Line data Source code
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 *v, int wanted);
42 : static int freev(struct vars *v, int err);
43 : static void makesearch(struct vars *v, struct nfa *nfa);
44 : static struct subre *parse(struct vars *v, int stopper, int type,
45 : struct state *init, struct state *final);
46 : static struct subre *parsebranch(struct vars *v, int stopper, int type,
47 : struct state *left, struct state *right,
48 : int partial);
49 : static struct subre *parseqatom(struct vars *v, int stopper, int type,
50 : struct state *lp, struct state *rp,
51 : struct subre *top);
52 : static void nonword(struct vars *v, int dir, struct state *lp,
53 : struct state *rp);
54 : static void word(struct vars *v, int dir, struct state *lp, struct state *rp);
55 : static void charclass(struct vars *v, enum char_classes cls, struct state *lp,
56 : struct state *rp);
57 : static void charclasscomplement(struct vars *v, enum char_classes cls,
58 : struct state *lp, struct state *rp);
59 : static int scannum(struct vars *v);
60 : static void repeat(struct vars *v, struct state *lp, struct state *rp,
61 : int m, int n);
62 : static void bracket(struct vars *v, struct state *lp, struct state *rp);
63 : static void cbracket(struct vars *v, struct state *lp, struct state *rp);
64 : static void brackpart(struct vars *v, struct state *lp, struct state *rp,
65 : bool *have_cclassc);
66 : static const chr *scanplain(struct vars *v);
67 : static void onechr(struct vars *v, chr c, struct state *lp, struct state *rp);
68 : static void optimizebracket(struct vars *v, struct state *lp, struct state *rp);
69 : static void wordchrs(struct vars *v);
70 : static void processlacon(struct vars *v, struct state *begin,
71 : struct state *end, int latype,
72 : struct state *lp, struct state *rp);
73 : static struct subre *subre(struct vars *v, int op, int flags,
74 : struct state *begin, struct state *end);
75 : static void freesubre(struct vars *v, struct subre *sr);
76 : static void freesubreandsiblings(struct vars *v, struct subre *sr);
77 : static void freesrnode(struct vars *v, struct subre *sr);
78 : static void removecaptures(struct vars *v, struct subre *t);
79 : static int numst(struct subre *t, int start);
80 : static void markst(struct subre *t);
81 : static void cleanst(struct vars *v);
82 : static long nfatree(struct vars *v, struct subre *t, FILE *f);
83 : static long nfanode(struct vars *v, struct subre *t,
84 : int converttosearch, FILE *f);
85 : static int newlacon(struct vars *v, struct state *begin, struct state *end,
86 : int latype);
87 : static void freelacons(struct subre *subs, int n);
88 : static void rfree(regex_t *re);
89 : static int rstacktoodeep(void);
90 :
91 : #ifdef REG_DEBUG
92 : static void dump(regex_t *re, FILE *f);
93 : static void dumpst(struct subre *t, FILE *f, int nfapresent);
94 : static void stdump(struct subre *t, FILE *f, int nfapresent);
95 : static const char *stid(struct subre *t, char *buf, size_t bufsize);
96 : #endif
97 : /* === regc_lex.c === */
98 : static void lexstart(struct vars *v);
99 : static void prefixes(struct vars *v);
100 : static int next(struct vars *v);
101 : static int lexescape(struct vars *v);
102 : static chr lexdigits(struct vars *v, int base, int minlen, int maxlen);
103 : static int brenext(struct vars *v, chr c);
104 : static void skip(struct vars *v);
105 : static chr newline(void);
106 : static chr chrnamed(struct vars *v, const chr *startp, const chr *endp,
107 : chr lastresort);
108 :
109 : /* === regc_color.c === */
110 : static void initcm(struct vars *v, struct colormap *cm);
111 : static void freecm(struct colormap *cm);
112 : static color maxcolor(struct colormap *cm);
113 : static color newcolor(struct colormap *cm);
114 : static void freecolor(struct colormap *cm, color co);
115 : static color pseudocolor(struct colormap *cm);
116 : static color subcolor(struct colormap *cm, chr c);
117 : static color subcolorhi(struct colormap *cm, color *pco);
118 : static color newsub(struct colormap *cm, color co);
119 : static int newhicolorrow(struct colormap *cm, int oldrow);
120 : static void newhicolorcols(struct colormap *cm);
121 : static void subcolorcvec(struct vars *v, struct cvec *cv, struct state *lp,
122 : struct state *rp);
123 : static void subcoloronechr(struct vars *v, chr ch, struct state *lp,
124 : struct state *rp, color *lastsubcolor);
125 : static void subcoloronerange(struct vars *v, chr from, chr to,
126 : struct state *lp, struct state *rp,
127 : color *lastsubcolor);
128 : static void subcoloronerow(struct vars *v, int rownum, struct state *lp,
129 : struct state *rp, color *lastsubcolor);
130 : static void okcolors(struct nfa *nfa, struct colormap *cm);
131 : static void colorchain(struct colormap *cm, struct arc *a);
132 : static void uncolorchain(struct colormap *cm, struct arc *a);
133 : static void rainbow(struct nfa *nfa, struct colormap *cm, int type, color but,
134 : struct state *from, struct state *to);
135 : static void colorcomplement(struct nfa *nfa, struct colormap *cm, int type,
136 : struct state *of, struct state *from,
137 : struct state *to);
138 :
139 : #ifdef REG_DEBUG
140 : static void dumpcolors(struct colormap *cm, FILE *f);
141 : static void dumpchr(chr c, FILE *f);
142 : #endif
143 : /* === regc_nfa.c === */
144 : static struct nfa *newnfa(struct vars *v, struct colormap *cm,
145 : struct nfa *parent);
146 : static void freenfa(struct nfa *nfa);
147 : static struct state *newstate(struct nfa *nfa);
148 : static struct state *newfstate(struct nfa *nfa, int flag);
149 : static void dropstate(struct nfa *nfa, struct state *s);
150 : static void freestate(struct nfa *nfa, struct state *s);
151 : static void newarc(struct nfa *nfa, int t, color co,
152 : struct state *from, struct state *to);
153 : static void createarc(struct nfa *nfa, int t, color co,
154 : struct state *from, struct state *to);
155 : static struct arc *allocarc(struct nfa *nfa);
156 : static void freearc(struct nfa *nfa, struct arc *victim);
157 : static void changearcsource(struct arc *a, struct state *newfrom);
158 : static void changearctarget(struct arc *a, struct state *newto);
159 : static int hasnonemptyout(struct state *s);
160 : static struct arc *findarc(struct state *s, int type, color co);
161 : static void cparc(struct nfa *nfa, struct arc *oa,
162 : struct state *from, struct state *to);
163 : static void sortins(struct nfa *nfa, struct state *s);
164 : static int sortins_cmp(const void *a, const void *b);
165 : static void sortouts(struct nfa *nfa, struct state *s);
166 : static int sortouts_cmp(const void *a, const void *b);
167 : static void moveins(struct nfa *nfa, struct state *oldState,
168 : struct state *newState);
169 : static void copyins(struct nfa *nfa, struct state *oldState,
170 : struct state *newState);
171 : static void mergeins(struct nfa *nfa, struct state *s,
172 : struct arc **arcarray, int arccount);
173 : static void moveouts(struct nfa *nfa, struct state *oldState,
174 : struct state *newState);
175 : static void copyouts(struct nfa *nfa, struct state *oldState,
176 : struct state *newState);
177 : static void cloneouts(struct nfa *nfa, struct state *old, struct state *from,
178 : struct state *to, int type);
179 : static void delsub(struct nfa *nfa, struct state *lp, struct state *rp);
180 : static void deltraverse(struct nfa *nfa, struct state *leftend,
181 : struct state *s);
182 : static void dupnfa(struct nfa *nfa, struct state *start, struct state *stop,
183 : struct state *from, struct state *to);
184 : static void duptraverse(struct nfa *nfa, struct state *s, struct state *stmp);
185 : static void removeconstraints(struct nfa *nfa, struct state *start, struct state *stop);
186 : static void removetraverse(struct nfa *nfa, struct state *s);
187 : static void cleartraverse(struct nfa *nfa, struct state *s);
188 : static struct state *single_color_transition(struct state *s1,
189 : struct state *s2);
190 : static void specialcolors(struct nfa *nfa);
191 : static long optimize(struct nfa *nfa, FILE *f);
192 : static void pullback(struct nfa *nfa, FILE *f);
193 : static int pull(struct nfa *nfa, struct arc *con,
194 : struct state **intermediates);
195 : static void pushfwd(struct nfa *nfa, FILE *f);
196 : static int push(struct nfa *nfa, struct arc *con,
197 : struct state **intermediates);
198 :
199 : #define INCOMPATIBLE 1 /* destroys arc */
200 : #define SATISFIED 2 /* constraint satisfied */
201 : #define COMPATIBLE 3 /* compatible but not satisfied yet */
202 : #define REPLACEARC 4 /* replace arc's color with constraint color */
203 : static int combine(struct nfa *nfa, struct arc *con, struct arc *a);
204 : static void fixempties(struct nfa *nfa, FILE *f);
205 : static struct state *emptyreachable(struct nfa *nfa, struct state *s,
206 : struct state *lastfound,
207 : struct arc **inarcsorig);
208 : static int isconstraintarc(struct arc *a);
209 : static int hasconstraintout(struct state *s);
210 : static void fixconstraintloops(struct nfa *nfa, FILE *f);
211 : static int findconstraintloop(struct nfa *nfa, struct state *s);
212 : static void breakconstraintloop(struct nfa *nfa, struct state *sinitial);
213 : static void clonesuccessorstates(struct nfa *nfa, struct state *ssource,
214 : struct state *sclone,
215 : struct state *spredecessor,
216 : struct arc *refarc, char *curdonemap,
217 : char *outerdonemap, int nstates);
218 : static void removecantmatch(struct nfa *nfa);
219 : static void cleanup(struct nfa *nfa);
220 : static void markreachable(struct nfa *nfa, struct state *s,
221 : struct state *okay, struct state *mark);
222 : static void markcanreach(struct nfa *nfa, struct state *s, struct state *okay,
223 : struct state *mark);
224 : static long analyze(struct nfa *nfa);
225 : static void checkmatchall(struct nfa *nfa);
226 : static bool checkmatchall_recurse(struct nfa *nfa, struct state *s,
227 : bool **haspaths);
228 : static bool check_out_colors_match(struct state *s, color co1, color co2);
229 : static bool check_in_colors_match(struct state *s, color co1, color co2);
230 : static void compact(struct nfa *nfa, struct cnfa *cnfa);
231 : static void carcsort(struct carc *first, size_t n);
232 : static int carc_cmp(const void *a, const void *b);
233 : static void freecnfa(struct cnfa *cnfa);
234 : static void dumpnfa(struct nfa *nfa, FILE *f);
235 :
236 : #ifdef REG_DEBUG
237 : static void dumpstate(struct state *s, FILE *f);
238 : static void dumparcs(struct state *s, FILE *f);
239 : static void dumparc(struct arc *a, struct state *s, FILE *f);
240 : static void dumpcnfa(struct cnfa *cnfa, FILE *f);
241 : static void dumpcstate(int st, struct cnfa *cnfa, FILE *f);
242 : #endif
243 : /* === regc_cvec.c === */
244 : static struct cvec *newcvec(int nchrs, int nranges);
245 : static struct cvec *clearcvec(struct cvec *cv);
246 : static void addchr(struct cvec *cv, chr c);
247 : static void addrange(struct cvec *cv, chr from, chr to);
248 : static struct cvec *getcvec(struct vars *v, int nchrs, int nranges);
249 : static void freecvec(struct cvec *cv);
250 :
251 : /* === regc_pg_locale.c === */
252 : static int regc_wc_isdigit(pg_wchar c);
253 : static int regc_wc_isalpha(pg_wchar c);
254 : static int regc_wc_isalnum(pg_wchar c);
255 : static int regc_wc_isword(pg_wchar c);
256 : static int regc_wc_isupper(pg_wchar c);
257 : static int regc_wc_islower(pg_wchar c);
258 : static int regc_wc_isgraph(pg_wchar c);
259 : static int regc_wc_isprint(pg_wchar c);
260 : static int regc_wc_ispunct(pg_wchar c);
261 : static int regc_wc_isspace(pg_wchar c);
262 : static pg_wchar regc_wc_toupper(pg_wchar c);
263 : static pg_wchar regc_wc_tolower(pg_wchar c);
264 :
265 : /* === regc_locale.c === */
266 : static chr element(struct vars *v, const chr *startp, const chr *endp);
267 : static struct cvec *range(struct vars *v, chr a, chr b, int cases);
268 : static int before(chr x, chr y);
269 : static struct cvec *eclass(struct vars *v, chr c, int cases);
270 : static enum char_classes lookupcclass(struct vars *v, const chr *startp,
271 : const chr *endp);
272 : static struct cvec *cclasscvec(struct vars *v, enum char_classes cclasscode,
273 : int cases);
274 : static int cclass_column_index(struct colormap *cm, chr c);
275 : static struct cvec *allcases(struct vars *v, chr c);
276 : static int cmp(const chr *x, const chr *y, size_t len);
277 : static int casecmp(const chr *x, const chr *y, size_t len);
278 :
279 :
280 : /* internal variables, bundled for easy passing around */
281 : struct vars
282 : {
283 : regex_t *re;
284 : const chr *now; /* scan pointer into string */
285 : const chr *stop; /* end of string */
286 : int err; /* error code (0 if none) */
287 : int cflags; /* copy of compile flags */
288 : int lasttype; /* type of previous token */
289 : int nexttype; /* type of next token */
290 : chr nextvalue; /* value (if any) of next token */
291 : int lexcon; /* lexical context type (see regc_lex.c) */
292 : int nsubexp; /* subexpression count */
293 : struct subre **subs; /* subRE pointer vector */
294 : size_t nsubs; /* length of vector */
295 : struct subre *sub10[10]; /* initial vector, enough for most */
296 : struct nfa *nfa; /* the NFA */
297 : struct colormap *cm; /* character color map */
298 : color nlcolor; /* color of newline */
299 : struct state *wordchrs; /* state in nfa holding word-char outarcs */
300 : struct subre *tree; /* subexpression tree */
301 : struct subre *treechain; /* all tree nodes allocated */
302 : struct subre *treefree; /* any free tree nodes */
303 : int ntree; /* number of tree nodes, plus one */
304 : struct cvec *cv; /* interface cvec */
305 : struct cvec *cv2; /* utility cvec */
306 : struct subre *lacons; /* lookaround-constraint vector */
307 : int nlacons; /* size of lacons[]; note that only slots
308 : * numbered 1 .. nlacons-1 are used */
309 : size_t spaceused; /* approx. space used for compilation */
310 : };
311 :
312 : /* parsing macros; most know that `v' is the struct vars pointer */
313 : #define NEXT() (next(v)) /* advance by one token */
314 : #define SEE(t) (v->nexttype == (t)) /* is next token this? */
315 : #define EAT(t) (SEE(t) && next(v)) /* if next is this, swallow it */
316 : #define VISERR(vv) ((vv)->err != 0) /* have we seen an error yet? */
317 : #define ISERR() VISERR(v)
318 : #define VERR(vv,e) ((vv)->nexttype = EOS, \
319 : (vv)->err = ((vv)->err ? (vv)->err : (e)))
320 : #define ERR(e) VERR(v, e) /* record an error */
321 : #define NOERR() {if (ISERR()) return;} /* if error seen, return */
322 : #define NOERRN() {if (ISERR()) return NULL;} /* NOERR with retval */
323 : #define NOERRZ() {if (ISERR()) return 0;} /* NOERR with retval */
324 : #define INSIST(c, e) do { if (!(c)) ERR(e); } while (0) /* error if c false */
325 : #define NOTE(b) (v->re->re_info |= (b)) /* note visible condition */
326 : #define EMPTYARC(x, y) newarc(v->nfa, EMPTY, 0, x, y)
327 :
328 : /* token type codes, some also used as NFA arc types */
329 : #define EMPTY 'n' /* no token present */
330 : #define EOS 'e' /* end of string */
331 : #define PLAIN 'p' /* ordinary character */
332 : #define DIGIT 'd' /* digit (in bound) */
333 : #define BACKREF 'b' /* back reference */
334 : #define COLLEL 'I' /* start of [. */
335 : #define ECLASS 'E' /* start of [= */
336 : #define CCLASS 'C' /* start of [: */
337 : #define END 'X' /* end of [. [= [: */
338 : #define CCLASSS 's' /* char class shorthand escape */
339 : #define CCLASSC 'c' /* complement char class shorthand escape */
340 : #define RANGE 'R' /* - within [] which might be range delim. */
341 : #define LACON 'L' /* lookaround constraint subRE */
342 : #define AHEAD 'a' /* color-lookahead arc */
343 : #define BEHIND 'r' /* color-lookbehind arc */
344 : #define WBDRY 'w' /* word boundary constraint */
345 : #define NWBDRY 'W' /* non-word-boundary constraint */
346 : #define CANTMATCH 'x' /* arc that cannot match anything */
347 : #define SBEGIN 'A' /* beginning of string (even if not BOL) */
348 : #define SEND 'Z' /* end of string (even if not EOL) */
349 :
350 : /* is an arc colored, and hence should belong to a color chain? */
351 : /* the test on "co" eliminates RAINBOW arcs, which we don't bother to chain */
352 : #define COLORED(a) \
353 : ((a)->co >= 0 && \
354 : ((a)->type == PLAIN || (a)->type == AHEAD || (a)->type == BEHIND))
355 :
356 :
357 : /* static function list */
358 : static const struct fns functions = {
359 : rfree, /* regfree insides */
360 : rstacktoodeep /* check for stack getting dangerously deep */
361 : };
362 :
363 :
364 :
365 : /*
366 : * pg_regcomp - compile regular expression
367 : *
368 : * Note: on failure, no resources remain allocated, so pg_regfree()
369 : * need not be applied to re.
370 : */
371 : int
372 4216 : pg_regcomp(regex_t *re,
373 : const chr *string,
374 : size_t len,
375 : int flags,
376 : Oid collation)
377 : {
378 : struct vars var;
379 4216 : struct vars *v = &var;
380 : struct guts *g;
381 : int i;
382 : size_t j;
383 :
384 : #ifdef REG_DEBUG
385 : FILE *debug = (flags & REG_PROGRESS) ? stdout : (FILE *) NULL;
386 : #else
387 4216 : FILE *debug = (FILE *) NULL;
388 : #endif
389 :
390 : #define CNOERR() { if (ISERR()) return freev(v, v->err); }
391 :
392 : /* sanity checks */
393 :
394 4216 : if (re == NULL || string == NULL)
395 0 : return REG_INVARG;
396 4216 : if ((flags & REG_QUOTE) &&
397 45 : (flags & (REG_ADVANCED | REG_EXPANDED | REG_NEWLINE)))
398 4 : return REG_INVARG;
399 4212 : if (!(flags & REG_EXTENDED) && (flags & REG_ADVF))
400 1 : return REG_INVARG;
401 :
402 : /* Initialize locale-dependent support */
403 4211 : pg_set_regex_collation(collation);
404 :
405 : /* initial setup (after which freev() is callable) */
406 4199 : v->re = re;
407 4199 : v->now = string;
408 4199 : v->stop = v->now + len;
409 4199 : v->err = 0;
410 4199 : v->cflags = flags;
411 4199 : v->nsubexp = 0;
412 4199 : v->subs = v->sub10;
413 4199 : v->nsubs = 10;
414 46189 : for (j = 0; j < v->nsubs; j++)
415 41990 : v->subs[j] = NULL;
416 4199 : v->nfa = NULL;
417 4199 : v->cm = NULL;
418 4199 : v->nlcolor = COLORLESS;
419 4199 : v->wordchrs = NULL;
420 4199 : v->tree = NULL;
421 4199 : v->treechain = NULL;
422 4199 : v->treefree = NULL;
423 4199 : v->cv = NULL;
424 4199 : v->cv2 = NULL;
425 4199 : v->lacons = NULL;
426 4199 : v->nlacons = 0;
427 4199 : v->spaceused = 0;
428 4199 : re->re_magic = REMAGIC;
429 4199 : re->re_info = 0; /* bits get set during parse */
430 4199 : re->re_csize = sizeof(chr);
431 4199 : re->re_collation = collation;
432 4199 : re->re_guts = NULL;
433 4199 : re->re_fns = VS(&functions);
434 :
435 : /* more complex setup, malloced things */
436 4199 : re->re_guts = VS(MALLOC(sizeof(struct guts)));
437 4199 : if (re->re_guts == NULL)
438 0 : return freev(v, REG_ESPACE);
439 4199 : g = (struct guts *) re->re_guts;
440 4199 : g->tree = NULL;
441 4199 : initcm(v, &g->cmap);
442 4199 : v->cm = &g->cmap;
443 4199 : g->lacons = NULL;
444 4199 : g->nlacons = 0;
445 4199 : ZAPCNFA(g->search);
446 4199 : v->nfa = newnfa(v, v->cm, (struct nfa *) NULL);
447 4199 : CNOERR();
448 : /* set up a reasonably-sized transient cvec for getcvec usage */
449 4199 : v->cv = newcvec(100, 20);
450 4199 : if (v->cv == NULL)
451 0 : return freev(v, REG_ESPACE);
452 :
453 : /* parsing */
454 4199 : lexstart(v); /* also handles prefixes */
455 4199 : if ((v->cflags & REG_NLSTOP) || (v->cflags & REG_NLANCH))
456 : {
457 : /* assign newline a unique color */
458 336 : v->nlcolor = subcolor(v->cm, newline());
459 336 : okcolors(v->nfa, v->cm);
460 : }
461 4199 : CNOERR();
462 4192 : v->tree = parse(v, EOS, PLAIN, v->nfa->init, v->nfa->final);
463 : assert(SEE(EOS)); /* even if error; ISERR() => SEE(EOS) */
464 4192 : CNOERR();
465 : assert(v->tree != NULL);
466 :
467 : /* finish setup of nfa and its subre tree */
468 4080 : specialcolors(v->nfa);
469 4080 : CNOERR();
470 : #ifdef REG_DEBUG
471 : if (debug != NULL)
472 : {
473 : fprintf(debug, "\n\n\n========= RAW ==========\n");
474 : dumpnfa(v->nfa, debug);
475 : dumpst(v->tree, debug, 1);
476 : }
477 : #endif
478 4080 : if (v->cflags & REG_NOSUB)
479 3097 : removecaptures(v, v->tree);
480 4080 : v->ntree = numst(v->tree, 1);
481 4080 : markst(v->tree);
482 4080 : cleanst(v);
483 : #ifdef REG_DEBUG
484 : if (debug != NULL)
485 : {
486 : fprintf(debug, "\n\n\n========= TREE FIXED ==========\n");
487 : dumpst(v->tree, debug, 1);
488 : }
489 : #endif
490 :
491 : /* build compacted NFAs for tree and lacons */
492 4080 : re->re_info |= nfatree(v, v->tree, debug);
493 4080 : CNOERR();
494 : assert(v->nlacons == 0 || v->lacons != NULL);
495 4120 : for (i = 1; i < v->nlacons; i++)
496 : {
497 43 : struct subre *lasub = &v->lacons[i];
498 :
499 : #ifdef REG_DEBUG
500 : if (debug != NULL)
501 : fprintf(debug, "\n\n\n========= LA%d ==========\n", i);
502 : #endif
503 :
504 : /* Prepend .* to pattern if it's a lookbehind LACON */
505 43 : nfanode(v, lasub, !LATYPE_IS_AHEAD(lasub->latype), debug);
506 : }
507 4077 : CNOERR();
508 4077 : if (v->tree->flags & SHORTER)
509 70 : NOTE(REG_USHORTEST);
510 :
511 : /* build compacted NFAs for tree, lacons, fast search */
512 : #ifdef REG_DEBUG
513 : if (debug != NULL)
514 : fprintf(debug, "\n\n\n========= SEARCH ==========\n");
515 : #endif
516 : /* can sacrifice main NFA now, so use it as work area */
517 4077 : (DISCARD) optimize(v->nfa, debug);
518 4077 : CNOERR();
519 4077 : makesearch(v, v->nfa);
520 4077 : CNOERR();
521 4077 : compact(v->nfa, &g->search);
522 4077 : CNOERR();
523 :
524 : /* looks okay, package it up */
525 4077 : re->re_nsub = v->nsubexp;
526 4077 : v->re = NULL; /* freev no longer frees re */
527 4077 : g->magic = GUTSMAGIC;
528 4077 : g->cflags = v->cflags;
529 4077 : g->info = re->re_info;
530 4077 : g->nsub = re->re_nsub;
531 4077 : g->tree = v->tree;
532 4077 : v->tree = NULL;
533 4077 : g->ntree = v->ntree;
534 4077 : g->compare = (v->cflags & REG_ICASE) ? casecmp : cmp;
535 4077 : g->lacons = v->lacons;
536 4077 : v->lacons = NULL;
537 4077 : g->nlacons = v->nlacons;
538 :
539 : #ifdef REG_DEBUG
540 : if (flags & REG_DUMP)
541 : {
542 : dump(re, stdout);
543 : fflush(stdout);
544 : }
545 : #endif
546 :
547 : assert(v->err == 0);
548 4077 : return freev(v, 0);
549 : }
550 :
551 : /*
552 : * moresubs - enlarge subRE vector
553 : */
554 : static void
555 12 : moresubs(struct vars *v,
556 : int wanted) /* want enough room for this one */
557 : {
558 : struct subre **p;
559 : size_t n;
560 :
561 : assert(wanted > 0 && (size_t) wanted >= v->nsubs);
562 12 : n = (size_t) wanted * 3 / 2 + 1;
563 :
564 12 : if (v->subs == v->sub10)
565 : {
566 6 : p = (struct subre **) MALLOC(n * sizeof(struct subre *));
567 6 : if (p != NULL)
568 6 : memcpy(VS(p), VS(v->subs),
569 6 : v->nsubs * sizeof(struct subre *));
570 : }
571 : else
572 6 : p = (struct subre **) REALLOC(v->subs, n * sizeof(struct subre *));
573 12 : if (p == NULL)
574 : {
575 0 : ERR(REG_ESPACE);
576 0 : return;
577 : }
578 12 : v->subs = p;
579 142 : for (p = &v->subs[v->nsubs]; v->nsubs < n; p++, v->nsubs++)
580 130 : *p = NULL;
581 : assert(v->nsubs == n);
582 : assert((size_t) wanted < v->nsubs);
583 : }
584 :
585 : /*
586 : * freev - free vars struct's substructures where necessary
587 : *
588 : * Optionally does error-number setting, and always returns error code
589 : * (if any), to make error-handling code terser.
590 : */
591 : static int
592 4199 : freev(struct vars *v,
593 : int err)
594 : {
595 4199 : if (v->re != NULL)
596 122 : rfree(v->re);
597 4199 : if (v->subs != v->sub10)
598 6 : FREE(v->subs);
599 4199 : if (v->nfa != NULL)
600 4199 : freenfa(v->nfa);
601 4199 : if (v->tree != NULL)
602 3 : freesubre(v, v->tree);
603 4199 : if (v->treechain != NULL)
604 112 : cleanst(v);
605 4199 : if (v->cv != NULL)
606 4199 : freecvec(v->cv);
607 4199 : if (v->cv2 != NULL)
608 0 : freecvec(v->cv2);
609 4199 : if (v->lacons != NULL)
610 0 : freelacons(v->lacons, v->nlacons);
611 4199 : ERR(err); /* nop if err==0 */
612 :
613 4199 : return v->err;
614 : }
615 :
616 : /*
617 : * makesearch - turn an NFA into a search NFA (implicit prepend of .*?)
618 : * NFA must have been optimize()d already.
619 : */
620 : static void
621 4086 : makesearch(struct vars *v,
622 : struct nfa *nfa)
623 : {
624 : struct arc *a;
625 : struct arc *b;
626 4086 : struct state *pre = nfa->pre;
627 : struct state *s;
628 : struct state *s2;
629 : struct state *slist;
630 :
631 : /* no loops are needed if it's anchored */
632 11608 : for (a = pre->outs; a != NULL; a = a->outchain)
633 : {
634 : assert(a->type == PLAIN);
635 9083 : if (a->co != nfa->bos[0] && a->co != nfa->bos[1])
636 1561 : break;
637 : }
638 4086 : if (a != NULL)
639 : {
640 : /* add implicit .* in front */
641 1561 : rainbow(nfa, v->cm, PLAIN, COLORLESS, pre, pre);
642 :
643 : /* and ^* and \A* too -- not always necessary, but harmless */
644 1561 : newarc(nfa, PLAIN, nfa->bos[0], pre, pre);
645 1561 : newarc(nfa, PLAIN, nfa->bos[1], pre, pre);
646 :
647 : /*
648 : * The pattern is still MATCHALL if it was before, but the max match
649 : * length is now infinity.
650 : */
651 1561 : if (nfa->flags & MATCHALL)
652 66 : nfa->maxmatchall = DUPINF;
653 : }
654 :
655 : /*
656 : * Now here's the subtle part. Because many REs have no lookback
657 : * constraints, often knowing when you were in the pre state tells you
658 : * little; it's the next state(s) that are informative. But some of them
659 : * may have other inarcs, i.e. it may be possible to make actual progress
660 : * and then return to one of them. We must de-optimize such cases,
661 : * splitting each such state into progress and no-progress states.
662 : */
663 :
664 : /* first, make a list of the states reachable from pre and elsewhere */
665 4086 : slist = NULL;
666 18928 : for (a = pre->outs; a != NULL; a = a->outchain)
667 : {
668 14842 : s = a->to;
669 50407 : for (b = s->ins; b != NULL; b = b->inchain)
670 : {
671 39490 : if (b->from != pre)
672 3925 : break;
673 : }
674 :
675 : /*
676 : * We want to mark states as being in the list already by having non
677 : * NULL tmp fields, but we can't just store the old slist value in tmp
678 : * because that doesn't work for the first such state. Instead, the
679 : * first list entry gets its own address in tmp.
680 : */
681 14842 : if (b != NULL && s->tmp == NULL)
682 : {
683 1489 : s->tmp = (slist != NULL) ? slist : s;
684 1489 : slist = s;
685 : }
686 : }
687 :
688 : /* do the splits */
689 5575 : for (s = slist; s != NULL; s = s2)
690 : {
691 1489 : s2 = newstate(nfa);
692 1489 : NOERR();
693 1489 : copyouts(nfa, s, s2);
694 1489 : NOERR();
695 221018 : for (a = s->ins; a != NULL; a = b)
696 : {
697 219529 : b = a->inchain;
698 219529 : if (a->from != pre)
699 : {
700 215604 : cparc(nfa, a, a->from, s2);
701 215604 : freearc(nfa, a);
702 : }
703 : }
704 1489 : s2 = (s->tmp != s) ? s->tmp : NULL;
705 1489 : s->tmp = NULL; /* clean up while we're at it */
706 : }
707 : }
708 :
709 : /*
710 : * parse - parse an RE
711 : *
712 : * This is actually just the top level, which parses a bunch of branches
713 : * tied together with '|'. If there's more than one, they appear in the
714 : * tree as the children of a '|' subre.
715 : */
716 : static struct subre *
717 7023 : parse(struct vars *v,
718 : int stopper, /* EOS or ')' */
719 : int type, /* LACON (lookaround subRE) or PLAIN */
720 : struct state *init, /* initial state */
721 : struct state *final) /* final state */
722 : {
723 : struct subre *branches; /* top level */
724 : struct subre *lastbranch; /* latest branch */
725 :
726 : assert(stopper == ')' || stopper == EOS);
727 :
728 7023 : branches = subre(v, '|', LONGER, init, final);
729 7023 : NOERRN();
730 7023 : lastbranch = NULL;
731 : do
732 : { /* a branch */
733 : struct subre *branch;
734 : struct state *left; /* scaffolding for branch */
735 : struct state *right;
736 :
737 7337 : left = newstate(v->nfa);
738 7337 : right = newstate(v->nfa);
739 7337 : NOERRN();
740 7337 : EMPTYARC(init, left);
741 7337 : EMPTYARC(right, final);
742 7337 : NOERRN();
743 7337 : branch = parsebranch(v, stopper, type, left, right, 0);
744 7337 : NOERRN();
745 7199 : if (lastbranch)
746 314 : lastbranch->sibling = branch;
747 : else
748 6885 : branches->child = branch;
749 7199 : branches->flags |= UP(branches->flags | branch->flags);
750 7199 : lastbranch = branch;
751 7199 : } while (EAT('|'));
752 : assert(SEE(stopper) || SEE(EOS));
753 :
754 6885 : if (!SEE(stopper))
755 : {
756 : assert(stopper == ')' && SEE(EOS));
757 8 : ERR(REG_EPAREN);
758 : }
759 :
760 : /* optimize out simple cases */
761 6885 : if (lastbranch == branches->child)
762 : { /* only one branch */
763 : assert(lastbranch->sibling == NULL);
764 6698 : freesrnode(v, branches);
765 6698 : branches = lastbranch;
766 : }
767 187 : else if (!MESSY(branches->flags))
768 : { /* no interesting innards */
769 105 : freesubreandsiblings(v, branches->child);
770 105 : branches->child = NULL;
771 105 : branches->op = '=';
772 : }
773 :
774 6885 : return branches;
775 : }
776 :
777 : /*
778 : * parsebranch - parse one branch of an RE
779 : *
780 : * This mostly manages concatenation, working closely with parseqatom().
781 : * Concatenated things are bundled up as much as possible, with separate
782 : * '.' nodes introduced only when necessary due to substructure.
783 : */
784 : static struct subre *
785 9589 : parsebranch(struct vars *v,
786 : int stopper, /* EOS or ')' */
787 : int type, /* LACON (lookaround subRE) or PLAIN */
788 : struct state *left, /* leftmost state */
789 : struct state *right, /* rightmost state */
790 : int partial) /* is this only part of a branch? */
791 : {
792 : struct state *lp; /* left end of current construct */
793 : int seencontent; /* is there anything in this branch yet? */
794 : struct subre *t;
795 :
796 9589 : lp = left;
797 9589 : seencontent = 0;
798 9589 : t = subre(v, '=', 0, left, right); /* op '=' is tentative */
799 9589 : NOERRN();
800 62150 : while (!SEE('|') && !SEE(stopper) && !SEE(EOS))
801 : {
802 52711 : if (seencontent)
803 : { /* implicit concat operator */
804 43228 : lp = newstate(v->nfa);
805 43228 : NOERRN();
806 43228 : moveins(v->nfa, right, lp);
807 : }
808 52711 : seencontent = 1;
809 :
810 : /* NB, recursion in parseqatom() may swallow rest of branch */
811 52711 : t = parseqatom(v, stopper, type, lp, right, t);
812 52711 : NOERRN();
813 : }
814 :
815 9439 : if (!seencontent)
816 : { /* empty branch */
817 106 : if (!partial)
818 106 : NOTE(REG_UUNSPEC);
819 : assert(lp == left);
820 106 : EMPTYARC(left, right);
821 : }
822 :
823 9439 : return t;
824 : }
825 :
826 : /*
827 : * parseqatom - parse one quantified atom or constraint of an RE
828 : *
829 : * The bookkeeping near the end cooperates very closely with parsebranch();
830 : * in particular, it contains a recursion that can involve parsing the rest
831 : * of the branch, making this function's name somewhat inaccurate.
832 : *
833 : * Usually, the return value is just "top", but in some cases where we
834 : * have parsed the rest of the branch, we may deem "top" redundant and
835 : * free it, returning some child subre instead.
836 : */
837 : static struct subre *
838 52711 : parseqatom(struct vars *v,
839 : int stopper, /* EOS or ')' */
840 : int type, /* LACON (lookaround subRE) or PLAIN */
841 : struct state *lp, /* left state to hang it on */
842 : struct state *rp, /* right state to hang it on */
843 : struct subre *top) /* subtree top */
844 : {
845 : struct state *s; /* temporaries for new states */
846 : struct state *s2;
847 :
848 : #define ARCV(t, val) newarc(v->nfa, t, val, lp, rp)
849 : int m,
850 : n;
851 : struct subre *atom; /* atom's subtree */
852 : struct subre *t;
853 : int cap; /* capturing parens? */
854 : int latype; /* lookaround constraint type */
855 : int subno; /* capturing-parens or backref number */
856 : int atomtype;
857 : int qprefer; /* quantifier short/long preference */
858 : int f;
859 : struct subre **atomp; /* where the pointer to atom is */
860 :
861 : /* initial bookkeeping */
862 52711 : atom = NULL;
863 : assert(lp->nouts == 0); /* must string new code */
864 : assert(rp->nins == 0); /* between lp and rp */
865 52711 : subno = 0; /* just to shut lint up */
866 :
867 : /* an atom or constraint... */
868 52711 : atomtype = v->nexttype;
869 52711 : switch (atomtype)
870 : {
871 : /* first, constraints, which end by returning */
872 2797 : case '^':
873 2797 : ARCV('^', 1);
874 2797 : if (v->cflags & REG_NLANCH)
875 257 : ARCV(BEHIND, v->nlcolor);
876 2797 : NEXT();
877 2797 : return top;
878 : break;
879 2300 : case '$':
880 2300 : ARCV('$', 1);
881 2300 : if (v->cflags & REG_NLANCH)
882 245 : ARCV(AHEAD, v->nlcolor);
883 2300 : NEXT();
884 2300 : return top;
885 : break;
886 13 : case SBEGIN:
887 13 : ARCV('^', 1); /* BOL */
888 13 : ARCV('^', 0); /* or BOS */
889 13 : NEXT();
890 13 : return top;
891 : break;
892 5 : case SEND:
893 5 : ARCV('$', 1); /* EOL */
894 5 : ARCV('$', 0); /* or EOS */
895 5 : NEXT();
896 5 : return top;
897 : break;
898 31 : case '<':
899 31 : wordchrs(v);
900 31 : s = newstate(v->nfa);
901 31 : NOERRN();
902 31 : nonword(v, BEHIND, lp, s);
903 31 : word(v, AHEAD, s, rp);
904 31 : NEXT();
905 31 : return top;
906 : break;
907 28 : case '>':
908 28 : wordchrs(v);
909 28 : s = newstate(v->nfa);
910 28 : NOERRN();
911 28 : word(v, BEHIND, lp, s);
912 28 : nonword(v, AHEAD, s, rp);
913 28 : NEXT();
914 28 : return top;
915 : break;
916 9 : case WBDRY:
917 9 : wordchrs(v);
918 9 : s = newstate(v->nfa);
919 9 : NOERRN();
920 9 : nonword(v, BEHIND, lp, s);
921 9 : word(v, AHEAD, s, rp);
922 9 : s = newstate(v->nfa);
923 9 : NOERRN();
924 9 : word(v, BEHIND, lp, s);
925 9 : nonword(v, AHEAD, s, rp);
926 9 : NEXT();
927 9 : return top;
928 : break;
929 19 : case NWBDRY:
930 19 : wordchrs(v);
931 19 : s = newstate(v->nfa);
932 19 : NOERRN();
933 19 : word(v, BEHIND, lp, s);
934 19 : word(v, AHEAD, s, rp);
935 19 : s = newstate(v->nfa);
936 19 : NOERRN();
937 19 : nonword(v, BEHIND, lp, s);
938 19 : nonword(v, AHEAD, s, rp);
939 19 : NEXT();
940 19 : return top;
941 : break;
942 134 : case LACON: /* lookaround constraint */
943 134 : latype = v->nextvalue;
944 134 : NEXT();
945 134 : s = newstate(v->nfa);
946 134 : s2 = newstate(v->nfa);
947 134 : NOERRN();
948 134 : t = parse(v, ')', LACON, s, s2);
949 134 : freesubre(v, t); /* internal structure irrelevant */
950 134 : NOERRN();
951 : assert(SEE(')'));
952 127 : NEXT();
953 127 : processlacon(v, s, s2, latype, lp, rp);
954 127 : return top;
955 : break;
956 : /* then errors, to get them out of the way */
957 34 : case '*':
958 : case '+':
959 : case '?':
960 : case '{':
961 34 : ERR(REG_BADRPT);
962 34 : return top;
963 : break;
964 0 : default:
965 0 : ERR(REG_ASSERT);
966 0 : return top;
967 : break;
968 : /* then plain characters, and minor variants on that theme */
969 3 : case ')': /* unbalanced paren */
970 3 : if ((v->cflags & REG_ADVANCED) != REG_EXTENDED)
971 : {
972 2 : ERR(REG_EPAREN);
973 2 : return top;
974 : }
975 : /* legal in EREs due to specification botch */
976 1 : NOTE(REG_UPBOTCH);
977 : /* fall through into case PLAIN */
978 : pg_fallthrough;
979 41967 : case PLAIN:
980 41967 : onechr(v, v->nextvalue, lp, rp);
981 41967 : okcolors(v->nfa, v->cm);
982 41967 : NOERRN();
983 41967 : NEXT();
984 41967 : break;
985 702 : case '[':
986 702 : if (v->nextvalue == 1)
987 561 : bracket(v, lp, rp);
988 : else
989 141 : cbracket(v, lp, rp);
990 : assert(SEE(']') || ISERR());
991 702 : NEXT();
992 702 : break;
993 209 : case CCLASSS:
994 209 : charclass(v, (enum char_classes) v->nextvalue, lp, rp);
995 209 : okcolors(v->nfa, v->cm);
996 209 : NEXT();
997 209 : break;
998 23 : case CCLASSC:
999 23 : charclasscomplement(v, (enum char_classes) v->nextvalue, lp, rp);
1000 : /* charclasscomplement() did okcolors() internally */
1001 23 : NEXT();
1002 23 : break;
1003 1628 : case '.':
1004 1628 : rainbow(v->nfa, v->cm, PLAIN,
1005 1628 : (v->cflags & REG_NLSTOP) ? v->nlcolor : COLORLESS,
1006 : lp, rp);
1007 1628 : NEXT();
1008 1628 : break;
1009 : /* and finally the ugly stuff */
1010 2697 : case '(': /* value flags as capturing or non */
1011 2697 : cap = (type == LACON) ? 0 : v->nextvalue;
1012 2697 : if (cap)
1013 : {
1014 2545 : v->nsubexp++;
1015 2545 : subno = v->nsubexp;
1016 2545 : if ((size_t) subno >= v->nsubs)
1017 12 : moresubs(v, subno);
1018 : }
1019 : else
1020 152 : atomtype = PLAIN; /* something that's not '(' */
1021 2697 : NEXT();
1022 :
1023 : /*
1024 : * Make separate endpoint states to keep this sub-NFA distinct
1025 : * from what surrounds it. We need to be sure that when we
1026 : * duplicate the sub-NFA for a backref, we get the right
1027 : * states/arcs and no others. In particular, letting a backref
1028 : * duplicate the sub-NFA from lp to rp would be quite wrong,
1029 : * because we may add quantification superstructure around this
1030 : * atom below. (Perhaps we could skip the extra states for
1031 : * non-capturing parens, but it seems not worth the trouble.)
1032 : */
1033 2697 : s = newstate(v->nfa);
1034 2697 : s2 = newstate(v->nfa);
1035 2697 : NOERRN();
1036 : /* We may not need these arcs, but keep things connected for now */
1037 2697 : EMPTYARC(lp, s);
1038 2697 : EMPTYARC(s2, rp);
1039 2697 : NOERRN();
1040 2697 : atom = parse(v, ')', type, s, s2);
1041 : assert(SEE(')') || ISERR());
1042 2697 : NEXT();
1043 2697 : NOERRN();
1044 2670 : if (cap)
1045 : {
1046 2522 : if (atom->capno == 0)
1047 : {
1048 : /* normal case: just mark the atom as capturing */
1049 2494 : atom->flags |= CAP;
1050 2494 : atom->capno = subno;
1051 : }
1052 : else
1053 : {
1054 : /* generate no-op wrapper node to handle "((x))" */
1055 28 : t = subre(v, '(', atom->flags | CAP, s, s2);
1056 28 : NOERRN();
1057 28 : t->capno = subno;
1058 28 : t->child = atom;
1059 28 : atom = t;
1060 : }
1061 : assert(v->subs[subno] == NULL);
1062 2522 : v->subs[subno] = atom;
1063 : }
1064 : /* postpone everything else pending possible {0} */
1065 2670 : break;
1066 113 : case BACKREF: /* the Feature From The Black Lagoon */
1067 113 : INSIST(type != LACON, REG_ESUBREG);
1068 113 : subno = v->nextvalue;
1069 : assert(subno > 0);
1070 113 : INSIST(subno < v->nsubs, REG_ESUBREG);
1071 113 : NOERRN();
1072 106 : INSIST(v->subs[subno] != NULL, REG_ESUBREG);
1073 106 : NOERRN();
1074 101 : atom = subre(v, 'b', BACKR, lp, rp);
1075 101 : NOERRN();
1076 101 : atom->backno = subno;
1077 101 : v->subs[subno]->flags |= BRUSE;
1078 101 : EMPTYARC(lp, rp); /* temporarily, so there's something */
1079 101 : NEXT();
1080 101 : break;
1081 : }
1082 :
1083 : /* ...and an atom may be followed by a quantifier */
1084 47300 : switch (v->nexttype)
1085 : {
1086 10666 : case '*':
1087 10666 : m = 0;
1088 10666 : n = DUPINF;
1089 10666 : qprefer = (v->nextvalue) ? LONGER : SHORTER;
1090 10666 : NEXT();
1091 10666 : break;
1092 473 : case '+':
1093 473 : m = 1;
1094 473 : n = DUPINF;
1095 473 : qprefer = (v->nextvalue) ? LONGER : SHORTER;
1096 473 : NEXT();
1097 473 : break;
1098 65 : case '?':
1099 65 : m = 0;
1100 65 : n = 1;
1101 65 : qprefer = (v->nextvalue) ? LONGER : SHORTER;
1102 65 : NEXT();
1103 65 : break;
1104 262 : case '{':
1105 262 : NEXT();
1106 262 : m = scannum(v);
1107 262 : if (EAT(','))
1108 : {
1109 122 : if (SEE(DIGIT))
1110 114 : n = scannum(v);
1111 : else
1112 8 : n = DUPINF;
1113 122 : if (m > n)
1114 : {
1115 4 : ERR(REG_BADBR);
1116 4 : return top;
1117 : }
1118 : /* {m,n} exercises preference, even if it's {m,m} */
1119 118 : qprefer = (v->nextvalue) ? LONGER : SHORTER;
1120 : }
1121 : else
1122 : {
1123 140 : n = m;
1124 : /* {m} passes operand's preference through */
1125 140 : qprefer = 0;
1126 : }
1127 258 : if (!SEE('}'))
1128 : { /* catches errors too */
1129 7 : ERR(REG_BADBR);
1130 7 : return top;
1131 : }
1132 251 : NEXT();
1133 251 : break;
1134 35834 : default: /* no quantifier */
1135 35834 : m = n = 1;
1136 35834 : qprefer = 0;
1137 35834 : break;
1138 : }
1139 :
1140 : /* annoying special case: {0} or {0,0} cancels everything */
1141 47289 : if (m == 0 && n == 0)
1142 : {
1143 : /*
1144 : * If we had capturing subexpression(s) within the atom, we don't want
1145 : * to destroy them, because it's legal (if useless) to back-ref them
1146 : * later. Hence, just unlink the atom from lp/rp and then ignore it.
1147 : */
1148 21 : if (atom != NULL && (atom->flags & CAP))
1149 : {
1150 18 : delsub(v->nfa, lp, atom->begin);
1151 18 : delsub(v->nfa, atom->end, rp);
1152 : }
1153 : else
1154 : {
1155 : /* Otherwise, we can clean up any subre infrastructure we made */
1156 3 : if (atom != NULL)
1157 1 : freesubre(v, atom);
1158 3 : delsub(v->nfa, lp, rp);
1159 : }
1160 21 : EMPTYARC(lp, rp);
1161 21 : return top;
1162 : }
1163 :
1164 : /* if not a messy case, avoid hard part */
1165 : assert(!MESSY(top->flags));
1166 47268 : f = top->flags | qprefer | ((atom != NULL) ? atom->flags : 0);
1167 47268 : if (atomtype != '(' && atomtype != BACKREF && !MESSY(UP(f)))
1168 : {
1169 44647 : if (!(m == 1 && n == 1))
1170 11096 : repeat(v, lp, rp, m, n);
1171 44647 : if (atom != NULL)
1172 129 : freesubre(v, atom);
1173 44647 : top->flags = f;
1174 44647 : return top;
1175 : }
1176 :
1177 : /*
1178 : * hard part: something messy
1179 : *
1180 : * That is, capturing parens, back reference, short/long clash, or an atom
1181 : * with substructure containing one of those.
1182 : */
1183 :
1184 : /* now we'll need a subre for the contents even if they're boring */
1185 2621 : if (atom == NULL)
1186 : {
1187 1 : atom = subre(v, '=', 0, lp, rp);
1188 1 : NOERRN();
1189 : }
1190 :
1191 : /*
1192 : * For what follows, we need the atom to have its own begin/end states
1193 : * that are distinct from lp/rp, so that we can wrap iteration structure
1194 : * around it. The parenthesized-atom case above already made suitable
1195 : * states (and we don't want to modify a capturing subre, since it's
1196 : * already recorded in v->subs[]). Otherwise, we need more states.
1197 : */
1198 2621 : if (atom->begin == lp || atom->end == rp)
1199 : {
1200 102 : s = newstate(v->nfa);
1201 102 : s2 = newstate(v->nfa);
1202 102 : NOERRN();
1203 102 : moveouts(v->nfa, lp, s);
1204 102 : moveins(v->nfa, rp, s2);
1205 102 : atom->begin = s;
1206 102 : atom->end = s2;
1207 : }
1208 : else
1209 : {
1210 : /* The atom's OK, but we must temporarily disconnect it from lp/rp */
1211 : /* (this removes the EMPTY arcs we made above) */
1212 2519 : delsub(v->nfa, lp, atom->begin);
1213 2519 : delsub(v->nfa, atom->end, rp);
1214 : }
1215 :
1216 : /*----------
1217 : * Prepare a general-purpose state skeleton.
1218 : *
1219 : * In the no-backrefs case, we want this:
1220 : *
1221 : * [lp] ---> [s] ---prefix---> ---atom---> ---rest---> [rp]
1222 : *
1223 : * where prefix is some repetitions of atom, and "rest" is the remainder
1224 : * of the branch. In the general case we need:
1225 : *
1226 : * [lp] ---> [s] ---iterator---> [s2] ---rest---> [rp]
1227 : *
1228 : * where the iterator wraps around the atom.
1229 : *
1230 : * We make the s state here for both cases; s2 is made below if needed
1231 : *----------
1232 : */
1233 2621 : s = newstate(v->nfa); /* set up starting state */
1234 2621 : NOERRN();
1235 2621 : EMPTYARC(lp, s);
1236 2621 : NOERRN();
1237 :
1238 : /* break remaining subRE into x{...} and what follows */
1239 2621 : t = subre(v, '.', COMBINE(qprefer, atom->flags), lp, rp);
1240 2621 : NOERRN();
1241 2621 : t->child = atom;
1242 2621 : atomp = &t->child;
1243 :
1244 : /*
1245 : * Here we should recurse to fill t->child->sibling ... but we must
1246 : * postpone that to the end. One reason is that t->child may be replaced
1247 : * below, and we don't want to worry about its sibling link.
1248 : */
1249 :
1250 : /*
1251 : * Convert top node to a concatenation of the prefix (top->child, covering
1252 : * whatever we parsed previously) and remaining (t). Note that the prefix
1253 : * could be empty, in which case this concatenation node is unnecessary.
1254 : * To keep things simple, we operate in a general way for now, and get rid
1255 : * of unnecessary subres below.
1256 : */
1257 : assert(top->op == '=' && top->child == NULL);
1258 2621 : top->child = subre(v, '=', top->flags, top->begin, lp);
1259 2621 : NOERRN();
1260 2621 : top->op = '.';
1261 2621 : top->child->sibling = t;
1262 : /* top->flags will get updated later */
1263 :
1264 : /* if it's a backref, now is the time to replicate the subNFA */
1265 2621 : if (atomtype == BACKREF)
1266 : {
1267 : assert(atom->begin->nouts == 1); /* just the EMPTY */
1268 101 : delsub(v->nfa, atom->begin, atom->end);
1269 : assert(v->subs[subno] != NULL);
1270 :
1271 : /*
1272 : * And here's why the recursion got postponed: it must wait until the
1273 : * skeleton is filled in, because it may hit a backref that wants to
1274 : * copy the filled-in skeleton.
1275 : */
1276 101 : dupnfa(v->nfa, v->subs[subno]->begin, v->subs[subno]->end,
1277 : atom->begin, atom->end);
1278 101 : NOERRN();
1279 :
1280 : /* The backref node's NFA should not enforce any constraints */
1281 101 : removeconstraints(v->nfa, atom->begin, atom->end);
1282 101 : NOERRN();
1283 : }
1284 :
1285 : /*
1286 : * It's quantifier time. If the atom is just a backref, we'll let it deal
1287 : * with quantifiers internally.
1288 : */
1289 2621 : if (atomtype == BACKREF)
1290 : {
1291 : /* special case: backrefs have internal quantifiers */
1292 101 : EMPTYARC(s, atom->begin); /* empty prefix */
1293 : /* just stuff everything into atom */
1294 101 : repeat(v, atom->begin, atom->end, m, n);
1295 101 : atom->min = (short) m;
1296 101 : atom->max = (short) n;
1297 101 : atom->flags |= COMBINE(qprefer, atom->flags);
1298 : /* rest of branch can be strung starting from atom->end */
1299 101 : s2 = atom->end;
1300 : }
1301 2520 : else if (m == 1 && n == 1 &&
1302 53 : (qprefer == 0 ||
1303 53 : (atom->flags & (LONGER | SHORTER | MIXED)) == 0 ||
1304 46 : qprefer == (atom->flags & (LONGER | SHORTER | MIXED))))
1305 : {
1306 : /* no/vacuous quantifier: done */
1307 2293 : EMPTYARC(s, atom->begin); /* empty prefix */
1308 : /* rest of branch can be strung starting from atom->end */
1309 2293 : s2 = atom->end;
1310 : }
1311 227 : else if (!(atom->flags & (CAP | BACKR)))
1312 : {
1313 : /*
1314 : * If there's no captures nor backrefs in the atom being repeated, we
1315 : * don't really care where the submatches of the iteration are, so we
1316 : * don't need an iteration node. Make a plain DFA node instead.
1317 : */
1318 8 : EMPTYARC(s, atom->begin); /* empty prefix */
1319 8 : repeat(v, atom->begin, atom->end, m, n);
1320 8 : f = COMBINE(qprefer, atom->flags);
1321 8 : t = subre(v, '=', f, atom->begin, atom->end);
1322 8 : NOERRN();
1323 8 : freesubre(v, atom);
1324 8 : *atomp = t;
1325 : /* rest of branch can be strung starting from t->end */
1326 8 : s2 = t->end;
1327 : }
1328 219 : else if (m > 0 && !(atom->flags & BACKR))
1329 : {
1330 : /*
1331 : * If there's no backrefs involved, we can turn x{m,n} into
1332 : * x{m-1,n-1}x, with capturing parens in only the second x. This is
1333 : * valid because we only care about capturing matches from the final
1334 : * iteration of the quantifier. It's a win because we can implement
1335 : * the backref-free left side as a plain DFA node, since we don't
1336 : * really care where its submatches are.
1337 : */
1338 136 : dupnfa(v->nfa, atom->begin, atom->end, s, atom->begin);
1339 : assert(m >= 1 && m != DUPINF && n >= 1);
1340 136 : repeat(v, s, atom->begin, m - 1, (n == DUPINF) ? n : n - 1);
1341 136 : f = COMBINE(qprefer, atom->flags);
1342 136 : t = subre(v, '.', f, s, atom->end); /* prefix and atom */
1343 136 : NOERRN();
1344 136 : t->child = subre(v, '=', PREF(f), s, atom->begin);
1345 136 : NOERRN();
1346 136 : t->child->sibling = atom;
1347 136 : *atomp = t;
1348 : /* rest of branch can be strung starting from atom->end */
1349 136 : s2 = atom->end;
1350 : }
1351 : else
1352 : {
1353 : /* general case: need an iteration node */
1354 83 : s2 = newstate(v->nfa);
1355 83 : NOERRN();
1356 83 : moveouts(v->nfa, atom->end, s2);
1357 83 : NOERRN();
1358 83 : dupnfa(v->nfa, atom->begin, atom->end, s, s2);
1359 83 : repeat(v, s, s2, m, n);
1360 83 : f = COMBINE(qprefer, atom->flags);
1361 83 : t = subre(v, '*', f, s, s2);
1362 83 : NOERRN();
1363 83 : t->min = (short) m;
1364 83 : t->max = (short) n;
1365 83 : t->child = atom;
1366 83 : *atomp = t;
1367 : /* rest of branch is to be strung from iteration's end state */
1368 : }
1369 :
1370 : /* and finally, look after that postponed recursion */
1371 2621 : t = top->child->sibling;
1372 2621 : if (!(SEE('|') || SEE(stopper) || SEE(EOS)))
1373 : {
1374 : /* parse all the rest of the branch, and insert in t->child->sibling */
1375 2252 : t->child->sibling = parsebranch(v, stopper, type, s2, rp, 1);
1376 2252 : NOERRN();
1377 : assert(SEE('|') || SEE(stopper) || SEE(EOS));
1378 :
1379 : /* here's the promised update of the flags */
1380 2240 : t->flags |= COMBINE(t->flags, t->child->sibling->flags);
1381 2240 : top->flags |= COMBINE(top->flags, t->flags);
1382 :
1383 : /* neither t nor top could be directly marked for capture as yet */
1384 : assert(t->capno == 0);
1385 : assert(top->capno == 0);
1386 :
1387 : /*
1388 : * At this point both top and t are concatenation (op == '.') subres,
1389 : * and we have top->child = prefix of branch, top->child->sibling = t,
1390 : * t->child = messy atom (with quantification superstructure if
1391 : * needed), t->child->sibling = rest of branch.
1392 : *
1393 : * If the messy atom was the first thing in the branch, then
1394 : * top->child is vacuous and we can get rid of one level of
1395 : * concatenation.
1396 : */
1397 : assert(top->child->op == '=');
1398 2240 : if (top->child->begin == top->child->end)
1399 : {
1400 : assert(!MESSY(top->child->flags));
1401 333 : freesubre(v, top->child);
1402 333 : top->child = t->child;
1403 333 : freesrnode(v, t);
1404 : }
1405 :
1406 : /*
1407 : * Otherwise, it's possible that t->child is not messy in itself, but
1408 : * we considered it messy because its greediness conflicts with what
1409 : * preceded it. Then it could be that the combination of t->child and
1410 : * the rest of the branch is also not messy, in which case we can get
1411 : * rid of the child concatenation by merging t->child and the rest of
1412 : * the branch into one plain DFA node.
1413 : */
1414 1907 : else if (t->child->op == '=' &&
1415 1855 : t->child->sibling->op == '=' &&
1416 1754 : !MESSY(UP(t->child->flags | t->child->sibling->flags)))
1417 : {
1418 0 : t->op = '=';
1419 0 : t->flags = COMBINE(t->child->flags, t->child->sibling->flags);
1420 0 : freesubreandsiblings(v, t->child);
1421 0 : t->child = NULL;
1422 : }
1423 : }
1424 : else
1425 : {
1426 : /*
1427 : * There's nothing left in the branch, so we don't need the second
1428 : * concatenation node 't'. Just link s2 straight to rp.
1429 : */
1430 369 : EMPTYARC(s2, rp);
1431 369 : top->child->sibling = t->child;
1432 369 : top->flags |= COMBINE(top->flags, top->child->sibling->flags);
1433 369 : freesrnode(v, t);
1434 :
1435 : /*
1436 : * Again, it could be that top->child is vacuous (if the messy atom
1437 : * was in fact the only thing in the branch). In that case we need no
1438 : * concatenation at all; just replace top with top->child->sibling.
1439 : */
1440 : assert(top->child->op == '=');
1441 369 : if (top->child->begin == top->child->end)
1442 : {
1443 : assert(!MESSY(top->child->flags));
1444 260 : t = top->child->sibling;
1445 260 : top->child->sibling = NULL;
1446 260 : freesubre(v, top);
1447 260 : top = t;
1448 : }
1449 : }
1450 :
1451 2609 : return top;
1452 : }
1453 :
1454 : /*
1455 : * nonword - generate arcs for non-word-character ahead or behind
1456 : */
1457 : static void
1458 115 : nonword(struct vars *v,
1459 : int dir, /* AHEAD or BEHIND */
1460 : struct state *lp,
1461 : struct state *rp)
1462 : {
1463 115 : int anchor = (dir == AHEAD) ? '$' : '^';
1464 :
1465 : assert(dir == AHEAD || dir == BEHIND);
1466 115 : newarc(v->nfa, anchor, 1, lp, rp);
1467 115 : newarc(v->nfa, anchor, 0, lp, rp);
1468 115 : colorcomplement(v->nfa, v->cm, dir, v->wordchrs, lp, rp);
1469 : /* (no need for special attention to \n) */
1470 115 : }
1471 :
1472 : /*
1473 : * word - generate arcs for word character ahead or behind
1474 : */
1475 : static void
1476 115 : word(struct vars *v,
1477 : int dir, /* AHEAD or BEHIND */
1478 : struct state *lp,
1479 : struct state *rp)
1480 : {
1481 : assert(dir == AHEAD || dir == BEHIND);
1482 115 : cloneouts(v->nfa, v->wordchrs, lp, rp, dir);
1483 : /* (no need for special attention to \n) */
1484 115 : }
1485 :
1486 : /*
1487 : * charclass - generate arcs for a character class
1488 : *
1489 : * This is used for both atoms (\w and sibling escapes) and for elements
1490 : * of bracket expressions. The caller is responsible for calling okcolors()
1491 : * at the end of processing the atom or bracket.
1492 : */
1493 : static void
1494 367 : charclass(struct vars *v,
1495 : enum char_classes cls,
1496 : struct state *lp,
1497 : struct state *rp)
1498 : {
1499 : struct cvec *cv;
1500 :
1501 : /* obtain possibly-cached cvec for char class */
1502 367 : NOTE(REG_ULOCALE);
1503 367 : cv = cclasscvec(v, cls, (v->cflags & REG_ICASE));
1504 367 : NOERR();
1505 :
1506 : /* build the arcs; this may cause color splitting */
1507 367 : subcolorcvec(v, cv, lp, rp);
1508 : }
1509 :
1510 : /*
1511 : * charclasscomplement - generate arcs for a complemented character class
1512 : *
1513 : * This is used for both atoms (\W and sibling escapes) and for elements
1514 : * of bracket expressions. In bracket expressions, it is the caller's
1515 : * responsibility that there not be any open subcolors when this is called.
1516 : */
1517 : static void
1518 39 : charclasscomplement(struct vars *v,
1519 : enum char_classes cls,
1520 : struct state *lp,
1521 : struct state *rp)
1522 : {
1523 : struct state *cstate;
1524 : struct cvec *cv;
1525 :
1526 : /* make dummy state to hang temporary arcs on */
1527 39 : cstate = newstate(v->nfa);
1528 39 : NOERR();
1529 :
1530 : /* obtain possibly-cached cvec for char class */
1531 39 : NOTE(REG_ULOCALE);
1532 39 : cv = cclasscvec(v, cls, (v->cflags & REG_ICASE));
1533 39 : NOERR();
1534 :
1535 : /* build arcs for char class; this may cause color splitting */
1536 39 : subcolorcvec(v, cv, cstate, cstate);
1537 39 : NOERR();
1538 :
1539 : /* clean up any subcolors in the arc set */
1540 39 : okcolors(v->nfa, v->cm);
1541 39 : NOERR();
1542 :
1543 : /* now build output arcs for the complement of the char class */
1544 39 : colorcomplement(v->nfa, v->cm, PLAIN, cstate, lp, rp);
1545 39 : NOERR();
1546 :
1547 : /* clean up dummy state */
1548 39 : dropstate(v->nfa, cstate);
1549 : }
1550 :
1551 : /*
1552 : * scannum - scan a number
1553 : */
1554 : static int /* value, <= DUPMAX */
1555 376 : scannum(struct vars *v)
1556 : {
1557 376 : int n = 0;
1558 :
1559 795 : while (SEE(DIGIT) && n < DUPMAX)
1560 : {
1561 419 : n = n * 10 + v->nextvalue;
1562 419 : NEXT();
1563 : }
1564 376 : if (SEE(DIGIT) || n > DUPMAX)
1565 : {
1566 2 : ERR(REG_BADBR);
1567 2 : return 0;
1568 : }
1569 374 : return n;
1570 : }
1571 :
1572 : /*
1573 : * repeat - replicate subNFA for quantifiers
1574 : *
1575 : * The sub-NFA strung from lp to rp is modified to represent m to n
1576 : * repetitions of its initial contents.
1577 : *
1578 : * The duplication sequences used here are chosen carefully so that any
1579 : * pointers starting out pointing into the subexpression end up pointing into
1580 : * the last occurrence. (Note that it may not be strung between the same
1581 : * left and right end states, however!) This used to be important for the
1582 : * subRE tree, although the important bits are now handled by the in-line
1583 : * code in parse(), and when this is called, it doesn't matter any more.
1584 : */
1585 : static void
1586 12703 : repeat(struct vars *v,
1587 : struct state *lp,
1588 : struct state *rp,
1589 : int m,
1590 : int n)
1591 : {
1592 : #define SOME 2
1593 : #define INF 3
1594 : #define PAIR(x, y) ((x)*4 + (y))
1595 : #define REDUCE(x) ( ((x) == DUPINF) ? INF : (((x) > 1) ? SOME : (x)) )
1596 12703 : const int rm = REDUCE(m);
1597 12703 : const int rn = REDUCE(n);
1598 : struct state *s;
1599 : struct state *s2;
1600 :
1601 12703 : switch (PAIR(rm, rn))
1602 : {
1603 15 : case PAIR(0, 0): /* empty string */
1604 15 : delsub(v->nfa, lp, rp);
1605 15 : EMPTYARC(lp, rp);
1606 15 : break;
1607 69 : case PAIR(0, 1): /* do as x| */
1608 69 : EMPTYARC(lp, rp);
1609 69 : break;
1610 2 : case PAIR(0, SOME): /* do as x{1,n}| */
1611 2 : repeat(v, lp, rp, 1, n);
1612 2 : NOERR();
1613 2 : EMPTYARC(lp, rp);
1614 2 : break;
1615 10781 : case PAIR(0, INF): /* loop x around */
1616 10781 : s = newstate(v->nfa);
1617 10781 : NOERR();
1618 10781 : moveouts(v->nfa, lp, s);
1619 10781 : moveins(v->nfa, rp, s);
1620 10781 : EMPTYARC(lp, s);
1621 10781 : EMPTYARC(s, rp);
1622 10781 : break;
1623 193 : case PAIR(1, 1): /* no action required */
1624 193 : break;
1625 349 : case PAIR(1, SOME): /* do as x{0,n-1}x = (x{1,n-1}|)x */
1626 349 : s = newstate(v->nfa);
1627 349 : NOERR();
1628 349 : moveouts(v->nfa, lp, s);
1629 349 : dupnfa(v->nfa, s, rp, lp, s);
1630 349 : NOERR();
1631 349 : repeat(v, lp, s, 1, n - 1);
1632 349 : NOERR();
1633 349 : EMPTYARC(lp, s);
1634 349 : break;
1635 366 : case PAIR(1, INF): /* add loopback arc */
1636 366 : s = newstate(v->nfa);
1637 366 : s2 = newstate(v->nfa);
1638 366 : NOERR();
1639 366 : moveouts(v->nfa, lp, s);
1640 366 : moveins(v->nfa, rp, s2);
1641 366 : EMPTYARC(lp, s);
1642 366 : EMPTYARC(s2, rp);
1643 366 : EMPTYARC(s2, s);
1644 366 : break;
1645 790 : case PAIR(SOME, SOME): /* do as x{m-1,n-1}x */
1646 790 : s = newstate(v->nfa);
1647 790 : NOERR();
1648 790 : moveouts(v->nfa, lp, s);
1649 790 : dupnfa(v->nfa, s, rp, lp, s);
1650 790 : NOERR();
1651 790 : repeat(v, lp, s, m - 1, n - 1);
1652 790 : break;
1653 138 : case PAIR(SOME, INF): /* do as x{m-1,}x */
1654 138 : s = newstate(v->nfa);
1655 138 : NOERR();
1656 138 : moveouts(v->nfa, lp, s);
1657 138 : dupnfa(v->nfa, s, rp, lp, s);
1658 138 : NOERR();
1659 138 : repeat(v, lp, s, m - 1, n);
1660 138 : break;
1661 0 : default:
1662 0 : ERR(REG_ASSERT);
1663 0 : break;
1664 : }
1665 : }
1666 :
1667 : /*
1668 : * bracket - handle non-complemented bracket expression
1669 : *
1670 : * Also called from cbracket for complemented bracket expressions.
1671 : */
1672 : static void
1673 702 : bracket(struct vars *v,
1674 : struct state *lp,
1675 : struct state *rp)
1676 : {
1677 : /*
1678 : * We can't process complemented char classes (e.g. \W) immediately while
1679 : * scanning the bracket expression, else color bookkeeping gets confused.
1680 : * Instead, remember whether we saw any in have_cclassc[], and process
1681 : * them at the end.
1682 : */
1683 : bool have_cclassc[NUM_CCLASSES];
1684 : bool any_cclassc;
1685 : int i;
1686 :
1687 702 : memset(have_cclassc, false, sizeof(have_cclassc));
1688 :
1689 : assert(SEE('['));
1690 702 : NEXT();
1691 1788 : while (!SEE(']') && !SEE(EOS))
1692 1086 : brackpart(v, lp, rp, have_cclassc);
1693 : assert(SEE(']') || ISERR());
1694 :
1695 : /* close up open subcolors from the positive bracket elements */
1696 702 : okcolors(v->nfa, v->cm);
1697 702 : NOERR();
1698 :
1699 : /* now handle any complemented elements */
1700 667 : any_cclassc = false;
1701 10005 : for (i = 0; i < NUM_CCLASSES; i++)
1702 : {
1703 9338 : if (have_cclassc[i])
1704 : {
1705 16 : charclasscomplement(v, (enum char_classes) i, lp, rp);
1706 16 : NOERR();
1707 16 : any_cclassc = true;
1708 : }
1709 : }
1710 :
1711 : /*
1712 : * If we had any complemented elements, see if we can optimize the bracket
1713 : * into a rainbow. Since a complemented element is the only way a WHITE
1714 : * arc could get into the result, there's no point in checking otherwise.
1715 : */
1716 667 : if (any_cclassc)
1717 16 : optimizebracket(v, lp, rp);
1718 : }
1719 :
1720 : /*
1721 : * cbracket - handle complemented bracket expression
1722 : *
1723 : * We do it by calling bracket() with dummy endpoints, and then complementing
1724 : * the result. The alternative would be to invoke rainbow(), and then delete
1725 : * arcs as the b.e. is seen... but that gets messy, and is really quite
1726 : * infeasible now that rainbow() just puts out one RAINBOW arc.
1727 : */
1728 : static void
1729 141 : cbracket(struct vars *v,
1730 : struct state *lp,
1731 : struct state *rp)
1732 : {
1733 141 : struct state *left = newstate(v->nfa);
1734 141 : struct state *right = newstate(v->nfa);
1735 :
1736 141 : NOERR();
1737 141 : bracket(v, left, right);
1738 :
1739 : /* in NLSTOP mode, ensure newline is not part of the result set */
1740 141 : if (v->cflags & REG_NLSTOP)
1741 2 : newarc(v->nfa, PLAIN, v->nlcolor, left, right);
1742 141 : NOERR();
1743 :
1744 : assert(lp->nouts == 0); /* all outarcs will be ours */
1745 :
1746 : /*
1747 : * Easy part of complementing, and all there is to do since the MCCE code
1748 : * was removed. Note that the result of colorcomplement() cannot be a
1749 : * rainbow, since we don't allow empty brackets; so there's no point in
1750 : * calling optimizebracket() again.
1751 : */
1752 141 : colorcomplement(v->nfa, v->cm, PLAIN, left, lp, rp);
1753 141 : NOERR();
1754 141 : dropstate(v->nfa, left);
1755 : assert(right->nins == 0);
1756 141 : freestate(v->nfa, right);
1757 : }
1758 :
1759 : /*
1760 : * brackpart - handle one item (or range) within a bracket expression
1761 : */
1762 : static void
1763 1086 : brackpart(struct vars *v,
1764 : struct state *lp,
1765 : struct state *rp,
1766 : bool *have_cclassc)
1767 : {
1768 : chr startc;
1769 : chr endc;
1770 : struct cvec *cv;
1771 : enum char_classes cls;
1772 : const chr *startp;
1773 : const chr *endp;
1774 :
1775 : /* parse something, get rid of special cases, take shortcuts */
1776 1086 : switch (v->nexttype)
1777 : {
1778 4 : case RANGE: /* a-b-c or other botch */
1779 4 : ERR(REG_ERANGE);
1780 4 : return;
1781 : break;
1782 878 : case PLAIN:
1783 878 : startc = v->nextvalue;
1784 878 : NEXT();
1785 : /* shortcut for ordinary chr (not range) */
1786 878 : if (!SEE(RANGE))
1787 : {
1788 565 : onechr(v, startc, lp, rp);
1789 565 : return;
1790 : }
1791 313 : NOERR();
1792 313 : break;
1793 10 : case COLLEL:
1794 10 : startp = v->now;
1795 10 : endp = scanplain(v);
1796 10 : INSIST(startp < endp, REG_ECOLLATE);
1797 10 : NOERR();
1798 8 : startc = element(v, startp, endp);
1799 8 : NOERR();
1800 6 : break;
1801 14 : case ECLASS:
1802 14 : startp = v->now;
1803 14 : endp = scanplain(v);
1804 14 : INSIST(startp < endp, REG_ECOLLATE);
1805 14 : NOERR();
1806 12 : startc = element(v, startp, endp);
1807 12 : NOERR();
1808 10 : cv = eclass(v, startc, (v->cflags & REG_ICASE));
1809 10 : NOERR();
1810 10 : subcolorcvec(v, cv, lp, rp);
1811 10 : return;
1812 : break;
1813 151 : case CCLASS:
1814 151 : startp = v->now;
1815 151 : endp = scanplain(v);
1816 151 : INSIST(startp < endp, REG_ECTYPE);
1817 151 : NOERR();
1818 149 : cls = lookupcclass(v, startp, endp);
1819 149 : NOERR();
1820 145 : charclass(v, cls, lp, rp);
1821 145 : return;
1822 : break;
1823 13 : case CCLASSS:
1824 13 : charclass(v, (enum char_classes) v->nextvalue, lp, rp);
1825 13 : NEXT();
1826 13 : return;
1827 : break;
1828 16 : case CCLASSC:
1829 : /* we cannot call charclasscomplement() immediately */
1830 16 : have_cclassc[v->nextvalue] = true;
1831 16 : NEXT();
1832 16 : return;
1833 : break;
1834 0 : default:
1835 0 : ERR(REG_ASSERT);
1836 0 : return;
1837 : break;
1838 : }
1839 :
1840 319 : if (SEE(RANGE))
1841 : {
1842 315 : NEXT();
1843 315 : switch (v->nexttype)
1844 : {
1845 305 : case PLAIN:
1846 : case RANGE:
1847 305 : endc = v->nextvalue;
1848 305 : NEXT();
1849 305 : NOERR();
1850 303 : break;
1851 2 : case COLLEL:
1852 2 : startp = v->now;
1853 2 : endp = scanplain(v);
1854 2 : INSIST(startp < endp, REG_ECOLLATE);
1855 2 : NOERR();
1856 2 : endc = element(v, startp, endp);
1857 2 : NOERR();
1858 2 : break;
1859 8 : default:
1860 8 : ERR(REG_ERANGE);
1861 8 : return;
1862 : break;
1863 : }
1864 : }
1865 : else
1866 4 : endc = startc;
1867 :
1868 : /*
1869 : * Ranges are unportable. Actually, standard C does guarantee that digits
1870 : * are contiguous, but making that an exception is just too complicated.
1871 : */
1872 309 : if (startc != endc)
1873 301 : NOTE(REG_UUNPORT);
1874 309 : cv = range(v, startc, endc, (v->cflags & REG_ICASE));
1875 309 : NOERR();
1876 307 : subcolorcvec(v, cv, lp, rp);
1877 : }
1878 :
1879 : /*
1880 : * scanplain - scan PLAIN contents of [. etc.
1881 : *
1882 : * Certain bits of trickery in regc_lex.c know that this code does not try
1883 : * to look past the final bracket of the [. etc.
1884 : */
1885 : static const chr * /* just after end of sequence */
1886 177 : scanplain(struct vars *v)
1887 : {
1888 : const chr *endp;
1889 :
1890 : assert(SEE(COLLEL) || SEE(ECLASS) || SEE(CCLASS));
1891 177 : NEXT();
1892 :
1893 177 : endp = v->now;
1894 953 : while (SEE(PLAIN))
1895 : {
1896 776 : endp = v->now;
1897 776 : NEXT();
1898 : }
1899 :
1900 : assert(SEE(END) || ISERR());
1901 177 : NEXT();
1902 :
1903 177 : return endp;
1904 : }
1905 :
1906 : /*
1907 : * onechr - fill in arcs for a plain character, and possible case complements
1908 : * This is mostly a shortcut for efficient handling of the common case.
1909 : */
1910 : static void
1911 42532 : onechr(struct vars *v,
1912 : chr c,
1913 : struct state *lp,
1914 : struct state *rp)
1915 : {
1916 42532 : if (!(v->cflags & REG_ICASE))
1917 : {
1918 41712 : color lastsubcolor = COLORLESS;
1919 :
1920 41712 : subcoloronechr(v, c, lp, rp, &lastsubcolor);
1921 41712 : return;
1922 : }
1923 :
1924 : /* rats, need general case anyway... */
1925 820 : subcolorcvec(v, allcases(v, c), lp, rp);
1926 : }
1927 :
1928 : /*
1929 : * optimizebracket - see if bracket expression can be converted to RAINBOW
1930 : *
1931 : * Cases such as "[\s\S]" can produce a set of arcs of all colors, which we
1932 : * can replace by a single RAINBOW arc for efficiency. (This might seem
1933 : * like a silly way to write ".", but it's seemingly a common locution in
1934 : * some other flavors of regex, so take the trouble to support it well.)
1935 : */
1936 : static void
1937 16 : optimizebracket(struct vars *v,
1938 : struct state *lp,
1939 : struct state *rp)
1940 : {
1941 : struct colordesc *cd;
1942 16 : struct colordesc *end = CDEND(v->cm);
1943 : struct arc *a;
1944 : bool israinbow;
1945 :
1946 : /*
1947 : * Scan lp's out-arcs and transiently mark the mentioned colors. We
1948 : * expect that all of lp's out-arcs are plain, non-RAINBOW arcs to rp.
1949 : * (Note: there shouldn't be any pseudocolors yet, but check anyway.)
1950 : */
1951 41 : for (a = lp->outs; a != NULL; a = a->outchain)
1952 : {
1953 : assert(a->type == PLAIN);
1954 : assert(a->co >= 0); /* i.e. not RAINBOW */
1955 : assert(a->to == rp);
1956 25 : cd = &v->cm->cd[a->co];
1957 : assert(!UNUSEDCOLOR(cd) && !(cd->flags & PSEUDO));
1958 25 : cd->flags |= COLMARK;
1959 : }
1960 :
1961 : /* Scan colors, clear transient marks, check for unmarked live colors */
1962 16 : israinbow = true;
1963 62 : for (cd = v->cm->cd; cd < end; cd++)
1964 : {
1965 46 : if (cd->flags & COLMARK)
1966 25 : cd->flags &= ~COLMARK;
1967 21 : else if (!UNUSEDCOLOR(cd) && !(cd->flags & PSEUDO))
1968 14 : israinbow = false;
1969 : }
1970 :
1971 : /* Can't do anything if not all colors have arcs */
1972 16 : if (!israinbow)
1973 13 : return;
1974 :
1975 : /* OK, drop existing arcs and replace with a rainbow */
1976 9 : while ((a = lp->outs) != NULL)
1977 6 : freearc(v->nfa, a);
1978 3 : newarc(v->nfa, PLAIN, RAINBOW, lp, rp);
1979 : }
1980 :
1981 : /*
1982 : * wordchrs - set up word-chr list for word-boundary stuff, if needed
1983 : *
1984 : * The list is kept as a bunch of circular arcs on an otherwise-unused state.
1985 : *
1986 : * Note that this must not be called while we have any open subcolors,
1987 : * else construction of the list would confuse color bookkeeping.
1988 : * Hence, we can't currently apply a similar optimization in
1989 : * charclass[complement](), as those need to be usable within bracket
1990 : * expressions.
1991 : */
1992 : static void
1993 87 : wordchrs(struct vars *v)
1994 : {
1995 : struct state *cstate;
1996 : struct cvec *cv;
1997 :
1998 87 : if (v->wordchrs != NULL)
1999 15 : return; /* done already */
2000 :
2001 : /* make dummy state to hang the cache arcs on */
2002 72 : cstate = newstate(v->nfa);
2003 72 : NOERR();
2004 :
2005 : /* obtain possibly-cached cvec for \w characters */
2006 72 : NOTE(REG_ULOCALE);
2007 72 : cv = cclasscvec(v, CC_WORD, (v->cflags & REG_ICASE));
2008 72 : NOERR();
2009 :
2010 : /* build the arcs; this may cause color splitting */
2011 72 : subcolorcvec(v, cv, cstate, cstate);
2012 72 : NOERR();
2013 :
2014 : /* close new open subcolors to ensure the cache entry is self-contained */
2015 72 : okcolors(v->nfa, v->cm);
2016 72 : NOERR();
2017 :
2018 : /* success! save the cache pointer */
2019 72 : v->wordchrs = cstate;
2020 : }
2021 :
2022 : /*
2023 : * processlacon - generate the NFA representation of a LACON
2024 : *
2025 : * In the general case this is just newlacon() + newarc(), but some cases
2026 : * can be optimized.
2027 : */
2028 : static void
2029 127 : processlacon(struct vars *v,
2030 : struct state *begin, /* start of parsed LACON sub-re */
2031 : struct state *end, /* end of parsed LACON sub-re */
2032 : int latype,
2033 : struct state *lp, /* left state to hang it on */
2034 : struct state *rp) /* right state to hang it on */
2035 : {
2036 : struct state *s1;
2037 : int n;
2038 :
2039 : /*
2040 : * Check for lookaround RE consisting of a single plain color arc (or set
2041 : * of arcs); this would typically be a simple chr or a bracket expression.
2042 : */
2043 127 : s1 = single_color_transition(begin, end);
2044 127 : switch (latype)
2045 : {
2046 35 : case LATYPE_AHEAD_POS:
2047 : /* If lookahead RE is just colorset C, convert to AHEAD(C) */
2048 35 : if (s1 != NULL)
2049 : {
2050 30 : cloneouts(v->nfa, s1, lp, rp, AHEAD);
2051 30 : return;
2052 : }
2053 5 : break;
2054 39 : case LATYPE_AHEAD_NEG:
2055 : /* If lookahead RE is just colorset C, convert to AHEAD(^C)|$ */
2056 39 : if (s1 != NULL)
2057 : {
2058 10 : colorcomplement(v->nfa, v->cm, AHEAD, s1, lp, rp);
2059 10 : newarc(v->nfa, '$', 1, lp, rp);
2060 10 : newarc(v->nfa, '$', 0, lp, rp);
2061 10 : return;
2062 : }
2063 29 : break;
2064 39 : case LATYPE_BEHIND_POS:
2065 : /* If lookbehind RE is just colorset C, convert to BEHIND(C) */
2066 39 : if (s1 != NULL)
2067 : {
2068 30 : cloneouts(v->nfa, s1, lp, rp, BEHIND);
2069 30 : return;
2070 : }
2071 9 : break;
2072 14 : case LATYPE_BEHIND_NEG:
2073 : /* If lookbehind RE is just colorset C, convert to BEHIND(^C)|^ */
2074 14 : if (s1 != NULL)
2075 : {
2076 14 : colorcomplement(v->nfa, v->cm, BEHIND, s1, lp, rp);
2077 14 : newarc(v->nfa, '^', 1, lp, rp);
2078 14 : newarc(v->nfa, '^', 0, lp, rp);
2079 14 : return;
2080 : }
2081 0 : break;
2082 43 : default:
2083 : assert(NOTREACHED);
2084 : }
2085 :
2086 : /* General case: we need a LACON subre and arc */
2087 43 : n = newlacon(v, begin, end, latype);
2088 43 : newarc(v->nfa, LACON, n, lp, rp);
2089 : }
2090 :
2091 : /*
2092 : * subre - allocate a subre
2093 : */
2094 : static struct subre *
2095 22347 : subre(struct vars *v,
2096 : int op,
2097 : int flags,
2098 : struct state *begin,
2099 : struct state *end)
2100 : {
2101 22347 : struct subre *ret = v->treefree;
2102 :
2103 : /*
2104 : * Checking for stack overflow here is sufficient to protect parse() and
2105 : * its recursive subroutines.
2106 : */
2107 22347 : if (STACK_TOO_DEEP(v->re))
2108 : {
2109 0 : ERR(REG_ETOOBIG);
2110 0 : return NULL;
2111 : }
2112 :
2113 22347 : if (ret != NULL)
2114 3089 : v->treefree = ret->child;
2115 : else
2116 : {
2117 19258 : ret = (struct subre *) MALLOC(sizeof(struct subre));
2118 19258 : if (ret == NULL)
2119 : {
2120 0 : ERR(REG_ESPACE);
2121 0 : return NULL;
2122 : }
2123 19258 : ret->chain = v->treechain;
2124 19258 : v->treechain = ret;
2125 : }
2126 :
2127 : assert(strchr("=b|.*(", op) != NULL);
2128 :
2129 22347 : ret->op = op;
2130 22347 : ret->flags = flags;
2131 22347 : ret->latype = (char) -1;
2132 22347 : ret->id = 0; /* will be assigned later */
2133 22347 : ret->capno = 0;
2134 22347 : ret->backno = 0;
2135 22347 : ret->min = ret->max = 1;
2136 22347 : ret->child = NULL;
2137 22347 : ret->sibling = NULL;
2138 22347 : ret->begin = begin;
2139 22347 : ret->end = end;
2140 22347 : ZAPCNFA(ret->cnfa);
2141 :
2142 22347 : return ret;
2143 : }
2144 :
2145 : /*
2146 : * freesubre - free a subRE subtree
2147 : *
2148 : * This frees child node(s) of the given subRE too,
2149 : * but not its siblings.
2150 : */
2151 : static void
2152 10093 : freesubre(struct vars *v, /* might be NULL */
2153 : struct subre *sr)
2154 : {
2155 10093 : if (sr == NULL)
2156 7 : return;
2157 :
2158 10086 : if (sr->child != NULL)
2159 697 : freesubreandsiblings(v, sr->child);
2160 :
2161 10086 : freesrnode(v, sr);
2162 : }
2163 :
2164 : /*
2165 : * freesubreandsiblings - free a subRE subtree
2166 : *
2167 : * This frees child node(s) of the given subRE too,
2168 : * as well as any following siblings.
2169 : */
2170 : static void
2171 4393 : freesubreandsiblings(struct vars *v, /* might be NULL */
2172 : struct subre *sr)
2173 : {
2174 12968 : while (sr != NULL)
2175 : {
2176 8575 : struct subre *next = sr->sibling;
2177 :
2178 8575 : freesubre(v, sr);
2179 8575 : sr = next;
2180 : }
2181 4393 : }
2182 :
2183 : /*
2184 : * freesrnode - free one node in a subRE subtree
2185 : */
2186 : static void
2187 17486 : freesrnode(struct vars *v, /* might be NULL */
2188 : struct subre *sr)
2189 : {
2190 17486 : if (sr == NULL)
2191 0 : return;
2192 :
2193 17486 : if (!NULLCNFA(sr->cnfa))
2194 1480 : freecnfa(&sr->cnfa);
2195 17486 : sr->flags = 0; /* in particular, not INUSE */
2196 17486 : sr->child = sr->sibling = NULL;
2197 17486 : sr->begin = sr->end = NULL;
2198 :
2199 17486 : if (v != NULL && v->treechain != NULL)
2200 : {
2201 : /* we're still parsing, maybe we can reuse the subre */
2202 16003 : sr->child = v->treefree;
2203 16003 : v->treefree = sr;
2204 : }
2205 : else
2206 1483 : FREE(sr);
2207 : }
2208 :
2209 : /*
2210 : * removecaptures - remove unnecessary capture subREs
2211 : *
2212 : * If the caller said that it doesn't care about subexpression match data,
2213 : * we may delete the "capture" markers on subREs that are not referenced
2214 : * by any backrefs, and then simplify anything that's become non-messy.
2215 : * Call this only if REG_NOSUB flag is set.
2216 : */
2217 : static void
2218 10461 : removecaptures(struct vars *v,
2219 : struct subre *t)
2220 : {
2221 : struct subre *t2;
2222 :
2223 : assert(t != NULL);
2224 :
2225 : /*
2226 : * If this isn't itself a backref target, clear capno and tentatively
2227 : * clear CAP flag.
2228 : */
2229 10461 : if (!(t->flags & BRUSE))
2230 : {
2231 10424 : t->capno = 0;
2232 10424 : t->flags &= ~CAP;
2233 : }
2234 :
2235 : /* Now recurse to children */
2236 17825 : for (t2 = t->child; t2 != NULL; t2 = t2->sibling)
2237 : {
2238 7364 : removecaptures(v, t2);
2239 : /* Propagate child CAP flag back up, if it's still set */
2240 7364 : if (t2->flags & CAP)
2241 74 : t->flags |= CAP;
2242 : }
2243 :
2244 : /*
2245 : * If t now contains neither captures nor backrefs, there's no longer any
2246 : * need to care where its sub-match boundaries are, so we can reduce it to
2247 : * a simple DFA node. (Note in particular that MIXED child greediness is
2248 : * not a hindrance here, so we don't use the MESSY() macro.)
2249 : */
2250 10461 : if ((t->flags & (CAP | BACKR)) == 0)
2251 : {
2252 10264 : if (t->child)
2253 3591 : freesubreandsiblings(v, t->child);
2254 10264 : t->child = NULL;
2255 10264 : t->op = '=';
2256 10264 : t->flags &= ~MIXED;
2257 : }
2258 10461 : }
2259 :
2260 : /*
2261 : * numst - number tree nodes (assigning "id" indexes)
2262 : */
2263 : static int /* next number */
2264 5983 : numst(struct subre *t,
2265 : int start) /* starting point for subtree numbers */
2266 : {
2267 : int i;
2268 : struct subre *t2;
2269 :
2270 : assert(t != NULL);
2271 :
2272 5983 : i = start;
2273 5983 : t->id = i++;
2274 7886 : for (t2 = t->child; t2 != NULL; t2 = t2->sibling)
2275 1903 : i = numst(t2, i);
2276 5983 : return i;
2277 : }
2278 :
2279 : /*
2280 : * markst - mark tree nodes as INUSE
2281 : *
2282 : * Note: this is a great deal more subtle than it looks. During initial
2283 : * parsing of a regex, all subres are linked into the treechain list;
2284 : * discarded ones are also linked into the treefree list for possible reuse.
2285 : * After we are done creating all subres required for a regex, we run markst()
2286 : * then cleanst(), which results in discarding all subres not reachable from
2287 : * v->tree. We then clear v->treechain, indicating that subres must be found
2288 : * by descending from v->tree. This changes the behavior of freesubre(): it
2289 : * will henceforth FREE() unwanted subres rather than sticking them into the
2290 : * treefree list. (Doing that any earlier would result in dangling links in
2291 : * the treechain list.) This all means that freev() will clean up correctly
2292 : * if invoked before or after markst()+cleanst(); but it would not work if
2293 : * called partway through this state conversion, so we mustn't error out
2294 : * in or between these two functions.
2295 : */
2296 : static void
2297 5983 : markst(struct subre *t)
2298 : {
2299 : struct subre *t2;
2300 :
2301 : assert(t != NULL);
2302 :
2303 5983 : t->flags |= INUSE;
2304 7886 : for (t2 = t->child; t2 != NULL; t2 = t2->sibling)
2305 1903 : markst(t2);
2306 5983 : }
2307 :
2308 : /*
2309 : * cleanst - free any tree nodes not marked INUSE
2310 : */
2311 : static void
2312 4192 : cleanst(struct vars *v)
2313 : {
2314 : struct subre *t;
2315 : struct subre *next;
2316 :
2317 23450 : for (t = v->treechain; t != NULL; t = next)
2318 : {
2319 19258 : next = t->chain;
2320 19258 : if (!(t->flags & INUSE))
2321 13275 : FREE(t);
2322 : }
2323 4192 : v->treechain = NULL;
2324 4192 : v->treefree = NULL; /* just on general principles */
2325 4192 : }
2326 :
2327 : /*
2328 : * nfatree - turn a subRE subtree into a tree of compacted NFAs
2329 : */
2330 : static long /* optimize results from top node */
2331 5983 : nfatree(struct vars *v,
2332 : struct subre *t,
2333 : FILE *f) /* for debug output */
2334 : {
2335 : struct subre *t2;
2336 :
2337 : assert(t != NULL && t->begin != NULL);
2338 :
2339 7886 : for (t2 = t->child; t2 != NULL; t2 = t2->sibling)
2340 1903 : (DISCARD) nfatree(v, t2, f);
2341 :
2342 5983 : return nfanode(v, t, 0, f);
2343 : }
2344 :
2345 : /*
2346 : * nfanode - do one NFA for nfatree or lacons
2347 : *
2348 : * If converttosearch is true, apply makesearch() to the NFA.
2349 : */
2350 : static long /* optimize results */
2351 6026 : nfanode(struct vars *v,
2352 : struct subre *t,
2353 : int converttosearch,
2354 : FILE *f) /* for debug output */
2355 : {
2356 : struct nfa *nfa;
2357 6026 : long ret = 0;
2358 :
2359 : assert(t->begin != NULL);
2360 :
2361 : #ifdef REG_DEBUG
2362 : if (f != NULL)
2363 : {
2364 : char idbuf[50];
2365 :
2366 : fprintf(f, "\n\n\n========= TREE NODE %s ==========\n",
2367 : stid(t, idbuf, sizeof(idbuf)));
2368 : }
2369 : #endif
2370 6026 : nfa = newnfa(v, v->cm, v->nfa);
2371 6026 : NOERRZ();
2372 6026 : dupnfa(nfa, t->begin, t->end, nfa->init, nfa->final);
2373 6026 : nfa->flags = v->nfa->flags;
2374 6026 : if (!ISERR())
2375 6026 : specialcolors(nfa);
2376 6026 : if (!ISERR())
2377 6026 : ret = optimize(nfa, f);
2378 6026 : if (converttosearch && !ISERR())
2379 9 : makesearch(v, nfa);
2380 6026 : if (!ISERR())
2381 6023 : compact(nfa, &t->cnfa);
2382 :
2383 6026 : freenfa(nfa);
2384 6026 : return ret;
2385 : }
2386 :
2387 : /*
2388 : * newlacon - allocate a lookaround-constraint subRE
2389 : */
2390 : static int /* lacon number */
2391 43 : newlacon(struct vars *v,
2392 : struct state *begin,
2393 : struct state *end,
2394 : int latype)
2395 : {
2396 : int n;
2397 : struct subre *newlacons;
2398 : struct subre *sub;
2399 :
2400 43 : if (v->nlacons == 0)
2401 : {
2402 29 : n = 1; /* skip 0th */
2403 29 : newlacons = (struct subre *) MALLOC(2 * sizeof(struct subre));
2404 : }
2405 : else
2406 : {
2407 14 : n = v->nlacons;
2408 14 : newlacons = (struct subre *) REALLOC(v->lacons,
2409 : (n + 1) * sizeof(struct subre));
2410 : }
2411 43 : if (newlacons == NULL)
2412 : {
2413 0 : ERR(REG_ESPACE);
2414 0 : return 0;
2415 : }
2416 43 : v->lacons = newlacons;
2417 43 : v->nlacons = n + 1;
2418 43 : sub = &v->lacons[n];
2419 43 : sub->begin = begin;
2420 43 : sub->end = end;
2421 43 : sub->latype = latype;
2422 43 : ZAPCNFA(sub->cnfa);
2423 43 : return n;
2424 : }
2425 :
2426 : /*
2427 : * freelacons - free lookaround-constraint subRE vector
2428 : */
2429 : static void
2430 10 : freelacons(struct subre *subs,
2431 : int n)
2432 : {
2433 : struct subre *sub;
2434 : int i;
2435 :
2436 : assert(n > 0);
2437 28 : for (sub = subs + 1, i = n - 1; i > 0; sub++, i--) /* no 0th */
2438 18 : if (!NULLCNFA(sub->cnfa))
2439 18 : freecnfa(&sub->cnfa);
2440 10 : FREE(subs);
2441 10 : }
2442 :
2443 : /*
2444 : * rfree - free a whole RE (insides of regfree)
2445 : */
2446 : static void
2447 772 : rfree(regex_t *re)
2448 : {
2449 : struct guts *g;
2450 :
2451 772 : if (re == NULL || re->re_magic != REMAGIC)
2452 0 : return;
2453 :
2454 772 : re->re_magic = 0; /* invalidate RE */
2455 772 : g = (struct guts *) re->re_guts;
2456 772 : re->re_guts = NULL;
2457 772 : re->re_fns = NULL;
2458 772 : if (g != NULL)
2459 : {
2460 772 : g->magic = 0;
2461 772 : freecm(&g->cmap);
2462 772 : if (g->tree != NULL)
2463 650 : freesubre((struct vars *) NULL, g->tree);
2464 772 : if (g->lacons != NULL)
2465 10 : freelacons(g->lacons, g->nlacons);
2466 772 : if (!NULLCNFA(g->search))
2467 650 : freecnfa(&g->search);
2468 772 : FREE(g);
2469 : }
2470 : }
2471 :
2472 : /*
2473 : * rstacktoodeep - check for stack getting dangerously deep
2474 : *
2475 : * Return nonzero to fail the operation with error code REG_ETOOBIG,
2476 : * zero to keep going
2477 : *
2478 : * The current implementation is Postgres-specific. If we ever get around
2479 : * to splitting the regex code out as a standalone library, there will need
2480 : * to be some API to let applications define a callback function for this.
2481 : */
2482 : static int
2483 11025644 : rstacktoodeep(void)
2484 : {
2485 11025644 : return stack_is_too_deep();
2486 : }
2487 :
2488 : #ifdef REG_DEBUG
2489 :
2490 : /*
2491 : * dump - dump an RE in human-readable form
2492 : */
2493 : static void
2494 : dump(regex_t *re,
2495 : FILE *f)
2496 : {
2497 : struct guts *g;
2498 : int i;
2499 :
2500 : if (re->re_magic != REMAGIC)
2501 : fprintf(f, "bad magic number (0x%x not 0x%x)\n", re->re_magic,
2502 : REMAGIC);
2503 : if (re->re_guts == NULL)
2504 : {
2505 : fprintf(f, "NULL guts!!!\n");
2506 : return;
2507 : }
2508 : g = (struct guts *) re->re_guts;
2509 : if (g->magic != GUTSMAGIC)
2510 : fprintf(f, "bad guts magic number (0x%x not 0x%x)\n", g->magic,
2511 : GUTSMAGIC);
2512 :
2513 : fprintf(f, "\n\n\n========= DUMP ==========\n");
2514 : fprintf(f, "nsub %d, info 0%lo, csize %d, ntree %d\n",
2515 : (int) re->re_nsub, re->re_info, re->re_csize, g->ntree);
2516 :
2517 : dumpcolors(&g->cmap, f);
2518 : if (!NULLCNFA(g->search))
2519 : {
2520 : fprintf(f, "\nsearch:\n");
2521 : dumpcnfa(&g->search, f);
2522 : }
2523 : for (i = 1; i < g->nlacons; i++)
2524 : {
2525 : struct subre *lasub = &g->lacons[i];
2526 : const char *latype;
2527 :
2528 : switch (lasub->latype)
2529 : {
2530 : case LATYPE_AHEAD_POS:
2531 : latype = "positive lookahead";
2532 : break;
2533 : case LATYPE_AHEAD_NEG:
2534 : latype = "negative lookahead";
2535 : break;
2536 : case LATYPE_BEHIND_POS:
2537 : latype = "positive lookbehind";
2538 : break;
2539 : case LATYPE_BEHIND_NEG:
2540 : latype = "negative lookbehind";
2541 : break;
2542 : default:
2543 : latype = "???";
2544 : break;
2545 : }
2546 : fprintf(f, "\nla%d (%s):\n", i, latype);
2547 : dumpcnfa(&lasub->cnfa, f);
2548 : }
2549 : fprintf(f, "\n");
2550 : dumpst(g->tree, f, 0);
2551 : }
2552 :
2553 : /*
2554 : * dumpst - dump a subRE tree
2555 : */
2556 : static void
2557 : dumpst(struct subre *t,
2558 : FILE *f,
2559 : int nfapresent) /* is the original NFA still around? */
2560 : {
2561 : if (t == NULL)
2562 : fprintf(f, "null tree\n");
2563 : else
2564 : stdump(t, f, nfapresent);
2565 : fflush(f);
2566 : }
2567 :
2568 : /*
2569 : * stdump - recursive guts of dumpst
2570 : */
2571 : static void
2572 : stdump(struct subre *t,
2573 : FILE *f,
2574 : int nfapresent) /* is the original NFA still around? */
2575 : {
2576 : char idbuf[50];
2577 : struct subre *t2;
2578 :
2579 : fprintf(f, "%s. `%c'", stid(t, idbuf, sizeof(idbuf)), t->op);
2580 : if (t->flags & LONGER)
2581 : fprintf(f, " longest");
2582 : if (t->flags & SHORTER)
2583 : fprintf(f, " shortest");
2584 : if (t->flags & MIXED)
2585 : fprintf(f, " hasmixed");
2586 : if (t->flags & CAP)
2587 : fprintf(f, " hascapture");
2588 : if (t->flags & BACKR)
2589 : fprintf(f, " hasbackref");
2590 : if (t->flags & BRUSE)
2591 : fprintf(f, " isreferenced");
2592 : if (!(t->flags & INUSE))
2593 : fprintf(f, " UNUSED");
2594 : if (t->latype != (char) -1)
2595 : fprintf(f, " latype(%d)", t->latype);
2596 : if (t->capno != 0)
2597 : fprintf(f, " capture(%d)", t->capno);
2598 : if (t->backno != 0)
2599 : fprintf(f, " backref(%d)", t->backno);
2600 : if (t->min != 1 || t->max != 1)
2601 : {
2602 : fprintf(f, " {%d,", t->min);
2603 : if (t->max != DUPINF)
2604 : fprintf(f, "%d", t->max);
2605 : fprintf(f, "}");
2606 : }
2607 : if (nfapresent)
2608 : fprintf(f, " %ld-%ld", (long) t->begin->no, (long) t->end->no);
2609 : if (t->child != NULL)
2610 : fprintf(f, " C:%s", stid(t->child, idbuf, sizeof(idbuf)));
2611 : /* printing second child isn't necessary, but it is often helpful */
2612 : if (t->child != NULL && t->child->sibling != NULL)
2613 : fprintf(f, " C2:%s", stid(t->child->sibling, idbuf, sizeof(idbuf)));
2614 : if (t->sibling != NULL)
2615 : fprintf(f, " S:%s", stid(t->sibling, idbuf, sizeof(idbuf)));
2616 : if (!NULLCNFA(t->cnfa))
2617 : {
2618 : fprintf(f, "\n");
2619 : dumpcnfa(&t->cnfa, f);
2620 : }
2621 : fprintf(f, "\n");
2622 : for (t2 = t->child; t2 != NULL; t2 = t2->sibling)
2623 : stdump(t2, f, nfapresent);
2624 : }
2625 :
2626 : /*
2627 : * stid - identify a subtree node for dumping
2628 : */
2629 : static const char * /* points to buf or constant string */
2630 : stid(struct subre *t,
2631 : char *buf,
2632 : size_t bufsize)
2633 : {
2634 : /* big enough for hex int or decimal t->id? */
2635 : if (bufsize < sizeof(void *) * 2 + 3 || bufsize < sizeof(t->id) * 3 + 1)
2636 : return "unable";
2637 : if (t->id != 0)
2638 : sprintf(buf, "%d", t->id);
2639 : else
2640 : sprintf(buf, "%p", t);
2641 : return buf;
2642 : }
2643 : #endif /* REG_DEBUG */
2644 :
2645 :
2646 : #include "regc_lex.c"
2647 : #include "regc_color.c"
2648 : #include "regc_nfa.c"
2649 : #include "regc_cvec.c"
2650 : #include "regc_pg_locale.c"
2651 : #include "regc_locale.c"
|
