| 1 | /* |
| 2 | * DFA routines |
| 3 | * This file is #included by regexec.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 | */ |
| 32 | |
| 33 | /* |
| 34 | - longest - longest-preferred matching engine |
| 35 | ^ static chr *longest(struct vars *, struct dfa *, chr *, chr *, int *); |
| 36 | */ |
| 37 | static chr * /* endpoint, or NULL */ |
| 38 | longest(v, d, start, stop, hitstopp) |
| 39 | struct vars *v; /* used only for debug and exec flags */ |
| 40 | struct dfa *d; |
| 41 | chr *start; /* where the match should start */ |
| 42 | chr *stop; /* match must end at or before here */ |
| 43 | int *hitstopp; /* record whether hit v->stop, if non-NULL */ |
| 44 | { |
| 45 | chr *cp; |
| 46 | chr *realstop = (stop == v->stop) ? stop : stop + 1; |
| 47 | color co; |
| 48 | struct sset *css; |
| 49 | struct sset *ss; |
| 50 | chr *post; |
| 51 | int i; |
| 52 | struct colormap *cm = d->cm; |
| 53 | |
| 54 | /* initialize */ |
| 55 | css = initialize(v, d, start); |
| 56 | cp = start; |
| 57 | if (hitstopp != NULL) |
| 58 | *hitstopp = 0; |
| 59 | |
| 60 | /* startup */ |
| 61 | FDEBUG(("+++ startup +++\n")); |
| 62 | if (cp == v->start) { |
| 63 | co = d->cnfa->bos[(v->eflags®_NOTBOL) ? 0 : 1]; |
| 64 | FDEBUG(("color %ld\n", (long)co)); |
| 65 | } else { |
| 66 | co = GETCOLOR(cm, *(cp - 1)); |
| 67 | FDEBUG(("char %c, color %ld\n", (char)*(cp-1), (long)co)); |
| 68 | } |
| 69 | css = miss(v, d, css, co, cp, start); |
| 70 | if (css == NULL) |
| 71 | return NULL; |
| 72 | css->lastseen = cp; |
| 73 | |
| 74 | /* main loop */ |
| 75 | if (v->eflags®_FTRACE) |
| 76 | while (cp < realstop) { |
| 77 | FDEBUG(("+++ at c%d +++\n", css - d->ssets)); |
| 78 | co = GETCOLOR(cm, *cp); |
| 79 | FDEBUG(("char %c, color %ld\n", (char)*cp, (long)co)); |
| 80 | ss = css->outs[co]; |
| 81 | if (ss == NULL) { |
| 82 | ss = miss(v, d, css, co, cp+1, start); |
| 83 | if (ss == NULL) |
| 84 | break; /* NOTE BREAK OUT */ |
| 85 | } |
| 86 | cp++; |
| 87 | ss->lastseen = cp; |
| 88 | css = ss; |
| 89 | } |
| 90 | else |
| 91 | while (cp < realstop) { |
| 92 | co = GETCOLOR(cm, *cp); |
| 93 | ss = css->outs[co]; |
| 94 | if (ss == NULL) { |
| 95 | ss = miss(v, d, css, co, cp+1, start); |
| 96 | if (ss == NULL) |
| 97 | break; /* NOTE BREAK OUT */ |
| 98 | } |
| 99 | cp++; |
| 100 | ss->lastseen = cp; |
| 101 | css = ss; |
| 102 | } |
| 103 | |
| 104 | /* shutdown */ |
| 105 | FDEBUG(("+++ shutdown at c%d +++\n", css - d->ssets)); |
| 106 | if (cp == v->stop && stop == v->stop) { |
| 107 | if (hitstopp != NULL) |
| 108 | *hitstopp = 1; |
| 109 | co = d->cnfa->eos[(v->eflags®_NOTEOL) ? 0 : 1]; |
| 110 | FDEBUG(("color %ld\n", (long)co)); |
| 111 | ss = miss(v, d, css, co, cp, start); |
| 112 | /* special case: match ended at eol? */ |
| 113 | if (ss != NULL && (ss->flags&POSTSTATE)) |
| 114 | return cp; |
| 115 | else if (ss != NULL) |
| 116 | ss->lastseen = cp; /* to be tidy */ |
| 117 | } |
| 118 | |
| 119 | /* find last match, if any */ |
| 120 | post = d->lastpost; |
| 121 | for (ss = d->ssets, i = d->nssused; i > 0; ss++, i--) |
| 122 | if ((ss->flags&POSTSTATE) && post != ss->lastseen && |
| 123 | (post == NULL || post < ss->lastseen)) |
| 124 | post = ss->lastseen; |
| 125 | if (post != NULL) /* found one */ |
| 126 | return post - 1; |
| 127 | |
| 128 | return NULL; |
| 129 | } |
| 130 | |
| 131 | /* |
| 132 | - shortest - shortest-preferred matching engine |
| 133 | ^ static chr *shortest(struct vars *, struct dfa *, chr *, chr *, chr *, |
| 134 | ^ chr **, int *); |
| 135 | */ |
| 136 | static chr * /* endpoint, or NULL */ |
| 137 | shortest(v, d, start, min, max, coldp, hitstopp) |
| 138 | struct vars *v; |
| 139 | struct dfa *d; |
| 140 | chr *start; /* where the match should start */ |
| 141 | chr *min; /* match must end at or after here */ |
| 142 | chr *max; /* match must end at or before here */ |
| 143 | chr **coldp; /* store coldstart pointer here, if nonNULL */ |
| 144 | int *hitstopp; /* record whether hit v->stop, if non-NULL */ |
| 145 | { |
| 146 | chr *cp; |
| 147 | chr *realmin = (min == v->stop) ? min : min + 1; |
| 148 | chr *realmax = (max == v->stop) ? max : max + 1; |
| 149 | color co; |
| 150 | struct sset *css; |
| 151 | struct sset *ss; |
| 152 | struct colormap *cm = d->cm; |
| 153 | |
| 154 | /* initialize */ |
| 155 | css = initialize(v, d, start); |
| 156 | cp = start; |
| 157 | if (hitstopp != NULL) |
| 158 | *hitstopp = 0; |
| 159 | |
| 160 | /* startup */ |
| 161 | FDEBUG(("--- startup ---\n")); |
| 162 | if (cp == v->start) { |
| 163 | co = d->cnfa->bos[(v->eflags®_NOTBOL) ? 0 : 1]; |
| 164 | FDEBUG(("color %ld\n", (long)co)); |
| 165 | } else { |
| 166 | co = GETCOLOR(cm, *(cp - 1)); |
| 167 | FDEBUG(("char %c, color %ld\n", (char)*(cp-1), (long)co)); |
| 168 | } |
| 169 | css = miss(v, d, css, co, cp, start); |
| 170 | if (css == NULL) |
| 171 | return NULL; |
| 172 | css->lastseen = cp; |
| 173 | ss = css; |
| 174 | |
| 175 | /* main loop */ |
| 176 | if (v->eflags®_FTRACE) |
| 177 | while (cp < realmax) { |
| 178 | FDEBUG(("--- at c%d ---\n", css - d->ssets)); |
| 179 | co = GETCOLOR(cm, *cp); |
| 180 | FDEBUG(("char %c, color %ld\n", (char)*cp, (long)co)); |
| 181 | ss = css->outs[co]; |
| 182 | if (ss == NULL) { |
| 183 | ss = miss(v, d, css, co, cp+1, start); |
| 184 | if (ss == NULL) |
| 185 | break; /* NOTE BREAK OUT */ |
| 186 | } |
| 187 | cp++; |
| 188 | ss->lastseen = cp; |
| 189 | css = ss; |
| 190 | if ((ss->flags&POSTSTATE) && cp >= realmin) |
| 191 | break; /* NOTE BREAK OUT */ |
| 192 | } |
| 193 | else |
| 194 | while (cp < realmax) { |
| 195 | co = GETCOLOR(cm, *cp); |
| 196 | ss = css->outs[co]; |
| 197 | if (ss == NULL) { |
| 198 | ss = miss(v, d, css, co, cp+1, start); |
| 199 | if (ss == NULL) |
| 200 | break; /* NOTE BREAK OUT */ |
| 201 | } |
| 202 | cp++; |
| 203 | ss->lastseen = cp; |
| 204 | css = ss; |
| 205 | if ((ss->flags&POSTSTATE) && cp >= realmin) |
| 206 | break; /* NOTE BREAK OUT */ |
| 207 | } |
| 208 | |
| 209 | if (ss == NULL) |
| 210 | return NULL; |
| 211 | |
| 212 | if (coldp != NULL) /* report last no-progress state set, if any */ |
| 213 | *coldp = lastcold(v, d); |
| 214 | |
| 215 | if ((ss->flags&POSTSTATE) && cp > min) { |
| 216 | assert(cp >= realmin); |
| 217 | cp--; |
| 218 | } else if (cp == v->stop && max == v->stop) { |
| 219 | co = d->cnfa->eos[(v->eflags®_NOTEOL) ? 0 : 1]; |
| 220 | FDEBUG(("color %ld\n", (long)co)); |
| 221 | ss = miss(v, d, css, co, cp, start); |
| 222 | /* match might have ended at eol */ |
| 223 | if ((ss == NULL || !(ss->flags&POSTSTATE)) && hitstopp != NULL) |
| 224 | *hitstopp = 1; |
| 225 | } |
| 226 | |
| 227 | if (ss == NULL || !(ss->flags&POSTSTATE)) |
| 228 | return NULL; |
| 229 | |
| 230 | return cp; |
| 231 | } |
| 232 | |
| 233 | /* |
| 234 | - lastcold - determine last point at which no progress had been made |
| 235 | ^ static chr *lastcold(struct vars *, struct dfa *); |
| 236 | */ |
| 237 | static chr * /* endpoint, or NULL */ |
| 238 | lastcold(v, d) |
| 239 | struct vars *v; |
| 240 | struct dfa *d; |
| 241 | { |
| 242 | struct sset *ss; |
| 243 | chr *nopr; |
| 244 | int i; |
| 245 | |
| 246 | nopr = d->lastnopr; |
| 247 | if (nopr == NULL) |
| 248 | nopr = v->start; |
| 249 | for (ss = d->ssets, i = d->nssused; i > 0; ss++, i--) |
| 250 | if ((ss->flags&NOPROGRESS) && nopr < ss->lastseen) |
| 251 | nopr = ss->lastseen; |
| 252 | return nopr; |
| 253 | } |
| 254 | |
| 255 | /* |
| 256 | - newdfa - set up a fresh DFA |
| 257 | ^ static struct dfa *newdfa(struct vars *, struct cnfa *, |
| 258 | ^ struct colormap *, struct smalldfa *); |
| 259 | */ |
| 260 | |
| 261 | /* FIXME Required for CW 8 on Mac since it's not in limits.h */ |
| 262 | #ifndef __CHAR_BIT__ |
| 263 | #define __CHAR_BIT__ 8 |
| 264 | #endif |
| 265 | |
| 266 | static struct dfa * |
| 267 | newdfa(v, cnfa, cm, small) |
| 268 | struct vars *v; |
| 269 | struct cnfa *cnfa; |
| 270 | struct colormap *cm; |
| 271 | struct smalldfa *small; /* preallocated space, may be NULL */ |
| 272 | { |
| 273 | struct dfa *d; |
| 274 | size_t nss = cnfa->nstates * 2; |
| 275 | int wordsper = (cnfa->nstates + UBITS - 1) / UBITS; |
| 276 | struct smalldfa *smallwas = small; |
| 277 | |
| 278 | assert(cnfa != NULL && cnfa->nstates != 0); |
| 279 | |
| 280 | if (nss <= FEWSTATES && cnfa->ncolors <= FEWCOLORS) { |
| 281 | assert(wordsper == 1); |
| 282 | if (small == NULL) { |
| 283 | small = (struct smalldfa *)MALLOC( |
| 284 | sizeof(struct smalldfa)); |
| 285 | if (small == NULL) { |
| 286 | ERR(REG_ESPACE); |
| 287 | return NULL; |
| 288 | } |
| 289 | } |
| 290 | d = &small->dfa; |
| 291 | d->ssets = small->ssets; |
| 292 | d->statesarea = small->statesarea; |
| 293 | d->work = &d->statesarea[nss]; |
| 294 | d->outsarea = small->outsarea; |
| 295 | d->incarea = small->incarea; |
| 296 | d->cptsmalloced = 0; |
| 297 | d->mallocarea = (smallwas == NULL) ? (char *)small : NULL; |
| 298 | } else { |
| 299 | d = (struct dfa *)MALLOC(sizeof(struct dfa)); |
| 300 | if (d == NULL) { |
| 301 | ERR(REG_ESPACE); |
| 302 | return NULL; |
| 303 | } |
| 304 | d->ssets = (struct sset *)MALLOC(nss * sizeof(struct sset)); |
| 305 | d->statesarea = (unsigned *)MALLOC((nss+WORK) * wordsper * |
| 306 | sizeof(unsigned)); |
| 307 | d->work = &d->statesarea[nss * wordsper]; |
| 308 | d->outsarea = (struct sset **)MALLOC(nss * cnfa->ncolors * |
| 309 | sizeof(struct sset *)); |
| 310 | d->incarea = (struct arcp *)MALLOC(nss * cnfa->ncolors * |
| 311 | sizeof(struct arcp)); |
| 312 | d->cptsmalloced = 1; |
| 313 | d->mallocarea = (char *)d; |
| 314 | if (d->ssets == NULL || d->statesarea == NULL || |
| 315 | d->outsarea == NULL || d->incarea == NULL) { |
| 316 | freedfa(d); |
| 317 | ERR(REG_ESPACE); |
| 318 | return NULL; |
| 319 | } |
| 320 | } |
| 321 | |
| 322 | d->nssets = (v->eflags®_SMALL) ? 7 : nss; |
| 323 | d->nssused = 0; |
| 324 | d->nstates = cnfa->nstates; |
| 325 | d->ncolors = cnfa->ncolors; |
| 326 | d->wordsper = wordsper; |
| 327 | d->cnfa = cnfa; |
| 328 | d->cm = cm; |
| 329 | d->lastpost = NULL; |
| 330 | d->lastnopr = NULL; |
| 331 | d->search = d->ssets; |
| 332 | |
| 333 | /* initialization of sset fields is done as needed */ |
| 334 | |
| 335 | return d; |
| 336 | } |
| 337 | |
| 338 | /* |
| 339 | - freedfa - free a DFA |
| 340 | ^ static VOID freedfa(struct dfa *); |
| 341 | */ |
| 342 | static VOID |
| 343 | freedfa(d) |
| 344 | struct dfa *d; |
| 345 | { |
| 346 | if (d->cptsmalloced) { |
| 347 | if (d->ssets != NULL) |
| 348 | FREE(d->ssets); |
| 349 | if (d->statesarea != NULL) |
| 350 | FREE(d->statesarea); |
| 351 | if (d->outsarea != NULL) |
| 352 | FREE(d->outsarea); |
| 353 | if (d->incarea != NULL) |
| 354 | FREE(d->incarea); |
| 355 | } |
| 356 | |
| 357 | if (d->mallocarea != NULL) |
| 358 | FREE(d->mallocarea); |
| 359 | } |
| 360 | |
| 361 | /* |
| 362 | - hash - construct a hash code for a bitvector |
| 363 | * There are probably better ways, but they're more expensive. |
| 364 | ^ static unsigned hash(unsigned *, int); |
| 365 | */ |
| 366 | static unsigned |
| 367 | hash(uv, n) |
| 368 | unsigned *uv; |
| 369 | int n; |
| 370 | { |
| 371 | int i; |
| 372 | unsigned h; |
| 373 | |
| 374 | h = 0; |
| 375 | for (i = 0; i < n; i++) |
| 376 | h ^= uv[i]; |
| 377 | return h; |
| 378 | } |
| 379 | |
| 380 | /* |
| 381 | - initialize - hand-craft a cache entry for startup, otherwise get ready |
| 382 | ^ static struct sset *initialize(struct vars *, struct dfa *, chr *); |
| 383 | */ |
| 384 | static struct sset * |
| 385 | initialize(v, d, start) |
| 386 | struct vars *v; /* used only for debug flags */ |
| 387 | struct dfa *d; |
| 388 | chr *start; |
| 389 | { |
| 390 | struct sset *ss; |
| 391 | int i; |
| 392 | |
| 393 | /* is previous one still there? */ |
| 394 | if (d->nssused > 0 && (d->ssets[0].flags&STARTER)) |
| 395 | ss = &d->ssets[0]; |
| 396 | else { /* no, must (re)build it */ |
| 397 | ss = getvacant(v, d, start, start); |
| 398 | for (i = 0; i < d->wordsper; i++) |
| 399 | ss->states[i] = 0; |
| 400 | BSET(ss->states, d->cnfa->pre); |
| 401 | ss->hash = HASH(ss->states, d->wordsper); |
| 402 | assert(d->cnfa->pre != d->cnfa->post); |
| 403 | ss->flags = STARTER|LOCKED|NOPROGRESS; |
| 404 | /* lastseen dealt with below */ |
| 405 | } |
| 406 | |
| 407 | for (i = 0; i < d->nssused; i++) |
| 408 | d->ssets[i].lastseen = NULL; |
| 409 | ss->lastseen = start; /* maybe untrue, but harmless */ |
| 410 | d->lastpost = NULL; |
| 411 | d->lastnopr = NULL; |
| 412 | return ss; |
| 413 | } |
| 414 | |
| 415 | /* |
| 416 | - miss - handle a cache miss |
| 417 | ^ static struct sset *miss(struct vars *, struct dfa *, struct sset *, |
| 418 | ^ pcolor, chr *, chr *); |
| 419 | */ |
| 420 | static struct sset * /* NULL if goes to empty set */ |
| 421 | miss(v, d, css, co, cp, start) |
| 422 | struct vars *v; /* used only for debug flags */ |
| 423 | struct dfa *d; |
| 424 | struct sset *css; |
| 425 | pcolor co; |
| 426 | chr *cp; /* next chr */ |
| 427 | chr *start; /* where the attempt got started */ |
| 428 | { |
| 429 | struct cnfa *cnfa = d->cnfa; |
| 430 | int i; |
| 431 | unsigned h; |
| 432 | struct carc *ca; |
| 433 | struct sset *p; |
| 434 | int ispost; |
| 435 | int noprogress; |
| 436 | int gotstate; |
| 437 | int dolacons; |
| 438 | int sawlacons; |
| 439 | |
| 440 | /* for convenience, we can be called even if it might not be a miss */ |
| 441 | if (css->outs[co] != NULL) { |
| 442 | FDEBUG(("hit\n")); |
| 443 | return css->outs[co]; |
| 444 | } |
| 445 | FDEBUG(("miss\n")); |
| 446 | |
| 447 | /* first, what set of states would we end up in? */ |
| 448 | for (i = 0; i < d->wordsper; i++) |
| 449 | d->work[i] = 0; |
| 450 | ispost = 0; |
| 451 | noprogress = 1; |
| 452 | gotstate = 0; |
| 453 | for (i = 0; i < d->nstates; i++) |
| 454 | if (ISBSET(css->states, i)) |
| 455 | for (ca = cnfa->states[i]+1; ca->co != COLORLESS; ca++) |
| 456 | if (ca->co == co) { |
| 457 | BSET(d->work, ca->to); |
| 458 | gotstate = 1; |
| 459 | if (ca->to == cnfa->post) |
| 460 | ispost = 1; |
| 461 | if (!cnfa->states[ca->to]->co) |
| 462 | noprogress = 0; |
| 463 | FDEBUG(("%d -> %d\n", i, ca->to)); |
| 464 | } |
| 465 | dolacons = (gotstate) ? (cnfa->flags&HASLACONS) : 0; |
| 466 | sawlacons = 0; |
| 467 | while (dolacons) { /* transitive closure */ |
| 468 | dolacons = 0; |
| 469 | for (i = 0; i < d->nstates; i++) |
| 470 | if (ISBSET(d->work, i)) |
| 471 | for (ca = cnfa->states[i]+1; ca->co != COLORLESS; |
| 472 | ca++) { |
| 473 | if (ca->co <= cnfa->ncolors) |
| 474 | continue; /* NOTE CONTINUE */ |
| 475 | sawlacons = 1; |
| 476 | if (ISBSET(d->work, ca->to)) |
| 477 | continue; /* NOTE CONTINUE */ |
| 478 | if (!lacon(v, cnfa, cp, ca->co)) |
| 479 | continue; /* NOTE CONTINUE */ |
| 480 | BSET(d->work, ca->to); |
| 481 | dolacons = 1; |
| 482 | if (ca->to == cnfa->post) |
| 483 | ispost = 1; |
| 484 | if (!cnfa->states[ca->to]->co) |
| 485 | noprogress = 0; |
| 486 | FDEBUG(("%d :> %d\n", i, ca->to)); |
| 487 | } |
| 488 | } |
| 489 | if (!gotstate) |
| 490 | return NULL; |
| 491 | h = HASH(d->work, d->wordsper); |
| 492 | |
| 493 | /* next, is that in the cache? */ |
| 494 | for (p = d->ssets, i = d->nssused; i > 0; p++, i--) |
| 495 | if (HIT(h, d->work, p, d->wordsper)) { |
| 496 | FDEBUG(("cached c%d\n", p - d->ssets)); |
| 497 | break; /* NOTE BREAK OUT */ |
| 498 | } |
| 499 | if (i == 0) { /* nope, need a new cache entry */ |
| 500 | p = getvacant(v, d, cp, start); |
| 501 | assert(p != css); |
| 502 | for (i = 0; i < d->wordsper; i++) |
| 503 | p->states[i] = d->work[i]; |
| 504 | p->hash = h; |
| 505 | p->flags = (ispost) ? POSTSTATE : 0; |
| 506 | if (noprogress) |
| 507 | p->flags |= NOPROGRESS; |
| 508 | /* lastseen to be dealt with by caller */ |
| 509 | } |
| 510 | |
| 511 | if (!sawlacons) { /* lookahead conds. always cache miss */ |
| 512 | FDEBUG(("c%d[%d]->c%d\n", css - d->ssets, co, p - d->ssets)); |
| 513 | css->outs[co] = p; |
| 514 | css->inchain[co] = p->ins; |
| 515 | p->ins.ss = css; |
| 516 | p->ins.co = (color)co; |
| 517 | } |
| 518 | return p; |
| 519 | } |
| 520 | |
| 521 | /* |
| 522 | - lacon - lookahead-constraint checker for miss() |
| 523 | ^ static int lacon(struct vars *, struct cnfa *, chr *, pcolor); |
| 524 | */ |
| 525 | static int /* predicate: constraint satisfied? */ |
| 526 | lacon(v, pcnfa, cp, co) |
| 527 | struct vars *v; |
| 528 | struct cnfa *pcnfa; /* parent cnfa */ |
| 529 | chr *cp; |
| 530 | pcolor co; /* "color" of the lookahead constraint */ |
| 531 | { |
| 532 | int n; |
| 533 | struct subre *sub; |
| 534 | struct dfa *d; |
| 535 | struct smalldfa sd; |
| 536 | chr *end; |
| 537 | |
| 538 | n = co - pcnfa->ncolors; |
| 539 | assert(n < v->g->nlacons && v->g->lacons != NULL); |
| 540 | FDEBUG(("=== testing lacon %d\n", n)); |
| 541 | sub = &v->g->lacons[n]; |
| 542 | d = newdfa(v, &sub->cnfa, &v->g->cmap, &sd); |
| 543 | if (d == NULL) { |
| 544 | ERR(REG_ESPACE); |
| 545 | return 0; |
| 546 | } |
| 547 | end = longest(v, d, cp, v->stop, (int *)NULL); |
| 548 | freedfa(d); |
| 549 | FDEBUG(("=== lacon %d match %d\n", n, (end != NULL))); |
| 550 | return (sub->subno) ? (end != NULL) : (end == NULL); |
| 551 | } |
| 552 | |
| 553 | /* |
| 554 | - getvacant - get a vacant state set |
| 555 | * This routine clears out the inarcs and outarcs, but does not otherwise |
| 556 | * clear the innards of the state set -- that's up to the caller. |
| 557 | ^ static struct sset *getvacant(struct vars *, struct dfa *, chr *, chr *); |
| 558 | */ |
| 559 | static struct sset * |
| 560 | getvacant(v, d, cp, start) |
| 561 | struct vars *v; /* used only for debug flags */ |
| 562 | struct dfa *d; |
| 563 | chr *cp; |
| 564 | chr *start; |
| 565 | { |
| 566 | int i; |
| 567 | struct sset *ss; |
| 568 | struct sset *p; |
| 569 | struct arcp ap; |
| 570 | struct arcp lastap; |
| 571 | color co; |
| 572 | lastap.ss = NULL; lastap.co = 0; // WX: suppress dummy gcc warnings |
| 573 | ss = pickss(v, d, cp, start); |
| 574 | assert(!(ss->flags&LOCKED)); |
| 575 | |
| 576 | /* clear out its inarcs, including self-referential ones */ |
| 577 | ap = ss->ins; |
| 578 | while ((p = ap.ss) != NULL) { |
| 579 | co = ap.co; |
| 580 | FDEBUG(("zapping c%d's %ld outarc\n", p - d->ssets, (long)co)); |
| 581 | p->outs[co] = NULL; |
| 582 | ap = p->inchain[co]; |
| 583 | p->inchain[co].ss = NULL; /* paranoia */ |
| 584 | } |
| 585 | ss->ins.ss = NULL; |
| 586 | |
| 587 | /* take it off the inarc chains of the ssets reached by its outarcs */ |
| 588 | for (i = 0; i < d->ncolors; i++) { |
| 589 | p = ss->outs[i]; |
| 590 | assert(p != ss); /* not self-referential */ |
| 591 | if (p == NULL) |
| 592 | continue; /* NOTE CONTINUE */ |
| 593 | FDEBUG(("del outarc %d from c%d's in chn\n", i, p - d->ssets)); |
| 594 | if (p->ins.ss == ss && p->ins.co == i) |
| 595 | p->ins = ss->inchain[i]; |
| 596 | else { |
| 597 | assert(p->ins.ss != NULL); |
| 598 | for (ap = p->ins; ap.ss != NULL && |
| 599 | !(ap.ss == ss && ap.co == i); |
| 600 | ap = ap.ss->inchain[ap.co]) |
| 601 | lastap = ap; |
| 602 | assert(ap.ss != NULL); |
| 603 | lastap.ss->inchain[lastap.co] = ss->inchain[i]; |
| 604 | } |
| 605 | ss->outs[i] = NULL; |
| 606 | ss->inchain[i].ss = NULL; |
| 607 | } |
| 608 | |
| 609 | /* if ss was a success state, may need to remember location */ |
| 610 | if ((ss->flags&POSTSTATE) && ss->lastseen != d->lastpost && |
| 611 | (d->lastpost == NULL || d->lastpost < ss->lastseen)) |
| 612 | d->lastpost = ss->lastseen; |
| 613 | |
| 614 | /* likewise for a no-progress state */ |
| 615 | if ((ss->flags&NOPROGRESS) && ss->lastseen != d->lastnopr && |
| 616 | (d->lastnopr == NULL || d->lastnopr < ss->lastseen)) |
| 617 | d->lastnopr = ss->lastseen; |
| 618 | |
| 619 | return ss; |
| 620 | } |
| 621 | |
| 622 | /* |
| 623 | - pickss - pick the next stateset to be used |
| 624 | ^ static struct sset *pickss(struct vars *, struct dfa *, chr *, chr *); |
| 625 | */ |
| 626 | static struct sset * |
| 627 | pickss(v, d, cp, start) |
| 628 | struct vars *v; /* used only for debug flags */ |
| 629 | struct dfa *d; |
| 630 | chr *cp; |
| 631 | chr *start; |
| 632 | { |
| 633 | int i; |
| 634 | struct sset *ss; |
| 635 | struct sset *end; |
| 636 | chr *ancient; |
| 637 | |
| 638 | /* shortcut for cases where cache isn't full */ |
| 639 | if (d->nssused < d->nssets) { |
| 640 | i = d->nssused; |
| 641 | d->nssused++; |
| 642 | ss = &d->ssets[i]; |
| 643 | FDEBUG(("new c%d\n", i)); |
| 644 | /* set up innards */ |
| 645 | ss->states = &d->statesarea[i * d->wordsper]; |
| 646 | ss->flags = 0; |
| 647 | ss->ins.ss = NULL; |
| 648 | ss->ins.co = WHITE; /* give it some value */ |
| 649 | ss->outs = &d->outsarea[i * d->ncolors]; |
| 650 | ss->inchain = &d->incarea[i * d->ncolors]; |
| 651 | for (i = 0; i < d->ncolors; i++) { |
| 652 | ss->outs[i] = NULL; |
| 653 | ss->inchain[i].ss = NULL; |
| 654 | } |
| 655 | return ss; |
| 656 | } |
| 657 | |
| 658 | /* look for oldest, or old enough anyway */ |
| 659 | if (cp - start > d->nssets*2/3) /* oldest 33% are expendable */ |
| 660 | ancient = cp - d->nssets*2/3; |
| 661 | else |
| 662 | ancient = start; |
| 663 | for (ss = d->search, end = &d->ssets[d->nssets]; ss < end; ss++) |
| 664 | if ((ss->lastseen == NULL || ss->lastseen < ancient) && |
| 665 | !(ss->flags&LOCKED)) { |
| 666 | d->search = ss + 1; |
| 667 | FDEBUG(("replacing c%d\n", ss - d->ssets)); |
| 668 | return ss; |
| 669 | } |
| 670 | for (ss = d->ssets, end = d->search; ss < end; ss++) |
| 671 | if ((ss->lastseen == NULL || ss->lastseen < ancient) && |
| 672 | !(ss->flags&LOCKED)) { |
| 673 | d->search = ss + 1; |
| 674 | FDEBUG(("replacing c%d\n", ss - d->ssets)); |
| 675 | return ss; |
| 676 | } |
| 677 | |
| 678 | /* nobody's old enough?!? -- something's really wrong */ |
| 679 | FDEBUG(("can't find victim to replace!\n")); |
| 680 | assert(NOTREACHED); |
| 681 | ERR(REG_ASSERT); |
| 682 | return d->ssets; |
| 683 | } |