Commit | Line | Data |
---|---|---|
bc78d348 KB |
1 | /* Extended regular expression matching and search library, |
2 | version 0.11. | |
3 | (Implements POSIX draft P10003.2/D11.2, except for | |
4 | internationalization features.) | |
5 | ||
6 | Copyright (C) 1985, 89, 90, 91, 92 Free Software Foundation, Inc. | |
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 | ||
29 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ | |
30 | #include <sys/types.h> | |
31 | ||
32 | /* The `emacs' switch turns on certain matching commands | |
33 | that make sense only in Emacs. */ | |
34 | #ifdef emacs | |
35 | ||
36 | #include "config.h" | |
37 | #include "lisp.h" | |
38 | #include "buffer.h" | |
39 | #include "syntax.h" | |
40 | ||
41 | /* Emacs uses `NULL' as a predicate. */ | |
42 | #undef NULL | |
43 | ||
44 | #else /* not emacs */ | |
45 | ||
46 | /* We used to test for `BSTRING' here, but only GCC and Emacs define | |
47 | `BSTRING', as far as I know, and neither of them use this code. */ | |
48 | #if USG || STDC_HEADERS | |
49 | #include <string.h> | |
50 | #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n)) | |
51 | #define bcopy(s, d, n) memcpy ((d), (s), (n)) | |
52 | #define bzero(s, n) memset ((s), 0, (n)) | |
53 | #else | |
54 | #include <strings.h> | |
55 | #endif | |
56 | ||
57 | #ifdef STDC_HEADERS | |
58 | #include <stdlib.h> | |
59 | #else | |
60 | char *malloc (); | |
61 | char *realloc (); | |
62 | #endif | |
63 | ||
64 | ||
65 | /* Define the syntax stuff for \<, \>, etc. */ | |
66 | ||
67 | /* This must be nonzero for the wordchar and notwordchar pattern | |
68 | commands in re_match_2. */ | |
69 | #ifndef Sword | |
70 | #define Sword 1 | |
71 | #endif | |
72 | ||
73 | #ifdef SYNTAX_TABLE | |
74 | ||
75 | extern char *re_syntax_table; | |
76 | ||
77 | #else /* not SYNTAX_TABLE */ | |
78 | ||
79 | /* How many characters in the character set. */ | |
80 | #define CHAR_SET_SIZE 256 | |
81 | ||
82 | static char re_syntax_table[CHAR_SET_SIZE]; | |
83 | ||
84 | static void | |
85 | init_syntax_once () | |
86 | { | |
87 | register int c; | |
88 | static int done = 0; | |
89 | ||
90 | if (done) | |
91 | return; | |
92 | ||
93 | bzero (re_syntax_table, sizeof re_syntax_table); | |
94 | ||
95 | for (c = 'a'; c <= 'z'; c++) | |
96 | re_syntax_table[c] = Sword; | |
97 | ||
98 | for (c = 'A'; c <= 'Z'; c++) | |
99 | re_syntax_table[c] = Sword; | |
100 | ||
101 | for (c = '0'; c <= '9'; c++) | |
102 | re_syntax_table[c] = Sword; | |
103 | ||
104 | re_syntax_table['_'] = Sword; | |
105 | ||
106 | done = 1; | |
107 | } | |
108 | ||
109 | #endif /* not SYNTAX_TABLE */ | |
110 | ||
111 | #define SYNTAX(c) re_syntax_table[c] | |
112 | ||
113 | #endif /* not emacs */ | |
114 | \f | |
115 | /* Get the interface, including the syntax bits. */ | |
116 | #include "regex.h" | |
117 | ||
118 | ||
119 | /* isalpha etc. are used for the character classes. */ | |
120 | #include <ctype.h> | |
121 | #ifndef isgraph | |
122 | #define isgraph(c) (isprint (c) && !isspace (c)) | |
123 | #endif | |
124 | #ifndef isblank | |
125 | #define isblank(c) ((c) == ' ' || (c) == '\t') | |
126 | #endif | |
127 | ||
128 | #ifndef NULL | |
129 | #define NULL 0 | |
130 | #endif | |
131 | ||
132 | /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
133 | since ours (we hope) works properly with all combinations of | |
134 | machines, compilers, `char' and `unsigned char' argument types. | |
135 | (Per Bothner suggested the basic approach.) */ | |
136 | #undef SIGN_EXTEND_CHAR | |
137 | #if __STDC__ | |
138 | #define SIGN_EXTEND_CHAR(c) ((signed char) (c)) | |
139 | #else | |
140 | /* As in Harbison and Steele. */ | |
141 | #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) | |
142 | #endif | |
143 | \f | |
144 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we | |
145 | use `alloca' instead of `malloc'. This is because using malloc in | |
146 | re_search* or re_match* could cause memory leaks when C-g is used in | |
147 | Emacs; also, malloc is slower and causes storage fragmentation. On | |
148 | the other hand, malloc is more portable, and easier to debug. | |
149 | ||
150 | Because we sometimes use alloca, some routines have to be macros, | |
151 | not functions -- `alloca'-allocated space disappears at the end of the | |
152 | function it is called in. */ | |
153 | ||
154 | #ifdef REGEX_MALLOC | |
155 | ||
156 | #define REGEX_ALLOCATE malloc | |
157 | #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) | |
158 | ||
159 | #else /* not REGEX_MALLOC */ | |
160 | ||
161 | /* Emacs already defines alloca, sometimes. */ | |
162 | #ifndef alloca | |
163 | ||
164 | /* Make alloca work the best possible way. */ | |
165 | #ifdef __GNUC__ | |
166 | #define alloca __builtin_alloca | |
167 | #else /* not __GNUC__ */ | |
168 | #if HAVE_ALLOCA_H | |
169 | #include <alloca.h> | |
170 | #else /* not __GNUC__ or HAVE_ALLOCA_H */ | |
171 | #ifndef _AIX /* Already did AIX, up at the top. */ | |
172 | char *alloca (); | |
173 | #endif /* not _AIX */ | |
174 | #endif /* not HAVE_ALLOCA_H */ | |
175 | #endif /* not __GNUC__ */ | |
176 | ||
177 | #endif /* not alloca */ | |
178 | ||
179 | #define REGEX_ALLOCATE alloca | |
180 | ||
181 | /* Assumes a `char *destination' variable. */ | |
182 | #define REGEX_REALLOCATE(source, osize, nsize) \ | |
183 | (destination = (char *) alloca (nsize), \ | |
184 | bcopy (source, destination, osize), \ | |
185 | destination) | |
186 | ||
187 | #endif /* not REGEX_MALLOC */ | |
188 | ||
189 | ||
190 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
191 | `string1' or just past its end. This works if PTR is NULL, which is | |
192 | a good thing. */ | |
193 | #define FIRST_STRING_P(ptr) \ | |
194 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) | |
195 | ||
196 | /* (Re)Allocate N items of type T using malloc, or fail. */ | |
197 | #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) | |
198 | #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) | |
199 | #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) | |
200 | ||
201 | #define BYTEWIDTH 8 /* In bits. */ | |
202 | ||
203 | #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) | |
204 | ||
205 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
206 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
207 | ||
208 | typedef char boolean; | |
209 | #define false 0 | |
210 | #define true 1 | |
211 | \f | |
212 | /* These are the command codes that appear in compiled regular | |
213 | expressions. Some opcodes are followed by argument bytes. A | |
214 | command code can specify any interpretation whatsoever for its | |
215 | arguments. Zero bytes may appear in the compiled regular expression. | |
216 | ||
217 | The value of `exactn' is needed in search.c (search_buffer) in Emacs. | |
218 | So regex.h defines a symbol `RE_EXACTN_VALUE' to be 1; the value of | |
219 | `exactn' we use here must also be 1. */ | |
220 | ||
221 | typedef enum | |
222 | { | |
223 | no_op = 0, | |
224 | ||
225 | /* Followed by one byte giving n, then by n literal bytes. */ | |
226 | exactn = 1, | |
227 | ||
228 | /* Matches any (more or less) character. */ | |
229 | anychar, | |
230 | ||
231 | /* Matches any one char belonging to specified set. First | |
232 | following byte is number of bitmap bytes. Then come bytes | |
233 | for a bitmap saying which chars are in. Bits in each byte | |
234 | are ordered low-bit-first. A character is in the set if its | |
235 | bit is 1. A character too large to have a bit in the map is | |
236 | automatically not in the set. */ | |
237 | charset, | |
238 | ||
239 | /* Same parameters as charset, but match any character that is | |
240 | not one of those specified. */ | |
241 | charset_not, | |
242 | ||
243 | /* Start remembering the text that is matched, for storing in a | |
244 | register. Followed by one byte with the register number, in | |
245 | the range 0 to one less than the pattern buffer's re_nsub | |
246 | field. Then followed by one byte with the number of groups | |
247 | inner to this one. (This last has to be part of the | |
248 | start_memory only because we need it in the on_failure_jump | |
249 | of re_match_2.) */ | |
250 | start_memory, | |
251 | ||
252 | /* Stop remembering the text that is matched and store it in a | |
253 | memory register. Followed by one byte with the register | |
254 | number, in the range 0 to one less than `re_nsub' in the | |
255 | pattern buffer, and one byte with the number of inner groups, | |
256 | just like `start_memory'. (We need the number of inner | |
257 | groups here because we don't have any easy way of finding the | |
258 | corresponding start_memory when we're at a stop_memory.) */ | |
259 | stop_memory, | |
260 | ||
261 | /* Match a duplicate of something remembered. Followed by one | |
262 | byte containing the register number. */ | |
263 | duplicate, | |
264 | ||
265 | /* Fail unless at beginning of line. */ | |
266 | begline, | |
267 | ||
268 | /* Fail unless at end of line. */ | |
269 | endline, | |
270 | ||
271 | /* Succeeds if at beginning of buffer (if emacs) or at beginning | |
272 | of string to be matched (if not). */ | |
273 | begbuf, | |
274 | ||
275 | /* Analogously, for end of buffer/string. */ | |
276 | endbuf, | |
277 | ||
278 | /* Followed by two byte relative address to which to jump. */ | |
279 | jump, | |
280 | ||
281 | /* Same as jump, but marks the end of an alternative. */ | |
282 | jump_past_alt, | |
283 | ||
284 | /* Followed by two-byte relative address of place to resume at | |
285 | in case of failure. */ | |
286 | on_failure_jump, | |
287 | ||
288 | /* Like on_failure_jump, but pushes a placeholder instead of the | |
289 | current string position when executed. */ | |
290 | on_failure_keep_string_jump, | |
291 | ||
292 | /* Throw away latest failure point and then jump to following | |
293 | two-byte relative address. */ | |
294 | pop_failure_jump, | |
295 | ||
296 | /* Change to pop_failure_jump if know won't have to backtrack to | |
297 | match; otherwise change to jump. This is used to jump | |
298 | back to the beginning of a repeat. If what follows this jump | |
299 | clearly won't match what the repeat does, such that we can be | |
300 | sure that there is no use backtracking out of repetitions | |
301 | already matched, then we change it to a pop_failure_jump. | |
302 | Followed by two-byte address. */ | |
303 | maybe_pop_jump, | |
304 | ||
305 | /* Jump to following two-byte address, and push a dummy failure | |
306 | point. This failure point will be thrown away if an attempt | |
307 | is made to use it for a failure. A `+' construct makes this | |
308 | before the first repeat. Also used as an intermediary kind | |
309 | of jump when compiling an alternative. */ | |
310 | dummy_failure_jump, | |
311 | ||
312 | /* Push a dummy failure point and continue. Used at the end of | |
313 | alternatives. */ | |
314 | push_dummy_failure, | |
315 | ||
316 | /* Followed by two-byte relative address and two-byte number n. | |
317 | After matching N times, jump to the address upon failure. */ | |
318 | succeed_n, | |
319 | ||
320 | /* Followed by two-byte relative address, and two-byte number n. | |
321 | Jump to the address N times, then fail. */ | |
322 | jump_n, | |
323 | ||
324 | /* Set the following two-byte relative address to the | |
325 | subsequent two-byte number. The address *includes* the two | |
326 | bytes of number. */ | |
327 | set_number_at, | |
328 | ||
329 | wordchar, /* Matches any word-constituent character. */ | |
330 | notwordchar, /* Matches any char that is not a word-constituent. */ | |
331 | ||
332 | wordbeg, /* Succeeds if at word beginning. */ | |
333 | wordend, /* Succeeds if at word end. */ | |
334 | ||
335 | wordbound, /* Succeeds if at a word boundary. */ | |
336 | notwordbound /* Succeeds if not at a word boundary. */ | |
337 | ||
338 | #ifdef emacs | |
339 | ,before_dot, /* Succeeds if before point. */ | |
340 | at_dot, /* Succeeds if at point. */ | |
341 | after_dot, /* Succeeds if after point. */ | |
342 | ||
343 | /* Matches any character whose syntax is specified. Followed by | |
344 | a byte which contains a syntax code, e.g., Sword. */ | |
345 | syntaxspec, | |
346 | ||
347 | /* Matches any character whose syntax is not that specified. */ | |
348 | notsyntaxspec | |
349 | #endif /* emacs */ | |
350 | } re_opcode_t; | |
351 | \f | |
352 | /* Common operations on the compiled pattern. */ | |
353 | ||
354 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
355 | ||
356 | #define STORE_NUMBER(destination, number) \ | |
357 | do { \ | |
358 | (destination)[0] = (number) & 0377; \ | |
359 | (destination)[1] = (number) >> 8; \ | |
360 | } while (0) | |
361 | ||
362 | /* Same as STORE_NUMBER, except increment DESTINATION to | |
363 | the byte after where the number is stored. Therefore, DESTINATION | |
364 | must be an lvalue. */ | |
365 | ||
366 | #define STORE_NUMBER_AND_INCR(destination, number) \ | |
367 | do { \ | |
368 | STORE_NUMBER (destination, number); \ | |
369 | (destination) += 2; \ | |
370 | } while (0) | |
371 | ||
372 | /* Put into DESTINATION a number stored in two contiguous bytes starting | |
373 | at SOURCE. */ | |
374 | ||
375 | #define EXTRACT_NUMBER(destination, source) \ | |
376 | do { \ | |
377 | (destination) = *(source) & 0377; \ | |
378 | (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
379 | } while (0) | |
380 | ||
381 | #ifdef DEBUG | |
382 | static void | |
383 | extract_number (dest, source) | |
384 | int *dest; | |
385 | unsigned char *source; | |
386 | { | |
387 | int temp = SIGN_EXTEND_CHAR (*(source + 1)); | |
388 | *dest = *source & 0377; | |
389 | *dest += temp << 8; | |
390 | } | |
391 | ||
392 | #ifndef EXTRACT_MACROS /* To debug the macros. */ | |
393 | #undef EXTRACT_NUMBER | |
394 | #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
395 | #endif /* not EXTRACT_MACROS */ | |
396 | ||
397 | #endif /* DEBUG */ | |
398 | ||
399 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
400 | SOURCE must be an lvalue. */ | |
401 | ||
402 | #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
403 | do { \ | |
404 | EXTRACT_NUMBER (destination, source); \ | |
405 | (source) += 2; \ | |
406 | } while (0) | |
407 | ||
408 | #ifdef DEBUG | |
409 | static void | |
410 | extract_number_and_incr (destination, source) | |
411 | int *destination; | |
412 | unsigned char **source; | |
413 | { | |
414 | extract_number (destination, *source); | |
415 | *source += 2; | |
416 | } | |
417 | ||
418 | #ifndef EXTRACT_MACROS | |
419 | #undef EXTRACT_NUMBER_AND_INCR | |
420 | #define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
421 | extract_number_and_incr (&dest, &src) | |
422 | #endif /* not EXTRACT_MACROS */ | |
423 | ||
424 | #endif /* DEBUG */ | |
425 | \f | |
426 | /* If DEBUG is defined, Regex prints many voluminous messages about what | |
427 | it is doing (if the variable `debug' is nonzero). If linked with the | |
428 | main program in `iregex.c', you can enter patterns and strings | |
429 | interactively. And if linked with the main program in `main.c' and | |
430 | the other test files, you can run the already-written tests. */ | |
431 | ||
432 | #ifdef DEBUG | |
433 | ||
434 | /* We use standard I/O for debugging. */ | |
435 | #include <stdio.h> | |
436 | ||
437 | /* It is useful to test things that ``must'' be true when debugging. */ | |
438 | #include <assert.h> | |
439 | ||
440 | static int debug = 0; | |
441 | ||
442 | #define DEBUG_STATEMENT(e) e | |
443 | #define DEBUG_PRINT1(x) if (debug) printf (x) | |
444 | #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) | |
445 | #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) | |
446 | #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ | |
447 | if (debug) print_partial_compiled_pattern (s, e) | |
448 | #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ | |
449 | if (debug) print_double_string (w, s1, sz1, s2, sz2) | |
450 | ||
451 | ||
452 | extern void printchar (); | |
453 | ||
454 | /* Print the fastmap in human-readable form. */ | |
455 | ||
456 | void | |
457 | print_fastmap (fastmap) | |
458 | char *fastmap; | |
459 | { | |
460 | unsigned was_a_range = 0; | |
461 | unsigned i = 0; | |
462 | ||
463 | while (i < (1 << BYTEWIDTH)) | |
464 | { | |
465 | if (fastmap[i++]) | |
466 | { | |
467 | was_a_range = 0; | |
468 | printchar (i - 1); | |
469 | while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
470 | { | |
471 | was_a_range = 1; | |
472 | i++; | |
473 | } | |
474 | if (was_a_range) | |
475 | { | |
476 | printf ("-"); | |
477 | printchar (i - 1); | |
478 | } | |
479 | } | |
480 | } | |
481 | putchar ('\n'); | |
482 | } | |
483 | ||
484 | ||
485 | /* Print a compiled pattern string in human-readable form, starting at | |
486 | the START pointer into it and ending just before the pointer END. */ | |
487 | ||
488 | void | |
489 | print_partial_compiled_pattern (start, end) | |
490 | unsigned char *start; | |
491 | unsigned char *end; | |
492 | { | |
493 | int mcnt, mcnt2; | |
494 | unsigned char *p = start; | |
495 | unsigned char *pend = end; | |
496 | ||
497 | if (start == NULL) | |
498 | { | |
499 | printf ("(null)\n"); | |
500 | return; | |
501 | } | |
502 | ||
503 | /* Loop over pattern commands. */ | |
504 | while (p < pend) | |
505 | { | |
506 | switch ((re_opcode_t) *p++) | |
507 | { | |
508 | case no_op: | |
509 | printf ("/no_op"); | |
510 | break; | |
511 | ||
512 | case exactn: | |
513 | mcnt = *p++; | |
514 | printf ("/exactn/%d", mcnt); | |
515 | do | |
516 | { | |
517 | putchar ('/'); | |
518 | printchar (*p++); | |
519 | } | |
520 | while (--mcnt); | |
521 | break; | |
522 | ||
523 | case start_memory: | |
524 | mcnt = *p++; | |
525 | printf ("/start_memory/%d/%d", mcnt, *p++); | |
526 | break; | |
527 | ||
528 | case stop_memory: | |
529 | mcnt = *p++; | |
530 | printf ("/stop_memory/%d/%d", mcnt, *p++); | |
531 | break; | |
532 | ||
533 | case duplicate: | |
534 | printf ("/duplicate/%d", *p++); | |
535 | break; | |
536 | ||
537 | case anychar: | |
538 | printf ("/anychar"); | |
539 | break; | |
540 | ||
541 | case charset: | |
542 | case charset_not: | |
543 | { | |
544 | register int c; | |
545 | ||
546 | printf ("/charset%s", | |
547 | (re_opcode_t) *(p - 1) == charset_not ? "_not" : ""); | |
548 | ||
549 | assert (p + *p < pend); | |
550 | ||
551 | for (c = 0; c < *p; c++) | |
552 | { | |
553 | unsigned bit; | |
554 | unsigned char map_byte = p[1 + c]; | |
555 | ||
556 | putchar ('/'); | |
557 | ||
558 | for (bit = 0; bit < BYTEWIDTH; bit++) | |
559 | if (map_byte & (1 << bit)) | |
560 | printchar (c * BYTEWIDTH + bit); | |
561 | } | |
562 | p += 1 + *p; | |
563 | break; | |
564 | } | |
565 | ||
566 | case begline: | |
567 | printf ("/begline"); | |
568 | break; | |
569 | ||
570 | case endline: | |
571 | printf ("/endline"); | |
572 | break; | |
573 | ||
574 | case on_failure_jump: | |
575 | extract_number_and_incr (&mcnt, &p); | |
576 | printf ("/on_failure_jump/0/%d", mcnt); | |
577 | break; | |
578 | ||
579 | case on_failure_keep_string_jump: | |
580 | extract_number_and_incr (&mcnt, &p); | |
581 | printf ("/on_failure_keep_string_jump/0/%d", mcnt); | |
582 | break; | |
583 | ||
584 | case dummy_failure_jump: | |
585 | extract_number_and_incr (&mcnt, &p); | |
586 | printf ("/dummy_failure_jump/0/%d", mcnt); | |
587 | break; | |
588 | ||
589 | case push_dummy_failure: | |
590 | printf ("/push_dummy_failure"); | |
591 | break; | |
592 | ||
593 | case maybe_pop_jump: | |
594 | extract_number_and_incr (&mcnt, &p); | |
595 | printf ("/maybe_pop_jump/0/%d", mcnt); | |
596 | break; | |
597 | ||
598 | case pop_failure_jump: | |
599 | extract_number_and_incr (&mcnt, &p); | |
600 | printf ("/pop_failure_jump/0/%d", mcnt); | |
601 | break; | |
602 | ||
603 | case jump_past_alt: | |
604 | extract_number_and_incr (&mcnt, &p); | |
605 | printf ("/jump_past_alt/0/%d", mcnt); | |
606 | break; | |
607 | ||
608 | case jump: | |
609 | extract_number_and_incr (&mcnt, &p); | |
610 | printf ("/jump/0/%d", mcnt); | |
611 | break; | |
612 | ||
613 | case succeed_n: | |
614 | extract_number_and_incr (&mcnt, &p); | |
615 | extract_number_and_incr (&mcnt2, &p); | |
616 | printf ("/succeed_n/0/%d/0/%d", mcnt, mcnt2); | |
617 | break; | |
618 | ||
619 | case jump_n: | |
620 | extract_number_and_incr (&mcnt, &p); | |
621 | extract_number_and_incr (&mcnt2, &p); | |
622 | printf ("/jump_n/0/%d/0/%d", mcnt, mcnt2); | |
623 | break; | |
624 | ||
625 | case set_number_at: | |
626 | extract_number_and_incr (&mcnt, &p); | |
627 | extract_number_and_incr (&mcnt2, &p); | |
628 | printf ("/set_number_at/0/%d/0/%d", mcnt, mcnt2); | |
629 | break; | |
630 | ||
631 | case wordbound: | |
632 | printf ("/wordbound"); | |
633 | break; | |
634 | ||
635 | case notwordbound: | |
636 | printf ("/notwordbound"); | |
637 | break; | |
638 | ||
639 | case wordbeg: | |
640 | printf ("/wordbeg"); | |
641 | break; | |
642 | ||
643 | case wordend: | |
644 | printf ("/wordend"); | |
645 | ||
646 | #ifdef emacs | |
647 | case before_dot: | |
648 | printf ("/before_dot"); | |
649 | break; | |
650 | ||
651 | case at_dot: | |
652 | printf ("/at_dot"); | |
653 | break; | |
654 | ||
655 | case after_dot: | |
656 | printf ("/after_dot"); | |
657 | break; | |
658 | ||
659 | case syntaxspec: | |
660 | printf ("/syntaxspec"); | |
661 | mcnt = *p++; | |
662 | printf ("/%d", mcnt); | |
663 | break; | |
664 | ||
665 | case notsyntaxspec: | |
666 | printf ("/notsyntaxspec"); | |
667 | mcnt = *p++; | |
668 | printf ("/%d", mcnt); | |
669 | break; | |
670 | #endif /* emacs */ | |
671 | ||
672 | case wordchar: | |
673 | printf ("/wordchar"); | |
674 | break; | |
675 | ||
676 | case notwordchar: | |
677 | printf ("/notwordchar"); | |
678 | break; | |
679 | ||
680 | case begbuf: | |
681 | printf ("/begbuf"); | |
682 | break; | |
683 | ||
684 | case endbuf: | |
685 | printf ("/endbuf"); | |
686 | break; | |
687 | ||
688 | default: | |
689 | printf ("?%d", *(p-1)); | |
690 | } | |
691 | } | |
692 | printf ("/\n"); | |
693 | } | |
694 | ||
695 | ||
696 | void | |
697 | print_compiled_pattern (bufp) | |
698 | struct re_pattern_buffer *bufp; | |
699 | { | |
700 | unsigned char *buffer = bufp->buffer; | |
701 | ||
702 | print_partial_compiled_pattern (buffer, buffer + bufp->used); | |
703 | printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated); | |
704 | ||
705 | if (bufp->fastmap_accurate && bufp->fastmap) | |
706 | { | |
707 | printf ("fastmap: "); | |
708 | print_fastmap (bufp->fastmap); | |
709 | } | |
710 | ||
711 | printf ("re_nsub: %d\t", bufp->re_nsub); | |
712 | printf ("regs_alloc: %d\t", bufp->regs_allocated); | |
713 | printf ("can_be_null: %d\t", bufp->can_be_null); | |
714 | printf ("newline_anchor: %d\n", bufp->newline_anchor); | |
715 | printf ("no_sub: %d\t", bufp->no_sub); | |
716 | printf ("not_bol: %d\t", bufp->not_bol); | |
717 | printf ("not_eol: %d\t", bufp->not_eol); | |
718 | printf ("syntax: %d\n", bufp->syntax); | |
719 | /* Perhaps we should print the translate table? */ | |
720 | } | |
721 | ||
722 | ||
723 | void | |
724 | print_double_string (where, string1, size1, string2, size2) | |
725 | const char *where; | |
726 | const char *string1; | |
727 | const char *string2; | |
728 | int size1; | |
729 | int size2; | |
730 | { | |
731 | unsigned this_char; | |
732 | ||
733 | if (where == NULL) | |
734 | printf ("(null)"); | |
735 | else | |
736 | { | |
737 | if (FIRST_STRING_P (where)) | |
738 | { | |
739 | for (this_char = where - string1; this_char < size1; this_char++) | |
740 | printchar (string1[this_char]); | |
741 | ||
742 | where = string2; | |
743 | } | |
744 | ||
745 | for (this_char = where - string2; this_char < size2; this_char++) | |
746 | printchar (string2[this_char]); | |
747 | } | |
748 | } | |
749 | ||
750 | #else /* not DEBUG */ | |
751 | ||
752 | #undef assert | |
753 | #define assert(e) | |
754 | ||
755 | #define DEBUG_STATEMENT(e) | |
756 | #define DEBUG_PRINT1(x) | |
757 | #define DEBUG_PRINT2(x1, x2) | |
758 | #define DEBUG_PRINT3(x1, x2, x3) | |
759 | #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) | |
760 | #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
761 | ||
762 | #endif /* not DEBUG */ | |
763 | \f | |
764 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
765 | also be assigned to arbitrarily: each pattern buffer stores its own | |
766 | syntax, so it can be changed between regex compilations. */ | |
767 | reg_syntax_t re_syntax_options = RE_SYNTAX_EMACS; | |
768 | ||
769 | ||
770 | /* Specify the precise syntax of regexps for compilation. This provides | |
771 | for compatibility for various utilities which historically have | |
772 | different, incompatible syntaxes. | |
773 | ||
774 | The argument SYNTAX is a bit mask comprised of the various bits | |
775 | defined in regex.h. We return the old syntax. */ | |
776 | ||
777 | reg_syntax_t | |
778 | re_set_syntax (syntax) | |
779 | reg_syntax_t syntax; | |
780 | { | |
781 | reg_syntax_t ret = re_syntax_options; | |
782 | ||
783 | re_syntax_options = syntax; | |
784 | return ret; | |
785 | } | |
786 | \f | |
787 | /* This table gives an error message for each of the error codes listed | |
788 | in regex.h. Obviously the order here has to be same as there. */ | |
789 | ||
790 | static const char *re_error_msg[] = | |
791 | { NULL, /* REG_NOERROR */ | |
792 | "No match", /* REG_NOMATCH */ | |
793 | "Invalid regular expression", /* REG_BADPAT */ | |
794 | "Invalid collation character", /* REG_ECOLLATE */ | |
795 | "Invalid character class name", /* REG_ECTYPE */ | |
796 | "Trailing backslash", /* REG_EESCAPE */ | |
797 | "Invalid back reference", /* REG_ESUBREG */ | |
798 | "Unmatched [ or [^", /* REG_EBRACK */ | |
799 | "Unmatched ( or \\(", /* REG_EPAREN */ | |
800 | "Unmatched \\{", /* REG_EBRACE */ | |
801 | "Invalid content of \\{\\}", /* REG_BADBR */ | |
802 | "Invalid range end", /* REG_ERANGE */ | |
803 | "Memory exhausted", /* REG_ESPACE */ | |
804 | "Invalid preceding regular expression", /* REG_BADRPT */ | |
805 | "Premature end of regular expression", /* REG_EEND */ | |
806 | "Regular expression too big", /* REG_ESIZE */ | |
807 | "Unmatched ) or \\)", /* REG_ERPAREN */ | |
808 | }; | |
809 | \f | |
810 | /* Subroutine declarations and macros for regex_compile. */ | |
811 | ||
812 | static void store_op1 (), store_op2 (); | |
813 | static void insert_op1 (), insert_op2 (); | |
814 | static boolean at_begline_loc_p (), at_endline_loc_p (); | |
815 | static boolean group_in_compile_stack (); | |
816 | static reg_errcode_t compile_range (); | |
817 | ||
818 | /* Fetch the next character in the uncompiled pattern---translating it | |
819 | if necessary. Also cast from a signed character in the constant | |
820 | string passed to us by the user to an unsigned char that we can use | |
821 | as an array index (in, e.g., `translate'). */ | |
822 | #define PATFETCH(c) \ | |
823 | do {if (p == pend) return REG_EEND; \ | |
824 | c = (unsigned char) *p++; \ | |
825 | if (translate) c = translate[c]; \ | |
826 | } while (0) | |
827 | ||
828 | /* Fetch the next character in the uncompiled pattern, with no | |
829 | translation. */ | |
830 | #define PATFETCH_RAW(c) \ | |
831 | do {if (p == pend) return REG_EEND; \ | |
832 | c = (unsigned char) *p++; \ | |
833 | } while (0) | |
834 | ||
835 | /* Go backwards one character in the pattern. */ | |
836 | #define PATUNFETCH p-- | |
837 | ||
838 | ||
839 | /* If `translate' is non-null, return translate[D], else just D. We | |
840 | cast the subscript to translate because some data is declared as | |
841 | `char *', to avoid warnings when a string constant is passed. But | |
842 | when we use a character as a subscript we must make it unsigned. */ | |
843 | #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d)) | |
844 | ||
845 | ||
846 | /* Macros for outputting the compiled pattern into `buffer'. */ | |
847 | ||
848 | /* If the buffer isn't allocated when it comes in, use this. */ | |
849 | #define INIT_BUF_SIZE 32 | |
850 | ||
851 | /* Make sure we have at least N more bytes of space in buffer. */ | |
852 | #define GET_BUFFER_SPACE(n) \ | |
853 | while (b - bufp->buffer + (n) > bufp->allocated) \ | |
854 | EXTEND_BUFFER () | |
855 | ||
856 | /* Make sure we have one more byte of buffer space and then add C to it. */ | |
857 | #define BUF_PUSH(c) \ | |
858 | do { \ | |
859 | GET_BUFFER_SPACE (1); \ | |
860 | *b++ = (unsigned char) (c); \ | |
861 | } while (0) | |
862 | ||
863 | ||
864 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
865 | #define BUF_PUSH_2(c1, c2) \ | |
866 | do { \ | |
867 | GET_BUFFER_SPACE (2); \ | |
868 | *b++ = (unsigned char) (c1); \ | |
869 | *b++ = (unsigned char) (c2); \ | |
870 | } while (0) | |
871 | ||
872 | ||
873 | /* As with BUF_PUSH_2, except for three bytes. */ | |
874 | #define BUF_PUSH_3(c1, c2, c3) \ | |
875 | do { \ | |
876 | GET_BUFFER_SPACE (3); \ | |
877 | *b++ = (unsigned char) (c1); \ | |
878 | *b++ = (unsigned char) (c2); \ | |
879 | *b++ = (unsigned char) (c3); \ | |
880 | } while (0) | |
881 | ||
882 | ||
883 | /* Store a jump with opcode OP at LOC to location TO. We store a | |
884 | relative address offset by the three bytes the jump itself occupies. */ | |
885 | #define STORE_JUMP(op, loc, to) \ | |
886 | store_op1 (op, loc, (to) - (loc) - 3) | |
887 | ||
888 | /* Likewise, for a two-argument jump. */ | |
889 | #define STORE_JUMP2(op, loc, to, arg) \ | |
890 | store_op2 (op, loc, (to) - (loc) - 3, arg) | |
891 | ||
892 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ | |
893 | #define INSERT_JUMP(op, loc, to) \ | |
894 | insert_op1 (op, loc, (to) - (loc) - 3, b) | |
895 | ||
896 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
897 | #define INSERT_JUMP2(op, loc, to, arg) \ | |
898 | insert_op2 (op, loc, (to) - (loc) - 3, arg, b) | |
899 | ||
900 | ||
901 | /* This is not an arbitrary limit: the arguments which represent offsets | |
902 | into the pattern are two bytes long. So if 2^16 bytes turns out to | |
903 | be too small, many things would have to change. */ | |
904 | #define MAX_BUF_SIZE (1L << 16) | |
905 | ||
906 | ||
907 | /* Extend the buffer by twice its current size via realloc and | |
908 | reset the pointers that pointed into the old block to point to the | |
909 | correct places in the new one. If extending the buffer results in it | |
910 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ | |
911 | #define EXTEND_BUFFER() \ | |
912 | do { \ | |
913 | unsigned char *old_buffer = bufp->buffer; \ | |
914 | if (bufp->allocated == MAX_BUF_SIZE) \ | |
915 | return REG_ESIZE; \ | |
916 | bufp->allocated <<= 1; \ | |
917 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
918 | bufp->allocated = MAX_BUF_SIZE; \ | |
919 | bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\ | |
920 | if (bufp->buffer == NULL) \ | |
921 | return REG_ESPACE; \ | |
922 | /* If the buffer moved, move all the pointers into it. */ \ | |
923 | if (old_buffer != bufp->buffer) \ | |
924 | { \ | |
925 | b = (b - old_buffer) + bufp->buffer; \ | |
926 | begalt = (begalt - old_buffer) + bufp->buffer; \ | |
927 | if (fixup_alt_jump) \ | |
928 | fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\ | |
929 | if (laststart) \ | |
930 | laststart = (laststart - old_buffer) + bufp->buffer; \ | |
931 | if (pending_exact) \ | |
932 | pending_exact = (pending_exact - old_buffer) + bufp->buffer; \ | |
933 | } \ | |
934 | } while (0) | |
935 | ||
936 | ||
937 | /* Since we have one byte reserved for the register number argument to | |
938 | {start,stop}_memory, the maximum number of groups we can report | |
939 | things about is what fits in that byte. */ | |
940 | #define MAX_REGNUM 255 | |
941 | ||
942 | /* But patterns can have more than `MAX_REGNUM' registers. We just | |
943 | ignore the excess. */ | |
944 | typedef unsigned regnum_t; | |
945 | ||
946 | ||
947 | /* Macros for the compile stack. */ | |
948 | ||
949 | /* Since offsets can go either forwards or backwards, this type needs to | |
950 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ | |
951 | typedef int pattern_offset_t; | |
952 | ||
953 | typedef struct | |
954 | { | |
955 | pattern_offset_t begalt_offset; | |
956 | pattern_offset_t fixup_alt_jump; | |
957 | pattern_offset_t inner_group_offset; | |
958 | pattern_offset_t laststart_offset; | |
959 | regnum_t regnum; | |
960 | } compile_stack_elt_t; | |
961 | ||
962 | ||
963 | typedef struct | |
964 | { | |
965 | compile_stack_elt_t *stack; | |
966 | unsigned size; | |
967 | unsigned avail; /* Offset of next open position. */ | |
968 | } compile_stack_type; | |
969 | ||
970 | ||
971 | #define INIT_COMPILE_STACK_SIZE 32 | |
972 | ||
973 | #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
974 | #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
975 | ||
976 | /* The next available element. */ | |
977 | #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) | |
978 | ||
979 | ||
980 | /* Set the bit for character C in a list. */ | |
981 | #define SET_LIST_BIT(c) \ | |
982 | (b[((unsigned char) (c)) / BYTEWIDTH] \ | |
983 | |= 1 << (((unsigned char) c) % BYTEWIDTH)) | |
984 | ||
985 | ||
986 | /* Get the next unsigned number in the uncompiled pattern. */ | |
987 | #define GET_UNSIGNED_NUMBER(num) \ | |
988 | { if (p != pend) \ | |
989 | { \ | |
990 | PATFETCH (c); \ | |
991 | while (isdigit (c)) \ | |
992 | { \ | |
993 | if (num < 0) \ | |
994 | num = 0; \ | |
995 | num = num * 10 + c - '0'; \ | |
996 | if (p == pend) \ | |
997 | break; \ | |
998 | PATFETCH (c); \ | |
999 | } \ | |
1000 | } \ | |
1001 | } | |
1002 | ||
1003 | #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ | |
1004 | ||
1005 | #define IS_CHAR_CLASS(string) \ | |
1006 | (STREQ (string, "alpha") || STREQ (string, "upper") \ | |
1007 | || STREQ (string, "lower") || STREQ (string, "digit") \ | |
1008 | || STREQ (string, "alnum") || STREQ (string, "xdigit") \ | |
1009 | || STREQ (string, "space") || STREQ (string, "print") \ | |
1010 | || STREQ (string, "punct") || STREQ (string, "graph") \ | |
1011 | || STREQ (string, "cntrl") || STREQ (string, "blank")) | |
1012 | \f | |
1013 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. | |
1014 | Returns one of error codes defined in `regex.h', or zero for success. | |
1015 | ||
1016 | Assumes the `allocated' (and perhaps `buffer') and `translate' | |
1017 | fields are set in BUFP on entry. | |
1018 | ||
1019 | If it succeeds, results are put in BUFP (if it returns an error, the | |
1020 | contents of BUFP are undefined): | |
1021 | `buffer' is the compiled pattern; | |
1022 | `syntax' is set to SYNTAX; | |
1023 | `used' is set to the length of the compiled pattern; | |
1024 | `fastmap_accurate' is set to zero; | |
1025 | `re_nsub' is set to the number of groups in PATTERN; | |
1026 | `not_bol' and `not_eol' are set to zero. | |
1027 | ||
1028 | The `fastmap' and `newline_anchor' fields are neither | |
1029 | examined nor set. */ | |
1030 | ||
1031 | static reg_errcode_t | |
1032 | regex_compile (pattern, size, syntax, bufp) | |
1033 | const char *pattern; | |
1034 | int size; | |
1035 | reg_syntax_t syntax; | |
1036 | struct re_pattern_buffer *bufp; | |
1037 | { | |
1038 | /* We fetch characters from PATTERN here. Even though PATTERN is | |
1039 | `char *' (i.e., signed), we declare these variables as unsigned, so | |
1040 | they can be reliably used as array indices. */ | |
1041 | register unsigned char c, c1; | |
1042 | ||
1043 | /* A random tempory spot in PATTERN. */ | |
1044 | const char *p1; | |
1045 | ||
1046 | /* Points to the end of the buffer, where we should append. */ | |
1047 | register unsigned char *b; | |
1048 | ||
1049 | /* Keeps track of unclosed groups. */ | |
1050 | compile_stack_type compile_stack; | |
1051 | ||
1052 | /* Points to the current (ending) position in the pattern. */ | |
1053 | const char *p = pattern; | |
1054 | const char *pend = pattern + size; | |
1055 | ||
1056 | /* How to translate the characters in the pattern. */ | |
1057 | char *translate = bufp->translate; | |
1058 | ||
1059 | /* Address of the count-byte of the most recently inserted `exactn' | |
1060 | command. This makes it possible to tell if a new exact-match | |
1061 | character can be added to that command or if the character requires | |
1062 | a new `exactn' command. */ | |
1063 | unsigned char *pending_exact = 0; | |
1064 | ||
1065 | /* Address of start of the most recently finished expression. | |
1066 | This tells, e.g., postfix * where to find the start of its | |
1067 | operand. Reset at the beginning of groups and alternatives. */ | |
1068 | unsigned char *laststart = 0; | |
1069 | ||
1070 | /* Address of beginning of regexp, or inside of last group. */ | |
1071 | unsigned char *begalt; | |
1072 | ||
1073 | /* Place in the uncompiled pattern (i.e., the {) to | |
1074 | which to go back if the interval is invalid. */ | |
1075 | const char *beg_interval; | |
1076 | ||
1077 | /* Address of the place where a forward jump should go to the end of | |
1078 | the containing expression. Each alternative of an `or' -- except the | |
1079 | last -- ends with a forward jump of this sort. */ | |
1080 | unsigned char *fixup_alt_jump = 0; | |
1081 | ||
1082 | /* Counts open-groups as they are encountered. Remembered for the | |
1083 | matching close-group on the compile stack, so the same register | |
1084 | number is put in the stop_memory as the start_memory. */ | |
1085 | regnum_t regnum = 0; | |
1086 | ||
1087 | #ifdef DEBUG | |
1088 | DEBUG_PRINT1 ("\nCompiling pattern: "); | |
1089 | if (debug) | |
1090 | { | |
1091 | unsigned debug_count; | |
1092 | ||
1093 | for (debug_count = 0; debug_count < size; debug_count++) | |
1094 | printchar (pattern[debug_count]); | |
1095 | putchar ('\n'); | |
1096 | } | |
1097 | #endif /* DEBUG */ | |
1098 | ||
1099 | /* Initialize the compile stack. */ | |
1100 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
1101 | if (compile_stack.stack == NULL) | |
1102 | return REG_ESPACE; | |
1103 | ||
1104 | compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
1105 | compile_stack.avail = 0; | |
1106 | ||
1107 | /* Initialize the pattern buffer. */ | |
1108 | bufp->syntax = syntax; | |
1109 | bufp->fastmap_accurate = 0; | |
1110 | bufp->not_bol = bufp->not_eol = 0; | |
1111 | ||
1112 | /* Set `used' to zero, so that if we return an error, the pattern | |
1113 | printer (for debugging) will think there's no pattern. We reset it | |
1114 | at the end. */ | |
1115 | bufp->used = 0; | |
1116 | ||
1117 | /* Always count groups, whether or not bufp->no_sub is set. */ | |
1118 | bufp->re_nsub = 0; | |
1119 | ||
1120 | #if !defined (emacs) && !defined (SYNTAX_TABLE) | |
1121 | /* Initialize the syntax table. */ | |
1122 | init_syntax_once (); | |
1123 | #endif | |
1124 | ||
1125 | if (bufp->allocated == 0) | |
1126 | { | |
1127 | if (bufp->buffer) | |
1128 | { /* If zero allocated, but buffer is non-null, try to realloc | |
1129 | enough space. This loses if buffer's address is bogus, but | |
1130 | that is the user's responsibility. */ | |
1131 | RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); | |
1132 | } | |
1133 | else | |
1134 | { /* Caller did not allocate a buffer. Do it for them. */ | |
1135 | bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); | |
1136 | } | |
1137 | if (!bufp->buffer) return REG_ESPACE; | |
1138 | ||
1139 | bufp->allocated = INIT_BUF_SIZE; | |
1140 | } | |
1141 | ||
1142 | begalt = b = bufp->buffer; | |
1143 | ||
1144 | /* Loop through the uncompiled pattern until we're at the end. */ | |
1145 | while (p != pend) | |
1146 | { | |
1147 | PATFETCH (c); | |
1148 | ||
1149 | switch (c) | |
1150 | { | |
1151 | case '^': | |
1152 | { | |
1153 | if ( /* If at start of pattern, it's an operator. */ | |
1154 | p == pattern + 1 | |
1155 | /* If context independent, it's an operator. */ | |
1156 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
1157 | /* Otherwise, depends on what's come before. */ | |
1158 | || at_begline_loc_p (pattern, p, syntax)) | |
1159 | BUF_PUSH (begline); | |
1160 | else | |
1161 | goto normal_char; | |
1162 | } | |
1163 | break; | |
1164 | ||
1165 | ||
1166 | case '$': | |
1167 | { | |
1168 | if ( /* If at end of pattern, it's an operator. */ | |
1169 | p == pend | |
1170 | /* If context independent, it's an operator. */ | |
1171 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
1172 | /* Otherwise, depends on what's next. */ | |
1173 | || at_endline_loc_p (p, pend, syntax)) | |
1174 | BUF_PUSH (endline); | |
1175 | else | |
1176 | goto normal_char; | |
1177 | } | |
1178 | break; | |
1179 | ||
1180 | ||
1181 | case '+': | |
1182 | case '?': | |
1183 | if ((syntax & RE_BK_PLUS_QM) | |
1184 | || (syntax & RE_LIMITED_OPS)) | |
1185 | goto normal_char; | |
1186 | handle_plus: | |
1187 | case '*': | |
1188 | /* If there is no previous pattern... */ | |
1189 | if (!laststart) | |
1190 | { | |
1191 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
1192 | return REG_BADRPT; | |
1193 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
1194 | goto normal_char; | |
1195 | } | |
1196 | ||
1197 | { | |
1198 | /* Are we optimizing this jump? */ | |
1199 | boolean keep_string_p = false; | |
1200 | ||
1201 | /* 1 means zero (many) matches is allowed. */ | |
1202 | char zero_times_ok = 0, many_times_ok = 0; | |
1203 | ||
1204 | /* If there is a sequence of repetition chars, collapse it | |
1205 | down to just one (the right one). We can't combine | |
1206 | interval operators with these because of, e.g., `a{2}*', | |
1207 | which should only match an even number of `a's. */ | |
1208 | ||
1209 | for (;;) | |
1210 | { | |
1211 | zero_times_ok |= c != '+'; | |
1212 | many_times_ok |= c != '?'; | |
1213 | ||
1214 | if (p == pend) | |
1215 | break; | |
1216 | ||
1217 | PATFETCH (c); | |
1218 | ||
1219 | if (c == '*' | |
1220 | || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) | |
1221 | ; | |
1222 | ||
1223 | else if (syntax & RE_BK_PLUS_QM && c == '\\') | |
1224 | { | |
1225 | if (p == pend) return REG_EESCAPE; | |
1226 | ||
1227 | PATFETCH (c1); | |
1228 | if (!(c1 == '+' || c1 == '?')) | |
1229 | { | |
1230 | PATUNFETCH; | |
1231 | PATUNFETCH; | |
1232 | break; | |
1233 | } | |
1234 | ||
1235 | c = c1; | |
1236 | } | |
1237 | else | |
1238 | { | |
1239 | PATUNFETCH; | |
1240 | break; | |
1241 | } | |
1242 | ||
1243 | /* If we get here, we found another repeat character. */ | |
1244 | } | |
1245 | ||
1246 | /* Star, etc. applied to an empty pattern is equivalent | |
1247 | to an empty pattern. */ | |
1248 | if (!laststart) | |
1249 | break; | |
1250 | ||
1251 | /* Now we know whether or not zero matches is allowed | |
1252 | and also whether or not two or more matches is allowed. */ | |
1253 | if (many_times_ok) | |
1254 | { /* More than one repetition is allowed, so put in at the | |
1255 | end a backward relative jump from `b' to before the next | |
1256 | jump we're going to put in below (which jumps from | |
1257 | laststart to after this jump). | |
1258 | ||
1259 | But if we are at the `*' in the exact sequence `.*\n', | |
1260 | insert an unconditional jump backwards to the ., | |
1261 | instead of the beginning of the loop. This way we only | |
1262 | push a failure point once, instead of every time | |
1263 | through the loop. */ | |
1264 | assert (p - 1 > pattern); | |
1265 | ||
1266 | /* Allocate the space for the jump. */ | |
1267 | GET_BUFFER_SPACE (3); | |
1268 | ||
1269 | /* We know we are not at the first character of the pattern, | |
1270 | because laststart was nonzero. And we've already | |
1271 | incremented `p', by the way, to be the character after | |
1272 | the `*'. Do we have to do something analogous here | |
1273 | for null bytes, because of RE_DOT_NOT_NULL? */ | |
1274 | if (TRANSLATE (*(p - 2)) == TRANSLATE ('.') | |
1275 | && p < pend && TRANSLATE (*p) == TRANSLATE ('\n') | |
1276 | && !(syntax & RE_DOT_NEWLINE)) | |
1277 | { /* We have .*\n. */ | |
1278 | STORE_JUMP (jump, b, laststart); | |
1279 | keep_string_p = true; | |
1280 | } | |
1281 | else | |
1282 | /* Anything else. */ | |
1283 | STORE_JUMP (maybe_pop_jump, b, laststart - 3); | |
1284 | ||
1285 | /* We've added more stuff to the buffer. */ | |
1286 | b += 3; | |
1287 | } | |
1288 | ||
1289 | /* On failure, jump from laststart to b + 3, which will be the | |
1290 | end of the buffer after this jump is inserted. */ | |
1291 | GET_BUFFER_SPACE (3); | |
1292 | INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump | |
1293 | : on_failure_jump, | |
1294 | laststart, b + 3); | |
1295 | pending_exact = 0; | |
1296 | b += 3; | |
1297 | ||
1298 | if (!zero_times_ok) | |
1299 | { | |
1300 | /* At least one repetition is required, so insert a | |
1301 | `dummy_failure_jump' before the initial | |
1302 | `on_failure_jump' instruction of the loop. This | |
1303 | effects a skip over that instruction the first time | |
1304 | we hit that loop. */ | |
1305 | GET_BUFFER_SPACE (3); | |
1306 | INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6); | |
1307 | b += 3; | |
1308 | } | |
1309 | } | |
1310 | break; | |
1311 | ||
1312 | ||
1313 | case '.': | |
1314 | laststart = b; | |
1315 | BUF_PUSH (anychar); | |
1316 | break; | |
1317 | ||
1318 | ||
1319 | case '[': | |
1320 | { | |
1321 | boolean had_char_class = false; | |
1322 | ||
1323 | if (p == pend) return REG_EBRACK; | |
1324 | ||
1325 | /* Ensure that we have enough space to push a charset: the | |
1326 | opcode, the length count, and the bitset; 34 bytes in all. */ | |
1327 | GET_BUFFER_SPACE (34); | |
1328 | ||
1329 | laststart = b; | |
1330 | ||
1331 | /* We test `*p == '^' twice, instead of using an if | |
1332 | statement, so we only need one BUF_PUSH. */ | |
1333 | BUF_PUSH (*p == '^' ? charset_not : charset); | |
1334 | if (*p == '^') | |
1335 | p++; | |
1336 | ||
1337 | /* Remember the first position in the bracket expression. */ | |
1338 | p1 = p; | |
1339 | ||
1340 | /* Push the number of bytes in the bitmap. */ | |
1341 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
1342 | ||
1343 | /* Clear the whole map. */ | |
1344 | bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); | |
1345 | ||
1346 | /* charset_not matches newline according to a syntax bit. */ | |
1347 | if ((re_opcode_t) b[-2] == charset_not | |
1348 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
1349 | SET_LIST_BIT ('\n'); | |
1350 | ||
1351 | /* Read in characters and ranges, setting map bits. */ | |
1352 | for (;;) | |
1353 | { | |
1354 | if (p == pend) return REG_EBRACK; | |
1355 | ||
1356 | PATFETCH (c); | |
1357 | ||
1358 | /* \ might escape characters inside [...] and [^...]. */ | |
1359 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
1360 | { | |
1361 | if (p == pend) return REG_EESCAPE; | |
1362 | ||
1363 | PATFETCH (c1); | |
1364 | SET_LIST_BIT (c1); | |
1365 | continue; | |
1366 | } | |
1367 | ||
1368 | /* Could be the end of the bracket expression. If it's | |
1369 | not (i.e., when the bracket expression is `[]' so | |
1370 | far), the ']' character bit gets set way below. */ | |
1371 | if (c == ']' && p != p1 + 1) | |
1372 | break; | |
1373 | ||
1374 | /* Look ahead to see if it's a range when the last thing | |
1375 | was a character class. */ | |
1376 | if (had_char_class && c == '-' && *p != ']') | |
1377 | return REG_ERANGE; | |
1378 | ||
1379 | /* Look ahead to see if it's a range when the last thing | |
1380 | was a character: if this is a hyphen not at the | |
1381 | beginning or the end of a list, then it's the range | |
1382 | operator. */ | |
1383 | if (c == '-' | |
1384 | && !(p - 2 >= pattern && p[-2] == '[') | |
1385 | && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') | |
1386 | && *p != ']') | |
1387 | { | |
1388 | reg_errcode_t ret | |
1389 | = compile_range (&p, pend, translate, syntax, b); | |
1390 | if (ret != REG_NOERROR) return ret; | |
1391 | } | |
1392 | ||
1393 | else if (p[0] == '-' && p[1] != ']') | |
1394 | { /* This handles ranges made up of characters only. */ | |
1395 | reg_errcode_t ret; | |
1396 | ||
1397 | /* Move past the `-'. */ | |
1398 | PATFETCH (c1); | |
1399 | ||
1400 | ret = compile_range (&p, pend, translate, syntax, b); | |
1401 | if (ret != REG_NOERROR) return ret; | |
1402 | } | |
1403 | ||
1404 | /* See if we're at the beginning of a possible character | |
1405 | class. */ | |
1406 | ||
1407 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
1408 | { /* Leave room for the null. */ | |
1409 | char str[CHAR_CLASS_MAX_LENGTH + 1]; | |
1410 | ||
1411 | PATFETCH (c); | |
1412 | c1 = 0; | |
1413 | ||
1414 | /* If pattern is `[[:'. */ | |
1415 | if (p == pend) return REG_EBRACK; | |
1416 | ||
1417 | for (;;) | |
1418 | { | |
1419 | PATFETCH (c); | |
1420 | if (c == ':' || c == ']' || p == pend | |
1421 | || c1 == CHAR_CLASS_MAX_LENGTH) | |
1422 | break; | |
1423 | str[c1++] = c; | |
1424 | } | |
1425 | str[c1] = '\0'; | |
1426 | ||
1427 | /* If isn't a word bracketed by `[:' and:`]': | |
1428 | undo the ending character, the letters, and leave | |
1429 | the leading `:' and `[' (but set bits for them). */ | |
1430 | if (c == ':' && *p == ']') | |
1431 | { | |
1432 | int ch; | |
1433 | boolean is_alnum = STREQ (str, "alnum"); | |
1434 | boolean is_alpha = STREQ (str, "alpha"); | |
1435 | boolean is_blank = STREQ (str, "blank"); | |
1436 | boolean is_cntrl = STREQ (str, "cntrl"); | |
1437 | boolean is_digit = STREQ (str, "digit"); | |
1438 | boolean is_graph = STREQ (str, "graph"); | |
1439 | boolean is_lower = STREQ (str, "lower"); | |
1440 | boolean is_print = STREQ (str, "print"); | |
1441 | boolean is_punct = STREQ (str, "punct"); | |
1442 | boolean is_space = STREQ (str, "space"); | |
1443 | boolean is_upper = STREQ (str, "upper"); | |
1444 | boolean is_xdigit = STREQ (str, "xdigit"); | |
1445 | ||
1446 | if (!IS_CHAR_CLASS (str)) return REG_ECTYPE; | |
1447 | ||
1448 | /* Throw away the ] at the end of the character | |
1449 | class. */ | |
1450 | PATFETCH (c); | |
1451 | ||
1452 | if (p == pend) return REG_EBRACK; | |
1453 | ||
1454 | for (ch = 0; ch < 1 << BYTEWIDTH; ch++) | |
1455 | { | |
1456 | if ( (is_alnum && isalnum (ch)) | |
1457 | || (is_alpha && isalpha (ch)) | |
1458 | || (is_blank && isblank (ch)) | |
1459 | || (is_cntrl && iscntrl (ch)) | |
1460 | || (is_digit && isdigit (ch)) | |
1461 | || (is_graph && isgraph (ch)) | |
1462 | || (is_lower && islower (ch)) | |
1463 | || (is_print && isprint (ch)) | |
1464 | || (is_punct && ispunct (ch)) | |
1465 | || (is_space && isspace (ch)) | |
1466 | || (is_upper && isupper (ch)) | |
1467 | || (is_xdigit && isxdigit (ch))) | |
1468 | SET_LIST_BIT (ch); | |
1469 | } | |
1470 | had_char_class = true; | |
1471 | } | |
1472 | else | |
1473 | { | |
1474 | c1++; | |
1475 | while (c1--) | |
1476 | PATUNFETCH; | |
1477 | SET_LIST_BIT ('['); | |
1478 | SET_LIST_BIT (':'); | |
1479 | had_char_class = false; | |
1480 | } | |
1481 | } | |
1482 | else | |
1483 | { | |
1484 | had_char_class = false; | |
1485 | SET_LIST_BIT (c); | |
1486 | } | |
1487 | } | |
1488 | ||
1489 | /* Discard any (non)matching list bytes that are all 0 at the | |
1490 | end of the map. Decrease the map-length byte too. */ | |
1491 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) | |
1492 | b[-1]--; | |
1493 | b += b[-1]; | |
1494 | } | |
1495 | break; | |
1496 | ||
1497 | ||
1498 | case '(': | |
1499 | if (syntax & RE_NO_BK_PARENS) | |
1500 | goto handle_open; | |
1501 | else | |
1502 | goto normal_char; | |
1503 | ||
1504 | ||
1505 | case ')': | |
1506 | if (syntax & RE_NO_BK_PARENS) | |
1507 | goto handle_close; | |
1508 | else | |
1509 | goto normal_char; | |
1510 | ||
1511 | ||
1512 | case '\n': | |
1513 | if (syntax & RE_NEWLINE_ALT) | |
1514 | goto handle_alt; | |
1515 | else | |
1516 | goto normal_char; | |
1517 | ||
1518 | ||
1519 | case '|': | |
1520 | if (syntax & RE_NO_BK_VBAR) | |
1521 | goto handle_alt; | |
1522 | else | |
1523 | goto normal_char; | |
1524 | ||
1525 | ||
1526 | case '{': | |
1527 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
1528 | goto handle_interval; | |
1529 | else | |
1530 | goto normal_char; | |
1531 | ||
1532 | ||
1533 | case '\\': | |
1534 | if (p == pend) return REG_EESCAPE; | |
1535 | ||
1536 | /* Do not translate the character after the \, so that we can | |
1537 | distinguish, e.g., \B from \b, even if we normally would | |
1538 | translate, e.g., B to b. */ | |
1539 | PATFETCH_RAW (c); | |
1540 | ||
1541 | switch (c) | |
1542 | { | |
1543 | case '(': | |
1544 | if (syntax & RE_NO_BK_PARENS) | |
1545 | goto normal_backslash; | |
1546 | ||
1547 | handle_open: | |
1548 | bufp->re_nsub++; | |
1549 | regnum++; | |
1550 | ||
1551 | if (COMPILE_STACK_FULL) | |
1552 | { | |
1553 | RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
1554 | compile_stack_elt_t); | |
1555 | if (compile_stack.stack == NULL) return REG_ESPACE; | |
1556 | ||
1557 | compile_stack.size <<= 1; | |
1558 | } | |
1559 | ||
1560 | /* These are the values to restore when we hit end of this | |
1561 | group. They are all relative offsets, so that if the | |
1562 | whole pattern moves because of realloc, they will still | |
1563 | be valid. */ | |
1564 | COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; | |
1565 | COMPILE_STACK_TOP.fixup_alt_jump | |
1566 | = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; | |
1567 | COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; | |
1568 | COMPILE_STACK_TOP.regnum = regnum; | |
1569 | ||
1570 | /* We will eventually replace the 0 with the number of | |
1571 | groups inner to this one. But do not push a | |
1572 | start_memory for groups beyond the last one we can | |
1573 | represent in the compiled pattern. */ | |
1574 | if (regnum <= MAX_REGNUM) | |
1575 | { | |
1576 | COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2; | |
1577 | BUF_PUSH_3 (start_memory, regnum, 0); | |
1578 | } | |
1579 | ||
1580 | compile_stack.avail++; | |
1581 | ||
1582 | fixup_alt_jump = 0; | |
1583 | laststart = 0; | |
1584 | begalt = b; | |
1585 | break; | |
1586 | ||
1587 | ||
1588 | case ')': | |
1589 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
1590 | ||
1591 | if (COMPILE_STACK_EMPTY) | |
1592 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
1593 | goto normal_backslash; | |
1594 | else | |
1595 | return REG_ERPAREN; | |
1596 | ||
1597 | handle_close: | |
1598 | if (fixup_alt_jump) | |
1599 | { /* Push a dummy failure point at the end of the | |
1600 | alternative for a possible future | |
1601 | `pop_failure_jump' to pop. See comments at | |
1602 | `push_dummy_failure' in `re_match_2'. */ | |
1603 | BUF_PUSH (push_dummy_failure); | |
1604 | ||
1605 | /* We allocated space for this jump when we assigned | |
1606 | to `fixup_alt_jump', in the `handle_alt' case below. */ | |
1607 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); | |
1608 | } | |
1609 | ||
1610 | /* See similar code for backslashed left paren above. */ | |
1611 | if (COMPILE_STACK_EMPTY) | |
1612 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
1613 | goto normal_char; | |
1614 | else | |
1615 | return REG_ERPAREN; | |
1616 | ||
1617 | /* Since we just checked for an empty stack above, this | |
1618 | ``can't happen''. */ | |
1619 | assert (compile_stack.avail != 0); | |
1620 | { | |
1621 | /* We don't just want to restore into `regnum', because | |
1622 | later groups should continue to be numbered higher, | |
1623 | as in `(ab)c(de)' -- the second group is #2. */ | |
1624 | regnum_t this_group_regnum; | |
1625 | ||
1626 | compile_stack.avail--; | |
1627 | begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; | |
1628 | fixup_alt_jump | |
1629 | = COMPILE_STACK_TOP.fixup_alt_jump | |
1630 | ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
1631 | : 0; | |
1632 | laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; | |
1633 | this_group_regnum = COMPILE_STACK_TOP.regnum; | |
1634 | ||
1635 | /* We're at the end of the group, so now we know how many | |
1636 | groups were inside this one. */ | |
1637 | if (this_group_regnum <= MAX_REGNUM) | |
1638 | { | |
1639 | unsigned char *inner_group_loc | |
1640 | = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset; | |
1641 | ||
1642 | *inner_group_loc = regnum - this_group_regnum; | |
1643 | BUF_PUSH_3 (stop_memory, this_group_regnum, | |
1644 | regnum - this_group_regnum); | |
1645 | } | |
1646 | } | |
1647 | break; | |
1648 | ||
1649 | ||
1650 | case '|': /* `\|'. */ | |
1651 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
1652 | goto normal_backslash; | |
1653 | handle_alt: | |
1654 | if (syntax & RE_LIMITED_OPS) | |
1655 | goto normal_char; | |
1656 | ||
1657 | /* Insert before the previous alternative a jump which | |
1658 | jumps to this alternative if the former fails. */ | |
1659 | GET_BUFFER_SPACE (3); | |
1660 | INSERT_JUMP (on_failure_jump, begalt, b + 6); | |
1661 | pending_exact = 0; | |
1662 | b += 3; | |
1663 | ||
1664 | /* The alternative before this one has a jump after it | |
1665 | which gets executed if it gets matched. Adjust that | |
1666 | jump so it will jump to this alternative's analogous | |
1667 | jump (put in below, which in turn will jump to the next | |
1668 | (if any) alternative's such jump, etc.). The last such | |
1669 | jump jumps to the correct final destination. A picture: | |
1670 | _____ _____ | |
1671 | | | | | | |
1672 | | v | v | |
1673 | a | b | c | |
1674 | ||
1675 | If we are at `b,' then fixup_alt_jump right now points to a | |
1676 | three-byte space after `a.' We'll put in the jump, set | |
1677 | fixup_alt_jump to right after `b,' and leave behind three | |
1678 | bytes which we'll fill in when we get to after `c.' */ | |
1679 | ||
1680 | if (fixup_alt_jump) | |
1681 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
1682 | ||
1683 | /* Mark and leave space for a jump after this alternative, | |
1684 | to be filled in later either by next alternative or | |
1685 | when know we're at the end of a series of alternatives. */ | |
1686 | fixup_alt_jump = b; | |
1687 | GET_BUFFER_SPACE (3); | |
1688 | b += 3; | |
1689 | ||
1690 | laststart = 0; | |
1691 | begalt = b; | |
1692 | break; | |
1693 | ||
1694 | ||
1695 | case '{': | |
1696 | /* If \{ is a literal. */ | |
1697 | if (!(syntax & RE_INTERVALS) | |
1698 | /* If we're at `\{' and it's not the open-interval | |
1699 | operator. */ | |
1700 | || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES)) | |
1701 | || (p - 2 == pattern && p == pend)) | |
1702 | goto normal_backslash; | |
1703 | ||
1704 | handle_interval: | |
1705 | { | |
1706 | /* If got here, then the syntax allows intervals. */ | |
1707 | ||
1708 | /* At least (most) this many matches must be made. */ | |
1709 | int lower_bound = -1, upper_bound = -1; | |
1710 | ||
1711 | beg_interval = p - 1; | |
1712 | ||
1713 | if (p == pend) | |
1714 | { | |
1715 | if (syntax & RE_NO_BK_BRACES) | |
1716 | goto unfetch_interval; | |
1717 | else | |
1718 | return REG_EBRACE; | |
1719 | } | |
1720 | ||
1721 | GET_UNSIGNED_NUMBER (lower_bound); | |
1722 | ||
1723 | if (c == ',') | |
1724 | { | |
1725 | GET_UNSIGNED_NUMBER (upper_bound); | |
1726 | if (upper_bound < 0) upper_bound = RE_DUP_MAX; | |
1727 | } | |
1728 | else | |
1729 | /* Interval such as `{1}' => match exactly once. */ | |
1730 | upper_bound = lower_bound; | |
1731 | ||
1732 | if (lower_bound < 0 || upper_bound > RE_DUP_MAX | |
1733 | || lower_bound > upper_bound) | |
1734 | { | |
1735 | if (syntax & RE_NO_BK_BRACES) | |
1736 | goto unfetch_interval; | |
1737 | else | |
1738 | return REG_BADBR; | |
1739 | } | |
1740 | ||
1741 | if (!(syntax & RE_NO_BK_BRACES)) | |
1742 | { | |
1743 | if (c != '\\') return REG_EBRACE; | |
1744 | ||
1745 | PATFETCH (c); | |
1746 | } | |
1747 | ||
1748 | if (c != '}') | |
1749 | { | |
1750 | if (syntax & RE_NO_BK_BRACES) | |
1751 | goto unfetch_interval; | |
1752 | else | |
1753 | return REG_BADBR; | |
1754 | } | |
1755 | ||
1756 | /* We just parsed a valid interval. */ | |
1757 | ||
1758 | /* If it's invalid to have no preceding re. */ | |
1759 | if (!laststart) | |
1760 | { | |
1761 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
1762 | return REG_BADRPT; | |
1763 | else if (syntax & RE_CONTEXT_INDEP_OPS) | |
1764 | laststart = b; | |
1765 | else | |
1766 | goto unfetch_interval; | |
1767 | } | |
1768 | ||
1769 | /* If the upper bound is zero, don't want to succeed at | |
1770 | all; jump from `laststart' to `b + 3', which will be | |
1771 | the end of the buffer after we insert the jump. */ | |
1772 | if (upper_bound == 0) | |
1773 | { | |
1774 | GET_BUFFER_SPACE (3); | |
1775 | INSERT_JUMP (jump, laststart, b + 3); | |
1776 | b += 3; | |
1777 | } | |
1778 | ||
1779 | /* Otherwise, we have a nontrivial interval. When | |
1780 | we're all done, the pattern will look like: | |
1781 | set_number_at <jump count> <upper bound> | |
1782 | set_number_at <succeed_n count> <lower bound> | |
1783 | succeed_n <after jump addr> <succed_n count> | |
1784 | <body of loop> | |
1785 | jump_n <succeed_n addr> <jump count> | |
1786 | (The upper bound and `jump_n' are omitted if | |
1787 | `upper_bound' is 1, though.) */ | |
1788 | else | |
1789 | { /* If the upper bound is > 1, we need to insert | |
1790 | more at the end of the loop. */ | |
1791 | unsigned nbytes = 10 + (upper_bound > 1) * 10; | |
1792 | ||
1793 | GET_BUFFER_SPACE (nbytes); | |
1794 | ||
1795 | /* Initialize lower bound of the `succeed_n', even | |
1796 | though it will be set during matching by its | |
1797 | attendant `set_number_at' (inserted next), | |
1798 | because `re_compile_fastmap' needs to know. | |
1799 | Jump to the `jump_n' we might insert below. */ | |
1800 | INSERT_JUMP2 (succeed_n, laststart, | |
1801 | b + 5 + (upper_bound > 1) * 5, | |
1802 | lower_bound); | |
1803 | b += 5; | |
1804 | ||
1805 | /* Code to initialize the lower bound. Insert | |
1806 | before the `succeed_n'. The `5' is the last two | |
1807 | bytes of this `set_number_at', plus 3 bytes of | |
1808 | the following `succeed_n'. */ | |
1809 | insert_op2 (set_number_at, laststart, 5, lower_bound, b); | |
1810 | b += 5; | |
1811 | ||
1812 | if (upper_bound > 1) | |
1813 | { /* More than one repetition is allowed, so | |
1814 | append a backward jump to the `succeed_n' | |
1815 | that starts this interval. | |
1816 | ||
1817 | When we've reached this during matching, | |
1818 | we'll have matched the interval once, so | |
1819 | jump back only `upper_bound - 1' times. */ | |
1820 | STORE_JUMP2 (jump_n, b, laststart + 5, | |
1821 | upper_bound - 1); | |
1822 | b += 5; | |
1823 | ||
1824 | /* The location we want to set is the second | |
1825 | parameter of the `jump_n'; that is `b-2' as | |
1826 | an absolute address. `laststart' will be | |
1827 | the `set_number_at' we're about to insert; | |
1828 | `laststart+3' the number to set, the source | |
1829 | for the relative address. But we are | |
1830 | inserting into the middle of the pattern -- | |
1831 | so everything is getting moved up by 5. | |
1832 | Conclusion: (b - 2) - (laststart + 3) + 5, | |
1833 | i.e., b - laststart. | |
1834 | ||
1835 | We insert this at the beginning of the loop | |
1836 | so that if we fail during matching, we'll | |
1837 | reinitialize the bounds. */ | |
1838 | insert_op2 (set_number_at, laststart, b - laststart, | |
1839 | upper_bound - 1, b); | |
1840 | b += 5; | |
1841 | } | |
1842 | } | |
1843 | pending_exact = 0; | |
1844 | beg_interval = NULL; | |
1845 | } | |
1846 | break; | |
1847 | ||
1848 | unfetch_interval: | |
1849 | /* If an invalid interval, match the characters as literals. */ | |
1850 | assert (beg_interval); | |
1851 | p = beg_interval; | |
1852 | beg_interval = NULL; | |
1853 | ||
1854 | /* normal_char and normal_backslash need `c'. */ | |
1855 | PATFETCH (c); | |
1856 | ||
1857 | if (!(syntax & RE_NO_BK_BRACES)) | |
1858 | { | |
1859 | if (p > pattern && p[-1] == '\\') | |
1860 | goto normal_backslash; | |
1861 | } | |
1862 | goto normal_char; | |
1863 | ||
1864 | #ifdef emacs | |
1865 | /* There is no way to specify the before_dot and after_dot | |
1866 | operators. rms says this is ok. --karl */ | |
1867 | case '=': | |
1868 | BUF_PUSH (at_dot); | |
1869 | break; | |
1870 | ||
1871 | case 's': | |
1872 | laststart = b; | |
1873 | PATFETCH (c); | |
1874 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
1875 | break; | |
1876 | ||
1877 | case 'S': | |
1878 | laststart = b; | |
1879 | PATFETCH (c); | |
1880 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
1881 | break; | |
1882 | #endif /* emacs */ | |
1883 | ||
1884 | ||
1885 | case 'w': | |
1886 | laststart = b; | |
1887 | BUF_PUSH (wordchar); | |
1888 | break; | |
1889 | ||
1890 | ||
1891 | case 'W': | |
1892 | laststart = b; | |
1893 | BUF_PUSH (notwordchar); | |
1894 | break; | |
1895 | ||
1896 | ||
1897 | case '<': | |
1898 | BUF_PUSH (wordbeg); | |
1899 | break; | |
1900 | ||
1901 | case '>': | |
1902 | BUF_PUSH (wordend); | |
1903 | break; | |
1904 | ||
1905 | case 'b': | |
1906 | BUF_PUSH (wordbound); | |
1907 | break; | |
1908 | ||
1909 | case 'B': | |
1910 | BUF_PUSH (notwordbound); | |
1911 | break; | |
1912 | ||
1913 | case '`': | |
1914 | BUF_PUSH (begbuf); | |
1915 | break; | |
1916 | ||
1917 | case '\'': | |
1918 | BUF_PUSH (endbuf); | |
1919 | break; | |
1920 | ||
1921 | case '1': case '2': case '3': case '4': case '5': | |
1922 | case '6': case '7': case '8': case '9': | |
1923 | if (syntax & RE_NO_BK_REFS) | |
1924 | goto normal_char; | |
1925 | ||
1926 | c1 = c - '0'; | |
1927 | ||
1928 | if (c1 > regnum) | |
1929 | return REG_ESUBREG; | |
1930 | ||
1931 | /* Can't back reference to a subexpression if inside of it. */ | |
1932 | if (group_in_compile_stack (compile_stack, c1)) | |
1933 | goto normal_char; | |
1934 | ||
1935 | laststart = b; | |
1936 | BUF_PUSH_2 (duplicate, c1); | |
1937 | break; | |
1938 | ||
1939 | ||
1940 | case '+': | |
1941 | case '?': | |
1942 | if (syntax & RE_BK_PLUS_QM) | |
1943 | goto handle_plus; | |
1944 | else | |
1945 | goto normal_backslash; | |
1946 | ||
1947 | default: | |
1948 | normal_backslash: | |
1949 | /* You might think it would be useful for \ to mean | |
1950 | not to translate; but if we don't translate it | |
1951 | it will never match anything. */ | |
1952 | c = TRANSLATE (c); | |
1953 | goto normal_char; | |
1954 | } | |
1955 | break; | |
1956 | ||
1957 | ||
1958 | default: | |
1959 | /* Expects the character in `c'. */ | |
1960 | normal_char: | |
1961 | /* If no exactn currently being built. */ | |
1962 | if (!pending_exact | |
1963 | ||
1964 | /* If last exactn not at current position. */ | |
1965 | || pending_exact + *pending_exact + 1 != b | |
1966 | ||
1967 | /* We have only one byte following the exactn for the count. */ | |
1968 | || *pending_exact == (1 << BYTEWIDTH) - 1 | |
1969 | ||
1970 | /* If followed by a repetition operator. */ | |
1971 | || *p == '*' || *p == '^' | |
1972 | || ((syntax & RE_BK_PLUS_QM) | |
1973 | ? *p == '\\' && (p[1] == '+' || p[1] == '?') | |
1974 | : (*p == '+' || *p == '?')) | |
1975 | || ((syntax & RE_INTERVALS) | |
1976 | && ((syntax & RE_NO_BK_BRACES) | |
1977 | ? *p == '{' | |
1978 | : (p[0] == '\\' && p[1] == '{')))) | |
1979 | { | |
1980 | /* Start building a new exactn. */ | |
1981 | ||
1982 | laststart = b; | |
1983 | ||
1984 | BUF_PUSH_2 (exactn, 0); | |
1985 | pending_exact = b - 1; | |
1986 | } | |
1987 | ||
1988 | BUF_PUSH (c); | |
1989 | (*pending_exact)++; | |
1990 | break; | |
1991 | } /* switch (c) */ | |
1992 | } /* while p != pend */ | |
1993 | ||
1994 | ||
1995 | /* Through the pattern now. */ | |
1996 | ||
1997 | if (fixup_alt_jump) | |
1998 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
1999 | ||
2000 | if (!COMPILE_STACK_EMPTY) | |
2001 | return REG_EPAREN; | |
2002 | ||
2003 | free (compile_stack.stack); | |
2004 | ||
2005 | /* We have succeeded; set the length of the buffer. */ | |
2006 | bufp->used = b - bufp->buffer; | |
2007 | ||
2008 | #ifdef DEBUG | |
2009 | if (debug) | |
2010 | { | |
2011 | DEBUG_PRINT1 ("\nCompiled pattern: "); | |
2012 | print_compiled_pattern (bufp); | |
2013 | } | |
2014 | #endif /* DEBUG */ | |
2015 | ||
2016 | return REG_NOERROR; | |
2017 | } /* regex_compile */ | |
2018 | \f | |
2019 | /* Subroutines for `regex_compile'. */ | |
2020 | ||
2021 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ | |
2022 | ||
2023 | static void | |
2024 | store_op1 (op, loc, arg) | |
2025 | re_opcode_t op; | |
2026 | unsigned char *loc; | |
2027 | int arg; | |
2028 | { | |
2029 | *loc = (unsigned char) op; | |
2030 | STORE_NUMBER (loc + 1, arg); | |
2031 | } | |
2032 | ||
2033 | ||
2034 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
2035 | ||
2036 | static void | |
2037 | store_op2 (op, loc, arg1, arg2) | |
2038 | re_opcode_t op; | |
2039 | unsigned char *loc; | |
2040 | int arg1, arg2; | |
2041 | { | |
2042 | *loc = (unsigned char) op; | |
2043 | STORE_NUMBER (loc + 1, arg1); | |
2044 | STORE_NUMBER (loc + 3, arg2); | |
2045 | } | |
2046 | ||
2047 | ||
2048 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
2049 | for OP followed by two-byte integer parameter ARG. */ | |
2050 | ||
2051 | static void | |
2052 | insert_op1 (op, loc, arg, end) | |
2053 | re_opcode_t op; | |
2054 | unsigned char *loc; | |
2055 | int arg; | |
2056 | unsigned char *end; | |
2057 | { | |
2058 | register unsigned char *pfrom = end; | |
2059 | register unsigned char *pto = end + 3; | |
2060 | ||
2061 | while (pfrom != loc) | |
2062 | *--pto = *--pfrom; | |
2063 | ||
2064 | store_op1 (op, loc, arg); | |
2065 | } | |
2066 | ||
2067 | ||
2068 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
2069 | ||
2070 | static void | |
2071 | insert_op2 (op, loc, arg1, arg2, end) | |
2072 | re_opcode_t op; | |
2073 | unsigned char *loc; | |
2074 | int arg1, arg2; | |
2075 | unsigned char *end; | |
2076 | { | |
2077 | register unsigned char *pfrom = end; | |
2078 | register unsigned char *pto = end + 5; | |
2079 | ||
2080 | while (pfrom != loc) | |
2081 | *--pto = *--pfrom; | |
2082 | ||
2083 | store_op2 (op, loc, arg1, arg2); | |
2084 | } | |
2085 | ||
2086 | ||
2087 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
2088 | after an alternative or a begin-subexpression. We assume there is at | |
2089 | least one character before the ^. */ | |
2090 | ||
2091 | static boolean | |
2092 | at_begline_loc_p (pattern, p, syntax) | |
2093 | const char *pattern, *p; | |
2094 | reg_syntax_t syntax; | |
2095 | { | |
2096 | const char *prev = p - 2; | |
2097 | boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; | |
2098 | ||
2099 | return | |
2100 | /* After a subexpression? */ | |
2101 | (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
2102 | /* After an alternative? */ | |
2103 | || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); | |
2104 | } | |
2105 | ||
2106 | ||
2107 | /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
2108 | at least one character after the $, i.e., `P < PEND'. */ | |
2109 | ||
2110 | static boolean | |
2111 | at_endline_loc_p (p, pend, syntax) | |
2112 | const char *p, *pend; | |
2113 | int syntax; | |
2114 | { | |
2115 | const char *next = p; | |
2116 | boolean next_backslash = *next == '\\'; | |
2117 | const char *next_next = p + 1 < pend ? p + 1 : NULL; | |
2118 | ||
2119 | return | |
2120 | /* Before a subexpression? */ | |
2121 | (syntax & RE_NO_BK_PARENS ? *next == ')' | |
2122 | : next_backslash && next_next && *next_next == ')') | |
2123 | /* Before an alternative? */ | |
2124 | || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
2125 | : next_backslash && next_next && *next_next == '|'); | |
2126 | } | |
2127 | ||
2128 | ||
2129 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and | |
2130 | false if it's not. */ | |
2131 | ||
2132 | static boolean | |
2133 | group_in_compile_stack (compile_stack, regnum) | |
2134 | compile_stack_type compile_stack; | |
2135 | regnum_t regnum; | |
2136 | { | |
2137 | int this_element; | |
2138 | ||
2139 | for (this_element = compile_stack.avail - 1; | |
2140 | this_element >= 0; | |
2141 | this_element--) | |
2142 | if (compile_stack.stack[this_element].regnum == regnum) | |
2143 | return true; | |
2144 | ||
2145 | return false; | |
2146 | } | |
2147 | ||
2148 | ||
2149 | /* Read the ending character of a range (in a bracket expression) from the | |
2150 | uncompiled pattern *P_PTR (which ends at PEND). We assume the | |
2151 | starting character is in `P[-2]'. (`P[-1]' is the character `-'.) | |
2152 | Then we set the translation of all bits between the starting and | |
2153 | ending characters (inclusive) in the compiled pattern B. | |
2154 | ||
2155 | Return an error code. | |
2156 | ||
2157 | We use these short variable names so we can use the same macros as | |
2158 | `regex_compile' itself. */ | |
2159 | ||
2160 | static reg_errcode_t | |
2161 | compile_range (p_ptr, pend, translate, syntax, b) | |
2162 | const char **p_ptr, *pend; | |
2163 | char *translate; | |
2164 | reg_syntax_t syntax; | |
2165 | unsigned char *b; | |
2166 | { | |
2167 | unsigned this_char; | |
2168 | ||
2169 | const char *p = *p_ptr; | |
2170 | ||
2171 | /* Even though the pattern is a signed `char *', we need to fetch into | |
2172 | `unsigned char's. Reason: if the high bit of the pattern character | |
2173 | is set, the range endpoints will be negative if we fetch into a | |
2174 | signed `char *'. */ | |
2175 | unsigned char range_end; | |
2176 | unsigned char range_start = p[-2]; | |
2177 | ||
2178 | if (p == pend) | |
2179 | return REG_ERANGE; | |
2180 | ||
2181 | PATFETCH (range_end); | |
2182 | ||
2183 | /* Have to increment the pointer into the pattern string, so the | |
2184 | caller isn't still at the ending character. */ | |
2185 | (*p_ptr)++; | |
2186 | ||
2187 | /* If the start is after the end, the range is empty. */ | |
2188 | if (range_start > range_end) | |
2189 | return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; | |
2190 | ||
2191 | /* Here we see why `this_char' has to be larger than an `unsigned | |
2192 | char' -- the range is inclusive, so if `range_end' == 0xff | |
2193 | (assuming 8-bit characters), we would otherwise go into an infinite | |
2194 | loop, since all characters <= 0xff. */ | |
2195 | for (this_char = range_start; this_char <= range_end; this_char++) | |
2196 | { | |
2197 | SET_LIST_BIT (TRANSLATE (this_char)); | |
2198 | } | |
2199 | ||
2200 | return REG_NOERROR; | |
2201 | } | |
2202 | \f | |
2203 | /* Failure stack declarations and macros; both re_compile_fastmap and | |
2204 | re_match_2 use a failure stack. These have to be macros because of | |
2205 | REGEX_ALLOCATE. */ | |
2206 | ||
2207 | ||
2208 | /* Number of failure points for which to initially allocate space | |
2209 | when matching. If this number is exceeded, we allocate more | |
2210 | space, so it is not a hard limit. */ | |
2211 | #ifndef INIT_FAILURE_ALLOC | |
2212 | #define INIT_FAILURE_ALLOC 5 | |
2213 | #endif | |
2214 | ||
2215 | /* Roughly the maximum number of failure points on the stack. Would be | |
2216 | exactly that if always used MAX_FAILURE_SPACE each time we failed. | |
2217 | This is a variable only so users of regex can assign to it; we never | |
2218 | change it ourselves. */ | |
2219 | int re_max_failures = 2000; | |
2220 | ||
2221 | typedef const unsigned char *fail_stack_elt_t; | |
2222 | ||
2223 | typedef struct | |
2224 | { | |
2225 | fail_stack_elt_t *stack; | |
2226 | unsigned size; | |
2227 | unsigned avail; /* Offset of next open position. */ | |
2228 | } fail_stack_type; | |
2229 | ||
2230 | #define FAIL_STACK_EMPTY() (fail_stack.avail == 0) | |
2231 | #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) | |
2232 | #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) | |
2233 | #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail]) | |
2234 | ||
2235 | ||
2236 | /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */ | |
2237 | ||
2238 | #define INIT_FAIL_STACK() \ | |
2239 | do { \ | |
2240 | fail_stack.stack = (fail_stack_elt_t *) \ | |
2241 | REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \ | |
2242 | \ | |
2243 | if (fail_stack.stack == NULL) \ | |
2244 | return -2; \ | |
2245 | \ | |
2246 | fail_stack.size = INIT_FAILURE_ALLOC; \ | |
2247 | fail_stack.avail = 0; \ | |
2248 | } while (0) | |
2249 | ||
2250 | ||
2251 | /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. | |
2252 | ||
2253 | Return 1 if succeeds, and 0 if either ran out of memory | |
2254 | allocating space for it or it was already too large. | |
2255 | ||
2256 | REGEX_REALLOCATE requires `destination' be declared. */ | |
2257 | ||
2258 | #define DOUBLE_FAIL_STACK(fail_stack) \ | |
2259 | ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \ | |
2260 | ? 0 \ | |
2261 | : ((fail_stack).stack = (fail_stack_elt_t *) \ | |
2262 | REGEX_REALLOCATE ((fail_stack).stack, \ | |
2263 | (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
2264 | ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \ | |
2265 | \ | |
2266 | (fail_stack).stack == NULL \ | |
2267 | ? 0 \ | |
2268 | : ((fail_stack).size <<= 1, \ | |
2269 | 1))) | |
2270 | ||
2271 | ||
2272 | /* Push PATTERN_OP on FAIL_STACK. | |
2273 | ||
2274 | Return 1 if was able to do so and 0 if ran out of memory allocating | |
2275 | space to do so. */ | |
2276 | #define PUSH_PATTERN_OP(pattern_op, fail_stack) \ | |
2277 | ((FAIL_STACK_FULL () \ | |
2278 | && !DOUBLE_FAIL_STACK (fail_stack)) \ | |
2279 | ? 0 \ | |
2280 | : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \ | |
2281 | 1)) | |
2282 | ||
2283 | /* This pushes an item onto the failure stack. Must be a four-byte | |
2284 | value. Assumes the variable `fail_stack'. Probably should only | |
2285 | be called from within `PUSH_FAILURE_POINT'. */ | |
2286 | #define PUSH_FAILURE_ITEM(item) \ | |
2287 | fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item | |
2288 | ||
2289 | /* The complement operation. Assumes `fail_stack' is nonempty. */ | |
2290 | #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail] | |
2291 | ||
2292 | /* Used to omit pushing failure point id's when we're not debugging. */ | |
2293 | #ifdef DEBUG | |
2294 | #define DEBUG_PUSH PUSH_FAILURE_ITEM | |
2295 | #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM () | |
2296 | #else | |
2297 | #define DEBUG_PUSH(item) | |
2298 | #define DEBUG_POP(item_addr) | |
2299 | #endif | |
2300 | ||
2301 | ||
2302 | /* Push the information about the state we will need | |
2303 | if we ever fail back to it. | |
2304 | ||
2305 | Requires variables fail_stack, regstart, regend, reg_info, and | |
2306 | num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be | |
2307 | declared. | |
2308 | ||
2309 | Does `return FAILURE_CODE' if runs out of memory. */ | |
2310 | ||
2311 | #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ | |
2312 | do { \ | |
2313 | char *destination; \ | |
2314 | /* Must be int, so when we don't save any registers, the arithmetic \ | |
2315 | of 0 + -1 isn't done as unsigned. */ \ | |
2316 | int this_reg; \ | |
2317 | \ | |
2318 | DEBUG_STATEMENT (failure_id++); \ | |
2319 | DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ | |
2320 | DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ | |
2321 | DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ | |
2322 | \ | |
2323 | DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \ | |
2324 | DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ | |
2325 | \ | |
2326 | /* Ensure we have enough space allocated for what we will push. */ \ | |
2327 | while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ | |
2328 | { \ | |
2329 | if (!DOUBLE_FAIL_STACK (fail_stack)) \ | |
2330 | return failure_code; \ | |
2331 | \ | |
2332 | DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ | |
2333 | (fail_stack).size); \ | |
2334 | DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ | |
2335 | } \ | |
2336 | \ | |
2337 | /* Push the info, starting with the registers. */ \ | |
2338 | DEBUG_PRINT1 ("\n"); \ | |
2339 | \ | |
2340 | for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ | |
2341 | this_reg++) \ | |
2342 | { \ | |
2343 | DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \ | |
2344 | DEBUG_STATEMENT (num_regs_pushed++); \ | |
2345 | \ | |
2346 | DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ | |
2347 | PUSH_FAILURE_ITEM (regstart[this_reg]); \ | |
2348 | \ | |
2349 | DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ | |
2350 | PUSH_FAILURE_ITEM (regend[this_reg]); \ | |
2351 | \ | |
2352 | DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \ | |
2353 | DEBUG_PRINT2 (" match_null=%d", \ | |
2354 | REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ | |
2355 | DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ | |
2356 | DEBUG_PRINT2 (" matched_something=%d", \ | |
2357 | MATCHED_SOMETHING (reg_info[this_reg])); \ | |
2358 | DEBUG_PRINT2 (" ever_matched=%d", \ | |
2359 | EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ | |
2360 | DEBUG_PRINT1 ("\n"); \ | |
2361 | PUSH_FAILURE_ITEM (reg_info[this_reg].word); \ | |
2362 | } \ | |
2363 | \ | |
2364 | DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\ | |
2365 | PUSH_FAILURE_ITEM (lowest_active_reg); \ | |
2366 | \ | |
2367 | DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\ | |
2368 | PUSH_FAILURE_ITEM (highest_active_reg); \ | |
2369 | \ | |
2370 | DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \ | |
2371 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ | |
2372 | PUSH_FAILURE_ITEM (pattern_place); \ | |
2373 | \ | |
2374 | DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \ | |
2375 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ | |
2376 | size2); \ | |
2377 | DEBUG_PRINT1 ("'\n"); \ | |
2378 | PUSH_FAILURE_ITEM (string_place); \ | |
2379 | \ | |
2380 | DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ | |
2381 | DEBUG_PUSH (failure_id); \ | |
2382 | } while (0) | |
2383 | ||
2384 | /* This is the number of items that are pushed and popped on the stack | |
2385 | for each register. */ | |
2386 | #define NUM_REG_ITEMS 3 | |
2387 | ||
2388 | /* Individual items aside from the registers. */ | |
2389 | #ifdef DEBUG | |
2390 | #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ | |
2391 | #else | |
2392 | #define NUM_NONREG_ITEMS 4 | |
2393 | #endif | |
2394 | ||
2395 | /* We push at most this many items on the stack. */ | |
2396 | #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS) | |
2397 | ||
2398 | /* We actually push this many items. */ | |
2399 | #define NUM_FAILURE_ITEMS \ | |
2400 | ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \ | |
2401 | + NUM_NONREG_ITEMS) | |
2402 | ||
2403 | /* How many items can still be added to the stack without overflowing it. */ | |
2404 | #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
2405 | ||
2406 | ||
2407 | /* Pops what PUSH_FAIL_STACK pushes. | |
2408 | ||
2409 | We restore into the parameters, all of which should be lvalues: | |
2410 | STR -- the saved data position. | |
2411 | PAT -- the saved pattern position. | |
2412 | LOW_REG, HIGH_REG -- the highest and lowest active registers. | |
2413 | REGSTART, REGEND -- arrays of string positions. | |
2414 | REG_INFO -- array of information about each subexpression. | |
2415 | ||
2416 | Also assumes the variables `fail_stack' and (if debugging), `bufp', | |
2417 | `pend', `string1', `size1', `string2', and `size2'. */ | |
2418 | ||
2419 | #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ | |
2420 | { \ | |
2421 | DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \ | |
2422 | int this_reg; \ | |
2423 | const unsigned char *string_temp; \ | |
2424 | \ | |
2425 | assert (!FAIL_STACK_EMPTY ()); \ | |
2426 | \ | |
2427 | /* Remove failure points and point to how many regs pushed. */ \ | |
2428 | DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
2429 | DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
2430 | DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ | |
2431 | \ | |
2432 | assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ | |
2433 | \ | |
2434 | DEBUG_POP (&failure_id); \ | |
2435 | DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ | |
2436 | \ | |
2437 | /* If the saved string location is NULL, it came from an \ | |
2438 | on_failure_keep_string_jump opcode, and we want to throw away the \ | |
2439 | saved NULL, thus retaining our current position in the string. */ \ | |
2440 | string_temp = POP_FAILURE_ITEM (); \ | |
2441 | if (string_temp != NULL) \ | |
2442 | str = (const char *) string_temp; \ | |
2443 | \ | |
2444 | DEBUG_PRINT2 (" Popping string 0x%x: `", str); \ | |
2445 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ | |
2446 | DEBUG_PRINT1 ("'\n"); \ | |
2447 | \ | |
2448 | pat = (unsigned char *) POP_FAILURE_ITEM (); \ | |
2449 | DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \ | |
2450 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
2451 | \ | |
2452 | /* Restore register info. */ \ | |
2453 | high_reg = (unsigned) POP_FAILURE_ITEM (); \ | |
2454 | DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \ | |
2455 | \ | |
2456 | low_reg = (unsigned) POP_FAILURE_ITEM (); \ | |
2457 | DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \ | |
2458 | \ | |
2459 | for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ | |
2460 | { \ | |
2461 | DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \ | |
2462 | \ | |
2463 | reg_info[this_reg].word = POP_FAILURE_ITEM (); \ | |
2464 | DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \ | |
2465 | \ | |
2466 | regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \ | |
2467 | DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \ | |
2468 | \ | |
2469 | regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \ | |
2470 | DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \ | |
2471 | } \ | |
2472 | } /* POP_FAILURE_POINT */ | |
2473 | \f | |
2474 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
2475 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
2476 | characters can start a string that matches the pattern. This fastmap | |
2477 | is used by re_search to skip quickly over impossible starting points. | |
2478 | ||
2479 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
2480 | area as BUFP->fastmap. | |
2481 | ||
2482 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in | |
2483 | the pattern buffer. | |
2484 | ||
2485 | Returns 0 if we succeed, -2 if an internal error. */ | |
2486 | ||
2487 | int | |
2488 | re_compile_fastmap (bufp) | |
2489 | struct re_pattern_buffer *bufp; | |
2490 | { | |
2491 | int j, k; | |
2492 | fail_stack_type fail_stack; | |
2493 | #ifndef REGEX_MALLOC | |
2494 | char *destination; | |
2495 | #endif | |
2496 | /* We don't push any register information onto the failure stack. */ | |
2497 | unsigned num_regs = 0; | |
2498 | ||
2499 | register char *fastmap = bufp->fastmap; | |
2500 | unsigned char *pattern = bufp->buffer; | |
2501 | unsigned long size = bufp->used; | |
2502 | const unsigned char *p = pattern; | |
2503 | register unsigned char *pend = pattern + size; | |
2504 | ||
2505 | /* Assume that each path through the pattern can be null until | |
2506 | proven otherwise. We set this false at the bottom of switch | |
2507 | statement, to which we get only if a particular path doesn't | |
2508 | match the empty string. */ | |
2509 | boolean path_can_be_null = true; | |
2510 | ||
2511 | /* We aren't doing a `succeed_n' to begin with. */ | |
2512 | boolean succeed_n_p = false; | |
2513 | ||
2514 | assert (fastmap != NULL && p != NULL); | |
2515 | ||
2516 | INIT_FAIL_STACK (); | |
2517 | bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ | |
2518 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ | |
2519 | bufp->can_be_null = 0; | |
2520 | ||
2521 | while (p != pend || !FAIL_STACK_EMPTY ()) | |
2522 | { | |
2523 | if (p == pend) | |
2524 | { | |
2525 | bufp->can_be_null |= path_can_be_null; | |
2526 | ||
2527 | /* Reset for next path. */ | |
2528 | path_can_be_null = true; | |
2529 | ||
2530 | p = fail_stack.stack[--fail_stack.avail]; | |
2531 | } | |
2532 | ||
2533 | /* We should never be about to go beyond the end of the pattern. */ | |
2534 | assert (p < pend); | |
2535 | ||
2536 | #ifdef SWITCH_ENUM_BUG | |
2537 | switch ((int) ((re_opcode_t) *p++)) | |
2538 | #else | |
2539 | switch ((re_opcode_t) *p++) | |
2540 | #endif | |
2541 | { | |
2542 | ||
2543 | /* I guess the idea here is to simply not bother with a fastmap | |
2544 | if a backreference is used, since it's too hard to figure out | |
2545 | the fastmap for the corresponding group. Setting | |
2546 | `can_be_null' stops `re_search_2' from using the fastmap, so | |
2547 | that is all we do. */ | |
2548 | case duplicate: | |
2549 | bufp->can_be_null = 1; | |
2550 | return 0; | |
2551 | ||
2552 | ||
2553 | /* Following are the cases which match a character. These end | |
2554 | with `break'. */ | |
2555 | ||
2556 | case exactn: | |
2557 | fastmap[p[1]] = 1; | |
2558 | break; | |
2559 | ||
2560 | ||
2561 | case charset: | |
2562 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
2563 | if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) | |
2564 | fastmap[j] = 1; | |
2565 | break; | |
2566 | ||
2567 | ||
2568 | case charset_not: | |
2569 | /* Chars beyond end of map must be allowed. */ | |
2570 | for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) | |
2571 | fastmap[j] = 1; | |
2572 | ||
2573 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
2574 | if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) | |
2575 | fastmap[j] = 1; | |
2576 | break; | |
2577 | ||
2578 | ||
2579 | case wordchar: | |
2580 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2581 | if (SYNTAX (j) == Sword) | |
2582 | fastmap[j] = 1; | |
2583 | break; | |
2584 | ||
2585 | ||
2586 | case notwordchar: | |
2587 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2588 | if (SYNTAX (j) != Sword) | |
2589 | fastmap[j] = 1; | |
2590 | break; | |
2591 | ||
2592 | ||
2593 | case anychar: | |
2594 | /* `.' matches anything ... */ | |
2595 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2596 | fastmap[j] = 1; | |
2597 | ||
2598 | /* ... except perhaps newline. */ | |
2599 | if (!(bufp->syntax & RE_DOT_NEWLINE)) | |
2600 | fastmap['\n'] = 0; | |
2601 | ||
2602 | /* Return if we have already set `can_be_null'; if we have, | |
2603 | then the fastmap is irrelevant. Something's wrong here. */ | |
2604 | else if (bufp->can_be_null) | |
2605 | return 0; | |
2606 | ||
2607 | /* Otherwise, have to check alternative paths. */ | |
2608 | break; | |
2609 | ||
2610 | ||
2611 | #ifdef emacs | |
2612 | case syntaxspec: | |
2613 | k = *p++; | |
2614 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2615 | if (SYNTAX (j) == (enum syntaxcode) k) | |
2616 | fastmap[j] = 1; | |
2617 | break; | |
2618 | ||
2619 | ||
2620 | case notsyntaxspec: | |
2621 | k = *p++; | |
2622 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
2623 | if (SYNTAX (j) != (enum syntaxcode) k) | |
2624 | fastmap[j] = 1; | |
2625 | break; | |
2626 | ||
2627 | ||
2628 | /* All cases after this match the empty string. These end with | |
2629 | `continue'. */ | |
2630 | ||
2631 | ||
2632 | case before_dot: | |
2633 | case at_dot: | |
2634 | case after_dot: | |
2635 | continue; | |
2636 | #endif /* not emacs */ | |
2637 | ||
2638 | ||
2639 | case no_op: | |
2640 | case begline: | |
2641 | case endline: | |
2642 | case begbuf: | |
2643 | case endbuf: | |
2644 | case wordbound: | |
2645 | case notwordbound: | |
2646 | case wordbeg: | |
2647 | case wordend: | |
2648 | case push_dummy_failure: | |
2649 | continue; | |
2650 | ||
2651 | ||
2652 | case jump_n: | |
2653 | case pop_failure_jump: | |
2654 | case maybe_pop_jump: | |
2655 | case jump: | |
2656 | case jump_past_alt: | |
2657 | case dummy_failure_jump: | |
2658 | EXTRACT_NUMBER_AND_INCR (j, p); | |
2659 | p += j; | |
2660 | if (j > 0) | |
2661 | continue; | |
2662 | ||
2663 | /* Jump backward implies we just went through the body of a | |
2664 | loop and matched nothing. Opcode jumped to should be | |
2665 | `on_failure_jump' or `succeed_n'. Just treat it like an | |
2666 | ordinary jump. For a * loop, it has pushed its failure | |
2667 | point already; if so, discard that as redundant. */ | |
2668 | if ((re_opcode_t) *p != on_failure_jump | |
2669 | && (re_opcode_t) *p != succeed_n) | |
2670 | continue; | |
2671 | ||
2672 | p++; | |
2673 | EXTRACT_NUMBER_AND_INCR (j, p); | |
2674 | p += j; | |
2675 | ||
2676 | /* If what's on the stack is where we are now, pop it. */ | |
2677 | if (!FAIL_STACK_EMPTY () | |
2678 | && fail_stack.stack[fail_stack.avail - 1] == p) | |
2679 | fail_stack.avail--; | |
2680 | ||
2681 | continue; | |
2682 | ||
2683 | ||
2684 | case on_failure_jump: | |
2685 | case on_failure_keep_string_jump: | |
2686 | handle_on_failure_jump: | |
2687 | EXTRACT_NUMBER_AND_INCR (j, p); | |
2688 | ||
2689 | /* For some patterns, e.g., `(a?)?', `p+j' here points to the | |
2690 | end of the pattern. We don't want to push such a point, | |
2691 | since when we restore it above, entering the switch will | |
2692 | increment `p' past the end of the pattern. We don't need | |
2693 | to push such a point since we obviously won't find any more | |
2694 | fastmap entries beyond `pend'. Such a pattern can match | |
2695 | the null string, though. */ | |
2696 | if (p + j < pend) | |
2697 | { | |
2698 | if (!PUSH_PATTERN_OP (p + j, fail_stack)) | |
2699 | return -2; | |
2700 | } | |
2701 | else | |
2702 | bufp->can_be_null = 1; | |
2703 | ||
2704 | if (succeed_n_p) | |
2705 | { | |
2706 | EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ | |
2707 | succeed_n_p = false; | |
2708 | } | |
2709 | ||
2710 | continue; | |
2711 | ||
2712 | ||
2713 | case succeed_n: | |
2714 | /* Get to the number of times to succeed. */ | |
2715 | p += 2; | |
2716 | ||
2717 | /* Increment p past the n for when k != 0. */ | |
2718 | EXTRACT_NUMBER_AND_INCR (k, p); | |
2719 | if (k == 0) | |
2720 | { | |
2721 | p -= 4; | |
2722 | succeed_n_p = true; /* Spaghetti code alert. */ | |
2723 | goto handle_on_failure_jump; | |
2724 | } | |
2725 | continue; | |
2726 | ||
2727 | ||
2728 | case set_number_at: | |
2729 | p += 4; | |
2730 | continue; | |
2731 | ||
2732 | ||
2733 | case start_memory: | |
2734 | case stop_memory: | |
2735 | p += 2; | |
2736 | continue; | |
2737 | ||
2738 | ||
2739 | default: | |
2740 | abort (); /* We have listed all the cases. */ | |
2741 | } /* switch *p++ */ | |
2742 | ||
2743 | /* Getting here means we have found the possible starting | |
2744 | characters for one path of the pattern -- and that the empty | |
2745 | string does not match. We need not follow this path further. | |
2746 | Instead, look at the next alternative (remembered on the | |
2747 | stack), or quit if no more. The test at the top of the loop | |
2748 | does these things. */ | |
2749 | path_can_be_null = false; | |
2750 | p = pend; | |
2751 | } /* while p */ | |
2752 | ||
2753 | /* Set `can_be_null' for the last path (also the first path, if the | |
2754 | pattern is empty). */ | |
2755 | bufp->can_be_null |= path_can_be_null; | |
2756 | return 0; | |
2757 | } /* re_compile_fastmap */ | |
2758 | \f | |
2759 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
2760 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
2761 | this memory for recording register information. STARTS and ENDS | |
2762 | must be allocated using the malloc library routine, and must each | |
2763 | be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
2764 | ||
2765 | If NUM_REGS == 0, then subsequent matches should allocate their own | |
2766 | register data. | |
2767 | ||
2768 | Unless this function is called, the first search or match using | |
2769 | PATTERN_BUFFER will allocate its own register data, without | |
2770 | freeing the old data. */ | |
2771 | ||
2772 | void | |
2773 | re_set_registers (bufp, regs, num_regs, starts, ends) | |
2774 | struct re_pattern_buffer *bufp; | |
2775 | struct re_registers *regs; | |
2776 | unsigned num_regs; | |
2777 | regoff_t *starts, *ends; | |
2778 | { | |
2779 | if (num_regs) | |
2780 | { | |
2781 | bufp->regs_allocated = REGS_REALLOCATE; | |
2782 | regs->num_regs = num_regs; | |
2783 | regs->start = starts; | |
2784 | regs->end = ends; | |
2785 | } | |
2786 | else | |
2787 | { | |
2788 | bufp->regs_allocated = REGS_UNALLOCATED; | |
2789 | regs->num_regs = 0; | |
2790 | regs->start = regs->end = (regoff_t) 0; | |
2791 | } | |
2792 | } | |
2793 | \f | |
2794 | /* Searching routines. */ | |
2795 | ||
2796 | /* Like re_search_2, below, but only one string is specified, and | |
2797 | doesn't let you say where to stop matching. */ | |
2798 | ||
2799 | int | |
2800 | re_search (bufp, string, size, startpos, range, regs) | |
2801 | struct re_pattern_buffer *bufp; | |
2802 | const char *string; | |
2803 | int size, startpos, range; | |
2804 | struct re_registers *regs; | |
2805 | { | |
2806 | return re_search_2 (bufp, NULL, 0, string, size, startpos, range, | |
2807 | regs, size); | |
2808 | } | |
2809 | ||
2810 | ||
2811 | /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
2812 | virtual concatenation of STRING1 and STRING2, starting first at index | |
2813 | STARTPOS, then at STARTPOS + 1, and so on. | |
2814 | ||
2815 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. | |
2816 | ||
2817 | RANGE is how far to scan while trying to match. RANGE = 0 means try | |
2818 | only at STARTPOS; in general, the last start tried is STARTPOS + | |
2819 | RANGE. | |
2820 | ||
2821 | In REGS, return the indices of the virtual concatenation of STRING1 | |
2822 | and STRING2 that matched the entire BUFP->buffer and its contained | |
2823 | subexpressions. | |
2824 | ||
2825 | Do not consider matching one past the index STOP in the virtual | |
2826 | concatenation of STRING1 and STRING2. | |
2827 | ||
2828 | We return either the position in the strings at which the match was | |
2829 | found, -1 if no match, or -2 if error (such as failure | |
2830 | stack overflow). */ | |
2831 | ||
2832 | int | |
2833 | re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop) | |
2834 | struct re_pattern_buffer *bufp; | |
2835 | const char *string1, *string2; | |
2836 | int size1, size2; | |
2837 | int startpos; | |
2838 | int range; | |
2839 | struct re_registers *regs; | |
2840 | int stop; | |
2841 | { | |
2842 | int val; | |
2843 | register char *fastmap = bufp->fastmap; | |
2844 | register char *translate = bufp->translate; | |
2845 | int total_size = size1 + size2; | |
2846 | int endpos = startpos + range; | |
2847 | ||
2848 | /* Check for out-of-range STARTPOS. */ | |
2849 | if (startpos < 0 || startpos > total_size) | |
2850 | return -1; | |
2851 | ||
2852 | /* Fix up RANGE if it might eventually take us outside | |
2853 | the virtual concatenation of STRING1 and STRING2. */ | |
2854 | if (endpos < -1) | |
2855 | range = -1 - startpos; | |
2856 | else if (endpos > total_size) | |
2857 | range = total_size - startpos; | |
2858 | ||
2859 | /* Update the fastmap now if not correct already. */ | |
2860 | if (fastmap && !bufp->fastmap_accurate) | |
2861 | if (re_compile_fastmap (bufp) == -2) | |
2862 | return -2; | |
2863 | ||
2864 | /* If the search isn't to be a backwards one, don't waste time in a | |
2865 | long search for a pattern that says it is anchored. */ | |
2866 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf | |
2867 | && range > 0) | |
2868 | { | |
2869 | if (startpos > 0) | |
2870 | return -1; | |
2871 | else | |
2872 | range = 1; | |
2873 | } | |
2874 | ||
2875 | for (;;) | |
2876 | { | |
2877 | /* If a fastmap is supplied, skip quickly over characters that | |
2878 | cannot be the start of a match. If the pattern can match the | |
2879 | null string, however, we don't need to skip characters; we want | |
2880 | the first null string. */ | |
2881 | if (fastmap && startpos < total_size && !bufp->can_be_null) | |
2882 | { | |
2883 | if (range > 0) /* Searching forwards. */ | |
2884 | { | |
2885 | register const char *d; | |
2886 | register int lim = 0; | |
2887 | int irange = range; | |
2888 | ||
2889 | if (startpos < size1 && startpos + range >= size1) | |
2890 | lim = range - (size1 - startpos); | |
2891 | ||
2892 | d = (startpos >= size1 ? string2 - size1 : string1) + startpos; | |
2893 | ||
2894 | /* Written out as an if-else to avoid testing `translate' | |
2895 | inside the loop. */ | |
2896 | if (translate) | |
2897 | while (range > lim | |
2898 | && !fastmap[(unsigned char) translate[*d++]]) | |
2899 | range--; | |
2900 | else | |
2901 | while (range > lim && !fastmap[(unsigned char) *d++]) | |
2902 | range--; | |
2903 | ||
2904 | startpos += irange - range; | |
2905 | } | |
2906 | else /* Searching backwards. */ | |
2907 | { | |
2908 | register char c = (size1 == 0 || startpos >= size1 | |
2909 | ? string2[startpos - size1] | |
2910 | : string1[startpos]); | |
2911 | ||
2912 | if (!fastmap[TRANSLATE (c)]) | |
2913 | goto advance; | |
2914 | } | |
2915 | } | |
2916 | ||
2917 | /* If can't match the null string, and that's all we have left, fail. */ | |
2918 | if (range >= 0 && startpos == total_size && fastmap | |
2919 | && !bufp->can_be_null) | |
2920 | return -1; | |
2921 | ||
2922 | val = re_match_2 (bufp, string1, size1, string2, size2, | |
2923 | startpos, regs, stop); | |
2924 | if (val >= 0) | |
2925 | return startpos; | |
2926 | ||
2927 | if (val == -2) | |
2928 | return -2; | |
2929 | ||
2930 | advance: | |
2931 | if (!range) | |
2932 | break; | |
2933 | else if (range > 0) | |
2934 | { | |
2935 | range--; | |
2936 | startpos++; | |
2937 | } | |
2938 | else | |
2939 | { | |
2940 | range++; | |
2941 | startpos--; | |
2942 | } | |
2943 | } | |
2944 | return -1; | |
2945 | } /* re_search_2 */ | |
2946 | \f | |
2947 | /* Declarations and macros for re_match_2. */ | |
2948 | ||
2949 | static int bcmp_translate (); | |
2950 | static boolean alt_match_null_string_p (), | |
2951 | common_op_match_null_string_p (), | |
2952 | group_match_null_string_p (); | |
2953 | ||
2954 | /* Structure for per-register (a.k.a. per-group) information. | |
2955 | This must not be longer than one word, because we push this value | |
2956 | onto the failure stack. Other register information, such as the | |
2957 | starting and ending positions (which are addresses), and the list of | |
2958 | inner groups (which is a bits list) are maintained in separate | |
2959 | variables. | |
2960 | ||
2961 | We are making a (strictly speaking) nonportable assumption here: that | |
2962 | the compiler will pack our bit fields into something that fits into | |
2963 | the type of `word', i.e., is something that fits into one item on the | |
2964 | failure stack. */ | |
2965 | typedef union | |
2966 | { | |
2967 | fail_stack_elt_t word; | |
2968 | struct | |
2969 | { | |
2970 | /* This field is one if this group can match the empty string, | |
2971 | zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ | |
2972 | #define MATCH_NULL_UNSET_VALUE 3 | |
2973 | unsigned match_null_string_p : 2; | |
2974 | unsigned is_active : 1; | |
2975 | unsigned matched_something : 1; | |
2976 | unsigned ever_matched_something : 1; | |
2977 | } bits; | |
2978 | } register_info_type; | |
2979 | ||
2980 | #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) | |
2981 | #define IS_ACTIVE(R) ((R).bits.is_active) | |
2982 | #define MATCHED_SOMETHING(R) ((R).bits.matched_something) | |
2983 | #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) | |
2984 | ||
2985 | ||
2986 | /* Call this when have matched something; it sets `matched' flags for the | |
2987 | registers corresponding to the group of which we currently are inside. | |
2988 | Also records whether this group ever matched something. We only care | |
2989 | about this information at `stop_memory', and then only about the | |
2990 | previous time through the loop (if the group is starred or whatever). | |
2991 | So it is ok to clear all the nonactive registers here. */ | |
2992 | #define SET_REGS_MATCHED() \ | |
2993 | do \ | |
2994 | { \ | |
2995 | unsigned r; \ | |
2996 | for (r = lowest_active_reg; r <= highest_active_reg; r++) \ | |
2997 | { \ | |
2998 | MATCHED_SOMETHING (reg_info[r]) \ | |
2999 | = EVER_MATCHED_SOMETHING (reg_info[r]) \ | |
3000 | = 1; \ | |
3001 | } \ | |
3002 | } \ | |
3003 | while (0) | |
3004 | ||
3005 | ||
3006 | /* This converts PTR, a pointer into one of the search strings `string1' | |
3007 | and `string2' into an offset from the beginning of that string. */ | |
3008 | #define POINTER_TO_OFFSET(ptr) \ | |
3009 | (FIRST_STRING_P (ptr) ? (ptr) - string1 : (ptr) - string2 + size1) | |
3010 | ||
3011 | /* Registers are set to a sentinel when they haven't yet matched. */ | |
3012 | #define REG_UNSET_VALUE ((char *) -1) | |
3013 | #define REG_UNSET(e) ((e) == REG_UNSET_VALUE) | |
3014 | ||
3015 | ||
3016 | /* Macros for dealing with the split strings in re_match_2. */ | |
3017 | ||
3018 | #define MATCHING_IN_FIRST_STRING (dend == end_match_1) | |
3019 | ||
3020 | /* Call before fetching a character with *d. This switches over to | |
3021 | string2 if necessary. */ | |
3022 | #define PREFETCH() \ | |
3023 | while (d == dend) \ | |
3024 | { \ | |
3025 | /* End of string2 => fail. */ \ | |
3026 | if (dend == end_match_2) \ | |
3027 | goto fail; \ | |
3028 | /* End of string1 => advance to string2. */ \ | |
3029 | d = string2; \ | |
3030 | dend = end_match_2; \ | |
3031 | } | |
3032 | ||
3033 | ||
3034 | /* Test if at very beginning or at very end of the virtual concatenation | |
3035 | of `string1' and `string2'. If only one string, it's `string2'. */ | |
3036 | #define AT_STRINGS_BEG() (d == (size1 ? string1 : string2) || !size2) | |
3037 | #define AT_STRINGS_END() (d == end2) | |
3038 | ||
3039 | ||
3040 | /* Test if D points to a character which is word-constituent. We have | |
3041 | two special cases to check for: if past the end of string1, look at | |
3042 | the first character in string2; and if before the beginning of | |
3043 | string2, look at the last character in string1. | |
3044 | ||
3045 | Assumes `string1' exists, so use in conjunction with AT_STRINGS_BEG (). */ | |
3046 | #define LETTER_P(d) \ | |
3047 | (SYNTAX ((d) == end1 ? *string2 \ | |
3048 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == Sword) | |
3049 | ||
3050 | /* Test if the character before D and the one at D differ with respect | |
3051 | to being word-constituent. */ | |
3052 | #define AT_WORD_BOUNDARY(d) \ | |
3053 | (AT_STRINGS_BEG () || AT_STRINGS_END () || LETTER_P (d - 1) != LETTER_P (d)) | |
3054 | ||
3055 | ||
3056 | /* Free everything we malloc. */ | |
3057 | #ifdef REGEX_MALLOC | |
3058 | #define FREE_VAR(var) if (var) free (var); var = NULL | |
3059 | #define FREE_VARIABLES() \ | |
3060 | do { \ | |
3061 | FREE_VAR (fail_stack.stack); \ | |
3062 | FREE_VAR (regstart); \ | |
3063 | FREE_VAR (regend); \ | |
3064 | FREE_VAR (old_regstart); \ | |
3065 | FREE_VAR (old_regend); \ | |
3066 | FREE_VAR (best_regstart); \ | |
3067 | FREE_VAR (best_regend); \ | |
3068 | FREE_VAR (reg_info); \ | |
3069 | FREE_VAR (reg_dummy); \ | |
3070 | FREE_VAR (reg_info_dummy); \ | |
3071 | } while (0) | |
3072 | #else /* not REGEX_MALLOC */ | |
3073 | /* Some MIPS systems (at least) want this to free alloca'd storage. */ | |
3074 | #define FREE_VARIABLES() alloca (0) | |
3075 | #endif /* not REGEX_MALLOC */ | |
3076 | ||
3077 | ||
3078 | /* These values must meet several constraints. They must not be valid | |
3079 | register values; since we have a limit of 255 registers (because | |
3080 | we use only one byte in the pattern for the register number), we can | |
3081 | use numbers larger than 255. They must differ by 1, because of | |
3082 | NUM_FAILURE_ITEMS above. And the value for the lowest register must | |
3083 | be larger than the value for the highest register, so we do not try | |
3084 | to actually save any registers when none are active. */ | |
3085 | #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) | |
3086 | #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) | |
3087 | \f | |
3088 | /* Matching routines. */ | |
3089 | ||
3090 | #ifndef emacs /* Emacs never uses this. */ | |
3091 | /* re_match is like re_match_2 except it takes only a single string. */ | |
3092 | ||
3093 | int | |
3094 | re_match (bufp, string, size, pos, regs) | |
3095 | struct re_pattern_buffer *bufp; | |
3096 | const char *string; | |
3097 | int size, pos; | |
3098 | struct re_registers *regs; | |
3099 | { | |
3100 | return re_match_2 (bufp, NULL, 0, string, size, pos, regs, size); | |
3101 | } | |
3102 | #endif /* not emacs */ | |
3103 | ||
3104 | ||
3105 | /* re_match_2 matches the compiled pattern in BUFP against the | |
3106 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
3107 | and SIZE2, respectively). We start matching at POS, and stop | |
3108 | matching at STOP. | |
3109 | ||
3110 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we | |
3111 | store offsets for the substring each group matched in REGS. See the | |
3112 | documentation for exactly how many groups we fill. | |
3113 | ||
3114 | We return -1 if no match, -2 if an internal error (such as the | |
3115 | failure stack overflowing). Otherwise, we return the length of the | |
3116 | matched substring. */ | |
3117 | ||
3118 | int | |
3119 | re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
3120 | struct re_pattern_buffer *bufp; | |
3121 | const char *string1, *string2; | |
3122 | int size1, size2; | |
3123 | int pos; | |
3124 | struct re_registers *regs; | |
3125 | int stop; | |
3126 | { | |
3127 | /* General temporaries. */ | |
3128 | int mcnt; | |
3129 | unsigned char *p1; | |
3130 | ||
3131 | /* Just past the end of the corresponding string. */ | |
3132 | const char *end1, *end2; | |
3133 | ||
3134 | /* Pointers into string1 and string2, just past the last characters in | |
3135 | each to consider matching. */ | |
3136 | const char *end_match_1, *end_match_2; | |
3137 | ||
3138 | /* Where we are in the data, and the end of the current string. */ | |
3139 | const char *d, *dend; | |
3140 | ||
3141 | /* Where we are in the pattern, and the end of the pattern. */ | |
3142 | unsigned char *p = bufp->buffer; | |
3143 | register unsigned char *pend = p + bufp->used; | |
3144 | ||
3145 | /* We use this to map every character in the string. */ | |
3146 | char *translate = bufp->translate; | |
3147 | ||
3148 | /* Failure point stack. Each place that can handle a failure further | |
3149 | down the line pushes a failure point on this stack. It consists of | |
3150 | restart, regend, and reg_info for all registers corresponding to | |
3151 | the subexpressions we're currently inside, plus the number of such | |
3152 | registers, and, finally, two char *'s. The first char * is where | |
3153 | to resume scanning the pattern; the second one is where to resume | |
3154 | scanning the strings. If the latter is zero, the failure point is | |
3155 | a ``dummy''; if a failure happens and the failure point is a dummy, | |
3156 | it gets discarded and the next next one is tried. */ | |
3157 | fail_stack_type fail_stack; | |
3158 | #ifdef DEBUG | |
3159 | static unsigned failure_id = 0; | |
3160 | #endif | |
3161 | ||
3162 | /* We fill all the registers internally, independent of what we | |
3163 | return, for use in backreferences. The number here includes | |
3164 | an element for register zero. */ | |
3165 | unsigned num_regs = bufp->re_nsub + 1; | |
3166 | ||
3167 | /* The currently active registers. */ | |
3168 | unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3169 | unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3170 | ||
3171 | /* Information on the contents of registers. These are pointers into | |
3172 | the input strings; they record just what was matched (on this | |
3173 | attempt) by a subexpression part of the pattern, that is, the | |
3174 | regnum-th regstart pointer points to where in the pattern we began | |
3175 | matching and the regnum-th regend points to right after where we | |
3176 | stopped matching the regnum-th subexpression. (The zeroth register | |
3177 | keeps track of what the whole pattern matches.) */ | |
3178 | const char **regstart, **regend; | |
3179 | ||
3180 | /* If a group that's operated upon by a repetition operator fails to | |
3181 | match anything, then the register for its start will need to be | |
3182 | restored because it will have been set to wherever in the string we | |
3183 | are when we last see its open-group operator. Similarly for a | |
3184 | register's end. */ | |
3185 | const char **old_regstart, **old_regend; | |
3186 | ||
3187 | /* The is_active field of reg_info helps us keep track of which (possibly | |
3188 | nested) subexpressions we are currently in. The matched_something | |
3189 | field of reg_info[reg_num] helps us tell whether or not we have | |
3190 | matched any of the pattern so far this time through the reg_num-th | |
3191 | subexpression. These two fields get reset each time through any | |
3192 | loop their register is in. */ | |
3193 | register_info_type *reg_info; | |
3194 | ||
3195 | /* The following record the register info as found in the above | |
3196 | variables when we find a match better than any we've seen before. | |
3197 | This happens as we backtrack through the failure points, which in | |
3198 | turn happens only if we have not yet matched the entire string. */ | |
3199 | unsigned best_regs_set = false; | |
3200 | const char **best_regstart, **best_regend; | |
3201 | ||
3202 | /* Logically, this is `best_regend[0]'. But we don't want to have to | |
3203 | allocate space for that if we're not allocating space for anything | |
3204 | else (see below). Also, we never need info about register 0 for | |
3205 | any of the other register vectors, and it seems rather a kludge to | |
3206 | treat `best_regend' differently than the rest. So we keep track of | |
3207 | the end of the best match so far in a separate variable. We | |
3208 | initialize this to NULL so that when we backtrack the first time | |
3209 | and need to test it, it's not garbage. */ | |
3210 | const char *match_end = NULL; | |
3211 | ||
3212 | /* Used when we pop values we don't care about. */ | |
3213 | const char **reg_dummy; | |
3214 | register_info_type *reg_info_dummy; | |
3215 | ||
3216 | #ifdef DEBUG | |
3217 | /* Counts the total number of registers pushed. */ | |
3218 | unsigned num_regs_pushed = 0; | |
3219 | #endif | |
3220 | ||
3221 | DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
3222 | ||
3223 | INIT_FAIL_STACK (); | |
3224 | ||
3225 | /* Do not bother to initialize all the register variables if there are | |
3226 | no groups in the pattern, as it takes a fair amount of time. If | |
3227 | there are groups, we include space for register 0 (the whole | |
3228 | pattern), even though we never use it, since it simplifies the | |
3229 | array indexing. We should fix this. */ | |
3230 | if (bufp->re_nsub) | |
3231 | { | |
3232 | regstart = REGEX_TALLOC (num_regs, const char *); | |
3233 | regend = REGEX_TALLOC (num_regs, const char *); | |
3234 | old_regstart = REGEX_TALLOC (num_regs, const char *); | |
3235 | old_regend = REGEX_TALLOC (num_regs, const char *); | |
3236 | best_regstart = REGEX_TALLOC (num_regs, const char *); | |
3237 | best_regend = REGEX_TALLOC (num_regs, const char *); | |
3238 | reg_info = REGEX_TALLOC (num_regs, register_info_type); | |
3239 | reg_dummy = REGEX_TALLOC (num_regs, const char *); | |
3240 | reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type); | |
3241 | ||
3242 | if (!(regstart && regend && old_regstart && old_regend && reg_info | |
3243 | && best_regstart && best_regend && reg_dummy && reg_info_dummy)) | |
3244 | { | |
3245 | FREE_VARIABLES (); | |
3246 | return -2; | |
3247 | } | |
3248 | } | |
3249 | #ifdef REGEX_MALLOC | |
3250 | else | |
3251 | { | |
3252 | /* We must initialize all our variables to NULL, so that | |
3253 | `FREE_VARIABLES' doesn't try to free them. Too bad this isn't | |
3254 | Lisp, so we could have a list of variables. As it is, */ | |
3255 | regstart = regend = old_regstart = old_regend = best_regstart | |
3256 | = best_regend = reg_dummy = NULL; | |
3257 | reg_info = reg_info_dummy = (register_info_type *) NULL; | |
3258 | } | |
3259 | #endif /* REGEX_MALLOC */ | |
3260 | ||
3261 | /* The starting position is bogus. */ | |
3262 | if (pos < 0 || pos > size1 + size2) | |
3263 | { | |
3264 | FREE_VARIABLES (); | |
3265 | return -1; | |
3266 | } | |
3267 | ||
3268 | /* Initialize subexpression text positions to -1 to mark ones that no | |
3269 | start_memory/stop_memory has been seen for. Also initialize the | |
3270 | register information struct. */ | |
3271 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3272 | { | |
3273 | regstart[mcnt] = regend[mcnt] | |
3274 | = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; | |
3275 | ||
3276 | REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; | |
3277 | IS_ACTIVE (reg_info[mcnt]) = 0; | |
3278 | MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
3279 | EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
3280 | } | |
3281 | ||
3282 | /* We move `string1' into `string2' if the latter's empty -- but not if | |
3283 | `string1' is null. */ | |
3284 | if (size2 == 0 && string1 != NULL) | |
3285 | { | |
3286 | string2 = string1; | |
3287 | size2 = size1; | |
3288 | string1 = 0; | |
3289 | size1 = 0; | |
3290 | } | |
3291 | end1 = string1 + size1; | |
3292 | end2 = string2 + size2; | |
3293 | ||
3294 | /* Compute where to stop matching, within the two strings. */ | |
3295 | if (stop <= size1) | |
3296 | { | |
3297 | end_match_1 = string1 + stop; | |
3298 | end_match_2 = string2; | |
3299 | } | |
3300 | else | |
3301 | { | |
3302 | end_match_1 = end1; | |
3303 | end_match_2 = string2 + stop - size1; | |
3304 | } | |
3305 | ||
3306 | /* `p' scans through the pattern as `d' scans through the data. | |
3307 | `dend' is the end of the input string that `d' points within. `d' | |
3308 | is advanced into the following input string whenever necessary, but | |
3309 | this happens before fetching; therefore, at the beginning of the | |
3310 | loop, `d' can be pointing at the end of a string, but it cannot | |
3311 | equal `string2'. */ | |
3312 | if (size1 > 0 && pos <= size1) | |
3313 | { | |
3314 | d = string1 + pos; | |
3315 | dend = end_match_1; | |
3316 | } | |
3317 | else | |
3318 | { | |
3319 | d = string2 + pos - size1; | |
3320 | dend = end_match_2; | |
3321 | } | |
3322 | ||
3323 | DEBUG_PRINT1 ("The compiled pattern is: "); | |
3324 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
3325 | DEBUG_PRINT1 ("The string to match is: `"); | |
3326 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
3327 | DEBUG_PRINT1 ("'\n"); | |
3328 | ||
3329 | /* This loops over pattern commands. It exits by returning from the | |
3330 | function if the match is complete, or it drops through if the match | |
3331 | fails at this starting point in the input data. */ | |
3332 | for (;;) | |
3333 | { | |
3334 | DEBUG_PRINT2 ("\n0x%x: ", p); | |
3335 | ||
3336 | if (p == pend) | |
3337 | { /* End of pattern means we might have succeeded. */ | |
3338 | DEBUG_PRINT1 ("End of pattern: "); | |
3339 | /* If not end of string, try backtracking. Otherwise done. */ | |
3340 | if (d != end_match_2) | |
3341 | { | |
3342 | DEBUG_PRINT1 ("backtracking.\n"); | |
3343 | ||
3344 | if (!FAIL_STACK_EMPTY ()) | |
3345 | { /* More failure points to try. */ | |
3346 | boolean same_str_p = (FIRST_STRING_P (match_end) | |
3347 | == MATCHING_IN_FIRST_STRING); | |
3348 | ||
3349 | /* If exceeds best match so far, save it. */ | |
3350 | if (!best_regs_set | |
3351 | || (same_str_p && d > match_end) | |
3352 | || (!same_str_p && !MATCHING_IN_FIRST_STRING)) | |
3353 | { | |
3354 | best_regs_set = true; | |
3355 | match_end = d; | |
3356 | ||
3357 | DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); | |
3358 | ||
3359 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3360 | { | |
3361 | best_regstart[mcnt] = regstart[mcnt]; | |
3362 | best_regend[mcnt] = regend[mcnt]; | |
3363 | } | |
3364 | } | |
3365 | goto fail; | |
3366 | } | |
3367 | ||
3368 | /* If no failure points, don't restore garbage. */ | |
3369 | else if (best_regs_set) | |
3370 | { | |
3371 | restore_best_regs: | |
3372 | /* Restore best match. It may happen that `dend == | |
3373 | end_match_1' while the restored d is in string2. | |
3374 | For example, the pattern `x.*y.*z' against the | |
3375 | strings `x-' and `y-z-', if the two strings are | |
3376 | not consecutive in memory. */ | |
3377 | d = match_end; | |
3378 | dend = ((d >= string1 && d <= end1) | |
3379 | ? end_match_1 : end_match_2); | |
3380 | ||
3381 | for (mcnt = 1; mcnt < num_regs; mcnt++) | |
3382 | { | |
3383 | regstart[mcnt] = best_regstart[mcnt]; | |
3384 | regend[mcnt] = best_regend[mcnt]; | |
3385 | } | |
3386 | } | |
3387 | } /* d != end_match_2 */ | |
3388 | ||
3389 | DEBUG_PRINT1 ("\nAccepting match.\n"); | |
3390 | ||
3391 | /* If caller wants register contents data back, do it. */ | |
3392 | if (regs && !bufp->no_sub) | |
3393 | { | |
3394 | /* Have the register data arrays been allocated? */ | |
3395 | if (bufp->regs_allocated == REGS_UNALLOCATED) | |
3396 | { /* No. So allocate them with malloc. We need one | |
3397 | extra element beyond `num_regs' for the `-1' marker | |
3398 | GNU code uses. */ | |
3399 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
3400 | regs->start = TALLOC (regs->num_regs, regoff_t); | |
3401 | regs->end = TALLOC (regs->num_regs, regoff_t); | |
3402 | if (regs->start == NULL || regs->end == NULL) | |
3403 | return -2; | |
3404 | bufp->regs_allocated = REGS_REALLOCATE; | |
3405 | } | |
3406 | else if (bufp->regs_allocated == REGS_REALLOCATE) | |
3407 | { /* Yes. If we need more elements than were already | |
3408 | allocated, reallocate them. If we need fewer, just | |
3409 | leave it alone. */ | |
3410 | if (regs->num_regs < num_regs + 1) | |
3411 | { | |
3412 | regs->num_regs = num_regs + 1; | |
3413 | RETALLOC (regs->start, regs->num_regs, regoff_t); | |
3414 | RETALLOC (regs->end, regs->num_regs, regoff_t); | |
3415 | if (regs->start == NULL || regs->end == NULL) | |
3416 | return -2; | |
3417 | } | |
3418 | } | |
3419 | else | |
3420 | assert (bufp->regs_allocated == REGS_FIXED); | |
3421 | ||
3422 | /* Convert the pointer data in `regstart' and `regend' to | |
3423 | indices. Register zero has to be set differently, | |
3424 | since we haven't kept track of any info for it. */ | |
3425 | if (regs->num_regs > 0) | |
3426 | { | |
3427 | regs->start[0] = pos; | |
3428 | regs->end[0] = (MATCHING_IN_FIRST_STRING ? d - string1 | |
3429 | : d - string2 + size1); | |
3430 | } | |
3431 | ||
3432 | /* Go through the first `min (num_regs, regs->num_regs)' | |
3433 | registers, since that is all we initialized. */ | |
3434 | for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++) | |
3435 | { | |
3436 | if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) | |
3437 | regs->start[mcnt] = regs->end[mcnt] = -1; | |
3438 | else | |
3439 | { | |
3440 | regs->start[mcnt] = POINTER_TO_OFFSET (regstart[mcnt]); | |
3441 | regs->end[mcnt] = POINTER_TO_OFFSET (regend[mcnt]); | |
3442 | } | |
3443 | } | |
3444 | ||
3445 | /* If the regs structure we return has more elements than | |
3446 | were in the pattern, set the extra elements to -1. If | |
3447 | we (re)allocated the registers, this is the case, | |
3448 | because we always allocate enough to have at least one | |
3449 | -1 at the end. */ | |
3450 | for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++) | |
3451 | regs->start[mcnt] = regs->end[mcnt] = -1; | |
3452 | } /* regs && !bufp->no_sub */ | |
3453 | ||
3454 | FREE_VARIABLES (); | |
3455 | DEBUG_PRINT2 ("%d registers pushed.\n", num_regs_pushed); | |
3456 | ||
3457 | mcnt = d - pos - (MATCHING_IN_FIRST_STRING | |
3458 | ? string1 | |
3459 | : string2 - size1); | |
3460 | ||
3461 | DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); | |
3462 | ||
3463 | return mcnt; | |
3464 | } | |
3465 | ||
3466 | /* Otherwise match next pattern command. */ | |
3467 | #ifdef SWITCH_ENUM_BUG | |
3468 | switch ((int) ((re_opcode_t) *p++)) | |
3469 | #else | |
3470 | switch ((re_opcode_t) *p++) | |
3471 | #endif | |
3472 | { | |
3473 | /* Ignore these. Used to ignore the n of succeed_n's which | |
3474 | currently have n == 0. */ | |
3475 | case no_op: | |
3476 | DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
3477 | break; | |
3478 | ||
3479 | ||
3480 | /* Match the next n pattern characters exactly. The following | |
3481 | byte in the pattern defines n, and the n bytes after that | |
3482 | are the characters to match. */ | |
3483 | case exactn: | |
3484 | mcnt = *p++; | |
3485 | DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); | |
3486 | ||
3487 | /* This is written out as an if-else so we don't waste time | |
3488 | testing `translate' inside the loop. */ | |
3489 | if (translate) | |
3490 | { | |
3491 | do | |
3492 | { | |
3493 | PREFETCH (); | |
3494 | if (translate[(unsigned char) *d++] != (char) *p++) | |
3495 | goto fail; | |
3496 | } | |
3497 | while (--mcnt); | |
3498 | } | |
3499 | else | |
3500 | { | |
3501 | do | |
3502 | { | |
3503 | PREFETCH (); | |
3504 | if (*d++ != (char) *p++) goto fail; | |
3505 | } | |
3506 | while (--mcnt); | |
3507 | } | |
3508 | SET_REGS_MATCHED (); | |
3509 | break; | |
3510 | ||
3511 | ||
3512 | /* Match any character except possibly a newline or a null. */ | |
3513 | case anychar: | |
3514 | DEBUG_PRINT1 ("EXECUTING anychar.\n"); | |
3515 | ||
3516 | PREFETCH (); | |
3517 | ||
3518 | if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n') | |
3519 | || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000')) | |
3520 | goto fail; | |
3521 | ||
3522 | SET_REGS_MATCHED (); | |
3523 | DEBUG_PRINT2 (" Matched `%d'.\n", *d); | |
3524 | d++; | |
3525 | break; | |
3526 | ||
3527 | ||
3528 | case charset: | |
3529 | case charset_not: | |
3530 | { | |
3531 | register unsigned char c; | |
3532 | boolean not = (re_opcode_t) *(p - 1) == charset_not; | |
3533 | ||
3534 | DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); | |
3535 | ||
3536 | PREFETCH (); | |
3537 | c = TRANSLATE (*d); /* The character to match. */ | |
3538 | ||
3539 | /* Cast to `unsigned' instead of `unsigned char' in case the | |
3540 | bit list is a full 32 bytes long. */ | |
3541 | if (c < (unsigned) (*p * BYTEWIDTH) | |
3542 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
3543 | not = !not; | |
3544 | ||
3545 | p += 1 + *p; | |
3546 | ||
3547 | if (!not) goto fail; | |
3548 | ||
3549 | SET_REGS_MATCHED (); | |
3550 | d++; | |
3551 | break; | |
3552 | } | |
3553 | ||
3554 | ||
3555 | /* The beginning of a group is represented by start_memory. | |
3556 | The arguments are the register number in the next byte, and the | |
3557 | number of groups inner to this one in the next. The text | |
3558 | matched within the group is recorded (in the internal | |
3559 | registers data structure) under the register number. */ | |
3560 | case start_memory: | |
3561 | DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]); | |
3562 | ||
3563 | /* Find out if this group can match the empty string. */ | |
3564 | p1 = p; /* To send to group_match_null_string_p. */ | |
3565 | ||
3566 | if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) | |
3567 | REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3568 | = group_match_null_string_p (&p1, pend, reg_info); | |
3569 | ||
3570 | /* Save the position in the string where we were the last time | |
3571 | we were at this open-group operator in case the group is | |
3572 | operated upon by a repetition operator, e.g., with `(a*)*b' | |
3573 | against `ab'; then we want to ignore where we are now in | |
3574 | the string in case this attempt to match fails. */ | |
3575 | old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3576 | ? REG_UNSET (regstart[*p]) ? d : regstart[*p] | |
3577 | : regstart[*p]; | |
3578 | DEBUG_PRINT2 (" old_regstart: %d\n", | |
3579 | POINTER_TO_OFFSET (old_regstart[*p])); | |
3580 | ||
3581 | regstart[*p] = d; | |
3582 | DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); | |
3583 | ||
3584 | IS_ACTIVE (reg_info[*p]) = 1; | |
3585 | MATCHED_SOMETHING (reg_info[*p]) = 0; | |
3586 | ||
3587 | /* This is the new highest active register. */ | |
3588 | highest_active_reg = *p; | |
3589 | ||
3590 | /* If nothing was active before, this is the new lowest active | |
3591 | register. */ | |
3592 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
3593 | lowest_active_reg = *p; | |
3594 | ||
3595 | /* Move past the register number and inner group count. */ | |
3596 | p += 2; | |
3597 | break; | |
3598 | ||
3599 | ||
3600 | /* The stop_memory opcode represents the end of a group. Its | |
3601 | arguments are the same as start_memory's: the register | |
3602 | number, and the number of inner groups. */ | |
3603 | case stop_memory: | |
3604 | DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]); | |
3605 | ||
3606 | /* We need to save the string position the last time we were at | |
3607 | this close-group operator in case the group is operated | |
3608 | upon by a repetition operator, e.g., with `((a*)*(b*)*)*' | |
3609 | against `aba'; then we want to ignore where we are now in | |
3610 | the string in case this attempt to match fails. */ | |
3611 | old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
3612 | ? REG_UNSET (regend[*p]) ? d : regend[*p] | |
3613 | : regend[*p]; | |
3614 | DEBUG_PRINT2 (" old_regend: %d\n", | |
3615 | POINTER_TO_OFFSET (old_regend[*p])); | |
3616 | ||
3617 | regend[*p] = d; | |
3618 | DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); | |
3619 | ||
3620 | /* This register isn't active anymore. */ | |
3621 | IS_ACTIVE (reg_info[*p]) = 0; | |
3622 | ||
3623 | /* If this was the only register active, nothing is active | |
3624 | anymore. */ | |
3625 | if (lowest_active_reg == highest_active_reg) | |
3626 | { | |
3627 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3628 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3629 | } | |
3630 | else | |
3631 | { /* We must scan for the new highest active register, since | |
3632 | it isn't necessarily one less than now: consider | |
3633 | (a(b)c(d(e)f)g). When group 3 ends, after the f), the | |
3634 | new highest active register is 1. */ | |
3635 | unsigned char r = *p - 1; | |
3636 | while (r > 0 && !IS_ACTIVE (reg_info[r])) | |
3637 | r--; | |
3638 | ||
3639 | /* If we end up at register zero, that means that we saved | |
3640 | the registers as the result of an `on_failure_jump', not | |
3641 | a `start_memory', and we jumped to past the innermost | |
3642 | `stop_memory'. For example, in ((.)*) we save | |
3643 | registers 1 and 2 as a result of the *, but when we pop | |
3644 | back to the second ), we are at the stop_memory 1. | |
3645 | Thus, nothing is active. */ | |
3646 | if (r == 0) | |
3647 | { | |
3648 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
3649 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
3650 | } | |
3651 | else | |
3652 | highest_active_reg = r; | |
3653 | } | |
3654 | ||
3655 | /* If just failed to match something this time around with a | |
3656 | group that's operated on by a repetition operator, try to | |
3657 | force exit from the ``loop,'' and restore the register | |
3658 | information for this group that we had before trying this | |
3659 | last match. */ | |
3660 | if ((!MATCHED_SOMETHING (reg_info[*p]) | |
3661 | || (re_opcode_t) p[-3] == start_memory) | |
3662 | && (p + 2) < pend) | |
3663 | { | |
3664 | boolean is_a_jump_n = false; | |
3665 | ||
3666 | p1 = p + 2; | |
3667 | mcnt = 0; | |
3668 | switch ((re_opcode_t) *p1++) | |
3669 | { | |
3670 | case jump_n: | |
3671 | is_a_jump_n = true; | |
3672 | case pop_failure_jump: | |
3673 | case maybe_pop_jump: | |
3674 | case jump: | |
3675 | case dummy_failure_jump: | |
3676 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
3677 | if (is_a_jump_n) | |
3678 | p1 += 2; | |
3679 | break; | |
3680 | ||
3681 | default: | |
3682 | /* do nothing */ ; | |
3683 | } | |
3684 | p1 += mcnt; | |
3685 | ||
3686 | /* If the next operation is a jump backwards in the pattern | |
3687 | to an on_failure_jump right before the start_memory | |
3688 | corresponding to this stop_memory, exit from the loop | |
3689 | by forcing a failure after pushing on the stack the | |
3690 | on_failure_jump's jump in the pattern, and d. */ | |
3691 | if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump | |
3692 | && (re_opcode_t) p1[3] == start_memory && p1[4] == *p) | |
3693 | { | |
3694 | /* If this group ever matched anything, then restore | |
3695 | what its registers were before trying this last | |
3696 | failed match, e.g., with `(a*)*b' against `ab' for | |
3697 | regstart[1], and, e.g., with `((a*)*(b*)*)*' | |
3698 | against `aba' for regend[3]. | |
3699 | ||
3700 | Also restore the registers for inner groups for, | |
3701 | e.g., `((a*)(b*))*' against `aba' (register 3 would | |
3702 | otherwise get trashed). */ | |
3703 | ||
3704 | if (EVER_MATCHED_SOMETHING (reg_info[*p])) | |
3705 | { | |
3706 | unsigned r; | |
3707 | ||
3708 | EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; | |
3709 | ||
3710 | /* Restore this and inner groups' (if any) registers. */ | |
3711 | for (r = *p; r < *p + *(p + 1); r++) | |
3712 | { | |
3713 | regstart[r] = old_regstart[r]; | |
3714 | ||
3715 | /* xx why this test? */ | |
3716 | if ((int) old_regend[r] >= (int) regstart[r]) | |
3717 | regend[r] = old_regend[r]; | |
3718 | } | |
3719 | } | |
3720 | p1++; | |
3721 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
3722 | PUSH_FAILURE_POINT (p1 + mcnt, d, -2); | |
3723 | ||
3724 | goto fail; | |
3725 | } | |
3726 | } | |
3727 | ||
3728 | /* Move past the register number and the inner group count. */ | |
3729 | p += 2; | |
3730 | break; | |
3731 | ||
3732 | ||
3733 | /* \<digit> has been turned into a `duplicate' command which is | |
3734 | followed by the numeric value of <digit> as the register number. */ | |
3735 | case duplicate: | |
3736 | { | |
3737 | register const char *d2, *dend2; | |
3738 | int regno = *p++; /* Get which register to match against. */ | |
3739 | DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); | |
3740 | ||
3741 | /* Can't back reference a group which we've never matched. */ | |
3742 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
3743 | goto fail; | |
3744 | ||
3745 | /* Where in input to try to start matching. */ | |
3746 | d2 = regstart[regno]; | |
3747 | ||
3748 | /* Where to stop matching; if both the place to start and | |
3749 | the place to stop matching are in the same string, then | |
3750 | set to the place to stop, otherwise, for now have to use | |
3751 | the end of the first string. */ | |
3752 | ||
3753 | dend2 = ((FIRST_STRING_P (regstart[regno]) | |
3754 | == FIRST_STRING_P (regend[regno])) | |
3755 | ? regend[regno] : end_match_1); | |
3756 | for (;;) | |
3757 | { | |
3758 | /* If necessary, advance to next segment in register | |
3759 | contents. */ | |
3760 | while (d2 == dend2) | |
3761 | { | |
3762 | if (dend2 == end_match_2) break; | |
3763 | if (dend2 == regend[regno]) break; | |
3764 | ||
3765 | /* End of string1 => advance to string2. */ | |
3766 | d2 = string2; | |
3767 | dend2 = regend[regno]; | |
3768 | } | |
3769 | /* At end of register contents => success */ | |
3770 | if (d2 == dend2) break; | |
3771 | ||
3772 | /* If necessary, advance to next segment in data. */ | |
3773 | PREFETCH (); | |
3774 | ||
3775 | /* How many characters left in this segment to match. */ | |
3776 | mcnt = dend - d; | |
3777 | ||
3778 | /* Want how many consecutive characters we can match in | |
3779 | one shot, so, if necessary, adjust the count. */ | |
3780 | if (mcnt > dend2 - d2) | |
3781 | mcnt = dend2 - d2; | |
3782 | ||
3783 | /* Compare that many; failure if mismatch, else move | |
3784 | past them. */ | |
3785 | if (translate | |
3786 | ? bcmp_translate (d, d2, mcnt, translate) | |
3787 | : bcmp (d, d2, mcnt)) | |
3788 | goto fail; | |
3789 | d += mcnt, d2 += mcnt; | |
3790 | } | |
3791 | } | |
3792 | break; | |
3793 | ||
3794 | ||
3795 | /* begline matches the empty string at the beginning of the string | |
3796 | (unless `not_bol' is set in `bufp'), and, if | |
3797 | `newline_anchor' is set, after newlines. */ | |
3798 | case begline: | |
3799 | DEBUG_PRINT1 ("EXECUTING begline.\n"); | |
3800 | ||
3801 | if (AT_STRINGS_BEG ()) | |
3802 | { | |
3803 | if (!bufp->not_bol) break; | |
3804 | } | |
3805 | else if (d[-1] == '\n' && bufp->newline_anchor) | |
3806 | { | |
3807 | break; | |
3808 | } | |
3809 | /* In all other cases, we fail. */ | |
3810 | goto fail; | |
3811 | ||
3812 | ||
3813 | /* endline is the dual of begline. */ | |
3814 | case endline: | |
3815 | DEBUG_PRINT1 ("EXECUTING endline.\n"); | |
3816 | ||
3817 | if (AT_STRINGS_END ()) | |
3818 | { | |
3819 | if (!bufp->not_eol) break; | |
3820 | } | |
3821 | ||
3822 | /* We have to ``prefetch'' the next character. */ | |
3823 | else if ((d == end1 ? *string2 : *d) == '\n' | |
3824 | && bufp->newline_anchor) | |
3825 | { | |
3826 | break; | |
3827 | } | |
3828 | goto fail; | |
3829 | ||
3830 | ||
3831 | /* Match at the very beginning of the data. */ | |
3832 | case begbuf: | |
3833 | DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
3834 | if (AT_STRINGS_BEG ()) | |
3835 | break; | |
3836 | goto fail; | |
3837 | ||
3838 | ||
3839 | /* Match at the very end of the data. */ | |
3840 | case endbuf: | |
3841 | DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
3842 | if (AT_STRINGS_END ()) | |
3843 | break; | |
3844 | goto fail; | |
3845 | ||
3846 | ||
3847 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It | |
3848 | pushes NULL as the value for the string on the stack. Then | |
3849 | `pop_failure_point' will keep the current value for the | |
3850 | string, instead of restoring it. To see why, consider | |
3851 | matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
3852 | then the . fails against the \n. But the next thing we want | |
3853 | to do is match the \n against the \n; if we restored the | |
3854 | string value, we would be back at the foo. | |
3855 | ||
3856 | Because this is used only in specific cases, we don't need to | |
3857 | check all the things that `on_failure_jump' does, to make | |
3858 | sure the right things get saved on the stack. Hence we don't | |
3859 | share its code. The only reason to push anything on the | |
3860 | stack at all is that otherwise we would have to change | |
3861 | `anychar's code to do something besides goto fail in this | |
3862 | case; that seems worse than this. */ | |
3863 | case on_failure_keep_string_jump: | |
3864 | DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); | |
3865 | ||
3866 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
3867 | DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); | |
3868 | ||
3869 | PUSH_FAILURE_POINT (p + mcnt, NULL, -2); | |
3870 | break; | |
3871 | ||
3872 | ||
3873 | /* Uses of on_failure_jump: | |
3874 | ||
3875 | Each alternative starts with an on_failure_jump that points | |
3876 | to the beginning of the next alternative. Each alternative | |
3877 | except the last ends with a jump that in effect jumps past | |
3878 | the rest of the alternatives. (They really jump to the | |
3879 | ending jump of the following alternative, because tensioning | |
3880 | these jumps is a hassle.) | |
3881 | ||
3882 | Repeats start with an on_failure_jump that points past both | |
3883 | the repetition text and either the following jump or | |
3884 | pop_failure_jump back to this on_failure_jump. */ | |
3885 | case on_failure_jump: | |
3886 | on_failure: | |
3887 | DEBUG_PRINT1 ("EXECUTING on_failure_jump"); | |
3888 | ||
3889 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
3890 | DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); | |
3891 | ||
3892 | /* If this on_failure_jump comes right before a group (i.e., | |
3893 | the original * applied to a group), save the information | |
3894 | for that group and all inner ones, so that if we fail back | |
3895 | to this point, the group's information will be correct. | |
3896 | For example, in \(a*\)*\1, we only need the preceding group, | |
3897 | and in \(\(a*\)b*\)\2, we need the inner group. */ | |
3898 | ||
3899 | /* We can't use `p' to check ahead because we push | |
3900 | a failure point to `p + mcnt' after we do this. */ | |
3901 | p1 = p; | |
3902 | ||
3903 | /* We need to skip no_op's before we look for the | |
3904 | start_memory in case this on_failure_jump is happening as | |
3905 | the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 | |
3906 | against aba. */ | |
3907 | while (p1 < pend && (re_opcode_t) *p1 == no_op) | |
3908 | p1++; | |
3909 | ||
3910 | if (p1 < pend && (re_opcode_t) *p1 == start_memory) | |
3911 | { | |
3912 | /* We have a new highest active register now. This will | |
3913 | get reset at the start_memory we are about to get to, | |
3914 | but we will have saved all the registers relevant to | |
3915 | this repetition op, as described above. */ | |
3916 | highest_active_reg = *(p1 + 1) + *(p1 + 2); | |
3917 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
3918 | lowest_active_reg = *(p1 + 1); | |
3919 | } | |
3920 | ||
3921 | DEBUG_PRINT1 (":\n"); | |
3922 | PUSH_FAILURE_POINT (p + mcnt, d, -2); | |
3923 | break; | |
3924 | ||
3925 | ||
3926 | /* A smart repeat ends with a maybe_pop_jump. | |
3927 | We change it either to a pop_failure_jump or a jump. */ | |
3928 | case maybe_pop_jump: | |
3929 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
3930 | DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); | |
3931 | { | |
3932 | register unsigned char *p2 = p; | |
3933 | ||
3934 | /* Compare the beginning of the repeat with what in the | |
3935 | pattern follows its end. If we can establish that there | |
3936 | is nothing that they would both match, i.e., that we | |
3937 | would have to backtrack because of (as in, e.g., `a*a') | |
3938 | then we can change to pop_failure_jump, because we'll | |
3939 | never have to backtrack. | |
3940 | ||
3941 | This is not true in the case of alternatives: in | |
3942 | `(a|ab)*' we do need to backtrack to the `ab' alternative | |
3943 | (e.g., if the string was `ab'). But instead of trying to | |
3944 | detect that here, the alternative has put on a dummy | |
3945 | failure point which is what we will end up popping. */ | |
3946 | ||
3947 | /* Skip over open/close-group commands. */ | |
3948 | while (p2 + 2 < pend | |
3949 | && ((re_opcode_t) *p2 == stop_memory | |
3950 | || (re_opcode_t) *p2 == start_memory)) | |
3951 | p2 += 3; /* Skip over args, too. */ | |
3952 | ||
3953 | /* If we're at the end of the pattern, we can change. */ | |
3954 | if (p2 == pend) | |
3955 | { | |
3956 | p[-3] = (unsigned char) pop_failure_jump; | |
3957 | DEBUG_PRINT1 | |
3958 | (" End of pattern: change to `pop_failure_jump'.\n"); | |
3959 | } | |
3960 | ||
3961 | else if ((re_opcode_t) *p2 == exactn | |
3962 | || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) | |
3963 | { | |
3964 | register unsigned char c | |
3965 | = *p2 == (unsigned char) endline ? '\n' : p2[2]; | |
3966 | p1 = p + mcnt; | |
3967 | ||
3968 | /* p1[0] ... p1[2] are the `on_failure_jump' corresponding | |
3969 | to the `maybe_finalize_jump' of this case. Examine what | |
3970 | follows. */ | |
3971 | if ((re_opcode_t) p1[3] == exactn && p1[5] != c) | |
3972 | p[-3] = (unsigned char) pop_failure_jump; | |
3973 | else if ((re_opcode_t) p1[3] == charset | |
3974 | || (re_opcode_t) p1[3] == charset_not) | |
3975 | { | |
3976 | int not = (re_opcode_t) p1[3] == charset_not; | |
3977 | ||
3978 | if (c < (unsigned char) (p1[4] * BYTEWIDTH) | |
3979 | && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
3980 | not = !not; | |
3981 | ||
3982 | /* `not' is equal to 1 if c would match, which means | |
3983 | that we can't change to pop_failure_jump. */ | |
3984 | if (!not) | |
3985 | { | |
3986 | p[-3] = (unsigned char) pop_failure_jump; | |
3987 | DEBUG_PRINT1 | |
3988 | (" No match: change to `pop_failure_jump'.\n"); | |
3989 | ||
3990 | } | |
3991 | } | |
3992 | } | |
3993 | } | |
3994 | p -= 2; /* Point at relative address again. */ | |
3995 | if ((re_opcode_t) p[-1] != pop_failure_jump) | |
3996 | { | |
3997 | p[-1] = (unsigned char) jump; | |
3998 | goto unconditional_jump; | |
3999 | } | |
4000 | /* Note fall through. */ | |
4001 | ||
4002 | ||
4003 | /* The end of a simple repeat has a pop_failure_jump back to | |
4004 | its matching on_failure_jump, where the latter will push a | |
4005 | failure point. The pop_failure_jump takes off failure | |
4006 | points put on by this pop_failure_jump's matching | |
4007 | on_failure_jump; we got through the pattern to here from the | |
4008 | matching on_failure_jump, so didn't fail. */ | |
4009 | case pop_failure_jump: | |
4010 | { | |
4011 | /* We need to pass separate storage for the lowest and | |
4012 | highest registers, even though we don't care about the | |
4013 | actual values. Otherwise, we will restore only one | |
4014 | register from the stack, since lowest will == highest in | |
4015 | `pop_failure_point'. */ | |
4016 | unsigned dummy_low_reg, dummy_high_reg; | |
4017 | unsigned char *pdummy; | |
4018 | const char *sdummy; | |
4019 | ||
4020 | DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); | |
4021 | POP_FAILURE_POINT (sdummy, pdummy, | |
4022 | dummy_low_reg, dummy_high_reg, | |
4023 | reg_dummy, reg_dummy, reg_info_dummy); | |
4024 | } | |
4025 | /* Note fall through. */ | |
4026 | ||
4027 | ||
4028 | /* Unconditionally jump (without popping any failure points). */ | |
4029 | case jump: | |
4030 | unconditional_jump: | |
4031 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ | |
4032 | DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); | |
4033 | p += mcnt; /* Do the jump. */ | |
4034 | DEBUG_PRINT2 ("(to 0x%x).\n", p); | |
4035 | break; | |
4036 | ||
4037 | ||
4038 | /* We need this opcode so we can detect where alternatives end | |
4039 | in `group_match_null_string_p' et al. */ | |
4040 | case jump_past_alt: | |
4041 | DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); | |
4042 | goto unconditional_jump; | |
4043 | ||
4044 | ||
4045 | /* Normally, the on_failure_jump pushes a failure point, which | |
4046 | then gets popped at pop_failure_jump. We will end up at | |
4047 | pop_failure_jump, also, and with a pattern of, say, `a+', we | |
4048 | are skipping over the on_failure_jump, so we have to push | |
4049 | something meaningless for pop_failure_jump to pop. */ | |
4050 | case dummy_failure_jump: | |
4051 | DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); | |
4052 | /* It doesn't matter what we push for the string here. What | |
4053 | the code at `fail' tests is the value for the pattern. */ | |
4054 | PUSH_FAILURE_POINT (0, 0, -2); | |
4055 | goto unconditional_jump; | |
4056 | ||
4057 | ||
4058 | /* At the end of an alternative, we need to push a dummy failure | |
4059 | point in case we are followed by a pop_failure_jump', because | |
4060 | we don't want the failure point for the alternative to be | |
4061 | popped. For example, matching `(a|ab)*' against `aab' | |
4062 | requires that we match the `ab' alternative. */ | |
4063 | case push_dummy_failure: | |
4064 | DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); | |
4065 | /* See comments just above at `dummy_failure_jump' about the | |
4066 | two zeroes. */ | |
4067 | PUSH_FAILURE_POINT (0, 0, -2); | |
4068 | break; | |
4069 | ||
4070 | /* Have to succeed matching what follows at least n times. | |
4071 | After that, handle like `on_failure_jump'. */ | |
4072 | case succeed_n: | |
4073 | EXTRACT_NUMBER (mcnt, p + 2); | |
4074 | DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); | |
4075 | ||
4076 | assert (mcnt >= 0); | |
4077 | /* Originally, this is how many times we HAVE to succeed. */ | |
4078 | if (mcnt > 0) | |
4079 | { | |
4080 | mcnt--; | |
4081 | p += 2; | |
4082 | STORE_NUMBER_AND_INCR (p, mcnt); | |
4083 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt); | |
4084 | } | |
4085 | else if (mcnt == 0) | |
4086 | { | |
4087 | DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2); | |
4088 | p[2] = (unsigned char) no_op; | |
4089 | p[3] = (unsigned char) no_op; | |
4090 | goto on_failure; | |
4091 | } | |
4092 | break; | |
4093 | ||
4094 | case jump_n: | |
4095 | EXTRACT_NUMBER (mcnt, p + 2); | |
4096 | DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); | |
4097 | ||
4098 | /* Originally, this is how many times we CAN jump. */ | |
4099 | if (mcnt) | |
4100 | { | |
4101 | mcnt--; | |
4102 | STORE_NUMBER (p + 2, mcnt); | |
4103 | goto unconditional_jump; | |
4104 | } | |
4105 | /* If don't have to jump any more, skip over the rest of command. */ | |
4106 | else | |
4107 | p += 4; | |
4108 | break; | |
4109 | ||
4110 | case set_number_at: | |
4111 | { | |
4112 | DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); | |
4113 | ||
4114 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4115 | p1 = p + mcnt; | |
4116 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4117 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); | |
4118 | STORE_NUMBER (p1, mcnt); | |
4119 | break; | |
4120 | } | |
4121 | ||
4122 | case wordbound: | |
4123 | DEBUG_PRINT1 ("EXECUTING wordbound.\n"); | |
4124 | if (AT_WORD_BOUNDARY (d)) | |
4125 | break; | |
4126 | goto fail; | |
4127 | ||
4128 | case notwordbound: | |
4129 | DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); | |
4130 | if (AT_WORD_BOUNDARY (d)) | |
4131 | goto fail; | |
4132 | break; | |
4133 | ||
4134 | case wordbeg: | |
4135 | DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); | |
4136 | if (LETTER_P (d) && (AT_STRINGS_BEG () || !LETTER_P (d - 1))) | |
4137 | break; | |
4138 | goto fail; | |
4139 | ||
4140 | case wordend: | |
4141 | DEBUG_PRINT1 ("EXECUTING wordend.\n"); | |
4142 | if (!AT_STRINGS_BEG () && LETTER_P (d - 1) | |
4143 | && (!LETTER_P (d) || AT_STRINGS_END ())) | |
4144 | break; | |
4145 | goto fail; | |
4146 | ||
4147 | #ifdef emacs | |
4148 | #ifdef emacs19 | |
4149 | case before_dot: | |
4150 | DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
4151 | if (PTR_CHAR_POS ((unsigned char *) d) >= point) | |
4152 | goto fail; | |
4153 | break; | |
4154 | ||
4155 | case at_dot: | |
4156 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
4157 | if (PTR_CHAR_POS ((unsigned char *) d) != point) | |
4158 | goto fail; | |
4159 | break; | |
4160 | ||
4161 | case after_dot: | |
4162 | DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
4163 | if (PTR_CHAR_POS ((unsigned char *) d) <= point) | |
4164 | goto fail; | |
4165 | break; | |
4166 | #else /* not emacs19 */ | |
4167 | case at_dot: | |
4168 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
4169 | if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point) | |
4170 | goto fail; | |
4171 | break; | |
4172 | #endif /* not emacs19 */ | |
4173 | ||
4174 | case syntaxspec: | |
4175 | DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); | |
4176 | mcnt = *p++; | |
4177 | goto matchsyntax; | |
4178 | ||
4179 | case wordchar: | |
4180 | DEBUG_PRINT1 ("EXECUTING wordchar.\n"); | |
4181 | mcnt = (int) Sword; | |
4182 | matchsyntax: | |
4183 | PREFETCH (); | |
4184 | if (SYNTAX (*d++) != (enum syntaxcode) mcnt) goto fail; | |
4185 | SET_REGS_MATCHED (); | |
4186 | break; | |
4187 | ||
4188 | case notsyntaxspec: | |
4189 | DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); | |
4190 | mcnt = *p++; | |
4191 | goto matchnotsyntax; | |
4192 | ||
4193 | case notwordchar: | |
4194 | DEBUG_PRINT1 ("EXECUTING notwordchar.\n"); | |
4195 | mcnt = (int) Sword; | |
4196 | matchnotsyntax: /* We goto here from notsyntaxspec. */ | |
4197 | PREFETCH (); | |
4198 | if (SYNTAX (*d++) == (enum syntaxcode) mcnt) goto fail; | |
4199 | SET_REGS_MATCHED (); | |
4200 | break; | |
4201 | ||
4202 | #else /* not emacs */ | |
4203 | case wordchar: | |
4204 | DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); | |
4205 | PREFETCH (); | |
4206 | if (!LETTER_P (d)) | |
4207 | goto fail; | |
4208 | SET_REGS_MATCHED (); | |
4209 | break; | |
4210 | ||
4211 | case notwordchar: | |
4212 | DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); | |
4213 | PREFETCH (); | |
4214 | if (LETTER_P (d)) | |
4215 | goto fail; | |
4216 | SET_REGS_MATCHED (); | |
4217 | break; | |
4218 | #endif /* not emacs */ | |
4219 | ||
4220 | default: | |
4221 | abort (); | |
4222 | } | |
4223 | continue; /* Successfully executed one pattern command; keep going. */ | |
4224 | ||
4225 | ||
4226 | /* We goto here if a matching operation fails. */ | |
4227 | fail: | |
4228 | if (!FAIL_STACK_EMPTY ()) | |
4229 | { /* A restart point is known. Restore to that state. */ | |
4230 | DEBUG_PRINT1 ("\nFAIL:\n"); | |
4231 | POP_FAILURE_POINT (d, p, | |
4232 | lowest_active_reg, highest_active_reg, | |
4233 | regstart, regend, reg_info); | |
4234 | ||
4235 | /* If this failure point is a dummy, try the next one. */ | |
4236 | if (!p) | |
4237 | goto fail; | |
4238 | ||
4239 | /* If we failed to the end of the pattern, don't examine *p. */ | |
4240 | assert (p <= pend); | |
4241 | if (p < pend) | |
4242 | { | |
4243 | boolean is_a_jump_n = false; | |
4244 | ||
4245 | /* If failed to a backwards jump that's part of a repetition | |
4246 | loop, need to pop this failure point and use the next one. */ | |
4247 | switch ((re_opcode_t) *p) | |
4248 | { | |
4249 | case jump_n: | |
4250 | is_a_jump_n = true; | |
4251 | case maybe_pop_jump: | |
4252 | case pop_failure_jump: | |
4253 | case jump: | |
4254 | p1 = p + 1; | |
4255 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4256 | p1 += mcnt; | |
4257 | ||
4258 | if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) | |
4259 | || (!is_a_jump_n | |
4260 | && (re_opcode_t) *p1 == on_failure_jump)) | |
4261 | goto fail; | |
4262 | break; | |
4263 | default: | |
4264 | /* do nothing */ ; | |
4265 | } | |
4266 | } | |
4267 | ||
4268 | if (d >= string1 && d <= end1) | |
4269 | dend = end_match_1; | |
4270 | } | |
4271 | else | |
4272 | break; /* Matching at this starting point really fails. */ | |
4273 | } /* for (;;) */ | |
4274 | ||
4275 | if (best_regs_set) | |
4276 | goto restore_best_regs; | |
4277 | ||
4278 | FREE_VARIABLES (); | |
4279 | ||
4280 | return -1; /* Failure to match. */ | |
4281 | } /* re_match_2 */ | |
4282 | \f | |
4283 | /* Subroutine definitions for re_match_2. */ | |
4284 | ||
4285 | ||
4286 | /* We are passed P pointing to a register number after a start_memory. | |
4287 | ||
4288 | Return true if the pattern up to the corresponding stop_memory can | |
4289 | match the empty string, and false otherwise. | |
4290 | ||
4291 | If we find the matching stop_memory, sets P to point to one past its number. | |
4292 | Otherwise, sets P to an undefined byte less than or equal to END. | |
4293 | ||
4294 | We don't handle duplicates properly (yet). */ | |
4295 | ||
4296 | static boolean | |
4297 | group_match_null_string_p (p, end, reg_info) | |
4298 | unsigned char **p, *end; | |
4299 | register_info_type *reg_info; | |
4300 | { | |
4301 | int mcnt; | |
4302 | /* Point to after the args to the start_memory. */ | |
4303 | unsigned char *p1 = *p + 2; | |
4304 | ||
4305 | while (p1 < end) | |
4306 | { | |
4307 | /* Skip over opcodes that can match nothing, and return true or | |
4308 | false, as appropriate, when we get to one that can't, or to the | |
4309 | matching stop_memory. */ | |
4310 | ||
4311 | switch ((re_opcode_t) *p1) | |
4312 | { | |
4313 | /* Could be either a loop or a series of alternatives. */ | |
4314 | case on_failure_jump: | |
4315 | p1++; | |
4316 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4317 | ||
4318 | /* If the next operation is not a jump backwards in the | |
4319 | pattern. */ | |
4320 | ||
4321 | if (mcnt >= 0) | |
4322 | { | |
4323 | /* Go through the on_failure_jumps of the alternatives, | |
4324 | seeing if any of the alternatives cannot match nothing. | |
4325 | The last alternative starts with only a jump, | |
4326 | whereas the rest start with on_failure_jump and end | |
4327 | with a jump, e.g., here is the pattern for `a|b|c': | |
4328 | ||
4329 | /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 | |
4330 | /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 | |
4331 | /exactn/1/c | |
4332 | ||
4333 | So, we have to first go through the first (n-1) | |
4334 | alternatives and then deal with the last one separately. */ | |
4335 | ||
4336 | ||
4337 | /* Deal with the first (n-1) alternatives, which start | |
4338 | with an on_failure_jump (see above) that jumps to right | |
4339 | past a jump_past_alt. */ | |
4340 | ||
4341 | while ((re_opcode_t) p1[mcnt-3] == jump_past_alt) | |
4342 | { | |
4343 | /* `mcnt' holds how many bytes long the alternative | |
4344 | is, including the ending `jump_past_alt' and | |
4345 | its number. */ | |
4346 | ||
4347 | if (!alt_match_null_string_p (p1, p1 + mcnt - 3, | |
4348 | reg_info)) | |
4349 | return false; | |
4350 | ||
4351 | /* Move to right after this alternative, including the | |
4352 | jump_past_alt. */ | |
4353 | p1 += mcnt; | |
4354 | ||
4355 | /* Break if it's the beginning of an n-th alternative | |
4356 | that doesn't begin with an on_failure_jump. */ | |
4357 | if ((re_opcode_t) *p1 != on_failure_jump) | |
4358 | break; | |
4359 | ||
4360 | /* Still have to check that it's not an n-th | |
4361 | alternative that starts with an on_failure_jump. */ | |
4362 | p1++; | |
4363 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4364 | if ((re_opcode_t) p1[mcnt-3] != jump_past_alt) | |
4365 | { | |
4366 | /* Get to the beginning of the n-th alternative. */ | |
4367 | p1 -= 3; | |
4368 | break; | |
4369 | } | |
4370 | } | |
4371 | ||
4372 | /* Deal with the last alternative: go back and get number | |
4373 | of the `jump_past_alt' just before it. `mcnt' contains | |
4374 | the length of the alternative. */ | |
4375 | EXTRACT_NUMBER (mcnt, p1 - 2); | |
4376 | ||
4377 | if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info)) | |
4378 | return false; | |
4379 | ||
4380 | p1 += mcnt; /* Get past the n-th alternative. */ | |
4381 | } /* if mcnt > 0 */ | |
4382 | break; | |
4383 | ||
4384 | ||
4385 | case stop_memory: | |
4386 | assert (p1[1] == **p); | |
4387 | *p = p1 + 2; | |
4388 | return true; | |
4389 | ||
4390 | ||
4391 | default: | |
4392 | if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
4393 | return false; | |
4394 | } | |
4395 | } /* while p1 < end */ | |
4396 | ||
4397 | return false; | |
4398 | } /* group_match_null_string_p */ | |
4399 | ||
4400 | ||
4401 | /* Similar to group_match_null_string_p, but doesn't deal with alternatives: | |
4402 | It expects P to be the first byte of a single alternative and END one | |
4403 | byte past the last. The alternative can contain groups. */ | |
4404 | ||
4405 | static boolean | |
4406 | alt_match_null_string_p (p, end, reg_info) | |
4407 | unsigned char *p, *end; | |
4408 | register_info_type *reg_info; | |
4409 | { | |
4410 | int mcnt; | |
4411 | unsigned char *p1 = p; | |
4412 | ||
4413 | while (p1 < end) | |
4414 | { | |
4415 | /* Skip over opcodes that can match nothing, and break when we get | |
4416 | to one that can't. */ | |
4417 | ||
4418 | switch ((re_opcode_t) *p1) | |
4419 | { | |
4420 | /* It's a loop. */ | |
4421 | case on_failure_jump: | |
4422 | p1++; | |
4423 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4424 | p1 += mcnt; | |
4425 | break; | |
4426 | ||
4427 | default: | |
4428 | if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
4429 | return false; | |
4430 | } | |
4431 | } /* while p1 < end */ | |
4432 | ||
4433 | return true; | |
4434 | } /* alt_match_null_string_p */ | |
4435 | ||
4436 | ||
4437 | /* Deals with the ops common to group_match_null_string_p and | |
4438 | alt_match_null_string_p. | |
4439 | ||
4440 | Sets P to one after the op and its arguments, if any. */ | |
4441 | ||
4442 | static boolean | |
4443 | common_op_match_null_string_p (p, end, reg_info) | |
4444 | unsigned char **p, *end; | |
4445 | register_info_type *reg_info; | |
4446 | { | |
4447 | int mcnt; | |
4448 | boolean ret; | |
4449 | int reg_no; | |
4450 | unsigned char *p1 = *p; | |
4451 | ||
4452 | switch ((re_opcode_t) *p1++) | |
4453 | { | |
4454 | case no_op: | |
4455 | case begline: | |
4456 | case endline: | |
4457 | case begbuf: | |
4458 | case endbuf: | |
4459 | case wordbeg: | |
4460 | case wordend: | |
4461 | case wordbound: | |
4462 | case notwordbound: | |
4463 | #ifdef emacs | |
4464 | case before_dot: | |
4465 | case at_dot: | |
4466 | case after_dot: | |
4467 | #endif | |
4468 | break; | |
4469 | ||
4470 | case start_memory: | |
4471 | reg_no = *p1; | |
4472 | assert (reg_no > 0 && reg_no <= MAX_REGNUM); | |
4473 | ret = group_match_null_string_p (&p1, end, reg_info); | |
4474 | ||
4475 | /* Have to set this here in case we're checking a group which | |
4476 | contains a group and a back reference to it. */ | |
4477 | ||
4478 | if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) | |
4479 | REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; | |
4480 | ||
4481 | if (!ret) | |
4482 | return false; | |
4483 | break; | |
4484 | ||
4485 | /* If this is an optimized succeed_n for zero times, make the jump. */ | |
4486 | case jump: | |
4487 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4488 | if (mcnt >= 0) | |
4489 | p1 += mcnt; | |
4490 | else | |
4491 | return false; | |
4492 | break; | |
4493 | ||
4494 | case succeed_n: | |
4495 | /* Get to the number of times to succeed. */ | |
4496 | p1 += 2; | |
4497 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4498 | ||
4499 | if (mcnt == 0) | |
4500 | { | |
4501 | p1 -= 4; | |
4502 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
4503 | p1 += mcnt; | |
4504 | } | |
4505 | else | |
4506 | return false; | |
4507 | break; | |
4508 | ||
4509 | case duplicate: | |
4510 | if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) | |
4511 | return false; | |
4512 | break; | |
4513 | ||
4514 | case set_number_at: | |
4515 | p1 += 4; | |
4516 | ||
4517 | default: | |
4518 | /* All other opcodes mean we cannot match the empty string. */ | |
4519 | return false; | |
4520 | } | |
4521 | ||
4522 | *p = p1; | |
4523 | return true; | |
4524 | } /* common_op_match_null_string_p */ | |
4525 | ||
4526 | ||
4527 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN | |
4528 | bytes; nonzero otherwise. */ | |
4529 | ||
4530 | static int | |
4531 | bcmp_translate (s1, s2, len, translate) | |
4532 | unsigned char *s1, *s2; | |
4533 | register int len; | |
4534 | char *translate; | |
4535 | { | |
4536 | register unsigned char *p1 = s1, *p2 = s2; | |
4537 | while (len) | |
4538 | { | |
4539 | if (translate[*p1++] != translate[*p2++]) return 1; | |
4540 | len--; | |
4541 | } | |
4542 | return 0; | |
4543 | } | |
4544 | \f | |
4545 | /* Entry points for GNU code. */ | |
4546 | ||
4547 | /* re_compile_pattern is the GNU regular expression compiler: it | |
4548 | compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
4549 | Returns 0 if the pattern was valid, otherwise an error string. | |
4550 | ||
4551 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields | |
4552 | are set in BUFP on entry. | |
4553 | ||
4554 | We call regex_compile to do the actual compilation. */ | |
4555 | ||
4556 | const char * | |
4557 | re_compile_pattern (pattern, length, bufp) | |
4558 | const char *pattern; | |
4559 | int length; | |
4560 | struct re_pattern_buffer *bufp; | |
4561 | { | |
4562 | reg_errcode_t ret; | |
4563 | ||
4564 | /* GNU code is written to assume at least RE_NREGS registers will be set | |
4565 | (and at least one extra will be -1). */ | |
4566 | bufp->regs_allocated = REGS_UNALLOCATED; | |
4567 | ||
4568 | /* And GNU code determines whether or not to get register information | |
4569 | by passing null for the REGS argument to re_match, etc., not by | |
4570 | setting no_sub. */ | |
4571 | bufp->no_sub = 0; | |
4572 | ||
4573 | /* Match anchors at newline. */ | |
4574 | bufp->newline_anchor = 1; | |
4575 | ||
4576 | ret = regex_compile (pattern, length, re_syntax_options, bufp); | |
4577 | ||
4578 | return re_error_msg[(int) ret]; | |
4579 | } | |
4580 | \f | |
4581 | /* Entry points compatible with 4.2 BSD regex library. We don't define | |
4582 | them if this is an Emacs or POSIX compilation. */ | |
4583 | ||
4584 | #if !defined (emacs) && !defined (_POSIX_SOURCE) | |
4585 | ||
4586 | /* BSD has one and only one pattern buffer. */ | |
4587 | static struct re_pattern_buffer re_comp_buf; | |
4588 | ||
4589 | char * | |
4590 | re_comp (s) | |
4591 | const char *s; | |
4592 | { | |
4593 | reg_errcode_t ret; | |
4594 | ||
4595 | if (!s) | |
4596 | { | |
4597 | if (!re_comp_buf.buffer) | |
4598 | return "No previous regular expression"; | |
4599 | return 0; | |
4600 | } | |
4601 | ||
4602 | if (!re_comp_buf.buffer) | |
4603 | { | |
4604 | re_comp_buf.buffer = (unsigned char *) malloc (200); | |
4605 | if (re_comp_buf.buffer == NULL) | |
4606 | return "Memory exhausted"; | |
4607 | re_comp_buf.allocated = 200; | |
4608 | ||
4609 | re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
4610 | if (re_comp_buf.fastmap == NULL) | |
4611 | return "Memory exhausted"; | |
4612 | } | |
4613 | ||
4614 | /* Since `re_exec' always passes NULL for the `regs' argument, we | |
4615 | don't need to initialize the pattern buffer fields which affect it. */ | |
4616 | ||
4617 | /* Match anchors at newlines. */ | |
4618 | re_comp_buf.newline_anchor = 1; | |
4619 | ||
4620 | ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); | |
4621 | ||
4622 | /* Yes, we're discarding `const' here. */ | |
4623 | return (char *) re_error_msg[(int) ret]; | |
4624 | } | |
4625 | ||
4626 | ||
4627 | int | |
4628 | re_exec (s) | |
4629 | const char *s; | |
4630 | { | |
4631 | const int len = strlen (s); | |
4632 | return | |
4633 | 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
4634 | } | |
4635 | #endif /* not emacs and not _POSIX_SOURCE */ | |
4636 | \f | |
4637 | /* POSIX.2 functions. Don't define these for Emacs. */ | |
4638 | ||
4639 | #ifndef emacs | |
4640 | ||
4641 | /* regcomp takes a regular expression as a string and compiles it. | |
4642 | ||
4643 | PREG is a regex_t *. We do not expect any fields to be initialized, | |
4644 | since POSIX says we shouldn't. Thus, we set | |
4645 | ||
4646 | `buffer' to the compiled pattern; | |
4647 | `used' to the length of the compiled pattern; | |
4648 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
4649 | REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
4650 | RE_SYNTAX_POSIX_BASIC; | |
4651 | `newline_anchor' to REG_NEWLINE being set in CFLAGS; | |
4652 | `fastmap' and `fastmap_accurate' to zero; | |
4653 | `re_nsub' to the number of subexpressions in PATTERN. | |
4654 | ||
4655 | PATTERN is the address of the pattern string. | |
4656 | ||
4657 | CFLAGS is a series of bits which affect compilation. | |
4658 | ||
4659 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
4660 | use POSIX basic syntax. | |
4661 | ||
4662 | If REG_NEWLINE is set, then . and [^...] don't match newline. | |
4663 | Also, regexec will try a match beginning after every newline. | |
4664 | ||
4665 | If REG_ICASE is set, then we considers upper- and lowercase | |
4666 | versions of letters to be equivalent when matching. | |
4667 | ||
4668 | If REG_NOSUB is set, then when PREG is passed to regexec, that | |
4669 | routine will report only success or failure, and nothing about the | |
4670 | registers. | |
4671 | ||
4672 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for | |
4673 | the return codes and their meanings.) */ | |
4674 | ||
4675 | int | |
4676 | regcomp (preg, pattern, cflags) | |
4677 | regex_t *preg; | |
4678 | const char *pattern; | |
4679 | int cflags; | |
4680 | { | |
4681 | reg_errcode_t ret; | |
4682 | unsigned syntax | |
4683 | = cflags & REG_EXTENDED ? RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; | |
4684 | ||
4685 | /* regex_compile will allocate the space for the compiled pattern. */ | |
4686 | preg->buffer = 0; | |
4687 | ||
4688 | /* Don't bother to use a fastmap when searching. This simplifies the | |
4689 | REG_NEWLINE case: if we used a fastmap, we'd have to put all the | |
4690 | characters after newlines into the fastmap. This way, we just try | |
4691 | every character. */ | |
4692 | preg->fastmap = 0; | |
4693 | ||
4694 | if (cflags & REG_ICASE) | |
4695 | { | |
4696 | unsigned i; | |
4697 | ||
4698 | preg->translate = (char *) malloc (CHAR_SET_SIZE); | |
4699 | if (preg->translate == NULL) | |
4700 | return (int) REG_ESPACE; | |
4701 | ||
4702 | /* Map uppercase characters to corresponding lowercase ones. */ | |
4703 | for (i = 0; i < CHAR_SET_SIZE; i++) | |
4704 | preg->translate[i] = isupper (i) ? tolower (i) : i; | |
4705 | } | |
4706 | else | |
4707 | preg->translate = NULL; | |
4708 | ||
4709 | /* If REG_NEWLINE is set, newlines are treated differently. */ | |
4710 | if (cflags & REG_NEWLINE) | |
4711 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
4712 | syntax &= ~RE_DOT_NEWLINE; | |
4713 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
4714 | /* It also changes the matching behavior. */ | |
4715 | preg->newline_anchor = 1; | |
4716 | } | |
4717 | else | |
4718 | preg->newline_anchor = 0; | |
4719 | ||
4720 | preg->no_sub = !!(cflags & REG_NOSUB); | |
4721 | ||
4722 | /* POSIX says a null character in the pattern terminates it, so we | |
4723 | can use strlen here in compiling the pattern. */ | |
4724 | ret = regex_compile (pattern, strlen (pattern), syntax, preg); | |
4725 | ||
4726 | /* POSIX doesn't distinguish between an unmatched open-group and an | |
4727 | unmatched close-group: both are REG_EPAREN. */ | |
4728 | if (ret == REG_ERPAREN) ret = REG_EPAREN; | |
4729 | ||
4730 | return (int) ret; | |
4731 | } | |
4732 | ||
4733 | ||
4734 | /* regexec searches for a given pattern, specified by PREG, in the | |
4735 | string STRING. | |
4736 | ||
4737 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to | |
4738 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at | |
4739 | least NMATCH elements, and we set them to the offsets of the | |
4740 | corresponding matched substrings. | |
4741 | ||
4742 | EFLAGS specifies `execution flags' which affect matching: if | |
4743 | REG_NOTBOL is set, then ^ does not match at the beginning of the | |
4744 | string; if REG_NOTEOL is set, then $ does not match at the end. | |
4745 | ||
4746 | We return 0 if we find a match and REG_NOMATCH if not. */ | |
4747 | ||
4748 | int | |
4749 | regexec (preg, string, nmatch, pmatch, eflags) | |
4750 | const regex_t *preg; | |
4751 | const char *string; | |
4752 | size_t nmatch; | |
4753 | regmatch_t pmatch[]; | |
4754 | int eflags; | |
4755 | { | |
4756 | int ret; | |
4757 | struct re_registers regs; | |
4758 | regex_t private_preg; | |
4759 | int len = strlen (string); | |
4760 | boolean want_reg_info = !preg->no_sub && nmatch > 0; | |
4761 | ||
4762 | private_preg = *preg; | |
4763 | ||
4764 | private_preg.not_bol = !!(eflags & REG_NOTBOL); | |
4765 | private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
4766 | ||
4767 | /* The user has told us exactly how many registers to return | |
4768 | information about, via `nmatch'. We have to pass that on to the | |
4769 | matching routines. */ | |
4770 | private_preg.regs_allocated = REGS_FIXED; | |
4771 | ||
4772 | if (want_reg_info) | |
4773 | { | |
4774 | regs.num_regs = nmatch; | |
4775 | regs.start = TALLOC (nmatch, regoff_t); | |
4776 | regs.end = TALLOC (nmatch, regoff_t); | |
4777 | if (regs.start == NULL || regs.end == NULL) | |
4778 | return (int) REG_NOMATCH; | |
4779 | } | |
4780 | ||
4781 | /* Perform the searching operation. */ | |
4782 | ret = re_search (&private_preg, string, len, | |
4783 | /* start: */ 0, /* range: */ len, | |
4784 | want_reg_info ? ®s : (struct re_registers *) 0); | |
4785 | ||
4786 | /* Copy the register information to the POSIX structure. */ | |
4787 | if (want_reg_info) | |
4788 | { | |
4789 | if (ret >= 0) | |
4790 | { | |
4791 | unsigned r; | |
4792 | ||
4793 | for (r = 0; r < nmatch; r++) | |
4794 | { | |
4795 | pmatch[r].rm_so = regs.start[r]; | |
4796 | pmatch[r].rm_eo = regs.end[r]; | |
4797 | } | |
4798 | } | |
4799 | ||
4800 | /* If we needed the temporary register info, free the space now. */ | |
4801 | free (regs.start); | |
4802 | free (regs.end); | |
4803 | } | |
4804 | ||
4805 | /* We want zero return to mean success, unlike `re_search'. */ | |
4806 | return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
4807 | } | |
4808 | ||
4809 | ||
4810 | /* Returns a message corresponding to an error code, ERRCODE, returned | |
4811 | from either regcomp or regexec. */ | |
4812 | ||
4813 | size_t | |
4814 | regerror (errcode, preg, errbuf, errbuf_size) | |
4815 | int errcode; | |
4816 | const regex_t *preg; | |
4817 | char *errbuf; | |
4818 | size_t errbuf_size; | |
4819 | { | |
4820 | const char *msg | |
4821 | = re_error_msg[errcode] == NULL ? "Success" : re_error_msg[errcode]; | |
4822 | size_t msg_size = strlen (msg) + 1; /* Includes the null. */ | |
4823 | ||
4824 | if (errbuf_size != 0) | |
4825 | { | |
4826 | if (msg_size > errbuf_size) | |
4827 | { | |
4828 | strncpy (errbuf, msg, errbuf_size - 1); | |
4829 | errbuf[errbuf_size - 1] = 0; | |
4830 | } | |
4831 | else | |
4832 | strcpy (errbuf, msg); | |
4833 | } | |
4834 | ||
4835 | return msg_size; | |
4836 | } | |
4837 | ||
4838 | ||
4839 | /* Free dynamically allocated space used by PREG. */ | |
4840 | ||
4841 | void | |
4842 | regfree (preg) | |
4843 | regex_t *preg; | |
4844 | { | |
4845 | if (preg->buffer != NULL) | |
4846 | free (preg->buffer); | |
4847 | preg->buffer = NULL; | |
4848 | ||
4849 | preg->allocated = 0; | |
4850 | preg->used = 0; | |
4851 | ||
4852 | if (preg->fastmap != NULL) | |
4853 | free (preg->fastmap); | |
4854 | preg->fastmap = NULL; | |
4855 | preg->fastmap_accurate = 0; | |
4856 | ||
4857 | if (preg->translate != NULL) | |
4858 | free (preg->translate); | |
4859 | preg->translate = NULL; | |
4860 | } | |
4861 | ||
4862 | #endif /* not emacs */ | |
4863 | \f | |
4864 | /* | |
4865 | Local variables: | |
4866 | make-backup-files: t | |
4867 | version-control: t | |
4868 | trim-versions-without-asking: nil | |
4869 | End: | |
4870 | */ |