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 struct subre *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 void charclass(struct vars *, enum char_classes,
50 struct state *, struct state *);
51 static void charclasscomplement(struct vars *, enum char_classes,
52 struct state *, struct state *);
53 static int scannum(struct vars *);
54 static void repeat(struct vars *, struct state *, struct state *, int, int);
55 static void bracket(struct vars *, struct state *, struct state *);
56 static void cbracket(struct vars *, struct state *, struct state *);
57 static void brackpart(struct vars *, struct state *, struct state *, bool *);
58 static const chr *scanplain(struct vars *);
59 static void onechr(struct vars *, chr, struct state *, struct state *);
60 static void optimizebracket(struct vars *, struct state *, struct state *);
61 static void wordchrs(struct vars *);
62 static void processlacon(struct vars *, struct state *, struct state *, int,
63 struct state *, struct state *);
64 static struct subre *subre(struct vars *, int, int, struct state *, struct state *);
65 static void freesubre(struct vars *, struct subre *);
66 static void freesubreandsiblings(struct vars *, struct subre *);
67 static void freesrnode(struct vars *, struct subre *);
68 static void optst(struct vars *, struct subre *);
69 static int numst(struct subre *, int);
70 static void markst(struct subre *);
71 static void cleanst(struct vars *);
72 static long nfatree(struct vars *, struct subre *, FILE *);
73 static long nfanode(struct vars *, struct subre *, int, FILE *);
74 static int newlacon(struct vars *, struct state *, struct state *, int);
75 static void freelacons(struct subre *, int);
76 static void rfree(regex_t *);
77 static int rcancelrequested(void);
78 static int rstacktoodeep(void);
79
80 #ifdef REG_DEBUG
81 static void dump(regex_t *, FILE *);
82 static void dumpst(struct subre *, FILE *, int);
83 static void stdump(struct subre *, FILE *, int);
84 static const char *stid(struct subre *, char *, size_t);
85 #endif
86 /* === regc_lex.c === */
87 static void lexstart(struct vars *);
88 static void prefixes(struct vars *);
89 static int next(struct vars *);
90 static int lexescape(struct vars *);
91 static chr lexdigits(struct vars *, int, int, int);
92 static int brenext(struct vars *, chr);
93 static void skip(struct vars *);
94 static chr newline(void);
95 static chr chrnamed(struct vars *, const chr *, const chr *, chr);
96
97 /* === regc_color.c === */
98 static void initcm(struct vars *, struct colormap *);
99 static void freecm(struct colormap *);
100 static color maxcolor(struct colormap *);
101 static color newcolor(struct colormap *);
102 static void freecolor(struct colormap *, color);
103 static color pseudocolor(struct colormap *);
104 static color subcolor(struct colormap *, chr);
105 static color subcolorhi(struct colormap *, color *);
106 static color newsub(struct colormap *, color);
107 static int newhicolorrow(struct colormap *, int);
108 static void newhicolorcols(struct colormap *);
109 static void subcolorcvec(struct vars *, struct cvec *, struct state *, struct state *);
110 static void subcoloronechr(struct vars *, chr, struct state *, struct state *, color *);
111 static void subcoloronerange(struct vars *, chr, chr, struct state *, struct state *, color *);
112 static void subcoloronerow(struct vars *, int, struct state *, struct state *, color *);
113 static void okcolors(struct nfa *, struct colormap *);
114 static void colorchain(struct colormap *, struct arc *);
115 static void uncolorchain(struct colormap *, struct arc *);
116 static void rainbow(struct nfa *, struct colormap *, int, color, struct state *, struct state *);
117 static void colorcomplement(struct nfa *, struct colormap *, int, struct state *, struct state *, struct state *);
118
119 #ifdef REG_DEBUG
120 static void dumpcolors(struct colormap *, FILE *);
121 static void dumpchr(chr, FILE *);
122 #endif
123 /* === regc_nfa.c === */
124 static struct nfa *newnfa(struct vars *, struct colormap *, struct nfa *);
125 static void freenfa(struct nfa *);
126 static struct state *newstate(struct nfa *);
127 static struct state *newfstate(struct nfa *, int flag);
128 static void dropstate(struct nfa *, struct state *);
129 static void freestate(struct nfa *, struct state *);
130 static void newarc(struct nfa *, int, color, struct state *, struct state *);
131 static void createarc(struct nfa *, int, color, struct state *, struct state *);
132 static struct arc *allocarc(struct nfa *);
133 static void freearc(struct nfa *, struct arc *);
134 static void changearcsource(struct arc *, struct state *);
135 static void changearctarget(struct arc *, struct state *);
136 static int hasnonemptyout(struct state *);
137 static struct arc *findarc(struct state *, int, color);
138 static void cparc(struct nfa *, struct arc *, struct state *, struct state *);
139 static void sortins(struct nfa *, struct state *);
140 static int sortins_cmp(const void *, const void *);
141 static void sortouts(struct nfa *, struct state *);
142 static int sortouts_cmp(const void *, const void *);
143 static void moveins(struct nfa *, struct state *, struct state *);
144 static void copyins(struct nfa *, struct state *, struct state *);
145 static void mergeins(struct nfa *, struct state *, struct arc **, int);
146 static void moveouts(struct nfa *, struct state *, struct state *);
147 static void copyouts(struct nfa *, struct state *, struct state *);
148 static void cloneouts(struct nfa *, struct state *, struct state *, struct state *, int);
149 static void delsub(struct nfa *, struct state *, struct state *);
150 static void deltraverse(struct nfa *, struct state *, struct state *);
151 static void dupnfa(struct nfa *, struct state *, struct state *, struct state *, struct state *);
152 static void duptraverse(struct nfa *, struct state *, struct state *);
153 static void removeconstraints(struct nfa *, struct state *, struct state *);
154 static void removetraverse(struct nfa *, struct state *);
155 static void cleartraverse(struct nfa *, struct state *);
156 static struct state *single_color_transition(struct state *, struct state *);
157 static void specialcolors(struct nfa *);
158 static long optimize(struct nfa *, FILE *);
159 static void pullback(struct nfa *, FILE *);
160 static int pull(struct nfa *, struct arc *, struct state **);
161 static void pushfwd(struct nfa *, FILE *);
162 static int push(struct nfa *, struct arc *, struct state **);
163
164 #define INCOMPATIBLE 1 /* destroys arc */
165 #define SATISFIED 2 /* constraint satisfied */
166 #define COMPATIBLE 3 /* compatible but not satisfied yet */
167 #define REPLACEARC 4 /* replace arc's color with constraint color */
168 static int combine(struct nfa *nfa, struct arc *con, struct arc *a);
169 static void fixempties(struct nfa *, FILE *);
170 static struct state *emptyreachable(struct nfa *, struct state *,
171 struct state *, struct arc **);
172 static int isconstraintarc(struct arc *);
173 static int hasconstraintout(struct state *);
174 static void fixconstraintloops(struct nfa *, FILE *);
175 static int findconstraintloop(struct nfa *, struct state *);
176 static void breakconstraintloop(struct nfa *, struct state *);
177 static void clonesuccessorstates(struct nfa *, struct state *, struct state *,
178 struct state *, struct arc *,
179 char *, char *, int);
180 static void cleanup(struct nfa *);
181 static void markreachable(struct nfa *, struct state *, struct state *, struct state *);
182 static void markcanreach(struct nfa *, struct state *, struct state *, struct state *);
183 static long analyze(struct nfa *);
184 static void checkmatchall(struct nfa *);
185 static bool checkmatchall_recurse(struct nfa *, struct state *, bool **);
186 static bool check_out_colors_match(struct state *, color, color);
187 static bool check_in_colors_match(struct state *, color, color);
188 static void compact(struct nfa *, struct cnfa *);
189 static void carcsort(struct carc *, size_t);
190 static int carc_cmp(const void *, const void *);
191 static void freecnfa(struct cnfa *);
192 static void dumpnfa(struct nfa *, FILE *);
193
194 #ifdef REG_DEBUG
195 static void dumpstate(struct state *, FILE *);
196 static void dumparcs(struct state *, FILE *);
197 static void dumparc(struct arc *, struct state *, FILE *);
198 static void dumpcnfa(struct cnfa *, FILE *);
199 static void dumpcstate(int, struct cnfa *, FILE *);
200 #endif
201 /* === regc_cvec.c === */
202 static struct cvec *newcvec(int, int);
203 static struct cvec *clearcvec(struct cvec *);
204 static void addchr(struct cvec *, chr);
205 static void addrange(struct cvec *, chr, chr);
206 static struct cvec *getcvec(struct vars *, int, int);
207 static void freecvec(struct cvec *);
208
209 /* === regc_pg_locale.c === */
210 static int pg_wc_isdigit(pg_wchar c);
211 static int pg_wc_isalpha(pg_wchar c);
212 static int pg_wc_isalnum(pg_wchar c);
213 static int pg_wc_isword(pg_wchar c);
214 static int pg_wc_isupper(pg_wchar c);
215 static int pg_wc_islower(pg_wchar c);
216 static int pg_wc_isgraph(pg_wchar c);
217 static int pg_wc_isprint(pg_wchar c);
218 static int pg_wc_ispunct(pg_wchar c);
219 static int pg_wc_isspace(pg_wchar c);
220 static pg_wchar pg_wc_toupper(pg_wchar c);
221 static pg_wchar pg_wc_tolower(pg_wchar c);
222
223 /* === regc_locale.c === */
224 static chr element(struct vars *, const chr *, const chr *);
225 static struct cvec *range(struct vars *, chr, chr, int);
226 static int before(chr, chr);
227 static struct cvec *eclass(struct vars *, chr, int);
228 static enum char_classes lookupcclass(struct vars *, const chr *, const chr *);
229 static struct cvec *cclasscvec(struct vars *, enum char_classes, int);
230 static int cclass_column_index(struct colormap *, chr);
231 static struct cvec *allcases(struct vars *, chr);
232 static int cmp(const chr *, const chr *, size_t);
233 static int casecmp(const chr *, const chr *, size_t);
234
235
236 /* internal variables, bundled for easy passing around */
237 struct vars
238 {
239 regex_t *re;
240 const chr *now; /* scan pointer into string */
241 const chr *stop; /* end of string */
242 int err; /* error code (0 if none) */
243 int cflags; /* copy of compile flags */
244 int lasttype; /* type of previous token */
245 int nexttype; /* type of next token */
246 chr nextvalue; /* value (if any) of next token */
247 int lexcon; /* lexical context type (see regc_lex.c) */
248 int nsubexp; /* subexpression count */
249 struct subre **subs; /* subRE pointer vector */
250 size_t nsubs; /* length of vector */
251 struct subre *sub10[10]; /* initial vector, enough for most */
252 struct nfa *nfa; /* the NFA */
253 struct colormap *cm; /* character color map */
254 color nlcolor; /* color of newline */
255 struct state *wordchrs; /* state in nfa holding word-char outarcs */
256 struct subre *tree; /* subexpression tree */
257 struct subre *treechain; /* all tree nodes allocated */
258 struct subre *treefree; /* any free tree nodes */
259 int ntree; /* number of tree nodes, plus one */
260 struct cvec *cv; /* interface cvec */
261 struct cvec *cv2; /* utility cvec */
262 struct subre *lacons; /* lookaround-constraint vector */
263 int nlacons; /* size of lacons[]; note that only slots
264 * numbered 1 .. nlacons-1 are used */
265 size_t spaceused; /* approx. space used for compilation */
266 };
267
268 /* parsing macros; most know that `v' is the struct vars pointer */
269 #define NEXT() (next(v)) /* advance by one token */
270 #define SEE(t) (v->nexttype == (t)) /* is next token this? */
271 #define EAT(t) (SEE(t) && next(v)) /* if next is this, swallow it */
272 #define VISERR(vv) ((vv)->err != 0) /* have we seen an error yet? */
273 #define ISERR() VISERR(v)
274 #define VERR(vv,e) ((vv)->nexttype = EOS, \
275 (vv)->err = ((vv)->err ? (vv)->err : (e)))
276 #define ERR(e) VERR(v, e) /* record an error */
277 #define NOERR() {if (ISERR()) return;} /* if error seen, return */
278 #define NOERRN() {if (ISERR()) return NULL;} /* NOERR with retval */
279 #define NOERRZ() {if (ISERR()) return 0;} /* NOERR with retval */
280 #define INSIST(c, e) do { if (!(c)) ERR(e); } while (0) /* error if c false */
281 #define NOTE(b) (v->re->re_info |= (b)) /* note visible condition */
282 #define EMPTYARC(x, y) newarc(v->nfa, EMPTY, 0, x, y)
283
284 /* token type codes, some also used as NFA arc types */
285 #define EMPTY 'n' /* no token present */
286 #define EOS 'e' /* end of string */
287 #define PLAIN 'p' /* ordinary character */
288 #define DIGIT 'd' /* digit (in bound) */
289 #define BACKREF 'b' /* back reference */
290 #define COLLEL 'I' /* start of [. */
291 #define ECLASS 'E' /* start of [= */
292 #define CCLASS 'C' /* start of [: */
293 #define END 'X' /* end of [. [= [: */
294 #define CCLASSS 's' /* char class shorthand escape */
295 #define CCLASSC 'c' /* complement char class shorthand escape */
296 #define RANGE 'R' /* - within [] which might be range delim. */
297 #define LACON 'L' /* lookaround constraint subRE */
298 #define AHEAD 'a' /* color-lookahead arc */
299 #define BEHIND 'r' /* color-lookbehind arc */
300 #define WBDRY 'w' /* word boundary constraint */
301 #define NWBDRY 'W' /* non-word-boundary constraint */
302 #define SBEGIN 'A' /* beginning of string (even if not BOL) */
303 #define SEND 'Z' /* end of string (even if not EOL) */
304
305 /* is an arc colored, and hence should belong to a color chain? */
306 /* the test on "co" eliminates RAINBOW arcs, which we don't bother to chain */
307 #define COLORED(a) \
308 ((a)->co >= 0 && \
309 ((a)->type == PLAIN || (a)->type == AHEAD || (a)->type == BEHIND))
310
311
312 /* static function list */
313 static const struct fns functions = {
314 rfree, /* regfree insides */
315 rcancelrequested, /* check for cancel request */
316 rstacktoodeep /* check for stack getting dangerously deep */
317 };
318
319
320
321 /*
322 * pg_regcomp - compile regular expression
323 *
324 * Note: on failure, no resources remain allocated, so pg_regfree()
325 * need not be applied to re.
326 */
327 int
pg_regcomp(regex_t * re,const chr * string,size_t len,int flags,Oid collation)328 pg_regcomp(regex_t *re,
329 const chr *string,
330 size_t len,
331 int flags,
332 Oid collation)
333 {
334 struct vars var;
335 struct vars *v = &var;
336 struct guts *g;
337 int i;
338 size_t j;
339
340 #ifdef REG_DEBUG
341 FILE *debug = (flags & REG_PROGRESS) ? stdout : (FILE *) NULL;
342 #else
343 FILE *debug = (FILE *) NULL;
344 #endif
345
346 #define CNOERR() { if (ISERR()) return freev(v, v->err); }
347
348 /* sanity checks */
349
350 if (re == NULL || string == NULL)
351 return REG_INVARG;
352 if ((flags & REG_QUOTE) &&
353 (flags & (REG_ADVANCED | REG_EXPANDED | REG_NEWLINE)))
354 return REG_INVARG;
355 if (!(flags & REG_EXTENDED) && (flags & REG_ADVF))
356 return REG_INVARG;
357
358 /* Initialize locale-dependent support */
359 pg_set_regex_collation(collation);
360
361 /* initial setup (after which freev() is callable) */
362 v->re = re;
363 v->now = string;
364 v->stop = v->now + len;
365 v->err = 0;
366 v->cflags = flags;
367 v->nsubexp = 0;
368 v->subs = v->sub10;
369 v->nsubs = 10;
370 for (j = 0; j < v->nsubs; j++)
371 v->subs[j] = NULL;
372 v->nfa = NULL;
373 v->cm = NULL;
374 v->nlcolor = COLORLESS;
375 v->wordchrs = NULL;
376 v->tree = NULL;
377 v->treechain = NULL;
378 v->treefree = NULL;
379 v->cv = NULL;
380 v->cv2 = NULL;
381 v->lacons = NULL;
382 v->nlacons = 0;
383 v->spaceused = 0;
384 re->re_magic = REMAGIC;
385 re->re_info = 0; /* bits get set during parse */
386 re->re_csize = sizeof(chr);
387 re->re_collation = collation;
388 re->re_guts = NULL;
389 re->re_fns = VS(&functions);
390
391 /* more complex setup, malloced things */
392 re->re_guts = VS(MALLOC(sizeof(struct guts)));
393 if (re->re_guts == NULL)
394 return freev(v, REG_ESPACE);
395 g = (struct guts *) re->re_guts;
396 g->tree = NULL;
397 initcm(v, &g->cmap);
398 v->cm = &g->cmap;
399 g->lacons = NULL;
400 g->nlacons = 0;
401 ZAPCNFA(g->search);
402 v->nfa = newnfa(v, v->cm, (struct nfa *) NULL);
403 CNOERR();
404 /* set up a reasonably-sized transient cvec for getcvec usage */
405 v->cv = newcvec(100, 20);
406 if (v->cv == NULL)
407 return freev(v, REG_ESPACE);
408
409 /* parsing */
410 lexstart(v); /* also handles prefixes */
411 if ((v->cflags & REG_NLSTOP) || (v->cflags & REG_NLANCH))
412 {
413 /* assign newline a unique color */
414 v->nlcolor = subcolor(v->cm, newline());
415 okcolors(v->nfa, v->cm);
416 }
417 CNOERR();
418 v->tree = parse(v, EOS, PLAIN, v->nfa->init, v->nfa->final);
419 assert(SEE(EOS)); /* even if error; ISERR() => SEE(EOS) */
420 CNOERR();
421 assert(v->tree != NULL);
422
423 /* finish setup of nfa and its subre tree */
424 specialcolors(v->nfa);
425 CNOERR();
426 #ifdef REG_DEBUG
427 if (debug != NULL)
428 {
429 fprintf(debug, "\n\n\n========= RAW ==========\n");
430 dumpnfa(v->nfa, debug);
431 dumpst(v->tree, debug, 1);
432 }
433 #endif
434 optst(v, v->tree);
435 v->ntree = numst(v->tree, 1);
436 markst(v->tree);
437 cleanst(v);
438 #ifdef REG_DEBUG
439 if (debug != NULL)
440 {
441 fprintf(debug, "\n\n\n========= TREE FIXED ==========\n");
442 dumpst(v->tree, debug, 1);
443 }
444 #endif
445
446 /* build compacted NFAs for tree and lacons */
447 re->re_info |= nfatree(v, v->tree, debug);
448 CNOERR();
449 assert(v->nlacons == 0 || v->lacons != NULL);
450 for (i = 1; i < v->nlacons; i++)
451 {
452 struct subre *lasub = &v->lacons[i];
453
454 #ifdef REG_DEBUG
455 if (debug != NULL)
456 fprintf(debug, "\n\n\n========= LA%d ==========\n", i);
457 #endif
458
459 /* Prepend .* to pattern if it's a lookbehind LACON */
460 nfanode(v, lasub, !LATYPE_IS_AHEAD(lasub->latype), debug);
461 }
462 CNOERR();
463 if (v->tree->flags & SHORTER)
464 NOTE(REG_USHORTEST);
465
466 /* build compacted NFAs for tree, lacons, fast search */
467 #ifdef REG_DEBUG
468 if (debug != NULL)
469 fprintf(debug, "\n\n\n========= SEARCH ==========\n");
470 #endif
471 /* can sacrifice main NFA now, so use it as work area */
472 (DISCARD) optimize(v->nfa, debug);
473 CNOERR();
474 makesearch(v, v->nfa);
475 CNOERR();
476 compact(v->nfa, &g->search);
477 CNOERR();
478
479 /* looks okay, package it up */
480 re->re_nsub = v->nsubexp;
481 v->re = NULL; /* freev no longer frees re */
482 g->magic = GUTSMAGIC;
483 g->cflags = v->cflags;
484 g->info = re->re_info;
485 g->nsub = re->re_nsub;
486 g->tree = v->tree;
487 v->tree = NULL;
488 g->ntree = v->ntree;
489 g->compare = (v->cflags & REG_ICASE) ? casecmp : cmp;
490 g->lacons = v->lacons;
491 v->lacons = NULL;
492 g->nlacons = v->nlacons;
493
494 #ifdef REG_DEBUG
495 if (flags & REG_DUMP)
496 {
497 dump(re, stdout);
498 fflush(stdout);
499 }
500 #endif
501
502 assert(v->err == 0);
503 return freev(v, 0);
504 }
505
506 /*
507 * moresubs - enlarge subRE vector
508 */
509 static void
moresubs(struct vars * v,int wanted)510 moresubs(struct vars *v,
511 int wanted) /* want enough room for this one */
512 {
513 struct subre **p;
514 size_t n;
515
516 assert(wanted > 0 && (size_t) wanted >= v->nsubs);
517 n = (size_t) wanted * 3 / 2 + 1;
518
519 if (v->subs == v->sub10)
520 {
521 p = (struct subre **) MALLOC(n * sizeof(struct subre *));
522 if (p != NULL)
523 memcpy(VS(p), VS(v->subs),
524 v->nsubs * sizeof(struct subre *));
525 }
526 else
527 p = (struct subre **) REALLOC(v->subs, n * sizeof(struct subre *));
528 if (p == NULL)
529 {
530 ERR(REG_ESPACE);
531 return;
532 }
533 v->subs = p;
534 for (p = &v->subs[v->nsubs]; v->nsubs < n; p++, v->nsubs++)
535 *p = NULL;
536 assert(v->nsubs == n);
537 assert((size_t) wanted < v->nsubs);
538 }
539
540 /*
541 * freev - free vars struct's substructures where necessary
542 *
543 * Optionally does error-number setting, and always returns error code
544 * (if any), to make error-handling code terser.
545 */
546 static int
freev(struct vars * v,int err)547 freev(struct vars *v,
548 int err)
549 {
550 if (v->re != NULL)
551 rfree(v->re);
552 if (v->subs != v->sub10)
553 FREE(v->subs);
554 if (v->nfa != NULL)
555 freenfa(v->nfa);
556 if (v->tree != NULL)
557 freesubre(v, v->tree);
558 if (v->treechain != NULL)
559 cleanst(v);
560 if (v->cv != NULL)
561 freecvec(v->cv);
562 if (v->cv2 != NULL)
563 freecvec(v->cv2);
564 if (v->lacons != NULL)
565 freelacons(v->lacons, v->nlacons);
566 ERR(err); /* nop if err==0 */
567
568 return v->err;
569 }
570
571 /*
572 * makesearch - turn an NFA into a search NFA (implicit prepend of .*?)
573 * NFA must have been optimize()d already.
574 */
575 static void
makesearch(struct vars * v,struct nfa * nfa)576 makesearch(struct vars *v,
577 struct nfa *nfa)
578 {
579 struct arc *a;
580 struct arc *b;
581 struct state *pre = nfa->pre;
582 struct state *s;
583 struct state *s2;
584 struct state *slist;
585
586 /* no loops are needed if it's anchored */
587 for (a = pre->outs; a != NULL; a = a->outchain)
588 {
589 assert(a->type == PLAIN);
590 if (a->co != nfa->bos[0] && a->co != nfa->bos[1])
591 break;
592 }
593 if (a != NULL)
594 {
595 /* add implicit .* in front */
596 rainbow(nfa, v->cm, PLAIN, COLORLESS, pre, pre);
597
598 /* and ^* and \A* too -- not always necessary, but harmless */
599 newarc(nfa, PLAIN, nfa->bos[0], pre, pre);
600 newarc(nfa, PLAIN, nfa->bos[1], pre, pre);
601
602 /*
603 * The pattern is still MATCHALL if it was before, but the max match
604 * length is now infinity.
605 */
606 if (nfa->flags & MATCHALL)
607 nfa->maxmatchall = DUPINF;
608 }
609
610 /*
611 * Now here's the subtle part. Because many REs have no lookback
612 * constraints, often knowing when you were in the pre state tells you
613 * little; it's the next state(s) that are informative. But some of them
614 * may have other inarcs, i.e. it may be possible to make actual progress
615 * and then return to one of them. We must de-optimize such cases,
616 * splitting each such state into progress and no-progress states.
617 */
618
619 /* first, make a list of the states reachable from pre and elsewhere */
620 slist = NULL;
621 for (a = pre->outs; a != NULL; a = a->outchain)
622 {
623 s = a->to;
624 for (b = s->ins; b != NULL; b = b->inchain)
625 {
626 if (b->from != pre)
627 break;
628 }
629
630 /*
631 * We want to mark states as being in the list already by having non
632 * NULL tmp fields, but we can't just store the old slist value in tmp
633 * because that doesn't work for the first such state. Instead, the
634 * first list entry gets its own address in tmp.
635 */
636 if (b != NULL && s->tmp == NULL)
637 {
638 s->tmp = (slist != NULL) ? slist : s;
639 slist = s;
640 }
641 }
642
643 /* do the splits */
644 for (s = slist; s != NULL; s = s2)
645 {
646 s2 = newstate(nfa);
647 NOERR();
648 copyouts(nfa, s, s2);
649 NOERR();
650 for (a = s->ins; a != NULL; a = b)
651 {
652 b = a->inchain;
653 if (a->from != pre)
654 {
655 cparc(nfa, a, a->from, s2);
656 freearc(nfa, a);
657 }
658 }
659 s2 = (s->tmp != s) ? s->tmp : NULL;
660 s->tmp = NULL; /* clean up while we're at it */
661 }
662 }
663
664 /*
665 * parse - parse an RE
666 *
667 * This is actually just the top level, which parses a bunch of branches
668 * tied together with '|'. If there's more than one, they appear in the
669 * tree as the children of a '|' subre.
670 */
671 static struct subre *
parse(struct vars * v,int stopper,int type,struct state * init,struct state * final)672 parse(struct vars *v,
673 int stopper, /* EOS or ')' */
674 int type, /* LACON (lookaround subRE) or PLAIN */
675 struct state *init, /* initial state */
676 struct state *final) /* final state */
677 {
678 struct subre *branches; /* top level */
679 struct subre *lastbranch; /* latest branch */
680
681 assert(stopper == ')' || stopper == EOS);
682
683 branches = subre(v, '|', LONGER, init, final);
684 NOERRN();
685 lastbranch = NULL;
686 do
687 { /* a branch */
688 struct subre *branch;
689 struct state *left; /* scaffolding for branch */
690 struct state *right;
691
692 left = newstate(v->nfa);
693 right = newstate(v->nfa);
694 NOERRN();
695 EMPTYARC(init, left);
696 EMPTYARC(right, final);
697 NOERRN();
698 branch = parsebranch(v, stopper, type, left, right, 0);
699 NOERRN();
700 if (lastbranch)
701 lastbranch->sibling = branch;
702 else
703 branches->child = branch;
704 branches->flags |= UP(branches->flags | branch->flags);
705 lastbranch = branch;
706 } while (EAT('|'));
707 assert(SEE(stopper) || SEE(EOS));
708
709 if (!SEE(stopper))
710 {
711 assert(stopper == ')' && SEE(EOS));
712 ERR(REG_EPAREN);
713 }
714
715 /* optimize out simple cases */
716 if (lastbranch == branches->child)
717 { /* only one branch */
718 assert(lastbranch->sibling == NULL);
719 freesrnode(v, branches);
720 branches = lastbranch;
721 }
722 else if (!MESSY(branches->flags))
723 { /* no interesting innards */
724 freesubreandsiblings(v, branches->child);
725 branches->child = NULL;
726 branches->op = '=';
727 }
728
729 return branches;
730 }
731
732 /*
733 * parsebranch - parse one branch of an RE
734 *
735 * This mostly manages concatenation, working closely with parseqatom().
736 * Concatenated things are bundled up as much as possible, with separate
737 * '.' nodes introduced only when necessary due to substructure.
738 */
739 static struct subre *
parsebranch(struct vars * v,int stopper,int type,struct state * left,struct state * right,int partial)740 parsebranch(struct vars *v,
741 int stopper, /* EOS or ')' */
742 int type, /* LACON (lookaround subRE) or PLAIN */
743 struct state *left, /* leftmost state */
744 struct state *right, /* rightmost state */
745 int partial) /* is this only part of a branch? */
746 {
747 struct state *lp; /* left end of current construct */
748 int seencontent; /* is there anything in this branch yet? */
749 struct subre *t;
750
751 lp = left;
752 seencontent = 0;
753 t = subre(v, '=', 0, left, right); /* op '=' is tentative */
754 NOERRN();
755 while (!SEE('|') && !SEE(stopper) && !SEE(EOS))
756 {
757 if (seencontent)
758 { /* implicit concat operator */
759 lp = newstate(v->nfa);
760 NOERRN();
761 moveins(v->nfa, right, lp);
762 }
763 seencontent = 1;
764
765 /* NB, recursion in parseqatom() may swallow rest of branch */
766 t = parseqatom(v, stopper, type, lp, right, t);
767 NOERRN();
768 }
769
770 if (!seencontent)
771 { /* empty branch */
772 if (!partial)
773 NOTE(REG_UUNSPEC);
774 assert(lp == left);
775 EMPTYARC(left, right);
776 }
777
778 return t;
779 }
780
781 /*
782 * parseqatom - parse one quantified atom or constraint of an RE
783 *
784 * The bookkeeping near the end cooperates very closely with parsebranch();
785 * in particular, it contains a recursion that can involve parsing the rest
786 * of the branch, making this function's name somewhat inaccurate.
787 *
788 * Usually, the return value is just "top", but in some cases where we
789 * have parsed the rest of the branch, we may deem "top" redundant and
790 * free it, returning some child subre instead.
791 */
792 static struct subre *
parseqatom(struct vars * v,int stopper,int type,struct state * lp,struct state * rp,struct subre * top)793 parseqatom(struct vars *v,
794 int stopper, /* EOS or ')' */
795 int type, /* LACON (lookaround subRE) or PLAIN */
796 struct state *lp, /* left state to hang it on */
797 struct state *rp, /* right state to hang it on */
798 struct subre *top) /* subtree top */
799 {
800 struct state *s; /* temporaries for new states */
801 struct state *s2;
802
803 #define ARCV(t, val) newarc(v->nfa, t, val, lp, rp)
804 int m,
805 n;
806 struct subre *atom; /* atom's subtree */
807 struct subre *t;
808 int cap; /* capturing parens? */
809 int latype; /* lookaround constraint type */
810 int subno; /* capturing-parens or backref number */
811 int atomtype;
812 int qprefer; /* quantifier short/long preference */
813 int f;
814 struct subre **atomp; /* where the pointer to atom is */
815
816 /* initial bookkeeping */
817 atom = NULL;
818 assert(lp->nouts == 0); /* must string new code */
819 assert(rp->nins == 0); /* between lp and rp */
820 subno = 0; /* just to shut lint up */
821
822 /* an atom or constraint... */
823 atomtype = v->nexttype;
824 switch (atomtype)
825 {
826 /* first, constraints, which end by returning */
827 case '^':
828 ARCV('^', 1);
829 if (v->cflags & REG_NLANCH)
830 ARCV(BEHIND, v->nlcolor);
831 NEXT();
832 return top;
833 break;
834 case '$':
835 ARCV('$', 1);
836 if (v->cflags & REG_NLANCH)
837 ARCV(AHEAD, v->nlcolor);
838 NEXT();
839 return top;
840 break;
841 case SBEGIN:
842 ARCV('^', 1); /* BOL */
843 ARCV('^', 0); /* or BOS */
844 NEXT();
845 return top;
846 break;
847 case SEND:
848 ARCV('$', 1); /* EOL */
849 ARCV('$', 0); /* or EOS */
850 NEXT();
851 return top;
852 break;
853 case '<':
854 wordchrs(v);
855 s = newstate(v->nfa);
856 NOERRN();
857 nonword(v, BEHIND, lp, s);
858 word(v, AHEAD, s, rp);
859 NEXT();
860 return top;
861 break;
862 case '>':
863 wordchrs(v);
864 s = newstate(v->nfa);
865 NOERRN();
866 word(v, BEHIND, lp, s);
867 nonword(v, AHEAD, s, rp);
868 NEXT();
869 return top;
870 break;
871 case WBDRY:
872 wordchrs(v);
873 s = newstate(v->nfa);
874 NOERRN();
875 nonword(v, BEHIND, lp, s);
876 word(v, AHEAD, s, rp);
877 s = newstate(v->nfa);
878 NOERRN();
879 word(v, BEHIND, lp, s);
880 nonword(v, AHEAD, s, rp);
881 NEXT();
882 return top;
883 break;
884 case NWBDRY:
885 wordchrs(v);
886 s = newstate(v->nfa);
887 NOERRN();
888 word(v, BEHIND, lp, s);
889 word(v, AHEAD, s, rp);
890 s = newstate(v->nfa);
891 NOERRN();
892 nonword(v, BEHIND, lp, s);
893 nonword(v, AHEAD, s, rp);
894 NEXT();
895 return top;
896 break;
897 case LACON: /* lookaround constraint */
898 latype = v->nextvalue;
899 NEXT();
900 s = newstate(v->nfa);
901 s2 = newstate(v->nfa);
902 NOERRN();
903 t = parse(v, ')', LACON, s, s2);
904 freesubre(v, t); /* internal structure irrelevant */
905 NOERRN();
906 assert(SEE(')'));
907 NEXT();
908 processlacon(v, s, s2, latype, lp, rp);
909 return top;
910 break;
911 /* then errors, to get them out of the way */
912 case '*':
913 case '+':
914 case '?':
915 case '{':
916 ERR(REG_BADRPT);
917 return top;
918 break;
919 default:
920 ERR(REG_ASSERT);
921 return top;
922 break;
923 /* then plain characters, and minor variants on that theme */
924 case ')': /* unbalanced paren */
925 if ((v->cflags & REG_ADVANCED) != REG_EXTENDED)
926 {
927 ERR(REG_EPAREN);
928 return top;
929 }
930 /* legal in EREs due to specification botch */
931 NOTE(REG_UPBOTCH);
932 /* fall through into case PLAIN */
933 /* FALLTHROUGH */
934 case PLAIN:
935 onechr(v, v->nextvalue, lp, rp);
936 okcolors(v->nfa, v->cm);
937 NOERRN();
938 NEXT();
939 break;
940 case '[':
941 if (v->nextvalue == 1)
942 bracket(v, lp, rp);
943 else
944 cbracket(v, lp, rp);
945 assert(SEE(']') || ISERR());
946 NEXT();
947 break;
948 case CCLASSS:
949 charclass(v, (enum char_classes) v->nextvalue, lp, rp);
950 okcolors(v->nfa, v->cm);
951 NEXT();
952 break;
953 case CCLASSC:
954 charclasscomplement(v, (enum char_classes) v->nextvalue, lp, rp);
955 /* charclasscomplement() did okcolors() internally */
956 NEXT();
957 break;
958 case '.':
959 rainbow(v->nfa, v->cm, PLAIN,
960 (v->cflags & REG_NLSTOP) ? v->nlcolor : COLORLESS,
961 lp, rp);
962 NEXT();
963 break;
964 /* and finally the ugly stuff */
965 case '(': /* value flags as capturing or non */
966 cap = (type == LACON) ? 0 : v->nextvalue;
967 if (cap)
968 {
969 v->nsubexp++;
970 subno = v->nsubexp;
971 if ((size_t) subno >= v->nsubs)
972 moresubs(v, subno);
973 }
974 else
975 atomtype = PLAIN; /* something that's not '(' */
976 NEXT();
977
978 /*
979 * Make separate endpoint states to keep this sub-NFA distinct
980 * from what surrounds it. We need to be sure that when we
981 * duplicate the sub-NFA for a backref, we get the right
982 * states/arcs and no others. In particular, letting a backref
983 * duplicate the sub-NFA from lp to rp would be quite wrong,
984 * because we may add quantification superstructure around this
985 * atom below. (Perhaps we could skip the extra states for
986 * non-capturing parens, but it seems not worth the trouble.)
987 */
988 s = newstate(v->nfa);
989 s2 = newstate(v->nfa);
990 NOERRN();
991 /* We may not need these arcs, but keep things connected for now */
992 EMPTYARC(lp, s);
993 EMPTYARC(s2, rp);
994 NOERRN();
995 atom = parse(v, ')', type, s, s2);
996 assert(SEE(')') || ISERR());
997 NEXT();
998 NOERRN();
999 if (cap)
1000 {
1001 if (atom->capno == 0)
1002 {
1003 /* normal case: just mark the atom as capturing */
1004 atom->flags |= CAP;
1005 atom->capno = subno;
1006 }
1007 else
1008 {
1009 /* generate no-op wrapper node to handle "((x))" */
1010 t = subre(v, '(', atom->flags | CAP, s, s2);
1011 NOERRN();
1012 t->capno = subno;
1013 t->child = atom;
1014 atom = t;
1015 }
1016 assert(v->subs[subno] == NULL);
1017 v->subs[subno] = atom;
1018 }
1019 /* postpone everything else pending possible {0} */
1020 break;
1021 case BACKREF: /* the Feature From The Black Lagoon */
1022 INSIST(type != LACON, REG_ESUBREG);
1023 INSIST(v->nextvalue < v->nsubs, REG_ESUBREG);
1024 INSIST(v->subs[v->nextvalue] != NULL, REG_ESUBREG);
1025 NOERRN();
1026 assert(v->nextvalue > 0);
1027 atom = subre(v, 'b', BACKR, lp, rp);
1028 NOERRN();
1029 subno = v->nextvalue;
1030 atom->backno = subno;
1031 EMPTYARC(lp, rp); /* temporarily, so there's something */
1032 NEXT();
1033 break;
1034 }
1035
1036 /* ...and an atom may be followed by a quantifier */
1037 switch (v->nexttype)
1038 {
1039 case '*':
1040 m = 0;
1041 n = DUPINF;
1042 qprefer = (v->nextvalue) ? LONGER : SHORTER;
1043 NEXT();
1044 break;
1045 case '+':
1046 m = 1;
1047 n = DUPINF;
1048 qprefer = (v->nextvalue) ? LONGER : SHORTER;
1049 NEXT();
1050 break;
1051 case '?':
1052 m = 0;
1053 n = 1;
1054 qprefer = (v->nextvalue) ? LONGER : SHORTER;
1055 NEXT();
1056 break;
1057 case '{':
1058 NEXT();
1059 m = scannum(v);
1060 if (EAT(','))
1061 {
1062 if (SEE(DIGIT))
1063 n = scannum(v);
1064 else
1065 n = DUPINF;
1066 if (m > n)
1067 {
1068 ERR(REG_BADBR);
1069 return top;
1070 }
1071 /* {m,n} exercises preference, even if it's {m,m} */
1072 qprefer = (v->nextvalue) ? LONGER : SHORTER;
1073 }
1074 else
1075 {
1076 n = m;
1077 /* {m} passes operand's preference through */
1078 qprefer = 0;
1079 }
1080 if (!SEE('}'))
1081 { /* catches errors too */
1082 ERR(REG_BADBR);
1083 return top;
1084 }
1085 NEXT();
1086 break;
1087 default: /* no quantifier */
1088 m = n = 1;
1089 qprefer = 0;
1090 break;
1091 }
1092
1093 /* annoying special case: {0} or {0,0} cancels everything */
1094 if (m == 0 && n == 0)
1095 {
1096 /*
1097 * If we had capturing subexpression(s) within the atom, we don't want
1098 * to destroy them, because it's legal (if useless) to back-ref them
1099 * later. Hence, just unlink the atom from lp/rp and then ignore it.
1100 */
1101 if (atom != NULL && (atom->flags & CAP))
1102 {
1103 delsub(v->nfa, lp, atom->begin);
1104 delsub(v->nfa, atom->end, rp);
1105 }
1106 else
1107 {
1108 /* Otherwise, we can clean up any subre infrastructure we made */
1109 if (atom != NULL)
1110 freesubre(v, atom);
1111 delsub(v->nfa, lp, rp);
1112 }
1113 EMPTYARC(lp, rp);
1114 return top;
1115 }
1116
1117 /* if not a messy case, avoid hard part */
1118 assert(!MESSY(top->flags));
1119 f = top->flags | qprefer | ((atom != NULL) ? atom->flags : 0);
1120 if (atomtype != '(' && atomtype != BACKREF && !MESSY(UP(f)))
1121 {
1122 if (!(m == 1 && n == 1))
1123 repeat(v, lp, rp, m, n);
1124 if (atom != NULL)
1125 freesubre(v, atom);
1126 top->flags = f;
1127 return top;
1128 }
1129
1130 /*
1131 * hard part: something messy
1132 *
1133 * That is, capturing parens, back reference, short/long clash, or an atom
1134 * with substructure containing one of those.
1135 */
1136
1137 /* now we'll need a subre for the contents even if they're boring */
1138 if (atom == NULL)
1139 {
1140 atom = subre(v, '=', 0, lp, rp);
1141 NOERRN();
1142 }
1143
1144 /*
1145 * For what follows, we need the atom to have its own begin/end states
1146 * that are distinct from lp/rp, so that we can wrap iteration structure
1147 * around it. The parenthesized-atom case above already made suitable
1148 * states (and we don't want to modify a capturing subre, since it's
1149 * already recorded in v->subs[]). Otherwise, we need more states.
1150 */
1151 if (atom->begin == lp || atom->end == rp)
1152 {
1153 s = newstate(v->nfa);
1154 s2 = newstate(v->nfa);
1155 NOERRN();
1156 moveouts(v->nfa, lp, s);
1157 moveins(v->nfa, rp, s2);
1158 atom->begin = s;
1159 atom->end = s2;
1160 }
1161 else
1162 {
1163 /* The atom's OK, but we must temporarily disconnect it from lp/rp */
1164 /* (this removes the EMPTY arcs we made above) */
1165 delsub(v->nfa, lp, atom->begin);
1166 delsub(v->nfa, atom->end, rp);
1167 }
1168
1169 /*----------
1170 * Prepare a general-purpose state skeleton.
1171 *
1172 * In the no-backrefs case, we want this:
1173 *
1174 * [lp] ---> [s] ---prefix---> ---atom---> ---rest---> [rp]
1175 *
1176 * where prefix is some repetitions of atom, and "rest" is the remainder
1177 * of the branch. In the general case we need:
1178 *
1179 * [lp] ---> [s] ---iterator---> [s2] ---rest---> [rp]
1180 *
1181 * where the iterator wraps around the atom.
1182 *
1183 * We make the s state here for both cases; s2 is made below if needed
1184 *----------
1185 */
1186 s = newstate(v->nfa); /* set up starting state */
1187 NOERRN();
1188 EMPTYARC(lp, s);
1189 NOERRN();
1190
1191 /* break remaining subRE into x{...} and what follows */
1192 t = subre(v, '.', COMBINE(qprefer, atom->flags), lp, rp);
1193 NOERRN();
1194 t->child = atom;
1195 atomp = &t->child;
1196
1197 /*
1198 * Here we should recurse to fill t->child->sibling ... but we must
1199 * postpone that to the end. One reason is that t->child may be replaced
1200 * below, and we don't want to worry about its sibling link.
1201 */
1202
1203 /*
1204 * Convert top node to a concatenation of the prefix (top->child, covering
1205 * whatever we parsed previously) and remaining (t). Note that the prefix
1206 * could be empty, in which case this concatenation node is unnecessary.
1207 * To keep things simple, we operate in a general way for now, and get rid
1208 * of unnecessary subres below.
1209 */
1210 assert(top->op == '=' && top->child == NULL);
1211 top->child = subre(v, '=', top->flags, top->begin, lp);
1212 NOERRN();
1213 top->op = '.';
1214 top->child->sibling = t;
1215 /* top->flags will get updated later */
1216
1217 /* if it's a backref, now is the time to replicate the subNFA */
1218 if (atomtype == BACKREF)
1219 {
1220 assert(atom->begin->nouts == 1); /* just the EMPTY */
1221 delsub(v->nfa, atom->begin, atom->end);
1222 assert(v->subs[subno] != NULL);
1223
1224 /*
1225 * And here's why the recursion got postponed: it must wait until the
1226 * skeleton is filled in, because it may hit a backref that wants to
1227 * copy the filled-in skeleton.
1228 */
1229 dupnfa(v->nfa, v->subs[subno]->begin, v->subs[subno]->end,
1230 atom->begin, atom->end);
1231 NOERRN();
1232
1233 /* The backref node's NFA should not enforce any constraints */
1234 removeconstraints(v->nfa, atom->begin, atom->end);
1235 NOERRN();
1236 }
1237
1238 /*
1239 * It's quantifier time. If the atom is just a backref, we'll let it deal
1240 * with quantifiers internally.
1241 */
1242 if (atomtype == BACKREF)
1243 {
1244 /* special case: backrefs have internal quantifiers */
1245 EMPTYARC(s, atom->begin); /* empty prefix */
1246 /* just stuff everything into atom */
1247 repeat(v, atom->begin, atom->end, m, n);
1248 atom->min = (short) m;
1249 atom->max = (short) n;
1250 atom->flags |= COMBINE(qprefer, atom->flags);
1251 /* rest of branch can be strung starting from atom->end */
1252 s2 = atom->end;
1253 }
1254 else if (m == 1 && n == 1 &&
1255 (qprefer == 0 ||
1256 (atom->flags & (LONGER | SHORTER | MIXED)) == 0 ||
1257 qprefer == (atom->flags & (LONGER | SHORTER | MIXED))))
1258 {
1259 /* no/vacuous quantifier: done */
1260 EMPTYARC(s, atom->begin); /* empty prefix */
1261 /* rest of branch can be strung starting from atom->end */
1262 s2 = atom->end;
1263 }
1264 else if (!(atom->flags & (CAP | BACKR)))
1265 {
1266 /*
1267 * If there's no captures nor backrefs in the atom being repeated, we
1268 * don't really care where the submatches of the iteration are, so we
1269 * don't need an iteration node. Make a plain DFA node instead.
1270 */
1271 EMPTYARC(s, atom->begin); /* empty prefix */
1272 repeat(v, atom->begin, atom->end, m, n);
1273 f = COMBINE(qprefer, atom->flags);
1274 t = subre(v, '=', f, atom->begin, atom->end);
1275 NOERRN();
1276 freesubre(v, atom);
1277 *atomp = t;
1278 /* rest of branch can be strung starting from t->end */
1279 s2 = t->end;
1280 }
1281 else if (m > 0 && !(atom->flags & BACKR))
1282 {
1283 /*
1284 * If there's no backrefs involved, we can turn x{m,n} into
1285 * x{m-1,n-1}x, with capturing parens in only the second x. This is
1286 * valid because we only care about capturing matches from the final
1287 * iteration of the quantifier. It's a win because we can implement
1288 * the backref-free left side as a plain DFA node, since we don't
1289 * really care where its submatches are.
1290 */
1291 dupnfa(v->nfa, atom->begin, atom->end, s, atom->begin);
1292 assert(m >= 1 && m != DUPINF && n >= 1);
1293 repeat(v, s, atom->begin, m - 1, (n == DUPINF) ? n : n - 1);
1294 f = COMBINE(qprefer, atom->flags);
1295 t = subre(v, '.', f, s, atom->end); /* prefix and atom */
1296 NOERRN();
1297 t->child = subre(v, '=', PREF(f), s, atom->begin);
1298 NOERRN();
1299 t->child->sibling = atom;
1300 *atomp = t;
1301 /* rest of branch can be strung starting from atom->end */
1302 s2 = atom->end;
1303 }
1304 else
1305 {
1306 /* general case: need an iteration node */
1307 s2 = newstate(v->nfa);
1308 NOERRN();
1309 moveouts(v->nfa, atom->end, s2);
1310 NOERRN();
1311 dupnfa(v->nfa, atom->begin, atom->end, s, s2);
1312 repeat(v, s, s2, m, n);
1313 f = COMBINE(qprefer, atom->flags);
1314 t = subre(v, '*', f, s, s2);
1315 NOERRN();
1316 t->min = (short) m;
1317 t->max = (short) n;
1318 t->child = atom;
1319 *atomp = t;
1320 /* rest of branch is to be strung from iteration's end state */
1321 }
1322
1323 /* and finally, look after that postponed recursion */
1324 t = top->child->sibling;
1325 if (!(SEE('|') || SEE(stopper) || SEE(EOS)))
1326 {
1327 /* parse all the rest of the branch, and insert in t->child->sibling */
1328 t->child->sibling = parsebranch(v, stopper, type, s2, rp, 1);
1329 NOERRN();
1330 assert(SEE('|') || SEE(stopper) || SEE(EOS));
1331
1332 /* here's the promised update of the flags */
1333 t->flags |= COMBINE(t->flags, t->child->sibling->flags);
1334 top->flags |= COMBINE(top->flags, t->flags);
1335
1336 /* neither t nor top could be directly marked for capture as yet */
1337 assert(t->capno == 0);
1338 assert(top->capno == 0);
1339
1340 /*
1341 * At this point both top and t are concatenation (op == '.') subres,
1342 * and we have top->child = prefix of branch, top->child->sibling = t,
1343 * t->child = messy atom (with quantification superstructure if
1344 * needed), t->child->sibling = rest of branch.
1345 *
1346 * If the messy atom was the first thing in the branch, then
1347 * top->child is vacuous and we can get rid of one level of
1348 * concatenation.
1349 */
1350 assert(top->child->op == '=');
1351 if (top->child->begin == top->child->end)
1352 {
1353 assert(!MESSY(top->child->flags));
1354 freesubre(v, top->child);
1355 top->child = t->child;
1356 freesrnode(v, t);
1357 }
1358
1359 /*
1360 * Otherwise, it's possible that t->child is not messy in itself, but
1361 * we considered it messy because its greediness conflicts with what
1362 * preceded it. Then it could be that the combination of t->child and
1363 * the rest of the branch is also not messy, in which case we can get
1364 * rid of the child concatenation by merging t->child and the rest of
1365 * the branch into one plain DFA node.
1366 */
1367 else if (t->child->op == '=' &&
1368 t->child->sibling->op == '=' &&
1369 !MESSY(UP(t->child->flags | t->child->sibling->flags)))
1370 {
1371 t->op = '=';
1372 t->flags = COMBINE(t->child->flags, t->child->sibling->flags);
1373 freesubreandsiblings(v, t->child);
1374 t->child = NULL;
1375 }
1376 }
1377 else
1378 {
1379 /*
1380 * There's nothing left in the branch, so we don't need the second
1381 * concatenation node 't'. Just link s2 straight to rp.
1382 */
1383 EMPTYARC(s2, rp);
1384 top->child->sibling = t->child;
1385 top->flags |= COMBINE(top->flags, top->child->sibling->flags);
1386 freesrnode(v, t);
1387
1388 /*
1389 * Again, it could be that top->child is vacuous (if the messy atom
1390 * was in fact the only thing in the branch). In that case we need no
1391 * concatenation at all; just replace top with top->child->sibling.
1392 */
1393 assert(top->child->op == '=');
1394 if (top->child->begin == top->child->end)
1395 {
1396 assert(!MESSY(top->child->flags));
1397 t = top->child->sibling;
1398 top->child->sibling = NULL;
1399 freesubre(v, top);
1400 top = t;
1401 }
1402 }
1403
1404 return top;
1405 }
1406
1407 /*
1408 * nonword - generate arcs for non-word-character ahead or behind
1409 */
1410 static void
nonword(struct vars * v,int dir,struct state * lp,struct state * rp)1411 nonword(struct vars *v,
1412 int dir, /* AHEAD or BEHIND */
1413 struct state *lp,
1414 struct state *rp)
1415 {
1416 int anchor = (dir == AHEAD) ? '$' : '^';
1417
1418 assert(dir == AHEAD || dir == BEHIND);
1419 newarc(v->nfa, anchor, 1, lp, rp);
1420 newarc(v->nfa, anchor, 0, lp, rp);
1421 colorcomplement(v->nfa, v->cm, dir, v->wordchrs, lp, rp);
1422 /* (no need for special attention to \n) */
1423 }
1424
1425 /*
1426 * word - generate arcs for word character ahead or behind
1427 */
1428 static void
word(struct vars * v,int dir,struct state * lp,struct state * rp)1429 word(struct vars *v,
1430 int dir, /* AHEAD or BEHIND */
1431 struct state *lp,
1432 struct state *rp)
1433 {
1434 assert(dir == AHEAD || dir == BEHIND);
1435 cloneouts(v->nfa, v->wordchrs, lp, rp, dir);
1436 /* (no need for special attention to \n) */
1437 }
1438
1439 /*
1440 * charclass - generate arcs for a character class
1441 *
1442 * This is used for both atoms (\w and sibling escapes) and for elements
1443 * of bracket expressions. The caller is responsible for calling okcolors()
1444 * at the end of processing the atom or bracket.
1445 */
1446 static void
charclass(struct vars * v,enum char_classes cls,struct state * lp,struct state * rp)1447 charclass(struct vars *v,
1448 enum char_classes cls,
1449 struct state *lp,
1450 struct state *rp)
1451 {
1452 struct cvec *cv;
1453
1454 /* obtain possibly-cached cvec for char class */
1455 NOTE(REG_ULOCALE);
1456 cv = cclasscvec(v, cls, (v->cflags & REG_ICASE));
1457 NOERR();
1458
1459 /* build the arcs; this may cause color splitting */
1460 subcolorcvec(v, cv, lp, rp);
1461 }
1462
1463 /*
1464 * charclasscomplement - generate arcs for a complemented character class
1465 *
1466 * This is used for both atoms (\W and sibling escapes) and for elements
1467 * of bracket expressions. In bracket expressions, it is the caller's
1468 * responsibility that there not be any open subcolors when this is called.
1469 */
1470 static void
charclasscomplement(struct vars * v,enum char_classes cls,struct state * lp,struct state * rp)1471 charclasscomplement(struct vars *v,
1472 enum char_classes cls,
1473 struct state *lp,
1474 struct state *rp)
1475 {
1476 struct state *cstate;
1477 struct cvec *cv;
1478
1479 /* make dummy state to hang temporary arcs on */
1480 cstate = newstate(v->nfa);
1481 NOERR();
1482
1483 /* obtain possibly-cached cvec for char class */
1484 NOTE(REG_ULOCALE);
1485 cv = cclasscvec(v, cls, (v->cflags & REG_ICASE));
1486 NOERR();
1487
1488 /* build arcs for char class; this may cause color splitting */
1489 subcolorcvec(v, cv, cstate, cstate);
1490 NOERR();
1491
1492 /* clean up any subcolors in the arc set */
1493 okcolors(v->nfa, v->cm);
1494 NOERR();
1495
1496 /* now build output arcs for the complement of the char class */
1497 colorcomplement(v->nfa, v->cm, PLAIN, cstate, lp, rp);
1498 NOERR();
1499
1500 /* clean up dummy state */
1501 dropstate(v->nfa, cstate);
1502 }
1503
1504 /*
1505 * scannum - scan a number
1506 */
1507 static int /* value, <= DUPMAX */
scannum(struct vars * v)1508 scannum(struct vars *v)
1509 {
1510 int n = 0;
1511
1512 while (SEE(DIGIT) && n < DUPMAX)
1513 {
1514 n = n * 10 + v->nextvalue;
1515 NEXT();
1516 }
1517 if (SEE(DIGIT) || n > DUPMAX)
1518 {
1519 ERR(REG_BADBR);
1520 return 0;
1521 }
1522 return n;
1523 }
1524
1525 /*
1526 * repeat - replicate subNFA for quantifiers
1527 *
1528 * The sub-NFA strung from lp to rp is modified to represent m to n
1529 * repetitions of its initial contents.
1530 *
1531 * The duplication sequences used here are chosen carefully so that any
1532 * pointers starting out pointing into the subexpression end up pointing into
1533 * the last occurrence. (Note that it may not be strung between the same
1534 * left and right end states, however!) This used to be important for the
1535 * subRE tree, although the important bits are now handled by the in-line
1536 * code in parse(), and when this is called, it doesn't matter any more.
1537 */
1538 static void
repeat(struct vars * v,struct state * lp,struct state * rp,int m,int n)1539 repeat(struct vars *v,
1540 struct state *lp,
1541 struct state *rp,
1542 int m,
1543 int n)
1544 {
1545 #define SOME 2
1546 #define INF 3
1547 #define PAIR(x, y) ((x)*4 + (y))
1548 #define REDUCE(x) ( ((x) == DUPINF) ? INF : (((x) > 1) ? SOME : (x)) )
1549 const int rm = REDUCE(m);
1550 const int rn = REDUCE(n);
1551 struct state *s;
1552 struct state *s2;
1553
1554 switch (PAIR(rm, rn))
1555 {
1556 case PAIR(0, 0): /* empty string */
1557 delsub(v->nfa, lp, rp);
1558 EMPTYARC(lp, rp);
1559 break;
1560 case PAIR(0, 1): /* do as x| */
1561 EMPTYARC(lp, rp);
1562 break;
1563 case PAIR(0, SOME): /* do as x{1,n}| */
1564 repeat(v, lp, rp, 1, n);
1565 NOERR();
1566 EMPTYARC(lp, rp);
1567 break;
1568 case PAIR(0, INF): /* loop x around */
1569 s = newstate(v->nfa);
1570 NOERR();
1571 moveouts(v->nfa, lp, s);
1572 moveins(v->nfa, rp, s);
1573 EMPTYARC(lp, s);
1574 EMPTYARC(s, rp);
1575 break;
1576 case PAIR(1, 1): /* no action required */
1577 break;
1578 case PAIR(1, SOME): /* do as x{0,n-1}x = (x{1,n-1}|)x */
1579 s = newstate(v->nfa);
1580 NOERR();
1581 moveouts(v->nfa, lp, s);
1582 dupnfa(v->nfa, s, rp, lp, s);
1583 NOERR();
1584 repeat(v, lp, s, 1, n - 1);
1585 NOERR();
1586 EMPTYARC(lp, s);
1587 break;
1588 case PAIR(1, INF): /* add loopback arc */
1589 s = newstate(v->nfa);
1590 s2 = newstate(v->nfa);
1591 NOERR();
1592 moveouts(v->nfa, lp, s);
1593 moveins(v->nfa, rp, s2);
1594 EMPTYARC(lp, s);
1595 EMPTYARC(s2, rp);
1596 EMPTYARC(s2, s);
1597 break;
1598 case PAIR(SOME, SOME): /* do as x{m-1,n-1}x */
1599 s = newstate(v->nfa);
1600 NOERR();
1601 moveouts(v->nfa, lp, s);
1602 dupnfa(v->nfa, s, rp, lp, s);
1603 NOERR();
1604 repeat(v, lp, s, m - 1, n - 1);
1605 break;
1606 case PAIR(SOME, INF): /* do as x{m-1,}x */
1607 s = newstate(v->nfa);
1608 NOERR();
1609 moveouts(v->nfa, lp, s);
1610 dupnfa(v->nfa, s, rp, lp, s);
1611 NOERR();
1612 repeat(v, lp, s, m - 1, n);
1613 break;
1614 default:
1615 ERR(REG_ASSERT);
1616 break;
1617 }
1618 }
1619
1620 /*
1621 * bracket - handle non-complemented bracket expression
1622 *
1623 * Also called from cbracket for complemented bracket expressions.
1624 */
1625 static void
bracket(struct vars * v,struct state * lp,struct state * rp)1626 bracket(struct vars *v,
1627 struct state *lp,
1628 struct state *rp)
1629 {
1630 /*
1631 * We can't process complemented char classes (e.g. \W) immediately while
1632 * scanning the bracket expression, else color bookkeeping gets confused.
1633 * Instead, remember whether we saw any in have_cclassc[], and process
1634 * them at the end.
1635 */
1636 bool have_cclassc[NUM_CCLASSES];
1637 bool any_cclassc;
1638 int i;
1639
1640 memset(have_cclassc, false, sizeof(have_cclassc));
1641
1642 assert(SEE('['));
1643 NEXT();
1644 while (!SEE(']') && !SEE(EOS))
1645 brackpart(v, lp, rp, have_cclassc);
1646 assert(SEE(']') || ISERR());
1647
1648 /* close up open subcolors from the positive bracket elements */
1649 okcolors(v->nfa, v->cm);
1650 NOERR();
1651
1652 /* now handle any complemented elements */
1653 any_cclassc = false;
1654 for (i = 0; i < NUM_CCLASSES; i++)
1655 {
1656 if (have_cclassc[i])
1657 {
1658 charclasscomplement(v, (enum char_classes) i, lp, rp);
1659 NOERR();
1660 any_cclassc = true;
1661 }
1662 }
1663
1664 /*
1665 * If we had any complemented elements, see if we can optimize the bracket
1666 * into a rainbow. Since a complemented element is the only way a WHITE
1667 * arc could get into the result, there's no point in checking otherwise.
1668 */
1669 if (any_cclassc)
1670 optimizebracket(v, lp, rp);
1671 }
1672
1673 /*
1674 * cbracket - handle complemented bracket expression
1675 *
1676 * We do it by calling bracket() with dummy endpoints, and then complementing
1677 * the result. The alternative would be to invoke rainbow(), and then delete
1678 * arcs as the b.e. is seen... but that gets messy, and is really quite
1679 * infeasible now that rainbow() just puts out one RAINBOW arc.
1680 */
1681 static void
cbracket(struct vars * v,struct state * lp,struct state * rp)1682 cbracket(struct vars *v,
1683 struct state *lp,
1684 struct state *rp)
1685 {
1686 struct state *left = newstate(v->nfa);
1687 struct state *right = newstate(v->nfa);
1688
1689 NOERR();
1690 bracket(v, left, right);
1691
1692 /* in NLSTOP mode, ensure newline is not part of the result set */
1693 if (v->cflags & REG_NLSTOP)
1694 newarc(v->nfa, PLAIN, v->nlcolor, left, right);
1695 NOERR();
1696
1697 assert(lp->nouts == 0); /* all outarcs will be ours */
1698
1699 /*
1700 * Easy part of complementing, and all there is to do since the MCCE code
1701 * was removed. Note that the result of colorcomplement() cannot be a
1702 * rainbow, since we don't allow empty brackets; so there's no point in
1703 * calling optimizebracket() again.
1704 */
1705 colorcomplement(v->nfa, v->cm, PLAIN, left, lp, rp);
1706 NOERR();
1707 dropstate(v->nfa, left);
1708 assert(right->nins == 0);
1709 freestate(v->nfa, right);
1710 }
1711
1712 /*
1713 * brackpart - handle one item (or range) within a bracket expression
1714 */
1715 static void
brackpart(struct vars * v,struct state * lp,struct state * rp,bool * have_cclassc)1716 brackpart(struct vars *v,
1717 struct state *lp,
1718 struct state *rp,
1719 bool *have_cclassc)
1720 {
1721 chr startc;
1722 chr endc;
1723 struct cvec *cv;
1724 enum char_classes cls;
1725 const chr *startp;
1726 const chr *endp;
1727
1728 /* parse something, get rid of special cases, take shortcuts */
1729 switch (v->nexttype)
1730 {
1731 case RANGE: /* a-b-c or other botch */
1732 ERR(REG_ERANGE);
1733 return;
1734 break;
1735 case PLAIN:
1736 startc = v->nextvalue;
1737 NEXT();
1738 /* shortcut for ordinary chr (not range) */
1739 if (!SEE(RANGE))
1740 {
1741 onechr(v, startc, lp, rp);
1742 return;
1743 }
1744 NOERR();
1745 break;
1746 case COLLEL:
1747 startp = v->now;
1748 endp = scanplain(v);
1749 INSIST(startp < endp, REG_ECOLLATE);
1750 NOERR();
1751 startc = element(v, startp, endp);
1752 NOERR();
1753 break;
1754 case ECLASS:
1755 startp = v->now;
1756 endp = scanplain(v);
1757 INSIST(startp < endp, REG_ECOLLATE);
1758 NOERR();
1759 startc = element(v, startp, endp);
1760 NOERR();
1761 cv = eclass(v, startc, (v->cflags & REG_ICASE));
1762 NOERR();
1763 subcolorcvec(v, cv, lp, rp);
1764 return;
1765 break;
1766 case CCLASS:
1767 startp = v->now;
1768 endp = scanplain(v);
1769 INSIST(startp < endp, REG_ECTYPE);
1770 NOERR();
1771 cls = lookupcclass(v, startp, endp);
1772 NOERR();
1773 charclass(v, cls, lp, rp);
1774 return;
1775 break;
1776 case CCLASSS:
1777 charclass(v, (enum char_classes) v->nextvalue, lp, rp);
1778 NEXT();
1779 return;
1780 break;
1781 case CCLASSC:
1782 /* we cannot call charclasscomplement() immediately */
1783 have_cclassc[v->nextvalue] = true;
1784 NEXT();
1785 return;
1786 break;
1787 default:
1788 ERR(REG_ASSERT);
1789 return;
1790 break;
1791 }
1792
1793 if (SEE(RANGE))
1794 {
1795 NEXT();
1796 switch (v->nexttype)
1797 {
1798 case PLAIN:
1799 case RANGE:
1800 endc = v->nextvalue;
1801 NEXT();
1802 NOERR();
1803 break;
1804 case COLLEL:
1805 startp = v->now;
1806 endp = scanplain(v);
1807 INSIST(startp < endp, REG_ECOLLATE);
1808 NOERR();
1809 endc = element(v, startp, endp);
1810 NOERR();
1811 break;
1812 default:
1813 ERR(REG_ERANGE);
1814 return;
1815 break;
1816 }
1817 }
1818 else
1819 endc = startc;
1820
1821 /*
1822 * Ranges are unportable. Actually, standard C does guarantee that digits
1823 * are contiguous, but making that an exception is just too complicated.
1824 */
1825 if (startc != endc)
1826 NOTE(REG_UUNPORT);
1827 cv = range(v, startc, endc, (v->cflags & REG_ICASE));
1828 NOERR();
1829 subcolorcvec(v, cv, lp, rp);
1830 }
1831
1832 /*
1833 * scanplain - scan PLAIN contents of [. etc.
1834 *
1835 * Certain bits of trickery in regc_lex.c know that this code does not try
1836 * to look past the final bracket of the [. etc.
1837 */
1838 static const chr * /* just after end of sequence */
scanplain(struct vars * v)1839 scanplain(struct vars *v)
1840 {
1841 const chr *endp;
1842
1843 assert(SEE(COLLEL) || SEE(ECLASS) || SEE(CCLASS));
1844 NEXT();
1845
1846 endp = v->now;
1847 while (SEE(PLAIN))
1848 {
1849 endp = v->now;
1850 NEXT();
1851 }
1852
1853 assert(SEE(END) || ISERR());
1854 NEXT();
1855
1856 return endp;
1857 }
1858
1859 /*
1860 * onechr - fill in arcs for a plain character, and possible case complements
1861 * This is mostly a shortcut for efficient handling of the common case.
1862 */
1863 static void
onechr(struct vars * v,chr c,struct state * lp,struct state * rp)1864 onechr(struct vars *v,
1865 chr c,
1866 struct state *lp,
1867 struct state *rp)
1868 {
1869 if (!(v->cflags & REG_ICASE))
1870 {
1871 color lastsubcolor = COLORLESS;
1872
1873 subcoloronechr(v, c, lp, rp, &lastsubcolor);
1874 return;
1875 }
1876
1877 /* rats, need general case anyway... */
1878 subcolorcvec(v, allcases(v, c), lp, rp);
1879 }
1880
1881 /*
1882 * optimizebracket - see if bracket expression can be converted to RAINBOW
1883 *
1884 * Cases such as "[\s\S]" can produce a set of arcs of all colors, which we
1885 * can replace by a single RAINBOW arc for efficiency. (This might seem
1886 * like a silly way to write ".", but it's seemingly a common locution in
1887 * some other flavors of regex, so take the trouble to support it well.)
1888 */
1889 static void
optimizebracket(struct vars * v,struct state * lp,struct state * rp)1890 optimizebracket(struct vars *v,
1891 struct state *lp,
1892 struct state *rp)
1893 {
1894 struct colordesc *cd;
1895 struct colordesc *end = CDEND(v->cm);
1896 struct arc *a;
1897 bool israinbow;
1898
1899 /*
1900 * Scan lp's out-arcs and transiently mark the mentioned colors. We
1901 * expect that all of lp's out-arcs are plain, non-RAINBOW arcs to rp.
1902 * (Note: there shouldn't be any pseudocolors yet, but check anyway.)
1903 */
1904 for (a = lp->outs; a != NULL; a = a->outchain)
1905 {
1906 assert(a->type == PLAIN);
1907 assert(a->co >= 0); /* i.e. not RAINBOW */
1908 assert(a->to == rp);
1909 cd = &v->cm->cd[a->co];
1910 assert(!UNUSEDCOLOR(cd) && !(cd->flags & PSEUDO));
1911 cd->flags |= COLMARK;
1912 }
1913
1914 /* Scan colors, clear transient marks, check for unmarked live colors */
1915 israinbow = true;
1916 for (cd = v->cm->cd; cd < end; cd++)
1917 {
1918 if (cd->flags & COLMARK)
1919 cd->flags &= ~COLMARK;
1920 else if (!UNUSEDCOLOR(cd) && !(cd->flags & PSEUDO))
1921 israinbow = false;
1922 }
1923
1924 /* Can't do anything if not all colors have arcs */
1925 if (!israinbow)
1926 return;
1927
1928 /* OK, drop existing arcs and replace with a rainbow */
1929 while ((a = lp->outs) != NULL)
1930 freearc(v->nfa, a);
1931 newarc(v->nfa, PLAIN, RAINBOW, lp, rp);
1932 }
1933
1934 /*
1935 * wordchrs - set up word-chr list for word-boundary stuff, if needed
1936 *
1937 * The list is kept as a bunch of circular arcs on an otherwise-unused state.
1938 *
1939 * Note that this must not be called while we have any open subcolors,
1940 * else construction of the list would confuse color bookkeeping.
1941 * Hence, we can't currently apply a similar optimization in
1942 * charclass[complement](), as those need to be usable within bracket
1943 * expressions.
1944 */
1945 static void
wordchrs(struct vars * v)1946 wordchrs(struct vars *v)
1947 {
1948 struct state *cstate;
1949 struct cvec *cv;
1950
1951 if (v->wordchrs != NULL)
1952 return; /* done already */
1953
1954 /* make dummy state to hang the cache arcs on */
1955 cstate = newstate(v->nfa);
1956 NOERR();
1957
1958 /* obtain possibly-cached cvec for \w characters */
1959 NOTE(REG_ULOCALE);
1960 cv = cclasscvec(v, CC_WORD, (v->cflags & REG_ICASE));
1961 NOERR();
1962
1963 /* build the arcs; this may cause color splitting */
1964 subcolorcvec(v, cv, cstate, cstate);
1965 NOERR();
1966
1967 /* close new open subcolors to ensure the cache entry is self-contained */
1968 okcolors(v->nfa, v->cm);
1969 NOERR();
1970
1971 /* success! save the cache pointer */
1972 v->wordchrs = cstate;
1973 }
1974
1975 /*
1976 * processlacon - generate the NFA representation of a LACON
1977 *
1978 * In the general case this is just newlacon() + newarc(), but some cases
1979 * can be optimized.
1980 */
1981 static void
processlacon(struct vars * v,struct state * begin,struct state * end,int latype,struct state * lp,struct state * rp)1982 processlacon(struct vars *v,
1983 struct state *begin, /* start of parsed LACON sub-re */
1984 struct state *end, /* end of parsed LACON sub-re */
1985 int latype,
1986 struct state *lp, /* left state to hang it on */
1987 struct state *rp) /* right state to hang it on */
1988 {
1989 struct state *s1;
1990 int n;
1991
1992 /*
1993 * Check for lookaround RE consisting of a single plain color arc (or set
1994 * of arcs); this would typically be a simple chr or a bracket expression.
1995 */
1996 s1 = single_color_transition(begin, end);
1997 switch (latype)
1998 {
1999 case LATYPE_AHEAD_POS:
2000 /* If lookahead RE is just colorset C, convert to AHEAD(C) */
2001 if (s1 != NULL)
2002 {
2003 cloneouts(v->nfa, s1, lp, rp, AHEAD);
2004 return;
2005 }
2006 break;
2007 case LATYPE_AHEAD_NEG:
2008 /* If lookahead RE is just colorset C, convert to AHEAD(^C)|$ */
2009 if (s1 != NULL)
2010 {
2011 colorcomplement(v->nfa, v->cm, AHEAD, s1, lp, rp);
2012 newarc(v->nfa, '$', 1, lp, rp);
2013 newarc(v->nfa, '$', 0, lp, rp);
2014 return;
2015 }
2016 break;
2017 case LATYPE_BEHIND_POS:
2018 /* If lookbehind RE is just colorset C, convert to BEHIND(C) */
2019 if (s1 != NULL)
2020 {
2021 cloneouts(v->nfa, s1, lp, rp, BEHIND);
2022 return;
2023 }
2024 break;
2025 case LATYPE_BEHIND_NEG:
2026 /* If lookbehind RE is just colorset C, convert to BEHIND(^C)|^ */
2027 if (s1 != NULL)
2028 {
2029 colorcomplement(v->nfa, v->cm, BEHIND, s1, lp, rp);
2030 newarc(v->nfa, '^', 1, lp, rp);
2031 newarc(v->nfa, '^', 0, lp, rp);
2032 return;
2033 }
2034 break;
2035 default:
2036 assert(NOTREACHED);
2037 }
2038
2039 /* General case: we need a LACON subre and arc */
2040 n = newlacon(v, begin, end, latype);
2041 newarc(v->nfa, LACON, n, lp, rp);
2042 }
2043
2044 /*
2045 * subre - allocate a subre
2046 */
2047 static struct subre *
subre(struct vars * v,int op,int flags,struct state * begin,struct state * end)2048 subre(struct vars *v,
2049 int op,
2050 int flags,
2051 struct state *begin,
2052 struct state *end)
2053 {
2054 struct subre *ret = v->treefree;
2055
2056 /*
2057 * Checking for stack overflow here is sufficient to protect parse() and
2058 * its recursive subroutines.
2059 */
2060 if (STACK_TOO_DEEP(v->re))
2061 {
2062 ERR(REG_ETOOBIG);
2063 return NULL;
2064 }
2065
2066 if (ret != NULL)
2067 v->treefree = ret->child;
2068 else
2069 {
2070 ret = (struct subre *) MALLOC(sizeof(struct subre));
2071 if (ret == NULL)
2072 {
2073 ERR(REG_ESPACE);
2074 return NULL;
2075 }
2076 ret->chain = v->treechain;
2077 v->treechain = ret;
2078 }
2079
2080 assert(strchr("=b|.*(", op) != NULL);
2081
2082 ret->op = op;
2083 ret->flags = flags;
2084 ret->latype = (char) -1;
2085 ret->id = 0; /* will be assigned later */
2086 ret->capno = 0;
2087 ret->backno = 0;
2088 ret->min = ret->max = 1;
2089 ret->child = NULL;
2090 ret->sibling = NULL;
2091 ret->begin = begin;
2092 ret->end = end;
2093 ZAPCNFA(ret->cnfa);
2094
2095 return ret;
2096 }
2097
2098 /*
2099 * freesubre - free a subRE subtree
2100 *
2101 * This frees child node(s) of the given subRE too,
2102 * but not its siblings.
2103 */
2104 static void
freesubre(struct vars * v,struct subre * sr)2105 freesubre(struct vars *v, /* might be NULL */
2106 struct subre *sr)
2107 {
2108 if (sr == NULL)
2109 return;
2110
2111 if (sr->child != NULL)
2112 freesubreandsiblings(v, sr->child);
2113
2114 freesrnode(v, sr);
2115 }
2116
2117 /*
2118 * freesubreandsiblings - free a subRE subtree
2119 *
2120 * This frees child node(s) of the given subRE too,
2121 * as well as any following siblings.
2122 */
2123 static void
freesubreandsiblings(struct vars * v,struct subre * sr)2124 freesubreandsiblings(struct vars *v, /* might be NULL */
2125 struct subre *sr)
2126 {
2127 while (sr != NULL)
2128 {
2129 struct subre *next = sr->sibling;
2130
2131 freesubre(v, sr);
2132 sr = next;
2133 }
2134 }
2135
2136 /*
2137 * freesrnode - free one node in a subRE subtree
2138 */
2139 static void
freesrnode(struct vars * v,struct subre * sr)2140 freesrnode(struct vars *v, /* might be NULL */
2141 struct subre *sr)
2142 {
2143 if (sr == NULL)
2144 return;
2145
2146 if (!NULLCNFA(sr->cnfa))
2147 freecnfa(&sr->cnfa);
2148 sr->flags = 0; /* in particular, not INUSE */
2149 sr->child = sr->sibling = NULL;
2150 sr->begin = sr->end = NULL;
2151
2152 if (v != NULL && v->treechain != NULL)
2153 {
2154 /* we're still parsing, maybe we can reuse the subre */
2155 sr->child = v->treefree;
2156 v->treefree = sr;
2157 }
2158 else
2159 FREE(sr);
2160 }
2161
2162 /*
2163 * optst - optimize a subRE subtree
2164 */
2165 static void
optst(struct vars * v,struct subre * t)2166 optst(struct vars *v,
2167 struct subre *t)
2168 {
2169 /*
2170 * DGP (2007-11-13): I assume it was the programmer's intent to eventually
2171 * come back and add code to optimize subRE trees, but the routine coded
2172 * just spends effort traversing the tree and doing nothing. We can do
2173 * nothing with less effort.
2174 */
2175 return;
2176 }
2177
2178 /*
2179 * numst - number tree nodes (assigning "id" indexes)
2180 */
2181 static int /* next number */
numst(struct subre * t,int start)2182 numst(struct subre *t,
2183 int start) /* starting point for subtree numbers */
2184 {
2185 int i;
2186 struct subre *t2;
2187
2188 assert(t != NULL);
2189
2190 i = start;
2191 t->id = i++;
2192 for (t2 = t->child; t2 != NULL; t2 = t2->sibling)
2193 i = numst(t2, i);
2194 return i;
2195 }
2196
2197 /*
2198 * markst - mark tree nodes as INUSE
2199 *
2200 * Note: this is a great deal more subtle than it looks. During initial
2201 * parsing of a regex, all subres are linked into the treechain list;
2202 * discarded ones are also linked into the treefree list for possible reuse.
2203 * After we are done creating all subres required for a regex, we run markst()
2204 * then cleanst(), which results in discarding all subres not reachable from
2205 * v->tree. We then clear v->treechain, indicating that subres must be found
2206 * by descending from v->tree. This changes the behavior of freesubre(): it
2207 * will henceforth FREE() unwanted subres rather than sticking them into the
2208 * treefree list. (Doing that any earlier would result in dangling links in
2209 * the treechain list.) This all means that freev() will clean up correctly
2210 * if invoked before or after markst()+cleanst(); but it would not work if
2211 * called partway through this state conversion, so we mustn't error out
2212 * in or between these two functions.
2213 */
2214 static void
markst(struct subre * t)2215 markst(struct subre *t)
2216 {
2217 struct subre *t2;
2218
2219 assert(t != NULL);
2220
2221 t->flags |= INUSE;
2222 for (t2 = t->child; t2 != NULL; t2 = t2->sibling)
2223 markst(t2);
2224 }
2225
2226 /*
2227 * cleanst - free any tree nodes not marked INUSE
2228 */
2229 static void
cleanst(struct vars * v)2230 cleanst(struct vars *v)
2231 {
2232 struct subre *t;
2233 struct subre *next;
2234
2235 for (t = v->treechain; t != NULL; t = next)
2236 {
2237 next = t->chain;
2238 if (!(t->flags & INUSE))
2239 FREE(t);
2240 }
2241 v->treechain = NULL;
2242 v->treefree = NULL; /* just on general principles */
2243 }
2244
2245 /*
2246 * nfatree - turn a subRE subtree into a tree of compacted NFAs
2247 */
2248 static long /* optimize results from top node */
nfatree(struct vars * v,struct subre * t,FILE * f)2249 nfatree(struct vars *v,
2250 struct subre *t,
2251 FILE *f) /* for debug output */
2252 {
2253 struct subre *t2;
2254
2255 assert(t != NULL && t->begin != NULL);
2256
2257 for (t2 = t->child; t2 != NULL; t2 = t2->sibling)
2258 (DISCARD) nfatree(v, t2, f);
2259
2260 return nfanode(v, t, 0, f);
2261 }
2262
2263 /*
2264 * nfanode - do one NFA for nfatree or lacons
2265 *
2266 * If converttosearch is true, apply makesearch() to the NFA.
2267 */
2268 static long /* optimize results */
nfanode(struct vars * v,struct subre * t,int converttosearch,FILE * f)2269 nfanode(struct vars *v,
2270 struct subre *t,
2271 int converttosearch,
2272 FILE *f) /* for debug output */
2273 {
2274 struct nfa *nfa;
2275 long ret = 0;
2276
2277 assert(t->begin != NULL);
2278
2279 #ifdef REG_DEBUG
2280 if (f != NULL)
2281 {
2282 char idbuf[50];
2283
2284 fprintf(f, "\n\n\n========= TREE NODE %s ==========\n",
2285 stid(t, idbuf, sizeof(idbuf)));
2286 }
2287 #endif
2288 nfa = newnfa(v, v->cm, v->nfa);
2289 NOERRZ();
2290 dupnfa(nfa, t->begin, t->end, nfa->init, nfa->final);
2291 if (!ISERR())
2292 specialcolors(nfa);
2293 if (!ISERR())
2294 ret = optimize(nfa, f);
2295 if (converttosearch && !ISERR())
2296 makesearch(v, nfa);
2297 if (!ISERR())
2298 compact(nfa, &t->cnfa);
2299
2300 freenfa(nfa);
2301 return ret;
2302 }
2303
2304 /*
2305 * newlacon - allocate a lookaround-constraint subRE
2306 */
2307 static int /* lacon number */
newlacon(struct vars * v,struct state * begin,struct state * end,int latype)2308 newlacon(struct vars *v,
2309 struct state *begin,
2310 struct state *end,
2311 int latype)
2312 {
2313 int n;
2314 struct subre *newlacons;
2315 struct subre *sub;
2316
2317 if (v->nlacons == 0)
2318 {
2319 n = 1; /* skip 0th */
2320 newlacons = (struct subre *) MALLOC(2 * sizeof(struct subre));
2321 }
2322 else
2323 {
2324 n = v->nlacons;
2325 newlacons = (struct subre *) REALLOC(v->lacons,
2326 (n + 1) * sizeof(struct subre));
2327 }
2328 if (newlacons == NULL)
2329 {
2330 ERR(REG_ESPACE);
2331 return 0;
2332 }
2333 v->lacons = newlacons;
2334 v->nlacons = n + 1;
2335 sub = &v->lacons[n];
2336 sub->begin = begin;
2337 sub->end = end;
2338 sub->latype = latype;
2339 ZAPCNFA(sub->cnfa);
2340 return n;
2341 }
2342
2343 /*
2344 * freelacons - free lookaround-constraint subRE vector
2345 */
2346 static void
freelacons(struct subre * subs,int n)2347 freelacons(struct subre *subs,
2348 int n)
2349 {
2350 struct subre *sub;
2351 int i;
2352
2353 assert(n > 0);
2354 for (sub = subs + 1, i = n - 1; i > 0; sub++, i--) /* no 0th */
2355 if (!NULLCNFA(sub->cnfa))
2356 freecnfa(&sub->cnfa);
2357 FREE(subs);
2358 }
2359
2360 /*
2361 * rfree - free a whole RE (insides of regfree)
2362 */
2363 static void
rfree(regex_t * re)2364 rfree(regex_t *re)
2365 {
2366 struct guts *g;
2367
2368 if (re == NULL || re->re_magic != REMAGIC)
2369 return;
2370
2371 re->re_magic = 0; /* invalidate RE */
2372 g = (struct guts *) re->re_guts;
2373 re->re_guts = NULL;
2374 re->re_fns = NULL;
2375 if (g != NULL)
2376 {
2377 g->magic = 0;
2378 freecm(&g->cmap);
2379 if (g->tree != NULL)
2380 freesubre((struct vars *) NULL, g->tree);
2381 if (g->lacons != NULL)
2382 freelacons(g->lacons, g->nlacons);
2383 if (!NULLCNFA(g->search))
2384 freecnfa(&g->search);
2385 FREE(g);
2386 }
2387 }
2388
2389 /*
2390 * rcancelrequested - check for external request to cancel regex operation
2391 *
2392 * Return nonzero to fail the operation with error code REG_CANCEL,
2393 * zero to keep going
2394 *
2395 * The current implementation is Postgres-specific. If we ever get around
2396 * to splitting the regex code out as a standalone library, there will need
2397 * to be some API to let applications define a callback function for this.
2398 */
2399 static int
rcancelrequested(void)2400 rcancelrequested(void)
2401 {
2402 return InterruptPending && (QueryCancelPending || ProcDiePending);
2403 }
2404
2405 /*
2406 * rstacktoodeep - check for stack getting dangerously deep
2407 *
2408 * Return nonzero to fail the operation with error code REG_ETOOBIG,
2409 * zero to keep going
2410 *
2411 * The current implementation is Postgres-specific. If we ever get around
2412 * to splitting the regex code out as a standalone library, there will need
2413 * to be some API to let applications define a callback function for this.
2414 */
2415 static int
rstacktoodeep(void)2416 rstacktoodeep(void)
2417 {
2418 return stack_is_too_deep();
2419 }
2420
2421 #ifdef REG_DEBUG
2422
2423 /*
2424 * dump - dump an RE in human-readable form
2425 */
2426 static void
dump(regex_t * re,FILE * f)2427 dump(regex_t *re,
2428 FILE *f)
2429 {
2430 struct guts *g;
2431 int i;
2432
2433 if (re->re_magic != REMAGIC)
2434 fprintf(f, "bad magic number (0x%x not 0x%x)\n", re->re_magic,
2435 REMAGIC);
2436 if (re->re_guts == NULL)
2437 {
2438 fprintf(f, "NULL guts!!!\n");
2439 return;
2440 }
2441 g = (struct guts *) re->re_guts;
2442 if (g->magic != GUTSMAGIC)
2443 fprintf(f, "bad guts magic number (0x%x not 0x%x)\n", g->magic,
2444 GUTSMAGIC);
2445
2446 fprintf(f, "\n\n\n========= DUMP ==========\n");
2447 fprintf(f, "nsub %d, info 0%lo, csize %d, ntree %d\n",
2448 (int) re->re_nsub, re->re_info, re->re_csize, g->ntree);
2449
2450 dumpcolors(&g->cmap, f);
2451 if (!NULLCNFA(g->search))
2452 {
2453 fprintf(f, "\nsearch:\n");
2454 dumpcnfa(&g->search, f);
2455 }
2456 for (i = 1; i < g->nlacons; i++)
2457 {
2458 struct subre *lasub = &g->lacons[i];
2459 const char *latype;
2460
2461 switch (lasub->latype)
2462 {
2463 case LATYPE_AHEAD_POS:
2464 latype = "positive lookahead";
2465 break;
2466 case LATYPE_AHEAD_NEG:
2467 latype = "negative lookahead";
2468 break;
2469 case LATYPE_BEHIND_POS:
2470 latype = "positive lookbehind";
2471 break;
2472 case LATYPE_BEHIND_NEG:
2473 latype = "negative lookbehind";
2474 break;
2475 default:
2476 latype = "???";
2477 break;
2478 }
2479 fprintf(f, "\nla%d (%s):\n", i, latype);
2480 dumpcnfa(&lasub->cnfa, f);
2481 }
2482 fprintf(f, "\n");
2483 dumpst(g->tree, f, 0);
2484 }
2485
2486 /*
2487 * dumpst - dump a subRE tree
2488 */
2489 static void
dumpst(struct subre * t,FILE * f,int nfapresent)2490 dumpst(struct subre *t,
2491 FILE *f,
2492 int nfapresent) /* is the original NFA still around? */
2493 {
2494 if (t == NULL)
2495 fprintf(f, "null tree\n");
2496 else
2497 stdump(t, f, nfapresent);
2498 fflush(f);
2499 }
2500
2501 /*
2502 * stdump - recursive guts of dumpst
2503 */
2504 static void
stdump(struct subre * t,FILE * f,int nfapresent)2505 stdump(struct subre *t,
2506 FILE *f,
2507 int nfapresent) /* is the original NFA still around? */
2508 {
2509 char idbuf[50];
2510 struct subre *t2;
2511
2512 fprintf(f, "%s. `%c'", stid(t, idbuf, sizeof(idbuf)), t->op);
2513 if (t->flags & LONGER)
2514 fprintf(f, " longest");
2515 if (t->flags & SHORTER)
2516 fprintf(f, " shortest");
2517 if (t->flags & MIXED)
2518 fprintf(f, " hasmixed");
2519 if (t->flags & CAP)
2520 fprintf(f, " hascapture");
2521 if (t->flags & BACKR)
2522 fprintf(f, " hasbackref");
2523 if (!(t->flags & INUSE))
2524 fprintf(f, " UNUSED");
2525 if (t->latype != (char) -1)
2526 fprintf(f, " latype(%d)", t->latype);
2527 if (t->capno != 0)
2528 fprintf(f, " capture(%d)", t->capno);
2529 if (t->backno != 0)
2530 fprintf(f, " backref(%d)", t->backno);
2531 if (t->min != 1 || t->max != 1)
2532 {
2533 fprintf(f, " {%d,", t->min);
2534 if (t->max != DUPINF)
2535 fprintf(f, "%d", t->max);
2536 fprintf(f, "}");
2537 }
2538 if (nfapresent)
2539 fprintf(f, " %ld-%ld", (long) t->begin->no, (long) t->end->no);
2540 if (t->child != NULL)
2541 fprintf(f, " C:%s", stid(t->child, idbuf, sizeof(idbuf)));
2542 /* printing second child isn't necessary, but it is often helpful */
2543 if (t->child != NULL && t->child->sibling != NULL)
2544 fprintf(f, " C2:%s", stid(t->child->sibling, idbuf, sizeof(idbuf)));
2545 if (t->sibling != NULL)
2546 fprintf(f, " S:%s", stid(t->sibling, idbuf, sizeof(idbuf)));
2547 if (!NULLCNFA(t->cnfa))
2548 {
2549 fprintf(f, "\n");
2550 dumpcnfa(&t->cnfa, f);
2551 }
2552 fprintf(f, "\n");
2553 for (t2 = t->child; t2 != NULL; t2 = t2->sibling)
2554 stdump(t2, f, nfapresent);
2555 }
2556
2557 /*
2558 * stid - identify a subtree node for dumping
2559 */
2560 static const char * /* points to buf or constant string */
stid(struct subre * t,char * buf,size_t bufsize)2561 stid(struct subre *t,
2562 char *buf,
2563 size_t bufsize)
2564 {
2565 /* big enough for hex int or decimal t->id? */
2566 if (bufsize < sizeof(void *) * 2 + 3 || bufsize < sizeof(t->id) * 3 + 1)
2567 return "unable";
2568 if (t->id != 0)
2569 sprintf(buf, "%d", t->id);
2570 else
2571 sprintf(buf, "%p", t);
2572 return buf;
2573 }
2574 #endif /* REG_DEBUG */
2575
2576
2577 #include "regc_lex.c"
2578 #include "regc_color.c"
2579 #include "regc_nfa.c"
2580 #include "regc_cvec.c"
2581 #include "regc_pg_locale.c"
2582 #include "regc_locale.c"
2583