1 | /* |
2 | * re_*comp and friends - compile REs |
3 | * This file #includes several others (see the bottom). |
4 | * |
5 | * Copyright (c) 1998, 1999 Henry Spencer. All rights reserved. |
6 | * |
7 | * Development of this software was funded, in part, by Cray Research Inc., |
8 | * UUNET Communications Services Inc., Sun Microsystems Inc., and Scriptics |
9 | * Corporation, none of whom are responsible for the results. The author |
10 | * thanks all of them. |
11 | * |
12 | * Redistribution and use in source and binary forms -- with or without |
13 | * modification -- are permitted for any purpose, provided that |
14 | * redistributions in source form retain this entire copyright notice and |
15 | * indicate the origin and nature of any modifications. |
16 | * |
17 | * I'd appreciate being given credit for this package in the documentation |
18 | * of software which uses it, but that is not a requirement. |
19 | * |
20 | * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, |
21 | * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY |
22 | * AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL |
23 | * HENRY SPENCER BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
24 | * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
25 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
26 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
27 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR |
28 | * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF |
29 | * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
30 | * |
31 | * src/backend/regex/regcomp.c |
32 | * |
33 | */ |
34 | |
35 | #include "regex/regguts.h" |
36 | |
37 | /* |
38 | * forward declarations, up here so forward datatypes etc. are defined early |
39 | */ |
40 | /* === regcomp.c === */ |
41 | static void moresubs(struct vars *, int); |
42 | static int freev(struct vars *, int); |
43 | static void makesearch(struct vars *, struct nfa *); |
44 | static struct subre *parse(struct vars *, int, int, struct state *, struct state *); |
45 | static struct subre *parsebranch(struct vars *, int, int, struct state *, struct state *, int); |
46 | static void parseqatom(struct vars *, int, int, struct state *, struct state *, struct subre *); |
47 | static void nonword(struct vars *, int, struct state *, struct state *); |
48 | static void word(struct vars *, int, struct state *, struct state *); |
49 | static int scannum(struct vars *); |
50 | static void repeat(struct vars *, struct state *, struct state *, int, int); |
51 | static void bracket(struct vars *, struct state *, struct state *); |
52 | static void cbracket(struct vars *, struct state *, struct state *); |
53 | static void brackpart(struct vars *, struct state *, struct state *); |
54 | static const chr *scanplain(struct vars *); |
55 | static void onechr(struct vars *, chr, struct state *, struct state *); |
56 | static void wordchrs(struct vars *); |
57 | static void processlacon(struct vars *, struct state *, struct state *, int, |
58 | struct state *, struct state *); |
59 | static struct subre *subre(struct vars *, int, int, struct state *, struct state *); |
60 | static void freesubre(struct vars *, struct subre *); |
61 | static void freesrnode(struct vars *, struct subre *); |
62 | static void optst(struct vars *, struct subre *); |
63 | static int numst(struct subre *, int); |
64 | static void markst(struct subre *); |
65 | static void cleanst(struct vars *); |
66 | static long nfatree(struct vars *, struct subre *, FILE *); |
67 | static long nfanode(struct vars *, struct subre *, int, FILE *); |
68 | static int newlacon(struct vars *, struct state *, struct state *, int); |
69 | static void freelacons(struct subre *, int); |
70 | static void rfree(regex_t *); |
71 | static int rcancelrequested(void); |
72 | static int rstacktoodeep(void); |
73 | |
74 | #ifdef REG_DEBUG |
75 | static void dump(regex_t *, FILE *); |
76 | static void dumpst(struct subre *, FILE *, int); |
77 | static void stdump(struct subre *, FILE *, int); |
78 | static const char *stid(struct subre *, char *, size_t); |
79 | #endif |
80 | /* === regc_lex.c === */ |
81 | static void lexstart(struct vars *); |
82 | static void prefixes(struct vars *); |
83 | static void lexnest(struct vars *, const chr *, const chr *); |
84 | static void lexword(struct vars *); |
85 | static int next(struct vars *); |
86 | static int lexescape(struct vars *); |
87 | static chr lexdigits(struct vars *, int, int, int); |
88 | static int brenext(struct vars *, chr); |
89 | static void skip(struct vars *); |
90 | static chr newline(void); |
91 | static chr chrnamed(struct vars *, const chr *, const chr *, chr); |
92 | |
93 | /* === regc_color.c === */ |
94 | static void initcm(struct vars *, struct colormap *); |
95 | static void freecm(struct colormap *); |
96 | static color maxcolor(struct colormap *); |
97 | static color newcolor(struct colormap *); |
98 | static void freecolor(struct colormap *, color); |
99 | static color pseudocolor(struct colormap *); |
100 | static color subcolor(struct colormap *, chr); |
101 | static color subcolorhi(struct colormap *, color *); |
102 | static color newsub(struct colormap *, color); |
103 | static int newhicolorrow(struct colormap *, int); |
104 | static void newhicolorcols(struct colormap *); |
105 | static void subcolorcvec(struct vars *, struct cvec *, struct state *, struct state *); |
106 | static void subcoloronechr(struct vars *, chr, struct state *, struct state *, color *); |
107 | static void subcoloronerange(struct vars *, chr, chr, struct state *, struct state *, color *); |
108 | static void subcoloronerow(struct vars *, int, struct state *, struct state *, color *); |
109 | static void okcolors(struct nfa *, struct colormap *); |
110 | static void colorchain(struct colormap *, struct arc *); |
111 | static void uncolorchain(struct colormap *, struct arc *); |
112 | static void rainbow(struct nfa *, struct colormap *, int, color, struct state *, struct state *); |
113 | static void colorcomplement(struct nfa *, struct colormap *, int, struct state *, struct state *, struct state *); |
114 | |
115 | #ifdef REG_DEBUG |
116 | static void dumpcolors(struct colormap *, FILE *); |
117 | static void dumpchr(chr, FILE *); |
118 | #endif |
119 | /* === regc_nfa.c === */ |
120 | static struct nfa *newnfa(struct vars *, struct colormap *, struct nfa *); |
121 | static void freenfa(struct nfa *); |
122 | static struct state *newstate(struct nfa *); |
123 | static struct state *newfstate(struct nfa *, int flag); |
124 | static void dropstate(struct nfa *, struct state *); |
125 | static void freestate(struct nfa *, struct state *); |
126 | static void destroystate(struct nfa *, struct state *); |
127 | static void newarc(struct nfa *, int, color, struct state *, struct state *); |
128 | static void createarc(struct nfa *, int, color, struct state *, struct state *); |
129 | static struct arc *allocarc(struct nfa *, struct state *); |
130 | static void freearc(struct nfa *, struct arc *); |
131 | static void changearctarget(struct arc *, struct state *); |
132 | static int hasnonemptyout(struct state *); |
133 | static struct arc *findarc(struct state *, int, color); |
134 | static void cparc(struct nfa *, struct arc *, struct state *, struct state *); |
135 | static void sortins(struct nfa *, struct state *); |
136 | static int sortins_cmp(const void *, const void *); |
137 | static void sortouts(struct nfa *, struct state *); |
138 | static int sortouts_cmp(const void *, const void *); |
139 | static void moveins(struct nfa *, struct state *, struct state *); |
140 | static void copyins(struct nfa *, struct state *, struct state *); |
141 | static void mergeins(struct nfa *, struct state *, struct arc **, int); |
142 | static void moveouts(struct nfa *, struct state *, struct state *); |
143 | static void copyouts(struct nfa *, struct state *, struct state *); |
144 | static void cloneouts(struct nfa *, struct state *, struct state *, struct state *, int); |
145 | static void delsub(struct nfa *, struct state *, struct state *); |
146 | static void deltraverse(struct nfa *, struct state *, struct state *); |
147 | static void dupnfa(struct nfa *, struct state *, struct state *, struct state *, struct state *); |
148 | static void duptraverse(struct nfa *, struct state *, struct state *); |
149 | static void cleartraverse(struct nfa *, struct state *); |
150 | static struct state *single_color_transition(struct state *, struct state *); |
151 | static void specialcolors(struct nfa *); |
152 | static long optimize(struct nfa *, FILE *); |
153 | static void pullback(struct nfa *, FILE *); |
154 | static int pull(struct nfa *, struct arc *, struct state **); |
155 | static void pushfwd(struct nfa *, FILE *); |
156 | static int push(struct nfa *, struct arc *, struct state **); |
157 | |
158 | #define INCOMPATIBLE 1 /* destroys arc */ |
159 | #define SATISFIED 2 /* constraint satisfied */ |
160 | #define COMPATIBLE 3 /* compatible but not satisfied yet */ |
161 | static int combine(struct arc *, struct arc *); |
162 | static void fixempties(struct nfa *, FILE *); |
163 | static struct state *emptyreachable(struct nfa *, struct state *, |
164 | struct state *, struct arc **); |
165 | static int isconstraintarc(struct arc *); |
166 | static int hasconstraintout(struct state *); |
167 | static void fixconstraintloops(struct nfa *, FILE *); |
168 | static int findconstraintloop(struct nfa *, struct state *); |
169 | static void breakconstraintloop(struct nfa *, struct state *); |
170 | static void clonesuccessorstates(struct nfa *, struct state *, struct state *, |
171 | struct state *, struct arc *, |
172 | char *, char *, int); |
173 | static void cleanup(struct nfa *); |
174 | static void markreachable(struct nfa *, struct state *, struct state *, struct state *); |
175 | static void markcanreach(struct nfa *, struct state *, struct state *, struct state *); |
176 | static long analyze(struct nfa *); |
177 | static void compact(struct nfa *, struct cnfa *); |
178 | static void carcsort(struct carc *, size_t); |
179 | static int carc_cmp(const void *, const void *); |
180 | static void freecnfa(struct cnfa *); |
181 | static void dumpnfa(struct nfa *, FILE *); |
182 | |
183 | #ifdef REG_DEBUG |
184 | static void dumpstate(struct state *, FILE *); |
185 | static void dumparcs(struct state *, FILE *); |
186 | static void dumparc(struct arc *, struct state *, FILE *); |
187 | static void dumpcnfa(struct cnfa *, FILE *); |
188 | static void dumpcstate(int, struct cnfa *, FILE *); |
189 | #endif |
190 | /* === regc_cvec.c === */ |
191 | static struct cvec *newcvec(int, int); |
192 | static struct cvec *clearcvec(struct cvec *); |
193 | static void addchr(struct cvec *, chr); |
194 | static void addrange(struct cvec *, chr, chr); |
195 | static struct cvec *getcvec(struct vars *, int, int); |
196 | static void freecvec(struct cvec *); |
197 | |
198 | /* === regc_pg_locale.c === */ |
199 | static int pg_wc_isdigit(pg_wchar c); |
200 | static int pg_wc_isalpha(pg_wchar c); |
201 | static int pg_wc_isalnum(pg_wchar c); |
202 | static int pg_wc_isupper(pg_wchar c); |
203 | static int pg_wc_islower(pg_wchar c); |
204 | static int pg_wc_isgraph(pg_wchar c); |
205 | static int pg_wc_isprint(pg_wchar c); |
206 | static int pg_wc_ispunct(pg_wchar c); |
207 | static int pg_wc_isspace(pg_wchar c); |
208 | static pg_wchar pg_wc_toupper(pg_wchar c); |
209 | static pg_wchar pg_wc_tolower(pg_wchar c); |
210 | |
211 | /* === regc_locale.c === */ |
212 | static chr element(struct vars *, const chr *, const chr *); |
213 | static struct cvec *range(struct vars *, chr, chr, int); |
214 | static int before(chr, chr); |
215 | static struct cvec *eclass(struct vars *, chr, int); |
216 | static struct cvec *cclass(struct vars *, const chr *, const chr *, int); |
217 | static int cclass_column_index(struct colormap *, chr); |
218 | static struct cvec *allcases(struct vars *, chr); |
219 | static int cmp(const chr *, const chr *, size_t); |
220 | static int casecmp(const chr *, const chr *, size_t); |
221 | |
222 | |
223 | /* internal variables, bundled for easy passing around */ |
224 | struct vars |
225 | { |
226 | regex_t *re; |
227 | const chr *now; /* scan pointer into string */ |
228 | const chr *stop; /* end of string */ |
229 | const chr *savenow; /* saved now and stop for "subroutine call" */ |
230 | const chr *savestop; |
231 | int err; /* error code (0 if none) */ |
232 | int cflags; /* copy of compile flags */ |
233 | int lasttype; /* type of previous token */ |
234 | int nexttype; /* type of next token */ |
235 | chr nextvalue; /* value (if any) of next token */ |
236 | int lexcon; /* lexical context type (see lex.c) */ |
237 | int nsubexp; /* subexpression count */ |
238 | struct subre **subs; /* subRE pointer vector */ |
239 | size_t nsubs; /* length of vector */ |
240 | struct subre *sub10[10]; /* initial vector, enough for most */ |
241 | struct nfa *nfa; /* the NFA */ |
242 | struct colormap *cm; /* character color map */ |
243 | color nlcolor; /* color of newline */ |
244 | struct state *wordchrs; /* state in nfa holding word-char outarcs */ |
245 | struct subre *tree; /* subexpression tree */ |
246 | struct subre *treechain; /* all tree nodes allocated */ |
247 | struct subre *treefree; /* any free tree nodes */ |
248 | int ntree; /* number of tree nodes, plus one */ |
249 | struct cvec *cv; /* interface cvec */ |
250 | struct cvec *cv2; /* utility cvec */ |
251 | struct subre *lacons; /* lookaround-constraint vector */ |
252 | int nlacons; /* size of lacons[]; note that only slots |
253 | * numbered 1 .. nlacons-1 are used */ |
254 | size_t spaceused; /* approx. space used for compilation */ |
255 | }; |
256 | |
257 | /* parsing macros; most know that `v' is the struct vars pointer */ |
258 | #define NEXT() (next(v)) /* advance by one token */ |
259 | #define SEE(t) (v->nexttype == (t)) /* is next token this? */ |
260 | #define EAT(t) (SEE(t) && next(v)) /* if next is this, swallow it */ |
261 | #define VISERR(vv) ((vv)->err != 0) /* have we seen an error yet? */ |
262 | #define ISERR() VISERR(v) |
263 | #define VERR(vv,e) ((vv)->nexttype = EOS, \ |
264 | (vv)->err = ((vv)->err ? (vv)->err : (e))) |
265 | #define ERR(e) VERR(v, e) /* record an error */ |
266 | #define NOERR() {if (ISERR()) return;} /* if error seen, return */ |
267 | #define NOERRN() {if (ISERR()) return NULL;} /* NOERR with retval */ |
268 | #define NOERRZ() {if (ISERR()) return 0;} /* NOERR with retval */ |
269 | #define INSIST(c, e) do { if (!(c)) ERR(e); } while (0) /* error if c false */ |
270 | #define NOTE(b) (v->re->re_info |= (b)) /* note visible condition */ |
271 | #define EMPTYARC(x, y) newarc(v->nfa, EMPTY, 0, x, y) |
272 | |
273 | /* token type codes, some also used as NFA arc types */ |
274 | #define EMPTY 'n' /* no token present */ |
275 | #define EOS 'e' /* end of string */ |
276 | #define PLAIN 'p' /* ordinary character */ |
277 | #define DIGIT 'd' /* digit (in bound) */ |
278 | #define BACKREF 'b' /* back reference */ |
279 | #define COLLEL 'I' /* start of [. */ |
280 | #define ECLASS 'E' /* start of [= */ |
281 | #define CCLASS 'C' /* start of [: */ |
282 | #define END 'X' /* end of [. [= [: */ |
283 | #define RANGE 'R' /* - within [] which might be range delim. */ |
284 | #define LACON 'L' /* lookaround constraint subRE */ |
285 | #define AHEAD 'a' /* color-lookahead arc */ |
286 | #define BEHIND 'r' /* color-lookbehind arc */ |
287 | #define WBDRY 'w' /* word boundary constraint */ |
288 | #define NWBDRY 'W' /* non-word-boundary constraint */ |
289 | #define SBEGIN 'A' /* beginning of string (even if not BOL) */ |
290 | #define SEND 'Z' /* end of string (even if not EOL) */ |
291 | #define PREFER 'P' /* length preference */ |
292 | |
293 | /* is an arc colored, and hence on a color chain? */ |
294 | #define COLORED(a) \ |
295 | ((a)->type == PLAIN || (a)->type == AHEAD || (a)->type == BEHIND) |
296 | |
297 | |
298 | /* static function list */ |
299 | static const struct fns functions = { |
300 | rfree, /* regfree insides */ |
301 | rcancelrequested, /* check for cancel request */ |
302 | rstacktoodeep /* check for stack getting dangerously deep */ |
303 | }; |
304 | |
305 | |
306 | |
307 | /* |
308 | * pg_regcomp - compile regular expression |
309 | * |
310 | * Note: on failure, no resources remain allocated, so pg_regfree() |
311 | * need not be applied to re. |
312 | */ |
313 | int |
314 | pg_regcomp(regex_t *re, |
315 | const chr *string, |
316 | size_t len, |
317 | int flags, |
318 | Oid collation) |
319 | { |
320 | struct vars var; |
321 | struct vars *v = &var; |
322 | struct guts *g; |
323 | int i; |
324 | size_t j; |
325 | |
326 | #ifdef REG_DEBUG |
327 | FILE *debug = (flags & REG_PROGRESS) ? stdout : (FILE *) NULL; |
328 | #else |
329 | FILE *debug = (FILE *) NULL; |
330 | #endif |
331 | |
332 | #define CNOERR() { if (ISERR()) return freev(v, v->err); } |
333 | |
334 | /* sanity checks */ |
335 | |
336 | if (re == NULL || string == NULL) |
337 | return REG_INVARG; |
338 | if ((flags & REG_QUOTE) && |
339 | (flags & (REG_ADVANCED | REG_EXPANDED | REG_NEWLINE))) |
340 | return REG_INVARG; |
341 | if (!(flags & REG_EXTENDED) && (flags & REG_ADVF)) |
342 | return REG_INVARG; |
343 | |
344 | /* Initialize locale-dependent support */ |
345 | pg_set_regex_collation(collation); |
346 | |
347 | /* initial setup (after which freev() is callable) */ |
348 | v->re = re; |
349 | v->now = string; |
350 | v->stop = v->now + len; |
351 | v->savenow = v->savestop = NULL; |
352 | v->err = 0; |
353 | v->cflags = flags; |
354 | v->nsubexp = 0; |
355 | v->subs = v->sub10; |
356 | v->nsubs = 10; |
357 | for (j = 0; j < v->nsubs; j++) |
358 | v->subs[j] = NULL; |
359 | v->nfa = NULL; |
360 | v->cm = NULL; |
361 | v->nlcolor = COLORLESS; |
362 | v->wordchrs = NULL; |
363 | v->tree = NULL; |
364 | v->treechain = NULL; |
365 | v->treefree = NULL; |
366 | v->cv = NULL; |
367 | v->cv2 = NULL; |
368 | v->lacons = NULL; |
369 | v->nlacons = 0; |
370 | v->spaceused = 0; |
371 | re->re_magic = REMAGIC; |
372 | re->re_info = 0; /* bits get set during parse */ |
373 | re->re_csize = sizeof(chr); |
374 | re->re_collation = collation; |
375 | re->re_guts = NULL; |
376 | re->re_fns = VS(&functions); |
377 | |
378 | /* more complex setup, malloced things */ |
379 | re->re_guts = VS(MALLOC(sizeof(struct guts))); |
380 | if (re->re_guts == NULL) |
381 | return freev(v, REG_ESPACE); |
382 | g = (struct guts *) re->re_guts; |
383 | g->tree = NULL; |
384 | initcm(v, &g->cmap); |
385 | v->cm = &g->cmap; |
386 | g->lacons = NULL; |
387 | g->nlacons = 0; |
388 | ZAPCNFA(g->search); |
389 | v->nfa = newnfa(v, v->cm, (struct nfa *) NULL); |
390 | CNOERR(); |
391 | /* set up a reasonably-sized transient cvec for getcvec usage */ |
392 | v->cv = newcvec(100, 20); |
393 | if (v->cv == NULL) |
394 | return freev(v, REG_ESPACE); |
395 | |
396 | /* parsing */ |
397 | lexstart(v); /* also handles prefixes */ |
398 | if ((v->cflags & REG_NLSTOP) || (v->cflags & REG_NLANCH)) |
399 | { |
400 | /* assign newline a unique color */ |
401 | v->nlcolor = subcolor(v->cm, newline()); |
402 | okcolors(v->nfa, v->cm); |
403 | } |
404 | CNOERR(); |
405 | v->tree = parse(v, EOS, PLAIN, v->nfa->init, v->nfa->final); |
406 | assert(SEE(EOS)); /* even if error; ISERR() => SEE(EOS) */ |
407 | CNOERR(); |
408 | assert(v->tree != NULL); |
409 | |
410 | /* finish setup of nfa and its subre tree */ |
411 | specialcolors(v->nfa); |
412 | CNOERR(); |
413 | #ifdef REG_DEBUG |
414 | if (debug != NULL) |
415 | { |
416 | fprintf(debug, "\n\n\n========= RAW ==========\n" ); |
417 | dumpnfa(v->nfa, debug); |
418 | dumpst(v->tree, debug, 1); |
419 | } |
420 | #endif |
421 | optst(v, v->tree); |
422 | v->ntree = numst(v->tree, 1); |
423 | markst(v->tree); |
424 | cleanst(v); |
425 | #ifdef REG_DEBUG |
426 | if (debug != NULL) |
427 | { |
428 | fprintf(debug, "\n\n\n========= TREE FIXED ==========\n" ); |
429 | dumpst(v->tree, debug, 1); |
430 | } |
431 | #endif |
432 | |
433 | /* build compacted NFAs for tree and lacons */ |
434 | re->re_info |= nfatree(v, v->tree, debug); |
435 | CNOERR(); |
436 | assert(v->nlacons == 0 || v->lacons != NULL); |
437 | for (i = 1; i < v->nlacons; i++) |
438 | { |
439 | struct subre *lasub = &v->lacons[i]; |
440 | |
441 | #ifdef REG_DEBUG |
442 | if (debug != NULL) |
443 | fprintf(debug, "\n\n\n========= LA%d ==========\n" , i); |
444 | #endif |
445 | |
446 | /* Prepend .* to pattern if it's a lookbehind LACON */ |
447 | nfanode(v, lasub, !LATYPE_IS_AHEAD(lasub->subno), debug); |
448 | } |
449 | CNOERR(); |
450 | if (v->tree->flags & SHORTER) |
451 | NOTE(REG_USHORTEST); |
452 | |
453 | /* build compacted NFAs for tree, lacons, fast search */ |
454 | #ifdef REG_DEBUG |
455 | if (debug != NULL) |
456 | fprintf(debug, "\n\n\n========= SEARCH ==========\n" ); |
457 | #endif |
458 | /* can sacrifice main NFA now, so use it as work area */ |
459 | (DISCARD) optimize(v->nfa, debug); |
460 | CNOERR(); |
461 | makesearch(v, v->nfa); |
462 | CNOERR(); |
463 | compact(v->nfa, &g->search); |
464 | CNOERR(); |
465 | |
466 | /* looks okay, package it up */ |
467 | re->re_nsub = v->nsubexp; |
468 | v->re = NULL; /* freev no longer frees re */ |
469 | g->magic = GUTSMAGIC; |
470 | g->cflags = v->cflags; |
471 | g->info = re->re_info; |
472 | g->nsub = re->re_nsub; |
473 | g->tree = v->tree; |
474 | v->tree = NULL; |
475 | g->ntree = v->ntree; |
476 | g->compare = (v->cflags & REG_ICASE) ? casecmp : cmp; |
477 | g->lacons = v->lacons; |
478 | v->lacons = NULL; |
479 | g->nlacons = v->nlacons; |
480 | |
481 | #ifdef REG_DEBUG |
482 | if (flags & REG_DUMP) |
483 | dump(re, stdout); |
484 | #endif |
485 | |
486 | assert(v->err == 0); |
487 | return freev(v, 0); |
488 | } |
489 | |
490 | /* |
491 | * moresubs - enlarge subRE vector |
492 | */ |
493 | static void |
494 | moresubs(struct vars *v, |
495 | int wanted) /* want enough room for this one */ |
496 | { |
497 | struct subre **p; |
498 | size_t n; |
499 | |
500 | assert(wanted > 0 && (size_t) wanted >= v->nsubs); |
501 | n = (size_t) wanted * 3 / 2 + 1; |
502 | |
503 | if (v->subs == v->sub10) |
504 | { |
505 | p = (struct subre **) MALLOC(n * sizeof(struct subre *)); |
506 | if (p != NULL) |
507 | memcpy(VS(p), VS(v->subs), |
508 | v->nsubs * sizeof(struct subre *)); |
509 | } |
510 | else |
511 | p = (struct subre **) REALLOC(v->subs, n * sizeof(struct subre *)); |
512 | if (p == NULL) |
513 | { |
514 | ERR(REG_ESPACE); |
515 | return; |
516 | } |
517 | v->subs = p; |
518 | for (p = &v->subs[v->nsubs]; v->nsubs < n; p++, v->nsubs++) |
519 | *p = NULL; |
520 | assert(v->nsubs == n); |
521 | assert((size_t) wanted < v->nsubs); |
522 | } |
523 | |
524 | /* |
525 | * freev - free vars struct's substructures where necessary |
526 | * |
527 | * Optionally does error-number setting, and always returns error code |
528 | * (if any), to make error-handling code terser. |
529 | */ |
530 | static int |
531 | freev(struct vars *v, |
532 | int err) |
533 | { |
534 | if (v->re != NULL) |
535 | rfree(v->re); |
536 | if (v->subs != v->sub10) |
537 | FREE(v->subs); |
538 | if (v->nfa != NULL) |
539 | freenfa(v->nfa); |
540 | if (v->tree != NULL) |
541 | freesubre(v, v->tree); |
542 | if (v->treechain != NULL) |
543 | cleanst(v); |
544 | if (v->cv != NULL) |
545 | freecvec(v->cv); |
546 | if (v->cv2 != NULL) |
547 | freecvec(v->cv2); |
548 | if (v->lacons != NULL) |
549 | freelacons(v->lacons, v->nlacons); |
550 | ERR(err); /* nop if err==0 */ |
551 | |
552 | return v->err; |
553 | } |
554 | |
555 | /* |
556 | * makesearch - turn an NFA into a search NFA (implicit prepend of .*?) |
557 | * NFA must have been optimize()d already. |
558 | */ |
559 | static void |
560 | makesearch(struct vars *v, |
561 | struct nfa *nfa) |
562 | { |
563 | struct arc *a; |
564 | struct arc *b; |
565 | struct state *pre = nfa->pre; |
566 | struct state *s; |
567 | struct state *s2; |
568 | struct state *slist; |
569 | |
570 | /* no loops are needed if it's anchored */ |
571 | for (a = pre->outs; a != NULL; a = a->outchain) |
572 | { |
573 | assert(a->type == PLAIN); |
574 | if (a->co != nfa->bos[0] && a->co != nfa->bos[1]) |
575 | break; |
576 | } |
577 | if (a != NULL) |
578 | { |
579 | /* add implicit .* in front */ |
580 | rainbow(nfa, v->cm, PLAIN, COLORLESS, pre, pre); |
581 | |
582 | /* and ^* and \A* too -- not always necessary, but harmless */ |
583 | newarc(nfa, PLAIN, nfa->bos[0], pre, pre); |
584 | newarc(nfa, PLAIN, nfa->bos[1], pre, pre); |
585 | } |
586 | |
587 | /* |
588 | * Now here's the subtle part. Because many REs have no lookback |
589 | * constraints, often knowing when you were in the pre state tells you |
590 | * little; it's the next state(s) that are informative. But some of them |
591 | * may have other inarcs, i.e. it may be possible to make actual progress |
592 | * and then return to one of them. We must de-optimize such cases, |
593 | * splitting each such state into progress and no-progress states. |
594 | */ |
595 | |
596 | /* first, make a list of the states reachable from pre and elsewhere */ |
597 | slist = NULL; |
598 | for (a = pre->outs; a != NULL; a = a->outchain) |
599 | { |
600 | s = a->to; |
601 | for (b = s->ins; b != NULL; b = b->inchain) |
602 | { |
603 | if (b->from != pre) |
604 | break; |
605 | } |
606 | |
607 | /* |
608 | * We want to mark states as being in the list already by having non |
609 | * NULL tmp fields, but we can't just store the old slist value in tmp |
610 | * because that doesn't work for the first such state. Instead, the |
611 | * first list entry gets its own address in tmp. |
612 | */ |
613 | if (b != NULL && s->tmp == NULL) |
614 | { |
615 | s->tmp = (slist != NULL) ? slist : s; |
616 | slist = s; |
617 | } |
618 | } |
619 | |
620 | /* do the splits */ |
621 | for (s = slist; s != NULL; s = s2) |
622 | { |
623 | s2 = newstate(nfa); |
624 | NOERR(); |
625 | copyouts(nfa, s, s2); |
626 | NOERR(); |
627 | for (a = s->ins; a != NULL; a = b) |
628 | { |
629 | b = a->inchain; |
630 | if (a->from != pre) |
631 | { |
632 | cparc(nfa, a, a->from, s2); |
633 | freearc(nfa, a); |
634 | } |
635 | } |
636 | s2 = (s->tmp != s) ? s->tmp : NULL; |
637 | s->tmp = NULL; /* clean up while we're at it */ |
638 | } |
639 | } |
640 | |
641 | /* |
642 | * parse - parse an RE |
643 | * |
644 | * This is actually just the top level, which parses a bunch of branches |
645 | * tied together with '|'. They appear in the tree as the left children |
646 | * of a chain of '|' subres. |
647 | */ |
648 | static struct subre * |
649 | parse(struct vars *v, |
650 | int stopper, /* EOS or ')' */ |
651 | int type, /* LACON (lookaround subRE) or PLAIN */ |
652 | struct state *init, /* initial state */ |
653 | struct state *final) /* final state */ |
654 | { |
655 | struct state *left; /* scaffolding for branch */ |
656 | struct state *right; |
657 | struct subre *branches; /* top level */ |
658 | struct subre *branch; /* current branch */ |
659 | struct subre *t; /* temporary */ |
660 | int firstbranch; /* is this the first branch? */ |
661 | |
662 | assert(stopper == ')' || stopper == EOS); |
663 | |
664 | branches = subre(v, '|', LONGER, init, final); |
665 | NOERRN(); |
666 | branch = branches; |
667 | firstbranch = 1; |
668 | do |
669 | { /* a branch */ |
670 | if (!firstbranch) |
671 | { |
672 | /* need a place to hang it */ |
673 | branch->right = subre(v, '|', LONGER, init, final); |
674 | NOERRN(); |
675 | branch = branch->right; |
676 | } |
677 | firstbranch = 0; |
678 | left = newstate(v->nfa); |
679 | right = newstate(v->nfa); |
680 | NOERRN(); |
681 | EMPTYARC(init, left); |
682 | EMPTYARC(right, final); |
683 | NOERRN(); |
684 | branch->left = parsebranch(v, stopper, type, left, right, 0); |
685 | NOERRN(); |
686 | branch->flags |= UP(branch->flags | branch->left->flags); |
687 | if ((branch->flags & ~branches->flags) != 0) /* new flags */ |
688 | for (t = branches; t != branch; t = t->right) |
689 | t->flags |= branch->flags; |
690 | } while (EAT('|')); |
691 | assert(SEE(stopper) || SEE(EOS)); |
692 | |
693 | if (!SEE(stopper)) |
694 | { |
695 | assert(stopper == ')' && SEE(EOS)); |
696 | ERR(REG_EPAREN); |
697 | } |
698 | |
699 | /* optimize out simple cases */ |
700 | if (branch == branches) |
701 | { /* only one branch */ |
702 | assert(branch->right == NULL); |
703 | t = branch->left; |
704 | branch->left = NULL; |
705 | freesubre(v, branches); |
706 | branches = t; |
707 | } |
708 | else if (!MESSY(branches->flags)) |
709 | { /* no interesting innards */ |
710 | freesubre(v, branches->left); |
711 | branches->left = NULL; |
712 | freesubre(v, branches->right); |
713 | branches->right = NULL; |
714 | branches->op = '='; |
715 | } |
716 | |
717 | return branches; |
718 | } |
719 | |
720 | /* |
721 | * parsebranch - parse one branch of an RE |
722 | * |
723 | * This mostly manages concatenation, working closely with parseqatom(). |
724 | * Concatenated things are bundled up as much as possible, with separate |
725 | * ',' nodes introduced only when necessary due to substructure. |
726 | */ |
727 | static struct subre * |
728 | parsebranch(struct vars *v, |
729 | int stopper, /* EOS or ')' */ |
730 | int type, /* LACON (lookaround subRE) or PLAIN */ |
731 | struct state *left, /* leftmost state */ |
732 | struct state *right, /* rightmost state */ |
733 | int partial) /* is this only part of a branch? */ |
734 | { |
735 | struct state *lp; /* left end of current construct */ |
736 | int seencontent; /* is there anything in this branch yet? */ |
737 | struct subre *t; |
738 | |
739 | lp = left; |
740 | seencontent = 0; |
741 | t = subre(v, '=', 0, left, right); /* op '=' is tentative */ |
742 | NOERRN(); |
743 | while (!SEE('|') && !SEE(stopper) && !SEE(EOS)) |
744 | { |
745 | if (seencontent) |
746 | { /* implicit concat operator */ |
747 | lp = newstate(v->nfa); |
748 | NOERRN(); |
749 | moveins(v->nfa, right, lp); |
750 | } |
751 | seencontent = 1; |
752 | |
753 | /* NB, recursion in parseqatom() may swallow rest of branch */ |
754 | parseqatom(v, stopper, type, lp, right, t); |
755 | NOERRN(); |
756 | } |
757 | |
758 | if (!seencontent) |
759 | { /* empty branch */ |
760 | if (!partial) |
761 | NOTE(REG_UUNSPEC); |
762 | assert(lp == left); |
763 | EMPTYARC(left, right); |
764 | } |
765 | |
766 | return t; |
767 | } |
768 | |
769 | /* |
770 | * parseqatom - parse one quantified atom or constraint of an RE |
771 | * |
772 | * The bookkeeping near the end cooperates very closely with parsebranch(); |
773 | * in particular, it contains a recursion that can involve parsing the rest |
774 | * of the branch, making this function's name somewhat inaccurate. |
775 | */ |
776 | static void |
777 | parseqatom(struct vars *v, |
778 | int stopper, /* EOS or ')' */ |
779 | int type, /* LACON (lookaround subRE) or PLAIN */ |
780 | struct state *lp, /* left state to hang it on */ |
781 | struct state *rp, /* right state to hang it on */ |
782 | struct subre *top) /* subtree top */ |
783 | { |
784 | struct state *s; /* temporaries for new states */ |
785 | struct state *s2; |
786 | |
787 | #define ARCV(t, val) newarc(v->nfa, t, val, lp, rp) |
788 | int m, |
789 | n; |
790 | struct subre *atom; /* atom's subtree */ |
791 | struct subre *t; |
792 | int cap; /* capturing parens? */ |
793 | int latype; /* lookaround constraint type */ |
794 | int subno; /* capturing-parens or backref number */ |
795 | int atomtype; |
796 | int qprefer; /* quantifier short/long preference */ |
797 | int f; |
798 | struct subre **atomp; /* where the pointer to atom is */ |
799 | |
800 | /* initial bookkeeping */ |
801 | atom = NULL; |
802 | assert(lp->nouts == 0); /* must string new code */ |
803 | assert(rp->nins == 0); /* between lp and rp */ |
804 | subno = 0; /* just to shut lint up */ |
805 | |
806 | /* an atom or constraint... */ |
807 | atomtype = v->nexttype; |
808 | switch (atomtype) |
809 | { |
810 | /* first, constraints, which end by returning */ |
811 | case '^': |
812 | ARCV('^', 1); |
813 | if (v->cflags & REG_NLANCH) |
814 | ARCV(BEHIND, v->nlcolor); |
815 | NEXT(); |
816 | return; |
817 | break; |
818 | case '$': |
819 | ARCV('$', 1); |
820 | if (v->cflags & REG_NLANCH) |
821 | ARCV(AHEAD, v->nlcolor); |
822 | NEXT(); |
823 | return; |
824 | break; |
825 | case SBEGIN: |
826 | ARCV('^', 1); /* BOL */ |
827 | ARCV('^', 0); /* or BOS */ |
828 | NEXT(); |
829 | return; |
830 | break; |
831 | case SEND: |
832 | ARCV('$', 1); /* EOL */ |
833 | ARCV('$', 0); /* or EOS */ |
834 | NEXT(); |
835 | return; |
836 | break; |
837 | case '<': |
838 | wordchrs(v); /* does NEXT() */ |
839 | s = newstate(v->nfa); |
840 | NOERR(); |
841 | nonword(v, BEHIND, lp, s); |
842 | word(v, AHEAD, s, rp); |
843 | return; |
844 | break; |
845 | case '>': |
846 | wordchrs(v); /* does NEXT() */ |
847 | s = newstate(v->nfa); |
848 | NOERR(); |
849 | word(v, BEHIND, lp, s); |
850 | nonword(v, AHEAD, s, rp); |
851 | return; |
852 | break; |
853 | case WBDRY: |
854 | wordchrs(v); /* does NEXT() */ |
855 | s = newstate(v->nfa); |
856 | NOERR(); |
857 | nonword(v, BEHIND, lp, s); |
858 | word(v, AHEAD, s, rp); |
859 | s = newstate(v->nfa); |
860 | NOERR(); |
861 | word(v, BEHIND, lp, s); |
862 | nonword(v, AHEAD, s, rp); |
863 | return; |
864 | break; |
865 | case NWBDRY: |
866 | wordchrs(v); /* does NEXT() */ |
867 | s = newstate(v->nfa); |
868 | NOERR(); |
869 | word(v, BEHIND, lp, s); |
870 | word(v, AHEAD, s, rp); |
871 | s = newstate(v->nfa); |
872 | NOERR(); |
873 | nonword(v, BEHIND, lp, s); |
874 | nonword(v, AHEAD, s, rp); |
875 | return; |
876 | break; |
877 | case LACON: /* lookaround constraint */ |
878 | latype = v->nextvalue; |
879 | NEXT(); |
880 | s = newstate(v->nfa); |
881 | s2 = newstate(v->nfa); |
882 | NOERR(); |
883 | t = parse(v, ')', LACON, s, s2); |
884 | freesubre(v, t); /* internal structure irrelevant */ |
885 | NOERR(); |
886 | assert(SEE(')')); |
887 | NEXT(); |
888 | processlacon(v, s, s2, latype, lp, rp); |
889 | return; |
890 | break; |
891 | /* then errors, to get them out of the way */ |
892 | case '*': |
893 | case '+': |
894 | case '?': |
895 | case '{': |
896 | ERR(REG_BADRPT); |
897 | return; |
898 | break; |
899 | default: |
900 | ERR(REG_ASSERT); |
901 | return; |
902 | break; |
903 | /* then plain characters, and minor variants on that theme */ |
904 | case ')': /* unbalanced paren */ |
905 | if ((v->cflags & REG_ADVANCED) != REG_EXTENDED) |
906 | { |
907 | ERR(REG_EPAREN); |
908 | return; |
909 | } |
910 | /* legal in EREs due to specification botch */ |
911 | NOTE(REG_UPBOTCH); |
912 | /* fall through into case PLAIN */ |
913 | /* FALLTHROUGH */ |
914 | case PLAIN: |
915 | onechr(v, v->nextvalue, lp, rp); |
916 | okcolors(v->nfa, v->cm); |
917 | NOERR(); |
918 | NEXT(); |
919 | break; |
920 | case '[': |
921 | if (v->nextvalue == 1) |
922 | bracket(v, lp, rp); |
923 | else |
924 | cbracket(v, lp, rp); |
925 | assert(SEE(']') || ISERR()); |
926 | NEXT(); |
927 | break; |
928 | case '.': |
929 | rainbow(v->nfa, v->cm, PLAIN, |
930 | (v->cflags & REG_NLSTOP) ? v->nlcolor : COLORLESS, |
931 | lp, rp); |
932 | NEXT(); |
933 | break; |
934 | /* and finally the ugly stuff */ |
935 | case '(': /* value flags as capturing or non */ |
936 | cap = (type == LACON) ? 0 : v->nextvalue; |
937 | if (cap) |
938 | { |
939 | v->nsubexp++; |
940 | subno = v->nsubexp; |
941 | if ((size_t) subno >= v->nsubs) |
942 | moresubs(v, subno); |
943 | assert((size_t) subno < v->nsubs); |
944 | } |
945 | else |
946 | atomtype = PLAIN; /* something that's not '(' */ |
947 | NEXT(); |
948 | /* need new endpoints because tree will contain pointers */ |
949 | s = newstate(v->nfa); |
950 | s2 = newstate(v->nfa); |
951 | NOERR(); |
952 | EMPTYARC(lp, s); |
953 | EMPTYARC(s2, rp); |
954 | NOERR(); |
955 | atom = parse(v, ')', type, s, s2); |
956 | assert(SEE(')') || ISERR()); |
957 | NEXT(); |
958 | NOERR(); |
959 | if (cap) |
960 | { |
961 | v->subs[subno] = atom; |
962 | t = subre(v, '(', atom->flags | CAP, lp, rp); |
963 | NOERR(); |
964 | t->subno = subno; |
965 | t->left = atom; |
966 | atom = t; |
967 | } |
968 | /* postpone everything else pending possible {0} */ |
969 | break; |
970 | case BACKREF: /* the Feature From The Black Lagoon */ |
971 | INSIST(type != LACON, REG_ESUBREG); |
972 | INSIST(v->nextvalue < v->nsubs, REG_ESUBREG); |
973 | INSIST(v->subs[v->nextvalue] != NULL, REG_ESUBREG); |
974 | NOERR(); |
975 | assert(v->nextvalue > 0); |
976 | atom = subre(v, 'b', BACKR, lp, rp); |
977 | NOERR(); |
978 | subno = v->nextvalue; |
979 | atom->subno = subno; |
980 | EMPTYARC(lp, rp); /* temporarily, so there's something */ |
981 | NEXT(); |
982 | break; |
983 | } |
984 | |
985 | /* ...and an atom may be followed by a quantifier */ |
986 | switch (v->nexttype) |
987 | { |
988 | case '*': |
989 | m = 0; |
990 | n = DUPINF; |
991 | qprefer = (v->nextvalue) ? LONGER : SHORTER; |
992 | NEXT(); |
993 | break; |
994 | case '+': |
995 | m = 1; |
996 | n = DUPINF; |
997 | qprefer = (v->nextvalue) ? LONGER : SHORTER; |
998 | NEXT(); |
999 | break; |
1000 | case '?': |
1001 | m = 0; |
1002 | n = 1; |
1003 | qprefer = (v->nextvalue) ? LONGER : SHORTER; |
1004 | NEXT(); |
1005 | break; |
1006 | case '{': |
1007 | NEXT(); |
1008 | m = scannum(v); |
1009 | if (EAT(',')) |
1010 | { |
1011 | if (SEE(DIGIT)) |
1012 | n = scannum(v); |
1013 | else |
1014 | n = DUPINF; |
1015 | if (m > n) |
1016 | { |
1017 | ERR(REG_BADBR); |
1018 | return; |
1019 | } |
1020 | /* {m,n} exercises preference, even if it's {m,m} */ |
1021 | qprefer = (v->nextvalue) ? LONGER : SHORTER; |
1022 | } |
1023 | else |
1024 | { |
1025 | n = m; |
1026 | /* {m} passes operand's preference through */ |
1027 | qprefer = 0; |
1028 | } |
1029 | if (!SEE('}')) |
1030 | { /* catches errors too */ |
1031 | ERR(REG_BADBR); |
1032 | return; |
1033 | } |
1034 | NEXT(); |
1035 | break; |
1036 | default: /* no quantifier */ |
1037 | m = n = 1; |
1038 | qprefer = 0; |
1039 | break; |
1040 | } |
1041 | |
1042 | /* annoying special case: {0} or {0,0} cancels everything */ |
1043 | if (m == 0 && n == 0) |
1044 | { |
1045 | if (atom != NULL) |
1046 | freesubre(v, atom); |
1047 | if (atomtype == '(') |
1048 | v->subs[subno] = NULL; |
1049 | delsub(v->nfa, lp, rp); |
1050 | EMPTYARC(lp, rp); |
1051 | return; |
1052 | } |
1053 | |
1054 | /* if not a messy case, avoid hard part */ |
1055 | assert(!MESSY(top->flags)); |
1056 | f = top->flags | qprefer | ((atom != NULL) ? atom->flags : 0); |
1057 | if (atomtype != '(' && atomtype != BACKREF && !MESSY(UP(f))) |
1058 | { |
1059 | if (!(m == 1 && n == 1)) |
1060 | repeat(v, lp, rp, m, n); |
1061 | if (atom != NULL) |
1062 | freesubre(v, atom); |
1063 | top->flags = f; |
1064 | return; |
1065 | } |
1066 | |
1067 | /* |
1068 | * hard part: something messy |
1069 | * |
1070 | * That is, capturing parens, back reference, short/long clash, or an atom |
1071 | * with substructure containing one of those. |
1072 | */ |
1073 | |
1074 | /* now we'll need a subre for the contents even if they're boring */ |
1075 | if (atom == NULL) |
1076 | { |
1077 | atom = subre(v, '=', 0, lp, rp); |
1078 | NOERR(); |
1079 | } |
1080 | |
1081 | /*---------- |
1082 | * Prepare a general-purpose state skeleton. |
1083 | * |
1084 | * In the no-backrefs case, we want this: |
1085 | * |
1086 | * [lp] ---> [s] ---prefix---> [begin] ---atom---> [end] ---rest---> [rp] |
1087 | * |
1088 | * where prefix is some repetitions of atom. In the general case we need |
1089 | * |
1090 | * [lp] ---> [s] ---iterator---> [s2] ---rest---> [rp] |
1091 | * |
1092 | * where the iterator wraps around [begin] ---atom---> [end] |
1093 | * |
1094 | * We make the s state here for both cases; s2 is made below if needed |
1095 | *---------- |
1096 | */ |
1097 | s = newstate(v->nfa); /* first, new endpoints for the atom */ |
1098 | s2 = newstate(v->nfa); |
1099 | NOERR(); |
1100 | moveouts(v->nfa, lp, s); |
1101 | moveins(v->nfa, rp, s2); |
1102 | NOERR(); |
1103 | atom->begin = s; |
1104 | atom->end = s2; |
1105 | s = newstate(v->nfa); /* set up starting state */ |
1106 | NOERR(); |
1107 | EMPTYARC(lp, s); |
1108 | NOERR(); |
1109 | |
1110 | /* break remaining subRE into x{...} and what follows */ |
1111 | t = subre(v, '.', COMBINE(qprefer, atom->flags), lp, rp); |
1112 | NOERR(); |
1113 | t->left = atom; |
1114 | atomp = &t->left; |
1115 | |
1116 | /* here we should recurse... but we must postpone that to the end */ |
1117 | |
1118 | /* split top into prefix and remaining */ |
1119 | assert(top->op == '=' && top->left == NULL && top->right == NULL); |
1120 | top->left = subre(v, '=', top->flags, top->begin, lp); |
1121 | NOERR(); |
1122 | top->op = '.'; |
1123 | top->right = t; |
1124 | |
1125 | /* if it's a backref, now is the time to replicate the subNFA */ |
1126 | if (atomtype == BACKREF) |
1127 | { |
1128 | assert(atom->begin->nouts == 1); /* just the EMPTY */ |
1129 | delsub(v->nfa, atom->begin, atom->end); |
1130 | assert(v->subs[subno] != NULL); |
1131 | |
1132 | /* |
1133 | * And here's why the recursion got postponed: it must wait until the |
1134 | * skeleton is filled in, because it may hit a backref that wants to |
1135 | * copy the filled-in skeleton. |
1136 | */ |
1137 | dupnfa(v->nfa, v->subs[subno]->begin, v->subs[subno]->end, |
1138 | atom->begin, atom->end); |
1139 | NOERR(); |
1140 | } |
1141 | |
1142 | /* |
1143 | * It's quantifier time. If the atom is just a backref, we'll let it deal |
1144 | * with quantifiers internally. |
1145 | */ |
1146 | if (atomtype == BACKREF) |
1147 | { |
1148 | /* special case: backrefs have internal quantifiers */ |
1149 | EMPTYARC(s, atom->begin); /* empty prefix */ |
1150 | /* just stuff everything into atom */ |
1151 | repeat(v, atom->begin, atom->end, m, n); |
1152 | atom->min = (short) m; |
1153 | atom->max = (short) n; |
1154 | atom->flags |= COMBINE(qprefer, atom->flags); |
1155 | /* rest of branch can be strung starting from atom->end */ |
1156 | s2 = atom->end; |
1157 | } |
1158 | else if (m == 1 && n == 1 && |
1159 | (qprefer == 0 || |
1160 | (atom->flags & (LONGER | SHORTER | MIXED)) == 0 || |
1161 | qprefer == (atom->flags & (LONGER | SHORTER | MIXED)))) |
1162 | { |
1163 | /* no/vacuous quantifier: done */ |
1164 | EMPTYARC(s, atom->begin); /* empty prefix */ |
1165 | /* rest of branch can be strung starting from atom->end */ |
1166 | s2 = atom->end; |
1167 | } |
1168 | else if (m > 0 && !(atom->flags & BACKR)) |
1169 | { |
1170 | /* |
1171 | * If there's no backrefs involved, we can turn x{m,n} into |
1172 | * x{m-1,n-1}x, with capturing parens in only the second x. This is |
1173 | * valid because we only care about capturing matches from the final |
1174 | * iteration of the quantifier. It's a win because we can implement |
1175 | * the backref-free left side as a plain DFA node, since we don't |
1176 | * really care where its submatches are. |
1177 | */ |
1178 | dupnfa(v->nfa, atom->begin, atom->end, s, atom->begin); |
1179 | assert(m >= 1 && m != DUPINF && n >= 1); |
1180 | repeat(v, s, atom->begin, m - 1, (n == DUPINF) ? n : n - 1); |
1181 | f = COMBINE(qprefer, atom->flags); |
1182 | t = subre(v, '.', f, s, atom->end); /* prefix and atom */ |
1183 | NOERR(); |
1184 | t->left = subre(v, '=', PREF(f), s, atom->begin); |
1185 | NOERR(); |
1186 | t->right = atom; |
1187 | *atomp = t; |
1188 | /* rest of branch can be strung starting from atom->end */ |
1189 | s2 = atom->end; |
1190 | } |
1191 | else |
1192 | { |
1193 | /* general case: need an iteration node */ |
1194 | s2 = newstate(v->nfa); |
1195 | NOERR(); |
1196 | moveouts(v->nfa, atom->end, s2); |
1197 | NOERR(); |
1198 | dupnfa(v->nfa, atom->begin, atom->end, s, s2); |
1199 | repeat(v, s, s2, m, n); |
1200 | f = COMBINE(qprefer, atom->flags); |
1201 | t = subre(v, '*', f, s, s2); |
1202 | NOERR(); |
1203 | t->min = (short) m; |
1204 | t->max = (short) n; |
1205 | t->left = atom; |
1206 | *atomp = t; |
1207 | /* rest of branch is to be strung from iteration's end state */ |
1208 | } |
1209 | |
1210 | /* and finally, look after that postponed recursion */ |
1211 | t = top->right; |
1212 | if (!(SEE('|') || SEE(stopper) || SEE(EOS))) |
1213 | t->right = parsebranch(v, stopper, type, s2, rp, 1); |
1214 | else |
1215 | { |
1216 | EMPTYARC(s2, rp); |
1217 | t->right = subre(v, '=', 0, s2, rp); |
1218 | } |
1219 | NOERR(); |
1220 | assert(SEE('|') || SEE(stopper) || SEE(EOS)); |
1221 | t->flags |= COMBINE(t->flags, t->right->flags); |
1222 | top->flags |= COMBINE(top->flags, t->flags); |
1223 | } |
1224 | |
1225 | /* |
1226 | * nonword - generate arcs for non-word-character ahead or behind |
1227 | */ |
1228 | static void |
1229 | nonword(struct vars *v, |
1230 | int dir, /* AHEAD or BEHIND */ |
1231 | struct state *lp, |
1232 | struct state *rp) |
1233 | { |
1234 | int anchor = (dir == AHEAD) ? '$' : '^'; |
1235 | |
1236 | assert(dir == AHEAD || dir == BEHIND); |
1237 | newarc(v->nfa, anchor, 1, lp, rp); |
1238 | newarc(v->nfa, anchor, 0, lp, rp); |
1239 | colorcomplement(v->nfa, v->cm, dir, v->wordchrs, lp, rp); |
1240 | /* (no need for special attention to \n) */ |
1241 | } |
1242 | |
1243 | /* |
1244 | * word - generate arcs for word character ahead or behind |
1245 | */ |
1246 | static void |
1247 | word(struct vars *v, |
1248 | int dir, /* AHEAD or BEHIND */ |
1249 | struct state *lp, |
1250 | struct state *rp) |
1251 | { |
1252 | assert(dir == AHEAD || dir == BEHIND); |
1253 | cloneouts(v->nfa, v->wordchrs, lp, rp, dir); |
1254 | /* (no need for special attention to \n) */ |
1255 | } |
1256 | |
1257 | /* |
1258 | * scannum - scan a number |
1259 | */ |
1260 | static int /* value, <= DUPMAX */ |
1261 | scannum(struct vars *v) |
1262 | { |
1263 | int n = 0; |
1264 | |
1265 | while (SEE(DIGIT) && n < DUPMAX) |
1266 | { |
1267 | n = n * 10 + v->nextvalue; |
1268 | NEXT(); |
1269 | } |
1270 | if (SEE(DIGIT) || n > DUPMAX) |
1271 | { |
1272 | ERR(REG_BADBR); |
1273 | return 0; |
1274 | } |
1275 | return n; |
1276 | } |
1277 | |
1278 | /* |
1279 | * repeat - replicate subNFA for quantifiers |
1280 | * |
1281 | * The sub-NFA strung from lp to rp is modified to represent m to n |
1282 | * repetitions of its initial contents. |
1283 | * |
1284 | * The duplication sequences used here are chosen carefully so that any |
1285 | * pointers starting out pointing into the subexpression end up pointing into |
1286 | * the last occurrence. (Note that it may not be strung between the same |
1287 | * left and right end states, however!) This used to be important for the |
1288 | * subRE tree, although the important bits are now handled by the in-line |
1289 | * code in parse(), and when this is called, it doesn't matter any more. |
1290 | */ |
1291 | static void |
1292 | repeat(struct vars *v, |
1293 | struct state *lp, |
1294 | struct state *rp, |
1295 | int m, |
1296 | int n) |
1297 | { |
1298 | #define SOME 2 |
1299 | #define INF 3 |
1300 | #define PAIR(x, y) ((x)*4 + (y)) |
1301 | #define REDUCE(x) ( ((x) == DUPINF) ? INF : (((x) > 1) ? SOME : (x)) ) |
1302 | const int rm = REDUCE(m); |
1303 | const int rn = REDUCE(n); |
1304 | struct state *s; |
1305 | struct state *s2; |
1306 | |
1307 | switch (PAIR(rm, rn)) |
1308 | { |
1309 | case PAIR(0, 0): /* empty string */ |
1310 | delsub(v->nfa, lp, rp); |
1311 | EMPTYARC(lp, rp); |
1312 | break; |
1313 | case PAIR(0, 1): /* do as x| */ |
1314 | EMPTYARC(lp, rp); |
1315 | break; |
1316 | case PAIR(0, SOME): /* do as x{1,n}| */ |
1317 | repeat(v, lp, rp, 1, n); |
1318 | NOERR(); |
1319 | EMPTYARC(lp, rp); |
1320 | break; |
1321 | case PAIR(0, INF): /* loop x around */ |
1322 | s = newstate(v->nfa); |
1323 | NOERR(); |
1324 | moveouts(v->nfa, lp, s); |
1325 | moveins(v->nfa, rp, s); |
1326 | EMPTYARC(lp, s); |
1327 | EMPTYARC(s, rp); |
1328 | break; |
1329 | case PAIR(1, 1): /* no action required */ |
1330 | break; |
1331 | case PAIR(1, SOME): /* do as x{0,n-1}x = (x{1,n-1}|)x */ |
1332 | s = newstate(v->nfa); |
1333 | NOERR(); |
1334 | moveouts(v->nfa, lp, s); |
1335 | dupnfa(v->nfa, s, rp, lp, s); |
1336 | NOERR(); |
1337 | repeat(v, lp, s, 1, n - 1); |
1338 | NOERR(); |
1339 | EMPTYARC(lp, s); |
1340 | break; |
1341 | case PAIR(1, INF): /* add loopback arc */ |
1342 | s = newstate(v->nfa); |
1343 | s2 = newstate(v->nfa); |
1344 | NOERR(); |
1345 | moveouts(v->nfa, lp, s); |
1346 | moveins(v->nfa, rp, s2); |
1347 | EMPTYARC(lp, s); |
1348 | EMPTYARC(s2, rp); |
1349 | EMPTYARC(s2, s); |
1350 | break; |
1351 | case PAIR(SOME, SOME): /* do as x{m-1,n-1}x */ |
1352 | s = newstate(v->nfa); |
1353 | NOERR(); |
1354 | moveouts(v->nfa, lp, s); |
1355 | dupnfa(v->nfa, s, rp, lp, s); |
1356 | NOERR(); |
1357 | repeat(v, lp, s, m - 1, n - 1); |
1358 | break; |
1359 | case PAIR(SOME, INF): /* do as x{m-1,}x */ |
1360 | s = newstate(v->nfa); |
1361 | NOERR(); |
1362 | moveouts(v->nfa, lp, s); |
1363 | dupnfa(v->nfa, s, rp, lp, s); |
1364 | NOERR(); |
1365 | repeat(v, lp, s, m - 1, n); |
1366 | break; |
1367 | default: |
1368 | ERR(REG_ASSERT); |
1369 | break; |
1370 | } |
1371 | } |
1372 | |
1373 | /* |
1374 | * bracket - handle non-complemented bracket expression |
1375 | * Also called from cbracket for complemented bracket expressions. |
1376 | */ |
1377 | static void |
1378 | bracket(struct vars *v, |
1379 | struct state *lp, |
1380 | struct state *rp) |
1381 | { |
1382 | assert(SEE('[')); |
1383 | NEXT(); |
1384 | while (!SEE(']') && !SEE(EOS)) |
1385 | brackpart(v, lp, rp); |
1386 | assert(SEE(']') || ISERR()); |
1387 | okcolors(v->nfa, v->cm); |
1388 | } |
1389 | |
1390 | /* |
1391 | * cbracket - handle complemented bracket expression |
1392 | * We do it by calling bracket() with dummy endpoints, and then complementing |
1393 | * the result. The alternative would be to invoke rainbow(), and then delete |
1394 | * arcs as the b.e. is seen... but that gets messy. |
1395 | */ |
1396 | static void |
1397 | cbracket(struct vars *v, |
1398 | struct state *lp, |
1399 | struct state *rp) |
1400 | { |
1401 | struct state *left = newstate(v->nfa); |
1402 | struct state *right = newstate(v->nfa); |
1403 | |
1404 | NOERR(); |
1405 | bracket(v, left, right); |
1406 | if (v->cflags & REG_NLSTOP) |
1407 | newarc(v->nfa, PLAIN, v->nlcolor, left, right); |
1408 | NOERR(); |
1409 | |
1410 | assert(lp->nouts == 0); /* all outarcs will be ours */ |
1411 | |
1412 | /* |
1413 | * Easy part of complementing, and all there is to do since the MCCE code |
1414 | * was removed. |
1415 | */ |
1416 | colorcomplement(v->nfa, v->cm, PLAIN, left, lp, rp); |
1417 | NOERR(); |
1418 | dropstate(v->nfa, left); |
1419 | assert(right->nins == 0); |
1420 | freestate(v->nfa, right); |
1421 | } |
1422 | |
1423 | /* |
1424 | * brackpart - handle one item (or range) within a bracket expression |
1425 | */ |
1426 | static void |
1427 | brackpart(struct vars *v, |
1428 | struct state *lp, |
1429 | struct state *rp) |
1430 | { |
1431 | chr startc; |
1432 | chr endc; |
1433 | struct cvec *cv; |
1434 | const chr *startp; |
1435 | const chr *endp; |
1436 | chr c[1]; |
1437 | |
1438 | /* parse something, get rid of special cases, take shortcuts */ |
1439 | switch (v->nexttype) |
1440 | { |
1441 | case RANGE: /* a-b-c or other botch */ |
1442 | ERR(REG_ERANGE); |
1443 | return; |
1444 | break; |
1445 | case PLAIN: |
1446 | c[0] = v->nextvalue; |
1447 | NEXT(); |
1448 | /* shortcut for ordinary chr (not range) */ |
1449 | if (!SEE(RANGE)) |
1450 | { |
1451 | onechr(v, c[0], lp, rp); |
1452 | return; |
1453 | } |
1454 | startc = element(v, c, c + 1); |
1455 | NOERR(); |
1456 | break; |
1457 | case COLLEL: |
1458 | startp = v->now; |
1459 | endp = scanplain(v); |
1460 | INSIST(startp < endp, REG_ECOLLATE); |
1461 | NOERR(); |
1462 | startc = element(v, startp, endp); |
1463 | NOERR(); |
1464 | break; |
1465 | case ECLASS: |
1466 | startp = v->now; |
1467 | endp = scanplain(v); |
1468 | INSIST(startp < endp, REG_ECOLLATE); |
1469 | NOERR(); |
1470 | startc = element(v, startp, endp); |
1471 | NOERR(); |
1472 | cv = eclass(v, startc, (v->cflags & REG_ICASE)); |
1473 | NOERR(); |
1474 | subcolorcvec(v, cv, lp, rp); |
1475 | return; |
1476 | break; |
1477 | case CCLASS: |
1478 | startp = v->now; |
1479 | endp = scanplain(v); |
1480 | INSIST(startp < endp, REG_ECTYPE); |
1481 | NOERR(); |
1482 | cv = cclass(v, startp, endp, (v->cflags & REG_ICASE)); |
1483 | NOERR(); |
1484 | subcolorcvec(v, cv, lp, rp); |
1485 | return; |
1486 | break; |
1487 | default: |
1488 | ERR(REG_ASSERT); |
1489 | return; |
1490 | break; |
1491 | } |
1492 | |
1493 | if (SEE(RANGE)) |
1494 | { |
1495 | NEXT(); |
1496 | switch (v->nexttype) |
1497 | { |
1498 | case PLAIN: |
1499 | case RANGE: |
1500 | c[0] = v->nextvalue; |
1501 | NEXT(); |
1502 | endc = element(v, c, c + 1); |
1503 | NOERR(); |
1504 | break; |
1505 | case COLLEL: |
1506 | startp = v->now; |
1507 | endp = scanplain(v); |
1508 | INSIST(startp < endp, REG_ECOLLATE); |
1509 | NOERR(); |
1510 | endc = element(v, startp, endp); |
1511 | NOERR(); |
1512 | break; |
1513 | default: |
1514 | ERR(REG_ERANGE); |
1515 | return; |
1516 | break; |
1517 | } |
1518 | } |
1519 | else |
1520 | endc = startc; |
1521 | |
1522 | /* |
1523 | * Ranges are unportable. Actually, standard C does guarantee that digits |
1524 | * are contiguous, but making that an exception is just too complicated. |
1525 | */ |
1526 | if (startc != endc) |
1527 | NOTE(REG_UUNPORT); |
1528 | cv = range(v, startc, endc, (v->cflags & REG_ICASE)); |
1529 | NOERR(); |
1530 | subcolorcvec(v, cv, lp, rp); |
1531 | } |
1532 | |
1533 | /* |
1534 | * scanplain - scan PLAIN contents of [. etc. |
1535 | * |
1536 | * Certain bits of trickery in lex.c know that this code does not try |
1537 | * to look past the final bracket of the [. etc. |
1538 | */ |
1539 | static const chr * /* just after end of sequence */ |
1540 | scanplain(struct vars *v) |
1541 | { |
1542 | const chr *endp; |
1543 | |
1544 | assert(SEE(COLLEL) || SEE(ECLASS) || SEE(CCLASS)); |
1545 | NEXT(); |
1546 | |
1547 | endp = v->now; |
1548 | while (SEE(PLAIN)) |
1549 | { |
1550 | endp = v->now; |
1551 | NEXT(); |
1552 | } |
1553 | |
1554 | assert(SEE(END) || ISERR()); |
1555 | NEXT(); |
1556 | |
1557 | return endp; |
1558 | } |
1559 | |
1560 | /* |
1561 | * onechr - fill in arcs for a plain character, and possible case complements |
1562 | * This is mostly a shortcut for efficient handling of the common case. |
1563 | */ |
1564 | static void |
1565 | onechr(struct vars *v, |
1566 | chr c, |
1567 | struct state *lp, |
1568 | struct state *rp) |
1569 | { |
1570 | if (!(v->cflags & REG_ICASE)) |
1571 | { |
1572 | color lastsubcolor = COLORLESS; |
1573 | |
1574 | subcoloronechr(v, c, lp, rp, &lastsubcolor); |
1575 | return; |
1576 | } |
1577 | |
1578 | /* rats, need general case anyway... */ |
1579 | subcolorcvec(v, allcases(v, c), lp, rp); |
1580 | } |
1581 | |
1582 | /* |
1583 | * wordchrs - set up word-chr list for word-boundary stuff, if needed |
1584 | * |
1585 | * The list is kept as a bunch of arcs between two dummy states; it's |
1586 | * disposed of by the unreachable-states sweep in NFA optimization. |
1587 | * Does NEXT(). Must not be called from any unusual lexical context. |
1588 | * This should be reconciled with the \w etc. handling in lex.c, and |
1589 | * should be cleaned up to reduce dependencies on input scanning. |
1590 | */ |
1591 | static void |
1592 | wordchrs(struct vars *v) |
1593 | { |
1594 | struct state *left; |
1595 | struct state *right; |
1596 | |
1597 | if (v->wordchrs != NULL) |
1598 | { |
1599 | NEXT(); /* for consistency */ |
1600 | return; |
1601 | } |
1602 | |
1603 | left = newstate(v->nfa); |
1604 | right = newstate(v->nfa); |
1605 | NOERR(); |
1606 | /* fine point: implemented with [::], and lexer will set REG_ULOCALE */ |
1607 | lexword(v); |
1608 | NEXT(); |
1609 | assert(v->savenow != NULL && SEE('[')); |
1610 | bracket(v, left, right); |
1611 | assert((v->savenow != NULL && SEE(']')) || ISERR()); |
1612 | NEXT(); |
1613 | NOERR(); |
1614 | v->wordchrs = left; |
1615 | } |
1616 | |
1617 | /* |
1618 | * processlacon - generate the NFA representation of a LACON |
1619 | * |
1620 | * In the general case this is just newlacon() + newarc(), but some cases |
1621 | * can be optimized. |
1622 | */ |
1623 | static void |
1624 | processlacon(struct vars *v, |
1625 | struct state *begin, /* start of parsed LACON sub-re */ |
1626 | struct state *end, /* end of parsed LACON sub-re */ |
1627 | int latype, |
1628 | struct state *lp, /* left state to hang it on */ |
1629 | struct state *rp) /* right state to hang it on */ |
1630 | { |
1631 | struct state *s1; |
1632 | int n; |
1633 | |
1634 | /* |
1635 | * Check for lookaround RE consisting of a single plain color arc (or set |
1636 | * of arcs); this would typically be a simple chr or a bracket expression. |
1637 | */ |
1638 | s1 = single_color_transition(begin, end); |
1639 | switch (latype) |
1640 | { |
1641 | case LATYPE_AHEAD_POS: |
1642 | /* If lookahead RE is just colorset C, convert to AHEAD(C) */ |
1643 | if (s1 != NULL) |
1644 | { |
1645 | cloneouts(v->nfa, s1, lp, rp, AHEAD); |
1646 | return; |
1647 | } |
1648 | break; |
1649 | case LATYPE_AHEAD_NEG: |
1650 | /* If lookahead RE is just colorset C, convert to AHEAD(^C)|$ */ |
1651 | if (s1 != NULL) |
1652 | { |
1653 | colorcomplement(v->nfa, v->cm, AHEAD, s1, lp, rp); |
1654 | newarc(v->nfa, '$', 1, lp, rp); |
1655 | newarc(v->nfa, '$', 0, lp, rp); |
1656 | return; |
1657 | } |
1658 | break; |
1659 | case LATYPE_BEHIND_POS: |
1660 | /* If lookbehind RE is just colorset C, convert to BEHIND(C) */ |
1661 | if (s1 != NULL) |
1662 | { |
1663 | cloneouts(v->nfa, s1, lp, rp, BEHIND); |
1664 | return; |
1665 | } |
1666 | break; |
1667 | case LATYPE_BEHIND_NEG: |
1668 | /* If lookbehind RE is just colorset C, convert to BEHIND(^C)|^ */ |
1669 | if (s1 != NULL) |
1670 | { |
1671 | colorcomplement(v->nfa, v->cm, BEHIND, s1, lp, rp); |
1672 | newarc(v->nfa, '^', 1, lp, rp); |
1673 | newarc(v->nfa, '^', 0, lp, rp); |
1674 | return; |
1675 | } |
1676 | break; |
1677 | default: |
1678 | assert(NOTREACHED); |
1679 | } |
1680 | |
1681 | /* General case: we need a LACON subre and arc */ |
1682 | n = newlacon(v, begin, end, latype); |
1683 | newarc(v->nfa, LACON, n, lp, rp); |
1684 | } |
1685 | |
1686 | /* |
1687 | * subre - allocate a subre |
1688 | */ |
1689 | static struct subre * |
1690 | subre(struct vars *v, |
1691 | int op, |
1692 | int flags, |
1693 | struct state *begin, |
1694 | struct state *end) |
1695 | { |
1696 | struct subre *ret = v->treefree; |
1697 | |
1698 | /* |
1699 | * Checking for stack overflow here is sufficient to protect parse() and |
1700 | * its recursive subroutines. |
1701 | */ |
1702 | if (STACK_TOO_DEEP(v->re)) |
1703 | { |
1704 | ERR(REG_ETOOBIG); |
1705 | return NULL; |
1706 | } |
1707 | |
1708 | if (ret != NULL) |
1709 | v->treefree = ret->left; |
1710 | else |
1711 | { |
1712 | ret = (struct subre *) MALLOC(sizeof(struct subre)); |
1713 | if (ret == NULL) |
1714 | { |
1715 | ERR(REG_ESPACE); |
1716 | return NULL; |
1717 | } |
1718 | ret->chain = v->treechain; |
1719 | v->treechain = ret; |
1720 | } |
1721 | |
1722 | assert(strchr("=b|.*(" , op) != NULL); |
1723 | |
1724 | ret->op = op; |
1725 | ret->flags = flags; |
1726 | ret->id = 0; /* will be assigned later */ |
1727 | ret->subno = 0; |
1728 | ret->min = ret->max = 1; |
1729 | ret->left = NULL; |
1730 | ret->right = NULL; |
1731 | ret->begin = begin; |
1732 | ret->end = end; |
1733 | ZAPCNFA(ret->cnfa); |
1734 | |
1735 | return ret; |
1736 | } |
1737 | |
1738 | /* |
1739 | * freesubre - free a subRE subtree |
1740 | */ |
1741 | static void |
1742 | freesubre(struct vars *v, /* might be NULL */ |
1743 | struct subre *sr) |
1744 | { |
1745 | if (sr == NULL) |
1746 | return; |
1747 | |
1748 | if (sr->left != NULL) |
1749 | freesubre(v, sr->left); |
1750 | if (sr->right != NULL) |
1751 | freesubre(v, sr->right); |
1752 | |
1753 | freesrnode(v, sr); |
1754 | } |
1755 | |
1756 | /* |
1757 | * freesrnode - free one node in a subRE subtree |
1758 | */ |
1759 | static void |
1760 | freesrnode(struct vars *v, /* might be NULL */ |
1761 | struct subre *sr) |
1762 | { |
1763 | if (sr == NULL) |
1764 | return; |
1765 | |
1766 | if (!NULLCNFA(sr->cnfa)) |
1767 | freecnfa(&sr->cnfa); |
1768 | sr->flags = 0; |
1769 | |
1770 | if (v != NULL && v->treechain != NULL) |
1771 | { |
1772 | /* we're still parsing, maybe we can reuse the subre */ |
1773 | sr->left = v->treefree; |
1774 | v->treefree = sr; |
1775 | } |
1776 | else |
1777 | FREE(sr); |
1778 | } |
1779 | |
1780 | /* |
1781 | * optst - optimize a subRE subtree |
1782 | */ |
1783 | static void |
1784 | optst(struct vars *v, |
1785 | struct subre *t) |
1786 | { |
1787 | /* |
1788 | * DGP (2007-11-13): I assume it was the programmer's intent to eventually |
1789 | * come back and add code to optimize subRE trees, but the routine coded |
1790 | * just spends effort traversing the tree and doing nothing. We can do |
1791 | * nothing with less effort. |
1792 | */ |
1793 | return; |
1794 | } |
1795 | |
1796 | /* |
1797 | * numst - number tree nodes (assigning "id" indexes) |
1798 | */ |
1799 | static int /* next number */ |
1800 | numst(struct subre *t, |
1801 | int start) /* starting point for subtree numbers */ |
1802 | { |
1803 | int i; |
1804 | |
1805 | assert(t != NULL); |
1806 | |
1807 | i = start; |
1808 | t->id = (short) i++; |
1809 | if (t->left != NULL) |
1810 | i = numst(t->left, i); |
1811 | if (t->right != NULL) |
1812 | i = numst(t->right, i); |
1813 | return i; |
1814 | } |
1815 | |
1816 | /* |
1817 | * markst - mark tree nodes as INUSE |
1818 | * |
1819 | * Note: this is a great deal more subtle than it looks. During initial |
1820 | * parsing of a regex, all subres are linked into the treechain list; |
1821 | * discarded ones are also linked into the treefree list for possible reuse. |
1822 | * After we are done creating all subres required for a regex, we run markst() |
1823 | * then cleanst(), which results in discarding all subres not reachable from |
1824 | * v->tree. We then clear v->treechain, indicating that subres must be found |
1825 | * by descending from v->tree. This changes the behavior of freesubre(): it |
1826 | * will henceforth FREE() unwanted subres rather than sticking them into the |
1827 | * treefree list. (Doing that any earlier would result in dangling links in |
1828 | * the treechain list.) This all means that freev() will clean up correctly |
1829 | * if invoked before or after markst()+cleanst(); but it would not work if |
1830 | * called partway through this state conversion, so we mustn't error out |
1831 | * in or between these two functions. |
1832 | */ |
1833 | static void |
1834 | markst(struct subre *t) |
1835 | { |
1836 | assert(t != NULL); |
1837 | |
1838 | t->flags |= INUSE; |
1839 | if (t->left != NULL) |
1840 | markst(t->left); |
1841 | if (t->right != NULL) |
1842 | markst(t->right); |
1843 | } |
1844 | |
1845 | /* |
1846 | * cleanst - free any tree nodes not marked INUSE |
1847 | */ |
1848 | static void |
1849 | cleanst(struct vars *v) |
1850 | { |
1851 | struct subre *t; |
1852 | struct subre *next; |
1853 | |
1854 | for (t = v->treechain; t != NULL; t = next) |
1855 | { |
1856 | next = t->chain; |
1857 | if (!(t->flags & INUSE)) |
1858 | FREE(t); |
1859 | } |
1860 | v->treechain = NULL; |
1861 | v->treefree = NULL; /* just on general principles */ |
1862 | } |
1863 | |
1864 | /* |
1865 | * nfatree - turn a subRE subtree into a tree of compacted NFAs |
1866 | */ |
1867 | static long /* optimize results from top node */ |
1868 | nfatree(struct vars *v, |
1869 | struct subre *t, |
1870 | FILE *f) /* for debug output */ |
1871 | { |
1872 | assert(t != NULL && t->begin != NULL); |
1873 | |
1874 | if (t->left != NULL) |
1875 | (DISCARD) nfatree(v, t->left, f); |
1876 | if (t->right != NULL) |
1877 | (DISCARD) nfatree(v, t->right, f); |
1878 | |
1879 | return nfanode(v, t, 0, f); |
1880 | } |
1881 | |
1882 | /* |
1883 | * nfanode - do one NFA for nfatree or lacons |
1884 | * |
1885 | * If converttosearch is true, apply makesearch() to the NFA. |
1886 | */ |
1887 | static long /* optimize results */ |
1888 | nfanode(struct vars *v, |
1889 | struct subre *t, |
1890 | int converttosearch, |
1891 | FILE *f) /* for debug output */ |
1892 | { |
1893 | struct nfa *nfa; |
1894 | long ret = 0; |
1895 | |
1896 | assert(t->begin != NULL); |
1897 | |
1898 | #ifdef REG_DEBUG |
1899 | if (f != NULL) |
1900 | { |
1901 | char idbuf[50]; |
1902 | |
1903 | fprintf(f, "\n\n\n========= TREE NODE %s ==========\n" , |
1904 | stid(t, idbuf, sizeof(idbuf))); |
1905 | } |
1906 | #endif |
1907 | nfa = newnfa(v, v->cm, v->nfa); |
1908 | NOERRZ(); |
1909 | dupnfa(nfa, t->begin, t->end, nfa->init, nfa->final); |
1910 | if (!ISERR()) |
1911 | specialcolors(nfa); |
1912 | if (!ISERR()) |
1913 | ret = optimize(nfa, f); |
1914 | if (converttosearch && !ISERR()) |
1915 | makesearch(v, nfa); |
1916 | if (!ISERR()) |
1917 | compact(nfa, &t->cnfa); |
1918 | |
1919 | freenfa(nfa); |
1920 | return ret; |
1921 | } |
1922 | |
1923 | /* |
1924 | * newlacon - allocate a lookaround-constraint subRE |
1925 | */ |
1926 | static int /* lacon number */ |
1927 | newlacon(struct vars *v, |
1928 | struct state *begin, |
1929 | struct state *end, |
1930 | int latype) |
1931 | { |
1932 | int n; |
1933 | struct subre *newlacons; |
1934 | struct subre *sub; |
1935 | |
1936 | if (v->nlacons == 0) |
1937 | { |
1938 | n = 1; /* skip 0th */ |
1939 | newlacons = (struct subre *) MALLOC(2 * sizeof(struct subre)); |
1940 | } |
1941 | else |
1942 | { |
1943 | n = v->nlacons; |
1944 | newlacons = (struct subre *) REALLOC(v->lacons, |
1945 | (n + 1) * sizeof(struct subre)); |
1946 | } |
1947 | if (newlacons == NULL) |
1948 | { |
1949 | ERR(REG_ESPACE); |
1950 | return 0; |
1951 | } |
1952 | v->lacons = newlacons; |
1953 | v->nlacons = n + 1; |
1954 | sub = &v->lacons[n]; |
1955 | sub->begin = begin; |
1956 | sub->end = end; |
1957 | sub->subno = latype; |
1958 | ZAPCNFA(sub->cnfa); |
1959 | return n; |
1960 | } |
1961 | |
1962 | /* |
1963 | * freelacons - free lookaround-constraint subRE vector |
1964 | */ |
1965 | static void |
1966 | freelacons(struct subre *subs, |
1967 | int n) |
1968 | { |
1969 | struct subre *sub; |
1970 | int i; |
1971 | |
1972 | assert(n > 0); |
1973 | for (sub = subs + 1, i = n - 1; i > 0; sub++, i--) /* no 0th */ |
1974 | if (!NULLCNFA(sub->cnfa)) |
1975 | freecnfa(&sub->cnfa); |
1976 | FREE(subs); |
1977 | } |
1978 | |
1979 | /* |
1980 | * rfree - free a whole RE (insides of regfree) |
1981 | */ |
1982 | static void |
1983 | rfree(regex_t *re) |
1984 | { |
1985 | struct guts *g; |
1986 | |
1987 | if (re == NULL || re->re_magic != REMAGIC) |
1988 | return; |
1989 | |
1990 | re->re_magic = 0; /* invalidate RE */ |
1991 | g = (struct guts *) re->re_guts; |
1992 | re->re_guts = NULL; |
1993 | re->re_fns = NULL; |
1994 | if (g != NULL) |
1995 | { |
1996 | g->magic = 0; |
1997 | freecm(&g->cmap); |
1998 | if (g->tree != NULL) |
1999 | freesubre((struct vars *) NULL, g->tree); |
2000 | if (g->lacons != NULL) |
2001 | freelacons(g->lacons, g->nlacons); |
2002 | if (!NULLCNFA(g->search)) |
2003 | freecnfa(&g->search); |
2004 | FREE(g); |
2005 | } |
2006 | } |
2007 | |
2008 | /* |
2009 | * rcancelrequested - check for external request to cancel regex operation |
2010 | * |
2011 | * Return nonzero to fail the operation with error code REG_CANCEL, |
2012 | * zero to keep going |
2013 | * |
2014 | * The current implementation is Postgres-specific. If we ever get around |
2015 | * to splitting the regex code out as a standalone library, there will need |
2016 | * to be some API to let applications define a callback function for this. |
2017 | */ |
2018 | static int |
2019 | rcancelrequested(void) |
2020 | { |
2021 | return InterruptPending && (QueryCancelPending || ProcDiePending); |
2022 | } |
2023 | |
2024 | /* |
2025 | * rstacktoodeep - check for stack getting dangerously deep |
2026 | * |
2027 | * Return nonzero to fail the operation with error code REG_ETOOBIG, |
2028 | * zero to keep going |
2029 | * |
2030 | * The current implementation is Postgres-specific. If we ever get around |
2031 | * to splitting the regex code out as a standalone library, there will need |
2032 | * to be some API to let applications define a callback function for this. |
2033 | */ |
2034 | static int |
2035 | rstacktoodeep(void) |
2036 | { |
2037 | return stack_is_too_deep(); |
2038 | } |
2039 | |
2040 | #ifdef REG_DEBUG |
2041 | |
2042 | /* |
2043 | * dump - dump an RE in human-readable form |
2044 | */ |
2045 | static void |
2046 | dump(regex_t *re, |
2047 | FILE *f) |
2048 | { |
2049 | struct guts *g; |
2050 | int i; |
2051 | |
2052 | if (re->re_magic != REMAGIC) |
2053 | fprintf(f, "bad magic number (0x%x not 0x%x)\n" , re->re_magic, |
2054 | REMAGIC); |
2055 | if (re->re_guts == NULL) |
2056 | { |
2057 | fprintf(f, "NULL guts!!!\n" ); |
2058 | return; |
2059 | } |
2060 | g = (struct guts *) re->re_guts; |
2061 | if (g->magic != GUTSMAGIC) |
2062 | fprintf(f, "bad guts magic number (0x%x not 0x%x)\n" , g->magic, |
2063 | GUTSMAGIC); |
2064 | |
2065 | fprintf(f, "\n\n\n========= DUMP ==========\n" ); |
2066 | fprintf(f, "nsub %d, info 0%lo, csize %d, ntree %d\n" , |
2067 | (int) re->re_nsub, re->re_info, re->re_csize, g->ntree); |
2068 | |
2069 | dumpcolors(&g->cmap, f); |
2070 | if (!NULLCNFA(g->search)) |
2071 | { |
2072 | fprintf(f, "\nsearch:\n" ); |
2073 | dumpcnfa(&g->search, f); |
2074 | } |
2075 | for (i = 1; i < g->nlacons; i++) |
2076 | { |
2077 | struct subre *lasub = &g->lacons[i]; |
2078 | const char *latype; |
2079 | |
2080 | switch (lasub->subno) |
2081 | { |
2082 | case LATYPE_AHEAD_POS: |
2083 | latype = "positive lookahead" ; |
2084 | break; |
2085 | case LATYPE_AHEAD_NEG: |
2086 | latype = "negative lookahead" ; |
2087 | break; |
2088 | case LATYPE_BEHIND_POS: |
2089 | latype = "positive lookbehind" ; |
2090 | break; |
2091 | case LATYPE_BEHIND_NEG: |
2092 | latype = "negative lookbehind" ; |
2093 | break; |
2094 | default: |
2095 | latype = "???" ; |
2096 | break; |
2097 | } |
2098 | fprintf(f, "\nla%d (%s):\n" , i, latype); |
2099 | dumpcnfa(&lasub->cnfa, f); |
2100 | } |
2101 | fprintf(f, "\n" ); |
2102 | dumpst(g->tree, f, 0); |
2103 | } |
2104 | |
2105 | /* |
2106 | * dumpst - dump a subRE tree |
2107 | */ |
2108 | static void |
2109 | dumpst(struct subre *t, |
2110 | FILE *f, |
2111 | int nfapresent) /* is the original NFA still around? */ |
2112 | { |
2113 | if (t == NULL) |
2114 | fprintf(f, "null tree\n" ); |
2115 | else |
2116 | stdump(t, f, nfapresent); |
2117 | fflush(f); |
2118 | } |
2119 | |
2120 | /* |
2121 | * stdump - recursive guts of dumpst |
2122 | */ |
2123 | static void |
2124 | stdump(struct subre *t, |
2125 | FILE *f, |
2126 | int nfapresent) /* is the original NFA still around? */ |
2127 | { |
2128 | char idbuf[50]; |
2129 | |
2130 | fprintf(f, "%s. `%c'" , stid(t, idbuf, sizeof(idbuf)), t->op); |
2131 | if (t->flags & LONGER) |
2132 | fprintf(f, " longest" ); |
2133 | if (t->flags & SHORTER) |
2134 | fprintf(f, " shortest" ); |
2135 | if (t->flags & MIXED) |
2136 | fprintf(f, " hasmixed" ); |
2137 | if (t->flags & CAP) |
2138 | fprintf(f, " hascapture" ); |
2139 | if (t->flags & BACKR) |
2140 | fprintf(f, " hasbackref" ); |
2141 | if (!(t->flags & INUSE)) |
2142 | fprintf(f, " UNUSED" ); |
2143 | if (t->subno != 0) |
2144 | fprintf(f, " (#%d)" , t->subno); |
2145 | if (t->min != 1 || t->max != 1) |
2146 | { |
2147 | fprintf(f, " {%d," , t->min); |
2148 | if (t->max != DUPINF) |
2149 | fprintf(f, "%d" , t->max); |
2150 | fprintf(f, "}" ); |
2151 | } |
2152 | if (nfapresent) |
2153 | fprintf(f, " %ld-%ld" , (long) t->begin->no, (long) t->end->no); |
2154 | if (t->left != NULL) |
2155 | fprintf(f, " L:%s" , stid(t->left, idbuf, sizeof(idbuf))); |
2156 | if (t->right != NULL) |
2157 | fprintf(f, " R:%s" , stid(t->right, idbuf, sizeof(idbuf))); |
2158 | if (!NULLCNFA(t->cnfa)) |
2159 | { |
2160 | fprintf(f, "\n" ); |
2161 | dumpcnfa(&t->cnfa, f); |
2162 | } |
2163 | fprintf(f, "\n" ); |
2164 | if (t->left != NULL) |
2165 | stdump(t->left, f, nfapresent); |
2166 | if (t->right != NULL) |
2167 | stdump(t->right, f, nfapresent); |
2168 | } |
2169 | |
2170 | /* |
2171 | * stid - identify a subtree node for dumping |
2172 | */ |
2173 | static const char * /* points to buf or constant string */ |
2174 | stid(struct subre *t, |
2175 | char *buf, |
2176 | size_t bufsize) |
2177 | { |
2178 | /* big enough for hex int or decimal t->id? */ |
2179 | if (bufsize < sizeof(void *) * 2 + 3 || bufsize < sizeof(t->id) * 3 + 1) |
2180 | return "unable" ; |
2181 | if (t->id != 0) |
2182 | sprintf(buf, "%d" , t->id); |
2183 | else |
2184 | sprintf(buf, "%p" , t); |
2185 | return buf; |
2186 | } |
2187 | #endif /* REG_DEBUG */ |
2188 | |
2189 | |
2190 | #include "regc_lex.c" |
2191 | #include "regc_color.c" |
2192 | #include "regc_nfa.c" |
2193 | #include "regc_cvec.c" |
2194 | #include "regc_pg_locale.c" |
2195 | #include "regc_locale.c" |
2196 | |