4365 | p1++; | |
4366 | ||
4367 | if (p1 < pend && (re_opcode_t) *p1 == start_memory) | |
4368 | { | |
4369 | /* We have a new highest active register now. This will | |
4370 | get reset at the start_memory we are about to get to, | |
4371 | but we will have saved all the registers relevant to | |
4372 | this repetition op, as described above. */ | |
4373 | highest_active_reg = *(p1 + 1) + *(p1 + 2); | |
4374 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
4375 | lowest_active_reg = *(p1 + 1); | |
4376 | } | |
4377 | ||
4378 | DEBUG_PRINT1 (":\n"); | |
4379 | PUSH_FAILURE_POINT (p + mcnt, d, -2); | |
4380 | break; | |
4381 | ||
4382 | ||
4383 | /* A smart repeat ends with `maybe_pop_jump'. | |
4384 | We change it to either `pop_failure_jump' or `jump'. */ | |
4385 | case maybe_pop_jump: | |
4386 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4387 | DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); | |
4388 | { | |
4389 | register unsigned char *p2 = p; | |
4390 | ||
4391 | /* Compare the beginning of the repeat with what in the | |
4392 | pattern follows its end. If we can establish that there | |
4393 | is nothing that they would both match, i.e., that we | |
4394 | would have to backtrack because of (as in, e.g., `a*a') | |
4395 | then we can change to pop_failure_jump, because we'll | |
4396 | never have to backtrack. | |
4397 | ||
4398 | This is not true in the case of alternatives: in | |
4399 | `(a|ab)*' we do need to backtrack to the `ab' alternative | |
4400 | (e.g., if the string was `ab'). But instead of trying to | |
4401 | detect that here, the alternative has put on a dummy | |
4402 | failure point which is what we will end up popping. */ | |
4403 | ||
4404 | /* Skip over open/close-group commands. | |
4405 | If what follows this loop is a ...+ construct, | |
4406 | look at what begins its body, since we will have to | |
4407 | match at least one of that. */ | |
4408 | while (1) | |
4409 | { | |
4410 | if (p2 + 2 < pend | |
4411 | && ((re_opcode_t) *p2 == stop_memory | |
4412 | || (re_opcode_t) *p2 == start_memory)) | |
4413 | p2 += 3; | |
4414 | else if (p2 + 6 < pend | |
4415 | && (re_opcode_t) *p2 == dummy_failure_jump) | |
4416 | p2 += 6; | |
4417 | else | |
4418 | break; | |
4419 | } | |
4420 | ||
4421 | p1 = p + mcnt; | |
4422 | /* p1[0] ... p1[2] are the `on_failure_jump' corresponding | |
4423 | to the `maybe_finalize_jump' of this case. Examine what | |
4424 | follows. */ | |
4425 | ||
4426 | /* If we're at the end of the pattern, we can change. */ | |
4427 | if (p2 == pend) | |
4428 | { | |
4429 | /* Consider what happens when matching ":\(.*\)" | |
4430 | against ":/". I don't really understand this code | |
4431 | yet. */ | |
4432 | p[-3] = (unsigned char) pop_failure_jump; | |
4433 | DEBUG_PRINT1 | |
4434 | (" End of pattern: change to `pop_failure_jump'.\n"); | |
4435 | } | |
4436 | ||
4437 | else if ((re_opcode_t) *p2 == exactn | |
4438 | || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) | |
4439 | { | |
4440 | register unsigned char c | |
4441 | = *p2 == (unsigned char) endline ? '\n' : p2[2]; | |
4442 | ||
4443 | if ((re_opcode_t) p1[3] == exactn && p1[5] != c) | |
4444 | { | |
4445 | p[-3] = (unsigned char) pop_failure_jump; | |
4446 | DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", | |
4447 | c, p1[5]); | |
4448 | } | |
4449 | ||
4450 | else if ((re_opcode_t) p1[3] == charset | |
4451 | || (re_opcode_t) p1[3] == charset_not) | |
4452 | { | |
4453 | int not = (re_opcode_t) p1[3] == charset_not; | |
4454 | ||
4455 | if (c < (unsigned char) (p1[4] * BYTEWIDTH) | |
4456 | && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
4457 | not = !not; | |
4458 | ||
4459 | /* `not' is equal to 1 if c would match, which means | |
4460 | that we can't change to pop_failure_jump. */ | |
4461 | if (!not) | |
4462 | { | |
4463 | p[-3] = (unsigned char) pop_failure_jump; | |
4464 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
4465 | } | |
4466 | } | |
4467 | } | |
4468 | else if ((re_opcode_t) *p2 == charset) | |
4469 | { | |
4470 | #ifdef DEBUG | |
4471 | register unsigned char c | |
4472 | = *p2 == (unsigned char) endline ? '\n' : p2[2]; | |
4473 | #endif | |
4474 | ||
4475 | if ((re_opcode_t) p1[3] == exactn | |
4476 | && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4] | |
4477 | && (p2[1 + p1[4] / BYTEWIDTH] | |
4478 | & (1 << (p1[4] % BYTEWIDTH))))) | |
4479 | { | |
4480 | p[-3] = (unsigned char) pop_failure_jump; | |
4481 | DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", | |
4482 | c, p1[5]); | |
4483 | } | |
4484 | ||
4485 | else if ((re_opcode_t) p1[3] == charset_not) | |
4486 | { | |
4487 | int idx; | |
4488 | /* We win if the charset_not inside the loop | |
4489 | lists every character listed in the charset after. */ | |
4490 | for (idx = 0; idx < (int) p2[1]; idx++) | |
4491 | if (! (p2[2 + idx] == 0 | |
4492 | || (idx < (int) p1[4] | |
4493 | && ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) | |
4494 | break; | |
4495 | ||
4496 | if (idx == p2[1]) | |
4497 | { | |
4498 | p[-3] = (unsigned char) pop_failure_jump; | |
4499 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
4500 | } | |
4501 | } | |
4502 | else if ((re_opcode_t) p1[3] == charset) | |
4503 | { | |
4504 | int idx; | |
4505 | /* We win if the charset inside the loop | |
4506 | has no overlap with the one after the loop. */ | |
4507 | for (idx = 0; | |
4508 | idx < (int) p2[1] && idx < (int) p1[4]; | |
4509 | idx++) | |
4510 | if ((p2[2 + idx] & p1[5 + idx]) != 0) | |
4511 | break; | |
4512 | ||
4513 | if (idx == p2[1] || idx == p1[4]) | |
4514 | { | |
4515 | p[-3] = (unsigned char) pop_failure_jump; | |
4516 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
4517 | } | |
4518 | } | |
4519 | } | |
4520 | } | |
4521 | p -= 2; /* Point at relative address again. */ | |
4522 | if ((re_opcode_t) p[-1] != pop_failure_jump) | |
4523 | { | |
4524 | p[-1] = (unsigned char) jump; | |
4525 | DEBUG_PRINT1 (" Match => jump.\n"); | |
4526 | goto unconditional_jump; | |
4527 | } | |
4528 | /* Note fall through. */ | |
4529 | ||
4530 | ||
4531 | /* The end of a simple repeat has a pop_failure_jump back to | |
4532 | its matching on_failure_jump, where the latter will push a | |
4533 | failure point. The pop_failure_jump takes off failure | |
4534 | points put on by this pop_failure_jump's matching | |
4535 | on_failure_jump; we got through the pattern to here from the | |
4536 | matching on_failure_jump, so didn't fail. */ | |
4537 | case pop_failure_jump: | |
4538 | { | |
4539 | /* We need to pass separate storage for the lowest and | |
4540 | highest registers, even though we don't care about the | |
4541 | actual values. Otherwise, we will restore only one | |
4542 | register from the stack, since lowest will == highest in | |
4543 | `pop_failure_point'. */ | |
4544 | unsigned dummy_low_reg, dummy_high_reg; | |
4545 | unsigned char *pdummy; | |
4546 | const char *sdummy; | |
4547 | ||
4548 | DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); | |
4549 | POP_FAILURE_POINT (sdummy, pdummy, | |
4550 | dummy_low_reg, dummy_high_reg, | |
4551 | reg_dummy, reg_dummy, reg_info_dummy); | |
4552 | } | |
4553 | /* Note fall through. */ | |
4554 | ||
4555 | ||
4556 | /* Unconditionally jump (without popping any failure points). */ | |
4557 | case jump: | |
4558 | unconditional_jump: | |
4559 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ | |
4560 | DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); | |
4561 | p += mcnt; /* Do the jump. */ | |
4562 | DEBUG_PRINT2 ("(to 0x%x).\n", p); | |
4563 | break; | |
4564 | ||
4565 | ||
4566 | /* We need this opcode so we can detect where alternatives end | |
4567 | in `group_match_null_string_p' et al. */ | |
4568 | case jump_past_alt: | |
4569 | DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); | |
4570 | goto unconditional_jump; | |
4571 | ||
4572 | ||
4573 | /* Normally, the on_failure_jump pushes a failure point, which | |
4574 | then gets popped at pop_failure_jump. We will end up at | |
4575 | pop_failure_jump, also, and with a pattern of, say, `a+', we | |
4576 | are skipping over the on_failure_jump, so we have to push | |
4577 | something meaningless for pop_failure_jump to pop. */ | |
4578 | case dummy_failure_jump: | |
4579 | DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); | |
4580 | /* It doesn't matter what we push for the string here. What | |
4581 | the code at `fail' tests is the value for the pattern. */ | |
4582 | PUSH_FAILURE_POINT (0, 0, -2); | |
4583 | goto unconditional_jump; | |
4584 | ||
4585 | ||
4586 | /* At the end of an alternative, we need to push a dummy failure | |
4587 | point in case we are followed by a `pop_failure_jump', because | |
4588 | we don't want the failure point for the alternative to be | |
4589 | popped. For example, matching `(a|ab)*' against `aab' | |
4590 | requires that we match the `ab' alternative. */ | |
4591 | case push_dummy_failure: | |
4592 | DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); | |
4593 | /* See comments just above at `dummy_failure_jump' about the | |
4594 | two zeroes. */ | |
4595 | PUSH_FAILURE_POINT (0, 0, -2); | |
4596 | break; | |
4597 | ||
4598 | /* Have to succeed matching what follows at least n times. | |
4599 | After that, handle like `on_failure_jump'. */ | |
4600 | case succeed_n: | |
4601 | EXTRACT_NUMBER (mcnt, p + 2); | |
4602 | DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); | |
4603 | ||
4604 | assert (mcnt >= 0); | |
4605 | /* Originally, this is how many times we HAVE to succeed. */ | |
4606 | if (mcnt > 0) | |
4607 | { | |
4608 | mcnt--; | |
4609 | p += 2; | |
4610 | STORE_NUMBER_AND_INCR (p, mcnt); | |
4611 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt); | |
4612 | } | |
4613 | else if (mcnt == 0) | |
4614 | { | |
4615 | DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2); | |
4616 | p[2] = (unsigned char) no_op; | |
4617 | p[3] = (unsigned char) no_op; | |
4618 | goto on_failure; | |
4619 | } | |
4620 | break; | |
4621 | ||
4622 | case jump_n: | |
4623 | EXTRACT_NUMBER (mcnt, p + 2); | |
4624 | DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); | |
4625 | ||
4626 | /* Originally, this is how many times we CAN jump. */ | |
4627 | if (mcnt) | |
4628 | { | |
4629 | mcnt--; | |
4630 | STORE_NUMBER (p + 2, mcnt); | |
4631 | goto unconditional_jump; | |
4632 | } | |
4633 | /* If don't have to jump any more, skip over the rest of command. */ | |
4634 | else | |
4635 | p += 4; | |
4636 | break; | |
4637 | ||
4638 | case set_number_at: | |
4639 | { | |
4640 | DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); | |
4641 | ||
4642 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4643 | p1 = p + mcnt; | |
4644 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4645 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); | |
4646 | STORE_NUMBER (p1, mcnt); | |
4647 | break; | |
4648 | } | |
4649 | ||
4650 | case wordbound: | |
4651 | DEBUG_PRINT1 ("EXECUTING wordbound.\n"); | |
4652 | if (AT_WORD_BOUNDARY (d)) | |
4653 | break; | |
4654 | goto fail; | |
4655 | ||
4656 | case notwordbound: | |
4657 | DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); | |
4658 | if (AT_WORD_BOUNDARY (d)) | |
4659 | goto fail; | |
4660 | break; | |
4661 | ||
4662 | case wordbeg: | |
4663 | DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); | |
4664 | if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) | |
4665 | break; | |
4666 | goto fail; | |
4667 | ||
4668 | case wordend: | |
4669 | DEBUG_PRINT1 ("EXECUTING wordend.\n"); | |
4670 | if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) | |
4671 | && (!WORDCHAR_P (d) || AT_STRINGS_END (d))) | |
4672 | break; | |
4673 | goto fail; | |
4674 | ||
4675 | #ifdef emacs | |
4676 | case before_dot: | |
4677 | DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
4678 | if (PTR_CHAR_POS ((unsigned char *) d) >= point) | |
4679 | goto fail; | |
4680 | break; | |
4681 | ||
4682 | case at_dot: | |
4683 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
4684 | if (PTR_CHAR_POS ((unsigned char *) d) != point) | |
4685 | goto fail; | |
4686 | break; | |
4687 | ||
4688 | case after_dot: | |
4689 | DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
4690 | if (PTR_CHAR_POS ((unsigned char *) d) <= point) | |
4691 | goto fail; | |
4692 | break; | |
fa9a63c5 RM |
4693 | |
4694 | case syntaxspec: | |
4695 | DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); | |
4696 | mcnt = *p++; | |
4697 | goto matchsyntax; | |
4698 | ||
4699 | case wordchar: | |
4700 | DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n"); | |
4701 | mcnt = (int) Sword; | |
4702 | matchsyntax: | |
4703 | PREFETCH (); | |
4704 | /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ | |
4705 | d++; | |
4706 | if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt) | |
4707 | goto fail; | |
4708 | SET_REGS_MATCHED (); | |
4709 | break; | |
4710 | ||
4711 | case notsyntaxspec: | |
4712 | DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); | |
4713 | mcnt = *p++; | |
4714 | goto matchnotsyntax; | |
4715 | ||
4716 | case notwordchar: | |
4717 | DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n"); | |
4718 | mcnt = (int) Sword; | |
4719 | matchnotsyntax: | |
4720 | PREFETCH (); | |
4721 | /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ | |
4722 | d++; | |
4723 | if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt) | |
4724 | goto fail; | |
4725 | SET_REGS_MATCHED (); | |
4726 | break; | |
4727 | ||
4728 | #else /* not emacs */ | |
4729 | case wordchar: | |
4730 | DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); | |
4731 | PREFETCH (); | |
4732 | if (!WORDCHAR_P (d)) | |
4733 | goto fail; | |
4734 | SET_REGS_MATCHED (); | |
4735 | d++; | |
4736 | break; | |
4737 | ||
4738 | case notwordchar: | |
4739 | DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); | |
4740 | PREFETCH (); | |
4741 | if (WORDCHAR_P (d)) | |
4742 | goto fail; | |
4743 | SET_REGS_MATCHED (); | |
4744 | d++; | |
4745 | break; | |
4746 | #endif /* not emacs */ | |
4747 | ||
4748 | default: | |
4749 | abort (); | |
4750 | } | |
4751 | continue; /* Successfully executed one pattern command; keep going. */ | |
4752 | ||
4753 | ||
4754 | /* We goto here if a matching operation fails. */ | |
4755 | fail: | |
4756 | if (!FAIL_STACK_EMPTY ()) | |
4757 | { /* A restart point is known. Restore to that state. */ | |
4758 | DEBUG_PRINT1 ("\nFAIL:\n"); | |
4759 | POP_FAILURE_POINT (d, p, | |
4760 | lowest_active_reg, highest_active_reg, | |
4761 | regstart, regend, reg_info); | |
4762 | ||
4763 | /* If this failure point is a dummy, try the next one. */ | |
4764 | if (!p) | |
4765 | goto fail; | |
4766 | ||
4767 | /* If we failed to the end of the pattern, don't examine *p. */ | |
4768 | assert (p <= pend); | |
4769 | if (p < pend) | |
4770 | { | |
4771 | boolean is_a_jump_n = false; | |
4772 | ||
4773 | /* If failed to a backwards jump that's part of a repetition | |
4774 | loop, need to pop this failure point and use the next one. */ | |
4775 | switch ((re_opcode_t) *p) | |
4776 | { | |
4777 | case jump_n: | |
4778 | is_a_jump_n = true; | |
4779 | case maybe_pop_jump: | |
4780 | case pop_failure_jump: | |
4781 | case jump: | |
4782 | p1 = p + 1; | |
4783 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4784 | p1 += mcnt; | |
4785 | ||
4786 | if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) | |
4787 | || (!is_a_jump_n | |
4788 | && (re_opcode_t) *p1 == on_failure_jump)) | |
4789 | goto fail; | |
4790 | break; | |
4791 | default: | |
4792 | /* do nothing */ ; | |
4793 | } | |
4794 | } | |
4795 | ||
4796 | if (d >= string1 && d <= end1) | |
4797 | dend = end_match_1; | |
4798 | } | |
4799 | else | |
4800 | break; /* Matching at this starting point really fails. */ | |
4801 | } /* for (;;) */ | |
4802 | ||
4803 | if (best_regs_set) | |
4804 | goto restore_best_regs; | |
4805 | ||
4806 | FREE_VARIABLES (); | |
4807 | ||
4808 | return -1; /* Failure to match. */ | |
4809 | } /* re_match_2 */ | |
4810 | \f | |
4811 | /* Subroutine definitions for re_match_2. */ | |
4812 | ||
4813 | ||
4814 | /* We are passed P pointing to a register number after a start_memory. | |
4815 | ||
4816 | Return true if the pattern up to the corresponding stop_memory can | |
4817 | match the empty string, and false otherwise. | |
4818 | ||
4819 | If we find the matching stop_memory, sets P to point to one past its number. | |
4820 | Otherwise, sets P to an undefined byte less than or equal to END. | |
4821 | ||
4822 | We don't handle duplicates properly (yet). */ | |
4823 | ||
4824 | static boolean | |
4825 | group_match_null_string_p (p, end, reg_info) | |
4826 | unsigned char **p, *end; | |
4827 | register_info_type *reg_info; | |
4828 | { | |
4829 | int mcnt; | |
4830 | /* Point to after the args to the start_memory. */ | |
4831 | unsigned char *p1 = *p + 2; | |
4832 | ||
4833 | while (p1 < end) | |
4834 | { | |
4835 | /* Skip over opcodes that can match nothing, and return true or | |
4836 | false, as appropriate, when we get to one that can't, or to the | |
4837 | matching stop_memory. */ | |
4838 | ||
4839 | switch ((re_opcode_t) *p1) | |
4840 | { | |
4841 | /* Could be either a loop or a series of alternatives. */ | |
4842 | case on_failure_jump: | |
4843 | p1++; | |
4844 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4845 | ||
4846 | /* If the next operation is not a jump backwards in the | |
4847 | pattern. */ | |
4848 | ||
4849 | if (mcnt >= 0) | |
4850 | { | |
4851 | /* Go through the on_failure_jumps of the alternatives, | |
4852 | seeing if any of the alternatives cannot match nothing. | |
4853 | The last alternative starts with only a jump, | |
4854 | whereas the rest start with on_failure_jump and end | |
4855 | with a jump, e.g., here is the pattern for `a|b|c': | |
4856 | ||
4857 | /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 | |
4858 | /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 | |
4859 | /exactn/1/c | |
4860 | ||
4861 | So, we have to first go through the first (n-1) | |
4862 | alternatives and then deal with the last one separately. */ | |
4863 | ||
4864 | ||
4865 | /* Deal with the first (n-1) alternatives, which start | |
4866 | with an on_failure_jump (see above) that jumps to right | |
4867 | past a jump_past_alt. */ | |
4868 | ||
4869 | while ((re_opcode_t) p1[mcnt-3] == jump_past_alt) | |
4870 | { | |
4871 | /* `mcnt' holds how many bytes long the alternative | |
4872 | is, including the ending `jump_past_alt' and | |
4873 | its number. */ | |
4874 | ||
4875 | if (!alt_match_null_string_p (p1, p1 + mcnt - 3, | |
4876 | reg_info)) | |
4877 | return false; | |
4878 | ||
4879 | /* Move to right after this alternative, including the | |
4880 | jump_past_alt. */ | |
4881 | p1 += mcnt; | |
4882 | ||
4883 | /* Break if it's the beginning of an n-th alternative | |
4884 | that doesn't begin with an on_failure_jump. */ | |
4885 | if ((re_opcode_t) *p1 != on_failure_jump) | |
4886 | break; | |
4887 | ||
4888 | /* Still have to check that it's not an n-th | |
4889 | alternative that starts with an on_failure_jump. */ | |
4890 | p1++; | |
4891 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4892 | if ((re_opcode_t) p1[mcnt-3] != jump_past_alt) | |
4893 | { | |
4894 | /* Get to the beginning of the n-th alternative. */ | |
4895 | p1 -= 3; | |
4896 | break; | |
4897 | } | |
4898 | } | |
4899 | ||
4900 | /* Deal with the last alternative: go back and get number | |
4901 | of the `jump_past_alt' just before it. `mcnt' contains | |
4902 | the length of the alternative. */ | |
4903 | EXTRACT_NUMBER (mcnt, p1 - 2); | |
4904 | ||
4905 | if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info)) | |
4906 | return false; | |
4907 | ||
4908 | p1 += mcnt; /* Get past the n-th alternative. */ | |
4909 | } /* if mcnt > 0 */ | |
4910 | break; | |
4911 | ||
4912 | ||
4913 | case stop_memory: | |
4914 | assert (p1[1] == **p); | |
4915 | *p = p1 + 2; | |
4916 | return true; | |
4917 | ||
4918 | ||
4919 | default: | |
4920 | if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
4921 | return false; | |
4922 | } | |
4923 | } /* while p1 < end */ | |
4924 | ||
4925 | return false; | |
4926 | } /* group_match_null_string_p */ | |
4927 | ||
4928 | ||
4929 | /* Similar to group_match_null_string_p, but doesn't deal with alternatives: | |
4930 | It expects P to be the first byte of a single alternative and END one | |
4931 | byte past the last. The alternative can contain groups. */ | |
4932 | ||
4933 | static boolean | |
4934 | alt_match_null_string_p (p, end, reg_info) | |
4935 | unsigned char *p, *end; | |
4936 | register_info_type *reg_info; | |
4937 | { | |
4938 | int mcnt; | |
4939 | unsigned char *p1 = p; | |
4940 | ||
4941 | while (p1 < end) | |
4942 | { | |
4943 | /* Skip over opcodes that can match nothing, and break when we get | |
4944 | to one that can't. */ | |
4945 | ||
4946 | switch ((re_opcode_t) *p1) | |
4947 | { | |
4948 | /* It's a loop. */ | |
4949 | case on_failure_jump: | |
4950 | p1++; | |
4951 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4952 | p1 += mcnt; | |
4953 | break; | |
4954 | ||
4955 | default: | |
4956 | if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
4957 | return false; | |
4958 | } | |
4959 | } /* while p1 < end */ | |
4960 | ||
4961 | return true; | |
4962 | } /* alt_match_null_string_p */ | |
4963 | ||
4964 | ||
4965 | /* Deals with the ops common to group_match_null_string_p and | |
4966 | alt_match_null_string_p. | |
4967 | ||
4968 | Sets P to one after the op and its arguments, if any. */ | |
4969 | ||
4970 | static boolean | |
4971 | common_op_match_null_string_p (p, end, reg_info) | |
4972 | unsigned char **p, *end; | |
4973 | register_info_type *reg_info; | |
4974 | { | |
4975 | int mcnt; | |
4976 | boolean ret; | |
4977 | int reg_no; | |
4978 | unsigned char *p1 = *p; | |
4979 | ||
4980 | switch ((re_opcode_t) *p1++) | |
4981 | { | |
4982 | case no_op: | |
4983 | case begline: | |
4984 | case endline: | |
4985 | case begbuf: | |
4986 | case endbuf: | |
4987 | case wordbeg: | |
4988 | case wordend: | |
4989 | case wordbound: | |
4990 | case notwordbound: | |
4991 | #ifdef emacs | |
4992 | case before_dot: | |
4993 | case at_dot: | |
4994 | case after_dot: | |
4995 | #endif | |
4996 | break; | |
4997 | ||
4998 | case start_memory: | |
4999 | reg_no = *p1; | |
5000 | assert (reg_no > 0 && reg_no <= MAX_REGNUM); | |
5001 | ret = group_match_null_string_p (&p1, end, reg_info); | |
5002 | ||
5003 | /* Have to set this here in case we're checking a group which | |
5004 | contains a group and a back reference to it. */ | |
5005 | ||
5006 | if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) | |
5007 | REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; | |
5008 | ||
5009 | if (!ret) | |
5010 | return false; | |
5011 | break; | |
5012 | ||
5013 | /* If this is an optimized succeed_n for zero times, make the jump. */ | |
5014 | case jump: | |
5015 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
5016 | if (mcnt >= 0) | |
5017 | p1 += mcnt; | |
5018 | else | |
5019 | return false; | |
5020 | break; | |
5021 | ||
5022 | case succeed_n: | |
5023 | /* Get to the number of times to succeed. */ | |
5024 | p1 += 2; | |
5025 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
5026 | ||
5027 | if (mcnt == 0) | |
5028 | { | |
5029 | p1 -= 4; | |
5030 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
5031 | p1 += mcnt; | |
5032 | } | |
5033 | else | |
5034 | return false; | |
5035 | break; | |
5036 | ||
5037 | case duplicate: | |
5038 | if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) | |
5039 | return false; | |
5040 | break; | |
5041 | ||
5042 | case set_number_at: | |
5043 | p1 += 4; | |
5044 | ||
5045 | default: | |
5046 | /* All other opcodes mean we cannot match the empty string. */ | |
5047 | return false; | |
5048 | } | |
5049 | ||
5050 | *p = p1; | |
5051 | return true; | |
5052 | } /* common_op_match_null_string_p */ | |
5053 | ||
5054 | ||
5055 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN | |
5056 | bytes; nonzero otherwise. */ | |
5057 | ||
5058 | static int | |
5059 | bcmp_translate (s1, s2, len, translate) | |
5060 | unsigned char *s1, *s2; | |
5061 | register int len; | |
5062 | char *translate; | |
5063 | { | |
5064 | register unsigned char *p1 = s1, *p2 = s2; | |
5065 | while (len) | |
5066 | { | |
5067 | if (translate[*p1++] != translate[*p2++]) return 1; | |
5068 | len--; | |
5069 | } | |
5070 | return 0; | |
5071 | } | |
5072 | \f | |
5073 | /* Entry points for GNU code. */ | |
5074 | ||
5075 | /* re_compile_pattern is the GNU regular expression compiler: it | |
5076 | compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
5077 | Returns 0 if the pattern was valid, otherwise an error string. | |
5078 | ||
5079 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields | |
5080 | are set in BUFP on entry. | |
5081 | ||
5082 | We call regex_compile to do the actual compilation. */ | |
5083 | ||
5084 | const char * | |
5085 | re_compile_pattern (pattern, length, bufp) | |
5086 | const char *pattern; | |
5087 | int length; | |
5088 | struct re_pattern_buffer *bufp; | |
5089 | { | |
5090 | reg_errcode_t ret; | |
5091 | ||
5092 | /* GNU code is written to assume at least RE_NREGS registers will be set | |
5093 | (and at least one extra will be -1). */ | |
5094 | bufp->regs_allocated = REGS_UNALLOCATED; | |
5095 | ||
5096 | /* And GNU code determines whether or not to get register information | |
5097 | by passing null for the REGS argument to re_match, etc., not by | |
5098 | setting no_sub. */ | |
5099 | bufp->no_sub = 0; | |
5100 | ||
5101 | /* Match anchors at newline. */ | |
5102 | bufp->newline_anchor = 1; | |
5103 | ||
5104 | ret = regex_compile (pattern, length, re_syntax_options, bufp); | |
5105 | ||
5106 | if (!ret) | |
5107 | return NULL; | |
5108 | return gettext (re_error_msgid[(int) ret]); | |
5109 | } | |
5110 | \f | |
5111 | /* Entry points compatible with 4.2 BSD regex library. We don't define | |
5112 | them unless specifically requested. */ | |
5113 | ||
5114 | #ifdef _REGEX_RE_COMP | |
5115 | ||
5116 | /* BSD has one and only one pattern buffer. */ | |
5117 | static struct re_pattern_buffer re_comp_buf; | |
5118 | ||
5119 | char * | |
5120 | re_comp (s) | |
5121 | const char *s; | |
5122 | { | |
5123 | reg_errcode_t ret; | |
5124 | ||
5125 | if (!s) | |
5126 | { | |
5127 | if (!re_comp_buf.buffer) | |
5128 | return gettext ("No previous regular expression"); | |
5129 | return 0; | |
5130 | } | |
5131 | ||
5132 | if (!re_comp_buf.buffer) | |
5133 | { | |
5134 | re_comp_buf.buffer = (unsigned char *) malloc (200); | |
5135 | if (re_comp_buf.buffer == NULL) | |
5136 | return gettext (re_error_msgid[(int) REG_ESPACE]); | |
5137 | re_comp_buf.allocated = 200; | |
5138 | ||
5139 | re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
5140 | if (re_comp_buf.fastmap == NULL) | |
5141 | return gettext (re_error_msgid[(int) REG_ESPACE]); | |
5142 | } | |
5143 | ||
5144 | /* Since `re_exec' always passes NULL for the `regs' argument, we | |
5145 | don't need to initialize the pattern buffer fields which affect it. */ | |
5146 | ||
5147 | /* Match anchors at newlines. */ | |
5148 | re_comp_buf.newline_anchor = 1; | |
5149 | ||
5150 | ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); | |
5151 | ||
5152 | if (!ret) | |
5153 | return NULL; | |
5154 | ||
5155 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ | |
5156 | return (char *) gettext (re_error_msgid[(int) ret]); | |
5157 | } | |
5158 | ||
5159 | ||
5160 | int | |
5161 | re_exec (s) | |
5162 | const char *s; | |
5163 | { | |
5164 | const int len = strlen (s); | |
5165 | return | |
5166 | 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
5167 | } | |
5168 | #endif /* _REGEX_RE_COMP */ | |
5169 | \f | |
5170 | /* POSIX.2 functions. Don't define these for Emacs. */ | |
5171 | ||
5172 | #ifndef emacs | |
5173 | ||
5174 | /* regcomp takes a regular expression as a string and compiles it. | |
5175 | ||
5176 | PREG is a regex_t *. We do not expect any fields to be initialized, | |
5177 | since POSIX says we shouldn't. Thus, we set | |
5178 | ||
5179 | `buffer' to the compiled pattern; | |
5180 | `used' to the length of the compiled pattern; | |
5181 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
5182 | REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
5183 | RE_SYNTAX_POSIX_BASIC; | |
5184 | `newline_anchor' to REG_NEWLINE being set in CFLAGS; | |
5185 | `fastmap' and `fastmap_accurate' to zero; | |
5186 | `re_nsub' to the number of subexpressions in PATTERN. | |
5187 | ||
5188 | PATTERN is the address of the pattern string. | |
5189 | ||
5190 | CFLAGS is a series of bits which affect compilation. | |
5191 | ||
5192 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
5193 | use POSIX basic syntax. | |
5194 | ||
5195 | If REG_NEWLINE is set, then . and [^...] don't match newline. | |
5196 | Also, regexec will try a match beginning after every newline. | |
5197 | ||
5198 | If REG_ICASE is set, then we considers upper- and lowercase | |
5199 | versions of letters to be equivalent when matching. | |
5200 | ||
5201 | If REG_NOSUB is set, then when PREG is passed to regexec, that | |
5202 | routine will report only success or failure, and nothing about the | |
5203 | registers. | |
5204 | ||
5205 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for | |
5206 | the return codes and their meanings.) */ | |
5207 | ||
5208 | int | |
5209 | regcomp (preg, pattern, cflags) | |
5210 | regex_t *preg; | |
5211 | const char *pattern; | |
5212 | int cflags; | |
5213 | { | |
5214 | reg_errcode_t ret; | |
5215 | unsigned syntax | |
5216 | = (cflags & REG_EXTENDED) ? | |
5217 | RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; | |
5218 | ||
5219 | /* regex_compile will allocate the space for the compiled pattern. */ | |
5220 | preg->buffer = 0; | |
5221 | preg->allocated = 0; | |
5222 | preg->used = 0; | |
5223 | ||
5224 | /* Don't bother to use a fastmap when searching. This simplifies the | |
5225 | REG_NEWLINE case: if we used a fastmap, we'd have to put all the | |
5226 | characters after newlines into the fastmap. This way, we just try | |
5227 | every character. */ | |
5228 | preg->fastmap = 0; | |
5229 | ||
5230 | if (cflags & REG_ICASE) | |
5231 | { | |
5232 | unsigned i; | |
5233 | ||
5234 | preg->translate = (char *) malloc (CHAR_SET_SIZE); | |
5235 | if (preg->translate == NULL) | |
5236 | return (int) REG_ESPACE; | |
5237 | ||
5238 | /* Map uppercase characters to corresponding lowercase ones. */ | |
5239 | for (i = 0; i < CHAR_SET_SIZE; i++) | |
5240 | preg->translate[i] = ISUPPER (i) ? tolower (i) : i; | |
5241 | } | |
5242 | else | |
5243 | preg->translate = NULL; | |
5244 | ||
5245 | /* If REG_NEWLINE is set, newlines are treated differently. */ | |
5246 | if (cflags & REG_NEWLINE) | |
5247 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
5248 | syntax &= ~RE_DOT_NEWLINE; | |
5249 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
5250 | /* It also changes the matching behavior. */ | |
5251 | preg->newline_anchor = 1; | |
5252 | } | |
5253 | else | |
5254 | preg->newline_anchor = 0; | |
5255 | ||
5256 | preg->no_sub = !!(cflags & REG_NOSUB); | |
5257 | ||
5258 | /* POSIX says a null character in the pattern terminates it, so we | |
5259 | can use strlen here in compiling the pattern. */ | |
5260 | ret = regex_compile (pattern, strlen (pattern), syntax, preg); | |
5261 | ||
5262 | /* POSIX doesn't distinguish between an unmatched open-group and an | |
5263 | unmatched close-group: both are REG_EPAREN. */ | |
5264 | if (ret == REG_ERPAREN) ret = REG_EPAREN; | |
5265 | ||
5266 | return (int) ret; | |
5267 | } | |
5268 | ||
5269 | ||
5270 | /* regexec searches for a given pattern, specified by PREG, in the | |
5271 | string STRING. | |
5272 | ||
5273 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to | |
5274 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at | |
5275 | least NMATCH elements, and we set them to the offsets of the | |
5276 | corresponding matched substrings. | |
5277 | ||
5278 | EFLAGS specifies `execution flags' which affect matching: if | |
5279 | REG_NOTBOL is set, then ^ does not match at the beginning of the | |
5280 | string; if REG_NOTEOL is set, then $ does not match at the end. | |
5281 | ||
5282 | We return 0 if we find a match and REG_NOMATCH if not. */ | |
5283 | ||
5284 | int | |
5285 | regexec (preg, string, nmatch, pmatch, eflags) | |
5286 | const regex_t *preg; | |
5287 | const char *string; | |
5288 | size_t nmatch; | |
5289 | regmatch_t pmatch[]; | |
5290 | int eflags; | |
5291 | { | |
5292 | int ret; | |
5293 | struct re_registers regs; | |
5294 | regex_t private_preg; | |
5295 | int len = strlen (string); | |
5296 | boolean want_reg_info = !preg->no_sub && nmatch > 0; | |
5297 | ||
5298 | private_preg = *preg; | |
5299 | ||
5300 | private_preg.not_bol = !!(eflags & REG_NOTBOL); | |
5301 | private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
5302 | ||
5303 | /* The user has told us exactly how many registers to return | |
5304 | information about, via `nmatch'. We have to pass that on to the | |
5305 | matching routines. */ | |
5306 | private_preg.regs_allocated = REGS_FIXED; | |
5307 | ||
5308 | if (want_reg_info) | |
5309 | { | |
5310 | regs.num_regs = nmatch; | |
5311 | regs.start = TALLOC (nmatch, regoff_t); | |
5312 | regs.end = TALLOC (nmatch, regoff_t); | |
5313 | if (regs.start == NULL || regs.end == NULL) | |
5314 | return (int) REG_NOMATCH; | |
5315 | } | |
5316 | ||
5317 | /* Perform the searching operation. */ | |
5318 | ret = re_search (&private_preg, string, len, | |
5319 | /* start: */ 0, /* range: */ len, | |
5320 | want_reg_info ? ®s : (struct re_registers *) 0); | |
5321 | ||
5322 | /* Copy the register information to the POSIX structure. */ | |
5323 | if (want_reg_info) | |
5324 | { | |
5325 | if (ret >= 0) | |
5326 | { | |
5327 | unsigned r; | |
5328 | ||
5329 | for (r = 0; r < nmatch; r++) | |
5330 | { | |
5331 | pmatch[r].rm_so = regs.start[r]; | |
5332 | pmatch[r].rm_eo = regs.end[r]; | |
5333 | } | |
5334 | } | |
5335 | ||
5336 | /* If we needed the temporary register info, free the space now. */ | |
5337 | free (regs.start); | |
5338 | free (regs.end); | |
5339 | } | |
5340 | ||
5341 | /* We want zero return to mean success, unlike `re_search'. */ | |
5342 | return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
5343 | } | |
5344 | ||
5345 | ||
5346 | /* Returns a message corresponding to an error code, ERRCODE, returned | |
5347 | from either regcomp or regexec. We don't use PREG here. */ | |
5348 | ||
5349 | size_t | |
5350 | regerror (errcode, preg, errbuf, errbuf_size) | |
5351 | int errcode; | |
5352 | const regex_t *preg; | |
5353 | char *errbuf; | |
5354 | size_t errbuf_size; | |
5355 | { | |
5356 | const char *msg; | |
5357 | size_t msg_size; | |
5358 | ||
5359 | if (errcode < 0 | |
5360 | || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0]))) | |
5361 | /* Only error codes returned by the rest of the code should be passed | |
5362 | to this routine. If we are given anything else, or if other regex | |
5363 | code generates an invalid error code, then the program has a bug. | |
5364 | Dump core so we can fix it. */ | |
5365 | abort (); | |
5366 | ||
5367 | msg = gettext (re_error_msgid[errcode]); | |
5368 | ||
5369 | msg_size = strlen (msg) + 1; /* Includes the null. */ | |
5370 | ||
5371 | if (errbuf_size != 0) | |
5372 | { | |
5373 | if (msg_size > errbuf_size) | |
5374 | { | |
5375 | strncpy (errbuf, msg, errbuf_size - 1); | |
5376 | errbuf[errbuf_size - 1] = 0; | |
5377 | } | |
5378 | else | |
5379 | strcpy (errbuf, msg); | |
5380 | } | |
5381 | ||
5382 | return msg_size; | |
5383 | } | |
5384 | ||
5385 | ||
5386 | /* Free dynamically allocated space used by PREG. */ | |
5387 | ||
5388 | void | |
5389 | regfree (preg) | |
5390 | regex_t *preg; | |
5391 | { | |
5392 | if (preg->buffer != NULL) | |
5393 | free (preg->buffer); | |
5394 | preg->buffer = NULL; | |
5395 | ||
5396 | preg->allocated = 0; | |
5397 | preg->used = 0; | |
5398 | ||
5399 | if (preg->fastmap != NULL) | |
5400 | free (preg->fastmap); | |
5401 | preg->fastmap = NULL; | |
5402 | preg->fastmap_accurate = 0; | |
5403 | ||
5404 | if (preg->translate != NULL) | |
5405 | free (preg->translate); | |
5406 | preg->translate = NULL; | |
5407 | } | |
5408 | ||
5409 | #endif /* not emacs */ | |
5410 | \f | |
5411 | /* | |
5412 | Local variables: | |
5413 | make-backup-files: t | |
5414 | version-control: t | |
5415 | trim-versions-without-asking: nil | |
5416 | End: | |
5417 | */ |