1 | /* |
2 | * NFA utilities. |
3 | * This file is #included by regcomp.c. |
4 | * |
5 | * Copyright (c) 1998, 1999 Henry Spencer. All rights reserved. |
6 | * |
7 | * Development of this software was funded, in part, by Cray Research Inc., |
8 | * UUNET Communications Services Inc., Sun Microsystems Inc., and Scriptics |
9 | * Corporation, none of whom are responsible for the results. The author |
10 | * thanks all of them. |
11 | * |
12 | * Redistribution and use in source and binary forms -- with or without |
13 | * modification -- are permitted for any purpose, provided that |
14 | * redistributions in source form retain this entire copyright notice and |
15 | * indicate the origin and nature of any modifications. |
16 | * |
17 | * I'd appreciate being given credit for this package in the documentation |
18 | * of software which uses it, but that is not a requirement. |
19 | * |
20 | * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, |
21 | * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY |
22 | * AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL |
23 | * HENRY SPENCER BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
24 | * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
25 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
26 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
27 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR |
28 | * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF |
29 | * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
30 | * |
31 | * src/backend/regex/regc_nfa.c |
32 | * |
33 | * |
34 | * One or two things that technically ought to be in here |
35 | * are actually in color.c, thanks to some incestuous relationships in |
36 | * the color chains. |
37 | */ |
38 | |
39 | #define NISERR() VISERR(nfa->v) |
40 | #define NERR(e) VERR(nfa->v, (e)) |
41 | |
42 | |
43 | /* |
44 | * newnfa - set up an NFA |
45 | */ |
46 | static struct nfa * /* the NFA, or NULL */ |
47 | newnfa(struct vars *v, |
48 | struct colormap *cm, |
49 | struct nfa *parent) /* NULL if primary NFA */ |
50 | { |
51 | struct nfa *nfa; |
52 | |
53 | nfa = (struct nfa *) MALLOC(sizeof(struct nfa)); |
54 | if (nfa == NULL) |
55 | { |
56 | ERR(REG_ESPACE); |
57 | return NULL; |
58 | } |
59 | |
60 | nfa->states = NULL; |
61 | nfa->slast = NULL; |
62 | nfa->free = NULL; |
63 | nfa->nstates = 0; |
64 | nfa->cm = cm; |
65 | nfa->v = v; |
66 | nfa->bos[0] = nfa->bos[1] = COLORLESS; |
67 | nfa->eos[0] = nfa->eos[1] = COLORLESS; |
68 | nfa->parent = parent; /* Precedes newfstate so parent is valid. */ |
69 | nfa->post = newfstate(nfa, '@'); /* number 0 */ |
70 | nfa->pre = newfstate(nfa, '>'); /* number 1 */ |
71 | |
72 | nfa->init = newstate(nfa); /* may become invalid later */ |
73 | nfa->final = newstate(nfa); |
74 | if (ISERR()) |
75 | { |
76 | freenfa(nfa); |
77 | return NULL; |
78 | } |
79 | rainbow(nfa, nfa->cm, PLAIN, COLORLESS, nfa->pre, nfa->init); |
80 | newarc(nfa, '^', 1, nfa->pre, nfa->init); |
81 | newarc(nfa, '^', 0, nfa->pre, nfa->init); |
82 | rainbow(nfa, nfa->cm, PLAIN, COLORLESS, nfa->final, nfa->post); |
83 | newarc(nfa, '$', 1, nfa->final, nfa->post); |
84 | newarc(nfa, '$', 0, nfa->final, nfa->post); |
85 | |
86 | if (ISERR()) |
87 | { |
88 | freenfa(nfa); |
89 | return NULL; |
90 | } |
91 | return nfa; |
92 | } |
93 | |
94 | /* |
95 | * freenfa - free an entire NFA |
96 | */ |
97 | static void |
98 | freenfa(struct nfa *nfa) |
99 | { |
100 | struct state *s; |
101 | |
102 | while ((s = nfa->states) != NULL) |
103 | { |
104 | s->nins = s->nouts = 0; /* don't worry about arcs */ |
105 | freestate(nfa, s); |
106 | } |
107 | while ((s = nfa->free) != NULL) |
108 | { |
109 | nfa->free = s->next; |
110 | destroystate(nfa, s); |
111 | } |
112 | |
113 | nfa->slast = NULL; |
114 | nfa->nstates = -1; |
115 | nfa->pre = NULL; |
116 | nfa->post = NULL; |
117 | FREE(nfa); |
118 | } |
119 | |
120 | /* |
121 | * newstate - allocate an NFA state, with zero flag value |
122 | */ |
123 | static struct state * /* NULL on error */ |
124 | newstate(struct nfa *nfa) |
125 | { |
126 | struct state *s; |
127 | |
128 | /* |
129 | * This is a handy place to check for operation cancel during regex |
130 | * compilation, since no code path will go very long without making a new |
131 | * state or arc. |
132 | */ |
133 | if (CANCEL_REQUESTED(nfa->v->re)) |
134 | { |
135 | NERR(REG_CANCEL); |
136 | return NULL; |
137 | } |
138 | |
139 | if (nfa->free != NULL) |
140 | { |
141 | s = nfa->free; |
142 | nfa->free = s->next; |
143 | } |
144 | else |
145 | { |
146 | if (nfa->v->spaceused >= REG_MAX_COMPILE_SPACE) |
147 | { |
148 | NERR(REG_ETOOBIG); |
149 | return NULL; |
150 | } |
151 | s = (struct state *) MALLOC(sizeof(struct state)); |
152 | if (s == NULL) |
153 | { |
154 | NERR(REG_ESPACE); |
155 | return NULL; |
156 | } |
157 | nfa->v->spaceused += sizeof(struct state); |
158 | s->oas.next = NULL; |
159 | s->free = NULL; |
160 | s->noas = 0; |
161 | } |
162 | |
163 | assert(nfa->nstates >= 0); |
164 | s->no = nfa->nstates++; |
165 | s->flag = 0; |
166 | if (nfa->states == NULL) |
167 | nfa->states = s; |
168 | s->nins = 0; |
169 | s->ins = NULL; |
170 | s->nouts = 0; |
171 | s->outs = NULL; |
172 | s->tmp = NULL; |
173 | s->next = NULL; |
174 | if (nfa->slast != NULL) |
175 | { |
176 | assert(nfa->slast->next == NULL); |
177 | nfa->slast->next = s; |
178 | } |
179 | s->prev = nfa->slast; |
180 | nfa->slast = s; |
181 | return s; |
182 | } |
183 | |
184 | /* |
185 | * newfstate - allocate an NFA state with a specified flag value |
186 | */ |
187 | static struct state * /* NULL on error */ |
188 | newfstate(struct nfa *nfa, int flag) |
189 | { |
190 | struct state *s; |
191 | |
192 | s = newstate(nfa); |
193 | if (s != NULL) |
194 | s->flag = (char) flag; |
195 | return s; |
196 | } |
197 | |
198 | /* |
199 | * dropstate - delete a state's inarcs and outarcs and free it |
200 | */ |
201 | static void |
202 | dropstate(struct nfa *nfa, |
203 | struct state *s) |
204 | { |
205 | struct arc *a; |
206 | |
207 | while ((a = s->ins) != NULL) |
208 | freearc(nfa, a); |
209 | while ((a = s->outs) != NULL) |
210 | freearc(nfa, a); |
211 | freestate(nfa, s); |
212 | } |
213 | |
214 | /* |
215 | * freestate - free a state, which has no in-arcs or out-arcs |
216 | */ |
217 | static void |
218 | freestate(struct nfa *nfa, |
219 | struct state *s) |
220 | { |
221 | assert(s != NULL); |
222 | assert(s->nins == 0 && s->nouts == 0); |
223 | |
224 | s->no = FREESTATE; |
225 | s->flag = 0; |
226 | if (s->next != NULL) |
227 | s->next->prev = s->prev; |
228 | else |
229 | { |
230 | assert(s == nfa->slast); |
231 | nfa->slast = s->prev; |
232 | } |
233 | if (s->prev != NULL) |
234 | s->prev->next = s->next; |
235 | else |
236 | { |
237 | assert(s == nfa->states); |
238 | nfa->states = s->next; |
239 | } |
240 | s->prev = NULL; |
241 | s->next = nfa->free; /* don't delete it, put it on the free list */ |
242 | nfa->free = s; |
243 | } |
244 | |
245 | /* |
246 | * destroystate - really get rid of an already-freed state |
247 | */ |
248 | static void |
249 | destroystate(struct nfa *nfa, |
250 | struct state *s) |
251 | { |
252 | struct arcbatch *ab; |
253 | struct arcbatch *abnext; |
254 | |
255 | assert(s->no == FREESTATE); |
256 | for (ab = s->oas.next; ab != NULL; ab = abnext) |
257 | { |
258 | abnext = ab->next; |
259 | FREE(ab); |
260 | nfa->v->spaceused -= sizeof(struct arcbatch); |
261 | } |
262 | s->ins = NULL; |
263 | s->outs = NULL; |
264 | s->next = NULL; |
265 | FREE(s); |
266 | nfa->v->spaceused -= sizeof(struct state); |
267 | } |
268 | |
269 | /* |
270 | * newarc - set up a new arc within an NFA |
271 | * |
272 | * This function checks to make sure that no duplicate arcs are created. |
273 | * In general we never want duplicates. |
274 | */ |
275 | static void |
276 | newarc(struct nfa *nfa, |
277 | int t, |
278 | color co, |
279 | struct state *from, |
280 | struct state *to) |
281 | { |
282 | struct arc *a; |
283 | |
284 | assert(from != NULL && to != NULL); |
285 | |
286 | /* |
287 | * This is a handy place to check for operation cancel during regex |
288 | * compilation, since no code path will go very long without making a new |
289 | * state or arc. |
290 | */ |
291 | if (CANCEL_REQUESTED(nfa->v->re)) |
292 | { |
293 | NERR(REG_CANCEL); |
294 | return; |
295 | } |
296 | |
297 | /* check for duplicate arc, using whichever chain is shorter */ |
298 | if (from->nouts <= to->nins) |
299 | { |
300 | for (a = from->outs; a != NULL; a = a->outchain) |
301 | if (a->to == to && a->co == co && a->type == t) |
302 | return; |
303 | } |
304 | else |
305 | { |
306 | for (a = to->ins; a != NULL; a = a->inchain) |
307 | if (a->from == from && a->co == co && a->type == t) |
308 | return; |
309 | } |
310 | |
311 | /* no dup, so create the arc */ |
312 | createarc(nfa, t, co, from, to); |
313 | } |
314 | |
315 | /* |
316 | * createarc - create a new arc within an NFA |
317 | * |
318 | * This function must *only* be used after verifying that there is no existing |
319 | * identical arc (same type/color/from/to). |
320 | */ |
321 | static void |
322 | createarc(struct nfa *nfa, |
323 | int t, |
324 | color co, |
325 | struct state *from, |
326 | struct state *to) |
327 | { |
328 | struct arc *a; |
329 | |
330 | /* the arc is physically allocated within its from-state */ |
331 | a = allocarc(nfa, from); |
332 | if (NISERR()) |
333 | return; |
334 | assert(a != NULL); |
335 | |
336 | a->type = t; |
337 | a->co = co; |
338 | a->to = to; |
339 | a->from = from; |
340 | |
341 | /* |
342 | * Put the new arc on the beginning, not the end, of the chains; it's |
343 | * simpler here, and freearc() is the same cost either way. See also the |
344 | * logic in moveins() and its cohorts, as well as fixempties(). |
345 | */ |
346 | a->inchain = to->ins; |
347 | a->inchainRev = NULL; |
348 | if (to->ins) |
349 | to->ins->inchainRev = a; |
350 | to->ins = a; |
351 | a->outchain = from->outs; |
352 | a->outchainRev = NULL; |
353 | if (from->outs) |
354 | from->outs->outchainRev = a; |
355 | from->outs = a; |
356 | |
357 | from->nouts++; |
358 | to->nins++; |
359 | |
360 | if (COLORED(a) && nfa->parent == NULL) |
361 | colorchain(nfa->cm, a); |
362 | } |
363 | |
364 | /* |
365 | * allocarc - allocate a new out-arc within a state |
366 | */ |
367 | static struct arc * /* NULL for failure */ |
368 | allocarc(struct nfa *nfa, |
369 | struct state *s) |
370 | { |
371 | struct arc *a; |
372 | |
373 | /* shortcut */ |
374 | if (s->free == NULL && s->noas < ABSIZE) |
375 | { |
376 | a = &s->oas.a[s->noas]; |
377 | s->noas++; |
378 | return a; |
379 | } |
380 | |
381 | /* if none at hand, get more */ |
382 | if (s->free == NULL) |
383 | { |
384 | struct arcbatch *newAb; |
385 | int i; |
386 | |
387 | if (nfa->v->spaceused >= REG_MAX_COMPILE_SPACE) |
388 | { |
389 | NERR(REG_ETOOBIG); |
390 | return NULL; |
391 | } |
392 | newAb = (struct arcbatch *) MALLOC(sizeof(struct arcbatch)); |
393 | if (newAb == NULL) |
394 | { |
395 | NERR(REG_ESPACE); |
396 | return NULL; |
397 | } |
398 | nfa->v->spaceused += sizeof(struct arcbatch); |
399 | newAb->next = s->oas.next; |
400 | s->oas.next = newAb; |
401 | |
402 | for (i = 0; i < ABSIZE; i++) |
403 | { |
404 | newAb->a[i].type = 0; |
405 | newAb->a[i].freechain = &newAb->a[i + 1]; |
406 | } |
407 | newAb->a[ABSIZE - 1].freechain = NULL; |
408 | s->free = &newAb->a[0]; |
409 | } |
410 | assert(s->free != NULL); |
411 | |
412 | a = s->free; |
413 | s->free = a->freechain; |
414 | return a; |
415 | } |
416 | |
417 | /* |
418 | * freearc - free an arc |
419 | */ |
420 | static void |
421 | freearc(struct nfa *nfa, |
422 | struct arc *victim) |
423 | { |
424 | struct state *from = victim->from; |
425 | struct state *to = victim->to; |
426 | struct arc *predecessor; |
427 | |
428 | assert(victim->type != 0); |
429 | |
430 | /* take it off color chain if necessary */ |
431 | if (COLORED(victim) && nfa->parent == NULL) |
432 | uncolorchain(nfa->cm, victim); |
433 | |
434 | /* take it off source's out-chain */ |
435 | assert(from != NULL); |
436 | predecessor = victim->outchainRev; |
437 | if (predecessor == NULL) |
438 | { |
439 | assert(from->outs == victim); |
440 | from->outs = victim->outchain; |
441 | } |
442 | else |
443 | { |
444 | assert(predecessor->outchain == victim); |
445 | predecessor->outchain = victim->outchain; |
446 | } |
447 | if (victim->outchain != NULL) |
448 | { |
449 | assert(victim->outchain->outchainRev == victim); |
450 | victim->outchain->outchainRev = predecessor; |
451 | } |
452 | from->nouts--; |
453 | |
454 | /* take it off target's in-chain */ |
455 | assert(to != NULL); |
456 | predecessor = victim->inchainRev; |
457 | if (predecessor == NULL) |
458 | { |
459 | assert(to->ins == victim); |
460 | to->ins = victim->inchain; |
461 | } |
462 | else |
463 | { |
464 | assert(predecessor->inchain == victim); |
465 | predecessor->inchain = victim->inchain; |
466 | } |
467 | if (victim->inchain != NULL) |
468 | { |
469 | assert(victim->inchain->inchainRev == victim); |
470 | victim->inchain->inchainRev = predecessor; |
471 | } |
472 | to->nins--; |
473 | |
474 | /* clean up and place on from-state's free list */ |
475 | victim->type = 0; |
476 | victim->from = NULL; /* precautions... */ |
477 | victim->to = NULL; |
478 | victim->inchain = NULL; |
479 | victim->inchainRev = NULL; |
480 | victim->outchain = NULL; |
481 | victim->outchainRev = NULL; |
482 | victim->freechain = from->free; |
483 | from->free = victim; |
484 | } |
485 | |
486 | /* |
487 | * changearctarget - flip an arc to have a different to state |
488 | * |
489 | * Caller must have verified that there is no pre-existing duplicate arc. |
490 | * |
491 | * Note that because we store arcs in their from state, we can't easily have |
492 | * a similar changearcsource function. |
493 | */ |
494 | static void |
495 | changearctarget(struct arc *a, struct state *newto) |
496 | { |
497 | struct state *oldto = a->to; |
498 | struct arc *predecessor; |
499 | |
500 | assert(oldto != newto); |
501 | |
502 | /* take it off old target's in-chain */ |
503 | assert(oldto != NULL); |
504 | predecessor = a->inchainRev; |
505 | if (predecessor == NULL) |
506 | { |
507 | assert(oldto->ins == a); |
508 | oldto->ins = a->inchain; |
509 | } |
510 | else |
511 | { |
512 | assert(predecessor->inchain == a); |
513 | predecessor->inchain = a->inchain; |
514 | } |
515 | if (a->inchain != NULL) |
516 | { |
517 | assert(a->inchain->inchainRev == a); |
518 | a->inchain->inchainRev = predecessor; |
519 | } |
520 | oldto->nins--; |
521 | |
522 | a->to = newto; |
523 | |
524 | /* prepend it to new target's in-chain */ |
525 | a->inchain = newto->ins; |
526 | a->inchainRev = NULL; |
527 | if (newto->ins) |
528 | newto->ins->inchainRev = a; |
529 | newto->ins = a; |
530 | newto->nins++; |
531 | } |
532 | |
533 | /* |
534 | * hasnonemptyout - Does state have a non-EMPTY out arc? |
535 | */ |
536 | static int |
537 | hasnonemptyout(struct state *s) |
538 | { |
539 | struct arc *a; |
540 | |
541 | for (a = s->outs; a != NULL; a = a->outchain) |
542 | { |
543 | if (a->type != EMPTY) |
544 | return 1; |
545 | } |
546 | return 0; |
547 | } |
548 | |
549 | /* |
550 | * findarc - find arc, if any, from given source with given type and color |
551 | * If there is more than one such arc, the result is random. |
552 | */ |
553 | static struct arc * |
554 | findarc(struct state *s, |
555 | int type, |
556 | color co) |
557 | { |
558 | struct arc *a; |
559 | |
560 | for (a = s->outs; a != NULL; a = a->outchain) |
561 | if (a->type == type && a->co == co) |
562 | return a; |
563 | return NULL; |
564 | } |
565 | |
566 | /* |
567 | * cparc - allocate a new arc within an NFA, copying details from old one |
568 | */ |
569 | static void |
570 | cparc(struct nfa *nfa, |
571 | struct arc *oa, |
572 | struct state *from, |
573 | struct state *to) |
574 | { |
575 | newarc(nfa, oa->type, oa->co, from, to); |
576 | } |
577 | |
578 | /* |
579 | * sortins - sort the in arcs of a state by from/color/type |
580 | */ |
581 | static void |
582 | sortins(struct nfa *nfa, |
583 | struct state *s) |
584 | { |
585 | struct arc **sortarray; |
586 | struct arc *a; |
587 | int n = s->nins; |
588 | int i; |
589 | |
590 | if (n <= 1) |
591 | return; /* nothing to do */ |
592 | /* make an array of arc pointers ... */ |
593 | sortarray = (struct arc **) MALLOC(n * sizeof(struct arc *)); |
594 | if (sortarray == NULL) |
595 | { |
596 | NERR(REG_ESPACE); |
597 | return; |
598 | } |
599 | i = 0; |
600 | for (a = s->ins; a != NULL; a = a->inchain) |
601 | sortarray[i++] = a; |
602 | assert(i == n); |
603 | /* ... sort the array */ |
604 | qsort(sortarray, n, sizeof(struct arc *), sortins_cmp); |
605 | /* ... and rebuild arc list in order */ |
606 | /* it seems worth special-casing first and last items to simplify loop */ |
607 | a = sortarray[0]; |
608 | s->ins = a; |
609 | a->inchain = sortarray[1]; |
610 | a->inchainRev = NULL; |
611 | for (i = 1; i < n - 1; i++) |
612 | { |
613 | a = sortarray[i]; |
614 | a->inchain = sortarray[i + 1]; |
615 | a->inchainRev = sortarray[i - 1]; |
616 | } |
617 | a = sortarray[i]; |
618 | a->inchain = NULL; |
619 | a->inchainRev = sortarray[i - 1]; |
620 | FREE(sortarray); |
621 | } |
622 | |
623 | static int |
624 | sortins_cmp(const void *a, const void *b) |
625 | { |
626 | const struct arc *aa = *((const struct arc *const *) a); |
627 | const struct arc *bb = *((const struct arc *const *) b); |
628 | |
629 | /* we check the fields in the order they are most likely to be different */ |
630 | if (aa->from->no < bb->from->no) |
631 | return -1; |
632 | if (aa->from->no > bb->from->no) |
633 | return 1; |
634 | if (aa->co < bb->co) |
635 | return -1; |
636 | if (aa->co > bb->co) |
637 | return 1; |
638 | if (aa->type < bb->type) |
639 | return -1; |
640 | if (aa->type > bb->type) |
641 | return 1; |
642 | return 0; |
643 | } |
644 | |
645 | /* |
646 | * sortouts - sort the out arcs of a state by to/color/type |
647 | */ |
648 | static void |
649 | sortouts(struct nfa *nfa, |
650 | struct state *s) |
651 | { |
652 | struct arc **sortarray; |
653 | struct arc *a; |
654 | int n = s->nouts; |
655 | int i; |
656 | |
657 | if (n <= 1) |
658 | return; /* nothing to do */ |
659 | /* make an array of arc pointers ... */ |
660 | sortarray = (struct arc **) MALLOC(n * sizeof(struct arc *)); |
661 | if (sortarray == NULL) |
662 | { |
663 | NERR(REG_ESPACE); |
664 | return; |
665 | } |
666 | i = 0; |
667 | for (a = s->outs; a != NULL; a = a->outchain) |
668 | sortarray[i++] = a; |
669 | assert(i == n); |
670 | /* ... sort the array */ |
671 | qsort(sortarray, n, sizeof(struct arc *), sortouts_cmp); |
672 | /* ... and rebuild arc list in order */ |
673 | /* it seems worth special-casing first and last items to simplify loop */ |
674 | a = sortarray[0]; |
675 | s->outs = a; |
676 | a->outchain = sortarray[1]; |
677 | a->outchainRev = NULL; |
678 | for (i = 1; i < n - 1; i++) |
679 | { |
680 | a = sortarray[i]; |
681 | a->outchain = sortarray[i + 1]; |
682 | a->outchainRev = sortarray[i - 1]; |
683 | } |
684 | a = sortarray[i]; |
685 | a->outchain = NULL; |
686 | a->outchainRev = sortarray[i - 1]; |
687 | FREE(sortarray); |
688 | } |
689 | |
690 | static int |
691 | sortouts_cmp(const void *a, const void *b) |
692 | { |
693 | const struct arc *aa = *((const struct arc *const *) a); |
694 | const struct arc *bb = *((const struct arc *const *) b); |
695 | |
696 | /* we check the fields in the order they are most likely to be different */ |
697 | if (aa->to->no < bb->to->no) |
698 | return -1; |
699 | if (aa->to->no > bb->to->no) |
700 | return 1; |
701 | if (aa->co < bb->co) |
702 | return -1; |
703 | if (aa->co > bb->co) |
704 | return 1; |
705 | if (aa->type < bb->type) |
706 | return -1; |
707 | if (aa->type > bb->type) |
708 | return 1; |
709 | return 0; |
710 | } |
711 | |
712 | /* |
713 | * Common decision logic about whether to use arc-by-arc operations or |
714 | * sort/merge. If there's just a few source arcs we cannot recoup the |
715 | * cost of sorting the destination arc list, no matter how large it is. |
716 | * Otherwise, limit the number of arc-by-arc comparisons to about 1000 |
717 | * (a somewhat arbitrary choice, but the breakeven point would probably |
718 | * be machine dependent anyway). |
719 | */ |
720 | #define BULK_ARC_OP_USE_SORT(nsrcarcs, ndestarcs) \ |
721 | ((nsrcarcs) < 4 ? 0 : ((nsrcarcs) > 32 || (ndestarcs) > 32)) |
722 | |
723 | /* |
724 | * moveins - move all in arcs of a state to another state |
725 | * |
726 | * You might think this could be done better by just updating the |
727 | * existing arcs, and you would be right if it weren't for the need |
728 | * for duplicate suppression, which makes it easier to just make new |
729 | * ones to exploit the suppression built into newarc. |
730 | * |
731 | * However, if we have a whole lot of arcs to deal with, retail duplicate |
732 | * checks become too slow. In that case we proceed by sorting and merging |
733 | * the arc lists, and then we can indeed just update the arcs in-place. |
734 | */ |
735 | static void |
736 | moveins(struct nfa *nfa, |
737 | struct state *oldState, |
738 | struct state *newState) |
739 | { |
740 | assert(oldState != newState); |
741 | |
742 | if (!BULK_ARC_OP_USE_SORT(oldState->nins, newState->nins)) |
743 | { |
744 | /* With not too many arcs, just do them one at a time */ |
745 | struct arc *a; |
746 | |
747 | while ((a = oldState->ins) != NULL) |
748 | { |
749 | cparc(nfa, a, a->from, newState); |
750 | freearc(nfa, a); |
751 | } |
752 | } |
753 | else |
754 | { |
755 | /* |
756 | * With many arcs, use a sort-merge approach. Note changearctarget() |
757 | * will put the arc onto the front of newState's chain, so it does not |
758 | * break our walk through the sorted part of the chain. |
759 | */ |
760 | struct arc *oa; |
761 | struct arc *na; |
762 | |
763 | /* |
764 | * Because we bypass newarc() in this code path, we'd better include a |
765 | * cancel check. |
766 | */ |
767 | if (CANCEL_REQUESTED(nfa->v->re)) |
768 | { |
769 | NERR(REG_CANCEL); |
770 | return; |
771 | } |
772 | |
773 | sortins(nfa, oldState); |
774 | sortins(nfa, newState); |
775 | if (NISERR()) |
776 | return; /* might have failed to sort */ |
777 | oa = oldState->ins; |
778 | na = newState->ins; |
779 | while (oa != NULL && na != NULL) |
780 | { |
781 | struct arc *a = oa; |
782 | |
783 | switch (sortins_cmp(&oa, &na)) |
784 | { |
785 | case -1: |
786 | /* newState does not have anything matching oa */ |
787 | oa = oa->inchain; |
788 | |
789 | /* |
790 | * Rather than doing createarc+freearc, we can just unlink |
791 | * and relink the existing arc struct. |
792 | */ |
793 | changearctarget(a, newState); |
794 | break; |
795 | case 0: |
796 | /* match, advance in both lists */ |
797 | oa = oa->inchain; |
798 | na = na->inchain; |
799 | /* ... and drop duplicate arc from oldState */ |
800 | freearc(nfa, a); |
801 | break; |
802 | case +1: |
803 | /* advance only na; oa might have a match later */ |
804 | na = na->inchain; |
805 | break; |
806 | default: |
807 | assert(NOTREACHED); |
808 | } |
809 | } |
810 | while (oa != NULL) |
811 | { |
812 | /* newState does not have anything matching oa */ |
813 | struct arc *a = oa; |
814 | |
815 | oa = oa->inchain; |
816 | changearctarget(a, newState); |
817 | } |
818 | } |
819 | |
820 | assert(oldState->nins == 0); |
821 | assert(oldState->ins == NULL); |
822 | } |
823 | |
824 | /* |
825 | * copyins - copy in arcs of a state to another state |
826 | */ |
827 | static void |
828 | copyins(struct nfa *nfa, |
829 | struct state *oldState, |
830 | struct state *newState) |
831 | { |
832 | assert(oldState != newState); |
833 | |
834 | if (!BULK_ARC_OP_USE_SORT(oldState->nins, newState->nins)) |
835 | { |
836 | /* With not too many arcs, just do them one at a time */ |
837 | struct arc *a; |
838 | |
839 | for (a = oldState->ins; a != NULL; a = a->inchain) |
840 | cparc(nfa, a, a->from, newState); |
841 | } |
842 | else |
843 | { |
844 | /* |
845 | * With many arcs, use a sort-merge approach. Note that createarc() |
846 | * will put new arcs onto the front of newState's chain, so it does |
847 | * not break our walk through the sorted part of the chain. |
848 | */ |
849 | struct arc *oa; |
850 | struct arc *na; |
851 | |
852 | /* |
853 | * Because we bypass newarc() in this code path, we'd better include a |
854 | * cancel check. |
855 | */ |
856 | if (CANCEL_REQUESTED(nfa->v->re)) |
857 | { |
858 | NERR(REG_CANCEL); |
859 | return; |
860 | } |
861 | |
862 | sortins(nfa, oldState); |
863 | sortins(nfa, newState); |
864 | if (NISERR()) |
865 | return; /* might have failed to sort */ |
866 | oa = oldState->ins; |
867 | na = newState->ins; |
868 | while (oa != NULL && na != NULL) |
869 | { |
870 | struct arc *a = oa; |
871 | |
872 | switch (sortins_cmp(&oa, &na)) |
873 | { |
874 | case -1: |
875 | /* newState does not have anything matching oa */ |
876 | oa = oa->inchain; |
877 | createarc(nfa, a->type, a->co, a->from, newState); |
878 | break; |
879 | case 0: |
880 | /* match, advance in both lists */ |
881 | oa = oa->inchain; |
882 | na = na->inchain; |
883 | break; |
884 | case +1: |
885 | /* advance only na; oa might have a match later */ |
886 | na = na->inchain; |
887 | break; |
888 | default: |
889 | assert(NOTREACHED); |
890 | } |
891 | } |
892 | while (oa != NULL) |
893 | { |
894 | /* newState does not have anything matching oa */ |
895 | struct arc *a = oa; |
896 | |
897 | oa = oa->inchain; |
898 | createarc(nfa, a->type, a->co, a->from, newState); |
899 | } |
900 | } |
901 | } |
902 | |
903 | /* |
904 | * mergeins - merge a list of inarcs into a state |
905 | * |
906 | * This is much like copyins, but the source arcs are listed in an array, |
907 | * and are not guaranteed unique. It's okay to clobber the array contents. |
908 | */ |
909 | static void |
910 | mergeins(struct nfa *nfa, |
911 | struct state *s, |
912 | struct arc **arcarray, |
913 | int arccount) |
914 | { |
915 | struct arc *na; |
916 | int i; |
917 | int j; |
918 | |
919 | if (arccount <= 0) |
920 | return; |
921 | |
922 | /* |
923 | * Because we bypass newarc() in this code path, we'd better include a |
924 | * cancel check. |
925 | */ |
926 | if (CANCEL_REQUESTED(nfa->v->re)) |
927 | { |
928 | NERR(REG_CANCEL); |
929 | return; |
930 | } |
931 | |
932 | /* Sort existing inarcs as well as proposed new ones */ |
933 | sortins(nfa, s); |
934 | if (NISERR()) |
935 | return; /* might have failed to sort */ |
936 | |
937 | qsort(arcarray, arccount, sizeof(struct arc *), sortins_cmp); |
938 | |
939 | /* |
940 | * arcarray very likely includes dups, so we must eliminate them. (This |
941 | * could be folded into the next loop, but it's not worth the trouble.) |
942 | */ |
943 | j = 0; |
944 | for (i = 1; i < arccount; i++) |
945 | { |
946 | switch (sortins_cmp(&arcarray[j], &arcarray[i])) |
947 | { |
948 | case -1: |
949 | /* non-dup */ |
950 | arcarray[++j] = arcarray[i]; |
951 | break; |
952 | case 0: |
953 | /* dup */ |
954 | break; |
955 | default: |
956 | /* trouble */ |
957 | assert(NOTREACHED); |
958 | } |
959 | } |
960 | arccount = j + 1; |
961 | |
962 | /* |
963 | * Now merge into s' inchain. Note that createarc() will put new arcs |
964 | * onto the front of s's chain, so it does not break our walk through the |
965 | * sorted part of the chain. |
966 | */ |
967 | i = 0; |
968 | na = s->ins; |
969 | while (i < arccount && na != NULL) |
970 | { |
971 | struct arc *a = arcarray[i]; |
972 | |
973 | switch (sortins_cmp(&a, &na)) |
974 | { |
975 | case -1: |
976 | /* s does not have anything matching a */ |
977 | createarc(nfa, a->type, a->co, a->from, s); |
978 | i++; |
979 | break; |
980 | case 0: |
981 | /* match, advance in both lists */ |
982 | i++; |
983 | na = na->inchain; |
984 | break; |
985 | case +1: |
986 | /* advance only na; array might have a match later */ |
987 | na = na->inchain; |
988 | break; |
989 | default: |
990 | assert(NOTREACHED); |
991 | } |
992 | } |
993 | while (i < arccount) |
994 | { |
995 | /* s does not have anything matching a */ |
996 | struct arc *a = arcarray[i]; |
997 | |
998 | createarc(nfa, a->type, a->co, a->from, s); |
999 | i++; |
1000 | } |
1001 | } |
1002 | |
1003 | /* |
1004 | * moveouts - move all out arcs of a state to another state |
1005 | */ |
1006 | static void |
1007 | moveouts(struct nfa *nfa, |
1008 | struct state *oldState, |
1009 | struct state *newState) |
1010 | { |
1011 | assert(oldState != newState); |
1012 | |
1013 | if (!BULK_ARC_OP_USE_SORT(oldState->nouts, newState->nouts)) |
1014 | { |
1015 | /* With not too many arcs, just do them one at a time */ |
1016 | struct arc *a; |
1017 | |
1018 | while ((a = oldState->outs) != NULL) |
1019 | { |
1020 | cparc(nfa, a, newState, a->to); |
1021 | freearc(nfa, a); |
1022 | } |
1023 | } |
1024 | else |
1025 | { |
1026 | /* |
1027 | * With many arcs, use a sort-merge approach. Note that createarc() |
1028 | * will put new arcs onto the front of newState's chain, so it does |
1029 | * not break our walk through the sorted part of the chain. |
1030 | */ |
1031 | struct arc *oa; |
1032 | struct arc *na; |
1033 | |
1034 | /* |
1035 | * Because we bypass newarc() in this code path, we'd better include a |
1036 | * cancel check. |
1037 | */ |
1038 | if (CANCEL_REQUESTED(nfa->v->re)) |
1039 | { |
1040 | NERR(REG_CANCEL); |
1041 | return; |
1042 | } |
1043 | |
1044 | sortouts(nfa, oldState); |
1045 | sortouts(nfa, newState); |
1046 | if (NISERR()) |
1047 | return; /* might have failed to sort */ |
1048 | oa = oldState->outs; |
1049 | na = newState->outs; |
1050 | while (oa != NULL && na != NULL) |
1051 | { |
1052 | struct arc *a = oa; |
1053 | |
1054 | switch (sortouts_cmp(&oa, &na)) |
1055 | { |
1056 | case -1: |
1057 | /* newState does not have anything matching oa */ |
1058 | oa = oa->outchain; |
1059 | createarc(nfa, a->type, a->co, newState, a->to); |
1060 | freearc(nfa, a); |
1061 | break; |
1062 | case 0: |
1063 | /* match, advance in both lists */ |
1064 | oa = oa->outchain; |
1065 | na = na->outchain; |
1066 | /* ... and drop duplicate arc from oldState */ |
1067 | freearc(nfa, a); |
1068 | break; |
1069 | case +1: |
1070 | /* advance only na; oa might have a match later */ |
1071 | na = na->outchain; |
1072 | break; |
1073 | default: |
1074 | assert(NOTREACHED); |
1075 | } |
1076 | } |
1077 | while (oa != NULL) |
1078 | { |
1079 | /* newState does not have anything matching oa */ |
1080 | struct arc *a = oa; |
1081 | |
1082 | oa = oa->outchain; |
1083 | createarc(nfa, a->type, a->co, newState, a->to); |
1084 | freearc(nfa, a); |
1085 | } |
1086 | } |
1087 | |
1088 | assert(oldState->nouts == 0); |
1089 | assert(oldState->outs == NULL); |
1090 | } |
1091 | |
1092 | /* |
1093 | * copyouts - copy out arcs of a state to another state |
1094 | */ |
1095 | static void |
1096 | copyouts(struct nfa *nfa, |
1097 | struct state *oldState, |
1098 | struct state *newState) |
1099 | { |
1100 | assert(oldState != newState); |
1101 | |
1102 | if (!BULK_ARC_OP_USE_SORT(oldState->nouts, newState->nouts)) |
1103 | { |
1104 | /* With not too many arcs, just do them one at a time */ |
1105 | struct arc *a; |
1106 | |
1107 | for (a = oldState->outs; a != NULL; a = a->outchain) |
1108 | cparc(nfa, a, newState, a->to); |
1109 | } |
1110 | else |
1111 | { |
1112 | /* |
1113 | * With many arcs, use a sort-merge approach. Note that createarc() |
1114 | * will put new arcs onto the front of newState's chain, so it does |
1115 | * not break our walk through the sorted part of the chain. |
1116 | */ |
1117 | struct arc *oa; |
1118 | struct arc *na; |
1119 | |
1120 | /* |
1121 | * Because we bypass newarc() in this code path, we'd better include a |
1122 | * cancel check. |
1123 | */ |
1124 | if (CANCEL_REQUESTED(nfa->v->re)) |
1125 | { |
1126 | NERR(REG_CANCEL); |
1127 | return; |
1128 | } |
1129 | |
1130 | sortouts(nfa, oldState); |
1131 | sortouts(nfa, newState); |
1132 | if (NISERR()) |
1133 | return; /* might have failed to sort */ |
1134 | oa = oldState->outs; |
1135 | na = newState->outs; |
1136 | while (oa != NULL && na != NULL) |
1137 | { |
1138 | struct arc *a = oa; |
1139 | |
1140 | switch (sortouts_cmp(&oa, &na)) |
1141 | { |
1142 | case -1: |
1143 | /* newState does not have anything matching oa */ |
1144 | oa = oa->outchain; |
1145 | createarc(nfa, a->type, a->co, newState, a->to); |
1146 | break; |
1147 | case 0: |
1148 | /* match, advance in both lists */ |
1149 | oa = oa->outchain; |
1150 | na = na->outchain; |
1151 | break; |
1152 | case +1: |
1153 | /* advance only na; oa might have a match later */ |
1154 | na = na->outchain; |
1155 | break; |
1156 | default: |
1157 | assert(NOTREACHED); |
1158 | } |
1159 | } |
1160 | while (oa != NULL) |
1161 | { |
1162 | /* newState does not have anything matching oa */ |
1163 | struct arc *a = oa; |
1164 | |
1165 | oa = oa->outchain; |
1166 | createarc(nfa, a->type, a->co, newState, a->to); |
1167 | } |
1168 | } |
1169 | } |
1170 | |
1171 | /* |
1172 | * cloneouts - copy out arcs of a state to another state pair, modifying type |
1173 | */ |
1174 | static void |
1175 | cloneouts(struct nfa *nfa, |
1176 | struct state *old, |
1177 | struct state *from, |
1178 | struct state *to, |
1179 | int type) |
1180 | { |
1181 | struct arc *a; |
1182 | |
1183 | assert(old != from); |
1184 | |
1185 | for (a = old->outs; a != NULL; a = a->outchain) |
1186 | newarc(nfa, type, a->co, from, to); |
1187 | } |
1188 | |
1189 | /* |
1190 | * delsub - delete a sub-NFA, updating subre pointers if necessary |
1191 | * |
1192 | * This uses a recursive traversal of the sub-NFA, marking already-seen |
1193 | * states using their tmp pointer. |
1194 | */ |
1195 | static void |
1196 | delsub(struct nfa *nfa, |
1197 | struct state *lp, /* the sub-NFA goes from here... */ |
1198 | struct state *rp) /* ...to here, *not* inclusive */ |
1199 | { |
1200 | assert(lp != rp); |
1201 | |
1202 | rp->tmp = rp; /* mark end */ |
1203 | |
1204 | deltraverse(nfa, lp, lp); |
1205 | if (NISERR()) |
1206 | return; /* asserts might not hold after failure */ |
1207 | assert(lp->nouts == 0 && rp->nins == 0); /* did the job */ |
1208 | assert(lp->no != FREESTATE && rp->no != FREESTATE); /* no more */ |
1209 | |
1210 | rp->tmp = NULL; /* unmark end */ |
1211 | lp->tmp = NULL; /* and begin, marked by deltraverse */ |
1212 | } |
1213 | |
1214 | /* |
1215 | * deltraverse - the recursive heart of delsub |
1216 | * This routine's basic job is to destroy all out-arcs of the state. |
1217 | */ |
1218 | static void |
1219 | deltraverse(struct nfa *nfa, |
1220 | struct state *leftend, |
1221 | struct state *s) |
1222 | { |
1223 | struct arc *a; |
1224 | struct state *to; |
1225 | |
1226 | /* Since this is recursive, it could be driven to stack overflow */ |
1227 | if (STACK_TOO_DEEP(nfa->v->re)) |
1228 | { |
1229 | NERR(REG_ETOOBIG); |
1230 | return; |
1231 | } |
1232 | |
1233 | if (s->nouts == 0) |
1234 | return; /* nothing to do */ |
1235 | if (s->tmp != NULL) |
1236 | return; /* already in progress */ |
1237 | |
1238 | s->tmp = s; /* mark as in progress */ |
1239 | |
1240 | while ((a = s->outs) != NULL) |
1241 | { |
1242 | to = a->to; |
1243 | deltraverse(nfa, leftend, to); |
1244 | if (NISERR()) |
1245 | return; /* asserts might not hold after failure */ |
1246 | assert(to->nouts == 0 || to->tmp != NULL); |
1247 | freearc(nfa, a); |
1248 | if (to->nins == 0 && to->tmp == NULL) |
1249 | { |
1250 | assert(to->nouts == 0); |
1251 | freestate(nfa, to); |
1252 | } |
1253 | } |
1254 | |
1255 | assert(s->no != FREESTATE); /* we're still here */ |
1256 | assert(s == leftend || s->nins != 0); /* and still reachable */ |
1257 | assert(s->nouts == 0); /* but have no outarcs */ |
1258 | |
1259 | s->tmp = NULL; /* we're done here */ |
1260 | } |
1261 | |
1262 | /* |
1263 | * dupnfa - duplicate sub-NFA |
1264 | * |
1265 | * Another recursive traversal, this time using tmp to point to duplicates |
1266 | * as well as mark already-seen states. (You knew there was a reason why |
1267 | * it's a state pointer, didn't you? :-)) |
1268 | */ |
1269 | static void |
1270 | dupnfa(struct nfa *nfa, |
1271 | struct state *start, /* duplicate of subNFA starting here */ |
1272 | struct state *stop, /* and stopping here */ |
1273 | struct state *from, /* stringing duplicate from here */ |
1274 | struct state *to) /* to here */ |
1275 | { |
1276 | if (start == stop) |
1277 | { |
1278 | newarc(nfa, EMPTY, 0, from, to); |
1279 | return; |
1280 | } |
1281 | |
1282 | stop->tmp = to; |
1283 | duptraverse(nfa, start, from); |
1284 | /* done, except for clearing out the tmp pointers */ |
1285 | |
1286 | stop->tmp = NULL; |
1287 | cleartraverse(nfa, start); |
1288 | } |
1289 | |
1290 | /* |
1291 | * duptraverse - recursive heart of dupnfa |
1292 | */ |
1293 | static void |
1294 | duptraverse(struct nfa *nfa, |
1295 | struct state *s, |
1296 | struct state *stmp) /* s's duplicate, or NULL */ |
1297 | { |
1298 | struct arc *a; |
1299 | |
1300 | /* Since this is recursive, it could be driven to stack overflow */ |
1301 | if (STACK_TOO_DEEP(nfa->v->re)) |
1302 | { |
1303 | NERR(REG_ETOOBIG); |
1304 | return; |
1305 | } |
1306 | |
1307 | if (s->tmp != NULL) |
1308 | return; /* already done */ |
1309 | |
1310 | s->tmp = (stmp == NULL) ? newstate(nfa) : stmp; |
1311 | if (s->tmp == NULL) |
1312 | { |
1313 | assert(NISERR()); |
1314 | return; |
1315 | } |
1316 | |
1317 | for (a = s->outs; a != NULL && !NISERR(); a = a->outchain) |
1318 | { |
1319 | duptraverse(nfa, a->to, (struct state *) NULL); |
1320 | if (NISERR()) |
1321 | break; |
1322 | assert(a->to->tmp != NULL); |
1323 | cparc(nfa, a, s->tmp, a->to->tmp); |
1324 | } |
1325 | } |
1326 | |
1327 | /* |
1328 | * cleartraverse - recursive cleanup for algorithms that leave tmp ptrs set |
1329 | */ |
1330 | static void |
1331 | cleartraverse(struct nfa *nfa, |
1332 | struct state *s) |
1333 | { |
1334 | struct arc *a; |
1335 | |
1336 | /* Since this is recursive, it could be driven to stack overflow */ |
1337 | if (STACK_TOO_DEEP(nfa->v->re)) |
1338 | { |
1339 | NERR(REG_ETOOBIG); |
1340 | return; |
1341 | } |
1342 | |
1343 | if (s->tmp == NULL) |
1344 | return; |
1345 | s->tmp = NULL; |
1346 | |
1347 | for (a = s->outs; a != NULL; a = a->outchain) |
1348 | cleartraverse(nfa, a->to); |
1349 | } |
1350 | |
1351 | /* |
1352 | * single_color_transition - does getting from s1 to s2 cross one PLAIN arc? |
1353 | * |
1354 | * If traversing from s1 to s2 requires a single PLAIN match (possibly of any |
1355 | * of a set of colors), return a state whose outarc list contains only PLAIN |
1356 | * arcs of those color(s). Otherwise return NULL. |
1357 | * |
1358 | * This is used before optimizing the NFA, so there may be EMPTY arcs, which |
1359 | * we should ignore; the possibility of an EMPTY is why the result state could |
1360 | * be different from s1. |
1361 | * |
1362 | * It's worth troubling to handle multiple parallel PLAIN arcs here because a |
1363 | * bracket construct such as [abc] might yield either one or several parallel |
1364 | * PLAIN arcs depending on earlier atoms in the expression. We'd rather that |
1365 | * that implementation detail not create user-visible performance differences. |
1366 | */ |
1367 | static struct state * |
1368 | single_color_transition(struct state *s1, struct state *s2) |
1369 | { |
1370 | struct arc *a; |
1371 | |
1372 | /* Ignore leading EMPTY arc, if any */ |
1373 | if (s1->nouts == 1 && s1->outs->type == EMPTY) |
1374 | s1 = s1->outs->to; |
1375 | /* Likewise for any trailing EMPTY arc */ |
1376 | if (s2->nins == 1 && s2->ins->type == EMPTY) |
1377 | s2 = s2->ins->from; |
1378 | /* Perhaps we could have a single-state loop in between, if so reject */ |
1379 | if (s1 == s2) |
1380 | return NULL; |
1381 | /* s1 must have at least one outarc... */ |
1382 | if (s1->outs == NULL) |
1383 | return NULL; |
1384 | /* ... and they must all be PLAIN arcs to s2 */ |
1385 | for (a = s1->outs; a != NULL; a = a->outchain) |
1386 | { |
1387 | if (a->type != PLAIN || a->to != s2) |
1388 | return NULL; |
1389 | } |
1390 | /* OK, return s1 as the possessor of the relevant outarcs */ |
1391 | return s1; |
1392 | } |
1393 | |
1394 | /* |
1395 | * specialcolors - fill in special colors for an NFA |
1396 | */ |
1397 | static void |
1398 | specialcolors(struct nfa *nfa) |
1399 | { |
1400 | /* false colors for BOS, BOL, EOS, EOL */ |
1401 | if (nfa->parent == NULL) |
1402 | { |
1403 | nfa->bos[0] = pseudocolor(nfa->cm); |
1404 | nfa->bos[1] = pseudocolor(nfa->cm); |
1405 | nfa->eos[0] = pseudocolor(nfa->cm); |
1406 | nfa->eos[1] = pseudocolor(nfa->cm); |
1407 | } |
1408 | else |
1409 | { |
1410 | assert(nfa->parent->bos[0] != COLORLESS); |
1411 | nfa->bos[0] = nfa->parent->bos[0]; |
1412 | assert(nfa->parent->bos[1] != COLORLESS); |
1413 | nfa->bos[1] = nfa->parent->bos[1]; |
1414 | assert(nfa->parent->eos[0] != COLORLESS); |
1415 | nfa->eos[0] = nfa->parent->eos[0]; |
1416 | assert(nfa->parent->eos[1] != COLORLESS); |
1417 | nfa->eos[1] = nfa->parent->eos[1]; |
1418 | } |
1419 | } |
1420 | |
1421 | /* |
1422 | * optimize - optimize an NFA |
1423 | * |
1424 | * The main goal of this function is not so much "optimization" (though it |
1425 | * does try to get rid of useless NFA states) as reducing the NFA to a form |
1426 | * the regex executor can handle. The executor, and indeed the cNFA format |
1427 | * that is its input, can only handle PLAIN and LACON arcs. The output of |
1428 | * the regex parser also includes EMPTY (do-nothing) arcs, as well as |
1429 | * ^, $, AHEAD, and BEHIND constraint arcs, which we must get rid of here. |
1430 | * We first get rid of EMPTY arcs and then deal with the constraint arcs. |
1431 | * The hardest part of either job is to get rid of circular loops of the |
1432 | * target arc type. We would have to do that in any case, though, as such a |
1433 | * loop would otherwise allow the executor to cycle through the loop endlessly |
1434 | * without making any progress in the input string. |
1435 | */ |
1436 | static long /* re_info bits */ |
1437 | optimize(struct nfa *nfa, |
1438 | FILE *f) /* for debug output; NULL none */ |
1439 | { |
1440 | #ifdef REG_DEBUG |
1441 | int verbose = (f != NULL) ? 1 : 0; |
1442 | |
1443 | if (verbose) |
1444 | fprintf(f, "\ninitial cleanup:\n" ); |
1445 | #endif |
1446 | cleanup(nfa); /* may simplify situation */ |
1447 | #ifdef REG_DEBUG |
1448 | if (verbose) |
1449 | dumpnfa(nfa, f); |
1450 | if (verbose) |
1451 | fprintf(f, "\nempties:\n" ); |
1452 | #endif |
1453 | fixempties(nfa, f); /* get rid of EMPTY arcs */ |
1454 | #ifdef REG_DEBUG |
1455 | if (verbose) |
1456 | fprintf(f, "\nconstraints:\n" ); |
1457 | #endif |
1458 | fixconstraintloops(nfa, f); /* get rid of constraint loops */ |
1459 | pullback(nfa, f); /* pull back constraints backward */ |
1460 | pushfwd(nfa, f); /* push fwd constraints forward */ |
1461 | #ifdef REG_DEBUG |
1462 | if (verbose) |
1463 | fprintf(f, "\nfinal cleanup:\n" ); |
1464 | #endif |
1465 | cleanup(nfa); /* final tidying */ |
1466 | #ifdef REG_DEBUG |
1467 | if (verbose) |
1468 | dumpnfa(nfa, f); |
1469 | #endif |
1470 | return analyze(nfa); /* and analysis */ |
1471 | } |
1472 | |
1473 | /* |
1474 | * pullback - pull back constraints backward to eliminate them |
1475 | */ |
1476 | static void |
1477 | pullback(struct nfa *nfa, |
1478 | FILE *f) /* for debug output; NULL none */ |
1479 | { |
1480 | struct state *s; |
1481 | struct state *nexts; |
1482 | struct arc *a; |
1483 | struct arc *nexta; |
1484 | struct state *intermediates; |
1485 | int progress; |
1486 | |
1487 | /* find and pull until there are no more */ |
1488 | do |
1489 | { |
1490 | progress = 0; |
1491 | for (s = nfa->states; s != NULL && !NISERR(); s = nexts) |
1492 | { |
1493 | nexts = s->next; |
1494 | intermediates = NULL; |
1495 | for (a = s->outs; a != NULL && !NISERR(); a = nexta) |
1496 | { |
1497 | nexta = a->outchain; |
1498 | if (a->type == '^' || a->type == BEHIND) |
1499 | if (pull(nfa, a, &intermediates)) |
1500 | progress = 1; |
1501 | } |
1502 | /* clear tmp fields of intermediate states created here */ |
1503 | while (intermediates != NULL) |
1504 | { |
1505 | struct state *ns = intermediates->tmp; |
1506 | |
1507 | intermediates->tmp = NULL; |
1508 | intermediates = ns; |
1509 | } |
1510 | /* if s is now useless, get rid of it */ |
1511 | if ((s->nins == 0 || s->nouts == 0) && !s->flag) |
1512 | dropstate(nfa, s); |
1513 | } |
1514 | if (progress && f != NULL) |
1515 | dumpnfa(nfa, f); |
1516 | } while (progress && !NISERR()); |
1517 | if (NISERR()) |
1518 | return; |
1519 | |
1520 | /* |
1521 | * Any ^ constraints we were able to pull to the start state can now be |
1522 | * replaced by PLAIN arcs referencing the BOS or BOL colors. There should |
1523 | * be no other ^ or BEHIND arcs left in the NFA, though we do not check |
1524 | * that here (compact() will fail if so). |
1525 | */ |
1526 | for (a = nfa->pre->outs; a != NULL; a = nexta) |
1527 | { |
1528 | nexta = a->outchain; |
1529 | if (a->type == '^') |
1530 | { |
1531 | assert(a->co == 0 || a->co == 1); |
1532 | newarc(nfa, PLAIN, nfa->bos[a->co], a->from, a->to); |
1533 | freearc(nfa, a); |
1534 | } |
1535 | } |
1536 | } |
1537 | |
1538 | /* |
1539 | * pull - pull a back constraint backward past its source state |
1540 | * |
1541 | * Returns 1 if successful (which it always is unless the source is the |
1542 | * start state or we have an internal error), 0 if nothing happened. |
1543 | * |
1544 | * A significant property of this function is that it deletes no pre-existing |
1545 | * states, and no outarcs of the constraint's from state other than the given |
1546 | * constraint arc. This makes the loops in pullback() safe, at the cost that |
1547 | * we may leave useless states behind. Therefore, we leave it to pullback() |
1548 | * to delete such states. |
1549 | * |
1550 | * If the from state has multiple back-constraint outarcs, and/or multiple |
1551 | * compatible constraint inarcs, we only need to create one new intermediate |
1552 | * state per combination of predecessor and successor states. *intermediates |
1553 | * points to a list of such intermediate states for this from state (chained |
1554 | * through their tmp fields). |
1555 | */ |
1556 | static int |
1557 | pull(struct nfa *nfa, |
1558 | struct arc *con, |
1559 | struct state **intermediates) |
1560 | { |
1561 | struct state *from = con->from; |
1562 | struct state *to = con->to; |
1563 | struct arc *a; |
1564 | struct arc *nexta; |
1565 | struct state *s; |
1566 | |
1567 | assert(from != to); /* should have gotten rid of this earlier */ |
1568 | if (from->flag) /* can't pull back beyond start */ |
1569 | return 0; |
1570 | if (from->nins == 0) |
1571 | { /* unreachable */ |
1572 | freearc(nfa, con); |
1573 | return 1; |
1574 | } |
1575 | |
1576 | /* |
1577 | * First, clone from state if necessary to avoid other outarcs. This may |
1578 | * seem wasteful, but it simplifies the logic, and we'll get rid of the |
1579 | * clone state again at the bottom. |
1580 | */ |
1581 | if (from->nouts > 1) |
1582 | { |
1583 | s = newstate(nfa); |
1584 | if (NISERR()) |
1585 | return 0; |
1586 | copyins(nfa, from, s); /* duplicate inarcs */ |
1587 | cparc(nfa, con, s, to); /* move constraint arc */ |
1588 | freearc(nfa, con); |
1589 | if (NISERR()) |
1590 | return 0; |
1591 | from = s; |
1592 | con = from->outs; |
1593 | } |
1594 | assert(from->nouts == 1); |
1595 | |
1596 | /* propagate the constraint into the from state's inarcs */ |
1597 | for (a = from->ins; a != NULL && !NISERR(); a = nexta) |
1598 | { |
1599 | nexta = a->inchain; |
1600 | switch (combine(con, a)) |
1601 | { |
1602 | case INCOMPATIBLE: /* destroy the arc */ |
1603 | freearc(nfa, a); |
1604 | break; |
1605 | case SATISFIED: /* no action needed */ |
1606 | break; |
1607 | case COMPATIBLE: /* swap the two arcs, more or less */ |
1608 | /* need an intermediate state, but might have one already */ |
1609 | for (s = *intermediates; s != NULL; s = s->tmp) |
1610 | { |
1611 | assert(s->nins > 0 && s->nouts > 0); |
1612 | if (s->ins->from == a->from && s->outs->to == to) |
1613 | break; |
1614 | } |
1615 | if (s == NULL) |
1616 | { |
1617 | s = newstate(nfa); |
1618 | if (NISERR()) |
1619 | return 0; |
1620 | s->tmp = *intermediates; |
1621 | *intermediates = s; |
1622 | } |
1623 | cparc(nfa, con, a->from, s); |
1624 | cparc(nfa, a, s, to); |
1625 | freearc(nfa, a); |
1626 | break; |
1627 | default: |
1628 | assert(NOTREACHED); |
1629 | break; |
1630 | } |
1631 | } |
1632 | |
1633 | /* remaining inarcs, if any, incorporate the constraint */ |
1634 | moveins(nfa, from, to); |
1635 | freearc(nfa, con); |
1636 | /* from state is now useless, but we leave it to pullback() to clean up */ |
1637 | return 1; |
1638 | } |
1639 | |
1640 | /* |
1641 | * pushfwd - push forward constraints forward to eliminate them |
1642 | */ |
1643 | static void |
1644 | pushfwd(struct nfa *nfa, |
1645 | FILE *f) /* for debug output; NULL none */ |
1646 | { |
1647 | struct state *s; |
1648 | struct state *nexts; |
1649 | struct arc *a; |
1650 | struct arc *nexta; |
1651 | struct state *intermediates; |
1652 | int progress; |
1653 | |
1654 | /* find and push until there are no more */ |
1655 | do |
1656 | { |
1657 | progress = 0; |
1658 | for (s = nfa->states; s != NULL && !NISERR(); s = nexts) |
1659 | { |
1660 | nexts = s->next; |
1661 | intermediates = NULL; |
1662 | for (a = s->ins; a != NULL && !NISERR(); a = nexta) |
1663 | { |
1664 | nexta = a->inchain; |
1665 | if (a->type == '$' || a->type == AHEAD) |
1666 | if (push(nfa, a, &intermediates)) |
1667 | progress = 1; |
1668 | } |
1669 | /* clear tmp fields of intermediate states created here */ |
1670 | while (intermediates != NULL) |
1671 | { |
1672 | struct state *ns = intermediates->tmp; |
1673 | |
1674 | intermediates->tmp = NULL; |
1675 | intermediates = ns; |
1676 | } |
1677 | /* if s is now useless, get rid of it */ |
1678 | if ((s->nins == 0 || s->nouts == 0) && !s->flag) |
1679 | dropstate(nfa, s); |
1680 | } |
1681 | if (progress && f != NULL) |
1682 | dumpnfa(nfa, f); |
1683 | } while (progress && !NISERR()); |
1684 | if (NISERR()) |
1685 | return; |
1686 | |
1687 | /* |
1688 | * Any $ constraints we were able to push to the post state can now be |
1689 | * replaced by PLAIN arcs referencing the EOS or EOL colors. There should |
1690 | * be no other $ or AHEAD arcs left in the NFA, though we do not check |
1691 | * that here (compact() will fail if so). |
1692 | */ |
1693 | for (a = nfa->post->ins; a != NULL; a = nexta) |
1694 | { |
1695 | nexta = a->inchain; |
1696 | if (a->type == '$') |
1697 | { |
1698 | assert(a->co == 0 || a->co == 1); |
1699 | newarc(nfa, PLAIN, nfa->eos[a->co], a->from, a->to); |
1700 | freearc(nfa, a); |
1701 | } |
1702 | } |
1703 | } |
1704 | |
1705 | /* |
1706 | * push - push a forward constraint forward past its destination state |
1707 | * |
1708 | * Returns 1 if successful (which it always is unless the destination is the |
1709 | * post state or we have an internal error), 0 if nothing happened. |
1710 | * |
1711 | * A significant property of this function is that it deletes no pre-existing |
1712 | * states, and no inarcs of the constraint's to state other than the given |
1713 | * constraint arc. This makes the loops in pushfwd() safe, at the cost that |
1714 | * we may leave useless states behind. Therefore, we leave it to pushfwd() |
1715 | * to delete such states. |
1716 | * |
1717 | * If the to state has multiple forward-constraint inarcs, and/or multiple |
1718 | * compatible constraint outarcs, we only need to create one new intermediate |
1719 | * state per combination of predecessor and successor states. *intermediates |
1720 | * points to a list of such intermediate states for this to state (chained |
1721 | * through their tmp fields). |
1722 | */ |
1723 | static int |
1724 | push(struct nfa *nfa, |
1725 | struct arc *con, |
1726 | struct state **intermediates) |
1727 | { |
1728 | struct state *from = con->from; |
1729 | struct state *to = con->to; |
1730 | struct arc *a; |
1731 | struct arc *nexta; |
1732 | struct state *s; |
1733 | |
1734 | assert(to != from); /* should have gotten rid of this earlier */ |
1735 | if (to->flag) /* can't push forward beyond end */ |
1736 | return 0; |
1737 | if (to->nouts == 0) |
1738 | { /* dead end */ |
1739 | freearc(nfa, con); |
1740 | return 1; |
1741 | } |
1742 | |
1743 | /* |
1744 | * First, clone to state if necessary to avoid other inarcs. This may |
1745 | * seem wasteful, but it simplifies the logic, and we'll get rid of the |
1746 | * clone state again at the bottom. |
1747 | */ |
1748 | if (to->nins > 1) |
1749 | { |
1750 | s = newstate(nfa); |
1751 | if (NISERR()) |
1752 | return 0; |
1753 | copyouts(nfa, to, s); /* duplicate outarcs */ |
1754 | cparc(nfa, con, from, s); /* move constraint arc */ |
1755 | freearc(nfa, con); |
1756 | if (NISERR()) |
1757 | return 0; |
1758 | to = s; |
1759 | con = to->ins; |
1760 | } |
1761 | assert(to->nins == 1); |
1762 | |
1763 | /* propagate the constraint into the to state's outarcs */ |
1764 | for (a = to->outs; a != NULL && !NISERR(); a = nexta) |
1765 | { |
1766 | nexta = a->outchain; |
1767 | switch (combine(con, a)) |
1768 | { |
1769 | case INCOMPATIBLE: /* destroy the arc */ |
1770 | freearc(nfa, a); |
1771 | break; |
1772 | case SATISFIED: /* no action needed */ |
1773 | break; |
1774 | case COMPATIBLE: /* swap the two arcs, more or less */ |
1775 | /* need an intermediate state, but might have one already */ |
1776 | for (s = *intermediates; s != NULL; s = s->tmp) |
1777 | { |
1778 | assert(s->nins > 0 && s->nouts > 0); |
1779 | if (s->ins->from == from && s->outs->to == a->to) |
1780 | break; |
1781 | } |
1782 | if (s == NULL) |
1783 | { |
1784 | s = newstate(nfa); |
1785 | if (NISERR()) |
1786 | return 0; |
1787 | s->tmp = *intermediates; |
1788 | *intermediates = s; |
1789 | } |
1790 | cparc(nfa, con, s, a->to); |
1791 | cparc(nfa, a, from, s); |
1792 | freearc(nfa, a); |
1793 | break; |
1794 | default: |
1795 | assert(NOTREACHED); |
1796 | break; |
1797 | } |
1798 | } |
1799 | |
1800 | /* remaining outarcs, if any, incorporate the constraint */ |
1801 | moveouts(nfa, to, from); |
1802 | freearc(nfa, con); |
1803 | /* to state is now useless, but we leave it to pushfwd() to clean up */ |
1804 | return 1; |
1805 | } |
1806 | |
1807 | /* |
1808 | * combine - constraint lands on an arc, what happens? |
1809 | * |
1810 | * #def INCOMPATIBLE 1 // destroys arc |
1811 | * #def SATISFIED 2 // constraint satisfied |
1812 | * #def COMPATIBLE 3 // compatible but not satisfied yet |
1813 | */ |
1814 | static int |
1815 | combine(struct arc *con, |
1816 | struct arc *a) |
1817 | { |
1818 | #define CA(ct,at) (((ct)<<CHAR_BIT) | (at)) |
1819 | |
1820 | switch (CA(con->type, a->type)) |
1821 | { |
1822 | case CA('^', PLAIN): /* newlines are handled separately */ |
1823 | case CA('$', PLAIN): |
1824 | return INCOMPATIBLE; |
1825 | break; |
1826 | case CA(AHEAD, PLAIN): /* color constraints meet colors */ |
1827 | case CA(BEHIND, PLAIN): |
1828 | if (con->co == a->co) |
1829 | return SATISFIED; |
1830 | return INCOMPATIBLE; |
1831 | break; |
1832 | case CA('^', '^'): /* collision, similar constraints */ |
1833 | case CA('$', '$'): |
1834 | case CA(AHEAD, AHEAD): |
1835 | case CA(BEHIND, BEHIND): |
1836 | if (con->co == a->co) /* true duplication */ |
1837 | return SATISFIED; |
1838 | return INCOMPATIBLE; |
1839 | break; |
1840 | case CA('^', BEHIND): /* collision, dissimilar constraints */ |
1841 | case CA(BEHIND, '^'): |
1842 | case CA('$', AHEAD): |
1843 | case CA(AHEAD, '$'): |
1844 | return INCOMPATIBLE; |
1845 | break; |
1846 | case CA('^', '$'): /* constraints passing each other */ |
1847 | case CA('^', AHEAD): |
1848 | case CA(BEHIND, '$'): |
1849 | case CA(BEHIND, AHEAD): |
1850 | case CA('$', '^'): |
1851 | case CA('$', BEHIND): |
1852 | case CA(AHEAD, '^'): |
1853 | case CA(AHEAD, BEHIND): |
1854 | case CA('^', LACON): |
1855 | case CA(BEHIND, LACON): |
1856 | case CA('$', LACON): |
1857 | case CA(AHEAD, LACON): |
1858 | return COMPATIBLE; |
1859 | break; |
1860 | } |
1861 | assert(NOTREACHED); |
1862 | return INCOMPATIBLE; /* for benefit of blind compilers */ |
1863 | } |
1864 | |
1865 | /* |
1866 | * fixempties - get rid of EMPTY arcs |
1867 | */ |
1868 | static void |
1869 | fixempties(struct nfa *nfa, |
1870 | FILE *f) /* for debug output; NULL none */ |
1871 | { |
1872 | struct state *s; |
1873 | struct state *s2; |
1874 | struct state *nexts; |
1875 | struct arc *a; |
1876 | struct arc *nexta; |
1877 | int totalinarcs; |
1878 | struct arc **inarcsorig; |
1879 | struct arc **arcarray; |
1880 | int arccount; |
1881 | int prevnins; |
1882 | int nskip; |
1883 | |
1884 | /* |
1885 | * First, get rid of any states whose sole out-arc is an EMPTY, since |
1886 | * they're basically just aliases for their successor. The parsing |
1887 | * algorithm creates enough of these that it's worth special-casing this. |
1888 | */ |
1889 | for (s = nfa->states; s != NULL && !NISERR(); s = nexts) |
1890 | { |
1891 | nexts = s->next; |
1892 | if (s->flag || s->nouts != 1) |
1893 | continue; |
1894 | a = s->outs; |
1895 | assert(a != NULL && a->outchain == NULL); |
1896 | if (a->type != EMPTY) |
1897 | continue; |
1898 | if (s != a->to) |
1899 | moveins(nfa, s, a->to); |
1900 | dropstate(nfa, s); |
1901 | } |
1902 | |
1903 | /* |
1904 | * Similarly, get rid of any state with a single EMPTY in-arc, by folding |
1905 | * it into its predecessor. |
1906 | */ |
1907 | for (s = nfa->states; s != NULL && !NISERR(); s = nexts) |
1908 | { |
1909 | nexts = s->next; |
1910 | /* while we're at it, ensure tmp fields are clear for next step */ |
1911 | assert(s->tmp == NULL); |
1912 | if (s->flag || s->nins != 1) |
1913 | continue; |
1914 | a = s->ins; |
1915 | assert(a != NULL && a->inchain == NULL); |
1916 | if (a->type != EMPTY) |
1917 | continue; |
1918 | if (s != a->from) |
1919 | moveouts(nfa, s, a->from); |
1920 | dropstate(nfa, s); |
1921 | } |
1922 | |
1923 | if (NISERR()) |
1924 | return; |
1925 | |
1926 | /* |
1927 | * For each remaining NFA state, find all other states from which it is |
1928 | * reachable by a chain of one or more EMPTY arcs. Then generate new arcs |
1929 | * that eliminate the need for each such chain. |
1930 | * |
1931 | * We could replace a chain of EMPTY arcs that leads from a "from" state |
1932 | * to a "to" state either by pushing non-EMPTY arcs forward (linking |
1933 | * directly from "from"'s predecessors to "to") or by pulling them back |
1934 | * (linking directly from "from" to "to"'s successors). We choose to |
1935 | * always do the former; this choice is somewhat arbitrary, but the |
1936 | * approach below requires that we uniformly do one or the other. |
1937 | * |
1938 | * Suppose we have a chain of N successive EMPTY arcs (where N can easily |
1939 | * approach the size of the NFA). All of the intermediate states must |
1940 | * have additional inarcs and outarcs, else they'd have been removed by |
1941 | * the steps above. Assuming their inarcs are mostly not empties, we will |
1942 | * add O(N^2) arcs to the NFA, since a non-EMPTY inarc leading to any one |
1943 | * state in the chain must be duplicated to lead to all its successor |
1944 | * states as well. So there is no hope of doing less than O(N^2) work; |
1945 | * however, we should endeavor to keep the big-O cost from being even |
1946 | * worse than that, which it can easily become without care. In |
1947 | * particular, suppose we were to copy all S1's inarcs forward to S2, and |
1948 | * then also to S3, and then later we consider pushing S2's inarcs forward |
1949 | * to S3. If we include the arcs already copied from S1 in that, we'd be |
1950 | * doing O(N^3) work. (The duplicate-arc elimination built into newarc() |
1951 | * and its cohorts would get rid of the extra arcs, but not without cost.) |
1952 | * |
1953 | * We can avoid this cost by treating only arcs that existed at the start |
1954 | * of this phase as candidates to be pushed forward. To identify those, |
1955 | * we remember the first inarc each state had to start with. We rely on |
1956 | * the fact that newarc() and friends put new arcs on the front of their |
1957 | * to-states' inchains, and that this phase never deletes arcs, so that |
1958 | * the original arcs must be the last arcs in their to-states' inchains. |
1959 | * |
1960 | * So the process here is that, for each state in the NFA, we gather up |
1961 | * all non-EMPTY inarcs of states that can reach the target state via |
1962 | * EMPTY arcs. We then sort, de-duplicate, and merge these arcs into the |
1963 | * target state's inchain. (We can safely use sort-merge for this as long |
1964 | * as we update each state's original-arcs pointer after we add arcs to |
1965 | * it; the sort step of mergeins probably changed the order of the old |
1966 | * arcs.) |
1967 | * |
1968 | * Another refinement worth making is that, because we only add non-EMPTY |
1969 | * arcs during this phase, and all added arcs have the same from-state as |
1970 | * the non-EMPTY arc they were cloned from, we know ahead of time that any |
1971 | * states having only EMPTY outarcs will be useless for lack of outarcs |
1972 | * after we drop the EMPTY arcs. (They cannot gain non-EMPTY outarcs if |
1973 | * they had none to start with.) So we need not bother to update the |
1974 | * inchains of such states at all. |
1975 | */ |
1976 | |
1977 | /* Remember the states' first original inarcs */ |
1978 | /* ... and while at it, count how many old inarcs there are altogether */ |
1979 | inarcsorig = (struct arc **) MALLOC(nfa->nstates * sizeof(struct arc *)); |
1980 | if (inarcsorig == NULL) |
1981 | { |
1982 | NERR(REG_ESPACE); |
1983 | return; |
1984 | } |
1985 | totalinarcs = 0; |
1986 | for (s = nfa->states; s != NULL; s = s->next) |
1987 | { |
1988 | inarcsorig[s->no] = s->ins; |
1989 | totalinarcs += s->nins; |
1990 | } |
1991 | |
1992 | /* |
1993 | * Create a workspace for accumulating the inarcs to be added to the |
1994 | * current target state. totalinarcs is probably a considerable |
1995 | * overestimate of the space needed, but the NFA is unlikely to be large |
1996 | * enough at this point to make it worth being smarter. |
1997 | */ |
1998 | arcarray = (struct arc **) MALLOC(totalinarcs * sizeof(struct arc *)); |
1999 | if (arcarray == NULL) |
2000 | { |
2001 | NERR(REG_ESPACE); |
2002 | FREE(inarcsorig); |
2003 | return; |
2004 | } |
2005 | |
2006 | /* And iterate over the target states */ |
2007 | for (s = nfa->states; s != NULL && !NISERR(); s = s->next) |
2008 | { |
2009 | /* Ignore target states without non-EMPTY outarcs, per note above */ |
2010 | if (!s->flag && !hasnonemptyout(s)) |
2011 | continue; |
2012 | |
2013 | /* Find predecessor states and accumulate their original inarcs */ |
2014 | arccount = 0; |
2015 | for (s2 = emptyreachable(nfa, s, s, inarcsorig); s2 != s; s2 = nexts) |
2016 | { |
2017 | /* Add s2's original inarcs to arcarray[], but ignore empties */ |
2018 | for (a = inarcsorig[s2->no]; a != NULL; a = a->inchain) |
2019 | { |
2020 | if (a->type != EMPTY) |
2021 | arcarray[arccount++] = a; |
2022 | } |
2023 | |
2024 | /* Reset the tmp fields as we walk back */ |
2025 | nexts = s2->tmp; |
2026 | s2->tmp = NULL; |
2027 | } |
2028 | s->tmp = NULL; |
2029 | assert(arccount <= totalinarcs); |
2030 | |
2031 | /* Remember how many original inarcs this state has */ |
2032 | prevnins = s->nins; |
2033 | |
2034 | /* Add non-duplicate inarcs to target state */ |
2035 | mergeins(nfa, s, arcarray, arccount); |
2036 | |
2037 | /* Now we must update the state's inarcsorig pointer */ |
2038 | nskip = s->nins - prevnins; |
2039 | a = s->ins; |
2040 | while (nskip-- > 0) |
2041 | a = a->inchain; |
2042 | inarcsorig[s->no] = a; |
2043 | } |
2044 | |
2045 | FREE(arcarray); |
2046 | FREE(inarcsorig); |
2047 | |
2048 | if (NISERR()) |
2049 | return; |
2050 | |
2051 | /* |
2052 | * Now remove all the EMPTY arcs, since we don't need them anymore. |
2053 | */ |
2054 | for (s = nfa->states; s != NULL; s = s->next) |
2055 | { |
2056 | for (a = s->outs; a != NULL; a = nexta) |
2057 | { |
2058 | nexta = a->outchain; |
2059 | if (a->type == EMPTY) |
2060 | freearc(nfa, a); |
2061 | } |
2062 | } |
2063 | |
2064 | /* |
2065 | * And remove any states that have become useless. (This cleanup is not |
2066 | * very thorough, and would be even less so if we tried to combine it with |
2067 | * the previous step; but cleanup() will take care of anything we miss.) |
2068 | */ |
2069 | for (s = nfa->states; s != NULL; s = nexts) |
2070 | { |
2071 | nexts = s->next; |
2072 | if ((s->nins == 0 || s->nouts == 0) && !s->flag) |
2073 | dropstate(nfa, s); |
2074 | } |
2075 | |
2076 | if (f != NULL) |
2077 | dumpnfa(nfa, f); |
2078 | } |
2079 | |
2080 | /* |
2081 | * emptyreachable - recursively find all states that can reach s by EMPTY arcs |
2082 | * |
2083 | * The return value is the last such state found. Its tmp field links back |
2084 | * to the next-to-last such state, and so on back to s, so that all these |
2085 | * states can be located without searching the whole NFA. |
2086 | * |
2087 | * Since this is only used in fixempties(), we pass in the inarcsorig[] array |
2088 | * maintained by that function. This lets us skip over all new inarcs, which |
2089 | * are certainly not EMPTY arcs. |
2090 | * |
2091 | * The maximum recursion depth here is equal to the length of the longest |
2092 | * loop-free chain of EMPTY arcs, which is surely no more than the size of |
2093 | * the NFA ... but that could still be enough to cause trouble. |
2094 | */ |
2095 | static struct state * |
2096 | emptyreachable(struct nfa *nfa, |
2097 | struct state *s, |
2098 | struct state *lastfound, |
2099 | struct arc **inarcsorig) |
2100 | { |
2101 | struct arc *a; |
2102 | |
2103 | /* Since this is recursive, it could be driven to stack overflow */ |
2104 | if (STACK_TOO_DEEP(nfa->v->re)) |
2105 | { |
2106 | NERR(REG_ETOOBIG); |
2107 | return lastfound; |
2108 | } |
2109 | |
2110 | s->tmp = lastfound; |
2111 | lastfound = s; |
2112 | for (a = inarcsorig[s->no]; a != NULL; a = a->inchain) |
2113 | { |
2114 | if (a->type == EMPTY && a->from->tmp == NULL) |
2115 | lastfound = emptyreachable(nfa, a->from, lastfound, inarcsorig); |
2116 | } |
2117 | return lastfound; |
2118 | } |
2119 | |
2120 | /* |
2121 | * isconstraintarc - detect whether an arc is of a constraint type |
2122 | */ |
2123 | static inline int |
2124 | isconstraintarc(struct arc *a) |
2125 | { |
2126 | switch (a->type) |
2127 | { |
2128 | case '^': |
2129 | case '$': |
2130 | case BEHIND: |
2131 | case AHEAD: |
2132 | case LACON: |
2133 | return 1; |
2134 | } |
2135 | return 0; |
2136 | } |
2137 | |
2138 | /* |
2139 | * hasconstraintout - does state have a constraint out arc? |
2140 | */ |
2141 | static int |
2142 | hasconstraintout(struct state *s) |
2143 | { |
2144 | struct arc *a; |
2145 | |
2146 | for (a = s->outs; a != NULL; a = a->outchain) |
2147 | { |
2148 | if (isconstraintarc(a)) |
2149 | return 1; |
2150 | } |
2151 | return 0; |
2152 | } |
2153 | |
2154 | /* |
2155 | * fixconstraintloops - get rid of loops containing only constraint arcs |
2156 | * |
2157 | * A loop of states that contains only constraint arcs is useless, since |
2158 | * passing around the loop represents no forward progress. Moreover, it |
2159 | * would cause infinite looping in pullback/pushfwd, so we need to get rid |
2160 | * of such loops before doing that. |
2161 | */ |
2162 | static void |
2163 | fixconstraintloops(struct nfa *nfa, |
2164 | FILE *f) /* for debug output; NULL none */ |
2165 | { |
2166 | struct state *s; |
2167 | struct state *nexts; |
2168 | struct arc *a; |
2169 | struct arc *nexta; |
2170 | int hasconstraints; |
2171 | |
2172 | /* |
2173 | * In the trivial case of a state that loops to itself, we can just drop |
2174 | * the constraint arc altogether. This is worth special-casing because |
2175 | * such loops are far more common than loops containing multiple states. |
2176 | * While we're at it, note whether any constraint arcs survive. |
2177 | */ |
2178 | hasconstraints = 0; |
2179 | for (s = nfa->states; s != NULL && !NISERR(); s = nexts) |
2180 | { |
2181 | nexts = s->next; |
2182 | /* while we're at it, ensure tmp fields are clear for next step */ |
2183 | assert(s->tmp == NULL); |
2184 | for (a = s->outs; a != NULL && !NISERR(); a = nexta) |
2185 | { |
2186 | nexta = a->outchain; |
2187 | if (isconstraintarc(a)) |
2188 | { |
2189 | if (a->to == s) |
2190 | freearc(nfa, a); |
2191 | else |
2192 | hasconstraints = 1; |
2193 | } |
2194 | } |
2195 | /* If we removed all the outarcs, the state is useless. */ |
2196 | if (s->nouts == 0 && !s->flag) |
2197 | dropstate(nfa, s); |
2198 | } |
2199 | |
2200 | /* Nothing to do if no remaining constraint arcs */ |
2201 | if (NISERR() || !hasconstraints) |
2202 | return; |
2203 | |
2204 | /* |
2205 | * Starting from each remaining NFA state, search outwards for a |
2206 | * constraint loop. If we find a loop, break the loop, then start the |
2207 | * search over. (We could possibly retain some state from the first scan, |
2208 | * but it would complicate things greatly, and multi-state constraint |
2209 | * loops are rare enough that it's not worth optimizing the case.) |
2210 | */ |
2211 | restart: |
2212 | for (s = nfa->states; s != NULL && !NISERR(); s = s->next) |
2213 | { |
2214 | if (findconstraintloop(nfa, s)) |
2215 | goto restart; |
2216 | } |
2217 | |
2218 | if (NISERR()) |
2219 | return; |
2220 | |
2221 | /* |
2222 | * Now remove any states that have become useless. (This cleanup is not |
2223 | * very thorough, and would be even less so if we tried to combine it with |
2224 | * the previous step; but cleanup() will take care of anything we miss.) |
2225 | * |
2226 | * Because findconstraintloop intentionally doesn't reset all tmp fields, |
2227 | * we have to clear them after it's done. This is a convenient place to |
2228 | * do that, too. |
2229 | */ |
2230 | for (s = nfa->states; s != NULL; s = nexts) |
2231 | { |
2232 | nexts = s->next; |
2233 | s->tmp = NULL; |
2234 | if ((s->nins == 0 || s->nouts == 0) && !s->flag) |
2235 | dropstate(nfa, s); |
2236 | } |
2237 | |
2238 | if (f != NULL) |
2239 | dumpnfa(nfa, f); |
2240 | } |
2241 | |
2242 | /* |
2243 | * findconstraintloop - recursively find a loop of constraint arcs |
2244 | * |
2245 | * If we find a loop, break it by calling breakconstraintloop(), then |
2246 | * return 1; otherwise return 0. |
2247 | * |
2248 | * State tmp fields are guaranteed all NULL on a success return, because |
2249 | * breakconstraintloop does that. After a failure return, any state that |
2250 | * is known not to be part of a loop is marked with s->tmp == s; this allows |
2251 | * us not to have to re-prove that fact on later calls. (This convention is |
2252 | * workable because we already eliminated single-state loops.) |
2253 | * |
2254 | * Note that the found loop doesn't necessarily include the first state we |
2255 | * are called on. Any loop reachable from that state will do. |
2256 | * |
2257 | * The maximum recursion depth here is one more than the length of the longest |
2258 | * loop-free chain of constraint arcs, which is surely no more than the size |
2259 | * of the NFA ... but that could still be enough to cause trouble. |
2260 | */ |
2261 | static int |
2262 | findconstraintloop(struct nfa *nfa, struct state *s) |
2263 | { |
2264 | struct arc *a; |
2265 | |
2266 | /* Since this is recursive, it could be driven to stack overflow */ |
2267 | if (STACK_TOO_DEEP(nfa->v->re)) |
2268 | { |
2269 | NERR(REG_ETOOBIG); |
2270 | return 1; /* to exit as quickly as possible */ |
2271 | } |
2272 | |
2273 | if (s->tmp != NULL) |
2274 | { |
2275 | /* Already proven uninteresting? */ |
2276 | if (s->tmp == s) |
2277 | return 0; |
2278 | /* Found a loop involving s */ |
2279 | breakconstraintloop(nfa, s); |
2280 | /* The tmp fields have been cleaned up by breakconstraintloop */ |
2281 | return 1; |
2282 | } |
2283 | for (a = s->outs; a != NULL; a = a->outchain) |
2284 | { |
2285 | if (isconstraintarc(a)) |
2286 | { |
2287 | struct state *sto = a->to; |
2288 | |
2289 | assert(sto != s); |
2290 | s->tmp = sto; |
2291 | if (findconstraintloop(nfa, sto)) |
2292 | return 1; |
2293 | } |
2294 | } |
2295 | |
2296 | /* |
2297 | * If we get here, no constraint loop exists leading out from s. Mark it |
2298 | * with s->tmp == s so we need not rediscover that fact again later. |
2299 | */ |
2300 | s->tmp = s; |
2301 | return 0; |
2302 | } |
2303 | |
2304 | /* |
2305 | * breakconstraintloop - break a loop of constraint arcs |
2306 | * |
2307 | * sinitial is any one member state of the loop. Each loop member's tmp |
2308 | * field links to its successor within the loop. (Note that this function |
2309 | * will reset all the tmp fields to NULL.) |
2310 | * |
2311 | * We can break the loop by, for any one state S1 in the loop, cloning its |
2312 | * loop successor state S2 (and possibly following states), and then moving |
2313 | * all S1->S2 constraint arcs to point to the cloned S2. The cloned S2 should |
2314 | * copy any non-constraint outarcs of S2. Constraint outarcs should be |
2315 | * dropped if they point back to S1, else they need to be copied as arcs to |
2316 | * similarly cloned states S3, S4, etc. In general, each cloned state copies |
2317 | * non-constraint outarcs, drops constraint outarcs that would lead to itself |
2318 | * or any earlier cloned state, and sends other constraint outarcs to newly |
2319 | * cloned states. No cloned state will have any inarcs that aren't constraint |
2320 | * arcs or do not lead from S1 or earlier-cloned states. It's okay to drop |
2321 | * constraint back-arcs since they would not take us to any state we've not |
2322 | * already been in; therefore, no new constraint loop is created. In this way |
2323 | * we generate a modified NFA that can still represent every useful state |
2324 | * sequence, but not sequences that represent state loops with no consumption |
2325 | * of input data. Note that the set of cloned states will certainly include |
2326 | * all of the loop member states other than S1, and it may also include |
2327 | * non-loop states that are reachable from S2 via constraint arcs. This is |
2328 | * important because there is no guarantee that findconstraintloop found a |
2329 | * maximal loop (and searching for one would be NP-hard, so don't try). |
2330 | * Frequently the "non-loop states" are actually part of a larger loop that |
2331 | * we didn't notice, and indeed there may be several overlapping loops. |
2332 | * This technique ensures convergence in such cases, while considering only |
2333 | * the originally-found loop does not. |
2334 | * |
2335 | * If there is only one S1->S2 constraint arc, then that constraint is |
2336 | * certainly satisfied when we enter any of the clone states. This means that |
2337 | * in the common case where many of the constraint arcs are identically |
2338 | * labeled, we can merge together clone states linked by a similarly-labeled |
2339 | * constraint: if we can get to the first one we can certainly get to the |
2340 | * second, so there's no need to distinguish. This greatly reduces the number |
2341 | * of new states needed, so we preferentially break the given loop at a state |
2342 | * pair where this is true. |
2343 | * |
2344 | * Furthermore, it's fairly common to find that a cloned successor state has |
2345 | * no outarcs, especially if we're a bit aggressive about removing unnecessary |
2346 | * outarcs. If that happens, then there is simply not any interesting state |
2347 | * that can be reached through the predecessor's loop arcs, which means we can |
2348 | * break the loop just by removing those loop arcs, with no new states added. |
2349 | */ |
2350 | static void |
2351 | breakconstraintloop(struct nfa *nfa, struct state *sinitial) |
2352 | { |
2353 | struct state *s; |
2354 | struct state *shead; |
2355 | struct state *stail; |
2356 | struct state *sclone; |
2357 | struct state *nexts; |
2358 | struct arc *refarc; |
2359 | struct arc *a; |
2360 | struct arc *nexta; |
2361 | |
2362 | /* |
2363 | * Start by identifying which loop step we want to break at. |
2364 | * Preferentially this is one with only one constraint arc. (XXX are |
2365 | * there any other secondary heuristics we want to use here?) Set refarc |
2366 | * to point to the selected lone constraint arc, if there is one. |
2367 | */ |
2368 | refarc = NULL; |
2369 | s = sinitial; |
2370 | do |
2371 | { |
2372 | nexts = s->tmp; |
2373 | assert(nexts != s); /* should not see any one-element loops */ |
2374 | if (refarc == NULL) |
2375 | { |
2376 | int narcs = 0; |
2377 | |
2378 | for (a = s->outs; a != NULL; a = a->outchain) |
2379 | { |
2380 | if (a->to == nexts && isconstraintarc(a)) |
2381 | { |
2382 | refarc = a; |
2383 | narcs++; |
2384 | } |
2385 | } |
2386 | assert(narcs > 0); |
2387 | if (narcs > 1) |
2388 | refarc = NULL; /* multiple constraint arcs here, no good */ |
2389 | } |
2390 | s = nexts; |
2391 | } while (s != sinitial); |
2392 | |
2393 | if (refarc) |
2394 | { |
2395 | /* break at the refarc */ |
2396 | shead = refarc->from; |
2397 | stail = refarc->to; |
2398 | assert(stail == shead->tmp); |
2399 | } |
2400 | else |
2401 | { |
2402 | /* for lack of a better idea, break after sinitial */ |
2403 | shead = sinitial; |
2404 | stail = sinitial->tmp; |
2405 | } |
2406 | |
2407 | /* |
2408 | * Reset the tmp fields so that we can use them for local storage in |
2409 | * clonesuccessorstates. (findconstraintloop won't mind, since it's just |
2410 | * going to abandon its search anyway.) |
2411 | */ |
2412 | for (s = nfa->states; s != NULL; s = s->next) |
2413 | s->tmp = NULL; |
2414 | |
2415 | /* |
2416 | * Recursively build clone state(s) as needed. |
2417 | */ |
2418 | sclone = newstate(nfa); |
2419 | if (sclone == NULL) |
2420 | { |
2421 | assert(NISERR()); |
2422 | return; |
2423 | } |
2424 | |
2425 | clonesuccessorstates(nfa, stail, sclone, shead, refarc, |
2426 | NULL, NULL, nfa->nstates); |
2427 | |
2428 | if (NISERR()) |
2429 | return; |
2430 | |
2431 | /* |
2432 | * It's possible that sclone has no outarcs at all, in which case it's |
2433 | * useless. (We don't try extremely hard to get rid of useless states |
2434 | * here, but this is an easy and fairly common case.) |
2435 | */ |
2436 | if (sclone->nouts == 0) |
2437 | { |
2438 | freestate(nfa, sclone); |
2439 | sclone = NULL; |
2440 | } |
2441 | |
2442 | /* |
2443 | * Move shead's constraint-loop arcs to point to sclone, or just drop them |
2444 | * if we discovered we don't need sclone. |
2445 | */ |
2446 | for (a = shead->outs; a != NULL; a = nexta) |
2447 | { |
2448 | nexta = a->outchain; |
2449 | if (a->to == stail && isconstraintarc(a)) |
2450 | { |
2451 | if (sclone) |
2452 | cparc(nfa, a, shead, sclone); |
2453 | freearc(nfa, a); |
2454 | if (NISERR()) |
2455 | break; |
2456 | } |
2457 | } |
2458 | } |
2459 | |
2460 | /* |
2461 | * clonesuccessorstates - create a tree of constraint-arc successor states |
2462 | * |
2463 | * ssource is the state to be cloned, and sclone is the state to copy its |
2464 | * outarcs into. sclone's inarcs, if any, should already be set up. |
2465 | * |
2466 | * spredecessor is the original predecessor state that we are trying to build |
2467 | * successors for (it may not be the immediate predecessor of ssource). |
2468 | * refarc, if not NULL, is the original constraint arc that is known to have |
2469 | * been traversed out of spredecessor to reach the successor(s). |
2470 | * |
2471 | * For each cloned successor state, we transiently create a "donemap" that is |
2472 | * a boolean array showing which source states we've already visited for this |
2473 | * clone state. This prevents infinite recursion as well as useless repeat |
2474 | * visits to the same state subtree (which can add up fast, since typical NFAs |
2475 | * have multiple redundant arc pathways). Each donemap is a char array |
2476 | * indexed by state number. The donemaps are all of the same size "nstates", |
2477 | * which is nfa->nstates as of the start of the recursion. This is enough to |
2478 | * have entries for all pre-existing states, but *not* entries for clone |
2479 | * states created during the recursion. That's okay since we have no need to |
2480 | * mark those. |
2481 | * |
2482 | * curdonemap is NULL when recursing to a new sclone state, or sclone's |
2483 | * donemap when we are recursing without having created a new state (which we |
2484 | * do when we decide we can merge a successor state into the current clone |
2485 | * state). outerdonemap is NULL at the top level and otherwise the parent |
2486 | * clone state's donemap. |
2487 | * |
2488 | * The successor states we create and fill here form a strict tree structure, |
2489 | * with each state having exactly one predecessor, except that the toplevel |
2490 | * state has no inarcs as yet (breakconstraintloop will add its inarcs from |
2491 | * spredecessor after we're done). Thus, we can examine sclone's inarcs back |
2492 | * to the root, plus refarc if any, to identify the set of constraints already |
2493 | * known valid at the current point. This allows us to avoid generating extra |
2494 | * successor states. |
2495 | */ |
2496 | static void |
2497 | clonesuccessorstates(struct nfa *nfa, |
2498 | struct state *ssource, |
2499 | struct state *sclone, |
2500 | struct state *spredecessor, |
2501 | struct arc *refarc, |
2502 | char *curdonemap, |
2503 | char *outerdonemap, |
2504 | int nstates) |
2505 | { |
2506 | char *donemap; |
2507 | struct arc *a; |
2508 | |
2509 | /* Since this is recursive, it could be driven to stack overflow */ |
2510 | if (STACK_TOO_DEEP(nfa->v->re)) |
2511 | { |
2512 | NERR(REG_ETOOBIG); |
2513 | return; |
2514 | } |
2515 | |
2516 | /* If this state hasn't already got a donemap, create one */ |
2517 | donemap = curdonemap; |
2518 | if (donemap == NULL) |
2519 | { |
2520 | donemap = (char *) MALLOC(nstates * sizeof(char)); |
2521 | if (donemap == NULL) |
2522 | { |
2523 | NERR(REG_ESPACE); |
2524 | return; |
2525 | } |
2526 | |
2527 | if (outerdonemap != NULL) |
2528 | { |
2529 | /* |
2530 | * Not at outermost recursion level, so copy the outer level's |
2531 | * donemap; this ensures that we see states in process of being |
2532 | * visited at outer levels, or already merged into predecessor |
2533 | * states, as ones we shouldn't traverse back to. |
2534 | */ |
2535 | memcpy(donemap, outerdonemap, nstates * sizeof(char)); |
2536 | } |
2537 | else |
2538 | { |
2539 | /* At outermost level, only spredecessor is off-limits */ |
2540 | memset(donemap, 0, nstates * sizeof(char)); |
2541 | assert(spredecessor->no < nstates); |
2542 | donemap[spredecessor->no] = 1; |
2543 | } |
2544 | } |
2545 | |
2546 | /* Mark ssource as visited in the donemap */ |
2547 | assert(ssource->no < nstates); |
2548 | assert(donemap[ssource->no] == 0); |
2549 | donemap[ssource->no] = 1; |
2550 | |
2551 | /* |
2552 | * We proceed by first cloning all of ssource's outarcs, creating new |
2553 | * clone states as needed but not doing more with them than that. Then in |
2554 | * a second pass, recurse to process the child clone states. This allows |
2555 | * us to have only one child clone state per reachable source state, even |
2556 | * when there are multiple outarcs leading to the same state. Also, when |
2557 | * we do visit a child state, its set of inarcs is known exactly, which |
2558 | * makes it safe to apply the constraint-is-already-checked optimization. |
2559 | * Also, this ensures that we've merged all the states we can into the |
2560 | * current clone before we recurse to any children, thus possibly saving |
2561 | * them from making extra images of those states. |
2562 | * |
2563 | * While this function runs, child clone states of the current state are |
2564 | * marked by setting their tmp fields to point to the original state they |
2565 | * were cloned from. This makes it possible to detect multiple outarcs |
2566 | * leading to the same state, and also makes it easy to distinguish clone |
2567 | * states from original states (which will have tmp == NULL). |
2568 | */ |
2569 | for (a = ssource->outs; a != NULL && !NISERR(); a = a->outchain) |
2570 | { |
2571 | struct state *sto = a->to; |
2572 | |
2573 | /* |
2574 | * We do not consider cloning successor states that have no constraint |
2575 | * outarcs; just link to them as-is. They cannot be part of a |
2576 | * constraint loop so there is no need to make copies. In particular, |
2577 | * this rule keeps us from trying to clone the post state, which would |
2578 | * be a bad idea. |
2579 | */ |
2580 | if (isconstraintarc(a) && hasconstraintout(sto)) |
2581 | { |
2582 | struct state *prevclone; |
2583 | int canmerge; |
2584 | struct arc *a2; |
2585 | |
2586 | /* |
2587 | * Back-link constraint arcs must not be followed. Nor is there a |
2588 | * need to revisit states previously merged into this clone. |
2589 | */ |
2590 | assert(sto->no < nstates); |
2591 | if (donemap[sto->no] != 0) |
2592 | continue; |
2593 | |
2594 | /* |
2595 | * Check whether we already have a child clone state for this |
2596 | * source state. |
2597 | */ |
2598 | prevclone = NULL; |
2599 | for (a2 = sclone->outs; a2 != NULL; a2 = a2->outchain) |
2600 | { |
2601 | if (a2->to->tmp == sto) |
2602 | { |
2603 | prevclone = a2->to; |
2604 | break; |
2605 | } |
2606 | } |
2607 | |
2608 | /* |
2609 | * If this arc is labeled the same as refarc, or the same as any |
2610 | * arc we must have traversed to get to sclone, then no additional |
2611 | * constraints need to be met to get to sto, so we should just |
2612 | * merge its outarcs into sclone. |
2613 | */ |
2614 | if (refarc && a->type == refarc->type && a->co == refarc->co) |
2615 | canmerge = 1; |
2616 | else |
2617 | { |
2618 | struct state *s; |
2619 | |
2620 | canmerge = 0; |
2621 | for (s = sclone; s->ins; s = s->ins->from) |
2622 | { |
2623 | if (s->nins == 1 && |
2624 | a->type == s->ins->type && a->co == s->ins->co) |
2625 | { |
2626 | canmerge = 1; |
2627 | break; |
2628 | } |
2629 | } |
2630 | } |
2631 | |
2632 | if (canmerge) |
2633 | { |
2634 | /* |
2635 | * We can merge into sclone. If we previously made a child |
2636 | * clone state, drop it; there's no need to visit it. (This |
2637 | * can happen if ssource has multiple pathways to sto, and we |
2638 | * only just now found one that is provably a no-op.) |
2639 | */ |
2640 | if (prevclone) |
2641 | dropstate(nfa, prevclone); /* kills our outarc, too */ |
2642 | |
2643 | /* Recurse to merge sto's outarcs into sclone */ |
2644 | clonesuccessorstates(nfa, |
2645 | sto, |
2646 | sclone, |
2647 | spredecessor, |
2648 | refarc, |
2649 | donemap, |
2650 | outerdonemap, |
2651 | nstates); |
2652 | /* sto should now be marked as previously visited */ |
2653 | assert(NISERR() || donemap[sto->no] == 1); |
2654 | } |
2655 | else if (prevclone) |
2656 | { |
2657 | /* |
2658 | * We already have a clone state for this successor, so just |
2659 | * make another arc to it. |
2660 | */ |
2661 | cparc(nfa, a, sclone, prevclone); |
2662 | } |
2663 | else |
2664 | { |
2665 | /* |
2666 | * We need to create a new successor clone state. |
2667 | */ |
2668 | struct state *stoclone; |
2669 | |
2670 | stoclone = newstate(nfa); |
2671 | if (stoclone == NULL) |
2672 | { |
2673 | assert(NISERR()); |
2674 | break; |
2675 | } |
2676 | /* Mark it as to what it's a clone of */ |
2677 | stoclone->tmp = sto; |
2678 | /* ... and add the outarc leading to it */ |
2679 | cparc(nfa, a, sclone, stoclone); |
2680 | } |
2681 | } |
2682 | else |
2683 | { |
2684 | /* |
2685 | * Non-constraint outarcs just get copied to sclone, as do outarcs |
2686 | * leading to states with no constraint outarc. |
2687 | */ |
2688 | cparc(nfa, a, sclone, sto); |
2689 | } |
2690 | } |
2691 | |
2692 | /* |
2693 | * If we are at outer level for this clone state, recurse to all its child |
2694 | * clone states, clearing their tmp fields as we go. (If we're not |
2695 | * outermost for sclone, leave this to be done by the outer call level.) |
2696 | * Note that if we have multiple outarcs leading to the same clone state, |
2697 | * it will only be recursed-to once. |
2698 | */ |
2699 | if (curdonemap == NULL) |
2700 | { |
2701 | for (a = sclone->outs; a != NULL && !NISERR(); a = a->outchain) |
2702 | { |
2703 | struct state *stoclone = a->to; |
2704 | struct state *sto = stoclone->tmp; |
2705 | |
2706 | if (sto != NULL) |
2707 | { |
2708 | stoclone->tmp = NULL; |
2709 | clonesuccessorstates(nfa, |
2710 | sto, |
2711 | stoclone, |
2712 | spredecessor, |
2713 | refarc, |
2714 | NULL, |
2715 | donemap, |
2716 | nstates); |
2717 | } |
2718 | } |
2719 | |
2720 | /* Don't forget to free sclone's donemap when done with it */ |
2721 | FREE(donemap); |
2722 | } |
2723 | } |
2724 | |
2725 | /* |
2726 | * cleanup - clean up NFA after optimizations |
2727 | */ |
2728 | static void |
2729 | cleanup(struct nfa *nfa) |
2730 | { |
2731 | struct state *s; |
2732 | struct state *nexts; |
2733 | int n; |
2734 | |
2735 | if (NISERR()) |
2736 | return; |
2737 | |
2738 | /* clear out unreachable or dead-end states */ |
2739 | /* use pre to mark reachable, then post to mark can-reach-post */ |
2740 | markreachable(nfa, nfa->pre, (struct state *) NULL, nfa->pre); |
2741 | markcanreach(nfa, nfa->post, nfa->pre, nfa->post); |
2742 | for (s = nfa->states; s != NULL && !NISERR(); s = nexts) |
2743 | { |
2744 | nexts = s->next; |
2745 | if (s->tmp != nfa->post && !s->flag) |
2746 | dropstate(nfa, s); |
2747 | } |
2748 | assert(NISERR() || nfa->post->nins == 0 || nfa->post->tmp == nfa->post); |
2749 | cleartraverse(nfa, nfa->pre); |
2750 | assert(NISERR() || nfa->post->nins == 0 || nfa->post->tmp == NULL); |
2751 | /* the nins==0 (final unreachable) case will be caught later */ |
2752 | |
2753 | /* renumber surviving states */ |
2754 | n = 0; |
2755 | for (s = nfa->states; s != NULL; s = s->next) |
2756 | s->no = n++; |
2757 | nfa->nstates = n; |
2758 | } |
2759 | |
2760 | /* |
2761 | * markreachable - recursive marking of reachable states |
2762 | */ |
2763 | static void |
2764 | markreachable(struct nfa *nfa, |
2765 | struct state *s, |
2766 | struct state *okay, /* consider only states with this mark */ |
2767 | struct state *mark) /* the value to mark with */ |
2768 | { |
2769 | struct arc *a; |
2770 | |
2771 | /* Since this is recursive, it could be driven to stack overflow */ |
2772 | if (STACK_TOO_DEEP(nfa->v->re)) |
2773 | { |
2774 | NERR(REG_ETOOBIG); |
2775 | return; |
2776 | } |
2777 | |
2778 | if (s->tmp != okay) |
2779 | return; |
2780 | s->tmp = mark; |
2781 | |
2782 | for (a = s->outs; a != NULL; a = a->outchain) |
2783 | markreachable(nfa, a->to, okay, mark); |
2784 | } |
2785 | |
2786 | /* |
2787 | * markcanreach - recursive marking of states which can reach here |
2788 | */ |
2789 | static void |
2790 | markcanreach(struct nfa *nfa, |
2791 | struct state *s, |
2792 | struct state *okay, /* consider only states with this mark */ |
2793 | struct state *mark) /* the value to mark with */ |
2794 | { |
2795 | struct arc *a; |
2796 | |
2797 | /* Since this is recursive, it could be driven to stack overflow */ |
2798 | if (STACK_TOO_DEEP(nfa->v->re)) |
2799 | { |
2800 | NERR(REG_ETOOBIG); |
2801 | return; |
2802 | } |
2803 | |
2804 | if (s->tmp != okay) |
2805 | return; |
2806 | s->tmp = mark; |
2807 | |
2808 | for (a = s->ins; a != NULL; a = a->inchain) |
2809 | markcanreach(nfa, a->from, okay, mark); |
2810 | } |
2811 | |
2812 | /* |
2813 | * analyze - ascertain potentially-useful facts about an optimized NFA |
2814 | */ |
2815 | static long /* re_info bits to be ORed in */ |
2816 | analyze(struct nfa *nfa) |
2817 | { |
2818 | struct arc *a; |
2819 | struct arc *aa; |
2820 | |
2821 | if (NISERR()) |
2822 | return 0; |
2823 | |
2824 | if (nfa->pre->outs == NULL) |
2825 | return REG_UIMPOSSIBLE; |
2826 | for (a = nfa->pre->outs; a != NULL; a = a->outchain) |
2827 | for (aa = a->to->outs; aa != NULL; aa = aa->outchain) |
2828 | if (aa->to == nfa->post) |
2829 | return REG_UEMPTYMATCH; |
2830 | return 0; |
2831 | } |
2832 | |
2833 | /* |
2834 | * compact - construct the compact representation of an NFA |
2835 | */ |
2836 | static void |
2837 | compact(struct nfa *nfa, |
2838 | struct cnfa *cnfa) |
2839 | { |
2840 | struct state *s; |
2841 | struct arc *a; |
2842 | size_t nstates; |
2843 | size_t narcs; |
2844 | struct carc *ca; |
2845 | struct carc *first; |
2846 | |
2847 | assert(!NISERR()); |
2848 | |
2849 | nstates = 0; |
2850 | narcs = 0; |
2851 | for (s = nfa->states; s != NULL; s = s->next) |
2852 | { |
2853 | nstates++; |
2854 | narcs += s->nouts + 1; /* need one extra for endmarker */ |
2855 | } |
2856 | |
2857 | cnfa->stflags = (char *) MALLOC(nstates * sizeof(char)); |
2858 | cnfa->states = (struct carc **) MALLOC(nstates * sizeof(struct carc *)); |
2859 | cnfa->arcs = (struct carc *) MALLOC(narcs * sizeof(struct carc)); |
2860 | if (cnfa->stflags == NULL || cnfa->states == NULL || cnfa->arcs == NULL) |
2861 | { |
2862 | if (cnfa->stflags != NULL) |
2863 | FREE(cnfa->stflags); |
2864 | if (cnfa->states != NULL) |
2865 | FREE(cnfa->states); |
2866 | if (cnfa->arcs != NULL) |
2867 | FREE(cnfa->arcs); |
2868 | NERR(REG_ESPACE); |
2869 | return; |
2870 | } |
2871 | cnfa->nstates = nstates; |
2872 | cnfa->pre = nfa->pre->no; |
2873 | cnfa->post = nfa->post->no; |
2874 | cnfa->bos[0] = nfa->bos[0]; |
2875 | cnfa->bos[1] = nfa->bos[1]; |
2876 | cnfa->eos[0] = nfa->eos[0]; |
2877 | cnfa->eos[1] = nfa->eos[1]; |
2878 | cnfa->ncolors = maxcolor(nfa->cm) + 1; |
2879 | cnfa->flags = 0; |
2880 | |
2881 | ca = cnfa->arcs; |
2882 | for (s = nfa->states; s != NULL; s = s->next) |
2883 | { |
2884 | assert((size_t) s->no < nstates); |
2885 | cnfa->stflags[s->no] = 0; |
2886 | cnfa->states[s->no] = ca; |
2887 | first = ca; |
2888 | for (a = s->outs; a != NULL; a = a->outchain) |
2889 | switch (a->type) |
2890 | { |
2891 | case PLAIN: |
2892 | ca->co = a->co; |
2893 | ca->to = a->to->no; |
2894 | ca++; |
2895 | break; |
2896 | case LACON: |
2897 | assert(s->no != cnfa->pre); |
2898 | ca->co = (color) (cnfa->ncolors + a->co); |
2899 | ca->to = a->to->no; |
2900 | ca++; |
2901 | cnfa->flags |= HASLACONS; |
2902 | break; |
2903 | default: |
2904 | NERR(REG_ASSERT); |
2905 | break; |
2906 | } |
2907 | carcsort(first, ca - first); |
2908 | ca->co = COLORLESS; |
2909 | ca->to = 0; |
2910 | ca++; |
2911 | } |
2912 | assert(ca == &cnfa->arcs[narcs]); |
2913 | assert(cnfa->nstates != 0); |
2914 | |
2915 | /* mark no-progress states */ |
2916 | for (a = nfa->pre->outs; a != NULL; a = a->outchain) |
2917 | cnfa->stflags[a->to->no] = CNFA_NOPROGRESS; |
2918 | cnfa->stflags[nfa->pre->no] = CNFA_NOPROGRESS; |
2919 | } |
2920 | |
2921 | /* |
2922 | * carcsort - sort compacted-NFA arcs by color |
2923 | */ |
2924 | static void |
2925 | carcsort(struct carc *first, size_t n) |
2926 | { |
2927 | if (n > 1) |
2928 | qsort(first, n, sizeof(struct carc), carc_cmp); |
2929 | } |
2930 | |
2931 | static int |
2932 | carc_cmp(const void *a, const void *b) |
2933 | { |
2934 | const struct carc *aa = (const struct carc *) a; |
2935 | const struct carc *bb = (const struct carc *) b; |
2936 | |
2937 | if (aa->co < bb->co) |
2938 | return -1; |
2939 | if (aa->co > bb->co) |
2940 | return +1; |
2941 | if (aa->to < bb->to) |
2942 | return -1; |
2943 | if (aa->to > bb->to) |
2944 | return +1; |
2945 | return 0; |
2946 | } |
2947 | |
2948 | /* |
2949 | * freecnfa - free a compacted NFA |
2950 | */ |
2951 | static void |
2952 | freecnfa(struct cnfa *cnfa) |
2953 | { |
2954 | assert(cnfa->nstates != 0); /* not empty already */ |
2955 | cnfa->nstates = 0; |
2956 | FREE(cnfa->stflags); |
2957 | FREE(cnfa->states); |
2958 | FREE(cnfa->arcs); |
2959 | } |
2960 | |
2961 | /* |
2962 | * dumpnfa - dump an NFA in human-readable form |
2963 | */ |
2964 | static void |
2965 | dumpnfa(struct nfa *nfa, |
2966 | FILE *f) |
2967 | { |
2968 | #ifdef REG_DEBUG |
2969 | struct state *s; |
2970 | int nstates = 0; |
2971 | int narcs = 0; |
2972 | |
2973 | fprintf(f, "pre %d, post %d" , nfa->pre->no, nfa->post->no); |
2974 | if (nfa->bos[0] != COLORLESS) |
2975 | fprintf(f, ", bos [%ld]" , (long) nfa->bos[0]); |
2976 | if (nfa->bos[1] != COLORLESS) |
2977 | fprintf(f, ", bol [%ld]" , (long) nfa->bos[1]); |
2978 | if (nfa->eos[0] != COLORLESS) |
2979 | fprintf(f, ", eos [%ld]" , (long) nfa->eos[0]); |
2980 | if (nfa->eos[1] != COLORLESS) |
2981 | fprintf(f, ", eol [%ld]" , (long) nfa->eos[1]); |
2982 | fprintf(f, "\n" ); |
2983 | for (s = nfa->states; s != NULL; s = s->next) |
2984 | { |
2985 | dumpstate(s, f); |
2986 | nstates++; |
2987 | narcs += s->nouts; |
2988 | } |
2989 | fprintf(f, "total of %d states, %d arcs\n" , nstates, narcs); |
2990 | if (nfa->parent == NULL) |
2991 | dumpcolors(nfa->cm, f); |
2992 | fflush(f); |
2993 | #endif |
2994 | } |
2995 | |
2996 | #ifdef REG_DEBUG /* subordinates of dumpnfa */ |
2997 | |
2998 | /* |
2999 | * dumpstate - dump an NFA state in human-readable form |
3000 | */ |
3001 | static void |
3002 | dumpstate(struct state *s, |
3003 | FILE *f) |
3004 | { |
3005 | struct arc *a; |
3006 | |
3007 | fprintf(f, "%d%s%c" , s->no, (s->tmp != NULL) ? "T" : "" , |
3008 | (s->flag) ? s->flag : '.'); |
3009 | if (s->prev != NULL && s->prev->next != s) |
3010 | fprintf(f, "\tstate chain bad\n" ); |
3011 | if (s->nouts == 0) |
3012 | fprintf(f, "\tno out arcs\n" ); |
3013 | else |
3014 | dumparcs(s, f); |
3015 | fflush(f); |
3016 | for (a = s->ins; a != NULL; a = a->inchain) |
3017 | { |
3018 | if (a->to != s) |
3019 | fprintf(f, "\tlink from %d to %d on %d's in-chain\n" , |
3020 | a->from->no, a->to->no, s->no); |
3021 | } |
3022 | } |
3023 | |
3024 | /* |
3025 | * dumparcs - dump out-arcs in human-readable form |
3026 | */ |
3027 | static void |
3028 | dumparcs(struct state *s, |
3029 | FILE *f) |
3030 | { |
3031 | int pos; |
3032 | struct arc *a; |
3033 | |
3034 | /* printing oldest arcs first is usually clearer */ |
3035 | a = s->outs; |
3036 | assert(a != NULL); |
3037 | while (a->outchain != NULL) |
3038 | a = a->outchain; |
3039 | pos = 1; |
3040 | do |
3041 | { |
3042 | dumparc(a, s, f); |
3043 | if (pos == 5) |
3044 | { |
3045 | fprintf(f, "\n" ); |
3046 | pos = 1; |
3047 | } |
3048 | else |
3049 | pos++; |
3050 | a = a->outchainRev; |
3051 | } while (a != NULL); |
3052 | if (pos != 1) |
3053 | fprintf(f, "\n" ); |
3054 | } |
3055 | |
3056 | /* |
3057 | * dumparc - dump one outarc in readable form, including prefixing tab |
3058 | */ |
3059 | static void |
3060 | dumparc(struct arc *a, |
3061 | struct state *s, |
3062 | FILE *f) |
3063 | { |
3064 | struct arc *aa; |
3065 | struct arcbatch *ab; |
3066 | |
3067 | fprintf(f, "\t" ); |
3068 | switch (a->type) |
3069 | { |
3070 | case PLAIN: |
3071 | fprintf(f, "[%ld]" , (long) a->co); |
3072 | break; |
3073 | case AHEAD: |
3074 | fprintf(f, ">%ld>" , (long) a->co); |
3075 | break; |
3076 | case BEHIND: |
3077 | fprintf(f, "<%ld<" , (long) a->co); |
3078 | break; |
3079 | case LACON: |
3080 | fprintf(f, ":%ld:" , (long) a->co); |
3081 | break; |
3082 | case '^': |
3083 | case '$': |
3084 | fprintf(f, "%c%d" , a->type, (int) a->co); |
3085 | break; |
3086 | case EMPTY: |
3087 | break; |
3088 | default: |
3089 | fprintf(f, "0x%x/0%lo" , a->type, (long) a->co); |
3090 | break; |
3091 | } |
3092 | if (a->from != s) |
3093 | fprintf(f, "?%d?" , a->from->no); |
3094 | for (ab = &a->from->oas; ab != NULL; ab = ab->next) |
3095 | { |
3096 | for (aa = &ab->a[0]; aa < &ab->a[ABSIZE]; aa++) |
3097 | if (aa == a) |
3098 | break; /* NOTE BREAK OUT */ |
3099 | if (aa < &ab->a[ABSIZE]) /* propagate break */ |
3100 | break; /* NOTE BREAK OUT */ |
3101 | } |
3102 | if (ab == NULL) |
3103 | fprintf(f, "?!?" ); /* not in allocated space */ |
3104 | fprintf(f, "->" ); |
3105 | if (a->to == NULL) |
3106 | { |
3107 | fprintf(f, "NULL" ); |
3108 | return; |
3109 | } |
3110 | fprintf(f, "%d" , a->to->no); |
3111 | for (aa = a->to->ins; aa != NULL; aa = aa->inchain) |
3112 | if (aa == a) |
3113 | break; /* NOTE BREAK OUT */ |
3114 | if (aa == NULL) |
3115 | fprintf(f, "?!?" ); /* missing from in-chain */ |
3116 | } |
3117 | #endif /* REG_DEBUG */ |
3118 | |
3119 | /* |
3120 | * dumpcnfa - dump a compacted NFA in human-readable form |
3121 | */ |
3122 | #ifdef REG_DEBUG |
3123 | static void |
3124 | dumpcnfa(struct cnfa *cnfa, |
3125 | FILE *f) |
3126 | { |
3127 | int st; |
3128 | |
3129 | fprintf(f, "pre %d, post %d" , cnfa->pre, cnfa->post); |
3130 | if (cnfa->bos[0] != COLORLESS) |
3131 | fprintf(f, ", bos [%ld]" , (long) cnfa->bos[0]); |
3132 | if (cnfa->bos[1] != COLORLESS) |
3133 | fprintf(f, ", bol [%ld]" , (long) cnfa->bos[1]); |
3134 | if (cnfa->eos[0] != COLORLESS) |
3135 | fprintf(f, ", eos [%ld]" , (long) cnfa->eos[0]); |
3136 | if (cnfa->eos[1] != COLORLESS) |
3137 | fprintf(f, ", eol [%ld]" , (long) cnfa->eos[1]); |
3138 | if (cnfa->flags & HASLACONS) |
3139 | fprintf(f, ", haslacons" ); |
3140 | fprintf(f, "\n" ); |
3141 | for (st = 0; st < cnfa->nstates; st++) |
3142 | dumpcstate(st, cnfa, f); |
3143 | fflush(f); |
3144 | } |
3145 | #endif |
3146 | |
3147 | #ifdef REG_DEBUG /* subordinates of dumpcnfa */ |
3148 | |
3149 | /* |
3150 | * dumpcstate - dump a compacted-NFA state in human-readable form |
3151 | */ |
3152 | static void |
3153 | dumpcstate(int st, |
3154 | struct cnfa *cnfa, |
3155 | FILE *f) |
3156 | { |
3157 | struct carc *ca; |
3158 | int pos; |
3159 | |
3160 | fprintf(f, "%d%s" , st, (cnfa->stflags[st] & CNFA_NOPROGRESS) ? ":" : "." ); |
3161 | pos = 1; |
3162 | for (ca = cnfa->states[st]; ca->co != COLORLESS; ca++) |
3163 | { |
3164 | if (ca->co < cnfa->ncolors) |
3165 | fprintf(f, "\t[%ld]->%d" , (long) ca->co, ca->to); |
3166 | else |
3167 | fprintf(f, "\t:%ld:->%d" , (long) (ca->co - cnfa->ncolors), ca->to); |
3168 | if (pos == 5) |
3169 | { |
3170 | fprintf(f, "\n" ); |
3171 | pos = 1; |
3172 | } |
3173 | else |
3174 | pos++; |
3175 | } |
3176 | if (ca == cnfa->states[st] || pos != 1) |
3177 | fprintf(f, "\n" ); |
3178 | fflush(f); |
3179 | } |
3180 | |
3181 | #endif /* REG_DEBUG */ |
3182 | |