1 /*
2 * re_*comp and friends - compile REs
3 * This file #includes several others (see the bottom).
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/regcomp.c
32 *
33 */
34
35 #include "regex/regguts.h"
36
37 /*
38 * forward declarations, up here so forward datatypes etc. are defined early
39 */
40 /* === regcomp.c === */
41 static void moresubs(struct vars *, int);
42 static int freev(struct vars *, int);
43 static void makesearch(struct vars *, struct nfa *);
44 static struct subre *parse(struct vars *, int, int, struct state *, struct state *);
45 static struct subre *parsebranch(struct vars *, int, int, struct state *, struct state *, int);
46 static void parseqatom(struct vars *, int, int, struct state *, struct state *, struct subre *);
47 static void nonword(struct vars *, int, struct state *, struct state *);
48 static void word(struct vars *, int, struct state *, struct state *);
49 static int scannum(struct vars *);
50 static void repeat(struct vars *, struct state *, struct state *, int, int);
51 static void bracket(struct vars *, struct state *, struct state *);
52 static void cbracket(struct vars *, struct state *, struct state *);
53 static void brackpart(struct vars *, struct state *, struct state *);
54 static const chr *scanplain(struct vars *);
55 static void onechr(struct vars *, chr, struct state *, struct state *);
56 static void wordchrs(struct vars *);
57 static void processlacon(struct vars *, struct state *, struct state *, int,
58 struct state *, struct state *);
59 static struct subre *subre(struct vars *, int, int, struct state *, struct state *);
60 static void freesubre(struct vars *, struct subre *);
61 static void freesrnode(struct vars *, struct subre *);
62 static void optst(struct vars *, struct subre *);
63 static int numst(struct subre *, int);
64 static void markst(struct subre *);
65 static void cleanst(struct vars *);
66 static long nfatree(struct vars *, struct subre *, FILE *);
67 static long nfanode(struct vars *, struct subre *, int, FILE *);
68 static int newlacon(struct vars *, struct state *, struct state *, int);
69 static void freelacons(struct subre *, int);
70 static void rfree(regex_t *);
71 static int rcancelrequested(void);
72 static int rstacktoodeep(void);
73
74 #ifdef REG_DEBUG
75 static void dump(regex_t *, FILE *);
76 static void dumpst(struct subre *, FILE *, int);
77 static void stdump(struct subre *, FILE *, int);
78 static const char *stid(struct subre *, char *, size_t);
79 #endif
80 /* === regc_lex.c === */
81 static void lexstart(struct vars *);
82 static void prefixes(struct vars *);
83 static void lexnest(struct vars *, const chr *, const chr *);
84 static void lexword(struct vars *);
85 static int next(struct vars *);
86 static int lexescape(struct vars *);
87 static chr lexdigits(struct vars *, int, int, int);
88 static int brenext(struct vars *, chr);
89 static void skip(struct vars *);
90 static chr newline(void);
91 static chr chrnamed(struct vars *, const chr *, const chr *, chr);
92
93 /* === regc_color.c === */
94 static void initcm(struct vars *, struct colormap *);
95 static void freecm(struct colormap *);
96 static color maxcolor(struct colormap *);
97 static color newcolor(struct colormap *);
98 static void freecolor(struct colormap *, color);
99 static color pseudocolor(struct colormap *);
100 static color subcolor(struct colormap *, chr);
101 static color subcolorhi(struct colormap *, color *);
102 static color newsub(struct colormap *, color);
103 static int newhicolorrow(struct colormap *, int);
104 static void newhicolorcols(struct colormap *);
105 static void subcolorcvec(struct vars *, struct cvec *, struct state *, struct state *);
106 static void subcoloronechr(struct vars *, chr, struct state *, struct state *, color *);
107 static void subcoloronerange(struct vars *, chr, chr, struct state *, struct state *, color *);
108 static void subcoloronerow(struct vars *, int, struct state *, struct state *, color *);
109 static void okcolors(struct nfa *, struct colormap *);
110 static void colorchain(struct colormap *, struct arc *);
111 static void uncolorchain(struct colormap *, struct arc *);
112 static void rainbow(struct nfa *, struct colormap *, int, color, struct state *, struct state *);
113 static void colorcomplement(struct nfa *, struct colormap *, int, struct state *, struct state *, struct state *);
114
115 #ifdef REG_DEBUG
116 static void dumpcolors(struct colormap *, FILE *);
117 static void dumpchr(chr, FILE *);
118 #endif
119 /* === regc_nfa.c === */
120 static struct nfa *newnfa(struct vars *, struct colormap *, struct nfa *);
121 static void freenfa(struct nfa *);
122 static struct state *newstate(struct nfa *);
123 static struct state *newfstate(struct nfa *, int flag);
124 static void dropstate(struct nfa *, struct state *);
125 static void freestate(struct nfa *, struct state *);
126 static void destroystate(struct nfa *, struct state *);
127 static void newarc(struct nfa *, int, color, struct state *, struct state *);
128 static void createarc(struct nfa *, int, color, struct state *, struct state *);
129 static struct arc *allocarc(struct nfa *, struct state *);
130 static void freearc(struct nfa *, struct arc *);
131 static void changearctarget(struct arc *, struct state *);
132 static int hasnonemptyout(struct state *);
133 static struct arc *findarc(struct state *, int, color);
134 static void cparc(struct nfa *, struct arc *, struct state *, struct state *);
135 static void sortins(struct nfa *, struct state *);
136 static int sortins_cmp(const void *, const void *);
137 static void sortouts(struct nfa *, struct state *);
138 static int sortouts_cmp(const void *, const void *);
139 static void moveins(struct nfa *, struct state *, struct state *);
140 static void copyins(struct nfa *, struct state *, struct state *);
141 static void mergeins(struct nfa *, struct state *, struct arc **, int);
142 static void moveouts(struct nfa *, struct state *, struct state *);
143 static void copyouts(struct nfa *, struct state *, struct state *);
144 static void cloneouts(struct nfa *, struct state *, struct state *, struct state *, int);
145 static void delsub(struct nfa *, struct state *, struct state *);
146 static void deltraverse(struct nfa *, struct state *, struct state *);
147 static void dupnfa(struct nfa *, struct state *, struct state *, struct state *, struct state *);
148 static void duptraverse(struct nfa *, struct state *, struct state *);
149 static void cleartraverse(struct nfa *, struct state *);
150 static struct state *single_color_transition(struct state *, struct state *);
151 static void specialcolors(struct nfa *);
152 static long optimize(struct nfa *, FILE *);
153 static void pullback(struct nfa *, FILE *);
154 static int pull(struct nfa *, struct arc *, struct state **);
155 static void pushfwd(struct nfa *, FILE *);
156 static int push(struct nfa *, struct arc *, struct state **);
157
158 #define INCOMPATIBLE 1 /* destroys arc */
159 #define SATISFIED 2 /* constraint satisfied */
160 #define COMPATIBLE 3 /* compatible but not satisfied yet */
161 static int combine(struct arc *, struct arc *);
162 static void fixempties(struct nfa *, FILE *);
163 static struct state *emptyreachable(struct nfa *, struct state *,
164 struct state *, struct arc **);
165 static int isconstraintarc(struct arc *);
166 static int hasconstraintout(struct state *);
167 static void fixconstraintloops(struct nfa *, FILE *);
168 static int findconstraintloop(struct nfa *, struct state *);
169 static void breakconstraintloop(struct nfa *, struct state *);
170 static void clonesuccessorstates(struct nfa *, struct state *, struct state *,
171 struct state *, struct arc *,
172 char *, char *, int);
173 static void cleanup(struct nfa *);
174 static void markreachable(struct nfa *, struct state *, struct state *, struct state *);
175 static void markcanreach(struct nfa *, struct state *, struct state *, struct state *);
176 static long analyze(struct nfa *);
177 static void compact(struct nfa *, struct cnfa *);
178 static void carcsort(struct carc *, size_t);
179 static int carc_cmp(const void *, const void *);
180 static void freecnfa(struct cnfa *);
181 static void dumpnfa(struct nfa *, FILE *);
182
183 #ifdef REG_DEBUG
184 static void dumpstate(struct state *, FILE *);
185 static void dumparcs(struct state *, FILE *);
186 static void dumparc(struct arc *, struct state *, FILE *);
187 static void dumpcnfa(struct cnfa *, FILE *);
188 static void dumpcstate(int, struct cnfa *, FILE *);
189 #endif
190 /* === regc_cvec.c === */
191 static struct cvec *newcvec(int, int);
192 static struct cvec *clearcvec(struct cvec *);
193 static void addchr(struct cvec *, chr);
194 static void addrange(struct cvec *, chr, chr);
195 static struct cvec *getcvec(struct vars *, int, int);
196 static void freecvec(struct cvec *);
197
198 /* === regc_pg_locale.c === */
199 static int pg_wc_isdigit(pg_wchar c);
200 static int pg_wc_isalpha(pg_wchar c);
201 static int pg_wc_isalnum(pg_wchar c);
202 static int pg_wc_isupper(pg_wchar c);
203 static int pg_wc_islower(pg_wchar c);
204 static int pg_wc_isgraph(pg_wchar c);
205 static int pg_wc_isprint(pg_wchar c);
206 static int pg_wc_ispunct(pg_wchar c);
207 static int pg_wc_isspace(pg_wchar c);
208 static pg_wchar pg_wc_toupper(pg_wchar c);
209 static pg_wchar pg_wc_tolower(pg_wchar c);
210
211 /* === regc_locale.c === */
212 static chr element(struct vars *, const chr *, const chr *);
213 static struct cvec *range(struct vars *, chr, chr, int);
214 static int before(chr, chr);
215 static struct cvec *eclass(struct vars *, chr, int);
216 static struct cvec *cclass(struct vars *, const chr *, const chr *, int);
217 static int cclass_column_index(struct colormap *, chr);
218 static struct cvec *allcases(struct vars *, chr);
219 static int cmp(const chr *, const chr *, size_t);
220 static int casecmp(const chr *, const chr *, size_t);
221
222
223 /* internal variables, bundled for easy passing around */
224 struct vars
225 {
226 regex_t *re;
227 const chr *now; /* scan pointer into string */
228 const chr *stop; /* end of string */
229 const chr *savenow; /* saved now and stop for "subroutine call" */
230 const chr *savestop;
231 int err; /* error code (0 if none) */
232 int cflags; /* copy of compile flags */
233 int lasttype; /* type of previous token */
234 int nexttype; /* type of next token */
235 chr nextvalue; /* value (if any) of next token */
236 int lexcon; /* lexical context type (see lex.c) */
237 int nsubexp; /* subexpression count */
238 struct subre **subs; /* subRE pointer vector */
239 size_t nsubs; /* length of vector */
240 struct subre *sub10[10]; /* initial vector, enough for most */
241 struct nfa *nfa; /* the NFA */
242 struct colormap *cm; /* character color map */
243 color nlcolor; /* color of newline */
244 struct state *wordchrs; /* state in nfa holding word-char outarcs */
245 struct subre *tree; /* subexpression tree */
246 struct subre *treechain; /* all tree nodes allocated */
247 struct subre *treefree; /* any free tree nodes */
248 int ntree; /* number of tree nodes, plus one */
249 struct cvec *cv; /* interface cvec */
250 struct cvec *cv2; /* utility cvec */
251 struct subre *lacons; /* lookaround-constraint vector */
252 int nlacons; /* size of lacons[]; note that only slots
253 * numbered 1 .. nlacons-1 are used */
254 size_t spaceused; /* approx. space used for compilation */
255 };
256
257 /* parsing macros; most know that `v' is the struct vars pointer */
258 #define NEXT() (next(v)) /* advance by one token */
259 #define SEE(t) (v->nexttype == (t)) /* is next token this? */
260 #define EAT(t) (SEE(t) && next(v)) /* if next is this, swallow it */
261 #define VISERR(vv) ((vv)->err != 0) /* have we seen an error yet? */
262 #define ISERR() VISERR(v)
263 #define VERR(vv,e) ((vv)->nexttype = EOS, \
264 (vv)->err = ((vv)->err ? (vv)->err : (e)))
265 #define ERR(e) VERR(v, e) /* record an error */
266 #define NOERR() {if (ISERR()) return;} /* if error seen, return */
267 #define NOERRN() {if (ISERR()) return NULL;} /* NOERR with retval */
268 #define NOERRZ() {if (ISERR()) return 0;} /* NOERR with retval */
269 #define INSIST(c, e) do { if (!(c)) ERR(e); } while (0) /* error if c false */
270 #define NOTE(b) (v->re->re_info |= (b)) /* note visible condition */
271 #define EMPTYARC(x, y) newarc(v->nfa, EMPTY, 0, x, y)
272
273 /* token type codes, some also used as NFA arc types */
274 #define EMPTY 'n' /* no token present */
275 #define EOS 'e' /* end of string */
276 #define PLAIN 'p' /* ordinary character */
277 #define DIGIT 'd' /* digit (in bound) */
278 #define BACKREF 'b' /* back reference */
279 #define COLLEL 'I' /* start of [. */
280 #define ECLASS 'E' /* start of [= */
281 #define CCLASS 'C' /* start of [: */
282 #define END 'X' /* end of [. [= [: */
283 #define RANGE 'R' /* - within [] which might be range delim. */
284 #define LACON 'L' /* lookaround constraint subRE */
285 #define AHEAD 'a' /* color-lookahead arc */
286 #define BEHIND 'r' /* color-lookbehind arc */
287 #define WBDRY 'w' /* word boundary constraint */
288 #define NWBDRY 'W' /* non-word-boundary constraint */
289 #define SBEGIN 'A' /* beginning of string (even if not BOL) */
290 #define SEND 'Z' /* end of string (even if not EOL) */
291 #define PREFER 'P' /* length preference */
292
293 /* is an arc colored, and hence on a color chain? */
294 #define COLORED(a) \
295 ((a)->type == PLAIN || (a)->type == AHEAD || (a)->type == BEHIND)
296
297
298 /* static function list */
299 static const struct fns functions = {
300 rfree, /* regfree insides */
301 rcancelrequested, /* check for cancel request */
302 rstacktoodeep /* check for stack getting dangerously deep */
303 };
304
305
306
307 /*
308 * pg_regcomp - compile regular expression
309 *
310 * Note: on failure, no resources remain allocated, so pg_regfree()
311 * need not be applied to re.
312 */
313 int
pg_regcomp(regex_t * re,const chr * string,size_t len,int flags,Oid collation)314 pg_regcomp(regex_t *re,
315 const chr *string,
316 size_t len,
317 int flags,
318 Oid collation)
319 {
320 struct vars var;
321 struct vars *v = &var;
322 struct guts *g;
323 int i;
324 size_t j;
325
326 #ifdef REG_DEBUG
327 FILE *debug = (flags & REG_PROGRESS) ? stdout : (FILE *) NULL;
328 #else
329 FILE *debug = (FILE *) NULL;
330 #endif
331
332 #define CNOERR() { if (ISERR()) return freev(v, v->err); }
333
334 /* sanity checks */
335
336 if (re == NULL || string == NULL)
337 return REG_INVARG;
338 if ((flags & REG_QUOTE) &&
339 (flags & (REG_ADVANCED | REG_EXPANDED | REG_NEWLINE)))
340 return REG_INVARG;
341 if (!(flags & REG_EXTENDED) && (flags & REG_ADVF))
342 return REG_INVARG;
343
344 /* Initialize locale-dependent support */
345 pg_set_regex_collation(collation);
346
347 /* initial setup (after which freev() is callable) */
348 v->re = re;
349 v->now = string;
350 v->stop = v->now + len;
351 v->savenow = v->savestop = NULL;
352 v->err = 0;
353 v->cflags = flags;
354 v->nsubexp = 0;
355 v->subs = v->sub10;
356 v->nsubs = 10;
357 for (j = 0; j < v->nsubs; j++)
358 v->subs[j] = NULL;
359 v->nfa = NULL;
360 v->cm = NULL;
361 v->nlcolor = COLORLESS;
362 v->wordchrs = NULL;
363 v->tree = NULL;
364 v->treechain = NULL;
365 v->treefree = NULL;
366 v->cv = NULL;
367 v->cv2 = NULL;
368 v->lacons = NULL;
369 v->nlacons = 0;
370 v->spaceused = 0;
371 re->re_magic = REMAGIC;
372 re->re_info = 0; /* bits get set during parse */
373 re->re_csize = sizeof(chr);
374 re->re_collation = collation;
375 re->re_guts = NULL;
376 re->re_fns = VS(&functions);
377
378 /* more complex setup, malloced things */
379 re->re_guts = VS(MALLOC(sizeof(struct guts)));
380 if (re->re_guts == NULL)
381 return freev(v, REG_ESPACE);
382 g = (struct guts *) re->re_guts;
383 g->tree = NULL;
384 initcm(v, &g->cmap);
385 v->cm = &g->cmap;
386 g->lacons = NULL;
387 g->nlacons = 0;
388 ZAPCNFA(g->search);
389 v->nfa = newnfa(v, v->cm, (struct nfa *) NULL);
390 CNOERR();
391 /* set up a reasonably-sized transient cvec for getcvec usage */
392 v->cv = newcvec(100, 20);
393 if (v->cv == NULL)
394 return freev(v, REG_ESPACE);
395
396 /* parsing */
397 lexstart(v); /* also handles prefixes */
398 if ((v->cflags & REG_NLSTOP) || (v->cflags & REG_NLANCH))
399 {
400 /* assign newline a unique color */
401 v->nlcolor = subcolor(v->cm, newline());
402 okcolors(v->nfa, v->cm);
403 }
404 CNOERR();
405 v->tree = parse(v, EOS, PLAIN, v->nfa->init, v->nfa->final);
406 assert(SEE(EOS)); /* even if error; ISERR() => SEE(EOS) */
407 CNOERR();
408 assert(v->tree != NULL);
409
410 /* finish setup of nfa and its subre tree */
411 specialcolors(v->nfa);
412 CNOERR();
413 #ifdef REG_DEBUG
414 if (debug != NULL)
415 {
416 fprintf(debug, "\n\n\n========= RAW ==========\n");
417 dumpnfa(v->nfa, debug);
418 dumpst(v->tree, debug, 1);
419 }
420 #endif
421 optst(v, v->tree);
422 v->ntree = numst(v->tree, 1);
423 markst(v->tree);
424 cleanst(v);
425 #ifdef REG_DEBUG
426 if (debug != NULL)
427 {
428 fprintf(debug, "\n\n\n========= TREE FIXED ==========\n");
429 dumpst(v->tree, debug, 1);
430 }
431 #endif
432
433 /* build compacted NFAs for tree and lacons */
434 re->re_info |= nfatree(v, v->tree, debug);
435 CNOERR();
436 assert(v->nlacons == 0 || v->lacons != NULL);
437 for (i = 1; i < v->nlacons; i++)
438 {
439 struct subre *lasub = &v->lacons[i];
440
441 #ifdef REG_DEBUG
442 if (debug != NULL)
443 fprintf(debug, "\n\n\n========= LA%d ==========\n", i);
444 #endif
445
446 /* Prepend .* to pattern if it's a lookbehind LACON */
447 nfanode(v, lasub, !LATYPE_IS_AHEAD(lasub->subno), debug);
448 }
449 CNOERR();
450 if (v->tree->flags & SHORTER)
451 NOTE(REG_USHORTEST);
452
453 /* build compacted NFAs for tree, lacons, fast search */
454 #ifdef REG_DEBUG
455 if (debug != NULL)
456 fprintf(debug, "\n\n\n========= SEARCH ==========\n");
457 #endif
458 /* can sacrifice main NFA now, so use it as work area */
459 (DISCARD) optimize(v->nfa, debug);
460 CNOERR();
461 makesearch(v, v->nfa);
462 CNOERR();
463 compact(v->nfa, &g->search);
464 CNOERR();
465
466 /* looks okay, package it up */
467 re->re_nsub = v->nsubexp;
468 v->re = NULL; /* freev no longer frees re */
469 g->magic = GUTSMAGIC;
470 g->cflags = v->cflags;
471 g->info = re->re_info;
472 g->nsub = re->re_nsub;
473 g->tree = v->tree;
474 v->tree = NULL;
475 g->ntree = v->ntree;
476 g->compare = (v->cflags & REG_ICASE) ? casecmp : cmp;
477 g->lacons = v->lacons;
478 v->lacons = NULL;
479 g->nlacons = v->nlacons;
480
481 #ifdef REG_DEBUG
482 if (flags & REG_DUMP)
483 dump(re, stdout);
484 #endif
485
486 assert(v->err == 0);
487 return freev(v, 0);
488 }
489
490 /*
491 * moresubs - enlarge subRE vector
492 */
493 static void
moresubs(struct vars * v,int wanted)494 moresubs(struct vars *v,
495 int wanted) /* want enough room for this one */
496 {
497 struct subre **p;
498 size_t n;
499
500 assert(wanted > 0 && (size_t) wanted >= v->nsubs);
501 n = (size_t) wanted * 3 / 2 + 1;
502
503 if (v->subs == v->sub10)
504 {
505 p = (struct subre **) MALLOC(n * sizeof(struct subre *));
506 if (p != NULL)
507 memcpy(VS(p), VS(v->subs),
508 v->nsubs * sizeof(struct subre *));
509 }
510 else
511 p = (struct subre **) REALLOC(v->subs, n * sizeof(struct subre *));
512 if (p == NULL)
513 {
514 ERR(REG_ESPACE);
515 return;
516 }
517 v->subs = p;
518 for (p = &v->subs[v->nsubs]; v->nsubs < n; p++, v->nsubs++)
519 *p = NULL;
520 assert(v->nsubs == n);
521 assert((size_t) wanted < v->nsubs);
522 }
523
524 /*
525 * freev - free vars struct's substructures where necessary
526 *
527 * Optionally does error-number setting, and always returns error code
528 * (if any), to make error-handling code terser.
529 */
530 static int
freev(struct vars * v,int err)531 freev(struct vars *v,
532 int err)
533 {
534 if (v->re != NULL)
535 rfree(v->re);
536 if (v->subs != v->sub10)
537 FREE(v->subs);
538 if (v->nfa != NULL)
539 freenfa(v->nfa);
540 if (v->tree != NULL)
541 freesubre(v, v->tree);
542 if (v->treechain != NULL)
543 cleanst(v);
544 if (v->cv != NULL)
545 freecvec(v->cv);
546 if (v->cv2 != NULL)
547 freecvec(v->cv2);
548 if (v->lacons != NULL)
549 freelacons(v->lacons, v->nlacons);
550 ERR(err); /* nop if err==0 */
551
552 return v->err;
553 }
554
555 /*
556 * makesearch - turn an NFA into a search NFA (implicit prepend of .*?)
557 * NFA must have been optimize()d already.
558 */
559 static void
makesearch(struct vars * v,struct nfa * nfa)560 makesearch(struct vars *v,
561 struct nfa *nfa)
562 {
563 struct arc *a;
564 struct arc *b;
565 struct state *pre = nfa->pre;
566 struct state *s;
567 struct state *s2;
568 struct state *slist;
569
570 /* no loops are needed if it's anchored */
571 for (a = pre->outs; a != NULL; a = a->outchain)
572 {
573 assert(a->type == PLAIN);
574 if (a->co != nfa->bos[0] && a->co != nfa->bos[1])
575 break;
576 }
577 if (a != NULL)
578 {
579 /* add implicit .* in front */
580 rainbow(nfa, v->cm, PLAIN, COLORLESS, pre, pre);
581
582 /* and ^* and \A* too -- not always necessary, but harmless */
583 newarc(nfa, PLAIN, nfa->bos[0], pre, pre);
584 newarc(nfa, PLAIN, nfa->bos[1], pre, pre);
585 }
586
587 /*
588 * Now here's the subtle part. Because many REs have no lookback
589 * constraints, often knowing when you were in the pre state tells you
590 * little; it's the next state(s) that are informative. But some of them
591 * may have other inarcs, i.e. it may be possible to make actual progress
592 * and then return to one of them. We must de-optimize such cases,
593 * splitting each such state into progress and no-progress states.
594 */
595
596 /* first, make a list of the states reachable from pre and elsewhere */
597 slist = NULL;
598 for (a = pre->outs; a != NULL; a = a->outchain)
599 {
600 s = a->to;
601 for (b = s->ins; b != NULL; b = b->inchain)
602 {
603 if (b->from != pre)
604 break;
605 }
606
607 /*
608 * We want to mark states as being in the list already by having non
609 * NULL tmp fields, but we can't just store the old slist value in tmp
610 * because that doesn't work for the first such state. Instead, the
611 * first list entry gets its own address in tmp.
612 */
613 if (b != NULL && s->tmp == NULL)
614 {
615 s->tmp = (slist != NULL) ? slist : s;
616 slist = s;
617 }
618 }
619
620 /* do the splits */
621 for (s = slist; s != NULL; s = s2)
622 {
623 s2 = newstate(nfa);
624 NOERR();
625 copyouts(nfa, s, s2);
626 NOERR();
627 for (a = s->ins; a != NULL; a = b)
628 {
629 b = a->inchain;
630 if (a->from != pre)
631 {
632 cparc(nfa, a, a->from, s2);
633 freearc(nfa, a);
634 }
635 }
636 s2 = (s->tmp != s) ? s->tmp : NULL;
637 s->tmp = NULL; /* clean up while we're at it */
638 }
639 }
640
641 /*
642 * parse - parse an RE
643 *
644 * This is actually just the top level, which parses a bunch of branches
645 * tied together with '|'. They appear in the tree as the left children
646 * of a chain of '|' subres.
647 */
648 static struct subre *
parse(struct vars * v,int stopper,int type,struct state * init,struct state * final)649 parse(struct vars *v,
650 int stopper, /* EOS or ')' */
651 int type, /* LACON (lookaround subRE) or PLAIN */
652 struct state *init, /* initial state */
653 struct state *final) /* final state */
654 {
655 struct state *left; /* scaffolding for branch */
656 struct state *right;
657 struct subre *branches; /* top level */
658 struct subre *branch; /* current branch */
659 struct subre *t; /* temporary */
660 int firstbranch; /* is this the first branch? */
661
662 assert(stopper == ')' || stopper == EOS);
663
664 branches = subre(v, '|', LONGER, init, final);
665 NOERRN();
666 branch = branches;
667 firstbranch = 1;
668 do
669 { /* a branch */
670 if (!firstbranch)
671 {
672 /* need a place to hang it */
673 branch->right = subre(v, '|', LONGER, init, final);
674 NOERRN();
675 branch = branch->right;
676 }
677 firstbranch = 0;
678 left = newstate(v->nfa);
679 right = newstate(v->nfa);
680 NOERRN();
681 EMPTYARC(init, left);
682 EMPTYARC(right, final);
683 NOERRN();
684 branch->left = parsebranch(v, stopper, type, left, right, 0);
685 NOERRN();
686 branch->flags |= UP(branch->flags | branch->left->flags);
687 if ((branch->flags & ~branches->flags) != 0) /* new flags */
688 for (t = branches; t != branch; t = t->right)
689 t->flags |= branch->flags;
690 } while (EAT('|'));
691 assert(SEE(stopper) || SEE(EOS));
692
693 if (!SEE(stopper))
694 {
695 assert(stopper == ')' && SEE(EOS));
696 ERR(REG_EPAREN);
697 }
698
699 /* optimize out simple cases */
700 if (branch == branches)
701 { /* only one branch */
702 assert(branch->right == NULL);
703 t = branch->left;
704 branch->left = NULL;
705 freesubre(v, branches);
706 branches = t;
707 }
708 else if (!MESSY(branches->flags))
709 { /* no interesting innards */
710 freesubre(v, branches->left);
711 branches->left = NULL;
712 freesubre(v, branches->right);
713 branches->right = NULL;
714 branches->op = '=';
715 }
716
717 return branches;
718 }
719
720 /*
721 * parsebranch - parse one branch of an RE
722 *
723 * This mostly manages concatenation, working closely with parseqatom().
724 * Concatenated things are bundled up as much as possible, with separate
725 * ',' nodes introduced only when necessary due to substructure.
726 */
727 static struct subre *
parsebranch(struct vars * v,int stopper,int type,struct state * left,struct state * right,int partial)728 parsebranch(struct vars *v,
729 int stopper, /* EOS or ')' */
730 int type, /* LACON (lookaround subRE) or PLAIN */
731 struct state *left, /* leftmost state */
732 struct state *right, /* rightmost state */
733 int partial) /* is this only part of a branch? */
734 {
735 struct state *lp; /* left end of current construct */
736 int seencontent; /* is there anything in this branch yet? */
737 struct subre *t;
738
739 lp = left;
740 seencontent = 0;
741 t = subre(v, '=', 0, left, right); /* op '=' is tentative */
742 NOERRN();
743 while (!SEE('|') && !SEE(stopper) && !SEE(EOS))
744 {
745 if (seencontent)
746 { /* implicit concat operator */
747 lp = newstate(v->nfa);
748 NOERRN();
749 moveins(v->nfa, right, lp);
750 }
751 seencontent = 1;
752
753 /* NB, recursion in parseqatom() may swallow rest of branch */
754 parseqatom(v, stopper, type, lp, right, t);
755 NOERRN();
756 }
757
758 if (!seencontent)
759 { /* empty branch */
760 if (!partial)
761 NOTE(REG_UUNSPEC);
762 assert(lp == left);
763 EMPTYARC(left, right);
764 }
765
766 return t;
767 }
768
769 /*
770 * parseqatom - parse one quantified atom or constraint of an RE
771 *
772 * The bookkeeping near the end cooperates very closely with parsebranch();
773 * in particular, it contains a recursion that can involve parsing the rest
774 * of the branch, making this function's name somewhat inaccurate.
775 */
776 static void
parseqatom(struct vars * v,int stopper,int type,struct state * lp,struct state * rp,struct subre * top)777 parseqatom(struct vars *v,
778 int stopper, /* EOS or ')' */
779 int type, /* LACON (lookaround subRE) or PLAIN */
780 struct state *lp, /* left state to hang it on */
781 struct state *rp, /* right state to hang it on */
782 struct subre *top) /* subtree top */
783 {
784 struct state *s; /* temporaries for new states */
785 struct state *s2;
786
787 #define ARCV(t, val) newarc(v->nfa, t, val, lp, rp)
788 int m,
789 n;
790 struct subre *atom; /* atom's subtree */
791 struct subre *t;
792 int cap; /* capturing parens? */
793 int latype; /* lookaround constraint type */
794 int subno; /* capturing-parens or backref number */
795 int atomtype;
796 int qprefer; /* quantifier short/long preference */
797 int f;
798 struct subre **atomp; /* where the pointer to atom is */
799
800 /* initial bookkeeping */
801 atom = NULL;
802 assert(lp->nouts == 0); /* must string new code */
803 assert(rp->nins == 0); /* between lp and rp */
804 subno = 0; /* just to shut lint up */
805
806 /* an atom or constraint... */
807 atomtype = v->nexttype;
808 switch (atomtype)
809 {
810 /* first, constraints, which end by returning */
811 case '^':
812 ARCV('^', 1);
813 if (v->cflags & REG_NLANCH)
814 ARCV(BEHIND, v->nlcolor);
815 NEXT();
816 return;
817 break;
818 case '$':
819 ARCV('$', 1);
820 if (v->cflags & REG_NLANCH)
821 ARCV(AHEAD, v->nlcolor);
822 NEXT();
823 return;
824 break;
825 case SBEGIN:
826 ARCV('^', 1); /* BOL */
827 ARCV('^', 0); /* or BOS */
828 NEXT();
829 return;
830 break;
831 case SEND:
832 ARCV('$', 1); /* EOL */
833 ARCV('$', 0); /* or EOS */
834 NEXT();
835 return;
836 break;
837 case '<':
838 wordchrs(v); /* does NEXT() */
839 s = newstate(v->nfa);
840 NOERR();
841 nonword(v, BEHIND, lp, s);
842 word(v, AHEAD, s, rp);
843 return;
844 break;
845 case '>':
846 wordchrs(v); /* does NEXT() */
847 s = newstate(v->nfa);
848 NOERR();
849 word(v, BEHIND, lp, s);
850 nonword(v, AHEAD, s, rp);
851 return;
852 break;
853 case WBDRY:
854 wordchrs(v); /* does NEXT() */
855 s = newstate(v->nfa);
856 NOERR();
857 nonword(v, BEHIND, lp, s);
858 word(v, AHEAD, s, rp);
859 s = newstate(v->nfa);
860 NOERR();
861 word(v, BEHIND, lp, s);
862 nonword(v, AHEAD, s, rp);
863 return;
864 break;
865 case NWBDRY:
866 wordchrs(v); /* does NEXT() */
867 s = newstate(v->nfa);
868 NOERR();
869 word(v, BEHIND, lp, s);
870 word(v, AHEAD, s, rp);
871 s = newstate(v->nfa);
872 NOERR();
873 nonword(v, BEHIND, lp, s);
874 nonword(v, AHEAD, s, rp);
875 return;
876 break;
877 case LACON: /* lookaround constraint */
878 latype = v->nextvalue;
879 NEXT();
880 s = newstate(v->nfa);
881 s2 = newstate(v->nfa);
882 NOERR();
883 t = parse(v, ')', LACON, s, s2);
884 freesubre(v, t); /* internal structure irrelevant */
885 NOERR();
886 assert(SEE(')'));
887 NEXT();
888 processlacon(v, s, s2, latype, lp, rp);
889 return;
890 break;
891 /* then errors, to get them out of the way */
892 case '*':
893 case '+':
894 case '?':
895 case '{':
896 ERR(REG_BADRPT);
897 return;
898 break;
899 default:
900 ERR(REG_ASSERT);
901 return;
902 break;
903 /* then plain characters, and minor variants on that theme */
904 case ')': /* unbalanced paren */
905 if ((v->cflags & REG_ADVANCED) != REG_EXTENDED)
906 {
907 ERR(REG_EPAREN);
908 return;
909 }
910 /* legal in EREs due to specification botch */
911 NOTE(REG_UPBOTCH);
912 /* fall through into case PLAIN */
913 /* FALLTHROUGH */
914 case PLAIN:
915 onechr(v, v->nextvalue, lp, rp);
916 okcolors(v->nfa, v->cm);
917 NOERR();
918 NEXT();
919 break;
920 case '[':
921 if (v->nextvalue == 1)
922 bracket(v, lp, rp);
923 else
924 cbracket(v, lp, rp);
925 assert(SEE(']') || ISERR());
926 NEXT();
927 break;
928 case '.':
929 rainbow(v->nfa, v->cm, PLAIN,
930 (v->cflags & REG_NLSTOP) ? v->nlcolor : COLORLESS,
931 lp, rp);
932 NEXT();
933 break;
934 /* and finally the ugly stuff */
935 case '(': /* value flags as capturing or non */
936 cap = (type == LACON) ? 0 : v->nextvalue;
937 if (cap)
938 {
939 v->nsubexp++;
940 subno = v->nsubexp;
941 if ((size_t) subno >= v->nsubs)
942 moresubs(v, subno);
943 assert((size_t) subno < v->nsubs);
944 }
945 else
946 atomtype = PLAIN; /* something that's not '(' */
947 NEXT();
948 /* need new endpoints because tree will contain pointers */
949 s = newstate(v->nfa);
950 s2 = newstate(v->nfa);
951 NOERR();
952 EMPTYARC(lp, s);
953 EMPTYARC(s2, rp);
954 NOERR();
955 atom = parse(v, ')', type, s, s2);
956 assert(SEE(')') || ISERR());
957 NEXT();
958 NOERR();
959 if (cap)
960 {
961 v->subs[subno] = atom;
962 t = subre(v, '(', atom->flags | CAP, s, s2);
963 NOERR();
964 t->subno = subno;
965 t->left = atom;
966 atom = t;
967 }
968 /* postpone everything else pending possible {0} */
969 break;
970 case BACKREF: /* the Feature From The Black Lagoon */
971 INSIST(type != LACON, REG_ESUBREG);
972 INSIST(v->nextvalue < v->nsubs, REG_ESUBREG);
973 INSIST(v->subs[v->nextvalue] != NULL, REG_ESUBREG);
974 NOERR();
975 assert(v->nextvalue > 0);
976 atom = subre(v, 'b', BACKR, lp, rp);
977 NOERR();
978 subno = v->nextvalue;
979 atom->subno = subno;
980 EMPTYARC(lp, rp); /* temporarily, so there's something */
981 NEXT();
982 break;
983 }
984
985 /* ...and an atom may be followed by a quantifier */
986 switch (v->nexttype)
987 {
988 case '*':
989 m = 0;
990 n = DUPINF;
991 qprefer = (v->nextvalue) ? LONGER : SHORTER;
992 NEXT();
993 break;
994 case '+':
995 m = 1;
996 n = DUPINF;
997 qprefer = (v->nextvalue) ? LONGER : SHORTER;
998 NEXT();
999 break;
1000 case '?':
1001 m = 0;
1002 n = 1;
1003 qprefer = (v->nextvalue) ? LONGER : SHORTER;
1004 NEXT();
1005 break;
1006 case '{':
1007 NEXT();
1008 m = scannum(v);
1009 if (EAT(','))
1010 {
1011 if (SEE(DIGIT))
1012 n = scannum(v);
1013 else
1014 n = DUPINF;
1015 if (m > n)
1016 {
1017 ERR(REG_BADBR);
1018 return;
1019 }
1020 /* {m,n} exercises preference, even if it's {m,m} */
1021 qprefer = (v->nextvalue) ? LONGER : SHORTER;
1022 }
1023 else
1024 {
1025 n = m;
1026 /* {m} passes operand's preference through */
1027 qprefer = 0;
1028 }
1029 if (!SEE('}'))
1030 { /* catches errors too */
1031 ERR(REG_BADBR);
1032 return;
1033 }
1034 NEXT();
1035 break;
1036 default: /* no quantifier */
1037 m = n = 1;
1038 qprefer = 0;
1039 break;
1040 }
1041
1042 /* annoying special case: {0} or {0,0} cancels everything */
1043 if (m == 0 && n == 0)
1044 {
1045 /*
1046 * If we had capturing subexpression(s) within the atom, we don't want
1047 * to destroy them, because it's legal (if useless) to back-ref them
1048 * later. Hence, just unlink the atom from lp/rp and then ignore it.
1049 */
1050 if (atom != NULL && (atom->flags & CAP))
1051 {
1052 delsub(v->nfa, lp, atom->begin);
1053 delsub(v->nfa, atom->end, rp);
1054 }
1055 else
1056 {
1057 /* Otherwise, we can clean up any subre infrastructure we made */
1058 if (atom != NULL)
1059 freesubre(v, atom);
1060 delsub(v->nfa, lp, rp);
1061 }
1062 EMPTYARC(lp, rp);
1063 return;
1064 }
1065
1066 /* if not a messy case, avoid hard part */
1067 assert(!MESSY(top->flags));
1068 f = top->flags | qprefer | ((atom != NULL) ? atom->flags : 0);
1069 if (atomtype != '(' && atomtype != BACKREF && !MESSY(UP(f)))
1070 {
1071 if (!(m == 1 && n == 1))
1072 repeat(v, lp, rp, m, n);
1073 if (atom != NULL)
1074 freesubre(v, atom);
1075 top->flags = f;
1076 return;
1077 }
1078
1079 /*
1080 * hard part: something messy
1081 *
1082 * That is, capturing parens, back reference, short/long clash, or an atom
1083 * with substructure containing one of those.
1084 */
1085
1086 /* now we'll need a subre for the contents even if they're boring */
1087 if (atom == NULL)
1088 {
1089 atom = subre(v, '=', 0, lp, rp);
1090 NOERR();
1091 }
1092
1093 /*----------
1094 * Prepare a general-purpose state skeleton.
1095 *
1096 * In the no-backrefs case, we want this:
1097 *
1098 * [lp] ---> [s] ---prefix---> [begin] ---atom---> [end] ---rest---> [rp]
1099 *
1100 * where prefix is some repetitions of atom. In the general case we need
1101 *
1102 * [lp] ---> [s] ---iterator---> [s2] ---rest---> [rp]
1103 *
1104 * where the iterator wraps around [begin] ---atom---> [end]
1105 *
1106 * We make the s state here for both cases; s2 is made below if needed
1107 *----------
1108 */
1109 s = newstate(v->nfa); /* first, new endpoints for the atom */
1110 s2 = newstate(v->nfa);
1111 NOERR();
1112 moveouts(v->nfa, lp, s);
1113 moveins(v->nfa, rp, s2);
1114 NOERR();
1115 atom->begin = s;
1116 atom->end = s2;
1117 s = newstate(v->nfa); /* set up starting state */
1118 NOERR();
1119 EMPTYARC(lp, s);
1120 NOERR();
1121
1122 /* break remaining subRE into x{...} and what follows */
1123 t = subre(v, '.', COMBINE(qprefer, atom->flags), lp, rp);
1124 NOERR();
1125 t->left = atom;
1126 atomp = &t->left;
1127
1128 /* here we should recurse... but we must postpone that to the end */
1129
1130 /* split top into prefix and remaining */
1131 assert(top->op == '=' && top->left == NULL && top->right == NULL);
1132 top->left = subre(v, '=', top->flags, top->begin, lp);
1133 NOERR();
1134 top->op = '.';
1135 top->right = t;
1136
1137 /* if it's a backref, now is the time to replicate the subNFA */
1138 if (atomtype == BACKREF)
1139 {
1140 assert(atom->begin->nouts == 1); /* just the EMPTY */
1141 delsub(v->nfa, atom->begin, atom->end);
1142 assert(v->subs[subno] != NULL);
1143
1144 /*
1145 * And here's why the recursion got postponed: it must wait until the
1146 * skeleton is filled in, because it may hit a backref that wants to
1147 * copy the filled-in skeleton.
1148 */
1149 dupnfa(v->nfa, v->subs[subno]->begin, v->subs[subno]->end,
1150 atom->begin, atom->end);
1151 NOERR();
1152 }
1153
1154 /*
1155 * It's quantifier time. If the atom is just a backref, we'll let it deal
1156 * with quantifiers internally.
1157 */
1158 if (atomtype == BACKREF)
1159 {
1160 /* special case: backrefs have internal quantifiers */
1161 EMPTYARC(s, atom->begin); /* empty prefix */
1162 /* just stuff everything into atom */
1163 repeat(v, atom->begin, atom->end, m, n);
1164 atom->min = (short) m;
1165 atom->max = (short) n;
1166 atom->flags |= COMBINE(qprefer, atom->flags);
1167 /* rest of branch can be strung starting from atom->end */
1168 s2 = atom->end;
1169 }
1170 else if (m == 1 && n == 1 &&
1171 (qprefer == 0 ||
1172 (atom->flags & (LONGER | SHORTER | MIXED)) == 0 ||
1173 qprefer == (atom->flags & (LONGER | SHORTER | MIXED))))
1174 {
1175 /* no/vacuous quantifier: done */
1176 EMPTYARC(s, atom->begin); /* empty prefix */
1177 /* rest of branch can be strung starting from atom->end */
1178 s2 = atom->end;
1179 }
1180 else if (m > 0 && !(atom->flags & BACKR))
1181 {
1182 /*
1183 * If there's no backrefs involved, we can turn x{m,n} into
1184 * x{m-1,n-1}x, with capturing parens in only the second x. This is
1185 * valid because we only care about capturing matches from the final
1186 * iteration of the quantifier. It's a win because we can implement
1187 * the backref-free left side as a plain DFA node, since we don't
1188 * really care where its submatches are.
1189 */
1190 dupnfa(v->nfa, atom->begin, atom->end, s, atom->begin);
1191 assert(m >= 1 && m != DUPINF && n >= 1);
1192 repeat(v, s, atom->begin, m - 1, (n == DUPINF) ? n : n - 1);
1193 f = COMBINE(qprefer, atom->flags);
1194 t = subre(v, '.', f, s, atom->end); /* prefix and atom */
1195 NOERR();
1196 t->left = subre(v, '=', PREF(f), s, atom->begin);
1197 NOERR();
1198 t->right = atom;
1199 *atomp = t;
1200 /* rest of branch can be strung starting from atom->end */
1201 s2 = atom->end;
1202 }
1203 else
1204 {
1205 /* general case: need an iteration node */
1206 s2 = newstate(v->nfa);
1207 NOERR();
1208 moveouts(v->nfa, atom->end, s2);
1209 NOERR();
1210 dupnfa(v->nfa, atom->begin, atom->end, s, s2);
1211 repeat(v, s, s2, m, n);
1212 f = COMBINE(qprefer, atom->flags);
1213 t = subre(v, '*', f, s, s2);
1214 NOERR();
1215 t->min = (short) m;
1216 t->max = (short) n;
1217 t->left = atom;
1218 *atomp = t;
1219 /* rest of branch is to be strung from iteration's end state */
1220 }
1221
1222 /* and finally, look after that postponed recursion */
1223 t = top->right;
1224 if (!(SEE('|') || SEE(stopper) || SEE(EOS)))
1225 t->right = parsebranch(v, stopper, type, s2, rp, 1);
1226 else
1227 {
1228 EMPTYARC(s2, rp);
1229 t->right = subre(v, '=', 0, s2, rp);
1230 }
1231 NOERR();
1232 assert(SEE('|') || SEE(stopper) || SEE(EOS));
1233 t->flags |= COMBINE(t->flags, t->right->flags);
1234 top->flags |= COMBINE(top->flags, t->flags);
1235 }
1236
1237 /*
1238 * nonword - generate arcs for non-word-character ahead or behind
1239 */
1240 static void
nonword(struct vars * v,int dir,struct state * lp,struct state * rp)1241 nonword(struct vars *v,
1242 int dir, /* AHEAD or BEHIND */
1243 struct state *lp,
1244 struct state *rp)
1245 {
1246 int anchor = (dir == AHEAD) ? '$' : '^';
1247
1248 assert(dir == AHEAD || dir == BEHIND);
1249 newarc(v->nfa, anchor, 1, lp, rp);
1250 newarc(v->nfa, anchor, 0, lp, rp);
1251 colorcomplement(v->nfa, v->cm, dir, v->wordchrs, lp, rp);
1252 /* (no need for special attention to \n) */
1253 }
1254
1255 /*
1256 * word - generate arcs for word character ahead or behind
1257 */
1258 static void
word(struct vars * v,int dir,struct state * lp,struct state * rp)1259 word(struct vars *v,
1260 int dir, /* AHEAD or BEHIND */
1261 struct state *lp,
1262 struct state *rp)
1263 {
1264 assert(dir == AHEAD || dir == BEHIND);
1265 cloneouts(v->nfa, v->wordchrs, lp, rp, dir);
1266 /* (no need for special attention to \n) */
1267 }
1268
1269 /*
1270 * scannum - scan a number
1271 */
1272 static int /* value, <= DUPMAX */
scannum(struct vars * v)1273 scannum(struct vars *v)
1274 {
1275 int n = 0;
1276
1277 while (SEE(DIGIT) && n < DUPMAX)
1278 {
1279 n = n * 10 + v->nextvalue;
1280 NEXT();
1281 }
1282 if (SEE(DIGIT) || n > DUPMAX)
1283 {
1284 ERR(REG_BADBR);
1285 return 0;
1286 }
1287 return n;
1288 }
1289
1290 /*
1291 * repeat - replicate subNFA for quantifiers
1292 *
1293 * The sub-NFA strung from lp to rp is modified to represent m to n
1294 * repetitions of its initial contents.
1295 *
1296 * The duplication sequences used here are chosen carefully so that any
1297 * pointers starting out pointing into the subexpression end up pointing into
1298 * the last occurrence. (Note that it may not be strung between the same
1299 * left and right end states, however!) This used to be important for the
1300 * subRE tree, although the important bits are now handled by the in-line
1301 * code in parse(), and when this is called, it doesn't matter any more.
1302 */
1303 static void
repeat(struct vars * v,struct state * lp,struct state * rp,int m,int n)1304 repeat(struct vars *v,
1305 struct state *lp,
1306 struct state *rp,
1307 int m,
1308 int n)
1309 {
1310 #define SOME 2
1311 #define INF 3
1312 #define PAIR(x, y) ((x)*4 + (y))
1313 #define REDUCE(x) ( ((x) == DUPINF) ? INF : (((x) > 1) ? SOME : (x)) )
1314 const int rm = REDUCE(m);
1315 const int rn = REDUCE(n);
1316 struct state *s;
1317 struct state *s2;
1318
1319 switch (PAIR(rm, rn))
1320 {
1321 case PAIR(0, 0): /* empty string */
1322 delsub(v->nfa, lp, rp);
1323 EMPTYARC(lp, rp);
1324 break;
1325 case PAIR(0, 1): /* do as x| */
1326 EMPTYARC(lp, rp);
1327 break;
1328 case PAIR(0, SOME): /* do as x{1,n}| */
1329 repeat(v, lp, rp, 1, n);
1330 NOERR();
1331 EMPTYARC(lp, rp);
1332 break;
1333 case PAIR(0, INF): /* loop x around */
1334 s = newstate(v->nfa);
1335 NOERR();
1336 moveouts(v->nfa, lp, s);
1337 moveins(v->nfa, rp, s);
1338 EMPTYARC(lp, s);
1339 EMPTYARC(s, rp);
1340 break;
1341 case PAIR(1, 1): /* no action required */
1342 break;
1343 case PAIR(1, SOME): /* do as x{0,n-1}x = (x{1,n-1}|)x */
1344 s = newstate(v->nfa);
1345 NOERR();
1346 moveouts(v->nfa, lp, s);
1347 dupnfa(v->nfa, s, rp, lp, s);
1348 NOERR();
1349 repeat(v, lp, s, 1, n - 1);
1350 NOERR();
1351 EMPTYARC(lp, s);
1352 break;
1353 case PAIR(1, INF): /* add loopback arc */
1354 s = newstate(v->nfa);
1355 s2 = newstate(v->nfa);
1356 NOERR();
1357 moveouts(v->nfa, lp, s);
1358 moveins(v->nfa, rp, s2);
1359 EMPTYARC(lp, s);
1360 EMPTYARC(s2, rp);
1361 EMPTYARC(s2, s);
1362 break;
1363 case PAIR(SOME, SOME): /* do as x{m-1,n-1}x */
1364 s = newstate(v->nfa);
1365 NOERR();
1366 moveouts(v->nfa, lp, s);
1367 dupnfa(v->nfa, s, rp, lp, s);
1368 NOERR();
1369 repeat(v, lp, s, m - 1, n - 1);
1370 break;
1371 case PAIR(SOME, INF): /* do as x{m-1,}x */
1372 s = newstate(v->nfa);
1373 NOERR();
1374 moveouts(v->nfa, lp, s);
1375 dupnfa(v->nfa, s, rp, lp, s);
1376 NOERR();
1377 repeat(v, lp, s, m - 1, n);
1378 break;
1379 default:
1380 ERR(REG_ASSERT);
1381 break;
1382 }
1383 }
1384
1385 /*
1386 * bracket - handle non-complemented bracket expression
1387 * Also called from cbracket for complemented bracket expressions.
1388 */
1389 static void
bracket(struct vars * v,struct state * lp,struct state * rp)1390 bracket(struct vars *v,
1391 struct state *lp,
1392 struct state *rp)
1393 {
1394 assert(SEE('['));
1395 NEXT();
1396 while (!SEE(']') && !SEE(EOS))
1397 brackpart(v, lp, rp);
1398 assert(SEE(']') || ISERR());
1399 okcolors(v->nfa, v->cm);
1400 }
1401
1402 /*
1403 * cbracket - handle complemented bracket expression
1404 * We do it by calling bracket() with dummy endpoints, and then complementing
1405 * the result. The alternative would be to invoke rainbow(), and then delete
1406 * arcs as the b.e. is seen... but that gets messy.
1407 */
1408 static void
cbracket(struct vars * v,struct state * lp,struct state * rp)1409 cbracket(struct vars *v,
1410 struct state *lp,
1411 struct state *rp)
1412 {
1413 struct state *left = newstate(v->nfa);
1414 struct state *right = newstate(v->nfa);
1415
1416 NOERR();
1417 bracket(v, left, right);
1418 if (v->cflags & REG_NLSTOP)
1419 newarc(v->nfa, PLAIN, v->nlcolor, left, right);
1420 NOERR();
1421
1422 assert(lp->nouts == 0); /* all outarcs will be ours */
1423
1424 /*
1425 * Easy part of complementing, and all there is to do since the MCCE code
1426 * was removed.
1427 */
1428 colorcomplement(v->nfa, v->cm, PLAIN, left, lp, rp);
1429 NOERR();
1430 dropstate(v->nfa, left);
1431 assert(right->nins == 0);
1432 freestate(v->nfa, right);
1433 }
1434
1435 /*
1436 * brackpart - handle one item (or range) within a bracket expression
1437 */
1438 static void
brackpart(struct vars * v,struct state * lp,struct state * rp)1439 brackpart(struct vars *v,
1440 struct state *lp,
1441 struct state *rp)
1442 {
1443 chr startc;
1444 chr endc;
1445 struct cvec *cv;
1446 const chr *startp;
1447 const chr *endp;
1448 chr c[1];
1449
1450 /* parse something, get rid of special cases, take shortcuts */
1451 switch (v->nexttype)
1452 {
1453 case RANGE: /* a-b-c or other botch */
1454 ERR(REG_ERANGE);
1455 return;
1456 break;
1457 case PLAIN:
1458 c[0] = v->nextvalue;
1459 NEXT();
1460 /* shortcut for ordinary chr (not range) */
1461 if (!SEE(RANGE))
1462 {
1463 onechr(v, c[0], lp, rp);
1464 return;
1465 }
1466 startc = element(v, c, c + 1);
1467 NOERR();
1468 break;
1469 case COLLEL:
1470 startp = v->now;
1471 endp = scanplain(v);
1472 INSIST(startp < endp, REG_ECOLLATE);
1473 NOERR();
1474 startc = element(v, startp, endp);
1475 NOERR();
1476 break;
1477 case ECLASS:
1478 startp = v->now;
1479 endp = scanplain(v);
1480 INSIST(startp < endp, REG_ECOLLATE);
1481 NOERR();
1482 startc = element(v, startp, endp);
1483 NOERR();
1484 cv = eclass(v, startc, (v->cflags & REG_ICASE));
1485 NOERR();
1486 subcolorcvec(v, cv, lp, rp);
1487 return;
1488 break;
1489 case CCLASS:
1490 startp = v->now;
1491 endp = scanplain(v);
1492 INSIST(startp < endp, REG_ECTYPE);
1493 NOERR();
1494 cv = cclass(v, startp, endp, (v->cflags & REG_ICASE));
1495 NOERR();
1496 subcolorcvec(v, cv, lp, rp);
1497 return;
1498 break;
1499 default:
1500 ERR(REG_ASSERT);
1501 return;
1502 break;
1503 }
1504
1505 if (SEE(RANGE))
1506 {
1507 NEXT();
1508 switch (v->nexttype)
1509 {
1510 case PLAIN:
1511 case RANGE:
1512 c[0] = v->nextvalue;
1513 NEXT();
1514 endc = element(v, c, c + 1);
1515 NOERR();
1516 break;
1517 case COLLEL:
1518 startp = v->now;
1519 endp = scanplain(v);
1520 INSIST(startp < endp, REG_ECOLLATE);
1521 NOERR();
1522 endc = element(v, startp, endp);
1523 NOERR();
1524 break;
1525 default:
1526 ERR(REG_ERANGE);
1527 return;
1528 break;
1529 }
1530 }
1531 else
1532 endc = startc;
1533
1534 /*
1535 * Ranges are unportable. Actually, standard C does guarantee that digits
1536 * are contiguous, but making that an exception is just too complicated.
1537 */
1538 if (startc != endc)
1539 NOTE(REG_UUNPORT);
1540 cv = range(v, startc, endc, (v->cflags & REG_ICASE));
1541 NOERR();
1542 subcolorcvec(v, cv, lp, rp);
1543 }
1544
1545 /*
1546 * scanplain - scan PLAIN contents of [. etc.
1547 *
1548 * Certain bits of trickery in lex.c know that this code does not try
1549 * to look past the final bracket of the [. etc.
1550 */
1551 static const chr * /* just after end of sequence */
scanplain(struct vars * v)1552 scanplain(struct vars *v)
1553 {
1554 const chr *endp;
1555
1556 assert(SEE(COLLEL) || SEE(ECLASS) || SEE(CCLASS));
1557 NEXT();
1558
1559 endp = v->now;
1560 while (SEE(PLAIN))
1561 {
1562 endp = v->now;
1563 NEXT();
1564 }
1565
1566 assert(SEE(END) || ISERR());
1567 NEXT();
1568
1569 return endp;
1570 }
1571
1572 /*
1573 * onechr - fill in arcs for a plain character, and possible case complements
1574 * This is mostly a shortcut for efficient handling of the common case.
1575 */
1576 static void
onechr(struct vars * v,chr c,struct state * lp,struct state * rp)1577 onechr(struct vars *v,
1578 chr c,
1579 struct state *lp,
1580 struct state *rp)
1581 {
1582 if (!(v->cflags & REG_ICASE))
1583 {
1584 color lastsubcolor = COLORLESS;
1585
1586 subcoloronechr(v, c, lp, rp, &lastsubcolor);
1587 return;
1588 }
1589
1590 /* rats, need general case anyway... */
1591 subcolorcvec(v, allcases(v, c), lp, rp);
1592 }
1593
1594 /*
1595 * wordchrs - set up word-chr list for word-boundary stuff, if needed
1596 *
1597 * The list is kept as a bunch of arcs between two dummy states; it's
1598 * disposed of by the unreachable-states sweep in NFA optimization.
1599 * Does NEXT(). Must not be called from any unusual lexical context.
1600 * This should be reconciled with the \w etc. handling in lex.c, and
1601 * should be cleaned up to reduce dependencies on input scanning.
1602 */
1603 static void
wordchrs(struct vars * v)1604 wordchrs(struct vars *v)
1605 {
1606 struct state *left;
1607 struct state *right;
1608
1609 if (v->wordchrs != NULL)
1610 {
1611 NEXT(); /* for consistency */
1612 return;
1613 }
1614
1615 left = newstate(v->nfa);
1616 right = newstate(v->nfa);
1617 NOERR();
1618 /* fine point: implemented with [::], and lexer will set REG_ULOCALE */
1619 lexword(v);
1620 NEXT();
1621 assert(v->savenow != NULL && SEE('['));
1622 bracket(v, left, right);
1623 assert((v->savenow != NULL && SEE(']')) || ISERR());
1624 NEXT();
1625 NOERR();
1626 v->wordchrs = left;
1627 }
1628
1629 /*
1630 * processlacon - generate the NFA representation of a LACON
1631 *
1632 * In the general case this is just newlacon() + newarc(), but some cases
1633 * can be optimized.
1634 */
1635 static void
processlacon(struct vars * v,struct state * begin,struct state * end,int latype,struct state * lp,struct state * rp)1636 processlacon(struct vars *v,
1637 struct state *begin, /* start of parsed LACON sub-re */
1638 struct state *end, /* end of parsed LACON sub-re */
1639 int latype,
1640 struct state *lp, /* left state to hang it on */
1641 struct state *rp) /* right state to hang it on */
1642 {
1643 struct state *s1;
1644 int n;
1645
1646 /*
1647 * Check for lookaround RE consisting of a single plain color arc (or set
1648 * of arcs); this would typically be a simple chr or a bracket expression.
1649 */
1650 s1 = single_color_transition(begin, end);
1651 switch (latype)
1652 {
1653 case LATYPE_AHEAD_POS:
1654 /* If lookahead RE is just colorset C, convert to AHEAD(C) */
1655 if (s1 != NULL)
1656 {
1657 cloneouts(v->nfa, s1, lp, rp, AHEAD);
1658 return;
1659 }
1660 break;
1661 case LATYPE_AHEAD_NEG:
1662 /* If lookahead RE is just colorset C, convert to AHEAD(^C)|$ */
1663 if (s1 != NULL)
1664 {
1665 colorcomplement(v->nfa, v->cm, AHEAD, s1, lp, rp);
1666 newarc(v->nfa, '$', 1, lp, rp);
1667 newarc(v->nfa, '$', 0, lp, rp);
1668 return;
1669 }
1670 break;
1671 case LATYPE_BEHIND_POS:
1672 /* If lookbehind RE is just colorset C, convert to BEHIND(C) */
1673 if (s1 != NULL)
1674 {
1675 cloneouts(v->nfa, s1, lp, rp, BEHIND);
1676 return;
1677 }
1678 break;
1679 case LATYPE_BEHIND_NEG:
1680 /* If lookbehind RE is just colorset C, convert to BEHIND(^C)|^ */
1681 if (s1 != NULL)
1682 {
1683 colorcomplement(v->nfa, v->cm, BEHIND, s1, lp, rp);
1684 newarc(v->nfa, '^', 1, lp, rp);
1685 newarc(v->nfa, '^', 0, lp, rp);
1686 return;
1687 }
1688 break;
1689 default:
1690 assert(NOTREACHED);
1691 }
1692
1693 /* General case: we need a LACON subre and arc */
1694 n = newlacon(v, begin, end, latype);
1695 newarc(v->nfa, LACON, n, lp, rp);
1696 }
1697
1698 /*
1699 * subre - allocate a subre
1700 */
1701 static struct subre *
subre(struct vars * v,int op,int flags,struct state * begin,struct state * end)1702 subre(struct vars *v,
1703 int op,
1704 int flags,
1705 struct state *begin,
1706 struct state *end)
1707 {
1708 struct subre *ret = v->treefree;
1709
1710 /*
1711 * Checking for stack overflow here is sufficient to protect parse() and
1712 * its recursive subroutines.
1713 */
1714 if (STACK_TOO_DEEP(v->re))
1715 {
1716 ERR(REG_ETOOBIG);
1717 return NULL;
1718 }
1719
1720 if (ret != NULL)
1721 v->treefree = ret->left;
1722 else
1723 {
1724 ret = (struct subre *) MALLOC(sizeof(struct subre));
1725 if (ret == NULL)
1726 {
1727 ERR(REG_ESPACE);
1728 return NULL;
1729 }
1730 ret->chain = v->treechain;
1731 v->treechain = ret;
1732 }
1733
1734 assert(strchr("=b|.*(", op) != NULL);
1735
1736 ret->op = op;
1737 ret->flags = flags;
1738 ret->id = 0; /* will be assigned later */
1739 ret->subno = 0;
1740 ret->min = ret->max = 1;
1741 ret->left = NULL;
1742 ret->right = NULL;
1743 ret->begin = begin;
1744 ret->end = end;
1745 ZAPCNFA(ret->cnfa);
1746
1747 return ret;
1748 }
1749
1750 /*
1751 * freesubre - free a subRE subtree
1752 */
1753 static void
freesubre(struct vars * v,struct subre * sr)1754 freesubre(struct vars *v, /* might be NULL */
1755 struct subre *sr)
1756 {
1757 if (sr == NULL)
1758 return;
1759
1760 if (sr->left != NULL)
1761 freesubre(v, sr->left);
1762 if (sr->right != NULL)
1763 freesubre(v, sr->right);
1764
1765 freesrnode(v, sr);
1766 }
1767
1768 /*
1769 * freesrnode - free one node in a subRE subtree
1770 */
1771 static void
freesrnode(struct vars * v,struct subre * sr)1772 freesrnode(struct vars *v, /* might be NULL */
1773 struct subre *sr)
1774 {
1775 if (sr == NULL)
1776 return;
1777
1778 if (!NULLCNFA(sr->cnfa))
1779 freecnfa(&sr->cnfa);
1780 sr->flags = 0;
1781
1782 if (v != NULL && v->treechain != NULL)
1783 {
1784 /* we're still parsing, maybe we can reuse the subre */
1785 sr->left = v->treefree;
1786 v->treefree = sr;
1787 }
1788 else
1789 FREE(sr);
1790 }
1791
1792 /*
1793 * optst - optimize a subRE subtree
1794 */
1795 static void
optst(struct vars * v,struct subre * t)1796 optst(struct vars *v,
1797 struct subre *t)
1798 {
1799 /*
1800 * DGP (2007-11-13): I assume it was the programmer's intent to eventually
1801 * come back and add code to optimize subRE trees, but the routine coded
1802 * just spends effort traversing the tree and doing nothing. We can do
1803 * nothing with less effort.
1804 */
1805 return;
1806 }
1807
1808 /*
1809 * numst - number tree nodes (assigning "id" indexes)
1810 */
1811 static int /* next number */
numst(struct subre * t,int start)1812 numst(struct subre *t,
1813 int start) /* starting point for subtree numbers */
1814 {
1815 int i;
1816
1817 assert(t != NULL);
1818
1819 i = start;
1820 t->id = (short) i++;
1821 if (t->left != NULL)
1822 i = numst(t->left, i);
1823 if (t->right != NULL)
1824 i = numst(t->right, i);
1825 return i;
1826 }
1827
1828 /*
1829 * markst - mark tree nodes as INUSE
1830 *
1831 * Note: this is a great deal more subtle than it looks. During initial
1832 * parsing of a regex, all subres are linked into the treechain list;
1833 * discarded ones are also linked into the treefree list for possible reuse.
1834 * After we are done creating all subres required for a regex, we run markst()
1835 * then cleanst(), which results in discarding all subres not reachable from
1836 * v->tree. We then clear v->treechain, indicating that subres must be found
1837 * by descending from v->tree. This changes the behavior of freesubre(): it
1838 * will henceforth FREE() unwanted subres rather than sticking them into the
1839 * treefree list. (Doing that any earlier would result in dangling links in
1840 * the treechain list.) This all means that freev() will clean up correctly
1841 * if invoked before or after markst()+cleanst(); but it would not work if
1842 * called partway through this state conversion, so we mustn't error out
1843 * in or between these two functions.
1844 */
1845 static void
markst(struct subre * t)1846 markst(struct subre *t)
1847 {
1848 assert(t != NULL);
1849
1850 t->flags |= INUSE;
1851 if (t->left != NULL)
1852 markst(t->left);
1853 if (t->right != NULL)
1854 markst(t->right);
1855 }
1856
1857 /*
1858 * cleanst - free any tree nodes not marked INUSE
1859 */
1860 static void
cleanst(struct vars * v)1861 cleanst(struct vars *v)
1862 {
1863 struct subre *t;
1864 struct subre *next;
1865
1866 for (t = v->treechain; t != NULL; t = next)
1867 {
1868 next = t->chain;
1869 if (!(t->flags & INUSE))
1870 FREE(t);
1871 }
1872 v->treechain = NULL;
1873 v->treefree = NULL; /* just on general principles */
1874 }
1875
1876 /*
1877 * nfatree - turn a subRE subtree into a tree of compacted NFAs
1878 */
1879 static long /* optimize results from top node */
nfatree(struct vars * v,struct subre * t,FILE * f)1880 nfatree(struct vars *v,
1881 struct subre *t,
1882 FILE *f) /* for debug output */
1883 {
1884 assert(t != NULL && t->begin != NULL);
1885
1886 if (t->left != NULL)
1887 (DISCARD) nfatree(v, t->left, f);
1888 if (t->right != NULL)
1889 (DISCARD) nfatree(v, t->right, f);
1890
1891 return nfanode(v, t, 0, f);
1892 }
1893
1894 /*
1895 * nfanode - do one NFA for nfatree or lacons
1896 *
1897 * If converttosearch is true, apply makesearch() to the NFA.
1898 */
1899 static long /* optimize results */
nfanode(struct vars * v,struct subre * t,int converttosearch,FILE * f)1900 nfanode(struct vars *v,
1901 struct subre *t,
1902 int converttosearch,
1903 FILE *f) /* for debug output */
1904 {
1905 struct nfa *nfa;
1906 long ret = 0;
1907
1908 assert(t->begin != NULL);
1909
1910 #ifdef REG_DEBUG
1911 if (f != NULL)
1912 {
1913 char idbuf[50];
1914
1915 fprintf(f, "\n\n\n========= TREE NODE %s ==========\n",
1916 stid(t, idbuf, sizeof(idbuf)));
1917 }
1918 #endif
1919 nfa = newnfa(v, v->cm, v->nfa);
1920 NOERRZ();
1921 dupnfa(nfa, t->begin, t->end, nfa->init, nfa->final);
1922 if (!ISERR())
1923 specialcolors(nfa);
1924 if (!ISERR())
1925 ret = optimize(nfa, f);
1926 if (converttosearch && !ISERR())
1927 makesearch(v, nfa);
1928 if (!ISERR())
1929 compact(nfa, &t->cnfa);
1930
1931 freenfa(nfa);
1932 return ret;
1933 }
1934
1935 /*
1936 * newlacon - allocate a lookaround-constraint subRE
1937 */
1938 static int /* lacon number */
newlacon(struct vars * v,struct state * begin,struct state * end,int latype)1939 newlacon(struct vars *v,
1940 struct state *begin,
1941 struct state *end,
1942 int latype)
1943 {
1944 int n;
1945 struct subre *newlacons;
1946 struct subre *sub;
1947
1948 if (v->nlacons == 0)
1949 {
1950 n = 1; /* skip 0th */
1951 newlacons = (struct subre *) MALLOC(2 * sizeof(struct subre));
1952 }
1953 else
1954 {
1955 n = v->nlacons;
1956 newlacons = (struct subre *) REALLOC(v->lacons,
1957 (n + 1) * sizeof(struct subre));
1958 }
1959 if (newlacons == NULL)
1960 {
1961 ERR(REG_ESPACE);
1962 return 0;
1963 }
1964 v->lacons = newlacons;
1965 v->nlacons = n + 1;
1966 sub = &v->lacons[n];
1967 sub->begin = begin;
1968 sub->end = end;
1969 sub->subno = latype;
1970 ZAPCNFA(sub->cnfa);
1971 return n;
1972 }
1973
1974 /*
1975 * freelacons - free lookaround-constraint subRE vector
1976 */
1977 static void
freelacons(struct subre * subs,int n)1978 freelacons(struct subre *subs,
1979 int n)
1980 {
1981 struct subre *sub;
1982 int i;
1983
1984 assert(n > 0);
1985 for (sub = subs + 1, i = n - 1; i > 0; sub++, i--) /* no 0th */
1986 if (!NULLCNFA(sub->cnfa))
1987 freecnfa(&sub->cnfa);
1988 FREE(subs);
1989 }
1990
1991 /*
1992 * rfree - free a whole RE (insides of regfree)
1993 */
1994 static void
rfree(regex_t * re)1995 rfree(regex_t *re)
1996 {
1997 struct guts *g;
1998
1999 if (re == NULL || re->re_magic != REMAGIC)
2000 return;
2001
2002 re->re_magic = 0; /* invalidate RE */
2003 g = (struct guts *) re->re_guts;
2004 re->re_guts = NULL;
2005 re->re_fns = NULL;
2006 if (g != NULL)
2007 {
2008 g->magic = 0;
2009 freecm(&g->cmap);
2010 if (g->tree != NULL)
2011 freesubre((struct vars *) NULL, g->tree);
2012 if (g->lacons != NULL)
2013 freelacons(g->lacons, g->nlacons);
2014 if (!NULLCNFA(g->search))
2015 freecnfa(&g->search);
2016 FREE(g);
2017 }
2018 }
2019
2020 /*
2021 * rcancelrequested - check for external request to cancel regex operation
2022 *
2023 * Return nonzero to fail the operation with error code REG_CANCEL,
2024 * zero to keep going
2025 *
2026 * The current implementation is Postgres-specific. If we ever get around
2027 * to splitting the regex code out as a standalone library, there will need
2028 * to be some API to let applications define a callback function for this.
2029 */
2030 static int
rcancelrequested(void)2031 rcancelrequested(void)
2032 {
2033 return InterruptPending && (QueryCancelPending || ProcDiePending);
2034 }
2035
2036 /*
2037 * rstacktoodeep - check for stack getting dangerously deep
2038 *
2039 * Return nonzero to fail the operation with error code REG_ETOOBIG,
2040 * zero to keep going
2041 *
2042 * The current implementation is Postgres-specific. If we ever get around
2043 * to splitting the regex code out as a standalone library, there will need
2044 * to be some API to let applications define a callback function for this.
2045 */
2046 static int
rstacktoodeep(void)2047 rstacktoodeep(void)
2048 {
2049 return stack_is_too_deep();
2050 }
2051
2052 #ifdef REG_DEBUG
2053
2054 /*
2055 * dump - dump an RE in human-readable form
2056 */
2057 static void
dump(regex_t * re,FILE * f)2058 dump(regex_t *re,
2059 FILE *f)
2060 {
2061 struct guts *g;
2062 int i;
2063
2064 if (re->re_magic != REMAGIC)
2065 fprintf(f, "bad magic number (0x%x not 0x%x)\n", re->re_magic,
2066 REMAGIC);
2067 if (re->re_guts == NULL)
2068 {
2069 fprintf(f, "NULL guts!!!\n");
2070 return;
2071 }
2072 g = (struct guts *) re->re_guts;
2073 if (g->magic != GUTSMAGIC)
2074 fprintf(f, "bad guts magic number (0x%x not 0x%x)\n", g->magic,
2075 GUTSMAGIC);
2076
2077 fprintf(f, "\n\n\n========= DUMP ==========\n");
2078 fprintf(f, "nsub %d, info 0%lo, csize %d, ntree %d\n",
2079 (int) re->re_nsub, re->re_info, re->re_csize, g->ntree);
2080
2081 dumpcolors(&g->cmap, f);
2082 if (!NULLCNFA(g->search))
2083 {
2084 fprintf(f, "\nsearch:\n");
2085 dumpcnfa(&g->search, f);
2086 }
2087 for (i = 1; i < g->nlacons; i++)
2088 {
2089 struct subre *lasub = &g->lacons[i];
2090 const char *latype;
2091
2092 switch (lasub->subno)
2093 {
2094 case LATYPE_AHEAD_POS:
2095 latype = "positive lookahead";
2096 break;
2097 case LATYPE_AHEAD_NEG:
2098 latype = "negative lookahead";
2099 break;
2100 case LATYPE_BEHIND_POS:
2101 latype = "positive lookbehind";
2102 break;
2103 case LATYPE_BEHIND_NEG:
2104 latype = "negative lookbehind";
2105 break;
2106 default:
2107 latype = "???";
2108 break;
2109 }
2110 fprintf(f, "\nla%d (%s):\n", i, latype);
2111 dumpcnfa(&lasub->cnfa, f);
2112 }
2113 fprintf(f, "\n");
2114 dumpst(g->tree, f, 0);
2115 }
2116
2117 /*
2118 * dumpst - dump a subRE tree
2119 */
2120 static void
dumpst(struct subre * t,FILE * f,int nfapresent)2121 dumpst(struct subre *t,
2122 FILE *f,
2123 int nfapresent) /* is the original NFA still around? */
2124 {
2125 if (t == NULL)
2126 fprintf(f, "null tree\n");
2127 else
2128 stdump(t, f, nfapresent);
2129 fflush(f);
2130 }
2131
2132 /*
2133 * stdump - recursive guts of dumpst
2134 */
2135 static void
stdump(struct subre * t,FILE * f,int nfapresent)2136 stdump(struct subre *t,
2137 FILE *f,
2138 int nfapresent) /* is the original NFA still around? */
2139 {
2140 char idbuf[50];
2141
2142 fprintf(f, "%s. `%c'", stid(t, idbuf, sizeof(idbuf)), t->op);
2143 if (t->flags & LONGER)
2144 fprintf(f, " longest");
2145 if (t->flags & SHORTER)
2146 fprintf(f, " shortest");
2147 if (t->flags & MIXED)
2148 fprintf(f, " hasmixed");
2149 if (t->flags & CAP)
2150 fprintf(f, " hascapture");
2151 if (t->flags & BACKR)
2152 fprintf(f, " hasbackref");
2153 if (!(t->flags & INUSE))
2154 fprintf(f, " UNUSED");
2155 if (t->subno != 0)
2156 fprintf(f, " (#%d)", t->subno);
2157 if (t->min != 1 || t->max != 1)
2158 {
2159 fprintf(f, " {%d,", t->min);
2160 if (t->max != DUPINF)
2161 fprintf(f, "%d", t->max);
2162 fprintf(f, "}");
2163 }
2164 if (nfapresent)
2165 fprintf(f, " %ld-%ld", (long) t->begin->no, (long) t->end->no);
2166 if (t->left != NULL)
2167 fprintf(f, " L:%s", stid(t->left, idbuf, sizeof(idbuf)));
2168 if (t->right != NULL)
2169 fprintf(f, " R:%s", stid(t->right, idbuf, sizeof(idbuf)));
2170 if (!NULLCNFA(t->cnfa))
2171 {
2172 fprintf(f, "\n");
2173 dumpcnfa(&t->cnfa, f);
2174 }
2175 fprintf(f, "\n");
2176 if (t->left != NULL)
2177 stdump(t->left, f, nfapresent);
2178 if (t->right != NULL)
2179 stdump(t->right, f, nfapresent);
2180 }
2181
2182 /*
2183 * stid - identify a subtree node for dumping
2184 */
2185 static const char * /* points to buf or constant string */
stid(struct subre * t,char * buf,size_t bufsize)2186 stid(struct subre *t,
2187 char *buf,
2188 size_t bufsize)
2189 {
2190 /* big enough for hex int or decimal t->id? */
2191 if (bufsize < sizeof(void *) * 2 + 3 || bufsize < sizeof(t->id) * 3 + 1)
2192 return "unable";
2193 if (t->id != 0)
2194 sprintf(buf, "%d", t->id);
2195 else
2196 sprintf(buf, "%p", t);
2197 return buf;
2198 }
2199 #endif /* REG_DEBUG */
2200
2201
2202 #include "regc_lex.c"
2203 #include "regc_color.c"
2204 #include "regc_nfa.c"
2205 #include "regc_cvec.c"
2206 #include "regc_pg_locale.c"
2207 #include "regc_locale.c"
2208