2 * re_*comp and friends - compile REs
3 * This file #includes several others (see the bottom).
5 * Copyright (c) 1998, 1999 Henry Spencer. All rights reserved.
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
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.
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.
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.
31 * src/backend/regex/regcomp.c
35 #include "regex/regguts.h"
38 * forward declarations, up here so forward datatypes etc. are defined early
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);
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);
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);
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 *);
116 static void dumpcolors(struct colormap *, FILE *);
117 static void dumpchr(chr, FILE *);
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 **);
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 *);
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 *);
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 *);
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);
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);
223 /* internal variables, bundled for easy passing around */
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" */
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 */
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)
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) */
292 /* is an arc colored, and hence on a color chain? */
294 ((a)->type == PLAIN || (a)->type == AHEAD || (a)->type == BEHIND)
297 /* static function list */
298 static const struct fns functions = {
299 rfree, /* regfree insides */
300 rcancelrequested, /* check for cancel request */
301 rstacktoodeep /* check for stack getting dangerously deep */
307 * pg_regcomp - compile regular expression
309 * Note: on failure, no resources remain allocated, so pg_regfree()
310 * need not be applied to re.
313 pg_regcomp(regex_t *re,
320 struct vars *v = &var;
326 FILE *debug = (flags & REG_PROGRESS) ? stdout : (FILE *) NULL;
328 FILE *debug = (FILE *) NULL;
331 #define CNOERR() { if (ISERR()) return freev(v, v->err); }
335 if (re == NULL || string == NULL)
337 if ((flags & REG_QUOTE) &&
338 (flags & (REG_ADVANCED | REG_EXPANDED | REG_NEWLINE)))
340 if (!(flags & REG_EXTENDED) && (flags & REG_ADVF))
343 /* Initialize locale-dependent support */
344 pg_set_regex_collation(collation);
346 /* initial setup (after which freev() is callable) */
349 v->stop = v->now + len;
350 v->savenow = v->savestop = NULL;
356 for (j = 0; j < v->nsubs; j++)
360 v->nlcolor = COLORLESS;
370 re->re_magic = REMAGIC;
371 re->re_info = 0; /* bits get set during parse */
372 re->re_csize = sizeof(chr);
373 re->re_collation = collation;
375 re->re_fns = VS(&functions);
377 /* more complex setup, malloced things */
378 re->re_guts = VS(MALLOC(sizeof(struct guts)));
379 if (re->re_guts == NULL)
380 return freev(v, REG_ESPACE);
381 g = (struct guts *) re->re_guts;
388 v->nfa = newnfa(v, v->cm, (struct nfa *) NULL);
390 /* set up a reasonably-sized transient cvec for getcvec usage */
391 v->cv = newcvec(100, 20);
393 return freev(v, REG_ESPACE);
396 lexstart(v); /* also handles prefixes */
397 if ((v->cflags & REG_NLSTOP) || (v->cflags & REG_NLANCH))
399 /* assign newline a unique color */
400 v->nlcolor = subcolor(v->cm, newline());
401 okcolors(v->nfa, v->cm);
404 v->tree = parse(v, EOS, PLAIN, v->nfa->init, v->nfa->final);
405 assert(SEE(EOS)); /* even if error; ISERR() => SEE(EOS) */
407 assert(v->tree != NULL);
409 /* finish setup of nfa and its subre tree */
410 specialcolors(v->nfa);
415 fprintf(debug, "\n\n\n========= RAW ==========\n");
416 dumpnfa(v->nfa, debug);
417 dumpst(v->tree, debug, 1);
421 v->ntree = numst(v->tree, 1);
427 fprintf(debug, "\n\n\n========= TREE FIXED ==========\n");
428 dumpst(v->tree, debug, 1);
432 /* build compacted NFAs for tree and lacons */
433 re->re_info |= nfatree(v, v->tree, debug);
435 assert(v->nlacons == 0 || v->lacons != NULL);
436 for (i = 1; i < v->nlacons; i++)
438 struct subre *lasub = &v->lacons[i];
442 fprintf(debug, "\n\n\n========= LA%d ==========\n", i);
445 /* Prepend .* to pattern if it's a lookbehind LACON */
446 nfanode(v, lasub, !LATYPE_IS_AHEAD(lasub->subno), debug);
449 if (v->tree->flags & SHORTER)
452 /* build compacted NFAs for tree, lacons, fast search */
455 fprintf(debug, "\n\n\n========= SEARCH ==========\n");
457 /* can sacrifice main NFA now, so use it as work area */
458 (DISCARD) optimize(v->nfa, debug);
460 makesearch(v, v->nfa);
462 compact(v->nfa, &g->search);
465 /* looks okay, package it up */
466 re->re_nsub = v->nsubexp;
467 v->re = NULL; /* freev no longer frees re */
468 g->magic = GUTSMAGIC;
469 g->cflags = v->cflags;
470 g->info = re->re_info;
471 g->nsub = re->re_nsub;
475 g->compare = (v->cflags & REG_ICASE) ? casecmp : cmp;
476 g->lacons = v->lacons;
478 g->nlacons = v->nlacons;
481 if (flags & REG_DUMP)
490 * moresubs - enlarge subRE vector
493 moresubs(struct vars *v,
494 int wanted) /* want enough room for this one */
499 assert(wanted > 0 && (size_t) wanted >= v->nsubs);
500 n = (size_t) wanted * 3 / 2 + 1;
502 if (v->subs == v->sub10)
504 p = (struct subre **) MALLOC(n * sizeof(struct subre *));
506 memcpy(VS(p), VS(v->subs),
507 v->nsubs * sizeof(struct subre *));
510 p = (struct subre **) REALLOC(v->subs, n * sizeof(struct subre *));
517 for (p = &v->subs[v->nsubs]; v->nsubs < n; p++, v->nsubs++)
519 assert(v->nsubs == n);
520 assert((size_t) wanted < v->nsubs);
524 * freev - free vars struct's substructures where necessary
526 * Optionally does error-number setting, and always returns error code
527 * (if any), to make error-handling code terser.
530 freev(struct vars *v,
535 if (v->subs != v->sub10)
540 freesubre(v, v->tree);
541 if (v->treechain != NULL)
547 if (v->lacons != NULL)
548 freelacons(v->lacons, v->nlacons);
549 ERR(err); /* nop if err==0 */
555 * makesearch - turn an NFA into a search NFA (implicit prepend of .*?)
556 * NFA must have been optimize()d already.
559 makesearch(struct vars *v,
564 struct state *pre = nfa->pre;
569 /* no loops are needed if it's anchored */
570 for (a = pre->outs; a != NULL; a = a->outchain)
572 assert(a->type == PLAIN);
573 if (a->co != nfa->bos[0] && a->co != nfa->bos[1])
578 /* add implicit .* in front */
579 rainbow(nfa, v->cm, PLAIN, COLORLESS, pre, pre);
581 /* and ^* and \A* too -- not always necessary, but harmless */
582 newarc(nfa, PLAIN, nfa->bos[0], pre, pre);
583 newarc(nfa, PLAIN, nfa->bos[1], pre, pre);
587 * Now here's the subtle part. Because many REs have no lookback
588 * constraints, often knowing when you were in the pre state tells you
589 * little; it's the next state(s) that are informative. But some of them
590 * may have other inarcs, i.e. it may be possible to make actual progress
591 * and then return to one of them. We must de-optimize such cases,
592 * splitting each such state into progress and no-progress states.
595 /* first, make a list of the states reachable from pre and elsewhere */
597 for (a = pre->outs; a != NULL; a = a->outchain)
600 for (b = s->ins; b != NULL; b = b->inchain)
607 * We want to mark states as being in the list already by having non
608 * NULL tmp fields, but we can't just store the old slist value in tmp
609 * because that doesn't work for the first such state. Instead, the
610 * first list entry gets its own address in tmp.
612 if (b != NULL && s->tmp == NULL)
614 s->tmp = (slist != NULL) ? slist : s;
620 for (s = slist; s != NULL; s = s2)
624 copyouts(nfa, s, s2);
626 for (a = s->ins; a != NULL; a = b)
631 cparc(nfa, a, a->from, s2);
635 s2 = (s->tmp != s) ? s->tmp : NULL;
636 s->tmp = NULL; /* clean up while we're at it */
641 * parse - parse an RE
643 * This is actually just the top level, which parses a bunch of branches
644 * tied together with '|'. They appear in the tree as the left children
645 * of a chain of '|' subres.
647 static struct subre *
648 parse(struct vars *v,
649 int stopper, /* EOS or ')' */
650 int type, /* LACON (lookaround subRE) or PLAIN */
651 struct state *init, /* initial state */
652 struct state *final) /* final state */
654 struct state *left; /* scaffolding for branch */
656 struct subre *branches; /* top level */
657 struct subre *branch; /* current branch */
658 struct subre *t; /* temporary */
659 int firstbranch; /* is this the first branch? */
661 assert(stopper == ')' || stopper == EOS);
663 branches = subre(v, '|', LONGER, init, final);
671 /* need a place to hang it */
672 branch->right = subre(v, '|', LONGER, init, final);
674 branch = branch->right;
677 left = newstate(v->nfa);
678 right = newstate(v->nfa);
680 EMPTYARC(init, left);
681 EMPTYARC(right, final);
683 branch->left = parsebranch(v, stopper, type, left, right, 0);
685 branch->flags |= UP(branch->flags | branch->left->flags);
686 if ((branch->flags & ~branches->flags) != 0) /* new flags */
687 for (t = branches; t != branch; t = t->right)
688 t->flags |= branch->flags;
690 assert(SEE(stopper) || SEE(EOS));
694 assert(stopper == ')' && SEE(EOS));
698 /* optimize out simple cases */
699 if (branch == branches)
700 { /* only one branch */
701 assert(branch->right == NULL);
704 freesubre(v, branches);
707 else if (!MESSY(branches->flags))
708 { /* no interesting innards */
709 freesubre(v, branches->left);
710 branches->left = NULL;
711 freesubre(v, branches->right);
712 branches->right = NULL;
720 * parsebranch - parse one branch of an RE
722 * This mostly manages concatenation, working closely with parseqatom().
723 * Concatenated things are bundled up as much as possible, with separate
724 * ',' nodes introduced only when necessary due to substructure.
726 static struct subre *
727 parsebranch(struct vars *v,
728 int stopper, /* EOS or ')' */
729 int type, /* LACON (lookaround subRE) or PLAIN */
730 struct state *left, /* leftmost state */
731 struct state *right, /* rightmost state */
732 int partial) /* is this only part of a branch? */
734 struct state *lp; /* left end of current construct */
735 int seencontent; /* is there anything in this branch yet? */
740 t = subre(v, '=', 0, left, right); /* op '=' is tentative */
742 while (!SEE('|') && !SEE(stopper) && !SEE(EOS))
745 { /* implicit concat operator */
746 lp = newstate(v->nfa);
748 moveins(v->nfa, right, lp);
752 /* NB, recursion in parseqatom() may swallow rest of branch */
753 parseqatom(v, stopper, type, lp, right, t);
762 EMPTYARC(left, right);
769 * parseqatom - parse one quantified atom or constraint of an RE
771 * The bookkeeping near the end cooperates very closely with parsebranch();
772 * in particular, it contains a recursion that can involve parsing the rest
773 * of the branch, making this function's name somewhat inaccurate.
776 parseqatom(struct vars *v,
777 int stopper, /* EOS or ')' */
778 int type, /* LACON (lookaround subRE) or PLAIN */
779 struct state *lp, /* left state to hang it on */
780 struct state *rp, /* right state to hang it on */
781 struct subre *top) /* subtree top */
783 struct state *s; /* temporaries for new states */
786 #define ARCV(t, val) newarc(v->nfa, t, val, lp, rp)
789 struct subre *atom; /* atom's subtree */
791 int cap; /* capturing parens? */
792 int latype; /* lookaround constraint type */
793 int subno; /* capturing-parens or backref number */
795 int qprefer; /* quantifier short/long preference */
797 struct subre **atomp; /* where the pointer to atom is */
799 /* initial bookkeeping */
801 assert(lp->nouts == 0); /* must string new code */
802 assert(rp->nins == 0); /* between lp and rp */
803 subno = 0; /* just to shut lint up */
805 /* an atom or constraint... */
806 atomtype = v->nexttype;
809 /* first, constraints, which end by returning */
812 if (v->cflags & REG_NLANCH)
813 ARCV(BEHIND, v->nlcolor);
819 if (v->cflags & REG_NLANCH)
820 ARCV(AHEAD, v->nlcolor);
825 ARCV('^', 1); /* BOL */
826 ARCV('^', 0); /* or BOS */
831 ARCV('$', 1); /* EOL */
832 ARCV('$', 0); /* or EOS */
837 wordchrs(v); /* does NEXT() */
838 s = newstate(v->nfa);
840 nonword(v, BEHIND, lp, s);
841 word(v, AHEAD, s, rp);
845 wordchrs(v); /* does NEXT() */
846 s = newstate(v->nfa);
848 word(v, BEHIND, lp, s);
849 nonword(v, AHEAD, s, rp);
853 wordchrs(v); /* does NEXT() */
854 s = newstate(v->nfa);
856 nonword(v, BEHIND, lp, s);
857 word(v, AHEAD, s, rp);
858 s = newstate(v->nfa);
860 word(v, BEHIND, lp, s);
861 nonword(v, AHEAD, s, rp);
865 wordchrs(v); /* does NEXT() */
866 s = newstate(v->nfa);
868 word(v, BEHIND, lp, s);
869 word(v, AHEAD, s, rp);
870 s = newstate(v->nfa);
872 nonword(v, BEHIND, lp, s);
873 nonword(v, AHEAD, s, rp);
876 case LACON: /* lookaround constraint */
877 latype = v->nextvalue;
879 s = newstate(v->nfa);
880 s2 = newstate(v->nfa);
882 t = parse(v, ')', LACON, s, s2);
883 freesubre(v, t); /* internal structure irrelevant */
887 processlacon(v, s, s2, latype, lp, rp);
890 /* then errors, to get them out of the way */
902 /* then plain characters, and minor variants on that theme */
903 case ')': /* unbalanced paren */
904 if ((v->cflags & REG_ADVANCED) != REG_EXTENDED)
909 /* legal in EREs due to specification botch */
911 /* fall through into case PLAIN */
914 onechr(v, v->nextvalue, lp, rp);
915 okcolors(v->nfa, v->cm);
920 if (v->nextvalue == 1)
924 assert(SEE(']') || ISERR());
928 rainbow(v->nfa, v->cm, PLAIN,
929 (v->cflags & REG_NLSTOP) ? v->nlcolor : COLORLESS,
933 /* and finally the ugly stuff */
934 case '(': /* value flags as capturing or non */
935 cap = (type == LACON) ? 0 : v->nextvalue;
940 if ((size_t) subno >= v->nsubs)
942 assert((size_t) subno < v->nsubs);
945 atomtype = PLAIN; /* something that's not '(' */
947 /* need new endpoints because tree will contain pointers */
948 s = newstate(v->nfa);
949 s2 = newstate(v->nfa);
954 atom = parse(v, ')', type, s, s2);
955 assert(SEE(')') || ISERR());
960 v->subs[subno] = atom;
961 t = subre(v, '(', atom->flags | CAP, lp, rp);
967 /* postpone everything else pending possible {0} */
969 case BACKREF: /* the Feature From The Black Lagoon */
970 INSIST(type != LACON, REG_ESUBREG);
971 INSIST(v->nextvalue < v->nsubs, REG_ESUBREG);
972 INSIST(v->subs[v->nextvalue] != NULL, REG_ESUBREG);
974 assert(v->nextvalue > 0);
975 atom = subre(v, 'b', BACKR, lp, rp);
977 subno = v->nextvalue;
979 EMPTYARC(lp, rp); /* temporarily, so there's something */
984 /* ...and an atom may be followed by a quantifier */
990 qprefer = (v->nextvalue) ? LONGER : SHORTER;
996 qprefer = (v->nextvalue) ? LONGER : SHORTER;
1002 qprefer = (v->nextvalue) ? LONGER : SHORTER;
1019 /* {m,n} exercises preference, even if it's {m,m} */
1020 qprefer = (v->nextvalue) ? LONGER : SHORTER;
1025 /* {m} passes operand's preference through */
1029 { /* catches errors too */
1035 default: /* no quantifier */
1041 /* annoying special case: {0} or {0,0} cancels everything */
1042 if (m == 0 && n == 0)
1046 if (atomtype == '(')
1047 v->subs[subno] = NULL;
1048 delsub(v->nfa, lp, rp);
1053 /* if not a messy case, avoid hard part */
1054 assert(!MESSY(top->flags));
1055 f = top->flags | qprefer | ((atom != NULL) ? atom->flags : 0);
1056 if (atomtype != '(' && atomtype != BACKREF && !MESSY(UP(f)))
1058 if (!(m == 1 && n == 1))
1059 repeat(v, lp, rp, m, n);
1067 * hard part: something messy
1069 * That is, capturing parens, back reference, short/long clash, or an atom
1070 * with substructure containing one of those.
1073 /* now we'll need a subre for the contents even if they're boring */
1076 atom = subre(v, '=', 0, lp, rp);
1081 * Prepare a general-purpose state skeleton.
1083 * In the no-backrefs case, we want this:
1085 * [lp] ---> [s] ---prefix---> [begin] ---atom---> [end] ---rest---> [rp]
1087 * where prefix is some repetitions of atom. In the general case we need
1089 * [lp] ---> [s] ---iterator---> [s2] ---rest---> [rp]
1091 * where the iterator wraps around [begin] ---atom---> [end]
1093 * We make the s state here for both cases; s2 is made below if needed
1096 s = newstate(v->nfa); /* first, new endpoints for the atom */
1097 s2 = newstate(v->nfa);
1099 moveouts(v->nfa, lp, s);
1100 moveins(v->nfa, rp, s2);
1104 s = newstate(v->nfa); /* set up starting state */
1109 /* break remaining subRE into x{...} and what follows */
1110 t = subre(v, '.', COMBINE(qprefer, atom->flags), lp, rp);
1115 /* here we should recurse... but we must postpone that to the end */
1117 /* split top into prefix and remaining */
1118 assert(top->op == '=' && top->left == NULL && top->right == NULL);
1119 top->left = subre(v, '=', top->flags, top->begin, lp);
1124 /* if it's a backref, now is the time to replicate the subNFA */
1125 if (atomtype == BACKREF)
1127 assert(atom->begin->nouts == 1); /* just the EMPTY */
1128 delsub(v->nfa, atom->begin, atom->end);
1129 assert(v->subs[subno] != NULL);
1132 * And here's why the recursion got postponed: it must wait until the
1133 * skeleton is filled in, because it may hit a backref that wants to
1134 * copy the filled-in skeleton.
1136 dupnfa(v->nfa, v->subs[subno]->begin, v->subs[subno]->end,
1137 atom->begin, atom->end);
1142 * It's quantifier time. If the atom is just a backref, we'll let it deal
1143 * with quantifiers internally.
1145 if (atomtype == BACKREF)
1147 /* special case: backrefs have internal quantifiers */
1148 EMPTYARC(s, atom->begin); /* empty prefix */
1149 /* just stuff everything into atom */
1150 repeat(v, atom->begin, atom->end, m, n);
1151 atom->min = (short) m;
1152 atom->max = (short) n;
1153 atom->flags |= COMBINE(qprefer, atom->flags);
1154 /* rest of branch can be strung starting from atom->end */
1157 else if (m == 1 && n == 1 &&
1159 (atom->flags & (LONGER | SHORTER | MIXED)) == 0 ||
1160 qprefer == (atom->flags & (LONGER | SHORTER | MIXED))))
1162 /* no/vacuous quantifier: done */
1163 EMPTYARC(s, atom->begin); /* empty prefix */
1164 /* rest of branch can be strung starting from atom->end */
1167 else if (m > 0 && !(atom->flags & BACKR))
1170 * If there's no backrefs involved, we can turn x{m,n} into
1171 * x{m-1,n-1}x, with capturing parens in only the second x. This is
1172 * valid because we only care about capturing matches from the final
1173 * iteration of the quantifier. It's a win because we can implement
1174 * the backref-free left side as a plain DFA node, since we don't
1175 * really care where its submatches are.
1177 dupnfa(v->nfa, atom->begin, atom->end, s, atom->begin);
1178 assert(m >= 1 && m != DUPINF && n >= 1);
1179 repeat(v, s, atom->begin, m - 1, (n == DUPINF) ? n : n - 1);
1180 f = COMBINE(qprefer, atom->flags);
1181 t = subre(v, '.', f, s, atom->end); /* prefix and atom */
1183 t->left = subre(v, '=', PREF(f), s, atom->begin);
1187 /* rest of branch can be strung starting from atom->end */
1192 /* general case: need an iteration node */
1193 s2 = newstate(v->nfa);
1195 moveouts(v->nfa, atom->end, s2);
1197 dupnfa(v->nfa, atom->begin, atom->end, s, s2);
1198 repeat(v, s, s2, m, n);
1199 f = COMBINE(qprefer, atom->flags);
1200 t = subre(v, '*', f, s, s2);
1206 /* rest of branch is to be strung from iteration's end state */
1209 /* and finally, look after that postponed recursion */
1211 if (!(SEE('|') || SEE(stopper) || SEE(EOS)))
1212 t->right = parsebranch(v, stopper, type, s2, rp, 1);
1216 t->right = subre(v, '=', 0, s2, rp);
1219 assert(SEE('|') || SEE(stopper) || SEE(EOS));
1220 t->flags |= COMBINE(t->flags, t->right->flags);
1221 top->flags |= COMBINE(top->flags, t->flags);
1225 * nonword - generate arcs for non-word-character ahead or behind
1228 nonword(struct vars *v,
1229 int dir, /* AHEAD or BEHIND */
1233 int anchor = (dir == AHEAD) ? '$' : '^';
1235 assert(dir == AHEAD || dir == BEHIND);
1236 newarc(v->nfa, anchor, 1, lp, rp);
1237 newarc(v->nfa, anchor, 0, lp, rp);
1238 colorcomplement(v->nfa, v->cm, dir, v->wordchrs, lp, rp);
1239 /* (no need for special attention to \n) */
1243 * word - generate arcs for word character ahead or behind
1246 word(struct vars *v,
1247 int dir, /* AHEAD or BEHIND */
1251 assert(dir == AHEAD || dir == BEHIND);
1252 cloneouts(v->nfa, v->wordchrs, lp, rp, dir);
1253 /* (no need for special attention to \n) */
1257 * scannum - scan a number
1259 static int /* value, <= DUPMAX */
1260 scannum(struct vars *v)
1264 while (SEE(DIGIT) && n < DUPMAX)
1266 n = n * 10 + v->nextvalue;
1269 if (SEE(DIGIT) || n > DUPMAX)
1278 * repeat - replicate subNFA for quantifiers
1280 * The sub-NFA strung from lp to rp is modified to represent m to n
1281 * repetitions of its initial contents.
1283 * The duplication sequences used here are chosen carefully so that any
1284 * pointers starting out pointing into the subexpression end up pointing into
1285 * the last occurrence. (Note that it may not be strung between the same
1286 * left and right end states, however!) This used to be important for the
1287 * subRE tree, although the important bits are now handled by the in-line
1288 * code in parse(), and when this is called, it doesn't matter any more.
1291 repeat(struct vars *v,
1299 #define PAIR(x, y) ((x)*4 + (y))
1300 #define REDUCE(x) ( ((x) == DUPINF) ? INF : (((x) > 1) ? SOME : (x)) )
1301 const int rm = REDUCE(m);
1302 const int rn = REDUCE(n);
1306 switch (PAIR(rm, rn))
1308 case PAIR(0, 0): /* empty string */
1309 delsub(v->nfa, lp, rp);
1312 case PAIR(0, 1): /* do as x| */
1315 case PAIR(0, SOME): /* do as x{1,n}| */
1316 repeat(v, lp, rp, 1, n);
1320 case PAIR(0, INF): /* loop x around */
1321 s = newstate(v->nfa);
1323 moveouts(v->nfa, lp, s);
1324 moveins(v->nfa, rp, s);
1328 case PAIR(1, 1): /* no action required */
1330 case PAIR(1, SOME): /* do as x{0,n-1}x = (x{1,n-1}|)x */
1331 s = newstate(v->nfa);
1333 moveouts(v->nfa, lp, s);
1334 dupnfa(v->nfa, s, rp, lp, s);
1336 repeat(v, lp, s, 1, n - 1);
1340 case PAIR(1, INF): /* add loopback arc */
1341 s = newstate(v->nfa);
1342 s2 = newstate(v->nfa);
1344 moveouts(v->nfa, lp, s);
1345 moveins(v->nfa, rp, s2);
1350 case PAIR(SOME, SOME): /* do as x{m-1,n-1}x */
1351 s = newstate(v->nfa);
1353 moveouts(v->nfa, lp, s);
1354 dupnfa(v->nfa, s, rp, lp, s);
1356 repeat(v, lp, s, m - 1, n - 1);
1358 case PAIR(SOME, INF): /* do as x{m-1,}x */
1359 s = newstate(v->nfa);
1361 moveouts(v->nfa, lp, s);
1362 dupnfa(v->nfa, s, rp, lp, s);
1364 repeat(v, lp, s, m - 1, n);
1373 * bracket - handle non-complemented bracket expression
1374 * Also called from cbracket for complemented bracket expressions.
1377 bracket(struct vars *v,
1383 while (!SEE(']') && !SEE(EOS))
1384 brackpart(v, lp, rp);
1385 assert(SEE(']') || ISERR());
1386 okcolors(v->nfa, v->cm);
1390 * cbracket - handle complemented bracket expression
1391 * We do it by calling bracket() with dummy endpoints, and then complementing
1392 * the result. The alternative would be to invoke rainbow(), and then delete
1393 * arcs as the b.e. is seen... but that gets messy.
1396 cbracket(struct vars *v,
1400 struct state *left = newstate(v->nfa);
1401 struct state *right = newstate(v->nfa);
1404 bracket(v, left, right);
1405 if (v->cflags & REG_NLSTOP)
1406 newarc(v->nfa, PLAIN, v->nlcolor, left, right);
1409 assert(lp->nouts == 0); /* all outarcs will be ours */
1412 * Easy part of complementing, and all there is to do since the MCCE code
1415 colorcomplement(v->nfa, v->cm, PLAIN, left, lp, rp);
1417 dropstate(v->nfa, left);
1418 assert(right->nins == 0);
1419 freestate(v->nfa, right);
1423 * brackpart - handle one item (or range) within a bracket expression
1426 brackpart(struct vars *v,
1437 /* parse something, get rid of special cases, take shortcuts */
1438 switch (v->nexttype)
1440 case RANGE: /* a-b-c or other botch */
1445 c[0] = v->nextvalue;
1447 /* shortcut for ordinary chr (not range) */
1450 onechr(v, c[0], lp, rp);
1453 startc = element(v, c, c + 1);
1458 endp = scanplain(v);
1459 INSIST(startp < endp, REG_ECOLLATE);
1461 startc = element(v, startp, endp);
1466 endp = scanplain(v);
1467 INSIST(startp < endp, REG_ECOLLATE);
1469 startc = element(v, startp, endp);
1471 cv = eclass(v, startc, (v->cflags & REG_ICASE));
1473 subcolorcvec(v, cv, lp, rp);
1478 endp = scanplain(v);
1479 INSIST(startp < endp, REG_ECTYPE);
1481 cv = cclass(v, startp, endp, (v->cflags & REG_ICASE));
1483 subcolorcvec(v, cv, lp, rp);
1495 switch (v->nexttype)
1499 c[0] = v->nextvalue;
1501 endc = element(v, c, c + 1);
1506 endp = scanplain(v);
1507 INSIST(startp < endp, REG_ECOLLATE);
1509 endc = element(v, startp, endp);
1522 * Ranges are unportable. Actually, standard C does guarantee that digits
1523 * are contiguous, but making that an exception is just too complicated.
1527 cv = range(v, startc, endc, (v->cflags & REG_ICASE));
1529 subcolorcvec(v, cv, lp, rp);
1533 * scanplain - scan PLAIN contents of [. etc.
1535 * Certain bits of trickery in lex.c know that this code does not try
1536 * to look past the final bracket of the [. etc.
1538 static const chr * /* just after end of sequence */
1539 scanplain(struct vars *v)
1543 assert(SEE(COLLEL) || SEE(ECLASS) || SEE(CCLASS));
1553 assert(SEE(END) || ISERR());
1560 * onechr - fill in arcs for a plain character, and possible case complements
1561 * This is mostly a shortcut for efficient handling of the common case.
1564 onechr(struct vars *v,
1569 if (!(v->cflags & REG_ICASE))
1571 color lastsubcolor = COLORLESS;
1573 subcoloronechr(v, c, lp, rp, &lastsubcolor);
1577 /* rats, need general case anyway... */
1578 subcolorcvec(v, allcases(v, c), lp, rp);
1582 * wordchrs - set up word-chr list for word-boundary stuff, if needed
1584 * The list is kept as a bunch of arcs between two dummy states; it's
1585 * disposed of by the unreachable-states sweep in NFA optimization.
1586 * Does NEXT(). Must not be called from any unusual lexical context.
1587 * This should be reconciled with the \w etc. handling in lex.c, and
1588 * should be cleaned up to reduce dependencies on input scanning.
1591 wordchrs(struct vars *v)
1594 struct state *right;
1596 if (v->wordchrs != NULL)
1598 NEXT(); /* for consistency */
1602 left = newstate(v->nfa);
1603 right = newstate(v->nfa);
1605 /* fine point: implemented with [::], and lexer will set REG_ULOCALE */
1608 assert(v->savenow != NULL && SEE('['));
1609 bracket(v, left, right);
1610 assert((v->savenow != NULL && SEE(']')) || ISERR());
1617 * processlacon - generate the NFA representation of a LACON
1619 * In the general case this is just newlacon() + newarc(), but some cases
1623 processlacon(struct vars *v,
1624 struct state *begin, /* start of parsed LACON sub-re */
1625 struct state *end, /* end of parsed LACON sub-re */
1627 struct state *lp, /* left state to hang it on */
1628 struct state *rp) /* right state to hang it on */
1634 * Check for lookaround RE consisting of a single plain color arc (or set
1635 * of arcs); this would typically be a simple chr or a bracket expression.
1637 s1 = single_color_transition(begin, end);
1640 case LATYPE_AHEAD_POS:
1641 /* If lookahead RE is just colorset C, convert to AHEAD(C) */
1644 cloneouts(v->nfa, s1, lp, rp, AHEAD);
1648 case LATYPE_AHEAD_NEG:
1649 /* If lookahead RE is just colorset C, convert to AHEAD(^C)|$ */
1652 colorcomplement(v->nfa, v->cm, AHEAD, s1, lp, rp);
1653 newarc(v->nfa, '$', 1, lp, rp);
1654 newarc(v->nfa, '$', 0, lp, rp);
1658 case LATYPE_BEHIND_POS:
1659 /* If lookbehind RE is just colorset C, convert to BEHIND(C) */
1662 cloneouts(v->nfa, s1, lp, rp, BEHIND);
1666 case LATYPE_BEHIND_NEG:
1667 /* If lookbehind RE is just colorset C, convert to BEHIND(^C)|^ */
1670 colorcomplement(v->nfa, v->cm, BEHIND, s1, lp, rp);
1671 newarc(v->nfa, '^', 1, lp, rp);
1672 newarc(v->nfa, '^', 0, lp, rp);
1680 /* General case: we need a LACON subre and arc */
1681 n = newlacon(v, begin, end, latype);
1682 newarc(v->nfa, LACON, n, lp, rp);
1686 * subre - allocate a subre
1688 static struct subre *
1689 subre(struct vars *v,
1692 struct state *begin,
1695 struct subre *ret = v->treefree;
1698 * Checking for stack overflow here is sufficient to protect parse() and
1699 * its recursive subroutines.
1701 if (STACK_TOO_DEEP(v->re))
1708 v->treefree = ret->left;
1711 ret = (struct subre *) MALLOC(sizeof(struct subre));
1717 ret->chain = v->treechain;
1721 assert(strchr("=b|.*(", op) != NULL);
1725 ret->id = 0; /* will be assigned later */
1727 ret->min = ret->max = 1;
1738 * freesubre - free a subRE subtree
1741 freesubre(struct vars *v, /* might be NULL */
1747 if (sr->left != NULL)
1748 freesubre(v, sr->left);
1749 if (sr->right != NULL)
1750 freesubre(v, sr->right);
1756 * freesrnode - free one node in a subRE subtree
1759 freesrnode(struct vars *v, /* might be NULL */
1765 if (!NULLCNFA(sr->cnfa))
1766 freecnfa(&sr->cnfa);
1769 if (v != NULL && v->treechain != NULL)
1771 /* we're still parsing, maybe we can reuse the subre */
1772 sr->left = v->treefree;
1780 * optst - optimize a subRE subtree
1783 optst(struct vars *v,
1787 * DGP (2007-11-13): I assume it was the programmer's intent to eventually
1788 * come back and add code to optimize subRE trees, but the routine coded
1789 * just spends effort traversing the tree and doing nothing. We can do
1790 * nothing with less effort.
1796 * numst - number tree nodes (assigning "id" indexes)
1798 static int /* next number */
1799 numst(struct subre *t,
1800 int start) /* starting point for subtree numbers */
1807 t->id = (short) i++;
1808 if (t->left != NULL)
1809 i = numst(t->left, i);
1810 if (t->right != NULL)
1811 i = numst(t->right, i);
1816 * markst - mark tree nodes as INUSE
1818 * Note: this is a great deal more subtle than it looks. During initial
1819 * parsing of a regex, all subres are linked into the treechain list;
1820 * discarded ones are also linked into the treefree list for possible reuse.
1821 * After we are done creating all subres required for a regex, we run markst()
1822 * then cleanst(), which results in discarding all subres not reachable from
1823 * v->tree. We then clear v->treechain, indicating that subres must be found
1824 * by descending from v->tree. This changes the behavior of freesubre(): it
1825 * will henceforth FREE() unwanted subres rather than sticking them into the
1826 * treefree list. (Doing that any earlier would result in dangling links in
1827 * the treechain list.) This all means that freev() will clean up correctly
1828 * if invoked before or after markst()+cleanst(); but it would not work if
1829 * called partway through this state conversion, so we mustn't error out
1830 * in or between these two functions.
1833 markst(struct subre *t)
1838 if (t->left != NULL)
1840 if (t->right != NULL)
1845 * cleanst - free any tree nodes not marked INUSE
1848 cleanst(struct vars *v)
1853 for (t = v->treechain; t != NULL; t = next)
1856 if (!(t->flags & INUSE))
1859 v->treechain = NULL;
1860 v->treefree = NULL; /* just on general principles */
1864 * nfatree - turn a subRE subtree into a tree of compacted NFAs
1866 static long /* optimize results from top node */
1867 nfatree(struct vars *v,
1869 FILE *f) /* for debug output */
1871 assert(t != NULL && t->begin != NULL);
1873 if (t->left != NULL)
1874 (DISCARD) nfatree(v, t->left, f);
1875 if (t->right != NULL)
1876 (DISCARD) nfatree(v, t->right, f);
1878 return nfanode(v, t, 0, f);
1882 * nfanode - do one NFA for nfatree or lacons
1884 * If converttosearch is true, apply makesearch() to the NFA.
1886 static long /* optimize results */
1887 nfanode(struct vars *v,
1889 int converttosearch,
1890 FILE *f) /* for debug output */
1895 assert(t->begin != NULL);
1902 fprintf(f, "\n\n\n========= TREE NODE %s ==========\n",
1903 stid(t, idbuf, sizeof(idbuf)));
1906 nfa = newnfa(v, v->cm, v->nfa);
1908 dupnfa(nfa, t->begin, t->end, nfa->init, nfa->final);
1912 ret = optimize(nfa, f);
1913 if (converttosearch && !ISERR())
1916 compact(nfa, &t->cnfa);
1923 * newlacon - allocate a lookaround-constraint subRE
1925 static int /* lacon number */
1926 newlacon(struct vars *v,
1927 struct state *begin,
1932 struct subre *newlacons;
1935 if (v->nlacons == 0)
1937 n = 1; /* skip 0th */
1938 newlacons = (struct subre *) MALLOC(2 * sizeof(struct subre));
1943 newlacons = (struct subre *) REALLOC(v->lacons,
1944 (n + 1) * sizeof(struct subre));
1946 if (newlacons == NULL)
1951 v->lacons = newlacons;
1953 sub = &v->lacons[n];
1956 sub->subno = latype;
1962 * freelacons - free lookaround-constraint subRE vector
1965 freelacons(struct subre *subs,
1972 for (sub = subs + 1, i = n - 1; i > 0; sub++, i--) /* no 0th */
1973 if (!NULLCNFA(sub->cnfa))
1974 freecnfa(&sub->cnfa);
1979 * rfree - free a whole RE (insides of regfree)
1986 if (re == NULL || re->re_magic != REMAGIC)
1989 re->re_magic = 0; /* invalidate RE */
1990 g = (struct guts *) re->re_guts;
1997 if (g->tree != NULL)
1998 freesubre((struct vars *) NULL, g->tree);
1999 if (g->lacons != NULL)
2000 freelacons(g->lacons, g->nlacons);
2001 if (!NULLCNFA(g->search))
2002 freecnfa(&g->search);
2008 * rcancelrequested - check for external request to cancel regex operation
2010 * Return nonzero to fail the operation with error code REG_CANCEL,
2011 * zero to keep going
2013 * The current implementation is Postgres-specific. If we ever get around
2014 * to splitting the regex code out as a standalone library, there will need
2015 * to be some API to let applications define a callback function for this.
2018 rcancelrequested(void)
2020 return InterruptPending && (QueryCancelPending || ProcDiePending);
2024 * rstacktoodeep - check for stack getting dangerously deep
2026 * Return nonzero to fail the operation with error code REG_ETOOBIG,
2027 * zero to keep going
2029 * The current implementation is Postgres-specific. If we ever get around
2030 * to splitting the regex code out as a standalone library, there will need
2031 * to be some API to let applications define a callback function for this.
2036 return stack_is_too_deep();
2042 * dump - dump an RE in human-readable form
2051 if (re->re_magic != REMAGIC)
2052 fprintf(f, "bad magic number (0x%x not 0x%x)\n", re->re_magic,
2054 if (re->re_guts == NULL)
2056 fprintf(f, "NULL guts!!!\n");
2059 g = (struct guts *) re->re_guts;
2060 if (g->magic != GUTSMAGIC)
2061 fprintf(f, "bad guts magic number (0x%x not 0x%x)\n", g->magic,
2064 fprintf(f, "\n\n\n========= DUMP ==========\n");
2065 fprintf(f, "nsub %d, info 0%lo, csize %d, ntree %d\n",
2066 (int) re->re_nsub, re->re_info, re->re_csize, g->ntree);
2068 dumpcolors(&g->cmap, f);
2069 if (!NULLCNFA(g->search))
2071 fprintf(f, "\nsearch:\n");
2072 dumpcnfa(&g->search, f);
2074 for (i = 1; i < g->nlacons; i++)
2076 struct subre *lasub = &g->lacons[i];
2079 switch (lasub->subno)
2081 case LATYPE_AHEAD_POS:
2082 latype = "positive lookahead";
2084 case LATYPE_AHEAD_NEG:
2085 latype = "negative lookahead";
2087 case LATYPE_BEHIND_POS:
2088 latype = "positive lookbehind";
2090 case LATYPE_BEHIND_NEG:
2091 latype = "negative lookbehind";
2097 fprintf(f, "\nla%d (%s):\n", i, latype);
2098 dumpcnfa(&lasub->cnfa, f);
2101 dumpst(g->tree, f, 0);
2105 * dumpst - dump a subRE tree
2108 dumpst(struct subre *t,
2110 int nfapresent) /* is the original NFA still around? */
2113 fprintf(f, "null tree\n");
2115 stdump(t, f, nfapresent);
2120 * stdump - recursive guts of dumpst
2123 stdump(struct subre *t,
2125 int nfapresent) /* is the original NFA still around? */
2129 fprintf(f, "%s. `%c'", stid(t, idbuf, sizeof(idbuf)), t->op);
2130 if (t->flags & LONGER)
2131 fprintf(f, " longest");
2132 if (t->flags & SHORTER)
2133 fprintf(f, " shortest");
2134 if (t->flags & MIXED)
2135 fprintf(f, " hasmixed");
2137 fprintf(f, " hascapture");
2138 if (t->flags & BACKR)
2139 fprintf(f, " hasbackref");
2140 if (!(t->flags & INUSE))
2141 fprintf(f, " UNUSED");
2143 fprintf(f, " (#%d)", t->subno);
2144 if (t->min != 1 || t->max != 1)
2146 fprintf(f, " {%d,", t->min);
2147 if (t->max != DUPINF)
2148 fprintf(f, "%d", t->max);
2152 fprintf(f, " %ld-%ld", (long) t->begin->no, (long) t->end->no);
2153 if (t->left != NULL)
2154 fprintf(f, " L:%s", stid(t->left, idbuf, sizeof(idbuf)));
2155 if (t->right != NULL)
2156 fprintf(f, " R:%s", stid(t->right, idbuf, sizeof(idbuf)));
2157 if (!NULLCNFA(t->cnfa))
2160 dumpcnfa(&t->cnfa, f);
2163 if (t->left != NULL)
2164 stdump(t->left, f, nfapresent);
2165 if (t->right != NULL)
2166 stdump(t->right, f, nfapresent);
2170 * stid - identify a subtree node for dumping
2172 static const char * /* points to buf or constant string */
2173 stid(struct subre *t,
2177 /* big enough for hex int or decimal t->id? */
2178 if (bufsize < sizeof(void *) * 2 + 3 || bufsize < sizeof(t->id) * 3 + 1)
2181 sprintf(buf, "%d", t->id);
2183 sprintf(buf, "%p", t);
2186 #endif /* REG_DEBUG */
2189 #include "regc_lex.c"
2190 #include "regc_color.c"
2191 #include "regc_nfa.c"
2192 #include "regc_cvec.c"
2193 #include "regc_pg_locale.c"
2194 #include "regc_locale.c"