Mercurial > emacs
annotate src/regex.c @ 5979:99e5ff29c408
*** empty log message ***
author | Richard M. Stallman <rms@gnu.org> |
---|---|
date | Thu, 17 Feb 1994 20:02:50 +0000 |
parents | e9cd26c4e4ff |
children | ab6eb517d144 |
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 | |
5841 | 259 #undef MAX |
260 #undef MIN | |
1155 | 261 #define MAX(a, b) ((a) > (b) ? (a) : (b)) |
262 #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
263 | |
264 typedef char boolean; | |
265 #define false 0 | |
266 #define true 1 | |
267 | |
268 /* These are the command codes that appear in compiled regular | |
269 expressions. Some opcodes are followed by argument bytes. A | |
270 command code can specify any interpretation whatsoever for its | |
271 arguments. Zero bytes may appear in the compiled regular expression. | |
272 | |
273 The value of `exactn' is needed in search.c (search_buffer) in Emacs. | |
274 So regex.h defines a symbol `RE_EXACTN_VALUE' to be 1; the value of | |
275 `exactn' we use here must also be 1. */ | |
276 | |
277 typedef enum | |
278 { | |
279 no_op = 0, | |
280 | |
281 /* Followed by one byte giving n, then by n literal bytes. */ | |
282 exactn = 1, | |
283 | |
284 /* Matches any (more or less) character. */ | |
285 anychar, | |
286 | |
287 /* Matches any one char belonging to specified set. First | |
288 following byte is number of bitmap bytes. Then come bytes | |
289 for a bitmap saying which chars are in. Bits in each byte | |
290 are ordered low-bit-first. A character is in the set if its | |
291 bit is 1. A character too large to have a bit in the map is | |
292 automatically not in the set. */ | |
293 charset, | |
294 | |
295 /* Same parameters as charset, but match any character that is | |
296 not one of those specified. */ | |
297 charset_not, | |
298 | |
299 /* Start remembering the text that is matched, for storing in a | |
300 register. Followed by one byte with the register number, in | |
301 the range 0 to one less than the pattern buffer's re_nsub | |
302 field. Then followed by one byte with the number of groups | |
303 inner to this one. (This last has to be part of the | |
304 start_memory only because we need it in the on_failure_jump | |
305 of re_match_2.) */ | |
306 start_memory, | |
307 | |
308 /* Stop remembering the text that is matched and store it in a | |
309 memory register. Followed by one byte with the register | |
310 number, in the range 0 to one less than `re_nsub' in the | |
311 pattern buffer, and one byte with the number of inner groups, | |
312 just like `start_memory'. (We need the number of inner | |
313 groups here because we don't have any easy way of finding the | |
314 corresponding start_memory when we're at a stop_memory.) */ | |
315 stop_memory, | |
316 | |
317 /* Match a duplicate of something remembered. Followed by one | |
318 byte containing the register number. */ | |
319 duplicate, | |
320 | |
321 /* Fail unless at beginning of line. */ | |
322 begline, | |
323 | |
324 /* Fail unless at end of line. */ | |
325 endline, | |
326 | |
327 /* Succeeds if at beginning of buffer (if emacs) or at beginning | |
328 of string to be matched (if not). */ | |
329 begbuf, | |
330 | |
331 /* Analogously, for end of buffer/string. */ | |
332 endbuf, | |
333 | |
334 /* Followed by two byte relative address to which to jump. */ | |
335 jump, | |
336 | |
337 /* Same as jump, but marks the end of an alternative. */ | |
338 jump_past_alt, | |
339 | |
340 /* Followed by two-byte relative address of place to resume at | |
341 in case of failure. */ | |
342 on_failure_jump, | |
343 | |
344 /* Like on_failure_jump, but pushes a placeholder instead of the | |
345 current string position when executed. */ | |
346 on_failure_keep_string_jump, | |
347 | |
348 /* Throw away latest failure point and then jump to following | |
349 two-byte relative address. */ | |
350 pop_failure_jump, | |
351 | |
352 /* Change to pop_failure_jump if know won't have to backtrack to | |
353 match; otherwise change to jump. This is used to jump | |
354 back to the beginning of a repeat. If what follows this jump | |
355 clearly won't match what the repeat does, such that we can be | |
356 sure that there is no use backtracking out of repetitions | |
357 already matched, then we change it to a pop_failure_jump. | |
358 Followed by two-byte address. */ | |
359 maybe_pop_jump, | |
360 | |
361 /* Jump to following two-byte address, and push a dummy failure | |
362 point. This failure point will be thrown away if an attempt | |
363 is made to use it for a failure. A `+' construct makes this | |
364 before the first repeat. Also used as an intermediary kind | |
365 of jump when compiling an alternative. */ | |
366 dummy_failure_jump, | |
367 | |
368 /* Push a dummy failure point and continue. Used at the end of | |
369 alternatives. */ | |
370 push_dummy_failure, | |
371 | |
372 /* Followed by two-byte relative address and two-byte number n. | |
373 After matching N times, jump to the address upon failure. */ | |
374 succeed_n, | |
375 | |
376 /* Followed by two-byte relative address, and two-byte number n. | |
377 Jump to the address N times, then fail. */ | |
378 jump_n, | |
379 | |
380 /* Set the following two-byte relative address to the | |
381 subsequent two-byte number. The address *includes* the two | |
382 bytes of number. */ | |
383 set_number_at, | |
384 | |
385 wordchar, /* Matches any word-constituent character. */ | |
386 notwordchar, /* Matches any char that is not a word-constituent. */ | |
387 | |
388 wordbeg, /* Succeeds if at word beginning. */ | |
389 wordend, /* Succeeds if at word end. */ | |
390 | |
391 wordbound, /* Succeeds if at a word boundary. */ | |
392 notwordbound /* Succeeds if not at a word boundary. */ | |
393 | |
394 #ifdef emacs | |
395 ,before_dot, /* Succeeds if before point. */ | |
396 at_dot, /* Succeeds if at point. */ | |
397 after_dot, /* Succeeds if after point. */ | |
398 | |
399 /* Matches any character whose syntax is specified. Followed by | |
400 a byte which contains a syntax code, e.g., Sword. */ | |
401 syntaxspec, | |
402 | |
403 /* Matches any character whose syntax is not that specified. */ | |
404 notsyntaxspec | |
405 #endif /* emacs */ | |
406 } re_opcode_t; | |
407 | |
408 /* Common operations on the compiled pattern. */ | |
409 | |
410 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
411 | |
412 #define STORE_NUMBER(destination, number) \ | |
413 do { \ | |
414 (destination)[0] = (number) & 0377; \ | |
415 (destination)[1] = (number) >> 8; \ | |
416 } while (0) | |
417 | |
418 /* Same as STORE_NUMBER, except increment DESTINATION to | |
419 the byte after where the number is stored. Therefore, DESTINATION | |
420 must be an lvalue. */ | |
421 | |
422 #define STORE_NUMBER_AND_INCR(destination, number) \ | |
423 do { \ | |
424 STORE_NUMBER (destination, number); \ | |
425 (destination) += 2; \ | |
426 } while (0) | |
427 | |
428 /* Put into DESTINATION a number stored in two contiguous bytes starting | |
429 at SOURCE. */ | |
430 | |
431 #define EXTRACT_NUMBER(destination, source) \ | |
432 do { \ | |
433 (destination) = *(source) & 0377; \ | |
434 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
435 } while (0) | |
436 | |
437 #ifdef DEBUG | |
438 static void | |
439 extract_number (dest, source) | |
440 int *dest; | |
441 unsigned char *source; | |
442 { | |
443 int temp = SIGN_EXTEND_CHAR (*(source + 1)); | |
444 *dest = *source & 0377; | |
445 *dest += temp << 8; | |
446 } | |
447 | |
448 #ifndef EXTRACT_MACROS /* To debug the macros. */ | |
449 #undef EXTRACT_NUMBER | |
450 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
451 #endif /* not EXTRACT_MACROS */ | |
452 | |
453 #endif /* DEBUG */ | |
454 | |
455 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
456 SOURCE must be an lvalue. */ | |
457 | |
458 #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
459 do { \ | |
460 EXTRACT_NUMBER (destination, source); \ | |
461 (source) += 2; \ | |
462 } while (0) | |
463 | |
464 #ifdef DEBUG | |
465 static void | |
466 extract_number_and_incr (destination, source) | |
467 int *destination; | |
468 unsigned char **source; | |
469 { | |
470 extract_number (destination, *source); | |
471 *source += 2; | |
472 } | |
473 | |
474 #ifndef EXTRACT_MACROS | |
475 #undef EXTRACT_NUMBER_AND_INCR | |
476 #define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
477 extract_number_and_incr (&dest, &src) | |
478 #endif /* not EXTRACT_MACROS */ | |
479 | |
480 #endif /* DEBUG */ | |
481 | |
482 /* If DEBUG is defined, Regex prints many voluminous messages about what | |
483 it is doing (if the variable `debug' is nonzero). If linked with the | |
484 main program in `iregex.c', you can enter patterns and strings | |
485 interactively. And if linked with the main program in `main.c' and | |
486 the other test files, you can run the already-written tests. */ | |
487 | |
488 #ifdef DEBUG | |
489 | |
490 /* We use standard I/O for debugging. */ | |
491 #include <stdio.h> | |
492 | |
493 /* It is useful to test things that ``must'' be true when debugging. */ | |
494 #include <assert.h> | |
495 | |
496 static int debug = 0; | |
497 | |
498 #define DEBUG_STATEMENT(e) e | |
499 #define DEBUG_PRINT1(x) if (debug) printf (x) | |
500 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) | |
501 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) | |
1637 | 502 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) |
1155 | 503 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ |
504 if (debug) print_partial_compiled_pattern (s, e) | |
505 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ | |
506 if (debug) print_double_string (w, s1, sz1, s2, sz2) | |
507 | |
508 | |
509 extern void printchar (); | |
510 | |
511 /* Print the fastmap in human-readable form. */ | |
512 | |
513 void | |
514 print_fastmap (fastmap) | |
515 char *fastmap; | |
516 { | |
517 unsigned was_a_range = 0; | |
518 unsigned i = 0; | |
519 | |
520 while (i < (1 << BYTEWIDTH)) | |
521 { | |
522 if (fastmap[i++]) | |
523 { | |
524 was_a_range = 0; | |
525 printchar (i - 1); | |
526 while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
527 { | |
528 was_a_range = 1; | |
529 i++; | |
530 } | |
531 if (was_a_range) | |
532 { | |
533 printf ("-"); | |
534 printchar (i - 1); | |
535 } | |
536 } | |
537 } | |
538 putchar ('\n'); | |
539 } | |
540 | |
541 | |
542 /* Print a compiled pattern string in human-readable form, starting at | |
543 the START pointer into it and ending just before the pointer END. */ | |
544 | |
545 void | |
546 print_partial_compiled_pattern (start, end) | |
547 unsigned char *start; | |
548 unsigned char *end; | |
549 { | |
550 int mcnt, mcnt2; | |
551 unsigned char *p = start; | |
552 unsigned char *pend = end; | |
553 | |
554 if (start == NULL) | |
555 { | |
556 printf ("(null)\n"); | |
557 return; | |
558 } | |
559 | |
560 /* Loop over pattern commands. */ | |
561 while (p < pend) | |
562 { | |
2615 | 563 printf ("%d:\t", p - start); |
564 | |
1155 | 565 switch ((re_opcode_t) *p++) |
566 { | |
567 case no_op: | |
568 printf ("/no_op"); | |
569 break; | |
570 | |
571 case exactn: | |
572 mcnt = *p++; | |
573 printf ("/exactn/%d", mcnt); | |
574 do | |
575 { | |
576 putchar ('/'); | |
577 printchar (*p++); | |
578 } | |
579 while (--mcnt); | |
580 break; | |
581 | |
582 case start_memory: | |
583 mcnt = *p++; | |
584 printf ("/start_memory/%d/%d", mcnt, *p++); | |
585 break; | |
586 | |
587 case stop_memory: | |
588 mcnt = *p++; | |
589 printf ("/stop_memory/%d/%d", mcnt, *p++); | |
590 break; | |
591 | |
592 case duplicate: | |
593 printf ("/duplicate/%d", *p++); | |
594 break; | |
595 | |
596 case anychar: | |
597 printf ("/anychar"); | |
598 break; | |
599 | |
600 case charset: | |
601 case charset_not: | |
602 { | |
2615 | 603 register int c, last = -100; |
604 register int in_range = 0; | |
605 | |
606 printf ("/charset [%s", | |
607 (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); | |
1155 | 608 |
609 assert (p + *p < pend); | |
610 | |
2615 | 611 for (c = 0; c < 256; c++) |
612 if (c / 8 < *p | |
613 && (p[1 + (c/8)] & (1 << (c % 8)))) | |
614 { | |
615 /* Are we starting a range? */ | |
616 if (last + 1 == c && ! in_range) | |
617 { | |
618 putchar ('-'); | |
619 in_range = 1; | |
620 } | |
621 /* Have we broken a range? */ | |
622 else if (last + 1 != c && in_range) | |
1155 | 623 { |
2615 | 624 printchar (last); |
625 in_range = 0; | |
626 } | |
1155 | 627 |
2615 | 628 if (! in_range) |
629 printchar (c); | |
630 | |
631 last = c; | |
1155 | 632 } |
2615 | 633 |
634 if (in_range) | |
635 printchar (last); | |
636 | |
637 putchar (']'); | |
638 | |
1155 | 639 p += 1 + *p; |
640 } | |
2615 | 641 break; |
1155 | 642 |
643 case begline: | |
644 printf ("/begline"); | |
645 break; | |
646 | |
647 case endline: | |
648 printf ("/endline"); | |
649 break; | |
650 | |
651 case on_failure_jump: | |
652 extract_number_and_incr (&mcnt, &p); | |
2615 | 653 printf ("/on_failure_jump to %d", p + mcnt - start); |
1155 | 654 break; |
655 | |
656 case on_failure_keep_string_jump: | |
657 extract_number_and_incr (&mcnt, &p); | |
2615 | 658 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start); |
1155 | 659 break; |
660 | |
661 case dummy_failure_jump: | |
662 extract_number_and_incr (&mcnt, &p); | |
2615 | 663 printf ("/dummy_failure_jump to %d", p + mcnt - start); |
1155 | 664 break; |
665 | |
666 case push_dummy_failure: | |
667 printf ("/push_dummy_failure"); | |
668 break; | |
669 | |
670 case maybe_pop_jump: | |
671 extract_number_and_incr (&mcnt, &p); | |
2615 | 672 printf ("/maybe_pop_jump to %d", p + mcnt - start); |
1155 | 673 break; |
674 | |
675 case pop_failure_jump: | |
676 extract_number_and_incr (&mcnt, &p); | |
2615 | 677 printf ("/pop_failure_jump to %d", p + mcnt - start); |
1155 | 678 break; |
679 | |
680 case jump_past_alt: | |
681 extract_number_and_incr (&mcnt, &p); | |
2615 | 682 printf ("/jump_past_alt to %d", p + mcnt - start); |
1155 | 683 break; |
684 | |
685 case jump: | |
686 extract_number_and_incr (&mcnt, &p); | |
2615 | 687 printf ("/jump to %d", p + mcnt - start); |
1155 | 688 break; |
689 | |
690 case succeed_n: | |
691 extract_number_and_incr (&mcnt, &p); | |
692 extract_number_and_incr (&mcnt2, &p); | |
2615 | 693 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2); |
1155 | 694 break; |
695 | |
696 case jump_n: | |
697 extract_number_and_incr (&mcnt, &p); | |
698 extract_number_and_incr (&mcnt2, &p); | |
2615 | 699 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2); |
1155 | 700 break; |
701 | |
702 case set_number_at: | |
703 extract_number_and_incr (&mcnt, &p); | |
704 extract_number_and_incr (&mcnt2, &p); | |
2615 | 705 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2); |
1155 | 706 break; |
707 | |
708 case wordbound: | |
709 printf ("/wordbound"); | |
710 break; | |
711 | |
712 case notwordbound: | |
713 printf ("/notwordbound"); | |
714 break; | |
715 | |
716 case wordbeg: | |
717 printf ("/wordbeg"); | |
718 break; | |
719 | |
720 case wordend: | |
721 printf ("/wordend"); | |
722 | |
723 #ifdef emacs | |
724 case before_dot: | |
725 printf ("/before_dot"); | |
726 break; | |
727 | |
728 case at_dot: | |
729 printf ("/at_dot"); | |
730 break; | |
731 | |
732 case after_dot: | |
733 printf ("/after_dot"); | |
734 break; | |
735 | |
736 case syntaxspec: | |
737 printf ("/syntaxspec"); | |
738 mcnt = *p++; | |
739 printf ("/%d", mcnt); | |
740 break; | |
741 | |
742 case notsyntaxspec: | |
743 printf ("/notsyntaxspec"); | |
744 mcnt = *p++; | |
745 printf ("/%d", mcnt); | |
746 break; | |
747 #endif /* emacs */ | |
748 | |
749 case wordchar: | |
750 printf ("/wordchar"); | |
751 break; | |
752 | |
753 case notwordchar: | |
754 printf ("/notwordchar"); | |
755 break; | |
756 | |
757 case begbuf: | |
758 printf ("/begbuf"); | |
759 break; | |
760 | |
761 case endbuf: | |
762 printf ("/endbuf"); | |
763 break; | |
764 | |
765 default: | |
766 printf ("?%d", *(p-1)); | |
767 } | |
2615 | 768 |
769 putchar ('\n'); | |
1155 | 770 } |
2615 | 771 |
772 printf ("%d:\tend of pattern.\n", p - start); | |
1155 | 773 } |
774 | |
775 | |
776 void | |
777 print_compiled_pattern (bufp) | |
778 struct re_pattern_buffer *bufp; | |
779 { | |
780 unsigned char *buffer = bufp->buffer; | |
781 | |
782 print_partial_compiled_pattern (buffer, buffer + bufp->used); | |
783 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated); | |
784 | |
785 if (bufp->fastmap_accurate && bufp->fastmap) | |
786 { | |
787 printf ("fastmap: "); | |
788 print_fastmap (bufp->fastmap); | |
789 } | |
790 | |
791 printf ("re_nsub: %d\t", bufp->re_nsub); | |
792 printf ("regs_alloc: %d\t", bufp->regs_allocated); | |
793 printf ("can_be_null: %d\t", bufp->can_be_null); | |
794 printf ("newline_anchor: %d\n", bufp->newline_anchor); | |
795 printf ("no_sub: %d\t", bufp->no_sub); | |
796 printf ("not_bol: %d\t", bufp->not_bol); | |
797 printf ("not_eol: %d\t", bufp->not_eol); | |
798 printf ("syntax: %d\n", bufp->syntax); | |
799 /* Perhaps we should print the translate table? */ | |
800 } | |
801 | |
802 | |
803 void | |
804 print_double_string (where, string1, size1, string2, size2) | |
805 const char *where; | |
806 const char *string1; | |
807 const char *string2; | |
808 int size1; | |
809 int size2; | |
810 { | |
811 unsigned this_char; | |
812 | |
813 if (where == NULL) | |
814 printf ("(null)"); | |
815 else | |
816 { | |
817 if (FIRST_STRING_P (where)) | |
818 { | |
819 for (this_char = where - string1; this_char < size1; this_char++) | |
820 printchar (string1[this_char]); | |
821 | |
822 where = string2; | |
823 } | |
824 | |
825 for (this_char = where - string2; this_char < size2; this_char++) | |
826 printchar (string2[this_char]); | |
827 } | |
828 } | |
829 | |
830 #else /* not DEBUG */ | |
831 | |
832 #undef assert | |
833 #define assert(e) | |
834 | |
835 #define DEBUG_STATEMENT(e) | |
836 #define DEBUG_PRINT1(x) | |
837 #define DEBUG_PRINT2(x1, x2) | |
838 #define DEBUG_PRINT3(x1, x2, x3) | |
1637 | 839 #define DEBUG_PRINT4(x1, x2, x3, x4) |
1155 | 840 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) |
841 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
842 | |
843 #endif /* not DEBUG */ | |
844 | |
845 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
846 also be assigned to arbitrarily: each pattern buffer stores its own | |
847 syntax, so it can be changed between regex compilations. */ | |
848 reg_syntax_t re_syntax_options = RE_SYNTAX_EMACS; | |
849 | |
850 | |
851 /* Specify the precise syntax of regexps for compilation. This provides | |
852 for compatibility for various utilities which historically have | |
853 different, incompatible syntaxes. | |
854 | |
855 The argument SYNTAX is a bit mask comprised of the various bits | |
856 defined in regex.h. We return the old syntax. */ | |
857 | |
858 reg_syntax_t | |
859 re_set_syntax (syntax) | |
860 reg_syntax_t syntax; | |
861 { | |
862 reg_syntax_t ret = re_syntax_options; | |
863 | |
864 re_syntax_options = syntax; | |
865 return ret; | |
866 } | |
867 | |
868 /* This table gives an error message for each of the error codes listed | |
869 in regex.h. Obviously the order here has to be same as there. */ | |
870 | |
871 static const char *re_error_msg[] = | |
872 { NULL, /* REG_NOERROR */ | |
873 "No match", /* REG_NOMATCH */ | |
874 "Invalid regular expression", /* REG_BADPAT */ | |
875 "Invalid collation character", /* REG_ECOLLATE */ | |
876 "Invalid character class name", /* REG_ECTYPE */ | |
877 "Trailing backslash", /* REG_EESCAPE */ | |
878 "Invalid back reference", /* REG_ESUBREG */ | |
879 "Unmatched [ or [^", /* REG_EBRACK */ | |
880 "Unmatched ( or \\(", /* REG_EPAREN */ | |
881 "Unmatched \\{", /* REG_EBRACE */ | |
882 "Invalid content of \\{\\}", /* REG_BADBR */ | |
883 "Invalid range end", /* REG_ERANGE */ | |
884 "Memory exhausted", /* REG_ESPACE */ | |
885 "Invalid preceding regular expression", /* REG_BADRPT */ | |
886 "Premature end of regular expression", /* REG_EEND */ | |
887 "Regular expression too big", /* REG_ESIZE */ | |
888 "Unmatched ) or \\)", /* REG_ERPAREN */ | |
889 }; | |
890 | |
2949 | 891 /* Avoiding alloca during matching, to placate r_alloc. */ |
892 | |
2952 | 893 /* Define MATCH_MAY_ALLOCATE if we need to make sure that the |
2949 | 894 searching and matching functions should not call alloca. On some |
895 systems, alloca is implemented in terms of malloc, and if we're | |
896 using the relocating allocator routines, then malloc could cause a | |
897 relocation, which might (if the strings being searched are in the | |
898 ralloc heap) shift the data out from underneath the regexp | |
3614 | 899 routines. |
900 | |
901 Here's another reason to avoid allocation: Emacs insists on | |
902 processing input from X in a signal handler; processing X input may | |
903 call malloc; if input arrives while a matching routine is calling | |
904 malloc, then we're scrod. But Emacs can't just block input while | |
905 calling matching routines; then we don't notice interrupts when | |
906 they come in. So, Emacs blocks input around all regexp calls | |
907 except the matching calls, which it leaves unprotected, in the | |
908 faith that they will not malloc. */ | |
2952 | 909 |
910 /* Normally, this is fine. */ | |
911 #define MATCH_MAY_ALLOCATE | |
912 | |
913 /* But under some circumstances, it's not. */ | |
3614 | 914 #if defined (emacs) || (defined (REL_ALLOC) && defined (C_ALLOCA)) |
2952 | 915 #undef MATCH_MAY_ALLOCATE |
2949 | 916 #endif |
917 | |
918 | |
919 /* Failure stack declarations and macros; both re_compile_fastmap and | |
920 re_match_2 use a failure stack. These have to be macros because of | |
921 REGEX_ALLOCATE. */ | |
922 | |
923 | |
924 /* Number of failure points for which to initially allocate space | |
925 when matching. If this number is exceeded, we allocate more | |
926 space, so it is not a hard limit. */ | |
927 #ifndef INIT_FAILURE_ALLOC | |
928 #define INIT_FAILURE_ALLOC 5 | |
929 #endif | |
930 | |
931 /* Roughly the maximum number of failure points on the stack. Would be | |
932 exactly that if always used MAX_FAILURE_SPACE each time we failed. | |
933 This is a variable only so users of regex can assign to it; we never | |
934 change it ourselves. */ | |
935 int re_max_failures = 2000; | |
936 | |
4918
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
937 typedef unsigned char *fail_stack_elt_t; |
2949 | 938 |
939 typedef struct | |
940 { | |
941 fail_stack_elt_t *stack; | |
942 unsigned size; | |
943 unsigned avail; /* Offset of next open position. */ | |
944 } fail_stack_type; | |
945 | |
946 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0) | |
947 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) | |
948 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) | |
949 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail]) | |
950 | |
951 | |
952 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */ | |
953 | |
2952 | 954 #ifdef MATCH_MAY_ALLOCATE |
2949 | 955 #define INIT_FAIL_STACK() \ |
956 do { \ | |
957 fail_stack.stack = (fail_stack_elt_t *) \ | |
958 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \ | |
959 \ | |
960 if (fail_stack.stack == NULL) \ | |
961 return -2; \ | |
962 \ | |
963 fail_stack.size = INIT_FAILURE_ALLOC; \ | |
964 fail_stack.avail = 0; \ | |
965 } while (0) | |
966 #else | |
967 #define INIT_FAIL_STACK() \ | |
968 do { \ | |
969 fail_stack.avail = 0; \ | |
970 } while (0) | |
971 #endif | |
972 | |
973 | |
974 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. | |
975 | |
976 Return 1 if succeeds, and 0 if either ran out of memory | |
977 allocating space for it or it was already too large. | |
978 | |
979 REGEX_REALLOCATE requires `destination' be declared. */ | |
980 | |
981 #define DOUBLE_FAIL_STACK(fail_stack) \ | |
982 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \ | |
983 ? 0 \ | |
984 : ((fail_stack).stack = (fail_stack_elt_t *) \ | |
985 REGEX_REALLOCATE ((fail_stack).stack, \ | |
986 (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
987 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \ | |
988 \ | |
989 (fail_stack).stack == NULL \ | |
990 ? 0 \ | |
991 : ((fail_stack).size <<= 1, \ | |
992 1))) | |
993 | |
994 | |
995 /* Push PATTERN_OP on FAIL_STACK. | |
996 | |
997 Return 1 if was able to do so and 0 if ran out of memory allocating | |
998 space to do so. */ | |
999 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \ | |
1000 ((FAIL_STACK_FULL () \ | |
1001 && !DOUBLE_FAIL_STACK (fail_stack)) \ | |
1002 ? 0 \ | |
1003 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \ | |
1004 1)) | |
1005 | |
1006 /* This pushes an item onto the failure stack. Must be a four-byte | |
1007 value. Assumes the variable `fail_stack'. Probably should only | |
1008 be called from within `PUSH_FAILURE_POINT'. */ | |
1009 #define PUSH_FAILURE_ITEM(item) \ | |
1010 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item | |
1011 | |
1012 /* The complement operation. Assumes `fail_stack' is nonempty. */ | |
1013 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail] | |
1014 | |
1015 /* Used to omit pushing failure point id's when we're not debugging. */ | |
1016 #ifdef DEBUG | |
1017 #define DEBUG_PUSH PUSH_FAILURE_ITEM | |
1018 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM () | |
1019 #else | |
1020 #define DEBUG_PUSH(item) | |
1021 #define DEBUG_POP(item_addr) | |
1022 #endif | |
1023 | |
1024 | |
1025 /* Push the information about the state we will need | |
1026 if we ever fail back to it. | |
1027 | |
1028 Requires variables fail_stack, regstart, regend, reg_info, and | |
1029 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be | |
1030 declared. | |
1031 | |
1032 Does `return FAILURE_CODE' if runs out of memory. */ | |
1033 | |
1034 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ | |
1035 do { \ | |
1036 char *destination; \ | |
1037 /* Must be int, so when we don't save any registers, the arithmetic \ | |
1038 of 0 + -1 isn't done as unsigned. */ \ | |
1039 int this_reg; \ | |
1040 \ | |
1041 DEBUG_STATEMENT (failure_id++); \ | |
1042 DEBUG_STATEMENT (nfailure_points_pushed++); \ | |
1043 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ | |
1044 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ | |
1045 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ | |
1046 \ | |
1047 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \ | |
1048 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ | |
1049 \ | |
1050 /* Ensure we have enough space allocated for what we will push. */ \ | |
1051 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ | |
1052 { \ | |
1053 if (!DOUBLE_FAIL_STACK (fail_stack)) \ | |
1054 return failure_code; \ | |
1055 \ | |
1056 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ | |
1057 (fail_stack).size); \ | |
1058 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ | |
1059 } \ | |
1060 \ | |
1061 /* Push the info, starting with the registers. */ \ | |
1062 DEBUG_PRINT1 ("\n"); \ | |
1063 \ | |
1064 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ | |
1065 this_reg++) \ | |
1066 { \ | |
1067 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \ | |
1068 DEBUG_STATEMENT (num_regs_pushed++); \ | |
1069 \ | |
1070 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ | |
1071 PUSH_FAILURE_ITEM (regstart[this_reg]); \ | |
1072 \ | |
1073 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ | |
1074 PUSH_FAILURE_ITEM (regend[this_reg]); \ | |
1075 \ | |
1076 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \ | |
1077 DEBUG_PRINT2 (" match_null=%d", \ | |
1078 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ | |
1079 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ | |
1080 DEBUG_PRINT2 (" matched_something=%d", \ | |
1081 MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1082 DEBUG_PRINT2 (" ever_matched=%d", \ | |
1083 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1084 DEBUG_PRINT1 ("\n"); \ | |
1085 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \ | |
1086 } \ | |
1087 \ | |
1088 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\ | |
1089 PUSH_FAILURE_ITEM (lowest_active_reg); \ | |
1090 \ | |
1091 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\ | |
1092 PUSH_FAILURE_ITEM (highest_active_reg); \ | |
1093 \ | |
1094 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \ | |
1095 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ | |
1096 PUSH_FAILURE_ITEM (pattern_place); \ | |
1097 \ | |
1098 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \ | |
1099 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ | |
1100 size2); \ | |
1101 DEBUG_PRINT1 ("'\n"); \ | |
1102 PUSH_FAILURE_ITEM (string_place); \ | |
1103 \ | |
1104 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ | |
1105 DEBUG_PUSH (failure_id); \ | |
1106 } while (0) | |
1107 | |
1108 /* This is the number of items that are pushed and popped on the stack | |
1109 for each register. */ | |
1110 #define NUM_REG_ITEMS 3 | |
1111 | |
1112 /* Individual items aside from the registers. */ | |
1113 #ifdef DEBUG | |
1114 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ | |
1115 #else | |
1116 #define NUM_NONREG_ITEMS 4 | |
1117 #endif | |
1118 | |
1119 /* We push at most this many items on the stack. */ | |
1120 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS) | |
1121 | |
1122 /* We actually push this many items. */ | |
1123 #define NUM_FAILURE_ITEMS \ | |
1124 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \ | |
1125 + NUM_NONREG_ITEMS) | |
1126 | |
1127 /* How many items can still be added to the stack without overflowing it. */ | |
1128 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
1129 | |
1130 | |
1131 /* Pops what PUSH_FAIL_STACK pushes. | |
1132 | |
1133 We restore into the parameters, all of which should be lvalues: | |
1134 STR -- the saved data position. | |
1135 PAT -- the saved pattern position. | |
1136 LOW_REG, HIGH_REG -- the highest and lowest active registers. | |
1137 REGSTART, REGEND -- arrays of string positions. | |
1138 REG_INFO -- array of information about each subexpression. | |
1139 | |
1140 Also assumes the variables `fail_stack' and (if debugging), `bufp', | |
1141 `pend', `string1', `size1', `string2', and `size2'. */ | |
1142 | |
1143 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ | |
1144 { \ | |
1145 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \ | |
1146 int this_reg; \ | |
1147 const unsigned char *string_temp; \ | |
1148 \ | |
1149 assert (!FAIL_STACK_EMPTY ()); \ | |
1150 \ | |
1151 /* Remove failure points and point to how many regs pushed. */ \ | |
1152 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
1153 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
1154 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ | |
1155 \ | |
1156 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ | |
1157 \ | |
1158 DEBUG_POP (&failure_id); \ | |
1159 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ | |
1160 \ | |
1161 /* If the saved string location is NULL, it came from an \ | |
1162 on_failure_keep_string_jump opcode, and we want to throw away the \ | |
1163 saved NULL, thus retaining our current position in the string. */ \ | |
1164 string_temp = POP_FAILURE_ITEM (); \ | |
1165 if (string_temp != NULL) \ | |
1166 str = (const char *) string_temp; \ | |
1167 \ | |
1168 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \ | |
1169 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ | |
1170 DEBUG_PRINT1 ("'\n"); \ | |
1171 \ | |
1172 pat = (unsigned char *) POP_FAILURE_ITEM (); \ | |
1173 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \ | |
1174 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
1175 \ | |
1176 /* Restore register info. */ \ | |
1177 high_reg = (unsigned) POP_FAILURE_ITEM (); \ | |
1178 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \ | |
1179 \ | |
1180 low_reg = (unsigned) POP_FAILURE_ITEM (); \ | |
1181 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \ | |
1182 \ | |
1183 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ | |
1184 { \ | |
1185 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \ | |
1186 \ | |
1187 reg_info[this_reg].word = POP_FAILURE_ITEM (); \ | |
1188 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \ | |
1189 \ | |
1190 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \ | |
1191 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ | |
1192 \ | |
1193 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \ | |
1194 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ | |
1195 } \ | |
1196 \ | |
1197 DEBUG_STATEMENT (nfailure_points_popped++); \ | |
1198 } /* POP_FAILURE_POINT */ | |
1199 | |
1200 | |
1201 | |
1202 /* Structure for per-register (a.k.a. per-group) information. | |
1203 This must not be longer than one word, because we push this value | |
1204 onto the failure stack. Other register information, such as the | |
1205 starting and ending positions (which are addresses), and the list of | |
1206 inner groups (which is a bits list) are maintained in separate | |
1207 variables. | |
1208 | |
1209 We are making a (strictly speaking) nonportable assumption here: that | |
1210 the compiler will pack our bit fields into something that fits into | |
1211 the type of `word', i.e., is something that fits into one item on the | |
1212 failure stack. */ | |
1213 typedef union | |
1214 { | |
1215 fail_stack_elt_t word; | |
1216 struct | |
1217 { | |
1218 /* This field is one if this group can match the empty string, | |
1219 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ | |
1220 #define MATCH_NULL_UNSET_VALUE 3 | |
1221 unsigned match_null_string_p : 2; | |
1222 unsigned is_active : 1; | |
1223 unsigned matched_something : 1; | |
1224 unsigned ever_matched_something : 1; | |
1225 } bits; | |
1226 } register_info_type; | |
1227 | |
1228 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) | |
1229 #define IS_ACTIVE(R) ((R).bits.is_active) | |
1230 #define MATCHED_SOMETHING(R) ((R).bits.matched_something) | |
1231 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) | |
1232 | |
1233 | |
1234 /* Call this when have matched a real character; it sets `matched' flags | |
1235 for the subexpressions which we are currently inside. Also records | |
1236 that those subexprs have matched. */ | |
1237 #define SET_REGS_MATCHED() \ | |
1238 do \ | |
1239 { \ | |
1240 unsigned r; \ | |
1241 for (r = lowest_active_reg; r <= highest_active_reg; r++) \ | |
1242 { \ | |
1243 MATCHED_SOMETHING (reg_info[r]) \ | |
1244 = EVER_MATCHED_SOMETHING (reg_info[r]) \ | |
1245 = 1; \ | |
1246 } \ | |
1247 } \ | |
1248 while (0) | |
1249 | |
1250 | |
1251 /* Registers are set to a sentinel when they haven't yet matched. */ | |
1252 #define REG_UNSET_VALUE ((char *) -1) | |
1253 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE) | |
1254 | |
1255 | |
1256 | |
2952 | 1257 /* How do we implement a missing MATCH_MAY_ALLOCATE? |
2949 | 1258 We make the fail stack a global thing, and then grow it to |
1259 re_max_failures when we compile. */ | |
2952 | 1260 #ifndef MATCH_MAY_ALLOCATE |
2949 | 1261 static fail_stack_type fail_stack; |
1262 | |
1263 static const char ** regstart, ** regend; | |
1264 static const char ** old_regstart, ** old_regend; | |
1265 static const char **best_regstart, **best_regend; | |
1266 static register_info_type *reg_info; | |
1267 static const char **reg_dummy; | |
1268 static register_info_type *reg_info_dummy; | |
1269 #endif | |
1270 | |
1271 | |
1155 | 1272 /* Subroutine declarations and macros for regex_compile. */ |
1273 | |
1274 static void store_op1 (), store_op2 (); | |
1275 static void insert_op1 (), insert_op2 (); | |
1276 static boolean at_begline_loc_p (), at_endline_loc_p (); | |
1277 static boolean group_in_compile_stack (); | |
1278 static reg_errcode_t compile_range (); | |
1279 | |
1280 /* Fetch the next character in the uncompiled pattern---translating it | |
1281 if necessary. Also cast from a signed character in the constant | |
1282 string passed to us by the user to an unsigned char that we can use | |
1283 as an array index (in, e.g., `translate'). */ | |
1284 #define PATFETCH(c) \ | |
1285 do {if (p == pend) return REG_EEND; \ | |
1286 c = (unsigned char) *p++; \ | |
1287 if (translate) c = translate[c]; \ | |
1288 } while (0) | |
1289 | |
1290 /* Fetch the next character in the uncompiled pattern, with no | |
1291 translation. */ | |
1292 #define PATFETCH_RAW(c) \ | |
1293 do {if (p == pend) return REG_EEND; \ | |
1294 c = (unsigned char) *p++; \ | |
1295 } while (0) | |
1296 | |
1297 /* Go backwards one character in the pattern. */ | |
1298 #define PATUNFETCH p-- | |
1299 | |
1300 | |
1301 /* If `translate' is non-null, return translate[D], else just D. We | |
1302 cast the subscript to translate because some data is declared as | |
1303 `char *', to avoid warnings when a string constant is passed. But | |
1304 when we use a character as a subscript we must make it unsigned. */ | |
1305 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d)) | |
1306 | |
1307 | |
1308 /* Macros for outputting the compiled pattern into `buffer'. */ | |
1309 | |
1310 /* If the buffer isn't allocated when it comes in, use this. */ | |
1311 #define INIT_BUF_SIZE 32 | |
1312 | |
1313 /* Make sure we have at least N more bytes of space in buffer. */ | |
1314 #define GET_BUFFER_SPACE(n) \ | |
1315 while (b - bufp->buffer + (n) > bufp->allocated) \ | |
1316 EXTEND_BUFFER () | |
1317 | |
1318 /* Make sure we have one more byte of buffer space and then add C to it. */ | |
1319 #define BUF_PUSH(c) \ | |
1320 do { \ | |
1321 GET_BUFFER_SPACE (1); \ | |
1322 *b++ = (unsigned char) (c); \ | |
1323 } while (0) | |
1324 | |
1325 | |
1326 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
1327 #define BUF_PUSH_2(c1, c2) \ | |
1328 do { \ | |
1329 GET_BUFFER_SPACE (2); \ | |
1330 *b++ = (unsigned char) (c1); \ | |
1331 *b++ = (unsigned char) (c2); \ | |
1332 } while (0) | |
1333 | |
1334 | |
1335 /* As with BUF_PUSH_2, except for three bytes. */ | |
1336 #define BUF_PUSH_3(c1, c2, c3) \ | |
1337 do { \ | |
1338 GET_BUFFER_SPACE (3); \ | |
1339 *b++ = (unsigned char) (c1); \ | |
1340 *b++ = (unsigned char) (c2); \ | |
1341 *b++ = (unsigned char) (c3); \ | |
1342 } while (0) | |
1343 | |
1344 | |
1345 /* Store a jump with opcode OP at LOC to location TO. We store a | |
1346 relative address offset by the three bytes the jump itself occupies. */ | |
1347 #define STORE_JUMP(op, loc, to) \ | |
1348 store_op1 (op, loc, (to) - (loc) - 3) | |
1349 | |
1350 /* Likewise, for a two-argument jump. */ | |
1351 #define STORE_JUMP2(op, loc, to, arg) \ | |
1352 store_op2 (op, loc, (to) - (loc) - 3, arg) | |
1353 | |
1354 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ | |
1355 #define INSERT_JUMP(op, loc, to) \ | |
1356 insert_op1 (op, loc, (to) - (loc) - 3, b) | |
1357 | |
1358 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
1359 #define INSERT_JUMP2(op, loc, to, arg) \ | |
1360 insert_op2 (op, loc, (to) - (loc) - 3, arg, b) | |
1361 | |
1362 | |
1363 /* This is not an arbitrary limit: the arguments which represent offsets | |
1364 into the pattern are two bytes long. So if 2^16 bytes turns out to | |
1365 be too small, many things would have to change. */ | |
1366 #define MAX_BUF_SIZE (1L << 16) | |
1367 | |
1368 | |
1369 /* Extend the buffer by twice its current size via realloc and | |
1370 reset the pointers that pointed into the old block to point to the | |
1371 correct places in the new one. If extending the buffer results in it | |
1372 being larger than MAX_BUF_SIZE, then flag memory exhausted. */ | |
1373 #define EXTEND_BUFFER() \ | |
1374 do { \ | |
1375 unsigned char *old_buffer = bufp->buffer; \ | |
1376 if (bufp->allocated == MAX_BUF_SIZE) \ | |
1377 return REG_ESIZE; \ | |
1378 bufp->allocated <<= 1; \ | |
1379 if (bufp->allocated > MAX_BUF_SIZE) \ | |
1380 bufp->allocated = MAX_BUF_SIZE; \ | |
1381 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\ | |
1382 if (bufp->buffer == NULL) \ | |
1383 return REG_ESPACE; \ | |
1384 /* If the buffer moved, move all the pointers into it. */ \ | |
1385 if (old_buffer != bufp->buffer) \ | |
1386 { \ | |
1387 b = (b - old_buffer) + bufp->buffer; \ | |
1388 begalt = (begalt - old_buffer) + bufp->buffer; \ | |
1389 if (fixup_alt_jump) \ | |
1390 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\ | |
1391 if (laststart) \ | |
1392 laststart = (laststart - old_buffer) + bufp->buffer; \ | |
1393 if (pending_exact) \ | |
1394 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \ | |
1395 } \ | |
1396 } while (0) | |
1397 | |
1398 | |
1399 /* Since we have one byte reserved for the register number argument to | |
1400 {start,stop}_memory, the maximum number of groups we can report | |
1401 things about is what fits in that byte. */ | |
1402 #define MAX_REGNUM 255 | |
1403 | |
1404 /* But patterns can have more than `MAX_REGNUM' registers. We just | |
1405 ignore the excess. */ | |
1406 typedef unsigned regnum_t; | |
1407 | |
1408 | |
1409 /* Macros for the compile stack. */ | |
1410 | |
1411 /* Since offsets can go either forwards or backwards, this type needs to | |
1412 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ | |
1413 typedef int pattern_offset_t; | |
1414 | |
1415 typedef struct | |
1416 { | |
1417 pattern_offset_t begalt_offset; | |
1418 pattern_offset_t fixup_alt_jump; | |
1419 pattern_offset_t inner_group_offset; | |
1420 pattern_offset_t laststart_offset; | |
1421 regnum_t regnum; | |
1422 } compile_stack_elt_t; | |
1423 | |
1424 | |
1425 typedef struct | |
1426 { | |
1427 compile_stack_elt_t *stack; | |
1428 unsigned size; | |
1429 unsigned avail; /* Offset of next open position. */ | |
1430 } compile_stack_type; | |
1431 | |
1432 | |
1433 #define INIT_COMPILE_STACK_SIZE 32 | |
1434 | |
1435 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
1436 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
1437 | |
1438 /* The next available element. */ | |
1439 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) | |
1440 | |
1441 | |
1442 /* Set the bit for character C in a list. */ | |
1443 #define SET_LIST_BIT(c) \ | |
1444 (b[((unsigned char) (c)) / BYTEWIDTH] \ | |
1445 |= 1 << (((unsigned char) c) % BYTEWIDTH)) | |
1446 | |
1447 | |
1448 /* Get the next unsigned number in the uncompiled pattern. */ | |
1449 #define GET_UNSIGNED_NUMBER(num) \ | |
1450 { if (p != pend) \ | |
1451 { \ | |
1452 PATFETCH (c); \ | |
1668 | 1453 while (ISDIGIT (c)) \ |
1155 | 1454 { \ |
1455 if (num < 0) \ | |
1456 num = 0; \ | |
1457 num = num * 10 + c - '0'; \ | |
1458 if (p == pend) \ | |
1459 break; \ | |
1460 PATFETCH (c); \ | |
1461 } \ | |
1462 } \ | |
1463 } | |
1464 | |
1465 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ | |
1466 | |
1467 #define IS_CHAR_CLASS(string) \ | |
1468 (STREQ (string, "alpha") || STREQ (string, "upper") \ | |
1469 || STREQ (string, "lower") || STREQ (string, "digit") \ | |
1470 || STREQ (string, "alnum") || STREQ (string, "xdigit") \ | |
1471 || STREQ (string, "space") || STREQ (string, "print") \ | |
1472 || STREQ (string, "punct") || STREQ (string, "graph") \ | |
1473 || STREQ (string, "cntrl") || STREQ (string, "blank")) | |
1474 | |
1475 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. | |
1476 Returns one of error codes defined in `regex.h', or zero for success. | |
1477 | |
1478 Assumes the `allocated' (and perhaps `buffer') and `translate' | |
1479 fields are set in BUFP on entry. | |
1480 | |
1481 If it succeeds, results are put in BUFP (if it returns an error, the | |
1482 contents of BUFP are undefined): | |
1483 `buffer' is the compiled pattern; | |
1484 `syntax' is set to SYNTAX; | |
1485 `used' is set to the length of the compiled pattern; | |
1637 | 1486 `fastmap_accurate' is zero; |
1487 `re_nsub' is the number of subexpressions in PATTERN; | |
1488 `not_bol' and `not_eol' are zero; | |
1155 | 1489 |
1490 The `fastmap' and `newline_anchor' fields are neither | |
1491 examined nor set. */ | |
1492 | |
1493 static reg_errcode_t | |
1494 regex_compile (pattern, size, syntax, bufp) | |
1495 const char *pattern; | |
1496 int size; | |
1497 reg_syntax_t syntax; | |
1498 struct re_pattern_buffer *bufp; | |
1499 { | |
1500 /* We fetch characters from PATTERN here. Even though PATTERN is | |
1501 `char *' (i.e., signed), we declare these variables as unsigned, so | |
1502 they can be reliably used as array indices. */ | |
1503 register unsigned char c, c1; | |
1504 | |
5841 | 1505 /* A random temporary spot in PATTERN. */ |
1155 | 1506 const char *p1; |
1507 | |
1508 /* Points to the end of the buffer, where we should append. */ | |
1509 register unsigned char *b; | |
1510 | |
1511 /* Keeps track of unclosed groups. */ | |
1512 compile_stack_type compile_stack; | |
1513 | |
1514 /* Points to the current (ending) position in the pattern. */ | |
1515 const char *p = pattern; | |
1516 const char *pend = pattern + size; | |
1517 | |
1518 /* How to translate the characters in the pattern. */ | |
1519 char *translate = bufp->translate; | |
1520 | |
1521 /* Address of the count-byte of the most recently inserted `exactn' | |
1522 command. This makes it possible to tell if a new exact-match | |
1523 character can be added to that command or if the character requires | |
1524 a new `exactn' command. */ | |
1525 unsigned char *pending_exact = 0; | |
1526 | |
1527 /* Address of start of the most recently finished expression. | |
1528 This tells, e.g., postfix * where to find the start of its | |
1529 operand. Reset at the beginning of groups and alternatives. */ | |
1530 unsigned char *laststart = 0; | |
1531 | |
1532 /* Address of beginning of regexp, or inside of last group. */ | |
1533 unsigned char *begalt; | |
1534 | |
1535 /* Place in the uncompiled pattern (i.e., the {) to | |
1536 which to go back if the interval is invalid. */ | |
1537 const char *beg_interval; | |
1538 | |
1539 /* Address of the place where a forward jump should go to the end of | |
1540 the containing expression. Each alternative of an `or' -- except the | |
1541 last -- ends with a forward jump of this sort. */ | |
1542 unsigned char *fixup_alt_jump = 0; | |
1543 | |
1544 /* Counts open-groups as they are encountered. Remembered for the | |
1545 matching close-group on the compile stack, so the same register | |
1546 number is put in the stop_memory as the start_memory. */ | |
1547 regnum_t regnum = 0; | |
1548 | |
1549 #ifdef DEBUG | |
1550 DEBUG_PRINT1 ("\nCompiling pattern: "); | |
1551 if (debug) | |
1552 { | |
1553 unsigned debug_count; | |
1554 | |
1555 for (debug_count = 0; debug_count < size; debug_count++) | |
1556 printchar (pattern[debug_count]); | |
1557 putchar ('\n'); | |
1558 } | |
1559 #endif /* DEBUG */ | |
1560 | |
1561 /* Initialize the compile stack. */ | |
1562 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
1563 if (compile_stack.stack == NULL) | |
1564 return REG_ESPACE; | |
1565 | |
1566 compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
1567 compile_stack.avail = 0; | |
1568 | |
1569 /* Initialize the pattern buffer. */ | |
1570 bufp->syntax = syntax; | |
1571 bufp->fastmap_accurate = 0; | |
1572 bufp->not_bol = bufp->not_eol = 0; | |
1573 | |
1574 /* Set `used' to zero, so that if we return an error, the pattern | |
1575 printer (for debugging) will think there's no pattern. We reset it | |
1576 at the end. */ | |
1577 bufp->used = 0; | |
1578 | |
1579 /* Always count groups, whether or not bufp->no_sub is set. */ | |
1580 bufp->re_nsub = 0; | |
1581 | |
1582 #if !defined (emacs) && !defined (SYNTAX_TABLE) | |
1583 /* Initialize the syntax table. */ | |
1584 init_syntax_once (); | |
1585 #endif | |
1586 | |
1587 if (bufp->allocated == 0) | |
1588 { | |
1589 if (bufp->buffer) | |
1590 { /* If zero allocated, but buffer is non-null, try to realloc | |
1591 enough space. This loses if buffer's address is bogus, but | |
1592 that is the user's responsibility. */ | |
1593 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); | |
1594 } | |
1595 else | |
1596 { /* Caller did not allocate a buffer. Do it for them. */ | |
1597 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); | |
1598 } | |
1599 if (!bufp->buffer) return REG_ESPACE; | |
1600 | |
1601 bufp->allocated = INIT_BUF_SIZE; | |
1602 } | |
1603 | |
1604 begalt = b = bufp->buffer; | |
1605 | |
1606 /* Loop through the uncompiled pattern until we're at the end. */ | |
1607 while (p != pend) | |
1608 { | |
1609 PATFETCH (c); | |
1610 | |
1611 switch (c) | |
1612 { | |
1613 case '^': | |
1614 { | |
1615 if ( /* If at start of pattern, it's an operator. */ | |
1616 p == pattern + 1 | |
1617 /* If context independent, it's an operator. */ | |
1618 || syntax & RE_CONTEXT_INDEP_ANCHORS | |
1619 /* Otherwise, depends on what's come before. */ | |
1620 || at_begline_loc_p (pattern, p, syntax)) | |
1621 BUF_PUSH (begline); | |
1622 else | |
1623 goto normal_char; | |
1624 } | |
1625 break; | |
1626 | |
1627 | |
1628 case '$': | |
1629 { | |
1630 if ( /* If at end of pattern, it's an operator. */ | |
1631 p == pend | |
1632 /* If context independent, it's an operator. */ | |
1633 || syntax & RE_CONTEXT_INDEP_ANCHORS | |
1634 /* Otherwise, depends on what's next. */ | |
1635 || at_endline_loc_p (p, pend, syntax)) | |
1636 BUF_PUSH (endline); | |
1637 else | |
1638 goto normal_char; | |
1639 } | |
1640 break; | |
1641 | |
1642 | |
1643 case '+': | |
1644 case '?': | |
1645 if ((syntax & RE_BK_PLUS_QM) | |
1646 || (syntax & RE_LIMITED_OPS)) | |
1647 goto normal_char; | |
1648 handle_plus: | |
1649 case '*': | |
1650 /* If there is no previous pattern... */ | |
1651 if (!laststart) | |
1652 { | |
1653 if (syntax & RE_CONTEXT_INVALID_OPS) | |
1654 return REG_BADRPT; | |
1655 else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
1656 goto normal_char; | |
1657 } | |
1658 | |
1659 { | |
1660 /* Are we optimizing this jump? */ | |
1661 boolean keep_string_p = false; | |
1662 | |
1663 /* 1 means zero (many) matches is allowed. */ | |
1664 char zero_times_ok = 0, many_times_ok = 0; | |
1665 | |
1666 /* If there is a sequence of repetition chars, collapse it | |
1667 down to just one (the right one). We can't combine | |
1668 interval operators with these because of, e.g., `a{2}*', | |
1669 which should only match an even number of `a's. */ | |
1670 | |
1671 for (;;) | |
1672 { | |
1673 zero_times_ok |= c != '+'; | |
1674 many_times_ok |= c != '?'; | |
1675 | |
1676 if (p == pend) | |
1677 break; | |
1678 | |
1679 PATFETCH (c); | |
1680 | |
1681 if (c == '*' | |
1682 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) | |
1683 ; | |
1684 | |
1685 else if (syntax & RE_BK_PLUS_QM && c == '\\') | |
1686 { | |
1687 if (p == pend) return REG_EESCAPE; | |
1688 | |
1689 PATFETCH (c1); | |
1690 if (!(c1 == '+' || c1 == '?')) | |
1691 { | |
1692 PATUNFETCH; | |
1693 PATUNFETCH; | |
1694 break; | |
1695 } | |
1696 | |
1697 c = c1; | |
1698 } | |
1699 else | |
1700 { | |
1701 PATUNFETCH; | |
1702 break; | |
1703 } | |
1704 | |
1705 /* If we get here, we found another repeat character. */ | |
1706 } | |
1707 | |
1708 /* Star, etc. applied to an empty pattern is equivalent | |
1709 to an empty pattern. */ | |
1710 if (!laststart) | |
1711 break; | |
1712 | |
1713 /* Now we know whether or not zero matches is allowed | |
1714 and also whether or not two or more matches is allowed. */ | |
1715 if (many_times_ok) | |
1716 { /* More than one repetition is allowed, so put in at the | |
1717 end a backward relative jump from `b' to before the next | |
1718 jump we're going to put in below (which jumps from | |
1719 laststart to after this jump). | |
1720 | |
1721 But if we are at the `*' in the exact sequence `.*\n', | |
1722 insert an unconditional jump backwards to the ., | |
1723 instead of the beginning of the loop. This way we only | |
1724 push a failure point once, instead of every time | |
1725 through the loop. */ | |
1726 assert (p - 1 > pattern); | |
1727 | |
1728 /* Allocate the space for the jump. */ | |
1729 GET_BUFFER_SPACE (3); | |
1730 | |
1731 /* We know we are not at the first character of the pattern, | |
1732 because laststart was nonzero. And we've already | |
1733 incremented `p', by the way, to be the character after | |
1734 the `*'. Do we have to do something analogous here | |
1735 for null bytes, because of RE_DOT_NOT_NULL? */ | |
1736 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.') | |
2453 | 1737 && zero_times_ok |
1155 | 1738 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n') |
1739 && !(syntax & RE_DOT_NEWLINE)) | |
1740 { /* We have .*\n. */ | |
1741 STORE_JUMP (jump, b, laststart); | |
1742 keep_string_p = true; | |
1743 } | |
1744 else | |
1745 /* Anything else. */ | |
1746 STORE_JUMP (maybe_pop_jump, b, laststart - 3); | |
1747 | |
1748 /* We've added more stuff to the buffer. */ | |
1749 b += 3; | |
1750 } | |
1751 | |
1752 /* On failure, jump from laststart to b + 3, which will be the | |
1753 end of the buffer after this jump is inserted. */ | |
1754 GET_BUFFER_SPACE (3); | |
1755 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump | |
1756 : on_failure_jump, | |
1757 laststart, b + 3); | |
1758 pending_exact = 0; | |
1759 b += 3; | |
1760 | |
1761 if (!zero_times_ok) | |
1762 { | |
1763 /* At least one repetition is required, so insert a | |
1764 `dummy_failure_jump' before the initial | |
1765 `on_failure_jump' instruction of the loop. This | |
1766 effects a skip over that instruction the first time | |
1767 we hit that loop. */ | |
1768 GET_BUFFER_SPACE (3); | |
1769 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6); | |
1770 b += 3; | |
1771 } | |
1772 } | |
1773 break; | |
1774 | |
1775 | |
1776 case '.': | |
1777 laststart = b; | |
1778 BUF_PUSH (anychar); | |
1779 break; | |
1780 | |
1781 | |
1782 case '[': | |
1783 { | |
1784 boolean had_char_class = false; | |
1785 | |
1786 if (p == pend) return REG_EBRACK; | |
1787 | |
1788 /* Ensure that we have enough space to push a charset: the | |
1789 opcode, the length count, and the bitset; 34 bytes in all. */ | |
1790 GET_BUFFER_SPACE (34); | |
1791 | |
1792 laststart = b; | |
1793 | |
1794 /* We test `*p == '^' twice, instead of using an if | |
1795 statement, so we only need one BUF_PUSH. */ | |
1796 BUF_PUSH (*p == '^' ? charset_not : charset); | |
1797 if (*p == '^') | |
1798 p++; | |
1799 | |
1800 /* Remember the first position in the bracket expression. */ | |
1801 p1 = p; | |
1802 | |
1803 /* Push the number of bytes in the bitmap. */ | |
1804 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
1805 | |
1806 /* Clear the whole map. */ | |
1807 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); | |
1808 | |
1809 /* charset_not matches newline according to a syntax bit. */ | |
1810 if ((re_opcode_t) b[-2] == charset_not | |
1811 && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
1812 SET_LIST_BIT ('\n'); | |
1813 | |
1814 /* Read in characters and ranges, setting map bits. */ | |
1815 for (;;) | |
1816 { | |
1817 if (p == pend) return REG_EBRACK; | |
1818 | |
1819 PATFETCH (c); | |
1820 | |
1821 /* \ might escape characters inside [...] and [^...]. */ | |
1822 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
1823 { | |
1824 if (p == pend) return REG_EESCAPE; | |
1825 | |
1826 PATFETCH (c1); | |
1827 SET_LIST_BIT (c1); | |
1828 continue; | |
1829 } | |
1830 | |
1831 /* Could be the end of the bracket expression. If it's | |
1832 not (i.e., when the bracket expression is `[]' so | |
1833 far), the ']' character bit gets set way below. */ | |
1834 if (c == ']' && p != p1 + 1) | |
1835 break; | |
1836 | |
1837 /* Look ahead to see if it's a range when the last thing | |
1838 was a character class. */ | |
1839 if (had_char_class && c == '-' && *p != ']') | |
1840 return REG_ERANGE; | |
1841 | |
1842 /* Look ahead to see if it's a range when the last thing | |
1843 was a character: if this is a hyphen not at the | |
1844 beginning or the end of a list, then it's the range | |
1845 operator. */ | |
1846 if (c == '-' | |
1847 && !(p - 2 >= pattern && p[-2] == '[') | |
1848 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') | |
1849 && *p != ']') | |
1850 { | |
1851 reg_errcode_t ret | |
1852 = compile_range (&p, pend, translate, syntax, b); | |
1853 if (ret != REG_NOERROR) return ret; | |
1854 } | |
1855 | |
1856 else if (p[0] == '-' && p[1] != ']') | |
1857 { /* This handles ranges made up of characters only. */ | |
1858 reg_errcode_t ret; | |
1859 | |
1860 /* Move past the `-'. */ | |
1861 PATFETCH (c1); | |
1862 | |
1863 ret = compile_range (&p, pend, translate, syntax, b); | |
1864 if (ret != REG_NOERROR) return ret; | |
1865 } | |
1866 | |
1867 /* See if we're at the beginning of a possible character | |
1868 class. */ | |
1869 | |
1870 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
1871 { /* Leave room for the null. */ | |
1872 char str[CHAR_CLASS_MAX_LENGTH + 1]; | |
1873 | |
1874 PATFETCH (c); | |
1875 c1 = 0; | |
1876 | |
1877 /* If pattern is `[[:'. */ | |
1878 if (p == pend) return REG_EBRACK; | |
1879 | |
1880 for (;;) | |
1881 { | |
1882 PATFETCH (c); | |
1883 if (c == ':' || c == ']' || p == pend | |
1884 || c1 == CHAR_CLASS_MAX_LENGTH) | |
1885 break; | |
1886 str[c1++] = c; | |
1887 } | |
1888 str[c1] = '\0'; | |
1889 | |
1890 /* If isn't a word bracketed by `[:' and:`]': | |
1891 undo the ending character, the letters, and leave | |
1892 the leading `:' and `[' (but set bits for them). */ | |
1893 if (c == ':' && *p == ']') | |
1894 { | |
1895 int ch; | |
1896 boolean is_alnum = STREQ (str, "alnum"); | |
1897 boolean is_alpha = STREQ (str, "alpha"); | |
1898 boolean is_blank = STREQ (str, "blank"); | |
1899 boolean is_cntrl = STREQ (str, "cntrl"); | |
1900 boolean is_digit = STREQ (str, "digit"); | |
1901 boolean is_graph = STREQ (str, "graph"); | |
1902 boolean is_lower = STREQ (str, "lower"); | |
1903 boolean is_print = STREQ (str, "print"); | |
1904 boolean is_punct = STREQ (str, "punct"); | |
1905 boolean is_space = STREQ (str, "space"); | |
1906 boolean is_upper = STREQ (str, "upper"); | |
1907 boolean is_xdigit = STREQ (str, "xdigit"); | |
1908 | |
1909 if (!IS_CHAR_CLASS (str)) return REG_ECTYPE; | |
1910 | |
1911 /* Throw away the ] at the end of the character | |
1912 class. */ | |
1913 PATFETCH (c); | |
1914 | |
1915 if (p == pend) return REG_EBRACK; | |
1916 | |
1917 for (ch = 0; ch < 1 << BYTEWIDTH; ch++) | |
1918 { | |
1668 | 1919 if ( (is_alnum && ISALNUM (ch)) |
1920 || (is_alpha && ISALPHA (ch)) | |
1921 || (is_blank && ISBLANK (ch)) | |
1922 || (is_cntrl && ISCNTRL (ch)) | |
1923 || (is_digit && ISDIGIT (ch)) | |
1924 || (is_graph && ISGRAPH (ch)) | |
1925 || (is_lower && ISLOWER (ch)) | |
1926 || (is_print && ISPRINT (ch)) | |
1927 || (is_punct && ISPUNCT (ch)) | |
1928 || (is_space && ISSPACE (ch)) | |
1929 || (is_upper && ISUPPER (ch)) | |
1930 || (is_xdigit && ISXDIGIT (ch))) | |
1155 | 1931 SET_LIST_BIT (ch); |
1932 } | |
1933 had_char_class = true; | |
1934 } | |
1935 else | |
1936 { | |
1937 c1++; | |
1938 while (c1--) | |
1939 PATUNFETCH; | |
1940 SET_LIST_BIT ('['); | |
1941 SET_LIST_BIT (':'); | |
1942 had_char_class = false; | |
1943 } | |
1944 } | |
1945 else | |
1946 { | |
1947 had_char_class = false; | |
1948 SET_LIST_BIT (c); | |
1949 } | |
1950 } | |
1951 | |
1952 /* Discard any (non)matching list bytes that are all 0 at the | |
1953 end of the map. Decrease the map-length byte too. */ | |
1954 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) | |
1955 b[-1]--; | |
1956 b += b[-1]; | |
1957 } | |
1958 break; | |
1959 | |
1960 | |
1961 case '(': | |
1962 if (syntax & RE_NO_BK_PARENS) | |
1963 goto handle_open; | |
1964 else | |
1965 goto normal_char; | |
1966 | |
1967 | |
1968 case ')': | |
1969 if (syntax & RE_NO_BK_PARENS) | |
1970 goto handle_close; | |
1971 else | |
1972 goto normal_char; | |
1973 | |
1974 | |
1975 case '\n': | |
1976 if (syntax & RE_NEWLINE_ALT) | |
1977 goto handle_alt; | |
1978 else | |
1979 goto normal_char; | |
1980 | |
1981 | |
1982 case '|': | |
1983 if (syntax & RE_NO_BK_VBAR) | |
1984 goto handle_alt; | |
1985 else | |
1986 goto normal_char; | |
1987 | |
1988 | |
1989 case '{': | |
1990 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
1991 goto handle_interval; | |
1992 else | |
1993 goto normal_char; | |
1994 | |
1995 | |
1996 case '\\': | |
1997 if (p == pend) return REG_EESCAPE; | |
1998 | |
1999 /* Do not translate the character after the \, so that we can | |
2000 distinguish, e.g., \B from \b, even if we normally would | |
2001 translate, e.g., B to b. */ | |
2002 PATFETCH_RAW (c); | |
2003 | |
2004 switch (c) | |
2005 { | |
2006 case '(': | |
2007 if (syntax & RE_NO_BK_PARENS) | |
2008 goto normal_backslash; | |
2009 | |
2010 handle_open: | |
2011 bufp->re_nsub++; | |
2012 regnum++; | |
2013 | |
2014 if (COMPILE_STACK_FULL) | |
2015 { | |
2016 RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
2017 compile_stack_elt_t); | |
2018 if (compile_stack.stack == NULL) return REG_ESPACE; | |
2019 | |
2020 compile_stack.size <<= 1; | |
2021 } | |
2022 | |
2023 /* These are the values to restore when we hit end of this | |
2024 group. They are all relative offsets, so that if the | |
2025 whole pattern moves because of realloc, they will still | |
2026 be valid. */ | |
2027 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; | |
2028 COMPILE_STACK_TOP.fixup_alt_jump | |
2029 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; | |
2030 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; | |
2031 COMPILE_STACK_TOP.regnum = regnum; | |
2032 | |
2033 /* We will eventually replace the 0 with the number of | |
2034 groups inner to this one. But do not push a | |
2035 start_memory for groups beyond the last one we can | |
2036 represent in the compiled pattern. */ | |
2037 if (regnum <= MAX_REGNUM) | |
2038 { | |
2039 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2; | |
2040 BUF_PUSH_3 (start_memory, regnum, 0); | |
2041 } | |
2042 | |
2043 compile_stack.avail++; | |
2044 | |
2045 fixup_alt_jump = 0; | |
2046 laststart = 0; | |
2047 begalt = b; | |
2453 | 2048 /* If we've reached MAX_REGNUM groups, then this open |
2049 won't actually generate any code, so we'll have to | |
2050 clear pending_exact explicitly. */ | |
2051 pending_exact = 0; | |
1155 | 2052 break; |
2053 | |
2054 | |
2055 case ')': | |
2056 if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
2057 | |
2058 if (COMPILE_STACK_EMPTY) | |
2059 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
2060 goto normal_backslash; | |
2061 else | |
2062 return REG_ERPAREN; | |
2063 | |
2064 handle_close: | |
2065 if (fixup_alt_jump) | |
2066 { /* Push a dummy failure point at the end of the | |
2067 alternative for a possible future | |
2068 `pop_failure_jump' to pop. See comments at | |
2069 `push_dummy_failure' in `re_match_2'. */ | |
2070 BUF_PUSH (push_dummy_failure); | |
2071 | |
2072 /* We allocated space for this jump when we assigned | |
2073 to `fixup_alt_jump', in the `handle_alt' case below. */ | |
2074 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); | |
2075 } | |
2076 | |
2077 /* See similar code for backslashed left paren above. */ | |
2078 if (COMPILE_STACK_EMPTY) | |
2079 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
2080 goto normal_char; | |
2081 else | |
2082 return REG_ERPAREN; | |
2083 | |
2084 /* Since we just checked for an empty stack above, this | |
2085 ``can't happen''. */ | |
2086 assert (compile_stack.avail != 0); | |
2087 { | |
2088 /* We don't just want to restore into `regnum', because | |
2089 later groups should continue to be numbered higher, | |
2090 as in `(ab)c(de)' -- the second group is #2. */ | |
2091 regnum_t this_group_regnum; | |
2092 | |
2093 compile_stack.avail--; | |
2094 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; | |
2095 fixup_alt_jump | |
2096 = COMPILE_STACK_TOP.fixup_alt_jump | |
2097 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
2098 : 0; | |
2099 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; | |
2100 this_group_regnum = COMPILE_STACK_TOP.regnum; | |
2453 | 2101 /* If we've reached MAX_REGNUM groups, then this open |
2102 won't actually generate any code, so we'll have to | |
2103 clear pending_exact explicitly. */ | |
2104 pending_exact = 0; | |
1155 | 2105 |
2106 /* We're at the end of the group, so now we know how many | |
2107 groups were inside this one. */ | |
2108 if (this_group_regnum <= MAX_REGNUM) | |
2109 { | |
2110 unsigned char *inner_group_loc | |
2111 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset; | |
2112 | |
2113 *inner_group_loc = regnum - this_group_regnum; | |
2114 BUF_PUSH_3 (stop_memory, this_group_regnum, | |
2115 regnum - this_group_regnum); | |
2116 } | |
2117 } | |
2118 break; | |
2119 | |
2120 | |
2121 case '|': /* `\|'. */ | |
2122 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
2123 goto normal_backslash; | |
2124 handle_alt: | |
2125 if (syntax & RE_LIMITED_OPS) | |
2126 goto normal_char; | |
2127 | |
2128 /* Insert before the previous alternative a jump which | |
2129 jumps to this alternative if the former fails. */ | |
2130 GET_BUFFER_SPACE (3); | |
2131 INSERT_JUMP (on_failure_jump, begalt, b + 6); | |
2132 pending_exact = 0; | |
2133 b += 3; | |
2134 | |
2135 /* The alternative before this one has a jump after it | |
2136 which gets executed if it gets matched. Adjust that | |
2137 jump so it will jump to this alternative's analogous | |
2138 jump (put in below, which in turn will jump to the next | |
2139 (if any) alternative's such jump, etc.). The last such | |
2140 jump jumps to the correct final destination. A picture: | |
2141 _____ _____ | |
2142 | | | | | |
2143 | v | v | |
2144 a | b | c | |
2145 | |
1637 | 2146 If we are at `b', then fixup_alt_jump right now points to a |
2147 three-byte space after `a'. We'll put in the jump, set | |
2148 fixup_alt_jump to right after `b', and leave behind three | |
2149 bytes which we'll fill in when we get to after `c'. */ | |
1155 | 2150 |
2151 if (fixup_alt_jump) | |
2152 STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
2153 | |
2154 /* Mark and leave space for a jump after this alternative, | |
2155 to be filled in later either by next alternative or | |
2156 when know we're at the end of a series of alternatives. */ | |
2157 fixup_alt_jump = b; | |
2158 GET_BUFFER_SPACE (3); | |
2159 b += 3; | |
2160 | |
2161 laststart = 0; | |
2162 begalt = b; | |
2163 break; | |
2164 | |
2165 | |
2166 case '{': | |
2167 /* If \{ is a literal. */ | |
2168 if (!(syntax & RE_INTERVALS) | |
2169 /* If we're at `\{' and it's not the open-interval | |
2170 operator. */ | |
2171 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES)) | |
2172 || (p - 2 == pattern && p == pend)) | |
2173 goto normal_backslash; | |
2174 | |
2175 handle_interval: | |
2176 { | |
2177 /* If got here, then the syntax allows intervals. */ | |
2178 | |
2179 /* At least (most) this many matches must be made. */ | |
2180 int lower_bound = -1, upper_bound = -1; | |
2181 | |
2182 beg_interval = p - 1; | |
2183 | |
2184 if (p == pend) | |
2185 { | |
2186 if (syntax & RE_NO_BK_BRACES) | |
2187 goto unfetch_interval; | |
2188 else | |
2189 return REG_EBRACE; | |
2190 } | |
2191 | |
2192 GET_UNSIGNED_NUMBER (lower_bound); | |
2193 | |
2194 if (c == ',') | |
2195 { | |
2196 GET_UNSIGNED_NUMBER (upper_bound); | |
2197 if (upper_bound < 0) upper_bound = RE_DUP_MAX; | |
2198 } | |
2199 else | |
2200 /* Interval such as `{1}' => match exactly once. */ | |
2201 upper_bound = lower_bound; | |
2202 | |
2203 if (lower_bound < 0 || upper_bound > RE_DUP_MAX | |
2204 || lower_bound > upper_bound) | |
2205 { | |
2206 if (syntax & RE_NO_BK_BRACES) | |
2207 goto unfetch_interval; | |
2208 else | |
2209 return REG_BADBR; | |
2210 } | |
2211 | |
2212 if (!(syntax & RE_NO_BK_BRACES)) | |
2213 { | |
2214 if (c != '\\') return REG_EBRACE; | |
2215 | |
2216 PATFETCH (c); | |
2217 } | |
2218 | |
2219 if (c != '}') | |
2220 { | |
2221 if (syntax & RE_NO_BK_BRACES) | |
2222 goto unfetch_interval; | |
2223 else | |
2224 return REG_BADBR; | |
2225 } | |
2226 | |
2227 /* We just parsed a valid interval. */ | |
2228 | |
2229 /* If it's invalid to have no preceding re. */ | |
2230 if (!laststart) | |
2231 { | |
2232 if (syntax & RE_CONTEXT_INVALID_OPS) | |
2233 return REG_BADRPT; | |
2234 else if (syntax & RE_CONTEXT_INDEP_OPS) | |
2235 laststart = b; | |
2236 else | |
2237 goto unfetch_interval; | |
2238 } | |
2239 | |
2240 /* If the upper bound is zero, don't want to succeed at | |
2241 all; jump from `laststart' to `b + 3', which will be | |
2242 the end of the buffer after we insert the jump. */ | |
2243 if (upper_bound == 0) | |
2244 { | |
2245 GET_BUFFER_SPACE (3); | |
2246 INSERT_JUMP (jump, laststart, b + 3); | |
2247 b += 3; | |
2248 } | |
2249 | |
2250 /* Otherwise, we have a nontrivial interval. When | |
2251 we're all done, the pattern will look like: | |
2252 set_number_at <jump count> <upper bound> | |
2253 set_number_at <succeed_n count> <lower bound> | |
5841 | 2254 succeed_n <after jump addr> <succeed_n count> |
1155 | 2255 <body of loop> |
2256 jump_n <succeed_n addr> <jump count> | |
2257 (The upper bound and `jump_n' are omitted if | |
2258 `upper_bound' is 1, though.) */ | |
2259 else | |
2260 { /* If the upper bound is > 1, we need to insert | |
2261 more at the end of the loop. */ | |
2262 unsigned nbytes = 10 + (upper_bound > 1) * 10; | |
2263 | |
2264 GET_BUFFER_SPACE (nbytes); | |
2265 | |
2266 /* Initialize lower bound of the `succeed_n', even | |
2267 though it will be set during matching by its | |
2268 attendant `set_number_at' (inserted next), | |
2269 because `re_compile_fastmap' needs to know. | |
2270 Jump to the `jump_n' we might insert below. */ | |
2271 INSERT_JUMP2 (succeed_n, laststart, | |
2272 b + 5 + (upper_bound > 1) * 5, | |
2273 lower_bound); | |
2274 b += 5; | |
2275 | |
2276 /* Code to initialize the lower bound. Insert | |
2277 before the `succeed_n'. The `5' is the last two | |
2278 bytes of this `set_number_at', plus 3 bytes of | |
2279 the following `succeed_n'. */ | |
2280 insert_op2 (set_number_at, laststart, 5, lower_bound, b); | |
2281 b += 5; | |
2282 | |
2283 if (upper_bound > 1) | |
2284 { /* More than one repetition is allowed, so | |
2285 append a backward jump to the `succeed_n' | |
2286 that starts this interval. | |
2287 | |
2288 When we've reached this during matching, | |
2289 we'll have matched the interval once, so | |
2290 jump back only `upper_bound - 1' times. */ | |
2291 STORE_JUMP2 (jump_n, b, laststart + 5, | |
2292 upper_bound - 1); | |
2293 b += 5; | |
2294 | |
2295 /* The location we want to set is the second | |
2296 parameter of the `jump_n'; that is `b-2' as | |
2297 an absolute address. `laststart' will be | |
2298 the `set_number_at' we're about to insert; | |
2299 `laststart+3' the number to set, the source | |
2300 for the relative address. But we are | |
2301 inserting into the middle of the pattern -- | |
2302 so everything is getting moved up by 5. | |
2303 Conclusion: (b - 2) - (laststart + 3) + 5, | |
2304 i.e., b - laststart. | |
2305 | |
2306 We insert this at the beginning of the loop | |
2307 so that if we fail during matching, we'll | |
2308 reinitialize the bounds. */ | |
2309 insert_op2 (set_number_at, laststart, b - laststart, | |
2310 upper_bound - 1, b); | |
2311 b += 5; | |
2312 } | |
2313 } | |
2314 pending_exact = 0; | |
2315 beg_interval = NULL; | |
2316 } | |
2317 break; | |
2318 | |
2319 unfetch_interval: | |
2320 /* If an invalid interval, match the characters as literals. */ | |
2321 assert (beg_interval); | |
2322 p = beg_interval; | |
2323 beg_interval = NULL; | |
2324 | |
2325 /* normal_char and normal_backslash need `c'. */ | |
2326 PATFETCH (c); | |
2327 | |
2328 if (!(syntax & RE_NO_BK_BRACES)) | |
2329 { | |
2330 if (p > pattern && p[-1] == '\\') | |
2331 goto normal_backslash; | |
2332 } | |
2333 goto normal_char; | |
2334 | |
2335 #ifdef emacs | |
2336 /* There is no way to specify the before_dot and after_dot | |
2337 operators. rms says this is ok. --karl */ | |
2338 case '=': | |
2339 BUF_PUSH (at_dot); | |
2340 break; | |
2341 | |
2342 case 's': | |
2343 laststart = b; | |
2344 PATFETCH (c); | |
2345 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
2346 break; | |
2347 | |
2348 case 'S': | |
2349 laststart = b; | |
2350 PATFETCH (c); | |
2351 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
2352 break; | |
2353 #endif /* emacs */ | |
2354 | |
2355 | |
2356 case 'w': | |
2357 laststart = b; | |
2358 BUF_PUSH (wordchar); | |
2359 break; | |
2360 | |
2361 | |
2362 case 'W': | |
2363 laststart = b; | |
2364 BUF_PUSH (notwordchar); | |
2365 break; | |
2366 | |
2367 | |
2368 case '<': | |
2369 BUF_PUSH (wordbeg); | |
2370 break; | |
2371 | |
2372 case '>': | |
2373 BUF_PUSH (wordend); | |
2374 break; | |
2375 | |
2376 case 'b': | |
2377 BUF_PUSH (wordbound); | |
2378 break; | |
2379 | |
2380 case 'B': | |
2381 BUF_PUSH (notwordbound); | |
2382 break; | |
2383 | |
2384 case '`': | |
2385 BUF_PUSH (begbuf); | |
2386 break; | |
2387 | |
2388 case '\'': | |
2389 BUF_PUSH (endbuf); | |
2390 break; | |
2391 | |
2392 case '1': case '2': case '3': case '4': case '5': | |
2393 case '6': case '7': case '8': case '9': | |
2394 if (syntax & RE_NO_BK_REFS) | |
2395 goto normal_char; | |
2396 | |
2397 c1 = c - '0'; | |
2398 | |
2399 if (c1 > regnum) | |
2400 return REG_ESUBREG; | |
2401 | |
2402 /* Can't back reference to a subexpression if inside of it. */ | |
2403 if (group_in_compile_stack (compile_stack, c1)) | |
2404 goto normal_char; | |
2405 | |
2406 laststart = b; | |
2407 BUF_PUSH_2 (duplicate, c1); | |
2408 break; | |
2409 | |
2410 | |
2411 case '+': | |
2412 case '?': | |
2413 if (syntax & RE_BK_PLUS_QM) | |
2414 goto handle_plus; | |
2415 else | |
2416 goto normal_backslash; | |
2417 | |
2418 default: | |
2419 normal_backslash: | |
2420 /* You might think it would be useful for \ to mean | |
2421 not to translate; but if we don't translate it | |
2422 it will never match anything. */ | |
2423 c = TRANSLATE (c); | |
2424 goto normal_char; | |
2425 } | |
2426 break; | |
2427 | |
2428 | |
2429 default: | |
2430 /* Expects the character in `c'. */ | |
2431 normal_char: | |
2432 /* If no exactn currently being built. */ | |
2433 if (!pending_exact | |
2434 | |
2435 /* If last exactn not at current position. */ | |
2436 || pending_exact + *pending_exact + 1 != b | |
2437 | |
2438 /* We have only one byte following the exactn for the count. */ | |
2439 || *pending_exact == (1 << BYTEWIDTH) - 1 | |
2440 | |
2441 /* If followed by a repetition operator. */ | |
2442 || *p == '*' || *p == '^' | |
2443 || ((syntax & RE_BK_PLUS_QM) | |
2444 ? *p == '\\' && (p[1] == '+' || p[1] == '?') | |
2445 : (*p == '+' || *p == '?')) | |
2446 || ((syntax & RE_INTERVALS) | |
2447 && ((syntax & RE_NO_BK_BRACES) | |
2448 ? *p == '{' | |
2449 : (p[0] == '\\' && p[1] == '{')))) | |
2450 { | |
2451 /* Start building a new exactn. */ | |
2452 | |
2453 laststart = b; | |
2454 | |
2455 BUF_PUSH_2 (exactn, 0); | |
2456 pending_exact = b - 1; | |
2457 } | |
2458 | |
2459 BUF_PUSH (c); | |
2460 (*pending_exact)++; | |
2461 break; | |
2462 } /* switch (c) */ | |
2463 } /* while p != pend */ | |
2464 | |
2465 | |
2466 /* Through the pattern now. */ | |
2467 | |
2468 if (fixup_alt_jump) | |
2469 STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
2470 | |
2471 if (!COMPILE_STACK_EMPTY) | |
2472 return REG_EPAREN; | |
2473 | |
2474 free (compile_stack.stack); | |
2475 | |
2476 /* We have succeeded; set the length of the buffer. */ | |
2477 bufp->used = b - bufp->buffer; | |
2478 | |
2479 #ifdef DEBUG | |
2480 if (debug) | |
2481 { | |
2615 | 2482 DEBUG_PRINT1 ("\nCompiled pattern: \n"); |
1155 | 2483 print_compiled_pattern (bufp); |
2484 } | |
2485 #endif /* DEBUG */ | |
2486 | |
2952 | 2487 #ifndef MATCH_MAY_ALLOCATE |
2949 | 2488 /* Initialize the failure stack to the largest possible stack. This |
2489 isn't necessary unless we're trying to avoid calling alloca in | |
2490 the search and match routines. */ | |
2491 { | |
2492 int num_regs = bufp->re_nsub + 1; | |
2493 | |
2494 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size | |
2495 is strictly greater than re_max_failures, the largest possible stack | |
2496 is 2 * re_max_failures failure points. */ | |
2497 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); | |
5841 | 2498 if (! fail_stack.stack) |
2949 | 2499 fail_stack.stack = |
2500 (fail_stack_elt_t *) malloc (fail_stack.size | |
2501 * sizeof (fail_stack_elt_t)); | |
2502 | |
2503 /* Initialize some other variables the matcher uses. */ | |
2504 RETALLOC_IF (regstart, num_regs, const char *); | |
2505 RETALLOC_IF (regend, num_regs, const char *); | |
2506 RETALLOC_IF (old_regstart, num_regs, const char *); | |
2507 RETALLOC_IF (old_regend, num_regs, const char *); | |
2508 RETALLOC_IF (best_regstart, num_regs, const char *); | |
2509 RETALLOC_IF (best_regend, num_regs, const char *); | |
2510 RETALLOC_IF (reg_info, num_regs, register_info_type); | |
2511 RETALLOC_IF (reg_dummy, num_regs, const char *); | |
2512 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type); | |
2513 } | |
2514 #endif | |
2515 | |
1155 | 2516 return REG_NOERROR; |
2517 } /* regex_compile */ | |
2518 | |
2519 /* Subroutines for `regex_compile'. */ | |
2520 | |
2521 /* Store OP at LOC followed by two-byte integer parameter ARG. */ | |
2522 | |
2523 static void | |
2524 store_op1 (op, loc, arg) | |
2525 re_opcode_t op; | |
2526 unsigned char *loc; | |
2527 int arg; | |
2528 { | |
2529 *loc = (unsigned char) op; | |
2530 STORE_NUMBER (loc + 1, arg); | |
2531 } | |
2532 | |
2533 | |
2534 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
2535 | |
2536 static void | |
2537 store_op2 (op, loc, arg1, arg2) | |
2538 re_opcode_t op; | |
2539 unsigned char *loc; | |
2540 int arg1, arg2; | |
2541 { | |
2542 *loc = (unsigned char) op; | |
2543 STORE_NUMBER (loc + 1, arg1); | |
2544 STORE_NUMBER (loc + 3, arg2); | |
2545 } | |
2546 | |
2547 | |
2548 /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
2549 for OP followed by two-byte integer parameter ARG. */ | |
2550 | |
2551 static void | |
2552 insert_op1 (op, loc, arg, end) | |
2553 re_opcode_t op; | |
2554 unsigned char *loc; | |
2555 int arg; | |
2556 unsigned char *end; | |
2557 { | |
2558 register unsigned char *pfrom = end; | |
2559 register unsigned char *pto = end + 3; | |
2560 | |
2561 while (pfrom != loc) | |
2562 *--pto = *--pfrom; | |
2563 | |
2564 store_op1 (op, loc, arg); | |
2565 } | |
2566 | |
2567 | |
2568 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
2569 | |
2570 static void | |
2571 insert_op2 (op, loc, arg1, arg2, end) | |
2572 re_opcode_t op; | |
2573 unsigned char *loc; | |
2574 int arg1, arg2; | |
2575 unsigned char *end; | |
2576 { | |
2577 register unsigned char *pfrom = end; | |
2578 register unsigned char *pto = end + 5; | |
2579 | |
2580 while (pfrom != loc) | |
2581 *--pto = *--pfrom; | |
2582 | |
2583 store_op2 (op, loc, arg1, arg2); | |
2584 } | |
2585 | |
2586 | |
2587 /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
2588 after an alternative or a begin-subexpression. We assume there is at | |
2589 least one character before the ^. */ | |
2590 | |
2591 static boolean | |
2592 at_begline_loc_p (pattern, p, syntax) | |
2593 const char *pattern, *p; | |
2594 reg_syntax_t syntax; | |
2595 { | |
2596 const char *prev = p - 2; | |
2597 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; | |
2598 | |
2599 return | |
2600 /* After a subexpression? */ | |
2601 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
2602 /* After an alternative? */ | |
2603 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); | |
2604 } | |
2605 | |
2606 | |
2607 /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
2608 at least one character after the $, i.e., `P < PEND'. */ | |
2609 | |
2610 static boolean | |
2611 at_endline_loc_p (p, pend, syntax) | |
2612 const char *p, *pend; | |
2613 int syntax; | |
2614 { | |
2615 const char *next = p; | |
2616 boolean next_backslash = *next == '\\'; | |
2617 const char *next_next = p + 1 < pend ? p + 1 : NULL; | |
2618 | |
2619 return | |
2620 /* Before a subexpression? */ | |
2621 (syntax & RE_NO_BK_PARENS ? *next == ')' | |
2622 : next_backslash && next_next && *next_next == ')') | |
2623 /* Before an alternative? */ | |
2624 || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
2625 : next_backslash && next_next && *next_next == '|'); | |
2626 } | |
2627 | |
2628 | |
2629 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and | |
2630 false if it's not. */ | |
2631 | |
2632 static boolean | |
2633 group_in_compile_stack (compile_stack, regnum) | |
2634 compile_stack_type compile_stack; | |
2635 regnum_t regnum; | |
2636 { | |
2637 int this_element; | |
2638 | |
2639 for (this_element = compile_stack.avail - 1; | |
2640 this_element >= 0; | |
2641 this_element--) | |
2642 if (compile_stack.stack[this_element].regnum == regnum) | |
2643 return true; | |
2644 | |
2645 return false; | |
2646 } | |
2647 | |
2648 | |
2649 /* Read the ending character of a range (in a bracket expression) from the | |
2650 uncompiled pattern *P_PTR (which ends at PEND). We assume the | |
2651 starting character is in `P[-2]'. (`P[-1]' is the character `-'.) | |
2652 Then we set the translation of all bits between the starting and | |
2653 ending characters (inclusive) in the compiled pattern B. | |
2654 | |
2655 Return an error code. | |
2656 | |
2657 We use these short variable names so we can use the same macros as | |
2658 `regex_compile' itself. */ | |
2659 | |
2660 static reg_errcode_t | |
2661 compile_range (p_ptr, pend, translate, syntax, b) | |
2662 const char **p_ptr, *pend; | |
2663 char *translate; | |
2664 reg_syntax_t syntax; | |
2665 unsigned char *b; | |
2666 { | |
2667 unsigned this_char; | |
2668 | |
2669 const char *p = *p_ptr; | |
1689 | 2670 int range_start, range_end; |
1155 | 2671 |
2672 if (p == pend) | |
2673 return REG_ERANGE; | |
2674 | |
1689 | 2675 /* Even though the pattern is a signed `char *', we need to fetch |
2676 with unsigned char *'s; if the high bit of the pattern character | |
2677 is set, the range endpoints will be negative if we fetch using a | |
2678 signed char *. | |
2679 | |
2680 We also want to fetch the endpoints without translating them; the | |
2681 appropriate translation is done in the bit-setting loop below. */ | |
2682 range_start = ((unsigned char *) p)[-2]; | |
2683 range_end = ((unsigned char *) p)[0]; | |
1155 | 2684 |
2685 /* Have to increment the pointer into the pattern string, so the | |
2686 caller isn't still at the ending character. */ | |
2687 (*p_ptr)++; | |
2688 | |
2689 /* If the start is after the end, the range is empty. */ | |
2690 if (range_start > range_end) | |
2691 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; | |
2692 | |
2693 /* Here we see why `this_char' has to be larger than an `unsigned | |
2694 char' -- the range is inclusive, so if `range_end' == 0xff | |
2695 (assuming 8-bit characters), we would otherwise go into an infinite | |
2696 loop, since all characters <= 0xff. */ | |
2697 for (this_char = range_start; this_char <= range_end; this_char++) | |
2698 { | |
2699 SET_LIST_BIT (TRANSLATE (this_char)); | |
2700 } | |
2701 | |
2702 return REG_NOERROR; | |
2703 } | |
2704 | |
2705 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
2706 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
2707 characters can start a string that matches the pattern. This fastmap | |
2708 is used by re_search to skip quickly over impossible starting points. | |
2709 | |
2710 The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
2711 area as BUFP->fastmap. | |
2712 | |
2713 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in | |
2714 the pattern buffer. | |
2715 | |
2716 Returns 0 if we succeed, -2 if an internal error. */ | |
2717 | |
2718 int | |
2719 re_compile_fastmap (bufp) | |
2720 struct re_pattern_buffer *bufp; | |
2721 { | |
2722 int j, k; | |
2952 | 2723 #ifdef MATCH_MAY_ALLOCATE |
1155 | 2724 fail_stack_type fail_stack; |
2949 | 2725 #endif |
1155 | 2726 #ifndef REGEX_MALLOC |
2727 char *destination; | |
2728 #endif | |
2729 /* We don't push any register information onto the failure stack. */ | |
2730 unsigned num_regs = 0; | |
2731 | |
2732 register char *fastmap = bufp->fastmap; | |
2733 unsigned char *pattern = bufp->buffer; | |
2734 unsigned long size = bufp->used; | |
4918
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
2735 unsigned char *p = pattern; |
1155 | 2736 register unsigned char *pend = pattern + size; |
2737 | |
2738 /* Assume that each path through the pattern can be null until | |
2739 proven otherwise. We set this false at the bottom of switch | |
2740 statement, to which we get only if a particular path doesn't | |
2741 match the empty string. */ | |
2742 boolean path_can_be_null = true; | |
2743 | |
2744 /* We aren't doing a `succeed_n' to begin with. */ | |
2745 boolean succeed_n_p = false; | |
2746 | |
2747 assert (fastmap != NULL && p != NULL); | |
2748 | |
2749 INIT_FAIL_STACK (); | |
2750 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ | |
2751 bufp->fastmap_accurate = 1; /* It will be when we're done. */ | |
2752 bufp->can_be_null = 0; | |
2753 | |
2754 while (p != pend || !FAIL_STACK_EMPTY ()) | |
2755 { | |
2756 if (p == pend) | |
2757 { | |
2758 bufp->can_be_null |= path_can_be_null; | |
2759 | |
2760 /* Reset for next path. */ | |
2761 path_can_be_null = true; | |
2762 | |
2763 p = fail_stack.stack[--fail_stack.avail]; | |
2764 } | |
2765 | |
2766 /* We should never be about to go beyond the end of the pattern. */ | |
2767 assert (p < pend); | |
2768 | |
2769 #ifdef SWITCH_ENUM_BUG | |
2770 switch ((int) ((re_opcode_t) *p++)) | |
2771 #else | |
2772 switch ((re_opcode_t) *p++) | |
2773 #endif | |
2774 { | |
2775 | |
2776 /* I guess the idea here is to simply not bother with a fastmap | |
2777 if a backreference is used, since it's too hard to figure out | |
2778 the fastmap for the corresponding group. Setting | |
2779 `can_be_null' stops `re_search_2' from using the fastmap, so | |
2780 that is all we do. */ | |
2781 case duplicate: | |
2782 bufp->can_be_null = 1; | |
2783 return 0; | |
2784 | |
2785 | |
2786 /* Following are the cases which match a character. These end | |
2787 with `break'. */ | |
2788 | |
2789 case exactn: | |
2790 fastmap[p[1]] = 1; | |
2791 break; | |
2792 | |
2793 | |
2794 case charset: | |
2795 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
2796 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) | |
2797 fastmap[j] = 1; | |
2798 break; | |
2799 | |
2800 | |
2801 case charset_not: | |
2802 /* Chars beyond end of map must be allowed. */ | |
2803 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) | |
2804 fastmap[j] = 1; | |
2805 | |
2806 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
2807 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) | |
2808 fastmap[j] = 1; | |
2809 break; | |
2810 | |
2811 | |
2812 case wordchar: | |
2813 for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2814 if (SYNTAX (j) == Sword) | |
2815 fastmap[j] = 1; | |
2816 break; | |
2817 | |
2818 | |
2819 case notwordchar: | |
2820 for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2821 if (SYNTAX (j) != Sword) | |
2822 fastmap[j] = 1; | |
2823 break; | |
2824 | |
2825 | |
2826 case anychar: | |
2827 /* `.' matches anything ... */ | |
2828 for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2829 fastmap[j] = 1; | |
2830 | |
2831 /* ... except perhaps newline. */ | |
2832 if (!(bufp->syntax & RE_DOT_NEWLINE)) | |
2833 fastmap['\n'] = 0; | |
2834 | |
2835 /* Return if we have already set `can_be_null'; if we have, | |
2836 then the fastmap is irrelevant. Something's wrong here. */ | |
2837 else if (bufp->can_be_null) | |
2838 return 0; | |
2839 | |
2840 /* Otherwise, have to check alternative paths. */ | |
2841 break; | |
2842 | |
2843 | |
2844 #ifdef emacs | |
2845 case syntaxspec: | |
2846 k = *p++; | |
2847 for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2848 if (SYNTAX (j) == (enum syntaxcode) k) | |
2849 fastmap[j] = 1; | |
2850 break; | |
2851 | |
2852 | |
2853 case notsyntaxspec: | |
2854 k = *p++; | |
2855 for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2856 if (SYNTAX (j) != (enum syntaxcode) k) | |
2857 fastmap[j] = 1; | |
2858 break; | |
2859 | |
2860 | |
2861 /* All cases after this match the empty string. These end with | |
2862 `continue'. */ | |
2863 | |
2864 | |
2865 case before_dot: | |
2866 case at_dot: | |
2867 case after_dot: | |
2868 continue; | |
2869 #endif /* not emacs */ | |
2870 | |
2871 | |
2872 case no_op: | |
2873 case begline: | |
2874 case endline: | |
2875 case begbuf: | |
2876 case endbuf: | |
2877 case wordbound: | |
2878 case notwordbound: | |
2879 case wordbeg: | |
2880 case wordend: | |
2881 case push_dummy_failure: | |
2882 continue; | |
2883 | |
2884 | |
2885 case jump_n: | |
2886 case pop_failure_jump: | |
2887 case maybe_pop_jump: | |
2888 case jump: | |
2889 case jump_past_alt: | |
2890 case dummy_failure_jump: | |
2891 EXTRACT_NUMBER_AND_INCR (j, p); | |
2892 p += j; | |
2893 if (j > 0) | |
2894 continue; | |
2895 | |
2896 /* Jump backward implies we just went through the body of a | |
2897 loop and matched nothing. Opcode jumped to should be | |
2898 `on_failure_jump' or `succeed_n'. Just treat it like an | |
2899 ordinary jump. For a * loop, it has pushed its failure | |
2900 point already; if so, discard that as redundant. */ | |
2901 if ((re_opcode_t) *p != on_failure_jump | |
2902 && (re_opcode_t) *p != succeed_n) | |
2903 continue; | |
2904 | |
2905 p++; | |
2906 EXTRACT_NUMBER_AND_INCR (j, p); | |
2907 p += j; | |
2908 | |
2909 /* If what's on the stack is where we are now, pop it. */ | |
2910 if (!FAIL_STACK_EMPTY () | |
2911 && fail_stack.stack[fail_stack.avail - 1] == p) | |
2912 fail_stack.avail--; | |
2913 | |
2914 continue; | |
2915 | |
2916 | |
2917 case on_failure_jump: | |
2918 case on_failure_keep_string_jump: | |
2919 handle_on_failure_jump: | |
2920 EXTRACT_NUMBER_AND_INCR (j, p); | |
2921 | |
2922 /* For some patterns, e.g., `(a?)?', `p+j' here points to the | |
2923 end of the pattern. We don't want to push such a point, | |
2924 since when we restore it above, entering the switch will | |
2925 increment `p' past the end of the pattern. We don't need | |
2926 to push such a point since we obviously won't find any more | |
2927 fastmap entries beyond `pend'. Such a pattern can match | |
2928 the null string, though. */ | |
2929 if (p + j < pend) | |
2930 { | |
2931 if (!PUSH_PATTERN_OP (p + j, fail_stack)) | |
2932 return -2; | |
2933 } | |
2934 else | |
2935 bufp->can_be_null = 1; | |
2936 | |
2937 if (succeed_n_p) | |
2938 { | |
2939 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ | |
2940 succeed_n_p = false; | |
2941 } | |
2942 | |
2943 continue; | |
2944 | |
2945 | |
2946 case succeed_n: | |
2947 /* Get to the number of times to succeed. */ | |
2948 p += 2; | |
2949 | |
2950 /* Increment p past the n for when k != 0. */ | |
2951 EXTRACT_NUMBER_AND_INCR (k, p); | |
2952 if (k == 0) | |
2953 { | |
2954 p -= 4; | |
2955 succeed_n_p = true; /* Spaghetti code alert. */ | |
2956 goto handle_on_failure_jump; | |
2957 } | |
2958 continue; | |
2959 | |
2960 | |
2961 case set_number_at: | |
2962 p += 4; | |
2963 continue; | |
2964 | |
2965 | |
2966 case start_memory: | |
2967 case stop_memory: | |
2968 p += 2; | |
2969 continue; | |
2970 | |
2971 | |
2972 default: | |
2973 abort (); /* We have listed all the cases. */ | |
2974 } /* switch *p++ */ | |
2975 | |
2976 /* Getting here means we have found the possible starting | |
2977 characters for one path of the pattern -- and that the empty | |
2978 string does not match. We need not follow this path further. | |
2979 Instead, look at the next alternative (remembered on the | |
2980 stack), or quit if no more. The test at the top of the loop | |
2981 does these things. */ | |
2982 path_can_be_null = false; | |
2983 p = pend; | |
2984 } /* while p */ | |
2985 | |
2986 /* Set `can_be_null' for the last path (also the first path, if the | |
2987 pattern is empty). */ | |
2988 bufp->can_be_null |= path_can_be_null; | |
2989 return 0; | |
2990 } /* re_compile_fastmap */ | |
2991 | |
2992 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
2993 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
2994 this memory for recording register information. STARTS and ENDS | |
2995 must be allocated using the malloc library routine, and must each | |
2996 be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
2997 | |
2998 If NUM_REGS == 0, then subsequent matches should allocate their own | |
2999 register data. | |
3000 | |
3001 Unless this function is called, the first search or match using | |
3002 PATTERN_BUFFER will allocate its own register data, without | |
3003 freeing the old data. */ | |
3004 | |
3005 void | |
3006 re_set_registers (bufp, regs, num_regs, starts, ends) | |
3007 struct re_pattern_buffer *bufp; | |
3008 struct re_registers *regs; | |
3009 unsigned num_regs; | |
3010 regoff_t *starts, *ends; | |
3011 { | |
3012 if (num_regs) | |
3013 { | |
3014 bufp->regs_allocated = REGS_REALLOCATE; | |
3015 regs->num_regs = num_regs; | |
3016 regs->start = starts; | |
3017 regs->end = ends; | |
3018 } | |
3019 else | |
3020 { | |
3021 bufp->regs_allocated = REGS_UNALLOCATED; | |
3022 regs->num_regs = 0; | |
5014
6062331f7430
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4918
diff
changeset
|
3023 regs->start = regs->end = (regoff_t *) 0; |
1155 | 3024 } |
3025 } | |
3026 | |
3027 /* Searching routines. */ | |
3028 | |
3029 /* Like re_search_2, below, but only one string is specified, and | |
3030 doesn't let you say where to stop matching. */ | |
3031 | |
3032 int | |
3033 re_search (bufp, string, size, startpos, range, regs) | |
3034 struct re_pattern_buffer *bufp; | |
3035 const char *string; | |
3036 int size, startpos, range; | |
3037 struct re_registers *regs; | |
3038 { | |
3039 return re_search_2 (bufp, NULL, 0, string, size, startpos, range, | |
3040 regs, size); | |
3041 } | |
3042 | |
3043 | |
3044 /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
3045 virtual concatenation of STRING1 and STRING2, starting first at index | |
3046 STARTPOS, then at STARTPOS + 1, and so on. | |
3047 | |
3048 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. | |
3049 | |
3050 RANGE is how far to scan while trying to match. RANGE = 0 means try | |
3051 only at STARTPOS; in general, the last start tried is STARTPOS + | |
3052 RANGE. | |
3053 | |
3054 In REGS, return the indices of the virtual concatenation of STRING1 | |
3055 and STRING2 that matched the entire BUFP->buffer and its contained | |
3056 subexpressions. | |
3057 | |
3058 Do not consider matching one past the index STOP in the virtual | |
3059 concatenation of STRING1 and STRING2. | |
3060 | |
3061 We return either the position in the strings at which the match was | |
3062 found, -1 if no match, or -2 if error (such as failure | |
3063 stack overflow). */ | |
3064 | |
3065 int | |
3066 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop) | |
3067 struct re_pattern_buffer *bufp; | |
3068 const char *string1, *string2; | |
3069 int size1, size2; | |
3070 int startpos; | |
3071 int range; | |
3072 struct re_registers *regs; | |
3073 int stop; | |
3074 { | |
3075 int val; | |
3076 register char *fastmap = bufp->fastmap; | |
3077 register char *translate = bufp->translate; | |
3078 int total_size = size1 + size2; | |
3079 int endpos = startpos + range; | |
3080 | |
3081 /* Check for out-of-range STARTPOS. */ | |
3082 if (startpos < 0 || startpos > total_size) | |
3083 return -1; | |
3084 | |
3085 /* Fix up RANGE if it might eventually take us outside | |
3086 the virtual concatenation of STRING1 and STRING2. */ | |
3087 if (endpos < -1) | |
3088 range = -1 - startpos; | |
3089 else if (endpos > total_size) | |
3090 range = total_size - startpos; | |
3091 | |
3092 /* If the search isn't to be a backwards one, don't waste time in a | |
1637 | 3093 search for a pattern that must be anchored. */ |
3094 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) | |
1155 | 3095 { |
3096 if (startpos > 0) | |
3097 return -1; | |
3098 else | |
3099 range = 1; | |
3100 } | |
3101 | |
1637 | 3102 /* Update the fastmap now if not correct already. */ |
3103 if (fastmap && !bufp->fastmap_accurate) | |
3104 if (re_compile_fastmap (bufp) == -2) | |
3105 return -2; | |
3106 | |
3107 /* Loop through the string, looking for a place to start matching. */ | |
1155 | 3108 for (;;) |
3109 { | |
3110 /* If a fastmap is supplied, skip quickly over characters that | |
3111 cannot be the start of a match. If the pattern can match the | |
3112 null string, however, we don't need to skip characters; we want | |
3113 the first null string. */ | |
3114 if (fastmap && startpos < total_size && !bufp->can_be_null) | |
3115 { | |
3116 if (range > 0) /* Searching forwards. */ | |
3117 { | |
3118 register const char *d; | |
3119 register int lim = 0; | |
3120 int irange = range; | |
3121 | |
3122 if (startpos < size1 && startpos + range >= size1) | |
3123 lim = range - (size1 - startpos); | |
3124 | |
3125 d = (startpos >= size1 ? string2 - size1 : string1) + startpos; | |
3126 | |
3127 /* Written out as an if-else to avoid testing `translate' | |
3128 inside the loop. */ | |
3129 if (translate) | |
3130 while (range > lim | |
2078 | 3131 && !fastmap[(unsigned char) |
3132 translate[(unsigned char) *d++]]) | |
1155 | 3133 range--; |
3134 else | |
3135 while (range > lim && !fastmap[(unsigned char) *d++]) | |
3136 range--; | |
3137 | |
3138 startpos += irange - range; | |
3139 } | |
3140 else /* Searching backwards. */ | |
3141 { | |
3142 register char c = (size1 == 0 || startpos >= size1 | |
3143 ? string2[startpos - size1] | |
3144 : string1[startpos]); | |
3145 | |
1637 | 3146 if (!fastmap[(unsigned char) TRANSLATE (c)]) |
1155 | 3147 goto advance; |
3148 } | |
3149 } | |
3150 | |
3151 /* If can't match the null string, and that's all we have left, fail. */ | |
3152 if (range >= 0 && startpos == total_size && fastmap | |
3153 && !bufp->can_be_null) | |
3154 return -1; | |
3155 | |
3156 val = re_match_2 (bufp, string1, size1, string2, size2, | |
3157 startpos, regs, stop); | |
3158 if (val >= 0) | |
3159 return startpos; | |
3160 | |
3161 if (val == -2) | |
3162 return -2; | |
3163 | |
3164 advance: | |
3165 if (!range) | |
3166 break; | |
3167 else if (range > 0) | |
3168 { | |
3169 range--; | |
3170 startpos++; | |
3171 } | |
3172 else | |
3173 { | |
3174 range++; | |
3175 startpos--; | |
3176 } | |
3177 } | |
3178 return -1; | |
3179 } /* re_search_2 */ | |
3180 | |
3181 /* Declarations and macros for re_match_2. */ | |
3182 | |
3183 static int bcmp_translate (); | |
3184 static boolean alt_match_null_string_p (), | |
3185 common_op_match_null_string_p (), | |
3186 group_match_null_string_p (); | |
3187 | |
3188 /* This converts PTR, a pointer into one of the search strings `string1' | |
3189 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
|
3190 #define POINTER_TO_OFFSET(ptr) \ |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3191 (FIRST_STRING_P (ptr) \ |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3192 ? ((regoff_t) ((ptr) - string1)) \ |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3193 : ((regoff_t) ((ptr) - string2 + size1))) |
1155 | 3194 |
3195 /* Macros for dealing with the split strings in re_match_2. */ | |
3196 | |
3197 #define MATCHING_IN_FIRST_STRING (dend == end_match_1) | |
3198 | |
3199 /* Call before fetching a character with *d. This switches over to | |
3200 string2 if necessary. */ | |
3201 #define PREFETCH() \ | |
3202 while (d == dend) \ | |
3203 { \ | |
3204 /* End of string2 => fail. */ \ | |
3205 if (dend == end_match_2) \ | |
3206 goto fail; \ | |
3207 /* End of string1 => advance to string2. */ \ | |
3208 d = string2; \ | |
3209 dend = end_match_2; \ | |
3210 } | |
3211 | |
3212 | |
3213 /* Test if at very beginning or at very end of the virtual concatenation | |
3214 of `string1' and `string2'. If only one string, it's `string2'. */ | |
1637 | 3215 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
3216 #define AT_STRINGS_END(d) ((d) == end2) | |
1155 | 3217 |
3218 | |
3219 /* Test if D points to a character which is word-constituent. We have | |
3220 two special cases to check for: if past the end of string1, look at | |
3221 the first character in string2; and if before the beginning of | |
1637 | 3222 string2, look at the last character in string1. */ |
3223 #define WORDCHAR_P(d) \ | |
1155 | 3224 (SYNTAX ((d) == end1 ? *string2 \ |
1637 | 3225 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
3226 == Sword) | |
1155 | 3227 |
3228 /* Test if the character before D and the one at D differ with respect | |
3229 to being word-constituent. */ | |
3230 #define AT_WORD_BOUNDARY(d) \ | |
1637 | 3231 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ |
3232 || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) | |
1155 | 3233 |
3234 | |
3235 /* Free everything we malloc. */ | |
2952 | 3236 #ifdef MATCH_MAY_ALLOCATE |
1155 | 3237 #ifdef REGEX_MALLOC |
3238 #define FREE_VAR(var) if (var) free (var); var = NULL | |
3239 #define FREE_VARIABLES() \ | |
3240 do { \ | |
3241 FREE_VAR (fail_stack.stack); \ | |
3242 FREE_VAR (regstart); \ | |
3243 FREE_VAR (regend); \ | |
3244 FREE_VAR (old_regstart); \ | |
3245 FREE_VAR (old_regend); \ | |
3246 FREE_VAR (best_regstart); \ | |
3247 FREE_VAR (best_regend); \ | |
3248 FREE_VAR (reg_info); \ | |
3249 FREE_VAR (reg_dummy); \ | |
3250 FREE_VAR (reg_info_dummy); \ | |
3251 } while (0) | |
3252 #else /* not REGEX_MALLOC */ | |
3253 /* Some MIPS systems (at least) want this to free alloca'd storage. */ | |
3254 #define FREE_VARIABLES() alloca (0) | |
3255 #endif /* not REGEX_MALLOC */ | |
2952 | 3256 #else |
3257 #define FREE_VARIABLES() /* Do nothing! */ | |
3258 #endif /* not MATCH_MAY_ALLOCATE */ | |
1155 | 3259 |
3260 /* These values must meet several constraints. They must not be valid | |
3261 register values; since we have a limit of 255 registers (because | |
3262 we use only one byte in the pattern for the register number), we can | |
3263 use numbers larger than 255. They must differ by 1, because of | |
3264 NUM_FAILURE_ITEMS above. And the value for the lowest register must | |
3265 be larger than the value for the highest register, so we do not try | |
3266 to actually save any registers when none are active. */ | |
3267 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) | |
3268 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) | |
3269 | |
3270 /* Matching routines. */ | |
3271 | |
3272 #ifndef emacs /* Emacs never uses this. */ | |
3273 /* re_match is like re_match_2 except it takes only a single string. */ | |
3274 | |
3275 int | |
3276 re_match (bufp, string, size, pos, regs) | |
3277 struct re_pattern_buffer *bufp; | |
3278 const char *string; | |
3279 int size, pos; | |
3280 struct re_registers *regs; | |
3281 { | |
3282 return re_match_2 (bufp, NULL, 0, string, size, pos, regs, size); | |
3283 } | |
3284 #endif /* not emacs */ | |
3285 | |
3286 | |
3287 /* re_match_2 matches the compiled pattern in BUFP against the | |
3288 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
3289 and SIZE2, respectively). We start matching at POS, and stop | |
3290 matching at STOP. | |
3291 | |
3292 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we | |
3293 store offsets for the substring each group matched in REGS. See the | |
3294 documentation for exactly how many groups we fill. | |
3295 | |
3296 We return -1 if no match, -2 if an internal error (such as the | |
3297 failure stack overflowing). Otherwise, we return the length of the | |
3298 matched substring. */ | |
3299 | |
3300 int | |
3301 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
3302 struct re_pattern_buffer *bufp; | |
3303 const char *string1, *string2; | |
3304 int size1, size2; | |
3305 int pos; | |
3306 struct re_registers *regs; | |
3307 int stop; | |
3308 { | |
3309 /* General temporaries. */ | |
3310 int mcnt; | |
3311 unsigned char *p1; | |
3312 | |
3313 /* Just past the end of the corresponding string. */ | |
3314 const char *end1, *end2; | |
3315 | |
3316 /* Pointers into string1 and string2, just past the last characters in | |
3317 each to consider matching. */ | |
3318 const char *end_match_1, *end_match_2; | |
3319 | |
3320 /* Where we are in the data, and the end of the current string. */ | |
3321 const char *d, *dend; | |
3322 | |
3323 /* Where we are in the pattern, and the end of the pattern. */ | |
3324 unsigned char *p = bufp->buffer; | |
3325 register unsigned char *pend = p + bufp->used; | |
3326 | |
5842 | 3327 /* Mark the opcode just after a start_memory, so we can test for an |
3328 empty subpattern when we get to the stop_memory. */ | |
3329 unsigned char *just_past_start_mem = 0; | |
3330 | |
1155 | 3331 /* We use this to map every character in the string. */ |
3332 char *translate = bufp->translate; | |
3333 | |
3334 /* Failure point stack. Each place that can handle a failure further | |
3335 down the line pushes a failure point on this stack. It consists of | |
3336 restart, regend, and reg_info for all registers corresponding to | |
3337 the subexpressions we're currently inside, plus the number of such | |
3338 registers, and, finally, two char *'s. The first char * is where | |
3339 to resume scanning the pattern; the second one is where to resume | |
3340 scanning the strings. If the latter is zero, the failure point is | |
3341 a ``dummy''; if a failure happens and the failure point is a dummy, | |
3342 it gets discarded and the next next one is tried. */ | |
2952 | 3343 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
1155 | 3344 fail_stack_type fail_stack; |
2949 | 3345 #endif |
1155 | 3346 #ifdef DEBUG |
3347 static unsigned failure_id = 0; | |
1637 | 3348 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
1155 | 3349 #endif |
3350 | |
3351 /* We fill all the registers internally, independent of what we | |
3352 return, for use in backreferences. The number here includes | |
3353 an element for register zero. */ | |
3354 unsigned num_regs = bufp->re_nsub + 1; | |
3355 | |
3356 /* The currently active registers. */ | |
3357 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3358 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3359 | |
3360 /* Information on the contents of registers. These are pointers into | |
3361 the input strings; they record just what was matched (on this | |
3362 attempt) by a subexpression part of the pattern, that is, the | |
3363 regnum-th regstart pointer points to where in the pattern we began | |
3364 matching and the regnum-th regend points to right after where we | |
3365 stopped matching the regnum-th subexpression. (The zeroth register | |
3366 keeps track of what the whole pattern matches.) */ | |
2952 | 3367 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
1155 | 3368 const char **regstart, **regend; |
2949 | 3369 #endif |
1155 | 3370 |
3371 /* If a group that's operated upon by a repetition operator fails to | |
3372 match anything, then the register for its start will need to be | |
3373 restored because it will have been set to wherever in the string we | |
3374 are when we last see its open-group operator. Similarly for a | |
3375 register's end. */ | |
2952 | 3376 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
1155 | 3377 const char **old_regstart, **old_regend; |
2949 | 3378 #endif |
1155 | 3379 |
3380 /* The is_active field of reg_info helps us keep track of which (possibly | |
3381 nested) subexpressions we are currently in. The matched_something | |
3382 field of reg_info[reg_num] helps us tell whether or not we have | |
3383 matched any of the pattern so far this time through the reg_num-th | |
3384 subexpression. These two fields get reset each time through any | |
3385 loop their register is in. */ | |
2952 | 3386 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
1155 | 3387 register_info_type *reg_info; |
2949 | 3388 #endif |
1155 | 3389 |
3390 /* The following record the register info as found in the above | |
3391 variables when we find a match better than any we've seen before. | |
3392 This happens as we backtrack through the failure points, which in | |
3393 turn happens only if we have not yet matched the entire string. */ | |
3394 unsigned best_regs_set = false; | |
2952 | 3395 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
1155 | 3396 const char **best_regstart, **best_regend; |
2949 | 3397 #endif |
1155 | 3398 |
3399 /* Logically, this is `best_regend[0]'. But we don't want to have to | |
3400 allocate space for that if we're not allocating space for anything | |
3401 else (see below). Also, we never need info about register 0 for | |
3402 any of the other register vectors, and it seems rather a kludge to | |
3403 treat `best_regend' differently than the rest. So we keep track of | |
3404 the end of the best match so far in a separate variable. We | |
3405 initialize this to NULL so that when we backtrack the first time | |
3406 and need to test it, it's not garbage. */ | |
3407 const char *match_end = NULL; | |
3408 | |
3409 /* Used when we pop values we don't care about. */ | |
2952 | 3410 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
1155 | 3411 const char **reg_dummy; |
3412 register_info_type *reg_info_dummy; | |
2949 | 3413 #endif |
1155 | 3414 |
3415 #ifdef DEBUG | |
3416 /* Counts the total number of registers pushed. */ | |
3417 unsigned num_regs_pushed = 0; | |
3418 #endif | |
3419 | |
3420 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
3421 | |
3422 INIT_FAIL_STACK (); | |
3423 | |
2952 | 3424 #ifdef MATCH_MAY_ALLOCATE |
1155 | 3425 /* Do not bother to initialize all the register variables if there are |
3426 no groups in the pattern, as it takes a fair amount of time. If | |
3427 there are groups, we include space for register 0 (the whole | |
3428 pattern), even though we never use it, since it simplifies the | |
3429 array indexing. We should fix this. */ | |
3430 if (bufp->re_nsub) | |
3431 { | |
3432 regstart = REGEX_TALLOC (num_regs, const char *); | |
3433 regend = REGEX_TALLOC (num_regs, const char *); | |
3434 old_regstart = REGEX_TALLOC (num_regs, const char *); | |
3435 old_regend = REGEX_TALLOC (num_regs, const char *); | |
3436 best_regstart = REGEX_TALLOC (num_regs, const char *); | |
3437 best_regend = REGEX_TALLOC (num_regs, const char *); | |
3438 reg_info = REGEX_TALLOC (num_regs, register_info_type); | |
3439 reg_dummy = REGEX_TALLOC (num_regs, const char *); | |
3440 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type); | |
3441 | |
3442 if (!(regstart && regend && old_regstart && old_regend && reg_info | |
3443 && best_regstart && best_regend && reg_dummy && reg_info_dummy)) | |
3444 { | |
3445 FREE_VARIABLES (); | |
3446 return -2; | |
3447 } | |
3448 } | |
2949 | 3449 #if defined (REGEX_MALLOC) |
1155 | 3450 else |
3451 { | |
3452 /* We must initialize all our variables to NULL, so that | |
1637 | 3453 `FREE_VARIABLES' doesn't try to free them. */ |
1155 | 3454 regstart = regend = old_regstart = old_regend = best_regstart |
3455 = best_regend = reg_dummy = NULL; | |
3456 reg_info = reg_info_dummy = (register_info_type *) NULL; | |
3457 } | |
3458 #endif /* REGEX_MALLOC */ | |
2952 | 3459 #endif /* MATCH_MAY_ALLOCATE */ |
1155 | 3460 |
3461 /* The starting position is bogus. */ | |
3462 if (pos < 0 || pos > size1 + size2) | |
3463 { | |
3464 FREE_VARIABLES (); | |
3465 return -1; | |
3466 } | |
3467 | |
3468 /* Initialize subexpression text positions to -1 to mark ones that no | |
3469 start_memory/stop_memory has been seen for. Also initialize the | |
3470 register information struct. */ | |
3471 for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3472 { | |
3473 regstart[mcnt] = regend[mcnt] | |
3474 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; | |
3475 | |
3476 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; | |
3477 IS_ACTIVE (reg_info[mcnt]) = 0; | |
3478 MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
3479 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
3480 } | |
3481 | |
3482 /* We move `string1' into `string2' if the latter's empty -- but not if | |
3483 `string1' is null. */ | |
3484 if (size2 == 0 && string1 != NULL) | |
3485 { | |
3486 string2 = string1; | |
3487 size2 = size1; | |
3488 string1 = 0; | |
3489 size1 = 0; | |
3490 } | |
3491 end1 = string1 + size1; | |
3492 end2 = string2 + size2; | |
3493 | |
3494 /* Compute where to stop matching, within the two strings. */ | |
3495 if (stop <= size1) | |
3496 { | |
3497 end_match_1 = string1 + stop; | |
3498 end_match_2 = string2; | |
3499 } | |
3500 else | |
3501 { | |
3502 end_match_1 = end1; | |
3503 end_match_2 = string2 + stop - size1; | |
3504 } | |
3505 | |
3506 /* `p' scans through the pattern as `d' scans through the data. | |
3507 `dend' is the end of the input string that `d' points within. `d' | |
3508 is advanced into the following input string whenever necessary, but | |
3509 this happens before fetching; therefore, at the beginning of the | |
3510 loop, `d' can be pointing at the end of a string, but it cannot | |
3511 equal `string2'. */ | |
3512 if (size1 > 0 && pos <= size1) | |
3513 { | |
3514 d = string1 + pos; | |
3515 dend = end_match_1; | |
3516 } | |
3517 else | |
3518 { | |
3519 d = string2 + pos - size1; | |
3520 dend = end_match_2; | |
3521 } | |
3522 | |
3523 DEBUG_PRINT1 ("The compiled pattern is: "); | |
3524 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
3525 DEBUG_PRINT1 ("The string to match is: `"); | |
3526 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
3527 DEBUG_PRINT1 ("'\n"); | |
3528 | |
3529 /* This loops over pattern commands. It exits by returning from the | |
3530 function if the match is complete, or it drops through if the match | |
3531 fails at this starting point in the input data. */ | |
3532 for (;;) | |
3533 { | |
3534 DEBUG_PRINT2 ("\n0x%x: ", p); | |
3535 | |
3536 if (p == pend) | |
3537 { /* End of pattern means we might have succeeded. */ | |
1637 | 3538 DEBUG_PRINT1 ("end of pattern ... "); |
3539 | |
3540 /* If we haven't matched the entire string, and we want the | |
3541 longest match, try backtracking. */ | |
1155 | 3542 if (d != end_match_2) |
3543 { | |
3544 DEBUG_PRINT1 ("backtracking.\n"); | |
3545 | |
3546 if (!FAIL_STACK_EMPTY ()) | |
3547 { /* More failure points to try. */ | |
3548 boolean same_str_p = (FIRST_STRING_P (match_end) | |
3549 == MATCHING_IN_FIRST_STRING); | |
3550 | |
3551 /* If exceeds best match so far, save it. */ | |
3552 if (!best_regs_set | |
3553 || (same_str_p && d > match_end) | |
3554 || (!same_str_p && !MATCHING_IN_FIRST_STRING)) | |
3555 { | |
3556 best_regs_set = true; | |
3557 match_end = d; | |
3558 | |
3559 DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); | |
3560 | |
3561 for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3562 { | |
3563 best_regstart[mcnt] = regstart[mcnt]; | |
3564 best_regend[mcnt] = regend[mcnt]; | |
3565 } | |
3566 } | |
3567 goto fail; | |
3568 } | |
3569 | |
3570 /* If no failure points, don't restore garbage. */ | |
3571 else if (best_regs_set) | |
3572 { | |
3573 restore_best_regs: | |
3574 /* Restore best match. It may happen that `dend == | |
3575 end_match_1' while the restored d is in string2. | |
3576 For example, the pattern `x.*y.*z' against the | |
3577 strings `x-' and `y-z-', if the two strings are | |
3578 not consecutive in memory. */ | |
1637 | 3579 DEBUG_PRINT1 ("Restoring best registers.\n"); |
3580 | |
1155 | 3581 d = match_end; |
3582 dend = ((d >= string1 && d <= end1) | |
3583 ? end_match_1 : end_match_2); | |
3584 | |
3585 for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3586 { | |
3587 regstart[mcnt] = best_regstart[mcnt]; | |
3588 regend[mcnt] = best_regend[mcnt]; | |
3589 } | |
3590 } | |
3591 } /* d != end_match_2 */ | |
3592 | |
1637 | 3593 DEBUG_PRINT1 ("Accepting match.\n"); |
1155 | 3594 |
3595 /* If caller wants register contents data back, do it. */ | |
3596 if (regs && !bufp->no_sub) | |
3597 { | |
3598 /* Have the register data arrays been allocated? */ | |
3599 if (bufp->regs_allocated == REGS_UNALLOCATED) | |
3600 { /* No. So allocate them with malloc. We need one | |
3601 extra element beyond `num_regs' for the `-1' marker | |
3602 GNU code uses. */ | |
3603 regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
3604 regs->start = TALLOC (regs->num_regs, regoff_t); | |
3605 regs->end = TALLOC (regs->num_regs, regoff_t); | |
3606 if (regs->start == NULL || regs->end == NULL) | |
3607 return -2; | |
3608 bufp->regs_allocated = REGS_REALLOCATE; | |
3609 } | |
3610 else if (bufp->regs_allocated == REGS_REALLOCATE) | |
3611 { /* Yes. If we need more elements than were already | |
3612 allocated, reallocate them. If we need fewer, just | |
3613 leave it alone. */ | |
3614 if (regs->num_regs < num_regs + 1) | |
3615 { | |
3616 regs->num_regs = num_regs + 1; | |
3617 RETALLOC (regs->start, regs->num_regs, regoff_t); | |
3618 RETALLOC (regs->end, regs->num_regs, regoff_t); | |
3619 if (regs->start == NULL || regs->end == NULL) | |
3620 return -2; | |
3621 } | |
3622 } | |
3623 else | |
2465 | 3624 { |
3625 /* These braces fend off a "empty body in an else-statement" | |
3626 warning under GCC when assert expands to nothing. */ | |
3627 assert (bufp->regs_allocated == REGS_FIXED); | |
3628 } | |
1155 | 3629 |
3630 /* Convert the pointer data in `regstart' and `regend' to | |
3631 indices. Register zero has to be set differently, | |
3632 since we haven't kept track of any info for it. */ | |
3633 if (regs->num_regs > 0) | |
3634 { | |
3635 regs->start[0] = pos; | |
4918
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3636 regs->end[0] = (MATCHING_IN_FIRST_STRING |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3637 ? ((regoff_t) (d - string1)) |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3638 : ((regoff_t) (d - string2 + size1))); |
1155 | 3639 } |
3640 | |
3641 /* Go through the first `min (num_regs, regs->num_regs)' | |
3642 registers, since that is all we initialized. */ | |
3643 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++) | |
3644 { | |
3645 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) | |
3646 regs->start[mcnt] = regs->end[mcnt] = -1; | |
3647 else | |
3648 { | |
4918
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3649 regs->start[mcnt] |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3650 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]); |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3651 regs->end[mcnt] |
e928d39564ad
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
4846
diff
changeset
|
3652 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]); |
1155 | 3653 } |
3654 } | |
3655 | |
3656 /* If the regs structure we return has more elements than | |
3657 were in the pattern, set the extra elements to -1. If | |
3658 we (re)allocated the registers, this is the case, | |
3659 because we always allocate enough to have at least one | |
3660 -1 at the end. */ | |
3661 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++) | |
3662 regs->start[mcnt] = regs->end[mcnt] = -1; | |
3663 } /* regs && !bufp->no_sub */ | |
3664 | |
3665 FREE_VARIABLES (); | |
1637 | 3666 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", |
3667 nfailure_points_pushed, nfailure_points_popped, | |
3668 nfailure_points_pushed - nfailure_points_popped); | |
3669 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); | |
1155 | 3670 |
3671 mcnt = d - pos - (MATCHING_IN_FIRST_STRING | |
3672 ? string1 | |
3673 : string2 - size1); | |
3674 | |
3675 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); | |
3676 | |
3677 return mcnt; | |
3678 } | |
3679 | |
3680 /* Otherwise match next pattern command. */ | |
3681 #ifdef SWITCH_ENUM_BUG | |
3682 switch ((int) ((re_opcode_t) *p++)) | |
3683 #else | |
3684 switch ((re_opcode_t) *p++) | |
3685 #endif | |
3686 { | |
3687 /* Ignore these. Used to ignore the n of succeed_n's which | |
3688 currently have n == 0. */ | |
3689 case no_op: | |
3690 DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
3691 break; | |
3692 | |
3693 | |
3694 /* Match the next n pattern characters exactly. The following | |
3695 byte in the pattern defines n, and the n bytes after that | |
3696 are the characters to match. */ | |
3697 case exactn: | |
3698 mcnt = *p++; | |
3699 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); | |
3700 | |
3701 /* This is written out as an if-else so we don't waste time | |
3702 testing `translate' inside the loop. */ | |
3703 if (translate) | |
3704 { | |
3705 do | |
3706 { | |
3707 PREFETCH (); | |
3708 if (translate[(unsigned char) *d++] != (char) *p++) | |
3709 goto fail; | |
3710 } | |
3711 while (--mcnt); | |
3712 } | |
3713 else | |
3714 { | |
3715 do | |
3716 { | |
3717 PREFETCH (); | |
3718 if (*d++ != (char) *p++) goto fail; | |
3719 } | |
3720 while (--mcnt); | |
3721 } | |
3722 SET_REGS_MATCHED (); | |
3723 break; | |
3724 | |
3725 | |
3726 /* Match any character except possibly a newline or a null. */ | |
3727 case anychar: | |
3728 DEBUG_PRINT1 ("EXECUTING anychar.\n"); | |
3729 | |
3730 PREFETCH (); | |
3731 | |
3732 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n') | |
3733 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000')) | |
3734 goto fail; | |
3735 | |
3736 SET_REGS_MATCHED (); | |
3737 DEBUG_PRINT2 (" Matched `%d'.\n", *d); | |
3738 d++; | |
3739 break; | |
3740 | |
3741 | |
3742 case charset: | |
3743 case charset_not: | |
3744 { | |
3745 register unsigned char c; | |
3746 boolean not = (re_opcode_t) *(p - 1) == charset_not; | |
3747 | |
3748 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); | |
3749 | |
3750 PREFETCH (); | |
3751 c = TRANSLATE (*d); /* The character to match. */ | |
3752 | |
3753 /* Cast to `unsigned' instead of `unsigned char' in case the | |
3754 bit list is a full 32 bytes long. */ | |
3755 if (c < (unsigned) (*p * BYTEWIDTH) | |
3756 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
3757 not = !not; | |
3758 | |
3759 p += 1 + *p; | |
3760 | |
3761 if (!not) goto fail; | |
3762 | |
3763 SET_REGS_MATCHED (); | |
3764 d++; | |
3765 break; | |
3766 } | |
3767 | |
3768 | |
3769 /* The beginning of a group is represented by start_memory. | |
3770 The arguments are the register number in the next byte, and the | |
3771 number of groups inner to this one in the next. The text | |
3772 matched within the group is recorded (in the internal | |
3773 registers data structure) under the register number. */ | |
3774 case start_memory: | |
3775 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]); | |
3776 | |
3777 /* Find out if this group can match the empty string. */ | |
3778 p1 = p; /* To send to group_match_null_string_p. */ | |
3779 | |
3780 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) | |
3781 REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3782 = group_match_null_string_p (&p1, pend, reg_info); | |
3783 | |
3784 /* Save the position in the string where we were the last time | |
3785 we were at this open-group operator in case the group is | |
3786 operated upon by a repetition operator, e.g., with `(a*)*b' | |
3787 against `ab'; then we want to ignore where we are now in | |
3788 the string in case this attempt to match fails. */ | |
3789 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3790 ? REG_UNSET (regstart[*p]) ? d : regstart[*p] | |
3791 : regstart[*p]; | |
3792 DEBUG_PRINT2 (" old_regstart: %d\n", | |
3793 POINTER_TO_OFFSET (old_regstart[*p])); | |
3794 | |
3795 regstart[*p] = d; | |
3796 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); | |
3797 | |
3798 IS_ACTIVE (reg_info[*p]) = 1; | |
3799 MATCHED_SOMETHING (reg_info[*p]) = 0; | |
3800 | |
3801 /* This is the new highest active register. */ | |
3802 highest_active_reg = *p; | |
3803 | |
3804 /* If nothing was active before, this is the new lowest active | |
3805 register. */ | |
3806 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
3807 lowest_active_reg = *p; | |
3808 | |
3809 /* Move past the register number and inner group count. */ | |
3810 p += 2; | |
5842 | 3811 just_past_start_mem = p; |
1155 | 3812 break; |
3813 | |
3814 | |
3815 /* The stop_memory opcode represents the end of a group. Its | |
3816 arguments are the same as start_memory's: the register | |
3817 number, and the number of inner groups. */ | |
3818 case stop_memory: | |
3819 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]); | |
3820 | |
3821 /* We need to save the string position the last time we were at | |
3822 this close-group operator in case the group is operated | |
3823 upon by a repetition operator, e.g., with `((a*)*(b*)*)*' | |
3824 against `aba'; then we want to ignore where we are now in | |
3825 the string in case this attempt to match fails. */ | |
3826 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3827 ? REG_UNSET (regend[*p]) ? d : regend[*p] | |
3828 : regend[*p]; | |
3829 DEBUG_PRINT2 (" old_regend: %d\n", | |
3830 POINTER_TO_OFFSET (old_regend[*p])); | |
3831 | |
3832 regend[*p] = d; | |
3833 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); | |
3834 | |
3835 /* This register isn't active anymore. */ | |
3836 IS_ACTIVE (reg_info[*p]) = 0; | |
3837 | |
3838 /* If this was the only register active, nothing is active | |
3839 anymore. */ | |
3840 if (lowest_active_reg == highest_active_reg) | |
3841 { | |
3842 lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3843 highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3844 } | |
3845 else | |
3846 { /* We must scan for the new highest active register, since | |
3847 it isn't necessarily one less than now: consider | |
3848 (a(b)c(d(e)f)g). When group 3 ends, after the f), the | |
3849 new highest active register is 1. */ | |
3850 unsigned char r = *p - 1; | |
3851 while (r > 0 && !IS_ACTIVE (reg_info[r])) | |
3852 r--; | |
3853 | |
3854 /* If we end up at register zero, that means that we saved | |
3855 the registers as the result of an `on_failure_jump', not | |
3856 a `start_memory', and we jumped to past the innermost | |
3857 `stop_memory'. For example, in ((.)*) we save | |
3858 registers 1 and 2 as a result of the *, but when we pop | |
3859 back to the second ), we are at the stop_memory 1. | |
3860 Thus, nothing is active. */ | |
3861 if (r == 0) | |
3862 { | |
3863 lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3864 highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3865 } | |
3866 else | |
3867 highest_active_reg = r; | |
3868 } | |
3869 | |
3870 /* If just failed to match something this time around with a | |
3871 group that's operated on by a repetition operator, try to | |
1637 | 3872 force exit from the ``loop'', and restore the register |
1155 | 3873 information for this group that we had before trying this |
3874 last match. */ | |
3875 if ((!MATCHED_SOMETHING (reg_info[*p]) | |
5842 | 3876 || just_past_start_mem == p - 1) |
1155 | 3877 && (p + 2) < pend) |
3878 { | |
3879 boolean is_a_jump_n = false; | |
3880 | |
3881 p1 = p + 2; | |
3882 mcnt = 0; | |
3883 switch ((re_opcode_t) *p1++) | |
3884 { | |
3885 case jump_n: | |
3886 is_a_jump_n = true; | |
3887 case pop_failure_jump: | |
3888 case maybe_pop_jump: | |
3889 case jump: | |
3890 case dummy_failure_jump: | |
3891 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
3892 if (is_a_jump_n) | |
3893 p1 += 2; | |
3894 break; | |
3895 | |
3896 default: | |
3897 /* do nothing */ ; | |
3898 } | |
3899 p1 += mcnt; | |
3900 | |
3901 /* If the next operation is a jump backwards in the pattern | |
3902 to an on_failure_jump right before the start_memory | |
3903 corresponding to this stop_memory, exit from the loop | |
3904 by forcing a failure after pushing on the stack the | |
3905 on_failure_jump's jump in the pattern, and d. */ | |
3906 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump | |
3907 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p) | |
3908 { | |
3909 /* If this group ever matched anything, then restore | |
3910 what its registers were before trying this last | |
3911 failed match, e.g., with `(a*)*b' against `ab' for | |
3912 regstart[1], and, e.g., with `((a*)*(b*)*)*' | |
3913 against `aba' for regend[3]. | |
3914 | |
3915 Also restore the registers for inner groups for, | |
3916 e.g., `((a*)(b*))*' against `aba' (register 3 would | |
3917 otherwise get trashed). */ | |
3918 | |
3919 if (EVER_MATCHED_SOMETHING (reg_info[*p])) | |
3920 { | |
3921 unsigned r; | |
3922 | |
3923 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; | |
3924 | |
3925 /* Restore this and inner groups' (if any) registers. */ | |
3926 for (r = *p; r < *p + *(p + 1); r++) | |
3927 { | |
3928 regstart[r] = old_regstart[r]; | |
3929 | |
3930 /* xx why this test? */ | |
3931 if ((int) old_regend[r] >= (int) regstart[r]) | |
3932 regend[r] = old_regend[r]; | |
3933 } | |
3934 } | |
3935 p1++; | |
3936 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
3937 PUSH_FAILURE_POINT (p1 + mcnt, d, -2); | |
3938 | |
3939 goto fail; | |
3940 } | |
3941 } | |
3942 | |
3943 /* Move past the register number and the inner group count. */ | |
3944 p += 2; | |
3945 break; | |
3946 | |
3947 | |
3948 /* \<digit> has been turned into a `duplicate' command which is | |
3949 followed by the numeric value of <digit> as the register number. */ | |
3950 case duplicate: | |
3951 { | |
3952 register const char *d2, *dend2; | |
3953 int regno = *p++; /* Get which register to match against. */ | |
3954 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); | |
3955 | |
3956 /* Can't back reference a group which we've never matched. */ | |
3957 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
3958 goto fail; | |
3959 | |
3960 /* Where in input to try to start matching. */ | |
3961 d2 = regstart[regno]; | |
3962 | |
3963 /* Where to stop matching; if both the place to start and | |
3964 the place to stop matching are in the same string, then | |
3965 set to the place to stop, otherwise, for now have to use | |
3966 the end of the first string. */ | |
3967 | |
3968 dend2 = ((FIRST_STRING_P (regstart[regno]) | |
3969 == FIRST_STRING_P (regend[regno])) | |
3970 ? regend[regno] : end_match_1); | |
3971 for (;;) | |
3972 { | |
3973 /* If necessary, advance to next segment in register | |
3974 contents. */ | |
3975 while (d2 == dend2) | |
3976 { | |
3977 if (dend2 == end_match_2) break; | |
3978 if (dend2 == regend[regno]) break; | |
3979 | |
3980 /* End of string1 => advance to string2. */ | |
3981 d2 = string2; | |
3982 dend2 = regend[regno]; | |
3983 } | |
3984 /* At end of register contents => success */ | |
3985 if (d2 == dend2) break; | |
3986 | |
3987 /* If necessary, advance to next segment in data. */ | |
3988 PREFETCH (); | |
3989 | |
3990 /* How many characters left in this segment to match. */ | |
3991 mcnt = dend - d; | |
3992 | |
3993 /* Want how many consecutive characters we can match in | |
3994 one shot, so, if necessary, adjust the count. */ | |
3995 if (mcnt > dend2 - d2) | |
3996 mcnt = dend2 - d2; | |
3997 | |
3998 /* Compare that many; failure if mismatch, else move | |
3999 past them. */ | |
4000 if (translate | |
4001 ? bcmp_translate (d, d2, mcnt, translate) | |
4002 : bcmp (d, d2, mcnt)) | |
4003 goto fail; | |
4004 d += mcnt, d2 += mcnt; | |
4005 } | |
4006 } | |
4007 break; | |
4008 | |
4009 | |
4010 /* begline matches the empty string at the beginning of the string | |
4011 (unless `not_bol' is set in `bufp'), and, if | |
4012 `newline_anchor' is set, after newlines. */ | |
4013 case begline: | |
4014 DEBUG_PRINT1 ("EXECUTING begline.\n"); | |
4015 | |
1637 | 4016 if (AT_STRINGS_BEG (d)) |
1155 | 4017 { |
4018 if (!bufp->not_bol) break; | |
4019 } | |
4020 else if (d[-1] == '\n' && bufp->newline_anchor) | |
4021 { | |
4022 break; | |
4023 } | |
4024 /* In all other cases, we fail. */ | |
4025 goto fail; | |
4026 | |
4027 | |
4028 /* endline is the dual of begline. */ | |
4029 case endline: | |
4030 DEBUG_PRINT1 ("EXECUTING endline.\n"); | |
4031 | |
1637 | 4032 if (AT_STRINGS_END (d)) |
1155 | 4033 { |
4034 if (!bufp->not_eol) break; | |
4035 } | |
4036 | |
4037 /* We have to ``prefetch'' the next character. */ | |
4038 else if ((d == end1 ? *string2 : *d) == '\n' | |
4039 && bufp->newline_anchor) | |
4040 { | |
4041 break; | |
4042 } | |
4043 goto fail; | |
4044 | |
4045 | |
4046 /* Match at the very beginning of the data. */ | |
4047 case begbuf: | |
4048 DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
1637 | 4049 if (AT_STRINGS_BEG (d)) |
1155 | 4050 break; |
4051 goto fail; | |
4052 | |
4053 | |
4054 /* Match at the very end of the data. */ | |
4055 case endbuf: | |
4056 DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
1637 | 4057 if (AT_STRINGS_END (d)) |
1155 | 4058 break; |
4059 goto fail; | |
4060 | |
4061 | |
4062 /* on_failure_keep_string_jump is used to optimize `.*\n'. It | |
4063 pushes NULL as the value for the string on the stack. Then | |
4064 `pop_failure_point' will keep the current value for the | |
4065 string, instead of restoring it. To see why, consider | |
4066 matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
4067 then the . fails against the \n. But the next thing we want | |
4068 to do is match the \n against the \n; if we restored the | |
4069 string value, we would be back at the foo. | |
4070 | |
4071 Because this is used only in specific cases, we don't need to | |
4072 check all the things that `on_failure_jump' does, to make | |
4073 sure the right things get saved on the stack. Hence we don't | |
4074 share its code. The only reason to push anything on the | |
4075 stack at all is that otherwise we would have to change | |
4076 `anychar's code to do something besides goto fail in this | |
4077 case; that seems worse than this. */ | |
4078 case on_failure_keep_string_jump: | |
4079 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); | |
4080 | |
4081 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4082 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); | |
4083 | |
4084 PUSH_FAILURE_POINT (p + mcnt, NULL, -2); | |
4085 break; | |
4086 | |
4087 | |
4088 /* Uses of on_failure_jump: | |
4089 | |
4090 Each alternative starts with an on_failure_jump that points | |
4091 to the beginning of the next alternative. Each alternative | |
4092 except the last ends with a jump that in effect jumps past | |
4093 the rest of the alternatives. (They really jump to the | |
4094 ending jump of the following alternative, because tensioning | |
4095 these jumps is a hassle.) | |
4096 | |
4097 Repeats start with an on_failure_jump that points past both | |
4098 the repetition text and either the following jump or | |
4099 pop_failure_jump back to this on_failure_jump. */ | |
4100 case on_failure_jump: | |
4101 on_failure: | |
4102 DEBUG_PRINT1 ("EXECUTING on_failure_jump"); | |
4103 | |
4104 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4105 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); | |
4106 | |
4107 /* If this on_failure_jump comes right before a group (i.e., | |
4108 the original * applied to a group), save the information | |
4109 for that group and all inner ones, so that if we fail back | |
4110 to this point, the group's information will be correct. | |
1637 | 4111 For example, in \(a*\)*\1, we need the preceding group, |
1155 | 4112 and in \(\(a*\)b*\)\2, we need the inner group. */ |
4113 | |
4114 /* We can't use `p' to check ahead because we push | |
4115 a failure point to `p + mcnt' after we do this. */ | |
4116 p1 = p; | |
4117 | |
4118 /* We need to skip no_op's before we look for the | |
4119 start_memory in case this on_failure_jump is happening as | |
4120 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 | |
4121 against aba. */ | |
4122 while (p1 < pend && (re_opcode_t) *p1 == no_op) | |
4123 p1++; | |
4124 | |
4125 if (p1 < pend && (re_opcode_t) *p1 == start_memory) | |
4126 { | |
4127 /* We have a new highest active register now. This will | |
4128 get reset at the start_memory we are about to get to, | |
4129 but we will have saved all the registers relevant to | |
4130 this repetition op, as described above. */ | |
4131 highest_active_reg = *(p1 + 1) + *(p1 + 2); | |
4132 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
4133 lowest_active_reg = *(p1 + 1); | |
4134 } | |
4135 | |
4136 DEBUG_PRINT1 (":\n"); | |
4137 PUSH_FAILURE_POINT (p + mcnt, d, -2); | |
4138 break; | |
4139 | |
4140 | |
1637 | 4141 /* A smart repeat ends with `maybe_pop_jump'. |
4142 We change it to either `pop_failure_jump' or `jump'. */ | |
1155 | 4143 case maybe_pop_jump: |
4144 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4145 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); | |
4146 { | |
4147 register unsigned char *p2 = p; | |
4148 | |
4149 /* Compare the beginning of the repeat with what in the | |
4150 pattern follows its end. If we can establish that there | |
4151 is nothing that they would both match, i.e., that we | |
4152 would have to backtrack because of (as in, e.g., `a*a') | |
4153 then we can change to pop_failure_jump, because we'll | |
4154 never have to backtrack. | |
4155 | |
4156 This is not true in the case of alternatives: in | |
4157 `(a|ab)*' we do need to backtrack to the `ab' alternative | |
4158 (e.g., if the string was `ab'). But instead of trying to | |
4159 detect that here, the alternative has put on a dummy | |
4160 failure point which is what we will end up popping. */ | |
4161 | |
3541
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4162 /* Skip over open/close-group commands. |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4163 If what follows this loop is a ...+ construct, |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4164 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
|
4165 match at least one of that. */ |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4166 while (1) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4167 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4168 if (p2 + 2 < pend |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4169 && ((re_opcode_t) *p2 == stop_memory |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4170 || (re_opcode_t) *p2 == start_memory)) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4171 p2 += 3; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4172 else if (p2 + 6 < pend |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4173 && (re_opcode_t) *p2 == dummy_failure_jump) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4174 p2 += 6; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4175 else |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4176 break; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4177 } |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4178 |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4179 p1 = p + mcnt; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4180 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4181 to the `maybe_finalize_jump' of this case. Examine what |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4182 follows. */ |
1155 | 4183 |
4184 /* If we're at the end of the pattern, we can change. */ | |
4185 if (p2 == pend) | |
1669 | 4186 { |
4187 /* Consider what happens when matching ":\(.*\)" | |
4188 against ":/". I don't really understand this code | |
4189 yet. */ | |
1155 | 4190 p[-3] = (unsigned char) pop_failure_jump; |
1669 | 4191 DEBUG_PRINT1 |
4192 (" End of pattern: change to `pop_failure_jump'.\n"); | |
1155 | 4193 } |
4194 | |
4195 else if ((re_opcode_t) *p2 == exactn | |
4196 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) | |
4197 { | |
4198 register unsigned char c | |
4199 = *p2 == (unsigned char) endline ? '\n' : p2[2]; | |
3541
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4200 |
1155 | 4201 if ((re_opcode_t) p1[3] == exactn && p1[5] != c) |
1637 | 4202 { |
4203 p[-3] = (unsigned char) pop_failure_jump; | |
4204 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", | |
4205 c, p1[5]); | |
4206 } | |
4207 | |
1155 | 4208 else if ((re_opcode_t) p1[3] == charset |
4209 || (re_opcode_t) p1[3] == charset_not) | |
4210 { | |
4211 int not = (re_opcode_t) p1[3] == charset_not; | |
4212 | |
4213 if (c < (unsigned char) (p1[4] * BYTEWIDTH) | |
4214 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
4215 not = !not; | |
4216 | |
4217 /* `not' is equal to 1 if c would match, which means | |
4218 that we can't change to pop_failure_jump. */ | |
4219 if (!not) | |
4220 { | |
4221 p[-3] = (unsigned char) pop_failure_jump; | |
1637 | 4222 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
1155 | 4223 } |
4224 } | |
4225 } | |
3541
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4226 else if ((re_opcode_t) *p2 == charset) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4227 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4228 register unsigned char c |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4229 = *p2 == (unsigned char) endline ? '\n' : p2[2]; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4230 |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4231 if ((re_opcode_t) p1[3] == exactn |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4232 && ! (p2[1] * BYTEWIDTH > p1[4] |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4233 && (p2[1 + p1[4] / BYTEWIDTH] |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4234 & (1 << (p1[4] % BYTEWIDTH))))) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4235 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4236 p[-3] = (unsigned char) pop_failure_jump; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4237 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4238 c, p1[5]); |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4239 } |
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 else if ((re_opcode_t) p1[3] == charset_not) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4242 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4243 int idx; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4244 /* We win if the charset_not inside the loop |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4245 lists every character listed in the charset after. */ |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4246 for (idx = 0; idx < p2[1]; idx++) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4247 if (! (p2[2 + idx] == 0 |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4248 || (idx < p1[4] |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4249 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4250 break; |
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 if (idx == p2[1]) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4253 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4254 p[-3] = (unsigned char) pop_failure_jump; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4255 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4256 } |
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 else if ((re_opcode_t) p1[3] == charset) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4259 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4260 int idx; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4261 /* We win if the charset inside the loop |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4262 has no overlap with the one after the loop. */ |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4263 for (idx = 0; idx < p2[1] && idx < p1[4]; idx++) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4264 if ((p2[2 + idx] & p1[5 + idx]) != 0) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4265 break; |
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 if (idx == p2[1] || idx == p1[4]) |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4268 { |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4269 p[-3] = (unsigned char) pop_failure_jump; |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4270 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4271 } |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4272 } |
cb4aa2f13edd
*** empty log message ***
Richard M. Stallman <rms@gnu.org>
parents:
2952
diff
changeset
|
4273 } |
1155 | 4274 } |
4275 p -= 2; /* Point at relative address again. */ | |
4276 if ((re_opcode_t) p[-1] != pop_failure_jump) | |
4277 { | |
4278 p[-1] = (unsigned char) jump; | |
1637 | 4279 DEBUG_PRINT1 (" Match => jump.\n"); |
1155 | 4280 goto unconditional_jump; |
4281 } | |
4282 /* Note fall through. */ | |
4283 | |
4284 | |
4285 /* The end of a simple repeat has a pop_failure_jump back to | |
4286 its matching on_failure_jump, where the latter will push a | |
4287 failure point. The pop_failure_jump takes off failure | |
4288 points put on by this pop_failure_jump's matching | |
4289 on_failure_jump; we got through the pattern to here from the | |
4290 matching on_failure_jump, so didn't fail. */ | |
4291 case pop_failure_jump: | |
4292 { | |
4293 /* We need to pass separate storage for the lowest and | |
4294 highest registers, even though we don't care about the | |
4295 actual values. Otherwise, we will restore only one | |
4296 register from the stack, since lowest will == highest in | |
4297 `pop_failure_point'. */ | |
4298 unsigned dummy_low_reg, dummy_high_reg; | |
4299 unsigned char *pdummy; | |
4300 const char *sdummy; | |
4301 | |
4302 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); | |
4303 POP_FAILURE_POINT (sdummy, pdummy, | |
4304 dummy_low_reg, dummy_high_reg, | |
4305 reg_dummy, reg_dummy, reg_info_dummy); | |
4306 } | |
4307 /* Note fall through. */ | |
4308 | |
4309 | |
4310 /* Unconditionally jump (without popping any failure points). */ | |
4311 case jump: | |
4312 unconditional_jump: | |
4313 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ | |
4314 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); | |
4315 p += mcnt; /* Do the jump. */ | |
4316 DEBUG_PRINT2 ("(to 0x%x).\n", p); | |
4317 break; | |
4318 | |
4319 | |
4320 /* We need this opcode so we can detect where alternatives end | |
4321 in `group_match_null_string_p' et al. */ | |
4322 case jump_past_alt: | |
4323 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); | |
4324 goto unconditional_jump; | |
4325 | |
4326 | |
4327 /* Normally, the on_failure_jump pushes a failure point, which | |
4328 then gets popped at pop_failure_jump. We will end up at | |
4329 pop_failure_jump, also, and with a pattern of, say, `a+', we | |
4330 are skipping over the on_failure_jump, so we have to push | |
4331 something meaningless for pop_failure_jump to pop. */ | |
4332 case dummy_failure_jump: | |
4333 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); | |
4334 /* It doesn't matter what we push for the string here. What | |
4335 the code at `fail' tests is the value for the pattern. */ | |
4336 PUSH_FAILURE_POINT (0, 0, -2); | |
4337 goto unconditional_jump; | |
4338 | |
4339 | |
4340 /* At the end of an alternative, we need to push a dummy failure | |
1637 | 4341 point in case we are followed by a `pop_failure_jump', because |
1155 | 4342 we don't want the failure point for the alternative to be |
4343 popped. For example, matching `(a|ab)*' against `aab' | |
4344 requires that we match the `ab' alternative. */ | |
4345 case push_dummy_failure: | |
4346 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); | |
4347 /* See comments just above at `dummy_failure_jump' about the | |
4348 two zeroes. */ | |
4349 PUSH_FAILURE_POINT (0, 0, -2); | |
4350 break; | |
4351 | |
4352 /* Have to succeed matching what follows at least n times. | |
4353 After that, handle like `on_failure_jump'. */ | |
4354 case succeed_n: | |
4355 EXTRACT_NUMBER (mcnt, p + 2); | |
4356 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); | |
4357 | |
4358 assert (mcnt >= 0); | |
4359 /* Originally, this is how many times we HAVE to succeed. */ | |
4360 if (mcnt > 0) | |
4361 { | |
4362 mcnt--; | |
4363 p += 2; | |
4364 STORE_NUMBER_AND_INCR (p, mcnt); | |
4365 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt); | |
4366 } | |
4367 else if (mcnt == 0) | |
4368 { | |
4369 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2); | |
4370 p[2] = (unsigned char) no_op; | |
4371 p[3] = (unsigned char) no_op; | |
4372 goto on_failure; | |
4373 } | |
4374 break; | |
4375 | |
4376 case jump_n: | |
4377 EXTRACT_NUMBER (mcnt, p + 2); | |
4378 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); | |
4379 | |
4380 /* Originally, this is how many times we CAN jump. */ | |
4381 if (mcnt) | |
4382 { | |
4383 mcnt--; | |
4384 STORE_NUMBER (p + 2, mcnt); | |
4385 goto unconditional_jump; | |
4386 } | |
4387 /* If don't have to jump any more, skip over the rest of command. */ | |
4388 else | |
4389 p += 4; | |
4390 break; | |
4391 | |
4392 case set_number_at: | |
4393 { | |
4394 DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); | |
4395 | |
4396 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4397 p1 = p + mcnt; | |
4398 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4399 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); | |
4400 STORE_NUMBER (p1, mcnt); | |
4401 break; | |
4402 } | |
4403 | |
4404 case wordbound: | |
4405 DEBUG_PRINT1 ("EXECUTING wordbound.\n"); | |
4406 if (AT_WORD_BOUNDARY (d)) | |
4407 break; | |
4408 goto fail; | |
4409 | |
4410 case notwordbound: | |
4411 DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); | |
4412 if (AT_WORD_BOUNDARY (d)) | |
4413 goto fail; | |
4414 break; | |
4415 | |
4416 case wordbeg: | |
4417 DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); | |
1637 | 4418 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) |
1155 | 4419 break; |
4420 goto fail; | |
4421 | |
4422 case wordend: | |
4423 DEBUG_PRINT1 ("EXECUTING wordend.\n"); | |
1637 | 4424 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) |
4425 && (!WORDCHAR_P (d) || AT_STRINGS_END (d))) | |
1155 | 4426 break; |
4427 goto fail; | |
4428 | |
4429 #ifdef emacs | |
4430 #ifdef emacs19 | |
4431 case before_dot: | |
4432 DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
4433 if (PTR_CHAR_POS ((unsigned char *) d) >= point) | |
4434 goto fail; | |
4435 break; | |
4436 | |
4437 case at_dot: | |
4438 DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
4439 if (PTR_CHAR_POS ((unsigned char *) d) != point) | |
4440 goto fail; | |
4441 break; | |
4442 | |
4443 case after_dot: | |
4444 DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
4445 if (PTR_CHAR_POS ((unsigned char *) d) <= point) | |
4446 goto fail; | |
4447 break; | |
4448 #else /* not emacs19 */ | |
4449 case at_dot: | |
4450 DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
4451 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point) | |
4452 goto fail; | |
4453 break; | |
4454 #endif /* not emacs19 */ | |
4455 | |
4456 case syntaxspec: | |
4457 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); | |
4458 mcnt = *p++; | |
4459 goto matchsyntax; | |
4460 | |
4461 case wordchar: | |
1637 | 4462 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n"); |
1155 | 4463 mcnt = (int) Sword; |
4464 matchsyntax: | |
4465 PREFETCH (); | |
1637 | 4466 if (SYNTAX (*d++) != (enum syntaxcode) mcnt) |
4467 goto fail; | |
1155 | 4468 SET_REGS_MATCHED (); |
4469 break; | |
4470 | |
4471 case notsyntaxspec: | |
4472 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); | |
4473 mcnt = *p++; | |
4474 goto matchnotsyntax; | |
4475 | |
4476 case notwordchar: | |
1637 | 4477 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n"); |
1155 | 4478 mcnt = (int) Sword; |
1637 | 4479 matchnotsyntax: |
1155 | 4480 PREFETCH (); |
1637 | 4481 if (SYNTAX (*d++) == (enum syntaxcode) mcnt) |
4482 goto fail; | |
1155 | 4483 SET_REGS_MATCHED (); |
4484 break; | |
4485 | |
4486 #else /* not emacs */ | |
4487 case wordchar: | |
4488 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); | |
4489 PREFETCH (); | |
1637 | 4490 if (!WORDCHAR_P (d)) |
1155 | 4491 goto fail; |
4492 SET_REGS_MATCHED (); | |
1637 | 4493 d++; |
1155 | 4494 break; |
4495 | |
4496 case notwordchar: | |
4497 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); | |
4498 PREFETCH (); | |
1637 | 4499 if (WORDCHAR_P (d)) |
1155 | 4500 goto fail; |
4501 SET_REGS_MATCHED (); | |
1637 | 4502 d++; |
1155 | 4503 break; |
4504 #endif /* not emacs */ | |
4505 | |
4506 default: | |
4507 abort (); | |
4508 } | |
4509 continue; /* Successfully executed one pattern command; keep going. */ | |
4510 | |
4511 | |
4512 /* We goto here if a matching operation fails. */ | |
4513 fail: | |
4514 if (!FAIL_STACK_EMPTY ()) | |
4515 { /* A restart point is known. Restore to that state. */ | |
4516 DEBUG_PRINT1 ("\nFAIL:\n"); | |
4517 POP_FAILURE_POINT (d, p, | |
4518 lowest_active_reg, highest_active_reg, | |
4519 regstart, regend, reg_info); | |
4520 | |
4521 /* If this failure point is a dummy, try the next one. */ | |
4522 if (!p) | |
4523 goto fail; | |
4524 | |
4525 /* If we failed to the end of the pattern, don't examine *p. */ | |
4526 assert (p <= pend); | |
4527 if (p < pend) | |
4528 { | |
4529 boolean is_a_jump_n = false; | |
4530 | |
4531 /* If failed to a backwards jump that's part of a repetition | |
4532 loop, need to pop this failure point and use the next one. */ | |
4533 switch ((re_opcode_t) *p) | |
4534 { | |
4535 case jump_n: | |
4536 is_a_jump_n = true; | |
4537 case maybe_pop_jump: | |
4538 case pop_failure_jump: | |
4539 case jump: | |
4540 p1 = p + 1; | |
4541 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4542 p1 += mcnt; | |
4543 | |
4544 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) | |
4545 || (!is_a_jump_n | |
4546 && (re_opcode_t) *p1 == on_failure_jump)) | |
4547 goto fail; | |
4548 break; | |
4549 default: | |
4550 /* do nothing */ ; | |
4551 } | |
4552 } | |
4553 | |
4554 if (d >= string1 && d <= end1) | |
4555 dend = end_match_1; | |
4556 } | |
4557 else | |
4558 break; /* Matching at this starting point really fails. */ | |
4559 } /* for (;;) */ | |
4560 | |
4561 if (best_regs_set) | |
4562 goto restore_best_regs; | |
4563 | |
4564 FREE_VARIABLES (); | |
4565 | |
4566 return -1; /* Failure to match. */ | |
4567 } /* re_match_2 */ | |
4568 | |
4569 /* Subroutine definitions for re_match_2. */ | |
4570 | |
4571 | |
4572 /* We are passed P pointing to a register number after a start_memory. | |
4573 | |
4574 Return true if the pattern up to the corresponding stop_memory can | |
4575 match the empty string, and false otherwise. | |
4576 | |
4577 If we find the matching stop_memory, sets P to point to one past its number. | |
4578 Otherwise, sets P to an undefined byte less than or equal to END. | |
4579 | |
4580 We don't handle duplicates properly (yet). */ | |
4581 | |
4582 static boolean | |
4583 group_match_null_string_p (p, end, reg_info) | |
4584 unsigned char **p, *end; | |
4585 register_info_type *reg_info; | |
4586 { | |
4587 int mcnt; | |
4588 /* Point to after the args to the start_memory. */ | |
4589 unsigned char *p1 = *p + 2; | |
4590 | |
4591 while (p1 < end) | |
4592 { | |
4593 /* Skip over opcodes that can match nothing, and return true or | |
4594 false, as appropriate, when we get to one that can't, or to the | |
4595 matching stop_memory. */ | |
4596 | |
4597 switch ((re_opcode_t) *p1) | |
4598 { | |
4599 /* Could be either a loop or a series of alternatives. */ | |
4600 case on_failure_jump: | |
4601 p1++; | |
4602 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4603 | |
4604 /* If the next operation is not a jump backwards in the | |
4605 pattern. */ | |
4606 | |
4607 if (mcnt >= 0) | |
4608 { | |
4609 /* Go through the on_failure_jumps of the alternatives, | |
4610 seeing if any of the alternatives cannot match nothing. | |
4611 The last alternative starts with only a jump, | |
4612 whereas the rest start with on_failure_jump and end | |
4613 with a jump, e.g., here is the pattern for `a|b|c': | |
4614 | |
4615 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 | |
4616 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 | |
4617 /exactn/1/c | |
4618 | |
4619 So, we have to first go through the first (n-1) | |
4620 alternatives and then deal with the last one separately. */ | |
4621 | |
4622 | |
4623 /* Deal with the first (n-1) alternatives, which start | |
4624 with an on_failure_jump (see above) that jumps to right | |
4625 past a jump_past_alt. */ | |
4626 | |
4627 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt) | |
4628 { | |
4629 /* `mcnt' holds how many bytes long the alternative | |
4630 is, including the ending `jump_past_alt' and | |
4631 its number. */ | |
4632 | |
4633 if (!alt_match_null_string_p (p1, p1 + mcnt - 3, | |
4634 reg_info)) | |
4635 return false; | |
4636 | |
4637 /* Move to right after this alternative, including the | |
4638 jump_past_alt. */ | |
4639 p1 += mcnt; | |
4640 | |
4641 /* Break if it's the beginning of an n-th alternative | |
4642 that doesn't begin with an on_failure_jump. */ | |
4643 if ((re_opcode_t) *p1 != on_failure_jump) | |
4644 break; | |
4645 | |
4646 /* Still have to check that it's not an n-th | |
4647 alternative that starts with an on_failure_jump. */ | |
4648 p1++; | |
4649 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4650 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt) | |
4651 { | |
4652 /* Get to the beginning of the n-th alternative. */ | |
4653 p1 -= 3; | |
4654 break; | |
4655 } | |
4656 } | |
4657 | |
4658 /* Deal with the last alternative: go back and get number | |
4659 of the `jump_past_alt' just before it. `mcnt' contains | |
4660 the length of the alternative. */ | |
4661 EXTRACT_NUMBER (mcnt, p1 - 2); | |
4662 | |
4663 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info)) | |
4664 return false; | |
4665 | |
4666 p1 += mcnt; /* Get past the n-th alternative. */ | |
4667 } /* if mcnt > 0 */ | |
4668 break; | |
4669 | |
4670 | |
4671 case stop_memory: | |
4672 assert (p1[1] == **p); | |
4673 *p = p1 + 2; | |
4674 return true; | |
4675 | |
4676 | |
4677 default: | |
4678 if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
4679 return false; | |
4680 } | |
4681 } /* while p1 < end */ | |
4682 | |
4683 return false; | |
4684 } /* group_match_null_string_p */ | |
4685 | |
4686 | |
4687 /* Similar to group_match_null_string_p, but doesn't deal with alternatives: | |
4688 It expects P to be the first byte of a single alternative and END one | |
4689 byte past the last. The alternative can contain groups. */ | |
4690 | |
4691 static boolean | |
4692 alt_match_null_string_p (p, end, reg_info) | |
4693 unsigned char *p, *end; | |
4694 register_info_type *reg_info; | |
4695 { | |
4696 int mcnt; | |
4697 unsigned char *p1 = p; | |
4698 | |
4699 while (p1 < end) | |
4700 { | |
4701 /* Skip over opcodes that can match nothing, and break when we get | |
4702 to one that can't. */ | |
4703 | |
4704 switch ((re_opcode_t) *p1) | |
4705 { | |
4706 /* It's a loop. */ | |
4707 case on_failure_jump: | |
4708 p1++; | |
4709 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4710 p1 += mcnt; | |
4711 break; | |
4712 | |
4713 default: | |
4714 if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
4715 return false; | |
4716 } | |
4717 } /* while p1 < end */ | |
4718 | |
4719 return true; | |
4720 } /* alt_match_null_string_p */ | |
4721 | |
4722 | |
4723 /* Deals with the ops common to group_match_null_string_p and | |
4724 alt_match_null_string_p. | |
4725 | |
4726 Sets P to one after the op and its arguments, if any. */ | |
4727 | |
4728 static boolean | |
4729 common_op_match_null_string_p (p, end, reg_info) | |
4730 unsigned char **p, *end; | |
4731 register_info_type *reg_info; | |
4732 { | |
4733 int mcnt; | |
4734 boolean ret; | |
4735 int reg_no; | |
4736 unsigned char *p1 = *p; | |
4737 | |
4738 switch ((re_opcode_t) *p1++) | |
4739 { | |
4740 case no_op: | |
4741 case begline: | |
4742 case endline: | |
4743 case begbuf: | |
4744 case endbuf: | |
4745 case wordbeg: | |
4746 case wordend: | |
4747 case wordbound: | |
4748 case notwordbound: | |
4749 #ifdef emacs | |
4750 case before_dot: | |
4751 case at_dot: | |
4752 case after_dot: | |
4753 #endif | |
4754 break; | |
4755 | |
4756 case start_memory: | |
4757 reg_no = *p1; | |
4758 assert (reg_no > 0 && reg_no <= MAX_REGNUM); | |
4759 ret = group_match_null_string_p (&p1, end, reg_info); | |
4760 | |
4761 /* Have to set this here in case we're checking a group which | |
4762 contains a group and a back reference to it. */ | |
4763 | |
4764 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) | |
4765 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; | |
4766 | |
4767 if (!ret) | |
4768 return false; | |
4769 break; | |
4770 | |
4771 /* If this is an optimized succeed_n for zero times, make the jump. */ | |
4772 case jump: | |
4773 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4774 if (mcnt >= 0) | |
4775 p1 += mcnt; | |
4776 else | |
4777 return false; | |
4778 break; | |
4779 | |
4780 case succeed_n: | |
4781 /* Get to the number of times to succeed. */ | |
4782 p1 += 2; | |
4783 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4784 | |
4785 if (mcnt == 0) | |
4786 { | |
4787 p1 -= 4; | |
4788 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4789 p1 += mcnt; | |
4790 } | |
4791 else | |
4792 return false; | |
4793 break; | |
4794 | |
4795 case duplicate: | |
4796 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) | |
4797 return false; | |
4798 break; | |
4799 | |
4800 case set_number_at: | |
4801 p1 += 4; | |
4802 | |
4803 default: | |
4804 /* All other opcodes mean we cannot match the empty string. */ | |
4805 return false; | |
4806 } | |
4807 | |
4808 *p = p1; | |
4809 return true; | |
4810 } /* common_op_match_null_string_p */ | |
4811 | |
4812 | |
4813 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN | |
4814 bytes; nonzero otherwise. */ | |
4815 | |
4816 static int | |
4817 bcmp_translate (s1, s2, len, translate) | |
4818 unsigned char *s1, *s2; | |
4819 register int len; | |
4820 char *translate; | |
4821 { | |
4822 register unsigned char *p1 = s1, *p2 = s2; | |
4823 while (len) | |
4824 { | |
4825 if (translate[*p1++] != translate[*p2++]) return 1; | |
4826 len--; | |
4827 } | |
4828 return 0; | |
4829 } | |
4830 | |
4831 /* Entry points for GNU code. */ | |
4832 | |
4833 /* re_compile_pattern is the GNU regular expression compiler: it | |
4834 compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
4835 Returns 0 if the pattern was valid, otherwise an error string. | |
4836 | |
4837 Assumes the `allocated' (and perhaps `buffer') and `translate' fields | |
4838 are set in BUFP on entry. | |
4839 | |
4840 We call regex_compile to do the actual compilation. */ | |
4841 | |
4842 const char * | |
4843 re_compile_pattern (pattern, length, bufp) | |
4844 const char *pattern; | |
4845 int length; | |
4846 struct re_pattern_buffer *bufp; | |
4847 { | |
4848 reg_errcode_t ret; | |
4849 | |
4850 /* GNU code is written to assume at least RE_NREGS registers will be set | |
4851 (and at least one extra will be -1). */ | |
4852 bufp->regs_allocated = REGS_UNALLOCATED; | |
4853 | |
4854 /* And GNU code determines whether or not to get register information | |
4855 by passing null for the REGS argument to re_match, etc., not by | |
4856 setting no_sub. */ | |
4857 bufp->no_sub = 0; | |
4858 | |
4859 /* Match anchors at newline. */ | |
4860 bufp->newline_anchor = 1; | |
4861 | |
4862 ret = regex_compile (pattern, length, re_syntax_options, bufp); | |
4863 | |
4864 return re_error_msg[(int) ret]; | |
4865 } | |
4866 | |
4867 /* Entry points compatible with 4.2 BSD regex library. We don't define | |
4868 them if this is an Emacs or POSIX compilation. */ | |
4869 | |
4870 #if !defined (emacs) && !defined (_POSIX_SOURCE) | |
4871 | |
4872 /* BSD has one and only one pattern buffer. */ | |
4873 static struct re_pattern_buffer re_comp_buf; | |
4874 | |
4875 char * | |
4876 re_comp (s) | |
4877 const char *s; | |
4878 { | |
4879 reg_errcode_t ret; | |
4880 | |
4881 if (!s) | |
4882 { | |
4883 if (!re_comp_buf.buffer) | |
4884 return "No previous regular expression"; | |
4885 return 0; | |
4886 } | |
4887 | |
4888 if (!re_comp_buf.buffer) | |
4889 { | |
4890 re_comp_buf.buffer = (unsigned char *) malloc (200); | |
4891 if (re_comp_buf.buffer == NULL) | |
4892 return "Memory exhausted"; | |
4893 re_comp_buf.allocated = 200; | |
4894 | |
4895 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
4896 if (re_comp_buf.fastmap == NULL) | |
4897 return "Memory exhausted"; | |
4898 } | |
4899 | |
4900 /* Since `re_exec' always passes NULL for the `regs' argument, we | |
4901 don't need to initialize the pattern buffer fields which affect it. */ | |
4902 | |
4903 /* Match anchors at newlines. */ | |
4904 re_comp_buf.newline_anchor = 1; | |
4905 | |
4906 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); | |
4907 | |
4908 /* Yes, we're discarding `const' here. */ | |
4909 return (char *) re_error_msg[(int) ret]; | |
4910 } | |
4911 | |
4912 | |
4913 int | |
4914 re_exec (s) | |
4915 const char *s; | |
4916 { | |
4917 const int len = strlen (s); | |
4918 return | |
4919 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
4920 } | |
4921 #endif /* not emacs and not _POSIX_SOURCE */ | |
4922 | |
4923 /* POSIX.2 functions. Don't define these for Emacs. */ | |
4924 | |
4925 #ifndef emacs | |
4926 | |
4927 /* regcomp takes a regular expression as a string and compiles it. | |
4928 | |
4929 PREG is a regex_t *. We do not expect any fields to be initialized, | |
4930 since POSIX says we shouldn't. Thus, we set | |
4931 | |
4932 `buffer' to the compiled pattern; | |
4933 `used' to the length of the compiled pattern; | |
4934 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
4935 REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
4936 RE_SYNTAX_POSIX_BASIC; | |
4937 `newline_anchor' to REG_NEWLINE being set in CFLAGS; | |
4938 `fastmap' and `fastmap_accurate' to zero; | |
4939 `re_nsub' to the number of subexpressions in PATTERN. | |
4940 | |
4941 PATTERN is the address of the pattern string. | |
4942 | |
4943 CFLAGS is a series of bits which affect compilation. | |
4944 | |
4945 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
4946 use POSIX basic syntax. | |
4947 | |
4948 If REG_NEWLINE is set, then . and [^...] don't match newline. | |
4949 Also, regexec will try a match beginning after every newline. | |
4950 | |
4951 If REG_ICASE is set, then we considers upper- and lowercase | |
4952 versions of letters to be equivalent when matching. | |
4953 | |
4954 If REG_NOSUB is set, then when PREG is passed to regexec, that | |
4955 routine will report only success or failure, and nothing about the | |
4956 registers. | |
4957 | |
4958 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for | |
4959 the return codes and their meanings.) */ | |
4960 | |
4961 int | |
4962 regcomp (preg, pattern, cflags) | |
4963 regex_t *preg; | |
4964 const char *pattern; | |
4965 int cflags; | |
4966 { | |
4967 reg_errcode_t ret; | |
4968 unsigned syntax | |
1642
340feb030df1
*** empty log message ***
David J. MacKenzie <djm@gnu.org>
parents:
1641
diff
changeset
|
4969 = (cflags & REG_EXTENDED) ? |
340feb030df1
*** empty log message ***
David J. MacKenzie <djm@gnu.org>
parents:
1641
diff
changeset
|
4970 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; |
1155 | 4971 |
4972 /* regex_compile will allocate the space for the compiled pattern. */ | |
4973 preg->buffer = 0; | |
1642
340feb030df1
*** empty log message ***
David J. MacKenzie <djm@gnu.org>
parents:
1641
diff
changeset
|
4974 preg->allocated = 0; |
2758 | 4975 preg->used = 0; |
1155 | 4976 |
4977 /* Don't bother to use a fastmap when searching. This simplifies the | |
4978 REG_NEWLINE case: if we used a fastmap, we'd have to put all the | |
4979 characters after newlines into the fastmap. This way, we just try | |
4980 every character. */ | |
4981 preg->fastmap = 0; | |
4982 | |
4983 if (cflags & REG_ICASE) | |
4984 { | |
4985 unsigned i; | |
4986 | |
4987 preg->translate = (char *) malloc (CHAR_SET_SIZE); | |
4988 if (preg->translate == NULL) | |
4989 return (int) REG_ESPACE; | |
4990 | |
4991 /* Map uppercase characters to corresponding lowercase ones. */ | |
4992 for (i = 0; i < CHAR_SET_SIZE; i++) | |
1668 | 4993 preg->translate[i] = ISUPPER (i) ? tolower (i) : i; |
1155 | 4994 } |
4995 else | |
4996 preg->translate = NULL; | |
4997 | |
4998 /* If REG_NEWLINE is set, newlines are treated differently. */ | |
4999 if (cflags & REG_NEWLINE) | |
5000 { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
5001 syntax &= ~RE_DOT_NEWLINE; | |
5002 syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
5003 /* It also changes the matching behavior. */ | |
5004 preg->newline_anchor = 1; | |
5005 } | |
5006 else | |
5007 preg->newline_anchor = 0; | |
5008 | |
5009 preg->no_sub = !!(cflags & REG_NOSUB); | |
5010 | |
5011 /* POSIX says a null character in the pattern terminates it, so we | |
5012 can use strlen here in compiling the pattern. */ | |
5013 ret = regex_compile (pattern, strlen (pattern), syntax, preg); | |
5014 | |
5015 /* POSIX doesn't distinguish between an unmatched open-group and an | |
5016 unmatched close-group: both are REG_EPAREN. */ | |
5017 if (ret == REG_ERPAREN) ret = REG_EPAREN; | |
5018 | |
5019 return (int) ret; | |
5020 } | |
5021 | |
5022 | |
5023 /* regexec searches for a given pattern, specified by PREG, in the | |
5024 string STRING. | |
5025 | |
5026 If NMATCH is zero or REG_NOSUB was set in the cflags argument to | |
5027 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at | |
5028 least NMATCH elements, and we set them to the offsets of the | |
5029 corresponding matched substrings. | |
5030 | |
5031 EFLAGS specifies `execution flags' which affect matching: if | |
5032 REG_NOTBOL is set, then ^ does not match at the beginning of the | |
5033 string; if REG_NOTEOL is set, then $ does not match at the end. | |
5034 | |
5035 We return 0 if we find a match and REG_NOMATCH if not. */ | |
5036 | |
5037 int | |
5038 regexec (preg, string, nmatch, pmatch, eflags) | |
5039 const regex_t *preg; | |
5040 const char *string; | |
5041 size_t nmatch; | |
5042 regmatch_t pmatch[]; | |
5043 int eflags; | |
5044 { | |
5045 int ret; | |
5046 struct re_registers regs; | |
5047 regex_t private_preg; | |
5048 int len = strlen (string); | |
5049 boolean want_reg_info = !preg->no_sub && nmatch > 0; | |
5050 | |
5051 private_preg = *preg; | |
5052 | |
5053 private_preg.not_bol = !!(eflags & REG_NOTBOL); | |
5054 private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
5055 | |
5056 /* The user has told us exactly how many registers to return | |
5057 information about, via `nmatch'. We have to pass that on to the | |
5058 matching routines. */ | |
5059 private_preg.regs_allocated = REGS_FIXED; | |
5060 | |
5061 if (want_reg_info) | |
5062 { | |
5063 regs.num_regs = nmatch; | |
5064 regs.start = TALLOC (nmatch, regoff_t); | |
5065 regs.end = TALLOC (nmatch, regoff_t); | |
5066 if (regs.start == NULL || regs.end == NULL) | |
5067 return (int) REG_NOMATCH; | |
5068 } | |
5069 | |
5070 /* Perform the searching operation. */ | |
5071 ret = re_search (&private_preg, string, len, | |
5072 /* start: */ 0, /* range: */ len, | |
5073 want_reg_info ? ®s : (struct re_registers *) 0); | |
5074 | |
5075 /* Copy the register information to the POSIX structure. */ | |
5076 if (want_reg_info) | |
5077 { | |
5078 if (ret >= 0) | |
5079 { | |
5080 unsigned r; | |
5081 | |
5082 for (r = 0; r < nmatch; r++) | |
5083 { | |
5084 pmatch[r].rm_so = regs.start[r]; | |
5085 pmatch[r].rm_eo = regs.end[r]; | |
5086 } | |
5087 } | |
5088 | |
5089 /* If we needed the temporary register info, free the space now. */ | |
5090 free (regs.start); | |
5091 free (regs.end); | |
5092 } | |
5093 | |
5094 /* We want zero return to mean success, unlike `re_search'. */ | |
5095 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
5096 } | |
5097 | |
5098 | |
5099 /* Returns a message corresponding to an error code, ERRCODE, returned | |
1637 | 5100 from either regcomp or regexec. We don't use PREG here. */ |
1155 | 5101 |
5102 size_t | |
5103 regerror (errcode, preg, errbuf, errbuf_size) | |
5104 int errcode; | |
5105 const regex_t *preg; | |
5106 char *errbuf; | |
5107 size_t errbuf_size; | |
5108 { | |
1738 | 5109 const char *msg; |
5110 size_t msg_size; | |
5111 | |
5112 if (errcode < 0 | |
5113 || errcode >= (sizeof (re_error_msg) / sizeof (re_error_msg[0]))) | |
5114 /* Only error codes returned by the rest of the code should be passed | |
5115 to this routine. If we are given anything else, or if other regex | |
5116 code generates an invalid error code, then the program has a bug. | |
5117 Dump core so we can fix it. */ | |
5118 abort (); | |
5119 | |
2453 | 5120 msg = re_error_msg[errcode]; |
5121 | |
5122 /* POSIX doesn't require that we do anything in this case, but why | |
5123 not be nice. */ | |
5124 if (! msg) | |
5125 msg = "Success"; | |
5126 | |
1738 | 5127 msg_size = strlen (msg) + 1; /* Includes the null. */ |
1155 | 5128 |
5129 if (errbuf_size != 0) | |
5130 { | |
5131 if (msg_size > errbuf_size) | |
5132 { | |
5133 strncpy (errbuf, msg, errbuf_size - 1); | |
5134 errbuf[errbuf_size - 1] = 0; | |
5135 } | |
5136 else | |
5137 strcpy (errbuf, msg); | |
5138 } | |
5139 | |
5140 return msg_size; | |
5141 } | |
5142 | |
5143 | |
5144 /* Free dynamically allocated space used by PREG. */ | |
5145 | |
5146 void | |
5147 regfree (preg) | |
5148 regex_t *preg; | |
5149 { | |
5150 if (preg->buffer != NULL) | |
5151 free (preg->buffer); | |
5152 preg->buffer = NULL; | |
5153 | |
5154 preg->allocated = 0; | |
5155 preg->used = 0; | |
5156 | |
5157 if (preg->fastmap != NULL) | |
5158 free (preg->fastmap); | |
5159 preg->fastmap = NULL; | |
5160 preg->fastmap_accurate = 0; | |
5161 | |
5162 if (preg->translate != NULL) | |
5163 free (preg->translate); | |
5164 preg->translate = NULL; | |
5165 } | |
5166 | |
5167 #endif /* not emacs */ | |
5168 | |
5169 /* | |
5170 Local variables: | |
5171 make-backup-files: t | |
5172 version-control: t | |
5173 trim-versions-without-asking: nil | |
5174 End: | |
5175 */ |