1 | /* |
2 | * re_*exec and friends - match REs |
3 | * |
4 | * Copyright (c) 1998, 1999 Henry Spencer. All rights reserved. |
5 | * |
6 | * Development of this software was funded, in part, by Cray Research Inc., |
7 | * UUNET Communications Services Inc., Sun Microsystems Inc., and Scriptics |
8 | * Corporation, none of whom are responsible for the results. The author |
9 | * thanks all of them. |
10 | * |
11 | * Redistribution and use in source and binary forms -- with or without |
12 | * modification -- are permitted for any purpose, provided that |
13 | * redistributions in source form retain this entire copyright notice and |
14 | * indicate the origin and nature of any modifications. |
15 | * |
16 | * I'd appreciate being given credit for this package in the documentation |
17 | * of software which uses it, but that is not a requirement. |
18 | * |
19 | * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, |
20 | * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY |
21 | * AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL |
22 | * HENRY SPENCER BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
23 | * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
24 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
25 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
26 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR |
27 | * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF |
28 | * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
29 | * |
30 | * src/backend/regex/regexec.c |
31 | * |
32 | */ |
33 | |
34 | #include "regex/regguts.h" |
35 | |
36 | |
37 | |
38 | /* lazy-DFA representation */ |
39 | struct arcp |
40 | { /* "pointer" to an outarc */ |
41 | struct sset *ss; |
42 | color co; |
43 | }; |
44 | |
45 | struct sset |
46 | { /* state set */ |
47 | unsigned *states; /* pointer to bitvector */ |
48 | unsigned hash; /* hash of bitvector */ |
49 | #define HASH(bv, nw) (((nw) == 1) ? *(bv) : hash(bv, nw)) |
50 | #define HIT(h,bv,ss,nw) ((ss)->hash == (h) && ((nw) == 1 || \ |
51 | memcmp(VS(bv), VS((ss)->states), (nw)*sizeof(unsigned)) == 0)) |
52 | int flags; |
53 | #define STARTER 01 /* the initial state set */ |
54 | #define POSTSTATE 02 /* includes the goal state */ |
55 | #define LOCKED 04 /* locked in cache */ |
56 | #define NOPROGRESS 010 /* zero-progress state set */ |
57 | struct arcp ins; /* chain of inarcs pointing here */ |
58 | chr *lastseen; /* last entered on arrival here */ |
59 | struct sset **outs; /* outarc vector indexed by color */ |
60 | struct arcp *inchain; /* chain-pointer vector for outarcs */ |
61 | }; |
62 | |
63 | struct dfa |
64 | { |
65 | int nssets; /* size of cache */ |
66 | int nssused; /* how many entries occupied yet */ |
67 | int nstates; /* number of states */ |
68 | int ncolors; /* length of outarc and inchain vectors */ |
69 | int wordsper; /* length of state-set bitvectors */ |
70 | struct sset *ssets; /* state-set cache */ |
71 | unsigned *statesarea; /* bitvector storage */ |
72 | unsigned *work; /* pointer to work area within statesarea */ |
73 | struct sset **outsarea; /* outarc-vector storage */ |
74 | struct arcp *incarea; /* inchain storage */ |
75 | struct cnfa *cnfa; |
76 | struct colormap *cm; |
77 | chr *lastpost; /* location of last cache-flushed success */ |
78 | chr *lastnopr; /* location of last cache-flushed NOPROGRESS */ |
79 | struct sset *search; /* replacement-search-pointer memory */ |
80 | int cptsmalloced; /* were the areas individually malloced? */ |
81 | char *mallocarea; /* self, or master malloced area, or NULL */ |
82 | }; |
83 | |
84 | #define WORK 1 /* number of work bitvectors needed */ |
85 | |
86 | /* setup for non-malloc allocation for small cases */ |
87 | #define FEWSTATES 20 /* must be less than UBITS */ |
88 | #define FEWCOLORS 15 |
89 | struct smalldfa |
90 | { |
91 | struct dfa dfa; |
92 | struct sset ssets[FEWSTATES * 2]; |
93 | unsigned statesarea[FEWSTATES * 2 + WORK]; |
94 | struct sset *outsarea[FEWSTATES * 2 * FEWCOLORS]; |
95 | struct arcp incarea[FEWSTATES * 2 * FEWCOLORS]; |
96 | }; |
97 | |
98 | #define DOMALLOC ((struct smalldfa *)NULL) /* force malloc */ |
99 | |
100 | |
101 | |
102 | /* internal variables, bundled for easy passing around */ |
103 | struct vars |
104 | { |
105 | regex_t *re; |
106 | struct guts *g; |
107 | int eflags; /* copies of arguments */ |
108 | size_t nmatch; |
109 | regmatch_t *pmatch; |
110 | rm_detail_t *details; |
111 | chr *start; /* start of string */ |
112 | chr *search_start; /* search start of string */ |
113 | chr *stop; /* just past end of string */ |
114 | int err; /* error code if any (0 none) */ |
115 | struct dfa **subdfas; /* per-tree-subre DFAs */ |
116 | struct dfa **ladfas; /* per-lacon-subre DFAs */ |
117 | struct sset **lblastcss; /* per-lacon-subre lookbehind restart data */ |
118 | chr **lblastcp; /* per-lacon-subre lookbehind restart data */ |
119 | struct smalldfa dfa1; |
120 | struct smalldfa dfa2; |
121 | }; |
122 | |
123 | #define VISERR(vv) ((vv)->err != 0) /* have we seen an error yet? */ |
124 | #define ISERR() VISERR(v) |
125 | #define VERR(vv,e) ((vv)->err = ((vv)->err ? (vv)->err : (e))) |
126 | #define ERR(e) VERR(v, e) /* record an error */ |
127 | #define NOERR() {if (ISERR()) return v->err;} /* if error seen, return it */ |
128 | #define OFF(p) ((p) - v->start) |
129 | #define LOFF(p) ((long)OFF(p)) |
130 | |
131 | |
132 | |
133 | /* |
134 | * forward declarations |
135 | */ |
136 | /* === regexec.c === */ |
137 | static struct dfa *getsubdfa(struct vars *, struct subre *); |
138 | static struct dfa *getladfa(struct vars *, int); |
139 | static int find(struct vars *, struct cnfa *, struct colormap *); |
140 | static int cfind(struct vars *, struct cnfa *, struct colormap *); |
141 | static int cfindloop(struct vars *, struct cnfa *, struct colormap *, struct dfa *, struct dfa *, chr **); |
142 | static void zapallsubs(regmatch_t *, size_t); |
143 | static void zaptreesubs(struct vars *, struct subre *); |
144 | static void subset(struct vars *, struct subre *, chr *, chr *); |
145 | static int cdissect(struct vars *, struct subre *, chr *, chr *); |
146 | static int ccondissect(struct vars *, struct subre *, chr *, chr *); |
147 | static int crevcondissect(struct vars *, struct subre *, chr *, chr *); |
148 | static int cbrdissect(struct vars *, struct subre *, chr *, chr *); |
149 | static int caltdissect(struct vars *, struct subre *, chr *, chr *); |
150 | static int citerdissect(struct vars *, struct subre *, chr *, chr *); |
151 | static int creviterdissect(struct vars *, struct subre *, chr *, chr *); |
152 | |
153 | /* === rege_dfa.c === */ |
154 | static chr *longest(struct vars *, struct dfa *, chr *, chr *, int *); |
155 | static chr *shortest(struct vars *, struct dfa *, chr *, chr *, chr *, chr **, int *); |
156 | static int matchuntil(struct vars *, struct dfa *, chr *, struct sset **, chr **); |
157 | static chr *lastcold(struct vars *, struct dfa *); |
158 | static struct dfa *newdfa(struct vars *, struct cnfa *, struct colormap *, struct smalldfa *); |
159 | static void freedfa(struct dfa *); |
160 | static unsigned hash(unsigned *, int); |
161 | static struct sset *initialize(struct vars *, struct dfa *, chr *); |
162 | static struct sset *miss(struct vars *, struct dfa *, struct sset *, color, chr *, chr *); |
163 | static int lacon(struct vars *, struct cnfa *, chr *, color); |
164 | static struct sset *getvacant(struct vars *, struct dfa *, chr *, chr *); |
165 | static struct sset *pickss(struct vars *, struct dfa *, chr *, chr *); |
166 | |
167 | |
168 | /* |
169 | * pg_regexec - match regular expression |
170 | */ |
171 | int |
172 | pg_regexec(regex_t *re, |
173 | const chr *string, |
174 | size_t len, |
175 | size_t search_start, |
176 | rm_detail_t *details, |
177 | size_t nmatch, |
178 | regmatch_t pmatch[], |
179 | int flags) |
180 | { |
181 | struct vars var; |
182 | register struct vars *v = &var; |
183 | int st; |
184 | size_t n; |
185 | size_t i; |
186 | int backref; |
187 | |
188 | #define LOCALMAT 20 |
189 | regmatch_t mat[LOCALMAT]; |
190 | |
191 | #define LOCALDFAS 40 |
192 | struct dfa *subdfas[LOCALDFAS]; |
193 | |
194 | /* sanity checks */ |
195 | if (re == NULL || string == NULL || re->re_magic != REMAGIC) |
196 | return REG_INVARG; |
197 | if (re->re_csize != sizeof(chr)) |
198 | return REG_MIXED; |
199 | |
200 | /* Initialize locale-dependent support */ |
201 | pg_set_regex_collation(re->re_collation); |
202 | |
203 | /* setup */ |
204 | v->re = re; |
205 | v->g = (struct guts *) re->re_guts; |
206 | if ((v->g->cflags & REG_EXPECT) && details == NULL) |
207 | return REG_INVARG; |
208 | if (v->g->info & REG_UIMPOSSIBLE) |
209 | return REG_NOMATCH; |
210 | backref = (v->g->info & REG_UBACKREF) ? 1 : 0; |
211 | v->eflags = flags; |
212 | if (v->g->cflags & REG_NOSUB) |
213 | nmatch = 0; /* override client */ |
214 | v->nmatch = nmatch; |
215 | if (backref) |
216 | { |
217 | /* need work area */ |
218 | if (v->g->nsub + 1 <= LOCALMAT) |
219 | v->pmatch = mat; |
220 | else |
221 | v->pmatch = (regmatch_t *) MALLOC((v->g->nsub + 1) * |
222 | sizeof(regmatch_t)); |
223 | if (v->pmatch == NULL) |
224 | return REG_ESPACE; |
225 | v->nmatch = v->g->nsub + 1; |
226 | } |
227 | else |
228 | v->pmatch = pmatch; |
229 | v->details = details; |
230 | v->start = (chr *) string; |
231 | v->search_start = (chr *) string + search_start; |
232 | v->stop = (chr *) string + len; |
233 | v->err = 0; |
234 | v->subdfas = NULL; |
235 | v->ladfas = NULL; |
236 | v->lblastcss = NULL; |
237 | v->lblastcp = NULL; |
238 | /* below this point, "goto cleanup" will behave sanely */ |
239 | |
240 | assert(v->g->ntree >= 0); |
241 | n = (size_t) v->g->ntree; |
242 | if (n <= LOCALDFAS) |
243 | v->subdfas = subdfas; |
244 | else |
245 | { |
246 | v->subdfas = (struct dfa **) MALLOC(n * sizeof(struct dfa *)); |
247 | if (v->subdfas == NULL) |
248 | { |
249 | st = REG_ESPACE; |
250 | goto cleanup; |
251 | } |
252 | } |
253 | for (i = 0; i < n; i++) |
254 | v->subdfas[i] = NULL; |
255 | |
256 | assert(v->g->nlacons >= 0); |
257 | n = (size_t) v->g->nlacons; |
258 | if (n > 0) |
259 | { |
260 | v->ladfas = (struct dfa **) MALLOC(n * sizeof(struct dfa *)); |
261 | if (v->ladfas == NULL) |
262 | { |
263 | st = REG_ESPACE; |
264 | goto cleanup; |
265 | } |
266 | for (i = 0; i < n; i++) |
267 | v->ladfas[i] = NULL; |
268 | v->lblastcss = (struct sset **) MALLOC(n * sizeof(struct sset *)); |
269 | v->lblastcp = (chr **) MALLOC(n * sizeof(chr *)); |
270 | if (v->lblastcss == NULL || v->lblastcp == NULL) |
271 | { |
272 | st = REG_ESPACE; |
273 | goto cleanup; |
274 | } |
275 | for (i = 0; i < n; i++) |
276 | { |
277 | v->lblastcss[i] = NULL; |
278 | v->lblastcp[i] = NULL; |
279 | } |
280 | } |
281 | |
282 | /* do it */ |
283 | assert(v->g->tree != NULL); |
284 | if (backref) |
285 | st = cfind(v, &v->g->tree->cnfa, &v->g->cmap); |
286 | else |
287 | st = find(v, &v->g->tree->cnfa, &v->g->cmap); |
288 | |
289 | /* copy (portion of) match vector over if necessary */ |
290 | if (st == REG_OKAY && v->pmatch != pmatch && nmatch > 0) |
291 | { |
292 | zapallsubs(pmatch, nmatch); |
293 | n = (nmatch < v->nmatch) ? nmatch : v->nmatch; |
294 | memcpy(VS(pmatch), VS(v->pmatch), n * sizeof(regmatch_t)); |
295 | } |
296 | |
297 | /* clean up */ |
298 | cleanup: |
299 | if (v->pmatch != pmatch && v->pmatch != mat) |
300 | FREE(v->pmatch); |
301 | if (v->subdfas != NULL) |
302 | { |
303 | n = (size_t) v->g->ntree; |
304 | for (i = 0; i < n; i++) |
305 | { |
306 | if (v->subdfas[i] != NULL) |
307 | freedfa(v->subdfas[i]); |
308 | } |
309 | if (v->subdfas != subdfas) |
310 | FREE(v->subdfas); |
311 | } |
312 | if (v->ladfas != NULL) |
313 | { |
314 | n = (size_t) v->g->nlacons; |
315 | for (i = 0; i < n; i++) |
316 | { |
317 | if (v->ladfas[i] != NULL) |
318 | freedfa(v->ladfas[i]); |
319 | } |
320 | FREE(v->ladfas); |
321 | } |
322 | if (v->lblastcss != NULL) |
323 | FREE(v->lblastcss); |
324 | if (v->lblastcp != NULL) |
325 | FREE(v->lblastcp); |
326 | |
327 | return st; |
328 | } |
329 | |
330 | /* |
331 | * getsubdfa - create or re-fetch the DFA for a tree subre node |
332 | * |
333 | * We only need to create the DFA once per overall regex execution. |
334 | * The DFA will be freed by the cleanup step in pg_regexec(). |
335 | */ |
336 | static struct dfa * |
337 | getsubdfa(struct vars *v, |
338 | struct subre *t) |
339 | { |
340 | if (v->subdfas[t->id] == NULL) |
341 | { |
342 | v->subdfas[t->id] = newdfa(v, &t->cnfa, &v->g->cmap, DOMALLOC); |
343 | if (ISERR()) |
344 | return NULL; |
345 | } |
346 | return v->subdfas[t->id]; |
347 | } |
348 | |
349 | /* |
350 | * getladfa - create or re-fetch the DFA for a LACON subre node |
351 | * |
352 | * Same as above, but for LACONs. |
353 | */ |
354 | static struct dfa * |
355 | getladfa(struct vars *v, |
356 | int n) |
357 | { |
358 | assert(n > 0 && n < v->g->nlacons && v->g->lacons != NULL); |
359 | |
360 | if (v->ladfas[n] == NULL) |
361 | { |
362 | struct subre *sub = &v->g->lacons[n]; |
363 | |
364 | v->ladfas[n] = newdfa(v, &sub->cnfa, &v->g->cmap, DOMALLOC); |
365 | if (ISERR()) |
366 | return NULL; |
367 | } |
368 | return v->ladfas[n]; |
369 | } |
370 | |
371 | /* |
372 | * find - find a match for the main NFA (no-complications case) |
373 | */ |
374 | static int |
375 | find(struct vars *v, |
376 | struct cnfa *cnfa, |
377 | struct colormap *cm) |
378 | { |
379 | struct dfa *s; |
380 | struct dfa *d; |
381 | chr *begin; |
382 | chr *end = NULL; |
383 | chr *cold; |
384 | chr *open; /* open and close of range of possible starts */ |
385 | chr *close; |
386 | int hitend; |
387 | int shorter = (v->g->tree->flags & SHORTER) ? 1 : 0; |
388 | |
389 | /* first, a shot with the search RE */ |
390 | s = newdfa(v, &v->g->search, cm, &v->dfa1); |
391 | assert(!(ISERR() && s != NULL)); |
392 | NOERR(); |
393 | MDEBUG(("\nsearch at %ld\n" , LOFF(v->start))); |
394 | cold = NULL; |
395 | close = shortest(v, s, v->search_start, v->search_start, v->stop, |
396 | &cold, (int *) NULL); |
397 | freedfa(s); |
398 | NOERR(); |
399 | if (v->g->cflags & REG_EXPECT) |
400 | { |
401 | assert(v->details != NULL); |
402 | if (cold != NULL) |
403 | v->details->rm_extend.rm_so = OFF(cold); |
404 | else |
405 | v->details->rm_extend.rm_so = OFF(v->stop); |
406 | v->details->rm_extend.rm_eo = OFF(v->stop); /* unknown */ |
407 | } |
408 | if (close == NULL) /* not found */ |
409 | return REG_NOMATCH; |
410 | if (v->nmatch == 0) /* found, don't need exact location */ |
411 | return REG_OKAY; |
412 | |
413 | /* find starting point and match */ |
414 | assert(cold != NULL); |
415 | open = cold; |
416 | cold = NULL; |
417 | MDEBUG(("between %ld and %ld\n" , LOFF(open), LOFF(close))); |
418 | d = newdfa(v, cnfa, cm, &v->dfa1); |
419 | assert(!(ISERR() && d != NULL)); |
420 | NOERR(); |
421 | for (begin = open; begin <= close; begin++) |
422 | { |
423 | MDEBUG(("\nfind trying at %ld\n" , LOFF(begin))); |
424 | if (shorter) |
425 | end = shortest(v, d, begin, begin, v->stop, |
426 | (chr **) NULL, &hitend); |
427 | else |
428 | end = longest(v, d, begin, v->stop, &hitend); |
429 | if (ISERR()) |
430 | { |
431 | freedfa(d); |
432 | return v->err; |
433 | } |
434 | if (hitend && cold == NULL) |
435 | cold = begin; |
436 | if (end != NULL) |
437 | break; /* NOTE BREAK OUT */ |
438 | } |
439 | assert(end != NULL); /* search RE succeeded so loop should */ |
440 | freedfa(d); |
441 | |
442 | /* and pin down details */ |
443 | assert(v->nmatch > 0); |
444 | v->pmatch[0].rm_so = OFF(begin); |
445 | v->pmatch[0].rm_eo = OFF(end); |
446 | if (v->g->cflags & REG_EXPECT) |
447 | { |
448 | if (cold != NULL) |
449 | v->details->rm_extend.rm_so = OFF(cold); |
450 | else |
451 | v->details->rm_extend.rm_so = OFF(v->stop); |
452 | v->details->rm_extend.rm_eo = OFF(v->stop); /* unknown */ |
453 | } |
454 | if (v->nmatch == 1) /* no need for submatches */ |
455 | return REG_OKAY; |
456 | |
457 | /* find submatches */ |
458 | zapallsubs(v->pmatch, v->nmatch); |
459 | return cdissect(v, v->g->tree, begin, end); |
460 | } |
461 | |
462 | /* |
463 | * cfind - find a match for the main NFA (with complications) |
464 | */ |
465 | static int |
466 | cfind(struct vars *v, |
467 | struct cnfa *cnfa, |
468 | struct colormap *cm) |
469 | { |
470 | struct dfa *s; |
471 | struct dfa *d; |
472 | chr *cold; |
473 | int ret; |
474 | |
475 | s = newdfa(v, &v->g->search, cm, &v->dfa1); |
476 | NOERR(); |
477 | d = newdfa(v, cnfa, cm, &v->dfa2); |
478 | if (ISERR()) |
479 | { |
480 | assert(d == NULL); |
481 | freedfa(s); |
482 | return v->err; |
483 | } |
484 | |
485 | ret = cfindloop(v, cnfa, cm, d, s, &cold); |
486 | |
487 | freedfa(d); |
488 | freedfa(s); |
489 | NOERR(); |
490 | if (v->g->cflags & REG_EXPECT) |
491 | { |
492 | assert(v->details != NULL); |
493 | if (cold != NULL) |
494 | v->details->rm_extend.rm_so = OFF(cold); |
495 | else |
496 | v->details->rm_extend.rm_so = OFF(v->stop); |
497 | v->details->rm_extend.rm_eo = OFF(v->stop); /* unknown */ |
498 | } |
499 | return ret; |
500 | } |
501 | |
502 | /* |
503 | * cfindloop - the heart of cfind |
504 | */ |
505 | static int |
506 | cfindloop(struct vars *v, |
507 | struct cnfa *cnfa, |
508 | struct colormap *cm, |
509 | struct dfa *d, |
510 | struct dfa *s, |
511 | chr **coldp) /* where to put coldstart pointer */ |
512 | { |
513 | chr *begin; |
514 | chr *end; |
515 | chr *cold; |
516 | chr *open; /* open and close of range of possible starts */ |
517 | chr *close; |
518 | chr *estart; |
519 | chr *estop; |
520 | int er; |
521 | int shorter = v->g->tree->flags & SHORTER; |
522 | int hitend; |
523 | |
524 | assert(d != NULL && s != NULL); |
525 | cold = NULL; |
526 | close = v->search_start; |
527 | do |
528 | { |
529 | /* Search with the search RE for match range at/beyond "close" */ |
530 | MDEBUG(("\ncsearch at %ld\n" , LOFF(close))); |
531 | close = shortest(v, s, close, close, v->stop, &cold, (int *) NULL); |
532 | if (ISERR()) |
533 | { |
534 | *coldp = cold; |
535 | return v->err; |
536 | } |
537 | if (close == NULL) |
538 | break; /* no more possible match anywhere */ |
539 | assert(cold != NULL); |
540 | open = cold; |
541 | cold = NULL; |
542 | /* Search for matches starting between "open" and "close" inclusive */ |
543 | MDEBUG(("cbetween %ld and %ld\n" , LOFF(open), LOFF(close))); |
544 | for (begin = open; begin <= close; begin++) |
545 | { |
546 | MDEBUG(("\ncfind trying at %ld\n" , LOFF(begin))); |
547 | estart = begin; |
548 | estop = v->stop; |
549 | for (;;) |
550 | { |
551 | /* Here we use the top node's detailed RE */ |
552 | if (shorter) |
553 | end = shortest(v, d, begin, estart, |
554 | estop, (chr **) NULL, &hitend); |
555 | else |
556 | end = longest(v, d, begin, estop, |
557 | &hitend); |
558 | if (ISERR()) |
559 | { |
560 | *coldp = cold; |
561 | return v->err; |
562 | } |
563 | if (hitend && cold == NULL) |
564 | cold = begin; |
565 | if (end == NULL) |
566 | break; /* no match with this begin point, try next */ |
567 | MDEBUG(("tentative end %ld\n" , LOFF(end))); |
568 | /* Dissect the potential match to see if it really matches */ |
569 | zapallsubs(v->pmatch, v->nmatch); |
570 | er = cdissect(v, v->g->tree, begin, end); |
571 | if (er == REG_OKAY) |
572 | { |
573 | if (v->nmatch > 0) |
574 | { |
575 | v->pmatch[0].rm_so = OFF(begin); |
576 | v->pmatch[0].rm_eo = OFF(end); |
577 | } |
578 | *coldp = cold; |
579 | return REG_OKAY; |
580 | } |
581 | if (er != REG_NOMATCH) |
582 | { |
583 | ERR(er); |
584 | *coldp = cold; |
585 | return er; |
586 | } |
587 | /* Try next longer/shorter match with same begin point */ |
588 | if (shorter) |
589 | { |
590 | if (end == estop) |
591 | break; /* no more, so try next begin point */ |
592 | estart = end + 1; |
593 | } |
594 | else |
595 | { |
596 | if (end == begin) |
597 | break; /* no more, so try next begin point */ |
598 | estop = end - 1; |
599 | } |
600 | } /* end loop over endpoint positions */ |
601 | } /* end loop over beginning positions */ |
602 | |
603 | /* |
604 | * If we get here, there is no possible match starting at or before |
605 | * "close", so consider matches beyond that. We'll do a fresh search |
606 | * with the search RE to find a new promising match range. |
607 | */ |
608 | close++; |
609 | } while (close < v->stop); |
610 | |
611 | *coldp = cold; |
612 | return REG_NOMATCH; |
613 | } |
614 | |
615 | /* |
616 | * zapallsubs - initialize all subexpression matches to "no match" |
617 | */ |
618 | static void |
619 | zapallsubs(regmatch_t *p, |
620 | size_t n) |
621 | { |
622 | size_t i; |
623 | |
624 | for (i = n - 1; i > 0; i--) |
625 | { |
626 | p[i].rm_so = -1; |
627 | p[i].rm_eo = -1; |
628 | } |
629 | } |
630 | |
631 | /* |
632 | * zaptreesubs - initialize subexpressions within subtree to "no match" |
633 | */ |
634 | static void |
635 | zaptreesubs(struct vars *v, |
636 | struct subre *t) |
637 | { |
638 | if (t->op == '(') |
639 | { |
640 | int n = t->subno; |
641 | |
642 | assert(n > 0); |
643 | if ((size_t) n < v->nmatch) |
644 | { |
645 | v->pmatch[n].rm_so = -1; |
646 | v->pmatch[n].rm_eo = -1; |
647 | } |
648 | } |
649 | |
650 | if (t->left != NULL) |
651 | zaptreesubs(v, t->left); |
652 | if (t->right != NULL) |
653 | zaptreesubs(v, t->right); |
654 | } |
655 | |
656 | /* |
657 | * subset - set subexpression match data for a successful subre |
658 | */ |
659 | static void |
660 | subset(struct vars *v, |
661 | struct subre *sub, |
662 | chr *begin, |
663 | chr *end) |
664 | { |
665 | int n = sub->subno; |
666 | |
667 | assert(n > 0); |
668 | if ((size_t) n >= v->nmatch) |
669 | return; |
670 | |
671 | MDEBUG(("setting %d\n" , n)); |
672 | v->pmatch[n].rm_so = OFF(begin); |
673 | v->pmatch[n].rm_eo = OFF(end); |
674 | } |
675 | |
676 | /* |
677 | * cdissect - check backrefs and determine subexpression matches |
678 | * |
679 | * cdissect recursively processes a subre tree to check matching of backrefs |
680 | * and/or identify submatch boundaries for capture nodes. The proposed match |
681 | * runs from "begin" to "end" (not including "end"), and we are basically |
682 | * "dissecting" it to see where the submatches are. |
683 | * |
684 | * Before calling any level of cdissect, the caller must have run the node's |
685 | * DFA and found that the proposed substring satisfies the DFA. (We make |
686 | * the caller do that because in concatenation and iteration nodes, it's |
687 | * much faster to check all the substrings against the child DFAs before we |
688 | * recurse.) Also, caller must have cleared subexpression match data via |
689 | * zaptreesubs (or zapallsubs at the top level). |
690 | */ |
691 | static int /* regexec return code */ |
692 | cdissect(struct vars *v, |
693 | struct subre *t, |
694 | chr *begin, /* beginning of relevant substring */ |
695 | chr *end) /* end of same */ |
696 | { |
697 | int er; |
698 | |
699 | assert(t != NULL); |
700 | MDEBUG(("cdissect %ld-%ld %c\n" , LOFF(begin), LOFF(end), t->op)); |
701 | |
702 | /* handy place to check for operation cancel */ |
703 | if (CANCEL_REQUESTED(v->re)) |
704 | return REG_CANCEL; |
705 | /* ... and stack overrun */ |
706 | if (STACK_TOO_DEEP(v->re)) |
707 | return REG_ETOOBIG; |
708 | |
709 | switch (t->op) |
710 | { |
711 | case '=': /* terminal node */ |
712 | assert(t->left == NULL && t->right == NULL); |
713 | er = REG_OKAY; /* no action, parent did the work */ |
714 | break; |
715 | case 'b': /* back reference */ |
716 | assert(t->left == NULL && t->right == NULL); |
717 | er = cbrdissect(v, t, begin, end); |
718 | break; |
719 | case '.': /* concatenation */ |
720 | assert(t->left != NULL && t->right != NULL); |
721 | if (t->left->flags & SHORTER) /* reverse scan */ |
722 | er = crevcondissect(v, t, begin, end); |
723 | else |
724 | er = ccondissect(v, t, begin, end); |
725 | break; |
726 | case '|': /* alternation */ |
727 | assert(t->left != NULL); |
728 | er = caltdissect(v, t, begin, end); |
729 | break; |
730 | case '*': /* iteration */ |
731 | assert(t->left != NULL); |
732 | if (t->left->flags & SHORTER) /* reverse scan */ |
733 | er = creviterdissect(v, t, begin, end); |
734 | else |
735 | er = citerdissect(v, t, begin, end); |
736 | break; |
737 | case '(': /* capturing */ |
738 | assert(t->left != NULL && t->right == NULL); |
739 | assert(t->subno > 0); |
740 | er = cdissect(v, t->left, begin, end); |
741 | if (er == REG_OKAY) |
742 | subset(v, t, begin, end); |
743 | break; |
744 | default: |
745 | er = REG_ASSERT; |
746 | break; |
747 | } |
748 | |
749 | /* |
750 | * We should never have a match failure unless backrefs lurk below; |
751 | * otherwise, either caller failed to check the DFA, or there's some |
752 | * inconsistency between the DFA and the node's innards. |
753 | */ |
754 | assert(er != REG_NOMATCH || (t->flags & BACKR)); |
755 | |
756 | return er; |
757 | } |
758 | |
759 | /* |
760 | * ccondissect - dissect match for concatenation node |
761 | */ |
762 | static int /* regexec return code */ |
763 | ccondissect(struct vars *v, |
764 | struct subre *t, |
765 | chr *begin, /* beginning of relevant substring */ |
766 | chr *end) /* end of same */ |
767 | { |
768 | struct dfa *d; |
769 | struct dfa *d2; |
770 | chr *mid; |
771 | int er; |
772 | |
773 | assert(t->op == '.'); |
774 | assert(t->left != NULL && t->left->cnfa.nstates > 0); |
775 | assert(t->right != NULL && t->right->cnfa.nstates > 0); |
776 | assert(!(t->left->flags & SHORTER)); |
777 | |
778 | d = getsubdfa(v, t->left); |
779 | NOERR(); |
780 | d2 = getsubdfa(v, t->right); |
781 | NOERR(); |
782 | MDEBUG(("cconcat %d\n" , t->id)); |
783 | |
784 | /* pick a tentative midpoint */ |
785 | mid = longest(v, d, begin, end, (int *) NULL); |
786 | NOERR(); |
787 | if (mid == NULL) |
788 | return REG_NOMATCH; |
789 | MDEBUG(("tentative midpoint %ld\n" , LOFF(mid))); |
790 | |
791 | /* iterate until satisfaction or failure */ |
792 | for (;;) |
793 | { |
794 | /* try this midpoint on for size */ |
795 | if (longest(v, d2, mid, end, (int *) NULL) == end) |
796 | { |
797 | er = cdissect(v, t->left, begin, mid); |
798 | if (er == REG_OKAY) |
799 | { |
800 | er = cdissect(v, t->right, mid, end); |
801 | if (er == REG_OKAY) |
802 | { |
803 | /* satisfaction */ |
804 | MDEBUG(("successful\n" )); |
805 | return REG_OKAY; |
806 | } |
807 | } |
808 | if (er != REG_NOMATCH) |
809 | return er; |
810 | } |
811 | NOERR(); |
812 | |
813 | /* that midpoint didn't work, find a new one */ |
814 | if (mid == begin) |
815 | { |
816 | /* all possibilities exhausted */ |
817 | MDEBUG(("%d no midpoint\n" , t->id)); |
818 | return REG_NOMATCH; |
819 | } |
820 | mid = longest(v, d, begin, mid - 1, (int *) NULL); |
821 | NOERR(); |
822 | if (mid == NULL) |
823 | { |
824 | /* failed to find a new one */ |
825 | MDEBUG(("%d failed midpoint\n" , t->id)); |
826 | return REG_NOMATCH; |
827 | } |
828 | MDEBUG(("%d: new midpoint %ld\n" , t->id, LOFF(mid))); |
829 | zaptreesubs(v, t->left); |
830 | zaptreesubs(v, t->right); |
831 | } |
832 | |
833 | /* can't get here */ |
834 | return REG_ASSERT; |
835 | } |
836 | |
837 | /* |
838 | * crevcondissect - dissect match for concatenation node, shortest-first |
839 | */ |
840 | static int /* regexec return code */ |
841 | crevcondissect(struct vars *v, |
842 | struct subre *t, |
843 | chr *begin, /* beginning of relevant substring */ |
844 | chr *end) /* end of same */ |
845 | { |
846 | struct dfa *d; |
847 | struct dfa *d2; |
848 | chr *mid; |
849 | int er; |
850 | |
851 | assert(t->op == '.'); |
852 | assert(t->left != NULL && t->left->cnfa.nstates > 0); |
853 | assert(t->right != NULL && t->right->cnfa.nstates > 0); |
854 | assert(t->left->flags & SHORTER); |
855 | |
856 | d = getsubdfa(v, t->left); |
857 | NOERR(); |
858 | d2 = getsubdfa(v, t->right); |
859 | NOERR(); |
860 | MDEBUG(("crevcon %d\n" , t->id)); |
861 | |
862 | /* pick a tentative midpoint */ |
863 | mid = shortest(v, d, begin, begin, end, (chr **) NULL, (int *) NULL); |
864 | NOERR(); |
865 | if (mid == NULL) |
866 | return REG_NOMATCH; |
867 | MDEBUG(("tentative midpoint %ld\n" , LOFF(mid))); |
868 | |
869 | /* iterate until satisfaction or failure */ |
870 | for (;;) |
871 | { |
872 | /* try this midpoint on for size */ |
873 | if (longest(v, d2, mid, end, (int *) NULL) == end) |
874 | { |
875 | er = cdissect(v, t->left, begin, mid); |
876 | if (er == REG_OKAY) |
877 | { |
878 | er = cdissect(v, t->right, mid, end); |
879 | if (er == REG_OKAY) |
880 | { |
881 | /* satisfaction */ |
882 | MDEBUG(("successful\n" )); |
883 | return REG_OKAY; |
884 | } |
885 | } |
886 | if (er != REG_NOMATCH) |
887 | return er; |
888 | } |
889 | NOERR(); |
890 | |
891 | /* that midpoint didn't work, find a new one */ |
892 | if (mid == end) |
893 | { |
894 | /* all possibilities exhausted */ |
895 | MDEBUG(("%d no midpoint\n" , t->id)); |
896 | return REG_NOMATCH; |
897 | } |
898 | mid = shortest(v, d, begin, mid + 1, end, (chr **) NULL, (int *) NULL); |
899 | NOERR(); |
900 | if (mid == NULL) |
901 | { |
902 | /* failed to find a new one */ |
903 | MDEBUG(("%d failed midpoint\n" , t->id)); |
904 | return REG_NOMATCH; |
905 | } |
906 | MDEBUG(("%d: new midpoint %ld\n" , t->id, LOFF(mid))); |
907 | zaptreesubs(v, t->left); |
908 | zaptreesubs(v, t->right); |
909 | } |
910 | |
911 | /* can't get here */ |
912 | return REG_ASSERT; |
913 | } |
914 | |
915 | /* |
916 | * cbrdissect - dissect match for backref node |
917 | */ |
918 | static int /* regexec return code */ |
919 | cbrdissect(struct vars *v, |
920 | struct subre *t, |
921 | chr *begin, /* beginning of relevant substring */ |
922 | chr *end) /* end of same */ |
923 | { |
924 | int n = t->subno; |
925 | size_t numreps; |
926 | size_t tlen; |
927 | size_t brlen; |
928 | chr *brstring; |
929 | chr *p; |
930 | int min = t->min; |
931 | int max = t->max; |
932 | |
933 | assert(t != NULL); |
934 | assert(t->op == 'b'); |
935 | assert(n >= 0); |
936 | assert((size_t) n < v->nmatch); |
937 | |
938 | MDEBUG(("cbackref n%d %d{%d-%d}\n" , t->id, n, min, max)); |
939 | |
940 | /* get the backreferenced string */ |
941 | if (v->pmatch[n].rm_so == -1) |
942 | return REG_NOMATCH; |
943 | brstring = v->start + v->pmatch[n].rm_so; |
944 | brlen = v->pmatch[n].rm_eo - v->pmatch[n].rm_so; |
945 | |
946 | /* special cases for zero-length strings */ |
947 | if (brlen == 0) |
948 | { |
949 | /* |
950 | * matches only if target is zero length, but any number of |
951 | * repetitions can be considered to be present |
952 | */ |
953 | if (begin == end && min <= max) |
954 | { |
955 | MDEBUG(("cbackref matched trivially\n" )); |
956 | return REG_OKAY; |
957 | } |
958 | return REG_NOMATCH; |
959 | } |
960 | if (begin == end) |
961 | { |
962 | /* matches only if zero repetitions are okay */ |
963 | if (min == 0) |
964 | { |
965 | MDEBUG(("cbackref matched trivially\n" )); |
966 | return REG_OKAY; |
967 | } |
968 | return REG_NOMATCH; |
969 | } |
970 | |
971 | /* |
972 | * check target length to see if it could possibly be an allowed number of |
973 | * repetitions of brstring |
974 | */ |
975 | assert(end > begin); |
976 | tlen = end - begin; |
977 | if (tlen % brlen != 0) |
978 | return REG_NOMATCH; |
979 | numreps = tlen / brlen; |
980 | if (numreps < min || (numreps > max && max != DUPINF)) |
981 | return REG_NOMATCH; |
982 | |
983 | /* okay, compare the actual string contents */ |
984 | p = begin; |
985 | while (numreps-- > 0) |
986 | { |
987 | if ((*v->g->compare) (brstring, p, brlen) != 0) |
988 | return REG_NOMATCH; |
989 | p += brlen; |
990 | } |
991 | |
992 | MDEBUG(("cbackref matched\n" )); |
993 | return REG_OKAY; |
994 | } |
995 | |
996 | /* |
997 | * caltdissect - dissect match for alternation node |
998 | */ |
999 | static int /* regexec return code */ |
1000 | caltdissect(struct vars *v, |
1001 | struct subre *t, |
1002 | chr *begin, /* beginning of relevant substring */ |
1003 | chr *end) /* end of same */ |
1004 | { |
1005 | struct dfa *d; |
1006 | int er; |
1007 | |
1008 | /* We loop, rather than tail-recurse, to handle a chain of alternatives */ |
1009 | while (t != NULL) |
1010 | { |
1011 | assert(t->op == '|'); |
1012 | assert(t->left != NULL && t->left->cnfa.nstates > 0); |
1013 | |
1014 | MDEBUG(("calt n%d\n" , t->id)); |
1015 | |
1016 | d = getsubdfa(v, t->left); |
1017 | NOERR(); |
1018 | if (longest(v, d, begin, end, (int *) NULL) == end) |
1019 | { |
1020 | MDEBUG(("calt matched\n" )); |
1021 | er = cdissect(v, t->left, begin, end); |
1022 | if (er != REG_NOMATCH) |
1023 | return er; |
1024 | } |
1025 | NOERR(); |
1026 | |
1027 | t = t->right; |
1028 | } |
1029 | |
1030 | return REG_NOMATCH; |
1031 | } |
1032 | |
1033 | /* |
1034 | * citerdissect - dissect match for iteration node |
1035 | */ |
1036 | static int /* regexec return code */ |
1037 | citerdissect(struct vars *v, |
1038 | struct subre *t, |
1039 | chr *begin, /* beginning of relevant substring */ |
1040 | chr *end) /* end of same */ |
1041 | { |
1042 | struct dfa *d; |
1043 | chr **endpts; |
1044 | chr *limit; |
1045 | int min_matches; |
1046 | size_t max_matches; |
1047 | int nverified; |
1048 | int k; |
1049 | int i; |
1050 | int er; |
1051 | |
1052 | assert(t->op == '*'); |
1053 | assert(t->left != NULL && t->left->cnfa.nstates > 0); |
1054 | assert(!(t->left->flags & SHORTER)); |
1055 | assert(begin <= end); |
1056 | |
1057 | /* |
1058 | * For the moment, assume the minimum number of matches is 1. If zero |
1059 | * matches are allowed, and the target string is empty, we are allowed to |
1060 | * match regardless of the contents of the iter node --- but we would |
1061 | * prefer to match once, so that capturing parens get set. (An example of |
1062 | * the concern here is a pattern like "()*\1", which historically this |
1063 | * code has allowed to succeed.) Therefore, we deal with the zero-matches |
1064 | * case at the bottom, after failing to find any other way to match. |
1065 | */ |
1066 | min_matches = t->min; |
1067 | if (min_matches <= 0) |
1068 | min_matches = 1; |
1069 | |
1070 | /* |
1071 | * We need workspace to track the endpoints of each sub-match. Normally |
1072 | * we consider only nonzero-length sub-matches, so there can be at most |
1073 | * end-begin of them. However, if min is larger than that, we will also |
1074 | * consider zero-length sub-matches in order to find enough matches. |
1075 | * |
1076 | * For convenience, endpts[0] contains the "begin" pointer and we store |
1077 | * sub-match endpoints in endpts[1..max_matches]. |
1078 | */ |
1079 | max_matches = end - begin; |
1080 | if (max_matches > t->max && t->max != DUPINF) |
1081 | max_matches = t->max; |
1082 | if (max_matches < min_matches) |
1083 | max_matches = min_matches; |
1084 | endpts = (chr **) MALLOC((max_matches + 1) * sizeof(chr *)); |
1085 | if (endpts == NULL) |
1086 | return REG_ESPACE; |
1087 | endpts[0] = begin; |
1088 | |
1089 | d = getsubdfa(v, t->left); |
1090 | if (ISERR()) |
1091 | { |
1092 | FREE(endpts); |
1093 | return v->err; |
1094 | } |
1095 | MDEBUG(("citer %d\n" , t->id)); |
1096 | |
1097 | /* |
1098 | * Our strategy is to first find a set of sub-match endpoints that are |
1099 | * valid according to the child node's DFA, and then recursively dissect |
1100 | * each sub-match to confirm validity. If any validity check fails, |
1101 | * backtrack the last sub-match and try again. And, when we next try for |
1102 | * a validity check, we need not recheck any successfully verified |
1103 | * sub-matches that we didn't move the endpoints of. nverified remembers |
1104 | * how many sub-matches are currently known okay. |
1105 | */ |
1106 | |
1107 | /* initialize to consider first sub-match */ |
1108 | nverified = 0; |
1109 | k = 1; |
1110 | limit = end; |
1111 | |
1112 | /* iterate until satisfaction or failure */ |
1113 | while (k > 0) |
1114 | { |
1115 | /* try to find an endpoint for the k'th sub-match */ |
1116 | endpts[k] = longest(v, d, endpts[k - 1], limit, (int *) NULL); |
1117 | if (ISERR()) |
1118 | { |
1119 | FREE(endpts); |
1120 | return v->err; |
1121 | } |
1122 | if (endpts[k] == NULL) |
1123 | { |
1124 | /* no match possible, so see if we can shorten previous one */ |
1125 | k--; |
1126 | goto backtrack; |
1127 | } |
1128 | MDEBUG(("%d: working endpoint %d: %ld\n" , |
1129 | t->id, k, LOFF(endpts[k]))); |
1130 | |
1131 | /* k'th sub-match can no longer be considered verified */ |
1132 | if (nverified >= k) |
1133 | nverified = k - 1; |
1134 | |
1135 | if (endpts[k] != end) |
1136 | { |
1137 | /* haven't reached end yet, try another iteration if allowed */ |
1138 | if (k >= max_matches) |
1139 | { |
1140 | /* must try to shorten some previous match */ |
1141 | k--; |
1142 | goto backtrack; |
1143 | } |
1144 | |
1145 | /* reject zero-length match unless necessary to achieve min */ |
1146 | if (endpts[k] == endpts[k - 1] && |
1147 | (k >= min_matches || min_matches - k < end - endpts[k])) |
1148 | goto backtrack; |
1149 | |
1150 | k++; |
1151 | limit = end; |
1152 | continue; |
1153 | } |
1154 | |
1155 | /* |
1156 | * We've identified a way to divide the string into k sub-matches that |
1157 | * works so far as the child DFA can tell. If k is an allowed number |
1158 | * of matches, start the slow part: recurse to verify each sub-match. |
1159 | * We always have k <= max_matches, needn't check that. |
1160 | */ |
1161 | if (k < min_matches) |
1162 | goto backtrack; |
1163 | |
1164 | MDEBUG(("%d: verifying %d..%d\n" , t->id, nverified + 1, k)); |
1165 | |
1166 | for (i = nverified + 1; i <= k; i++) |
1167 | { |
1168 | zaptreesubs(v, t->left); |
1169 | er = cdissect(v, t->left, endpts[i - 1], endpts[i]); |
1170 | if (er == REG_OKAY) |
1171 | { |
1172 | nverified = i; |
1173 | continue; |
1174 | } |
1175 | if (er == REG_NOMATCH) |
1176 | break; |
1177 | /* oops, something failed */ |
1178 | FREE(endpts); |
1179 | return er; |
1180 | } |
1181 | |
1182 | if (i > k) |
1183 | { |
1184 | /* satisfaction */ |
1185 | MDEBUG(("%d successful\n" , t->id)); |
1186 | FREE(endpts); |
1187 | return REG_OKAY; |
1188 | } |
1189 | |
1190 | /* match failed to verify, so backtrack */ |
1191 | |
1192 | backtrack: |
1193 | |
1194 | /* |
1195 | * Must consider shorter versions of the current sub-match. However, |
1196 | * we'll only ask for a zero-length match if necessary. |
1197 | */ |
1198 | while (k > 0) |
1199 | { |
1200 | chr *prev_end = endpts[k - 1]; |
1201 | |
1202 | if (endpts[k] > prev_end) |
1203 | { |
1204 | limit = endpts[k] - 1; |
1205 | if (limit > prev_end || |
1206 | (k < min_matches && min_matches - k >= end - prev_end)) |
1207 | { |
1208 | /* break out of backtrack loop, continue the outer one */ |
1209 | break; |
1210 | } |
1211 | } |
1212 | /* can't shorten k'th sub-match any more, consider previous one */ |
1213 | k--; |
1214 | } |
1215 | } |
1216 | |
1217 | /* all possibilities exhausted */ |
1218 | FREE(endpts); |
1219 | |
1220 | /* |
1221 | * Now consider the possibility that we can match to a zero-length string |
1222 | * by using zero repetitions. |
1223 | */ |
1224 | if (t->min == 0 && begin == end) |
1225 | { |
1226 | MDEBUG(("%d allowing zero matches\n" , t->id)); |
1227 | return REG_OKAY; |
1228 | } |
1229 | |
1230 | MDEBUG(("%d failed\n" , t->id)); |
1231 | return REG_NOMATCH; |
1232 | } |
1233 | |
1234 | /* |
1235 | * creviterdissect - dissect match for iteration node, shortest-first |
1236 | */ |
1237 | static int /* regexec return code */ |
1238 | creviterdissect(struct vars *v, |
1239 | struct subre *t, |
1240 | chr *begin, /* beginning of relevant substring */ |
1241 | chr *end) /* end of same */ |
1242 | { |
1243 | struct dfa *d; |
1244 | chr **endpts; |
1245 | chr *limit; |
1246 | int min_matches; |
1247 | size_t max_matches; |
1248 | int nverified; |
1249 | int k; |
1250 | int i; |
1251 | int er; |
1252 | |
1253 | assert(t->op == '*'); |
1254 | assert(t->left != NULL && t->left->cnfa.nstates > 0); |
1255 | assert(t->left->flags & SHORTER); |
1256 | assert(begin <= end); |
1257 | |
1258 | /* |
1259 | * If zero matches are allowed, and target string is empty, just declare |
1260 | * victory. OTOH, if target string isn't empty, zero matches can't work |
1261 | * so we pretend the min is 1. |
1262 | */ |
1263 | min_matches = t->min; |
1264 | if (min_matches <= 0) |
1265 | { |
1266 | if (begin == end) |
1267 | return REG_OKAY; |
1268 | min_matches = 1; |
1269 | } |
1270 | |
1271 | /* |
1272 | * We need workspace to track the endpoints of each sub-match. Normally |
1273 | * we consider only nonzero-length sub-matches, so there can be at most |
1274 | * end-begin of them. However, if min is larger than that, we will also |
1275 | * consider zero-length sub-matches in order to find enough matches. |
1276 | * |
1277 | * For convenience, endpts[0] contains the "begin" pointer and we store |
1278 | * sub-match endpoints in endpts[1..max_matches]. |
1279 | */ |
1280 | max_matches = end - begin; |
1281 | if (max_matches > t->max && t->max != DUPINF) |
1282 | max_matches = t->max; |
1283 | if (max_matches < min_matches) |
1284 | max_matches = min_matches; |
1285 | endpts = (chr **) MALLOC((max_matches + 1) * sizeof(chr *)); |
1286 | if (endpts == NULL) |
1287 | return REG_ESPACE; |
1288 | endpts[0] = begin; |
1289 | |
1290 | d = getsubdfa(v, t->left); |
1291 | if (ISERR()) |
1292 | { |
1293 | FREE(endpts); |
1294 | return v->err; |
1295 | } |
1296 | MDEBUG(("creviter %d\n" , t->id)); |
1297 | |
1298 | /* |
1299 | * Our strategy is to first find a set of sub-match endpoints that are |
1300 | * valid according to the child node's DFA, and then recursively dissect |
1301 | * each sub-match to confirm validity. If any validity check fails, |
1302 | * backtrack the last sub-match and try again. And, when we next try for |
1303 | * a validity check, we need not recheck any successfully verified |
1304 | * sub-matches that we didn't move the endpoints of. nverified remembers |
1305 | * how many sub-matches are currently known okay. |
1306 | */ |
1307 | |
1308 | /* initialize to consider first sub-match */ |
1309 | nverified = 0; |
1310 | k = 1; |
1311 | limit = begin; |
1312 | |
1313 | /* iterate until satisfaction or failure */ |
1314 | while (k > 0) |
1315 | { |
1316 | /* disallow zero-length match unless necessary to achieve min */ |
1317 | if (limit == endpts[k - 1] && |
1318 | limit != end && |
1319 | (k >= min_matches || min_matches - k < end - limit)) |
1320 | limit++; |
1321 | |
1322 | /* if this is the last allowed sub-match, it must reach to the end */ |
1323 | if (k >= max_matches) |
1324 | limit = end; |
1325 | |
1326 | /* try to find an endpoint for the k'th sub-match */ |
1327 | endpts[k] = shortest(v, d, endpts[k - 1], limit, end, |
1328 | (chr **) NULL, (int *) NULL); |
1329 | if (ISERR()) |
1330 | { |
1331 | FREE(endpts); |
1332 | return v->err; |
1333 | } |
1334 | if (endpts[k] == NULL) |
1335 | { |
1336 | /* no match possible, so see if we can lengthen previous one */ |
1337 | k--; |
1338 | goto backtrack; |
1339 | } |
1340 | MDEBUG(("%d: working endpoint %d: %ld\n" , |
1341 | t->id, k, LOFF(endpts[k]))); |
1342 | |
1343 | /* k'th sub-match can no longer be considered verified */ |
1344 | if (nverified >= k) |
1345 | nverified = k - 1; |
1346 | |
1347 | if (endpts[k] != end) |
1348 | { |
1349 | /* haven't reached end yet, try another iteration if allowed */ |
1350 | if (k >= max_matches) |
1351 | { |
1352 | /* must try to lengthen some previous match */ |
1353 | k--; |
1354 | goto backtrack; |
1355 | } |
1356 | |
1357 | k++; |
1358 | limit = endpts[k - 1]; |
1359 | continue; |
1360 | } |
1361 | |
1362 | /* |
1363 | * We've identified a way to divide the string into k sub-matches that |
1364 | * works so far as the child DFA can tell. If k is an allowed number |
1365 | * of matches, start the slow part: recurse to verify each sub-match. |
1366 | * We always have k <= max_matches, needn't check that. |
1367 | */ |
1368 | if (k < min_matches) |
1369 | goto backtrack; |
1370 | |
1371 | MDEBUG(("%d: verifying %d..%d\n" , t->id, nverified + 1, k)); |
1372 | |
1373 | for (i = nverified + 1; i <= k; i++) |
1374 | { |
1375 | zaptreesubs(v, t->left); |
1376 | er = cdissect(v, t->left, endpts[i - 1], endpts[i]); |
1377 | if (er == REG_OKAY) |
1378 | { |
1379 | nverified = i; |
1380 | continue; |
1381 | } |
1382 | if (er == REG_NOMATCH) |
1383 | break; |
1384 | /* oops, something failed */ |
1385 | FREE(endpts); |
1386 | return er; |
1387 | } |
1388 | |
1389 | if (i > k) |
1390 | { |
1391 | /* satisfaction */ |
1392 | MDEBUG(("%d successful\n" , t->id)); |
1393 | FREE(endpts); |
1394 | return REG_OKAY; |
1395 | } |
1396 | |
1397 | /* match failed to verify, so backtrack */ |
1398 | |
1399 | backtrack: |
1400 | |
1401 | /* |
1402 | * Must consider longer versions of the current sub-match. |
1403 | */ |
1404 | while (k > 0) |
1405 | { |
1406 | if (endpts[k] < end) |
1407 | { |
1408 | limit = endpts[k] + 1; |
1409 | /* break out of backtrack loop, continue the outer one */ |
1410 | break; |
1411 | } |
1412 | /* can't lengthen k'th sub-match any more, consider previous one */ |
1413 | k--; |
1414 | } |
1415 | } |
1416 | |
1417 | /* all possibilities exhausted */ |
1418 | MDEBUG(("%d failed\n" , t->id)); |
1419 | FREE(endpts); |
1420 | return REG_NOMATCH; |
1421 | } |
1422 | |
1423 | |
1424 | |
1425 | #include "rege_dfa.c" |
1426 | |