1/* Extended regular expression matching and search library.
2 Copyright (C) 2002-2018 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4 Contributed by Isamu Hasegawa <isamu@yamato.ibm.com>.
5
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2.1 of the License, or (at your option) any later version.
10
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; if not, see
18 <http://www.gnu.org/licenses/>. */
19
20#include <stdint.h>
21
22static reg_errcode_t match_ctx_init (re_match_context_t *cache, int eflags,
23 int n);
24static void match_ctx_clean (re_match_context_t *mctx);
25static void match_ctx_free (re_match_context_t *cache);
26static reg_errcode_t match_ctx_add_entry (re_match_context_t *cache, int node,
27 int str_idx, int from, int to);
28static int search_cur_bkref_entry (const re_match_context_t *mctx,
29 int str_idx);
30static reg_errcode_t match_ctx_add_subtop (re_match_context_t *mctx, int node,
31 int str_idx);
32static re_sub_match_last_t * match_ctx_add_sublast (re_sub_match_top_t *subtop,
33 int node, int str_idx);
34static void sift_ctx_init (re_sift_context_t *sctx, re_dfastate_t **sifted_sts,
35 re_dfastate_t **limited_sts, int last_node,
36 int last_str_idx);
37static reg_errcode_t re_search_internal (const regex_t *preg,
38 const char *string, int length,
39 int start, int range, int stop,
40 size_t nmatch, regmatch_t pmatch[],
41 int eflags);
42static int re_search_2_stub (struct re_pattern_buffer *bufp,
43 const char *string1, int length1,
44 const char *string2, int length2,
45 int start, int range, struct re_registers *regs,
46 int stop, int ret_len);
47static int re_search_stub (struct re_pattern_buffer *bufp,
48 const char *string, int length, int start,
49 int range, int stop, struct re_registers *regs,
50 int ret_len);
51static unsigned re_copy_regs (struct re_registers *regs, regmatch_t *pmatch,
52 int nregs, int regs_allocated);
53static reg_errcode_t prune_impossible_nodes (re_match_context_t *mctx);
54static int check_matching (re_match_context_t *mctx, int fl_longest_match,
55 int *p_match_first);
56static int check_halt_state_context (const re_match_context_t *mctx,
57 const re_dfastate_t *state, int idx);
58static void update_regs (const re_dfa_t *dfa, regmatch_t *pmatch,
59 regmatch_t *prev_idx_match, int cur_node,
60 int cur_idx, int nmatch);
61static reg_errcode_t push_fail_stack (struct re_fail_stack_t *fs,
62 int str_idx, int dest_node, int nregs,
63 regmatch_t *regs,
64 re_node_set *eps_via_nodes);
65static reg_errcode_t set_regs (const regex_t *preg,
66 const re_match_context_t *mctx,
67 size_t nmatch, regmatch_t *pmatch,
68 int fl_backtrack);
69static reg_errcode_t free_fail_stack_return (struct re_fail_stack_t *fs);
70
71#ifdef RE_ENABLE_I18N
72static int sift_states_iter_mb (const re_match_context_t *mctx,
73 re_sift_context_t *sctx,
74 int node_idx, int str_idx, int max_str_idx);
75#endif /* RE_ENABLE_I18N */
76static reg_errcode_t sift_states_backward (const re_match_context_t *mctx,
77 re_sift_context_t *sctx);
78static reg_errcode_t build_sifted_states (const re_match_context_t *mctx,
79 re_sift_context_t *sctx, int str_idx,
80 re_node_set *cur_dest);
81static reg_errcode_t update_cur_sifted_state (const re_match_context_t *mctx,
82 re_sift_context_t *sctx,
83 int str_idx,
84 re_node_set *dest_nodes);
85static reg_errcode_t add_epsilon_src_nodes (const re_dfa_t *dfa,
86 re_node_set *dest_nodes,
87 const re_node_set *candidates);
88static int check_dst_limits (const re_match_context_t *mctx,
89 re_node_set *limits,
90 int dst_node, int dst_idx, int src_node,
91 int src_idx);
92static int check_dst_limits_calc_pos_1 (const re_match_context_t *mctx,
93 int boundaries, int subexp_idx,
94 int from_node, int bkref_idx);
95static int check_dst_limits_calc_pos (const re_match_context_t *mctx,
96 int limit, int subexp_idx,
97 int node, int str_idx,
98 int bkref_idx);
99static reg_errcode_t check_subexp_limits (const re_dfa_t *dfa,
100 re_node_set *dest_nodes,
101 const re_node_set *candidates,
102 re_node_set *limits,
103 struct re_backref_cache_entry *bkref_ents,
104 int str_idx);
105static reg_errcode_t sift_states_bkref (const re_match_context_t *mctx,
106 re_sift_context_t *sctx,
107 int str_idx,
108 const re_node_set *candidates);
109static reg_errcode_t merge_state_array (const re_dfa_t *dfa,
110 re_dfastate_t **dst,
111 re_dfastate_t **src, int num);
112static re_dfastate_t *find_recover_state (reg_errcode_t *err,
113 re_match_context_t *mctx);
114static re_dfastate_t *transit_state (reg_errcode_t *err,
115 re_match_context_t *mctx,
116 re_dfastate_t *state);
117static re_dfastate_t *merge_state_with_log (reg_errcode_t *err,
118 re_match_context_t *mctx,
119 re_dfastate_t *next_state);
120static reg_errcode_t check_subexp_matching_top (re_match_context_t *mctx,
121 re_node_set *cur_nodes,
122 int str_idx);
123#if 0
124static re_dfastate_t *transit_state_sb (reg_errcode_t *err,
125 re_match_context_t *mctx,
126 re_dfastate_t *pstate);
127#endif
128#ifdef RE_ENABLE_I18N
129static reg_errcode_t transit_state_mb (re_match_context_t *mctx,
130 re_dfastate_t *pstate);
131#endif /* RE_ENABLE_I18N */
132static reg_errcode_t transit_state_bkref (re_match_context_t *mctx,
133 const re_node_set *nodes);
134static reg_errcode_t get_subexp (re_match_context_t *mctx,
135 int bkref_node, int bkref_str_idx);
136static reg_errcode_t get_subexp_sub (re_match_context_t *mctx,
137 const re_sub_match_top_t *sub_top,
138 re_sub_match_last_t *sub_last,
139 int bkref_node, int bkref_str);
140static int find_subexp_node (const re_dfa_t *dfa, const re_node_set *nodes,
141 int subexp_idx, int type);
142static reg_errcode_t check_arrival (re_match_context_t *mctx,
143 state_array_t *path, int top_node,
144 int top_str, int last_node, int last_str,
145 int type);
146static reg_errcode_t check_arrival_add_next_nodes (re_match_context_t *mctx,
147 int str_idx,
148 re_node_set *cur_nodes,
149 re_node_set *next_nodes);
150static reg_errcode_t check_arrival_expand_ecl (const re_dfa_t *dfa,
151 re_node_set *cur_nodes,
152 int ex_subexp, int type);
153static reg_errcode_t check_arrival_expand_ecl_sub (const re_dfa_t *dfa,
154 re_node_set *dst_nodes,
155 int target, int ex_subexp,
156 int type);
157static reg_errcode_t expand_bkref_cache (re_match_context_t *mctx,
158 re_node_set *cur_nodes, int cur_str,
159 int subexp_num, int type);
160static int build_trtable (const re_dfa_t *dfa, re_dfastate_t *state);
161#ifdef RE_ENABLE_I18N
162static int check_node_accept_bytes (const re_dfa_t *dfa, int node_idx,
163 const re_string_t *input, int idx);
164# ifdef _LIBC
165static unsigned int find_collation_sequence_value (const unsigned char *mbs,
166 size_t name_len);
167# endif /* _LIBC */
168#endif /* RE_ENABLE_I18N */
169static int group_nodes_into_DFAstates (const re_dfa_t *dfa,
170 const re_dfastate_t *state,
171 re_node_set *states_node,
172 bitset_t *states_ch);
173static int check_node_accept (const re_match_context_t *mctx,
174 const re_token_t *node, int idx);
175static reg_errcode_t extend_buffers (re_match_context_t *mctx, int min_len);
176
177/* Entry point for POSIX code. */
178
179/* regexec searches for a given pattern, specified by PREG, in the
180 string STRING.
181
182 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
183 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
184 least NMATCH elements, and we set them to the offsets of the
185 corresponding matched substrings.
186
187 EFLAGS specifies `execution flags' which affect matching: if
188 REG_NOTBOL is set, then ^ does not match at the beginning of the
189 string; if REG_NOTEOL is set, then $ does not match at the end.
190
191 We return 0 if we find a match and REG_NOMATCH if not. */
192
193int
194regexec (const regex_t *__restrict preg, const char *__restrict string,
195 size_t nmatch, regmatch_t pmatch[], int eflags)
196{
197 reg_errcode_t err;
198 int start, length;
199 re_dfa_t *dfa = (re_dfa_t *) preg->buffer;
200
201 if (eflags & ~(REG_NOTBOL | REG_NOTEOL | REG_STARTEND))
202 return REG_BADPAT;
203
204 if (eflags & REG_STARTEND)
205 {
206 start = pmatch[0].rm_so;
207 length = pmatch[0].rm_eo;
208 }
209 else
210 {
211 start = 0;
212 length = strlen (string);
213 }
214
215 __libc_lock_lock (dfa->lock);
216 if (preg->no_sub)
217 err = re_search_internal (preg, string, length, start, length - start,
218 length, 0, NULL, eflags);
219 else
220 err = re_search_internal (preg, string, length, start, length - start,
221 length, nmatch, pmatch, eflags);
222 __libc_lock_unlock (dfa->lock);
223 return err != REG_NOERROR;
224}
225
226#ifdef _LIBC
227libc_hidden_def (__regexec)
228
229# include <shlib-compat.h>
230versioned_symbol (libc, __regexec, regexec, GLIBC_2_3_4);
231
232# if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_3_4)
233__typeof__ (__regexec) __compat_regexec;
234
235int
236attribute_compat_text_section
237__compat_regexec (const regex_t *__restrict preg,
238 const char *__restrict string, size_t nmatch,
239 regmatch_t pmatch[], int eflags)
240{
241 return regexec (preg, string, nmatch, pmatch,
242 eflags & (REG_NOTBOL | REG_NOTEOL));
243}
244compat_symbol (libc, __compat_regexec, regexec, GLIBC_2_0);
245# endif
246#endif
247
248/* Entry points for GNU code. */
249
250/* re_match, re_search, re_match_2, re_search_2
251
252 The former two functions operate on STRING with length LENGTH,
253 while the later two operate on concatenation of STRING1 and STRING2
254 with lengths LENGTH1 and LENGTH2, respectively.
255
256 re_match() matches the compiled pattern in BUFP against the string,
257 starting at index START.
258
259 re_search() first tries matching at index START, then it tries to match
260 starting from index START + 1, and so on. The last start position tried
261 is START + RANGE. (Thus RANGE = 0 forces re_search to operate the same
262 way as re_match().)
263
264 The parameter STOP of re_{match,search}_2 specifies that no match exceeding
265 the first STOP characters of the concatenation of the strings should be
266 concerned.
267
268 If REGS is not NULL, and BUFP->no_sub is not set, the offsets of the match
269 and all groups is stored in REGS. (For the "_2" variants, the offsets are
270 computed relative to the concatenation, not relative to the individual
271 strings.)
272
273 On success, re_match* functions return the length of the match, re_search*
274 return the position of the start of the match. Return value -1 means no
275 match was found and -2 indicates an internal error. */
276
277int
278re_match (struct re_pattern_buffer *bufp, const char *string, int length,
279 int start, struct re_registers *regs)
280{
281 return re_search_stub (bufp, string, length, start, 0, length, regs, 1);
282}
283#ifdef _LIBC
284weak_alias (__re_match, re_match)
285#endif
286
287int
288re_search (struct re_pattern_buffer *bufp, const char *string, int length,
289 int start, int range, struct re_registers *regs)
290{
291 return re_search_stub (bufp, string, length, start, range, length, regs, 0);
292}
293#ifdef _LIBC
294weak_alias (__re_search, re_search)
295#endif
296
297int
298re_match_2 (struct re_pattern_buffer *bufp, const char *string1, int length1,
299 const char *string2, int length2, int start,
300 struct re_registers *regs, int stop)
301{
302 return re_search_2_stub (bufp, string1, length1, string2, length2,
303 start, 0, regs, stop, 1);
304}
305#ifdef _LIBC
306weak_alias (__re_match_2, re_match_2)
307#endif
308
309int
310re_search_2 (struct re_pattern_buffer *bufp, const char *string1, int length1,
311 const char *string2, int length2, int start, int range,
312 struct re_registers *regs, int stop)
313{
314 return re_search_2_stub (bufp, string1, length1, string2, length2,
315 start, range, regs, stop, 0);
316}
317#ifdef _LIBC
318weak_alias (__re_search_2, re_search_2)
319#endif
320
321static int
322re_search_2_stub (struct re_pattern_buffer *bufp, const char *string1,
323 int length1, const char *string2, int length2, int start,
324 int range, struct re_registers *regs,
325 int stop, int ret_len)
326{
327 const char *str;
328 int rval;
329 int len = length1 + length2;
330 char *s = NULL;
331
332 if (BE (length1 < 0 || length2 < 0 || stop < 0 || len < length1, 0))
333 return -2;
334
335 /* Concatenate the strings. */
336 if (length2 > 0)
337 if (length1 > 0)
338 {
339 s = re_malloc (char, len);
340
341 if (BE (s == NULL, 0))
342 return -2;
343#ifdef _LIBC
344 memcpy (__mempcpy (s, string1, length1), string2, length2);
345#else
346 memcpy (s, string1, length1);
347 memcpy (s + length1, string2, length2);
348#endif
349 str = s;
350 }
351 else
352 str = string2;
353 else
354 str = string1;
355
356 rval = re_search_stub (bufp, str, len, start, range, stop, regs, ret_len);
357 re_free (s);
358 return rval;
359}
360
361/* The parameters have the same meaning as those of re_search.
362 Additional parameters:
363 If RET_LEN is nonzero the length of the match is returned (re_match style);
364 otherwise the position of the match is returned. */
365
366static int
367re_search_stub (struct re_pattern_buffer *bufp, const char *string, int length,
368 int start, int range, int stop, struct re_registers *regs,
369 int ret_len)
370{
371 reg_errcode_t result;
372 regmatch_t *pmatch;
373 int nregs, rval;
374 int eflags = 0;
375 re_dfa_t *dfa = (re_dfa_t *) bufp->buffer;
376
377 /* Check for out-of-range. */
378 if (BE (start < 0 || start > length, 0))
379 return -1;
380 if (BE (start + range > length, 0))
381 range = length - start;
382 else if (BE (start + range < 0, 0))
383 range = -start;
384
385 __libc_lock_lock (dfa->lock);
386
387 eflags |= (bufp->not_bol) ? REG_NOTBOL : 0;
388 eflags |= (bufp->not_eol) ? REG_NOTEOL : 0;
389
390 /* Compile fastmap if we haven't yet. */
391 if (range > 0 && bufp->fastmap != NULL && !bufp->fastmap_accurate)
392 re_compile_fastmap (bufp);
393
394 if (BE (bufp->no_sub, 0))
395 regs = NULL;
396
397 /* We need at least 1 register. */
398 if (regs == NULL)
399 nregs = 1;
400 else if (BE (bufp->regs_allocated == REGS_FIXED &&
401 regs->num_regs < bufp->re_nsub + 1, 0))
402 {
403 nregs = regs->num_regs;
404 if (BE (nregs < 1, 0))
405 {
406 /* Nothing can be copied to regs. */
407 regs = NULL;
408 nregs = 1;
409 }
410 }
411 else
412 nregs = bufp->re_nsub + 1;
413 pmatch = re_malloc (regmatch_t, nregs);
414 if (BE (pmatch == NULL, 0))
415 {
416 rval = -2;
417 goto out;
418 }
419
420 result = re_search_internal (bufp, string, length, start, range, stop,
421 nregs, pmatch, eflags);
422
423 rval = 0;
424
425 /* I hope we needn't fill ther regs with -1's when no match was found. */
426 if (result != REG_NOERROR)
427 rval = -1;
428 else if (regs != NULL)
429 {
430 /* If caller wants register contents data back, copy them. */
431 bufp->regs_allocated = re_copy_regs (regs, pmatch, nregs,
432 bufp->regs_allocated);
433 if (BE (bufp->regs_allocated == REGS_UNALLOCATED, 0))
434 rval = -2;
435 }
436
437 if (BE (rval == 0, 1))
438 {
439 if (ret_len)
440 {
441 assert (pmatch[0].rm_so == start);
442 rval = pmatch[0].rm_eo - start;
443 }
444 else
445 rval = pmatch[0].rm_so;
446 }
447 re_free (pmatch);
448 out:
449 __libc_lock_unlock (dfa->lock);
450 return rval;
451}
452
453static unsigned
454re_copy_regs (struct re_registers *regs, regmatch_t *pmatch, int nregs,
455 int regs_allocated)
456{
457 int rval = REGS_REALLOCATE;
458 int i;
459 int need_regs = nregs + 1;
460 /* We need one extra element beyond `num_regs' for the `-1' marker GNU code
461 uses. */
462
463 /* Have the register data arrays been allocated? */
464 if (regs_allocated == REGS_UNALLOCATED)
465 { /* No. So allocate them with malloc. */
466 regs->start = re_malloc (regoff_t, need_regs);
467 if (BE (regs->start == NULL, 0))
468 return REGS_UNALLOCATED;
469 regs->end = re_malloc (regoff_t, need_regs);
470 if (BE (regs->end == NULL, 0))
471 {
472 re_free (regs->start);
473 return REGS_UNALLOCATED;
474 }
475 regs->num_regs = need_regs;
476 }
477 else if (regs_allocated == REGS_REALLOCATE)
478 { /* Yes. If we need more elements than were already
479 allocated, reallocate them. If we need fewer, just
480 leave it alone. */
481 if (BE (need_regs > regs->num_regs, 0))
482 {
483 regoff_t *new_start = re_realloc (regs->start, regoff_t, need_regs);
484 regoff_t *new_end;
485 if (BE (new_start == NULL, 0))
486 return REGS_UNALLOCATED;
487 new_end = re_realloc (regs->end, regoff_t, need_regs);
488 if (BE (new_end == NULL, 0))
489 {
490 re_free (new_start);
491 return REGS_UNALLOCATED;
492 }
493 regs->start = new_start;
494 regs->end = new_end;
495 regs->num_regs = need_regs;
496 }
497 }
498 else
499 {
500 assert (regs_allocated == REGS_FIXED);
501 /* This function may not be called with REGS_FIXED and nregs too big. */
502 assert (regs->num_regs >= nregs);
503 rval = REGS_FIXED;
504 }
505
506 /* Copy the regs. */
507 for (i = 0; i < nregs; ++i)
508 {
509 regs->start[i] = pmatch[i].rm_so;
510 regs->end[i] = pmatch[i].rm_eo;
511 }
512 for ( ; i < regs->num_regs; ++i)
513 regs->start[i] = regs->end[i] = -1;
514
515 return rval;
516}
517
518/* Set REGS to hold NUM_REGS registers, storing them in STARTS and
519 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
520 this memory for recording register information. STARTS and ENDS
521 must be allocated using the malloc library routine, and must each
522 be at least NUM_REGS * sizeof (regoff_t) bytes long.
523
524 If NUM_REGS == 0, then subsequent matches should allocate their own
525 register data.
526
527 Unless this function is called, the first search or match using
528 PATTERN_BUFFER will allocate its own register data, without
529 freeing the old data. */
530
531void
532re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs,
533 unsigned num_regs, regoff_t *starts, regoff_t *ends)
534{
535 if (num_regs)
536 {
537 bufp->regs_allocated = REGS_REALLOCATE;
538 regs->num_regs = num_regs;
539 regs->start = starts;
540 regs->end = ends;
541 }
542 else
543 {
544 bufp->regs_allocated = REGS_UNALLOCATED;
545 regs->num_regs = 0;
546 regs->start = regs->end = (regoff_t *) 0;
547 }
548}
549#ifdef _LIBC
550weak_alias (__re_set_registers, re_set_registers)
551#endif
552
553/* Entry points compatible with 4.2 BSD regex library. We don't define
554 them unless specifically requested. */
555
556#if defined _REGEX_RE_COMP || defined _LIBC
557int
558# ifdef _LIBC
559weak_function
560# endif
561re_exec (const char *s)
562{
563 return 0 == regexec (&re_comp_buf, s, 0, NULL, 0);
564}
565#endif /* _REGEX_RE_COMP */
566
567/* Internal entry point. */
568
569/* Searches for a compiled pattern PREG in the string STRING, whose
570 length is LENGTH. NMATCH, PMATCH, and EFLAGS have the same
571 meaning as with regexec. START, and RANGE have the same meanings
572 with re_search.
573 Return REG_NOERROR if we find a match, and REG_NOMATCH if not,
574 otherwise return the error code.
575 Note: We assume front end functions already check ranges.
576 (START + RANGE >= 0 && START + RANGE <= LENGTH) */
577
578static reg_errcode_t
579__attribute_warn_unused_result__
580re_search_internal (const regex_t *preg, const char *string, int length,
581 int start, int range, int stop, size_t nmatch,
582 regmatch_t pmatch[], int eflags)
583{
584 reg_errcode_t err;
585 const re_dfa_t *dfa = (const re_dfa_t *) preg->buffer;
586 int left_lim, right_lim, incr;
587 int fl_longest_match, match_first, match_kind, match_last = -1;
588 int extra_nmatch;
589 int sb, ch;
590#if defined _LIBC || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L)
591 re_match_context_t mctx = { .dfa = dfa };
592#else
593 re_match_context_t mctx;
594#endif
595 char *fastmap = (preg->fastmap != NULL && preg->fastmap_accurate
596 && range && !preg->can_be_null) ? preg->fastmap : NULL;
597 RE_TRANSLATE_TYPE t = preg->translate;
598
599#if !(defined _LIBC || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L))
600 memset (&mctx, '\0', sizeof (re_match_context_t));
601 mctx.dfa = dfa;
602#endif
603
604 extra_nmatch = (nmatch > preg->re_nsub) ? nmatch - (preg->re_nsub + 1) : 0;
605 nmatch -= extra_nmatch;
606
607 /* Check if the DFA haven't been compiled. */
608 if (BE (preg->used == 0 || dfa->init_state == NULL
609 || dfa->init_state_word == NULL || dfa->init_state_nl == NULL
610 || dfa->init_state_begbuf == NULL, 0))
611 return REG_NOMATCH;
612
613#ifdef DEBUG
614 /* We assume front-end functions already check them. */
615 assert (start + range >= 0 && start + range <= length);
616#endif
617
618 /* If initial states with non-begbuf contexts have no elements,
619 the regex must be anchored. If preg->newline_anchor is set,
620 we'll never use init_state_nl, so do not check it. */
621 if (dfa->init_state->nodes.nelem == 0
622 && dfa->init_state_word->nodes.nelem == 0
623 && (dfa->init_state_nl->nodes.nelem == 0
624 || !preg->newline_anchor))
625 {
626 if (start != 0 && start + range != 0)
627 return REG_NOMATCH;
628 start = range = 0;
629 }
630
631 /* We must check the longest matching, if nmatch > 0. */
632 fl_longest_match = (nmatch != 0 || dfa->nbackref);
633
634 err = re_string_allocate (&mctx.input, string, length, dfa->nodes_len + 1,
635 preg->translate, preg->syntax & RE_ICASE, dfa);
636 if (BE (err != REG_NOERROR, 0))
637 goto free_return;
638 mctx.input.stop = stop;
639 mctx.input.raw_stop = stop;
640 mctx.input.newline_anchor = preg->newline_anchor;
641
642 err = match_ctx_init (&mctx, eflags, dfa->nbackref * 2);
643 if (BE (err != REG_NOERROR, 0))
644 goto free_return;
645
646 /* We will log all the DFA states through which the dfa pass,
647 if nmatch > 1, or this dfa has "multibyte node", which is a
648 back-reference or a node which can accept multibyte character or
649 multi character collating element. */
650 if (nmatch > 1 || dfa->has_mb_node)
651 {
652 /* Avoid overflow. */
653 if (BE (SIZE_MAX / sizeof (re_dfastate_t *) <= mctx.input.bufs_len, 0))
654 {
655 err = REG_ESPACE;
656 goto free_return;
657 }
658
659 mctx.state_log = re_malloc (re_dfastate_t *, mctx.input.bufs_len + 1);
660 if (BE (mctx.state_log == NULL, 0))
661 {
662 err = REG_ESPACE;
663 goto free_return;
664 }
665 }
666 else
667 mctx.state_log = NULL;
668
669 match_first = start;
670 mctx.input.tip_context = (eflags & REG_NOTBOL) ? CONTEXT_BEGBUF
671 : CONTEXT_NEWLINE | CONTEXT_BEGBUF;
672
673 /* Check incrementally whether of not the input string match. */
674 incr = (range < 0) ? -1 : 1;
675 left_lim = (range < 0) ? start + range : start;
676 right_lim = (range < 0) ? start : start + range;
677 sb = dfa->mb_cur_max == 1;
678 match_kind =
679 (fastmap
680 ? ((sb || !(preg->syntax & RE_ICASE || t) ? 4 : 0)
681 | (range >= 0 ? 2 : 0)
682 | (t != NULL ? 1 : 0))
683 : 8);
684
685 for (;; match_first += incr)
686 {
687 err = REG_NOMATCH;
688 if (match_first < left_lim || right_lim < match_first)
689 goto free_return;
690
691 /* Advance as rapidly as possible through the string, until we
692 find a plausible place to start matching. This may be done
693 with varying efficiency, so there are various possibilities:
694 only the most common of them are specialized, in order to
695 save on code size. We use a switch statement for speed. */
696 switch (match_kind)
697 {
698 case 8:
699 /* No fastmap. */
700 break;
701
702 case 7:
703 /* Fastmap with single-byte translation, match forward. */
704 while (BE (match_first < right_lim, 1)
705 && !fastmap[t[(unsigned char) string[match_first]]])
706 ++match_first;
707 goto forward_match_found_start_or_reached_end;
708
709 case 6:
710 /* Fastmap without translation, match forward. */
711 while (BE (match_first < right_lim, 1)
712 && !fastmap[(unsigned char) string[match_first]])
713 ++match_first;
714
715 forward_match_found_start_or_reached_end:
716 if (BE (match_first == right_lim, 0))
717 {
718 ch = match_first >= length
719 ? 0 : (unsigned char) string[match_first];
720 if (!fastmap[t ? t[ch] : ch])
721 goto free_return;
722 }
723 break;
724
725 case 4:
726 case 5:
727 /* Fastmap without multi-byte translation, match backwards. */
728 while (match_first >= left_lim)
729 {
730 ch = match_first >= length
731 ? 0 : (unsigned char) string[match_first];
732 if (fastmap[t ? t[ch] : ch])
733 break;
734 --match_first;
735 }
736 if (match_first < left_lim)
737 goto free_return;
738 break;
739
740 default:
741 /* In this case, we can't determine easily the current byte,
742 since it might be a component byte of a multibyte
743 character. Then we use the constructed buffer instead. */
744 for (;;)
745 {
746 /* If MATCH_FIRST is out of the valid range, reconstruct the
747 buffers. */
748 unsigned int offset = match_first - mctx.input.raw_mbs_idx;
749 if (BE (offset >= (unsigned int) mctx.input.valid_raw_len, 0))
750 {
751 err = re_string_reconstruct (&mctx.input, match_first,
752 eflags);
753 if (BE (err != REG_NOERROR, 0))
754 goto free_return;
755
756 offset = match_first - mctx.input.raw_mbs_idx;
757 }
758 /* If MATCH_FIRST is out of the buffer, leave it as '\0'.
759 Note that MATCH_FIRST must not be smaller than 0. */
760 ch = (match_first >= length
761 ? 0 : re_string_byte_at (&mctx.input, offset));
762 if (fastmap[ch])
763 break;
764 match_first += incr;
765 if (match_first < left_lim || match_first > right_lim)
766 {
767 err = REG_NOMATCH;
768 goto free_return;
769 }
770 }
771 break;
772 }
773
774 /* Reconstruct the buffers so that the matcher can assume that
775 the matching starts from the beginning of the buffer. */
776 err = re_string_reconstruct (&mctx.input, match_first, eflags);
777 if (BE (err != REG_NOERROR, 0))
778 goto free_return;
779
780#ifdef RE_ENABLE_I18N
781 /* Don't consider this char as a possible match start if it part,
782 yet isn't the head, of a multibyte character. */
783 if (!sb && !re_string_first_byte (&mctx.input, 0))
784 continue;
785#endif
786
787 /* It seems to be appropriate one, then use the matcher. */
788 /* We assume that the matching starts from 0. */
789 mctx.state_log_top = mctx.nbkref_ents = mctx.max_mb_elem_len = 0;
790 match_last = check_matching (&mctx, fl_longest_match,
791 range >= 0 ? &match_first : NULL);
792 if (match_last != -1)
793 {
794 if (BE (match_last == -2, 0))
795 {
796 err = REG_ESPACE;
797 goto free_return;
798 }
799 else
800 {
801 mctx.match_last = match_last;
802 if ((!preg->no_sub && nmatch > 1) || dfa->nbackref)
803 {
804 re_dfastate_t *pstate = mctx.state_log[match_last];
805 mctx.last_node = check_halt_state_context (&mctx, pstate,
806 match_last);
807 }
808 if ((!preg->no_sub && nmatch > 1 && dfa->has_plural_match)
809 || dfa->nbackref)
810 {
811 err = prune_impossible_nodes (&mctx);
812 if (err == REG_NOERROR)
813 break;
814 if (BE (err != REG_NOMATCH, 0))
815 goto free_return;
816 match_last = -1;
817 }
818 else
819 break; /* We found a match. */
820 }
821 }
822
823 match_ctx_clean (&mctx);
824 }
825
826#ifdef DEBUG
827 assert (match_last != -1);
828 assert (err == REG_NOERROR);
829#endif
830
831 /* Set pmatch[] if we need. */
832 if (nmatch > 0)
833 {
834 int reg_idx;
835
836 /* Initialize registers. */
837 for (reg_idx = 1; reg_idx < nmatch; ++reg_idx)
838 pmatch[reg_idx].rm_so = pmatch[reg_idx].rm_eo = -1;
839
840 /* Set the points where matching start/end. */
841 pmatch[0].rm_so = 0;
842 pmatch[0].rm_eo = mctx.match_last;
843
844 if (!preg->no_sub && nmatch > 1)
845 {
846 err = set_regs (preg, &mctx, nmatch, pmatch,
847 dfa->has_plural_match && dfa->nbackref > 0);
848 if (BE (err != REG_NOERROR, 0))
849 goto free_return;
850 }
851
852 /* At last, add the offset to each register, since we slid
853 the buffers so that we could assume that the matching starts
854 from 0. */
855 for (reg_idx = 0; reg_idx < nmatch; ++reg_idx)
856 if (pmatch[reg_idx].rm_so != -1)
857 {
858#ifdef RE_ENABLE_I18N
859 if (BE (mctx.input.offsets_needed != 0, 0))
860 {
861 pmatch[reg_idx].rm_so =
862 (pmatch[reg_idx].rm_so == mctx.input.valid_len
863 ? mctx.input.valid_raw_len
864 : mctx.input.offsets[pmatch[reg_idx].rm_so]);
865 pmatch[reg_idx].rm_eo =
866 (pmatch[reg_idx].rm_eo == mctx.input.valid_len
867 ? mctx.input.valid_raw_len
868 : mctx.input.offsets[pmatch[reg_idx].rm_eo]);
869 }
870#else
871 assert (mctx.input.offsets_needed == 0);
872#endif
873 pmatch[reg_idx].rm_so += match_first;
874 pmatch[reg_idx].rm_eo += match_first;
875 }
876 for (reg_idx = 0; reg_idx < extra_nmatch; ++reg_idx)
877 {
878 pmatch[nmatch + reg_idx].rm_so = -1;
879 pmatch[nmatch + reg_idx].rm_eo = -1;
880 }
881
882 if (dfa->subexp_map)
883 for (reg_idx = 0; reg_idx + 1 < nmatch; reg_idx++)
884 if (dfa->subexp_map[reg_idx] != reg_idx)
885 {
886 pmatch[reg_idx + 1].rm_so
887 = pmatch[dfa->subexp_map[reg_idx] + 1].rm_so;
888 pmatch[reg_idx + 1].rm_eo
889 = pmatch[dfa->subexp_map[reg_idx] + 1].rm_eo;
890 }
891 }
892
893 free_return:
894 re_free (mctx.state_log);
895 if (dfa->nbackref)
896 match_ctx_free (&mctx);
897 re_string_destruct (&mctx.input);
898 return err;
899}
900
901static reg_errcode_t
902__attribute_warn_unused_result__
903prune_impossible_nodes (re_match_context_t *mctx)
904{
905 const re_dfa_t *const dfa = mctx->dfa;
906 int halt_node, match_last;
907 reg_errcode_t ret;
908 re_dfastate_t **sifted_states;
909 re_dfastate_t **lim_states = NULL;
910 re_sift_context_t sctx;
911#ifdef DEBUG
912 assert (mctx->state_log != NULL);
913#endif
914 match_last = mctx->match_last;
915 halt_node = mctx->last_node;
916
917 /* Avoid overflow. */
918 if (BE (SIZE_MAX / sizeof (re_dfastate_t *) <= match_last, 0))
919 return REG_ESPACE;
920
921 sifted_states = re_malloc (re_dfastate_t *, match_last + 1);
922 if (BE (sifted_states == NULL, 0))
923 {
924 ret = REG_ESPACE;
925 goto free_return;
926 }
927 if (dfa->nbackref)
928 {
929 lim_states = re_malloc (re_dfastate_t *, match_last + 1);
930 if (BE (lim_states == NULL, 0))
931 {
932 ret = REG_ESPACE;
933 goto free_return;
934 }
935 while (1)
936 {
937 memset (lim_states, '\0',
938 sizeof (re_dfastate_t *) * (match_last + 1));
939 sift_ctx_init (&sctx, sifted_states, lim_states, halt_node,
940 match_last);
941 ret = sift_states_backward (mctx, &sctx);
942 re_node_set_free (&sctx.limits);
943 if (BE (ret != REG_NOERROR, 0))
944 goto free_return;
945 if (sifted_states[0] != NULL || lim_states[0] != NULL)
946 break;
947 do
948 {
949 --match_last;
950 if (match_last < 0)
951 {
952 ret = REG_NOMATCH;
953 goto free_return;
954 }
955 } while (mctx->state_log[match_last] == NULL
956 || !mctx->state_log[match_last]->halt);
957 halt_node = check_halt_state_context (mctx,
958 mctx->state_log[match_last],
959 match_last);
960 }
961 ret = merge_state_array (dfa, sifted_states, lim_states,
962 match_last + 1);
963 re_free (lim_states);
964 lim_states = NULL;
965 if (BE (ret != REG_NOERROR, 0))
966 goto free_return;
967 }
968 else
969 {
970 sift_ctx_init (&sctx, sifted_states, lim_states, halt_node, match_last);
971 ret = sift_states_backward (mctx, &sctx);
972 re_node_set_free (&sctx.limits);
973 if (BE (ret != REG_NOERROR, 0))
974 goto free_return;
975 if (sifted_states[0] == NULL)
976 {
977 ret = REG_NOMATCH;
978 goto free_return;
979 }
980 }
981 re_free (mctx->state_log);
982 mctx->state_log = sifted_states;
983 sifted_states = NULL;
984 mctx->last_node = halt_node;
985 mctx->match_last = match_last;
986 ret = REG_NOERROR;
987 free_return:
988 re_free (sifted_states);
989 re_free (lim_states);
990 return ret;
991}
992
993/* Acquire an initial state and return it.
994 We must select appropriate initial state depending on the context,
995 since initial states may have constraints like "\<", "^", etc.. */
996
997static inline re_dfastate_t *
998__attribute ((always_inline))
999acquire_init_state_context (reg_errcode_t *err, const re_match_context_t *mctx,
1000 int idx)
1001{
1002 const re_dfa_t *const dfa = mctx->dfa;
1003 if (dfa->init_state->has_constraint)
1004 {
1005 unsigned int context;
1006 context = re_string_context_at (&mctx->input, idx - 1, mctx->eflags);
1007 if (IS_WORD_CONTEXT (context))
1008 return dfa->init_state_word;
1009 else if (IS_ORDINARY_CONTEXT (context))
1010 return dfa->init_state;
1011 else if (IS_BEGBUF_CONTEXT (context) && IS_NEWLINE_CONTEXT (context))
1012 return dfa->init_state_begbuf;
1013 else if (IS_NEWLINE_CONTEXT (context))
1014 return dfa->init_state_nl;
1015 else if (IS_BEGBUF_CONTEXT (context))
1016 {
1017 /* It is relatively rare case, then calculate on demand. */
1018 return re_acquire_state_context (err, dfa,
1019 dfa->init_state->entrance_nodes,
1020 context);
1021 }
1022 else
1023 /* Must not happen? */
1024 return dfa->init_state;
1025 }
1026 else
1027 return dfa->init_state;
1028}
1029
1030/* Check whether the regular expression match input string INPUT or not,
1031 and return the index where the matching end, return -1 if not match,
1032 or return -2 in case of an error.
1033 FL_LONGEST_MATCH means we want the POSIX longest matching.
1034 If P_MATCH_FIRST is not NULL, and the match fails, it is set to the
1035 next place where we may want to try matching.
1036 Note that the matcher assume that the maching starts from the current
1037 index of the buffer. */
1038
1039static int
1040__attribute_warn_unused_result__
1041check_matching (re_match_context_t *mctx, int fl_longest_match,
1042 int *p_match_first)
1043{
1044 const re_dfa_t *const dfa = mctx->dfa;
1045 reg_errcode_t err;
1046 int match = 0;
1047 int match_last = -1;
1048 int cur_str_idx = re_string_cur_idx (&mctx->input);
1049 re_dfastate_t *cur_state;
1050 int at_init_state = p_match_first != NULL;
1051 int next_start_idx = cur_str_idx;
1052
1053 err = REG_NOERROR;
1054 cur_state = acquire_init_state_context (&err, mctx, cur_str_idx);
1055 /* An initial state must not be NULL (invalid). */
1056 if (BE (cur_state == NULL, 0))
1057 {
1058 assert (err == REG_ESPACE);
1059 return -2;
1060 }
1061
1062 if (mctx->state_log != NULL)
1063 {
1064 mctx->state_log[cur_str_idx] = cur_state;
1065
1066 /* Check OP_OPEN_SUBEXP in the initial state in case that we use them
1067 later. E.g. Processing back references. */
1068 if (BE (dfa->nbackref, 0))
1069 {
1070 at_init_state = 0;
1071 err = check_subexp_matching_top (mctx, &cur_state->nodes, 0);
1072 if (BE (err != REG_NOERROR, 0))
1073 return err;
1074
1075 if (cur_state->has_backref)
1076 {
1077 err = transit_state_bkref (mctx, &cur_state->nodes);
1078 if (BE (err != REG_NOERROR, 0))
1079 return err;
1080 }
1081 }
1082 }
1083
1084 /* If the RE accepts NULL string. */
1085 if (BE (cur_state->halt, 0))
1086 {
1087 if (!cur_state->has_constraint
1088 || check_halt_state_context (mctx, cur_state, cur_str_idx))
1089 {
1090 if (!fl_longest_match)
1091 return cur_str_idx;
1092 else
1093 {
1094 match_last = cur_str_idx;
1095 match = 1;
1096 }
1097 }
1098 }
1099
1100 while (!re_string_eoi (&mctx->input))
1101 {
1102 re_dfastate_t *old_state = cur_state;
1103 int next_char_idx = re_string_cur_idx (&mctx->input) + 1;
1104
1105 if ((BE (next_char_idx >= mctx->input.bufs_len, 0)
1106 && mctx->input.bufs_len < mctx->input.len)
1107 || (BE (next_char_idx >= mctx->input.valid_len, 0)
1108 && mctx->input.valid_len < mctx->input.len))
1109 {
1110 err = extend_buffers (mctx, next_char_idx + 1);
1111 if (BE (err != REG_NOERROR, 0))
1112 {
1113 assert (err == REG_ESPACE);
1114 return -2;
1115 }
1116 }
1117
1118 cur_state = transit_state (&err, mctx, cur_state);
1119 if (mctx->state_log != NULL)
1120 cur_state = merge_state_with_log (&err, mctx, cur_state);
1121
1122 if (cur_state == NULL)
1123 {
1124 /* Reached the invalid state or an error. Try to recover a valid
1125 state using the state log, if available and if we have not
1126 already found a valid (even if not the longest) match. */
1127 if (BE (err != REG_NOERROR, 0))
1128 return -2;
1129
1130 if (mctx->state_log == NULL
1131 || (match && !fl_longest_match)
1132 || (cur_state = find_recover_state (&err, mctx)) == NULL)
1133 break;
1134 }
1135
1136 if (BE (at_init_state, 0))
1137 {
1138 if (old_state == cur_state)
1139 next_start_idx = next_char_idx;
1140 else
1141 at_init_state = 0;
1142 }
1143
1144 if (cur_state->halt)
1145 {
1146 /* Reached a halt state.
1147 Check the halt state can satisfy the current context. */
1148 if (!cur_state->has_constraint
1149 || check_halt_state_context (mctx, cur_state,
1150 re_string_cur_idx (&mctx->input)))
1151 {
1152 /* We found an appropriate halt state. */
1153 match_last = re_string_cur_idx (&mctx->input);
1154 match = 1;
1155
1156 /* We found a match, do not modify match_first below. */
1157 p_match_first = NULL;
1158 if (!fl_longest_match)
1159 break;
1160 }
1161 }
1162 }
1163
1164 if (p_match_first)
1165 *p_match_first += next_start_idx;
1166
1167 return match_last;
1168}
1169
1170/* Check NODE match the current context. */
1171
1172static int
1173check_halt_node_context (const re_dfa_t *dfa, int node, unsigned int context)
1174{
1175 re_token_type_t type = dfa->nodes[node].type;
1176 unsigned int constraint = dfa->nodes[node].constraint;
1177 if (type != END_OF_RE)
1178 return 0;
1179 if (!constraint)
1180 return 1;
1181 if (NOT_SATISFY_NEXT_CONSTRAINT (constraint, context))
1182 return 0;
1183 return 1;
1184}
1185
1186/* Check the halt state STATE match the current context.
1187 Return 0 if not match, if the node, STATE has, is a halt node and
1188 match the context, return the node. */
1189
1190static int
1191check_halt_state_context (const re_match_context_t *mctx,
1192 const re_dfastate_t *state, int idx)
1193{
1194 int i;
1195 unsigned int context;
1196#ifdef DEBUG
1197 assert (state->halt);
1198#endif
1199 context = re_string_context_at (&mctx->input, idx, mctx->eflags);
1200 for (i = 0; i < state->nodes.nelem; ++i)
1201 if (check_halt_node_context (mctx->dfa, state->nodes.elems[i], context))
1202 return state->nodes.elems[i];
1203 return 0;
1204}
1205
1206/* Compute the next node to which "NFA" transit from NODE("NFA" is a NFA
1207 corresponding to the DFA).
1208 Return the destination node, and update EPS_VIA_NODES, return -1 in case
1209 of errors. */
1210
1211static int
1212proceed_next_node (const re_match_context_t *mctx, int nregs, regmatch_t *regs,
1213 int *pidx, int node, re_node_set *eps_via_nodes,
1214 struct re_fail_stack_t *fs)
1215{
1216 const re_dfa_t *const dfa = mctx->dfa;
1217 int i, err;
1218 if (IS_EPSILON_NODE (dfa->nodes[node].type))
1219 {
1220 re_node_set *cur_nodes = &mctx->state_log[*pidx]->nodes;
1221 re_node_set *edests = &dfa->edests[node];
1222 int dest_node;
1223 err = re_node_set_insert (eps_via_nodes, node);
1224 if (BE (err < 0, 0))
1225 return -2;
1226 /* Pick up a valid destination, or return -1 if none is found. */
1227 for (dest_node = -1, i = 0; i < edests->nelem; ++i)
1228 {
1229 int candidate = edests->elems[i];
1230 if (!re_node_set_contains (cur_nodes, candidate))
1231 continue;
1232 if (dest_node == -1)
1233 dest_node = candidate;
1234
1235 else
1236 {
1237 /* In order to avoid infinite loop like "(a*)*", return the second
1238 epsilon-transition if the first was already considered. */
1239 if (re_node_set_contains (eps_via_nodes, dest_node))
1240 return candidate;
1241
1242 /* Otherwise, push the second epsilon-transition on the fail stack. */
1243 else if (fs != NULL
1244 && push_fail_stack (fs, *pidx, candidate, nregs, regs,
1245 eps_via_nodes))
1246 return -2;
1247
1248 /* We know we are going to exit. */
1249 break;
1250 }
1251 }
1252 return dest_node;
1253 }
1254 else
1255 {
1256 int naccepted = 0;
1257 re_token_type_t type = dfa->nodes[node].type;
1258
1259#ifdef RE_ENABLE_I18N
1260 if (dfa->nodes[node].accept_mb)
1261 naccepted = check_node_accept_bytes (dfa, node, &mctx->input, *pidx);
1262 else
1263#endif /* RE_ENABLE_I18N */
1264 if (type == OP_BACK_REF)
1265 {
1266 int subexp_idx = dfa->nodes[node].opr.idx + 1;
1267 naccepted = regs[subexp_idx].rm_eo - regs[subexp_idx].rm_so;
1268 if (fs != NULL)
1269 {
1270 if (regs[subexp_idx].rm_so == -1 || regs[subexp_idx].rm_eo == -1)
1271 return -1;
1272 else if (naccepted)
1273 {
1274 char *buf = (char *) re_string_get_buffer (&mctx->input);
1275 if (memcmp (buf + regs[subexp_idx].rm_so, buf + *pidx,
1276 naccepted) != 0)
1277 return -1;
1278 }
1279 }
1280
1281 if (naccepted == 0)
1282 {
1283 int dest_node;
1284 err = re_node_set_insert (eps_via_nodes, node);
1285 if (BE (err < 0, 0))
1286 return -2;
1287 dest_node = dfa->edests[node].elems[0];
1288 if (re_node_set_contains (&mctx->state_log[*pidx]->nodes,
1289 dest_node))
1290 return dest_node;
1291 }
1292 }
1293
1294 if (naccepted != 0
1295 || check_node_accept (mctx, dfa->nodes + node, *pidx))
1296 {
1297 int dest_node = dfa->nexts[node];
1298 *pidx = (naccepted == 0) ? *pidx + 1 : *pidx + naccepted;
1299 if (fs && (*pidx > mctx->match_last || mctx->state_log[*pidx] == NULL
1300 || !re_node_set_contains (&mctx->state_log[*pidx]->nodes,
1301 dest_node)))
1302 return -1;
1303 re_node_set_empty (eps_via_nodes);
1304 return dest_node;
1305 }
1306 }
1307 return -1;
1308}
1309
1310static reg_errcode_t
1311__attribute_warn_unused_result__
1312push_fail_stack (struct re_fail_stack_t *fs, int str_idx, int dest_node,
1313 int nregs, regmatch_t *regs, re_node_set *eps_via_nodes)
1314{
1315 reg_errcode_t err;
1316 int num = fs->num++;
1317 if (fs->num == fs->alloc)
1318 {
1319 struct re_fail_stack_ent_t *new_array;
1320 new_array = realloc (fs->stack, (sizeof (struct re_fail_stack_ent_t)
1321 * fs->alloc * 2));
1322 if (new_array == NULL)
1323 return REG_ESPACE;
1324 fs->alloc *= 2;
1325 fs->stack = new_array;
1326 }
1327 fs->stack[num].idx = str_idx;
1328 fs->stack[num].node = dest_node;
1329 fs->stack[num].regs = re_malloc (regmatch_t, nregs);
1330 if (fs->stack[num].regs == NULL)
1331 return REG_ESPACE;
1332 memcpy (fs->stack[num].regs, regs, sizeof (regmatch_t) * nregs);
1333 err = re_node_set_init_copy (&fs->stack[num].eps_via_nodes, eps_via_nodes);
1334 return err;
1335}
1336
1337static int
1338pop_fail_stack (struct re_fail_stack_t *fs, int *pidx, int nregs,
1339 regmatch_t *regs, re_node_set *eps_via_nodes)
1340{
1341 int num = --fs->num;
1342 assert (num >= 0);
1343 *pidx = fs->stack[num].idx;
1344 memcpy (regs, fs->stack[num].regs, sizeof (regmatch_t) * nregs);
1345 re_node_set_free (eps_via_nodes);
1346 re_free (fs->stack[num].regs);
1347 *eps_via_nodes = fs->stack[num].eps_via_nodes;
1348 return fs->stack[num].node;
1349}
1350
1351/* Set the positions where the subexpressions are starts/ends to registers
1352 PMATCH.
1353 Note: We assume that pmatch[0] is already set, and
1354 pmatch[i].rm_so == pmatch[i].rm_eo == -1 for 0 < i < nmatch. */
1355
1356static reg_errcode_t
1357__attribute_warn_unused_result__
1358set_regs (const regex_t *preg, const re_match_context_t *mctx, size_t nmatch,
1359 regmatch_t *pmatch, int fl_backtrack)
1360{
1361 const re_dfa_t *dfa = (const re_dfa_t *) preg->buffer;
1362 int idx, cur_node;
1363 re_node_set eps_via_nodes;
1364 struct re_fail_stack_t *fs;
1365 struct re_fail_stack_t fs_body = { 0, 2, NULL };
1366 regmatch_t *prev_idx_match;
1367 int prev_idx_match_malloced = 0;
1368
1369#ifdef DEBUG
1370 assert (nmatch > 1);
1371 assert (mctx->state_log != NULL);
1372#endif
1373 if (fl_backtrack)
1374 {
1375 fs = &fs_body;
1376 fs->stack = re_malloc (struct re_fail_stack_ent_t, fs->alloc);
1377 if (fs->stack == NULL)
1378 return REG_ESPACE;
1379 }
1380 else
1381 fs = NULL;
1382
1383 cur_node = dfa->init_node;
1384 re_node_set_init_empty (&eps_via_nodes);
1385
1386 if (__libc_use_alloca (nmatch * sizeof (regmatch_t)))
1387 prev_idx_match = (regmatch_t *) alloca (nmatch * sizeof (regmatch_t));
1388 else
1389 {
1390 prev_idx_match = re_malloc (regmatch_t, nmatch);
1391 if (prev_idx_match == NULL)
1392 {
1393 free_fail_stack_return (fs);
1394 return REG_ESPACE;
1395 }
1396 prev_idx_match_malloced = 1;
1397 }
1398 memcpy (prev_idx_match, pmatch, sizeof (regmatch_t) * nmatch);
1399
1400 for (idx = pmatch[0].rm_so; idx <= pmatch[0].rm_eo ;)
1401 {
1402 update_regs (dfa, pmatch, prev_idx_match, cur_node, idx, nmatch);
1403
1404 if (idx == pmatch[0].rm_eo && cur_node == mctx->last_node)
1405 {
1406 int reg_idx;
1407 if (fs)
1408 {
1409 for (reg_idx = 0; reg_idx < nmatch; ++reg_idx)
1410 if (pmatch[reg_idx].rm_so > -1 && pmatch[reg_idx].rm_eo == -1)
1411 break;
1412 if (reg_idx == nmatch)
1413 {
1414 re_node_set_free (&eps_via_nodes);
1415 if (prev_idx_match_malloced)
1416 re_free (prev_idx_match);
1417 return free_fail_stack_return (fs);
1418 }
1419 cur_node = pop_fail_stack (fs, &idx, nmatch, pmatch,
1420 &eps_via_nodes);
1421 }
1422 else
1423 {
1424 re_node_set_free (&eps_via_nodes);
1425 if (prev_idx_match_malloced)
1426 re_free (prev_idx_match);
1427 return REG_NOERROR;
1428 }
1429 }
1430
1431 /* Proceed to next node. */
1432 cur_node = proceed_next_node (mctx, nmatch, pmatch, &idx, cur_node,
1433 &eps_via_nodes, fs);
1434
1435 if (BE (cur_node < 0, 0))
1436 {
1437 if (BE (cur_node == -2, 0))
1438 {
1439 re_node_set_free (&eps_via_nodes);
1440 if (prev_idx_match_malloced)
1441 re_free (prev_idx_match);
1442 free_fail_stack_return (fs);
1443 return REG_ESPACE;
1444 }
1445 if (fs)
1446 cur_node = pop_fail_stack (fs, &idx, nmatch, pmatch,
1447 &eps_via_nodes);
1448 else
1449 {
1450 re_node_set_free (&eps_via_nodes);
1451 if (prev_idx_match_malloced)
1452 re_free (prev_idx_match);
1453 return REG_NOMATCH;
1454 }
1455 }
1456 }
1457 re_node_set_free (&eps_via_nodes);
1458 if (prev_idx_match_malloced)
1459 re_free (prev_idx_match);
1460 return free_fail_stack_return (fs);
1461}
1462
1463static reg_errcode_t
1464free_fail_stack_return (struct re_fail_stack_t *fs)
1465{
1466 if (fs)
1467 {
1468 int fs_idx;
1469 for (fs_idx = 0; fs_idx < fs->num; ++fs_idx)
1470 {
1471 re_node_set_free (&fs->stack[fs_idx].eps_via_nodes);
1472 re_free (fs->stack[fs_idx].regs);
1473 }
1474 re_free (fs->stack);
1475 }
1476 return REG_NOERROR;
1477}
1478
1479static void
1480update_regs (const re_dfa_t *dfa, regmatch_t *pmatch,
1481 regmatch_t *prev_idx_match, int cur_node, int cur_idx, int nmatch)
1482{
1483 int type = dfa->nodes[cur_node].type;
1484 if (type == OP_OPEN_SUBEXP)
1485 {
1486 int reg_num = dfa->nodes[cur_node].opr.idx + 1;
1487
1488 /* We are at the first node of this sub expression. */
1489 if (reg_num < nmatch)
1490 {
1491 pmatch[reg_num].rm_so = cur_idx;
1492 pmatch[reg_num].rm_eo = -1;
1493 }
1494 }
1495 else if (type == OP_CLOSE_SUBEXP)
1496 {
1497 int reg_num = dfa->nodes[cur_node].opr.idx + 1;
1498 if (reg_num < nmatch)
1499 {
1500 /* We are at the last node of this sub expression. */
1501 if (pmatch[reg_num].rm_so < cur_idx)
1502 {
1503 pmatch[reg_num].rm_eo = cur_idx;
1504 /* This is a non-empty match or we are not inside an optional
1505 subexpression. Accept this right away. */
1506 memcpy (prev_idx_match, pmatch, sizeof (regmatch_t) * nmatch);
1507 }
1508 else
1509 {
1510 if (dfa->nodes[cur_node].opt_subexp
1511 && prev_idx_match[reg_num].rm_so != -1)
1512 /* We transited through an empty match for an optional
1513 subexpression, like (a?)*, and this is not the subexp's
1514 first match. Copy back the old content of the registers
1515 so that matches of an inner subexpression are undone as
1516 well, like in ((a?))*. */
1517 memcpy (pmatch, prev_idx_match, sizeof (regmatch_t) * nmatch);
1518 else
1519 /* We completed a subexpression, but it may be part of
1520 an optional one, so do not update PREV_IDX_MATCH. */
1521 pmatch[reg_num].rm_eo = cur_idx;
1522 }
1523 }
1524 }
1525}
1526
1527/* This function checks the STATE_LOG from the SCTX->last_str_idx to 0
1528 and sift the nodes in each states according to the following rules.
1529 Updated state_log will be wrote to STATE_LOG.
1530
1531 Rules: We throw away the Node `a' in the STATE_LOG[STR_IDX] if...
1532 1. When STR_IDX == MATCH_LAST(the last index in the state_log):
1533 If `a' isn't the LAST_NODE and `a' can't epsilon transit to
1534 the LAST_NODE, we throw away the node `a'.
1535 2. When 0 <= STR_IDX < MATCH_LAST and `a' accepts
1536 string `s' and transit to `b':
1537 i. If 'b' isn't in the STATE_LOG[STR_IDX+strlen('s')], we throw
1538 away the node `a'.
1539 ii. If 'b' is in the STATE_LOG[STR_IDX+strlen('s')] but 'b' is
1540 thrown away, we throw away the node `a'.
1541 3. When 0 <= STR_IDX < MATCH_LAST and 'a' epsilon transit to 'b':
1542 i. If 'b' isn't in the STATE_LOG[STR_IDX], we throw away the
1543 node `a'.
1544 ii. If 'b' is in the STATE_LOG[STR_IDX] but 'b' is thrown away,
1545 we throw away the node `a'. */
1546
1547#define STATE_NODE_CONTAINS(state,node) \
1548 ((state) != NULL && re_node_set_contains (&(state)->nodes, node))
1549
1550static reg_errcode_t
1551sift_states_backward (const re_match_context_t *mctx, re_sift_context_t *sctx)
1552{
1553 reg_errcode_t err;
1554 int null_cnt = 0;
1555 int str_idx = sctx->last_str_idx;
1556 re_node_set cur_dest;
1557
1558#ifdef DEBUG
1559 assert (mctx->state_log != NULL && mctx->state_log[str_idx] != NULL);
1560#endif
1561
1562 /* Build sifted state_log[str_idx]. It has the nodes which can epsilon
1563 transit to the last_node and the last_node itself. */
1564 err = re_node_set_init_1 (&cur_dest, sctx->last_node);
1565 if (BE (err != REG_NOERROR, 0))
1566 return err;
1567 err = update_cur_sifted_state (mctx, sctx, str_idx, &cur_dest);
1568 if (BE (err != REG_NOERROR, 0))
1569 goto free_return;
1570
1571 /* Then check each states in the state_log. */
1572 while (str_idx > 0)
1573 {
1574 /* Update counters. */
1575 null_cnt = (sctx->sifted_states[str_idx] == NULL) ? null_cnt + 1 : 0;
1576 if (null_cnt > mctx->max_mb_elem_len)
1577 {
1578 memset (sctx->sifted_states, '\0',
1579 sizeof (re_dfastate_t *) * str_idx);
1580 re_node_set_free (&cur_dest);
1581 return REG_NOERROR;
1582 }
1583 re_node_set_empty (&cur_dest);
1584 --str_idx;
1585
1586 if (mctx->state_log[str_idx])
1587 {
1588 err = build_sifted_states (mctx, sctx, str_idx, &cur_dest);
1589 if (BE (err != REG_NOERROR, 0))
1590 goto free_return;
1591 }
1592
1593 /* Add all the nodes which satisfy the following conditions:
1594 - It can epsilon transit to a node in CUR_DEST.
1595 - It is in CUR_SRC.
1596 And update state_log. */
1597 err = update_cur_sifted_state (mctx, sctx, str_idx, &cur_dest);
1598 if (BE (err != REG_NOERROR, 0))
1599 goto free_return;
1600 }
1601 err = REG_NOERROR;
1602 free_return:
1603 re_node_set_free (&cur_dest);
1604 return err;
1605}
1606
1607static reg_errcode_t
1608__attribute_warn_unused_result__
1609build_sifted_states (const re_match_context_t *mctx, re_sift_context_t *sctx,
1610 int str_idx, re_node_set *cur_dest)
1611{
1612 const re_dfa_t *const dfa = mctx->dfa;
1613 const re_node_set *cur_src = &mctx->state_log[str_idx]->non_eps_nodes;
1614 int i;
1615
1616 /* Then build the next sifted state.
1617 We build the next sifted state on `cur_dest', and update
1618 `sifted_states[str_idx]' with `cur_dest'.
1619 Note:
1620 `cur_dest' is the sifted state from `state_log[str_idx + 1]'.
1621 `cur_src' points the node_set of the old `state_log[str_idx]'
1622 (with the epsilon nodes pre-filtered out). */
1623 for (i = 0; i < cur_src->nelem; i++)
1624 {
1625 int prev_node = cur_src->elems[i];
1626 int naccepted = 0;
1627 int ret;
1628
1629#ifdef DEBUG
1630 re_token_type_t type = dfa->nodes[prev_node].type;
1631 assert (!IS_EPSILON_NODE (type));
1632#endif
1633#ifdef RE_ENABLE_I18N
1634 /* If the node may accept `multi byte'. */
1635 if (dfa->nodes[prev_node].accept_mb)
1636 naccepted = sift_states_iter_mb (mctx, sctx, prev_node,
1637 str_idx, sctx->last_str_idx);
1638#endif /* RE_ENABLE_I18N */
1639
1640 /* We don't check backreferences here.
1641 See update_cur_sifted_state(). */
1642 if (!naccepted
1643 && check_node_accept (mctx, dfa->nodes + prev_node, str_idx)
1644 && STATE_NODE_CONTAINS (sctx->sifted_states[str_idx + 1],
1645 dfa->nexts[prev_node]))
1646 naccepted = 1;
1647
1648 if (naccepted == 0)
1649 continue;
1650
1651 if (sctx->limits.nelem)
1652 {
1653 int to_idx = str_idx + naccepted;
1654 if (check_dst_limits (mctx, &sctx->limits,
1655 dfa->nexts[prev_node], to_idx,
1656 prev_node, str_idx))
1657 continue;
1658 }
1659 ret = re_node_set_insert (cur_dest, prev_node);
1660 if (BE (ret == -1, 0))
1661 return REG_ESPACE;
1662 }
1663
1664 return REG_NOERROR;
1665}
1666
1667/* Helper functions. */
1668
1669static reg_errcode_t
1670clean_state_log_if_needed (re_match_context_t *mctx, int next_state_log_idx)
1671{
1672 int top = mctx->state_log_top;
1673
1674 if ((next_state_log_idx >= mctx->input.bufs_len
1675 && mctx->input.bufs_len < mctx->input.len)
1676 || (next_state_log_idx >= mctx->input.valid_len
1677 && mctx->input.valid_len < mctx->input.len))
1678 {
1679 reg_errcode_t err;
1680 err = extend_buffers (mctx, next_state_log_idx + 1);
1681 if (BE (err != REG_NOERROR, 0))
1682 return err;
1683 }
1684
1685 if (top < next_state_log_idx)
1686 {
1687 memset (mctx->state_log + top + 1, '\0',
1688 sizeof (re_dfastate_t *) * (next_state_log_idx - top));
1689 mctx->state_log_top = next_state_log_idx;
1690 }
1691 return REG_NOERROR;
1692}
1693
1694static reg_errcode_t
1695merge_state_array (const re_dfa_t *dfa, re_dfastate_t **dst,
1696 re_dfastate_t **src, int num)
1697{
1698 int st_idx;
1699 reg_errcode_t err;
1700 for (st_idx = 0; st_idx < num; ++st_idx)
1701 {
1702 if (dst[st_idx] == NULL)
1703 dst[st_idx] = src[st_idx];
1704 else if (src[st_idx] != NULL)
1705 {
1706 re_node_set merged_set;
1707 err = re_node_set_init_union (&merged_set, &dst[st_idx]->nodes,
1708 &src[st_idx]->nodes);
1709 if (BE (err != REG_NOERROR, 0))
1710 return err;
1711 dst[st_idx] = re_acquire_state (&err, dfa, &merged_set);
1712 re_node_set_free (&merged_set);
1713 if (BE (err != REG_NOERROR, 0))
1714 return err;
1715 }
1716 }
1717 return REG_NOERROR;
1718}
1719
1720static reg_errcode_t
1721update_cur_sifted_state (const re_match_context_t *mctx,
1722 re_sift_context_t *sctx, int str_idx,
1723 re_node_set *dest_nodes)
1724{
1725 const re_dfa_t *const dfa = mctx->dfa;
1726 reg_errcode_t err = REG_NOERROR;
1727 const re_node_set *candidates;
1728 candidates = ((mctx->state_log[str_idx] == NULL) ? NULL
1729 : &mctx->state_log[str_idx]->nodes);
1730
1731 if (dest_nodes->nelem == 0)
1732 sctx->sifted_states[str_idx] = NULL;
1733 else
1734 {
1735 if (candidates)
1736 {
1737 /* At first, add the nodes which can epsilon transit to a node in
1738 DEST_NODE. */
1739 err = add_epsilon_src_nodes (dfa, dest_nodes, candidates);
1740 if (BE (err != REG_NOERROR, 0))
1741 return err;
1742
1743 /* Then, check the limitations in the current sift_context. */
1744 if (sctx->limits.nelem)
1745 {
1746 err = check_subexp_limits (dfa, dest_nodes, candidates, &sctx->limits,
1747 mctx->bkref_ents, str_idx);
1748 if (BE (err != REG_NOERROR, 0))
1749 return err;
1750 }
1751 }
1752
1753 sctx->sifted_states[str_idx] = re_acquire_state (&err, dfa, dest_nodes);
1754 if (BE (err != REG_NOERROR, 0))
1755 return err;
1756 }
1757
1758 if (candidates && mctx->state_log[str_idx]->has_backref)
1759 {
1760 err = sift_states_bkref (mctx, sctx, str_idx, candidates);
1761 if (BE (err != REG_NOERROR, 0))
1762 return err;
1763 }
1764 return REG_NOERROR;
1765}
1766
1767static reg_errcode_t
1768__attribute_warn_unused_result__
1769add_epsilon_src_nodes (const re_dfa_t *dfa, re_node_set *dest_nodes,
1770 const re_node_set *candidates)
1771{
1772 reg_errcode_t err = REG_NOERROR;
1773 int i;
1774
1775 re_dfastate_t *state = re_acquire_state (&err, dfa, dest_nodes);
1776 if (BE (err != REG_NOERROR, 0))
1777 return err;
1778
1779 if (!state->inveclosure.alloc)
1780 {
1781 err = re_node_set_alloc (&state->inveclosure, dest_nodes->nelem);
1782 if (BE (err != REG_NOERROR, 0))
1783 return REG_ESPACE;
1784 for (i = 0; i < dest_nodes->nelem; i++)
1785 {
1786 err = re_node_set_merge (&state->inveclosure,
1787 dfa->inveclosures + dest_nodes->elems[i]);
1788 if (BE (err != REG_NOERROR, 0))
1789 return REG_ESPACE;
1790 }
1791 }
1792 return re_node_set_add_intersect (dest_nodes, candidates,
1793 &state->inveclosure);
1794}
1795
1796static reg_errcode_t
1797sub_epsilon_src_nodes (const re_dfa_t *dfa, int node, re_node_set *dest_nodes,
1798 const re_node_set *candidates)
1799{
1800 int ecl_idx;
1801 reg_errcode_t err;
1802 re_node_set *inv_eclosure = dfa->inveclosures + node;
1803 re_node_set except_nodes;
1804 re_node_set_init_empty (&except_nodes);
1805 for (ecl_idx = 0; ecl_idx < inv_eclosure->nelem; ++ecl_idx)
1806 {
1807 int cur_node = inv_eclosure->elems[ecl_idx];
1808 if (cur_node == node)
1809 continue;
1810 if (IS_EPSILON_NODE (dfa->nodes[cur_node].type))
1811 {
1812 int edst1 = dfa->edests[cur_node].elems[0];
1813 int edst2 = ((dfa->edests[cur_node].nelem > 1)
1814 ? dfa->edests[cur_node].elems[1] : -1);
1815 if ((!re_node_set_contains (inv_eclosure, edst1)
1816 && re_node_set_contains (dest_nodes, edst1))
1817 || (edst2 > 0
1818 && !re_node_set_contains (inv_eclosure, edst2)
1819 && re_node_set_contains (dest_nodes, edst2)))
1820 {
1821 err = re_node_set_add_intersect (&except_nodes, candidates,
1822 dfa->inveclosures + cur_node);
1823 if (BE (err != REG_NOERROR, 0))
1824 {
1825 re_node_set_free (&except_nodes);
1826 return err;
1827 }
1828 }
1829 }
1830 }
1831 for (ecl_idx = 0; ecl_idx < inv_eclosure->nelem; ++ecl_idx)
1832 {
1833 int cur_node = inv_eclosure->elems[ecl_idx];
1834 if (!re_node_set_contains (&except_nodes, cur_node))
1835 {
1836 int idx = re_node_set_contains (dest_nodes, cur_node) - 1;
1837 re_node_set_remove_at (dest_nodes, idx);
1838 }
1839 }
1840 re_node_set_free (&except_nodes);
1841 return REG_NOERROR;
1842}
1843
1844static int
1845check_dst_limits (const re_match_context_t *mctx, re_node_set *limits,
1846 int dst_node, int dst_idx, int src_node, int src_idx)
1847{
1848 const re_dfa_t *const dfa = mctx->dfa;
1849 int lim_idx, src_pos, dst_pos;
1850
1851 int dst_bkref_idx = search_cur_bkref_entry (mctx, dst_idx);
1852 int src_bkref_idx = search_cur_bkref_entry (mctx, src_idx);
1853 for (lim_idx = 0; lim_idx < limits->nelem; ++lim_idx)
1854 {
1855 int subexp_idx;
1856 struct re_backref_cache_entry *ent;
1857 ent = mctx->bkref_ents + limits->elems[lim_idx];
1858 subexp_idx = dfa->nodes[ent->node].opr.idx;
1859
1860 dst_pos = check_dst_limits_calc_pos (mctx, limits->elems[lim_idx],
1861 subexp_idx, dst_node, dst_idx,
1862 dst_bkref_idx);
1863 src_pos = check_dst_limits_calc_pos (mctx, limits->elems[lim_idx],
1864 subexp_idx, src_node, src_idx,
1865 src_bkref_idx);
1866
1867 /* In case of:
1868 <src> <dst> ( <subexp> )
1869 ( <subexp> ) <src> <dst>
1870 ( <subexp1> <src> <subexp2> <dst> <subexp3> ) */
1871 if (src_pos == dst_pos)
1872 continue; /* This is unrelated limitation. */
1873 else
1874 return 1;
1875 }
1876 return 0;
1877}
1878
1879static int
1880check_dst_limits_calc_pos_1 (const re_match_context_t *mctx, int boundaries,
1881 int subexp_idx, int from_node, int bkref_idx)
1882{
1883 const re_dfa_t *const dfa = mctx->dfa;
1884 const re_node_set *eclosures = dfa->eclosures + from_node;
1885 int node_idx;
1886
1887 /* Else, we are on the boundary: examine the nodes on the epsilon
1888 closure. */
1889 for (node_idx = 0; node_idx < eclosures->nelem; ++node_idx)
1890 {
1891 int node = eclosures->elems[node_idx];
1892 switch (dfa->nodes[node].type)
1893 {
1894 case OP_BACK_REF:
1895 if (bkref_idx != -1)
1896 {
1897 struct re_backref_cache_entry *ent = mctx->bkref_ents + bkref_idx;
1898 do
1899 {
1900 int dst, cpos;
1901
1902 if (ent->node != node)
1903 continue;
1904
1905 if (subexp_idx < BITSET_WORD_BITS
1906 && !(ent->eps_reachable_subexps_map
1907 & ((bitset_word_t) 1 << subexp_idx)))
1908 continue;
1909
1910 /* Recurse trying to reach the OP_OPEN_SUBEXP and
1911 OP_CLOSE_SUBEXP cases below. But, if the
1912 destination node is the same node as the source
1913 node, don't recurse because it would cause an
1914 infinite loop: a regex that exhibits this behavior
1915 is ()\1*\1* */
1916 dst = dfa->edests[node].elems[0];
1917 if (dst == from_node)
1918 {
1919 if (boundaries & 1)
1920 return -1;
1921 else /* if (boundaries & 2) */
1922 return 0;
1923 }
1924
1925 cpos =
1926 check_dst_limits_calc_pos_1 (mctx, boundaries, subexp_idx,
1927 dst, bkref_idx);
1928 if (cpos == -1 /* && (boundaries & 1) */)
1929 return -1;
1930 if (cpos == 0 && (boundaries & 2))
1931 return 0;
1932
1933 if (subexp_idx < BITSET_WORD_BITS)
1934 ent->eps_reachable_subexps_map
1935 &= ~((bitset_word_t) 1 << subexp_idx);
1936 }
1937 while (ent++->more);
1938 }
1939 break;
1940
1941 case OP_OPEN_SUBEXP:
1942 if ((boundaries & 1) && subexp_idx == dfa->nodes[node].opr.idx)
1943 return -1;
1944 break;
1945
1946 case OP_CLOSE_SUBEXP:
1947 if ((boundaries & 2) && subexp_idx == dfa->nodes[node].opr.idx)
1948 return 0;
1949 break;
1950
1951 default:
1952 break;
1953 }
1954 }
1955
1956 return (boundaries & 2) ? 1 : 0;
1957}
1958
1959static int
1960check_dst_limits_calc_pos (const re_match_context_t *mctx, int limit,
1961 int subexp_idx, int from_node, int str_idx,
1962 int bkref_idx)
1963{
1964 struct re_backref_cache_entry *lim = mctx->bkref_ents + limit;
1965 int boundaries;
1966
1967 /* If we are outside the range of the subexpression, return -1 or 1. */
1968 if (str_idx < lim->subexp_from)
1969 return -1;
1970
1971 if (lim->subexp_to < str_idx)
1972 return 1;
1973
1974 /* If we are within the subexpression, return 0. */
1975 boundaries = (str_idx == lim->subexp_from);
1976 boundaries |= (str_idx == lim->subexp_to) << 1;
1977 if (boundaries == 0)
1978 return 0;
1979
1980 /* Else, examine epsilon closure. */
1981 return check_dst_limits_calc_pos_1 (mctx, boundaries, subexp_idx,
1982 from_node, bkref_idx);
1983}
1984
1985/* Check the limitations of sub expressions LIMITS, and remove the nodes
1986 which are against limitations from DEST_NODES. */
1987
1988static reg_errcode_t
1989check_subexp_limits (const re_dfa_t *dfa, re_node_set *dest_nodes,
1990 const re_node_set *candidates, re_node_set *limits,
1991 struct re_backref_cache_entry *bkref_ents, int str_idx)
1992{
1993 reg_errcode_t err;
1994 int node_idx, lim_idx;
1995
1996 for (lim_idx = 0; lim_idx < limits->nelem; ++lim_idx)
1997 {
1998 int subexp_idx;
1999 struct re_backref_cache_entry *ent;
2000 ent = bkref_ents + limits->elems[lim_idx];
2001
2002 if (str_idx <= ent->subexp_from || ent->str_idx < str_idx)
2003 continue; /* This is unrelated limitation. */
2004
2005 subexp_idx = dfa->nodes[ent->node].opr.idx;
2006 if (ent->subexp_to == str_idx)
2007 {
2008 int ops_node = -1;
2009 int cls_node = -1;
2010 for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx)
2011 {
2012 int node = dest_nodes->elems[node_idx];
2013 re_token_type_t type = dfa->nodes[node].type;
2014 if (type == OP_OPEN_SUBEXP
2015 && subexp_idx == dfa->nodes[node].opr.idx)
2016 ops_node = node;
2017 else if (type == OP_CLOSE_SUBEXP
2018 && subexp_idx == dfa->nodes[node].opr.idx)
2019 cls_node = node;
2020 }
2021
2022 /* Check the limitation of the open subexpression. */
2023 /* Note that (ent->subexp_to = str_idx != ent->subexp_from). */
2024 if (ops_node >= 0)
2025 {
2026 err = sub_epsilon_src_nodes (dfa, ops_node, dest_nodes,
2027 candidates);
2028 if (BE (err != REG_NOERROR, 0))
2029 return err;
2030 }
2031
2032 /* Check the limitation of the close subexpression. */
2033 if (cls_node >= 0)
2034 for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx)
2035 {
2036 int node = dest_nodes->elems[node_idx];
2037 if (!re_node_set_contains (dfa->inveclosures + node,
2038 cls_node)
2039 && !re_node_set_contains (dfa->eclosures + node,
2040 cls_node))
2041 {
2042 /* It is against this limitation.
2043 Remove it form the current sifted state. */
2044 err = sub_epsilon_src_nodes (dfa, node, dest_nodes,
2045 candidates);
2046 if (BE (err != REG_NOERROR, 0))
2047 return err;
2048 --node_idx;
2049 }
2050 }
2051 }
2052 else /* (ent->subexp_to != str_idx) */
2053 {
2054 for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx)
2055 {
2056 int node = dest_nodes->elems[node_idx];
2057 re_token_type_t type = dfa->nodes[node].type;
2058 if (type == OP_CLOSE_SUBEXP || type == OP_OPEN_SUBEXP)
2059 {
2060 if (subexp_idx != dfa->nodes[node].opr.idx)
2061 continue;
2062 /* It is against this limitation.
2063 Remove it form the current sifted state. */
2064 err = sub_epsilon_src_nodes (dfa, node, dest_nodes,
2065 candidates);
2066 if (BE (err != REG_NOERROR, 0))
2067 return err;
2068 }
2069 }
2070 }
2071 }
2072 return REG_NOERROR;
2073}
2074
2075static reg_errcode_t
2076__attribute_warn_unused_result__
2077sift_states_bkref (const re_match_context_t *mctx, re_sift_context_t *sctx,
2078 int str_idx, const re_node_set *candidates)
2079{
2080 const re_dfa_t *const dfa = mctx->dfa;
2081 reg_errcode_t err;
2082 int node_idx, node;
2083 re_sift_context_t local_sctx;
2084 int first_idx = search_cur_bkref_entry (mctx, str_idx);
2085
2086 if (first_idx == -1)
2087 return REG_NOERROR;
2088
2089 local_sctx.sifted_states = NULL; /* Mark that it hasn't been initialized. */
2090
2091 for (node_idx = 0; node_idx < candidates->nelem; ++node_idx)
2092 {
2093 int enabled_idx;
2094 re_token_type_t type;
2095 struct re_backref_cache_entry *entry;
2096 node = candidates->elems[node_idx];
2097 type = dfa->nodes[node].type;
2098 /* Avoid infinite loop for the REs like "()\1+". */
2099 if (node == sctx->last_node && str_idx == sctx->last_str_idx)
2100 continue;
2101 if (type != OP_BACK_REF)
2102 continue;
2103
2104 entry = mctx->bkref_ents + first_idx;
2105 enabled_idx = first_idx;
2106 do
2107 {
2108 int subexp_len;
2109 int to_idx;
2110 int dst_node;
2111 int ret;
2112 re_dfastate_t *cur_state;
2113
2114 if (entry->node != node)
2115 continue;
2116 subexp_len = entry->subexp_to - entry->subexp_from;
2117 to_idx = str_idx + subexp_len;
2118 dst_node = (subexp_len ? dfa->nexts[node]
2119 : dfa->edests[node].elems[0]);
2120
2121 if (to_idx > sctx->last_str_idx
2122 || sctx->sifted_states[to_idx] == NULL
2123 || !STATE_NODE_CONTAINS (sctx->sifted_states[to_idx], dst_node)
2124 || check_dst_limits (mctx, &sctx->limits, node,
2125 str_idx, dst_node, to_idx))
2126 continue;
2127
2128 if (local_sctx.sifted_states == NULL)
2129 {
2130 local_sctx = *sctx;
2131 err = re_node_set_init_copy (&local_sctx.limits, &sctx->limits);
2132 if (BE (err != REG_NOERROR, 0))
2133 goto free_return;
2134 }
2135 local_sctx.last_node = node;
2136 local_sctx.last_str_idx = str_idx;
2137 ret = re_node_set_insert (&local_sctx.limits, enabled_idx);
2138 if (BE (ret < 0, 0))
2139 {
2140 err = REG_ESPACE;
2141 goto free_return;
2142 }
2143 cur_state = local_sctx.sifted_states[str_idx];
2144 err = sift_states_backward (mctx, &local_sctx);
2145 if (BE (err != REG_NOERROR, 0))
2146 goto free_return;
2147 if (sctx->limited_states != NULL)
2148 {
2149 err = merge_state_array (dfa, sctx->limited_states,
2150 local_sctx.sifted_states,
2151 str_idx + 1);
2152 if (BE (err != REG_NOERROR, 0))
2153 goto free_return;
2154 }
2155 local_sctx.sifted_states[str_idx] = cur_state;
2156 re_node_set_remove (&local_sctx.limits, enabled_idx);
2157
2158 /* mctx->bkref_ents may have changed, reload the pointer. */
2159 entry = mctx->bkref_ents + enabled_idx;
2160 }
2161 while (enabled_idx++, entry++->more);
2162 }
2163 err = REG_NOERROR;
2164 free_return:
2165 if (local_sctx.sifted_states != NULL)
2166 {
2167 re_node_set_free (&local_sctx.limits);
2168 }
2169
2170 return err;
2171}
2172
2173
2174#ifdef RE_ENABLE_I18N
2175static int
2176sift_states_iter_mb (const re_match_context_t *mctx, re_sift_context_t *sctx,
2177 int node_idx, int str_idx, int max_str_idx)
2178{
2179 const re_dfa_t *const dfa = mctx->dfa;
2180 int naccepted;
2181 /* Check the node can accept `multi byte'. */
2182 naccepted = check_node_accept_bytes (dfa, node_idx, &mctx->input, str_idx);
2183 if (naccepted > 0 && str_idx + naccepted <= max_str_idx &&
2184 !STATE_NODE_CONTAINS (sctx->sifted_states[str_idx + naccepted],
2185 dfa->nexts[node_idx]))
2186 /* The node can't accept the `multi byte', or the
2187 destination was already thrown away, then the node
2188 could't accept the current input `multi byte'. */
2189 naccepted = 0;
2190 /* Otherwise, it is sure that the node could accept
2191 `naccepted' bytes input. */
2192 return naccepted;
2193}
2194#endif /* RE_ENABLE_I18N */
2195
2196
2197/* Functions for state transition. */
2198
2199/* Return the next state to which the current state STATE will transit by
2200 accepting the current input byte, and update STATE_LOG if necessary.
2201 If STATE can accept a multibyte char/collating element/back reference
2202 update the destination of STATE_LOG. */
2203
2204static re_dfastate_t *
2205__attribute_warn_unused_result__
2206transit_state (reg_errcode_t *err, re_match_context_t *mctx,
2207 re_dfastate_t *state)
2208{
2209 re_dfastate_t **trtable;
2210 unsigned char ch;
2211
2212#ifdef RE_ENABLE_I18N
2213 /* If the current state can accept multibyte. */
2214 if (BE (state->accept_mb, 0))
2215 {
2216 *err = transit_state_mb (mctx, state);
2217 if (BE (*err != REG_NOERROR, 0))
2218 return NULL;
2219 }
2220#endif /* RE_ENABLE_I18N */
2221
2222 /* Then decide the next state with the single byte. */
2223#if 0
2224 if (0)
2225 /* don't use transition table */
2226 return transit_state_sb (err, mctx, state);
2227#endif
2228
2229 /* Use transition table */
2230 ch = re_string_fetch_byte (&mctx->input);
2231 for (;;)
2232 {
2233 trtable = state->trtable;
2234 if (BE (trtable != NULL, 1))
2235 return trtable[ch];
2236
2237 trtable = state->word_trtable;
2238 if (BE (trtable != NULL, 1))
2239 {
2240 unsigned int context;
2241 context
2242 = re_string_context_at (&mctx->input,
2243 re_string_cur_idx (&mctx->input) - 1,
2244 mctx->eflags);
2245 if (IS_WORD_CONTEXT (context))
2246 return trtable[ch + SBC_MAX];
2247 else
2248 return trtable[ch];
2249 }
2250
2251 if (!build_trtable (mctx->dfa, state))
2252 {
2253 *err = REG_ESPACE;
2254 return NULL;
2255 }
2256
2257 /* Retry, we now have a transition table. */
2258 }
2259}
2260
2261/* Update the state_log if we need */
2262re_dfastate_t *
2263merge_state_with_log (reg_errcode_t *err, re_match_context_t *mctx,
2264 re_dfastate_t *next_state)
2265{
2266 const re_dfa_t *const dfa = mctx->dfa;
2267 int cur_idx = re_string_cur_idx (&mctx->input);
2268
2269 if (cur_idx > mctx->state_log_top)
2270 {
2271 mctx->state_log[cur_idx] = next_state;
2272 mctx->state_log_top = cur_idx;
2273 }
2274 else if (mctx->state_log[cur_idx] == 0)
2275 {
2276 mctx->state_log[cur_idx] = next_state;
2277 }
2278 else
2279 {
2280 re_dfastate_t *pstate;
2281 unsigned int context;
2282 re_node_set next_nodes, *log_nodes, *table_nodes = NULL;
2283 /* If (state_log[cur_idx] != 0), it implies that cur_idx is
2284 the destination of a multibyte char/collating element/
2285 back reference. Then the next state is the union set of
2286 these destinations and the results of the transition table. */
2287 pstate = mctx->state_log[cur_idx];
2288 log_nodes = pstate->entrance_nodes;
2289 if (next_state != NULL)
2290 {
2291 table_nodes = next_state->entrance_nodes;
2292 *err = re_node_set_init_union (&next_nodes, table_nodes,
2293 log_nodes);
2294 if (BE (*err != REG_NOERROR, 0))
2295 return NULL;
2296 }
2297 else
2298 next_nodes = *log_nodes;
2299 /* Note: We already add the nodes of the initial state,
2300 then we don't need to add them here. */
2301
2302 context = re_string_context_at (&mctx->input,
2303 re_string_cur_idx (&mctx->input) - 1,
2304 mctx->eflags);
2305 next_state = mctx->state_log[cur_idx]
2306 = re_acquire_state_context (err, dfa, &next_nodes, context);
2307 /* We don't need to check errors here, since the return value of
2308 this function is next_state and ERR is already set. */
2309
2310 if (table_nodes != NULL)
2311 re_node_set_free (&next_nodes);
2312 }
2313
2314 if (BE (dfa->nbackref, 0) && next_state != NULL)
2315 {
2316 /* Check OP_OPEN_SUBEXP in the current state in case that we use them
2317 later. We must check them here, since the back references in the
2318 next state might use them. */
2319 *err = check_subexp_matching_top (mctx, &next_state->nodes,
2320 cur_idx);
2321 if (BE (*err != REG_NOERROR, 0))
2322 return NULL;
2323
2324 /* If the next state has back references. */
2325 if (next_state->has_backref)
2326 {
2327 *err = transit_state_bkref (mctx, &next_state->nodes);
2328 if (BE (*err != REG_NOERROR, 0))
2329 return NULL;
2330 next_state = mctx->state_log[cur_idx];
2331 }
2332 }
2333
2334 return next_state;
2335}
2336
2337/* Skip bytes in the input that correspond to part of a
2338 multi-byte match, then look in the log for a state
2339 from which to restart matching. */
2340re_dfastate_t *
2341find_recover_state (reg_errcode_t *err, re_match_context_t *mctx)
2342{
2343 re_dfastate_t *cur_state;
2344 do
2345 {
2346 int max = mctx->state_log_top;
2347 int cur_str_idx = re_string_cur_idx (&mctx->input);
2348
2349 do
2350 {
2351 if (++cur_str_idx > max)
2352 return NULL;
2353 re_string_skip_bytes (&mctx->input, 1);
2354 }
2355 while (mctx->state_log[cur_str_idx] == NULL);
2356
2357 cur_state = merge_state_with_log (err, mctx, NULL);
2358 }
2359 while (*err == REG_NOERROR && cur_state == NULL);
2360 return cur_state;
2361}
2362
2363/* Helper functions for transit_state. */
2364
2365/* From the node set CUR_NODES, pick up the nodes whose types are
2366 OP_OPEN_SUBEXP and which have corresponding back references in the regular
2367 expression. And register them to use them later for evaluating the
2368 corresponding back references. */
2369
2370static reg_errcode_t
2371check_subexp_matching_top (re_match_context_t *mctx, re_node_set *cur_nodes,
2372 int str_idx)
2373{
2374 const re_dfa_t *const dfa = mctx->dfa;
2375 int node_idx;
2376 reg_errcode_t err;
2377
2378 /* TODO: This isn't efficient.
2379 Because there might be more than one nodes whose types are
2380 OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all
2381 nodes.
2382 E.g. RE: (a){2} */
2383 for (node_idx = 0; node_idx < cur_nodes->nelem; ++node_idx)
2384 {
2385 int node = cur_nodes->elems[node_idx];
2386 if (dfa->nodes[node].type == OP_OPEN_SUBEXP
2387 && dfa->nodes[node].opr.idx < BITSET_WORD_BITS
2388 && (dfa->used_bkref_map
2389 & ((bitset_word_t) 1 << dfa->nodes[node].opr.idx)))
2390 {
2391 err = match_ctx_add_subtop (mctx, node, str_idx);
2392 if (BE (err != REG_NOERROR, 0))
2393 return err;
2394 }
2395 }
2396 return REG_NOERROR;
2397}
2398
2399#if 0
2400/* Return the next state to which the current state STATE will transit by
2401 accepting the current input byte. */
2402
2403static re_dfastate_t *
2404transit_state_sb (reg_errcode_t *err, re_match_context_t *mctx,
2405 re_dfastate_t *state)
2406{
2407 const re_dfa_t *const dfa = mctx->dfa;
2408 re_node_set next_nodes;
2409 re_dfastate_t *next_state;
2410 int node_cnt, cur_str_idx = re_string_cur_idx (&mctx->input);
2411 unsigned int context;
2412
2413 *err = re_node_set_alloc (&next_nodes, state->nodes.nelem + 1);
2414 if (BE (*err != REG_NOERROR, 0))
2415 return NULL;
2416 for (node_cnt = 0; node_cnt < state->nodes.nelem; ++node_cnt)
2417 {
2418 int cur_node = state->nodes.elems[node_cnt];
2419 if (check_node_accept (mctx, dfa->nodes + cur_node, cur_str_idx))
2420 {
2421 *err = re_node_set_merge (&next_nodes,
2422 dfa->eclosures + dfa->nexts[cur_node]);
2423 if (BE (*err != REG_NOERROR, 0))
2424 {
2425 re_node_set_free (&next_nodes);
2426 return NULL;
2427 }
2428 }
2429 }
2430 context = re_string_context_at (&mctx->input, cur_str_idx, mctx->eflags);
2431 next_state = re_acquire_state_context (err, dfa, &next_nodes, context);
2432 /* We don't need to check errors here, since the return value of
2433 this function is next_state and ERR is already set. */
2434
2435 re_node_set_free (&next_nodes);
2436 re_string_skip_bytes (&mctx->input, 1);
2437 return next_state;
2438}
2439#endif
2440
2441#ifdef RE_ENABLE_I18N
2442static reg_errcode_t
2443transit_state_mb (re_match_context_t *mctx, re_dfastate_t *pstate)
2444{
2445 const re_dfa_t *const dfa = mctx->dfa;
2446 reg_errcode_t err;
2447 int i;
2448
2449 for (i = 0; i < pstate->nodes.nelem; ++i)
2450 {
2451 re_node_set dest_nodes, *new_nodes;
2452 int cur_node_idx = pstate->nodes.elems[i];
2453 int naccepted, dest_idx;
2454 unsigned int context;
2455 re_dfastate_t *dest_state;
2456
2457 if (!dfa->nodes[cur_node_idx].accept_mb)
2458 continue;
2459
2460 if (dfa->nodes[cur_node_idx].constraint)
2461 {
2462 context = re_string_context_at (&mctx->input,
2463 re_string_cur_idx (&mctx->input),
2464 mctx->eflags);
2465 if (NOT_SATISFY_NEXT_CONSTRAINT (dfa->nodes[cur_node_idx].constraint,
2466 context))
2467 continue;
2468 }
2469
2470 /* How many bytes the node can accept? */
2471 naccepted = check_node_accept_bytes (dfa, cur_node_idx, &mctx->input,
2472 re_string_cur_idx (&mctx->input));
2473 if (naccepted == 0)
2474 continue;
2475
2476 /* The node can accepts `naccepted' bytes. */
2477 dest_idx = re_string_cur_idx (&mctx->input) + naccepted;
2478 mctx->max_mb_elem_len = ((mctx->max_mb_elem_len < naccepted) ? naccepted
2479 : mctx->max_mb_elem_len);
2480 err = clean_state_log_if_needed (mctx, dest_idx);
2481 if (BE (err != REG_NOERROR, 0))
2482 return err;
2483#ifdef DEBUG
2484 assert (dfa->nexts[cur_node_idx] != -1);
2485#endif
2486 new_nodes = dfa->eclosures + dfa->nexts[cur_node_idx];
2487
2488 dest_state = mctx->state_log[dest_idx];
2489 if (dest_state == NULL)
2490 dest_nodes = *new_nodes;
2491 else
2492 {
2493 err = re_node_set_init_union (&dest_nodes,
2494 dest_state->entrance_nodes, new_nodes);
2495 if (BE (err != REG_NOERROR, 0))
2496 return err;
2497 }
2498 context = re_string_context_at (&mctx->input, dest_idx - 1,
2499 mctx->eflags);
2500 mctx->state_log[dest_idx]
2501 = re_acquire_state_context (&err, dfa, &dest_nodes, context);
2502 if (dest_state != NULL)
2503 re_node_set_free (&dest_nodes);
2504 if (BE (mctx->state_log[dest_idx] == NULL && err != REG_NOERROR, 0))
2505 return err;
2506 }
2507 return REG_NOERROR;
2508}
2509#endif /* RE_ENABLE_I18N */
2510
2511static reg_errcode_t
2512transit_state_bkref (re_match_context_t *mctx, const re_node_set *nodes)
2513{
2514 const re_dfa_t *const dfa = mctx->dfa;
2515 reg_errcode_t err;
2516 int i;
2517 int cur_str_idx = re_string_cur_idx (&mctx->input);
2518
2519 for (i = 0; i < nodes->nelem; ++i)
2520 {
2521 int dest_str_idx, prev_nelem, bkc_idx;
2522 int node_idx = nodes->elems[i];
2523 unsigned int context;
2524 const re_token_t *node = dfa->nodes + node_idx;
2525 re_node_set *new_dest_nodes;
2526
2527 /* Check whether `node' is a backreference or not. */
2528 if (node->type != OP_BACK_REF)
2529 continue;
2530
2531 if (node->constraint)
2532 {
2533 context = re_string_context_at (&mctx->input, cur_str_idx,
2534 mctx->eflags);
2535 if (NOT_SATISFY_NEXT_CONSTRAINT (node->constraint, context))
2536 continue;
2537 }
2538
2539 /* `node' is a backreference.
2540 Check the substring which the substring matched. */
2541 bkc_idx = mctx->nbkref_ents;
2542 err = get_subexp (mctx, node_idx, cur_str_idx);
2543 if (BE (err != REG_NOERROR, 0))
2544 goto free_return;
2545
2546 /* And add the epsilon closures (which is `new_dest_nodes') of
2547 the backreference to appropriate state_log. */
2548#ifdef DEBUG
2549 assert (dfa->nexts[node_idx] != -1);
2550#endif
2551 for (; bkc_idx < mctx->nbkref_ents; ++bkc_idx)
2552 {
2553 int subexp_len;
2554 re_dfastate_t *dest_state;
2555 struct re_backref_cache_entry *bkref_ent;
2556 bkref_ent = mctx->bkref_ents + bkc_idx;
2557 if (bkref_ent->node != node_idx || bkref_ent->str_idx != cur_str_idx)
2558 continue;
2559 subexp_len = bkref_ent->subexp_to - bkref_ent->subexp_from;
2560 new_dest_nodes = (subexp_len == 0
2561 ? dfa->eclosures + dfa->edests[node_idx].elems[0]
2562 : dfa->eclosures + dfa->nexts[node_idx]);
2563 dest_str_idx = (cur_str_idx + bkref_ent->subexp_to
2564 - bkref_ent->subexp_from);
2565 context = re_string_context_at (&mctx->input, dest_str_idx - 1,
2566 mctx->eflags);
2567 dest_state = mctx->state_log[dest_str_idx];
2568 prev_nelem = ((mctx->state_log[cur_str_idx] == NULL) ? 0
2569 : mctx->state_log[cur_str_idx]->nodes.nelem);
2570 /* Add `new_dest_node' to state_log. */
2571 if (dest_state == NULL)
2572 {
2573 mctx->state_log[dest_str_idx]
2574 = re_acquire_state_context (&err, dfa, new_dest_nodes,
2575 context);
2576 if (BE (mctx->state_log[dest_str_idx] == NULL
2577 && err != REG_NOERROR, 0))
2578 goto free_return;
2579 }
2580 else
2581 {
2582 re_node_set dest_nodes;
2583 err = re_node_set_init_union (&dest_nodes,
2584 dest_state->entrance_nodes,
2585 new_dest_nodes);
2586 if (BE (err != REG_NOERROR, 0))
2587 {
2588 re_node_set_free (&dest_nodes);
2589 goto free_return;
2590 }
2591 mctx->state_log[dest_str_idx]
2592 = re_acquire_state_context (&err, dfa, &dest_nodes, context);
2593 re_node_set_free (&dest_nodes);
2594 if (BE (mctx->state_log[dest_str_idx] == NULL
2595 && err != REG_NOERROR, 0))
2596 goto free_return;
2597 }
2598 /* We need to check recursively if the backreference can epsilon
2599 transit. */
2600 if (subexp_len == 0
2601 && mctx->state_log[cur_str_idx]->nodes.nelem > prev_nelem)
2602 {
2603 err = check_subexp_matching_top (mctx, new_dest_nodes,
2604 cur_str_idx);
2605 if (BE (err != REG_NOERROR, 0))
2606 goto free_return;
2607 err = transit_state_bkref (mctx, new_dest_nodes);
2608 if (BE (err != REG_NOERROR, 0))
2609 goto free_return;
2610 }
2611 }
2612 }
2613 err = REG_NOERROR;
2614 free_return:
2615 return err;
2616}
2617
2618/* Enumerate all the candidates which the backreference BKREF_NODE can match
2619 at BKREF_STR_IDX, and register them by match_ctx_add_entry().
2620 Note that we might collect inappropriate candidates here.
2621 However, the cost of checking them strictly here is too high, then we
2622 delay these checking for prune_impossible_nodes(). */
2623
2624static reg_errcode_t
2625__attribute_warn_unused_result__
2626get_subexp (re_match_context_t *mctx, int bkref_node, int bkref_str_idx)
2627{
2628 const re_dfa_t *const dfa = mctx->dfa;
2629 int subexp_num, sub_top_idx;
2630 const char *buf = (const char *) re_string_get_buffer (&mctx->input);
2631 /* Return if we have already checked BKREF_NODE at BKREF_STR_IDX. */
2632 int cache_idx = search_cur_bkref_entry (mctx, bkref_str_idx);
2633 if (cache_idx != -1)
2634 {
2635 const struct re_backref_cache_entry *entry
2636 = mctx->bkref_ents + cache_idx;
2637 do
2638 if (entry->node == bkref_node)
2639 return REG_NOERROR; /* We already checked it. */
2640 while (entry++->more);
2641 }
2642
2643 subexp_num = dfa->nodes[bkref_node].opr.idx;
2644
2645 /* For each sub expression */
2646 for (sub_top_idx = 0; sub_top_idx < mctx->nsub_tops; ++sub_top_idx)
2647 {
2648 reg_errcode_t err;
2649 re_sub_match_top_t *sub_top = mctx->sub_tops[sub_top_idx];
2650 re_sub_match_last_t *sub_last;
2651 int sub_last_idx, sl_str, bkref_str_off;
2652
2653 if (dfa->nodes[sub_top->node].opr.idx != subexp_num)
2654 continue; /* It isn't related. */
2655
2656 sl_str = sub_top->str_idx;
2657 bkref_str_off = bkref_str_idx;
2658 /* At first, check the last node of sub expressions we already
2659 evaluated. */
2660 for (sub_last_idx = 0; sub_last_idx < sub_top->nlasts; ++sub_last_idx)
2661 {
2662 int sl_str_diff;
2663 sub_last = sub_top->lasts[sub_last_idx];
2664 sl_str_diff = sub_last->str_idx - sl_str;
2665 /* The matched string by the sub expression match with the substring
2666 at the back reference? */
2667 if (sl_str_diff > 0)
2668 {
2669 if (BE (bkref_str_off + sl_str_diff > mctx->input.valid_len, 0))
2670 {
2671 /* Not enough chars for a successful match. */
2672 if (bkref_str_off + sl_str_diff > mctx->input.len)
2673 break;
2674
2675 err = clean_state_log_if_needed (mctx,
2676 bkref_str_off
2677 + sl_str_diff);
2678 if (BE (err != REG_NOERROR, 0))
2679 return err;
2680 buf = (const char *) re_string_get_buffer (&mctx->input);
2681 }
2682 if (memcmp (buf + bkref_str_off, buf + sl_str, sl_str_diff) != 0)
2683 /* We don't need to search this sub expression any more. */
2684 break;
2685 }
2686 bkref_str_off += sl_str_diff;
2687 sl_str += sl_str_diff;
2688 err = get_subexp_sub (mctx, sub_top, sub_last, bkref_node,
2689 bkref_str_idx);
2690
2691 /* Reload buf, since the preceding call might have reallocated
2692 the buffer. */
2693 buf = (const char *) re_string_get_buffer (&mctx->input);
2694
2695 if (err == REG_NOMATCH)
2696 continue;
2697 if (BE (err != REG_NOERROR, 0))
2698 return err;
2699 }
2700
2701 if (sub_last_idx < sub_top->nlasts)
2702 continue;
2703 if (sub_last_idx > 0)
2704 ++sl_str;
2705 /* Then, search for the other last nodes of the sub expression. */
2706 for (; sl_str <= bkref_str_idx; ++sl_str)
2707 {
2708 int cls_node, sl_str_off;
2709 const re_node_set *nodes;
2710 sl_str_off = sl_str - sub_top->str_idx;
2711 /* The matched string by the sub expression match with the substring
2712 at the back reference? */
2713 if (sl_str_off > 0)
2714 {
2715 if (BE (bkref_str_off >= mctx->input.valid_len, 0))
2716 {
2717 /* If we are at the end of the input, we cannot match. */
2718 if (bkref_str_off >= mctx->input.len)
2719 break;
2720
2721 err = extend_buffers (mctx, bkref_str_off + 1);
2722 if (BE (err != REG_NOERROR, 0))
2723 return err;
2724
2725 buf = (const char *) re_string_get_buffer (&mctx->input);
2726 }
2727 if (buf [bkref_str_off++] != buf[sl_str - 1])
2728 break; /* We don't need to search this sub expression
2729 any more. */
2730 }
2731 if (mctx->state_log[sl_str] == NULL)
2732 continue;
2733 /* Does this state have a ')' of the sub expression? */
2734 nodes = &mctx->state_log[sl_str]->nodes;
2735 cls_node = find_subexp_node (dfa, nodes, subexp_num,
2736 OP_CLOSE_SUBEXP);
2737 if (cls_node == -1)
2738 continue; /* No. */
2739 if (sub_top->path == NULL)
2740 {
2741 sub_top->path = calloc (sizeof (state_array_t),
2742 sl_str - sub_top->str_idx + 1);
2743 if (sub_top->path == NULL)
2744 return REG_ESPACE;
2745 }
2746 /* Can the OP_OPEN_SUBEXP node arrive the OP_CLOSE_SUBEXP node
2747 in the current context? */
2748 err = check_arrival (mctx, sub_top->path, sub_top->node,
2749 sub_top->str_idx, cls_node, sl_str,
2750 OP_CLOSE_SUBEXP);
2751 if (err == REG_NOMATCH)
2752 continue;
2753 if (BE (err != REG_NOERROR, 0))
2754 return err;
2755 sub_last = match_ctx_add_sublast (sub_top, cls_node, sl_str);
2756 if (BE (sub_last == NULL, 0))
2757 return REG_ESPACE;
2758 err = get_subexp_sub (mctx, sub_top, sub_last, bkref_node,
2759 bkref_str_idx);
2760 if (err == REG_NOMATCH)
2761 continue;
2762 }
2763 }
2764 return REG_NOERROR;
2765}
2766
2767/* Helper functions for get_subexp(). */
2768
2769/* Check SUB_LAST can arrive to the back reference BKREF_NODE at BKREF_STR.
2770 If it can arrive, register the sub expression expressed with SUB_TOP
2771 and SUB_LAST. */
2772
2773static reg_errcode_t
2774get_subexp_sub (re_match_context_t *mctx, const re_sub_match_top_t *sub_top,
2775 re_sub_match_last_t *sub_last, int bkref_node, int bkref_str)
2776{
2777 reg_errcode_t err;
2778 int to_idx;
2779 /* Can the subexpression arrive the back reference? */
2780 err = check_arrival (mctx, &sub_last->path, sub_last->node,
2781 sub_last->str_idx, bkref_node, bkref_str,
2782 OP_OPEN_SUBEXP);
2783 if (err != REG_NOERROR)
2784 return err;
2785 err = match_ctx_add_entry (mctx, bkref_node, bkref_str, sub_top->str_idx,
2786 sub_last->str_idx);
2787 if (BE (err != REG_NOERROR, 0))
2788 return err;
2789 to_idx = bkref_str + sub_last->str_idx - sub_top->str_idx;
2790 return clean_state_log_if_needed (mctx, to_idx);
2791}
2792
2793/* Find the first node which is '(' or ')' and whose index is SUBEXP_IDX.
2794 Search '(' if FL_OPEN, or search ')' otherwise.
2795 TODO: This function isn't efficient...
2796 Because there might be more than one nodes whose types are
2797 OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all
2798 nodes.
2799 E.g. RE: (a){2} */
2800
2801static int
2802find_subexp_node (const re_dfa_t *dfa, const re_node_set *nodes,
2803 int subexp_idx, int type)
2804{
2805 int cls_idx;
2806 for (cls_idx = 0; cls_idx < nodes->nelem; ++cls_idx)
2807 {
2808 int cls_node = nodes->elems[cls_idx];
2809 const re_token_t *node = dfa->nodes + cls_node;
2810 if (node->type == type
2811 && node->opr.idx == subexp_idx)
2812 return cls_node;
2813 }
2814 return -1;
2815}
2816
2817/* Check whether the node TOP_NODE at TOP_STR can arrive to the node
2818 LAST_NODE at LAST_STR. We record the path onto PATH since it will be
2819 heavily reused.
2820 Return REG_NOERROR if it can arrive, or REG_NOMATCH otherwise. */
2821
2822static reg_errcode_t
2823__attribute_warn_unused_result__
2824check_arrival (re_match_context_t *mctx, state_array_t *path, int top_node,
2825 int top_str, int last_node, int last_str, int type)
2826{
2827 const re_dfa_t *const dfa = mctx->dfa;
2828 reg_errcode_t err = REG_NOERROR;
2829 int subexp_num, backup_cur_idx, str_idx, null_cnt;
2830 re_dfastate_t *cur_state = NULL;
2831 re_node_set *cur_nodes, next_nodes;
2832 re_dfastate_t **backup_state_log;
2833 unsigned int context;
2834
2835 subexp_num = dfa->nodes[top_node].opr.idx;
2836 /* Extend the buffer if we need. */
2837 if (BE (path->alloc < last_str + mctx->max_mb_elem_len + 1, 0))
2838 {
2839 re_dfastate_t **new_array;
2840 int old_alloc = path->alloc;
2841 path->alloc += last_str + mctx->max_mb_elem_len + 1;
2842 new_array = re_realloc (path->array, re_dfastate_t *, path->alloc);
2843 if (BE (new_array == NULL, 0))
2844 {
2845 path->alloc = old_alloc;
2846 return REG_ESPACE;
2847 }
2848 path->array = new_array;
2849 memset (new_array + old_alloc, '\0',
2850 sizeof (re_dfastate_t *) * (path->alloc - old_alloc));
2851 }
2852
2853 str_idx = path->next_idx ?: top_str;
2854
2855 /* Temporary modify MCTX. */
2856 backup_state_log = mctx->state_log;
2857 backup_cur_idx = mctx->input.cur_idx;
2858 mctx->state_log = path->array;
2859 mctx->input.cur_idx = str_idx;
2860
2861 /* Setup initial node set. */
2862 context = re_string_context_at (&mctx->input, str_idx - 1, mctx->eflags);
2863 if (str_idx == top_str)
2864 {
2865 err = re_node_set_init_1 (&next_nodes, top_node);
2866 if (BE (err != REG_NOERROR, 0))
2867 return err;
2868 err = check_arrival_expand_ecl (dfa, &next_nodes, subexp_num, type);
2869 if (BE (err != REG_NOERROR, 0))
2870 {
2871 re_node_set_free (&next_nodes);
2872 return err;
2873 }
2874 }
2875 else
2876 {
2877 cur_state = mctx->state_log[str_idx];
2878 if (cur_state && cur_state->has_backref)
2879 {
2880 err = re_node_set_init_copy (&next_nodes, &cur_state->nodes);
2881 if (BE (err != REG_NOERROR, 0))
2882 return err;
2883 }
2884 else
2885 re_node_set_init_empty (&next_nodes);
2886 }
2887 if (str_idx == top_str || (cur_state && cur_state->has_backref))
2888 {
2889 if (next_nodes.nelem)
2890 {
2891 err = expand_bkref_cache (mctx, &next_nodes, str_idx,
2892 subexp_num, type);
2893 if (BE (err != REG_NOERROR, 0))
2894 {
2895 re_node_set_free (&next_nodes);
2896 return err;
2897 }
2898 }
2899 cur_state = re_acquire_state_context (&err, dfa, &next_nodes, context);
2900 if (BE (cur_state == NULL && err != REG_NOERROR, 0))
2901 {
2902 re_node_set_free (&next_nodes);
2903 return err;
2904 }
2905 mctx->state_log[str_idx] = cur_state;
2906 }
2907
2908 for (null_cnt = 0; str_idx < last_str && null_cnt <= mctx->max_mb_elem_len;)
2909 {
2910 re_node_set_empty (&next_nodes);
2911 if (mctx->state_log[str_idx + 1])
2912 {
2913 err = re_node_set_merge (&next_nodes,
2914 &mctx->state_log[str_idx + 1]->nodes);
2915 if (BE (err != REG_NOERROR, 0))
2916 {
2917 re_node_set_free (&next_nodes);
2918 return err;
2919 }
2920 }
2921 if (cur_state)
2922 {
2923 err = check_arrival_add_next_nodes (mctx, str_idx,
2924 &cur_state->non_eps_nodes,
2925 &next_nodes);
2926 if (BE (err != REG_NOERROR, 0))
2927 {
2928 re_node_set_free (&next_nodes);
2929 return err;
2930 }
2931 }
2932 ++str_idx;
2933 if (next_nodes.nelem)
2934 {
2935 err = check_arrival_expand_ecl (dfa, &next_nodes, subexp_num, type);
2936 if (BE (err != REG_NOERROR, 0))
2937 {
2938 re_node_set_free (&next_nodes);
2939 return err;
2940 }
2941 err = expand_bkref_cache (mctx, &next_nodes, str_idx,
2942 subexp_num, type);
2943 if (BE (err != REG_NOERROR, 0))
2944 {
2945 re_node_set_free (&next_nodes);
2946 return err;
2947 }
2948 }
2949 context = re_string_context_at (&mctx->input, str_idx - 1, mctx->eflags);
2950 cur_state = re_acquire_state_context (&err, dfa, &next_nodes, context);
2951 if (BE (cur_state == NULL && err != REG_NOERROR, 0))
2952 {
2953 re_node_set_free (&next_nodes);
2954 return err;
2955 }
2956 mctx->state_log[str_idx] = cur_state;
2957 null_cnt = cur_state == NULL ? null_cnt + 1 : 0;
2958 }
2959 re_node_set_free (&next_nodes);
2960 cur_nodes = (mctx->state_log[last_str] == NULL ? NULL
2961 : &mctx->state_log[last_str]->nodes);
2962 path->next_idx = str_idx;
2963
2964 /* Fix MCTX. */
2965 mctx->state_log = backup_state_log;
2966 mctx->input.cur_idx = backup_cur_idx;
2967
2968 /* Then check the current node set has the node LAST_NODE. */
2969 if (cur_nodes != NULL && re_node_set_contains (cur_nodes, last_node))
2970 return REG_NOERROR;
2971
2972 return REG_NOMATCH;
2973}
2974
2975/* Helper functions for check_arrival. */
2976
2977/* Calculate the destination nodes of CUR_NODES at STR_IDX, and append them
2978 to NEXT_NODES.
2979 TODO: This function is similar to the functions transit_state*(),
2980 however this function has many additional works.
2981 Can't we unify them? */
2982
2983static reg_errcode_t
2984__attribute_warn_unused_result__
2985check_arrival_add_next_nodes (re_match_context_t *mctx, int str_idx,
2986 re_node_set *cur_nodes, re_node_set *next_nodes)
2987{
2988 const re_dfa_t *const dfa = mctx->dfa;
2989 int result;
2990 int cur_idx;
2991#ifdef RE_ENABLE_I18N
2992 reg_errcode_t err = REG_NOERROR;
2993#endif
2994 re_node_set union_set;
2995 re_node_set_init_empty (&union_set);
2996 for (cur_idx = 0; cur_idx < cur_nodes->nelem; ++cur_idx)
2997 {
2998 int naccepted = 0;
2999 int cur_node = cur_nodes->elems[cur_idx];
3000#ifdef DEBUG
3001 re_token_type_t type = dfa->nodes[cur_node].type;
3002 assert (!IS_EPSILON_NODE (type));
3003#endif
3004#ifdef RE_ENABLE_I18N
3005 /* If the node may accept `multi byte'. */
3006 if (dfa->nodes[cur_node].accept_mb)
3007 {
3008 naccepted = check_node_accept_bytes (dfa, cur_node, &mctx->input,
3009 str_idx);
3010 if (naccepted > 1)
3011 {
3012 re_dfastate_t *dest_state;
3013 int next_node = dfa->nexts[cur_node];
3014 int next_idx = str_idx + naccepted;
3015 dest_state = mctx->state_log[next_idx];
3016 re_node_set_empty (&union_set);
3017 if (dest_state)
3018 {
3019 err = re_node_set_merge (&union_set, &dest_state->nodes);
3020 if (BE (err != REG_NOERROR, 0))
3021 {
3022 re_node_set_free (&union_set);
3023 return err;
3024 }
3025 }
3026 result = re_node_set_insert (&union_set, next_node);
3027 if (BE (result < 0, 0))
3028 {
3029 re_node_set_free (&union_set);
3030 return REG_ESPACE;
3031 }
3032 mctx->state_log[next_idx] = re_acquire_state (&err, dfa,
3033 &union_set);
3034 if (BE (mctx->state_log[next_idx] == NULL
3035 && err != REG_NOERROR, 0))
3036 {
3037 re_node_set_free (&union_set);
3038 return err;
3039 }
3040 }
3041 }
3042#endif /* RE_ENABLE_I18N */
3043 if (naccepted
3044 || check_node_accept (mctx, dfa->nodes + cur_node, str_idx))
3045 {
3046 result = re_node_set_insert (next_nodes, dfa->nexts[cur_node]);
3047 if (BE (result < 0, 0))
3048 {
3049 re_node_set_free (&union_set);
3050 return REG_ESPACE;
3051 }
3052 }
3053 }
3054 re_node_set_free (&union_set);
3055 return REG_NOERROR;
3056}
3057
3058/* For all the nodes in CUR_NODES, add the epsilon closures of them to
3059 CUR_NODES, however exclude the nodes which are:
3060 - inside the sub expression whose number is EX_SUBEXP, if FL_OPEN.
3061 - out of the sub expression whose number is EX_SUBEXP, if !FL_OPEN.
3062*/
3063
3064static reg_errcode_t
3065check_arrival_expand_ecl (const re_dfa_t *dfa, re_node_set *cur_nodes,
3066 int ex_subexp, int type)
3067{
3068 reg_errcode_t err;
3069 int idx, outside_node;
3070 re_node_set new_nodes;
3071#ifdef DEBUG
3072 assert (cur_nodes->nelem);
3073#endif
3074 err = re_node_set_alloc (&new_nodes, cur_nodes->nelem);
3075 if (BE (err != REG_NOERROR, 0))
3076 return err;
3077 /* Create a new node set NEW_NODES with the nodes which are epsilon
3078 closures of the node in CUR_NODES. */
3079
3080 for (idx = 0; idx < cur_nodes->nelem; ++idx)
3081 {
3082 int cur_node = cur_nodes->elems[idx];
3083 const re_node_set *eclosure = dfa->eclosures + cur_node;
3084 outside_node = find_subexp_node (dfa, eclosure, ex_subexp, type);
3085 if (outside_node == -1)
3086 {
3087 /* There are no problematic nodes, just merge them. */
3088 err = re_node_set_merge (&new_nodes, eclosure);
3089 if (BE (err != REG_NOERROR, 0))
3090 {
3091 re_node_set_free (&new_nodes);
3092 return err;
3093 }
3094 }
3095 else
3096 {
3097 /* There are problematic nodes, re-calculate incrementally. */
3098 err = check_arrival_expand_ecl_sub (dfa, &new_nodes, cur_node,
3099 ex_subexp, type);
3100 if (BE (err != REG_NOERROR, 0))
3101 {
3102 re_node_set_free (&new_nodes);
3103 return err;
3104 }
3105 }
3106 }
3107 re_node_set_free (cur_nodes);
3108 *cur_nodes = new_nodes;
3109 return REG_NOERROR;
3110}
3111
3112/* Helper function for check_arrival_expand_ecl.
3113 Check incrementally the epsilon closure of TARGET, and if it isn't
3114 problematic append it to DST_NODES. */
3115
3116static reg_errcode_t
3117__attribute_warn_unused_result__
3118check_arrival_expand_ecl_sub (const re_dfa_t *dfa, re_node_set *dst_nodes,
3119 int target, int ex_subexp, int type)
3120{
3121 int cur_node;
3122 for (cur_node = target; !re_node_set_contains (dst_nodes, cur_node);)
3123 {
3124 int err;
3125
3126 if (dfa->nodes[cur_node].type == type
3127 && dfa->nodes[cur_node].opr.idx == ex_subexp)
3128 {
3129 if (type == OP_CLOSE_SUBEXP)
3130 {
3131 err = re_node_set_insert (dst_nodes, cur_node);
3132 if (BE (err == -1, 0))
3133 return REG_ESPACE;
3134 }
3135 break;
3136 }
3137 err = re_node_set_insert (dst_nodes, cur_node);
3138 if (BE (err == -1, 0))
3139 return REG_ESPACE;
3140 if (dfa->edests[cur_node].nelem == 0)
3141 break;
3142 if (dfa->edests[cur_node].nelem == 2)
3143 {
3144 err = check_arrival_expand_ecl_sub (dfa, dst_nodes,
3145 dfa->edests[cur_node].elems[1],
3146 ex_subexp, type);
3147 if (BE (err != REG_NOERROR, 0))
3148 return err;
3149 }
3150 cur_node = dfa->edests[cur_node].elems[0];
3151 }
3152 return REG_NOERROR;
3153}
3154
3155
3156/* For all the back references in the current state, calculate the
3157 destination of the back references by the appropriate entry
3158 in MCTX->BKREF_ENTS. */
3159
3160static reg_errcode_t
3161__attribute_warn_unused_result__
3162expand_bkref_cache (re_match_context_t *mctx, re_node_set *cur_nodes,
3163 int cur_str, int subexp_num, int type)
3164{
3165 const re_dfa_t *const dfa = mctx->dfa;
3166 reg_errcode_t err;
3167 int cache_idx_start = search_cur_bkref_entry (mctx, cur_str);
3168 struct re_backref_cache_entry *ent;
3169
3170 if (cache_idx_start == -1)
3171 return REG_NOERROR;
3172
3173 restart:
3174 ent = mctx->bkref_ents + cache_idx_start;
3175 do
3176 {
3177 int to_idx, next_node;
3178
3179 /* Is this entry ENT is appropriate? */
3180 if (!re_node_set_contains (cur_nodes, ent->node))
3181 continue; /* No. */
3182
3183 to_idx = cur_str + ent->subexp_to - ent->subexp_from;
3184 /* Calculate the destination of the back reference, and append it
3185 to MCTX->STATE_LOG. */
3186 if (to_idx == cur_str)
3187 {
3188 /* The backreference did epsilon transit, we must re-check all the
3189 node in the current state. */
3190 re_node_set new_dests;
3191 reg_errcode_t err2, err3;
3192 next_node = dfa->edests[ent->node].elems[0];
3193 if (re_node_set_contains (cur_nodes, next_node))
3194 continue;
3195 err = re_node_set_init_1 (&new_dests, next_node);
3196 err2 = check_arrival_expand_ecl (dfa, &new_dests, subexp_num, type);
3197 err3 = re_node_set_merge (cur_nodes, &new_dests);
3198 re_node_set_free (&new_dests);
3199 if (BE (err != REG_NOERROR || err2 != REG_NOERROR
3200 || err3 != REG_NOERROR, 0))
3201 {
3202 err = (err != REG_NOERROR ? err
3203 : (err2 != REG_NOERROR ? err2 : err3));
3204 return err;
3205 }
3206 /* TODO: It is still inefficient... */
3207 goto restart;
3208 }
3209 else
3210 {
3211 re_node_set union_set;
3212 next_node = dfa->nexts[ent->node];
3213 if (mctx->state_log[to_idx])
3214 {
3215 int ret;
3216 if (re_node_set_contains (&mctx->state_log[to_idx]->nodes,
3217 next_node))
3218 continue;
3219 err = re_node_set_init_copy (&union_set,
3220 &mctx->state_log[to_idx]->nodes);
3221 ret = re_node_set_insert (&union_set, next_node);
3222 if (BE (err != REG_NOERROR || ret < 0, 0))
3223 {
3224 re_node_set_free (&union_set);
3225 err = err != REG_NOERROR ? err : REG_ESPACE;
3226 return err;
3227 }
3228 }
3229 else
3230 {
3231 err = re_node_set_init_1 (&union_set, next_node);
3232 if (BE (err != REG_NOERROR, 0))
3233 return err;
3234 }
3235 mctx->state_log[to_idx] = re_acquire_state (&err, dfa, &union_set);
3236 re_node_set_free (&union_set);
3237 if (BE (mctx->state_log[to_idx] == NULL
3238 && err != REG_NOERROR, 0))
3239 return err;
3240 }
3241 }
3242 while (ent++->more);
3243 return REG_NOERROR;
3244}
3245
3246/* Build transition table for the state.
3247 Return 1 if succeeded, otherwise return NULL. */
3248
3249static int
3250build_trtable (const re_dfa_t *dfa, re_dfastate_t *state)
3251{
3252 reg_errcode_t err;
3253 int i, j, ch, need_word_trtable = 0;
3254 bitset_word_t elem, mask;
3255 bool dests_node_malloced = false;
3256 bool dest_states_malloced = false;
3257 int ndests; /* Number of the destination states from `state'. */
3258 re_dfastate_t **trtable;
3259 re_dfastate_t **dest_states = NULL, **dest_states_word, **dest_states_nl;
3260 re_node_set follows, *dests_node;
3261 bitset_t *dests_ch;
3262 bitset_t acceptable;
3263
3264 struct dests_alloc
3265 {
3266 re_node_set dests_node[SBC_MAX];
3267 bitset_t dests_ch[SBC_MAX];
3268 } *dests_alloc;
3269
3270 /* We build DFA states which corresponds to the destination nodes
3271 from `state'. `dests_node[i]' represents the nodes which i-th
3272 destination state contains, and `dests_ch[i]' represents the
3273 characters which i-th destination state accepts. */
3274 if (__libc_use_alloca (sizeof (struct dests_alloc)))
3275 dests_alloc = (struct dests_alloc *) alloca (sizeof (struct dests_alloc));
3276 else
3277 {
3278 dests_alloc = re_malloc (struct dests_alloc, 1);
3279 if (BE (dests_alloc == NULL, 0))
3280 return 0;
3281 dests_node_malloced = true;
3282 }
3283 dests_node = dests_alloc->dests_node;
3284 dests_ch = dests_alloc->dests_ch;
3285
3286 /* Initialize transiton table. */
3287 state->word_trtable = state->trtable = NULL;
3288
3289 /* At first, group all nodes belonging to `state' into several
3290 destinations. */
3291 ndests = group_nodes_into_DFAstates (dfa, state, dests_node, dests_ch);
3292 if (BE (ndests <= 0, 0))
3293 {
3294 if (dests_node_malloced)
3295 free (dests_alloc);
3296 /* Return 0 in case of an error, 1 otherwise. */
3297 if (ndests == 0)
3298 {
3299 state->trtable = (re_dfastate_t **)
3300 calloc (sizeof (re_dfastate_t *), SBC_MAX);
3301 if (BE (state->trtable == NULL, 0))
3302 return 0;
3303 return 1;
3304 }
3305 return 0;
3306 }
3307
3308 err = re_node_set_alloc (&follows, ndests + 1);
3309 if (BE (err != REG_NOERROR, 0))
3310 goto out_free;
3311
3312 /* Avoid arithmetic overflow in size calculation. */
3313 if (BE ((((SIZE_MAX - (sizeof (re_node_set) + sizeof (bitset_t)) * SBC_MAX)
3314 / (3 * sizeof (re_dfastate_t *)))
3315 < ndests),
3316 0))
3317 goto out_free;
3318
3319 if (__libc_use_alloca ((sizeof (re_node_set) + sizeof (bitset_t)) * SBC_MAX
3320 + ndests * 3 * sizeof (re_dfastate_t *)))
3321 dest_states = (re_dfastate_t **)
3322 alloca (ndests * 3 * sizeof (re_dfastate_t *));
3323 else
3324 {
3325 dest_states = (re_dfastate_t **)
3326 malloc (ndests * 3 * sizeof (re_dfastate_t *));
3327 if (BE (dest_states == NULL, 0))
3328 {
3329out_free:
3330 if (dest_states_malloced)
3331 free (dest_states);
3332 re_node_set_free (&follows);
3333 for (i = 0; i < ndests; ++i)
3334 re_node_set_free (dests_node + i);
3335 if (dests_node_malloced)
3336 free (dests_alloc);
3337 return 0;
3338 }
3339 dest_states_malloced = true;
3340 }
3341 dest_states_word = dest_states + ndests;
3342 dest_states_nl = dest_states_word + ndests;
3343 bitset_empty (acceptable);
3344
3345 /* Then build the states for all destinations. */
3346 for (i = 0; i < ndests; ++i)
3347 {
3348 int next_node;
3349 re_node_set_empty (&follows);
3350 /* Merge the follows of this destination states. */
3351 for (j = 0; j < dests_node[i].nelem; ++j)
3352 {
3353 next_node = dfa->nexts[dests_node[i].elems[j]];
3354 if (next_node != -1)
3355 {
3356 err = re_node_set_merge (&follows, dfa->eclosures + next_node);
3357 if (BE (err != REG_NOERROR, 0))
3358 goto out_free;
3359 }
3360 }
3361 dest_states[i] = re_acquire_state_context (&err, dfa, &follows, 0);
3362 if (BE (dest_states[i] == NULL && err != REG_NOERROR, 0))
3363 goto out_free;
3364 /* If the new state has context constraint,
3365 build appropriate states for these contexts. */
3366 if (dest_states[i]->has_constraint)
3367 {
3368 dest_states_word[i] = re_acquire_state_context (&err, dfa, &follows,
3369 CONTEXT_WORD);
3370 if (BE (dest_states_word[i] == NULL && err != REG_NOERROR, 0))
3371 goto out_free;
3372
3373 if (dest_states[i] != dest_states_word[i] && dfa->mb_cur_max > 1)
3374 need_word_trtable = 1;
3375
3376 dest_states_nl[i] = re_acquire_state_context (&err, dfa, &follows,
3377 CONTEXT_NEWLINE);
3378 if (BE (dest_states_nl[i] == NULL && err != REG_NOERROR, 0))
3379 goto out_free;
3380 }
3381 else
3382 {
3383 dest_states_word[i] = dest_states[i];
3384 dest_states_nl[i] = dest_states[i];
3385 }
3386 bitset_merge (acceptable, dests_ch[i]);
3387 }
3388
3389 if (!BE (need_word_trtable, 0))
3390 {
3391 /* We don't care about whether the following character is a word
3392 character, or we are in a single-byte character set so we can
3393 discern by looking at the character code: allocate a
3394 256-entry transition table. */
3395 trtable = state->trtable =
3396 (re_dfastate_t **) calloc (sizeof (re_dfastate_t *), SBC_MAX);
3397 if (BE (trtable == NULL, 0))
3398 goto out_free;
3399
3400 /* For all characters ch...: */
3401 for (i = 0; i < BITSET_WORDS; ++i)
3402 for (ch = i * BITSET_WORD_BITS, elem = acceptable[i], mask = 1;
3403 elem;
3404 mask <<= 1, elem >>= 1, ++ch)
3405 if (BE (elem & 1, 0))
3406 {
3407 /* There must be exactly one destination which accepts
3408 character ch. See group_nodes_into_DFAstates. */
3409 for (j = 0; (dests_ch[j][i] & mask) == 0; ++j)
3410 ;
3411
3412 /* j-th destination accepts the word character ch. */
3413 if (dfa->word_char[i] & mask)
3414 trtable[ch] = dest_states_word[j];
3415 else
3416 trtable[ch] = dest_states[j];
3417 }
3418 }
3419 else
3420 {
3421 /* We care about whether the following character is a word
3422 character, and we are in a multi-byte character set: discern
3423 by looking at the character code: build two 256-entry
3424 transition tables, one starting at trtable[0] and one
3425 starting at trtable[SBC_MAX]. */
3426 trtable = state->word_trtable =
3427 (re_dfastate_t **) calloc (sizeof (re_dfastate_t *), 2 * SBC_MAX);
3428 if (BE (trtable == NULL, 0))
3429 goto out_free;
3430
3431 /* For all characters ch...: */
3432 for (i = 0; i < BITSET_WORDS; ++i)
3433 for (ch = i * BITSET_WORD_BITS, elem = acceptable[i], mask = 1;
3434 elem;
3435 mask <<= 1, elem >>= 1, ++ch)
3436 if (BE (elem & 1, 0))
3437 {
3438 /* There must be exactly one destination which accepts
3439 character ch. See group_nodes_into_DFAstates. */
3440 for (j = 0; (dests_ch[j][i] & mask) == 0; ++j)
3441 ;
3442
3443 /* j-th destination accepts the word character ch. */
3444 trtable[ch] = dest_states[j];
3445 trtable[ch + SBC_MAX] = dest_states_word[j];
3446 }
3447 }
3448
3449 /* new line */
3450 if (bitset_contain (acceptable, NEWLINE_CHAR))
3451 {
3452 /* The current state accepts newline character. */
3453 for (j = 0; j < ndests; ++j)
3454 if (bitset_contain (dests_ch[j], NEWLINE_CHAR))
3455 {
3456 /* k-th destination accepts newline character. */
3457 trtable[NEWLINE_CHAR] = dest_states_nl[j];
3458 if (need_word_trtable)
3459 trtable[NEWLINE_CHAR + SBC_MAX] = dest_states_nl[j];
3460 /* There must be only one destination which accepts
3461 newline. See group_nodes_into_DFAstates. */
3462 break;
3463 }
3464 }
3465
3466 if (dest_states_malloced)
3467 free (dest_states);
3468
3469 re_node_set_free (&follows);
3470 for (i = 0; i < ndests; ++i)
3471 re_node_set_free (dests_node + i);
3472
3473 if (dests_node_malloced)
3474 free (dests_alloc);
3475
3476 return 1;
3477}
3478
3479/* Group all nodes belonging to STATE into several destinations.
3480 Then for all destinations, set the nodes belonging to the destination
3481 to DESTS_NODE[i] and set the characters accepted by the destination
3482 to DEST_CH[i]. This function return the number of destinations. */
3483
3484static int
3485group_nodes_into_DFAstates (const re_dfa_t *dfa, const re_dfastate_t *state,
3486 re_node_set *dests_node, bitset_t *dests_ch)
3487{
3488 reg_errcode_t err;
3489 int result;
3490 int i, j, k;
3491 int ndests; /* Number of the destinations from `state'. */
3492 bitset_t accepts; /* Characters a node can accept. */
3493 const re_node_set *cur_nodes = &state->nodes;
3494 bitset_empty (accepts);
3495 ndests = 0;
3496
3497 /* For all the nodes belonging to `state', */
3498 for (i = 0; i < cur_nodes->nelem; ++i)
3499 {
3500 re_token_t *node = &dfa->nodes[cur_nodes->elems[i]];
3501 re_token_type_t type = node->type;
3502 unsigned int constraint = node->constraint;
3503
3504 /* Enumerate all single byte character this node can accept. */
3505 if (type == CHARACTER)
3506 bitset_set (accepts, node->opr.c);
3507 else if (type == SIMPLE_BRACKET)
3508 {
3509 bitset_merge (accepts, node->opr.sbcset);
3510 }
3511 else if (type == OP_PERIOD)
3512 {
3513#ifdef RE_ENABLE_I18N
3514 if (dfa->mb_cur_max > 1)
3515 bitset_merge (accepts, dfa->sb_char);
3516 else
3517#endif
3518 bitset_set_all (accepts);
3519 if (!(dfa->syntax & RE_DOT_NEWLINE))
3520 bitset_clear (accepts, '\n');
3521 if (dfa->syntax & RE_DOT_NOT_NULL)
3522 bitset_clear (accepts, '\0');
3523 }
3524#ifdef RE_ENABLE_I18N
3525 else if (type == OP_UTF8_PERIOD)
3526 {
3527 memset (accepts, '\xff', sizeof (bitset_t) / 2);
3528 if (!(dfa->syntax & RE_DOT_NEWLINE))
3529 bitset_clear (accepts, '\n');
3530 if (dfa->syntax & RE_DOT_NOT_NULL)
3531 bitset_clear (accepts, '\0');
3532 }
3533#endif
3534 else
3535 continue;
3536
3537 /* Check the `accepts' and sift the characters which are not
3538 match it the context. */
3539 if (constraint)
3540 {
3541 if (constraint & NEXT_NEWLINE_CONSTRAINT)
3542 {
3543 bool accepts_newline = bitset_contain (accepts, NEWLINE_CHAR);
3544 bitset_empty (accepts);
3545 if (accepts_newline)
3546 bitset_set (accepts, NEWLINE_CHAR);
3547 else
3548 continue;
3549 }
3550 if (constraint & NEXT_ENDBUF_CONSTRAINT)
3551 {
3552 bitset_empty (accepts);
3553 continue;
3554 }
3555
3556 if (constraint & NEXT_WORD_CONSTRAINT)
3557 {
3558 bitset_word_t any_set = 0;
3559 if (type == CHARACTER && !node->word_char)
3560 {
3561 bitset_empty (accepts);
3562 continue;
3563 }
3564#ifdef RE_ENABLE_I18N
3565 if (dfa->mb_cur_max > 1)
3566 for (j = 0; j < BITSET_WORDS; ++j)
3567 any_set |= (accepts[j] &= (dfa->word_char[j] | ~dfa->sb_char[j]));
3568 else
3569#endif
3570 for (j = 0; j < BITSET_WORDS; ++j)
3571 any_set |= (accepts[j] &= dfa->word_char[j]);
3572 if (!any_set)
3573 continue;
3574 }
3575 if (constraint & NEXT_NOTWORD_CONSTRAINT)
3576 {
3577 bitset_word_t any_set = 0;
3578 if (type == CHARACTER && node->word_char)
3579 {
3580 bitset_empty (accepts);
3581 continue;
3582 }
3583#ifdef RE_ENABLE_I18N
3584 if (dfa->mb_cur_max > 1)
3585 for (j = 0; j < BITSET_WORDS; ++j)
3586 any_set |= (accepts[j] &= ~(dfa->word_char[j] & dfa->sb_char[j]));
3587 else
3588#endif
3589 for (j = 0; j < BITSET_WORDS; ++j)
3590 any_set |= (accepts[j] &= ~dfa->word_char[j]);
3591 if (!any_set)
3592 continue;
3593 }
3594 }
3595
3596 /* Then divide `accepts' into DFA states, or create a new
3597 state. Above, we make sure that accepts is not empty. */
3598 for (j = 0; j < ndests; ++j)
3599 {
3600 bitset_t intersec; /* Intersection sets, see below. */
3601 bitset_t remains;
3602 /* Flags, see below. */
3603 bitset_word_t has_intersec, not_subset, not_consumed;
3604
3605 /* Optimization, skip if this state doesn't accept the character. */
3606 if (type == CHARACTER && !bitset_contain (dests_ch[j], node->opr.c))
3607 continue;
3608
3609 /* Enumerate the intersection set of this state and `accepts'. */
3610 has_intersec = 0;
3611 for (k = 0; k < BITSET_WORDS; ++k)
3612 has_intersec |= intersec[k] = accepts[k] & dests_ch[j][k];
3613 /* And skip if the intersection set is empty. */
3614 if (!has_intersec)
3615 continue;
3616
3617 /* Then check if this state is a subset of `accepts'. */
3618 not_subset = not_consumed = 0;
3619 for (k = 0; k < BITSET_WORDS; ++k)
3620 {
3621 not_subset |= remains[k] = ~accepts[k] & dests_ch[j][k];
3622 not_consumed |= accepts[k] = accepts[k] & ~dests_ch[j][k];
3623 }
3624
3625 /* If this state isn't a subset of `accepts', create a
3626 new group state, which has the `remains'. */
3627 if (not_subset)
3628 {
3629 bitset_copy (dests_ch[ndests], remains);
3630 bitset_copy (dests_ch[j], intersec);
3631 err = re_node_set_init_copy (dests_node + ndests, &dests_node[j]);
3632 if (BE (err != REG_NOERROR, 0))
3633 goto error_return;
3634 ++ndests;
3635 }
3636
3637 /* Put the position in the current group. */
3638 result = re_node_set_insert (&dests_node[j], cur_nodes->elems[i]);
3639 if (BE (result < 0, 0))
3640 goto error_return;
3641
3642 /* If all characters are consumed, go to next node. */
3643 if (!not_consumed)
3644 break;
3645 }
3646 /* Some characters remain, create a new group. */
3647 if (j == ndests)
3648 {
3649 bitset_copy (dests_ch[ndests], accepts);
3650 err = re_node_set_init_1 (dests_node + ndests, cur_nodes->elems[i]);
3651 if (BE (err != REG_NOERROR, 0))
3652 goto error_return;
3653 ++ndests;
3654 bitset_empty (accepts);
3655 }
3656 }
3657 return ndests;
3658 error_return:
3659 for (j = 0; j < ndests; ++j)
3660 re_node_set_free (dests_node + j);
3661 return -1;
3662}
3663
3664#ifdef RE_ENABLE_I18N
3665/* Check how many bytes the node `dfa->nodes[node_idx]' accepts.
3666 Return the number of the bytes the node accepts.
3667 STR_IDX is the current index of the input string.
3668
3669 This function handles the nodes which can accept one character, or
3670 one collating element like '.', '[a-z]', opposite to the other nodes
3671 can only accept one byte. */
3672
3673# ifdef _LIBC
3674# include <locale/weight.h>
3675# endif
3676
3677static int
3678check_node_accept_bytes (const re_dfa_t *dfa, int node_idx,
3679 const re_string_t *input, int str_idx)
3680{
3681 const re_token_t *node = dfa->nodes + node_idx;
3682 int char_len, elem_len;
3683 int i;
3684
3685 if (BE (node->type == OP_UTF8_PERIOD, 0))
3686 {
3687 unsigned char c = re_string_byte_at (input, str_idx), d;
3688 if (BE (c < 0xc2, 1))
3689 return 0;
3690
3691 if (str_idx + 2 > input->len)
3692 return 0;
3693
3694 d = re_string_byte_at (input, str_idx + 1);
3695 if (c < 0xe0)
3696 return (d < 0x80 || d > 0xbf) ? 0 : 2;
3697 else if (c < 0xf0)
3698 {
3699 char_len = 3;
3700 if (c == 0xe0 && d < 0xa0)
3701 return 0;
3702 }
3703 else if (c < 0xf8)
3704 {
3705 char_len = 4;
3706 if (c == 0xf0 && d < 0x90)
3707 return 0;
3708 }
3709 else if (c < 0xfc)
3710 {
3711 char_len = 5;
3712 if (c == 0xf8 && d < 0x88)
3713 return 0;
3714 }
3715 else if (c < 0xfe)
3716 {
3717 char_len = 6;
3718 if (c == 0xfc && d < 0x84)
3719 return 0;
3720 }
3721 else
3722 return 0;
3723
3724 if (str_idx + char_len > input->len)
3725 return 0;
3726
3727 for (i = 1; i < char_len; ++i)
3728 {
3729 d = re_string_byte_at (input, str_idx + i);
3730 if (d < 0x80 || d > 0xbf)
3731 return 0;
3732 }
3733 return char_len;
3734 }
3735
3736 char_len = re_string_char_size_at (input, str_idx);
3737 if (node->type == OP_PERIOD)
3738 {
3739 if (char_len <= 1)
3740 return 0;
3741 /* FIXME: I don't think this if is needed, as both '\n'
3742 and '\0' are char_len == 1. */
3743 /* '.' accepts any one character except the following two cases. */
3744 if ((!(dfa->syntax & RE_DOT_NEWLINE) &&
3745 re_string_byte_at (input, str_idx) == '\n') ||
3746 ((dfa->syntax & RE_DOT_NOT_NULL) &&
3747 re_string_byte_at (input, str_idx) == '\0'))
3748 return 0;
3749 return char_len;
3750 }
3751
3752 elem_len = re_string_elem_size_at (input, str_idx);
3753 if ((elem_len <= 1 && char_len <= 1) || char_len == 0)
3754 return 0;
3755
3756 if (node->type == COMPLEX_BRACKET)
3757 {
3758 const re_charset_t *cset = node->opr.mbcset;
3759# ifdef _LIBC
3760 const unsigned char *pin
3761 = ((const unsigned char *) re_string_get_buffer (input) + str_idx);
3762 int j;
3763 uint32_t nrules;
3764# endif /* _LIBC */
3765 int match_len = 0;
3766 wchar_t wc = ((cset->nranges || cset->nchar_classes || cset->nmbchars)
3767 ? re_string_wchar_at (input, str_idx) : 0);
3768
3769 /* match with multibyte character? */
3770 for (i = 0; i < cset->nmbchars; ++i)
3771 if (wc == cset->mbchars[i])
3772 {
3773 match_len = char_len;
3774 goto check_node_accept_bytes_match;
3775 }
3776 /* match with character_class? */
3777 for (i = 0; i < cset->nchar_classes; ++i)
3778 {
3779 wctype_t wt = cset->char_classes[i];
3780 if (__iswctype (wc, wt))
3781 {
3782 match_len = char_len;
3783 goto check_node_accept_bytes_match;
3784 }
3785 }
3786
3787# ifdef _LIBC
3788 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3789 if (nrules != 0)
3790 {
3791 unsigned int in_collseq = 0;
3792 const int32_t *table, *indirect;
3793 const unsigned char *weights, *extra;
3794 const char *collseqwc;
3795
3796 /* match with collating_symbol? */
3797 if (cset->ncoll_syms)
3798 extra = (const unsigned char *)
3799 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
3800 for (i = 0; i < cset->ncoll_syms; ++i)
3801 {
3802 const unsigned char *coll_sym = extra + cset->coll_syms[i];
3803 /* Compare the length of input collating element and
3804 the length of current collating element. */
3805 if (*coll_sym != elem_len)
3806 continue;
3807 /* Compare each bytes. */
3808 for (j = 0; j < *coll_sym; j++)
3809 if (pin[j] != coll_sym[1 + j])
3810 break;
3811 if (j == *coll_sym)
3812 {
3813 /* Match if every bytes is equal. */
3814 match_len = j;
3815 goto check_node_accept_bytes_match;
3816 }
3817 }
3818
3819 if (cset->nranges)
3820 {
3821 if (elem_len <= char_len)
3822 {
3823 collseqwc = _NL_CURRENT (LC_COLLATE, _NL_COLLATE_COLLSEQWC);
3824 in_collseq = __collseq_table_lookup (collseqwc, wc);
3825 }
3826 else
3827 in_collseq = find_collation_sequence_value (pin, elem_len);
3828 }
3829 /* match with range expression? */
3830 for (i = 0; i < cset->nranges; ++i)
3831 if (cset->range_starts[i] <= in_collseq
3832 && in_collseq <= cset->range_ends[i])
3833 {
3834 match_len = elem_len;
3835 goto check_node_accept_bytes_match;
3836 }
3837
3838 /* match with equivalence_class? */
3839 if (cset->nequiv_classes)
3840 {
3841 const unsigned char *cp = pin;
3842 table = (const int32_t *)
3843 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
3844 weights = (const unsigned char *)
3845 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
3846 extra = (const unsigned char *)
3847 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
3848 indirect = (const int32_t *)
3849 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
3850 int32_t idx = findidx (table, indirect, extra, &cp, elem_len);
3851 int32_t rule = idx >> 24;
3852 idx &= 0xffffff;
3853 if (idx > 0)
3854 {
3855 size_t weight_len = weights[idx];
3856 for (i = 0; i < cset->nequiv_classes; ++i)
3857 {
3858 int32_t equiv_class_idx = cset->equiv_classes[i];
3859 int32_t equiv_class_rule = equiv_class_idx >> 24;
3860 equiv_class_idx &= 0xffffff;
3861 if (weights[equiv_class_idx] == weight_len
3862 && equiv_class_rule == rule
3863 && memcmp (weights + idx + 1,
3864 weights + equiv_class_idx + 1,
3865 weight_len) == 0)
3866 {
3867 match_len = elem_len;
3868 goto check_node_accept_bytes_match;
3869 }
3870 }
3871 }
3872 }
3873 }
3874 else
3875# endif /* _LIBC */
3876 {
3877 /* match with range expression? */
3878#if __GNUC__ >= 2
3879 wchar_t cmp_buf[] = {L'\0', L'\0', wc, L'\0', L'\0', L'\0'};
3880#else
3881 wchar_t cmp_buf[] = {L'\0', L'\0', L'\0', L'\0', L'\0', L'\0'};
3882 cmp_buf[2] = wc;
3883#endif
3884 for (i = 0; i < cset->nranges; ++i)
3885 {
3886 cmp_buf[0] = cset->range_starts[i];
3887 cmp_buf[4] = cset->range_ends[i];
3888 if (__wcscoll (cmp_buf, cmp_buf + 2) <= 0
3889 && __wcscoll (cmp_buf + 2, cmp_buf + 4) <= 0)
3890 {
3891 match_len = char_len;
3892 goto check_node_accept_bytes_match;
3893 }
3894 }
3895 }
3896 check_node_accept_bytes_match:
3897 if (!cset->non_match)
3898 return match_len;
3899 else
3900 {
3901 if (match_len > 0)
3902 return 0;
3903 else
3904 return (elem_len > char_len) ? elem_len : char_len;
3905 }
3906 }
3907 return 0;
3908}
3909
3910# ifdef _LIBC
3911static unsigned int
3912find_collation_sequence_value (const unsigned char *mbs, size_t mbs_len)
3913{
3914 uint32_t nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3915 if (nrules == 0)
3916 {
3917 if (mbs_len == 1)
3918 {
3919 /* No valid character. Match it as a single byte character. */
3920 const unsigned char *collseq = (const unsigned char *)
3921 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_COLLSEQMB);
3922 return collseq[mbs[0]];
3923 }
3924 return UINT_MAX;
3925 }
3926 else
3927 {
3928 int32_t idx;
3929 const unsigned char *extra = (const unsigned char *)
3930 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
3931 int32_t extrasize = (const unsigned char *)
3932 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB + 1) - extra;
3933
3934 for (idx = 0; idx < extrasize;)
3935 {
3936 int mbs_cnt, found = 0;
3937 int32_t elem_mbs_len;
3938 /* Skip the name of collating element name. */
3939 idx = idx + extra[idx] + 1;
3940 elem_mbs_len = extra[idx++];
3941 if (mbs_len == elem_mbs_len)
3942 {
3943 for (mbs_cnt = 0; mbs_cnt < elem_mbs_len; ++mbs_cnt)
3944 if (extra[idx + mbs_cnt] != mbs[mbs_cnt])
3945 break;
3946 if (mbs_cnt == elem_mbs_len)
3947 /* Found the entry. */
3948 found = 1;
3949 }
3950 /* Skip the byte sequence of the collating element. */
3951 idx += elem_mbs_len;
3952 /* Adjust for the alignment. */
3953 idx = (idx + 3) & ~3;
3954 /* Skip the collation sequence value. */
3955 idx += sizeof (uint32_t);
3956 /* Skip the wide char sequence of the collating element. */
3957 idx = idx + sizeof (uint32_t) * (*(int32_t *) (extra + idx) + 1);
3958 /* If we found the entry, return the sequence value. */
3959 if (found)
3960 return *(uint32_t *) (extra + idx);
3961 /* Skip the collation sequence value. */
3962 idx += sizeof (uint32_t);
3963 }
3964 return UINT_MAX;
3965 }
3966}
3967# endif /* _LIBC */
3968#endif /* RE_ENABLE_I18N */
3969
3970/* Check whether the node accepts the byte which is IDX-th
3971 byte of the INPUT. */
3972
3973static int
3974check_node_accept (const re_match_context_t *mctx, const re_token_t *node,
3975 int idx)
3976{
3977 unsigned char ch;
3978 ch = re_string_byte_at (&mctx->input, idx);
3979 switch (node->type)
3980 {
3981 case CHARACTER:
3982 if (node->opr.c != ch)
3983 return 0;
3984 break;
3985
3986 case SIMPLE_BRACKET:
3987 if (!bitset_contain (node->opr.sbcset, ch))
3988 return 0;
3989 break;
3990
3991#ifdef RE_ENABLE_I18N
3992 case OP_UTF8_PERIOD:
3993 if (ch >= 0x80)
3994 return 0;
3995 /* FALLTHROUGH */
3996#endif
3997 case OP_PERIOD:
3998 if ((ch == '\n' && !(mctx->dfa->syntax & RE_DOT_NEWLINE))
3999 || (ch == '\0' && (mctx->dfa->syntax & RE_DOT_NOT_NULL)))
4000 return 0;
4001 break;
4002
4003 default:
4004 return 0;
4005 }
4006
4007 if (node->constraint)
4008 {
4009 /* The node has constraints. Check whether the current context
4010 satisfies the constraints. */
4011 unsigned int context = re_string_context_at (&mctx->input, idx,
4012 mctx->eflags);
4013 if (NOT_SATISFY_NEXT_CONSTRAINT (node->constraint, context))
4014 return 0;
4015 }
4016
4017 return 1;
4018}
4019
4020/* Extend the buffers, if the buffers have run out. */
4021
4022static reg_errcode_t
4023__attribute_warn_unused_result__
4024extend_buffers (re_match_context_t *mctx, int min_len)
4025{
4026 reg_errcode_t ret;
4027 re_string_t *pstr = &mctx->input;
4028
4029 /* Avoid overflow. */
4030 if (BE (INT_MAX / 2 / sizeof (re_dfastate_t *) <= pstr->bufs_len, 0))
4031 return REG_ESPACE;
4032
4033 /* Double the lengthes of the buffers, but allocate at least MIN_LEN. */
4034 ret = re_string_realloc_buffers (pstr,
4035 MAX (min_len,
4036 MIN (pstr->len, pstr->bufs_len * 2)));
4037 if (BE (ret != REG_NOERROR, 0))
4038 return ret;
4039
4040 if (mctx->state_log != NULL)
4041 {
4042 /* And double the length of state_log. */
4043 /* XXX We have no indication of the size of this buffer. If this
4044 allocation fail we have no indication that the state_log array
4045 does not have the right size. */
4046 re_dfastate_t **new_array = re_realloc (mctx->state_log, re_dfastate_t *,
4047 pstr->bufs_len + 1);
4048 if (BE (new_array == NULL, 0))
4049 return REG_ESPACE;
4050 mctx->state_log = new_array;
4051 }
4052
4053 /* Then reconstruct the buffers. */
4054 if (pstr->icase)
4055 {
4056#ifdef RE_ENABLE_I18N
4057 if (pstr->mb_cur_max > 1)
4058 {
4059 ret = build_wcs_upper_buffer (pstr);
4060 if (BE (ret != REG_NOERROR, 0))
4061 return ret;
4062 }
4063 else
4064#endif /* RE_ENABLE_I18N */
4065 build_upper_buffer (pstr);
4066 }
4067 else
4068 {
4069#ifdef RE_ENABLE_I18N
4070 if (pstr->mb_cur_max > 1)
4071 build_wcs_buffer (pstr);
4072 else
4073#endif /* RE_ENABLE_I18N */
4074 {
4075 if (pstr->trans != NULL)
4076 re_string_translate_buffer (pstr);
4077 }
4078 }
4079 return REG_NOERROR;
4080}
4081
4082
4083/* Functions for matching context. */
4084
4085/* Initialize MCTX. */
4086
4087static reg_errcode_t
4088__attribute_warn_unused_result__
4089match_ctx_init (re_match_context_t *mctx, int eflags, int n)
4090{
4091 mctx->eflags = eflags;
4092 mctx->match_last = -1;
4093 if (n > 0)
4094 {
4095 mctx->bkref_ents = re_malloc (struct re_backref_cache_entry, n);
4096 mctx->sub_tops = re_malloc (re_sub_match_top_t *, n);
4097 if (BE (mctx->bkref_ents == NULL || mctx->sub_tops == NULL, 0))
4098 return REG_ESPACE;
4099 }
4100 /* Already zero-ed by the caller.
4101 else
4102 mctx->bkref_ents = NULL;
4103 mctx->nbkref_ents = 0;
4104 mctx->nsub_tops = 0; */
4105 mctx->abkref_ents = n;
4106 mctx->max_mb_elem_len = 1;
4107 mctx->asub_tops = n;
4108 return REG_NOERROR;
4109}
4110
4111/* Clean the entries which depend on the current input in MCTX.
4112 This function must be invoked when the matcher changes the start index
4113 of the input, or changes the input string. */
4114
4115static void
4116match_ctx_clean (re_match_context_t *mctx)
4117{
4118 int st_idx;
4119 for (st_idx = 0; st_idx < mctx->nsub_tops; ++st_idx)
4120 {
4121 int sl_idx;
4122 re_sub_match_top_t *top = mctx->sub_tops[st_idx];
4123 for (sl_idx = 0; sl_idx < top->nlasts; ++sl_idx)
4124 {
4125 re_sub_match_last_t *last = top->lasts[sl_idx];
4126 re_free (last->path.array);
4127 re_free (last);
4128 }
4129 re_free (top->lasts);
4130 if (top->path)
4131 {
4132 re_free (top->path->array);
4133 re_free (top->path);
4134 }
4135 free (top);
4136 }
4137
4138 mctx->nsub_tops = 0;
4139 mctx->nbkref_ents = 0;
4140}
4141
4142/* Free all the memory associated with MCTX. */
4143
4144static void
4145match_ctx_free (re_match_context_t *mctx)
4146{
4147 /* First, free all the memory associated with MCTX->SUB_TOPS. */
4148 match_ctx_clean (mctx);
4149 re_free (mctx->sub_tops);
4150 re_free (mctx->bkref_ents);
4151}
4152
4153/* Add a new backreference entry to MCTX.
4154 Note that we assume that caller never call this function with duplicate
4155 entry, and call with STR_IDX which isn't smaller than any existing entry.
4156*/
4157
4158static reg_errcode_t
4159__attribute_warn_unused_result__
4160match_ctx_add_entry (re_match_context_t *mctx, int node, int str_idx, int from,
4161 int to)
4162{
4163 if (mctx->nbkref_ents >= mctx->abkref_ents)
4164 {
4165 struct re_backref_cache_entry* new_entry;
4166 new_entry = re_realloc (mctx->bkref_ents, struct re_backref_cache_entry,
4167 mctx->abkref_ents * 2);
4168 if (BE (new_entry == NULL, 0))
4169 {
4170 re_free (mctx->bkref_ents);
4171 return REG_ESPACE;
4172 }
4173 mctx->bkref_ents = new_entry;
4174 memset (mctx->bkref_ents + mctx->nbkref_ents, '\0',
4175 sizeof (struct re_backref_cache_entry) * mctx->abkref_ents);
4176 mctx->abkref_ents *= 2;
4177 }
4178 if (mctx->nbkref_ents > 0
4179 && mctx->bkref_ents[mctx->nbkref_ents - 1].str_idx == str_idx)
4180 mctx->bkref_ents[mctx->nbkref_ents - 1].more = 1;
4181
4182 mctx->bkref_ents[mctx->nbkref_ents].node = node;
4183 mctx->bkref_ents[mctx->nbkref_ents].str_idx = str_idx;
4184 mctx->bkref_ents[mctx->nbkref_ents].subexp_from = from;
4185 mctx->bkref_ents[mctx->nbkref_ents].subexp_to = to;
4186
4187 /* This is a cache that saves negative results of check_dst_limits_calc_pos.
4188 If bit N is clear, means that this entry won't epsilon-transition to
4189 an OP_OPEN_SUBEXP or OP_CLOSE_SUBEXP for the N+1-th subexpression. If
4190 it is set, check_dst_limits_calc_pos_1 will recurse and try to find one
4191 such node.
4192
4193 A backreference does not epsilon-transition unless it is empty, so set
4194 to all zeros if FROM != TO. */
4195 mctx->bkref_ents[mctx->nbkref_ents].eps_reachable_subexps_map
4196 = (from == to ? ~0 : 0);
4197
4198 mctx->bkref_ents[mctx->nbkref_ents++].more = 0;
4199 if (mctx->max_mb_elem_len < to - from)
4200 mctx->max_mb_elem_len = to - from;
4201 return REG_NOERROR;
4202}
4203
4204/* Search for the first entry which has the same str_idx, or -1 if none is
4205 found. Note that MCTX->BKREF_ENTS is already sorted by MCTX->STR_IDX. */
4206
4207static int
4208search_cur_bkref_entry (const re_match_context_t *mctx, int str_idx)
4209{
4210 int left, right, mid, last;
4211 last = right = mctx->nbkref_ents;
4212 for (left = 0; left < right;)
4213 {
4214 mid = (left + right) / 2;
4215 if (mctx->bkref_ents[mid].str_idx < str_idx)
4216 left = mid + 1;
4217 else
4218 right = mid;
4219 }
4220 if (left < last && mctx->bkref_ents[left].str_idx == str_idx)
4221 return left;
4222 else
4223 return -1;
4224}
4225
4226/* Register the node NODE, whose type is OP_OPEN_SUBEXP, and which matches
4227 at STR_IDX. */
4228
4229static reg_errcode_t
4230__attribute_warn_unused_result__
4231match_ctx_add_subtop (re_match_context_t *mctx, int node, int str_idx)
4232{
4233#ifdef DEBUG
4234 assert (mctx->sub_tops != NULL);
4235 assert (mctx->asub_tops > 0);
4236#endif
4237 if (BE (mctx->nsub_tops == mctx->asub_tops, 0))
4238 {
4239 int new_asub_tops = mctx->asub_tops * 2;
4240 re_sub_match_top_t **new_array = re_realloc (mctx->sub_tops,
4241 re_sub_match_top_t *,
4242 new_asub_tops);
4243 if (BE (new_array == NULL, 0))
4244 return REG_ESPACE;
4245 mctx->sub_tops = new_array;
4246 mctx->asub_tops = new_asub_tops;
4247 }
4248 mctx->sub_tops[mctx->nsub_tops] = calloc (1, sizeof (re_sub_match_top_t));
4249 if (BE (mctx->sub_tops[mctx->nsub_tops] == NULL, 0))
4250 return REG_ESPACE;
4251 mctx->sub_tops[mctx->nsub_tops]->node = node;
4252 mctx->sub_tops[mctx->nsub_tops++]->str_idx = str_idx;
4253 return REG_NOERROR;
4254}
4255
4256/* Register the node NODE, whose type is OP_CLOSE_SUBEXP, and which matches
4257 at STR_IDX, whose corresponding OP_OPEN_SUBEXP is SUB_TOP. */
4258
4259static re_sub_match_last_t *
4260match_ctx_add_sublast (re_sub_match_top_t *subtop, int node, int str_idx)
4261{
4262 re_sub_match_last_t *new_entry;
4263 if (BE (subtop->nlasts == subtop->alasts, 0))
4264 {
4265 int new_alasts = 2 * subtop->alasts + 1;
4266 re_sub_match_last_t **new_array = re_realloc (subtop->lasts,
4267 re_sub_match_last_t *,
4268 new_alasts);
4269 if (BE (new_array == NULL, 0))
4270 return NULL;
4271 subtop->lasts = new_array;
4272 subtop->alasts = new_alasts;
4273 }
4274 new_entry = calloc (1, sizeof (re_sub_match_last_t));
4275 if (BE (new_entry != NULL, 1))
4276 {
4277 subtop->lasts[subtop->nlasts] = new_entry;
4278 new_entry->node = node;
4279 new_entry->str_idx = str_idx;
4280 ++subtop->nlasts;
4281 }
4282 return new_entry;
4283}
4284
4285static void
4286sift_ctx_init (re_sift_context_t *sctx, re_dfastate_t **sifted_sts,
4287 re_dfastate_t **limited_sts, int last_node, int last_str_idx)
4288{
4289 sctx->sifted_states = sifted_sts;
4290 sctx->limited_states = limited_sts;
4291 sctx->last_node = last_node;
4292 sctx->last_str_idx = last_str_idx;
4293 re_node_set_init_empty (&sctx->limits);
4294}
4295