Mercurial > emacs
annotate src/regex.c @ 5709:85119c3098a0
Fix rmail summary commands to work with rmail-mail-new-frame.
author | Karl Heuer <kwzh@gnu.org> |
---|---|
date | Mon, 31 Jan 1994 07:05:45 +0000 |
parents | b724b2f0fbc7 |
children | 2f86788715e5 |
rev | line source |
---|---|
1155 | 1 /* Extended regular expression matching and search library, |
2454 | 2 version 0.12. |
1155 | 3 (Implements POSIX draft P10003.2/D11.2, except for |
4 internationalization features.) | |
5 | |
1738 | 6 Copyright (C) 1993 Free Software Foundation, Inc. |
1155 | 7 |
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 | |
4846 | 29 #ifdef HAVE_CONFIG_H |
30 #if defined (CONFIG_BROKETS) | |
31 /* We use <config.h> instead of "config.h" so that a compilation | |
32 using -I. -I$srcdir will use ./config.h rather than $srcdir/config.h | |
33 (which it would do because it found this file in $srcdir). */ | |
34 #include <config.h> | |
35 #else | |
36 #include "config.h" | |
37 #endif | |
38 #endif | |
39 | |
1155 | 40 /* We need this for `regex.h', and perhaps for the Emacs include files. */ |
41 #include <sys/types.h> | |
42 | |
1669 | 43 #ifdef HAVE_CONFIG_H |
1645
fb092d69da76
*** empty log message ***
David J. MacKenzie <djm@gnu.org>
parents:
1642
diff
changeset
|
44 #include "config.h" |
fb092d69da76
*** empty log message ***
David J. MacKenzie <djm@gnu.org>
parents:
1642
diff
changeset
|
45 #endif |
fb092d69da76
*** empty log message ***
David J. MacKenzie <djm@gnu.org>
parents:
1642
diff
changeset
|
46 |
1155 | 47 /* The `emacs' switch turns on certain matching commands |
48 that make sense only in Emacs. */ | |
49 #ifdef emacs | |
50 | |
51 #include "lisp.h" | |
52 #include "buffer.h" | |
53 #include "syntax.h" | |
54 | |
55 /* Emacs uses `NULL' as a predicate. */ | |
56 #undef NULL | |
57 | |
58 #else /* not emacs */ | |
59 | |
3766 | 60 #ifdef STDC_HEADERS |
61 #include <stdlib.h> | |
62 #else | |
63 char *malloc (); | |
64 char *realloc (); | |
65 #endif | |
66 | |
67 | |
1155 | 68 /* We used to test for `BSTRING' here, but only GCC and Emacs define |
69 `BSTRING', as far as I know, and neither of them use this code. */ | |
1641
47ae0840b2b9
*** empty log message ***
David J. MacKenzie <djm@gnu.org>
parents:
1637
diff
changeset
|
70 #if HAVE_STRING_H || STDC_HEADERS |
1155 | 71 #include <string.h> |
1637 | 72 #ifndef bcmp |
1155 | 73 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n)) |
1637 | 74 #endif |
75 #ifndef bcopy | |
1155 | 76 #define bcopy(s, d, n) memcpy ((d), (s), (n)) |
1637 | 77 #endif |
78 #ifndef bzero | |
1155 | 79 #define bzero(s, n) memset ((s), 0, (n)) |
1637 | 80 #endif |
1155 | 81 #else |
82 #include <strings.h> | |
83 #endif | |
84 | |
85 /* Define the syntax stuff for \<, \>, etc. */ | |
86 | |
87 /* This must be nonzero for the wordchar and notwordchar pattern | |
88 commands in re_match_2. */ | |
89 #ifndef Sword | |
90 #define Sword 1 | |
91 #endif | |
92 | |
93 #ifdef SYNTAX_TABLE | |
94 | |
95 extern char *re_syntax_table; | |
96 | |
97 #else /* not SYNTAX_TABLE */ | |
98 | |
99 /* How many characters in the character set. */ | |
100 #define CHAR_SET_SIZE 256 | |
101 | |
102 static char re_syntax_table[CHAR_SET_SIZE]; | |
103 | |
104 static void | |
105 init_syntax_once () | |
106 { | |
107 register int c; | |
108 static int done = 0; | |
109 | |
110 if (done) | |
111 return; | |
112 | |
113 bzero (re_syntax_table, sizeof re_syntax_table); | |
114 | |
115 for (c = 'a'; c <= 'z'; c++) | |
116 re_syntax_table[c] = Sword; | |
117 | |
118 for (c = 'A'; c <= 'Z'; c++) | |
119 re_syntax_table[c] = Sword; | |
120 | |
121 for (c = '0'; c <= '9'; c++) | |
122 re_syntax_table[c] = Sword; | |
123 | |
124 re_syntax_table['_'] = Sword; | |
125 | |
126 done = 1; | |
127 } | |
128 | |
129 #endif /* not SYNTAX_TABLE */ | |
130 | |
131 #define SYNTAX(c) re_syntax_table[c] | |
132 | |
133 #endif /* not emacs */ | |
134 | |
135 /* Get the interface, including the syntax bits. */ | |
136 #include "regex.h" | |
137 | |
138 /* isalpha etc. are used for the character classes. */ | |
139 #include <ctype.h> | |
1668 | 140 |
2465 | 141 /* Jim Meyering writes: |
142 | |
143 "... Some ctype macros are valid only for character codes that | |
144 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when | |
145 using /bin/cc or gcc but without giving an ansi option). So, all | |
146 ctype uses should be through macros like ISPRINT... If | |
147 STDC_HEADERS is defined, then autoconf has verified that the ctype | |
148 macros don't need to be guarded with references to isascii. ... | |
149 Defining isascii to 1 should let any compiler worth its salt | |
150 eliminate the && through constant folding." */ | |
5076 | 151 |
152 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII)) | |
153 #define ISASCII(c) 1 | |
154 #else | |
155 #define ISASCII(c) isascii(c) | |
1668 | 156 #endif |
157 | |
158 #ifdef isblank | |
5076 | 159 #define ISBLANK(c) (ISASCII (c) && isblank (c)) |
1668 | 160 #else |
161 #define ISBLANK(c) ((c) == ' ' || (c) == '\t') | |
1155 | 162 #endif |
1668 | 163 #ifdef isgraph |
5076 | 164 #define ISGRAPH(c) (ISASCII (c) && isgraph (c)) |
1668 | 165 #else |
5076 | 166 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) |
1155 | 167 #endif |
168 | |
5076 | 169 #define ISPRINT(c) (ISASCII (c) && isprint (c)) |
170 #define ISDIGIT(c) (ISASCII (c) && isdigit (c)) | |
171 #define ISALNUM(c) (ISASCII (c) && isalnum (c)) | |
172 #define ISALPHA(c) (ISASCII (c) && isalpha (c)) | |
173 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) | |
174 #define ISLOWER(c) (ISASCII (c) && islower (c)) | |
175 #define ISPUNCT(c) (ISASCII (c) && ispunct (c)) | |
176 #define ISSPACE(c) (ISASCII (c) && isspace (c)) | |
177 #define ISUPPER(c) (ISASCII (c) && isupper (c)) | |
178 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) | |
1668 | 179 |
1155 | 180 #ifndef NULL |
181 #define NULL 0 | |
182 #endif | |
183 | |
184 /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
185 since ours (we hope) works properly with all combinations of | |
186 machines, compilers, `char' and `unsigned char' argument types. | |
187 (Per Bothner suggested the basic approach.) */ | |
188 #undef SIGN_EXTEND_CHAR | |
189 #if __STDC__ | |
190 #define SIGN_EXTEND_CHAR(c) ((signed char) (c)) | |
1637 | 191 #else /* not __STDC__ */ |
1155 | 192 /* As in Harbison and Steele. */ |
193 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) | |
194 #endif | |
195 | |
196 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we | |
197 use `alloca' instead of `malloc'. This is because using malloc in | |
198 re_search* or re_match* could cause memory leaks when C-g is used in | |
199 Emacs; also, malloc is slower and causes storage fragmentation. On | |
200 the other hand, malloc is more portable, and easier to debug. | |
201 | |
202 Because we sometimes use alloca, some routines have to be macros, | |
203 not functions -- `alloca'-allocated space disappears at the end of the | |
204 function it is called in. */ | |
205 | |
206 #ifdef REGEX_MALLOC | |
207 | |
208 #define REGEX_ALLOCATE malloc | |
209 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) | |
210 | |
211 #else /* not REGEX_MALLOC */ | |
212 | |
213 /* Emacs already defines alloca, sometimes. */ | |
214 #ifndef alloca | |
215 | |
216 /* Make alloca work the best possible way. */ | |
217 #ifdef __GNUC__ | |
218 #define alloca __builtin_alloca | |
219 #else /* not __GNUC__ */ | |
220 #if HAVE_ALLOCA_H | |
221 #include <alloca.h> | |
222 #else /* not __GNUC__ or HAVE_ALLOCA_H */ | |
223 #ifndef _AIX /* Already did AIX, up at the top. */ | |
224 char *alloca (); | |
225 #endif /* not _AIX */ | |
226 #endif /* not HAVE_ALLOCA_H */ | |
227 #endif /* not __GNUC__ */ | |
228 | |
229 #endif /* not alloca */ | |
230 | |
231 #define REGEX_ALLOCATE alloca | |
232 | |
233 /* Assumes a `char *destination' variable. */ | |
234 #define REGEX_REALLOCATE(source, osize, nsize) \ | |
235 (destination = (char *) alloca (nsize), \ | |
236 bcopy (source, destination, osize), \ | |
237 destination) | |
238 | |
239 #endif /* not REGEX_MALLOC */ | |
240 | |
241 | |
242 /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
243 `string1' or just past its end. This works if PTR is NULL, which is | |
244 a good thing. */ | |
245 #define FIRST_STRING_P(ptr) \ | |
246 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) | |
247 | |
248 /* (Re)Allocate N items of type T using malloc, or fail. */ | |
249 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) | |
250 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) | |
2949 | 251 #define RETALLOC_IF(addr, n, t) \ |
252 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) | |
1155 | 253 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) |
254 | |
255 #define BYTEWIDTH 8 /* In bits. */ | |
256 | |
257 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) | |
258 | |
259 #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
260 #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
261 | |
262 typedef char boolean; | |
263 #define false 0 | |
264 #define true 1 | |
265 | |
266 /* These are the command codes that appear in compiled regular | |
267 expressions. Some opcodes are followed by argument bytes. A | |
268 command code can specify any interpretation whatsoever for its | |
269 arguments. Zero bytes may appear in the compiled regular expression. | |
270 | |
271 The value of `exactn' is needed in search.c (search_buffer) in Emacs. | |
272 So regex.h defines a symbol `RE_EXACTN_VALUE' to be 1; the value of | |
273 `exactn' we use here must also be 1. */ | |
274 | |
275 typedef enum | |
276 { | |
277 no_op = 0, | |
278 | |
279 /* Followed by one byte giving n, then by n literal bytes. */ | |
280 exactn = 1, | |
281 | |
282 /* Matches any (more or less) character. */ | |
283 anychar, | |
284 | |
285 /* Matches any one char belonging to specified set. First | |
286 following byte is number of bitmap bytes. Then come bytes | |
287 for a bitmap saying which chars are in. Bits in each byte | |
288 are ordered low-bit-first. A character is in the set if its | |
289 bit is 1. A character too large to have a bit in the map is | |
290 automatically not in the set. */ | |
291 charset, | |
292 | |
293 /* Same parameters as charset, but match any character that is | |
294 not one of those specified. */ | |
295 charset_not, | |
296 | |
297 /* Start remembering the text that is matched, for storing in a | |
298 register. Followed by one byte with the register number, in | |
299 the range 0 to one less than the pattern buffer's re_nsub | |
300 field. Then followed by one byte with the number of groups | |
301 inner to this one. (This last has to be part of the | |
302 start_memory only because we need it in the on_failure_jump | |
303 of re_match_2.) */ | |
304 start_memory, | |
305 | |
306 /* Stop remembering the text that is matched and store it in a | |
307 memory register. Followed by one byte with the register | |
308 number, in the range 0 to one less than `re_nsub' in the | |
309 pattern buffer, and one byte with the number of inner groups, | |
310 just like `start_memory'. (We need the number of inner | |
311 groups here because we don't have any easy way of finding the | |
312 corresponding start_memory when we're at a stop_memory.) */ | |
313 stop_memory, | |
314 | |
315 /* Match a duplicate of something remembered. Followed by one | |
316 byte containing the register number. */ | |
317 duplicate, | |
318 | |
319 /* Fail unless at beginning of line. */ | |
320 begline, | |
321 | |
322 /* Fail unless at end of line. */ | |
323 endline, | |
324 | |
325 /* Succeeds if at beginning of buffer (if emacs) or at beginning | |
326 of string to be matched (if not). */ | |
327 begbuf, | |
328 | |
329 /* Analogously, for end of buffer/string. */ | |
330 endbuf, | |
331 | |
332 /* Followed by two byte relative address to which to jump. */ | |
333 jump, | |
334 | |
335 /* Same as jump, but marks the end of an alternative. */ | |
336 jump_past_alt, | |
337 | |
338 /* Followed by two-byte relative address of place to resume at | |
339 in case of failure. */ | |
340 on_failure_jump, | |
341 | |
342 /* Like on_failure_jump, but pushes a placeholder instead of the | |
343 current string position when executed. */ | |
344 on_failure_keep_string_jump, | |
345 | |
346 /* Throw away latest failure point and then jump to following | |
347 two-byte relative address. */ | |
348 pop_failure_jump, | |
349 | |
350 /* Change to pop_failure_jump if know won't have to backtrack to | |
351 match; otherwise change to jump. This is used to jump | |
352 back to the beginning of a repeat. If what follows this jump | |
353 clearly won't match what the repeat does, such that we can be | |
354 sure that there is no use backtracking out of repetitions | |
355 already matched, then we change it to a pop_failure_jump. | |
356 Followed by two-byte address. */ | |
357 maybe_pop_jump, | |
358 | |
359 /* Jump to following two-byte address, and push a dummy failure | |
360 point. This failure point will be thrown away if an attempt | |
361 is made to use it for a failure. A `+' construct makes this | |
362 before the first repeat. Also used as an intermediary kind | |
363 of jump when compiling an alternative. */ | |
364 dummy_failure_jump, | |
365 | |
366 /* Push a dummy failure point and continue. Used at the end of | |
367 alternatives. */ | |
368 push_dummy_failure, | |
369 | |
370 /* Followed by two-byte relative address and two-byte number n. | |
371 After matching N times, jump to the address upon failure. */ | |
372 succeed_n, | |
373 | |
374 /* Followed by two-byte relative address, and two-byte number n. | |
375 Jump to the address N times, then fail. */ | |
376 jump_n, | |
377 | |
378 /* Set the following two-byte relative address to the | |
379 subsequent two-byte number. The address *includes* the two | |
380 bytes of number. */ | |
381 set_number_at, | |
382 | |
383 wordchar, /* Matches any word-constituent character. */ | |
384 notwordchar, /* Matches any char that is not a word-constituent. */ | |
385 | |
386 wordbeg, /* Succeeds if at word beginning. */ | |
387 wordend, /* Succeeds if at word end. */ | |
388 | |
389 wordbound, /* Succeeds if at a word boundary. */ | |
390 notwordbound /* Succeeds if not at a word boundary. */ | |
391 | |
392 #ifdef emacs | |
393 ,before_dot, /* Succeeds if before point. */ | |
394 at_dot, /* Succeeds if at point. */ | |
395 after_dot, /* Succeeds if after point. */ | |
396 | |
397 /* Matches any character whose syntax is specified. Followed by | |
398 a byte which contains a syntax code, e.g., Sword. */ | |
399 syntaxspec, | |
400 | |
401 /* Matches any character whose syntax is not that specified. */ | |
402 notsyntaxspec | |
403 #endif /* emacs */ | |
404 } re_opcode_t; | |
405 | |
406 /* Common operations on the compiled pattern. */ | |
407 | |
408 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
409 | |
410 #define STORE_NUMBER(destination, number) \ | |
411 do { \ | |
412 (destination)[0] = (number) & 0377; \ | |
413 (destination)[1] = (number) >> 8; \ | |
414 } while (0) | |
415 | |
416 /* Same as STORE_NUMBER, except increment DESTINATION to | |
417 the byte after where the number is stored. Therefore, DESTINATION | |
418 must be an lvalue. */ | |
419 | |
420 #define STORE_NUMBER_AND_INCR(destination, number) \ | |
421 do { \ | |
422 STORE_NUMBER (destination, number); \ | |
423 (destination) += 2; \ | |
424 } while (0) | |
425 | |
426 /* Put into DESTINATION a number stored in two contiguous bytes starting | |
427 at SOURCE. */ | |
428 | |
429 #define EXTRACT_NUMBER(destination, source) \ | |
430 do { \ | |
431 (destination) = *(source) & 0377; \ | |
432 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
433 } while (0) | |
434 | |
435 #ifdef DEBUG | |
436 static void | |
437 extract_number (dest, source) | |
438 int *dest; | |
439 unsigned char *source; | |
440 { | |
441 int temp = SIGN_EXTEND_CHAR (*(source + 1)); | |
442 *dest = *source & 0377; | |
443 *dest += temp << 8; | |
444 } | |
445 | |
446 #ifndef EXTRACT_MACROS /* To debug the macros. */ | |
447 #undef EXTRACT_NUMBER | |
448 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
449 #endif /* not EXTRACT_MACROS */ | |
450 | |
451 #endif /* DEBUG */ | |
452 | |
453 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
454 SOURCE must be an lvalue. */ | |
455 | |
456 #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
457 do { \ | |
458 EXTRACT_NUMBER (destination, source); \ | |
459 (source) += 2; \ | |
460 } while (0) | |
461 | |
462 #ifdef DEBUG | |
463 static void | |
464 extract_number_and_incr (destination, source) | |
465 int *destination; | |
466 unsigned char **source; | |
467 { | |
468 extract_number (destination, *source); | |
469 *source += 2; | |
470 } | |
471 | |
472 #ifndef EXTRACT_MACROS | |
473 #undef EXTRACT_NUMBER_AND_INCR | |
474 #define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
475 extract_number_and_incr (&dest, &src) | |
476 #endif /* not EXTRACT_MACROS */ | |
477 | |
478 #endif /* DEBUG */ | |
479 | |
480 /* If DEBUG is defined, Regex prints many voluminous messages about what | |
481 it is doing (if the variable `debug' is nonzero). If linked with the | |
482 main program in `iregex.c', you can enter patterns and strings | |
483 interactively. And if linked with the main program in `main.c' and | |
484 the other test files, you can run the already-written tests. */ | |
485 | |
486 #ifdef DEBUG | |
487 | |
488 /* We use standard I/O for debugging. */ | |
489 #include <stdio.h> | |
490 | |
491 /* It is useful to test things that ``must'' be true when debugging. */ | |
492 #include <assert.h> | |
493 | |
494 static int debug = 0; | |
495 | |
496 #define DEBUG_STATEMENT(e) e | |
497 #define DEBUG_PRINT1(x) if (debug) printf (x) | |
498 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) | |
499 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) | |
1637 | 500 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) |
1155 | 501 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ |
502 if (debug) print_partial_compiled_pattern (s, e) | |
503 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ | |
504 if (debug) print_double_string (w, s1, sz1, s2, sz2) | |
505 | |
506 | |
507 extern void printchar (); | |
508 | |
509 /* Print the fastmap in human-readable form. */ | |
510 | |
511 void | |
512 print_fastmap (fastmap) | |
513 char *fastmap; | |
514 { | |
515 unsigned was_a_range = 0; | |
516 unsigned i = 0; | |
517 | |
518 while (i < (1 << BYTEWIDTH)) | |
519 { | |
520 if (fastmap[i++]) | |
521 { | |
522 was_a_range = 0; | |
523 printchar (i - 1); | |
524 while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
525 { | |
526 was_a_range = 1; | |
527 i++; | |
528 } | |
529 if (was_a_range) | |
530 { | |
531 printf ("-"); | |
532 printchar (i - 1); | |
533 } | |
534 } | |
535 } | |
536 putchar ('\n'); | |
537 } | |
538 | |
539 | |
540 /* Print a compiled pattern string in human-readable form, starting at | |
541 the START pointer into it and ending just before the pointer END. */ | |
542 | |
543 void | |
544 print_partial_compiled_pattern (start, end) | |
545 unsigned char *start; | |
546 unsigned char *end; | |
547 { | |
548 int mcnt, mcnt2; | |
549 unsigned char *p = start; | |
550 unsigned char *pend = end; | |
551 | |
552 if (start == NULL) | |
553 { | |
554 printf ("(null)\n"); | |
555 return; | |
556 } | |
557 | |
558 /* Loop over pattern commands. */ | |
559 while (p < pend) | |
560 { | |
2615 | 561 printf ("%d:\t", p - start); |
562 | |
1155 | 563 switch ((re_opcode_t) *p++) |
564 { | |
565 case no_op: | |
566 printf ("/no_op"); | |
567 break; | |
568 | |
569 case exactn: | |
570 mcnt = *p++; | |
571 printf ("/exactn/%d", mcnt); | |
572 do | |
573 { | |
574 putchar ('/'); | |
575 printchar (*p++); | |
576 } | |
577 while (--mcnt); | |
578 break; | |
579 | |
580 case start_memory: | |
581 mcnt = *p++; | |
582 printf ("/start_memory/%d/%d", mcnt, *p++); | |
583 break; | |
584 | |
585 case stop_memory: | |
586 mcnt = *p++; | |
587 printf ("/stop_memory/%d/%d", mcnt, *p++); | |
588 break; | |
589 | |
590 case duplicate: | |
591 printf ("/duplicate/%d", *p++); | |
592 break; | |
593 | |
594 case anychar: | |
595 printf ("/anychar"); | |
596 break; | |
597 | |
598 case charset: | |
599 case charset_not: | |
600 { | |
2615 | 601 register int c, last = -100; |
602 register int in_range = 0; | |
603 | |
604 printf ("/charset [%s", | |
605 (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); | |
1155 | 606 |
607 assert (p + *p < pend); | |
608 | |
2615 | 609 for (c = 0; c < 256; c++) |
610 if (c / 8 < *p | |
611 && (p[1 + (c/8)] & (1 << (c % 8)))) | |
612 { | |
613 /* Are we starting a range? */ | |
614 if (last + 1 == c && ! in_range) | |
615 { | |
616 putchar ('-'); | |
617 in_range = 1; | |
618 } | |
619 /* Have we broken a range? */ | |
620 else if (last + 1 != c && in_range) | |
1155 | 621 { |
2615 | 622 printchar (last); |
623 in_range = 0; | |
624 } | |
1155 | 625 |
2615 | 626 if (! in_range) |
627 printchar (c); | |
628 | |
629 last = c; | |
1155 | 630 } |
2615 | 631 |
632 if (in_range) | |
633 printchar (last); | |
634 | |
635 putchar (']'); | |
636 | |
1155 | 637 p += 1 + *p; |
638 } | |
2615 | 639 break; |
1155 | 640 |
641 case begline: | |
642 printf ("/begline"); | |
643 break; | |
644 | |
645 case endline: | |
646 printf ("/endline"); | |
647 break; | |
648 | |
649 case on_failure_jump: | |
650 extract_number_and_incr (&mcnt, &p); | |
2615 | 651 printf ("/on_failure_jump to %d", p + mcnt - start); |
1155 | 652 break; |
653 | |
654 case on_failure_keep_string_jump: | |
655 extract_number_and_incr (&mcnt, &p); | |
2615 | 656 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start); |
1155 | 657 break; |
658 | |
659 case dummy_failure_jump: | |
660 extract_number_and_incr (&mcnt, &p); | |
2615 | 661 printf ("/dummy_failure_jump to %d", p + mcnt - start); |
1155 | 662 break; |
663 | |
664 case push_dummy_failure: | |
665 printf ("/push_dummy_failure"); | |
666 break; | |
667 | |
668 case maybe_pop_jump: | |
669 extract_number_and_incr (&mcnt, &p); | |
2615 | 670 printf ("/maybe_pop_jump to %d", p + mcnt - start); |
1155 | 671 break; |
672 | |
673 case pop_failure_jump: | |
674 extract_number_and_incr (&mcnt, &p); | |
2615 | 675 printf ("/pop_failure_jump to %d", p + mcnt - start); |
1155 | 676 break; |
677 | |
678 case jump_past_alt: | |
679 extract_number_and_incr (&mcnt, &p); | |
2615 | 680 printf ("/jump_past_alt to %d", p + mcnt - start); |
1155 | 681 break; |
682 | |
683 case jump: | |
684 extract_number_and_incr (&mcnt, &p); | |
2615 | 685 printf ("/jump to %d", p + mcnt - start); |
1155 | 686 break; |
687 | |
688 case succeed_n: | |
689 extract_number_and_incr (&mcnt, &p); | |
690 extract_number_and_incr (&mcnt2, &p); | |
2615 | 691 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2); |
1155 | 692 break; |
693 | |
694 case jump_n: | |
695 extract_number_and_incr (&mcnt, &p); | |
696 extract_number_and_incr (&mcnt2, &p); | |
2615 | 697 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2); |
1155 | 698 break; |
699 | |
700 case set_number_at: | |
701 extract_number_and_incr (&mcnt, &p); | |
702 extract_number_and_incr (&mcnt2, &p); | |
2615 | 703 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2); |
1155 | 704 break; |
705 | |
706 case wordbound: | |
707 printf ("/wordbound"); | |
708 break; | |
709 | |
710 case notwordbound: | |
711 printf ("/notwordbound"); | |
712 break; | |
713 | |
714 case wordbeg: | |
715 printf ("/wordbeg"); | |
716 break; | |
717 | |
718 case wordend: | |
719 printf ("/wordend"); | |
720 | |
721 #ifdef emacs | |
722 case before_dot: | |
723 printf ("/before_dot"); | |
724 break; | |
725 | |
726 case at_dot: | |
727 printf ("/at_dot"); | |
728 break; | |
729 | |
730 case after_dot: | |
731 printf ("/after_dot"); | |
732 break; | |
733 | |
734 case syntaxspec: | |
735 printf ("/syntaxspec"); | |
736 mcnt = *p++; | |
737 printf ("/%d", mcnt); | |
738 break; | |
739 | |
740 case notsyntaxspec: | |
741 printf ("/notsyntaxspec"); | |
742 mcnt = *p++; | |
743 printf ("/%d", mcnt); | |
744 break; | |
745 #endif /* emacs */ | |
746 | |
747 case wordchar: | |
748 printf ("/wordchar"); | |
749 break; | |
750 | |
751 case notwordchar: | |
752 printf ("/notwordchar"); | |
753 break; | |
754 | |
755 case begbuf: | |
756 printf ("/begbuf"); | |
757 break; | |
758 | |
759 case endbuf: | |
760 printf ("/endbuf"); | |
761 break; | |
762 | |
763 default: | |
764 printf ("?%d", *(p-1)); | |
765 } | |
2615 | 766 |
767 putchar ('\n'); | |
1155 | 768 } |
2615 | 769 |
770 printf ("%d:\tend of pattern.\n", p - start); | |
1155 | 771 } |
772 | |
773 | |
774 void | |
775 print_compiled_pattern (bufp) | |
776 struct re_pattern_buffer *bufp; | |
777 { | |
778 unsigned char *buffer = bufp->buffer; | |
779 | |
780 print_partial_compiled_pattern (buffer, buffer + bufp->used); | |
781 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated); | |
782 | |
783 if (bufp->fastmap_accurate && bufp->fastmap) | |
784 { | |
785 printf ("fastmap: "); | |
786 print_fastmap (bufp->fastmap); | |
787 } | |
788 | |
789 printf ("re_nsub: %d\t", bufp->re_nsub); | |
790 printf ("regs_alloc: %d\t", bufp->regs_allocated); | |
791 printf ("can_be_null: %d\t", bufp->can_be_null); | |
792 printf ("newline_anchor: %d\n", bufp->newline_anchor); | |
793 printf ("no_sub: %d\t", bufp->no_sub); | |
794 printf ("not_bol: %d\t", bufp->not_bol); | |
795 printf ("not_eol: %d\t", bufp->not_eol); | |
796 printf ("syntax: %d\n", bufp->syntax); | |
797 /* Perhaps we should print the translate table? */ | |
798 } | |
799 | |
800 | |
801 void | |
802 print_double_string (where, string1, size1, string2, size2) | |
803 const char *where; | |
804 const char *string1; | |
805 const char *string2; | |
806 int size1; | |
807 int size2; | |
808 { | |
809 unsigned this_char; | |
810 | |
811 if (where == NULL) | |
812 printf ("(null)"); | |
813 else | |
814 { | |
815 if (FIRST_STRING_P (where)) | |
816 { | |
817 for (this_char = where - string1; this_char < size1; this_char++) | |
818 printchar (string1[this_char]); | |
819 | |
820 where = string2; | |
821 } | |
822 | |
823 for (this_char = where - string2; this_char < size2; this_char++) | |
824 printchar (string2[this_char]); | |
825 } | |
826 } | |
827 | |
828 #else /* not DEBUG */ | |
829 | |
830 #undef assert | |
831 #define assert(e) | |
832 | |
833 #define DEBUG_STATEMENT(e) | |
834 #define DEBUG_PRINT1(x) | |
835 #define DEBUG_PRINT2(x1, x2) | |
836 #define DEBUG_PRINT3(x1, x2, x3) | |
1637 | 837 #define DEBUG_PRINT4(x1, x2, x3, x4) |
1155 | 838 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) |
839 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
840 | |
841 #endif /* not DEBUG */ | |
842 | |
843 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
844 also be assigned to arbitrarily: each pattern buffer stores its own | |
845 syntax, so it can be changed between regex compilations. */ | |
846 reg_syntax_t re_syntax_options = RE_SYNTAX_EMACS; | |
847 | |
848 | |
849 /* Specify the precise syntax of regexps for compilation. This provides | |
850 for compatibility for various utilities which historically have | |
851 different, incompatible syntaxes. | |
852 | |
853 The argument SYNTAX is a bit mask comprised of the various bits | |
854 defined in regex.h. We return the old syntax. */ | |
855 | |
856 reg_syntax_t | |
857 re_set_syntax (syntax) | |
858 reg_syntax_t syntax; | |
859 { | |
860 reg_syntax_t ret = re_syntax_options; | |
861 | |
862 re_syntax_options = syntax; | |
863 return ret; | |
864 } | |
865 | |
866 /* This table gives an error message for each of the error codes listed | |
867 in regex.h. Obviously the order here has to be same as there. */ | |
868 | |
869 static const char *re_error_msg[] = | |
870 { NULL, /* REG_NOERROR */ | |
871 "No match", /* REG_NOMATCH */ | |
872 "Invalid regular expression", /* REG_BADPAT */ | |
873 "Invalid collation character", /* REG_ECOLLATE */ | |
874 "Invalid character class name", /* REG_ECTYPE */ | |
875 "Trailing backslash", /* REG_EESCAPE */ | |
876 "Invalid back reference", /* REG_ESUBREG */ | |
877 "Unmatched [ or [^", /* REG_EBRACK */ | |
878 "Unmatched ( or \\(", /* REG_EPAREN */ | |
879 "Unmatched \\{", /* REG_EBRACE */ | |
880 "Invalid content of \\{\\}", /* REG_BADBR */ | |
881 "Invalid range end", /* REG_ERANGE */ | |
882 "Memory exhausted", /* REG_ESPACE */ | |
883 "Invalid preceding regular expression", /* REG_BADRPT */ | |
884 "Premature end of regular expression", /* REG_EEND */ | |
885 "Regular expression too big", /* REG_ESIZE */ | |
886 "Unmatched ) or \\)", /* REG_ERPAREN */ | |
887 }; | |
888 | |
2949 | 889 /* Avoiding alloca during matching, to placate r_alloc. */ |
890 | |
2952 | 891 /* Define MATCH_MAY_ALLOCATE if we need to make sure that the |
2949 | 892 searching and matching functions should not call alloca. On some |
893 systems, alloca is implemented in terms of malloc, and if we're | |
894 using the relocating allocator routines, then malloc could cause a | |
895 relocation, which might (if the strings being searched are in the | |
896 ralloc heap) shift the data out from underneath the regexp | |
3614 | 897 routines. |
898 | |
899 Here's another reason to avoid allocation: Emacs insists on | |
900 processing input from X in a signal handler; processing X input may | |
901 call malloc; if input arrives while a matching routine is calling | |
902 malloc, then we're scrod. But Emacs can't just block input while | |
903 calling matching routines; then we don't notice interrupts when | |
904 they come in. So, Emacs blocks input around all regexp calls | |
905 except the matching calls, which it leaves unprotected, in the | |
906 faith that they will not malloc. */ | |
2952 | 907 |
908 /* Normally, this is fine. */ | |
909 #define MATCH_MAY_ALLOCATE | |
910 | |
911 /* But under some circumstances, it's not. */ | |
3614 | 912 #if defined (emacs) || (defined (REL_ALLOC) && defined (C_ALLOCA)) |
2952 | 913 #undef MATCH_MAY_ALLOCATE |
2949 | 914 #endif |
915 | |
916 | |
917 /* Failure stack declarations and macros; both re_compile_fastmap and | |
918 re_match_2 use a failure stack. These have to be macros because of | |
919 REGEX_ALLOCATE. */ | |
920 | |
921 | |
922 /* Number of failure points for which to initially allocate space | |
923 when matching. If this number is exceeded, we allocate more | |
924 space, so it is not a hard limit. */ | |
925 #ifndef INIT_FAILURE_ALLOC | |
926 #define INIT_FAILURE_ALLOC 5 | |
927 #endif | |
928 | |
929 /* Roughly the maximum number of failure points on the stack. Would be | |
930 exactly that if always used MAX_FAILURE_SPACE each time we failed. | |
931 This is a variable only so users of regex can assign to it; we never | |
932 change it ourselves. */ | |
933 int re_max_failures = 2000; | |
934 | |
4918
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
935 typedef unsigned char *fail_stack_elt_t; |
2949 | 936 |
937 typedef struct | |
938 { | |
939 fail_stack_elt_t *stack; | |
940 unsigned size; | |
941 unsigned avail; /* Offset of next open position. */ | |
942 } fail_stack_type; | |
943 | |
944 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0) | |
945 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) | |
946 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) | |
947 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail]) | |
948 | |
949 | |
950 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */ | |
951 | |
2952 | 952 #ifdef MATCH_MAY_ALLOCATE |
2949 | 953 #define INIT_FAIL_STACK() \ |
954 do { \ | |
955 fail_stack.stack = (fail_stack_elt_t *) \ | |
956 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \ | |
957 \ | |
958 if (fail_stack.stack == NULL) \ | |
959 return -2; \ | |
960 \ | |
961 fail_stack.size = INIT_FAILURE_ALLOC; \ | |
962 fail_stack.avail = 0; \ | |
963 } while (0) | |
964 #else | |
965 #define INIT_FAIL_STACK() \ | |
966 do { \ | |
967 fail_stack.avail = 0; \ | |
968 } while (0) | |
969 #endif | |
970 | |
971 | |
972 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. | |
973 | |
974 Return 1 if succeeds, and 0 if either ran out of memory | |
975 allocating space for it or it was already too large. | |
976 | |
977 REGEX_REALLOCATE requires `destination' be declared. */ | |
978 | |
979 #define DOUBLE_FAIL_STACK(fail_stack) \ | |
980 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \ | |
981 ? 0 \ | |
982 : ((fail_stack).stack = (fail_stack_elt_t *) \ | |
983 REGEX_REALLOCATE ((fail_stack).stack, \ | |
984 (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
985 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \ | |
986 \ | |
987 (fail_stack).stack == NULL \ | |
988 ? 0 \ | |
989 : ((fail_stack).size <<= 1, \ | |
990 1))) | |
991 | |
992 | |
993 /* Push PATTERN_OP on FAIL_STACK. | |
994 | |
995 Return 1 if was able to do so and 0 if ran out of memory allocating | |
996 space to do so. */ | |
997 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \ | |
998 ((FAIL_STACK_FULL () \ | |
999 && !DOUBLE_FAIL_STACK (fail_stack)) \ | |
1000 ? 0 \ | |
1001 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \ | |
1002 1)) | |
1003 | |
1004 /* This pushes an item onto the failure stack. Must be a four-byte | |
1005 value. Assumes the variable `fail_stack'. Probably should only | |
1006 be called from within `PUSH_FAILURE_POINT'. */ | |
1007 #define PUSH_FAILURE_ITEM(item) \ | |
1008 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item | |
1009 | |
1010 /* The complement operation. Assumes `fail_stack' is nonempty. */ | |
1011 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail] | |
1012 | |
1013 /* Used to omit pushing failure point id's when we're not debugging. */ | |
1014 #ifdef DEBUG | |
1015 #define DEBUG_PUSH PUSH_FAILURE_ITEM | |
1016 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM () | |
1017 #else | |
1018 #define DEBUG_PUSH(item) | |
1019 #define DEBUG_POP(item_addr) | |
1020 #endif | |
1021 | |
1022 | |
1023 /* Push the information about the state we will need | |
1024 if we ever fail back to it. | |
1025 | |
1026 Requires variables fail_stack, regstart, regend, reg_info, and | |
1027 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be | |
1028 declared. | |
1029 | |
1030 Does `return FAILURE_CODE' if runs out of memory. */ | |
1031 | |
1032 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ | |
1033 do { \ | |
1034 char *destination; \ | |
1035 /* Must be int, so when we don't save any registers, the arithmetic \ | |
1036 of 0 + -1 isn't done as unsigned. */ \ | |
1037 int this_reg; \ | |
1038 \ | |
1039 DEBUG_STATEMENT (failure_id++); \ | |
1040 DEBUG_STATEMENT (nfailure_points_pushed++); \ | |
1041 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ | |
1042 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ | |
1043 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ | |
1044 \ | |
1045 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \ | |
1046 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ | |
1047 \ | |
1048 /* Ensure we have enough space allocated for what we will push. */ \ | |
1049 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ | |
1050 { \ | |
1051 if (!DOUBLE_FAIL_STACK (fail_stack)) \ | |
1052 return failure_code; \ | |
1053 \ | |
1054 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ | |
1055 (fail_stack).size); \ | |
1056 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ | |
1057 } \ | |
1058 \ | |
1059 /* Push the info, starting with the registers. */ \ | |
1060 DEBUG_PRINT1 ("\n"); \ | |
1061 \ | |
1062 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ | |
1063 this_reg++) \ | |
1064 { \ | |
1065 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \ | |
1066 DEBUG_STATEMENT (num_regs_pushed++); \ | |
1067 \ | |
1068 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ | |
1069 PUSH_FAILURE_ITEM (regstart[this_reg]); \ | |
1070 \ | |
1071 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ | |
1072 PUSH_FAILURE_ITEM (regend[this_reg]); \ | |
1073 \ | |
1074 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \ | |
1075 DEBUG_PRINT2 (" match_null=%d", \ | |
1076 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ | |
1077 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ | |
1078 DEBUG_PRINT2 (" matched_something=%d", \ | |
1079 MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1080 DEBUG_PRINT2 (" ever_matched=%d", \ | |
1081 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1082 DEBUG_PRINT1 ("\n"); \ | |
1083 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \ | |
1084 } \ | |
1085 \ | |
1086 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\ | |
1087 PUSH_FAILURE_ITEM (lowest_active_reg); \ | |
1088 \ | |
1089 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\ | |
1090 PUSH_FAILURE_ITEM (highest_active_reg); \ | |
1091 \ | |
1092 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \ | |
1093 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ | |
1094 PUSH_FAILURE_ITEM (pattern_place); \ | |
1095 \ | |
1096 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \ | |
1097 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ | |
1098 size2); \ | |
1099 DEBUG_PRINT1 ("'\n"); \ | |
1100 PUSH_FAILURE_ITEM (string_place); \ | |
1101 \ | |
1102 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ | |
1103 DEBUG_PUSH (failure_id); \ | |
1104 } while (0) | |
1105 | |
1106 /* This is the number of items that are pushed and popped on the stack | |
1107 for each register. */ | |
1108 #define NUM_REG_ITEMS 3 | |
1109 | |
1110 /* Individual items aside from the registers. */ | |
1111 #ifdef DEBUG | |
1112 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ | |
1113 #else | |
1114 #define NUM_NONREG_ITEMS 4 | |
1115 #endif | |
1116 | |
1117 /* We push at most this many items on the stack. */ | |
1118 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS) | |
1119 | |
1120 /* We actually push this many items. */ | |
1121 #define NUM_FAILURE_ITEMS \ | |
1122 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \ | |
1123 + NUM_NONREG_ITEMS) | |
1124 | |
1125 /* How many items can still be added to the stack without overflowing it. */ | |
1126 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
1127 | |
1128 | |
1129 /* Pops what PUSH_FAIL_STACK pushes. | |
1130 | |
1131 We restore into the parameters, all of which should be lvalues: | |
1132 STR -- the saved data position. | |
1133 PAT -- the saved pattern position. | |
1134 LOW_REG, HIGH_REG -- the highest and lowest active registers. | |
1135 REGSTART, REGEND -- arrays of string positions. | |
1136 REG_INFO -- array of information about each subexpression. | |
1137 | |
1138 Also assumes the variables `fail_stack' and (if debugging), `bufp', | |
1139 `pend', `string1', `size1', `string2', and `size2'. */ | |
1140 | |
1141 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ | |
1142 { \ | |
1143 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \ | |
1144 int this_reg; \ | |
1145 const unsigned char *string_temp; \ | |
1146 \ | |
1147 assert (!FAIL_STACK_EMPTY ()); \ | |
1148 \ | |
1149 /* Remove failure points and point to how many regs pushed. */ \ | |
1150 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
1151 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
1152 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ | |
1153 \ | |
1154 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ | |
1155 \ | |
1156 DEBUG_POP (&failure_id); \ | |
1157 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ | |
1158 \ | |
1159 /* If the saved string location is NULL, it came from an \ | |
1160 on_failure_keep_string_jump opcode, and we want to throw away the \ | |
1161 saved NULL, thus retaining our current position in the string. */ \ | |
1162 string_temp = POP_FAILURE_ITEM (); \ | |
1163 if (string_temp != NULL) \ | |
1164 str = (const char *) string_temp; \ | |
1165 \ | |
1166 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \ | |
1167 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ | |
1168 DEBUG_PRINT1 ("'\n"); \ | |
1169 \ | |
1170 pat = (unsigned char *) POP_FAILURE_ITEM (); \ | |
1171 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \ | |
1172 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
1173 \ | |
1174 /* Restore register info. */ \ | |
1175 high_reg = (unsigned) POP_FAILURE_ITEM (); \ | |
1176 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \ | |
1177 \ | |
1178 low_reg = (unsigned) POP_FAILURE_ITEM (); \ | |
1179 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \ | |
1180 \ | |
1181 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ | |
1182 { \ | |
1183 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \ | |
1184 \ | |
1185 reg_info[this_reg].word = POP_FAILURE_ITEM (); \ | |
1186 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \ | |
1187 \ | |
1188 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \ | |
1189 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ | |
1190 \ | |
1191 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \ | |
1192 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ | |
1193 } \ | |
1194 \ | |
1195 DEBUG_STATEMENT (nfailure_points_popped++); \ | |
1196 } /* POP_FAILURE_POINT */ | |
1197 | |
1198 | |
1199 | |
1200 /* Structure for per-register (a.k.a. per-group) information. | |
1201 This must not be longer than one word, because we push this value | |
1202 onto the failure stack. Other register information, such as the | |
1203 starting and ending positions (which are addresses), and the list of | |
1204 inner groups (which is a bits list) are maintained in separate | |
1205 variables. | |
1206 | |
1207 We are making a (strictly speaking) nonportable assumption here: that | |
1208 the compiler will pack our bit fields into something that fits into | |
1209 the type of `word', i.e., is something that fits into one item on the | |
1210 failure stack. */ | |
1211 typedef union | |
1212 { | |
1213 fail_stack_elt_t word; | |
1214 struct | |
1215 { | |
1216 /* This field is one if this group can match the empty string, | |
1217 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ | |
1218 #define MATCH_NULL_UNSET_VALUE 3 | |
1219 unsigned match_null_string_p : 2; | |
1220 unsigned is_active : 1; | |
1221 unsigned matched_something : 1; | |
1222 unsigned ever_matched_something : 1; | |
1223 } bits; | |
1224 } register_info_type; | |
1225 | |
1226 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) | |
1227 #define IS_ACTIVE(R) ((R).bits.is_active) | |
1228 #define MATCHED_SOMETHING(R) ((R).bits.matched_something) | |
1229 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) | |
1230 | |
1231 | |
1232 /* Call this when have matched a real character; it sets `matched' flags | |
1233 for the subexpressions which we are currently inside. Also records | |
1234 that those subexprs have matched. */ | |
1235 #define SET_REGS_MATCHED() \ | |
1236 do \ | |
1237 { \ | |
1238 unsigned r; \ | |
1239 for (r = lowest_active_reg; r <= highest_active_reg; r++) \ | |
1240 { \ | |
1241 MATCHED_SOMETHING (reg_info[r]) \ | |
1242 = EVER_MATCHED_SOMETHING (reg_info[r]) \ | |
1243 = 1; \ | |
1244 } \ | |
1245 } \ | |
1246 while (0) | |
1247 | |
1248 | |
1249 /* Registers are set to a sentinel when they haven't yet matched. */ | |
1250 #define REG_UNSET_VALUE ((char *) -1) | |
1251 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE) | |
1252 | |
1253 | |
1254 | |
2952 | 1255 /* How do we implement a missing MATCH_MAY_ALLOCATE? |
2949 | 1256 We make the fail stack a global thing, and then grow it to |
1257 re_max_failures when we compile. */ | |
2952 | 1258 #ifndef MATCH_MAY_ALLOCATE |
2949 | 1259 static fail_stack_type fail_stack; |
1260 | |
1261 static const char ** regstart, ** regend; | |
1262 static const char ** old_regstart, ** old_regend; | |
1263 static const char **best_regstart, **best_regend; | |
1264 static register_info_type *reg_info; | |
1265 static const char **reg_dummy; | |
1266 static register_info_type *reg_info_dummy; | |
1267 #endif | |
1268 | |
1269 | |
1155 | 1270 /* Subroutine declarations and macros for regex_compile. */ |
1271 | |
1272 static void store_op1 (), store_op2 (); | |
1273 static void insert_op1 (), insert_op2 (); | |
1274 static boolean at_begline_loc_p (), at_endline_loc_p (); | |
1275 static boolean group_in_compile_stack (); | |
1276 static reg_errcode_t compile_range (); | |
1277 | |
1278 /* Fetch the next character in the uncompiled pattern---translating it | |
1279 if necessary. Also cast from a signed character in the constant | |
1280 string passed to us by the user to an unsigned char that we can use | |
1281 as an array index (in, e.g., `translate'). */ | |
1282 #define PATFETCH(c) \ | |
1283 do {if (p == pend) return REG_EEND; \ | |
1284 c = (unsigned char) *p++; \ | |
1285 if (translate) c = translate[c]; \ | |
1286 } while (0) | |
1287 | |
1288 /* Fetch the next character in the uncompiled pattern, with no | |
1289 translation. */ | |
1290 #define PATFETCH_RAW(c) \ | |
1291 do {if (p == pend) return REG_EEND; \ | |
1292 c = (unsigned char) *p++; \ | |
1293 } while (0) | |
1294 | |
1295 /* Go backwards one character in the pattern. */ | |
1296 #define PATUNFETCH p-- | |
1297 | |
1298 | |
1299 /* If `translate' is non-null, return translate[D], else just D. We | |
1300 cast the subscript to translate because some data is declared as | |
1301 `char *', to avoid warnings when a string constant is passed. But | |
1302 when we use a character as a subscript we must make it unsigned. */ | |
1303 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d)) | |
1304 | |
1305 | |
1306 /* Macros for outputting the compiled pattern into `buffer'. */ | |
1307 | |
1308 /* If the buffer isn't allocated when it comes in, use this. */ | |
1309 #define INIT_BUF_SIZE 32 | |
1310 | |
1311 /* Make sure we have at least N more bytes of space in buffer. */ | |
1312 #define GET_BUFFER_SPACE(n) \ | |
1313 while (b - bufp->buffer + (n) > bufp->allocated) \ | |
1314 EXTEND_BUFFER () | |
1315 | |
1316 /* Make sure we have one more byte of buffer space and then add C to it. */ | |
1317 #define BUF_PUSH(c) \ | |
1318 do { \ | |
1319 GET_BUFFER_SPACE (1); \ | |
1320 *b++ = (unsigned char) (c); \ | |
1321 } while (0) | |
1322 | |
1323 | |
1324 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
1325 #define BUF_PUSH_2(c1, c2) \ | |
1326 do { \ | |
1327 GET_BUFFER_SPACE (2); \ | |
1328 *b++ = (unsigned char) (c1); \ | |
1329 *b++ = (unsigned char) (c2); \ | |
1330 } while (0) | |
1331 | |
1332 | |
1333 /* As with BUF_PUSH_2, except for three bytes. */ | |
1334 #define BUF_PUSH_3(c1, c2, c3) \ | |
1335 do { \ | |
1336 GET_BUFFER_SPACE (3); \ | |
1337 *b++ = (unsigned char) (c1); \ | |
1338 *b++ = (unsigned char) (c2); \ | |
1339 *b++ = (unsigned char) (c3); \ | |
1340 } while (0) | |
1341 | |
1342 | |
1343 /* Store a jump with opcode OP at LOC to location TO. We store a | |
1344 relative address offset by the three bytes the jump itself occupies. */ | |
1345 #define STORE_JUMP(op, loc, to) \ | |
1346 store_op1 (op, loc, (to) - (loc) - 3) | |
1347 | |
1348 /* Likewise, for a two-argument jump. */ | |
1349 #define STORE_JUMP2(op, loc, to, arg) \ | |
1350 store_op2 (op, loc, (to) - (loc) - 3, arg) | |
1351 | |
1352 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ | |
1353 #define INSERT_JUMP(op, loc, to) \ | |
1354 insert_op1 (op, loc, (to) - (loc) - 3, b) | |
1355 | |
1356 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
1357 #define INSERT_JUMP2(op, loc, to, arg) \ | |
1358 insert_op2 (op, loc, (to) - (loc) - 3, arg, b) | |
1359 | |
1360 | |
1361 /* This is not an arbitrary limit: the arguments which represent offsets | |
1362 into the pattern are two bytes long. So if 2^16 bytes turns out to | |
1363 be too small, many things would have to change. */ | |
1364 #define MAX_BUF_SIZE (1L << 16) | |
1365 | |
1366 | |
1367 /* Extend the buffer by twice its current size via realloc and | |
1368 reset the pointers that pointed into the old block to point to the | |
1369 correct places in the new one. If extending the buffer results in it | |
1370 being larger than MAX_BUF_SIZE, then flag memory exhausted. */ | |
1371 #define EXTEND_BUFFER() \ | |
1372 do { \ | |
1373 unsigned char *old_buffer = bufp->buffer; \ | |
1374 if (bufp->allocated == MAX_BUF_SIZE) \ | |
1375 return REG_ESIZE; \ | |
1376 bufp->allocated <<= 1; \ | |
1377 if (bufp->allocated > MAX_BUF_SIZE) \ | |
1378 bufp->allocated = MAX_BUF_SIZE; \ | |
1379 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\ | |
1380 if (bufp->buffer == NULL) \ | |
1381 return REG_ESPACE; \ | |
1382 /* If the buffer moved, move all the pointers into it. */ \ | |
1383 if (old_buffer != bufp->buffer) \ | |
1384 { \ | |
1385 b = (b - old_buffer) + bufp->buffer; \ | |
1386 begalt = (begalt - old_buffer) + bufp->buffer; \ | |
1387 if (fixup_alt_jump) \ | |
1388 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\ | |
1389 if (laststart) \ | |
1390 laststart = (laststart - old_buffer) + bufp->buffer; \ | |
1391 if (pending_exact) \ | |
1392 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \ | |
1393 } \ | |
1394 } while (0) | |
1395 | |
1396 | |
1397 /* Since we have one byte reserved for the register number argument to | |
1398 {start,stop}_memory, the maximum number of groups we can report | |
1399 things about is what fits in that byte. */ | |
1400 #define MAX_REGNUM 255 | |
1401 | |
1402 /* But patterns can have more than `MAX_REGNUM' registers. We just | |
1403 ignore the excess. */ | |
1404 typedef unsigned regnum_t; | |
1405 | |
1406 | |
1407 /* Macros for the compile stack. */ | |
1408 | |
1409 /* Since offsets can go either forwards or backwards, this type needs to | |
1410 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ | |
1411 typedef int pattern_offset_t; | |
1412 | |
1413 typedef struct | |
1414 { | |
1415 pattern_offset_t begalt_offset; | |
1416 pattern_offset_t fixup_alt_jump; | |
1417 pattern_offset_t inner_group_offset; | |
1418 pattern_offset_t laststart_offset; | |
1419 regnum_t regnum; | |
1420 } compile_stack_elt_t; | |
1421 | |
1422 | |
1423 typedef struct | |
1424 { | |
1425 compile_stack_elt_t *stack; | |
1426 unsigned size; | |
1427 unsigned avail; /* Offset of next open position. */ | |
1428 } compile_stack_type; | |
1429 | |
1430 | |
1431 #define INIT_COMPILE_STACK_SIZE 32 | |
1432 | |
1433 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
1434 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
1435 | |
1436 /* The next available element. */ | |
1437 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) | |
1438 | |
1439 | |
1440 /* Set the bit for character C in a list. */ | |
1441 #define SET_LIST_BIT(c) \ | |
1442 (b[((unsigned char) (c)) / BYTEWIDTH] \ | |
1443 |= 1 << (((unsigned char) c) % BYTEWIDTH)) | |
1444 | |
1445 | |
1446 /* Get the next unsigned number in the uncompiled pattern. */ | |
1447 #define GET_UNSIGNED_NUMBER(num) \ | |
1448 { if (p != pend) \ | |
1449 { \ | |
1450 PATFETCH (c); \ | |
1668 | 1451 while (ISDIGIT (c)) \ |
1155 | 1452 { \ |
1453 if (num < 0) \ | |
1454 num = 0; \ | |
1455 num = num * 10 + c - '0'; \ | |
1456 if (p == pend) \ | |
1457 break; \ | |
1458 PATFETCH (c); \ | |
1459 } \ | |
1460 } \ | |
1461 } | |
1462 | |
1463 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ | |
1464 | |
1465 #define IS_CHAR_CLASS(string) \ | |
1466 (STREQ (string, "alpha") || STREQ (string, "upper") \ | |
1467 || STREQ (string, "lower") || STREQ (string, "digit") \ | |
1468 || STREQ (string, "alnum") || STREQ (string, "xdigit") \ | |
1469 || STREQ (string, "space") || STREQ (string, "print") \ | |
1470 || STREQ (string, "punct") || STREQ (string, "graph") \ | |
1471 || STREQ (string, "cntrl") || STREQ (string, "blank")) | |
1472 | |
1473 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. | |
1474 Returns one of error codes defined in `regex.h', or zero for success. | |
1475 | |
1476 Assumes the `allocated' (and perhaps `buffer') and `translate' | |
1477 fields are set in BUFP on entry. | |
1478 | |
1479 If it succeeds, results are put in BUFP (if it returns an error, the | |
1480 contents of BUFP are undefined): | |
1481 `buffer' is the compiled pattern; | |
1482 `syntax' is set to SYNTAX; | |
1483 `used' is set to the length of the compiled pattern; | |
1637 | 1484 `fastmap_accurate' is zero; |
1485 `re_nsub' is the number of subexpressions in PATTERN; | |
1486 `not_bol' and `not_eol' are zero; | |
1155 | 1487 |
1488 The `fastmap' and `newline_anchor' fields are neither | |
1489 examined nor set. */ | |
1490 | |
1491 static reg_errcode_t | |
1492 regex_compile (pattern, size, syntax, bufp) | |
1493 const char *pattern; | |
1494 int size; | |
1495 reg_syntax_t syntax; | |
1496 struct re_pattern_buffer *bufp; | |
1497 { | |
1498 /* We fetch characters from PATTERN here. Even though PATTERN is | |
1499 `char *' (i.e., signed), we declare these variables as unsigned, so | |
1500 they can be reliably used as array indices. */ | |
1501 register unsigned char c, c1; | |
1502 | |
1503 /* A random tempory spot in PATTERN. */ | |
1504 const char *p1; | |
1505 | |
1506 /* Points to the end of the buffer, where we should append. */ | |
1507 register unsigned char *b; | |
1508 | |
1509 /* Keeps track of unclosed groups. */ | |
1510 compile_stack_type compile_stack; | |
1511 | |
1512 /* Points to the current (ending) position in the pattern. */ | |
1513 const char *p = pattern; | |
1514 const char *pend = pattern + size; | |
1515 | |
1516 /* How to translate the characters in the pattern. */ | |
1517 char *translate = bufp->translate; | |
1518 | |
1519 /* Address of the count-byte of the most recently inserted `exactn' | |
1520 command. This makes it possible to tell if a new exact-match | |
1521 character can be added to that command or if the character requires | |
1522 a new `exactn' command. */ | |
1523 unsigned char *pending_exact = 0; | |
1524 | |
1525 /* Address of start of the most recently finished expression. | |
1526 This tells, e.g., postfix * where to find the start of its | |
1527 operand. Reset at the beginning of groups and alternatives. */ | |
1528 unsigned char *laststart = 0; | |
1529 | |
1530 /* Address of beginning of regexp, or inside of last group. */ | |
1531 unsigned char *begalt; | |
1532 | |
1533 /* Place in the uncompiled pattern (i.e., the {) to | |
1534 which to go back if the interval is invalid. */ | |
1535 const char *beg_interval; | |
1536 | |
1537 /* Address of the place where a forward jump should go to the end of | |
1538 the containing expression. Each alternative of an `or' -- except the | |
1539 last -- ends with a forward jump of this sort. */ | |
1540 unsigned char *fixup_alt_jump = 0; | |
1541 | |
1542 /* Counts open-groups as they are encountered. Remembered for the | |
1543 matching close-group on the compile stack, so the same register | |
1544 number is put in the stop_memory as the start_memory. */ | |
1545 regnum_t regnum = 0; | |
1546 | |
1547 #ifdef DEBUG | |
1548 DEBUG_PRINT1 ("\nCompiling pattern: "); | |
1549 if (debug) | |
1550 { | |
1551 unsigned debug_count; | |
1552 | |
1553 for (debug_count = 0; debug_count < size; debug_count++) | |
1554 printchar (pattern[debug_count]); | |
1555 putchar ('\n'); | |
1556 } | |
1557 #endif /* DEBUG */ | |
1558 | |
1559 /* Initialize the compile stack. */ | |
1560 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
1561 if (compile_stack.stack == NULL) | |
1562 return REG_ESPACE; | |
1563 | |
1564 compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
1565 compile_stack.avail = 0; | |
1566 | |
1567 /* Initialize the pattern buffer. */ | |
1568 bufp->syntax = syntax; | |
1569 bufp->fastmap_accurate = 0; | |
1570 bufp->not_bol = bufp->not_eol = 0; | |
1571 | |
1572 /* Set `used' to zero, so that if we return an error, the pattern | |
1573 printer (for debugging) will think there's no pattern. We reset it | |
1574 at the end. */ | |
1575 bufp->used = 0; | |
1576 | |
1577 /* Always count groups, whether or not bufp->no_sub is set. */ | |
1578 bufp->re_nsub = 0; | |
1579 | |
1580 #if !defined (emacs) && !defined (SYNTAX_TABLE) | |
1581 /* Initialize the syntax table. */ | |
1582 init_syntax_once (); | |
1583 #endif | |
1584 | |
1585 if (bufp->allocated == 0) | |
1586 { | |
1587 if (bufp->buffer) | |
1588 { /* If zero allocated, but buffer is non-null, try to realloc | |
1589 enough space. This loses if buffer's address is bogus, but | |
1590 that is the user's responsibility. */ | |
1591 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); | |
1592 } | |
1593 else | |
1594 { /* Caller did not allocate a buffer. Do it for them. */ | |
1595 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); | |
1596 } | |
1597 if (!bufp->buffer) return REG_ESPACE; | |
1598 | |
1599 bufp->allocated = INIT_BUF_SIZE; | |
1600 } | |
1601 | |
1602 begalt = b = bufp->buffer; | |
1603 | |
1604 /* Loop through the uncompiled pattern until we're at the end. */ | |
1605 while (p != pend) | |
1606 { | |
1607 PATFETCH (c); | |
1608 | |
1609 switch (c) | |
1610 { | |
1611 case '^': | |
1612 { | |
1613 if ( /* If at start of pattern, it's an operator. */ | |
1614 p == pattern + 1 | |
1615 /* If context independent, it's an operator. */ | |
1616 || syntax & RE_CONTEXT_INDEP_ANCHORS | |
1617 /* Otherwise, depends on what's come before. */ | |
1618 || at_begline_loc_p (pattern, p, syntax)) | |
1619 BUF_PUSH (begline); | |
1620 else | |
1621 goto normal_char; | |
1622 } | |
1623 break; | |
1624 | |
1625 | |
1626 case '$': | |
1627 { | |
1628 if ( /* If at end of pattern, it's an operator. */ | |
1629 p == pend | |
1630 /* If context independent, it's an operator. */ | |
1631 || syntax & RE_CONTEXT_INDEP_ANCHORS | |
1632 /* Otherwise, depends on what's next. */ | |
1633 || at_endline_loc_p (p, pend, syntax)) | |
1634 BUF_PUSH (endline); | |
1635 else | |
1636 goto normal_char; | |
1637 } | |
1638 break; | |
1639 | |
1640 | |
1641 case '+': | |
1642 case '?': | |
1643 if ((syntax & RE_BK_PLUS_QM) | |
1644 || (syntax & RE_LIMITED_OPS)) | |
1645 goto normal_char; | |
1646 handle_plus: | |
1647 case '*': | |
1648 /* If there is no previous pattern... */ | |
1649 if (!laststart) | |
1650 { | |
1651 if (syntax & RE_CONTEXT_INVALID_OPS) | |
1652 return REG_BADRPT; | |
1653 else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
1654 goto normal_char; | |
1655 } | |
1656 | |
1657 { | |
1658 /* Are we optimizing this jump? */ | |
1659 boolean keep_string_p = false; | |
1660 | |
1661 /* 1 means zero (many) matches is allowed. */ | |
1662 char zero_times_ok = 0, many_times_ok = 0; | |
1663 | |
1664 /* If there is a sequence of repetition chars, collapse it | |
1665 down to just one (the right one). We can't combine | |
1666 interval operators with these because of, e.g., `a{2}*', | |
1667 which should only match an even number of `a's. */ | |
1668 | |
1669 for (;;) | |
1670 { | |
1671 zero_times_ok |= c != '+'; | |
1672 many_times_ok |= c != '?'; | |
1673 | |
1674 if (p == pend) | |
1675 break; | |
1676 | |
1677 PATFETCH (c); | |
1678 | |
1679 if (c == '*' | |
1680 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) | |
1681 ; | |
1682 | |
1683 else if (syntax & RE_BK_PLUS_QM && c == '\\') | |
1684 { | |
1685 if (p == pend) return REG_EESCAPE; | |
1686 | |
1687 PATFETCH (c1); | |
1688 if (!(c1 == '+' || c1 == '?')) | |
1689 { | |
1690 PATUNFETCH; | |
1691 PATUNFETCH; | |
1692 break; | |
1693 } | |
1694 | |
1695 c = c1; | |
1696 } | |
1697 else | |
1698 { | |
1699 PATUNFETCH; | |
1700 break; | |
1701 } | |
1702 | |
1703 /* If we get here, we found another repeat character. */ | |
1704 } | |
1705 | |
1706 /* Star, etc. applied to an empty pattern is equivalent | |
1707 to an empty pattern. */ | |
1708 if (!laststart) | |
1709 break; | |
1710 | |
1711 /* Now we know whether or not zero matches is allowed | |
1712 and also whether or not two or more matches is allowed. */ | |
1713 if (many_times_ok) | |
1714 { /* More than one repetition is allowed, so put in at the | |
1715 end a backward relative jump from `b' to before the next | |
1716 jump we're going to put in below (which jumps from | |
1717 laststart to after this jump). | |
1718 | |
1719 But if we are at the `*' in the exact sequence `.*\n', | |
1720 insert an unconditional jump backwards to the ., | |
1721 instead of the beginning of the loop. This way we only | |
1722 push a failure point once, instead of every time | |
1723 through the loop. */ | |
1724 assert (p - 1 > pattern); | |
1725 | |
1726 /* Allocate the space for the jump. */ | |
1727 GET_BUFFER_SPACE (3); | |
1728 | |
1729 /* We know we are not at the first character of the pattern, | |
1730 because laststart was nonzero. And we've already | |
1731 incremented `p', by the way, to be the character after | |
1732 the `*'. Do we have to do something analogous here | |
1733 for null bytes, because of RE_DOT_NOT_NULL? */ | |
1734 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.') | |
2453 | 1735 && zero_times_ok |
1155 | 1736 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n') |
1737 && !(syntax & RE_DOT_NEWLINE)) | |
1738 { /* We have .*\n. */ | |
1739 STORE_JUMP (jump, b, laststart); | |
1740 keep_string_p = true; | |
1741 } | |
1742 else | |
1743 /* Anything else. */ | |
1744 STORE_JUMP (maybe_pop_jump, b, laststart - 3); | |
1745 | |
1746 /* We've added more stuff to the buffer. */ | |
1747 b += 3; | |
1748 } | |
1749 | |
1750 /* On failure, jump from laststart to b + 3, which will be the | |
1751 end of the buffer after this jump is inserted. */ | |
1752 GET_BUFFER_SPACE (3); | |
1753 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump | |
1754 : on_failure_jump, | |
1755 laststart, b + 3); | |
1756 pending_exact = 0; | |
1757 b += 3; | |
1758 | |
1759 if (!zero_times_ok) | |
1760 { | |
1761 /* At least one repetition is required, so insert a | |
1762 `dummy_failure_jump' before the initial | |
1763 `on_failure_jump' instruction of the loop. This | |
1764 effects a skip over that instruction the first time | |
1765 we hit that loop. */ | |
1766 GET_BUFFER_SPACE (3); | |
1767 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6); | |
1768 b += 3; | |
1769 } | |
1770 } | |
1771 break; | |
1772 | |
1773 | |
1774 case '.': | |
1775 laststart = b; | |
1776 BUF_PUSH (anychar); | |
1777 break; | |
1778 | |
1779 | |
1780 case '[': | |
1781 { | |
1782 boolean had_char_class = false; | |
1783 | |
1784 if (p == pend) return REG_EBRACK; | |
1785 | |
1786 /* Ensure that we have enough space to push a charset: the | |
1787 opcode, the length count, and the bitset; 34 bytes in all. */ | |
1788 GET_BUFFER_SPACE (34); | |
1789 | |
1790 laststart = b; | |
1791 | |
1792 /* We test `*p == '^' twice, instead of using an if | |
1793 statement, so we only need one BUF_PUSH. */ | |
1794 BUF_PUSH (*p == '^' ? charset_not : charset); | |
1795 if (*p == '^') | |
1796 p++; | |
1797 | |
1798 /* Remember the first position in the bracket expression. */ | |
1799 p1 = p; | |
1800 | |
1801 /* Push the number of bytes in the bitmap. */ | |
1802 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
1803 | |
1804 /* Clear the whole map. */ | |
1805 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); | |
1806 | |
1807 /* charset_not matches newline according to a syntax bit. */ | |
1808 if ((re_opcode_t) b[-2] == charset_not | |
1809 && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
1810 SET_LIST_BIT ('\n'); | |
1811 | |
1812 /* Read in characters and ranges, setting map bits. */ | |
1813 for (;;) | |
1814 { | |
1815 if (p == pend) return REG_EBRACK; | |
1816 | |
1817 PATFETCH (c); | |
1818 | |
1819 /* \ might escape characters inside [...] and [^...]. */ | |
1820 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
1821 { | |
1822 if (p == pend) return REG_EESCAPE; | |
1823 | |
1824 PATFETCH (c1); | |
1825 SET_LIST_BIT (c1); | |
1826 continue; | |
1827 } | |
1828 | |
1829 /* Could be the end of the bracket expression. If it's | |
1830 not (i.e., when the bracket expression is `[]' so | |
1831 far), the ']' character bit gets set way below. */ | |
1832 if (c == ']' && p != p1 + 1) | |
1833 break; | |
1834 | |
1835 /* Look ahead to see if it's a range when the last thing | |
1836 was a character class. */ | |
1837 if (had_char_class && c == '-' && *p != ']') | |
1838 return REG_ERANGE; | |
1839 | |
1840 /* Look ahead to see if it's a range when the last thing | |
1841 was a character: if this is a hyphen not at the | |
1842 beginning or the end of a list, then it's the range | |
1843 operator. */ | |
1844 if (c == '-' | |
1845 && !(p - 2 >= pattern && p[-2] == '[') | |
1846 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') | |
1847 && *p != ']') | |
1848 { | |
1849 reg_errcode_t ret | |
1850 = compile_range (&p, pend, translate, syntax, b); | |
1851 if (ret != REG_NOERROR) return ret; | |
1852 } | |
1853 | |
1854 else if (p[0] == '-' && p[1] != ']') | |
1855 { /* This handles ranges made up of characters only. */ | |
1856 reg_errcode_t ret; | |
1857 | |
1858 /* Move past the `-'. */ | |
1859 PATFETCH (c1); | |
1860 | |
1861 ret = compile_range (&p, pend, translate, syntax, b); | |
1862 if (ret != REG_NOERROR) return ret; | |
1863 } | |
1864 | |
1865 /* See if we're at the beginning of a possible character | |
1866 class. */ | |
1867 | |
1868 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
1869 { /* Leave room for the null. */ | |
1870 char str[CHAR_CLASS_MAX_LENGTH + 1]; | |
1871 | |
1872 PATFETCH (c); | |
1873 c1 = 0; | |
1874 | |
1875 /* If pattern is `[[:'. */ | |
1876 if (p == pend) return REG_EBRACK; | |
1877 | |
1878 for (;;) | |
1879 { | |
1880 PATFETCH (c); | |
1881 if (c == ':' || c == ']' || p == pend | |
1882 || c1 == CHAR_CLASS_MAX_LENGTH) | |
1883 break; | |
1884 str[c1++] = c; | |
1885 } | |
1886 str[c1] = '\0'; | |
1887 | |
1888 /* If isn't a word bracketed by `[:' and:`]': | |
1889 undo the ending character, the letters, and leave | |
1890 the leading `:' and `[' (but set bits for them). */ | |
1891 if (c == ':' && *p == ']') | |
1892 { | |
1893 int ch; | |
1894 boolean is_alnum = STREQ (str, "alnum"); | |
1895 boolean is_alpha = STREQ (str, "alpha"); | |
1896 boolean is_blank = STREQ (str, "blank"); | |
1897 boolean is_cntrl = STREQ (str, "cntrl"); | |
1898 boolean is_digit = STREQ (str, "digit"); | |
1899 boolean is_graph = STREQ (str, "graph"); | |
1900 boolean is_lower = STREQ (str, "lower"); | |
1901 boolean is_print = STREQ (str, "print"); | |
1902 boolean is_punct = STREQ (str, "punct"); | |
1903 boolean is_space = STREQ (str, "space"); | |
1904 boolean is_upper = STREQ (str, "upper"); | |
1905 boolean is_xdigit = STREQ (str, "xdigit"); | |
1906 | |
1907 if (!IS_CHAR_CLASS (str)) return REG_ECTYPE; | |
1908 | |
1909 /* Throw away the ] at the end of the character | |
1910 class. */ | |
1911 PATFETCH (c); | |
1912 | |
1913 if (p == pend) return REG_EBRACK; | |
1914 | |
1915 for (ch = 0; ch < 1 << BYTEWIDTH; ch++) | |
1916 { | |
1668 | 1917 if ( (is_alnum && ISALNUM (ch)) |
1918 || (is_alpha && ISALPHA (ch)) | |
1919 || (is_blank && ISBLANK (ch)) | |
1920 || (is_cntrl && ISCNTRL (ch)) | |
1921 || (is_digit && ISDIGIT (ch)) | |
1922 || (is_graph && ISGRAPH (ch)) | |
1923 || (is_lower && ISLOWER (ch)) | |
1924 || (is_print && ISPRINT (ch)) | |
1925 || (is_punct && ISPUNCT (ch)) | |
1926 || (is_space && ISSPACE (ch)) | |
1927 || (is_upper && ISUPPER (ch)) | |
1928 || (is_xdigit && ISXDIGIT (ch))) | |
1155 | 1929 SET_LIST_BIT (ch); |
1930 } | |
1931 had_char_class = true; | |
1932 } | |
1933 else | |
1934 { | |
1935 c1++; | |
1936 while (c1--) | |
1937 PATUNFETCH; | |
1938 SET_LIST_BIT ('['); | |
1939 SET_LIST_BIT (':'); | |
1940 had_char_class = false; | |
1941 } | |
1942 } | |
1943 else | |
1944 { | |
1945 had_char_class = false; | |
1946 SET_LIST_BIT (c); | |
1947 } | |
1948 } | |
1949 | |
1950 /* Discard any (non)matching list bytes that are all 0 at the | |
1951 end of the map. Decrease the map-length byte too. */ | |
1952 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) | |
1953 b[-1]--; | |
1954 b += b[-1]; | |
1955 } | |
1956 break; | |
1957 | |
1958 | |
1959 case '(': | |
1960 if (syntax & RE_NO_BK_PARENS) | |
1961 goto handle_open; | |
1962 else | |
1963 goto normal_char; | |
1964 | |
1965 | |
1966 case ')': | |
1967 if (syntax & RE_NO_BK_PARENS) | |
1968 goto handle_close; | |
1969 else | |
1970 goto normal_char; | |
1971 | |
1972 | |
1973 case '\n': | |
1974 if (syntax & RE_NEWLINE_ALT) | |
1975 goto handle_alt; | |
1976 else | |
1977 goto normal_char; | |
1978 | |
1979 | |
1980 case '|': | |
1981 if (syntax & RE_NO_BK_VBAR) | |
1982 goto handle_alt; | |
1983 else | |
1984 goto normal_char; | |
1985 | |
1986 | |
1987 case '{': | |
1988 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
1989 goto handle_interval; | |
1990 else | |
1991 goto normal_char; | |
1992 | |
1993 | |
1994 case '\\': | |
1995 if (p == pend) return REG_EESCAPE; | |
1996 | |
1997 /* Do not translate the character after the \, so that we can | |
1998 distinguish, e.g., \B from \b, even if we normally would | |
1999 translate, e.g., B to b. */ | |
2000 PATFETCH_RAW (c); | |
2001 | |
2002 switch (c) | |
2003 { | |
2004 case '(': | |
2005 if (syntax & RE_NO_BK_PARENS) | |
2006 goto normal_backslash; | |
2007 | |
2008 handle_open: | |
2009 bufp->re_nsub++; | |
2010 regnum++; | |
2011 | |
2012 if (COMPILE_STACK_FULL) | |
2013 { | |
2014 RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
2015 compile_stack_elt_t); | |
2016 if (compile_stack.stack == NULL) return REG_ESPACE; | |
2017 | |
2018 compile_stack.size <<= 1; | |
2019 } | |
2020 | |
2021 /* These are the values to restore when we hit end of this | |
2022 group. They are all relative offsets, so that if the | |
2023 whole pattern moves because of realloc, they will still | |
2024 be valid. */ | |
2025 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; | |
2026 COMPILE_STACK_TOP.fixup_alt_jump | |
2027 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; | |
2028 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; | |
2029 COMPILE_STACK_TOP.regnum = regnum; | |
2030 | |
2031 /* We will eventually replace the 0 with the number of | |
2032 groups inner to this one. But do not push a | |
2033 start_memory for groups beyond the last one we can | |
2034 represent in the compiled pattern. */ | |
2035 if (regnum <= MAX_REGNUM) | |
2036 { | |
2037 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2; | |
2038 BUF_PUSH_3 (start_memory, regnum, 0); | |
2039 } | |
2040 | |
2041 compile_stack.avail++; | |
2042 | |
2043 fixup_alt_jump = 0; | |
2044 laststart = 0; | |
2045 begalt = b; | |
2453 | 2046 /* If we've reached MAX_REGNUM groups, then this open |
2047 won't actually generate any code, so we'll have to | |
2048 clear pending_exact explicitly. */ | |
2049 pending_exact = 0; | |
1155 | 2050 break; |
2051 | |
2052 | |
2053 case ')': | |
2054 if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
2055 | |
2056 if (COMPILE_STACK_EMPTY) | |
2057 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
2058 goto normal_backslash; | |
2059 else | |
2060 return REG_ERPAREN; | |
2061 | |
2062 handle_close: | |
2063 if (fixup_alt_jump) | |
2064 { /* Push a dummy failure point at the end of the | |
2065 alternative for a possible future | |
2066 `pop_failure_jump' to pop. See comments at | |
2067 `push_dummy_failure' in `re_match_2'. */ | |
2068 BUF_PUSH (push_dummy_failure); | |
2069 | |
2070 /* We allocated space for this jump when we assigned | |
2071 to `fixup_alt_jump', in the `handle_alt' case below. */ | |
2072 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); | |
2073 } | |
2074 | |
2075 /* See similar code for backslashed left paren above. */ | |
2076 if (COMPILE_STACK_EMPTY) | |
2077 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
2078 goto normal_char; | |
2079 else | |
2080 return REG_ERPAREN; | |
2081 | |
2082 /* Since we just checked for an empty stack above, this | |
2083 ``can't happen''. */ | |
2084 assert (compile_stack.avail != 0); | |
2085 { | |
2086 /* We don't just want to restore into `regnum', because | |
2087 later groups should continue to be numbered higher, | |
2088 as in `(ab)c(de)' -- the second group is #2. */ | |
2089 regnum_t this_group_regnum; | |
2090 | |
2091 compile_stack.avail--; | |
2092 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; | |
2093 fixup_alt_jump | |
2094 = COMPILE_STACK_TOP.fixup_alt_jump | |
2095 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
2096 : 0; | |
2097 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; | |
2098 this_group_regnum = COMPILE_STACK_TOP.regnum; | |
2453 | 2099 /* If we've reached MAX_REGNUM groups, then this open |
2100 won't actually generate any code, so we'll have to | |
2101 clear pending_exact explicitly. */ | |
2102 pending_exact = 0; | |
1155 | 2103 |
2104 /* We're at the end of the group, so now we know how many | |
2105 groups were inside this one. */ | |
2106 if (this_group_regnum <= MAX_REGNUM) | |
2107 { | |
2108 unsigned char *inner_group_loc | |
2109 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset; | |
2110 | |
2111 *inner_group_loc = regnum - this_group_regnum; | |
2112 BUF_PUSH_3 (stop_memory, this_group_regnum, | |
2113 regnum - this_group_regnum); | |
2114 } | |
2115 } | |
2116 break; | |
2117 | |
2118 | |
2119 case '|': /* `\|'. */ | |
2120 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
2121 goto normal_backslash; | |
2122 handle_alt: | |
2123 if (syntax & RE_LIMITED_OPS) | |
2124 goto normal_char; | |
2125 | |
2126 /* Insert before the previous alternative a jump which | |
2127 jumps to this alternative if the former fails. */ | |
2128 GET_BUFFER_SPACE (3); | |
2129 INSERT_JUMP (on_failure_jump, begalt, b + 6); | |
2130 pending_exact = 0; | |
2131 b += 3; | |
2132 | |
2133 /* The alternative before this one has a jump after it | |
2134 which gets executed if it gets matched. Adjust that | |
2135 jump so it will jump to this alternative's analogous | |
2136 jump (put in below, which in turn will jump to the next | |
2137 (if any) alternative's such jump, etc.). The last such | |
2138 jump jumps to the correct final destination. A picture: | |
2139 _____ _____ | |
2140 | | | | | |
2141 | v | v | |
2142 a | b | c | |
2143 | |
1637 | 2144 If we are at `b', then fixup_alt_jump right now points to a |
2145 three-byte space after `a'. We'll put in the jump, set | |
2146 fixup_alt_jump to right after `b', and leave behind three | |
2147 bytes which we'll fill in when we get to after `c'. */ | |
1155 | 2148 |
2149 if (fixup_alt_jump) | |
2150 STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
2151 | |
2152 /* Mark and leave space for a jump after this alternative, | |
2153 to be filled in later either by next alternative or | |
2154 when know we're at the end of a series of alternatives. */ | |
2155 fixup_alt_jump = b; | |
2156 GET_BUFFER_SPACE (3); | |
2157 b += 3; | |
2158 | |
2159 laststart = 0; | |
2160 begalt = b; | |
2161 break; | |
2162 | |
2163 | |
2164 case '{': | |
2165 /* If \{ is a literal. */ | |
2166 if (!(syntax & RE_INTERVALS) | |
2167 /* If we're at `\{' and it's not the open-interval | |
2168 operator. */ | |
2169 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES)) | |
2170 || (p - 2 == pattern && p == pend)) | |
2171 goto normal_backslash; | |
2172 | |
2173 handle_interval: | |
2174 { | |
2175 /* If got here, then the syntax allows intervals. */ | |
2176 | |
2177 /* At least (most) this many matches must be made. */ | |
2178 int lower_bound = -1, upper_bound = -1; | |
2179 | |
2180 beg_interval = p - 1; | |
2181 | |
2182 if (p == pend) | |
2183 { | |
2184 if (syntax & RE_NO_BK_BRACES) | |
2185 goto unfetch_interval; | |
2186 else | |
2187 return REG_EBRACE; | |
2188 } | |
2189 | |
2190 GET_UNSIGNED_NUMBER (lower_bound); | |
2191 | |
2192 if (c == ',') | |
2193 { | |
2194 GET_UNSIGNED_NUMBER (upper_bound); | |
2195 if (upper_bound < 0) upper_bound = RE_DUP_MAX; | |
2196 } | |
2197 else | |
2198 /* Interval such as `{1}' => match exactly once. */ | |
2199 upper_bound = lower_bound; | |
2200 | |
2201 if (lower_bound < 0 || upper_bound > RE_DUP_MAX | |
2202 || lower_bound > upper_bound) | |
2203 { | |
2204 if (syntax & RE_NO_BK_BRACES) | |
2205 goto unfetch_interval; | |
2206 else | |
2207 return REG_BADBR; | |
2208 } | |
2209 | |
2210 if (!(syntax & RE_NO_BK_BRACES)) | |
2211 { | |
2212 if (c != '\\') return REG_EBRACE; | |
2213 | |
2214 PATFETCH (c); | |
2215 } | |
2216 | |
2217 if (c != '}') | |
2218 { | |
2219 if (syntax & RE_NO_BK_BRACES) | |
2220 goto unfetch_interval; | |
2221 else | |
2222 return REG_BADBR; | |
2223 } | |
2224 | |
2225 /* We just parsed a valid interval. */ | |
2226 | |
2227 /* If it's invalid to have no preceding re. */ | |
2228 if (!laststart) | |
2229 { | |
2230 if (syntax & RE_CONTEXT_INVALID_OPS) | |
2231 return REG_BADRPT; | |
2232 else if (syntax & RE_CONTEXT_INDEP_OPS) | |
2233 laststart = b; | |
2234 else | |
2235 goto unfetch_interval; | |
2236 } | |
2237 | |
2238 /* If the upper bound is zero, don't want to succeed at | |
2239 all; jump from `laststart' to `b + 3', which will be | |
2240 the end of the buffer after we insert the jump. */ | |
2241 if (upper_bound == 0) | |
2242 { | |
2243 GET_BUFFER_SPACE (3); | |
2244 INSERT_JUMP (jump, laststart, b + 3); | |
2245 b += 3; | |
2246 } | |
2247 | |
2248 /* Otherwise, we have a nontrivial interval. When | |
2249 we're all done, the pattern will look like: | |
2250 set_number_at <jump count> <upper bound> | |
2251 set_number_at <succeed_n count> <lower bound> | |
2252 succeed_n <after jump addr> <succed_n count> | |
2253 <body of loop> | |
2254 jump_n <succeed_n addr> <jump count> | |
2255 (The upper bound and `jump_n' are omitted if | |
2256 `upper_bound' is 1, though.) */ | |
2257 else | |
2258 { /* If the upper bound is > 1, we need to insert | |
2259 more at the end of the loop. */ | |
2260 unsigned nbytes = 10 + (upper_bound > 1) * 10; | |
2261 | |
2262 GET_BUFFER_SPACE (nbytes); | |
2263 | |
2264 /* Initialize lower bound of the `succeed_n', even | |
2265 though it will be set during matching by its | |
2266 attendant `set_number_at' (inserted next), | |
2267 because `re_compile_fastmap' needs to know. | |
2268 Jump to the `jump_n' we might insert below. */ | |
2269 INSERT_JUMP2 (succeed_n, laststart, | |
2270 b + 5 + (upper_bound > 1) * 5, | |
2271 lower_bound); | |
2272 b += 5; | |
2273 | |
2274 /* Code to initialize the lower bound. Insert | |
2275 before the `succeed_n'. The `5' is the last two | |
2276 bytes of this `set_number_at', plus 3 bytes of | |
2277 the following `succeed_n'. */ | |
2278 insert_op2 (set_number_at, laststart, 5, lower_bound, b); | |
2279 b += 5; | |
2280 | |
2281 if (upper_bound > 1) | |
2282 { /* More than one repetition is allowed, so | |
2283 append a backward jump to the `succeed_n' | |
2284 that starts this interval. | |
2285 | |
2286 When we've reached this during matching, | |
2287 we'll have matched the interval once, so | |
2288 jump back only `upper_bound - 1' times. */ | |
2289 STORE_JUMP2 (jump_n, b, laststart + 5, | |
2290 upper_bound - 1); | |
2291 b += 5; | |
2292 | |
2293 /* The location we want to set is the second | |
2294 parameter of the `jump_n'; that is `b-2' as | |
2295 an absolute address. `laststart' will be | |
2296 the `set_number_at' we're about to insert; | |
2297 `laststart+3' the number to set, the source | |
2298 for the relative address. But we are | |
2299 inserting into the middle of the pattern -- | |
2300 so everything is getting moved up by 5. | |
2301 Conclusion: (b - 2) - (laststart + 3) + 5, | |
2302 i.e., b - laststart. | |
2303 | |
2304 We insert this at the beginning of the loop | |
2305 so that if we fail during matching, we'll | |
2306 reinitialize the bounds. */ | |
2307 insert_op2 (set_number_at, laststart, b - laststart, | |
2308 upper_bound - 1, b); | |
2309 b += 5; | |
2310 } | |
2311 } | |
2312 pending_exact = 0; | |
2313 beg_interval = NULL; | |
2314 } | |
2315 break; | |
2316 | |
2317 unfetch_interval: | |
2318 /* If an invalid interval, match the characters as literals. */ | |
2319 assert (beg_interval); | |
2320 p = beg_interval; | |
2321 beg_interval = NULL; | |
2322 | |
2323 /* normal_char and normal_backslash need `c'. */ | |
2324 PATFETCH (c); | |
2325 | |
2326 if (!(syntax & RE_NO_BK_BRACES)) | |
2327 { | |
2328 if (p > pattern && p[-1] == '\\') | |
2329 goto normal_backslash; | |
2330 } | |
2331 goto normal_char; | |
2332 | |
2333 #ifdef emacs | |
2334 /* There is no way to specify the before_dot and after_dot | |
2335 operators. rms says this is ok. --karl */ | |
2336 case '=': | |
2337 BUF_PUSH (at_dot); | |
2338 break; | |
2339 | |
2340 case 's': | |
2341 laststart = b; | |
2342 PATFETCH (c); | |
2343 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
2344 break; | |
2345 | |
2346 case 'S': | |
2347 laststart = b; | |
2348 PATFETCH (c); | |
2349 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
2350 break; | |
2351 #endif /* emacs */ | |
2352 | |
2353 | |
2354 case 'w': | |
2355 laststart = b; | |
2356 BUF_PUSH (wordchar); | |
2357 break; | |
2358 | |
2359 | |
2360 case 'W': | |
2361 laststart = b; | |
2362 BUF_PUSH (notwordchar); | |
2363 break; | |
2364 | |
2365 | |
2366 case '<': | |
2367 BUF_PUSH (wordbeg); | |
2368 break; | |
2369 | |
2370 case '>': | |
2371 BUF_PUSH (wordend); | |
2372 break; | |
2373 | |
2374 case 'b': | |
2375 BUF_PUSH (wordbound); | |
2376 break; | |
2377 | |
2378 case 'B': | |
2379 BUF_PUSH (notwordbound); | |
2380 break; | |
2381 | |
2382 case '`': | |
2383 BUF_PUSH (begbuf); | |
2384 break; | |
2385 | |
2386 case '\'': | |
2387 BUF_PUSH (endbuf); | |
2388 break; | |
2389 | |
2390 case '1': case '2': case '3': case '4': case '5': | |
2391 case '6': case '7': case '8': case '9': | |
2392 if (syntax & RE_NO_BK_REFS) | |
2393 goto normal_char; | |
2394 | |
2395 c1 = c - '0'; | |
2396 | |
2397 if (c1 > regnum) | |
2398 return REG_ESUBREG; | |
2399 | |
2400 /* Can't back reference to a subexpression if inside of it. */ | |
2401 if (group_in_compile_stack (compile_stack, c1)) | |
2402 goto normal_char; | |
2403 | |
2404 laststart = b; | |
2405 BUF_PUSH_2 (duplicate, c1); | |
2406 break; | |
2407 | |
2408 | |
2409 case '+': | |
2410 case '?': | |
2411 if (syntax & RE_BK_PLUS_QM) | |
2412 goto handle_plus; | |
2413 else | |
2414 goto normal_backslash; | |
2415 | |
2416 default: | |
2417 normal_backslash: | |
2418 /* You might think it would be useful for \ to mean | |
2419 not to translate; but if we don't translate it | |
2420 it will never match anything. */ | |
2421 c = TRANSLATE (c); | |
2422 goto normal_char; | |
2423 } | |
2424 break; | |
2425 | |
2426 | |
2427 default: | |
2428 /* Expects the character in `c'. */ | |
2429 normal_char: | |
2430 /* If no exactn currently being built. */ | |
2431 if (!pending_exact | |
2432 | |
2433 /* If last exactn not at current position. */ | |
2434 || pending_exact + *pending_exact + 1 != b | |
2435 | |
2436 /* We have only one byte following the exactn for the count. */ | |
2437 || *pending_exact == (1 << BYTEWIDTH) - 1 | |
2438 | |
2439 /* If followed by a repetition operator. */ | |
2440 || *p == '*' || *p == '^' | |
2441 || ((syntax & RE_BK_PLUS_QM) | |
2442 ? *p == '\\' && (p[1] == '+' || p[1] == '?') | |
2443 : (*p == '+' || *p == '?')) | |
2444 || ((syntax & RE_INTERVALS) | |
2445 && ((syntax & RE_NO_BK_BRACES) | |
2446 ? *p == '{' | |
2447 : (p[0] == '\\' && p[1] == '{')))) | |
2448 { | |
2449 /* Start building a new exactn. */ | |
2450 | |
2451 laststart = b; | |
2452 | |
2453 BUF_PUSH_2 (exactn, 0); | |
2454 pending_exact = b - 1; | |
2455 } | |
2456 | |
2457 BUF_PUSH (c); | |
2458 (*pending_exact)++; | |
2459 break; | |
2460 } /* switch (c) */ | |
2461 } /* while p != pend */ | |
2462 | |
2463 | |
2464 /* Through the pattern now. */ | |
2465 | |
2466 if (fixup_alt_jump) | |
2467 STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
2468 | |
2469 if (!COMPILE_STACK_EMPTY) | |
2470 return REG_EPAREN; | |
2471 | |
2472 free (compile_stack.stack); | |
2473 | |
2474 /* We have succeeded; set the length of the buffer. */ | |
2475 bufp->used = b - bufp->buffer; | |
2476 | |
2477 #ifdef DEBUG | |
2478 if (debug) | |
2479 { | |
2615 | 2480 DEBUG_PRINT1 ("\nCompiled pattern: \n"); |
1155 | 2481 print_compiled_pattern (bufp); |
2482 } | |
2483 #endif /* DEBUG */ | |
2484 | |
2952 | 2485 #ifndef MATCH_MAY_ALLOCATE |
2949 | 2486 /* Initialize the failure stack to the largest possible stack. This |
2487 isn't necessary unless we're trying to avoid calling alloca in | |
2488 the search and match routines. */ | |
2489 { | |
2490 int num_regs = bufp->re_nsub + 1; | |
2491 | |
2492 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size | |
2493 is strictly greater than re_max_failures, the largest possible stack | |
2494 is 2 * re_max_failures failure points. */ | |
2495 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); | |
2496 if (fail_stack.stack) | |
2497 fail_stack.stack = | |
2498 (fail_stack_elt_t *) realloc (fail_stack.stack, | |
2499 (fail_stack.size | |
2500 * sizeof (fail_stack_elt_t))); | |
2501 else | |
2502 fail_stack.stack = | |
2503 (fail_stack_elt_t *) malloc (fail_stack.size | |
2504 * sizeof (fail_stack_elt_t)); | |
2505 | |
2506 /* Initialize some other variables the matcher uses. */ | |
2507 RETALLOC_IF (regstart, num_regs, const char *); | |
2508 RETALLOC_IF (regend, num_regs, const char *); | |
2509 RETALLOC_IF (old_regstart, num_regs, const char *); | |
2510 RETALLOC_IF (old_regend, num_regs, const char *); | |
2511 RETALLOC_IF (best_regstart, num_regs, const char *); | |
2512 RETALLOC_IF (best_regend, num_regs, const char *); | |
2513 RETALLOC_IF (reg_info, num_regs, register_info_type); | |
2514 RETALLOC_IF (reg_dummy, num_regs, const char *); | |
2515 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type); | |
2516 } | |
2517 #endif | |
2518 | |
1155 | 2519 return REG_NOERROR; |
2520 } /* regex_compile */ | |
2521 | |
2522 /* Subroutines for `regex_compile'. */ | |
2523 | |
2524 /* Store OP at LOC followed by two-byte integer parameter ARG. */ | |
2525 | |
2526 static void | |
2527 store_op1 (op, loc, arg) | |
2528 re_opcode_t op; | |
2529 unsigned char *loc; | |
2530 int arg; | |
2531 { | |
2532 *loc = (unsigned char) op; | |
2533 STORE_NUMBER (loc + 1, arg); | |
2534 } | |
2535 | |
2536 | |
2537 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
2538 | |
2539 static void | |
2540 store_op2 (op, loc, arg1, arg2) | |
2541 re_opcode_t op; | |
2542 unsigned char *loc; | |
2543 int arg1, arg2; | |
2544 { | |
2545 *loc = (unsigned char) op; | |
2546 STORE_NUMBER (loc + 1, arg1); | |
2547 STORE_NUMBER (loc + 3, arg2); | |
2548 } | |
2549 | |
2550 | |
2551 /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
2552 for OP followed by two-byte integer parameter ARG. */ | |
2553 | |
2554 static void | |
2555 insert_op1 (op, loc, arg, end) | |
2556 re_opcode_t op; | |
2557 unsigned char *loc; | |
2558 int arg; | |
2559 unsigned char *end; | |
2560 { | |
2561 register unsigned char *pfrom = end; | |
2562 register unsigned char *pto = end + 3; | |
2563 | |
2564 while (pfrom != loc) | |
2565 *--pto = *--pfrom; | |
2566 | |
2567 store_op1 (op, loc, arg); | |
2568 } | |
2569 | |
2570 | |
2571 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
2572 | |
2573 static void | |
2574 insert_op2 (op, loc, arg1, arg2, end) | |
2575 re_opcode_t op; | |
2576 unsigned char *loc; | |
2577 int arg1, arg2; | |
2578 unsigned char *end; | |
2579 { | |
2580 register unsigned char *pfrom = end; | |
2581 register unsigned char *pto = end + 5; | |
2582 | |
2583 while (pfrom != loc) | |
2584 *--pto = *--pfrom; | |
2585 | |
2586 store_op2 (op, loc, arg1, arg2); | |
2587 } | |
2588 | |
2589 | |
2590 /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
2591 after an alternative or a begin-subexpression. We assume there is at | |
2592 least one character before the ^. */ | |
2593 | |
2594 static boolean | |
2595 at_begline_loc_p (pattern, p, syntax) | |
2596 const char *pattern, *p; | |
2597 reg_syntax_t syntax; | |
2598 { | |
2599 const char *prev = p - 2; | |
2600 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; | |
2601 | |
2602 return | |
2603 /* After a subexpression? */ | |
2604 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
2605 /* After an alternative? */ | |
2606 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); | |
2607 } | |
2608 | |
2609 | |
2610 /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
2611 at least one character after the $, i.e., `P < PEND'. */ | |
2612 | |
2613 static boolean | |
2614 at_endline_loc_p (p, pend, syntax) | |
2615 const char *p, *pend; | |
2616 int syntax; | |
2617 { | |
2618 const char *next = p; | |
2619 boolean next_backslash = *next == '\\'; | |
2620 const char *next_next = p + 1 < pend ? p + 1 : NULL; | |
2621 | |
2622 return | |
2623 /* Before a subexpression? */ | |
2624 (syntax & RE_NO_BK_PARENS ? *next == ')' | |
2625 : next_backslash && next_next && *next_next == ')') | |
2626 /* Before an alternative? */ | |
2627 || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
2628 : next_backslash && next_next && *next_next == '|'); | |
2629 } | |
2630 | |
2631 | |
2632 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and | |
2633 false if it's not. */ | |
2634 | |
2635 static boolean | |
2636 group_in_compile_stack (compile_stack, regnum) | |
2637 compile_stack_type compile_stack; | |
2638 regnum_t regnum; | |
2639 { | |
2640 int this_element; | |
2641 | |
2642 for (this_element = compile_stack.avail - 1; | |
2643 this_element >= 0; | |
2644 this_element--) | |
2645 if (compile_stack.stack[this_element].regnum == regnum) | |
2646 return true; | |
2647 | |
2648 return false; | |
2649 } | |
2650 | |
2651 | |
2652 /* Read the ending character of a range (in a bracket expression) from the | |
2653 uncompiled pattern *P_PTR (which ends at PEND). We assume the | |
2654 starting character is in `P[-2]'. (`P[-1]' is the character `-'.) | |
2655 Then we set the translation of all bits between the starting and | |
2656 ending characters (inclusive) in the compiled pattern B. | |
2657 | |
2658 Return an error code. | |
2659 | |
2660 We use these short variable names so we can use the same macros as | |
2661 `regex_compile' itself. */ | |
2662 | |
2663 static reg_errcode_t | |
2664 compile_range (p_ptr, pend, translate, syntax, b) | |
2665 const char **p_ptr, *pend; | |
2666 char *translate; | |
2667 reg_syntax_t syntax; | |
2668 unsigned char *b; | |
2669 { | |
2670 unsigned this_char; | |
2671 | |
2672 const char *p = *p_ptr; | |
1689 | 2673 int range_start, range_end; |
1155 | 2674 |
2675 if (p == pend) | |
2676 return REG_ERANGE; | |
2677 | |
1689 | 2678 /* Even though the pattern is a signed `char *', we need to fetch |
2679 with unsigned char *'s; if the high bit of the pattern character | |
2680 is set, the range endpoints will be negative if we fetch using a | |
2681 signed char *. | |
2682 | |
2683 We also want to fetch the endpoints without translating them; the | |
2684 appropriate translation is done in the bit-setting loop below. */ | |
2685 range_start = ((unsigned char *) p)[-2]; | |
2686 range_end = ((unsigned char *) p)[0]; | |
1155 | 2687 |
2688 /* Have to increment the pointer into the pattern string, so the | |
2689 caller isn't still at the ending character. */ | |
2690 (*p_ptr)++; | |
2691 | |
2692 /* If the start is after the end, the range is empty. */ | |
2693 if (range_start > range_end) | |
2694 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; | |
2695 | |
2696 /* Here we see why `this_char' has to be larger than an `unsigned | |
2697 char' -- the range is inclusive, so if `range_end' == 0xff | |
2698 (assuming 8-bit characters), we would otherwise go into an infinite | |
2699 loop, since all characters <= 0xff. */ | |
2700 for (this_char = range_start; this_char <= range_end; this_char++) | |
2701 { | |
2702 SET_LIST_BIT (TRANSLATE (this_char)); | |
2703 } | |
2704 | |
2705 return REG_NOERROR; | |
2706 } | |
2707 | |
2708 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
2709 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
2710 characters can start a string that matches the pattern. This fastmap | |
2711 is used by re_search to skip quickly over impossible starting points. | |
2712 | |
2713 The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
2714 area as BUFP->fastmap. | |
2715 | |
2716 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in | |
2717 the pattern buffer. | |
2718 | |
2719 Returns 0 if we succeed, -2 if an internal error. */ | |
2720 | |
2721 int | |
2722 re_compile_fastmap (bufp) | |
2723 struct re_pattern_buffer *bufp; | |
2724 { | |
2725 int j, k; | |
2952 | 2726 #ifdef MATCH_MAY_ALLOCATE |
1155 | 2727 fail_stack_type fail_stack; |
2949 | 2728 #endif |
1155 | 2729 #ifndef REGEX_MALLOC |
2730 char *destination; | |
2731 #endif | |
2732 /* We don't push any register information onto the failure stack. */ | |
2733 unsigned num_regs = 0; | |
2734 | |
2735 register char *fastmap = bufp->fastmap; | |
2736 unsigned char *pattern = bufp->buffer; | |
2737 unsigned long size = bufp->used; | |
4918
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
2738 unsigned char *p = pattern; |
1155 | 2739 register unsigned char *pend = pattern + size; |
2740 | |
2741 /* Assume that each path through the pattern can be null until | |
2742 proven otherwise. We set this false at the bottom of switch | |
2743 statement, to which we get only if a particular path doesn't | |
2744 match the empty string. */ | |
2745 boolean path_can_be_null = true; | |
2746 | |
2747 /* We aren't doing a `succeed_n' to begin with. */ | |
2748 boolean succeed_n_p = false; | |
2749 | |
2750 assert (fastmap != NULL && p != NULL); | |
2751 | |
2752 INIT_FAIL_STACK (); | |
2753 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ | |
2754 bufp->fastmap_accurate = 1; /* It will be when we're done. */ | |
2755 bufp->can_be_null = 0; | |
2756 | |
2757 while (p != pend || !FAIL_STACK_EMPTY ()) | |
2758 { | |
2759 if (p == pend) | |
2760 { | |
2761 bufp->can_be_null |= path_can_be_null; | |
2762 | |
2763 /* Reset for next path. */ | |
2764 path_can_be_null = true; | |
2765 | |
2766 p = fail_stack.stack[--fail_stack.avail]; | |
2767 } | |
2768 | |
2769 /* We should never be about to go beyond the end of the pattern. */ | |
2770 assert (p < pend); | |
2771 | |
2772 #ifdef SWITCH_ENUM_BUG | |
2773 switch ((int) ((re_opcode_t) *p++)) | |
2774 #else | |
2775 switch ((re_opcode_t) *p++) | |
2776 #endif | |
2777 { | |
2778 | |
2779 /* I guess the idea here is to simply not bother with a fastmap | |
2780 if a backreference is used, since it's too hard to figure out | |
2781 the fastmap for the corresponding group. Setting | |
2782 `can_be_null' stops `re_search_2' from using the fastmap, so | |
2783 that is all we do. */ | |
2784 case duplicate: | |
2785 bufp->can_be_null = 1; | |
2786 return 0; | |
2787 | |
2788 | |
2789 /* Following are the cases which match a character. These end | |
2790 with `break'. */ | |
2791 | |
2792 case exactn: | |
2793 fastmap[p[1]] = 1; | |
2794 break; | |
2795 | |
2796 | |
2797 case charset: | |
2798 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
2799 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) | |
2800 fastmap[j] = 1; | |
2801 break; | |
2802 | |
2803 | |
2804 case charset_not: | |
2805 /* Chars beyond end of map must be allowed. */ | |
2806 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) | |
2807 fastmap[j] = 1; | |
2808 | |
2809 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
2810 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) | |
2811 fastmap[j] = 1; | |
2812 break; | |
2813 | |
2814 | |
2815 case wordchar: | |
2816 for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2817 if (SYNTAX (j) == Sword) | |
2818 fastmap[j] = 1; | |
2819 break; | |
2820 | |
2821 | |
2822 case notwordchar: | |
2823 for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2824 if (SYNTAX (j) != Sword) | |
2825 fastmap[j] = 1; | |
2826 break; | |
2827 | |
2828 | |
2829 case anychar: | |
2830 /* `.' matches anything ... */ | |
2831 for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2832 fastmap[j] = 1; | |
2833 | |
2834 /* ... except perhaps newline. */ | |
2835 if (!(bufp->syntax & RE_DOT_NEWLINE)) | |
2836 fastmap['\n'] = 0; | |
2837 | |
2838 /* Return if we have already set `can_be_null'; if we have, | |
2839 then the fastmap is irrelevant. Something's wrong here. */ | |
2840 else if (bufp->can_be_null) | |
2841 return 0; | |
2842 | |
2843 /* Otherwise, have to check alternative paths. */ | |
2844 break; | |
2845 | |
2846 | |
2847 #ifdef emacs | |
2848 case syntaxspec: | |
2849 k = *p++; | |
2850 for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2851 if (SYNTAX (j) == (enum syntaxcode) k) | |
2852 fastmap[j] = 1; | |
2853 break; | |
2854 | |
2855 | |
2856 case notsyntaxspec: | |
2857 k = *p++; | |
2858 for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2859 if (SYNTAX (j) != (enum syntaxcode) k) | |
2860 fastmap[j] = 1; | |
2861 break; | |
2862 | |
2863 | |
2864 /* All cases after this match the empty string. These end with | |
2865 `continue'. */ | |
2866 | |
2867 | |
2868 case before_dot: | |
2869 case at_dot: | |
2870 case after_dot: | |
2871 continue; | |
2872 #endif /* not emacs */ | |
2873 | |
2874 | |
2875 case no_op: | |
2876 case begline: | |
2877 case endline: | |
2878 case begbuf: | |
2879 case endbuf: | |
2880 case wordbound: | |
2881 case notwordbound: | |
2882 case wordbeg: | |
2883 case wordend: | |
2884 case push_dummy_failure: | |
2885 continue; | |
2886 | |
2887 | |
2888 case jump_n: | |
2889 case pop_failure_jump: | |
2890 case maybe_pop_jump: | |
2891 case jump: | |
2892 case jump_past_alt: | |
2893 case dummy_failure_jump: | |
2894 EXTRACT_NUMBER_AND_INCR (j, p); | |
2895 p += j; | |
2896 if (j > 0) | |
2897 continue; | |
2898 | |
2899 /* Jump backward implies we just went through the body of a | |
2900 loop and matched nothing. Opcode jumped to should be | |
2901 `on_failure_jump' or `succeed_n'. Just treat it like an | |
2902 ordinary jump. For a * loop, it has pushed its failure | |
2903 point already; if so, discard that as redundant. */ | |
2904 if ((re_opcode_t) *p != on_failure_jump | |
2905 && (re_opcode_t) *p != succeed_n) | |
2906 continue; | |
2907 | |
2908 p++; | |
2909 EXTRACT_NUMBER_AND_INCR (j, p); | |
2910 p += j; | |
2911 | |
2912 /* If what's on the stack is where we are now, pop it. */ | |
2913 if (!FAIL_STACK_EMPTY () | |
2914 && fail_stack.stack[fail_stack.avail - 1] == p) | |
2915 fail_stack.avail--; | |
2916 | |
2917 continue; | |
2918 | |
2919 | |
2920 case on_failure_jump: | |
2921 case on_failure_keep_string_jump: | |
2922 handle_on_failure_jump: | |
2923 EXTRACT_NUMBER_AND_INCR (j, p); | |
2924 | |
2925 /* For some patterns, e.g., `(a?)?', `p+j' here points to the | |
2926 end of the pattern. We don't want to push such a point, | |
2927 since when we restore it above, entering the switch will | |
2928 increment `p' past the end of the pattern. We don't need | |
2929 to push such a point since we obviously won't find any more | |
2930 fastmap entries beyond `pend'. Such a pattern can match | |
2931 the null string, though. */ | |
2932 if (p + j < pend) | |
2933 { | |
2934 if (!PUSH_PATTERN_OP (p + j, fail_stack)) | |
2935 return -2; | |
2936 } | |
2937 else | |
2938 bufp->can_be_null = 1; | |
2939 | |
2940 if (succeed_n_p) | |
2941 { | |
2942 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ | |
2943 succeed_n_p = false; | |
2944 } | |
2945 | |
2946 continue; | |
2947 | |
2948 | |
2949 case succeed_n: | |
2950 /* Get to the number of times to succeed. */ | |
2951 p += 2; | |
2952 | |
2953 /* Increment p past the n for when k != 0. */ | |
2954 EXTRACT_NUMBER_AND_INCR (k, p); | |
2955 if (k == 0) | |
2956 { | |
2957 p -= 4; | |
2958 succeed_n_p = true; /* Spaghetti code alert. */ | |
2959 goto handle_on_failure_jump; | |
2960 } | |
2961 continue; | |
2962 | |
2963 | |
2964 case set_number_at: | |
2965 p += 4; | |
2966 continue; | |
2967 | |
2968 | |
2969 case start_memory: | |
2970 case stop_memory: | |
2971 p += 2; | |
2972 continue; | |
2973 | |
2974 | |
2975 default: | |
2976 abort (); /* We have listed all the cases. */ | |
2977 } /* switch *p++ */ | |
2978 | |
2979 /* Getting here means we have found the possible starting | |
2980 characters for one path of the pattern -- and that the empty | |
2981 string does not match. We need not follow this path further. | |
2982 Instead, look at the next alternative (remembered on the | |
2983 stack), or quit if no more. The test at the top of the loop | |
2984 does these things. */ | |
2985 path_can_be_null = false; | |
2986 p = pend; | |
2987 } /* while p */ | |
2988 | |
2989 /* Set `can_be_null' for the last path (also the first path, if the | |
2990 pattern is empty). */ | |
2991 bufp->can_be_null |= path_can_be_null; | |
2992 return 0; | |
2993 } /* re_compile_fastmap */ | |
2994 | |
2995 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
2996 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
2997 this memory for recording register information. STARTS and ENDS | |
2998 must be allocated using the malloc library routine, and must each | |
2999 be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
3000 | |
3001 If NUM_REGS == 0, then subsequent matches should allocate their own | |
3002 register data. | |
3003 | |
3004 Unless this function is called, the first search or match using | |
3005 PATTERN_BUFFER will allocate its own register data, without | |
3006 freeing the old data. */ | |
3007 | |
3008 void | |
3009 re_set_registers (bufp, regs, num_regs, starts, ends) | |
3010 struct re_pattern_buffer *bufp; | |
3011 struct re_registers *regs; | |
3012 unsigned num_regs; | |
3013 regoff_t *starts, *ends; | |
3014 { | |
3015 if (num_regs) | |
3016 { | |
3017 bufp->regs_allocated = REGS_REALLOCATE; | |
3018 regs->num_regs = num_regs; | |
3019 regs->start = starts; | |
3020 regs->end = ends; | |
3021 } | |
3022 else | |
3023 { | |
3024 bufp->regs_allocated = REGS_UNALLOCATED; | |
3025 regs->num_regs = 0; | |
5014
6062331f7430
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4918
diff
changeset
|
3026 regs->start = regs->end = (regoff_t *) 0; |
1155 | 3027 } |
3028 } | |
3029 | |
3030 /* Searching routines. */ | |
3031 | |
3032 /* Like re_search_2, below, but only one string is specified, and | |
3033 doesn't let you say where to stop matching. */ | |
3034 | |
3035 int | |
3036 re_search (bufp, string, size, startpos, range, regs) | |
3037 struct re_pattern_buffer *bufp; | |
3038 const char *string; | |
3039 int size, startpos, range; | |
3040 struct re_registers *regs; | |
3041 { | |
3042 return re_search_2 (bufp, NULL, 0, string, size, startpos, range, | |
3043 regs, size); | |
3044 } | |
3045 | |
3046 | |
3047 /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
3048 virtual concatenation of STRING1 and STRING2, starting first at index | |
3049 STARTPOS, then at STARTPOS + 1, and so on. | |
3050 | |
3051 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. | |
3052 | |
3053 RANGE is how far to scan while trying to match. RANGE = 0 means try | |
3054 only at STARTPOS; in general, the last start tried is STARTPOS + | |
3055 RANGE. | |
3056 | |
3057 In REGS, return the indices of the virtual concatenation of STRING1 | |
3058 and STRING2 that matched the entire BUFP->buffer and its contained | |
3059 subexpressions. | |
3060 | |
3061 Do not consider matching one past the index STOP in the virtual | |
3062 concatenation of STRING1 and STRING2. | |
3063 | |
3064 We return either the position in the strings at which the match was | |
3065 found, -1 if no match, or -2 if error (such as failure | |
3066 stack overflow). */ | |
3067 | |
3068 int | |
3069 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop) | |
3070 struct re_pattern_buffer *bufp; | |
3071 const char *string1, *string2; | |
3072 int size1, size2; | |
3073 int startpos; | |
3074 int range; | |
3075 struct re_registers *regs; | |
3076 int stop; | |
3077 { | |
3078 int val; | |
3079 register char *fastmap = bufp->fastmap; | |
3080 register char *translate = bufp->translate; | |
3081 int total_size = size1 + size2; | |
3082 int endpos = startpos + range; | |
3083 | |
3084 /* Check for out-of-range STARTPOS. */ | |
3085 if (startpos < 0 || startpos > total_size) | |
3086 return -1; | |
3087 | |
3088 /* Fix up RANGE if it might eventually take us outside | |
3089 the virtual concatenation of STRING1 and STRING2. */ | |
3090 if (endpos < -1) | |
3091 range = -1 - startpos; | |
3092 else if (endpos > total_size) | |
3093 range = total_size - startpos; | |
3094 | |
3095 /* If the search isn't to be a backwards one, don't waste time in a | |
1637 | 3096 search for a pattern that must be anchored. */ |
3097 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) | |
1155 | 3098 { |
3099 if (startpos > 0) | |
3100 return -1; | |
3101 else | |
3102 range = 1; | |
3103 } | |
3104 | |
1637 | 3105 /* Update the fastmap now if not correct already. */ |
3106 if (fastmap && !bufp->fastmap_accurate) | |
3107 if (re_compile_fastmap (bufp) == -2) | |
3108 return -2; | |
3109 | |
3110 /* Loop through the string, looking for a place to start matching. */ | |
1155 | 3111 for (;;) |
3112 { | |
3113 /* If a fastmap is supplied, skip quickly over characters that | |
3114 cannot be the start of a match. If the pattern can match the | |
3115 null string, however, we don't need to skip characters; we want | |
3116 the first null string. */ | |
3117 if (fastmap && startpos < total_size && !bufp->can_be_null) | |
3118 { | |
3119 if (range > 0) /* Searching forwards. */ | |
3120 { | |
3121 register const char *d; | |
3122 register int lim = 0; | |
3123 int irange = range; | |
3124 | |
3125 if (startpos < size1 && startpos + range >= size1) | |
3126 lim = range - (size1 - startpos); | |
3127 | |
3128 d = (startpos >= size1 ? string2 - size1 : string1) + startpos; | |
3129 | |
3130 /* Written out as an if-else to avoid testing `translate' | |
3131 inside the loop. */ | |
3132 if (translate) | |
3133 while (range > lim | |
2078 | 3134 && !fastmap[(unsigned char) |
3135 translate[(unsigned char) *d++]]) | |
1155 | 3136 range--; |
3137 else | |
3138 while (range > lim && !fastmap[(unsigned char) *d++]) | |
3139 range--; | |
3140 | |
3141 startpos += irange - range; | |
3142 } | |
3143 else /* Searching backwards. */ | |
3144 { | |
3145 register char c = (size1 == 0 || startpos >= size1 | |
3146 ? string2[startpos - size1] | |
3147 : string1[startpos]); | |
3148 | |
1637 | 3149 if (!fastmap[(unsigned char) TRANSLATE (c)]) |
1155 | 3150 goto advance; |
3151 } | |
3152 } | |
3153 | |
3154 /* If can't match the null string, and that's all we have left, fail. */ | |
3155 if (range >= 0 && startpos == total_size && fastmap | |
3156 && !bufp->can_be_null) | |
3157 return -1; | |
3158 | |
3159 val = re_match_2 (bufp, string1, size1, string2, size2, | |
3160 startpos, regs, stop); | |
3161 if (val >= 0) | |
3162 return startpos; | |
3163 | |
3164 if (val == -2) | |
3165 return -2; | |
3166 | |
3167 advance: | |
3168 if (!range) | |
3169 break; | |
3170 else if (range > 0) | |
3171 { | |
3172 range--; | |
3173 startpos++; | |
3174 } | |
3175 else | |
3176 { | |
3177 range++; | |
3178 startpos--; | |
3179 } | |
3180 } | |
3181 return -1; | |
3182 } /* re_search_2 */ | |
3183 | |
3184 /* Declarations and macros for re_match_2. */ | |
3185 | |
3186 static int bcmp_translate (); | |
3187 static boolean alt_match_null_string_p (), | |
3188 common_op_match_null_string_p (), | |
3189 group_match_null_string_p (); | |
3190 | |
3191 /* This converts PTR, a pointer into one of the search strings `string1' | |
3192 and `string2' into an offset from the beginning of that string. */ | |
4918
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3193 #define POINTER_TO_OFFSET(ptr) \ |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3194 (FIRST_STRING_P (ptr) \ |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3195 ? ((regoff_t) ((ptr) - string1)) \ |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3196 : ((regoff_t) ((ptr) - string2 + size1))) |
1155 | 3197 |
3198 /* Macros for dealing with the split strings in re_match_2. */ | |
3199 | |
3200 #define MATCHING_IN_FIRST_STRING (dend == end_match_1) | |
3201 | |
3202 /* Call before fetching a character with *d. This switches over to | |
3203 string2 if necessary. */ | |
3204 #define PREFETCH() \ | |
3205 while (d == dend) \ | |
3206 { \ | |
3207 /* End of string2 => fail. */ \ | |
3208 if (dend == end_match_2) \ | |
3209 goto fail; \ | |
3210 /* End of string1 => advance to string2. */ \ | |
3211 d = string2; \ | |
3212 dend = end_match_2; \ | |
3213 } | |
3214 | |
3215 | |
3216 /* Test if at very beginning or at very end of the virtual concatenation | |
3217 of `string1' and `string2'. If only one string, it's `string2'. */ | |
1637 | 3218 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
3219 #define AT_STRINGS_END(d) ((d) == end2) | |
1155 | 3220 |
3221 | |
3222 /* Test if D points to a character which is word-constituent. We have | |
3223 two special cases to check for: if past the end of string1, look at | |
3224 the first character in string2; and if before the beginning of | |
1637 | 3225 string2, look at the last character in string1. */ |
3226 #define WORDCHAR_P(d) \ | |
1155 | 3227 (SYNTAX ((d) == end1 ? *string2 \ |
1637 | 3228 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
3229 == Sword) | |
1155 | 3230 |
3231 /* Test if the character before D and the one at D differ with respect | |
3232 to being word-constituent. */ | |
3233 #define AT_WORD_BOUNDARY(d) \ | |
1637 | 3234 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ |
3235 || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) | |
1155 | 3236 |
3237 | |
3238 /* Free everything we malloc. */ | |
2952 | 3239 #ifdef MATCH_MAY_ALLOCATE |
1155 | 3240 #ifdef REGEX_MALLOC |
3241 #define FREE_VAR(var) if (var) free (var); var = NULL | |
3242 #define FREE_VARIABLES() \ | |
3243 do { \ | |
3244 FREE_VAR (fail_stack.stack); \ | |
3245 FREE_VAR (regstart); \ | |
3246 FREE_VAR (regend); \ | |
3247 FREE_VAR (old_regstart); \ | |
3248 FREE_VAR (old_regend); \ | |
3249 FREE_VAR (best_regstart); \ | |
3250 FREE_VAR (best_regend); \ | |
3251 FREE_VAR (reg_info); \ | |
3252 FREE_VAR (reg_dummy); \ | |
3253 FREE_VAR (reg_info_dummy); \ | |
3254 } while (0) | |
3255 #else /* not REGEX_MALLOC */ | |
3256 /* Some MIPS systems (at least) want this to free alloca'd storage. */ | |
3257 #define FREE_VARIABLES() alloca (0) | |
3258 #endif /* not REGEX_MALLOC */ | |
2952 | 3259 #else |
3260 #define FREE_VARIABLES() /* Do nothing! */ | |
3261 #endif /* not MATCH_MAY_ALLOCATE */ | |
1155 | 3262 |
3263 /* These values must meet several constraints. They must not be valid | |
3264 register values; since we have a limit of 255 registers (because | |
3265 we use only one byte in the pattern for the register number), we can | |
3266 use numbers larger than 255. They must differ by 1, because of | |
3267 NUM_FAILURE_ITEMS above. And the value for the lowest register must | |
3268 be larger than the value for the highest register, so we do not try | |
3269 to actually save any registers when none are active. */ | |
3270 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) | |
3271 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) | |
3272 | |
3273 /* Matching routines. */ | |
3274 | |
3275 #ifndef emacs /* Emacs never uses this. */ | |
3276 /* re_match is like re_match_2 except it takes only a single string. */ | |
3277 | |
3278 int | |
3279 re_match (bufp, string, size, pos, regs) | |
3280 struct re_pattern_buffer *bufp; | |
3281 const char *string; | |
3282 int size, pos; | |
3283 struct re_registers *regs; | |
3284 { | |
3285 return re_match_2 (bufp, NULL, 0, string, size, pos, regs, size); | |
3286 } | |
3287 #endif /* not emacs */ | |
3288 | |
3289 | |
3290 /* re_match_2 matches the compiled pattern in BUFP against the | |
3291 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
3292 and SIZE2, respectively). We start matching at POS, and stop | |
3293 matching at STOP. | |
3294 | |
3295 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we | |
3296 store offsets for the substring each group matched in REGS. See the | |
3297 documentation for exactly how many groups we fill. | |
3298 | |
3299 We return -1 if no match, -2 if an internal error (such as the | |
3300 failure stack overflowing). Otherwise, we return the length of the | |
3301 matched substring. */ | |
3302 | |
3303 int | |
3304 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
3305 struct re_pattern_buffer *bufp; | |
3306 const char *string1, *string2; | |
3307 int size1, size2; | |
3308 int pos; | |
3309 struct re_registers *regs; | |
3310 int stop; | |
3311 { | |
3312 /* General temporaries. */ | |
3313 int mcnt; | |
3314 unsigned char *p1; | |
3315 | |
3316 /* Just past the end of the corresponding string. */ | |
3317 const char *end1, *end2; | |
3318 | |
3319 /* Pointers into string1 and string2, just past the last characters in | |
3320 each to consider matching. */ | |
3321 const char *end_match_1, *end_match_2; | |
3322 | |
3323 /* Where we are in the data, and the end of the current string. */ | |
3324 const char *d, *dend; | |
3325 | |
3326 /* Where we are in the pattern, and the end of the pattern. */ | |
3327 unsigned char *p = bufp->buffer; | |
3328 register unsigned char *pend = p + bufp->used; | |
3329 | |
3330 /* We use this to map every character in the string. */ | |
3331 char *translate = bufp->translate; | |
3332 | |
3333 /* Failure point stack. Each place that can handle a failure further | |
3334 down the line pushes a failure point on this stack. It consists of | |
3335 restart, regend, and reg_info for all registers corresponding to | |
3336 the subexpressions we're currently inside, plus the number of such | |
3337 registers, and, finally, two char *'s. The first char * is where | |
3338 to resume scanning the pattern; the second one is where to resume | |
3339 scanning the strings. If the latter is zero, the failure point is | |
3340 a ``dummy''; if a failure happens and the failure point is a dummy, | |
3341 it gets discarded and the next next one is tried. */ | |
2952 | 3342 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
1155 | 3343 fail_stack_type fail_stack; |
2949 | 3344 #endif |
1155 | 3345 #ifdef DEBUG |
3346 static unsigned failure_id = 0; | |
1637 | 3347 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
1155 | 3348 #endif |
3349 | |
3350 /* We fill all the registers internally, independent of what we | |
3351 return, for use in backreferences. The number here includes | |
3352 an element for register zero. */ | |
3353 unsigned num_regs = bufp->re_nsub + 1; | |
3354 | |
3355 /* The currently active registers. */ | |
3356 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3357 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3358 | |
3359 /* Information on the contents of registers. These are pointers into | |
3360 the input strings; they record just what was matched (on this | |
3361 attempt) by a subexpression part of the pattern, that is, the | |
3362 regnum-th regstart pointer points to where in the pattern we began | |
3363 matching and the regnum-th regend points to right after where we | |
3364 stopped matching the regnum-th subexpression. (The zeroth register | |
3365 keeps track of what the whole pattern matches.) */ | |
2952 | 3366 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
1155 | 3367 const char **regstart, **regend; |
2949 | 3368 #endif |
1155 | 3369 |
3370 /* If a group that's operated upon by a repetition operator fails to | |
3371 match anything, then the register for its start will need to be | |
3372 restored because it will have been set to wherever in the string we | |
3373 are when we last see its open-group operator. Similarly for a | |
3374 register's end. */ | |
2952 | 3375 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
1155 | 3376 const char **old_regstart, **old_regend; |
2949 | 3377 #endif |
1155 | 3378 |
3379 /* The is_active field of reg_info helps us keep track of which (possibly | |
3380 nested) subexpressions we are currently in. The matched_something | |
3381 field of reg_info[reg_num] helps us tell whether or not we have | |
3382 matched any of the pattern so far this time through the reg_num-th | |
3383 subexpression. These two fields get reset each time through any | |
3384 loop their register is in. */ | |
2952 | 3385 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
1155 | 3386 register_info_type *reg_info; |
2949 | 3387 #endif |
1155 | 3388 |
3389 /* The following record the register info as found in the above | |
3390 variables when we find a match better than any we've seen before. | |
3391 This happens as we backtrack through the failure points, which in | |
3392 turn happens only if we have not yet matched the entire string. */ | |
3393 unsigned best_regs_set = false; | |
2952 | 3394 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
1155 | 3395 const char **best_regstart, **best_regend; |
2949 | 3396 #endif |
1155 | 3397 |
3398 /* Logically, this is `best_regend[0]'. But we don't want to have to | |
3399 allocate space for that if we're not allocating space for anything | |
3400 else (see below). Also, we never need info about register 0 for | |
3401 any of the other register vectors, and it seems rather a kludge to | |
3402 treat `best_regend' differently than the rest. So we keep track of | |
3403 the end of the best match so far in a separate variable. We | |
3404 initialize this to NULL so that when we backtrack the first time | |
3405 and need to test it, it's not garbage. */ | |
3406 const char *match_end = NULL; | |
3407 | |
3408 /* Used when we pop values we don't care about. */ | |
2952 | 3409 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
1155 | 3410 const char **reg_dummy; |
3411 register_info_type *reg_info_dummy; | |
2949 | 3412 #endif |
1155 | 3413 |
3414 #ifdef DEBUG | |
3415 /* Counts the total number of registers pushed. */ | |
3416 unsigned num_regs_pushed = 0; | |
3417 #endif | |
3418 | |
3419 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
3420 | |
3421 INIT_FAIL_STACK (); | |
3422 | |
2952 | 3423 #ifdef MATCH_MAY_ALLOCATE |
1155 | 3424 /* Do not bother to initialize all the register variables if there are |
3425 no groups in the pattern, as it takes a fair amount of time. If | |
3426 there are groups, we include space for register 0 (the whole | |
3427 pattern), even though we never use it, since it simplifies the | |
3428 array indexing. We should fix this. */ | |
3429 if (bufp->re_nsub) | |
3430 { | |
3431 regstart = REGEX_TALLOC (num_regs, const char *); | |
3432 regend = REGEX_TALLOC (num_regs, const char *); | |
3433 old_regstart = REGEX_TALLOC (num_regs, const char *); | |
3434 old_regend = REGEX_TALLOC (num_regs, const char *); | |
3435 best_regstart = REGEX_TALLOC (num_regs, const char *); | |
3436 best_regend = REGEX_TALLOC (num_regs, const char *); | |
3437 reg_info = REGEX_TALLOC (num_regs, register_info_type); | |
3438 reg_dummy = REGEX_TALLOC (num_regs, const char *); | |
3439 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type); | |
3440 | |
3441 if (!(regstart && regend && old_regstart && old_regend && reg_info | |
3442 && best_regstart && best_regend && reg_dummy && reg_info_dummy)) | |
3443 { | |
3444 FREE_VARIABLES (); | |
3445 return -2; | |
3446 } | |
3447 } | |
2949 | 3448 #if defined (REGEX_MALLOC) |
1155 | 3449 else |
3450 { | |
3451 /* We must initialize all our variables to NULL, so that | |
1637 | 3452 `FREE_VARIABLES' doesn't try to free them. */ |
1155 | 3453 regstart = regend = old_regstart = old_regend = best_regstart |
3454 = best_regend = reg_dummy = NULL; | |
3455 reg_info = reg_info_dummy = (register_info_type *) NULL; | |
3456 } | |
3457 #endif /* REGEX_MALLOC */ | |
2952 | 3458 #endif /* MATCH_MAY_ALLOCATE */ |
1155 | 3459 |
3460 /* The starting position is bogus. */ | |
3461 if (pos < 0 || pos > size1 + size2) | |
3462 { | |
3463 FREE_VARIABLES (); | |
3464 return -1; | |
3465 } | |
3466 | |
3467 /* Initialize subexpression text positions to -1 to mark ones that no | |
3468 start_memory/stop_memory has been seen for. Also initialize the | |
3469 register information struct. */ | |
3470 for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3471 { | |
3472 regstart[mcnt] = regend[mcnt] | |
3473 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; | |
3474 | |
3475 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; | |
3476 IS_ACTIVE (reg_info[mcnt]) = 0; | |
3477 MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
3478 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
3479 } | |
3480 | |
3481 /* We move `string1' into `string2' if the latter's empty -- but not if | |
3482 `string1' is null. */ | |
3483 if (size2 == 0 && string1 != NULL) | |
3484 { | |
3485 string2 = string1; | |
3486 size2 = size1; | |
3487 string1 = 0; | |
3488 size1 = 0; | |
3489 } | |
3490 end1 = string1 + size1; | |
3491 end2 = string2 + size2; | |
3492 | |
3493 /* Compute where to stop matching, within the two strings. */ | |
3494 if (stop <= size1) | |
3495 { | |
3496 end_match_1 = string1 + stop; | |
3497 end_match_2 = string2; | |
3498 } | |
3499 else | |
3500 { | |
3501 end_match_1 = end1; | |
3502 end_match_2 = string2 + stop - size1; | |
3503 } | |
3504 | |
3505 /* `p' scans through the pattern as `d' scans through the data. | |
3506 `dend' is the end of the input string that `d' points within. `d' | |
3507 is advanced into the following input string whenever necessary, but | |
3508 this happens before fetching; therefore, at the beginning of the | |
3509 loop, `d' can be pointing at the end of a string, but it cannot | |
3510 equal `string2'. */ | |
3511 if (size1 > 0 && pos <= size1) | |
3512 { | |
3513 d = string1 + pos; | |
3514 dend = end_match_1; | |
3515 } | |
3516 else | |
3517 { | |
3518 d = string2 + pos - size1; | |
3519 dend = end_match_2; | |
3520 } | |
3521 | |
3522 DEBUG_PRINT1 ("The compiled pattern is: "); | |
3523 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
3524 DEBUG_PRINT1 ("The string to match is: `"); | |
3525 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
3526 DEBUG_PRINT1 ("'\n"); | |
3527 | |
3528 /* This loops over pattern commands. It exits by returning from the | |
3529 function if the match is complete, or it drops through if the match | |
3530 fails at this starting point in the input data. */ | |
3531 for (;;) | |
3532 { | |
3533 DEBUG_PRINT2 ("\n0x%x: ", p); | |
3534 | |
3535 if (p == pend) | |
3536 { /* End of pattern means we might have succeeded. */ | |
1637 | 3537 DEBUG_PRINT1 ("end of pattern ... "); |
3538 | |
3539 /* If we haven't matched the entire string, and we want the | |
3540 longest match, try backtracking. */ | |
1155 | 3541 if (d != end_match_2) |
3542 { | |
3543 DEBUG_PRINT1 ("backtracking.\n"); | |
3544 | |
3545 if (!FAIL_STACK_EMPTY ()) | |
3546 { /* More failure points to try. */ | |
3547 boolean same_str_p = (FIRST_STRING_P (match_end) | |
3548 == MATCHING_IN_FIRST_STRING); | |
3549 | |
3550 /* If exceeds best match so far, save it. */ | |
3551 if (!best_regs_set | |
3552 || (same_str_p && d > match_end) | |
3553 || (!same_str_p && !MATCHING_IN_FIRST_STRING)) | |
3554 { | |
3555 best_regs_set = true; | |
3556 match_end = d; | |
3557 | |
3558 DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); | |
3559 | |
3560 for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3561 { | |
3562 best_regstart[mcnt] = regstart[mcnt]; | |
3563 best_regend[mcnt] = regend[mcnt]; | |
3564 } | |
3565 } | |
3566 goto fail; | |
3567 } | |
3568 | |
3569 /* If no failure points, don't restore garbage. */ | |
3570 else if (best_regs_set) | |
3571 { | |
3572 restore_best_regs: | |
3573 /* Restore best match. It may happen that `dend == | |
3574 end_match_1' while the restored d is in string2. | |
3575 For example, the pattern `x.*y.*z' against the | |
3576 strings `x-' and `y-z-', if the two strings are | |
3577 not consecutive in memory. */ | |
1637 | 3578 DEBUG_PRINT1 ("Restoring best registers.\n"); |
3579 | |
1155 | 3580 d = match_end; |
3581 dend = ((d >= string1 && d <= end1) | |
3582 ? end_match_1 : end_match_2); | |
3583 | |
3584 for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3585 { | |
3586 regstart[mcnt] = best_regstart[mcnt]; | |
3587 regend[mcnt] = best_regend[mcnt]; | |
3588 } | |
3589 } | |
3590 } /* d != end_match_2 */ | |
3591 | |
1637 | 3592 DEBUG_PRINT1 ("Accepting match.\n"); |
1155 | 3593 |
3594 /* If caller wants register contents data back, do it. */ | |
3595 if (regs && !bufp->no_sub) | |
3596 { | |
3597 /* Have the register data arrays been allocated? */ | |
3598 if (bufp->regs_allocated == REGS_UNALLOCATED) | |
3599 { /* No. So allocate them with malloc. We need one | |
3600 extra element beyond `num_regs' for the `-1' marker | |
3601 GNU code uses. */ | |
3602 regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
3603 regs->start = TALLOC (regs->num_regs, regoff_t); | |
3604 regs->end = TALLOC (regs->num_regs, regoff_t); | |
3605 if (regs->start == NULL || regs->end == NULL) | |
3606 return -2; | |
3607 bufp->regs_allocated = REGS_REALLOCATE; | |
3608 } | |
3609 else if (bufp->regs_allocated == REGS_REALLOCATE) | |
3610 { /* Yes. If we need more elements than were already | |
3611 allocated, reallocate them. If we need fewer, just | |
3612 leave it alone. */ | |
3613 if (regs->num_regs < num_regs + 1) | |
3614 { | |
3615 regs->num_regs = num_regs + 1; | |
3616 RETALLOC (regs->start, regs->num_regs, regoff_t); | |
3617 RETALLOC (regs->end, regs->num_regs, regoff_t); | |
3618 if (regs->start == NULL || regs->end == NULL) | |
3619 return -2; | |
3620 } | |
3621 } | |
3622 else | |
2465 | 3623 { |
3624 /* These braces fend off a "empty body in an else-statement" | |
3625 warning under GCC when assert expands to nothing. */ | |
3626 assert (bufp->regs_allocated == REGS_FIXED); | |
3627 } | |
1155 | 3628 |
3629 /* Convert the pointer data in `regstart' and `regend' to | |
3630 indices. Register zero has to be set differently, | |
3631 since we haven't kept track of any info for it. */ | |
3632 if (regs->num_regs > 0) | |
3633 { | |
3634 regs->start[0] = pos; | |
4918
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3635 regs->end[0] = (MATCHING_IN_FIRST_STRING |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3636 ? ((regoff_t) (d - string1)) |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3637 : ((regoff_t) (d - string2 + size1))); |
1155 | 3638 } |
3639 | |
3640 /* Go through the first `min (num_regs, regs->num_regs)' | |
3641 registers, since that is all we initialized. */ | |
3642 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++) | |
3643 { | |
3644 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) | |
3645 regs->start[mcnt] = regs->end[mcnt] = -1; | |
3646 else | |
3647 { | |
4918
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3648 regs->start[mcnt] |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3649 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]); |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3650 regs->end[mcnt] |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3651 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]); |
1155 | 3652 } |
3653 } | |
3654 | |
3655 /* If the regs structure we return has more elements than | |
3656 were in the pattern, set the extra elements to -1. If | |
3657 we (re)allocated the registers, this is the case, | |
3658 because we always allocate enough to have at least one | |
3659 -1 at the end. */ | |
3660 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++) | |
3661 regs->start[mcnt] = regs->end[mcnt] = -1; | |
3662 } /* regs && !bufp->no_sub */ | |
3663 | |
3664 FREE_VARIABLES (); | |
1637 | 3665 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", |
3666 nfailure_points_pushed, nfailure_points_popped, | |
3667 nfailure_points_pushed - nfailure_points_popped); | |
3668 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); | |
1155 | 3669 |
3670 mcnt = d - pos - (MATCHING_IN_FIRST_STRING | |
3671 ? string1 | |
3672 : string2 - size1); | |
3673 | |
3674 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); | |
3675 | |
3676 return mcnt; | |
3677 } | |
3678 | |
3679 /* Otherwise match next pattern command. */ | |
3680 #ifdef SWITCH_ENUM_BUG | |
3681 switch ((int) ((re_opcode_t) *p++)) | |
3682 #else | |
3683 switch ((re_opcode_t) *p++) | |
3684 #endif | |
3685 { | |
3686 /* Ignore these. Used to ignore the n of succeed_n's which | |
3687 currently have n == 0. */ | |
3688 case no_op: | |
3689 DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
3690 break; | |
3691 | |
3692 | |
3693 /* Match the next n pattern characters exactly. The following | |
3694 byte in the pattern defines n, and the n bytes after that | |
3695 are the characters to match. */ | |
3696 case exactn: | |
3697 mcnt = *p++; | |
3698 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); | |
3699 | |
3700 /* This is written out as an if-else so we don't waste time | |
3701 testing `translate' inside the loop. */ | |
3702 if (translate) | |
3703 { | |
3704 do | |
3705 { | |
3706 PREFETCH (); | |
3707 if (translate[(unsigned char) *d++] != (char) *p++) | |
3708 goto fail; | |
3709 } | |
3710 while (--mcnt); | |
3711 } | |
3712 else | |
3713 { | |
3714 do | |
3715 { | |
3716 PREFETCH (); | |
3717 if (*d++ != (char) *p++) goto fail; | |
3718 } | |
3719 while (--mcnt); | |
3720 } | |
3721 SET_REGS_MATCHED (); | |
3722 break; | |
3723 | |
3724 | |
3725 /* Match any character except possibly a newline or a null. */ | |
3726 case anychar: | |
3727 DEBUG_PRINT1 ("EXECUTING anychar.\n"); | |
3728 | |
3729 PREFETCH (); | |
3730 | |
3731 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n') | |
3732 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000')) | |
3733 goto fail; | |
3734 | |
3735 SET_REGS_MATCHED (); | |
3736 DEBUG_PRINT2 (" Matched `%d'.\n", *d); | |
3737 d++; | |
3738 break; | |
3739 | |
3740 | |
3741 case charset: | |
3742 case charset_not: | |
3743 { | |
3744 register unsigned char c; | |
3745 boolean not = (re_opcode_t) *(p - 1) == charset_not; | |
3746 | |
3747 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); | |
3748 | |
3749 PREFETCH (); | |
3750 c = TRANSLATE (*d); /* The character to match. */ | |
3751 | |
3752 /* Cast to `unsigned' instead of `unsigned char' in case the | |
3753 bit list is a full 32 bytes long. */ | |
3754 if (c < (unsigned) (*p * BYTEWIDTH) | |
3755 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
3756 not = !not; | |
3757 | |
3758 p += 1 + *p; | |
3759 | |
3760 if (!not) goto fail; | |
3761 | |
3762 SET_REGS_MATCHED (); | |
3763 d++; | |
3764 break; | |
3765 } | |
3766 | |
3767 | |
3768 /* The beginning of a group is represented by start_memory. | |
3769 The arguments are the register number in the next byte, and the | |
3770 number of groups inner to this one in the next. The text | |
3771 matched within the group is recorded (in the internal | |
3772 registers data structure) under the register number. */ | |
3773 case start_memory: | |
3774 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]); | |
3775 | |
3776 /* Find out if this group can match the empty string. */ | |
3777 p1 = p; /* To send to group_match_null_string_p. */ | |
3778 | |
3779 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) | |
3780 REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3781 = group_match_null_string_p (&p1, pend, reg_info); | |
3782 | |
3783 /* Save the position in the string where we were the last time | |
3784 we were at this open-group operator in case the group is | |
3785 operated upon by a repetition operator, e.g., with `(a*)*b' | |
3786 against `ab'; then we want to ignore where we are now in | |
3787 the string in case this attempt to match fails. */ | |
3788 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3789 ? REG_UNSET (regstart[*p]) ? d : regstart[*p] | |
3790 : regstart[*p]; | |
3791 DEBUG_PRINT2 (" old_regstart: %d\n", | |
3792 POINTER_TO_OFFSET (old_regstart[*p])); | |
3793 | |
3794 regstart[*p] = d; | |
3795 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); | |
3796 | |
3797 IS_ACTIVE (reg_info[*p]) = 1; | |
3798 MATCHED_SOMETHING (reg_info[*p]) = 0; | |
3799 | |
3800 /* This is the new highest active register. */ | |
3801 highest_active_reg = *p; | |
3802 | |
3803 /* If nothing was active before, this is the new lowest active | |
3804 register. */ | |
3805 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
3806 lowest_active_reg = *p; | |
3807 | |
3808 /* Move past the register number and inner group count. */ | |
3809 p += 2; | |
3810 break; | |
3811 | |
3812 | |
3813 /* The stop_memory opcode represents the end of a group. Its | |
3814 arguments are the same as start_memory's: the register | |
3815 number, and the number of inner groups. */ | |
3816 case stop_memory: | |
3817 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]); | |
3818 | |
3819 /* We need to save the string position the last time we were at | |
3820 this close-group operator in case the group is operated | |
3821 upon by a repetition operator, e.g., with `((a*)*(b*)*)*' | |
3822 against `aba'; then we want to ignore where we are now in | |
3823 the string in case this attempt to match fails. */ | |
3824 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3825 ? REG_UNSET (regend[*p]) ? d : regend[*p] | |
3826 : regend[*p]; | |
3827 DEBUG_PRINT2 (" old_regend: %d\n", | |
3828 POINTER_TO_OFFSET (old_regend[*p])); | |
3829 | |
3830 regend[*p] = d; | |
3831 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); | |
3832 | |
3833 /* This register isn't active anymore. */ | |
3834 IS_ACTIVE (reg_info[*p]) = 0; | |
3835 | |
3836 /* If this was the only register active, nothing is active | |
3837 anymore. */ | |
3838 if (lowest_active_reg == highest_active_reg) | |
3839 { | |
3840 lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3841 highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3842 } | |
3843 else | |
3844 { /* We must scan for the new highest active register, since | |
3845 it isn't necessarily one less than now: consider | |
3846 (a(b)c(d(e)f)g). When group 3 ends, after the f), the | |
3847 new highest active register is 1. */ | |
3848 unsigned char r = *p - 1; | |
3849 while (r > 0 && !IS_ACTIVE (reg_info[r])) | |
3850 r--; | |
3851 | |
3852 /* If we end up at register zero, that means that we saved | |
3853 the registers as the result of an `on_failure_jump', not | |
3854 a `start_memory', and we jumped to past the innermost | |
3855 `stop_memory'. For example, in ((.)*) we save | |
3856 registers 1 and 2 as a result of the *, but when we pop | |
3857 back to the second ), we are at the stop_memory 1. | |
3858 Thus, nothing is active. */ | |
3859 if (r == 0) | |
3860 { | |
3861 lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3862 highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3863 } | |
3864 else | |
3865 highest_active_reg = r; | |
3866 } | |
3867 | |
3868 /* If just failed to match something this time around with a | |
3869 group that's operated on by a repetition operator, try to | |
1637 | 3870 force exit from the ``loop'', and restore the register |
1155 | 3871 information for this group that we had before trying this |
3872 last match. */ | |
3873 if ((!MATCHED_SOMETHING (reg_info[*p]) | |
3874 || (re_opcode_t) p[-3] == start_memory) | |
3875 && (p + 2) < pend) | |
3876 { | |
3877 boolean is_a_jump_n = false; | |
3878 | |
3879 p1 = p + 2; | |
3880 mcnt = 0; | |
3881 switch ((re_opcode_t) *p1++) | |
3882 { | |
3883 case jump_n: | |
3884 is_a_jump_n = true; | |
3885 case pop_failure_jump: | |
3886 case maybe_pop_jump: | |
3887 case jump: | |
3888 case dummy_failure_jump: | |
3889 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
3890 if (is_a_jump_n) | |
3891 p1 += 2; | |
3892 break; | |
3893 | |
3894 default: | |
3895 /* do nothing */ ; | |
3896 } | |
3897 p1 += mcnt; | |
3898 | |
3899 /* If the next operation is a jump backwards in the pattern | |
3900 to an on_failure_jump right before the start_memory | |
3901 corresponding to this stop_memory, exit from the loop | |
3902 by forcing a failure after pushing on the stack the | |
3903 on_failure_jump's jump in the pattern, and d. */ | |
3904 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump | |
3905 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p) | |
3906 { | |
3907 /* If this group ever matched anything, then restore | |
3908 what its registers were before trying this last | |
3909 failed match, e.g., with `(a*)*b' against `ab' for | |
3910 regstart[1], and, e.g., with `((a*)*(b*)*)*' | |
3911 against `aba' for regend[3]. | |
3912 | |
3913 Also restore the registers for inner groups for, | |
3914 e.g., `((a*)(b*))*' against `aba' (register 3 would | |
3915 otherwise get trashed). */ | |
3916 | |
3917 if (EVER_MATCHED_SOMETHING (reg_info[*p])) | |
3918 { | |
3919 unsigned r; | |
3920 | |
3921 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; | |
3922 | |
3923 /* Restore this and inner groups' (if any) registers. */ | |
3924 for (r = *p; r < *p + *(p + 1); r++) | |
3925 { | |
3926 regstart[r] = old_regstart[r]; | |
3927 | |
3928 /* xx why this test? */ | |
3929 if ((int) old_regend[r] >= (int) regstart[r]) | |
3930 regend[r] = old_regend[r]; | |
3931 } | |
3932 } | |
3933 p1++; | |
3934 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
3935 PUSH_FAILURE_POINT (p1 + mcnt, d, -2); | |
3936 | |
3937 goto fail; | |
3938 } | |
3939 } | |
3940 | |
3941 /* Move past the register number and the inner group count. */ | |
3942 p += 2; | |
3943 break; | |
3944 | |
3945 | |
3946 /* \<digit> has been turned into a `duplicate' command which is | |
3947 followed by the numeric value of <digit> as the register number. */ | |
3948 case duplicate: | |
3949 { | |
3950 register const char *d2, *dend2; | |
3951 int regno = *p++; /* Get which register to match against. */ | |
3952 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); | |
3953 | |
3954 /* Can't back reference a group which we've never matched. */ | |
3955 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
3956 goto fail; | |
3957 | |
3958 /* Where in input to try to start matching. */ | |
3959 d2 = regstart[regno]; | |
3960 | |
3961 /* Where to stop matching; if both the place to start and | |
3962 the place to stop matching are in the same string, then | |
3963 set to the place to stop, otherwise, for now have to use | |
3964 the end of the first string. */ | |
3965 | |
3966 dend2 = ((FIRST_STRING_P (regstart[regno]) | |
3967 == FIRST_STRING_P (regend[regno])) | |
3968 ? regend[regno] : end_match_1); | |
3969 for (;;) | |
3970 { | |
3971 /* If necessary, advance to next segment in register | |
3972 contents. */ | |
3973 while (d2 == dend2) | |
3974 { | |
3975 if (dend2 == end_match_2) break; | |
3976 if (dend2 == regend[regno]) break; | |
3977 | |
3978 /* End of string1 => advance to string2. */ | |
3979 d2 = string2; | |
3980 dend2 = regend[regno]; | |
3981 } | |
3982 /* At end of register contents => success */ | |
3983 if (d2 == dend2) break; | |
3984 | |
3985 /* If necessary, advance to next segment in data. */ | |
3986 PREFETCH (); | |
3987 | |
3988 /* How many characters left in this segment to match. */ | |
3989 mcnt = dend - d; | |
3990 | |
3991 /* Want how many consecutive characters we can match in | |
3992 one shot, so, if necessary, adjust the count. */ | |
3993 if (mcnt > dend2 - d2) | |
3994 mcnt = dend2 - d2; | |
3995 | |
3996 /* Compare that many; failure if mismatch, else move | |
3997 past them. */ | |
3998 if (translate | |
3999 ? bcmp_translate (d, d2, mcnt, translate) | |
4000 : bcmp (d, d2, mcnt)) | |
4001 goto fail; | |
4002 d += mcnt, d2 += mcnt; | |
4003 } | |
4004 } | |
4005 break; | |
4006 | |
4007 | |
4008 /* begline matches the empty string at the beginning of the string | |
4009 (unless `not_bol' is set in `bufp'), and, if | |
4010 `newline_anchor' is set, after newlines. */ | |
4011 case begline: | |
4012 DEBUG_PRINT1 ("EXECUTING begline.\n"); | |
4013 | |
1637 | 4014 if (AT_STRINGS_BEG (d)) |
1155 | 4015 { |
4016 if (!bufp->not_bol) break; | |
4017 } | |
4018 else if (d[-1] == '\n' && bufp->newline_anchor) | |
4019 { | |
4020 break; | |
4021 } | |
4022 /* In all other cases, we fail. */ | |
4023 goto fail; | |
4024 | |
4025 | |
4026 /* endline is the dual of begline. */ | |
4027 case endline: | |
4028 DEBUG_PRINT1 ("EXECUTING endline.\n"); | |
4029 | |
1637 | 4030 if (AT_STRINGS_END (d)) |
1155 | 4031 { |
4032 if (!bufp->not_eol) break; | |
4033 } | |
4034 | |
4035 /* We have to ``prefetch'' the next character. */ | |
4036 else if ((d == end1 ? *string2 : *d) == '\n' | |
4037 && bufp->newline_anchor) | |
4038 { | |
4039 break; | |
4040 } | |
4041 goto fail; | |
4042 | |
4043 | |
4044 /* Match at the very beginning of the data. */ | |
4045 case begbuf: | |
4046 DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
1637 | 4047 if (AT_STRINGS_BEG (d)) |
1155 | 4048 break; |
4049 goto fail; | |
4050 | |
4051 | |
4052 /* Match at the very end of the data. */ | |
4053 case endbuf: | |
4054 DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
1637 | 4055 if (AT_STRINGS_END (d)) |
1155 | 4056 break; |
4057 goto fail; | |
4058 | |
4059 | |
4060 /* on_failure_keep_string_jump is used to optimize `.*\n'. It | |
4061 pushes NULL as the value for the string on the stack. Then | |
4062 `pop_failure_point' will keep the current value for the | |
4063 string, instead of restoring it. To see why, consider | |
4064 matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
4065 then the . fails against the \n. But the next thing we want | |
4066 to do is match the \n against the \n; if we restored the | |
4067 string value, we would be back at the foo. | |
4068 | |
4069 Because this is used only in specific cases, we don't need to | |
4070 check all the things that `on_failure_jump' does, to make | |
4071 sure the right things get saved on the stack. Hence we don't | |
4072 share its code. The only reason to push anything on the | |
4073 stack at all is that otherwise we would have to change | |
4074 `anychar's code to do something besides goto fail in this | |
4075 case; that seems worse than this. */ | |
4076 case on_failure_keep_string_jump: | |
4077 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); | |
4078 | |
4079 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4080 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); | |
4081 | |
4082 PUSH_FAILURE_POINT (p + mcnt, NULL, -2); | |
4083 break; | |
4084 | |
4085 | |
4086 /* Uses of on_failure_jump: | |
4087 | |
4088 Each alternative starts with an on_failure_jump that points | |
4089 to the beginning of the next alternative. Each alternative | |
4090 except the last ends with a jump that in effect jumps past | |
4091 the rest of the alternatives. (They really jump to the | |
4092 ending jump of the following alternative, because tensioning | |
4093 these jumps is a hassle.) | |
4094 | |
4095 Repeats start with an on_failure_jump that points past both | |
4096 the repetition text and either the following jump or | |
4097 pop_failure_jump back to this on_failure_jump. */ | |
4098 case on_failure_jump: | |
4099 on_failure: | |
4100 DEBUG_PRINT1 ("EXECUTING on_failure_jump"); | |
4101 | |
4102 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4103 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); | |
4104 | |
4105 /* If this on_failure_jump comes right before a group (i.e., | |
4106 the original * applied to a group), save the information | |
4107 for that group and all inner ones, so that if we fail back | |
4108 to this point, the group's information will be correct. | |
1637 | 4109 For example, in \(a*\)*\1, we need the preceding group, |
1155 | 4110 and in \(\(a*\)b*\)\2, we need the inner group. */ |
4111 | |
4112 /* We can't use `p' to check ahead because we push | |
4113 a failure point to `p + mcnt' after we do this. */ | |
4114 p1 = p; | |
4115 | |
4116 /* We need to skip no_op's before we look for the | |
4117 start_memory in case this on_failure_jump is happening as | |
4118 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 | |
4119 against aba. */ | |
4120 while (p1 < pend && (re_opcode_t) *p1 == no_op) | |
4121 p1++; | |
4122 | |
4123 if (p1 < pend && (re_opcode_t) *p1 == start_memory) | |
4124 { | |
4125 /* We have a new highest active register now. This will | |
4126 get reset at the start_memory we are about to get to, | |
4127 but we will have saved all the registers relevant to | |
4128 this repetition op, as described above. */ | |
4129 highest_active_reg = *(p1 + 1) + *(p1 + 2); | |
4130 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
4131 lowest_active_reg = *(p1 + 1); | |
4132 } | |
4133 | |
4134 DEBUG_PRINT1 (":\n"); | |
4135 PUSH_FAILURE_POINT (p + mcnt, d, -2); | |
4136 break; | |
4137 | |
4138 | |
1637 | 4139 /* A smart repeat ends with `maybe_pop_jump'. |
4140 We change it to either `pop_failure_jump' or `jump'. */ | |
1155 | 4141 case maybe_pop_jump: |
4142 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4143 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); | |
4144 { | |
4145 register unsigned char *p2 = p; | |
4146 | |
4147 /* Compare the beginning of the repeat with what in the | |
4148 pattern follows its end. If we can establish that there | |
4149 is nothing that they would both match, i.e., that we | |
4150 would have to backtrack because of (as in, e.g., `a*a') | |
4151 then we can change to pop_failure_jump, because we'll | |
4152 never have to backtrack. | |
4153 | |
4154 This is not true in the case of alternatives: in | |
4155 `(a|ab)*' we do need to backtrack to the `ab' alternative | |
4156 (e.g., if the string was `ab'). But instead of trying to | |
4157 detect that here, the alternative has put on a dummy | |
4158 failure point which is what we will end up popping. */ | |
4159 | |
3541
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4160 /* Skip over open/close-group commands. |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4161 If what follows this loop is a ...+ construct, |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4162 look at what begins its body, since we will have to |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4163 match at least one of that. */ |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4164 while (1) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4165 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4166 if (p2 + 2 < pend |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4167 && ((re_opcode_t) *p2 == stop_memory |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4168 || (re_opcode_t) *p2 == start_memory)) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4169 p2 += 3; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4170 else if (p2 + 6 < pend |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4171 && (re_opcode_t) *p2 == dummy_failure_jump) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4172 p2 += 6; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4173 else |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4174 break; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4175 } |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4176 |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4177 p1 = p + mcnt; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4178 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4179 to the `maybe_finalize_jump' of this case. Examine what |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4180 follows. */ |
1155 | 4181 |
4182 /* If we're at the end of the pattern, we can change. */ | |
4183 if (p2 == pend) | |
1669 | 4184 { |
4185 /* Consider what happens when matching ":\(.*\)" | |
4186 against ":/". I don't really understand this code | |
4187 yet. */ | |
1155 | 4188 p[-3] = (unsigned char) pop_failure_jump; |
1669 | 4189 DEBUG_PRINT1 |
4190 (" End of pattern: change to `pop_failure_jump'.\n"); | |
1155 | 4191 } |
4192 | |
4193 else if ((re_opcode_t) *p2 == exactn | |
4194 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) | |
4195 { | |
4196 register unsigned char c | |
4197 = *p2 == (unsigned char) endline ? '\n' : p2[2]; | |
3541
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4198 |
1155 | 4199 if ((re_opcode_t) p1[3] == exactn && p1[5] != c) |
1637 | 4200 { |
4201 p[-3] = (unsigned char) pop_failure_jump; | |
4202 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", | |
4203 c, p1[5]); | |
4204 } | |
4205 | |
1155 | 4206 else if ((re_opcode_t) p1[3] == charset |
4207 || (re_opcode_t) p1[3] == charset_not) | |
4208 { | |
4209 int not = (re_opcode_t) p1[3] == charset_not; | |
4210 | |
4211 if (c < (unsigned char) (p1[4] * BYTEWIDTH) | |
4212 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
4213 not = !not; | |
4214 | |
4215 /* `not' is equal to 1 if c would match, which means | |
4216 that we can't change to pop_failure_jump. */ | |
4217 if (!not) | |
4218 { | |
4219 p[-3] = (unsigned char) pop_failure_jump; | |
1637 | 4220 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
1155 | 4221 } |
4222 } | |
4223 } | |
3541
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4224 else if ((re_opcode_t) *p2 == charset) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4225 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4226 register unsigned char c |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4227 = *p2 == (unsigned char) endline ? '\n' : p2[2]; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4228 |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4229 if ((re_opcode_t) p1[3] == exactn |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4230 && ! (p2[1] * BYTEWIDTH > p1[4] |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4231 && (p2[1 + p1[4] / BYTEWIDTH] |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4232 & (1 << (p1[4] % BYTEWIDTH))))) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4233 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4234 p[-3] = (unsigned char) pop_failure_jump; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4235 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4236 c, p1[5]); |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4237 } |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4238 |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4239 else if ((re_opcode_t) p1[3] == charset_not) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4240 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4241 int idx; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4242 /* We win if the charset_not inside the loop |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4243 lists every character listed in the charset after. */ |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4244 for (idx = 0; idx < p2[1]; idx++) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4245 if (! (p2[2 + idx] == 0 |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4246 || (idx < p1[4] |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4247 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4248 break; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4249 |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4250 if (idx == p2[1]) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4251 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4252 p[-3] = (unsigned char) pop_failure_jump; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4253 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4254 } |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4255 } |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4256 else if ((re_opcode_t) p1[3] == charset) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4257 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4258 int idx; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4259 /* We win if the charset inside the loop |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4260 has no overlap with the one after the loop. */ |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4261 for (idx = 0; idx < p2[1] && idx < p1[4]; idx++) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4262 if ((p2[2 + idx] & p1[5 + idx]) != 0) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4263 break; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4264 |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4265 if (idx == p2[1] || idx == p1[4]) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4266 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4267 p[-3] = (unsigned char) pop_failure_jump; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4268 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4269 } |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4270 } |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4271 } |
1155 | 4272 } |
4273 p -= 2; /* Point at relative address again. */ | |
4274 if ((re_opcode_t) p[-1] != pop_failure_jump) | |
4275 { | |
4276 p[-1] = (unsigned char) jump; | |
1637 | 4277 DEBUG_PRINT1 (" Match => jump.\n"); |
1155 | 4278 goto unconditional_jump; |
4279 } | |
4280 /* Note fall through. */ | |
4281 | |
4282 | |
4283 /* The end of a simple repeat has a pop_failure_jump back to | |
4284 its matching on_failure_jump, where the latter will push a | |
4285 failure point. The pop_failure_jump takes off failure | |
4286 points put on by this pop_failure_jump's matching | |
4287 on_failure_jump; we got through the pattern to here from the | |
4288 matching on_failure_jump, so didn't fail. */ | |
4289 case pop_failure_jump: | |
4290 { | |
4291 /* We need to pass separate storage for the lowest and | |
4292 highest registers, even though we don't care about the | |
4293 actual values. Otherwise, we will restore only one | |
4294 register from the stack, since lowest will == highest in | |
4295 `pop_failure_point'. */ | |
4296 unsigned dummy_low_reg, dummy_high_reg; | |
4297 unsigned char *pdummy; | |
4298 const char *sdummy; | |
4299 | |
4300 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); | |
4301 POP_FAILURE_POINT (sdummy, pdummy, | |
4302 dummy_low_reg, dummy_high_reg, | |
4303 reg_dummy, reg_dummy, reg_info_dummy); | |
4304 } | |
4305 /* Note fall through. */ | |
4306 | |
4307 | |
4308 /* Unconditionally jump (without popping any failure points). */ | |
4309 case jump: | |
4310 unconditional_jump: | |
4311 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ | |
4312 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); | |
4313 p += mcnt; /* Do the jump. */ | |
4314 DEBUG_PRINT2 ("(to 0x%x).\n", p); | |
4315 break; | |
4316 | |
4317 | |
4318 /* We need this opcode so we can detect where alternatives end | |
4319 in `group_match_null_string_p' et al. */ | |
4320 case jump_past_alt: | |
4321 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); | |
4322 goto unconditional_jump; | |
4323 | |
4324 | |
4325 /* Normally, the on_failure_jump pushes a failure point, which | |
4326 then gets popped at pop_failure_jump. We will end up at | |
4327 pop_failure_jump, also, and with a pattern of, say, `a+', we | |
4328 are skipping over the on_failure_jump, so we have to push | |
4329 something meaningless for pop_failure_jump to pop. */ | |
4330 case dummy_failure_jump: | |
4331 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); | |
4332 /* It doesn't matter what we push for the string here. What | |
4333 the code at `fail' tests is the value for the pattern. */ | |
4334 PUSH_FAILURE_POINT (0, 0, -2); | |
4335 goto unconditional_jump; | |
4336 | |
4337 | |
4338 /* At the end of an alternative, we need to push a dummy failure | |
1637 | 4339 point in case we are followed by a `pop_failure_jump', because |
1155 | 4340 we don't want the failure point for the alternative to be |
4341 popped. For example, matching `(a|ab)*' against `aab' | |
4342 requires that we match the `ab' alternative. */ | |
4343 case push_dummy_failure: | |
4344 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); | |
4345 /* See comments just above at `dummy_failure_jump' about the | |
4346 two zeroes. */ | |
4347 PUSH_FAILURE_POINT (0, 0, -2); | |
4348 break; | |
4349 | |
4350 /* Have to succeed matching what follows at least n times. | |
4351 After that, handle like `on_failure_jump'. */ | |
4352 case succeed_n: | |
4353 EXTRACT_NUMBER (mcnt, p + 2); | |
4354 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); | |
4355 | |
4356 assert (mcnt >= 0); | |
4357 /* Originally, this is how many times we HAVE to succeed. */ | |
4358 if (mcnt > 0) | |
4359 { | |
4360 mcnt--; | |
4361 p += 2; | |
4362 STORE_NUMBER_AND_INCR (p, mcnt); | |
4363 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt); | |
4364 } | |
4365 else if (mcnt == 0) | |
4366 { | |
4367 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2); | |
4368 p[2] = (unsigned char) no_op; | |
4369 p[3] = (unsigned char) no_op; | |
4370 goto on_failure; | |
4371 } | |
4372 break; | |
4373 | |
4374 case jump_n: | |
4375 EXTRACT_NUMBER (mcnt, p + 2); | |
4376 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); | |
4377 | |
4378 /* Originally, this is how many times we CAN jump. */ | |
4379 if (mcnt) | |
4380 { | |
4381 mcnt--; | |
4382 STORE_NUMBER (p + 2, mcnt); | |
4383 goto unconditional_jump; | |
4384 } | |
4385 /* If don't have to jump any more, skip over the rest of command. */ | |
4386 else | |
4387 p += 4; | |
4388 break; | |
4389 | |
4390 case set_number_at: | |
4391 { | |
4392 DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); | |
4393 | |
4394 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4395 p1 = p + mcnt; | |
4396 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4397 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); | |
4398 STORE_NUMBER (p1, mcnt); | |
4399 break; | |
4400 } | |
4401 | |
4402 case wordbound: | |
4403 DEBUG_PRINT1 ("EXECUTING wordbound.\n"); | |
4404 if (AT_WORD_BOUNDARY (d)) | |
4405 break; | |
4406 goto fail; | |
4407 | |
4408 case notwordbound: | |
4409 DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); | |
4410 if (AT_WORD_BOUNDARY (d)) | |
4411 goto fail; | |
4412 break; | |
4413 | |
4414 case wordbeg: | |
4415 DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); | |
1637 | 4416 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) |
1155 | 4417 break; |
4418 goto fail; | |
4419 | |
4420 case wordend: | |
4421 DEBUG_PRINT1 ("EXECUTING wordend.\n"); | |
1637 | 4422 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) |
4423 && (!WORDCHAR_P (d) || AT_STRINGS_END (d))) | |
1155 | 4424 break; |
4425 goto fail; | |
4426 | |
4427 #ifdef emacs | |
4428 #ifdef emacs19 | |
4429 case before_dot: | |
4430 DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
4431 if (PTR_CHAR_POS ((unsigned char *) d) >= point) | |
4432 goto fail; | |
4433 break; | |
4434 | |
4435 case at_dot: | |
4436 DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
4437 if (PTR_CHAR_POS ((unsigned char *) d) != point) | |
4438 goto fail; | |
4439 break; | |
4440 | |
4441 case after_dot: | |
4442 DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
4443 if (PTR_CHAR_POS ((unsigned char *) d) <= point) | |
4444 goto fail; | |
4445 break; | |
4446 #else /* not emacs19 */ | |
4447 case at_dot: | |
4448 DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
4449 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point) | |
4450 goto fail; | |
4451 break; | |
4452 #endif /* not emacs19 */ | |
4453 | |
4454 case syntaxspec: | |
4455 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); | |
4456 mcnt = *p++; | |
4457 goto matchsyntax; | |
4458 | |
4459 case wordchar: | |
1637 | 4460 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n"); |
1155 | 4461 mcnt = (int) Sword; |
4462 matchsyntax: | |
4463 PREFETCH (); | |
1637 | 4464 if (SYNTAX (*d++) != (enum syntaxcode) mcnt) |
4465 goto fail; | |
1155 | 4466 SET_REGS_MATCHED (); |
4467 break; | |
4468 | |
4469 case notsyntaxspec: | |
4470 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); | |
4471 mcnt = *p++; | |
4472 goto matchnotsyntax; | |
4473 | |
4474 case notwordchar: | |
1637 | 4475 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n"); |
1155 | 4476 mcnt = (int) Sword; |
1637 | 4477 matchnotsyntax: |
1155 | 4478 PREFETCH (); |
1637 | 4479 if (SYNTAX (*d++) == (enum syntaxcode) mcnt) |
4480 goto fail; | |
1155 | 4481 SET_REGS_MATCHED (); |
4482 break; | |
4483 | |
4484 #else /* not emacs */ | |
4485 case wordchar: | |
4486 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); | |
4487 PREFETCH (); | |
1637 | 4488 if (!WORDCHAR_P (d)) |
1155 | 4489 goto fail; |
4490 SET_REGS_MATCHED (); | |
1637 | 4491 d++; |
1155 | 4492 break; |
4493 | |
4494 case notwordchar: | |
4495 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); | |
4496 PREFETCH (); | |
1637 | 4497 if (WORDCHAR_P (d)) |
1155 | 4498 goto fail; |
4499 SET_REGS_MATCHED (); | |
1637 | 4500 d++; |
1155 | 4501 break; |
4502 #endif /* not emacs */ | |
4503 | |
4504 default: | |
4505 abort (); | |
4506 } | |
4507 continue; /* Successfully executed one pattern command; keep going. */ | |
4508 | |
4509 | |
4510 /* We goto here if a matching operation fails. */ | |
4511 fail: | |
4512 if (!FAIL_STACK_EMPTY ()) | |
4513 { /* A restart point is known. Restore to that state. */ | |
4514 DEBUG_PRINT1 ("\nFAIL:\n"); | |
4515 POP_FAILURE_POINT (d, p, | |
4516 lowest_active_reg, highest_active_reg, | |
4517 regstart, regend, reg_info); | |
4518 | |
4519 /* If this failure point is a dummy, try the next one. */ | |
4520 if (!p) | |
4521 goto fail; | |
4522 | |
4523 /* If we failed to the end of the pattern, don't examine *p. */ | |
4524 assert (p <= pend); | |
4525 if (p < pend) | |
4526 { | |
4527 boolean is_a_jump_n = false; | |
4528 | |
4529 /* If failed to a backwards jump that's part of a repetition | |
4530 loop, need to pop this failure point and use the next one. */ | |
4531 switch ((re_opcode_t) *p) | |
4532 { | |
4533 case jump_n: | |
4534 is_a_jump_n = true; | |
4535 case maybe_pop_jump: | |
4536 case pop_failure_jump: | |
4537 case jump: | |
4538 p1 = p + 1; | |
4539 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4540 p1 += mcnt; | |
4541 | |
4542 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) | |
4543 || (!is_a_jump_n | |
4544 && (re_opcode_t) *p1 == on_failure_jump)) | |
4545 goto fail; | |
4546 break; | |
4547 default: | |
4548 /* do nothing */ ; | |
4549 } | |
4550 } | |
4551 | |
4552 if (d >= string1 && d <= end1) | |
4553 dend = end_match_1; | |
4554 } | |
4555 else | |
4556 break; /* Matching at this starting point really fails. */ | |
4557 } /* for (;;) */ | |
4558 | |
4559 if (best_regs_set) | |
4560 goto restore_best_regs; | |
4561 | |
4562 FREE_VARIABLES (); | |
4563 | |
4564 return -1; /* Failure to match. */ | |
4565 } /* re_match_2 */ | |
4566 | |
4567 /* Subroutine definitions for re_match_2. */ | |
4568 | |
4569 | |
4570 /* We are passed P pointing to a register number after a start_memory. | |
4571 | |
4572 Return true if the pattern up to the corresponding stop_memory can | |
4573 match the empty string, and false otherwise. | |
4574 | |
4575 If we find the matching stop_memory, sets P to point to one past its number. | |
4576 Otherwise, sets P to an undefined byte less than or equal to END. | |
4577 | |
4578 We don't handle duplicates properly (yet). */ | |
4579 | |
4580 static boolean | |
4581 group_match_null_string_p (p, end, reg_info) | |
4582 unsigned char **p, *end; | |
4583 register_info_type *reg_info; | |
4584 { | |
4585 int mcnt; | |
4586 /* Point to after the args to the start_memory. */ | |
4587 unsigned char *p1 = *p + 2; | |
4588 | |
4589 while (p1 < end) | |
4590 { | |
4591 /* Skip over opcodes that can match nothing, and return true or | |
4592 false, as appropriate, when we get to one that can't, or to the | |
4593 matching stop_memory. */ | |
4594 | |
4595 switch ((re_opcode_t) *p1) | |
4596 { | |
4597 /* Could be either a loop or a series of alternatives. */ | |
4598 case on_failure_jump: | |
4599 p1++; | |
4600 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4601 | |
4602 /* If the next operation is not a jump backwards in the | |
4603 pattern. */ | |
4604 | |
4605 if (mcnt >= 0) | |
4606 { | |
4607 /* Go through the on_failure_jumps of the alternatives, | |
4608 seeing if any of the alternatives cannot match nothing. | |
4609 The last alternative starts with only a jump, | |
4610 whereas the rest start with on_failure_jump and end | |
4611 with a jump, e.g., here is the pattern for `a|b|c': | |
4612 | |
4613 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 | |
4614 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 | |
4615 /exactn/1/c | |
4616 | |
4617 So, we have to first go through the first (n-1) | |
4618 alternatives and then deal with the last one separately. */ | |
4619 | |
4620 | |
4621 /* Deal with the first (n-1) alternatives, which start | |
4622 with an on_failure_jump (see above) that jumps to right | |
4623 past a jump_past_alt. */ | |
4624 | |
4625 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt) | |
4626 { | |
4627 /* `mcnt' holds how many bytes long the alternative | |
4628 is, including the ending `jump_past_alt' and | |
4629 its number. */ | |
4630 | |
4631 if (!alt_match_null_string_p (p1, p1 + mcnt - 3, | |
4632 reg_info)) | |
4633 return false; | |
4634 | |
4635 /* Move to right after this alternative, including the | |
4636 jump_past_alt. */ | |
4637 p1 += mcnt; | |
4638 | |
4639 /* Break if it's the beginning of an n-th alternative | |
4640 that doesn't begin with an on_failure_jump. */ | |
4641 if ((re_opcode_t) *p1 != on_failure_jump) | |
4642 break; | |
4643 | |
4644 /* Still have to check that it's not an n-th | |
4645 alternative that starts with an on_failure_jump. */ | |
4646 p1++; | |
4647 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4648 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt) | |
4649 { | |
4650 /* Get to the beginning of the n-th alternative. */ | |
4651 p1 -= 3; | |
4652 break; | |
4653 } | |
4654 } | |
4655 | |
4656 /* Deal with the last alternative: go back and get number | |
4657 of the `jump_past_alt' just before it. `mcnt' contains | |
4658 the length of the alternative. */ | |
4659 EXTRACT_NUMBER (mcnt, p1 - 2); | |
4660 | |
4661 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info)) | |
4662 return false; | |
4663 | |
4664 p1 += mcnt; /* Get past the n-th alternative. */ | |
4665 } /* if mcnt > 0 */ | |
4666 break; | |
4667 | |
4668 | |
4669 case stop_memory: | |
4670 assert (p1[1] == **p); | |
4671 *p = p1 + 2; | |
4672 return true; | |
4673 | |
4674 | |
4675 default: | |
4676 if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
4677 return false; | |
4678 } | |
4679 } /* while p1 < end */ | |
4680 | |
4681 return false; | |
4682 } /* group_match_null_string_p */ | |
4683 | |
4684 | |
4685 /* Similar to group_match_null_string_p, but doesn't deal with alternatives: | |
4686 It expects P to be the first byte of a single alternative and END one | |
4687 byte past the last. The alternative can contain groups. */ | |
4688 | |
4689 static boolean | |
4690 alt_match_null_string_p (p, end, reg_info) | |
4691 unsigned char *p, *end; | |
4692 register_info_type *reg_info; | |
4693 { | |
4694 int mcnt; | |
4695 unsigned char *p1 = p; | |
4696 | |
4697 while (p1 < end) | |
4698 { | |
4699 /* Skip over opcodes that can match nothing, and break when we get | |
4700 to one that can't. */ | |
4701 | |
4702 switch ((re_opcode_t) *p1) | |
4703 { | |
4704 /* It's a loop. */ | |
4705 case on_failure_jump: | |
4706 p1++; | |
4707 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4708 p1 += mcnt; | |
4709 break; | |
4710 | |
4711 default: | |
4712 if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
4713 return false; | |
4714 } | |
4715 } /* while p1 < end */ | |
4716 | |
4717 return true; | |
4718 } /* alt_match_null_string_p */ | |
4719 | |
4720 | |
4721 /* Deals with the ops common to group_match_null_string_p and | |
4722 alt_match_null_string_p. | |
4723 | |
4724 Sets P to one after the op and its arguments, if any. */ | |
4725 | |
4726 static boolean | |
4727 common_op_match_null_string_p (p, end, reg_info) | |
4728 unsigned char **p, *end; | |
4729 register_info_type *reg_info; | |
4730 { | |
4731 int mcnt; | |
4732 boolean ret; | |
4733 int reg_no; | |
4734 unsigned char *p1 = *p; | |
4735 | |
4736 switch ((re_opcode_t) *p1++) | |
4737 { | |
4738 case no_op: | |
4739 case begline: | |
4740 case endline: | |
4741 case begbuf: | |
4742 case endbuf: | |
4743 case wordbeg: | |
4744 case wordend: | |
4745 case wordbound: | |
4746 case notwordbound: | |
4747 #ifdef emacs | |
4748 case before_dot: | |
4749 case at_dot: | |
4750 case after_dot: | |
4751 #endif | |
4752 break; | |
4753 | |
4754 case start_memory: | |
4755 reg_no = *p1; | |
4756 assert (reg_no > 0 && reg_no <= MAX_REGNUM); | |
4757 ret = group_match_null_string_p (&p1, end, reg_info); | |
4758 | |
4759 /* Have to set this here in case we're checking a group which | |
4760 contains a group and a back reference to it. */ | |
4761 | |
4762 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) | |
4763 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; | |
4764 | |
4765 if (!ret) | |
4766 return false; | |
4767 break; | |
4768 | |
4769 /* If this is an optimized succeed_n for zero times, make the jump. */ | |
4770 case jump: | |
4771 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4772 if (mcnt >= 0) | |
4773 p1 += mcnt; | |
4774 else | |
4775 return false; | |
4776 break; | |
4777 | |
4778 case succeed_n: | |
4779 /* Get to the number of times to succeed. */ | |
4780 p1 += 2; | |
4781 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4782 | |
4783 if (mcnt == 0) | |
4784 { | |
4785 p1 -= 4; | |
4786 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4787 p1 += mcnt; | |
4788 } | |
4789 else | |
4790 return false; | |
4791 break; | |
4792 | |
4793 case duplicate: | |
4794 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) | |
4795 return false; | |
4796 break; | |
4797 | |
4798 case set_number_at: | |
4799 p1 += 4; | |
4800 | |
4801 default: | |
4802 /* All other opcodes mean we cannot match the empty string. */ | |
4803 return false; | |
4804 } | |
4805 | |
4806 *p = p1; | |
4807 return true; | |
4808 } /* common_op_match_null_string_p */ | |
4809 | |
4810 | |
4811 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN | |
4812 bytes; nonzero otherwise. */ | |
4813 | |
4814 static int | |
4815 bcmp_translate (s1, s2, len, translate) | |
4816 unsigned char *s1, *s2; | |
4817 register int len; | |
4818 char *translate; | |
4819 { | |
4820 register unsigned char *p1 = s1, *p2 = s2; | |
4821 while (len) | |
4822 { | |
4823 if (translate[*p1++] != translate[*p2++]) return 1; | |
4824 len--; | |
4825 } | |
4826 return 0; | |
4827 } | |
4828 | |
4829 /* Entry points for GNU code. */ | |
4830 | |
4831 /* re_compile_pattern is the GNU regular expression compiler: it | |
4832 compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
4833 Returns 0 if the pattern was valid, otherwise an error string. | |
4834 | |
4835 Assumes the `allocated' (and perhaps `buffer') and `translate' fields | |
4836 are set in BUFP on entry. | |
4837 | |
4838 We call regex_compile to do the actual compilation. */ | |
4839 | |
4840 const char * | |
4841 re_compile_pattern (pattern, length, bufp) | |
4842 const char *pattern; | |
4843 int length; | |
4844 struct re_pattern_buffer *bufp; | |
4845 { | |
4846 reg_errcode_t ret; | |
4847 | |
4848 /* GNU code is written to assume at least RE_NREGS registers will be set | |
4849 (and at least one extra will be -1). */ | |
4850 bufp->regs_allocated = REGS_UNALLOCATED; | |
4851 | |
4852 /* And GNU code determines whether or not to get register information | |
4853 by passing null for the REGS argument to re_match, etc., not by | |
4854 setting no_sub. */ | |
4855 bufp->no_sub = 0; | |
4856 | |
4857 /* Match anchors at newline. */ | |
4858 bufp->newline_anchor = 1; | |
4859 | |
4860 ret = regex_compile (pattern, length, re_syntax_options, bufp); | |
4861 | |
4862 return re_error_msg[(int) ret]; | |
4863 } | |
4864 | |
4865 /* Entry points compatible with 4.2 BSD regex library. We don't define | |
4866 them if this is an Emacs or POSIX compilation. */ | |
4867 | |
4868 #if !defined (emacs) && !defined (_POSIX_SOURCE) | |
4869 | |
4870 /* BSD has one and only one pattern buffer. */ | |
4871 static struct re_pattern_buffer re_comp_buf; | |
4872 | |
4873 char * | |
4874 re_comp (s) | |
4875 const char *s; | |
4876 { | |
4877 reg_errcode_t ret; | |
4878 | |
4879 if (!s) | |
4880 { | |
4881 if (!re_comp_buf.buffer) | |
4882 return "No previous regular expression"; | |
4883 return 0; | |
4884 } | |
4885 | |
4886 if (!re_comp_buf.buffer) | |
4887 { | |
4888 re_comp_buf.buffer = (unsigned char *) malloc (200); | |
4889 if (re_comp_buf.buffer == NULL) | |
4890 return "Memory exhausted"; | |
4891 re_comp_buf.allocated = 200; | |
4892 | |
4893 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
4894 if (re_comp_buf.fastmap == NULL) | |
4895 return "Memory exhausted"; | |
4896 } | |
4897 | |
4898 /* Since `re_exec' always passes NULL for the `regs' argument, we | |
4899 don't need to initialize the pattern buffer fields which affect it. */ | |
4900 | |
4901 /* Match anchors at newlines. */ | |
4902 re_comp_buf.newline_anchor = 1; | |
4903 | |
4904 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); | |
4905 | |
4906 /* Yes, we're discarding `const' here. */ | |
4907 return (char *) re_error_msg[(int) ret]; | |
4908 } | |
4909 | |
4910 | |
4911 int | |
4912 re_exec (s) | |
4913 const char *s; | |
4914 { | |
4915 const int len = strlen (s); | |
4916 return | |
4917 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
4918 } | |
4919 #endif /* not emacs and not _POSIX_SOURCE */ | |
4920 | |
4921 /* POSIX.2 functions. Don't define these for Emacs. */ | |
4922 | |
4923 #ifndef emacs | |
4924 | |
4925 /* regcomp takes a regular expression as a string and compiles it. | |
4926 | |
4927 PREG is a regex_t *. We do not expect any fields to be initialized, | |
4928 since POSIX says we shouldn't. Thus, we set | |
4929 | |
4930 `buffer' to the compiled pattern; | |
4931 `used' to the length of the compiled pattern; | |
4932 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
4933 REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
4934 RE_SYNTAX_POSIX_BASIC; | |
4935 `newline_anchor' to REG_NEWLINE being set in CFLAGS; | |
4936 `fastmap' and `fastmap_accurate' to zero; | |
4937 `re_nsub' to the number of subexpressions in PATTERN. | |
4938 | |
4939 PATTERN is the address of the pattern string. | |
4940 | |
4941 CFLAGS is a series of bits which affect compilation. | |
4942 | |
4943 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
4944 use POSIX basic syntax. | |
4945 | |
4946 If REG_NEWLINE is set, then . and [^...] don't match newline. | |
4947 Also, regexec will try a match beginning after every newline. | |
4948 | |
4949 If REG_ICASE is set, then we considers upper- and lowercase | |
4950 versions of letters to be equivalent when matching. | |
4951 | |
4952 If REG_NOSUB is set, then when PREG is passed to regexec, that | |
4953 routine will report only success or failure, and nothing about the | |
4954 registers. | |
4955 | |
4956 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for | |
4957 the return codes and their meanings.) */ | |
4958 | |
4959 int | |
4960 regcomp (preg, pattern, cflags) | |
4961 regex_t *preg; | |
4962 const char *pattern; | |
4963 int cflags; | |
4964 { | |
4965 reg_errcode_t ret; | |
4966 unsigned syntax | |
1642
340feb030df1
*** empty log message ***
David J. MacKenzie <djm@gnu.org>
parents:
1641
diff
changeset
|
4967 = (cflags & REG_EXTENDED) ? |
340feb030df1
*** empty log message ***
David J. MacKenzie <djm@gnu.org>
parents:
1641
diff
changeset
|
4968 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; |
1155 | 4969 |
4970 /* regex_compile will allocate the space for the compiled pattern. */ | |
4971 preg->buffer = 0; | |
1642
340feb030df1
*** empty log message ***
David J. MacKenzie <djm@gnu.org>
parents:
1641
diff
changeset
|
4972 preg->allocated = 0; |
2758 | 4973 preg->used = 0; |
1155 | 4974 |
4975 /* Don't bother to use a fastmap when searching. This simplifies the | |
4976 REG_NEWLINE case: if we used a fastmap, we'd have to put all the | |
4977 characters after newlines into the fastmap. This way, we just try | |
4978 every character. */ | |
4979 preg->fastmap = 0; | |
4980 | |
4981 if (cflags & REG_ICASE) | |
4982 { | |
4983 unsigned i; | |
4984 | |
4985 preg->translate = (char *) malloc (CHAR_SET_SIZE); | |
4986 if (preg->translate == NULL) | |
4987 return (int) REG_ESPACE; | |
4988 | |
4989 /* Map uppercase characters to corresponding lowercase ones. */ | |
4990 for (i = 0; i < CHAR_SET_SIZE; i++) | |
1668 | 4991 preg->translate[i] = ISUPPER (i) ? tolower (i) : i; |
1155 | 4992 } |
4993 else | |
4994 preg->translate = NULL; | |
4995 | |
4996 /* If REG_NEWLINE is set, newlines are treated differently. */ | |
4997 if (cflags & REG_NEWLINE) | |
4998 { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
4999 syntax &= ~RE_DOT_NEWLINE; | |
5000 syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
5001 /* It also changes the matching behavior. */ | |
5002 preg->newline_anchor = 1; | |
5003 } | |
5004 else | |
5005 preg->newline_anchor = 0; | |
5006 | |
5007 preg->no_sub = !!(cflags & REG_NOSUB); | |
5008 | |
5009 /* POSIX says a null character in the pattern terminates it, so we | |
5010 can use strlen here in compiling the pattern. */ | |
5011 ret = regex_compile (pattern, strlen (pattern), syntax, preg); | |
5012 | |
5013 /* POSIX doesn't distinguish between an unmatched open-group and an | |
5014 unmatched close-group: both are REG_EPAREN. */ | |
5015 if (ret == REG_ERPAREN) ret = REG_EPAREN; | |
5016 | |
5017 return (int) ret; | |
5018 } | |
5019 | |
5020 | |
5021 /* regexec searches for a given pattern, specified by PREG, in the | |
5022 string STRING. | |
5023 | |
5024 If NMATCH is zero or REG_NOSUB was set in the cflags argument to | |
5025 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at | |
5026 least NMATCH elements, and we set them to the offsets of the | |
5027 corresponding matched substrings. | |
5028 | |
5029 EFLAGS specifies `execution flags' which affect matching: if | |
5030 REG_NOTBOL is set, then ^ does not match at the beginning of the | |
5031 string; if REG_NOTEOL is set, then $ does not match at the end. | |
5032 | |
5033 We return 0 if we find a match and REG_NOMATCH if not. */ | |
5034 | |
5035 int | |
5036 regexec (preg, string, nmatch, pmatch, eflags) | |
5037 const regex_t *preg; | |
5038 const char *string; | |
5039 size_t nmatch; | |
5040 regmatch_t pmatch[]; | |
5041 int eflags; | |
5042 { | |
5043 int ret; | |
5044 struct re_registers regs; | |
5045 regex_t private_preg; | |
5046 int len = strlen (string); | |
5047 boolean want_reg_info = !preg->no_sub && nmatch > 0; | |
5048 | |
5049 private_preg = *preg; | |
5050 | |
5051 private_preg.not_bol = !!(eflags & REG_NOTBOL); | |
5052 private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
5053 | |
5054 /* The user has told us exactly how many registers to return | |
5055 information about, via `nmatch'. We have to pass that on to the | |
5056 matching routines. */ | |
5057 private_preg.regs_allocated = REGS_FIXED; | |
5058 | |
5059 if (want_reg_info) | |
5060 { | |
5061 regs.num_regs = nmatch; | |
5062 regs.start = TALLOC (nmatch, regoff_t); | |
5063 regs.end = TALLOC (nmatch, regoff_t); | |
5064 if (regs.start == NULL || regs.end == NULL) | |
5065 return (int) REG_NOMATCH; | |
5066 } | |
5067 | |
5068 /* Perform the searching operation. */ | |
5069 ret = re_search (&private_preg, string, len, | |
5070 /* start: */ 0, /* range: */ len, | |
5071 want_reg_info ? ®s : (struct re_registers *) 0); | |
5072 | |
5073 /* Copy the register information to the POSIX structure. */ | |
5074 if (want_reg_info) | |
5075 { | |
5076 if (ret >= 0) | |
5077 { | |
5078 unsigned r; | |
5079 | |
5080 for (r = 0; r < nmatch; r++) | |
5081 { | |
5082 pmatch[r].rm_so = regs.start[r]; | |
5083 pmatch[r].rm_eo = regs.end[r]; | |
5084 } | |
5085 } | |
5086 | |
5087 /* If we needed the temporary register info, free the space now. */ | |
5088 free (regs.start); | |
5089 free (regs.end); | |
5090 } | |
5091 | |
5092 /* We want zero return to mean success, unlike `re_search'. */ | |
5093 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
5094 } | |
5095 | |
5096 | |
5097 /* Returns a message corresponding to an error code, ERRCODE, returned | |
1637 | 5098 from either regcomp or regexec. We don't use PREG here. */ |
1155 | 5099 |
5100 size_t | |
5101 regerror (errcode, preg, errbuf, errbuf_size) | |
5102 int errcode; | |
5103 const regex_t *preg; | |
5104 char *errbuf; | |
5105 size_t errbuf_size; | |
5106 { | |
1738 | 5107 const char *msg; |
5108 size_t msg_size; | |
5109 | |
5110 if (errcode < 0 | |
5111 || errcode >= (sizeof (re_error_msg) / sizeof (re_error_msg[0]))) | |
5112 /* Only error codes returned by the rest of the code should be passed | |
5113 to this routine. If we are given anything else, or if other regex | |
5114 code generates an invalid error code, then the program has a bug. | |
5115 Dump core so we can fix it. */ | |
5116 abort (); | |
5117 | |
2453 | 5118 msg = re_error_msg[errcode]; |
5119 | |
5120 /* POSIX doesn't require that we do anything in this case, but why | |
5121 not be nice. */ | |
5122 if (! msg) | |
5123 msg = "Success"; | |
5124 | |
1738 | 5125 msg_size = strlen (msg) + 1; /* Includes the null. */ |
1155 | 5126 |
5127 if (errbuf_size != 0) | |
5128 { | |
5129 if (msg_size > errbuf_size) | |
5130 { | |
5131 strncpy (errbuf, msg, errbuf_size - 1); | |
5132 errbuf[errbuf_size - 1] = 0; | |
5133 } | |
5134 else | |
5135 strcpy (errbuf, msg); | |
5136 } | |
5137 | |
5138 return msg_size; | |
5139 } | |
5140 | |
5141 | |
5142 /* Free dynamically allocated space used by PREG. */ | |
5143 | |
5144 void | |
5145 regfree (preg) | |
5146 regex_t *preg; | |
5147 { | |
5148 if (preg->buffer != NULL) | |
5149 free (preg->buffer); | |
5150 preg->buffer = NULL; | |
5151 | |
5152 preg->allocated = 0; | |
5153 preg->used = 0; | |
5154 | |
5155 if (preg->fastmap != NULL) | |
5156 free (preg->fastmap); | |
5157 preg->fastmap = NULL; | |
5158 preg->fastmap_accurate = 0; | |
5159 | |
5160 if (preg->translate != NULL) | |
5161 free (preg->translate); | |
5162 preg->translate = NULL; | |
5163 } | |
5164 | |
5165 #endif /* not emacs */ | |
5166 | |
5167 /* | |
5168 Local variables: | |
5169 make-backup-files: t | |
5170 version-control: t | |
5171 trim-versions-without-asking: nil | |
5172 End: | |
5173 */ |