1 /* Extended regular expression matching and search library.
2 Copyright (C) 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4 Contributed by Isamu Hasegawa <isamu@yamato.ibm.com>.
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.
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.
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/>. */
20 static reg_errcode_t match_ctx_init (re_match_context_t *cache, int eflags,
21 int n) internal_function;
22 static void match_ctx_clean (re_match_context_t *mctx) internal_function;
23 static void match_ctx_free (re_match_context_t *cache) internal_function;
24 static reg_errcode_t match_ctx_add_entry (re_match_context_t *cache, int node,
25 int str_idx, int from, int to)
27 static int search_cur_bkref_entry (const re_match_context_t *mctx, int str_idx)
29 static reg_errcode_t match_ctx_add_subtop (re_match_context_t *mctx, int node,
30 int str_idx) internal_function;
31 static re_sub_match_last_t * match_ctx_add_sublast (re_sub_match_top_t *subtop,
32 int node, int str_idx)
34 static void sift_ctx_init (re_sift_context_t *sctx, re_dfastate_t **sifted_sts,
35 re_dfastate_t **limited_sts, int last_node,
38 static reg_errcode_t re_search_internal (const regex_t *preg,
39 const char *string, int length,
40 int start, int range, int stop,
41 size_t nmatch, regmatch_t pmatch[],
42 int eflags) internal_function;
43 static int re_search_2_stub (struct re_pattern_buffer *bufp,
44 const char *string1, int length1,
45 const char *string2, int length2,
46 int start, int range, struct re_registers *regs,
47 int stop, int ret_len) internal_function;
48 static int re_search_stub (struct re_pattern_buffer *bufp,
49 const char *string, int length, int start,
50 int range, int stop, struct re_registers *regs,
51 int ret_len) internal_function;
52 static unsigned re_copy_regs (struct re_registers *regs, regmatch_t *pmatch,
53 int nregs, int regs_allocated) internal_function;
54 static reg_errcode_t prune_impossible_nodes (re_match_context_t *mctx)
56 static int check_matching (re_match_context_t *mctx, int fl_longest_match,
57 int *p_match_first) internal_function;
58 static int check_halt_state_context (const re_match_context_t *mctx,
59 const re_dfastate_t *state, int idx)
61 static void update_regs (const re_dfa_t *dfa, regmatch_t *pmatch,
62 regmatch_t *prev_idx_match, int cur_node,
63 int cur_idx, int nmatch) internal_function;
64 static reg_errcode_t push_fail_stack (struct re_fail_stack_t *fs,
65 int str_idx, int dest_node, int nregs,
67 re_node_set *eps_via_nodes)
69 static reg_errcode_t set_regs (const regex_t *preg,
70 const re_match_context_t *mctx,
71 size_t nmatch, regmatch_t *pmatch,
72 int fl_backtrack) internal_function;
73 static reg_errcode_t free_fail_stack_return (struct re_fail_stack_t *fs)
77 static int sift_states_iter_mb (const re_match_context_t *mctx,
78 re_sift_context_t *sctx,
79 int node_idx, int str_idx, int max_str_idx)
82 static reg_errcode_t sift_states_backward (const re_match_context_t *mctx,
83 re_sift_context_t *sctx)
85 static reg_errcode_t build_sifted_states (const re_match_context_t *mctx,
86 re_sift_context_t *sctx, int str_idx,
87 re_node_set *cur_dest)
89 static reg_errcode_t update_cur_sifted_state (const re_match_context_t *mctx,
90 re_sift_context_t *sctx,
92 re_node_set *dest_nodes)
94 static reg_errcode_t add_epsilon_src_nodes (const re_dfa_t *dfa,
95 re_node_set *dest_nodes,
96 const re_node_set *candidates)
98 static int check_dst_limits (const re_match_context_t *mctx,
100 int dst_node, int dst_idx, int src_node,
101 int src_idx) internal_function;
102 static int check_dst_limits_calc_pos_1 (const re_match_context_t *mctx,
103 int boundaries, int subexp_idx,
104 int from_node, int bkref_idx)
106 static int check_dst_limits_calc_pos (const re_match_context_t *mctx,
107 int limit, int subexp_idx,
108 int node, int str_idx,
109 int bkref_idx) internal_function;
110 static reg_errcode_t check_subexp_limits (const re_dfa_t *dfa,
111 re_node_set *dest_nodes,
112 const re_node_set *candidates,
114 struct re_backref_cache_entry *bkref_ents,
115 int str_idx) internal_function;
116 static reg_errcode_t sift_states_bkref (const re_match_context_t *mctx,
117 re_sift_context_t *sctx,
118 int str_idx, const re_node_set *candidates)
120 static reg_errcode_t merge_state_array (const re_dfa_t *dfa,
122 re_dfastate_t **src, int num)
124 static re_dfastate_t *find_recover_state (reg_errcode_t *err,
125 re_match_context_t *mctx) internal_function;
126 static re_dfastate_t *transit_state (reg_errcode_t *err,
127 re_match_context_t *mctx,
128 re_dfastate_t *state) internal_function;
129 static re_dfastate_t *merge_state_with_log (reg_errcode_t *err,
130 re_match_context_t *mctx,
131 re_dfastate_t *next_state)
133 static reg_errcode_t check_subexp_matching_top (re_match_context_t *mctx,
134 re_node_set *cur_nodes,
135 int str_idx) internal_function;
137 static re_dfastate_t *transit_state_sb (reg_errcode_t *err,
138 re_match_context_t *mctx,
139 re_dfastate_t *pstate)
142 #ifdef RE_ENABLE_I18N
143 static reg_errcode_t transit_state_mb (re_match_context_t *mctx,
144 re_dfastate_t *pstate)
147 static reg_errcode_t transit_state_bkref (re_match_context_t *mctx,
148 const re_node_set *nodes)
150 static reg_errcode_t get_subexp (re_match_context_t *mctx,
151 int bkref_node, int bkref_str_idx)
153 static reg_errcode_t get_subexp_sub (re_match_context_t *mctx,
154 const re_sub_match_top_t *sub_top,
155 re_sub_match_last_t *sub_last,
156 int bkref_node, int bkref_str)
158 static int find_subexp_node (const re_dfa_t *dfa, const re_node_set *nodes,
159 int subexp_idx, int type) internal_function;
160 static reg_errcode_t check_arrival (re_match_context_t *mctx,
161 state_array_t *path, int top_node,
162 int top_str, int last_node, int last_str,
163 int type) internal_function;
164 static reg_errcode_t check_arrival_add_next_nodes (re_match_context_t *mctx,
166 re_node_set *cur_nodes,
167 re_node_set *next_nodes)
169 static reg_errcode_t check_arrival_expand_ecl (const re_dfa_t *dfa,
170 re_node_set *cur_nodes,
171 int ex_subexp, int type)
173 static reg_errcode_t check_arrival_expand_ecl_sub (const re_dfa_t *dfa,
174 re_node_set *dst_nodes,
175 int target, int ex_subexp,
176 int type) internal_function;
177 static reg_errcode_t expand_bkref_cache (re_match_context_t *mctx,
178 re_node_set *cur_nodes, int cur_str,
179 int subexp_num, int type)
181 static int build_trtable (const re_dfa_t *dfa,
182 re_dfastate_t *state) internal_function;
183 #ifdef RE_ENABLE_I18N
184 static int check_node_accept_bytes (const re_dfa_t *dfa, int node_idx,
185 const re_string_t *input, int idx)
188 static int group_nodes_into_DFAstates (const re_dfa_t *dfa,
189 const re_dfastate_t *state,
190 re_node_set *states_node,
191 bitset_t *states_ch) internal_function;
192 static int check_node_accept (const re_match_context_t *mctx,
193 const re_token_t *node, int idx)
195 static reg_errcode_t extend_buffers (re_match_context_t *mctx)
198 /* Entry point for POSIX code. */
200 /* regexec searches for a given pattern, specified by PREG, in the
203 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
204 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
205 least NMATCH elements, and we set them to the offsets of the
206 corresponding matched substrings.
208 EFLAGS specifies `execution flags' which affect matching: if
209 REG_NOTBOL is set, then ^ does not match at the beginning of the
210 string; if REG_NOTEOL is set, then $ does not match at the end.
212 We return 0 if we find a match and REG_NOMATCH if not. */
215 regexec (const regex_t *__restrict preg, const char *__restrict string,
216 size_t nmatch, regmatch_t pmatch[], int eflags)
220 #ifdef __UCLIBC_HAS_THREADS__
221 re_dfa_t *dfa = (re_dfa_t *) preg->buffer;
224 if (eflags & ~(REG_NOTBOL | REG_NOTEOL | REG_STARTEND))
227 if (eflags & REG_STARTEND)
229 start = pmatch[0].rm_so;
230 length = pmatch[0].rm_eo;
235 length = strlen (string);
238 __libc_lock_lock (dfa->lock);
240 err = re_search_internal (preg, string, length, start, length - start,
241 length, 0, NULL, eflags);
243 err = re_search_internal (preg, string, length, start, length - start,
244 length, nmatch, pmatch, eflags);
245 __libc_lock_unlock (dfa->lock);
246 return err != REG_NOERROR;
248 libc_hidden_def(regexec)
250 /* Entry points for GNU code. */
252 /* re_match, re_search, re_match_2, re_search_2
254 The former two functions operate on STRING with length LENGTH,
255 while the later two operate on concatenation of STRING1 and STRING2
256 with lengths LENGTH1 and LENGTH2, respectively.
258 re_match() matches the compiled pattern in BUFP against the string,
259 starting at index START.
261 re_search() first tries matching at index START, then it tries to match
262 starting from index START + 1, and so on. The last start position tried
263 is START + RANGE. (Thus RANGE = 0 forces re_search to operate the same
266 The parameter STOP of re_{match,search}_2 specifies that no match exceeding
267 the first STOP characters of the concatenation of the strings should be
270 If REGS is not NULL, and BUFP->no_sub is not set, the offsets of the match
271 and all groups is stroed in REGS. (For the "_2" variants, the offsets are
272 computed relative to the concatenation, not relative to the individual
275 On success, re_match* functions return the length of the match, re_search*
276 return the position of the start of the match. Return value -1 means no
277 match was found and -2 indicates an internal error. */
280 re_match (struct re_pattern_buffer *bufp, const char *string, int length,
281 int start, struct re_registers *regs)
283 return re_search_stub (bufp, string, length, start, 0, length, regs, 1);
287 re_search (struct re_pattern_buffer *bufp, const char *string, int length,
288 int start, int range, struct re_registers *regs)
290 return re_search_stub (bufp, string, length, start, range, length, regs, 0);
292 libc_hidden_def(re_search)
295 re_match_2 (struct re_pattern_buffer *bufp, const char *string1, int length1,
296 const char *string2, int length2, int start,
297 struct re_registers *regs, int stop)
299 return re_search_2_stub (bufp, string1, length1, string2, length2,
300 start, 0, regs, stop, 1);
304 re_search_2 (struct re_pattern_buffer *bufp, const char *string1, int lenght1,
305 const char *string2, int length2, int start, int range,
306 struct re_registers *regs, int stop)
308 return re_search_2_stub (bufp, string1, lenght1, string2, length2,
309 start, range, regs, stop, 0);
311 libc_hidden_def(re_search_2)
313 static int internal_function
314 re_search_2_stub (struct re_pattern_buffer *bufp, const char *string1,
315 int length1, const char *string2, int length2, int start,
316 int range, struct re_registers *regs, int stop, int ret_len)
320 int len = length1 + length2;
323 if (BE (length1 < 0 || length2 < 0 || stop < 0, 0))
326 /* Concatenate the strings. */
330 char *s = re_malloc (char, len);
332 if (BE (s == NULL, 0))
334 memcpy (s, string1, length1);
335 memcpy (s + length1, string2, length2);
344 rval = re_search_stub (bufp, str, len, start, range, stop, regs,
347 re_free ((char *) str);
351 /* The parameters have the same meaning as those of re_search.
352 Additional parameters:
353 If RET_LEN is nonzero the length of the match is returned (re_match style);
354 otherwise the position of the match is returned. */
356 static int internal_function
357 re_search_stub (struct re_pattern_buffer *bufp, const char *string, int length,
358 int start, int range, int stop, struct re_registers *regs,
361 reg_errcode_t result;
365 #ifdef __UCLIBC_HAS_THREADS__
366 re_dfa_t *dfa = (re_dfa_t *) bufp->buffer;
368 /* Check for out-of-range. */
369 if (BE (start < 0 || start > length, 0))
371 if (BE (start + range > length, 0))
372 range = length - start;
373 else if (BE (start + range < 0, 0))
376 __libc_lock_lock (dfa->lock);
378 eflags |= (bufp->not_bol) ? REG_NOTBOL : 0;
379 eflags |= (bufp->not_eol) ? REG_NOTEOL : 0;
381 /* Compile fastmap if we haven't yet. */
382 if (range > 0 && bufp->fastmap != NULL && !bufp->fastmap_accurate)
383 re_compile_fastmap (bufp);
385 if (BE (bufp->no_sub, 0))
388 /* We need at least 1 register. */
391 else if (BE (bufp->regs_allocated == REGS_FIXED &&
392 regs->num_regs < bufp->re_nsub + 1, 0))
394 nregs = regs->num_regs;
395 if (BE (nregs < 1, 0))
397 /* Nothing can be copied to regs. */
403 nregs = bufp->re_nsub + 1;
404 pmatch = re_malloc (regmatch_t, nregs);
405 if (BE (pmatch == NULL, 0))
411 result = re_search_internal (bufp, string, length, start, range, stop,
412 nregs, pmatch, eflags);
416 /* I hope we needn't fill ther regs with -1's when no match was found. */
417 if (result != REG_NOERROR)
419 else if (regs != NULL)
421 /* If caller wants register contents data back, copy them. */
422 bufp->regs_allocated = re_copy_regs (regs, pmatch, nregs,
423 bufp->regs_allocated);
424 if (BE (bufp->regs_allocated == REGS_UNALLOCATED, 0))
428 if (BE (rval == 0, 1))
432 assert (pmatch[0].rm_so == start);
433 rval = pmatch[0].rm_eo - start;
436 rval = pmatch[0].rm_so;
440 __libc_lock_unlock (dfa->lock);
444 static unsigned internal_function
445 re_copy_regs (struct re_registers *regs, regmatch_t *pmatch, int nregs,
448 int rval = REGS_REALLOCATE;
450 int need_regs = nregs + 1;
451 /* We need one extra element beyond `num_regs' for the `-1' marker GNU code
454 /* Have the register data arrays been allocated? */
455 if (regs_allocated == REGS_UNALLOCATED)
456 { /* No. So allocate them with malloc. */
457 regs->start = re_malloc (regoff_t, need_regs);
458 regs->end = re_malloc (regoff_t, need_regs);
459 if (BE (regs->start == NULL, 0) || BE (regs->end == NULL, 0))
460 return REGS_UNALLOCATED;
461 regs->num_regs = need_regs;
463 else if (regs_allocated == REGS_REALLOCATE)
464 { /* Yes. If we need more elements than were already
465 allocated, reallocate them. If we need fewer, just
467 if (BE (need_regs > regs->num_regs, 0))
469 regoff_t *new_start = re_realloc (regs->start, regoff_t, need_regs);
470 regoff_t *new_end = re_realloc (regs->end, regoff_t, need_regs);
471 if (BE (new_start == NULL, 0) || BE (new_end == NULL, 0))
472 return REGS_UNALLOCATED;
473 regs->start = new_start;
475 regs->num_regs = need_regs;
480 assert (regs_allocated == REGS_FIXED);
481 /* This function may not be called with REGS_FIXED and nregs too big. */
482 assert (regs->num_regs >= nregs);
487 for (i = 0; i < nregs; ++i)
489 regs->start[i] = pmatch[i].rm_so;
490 regs->end[i] = pmatch[i].rm_eo;
492 for ( ; i < regs->num_regs; ++i)
493 regs->start[i] = regs->end[i] = -1;
498 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
499 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
500 this memory for recording register information. STARTS and ENDS
501 must be allocated using the malloc library routine, and must each
502 be at least NUM_REGS * sizeof (regoff_t) bytes long.
504 If NUM_REGS == 0, then subsequent matches should allocate their own
507 Unless this function is called, the first search or match using
508 PATTERN_BUFFER will allocate its own register data, without
509 freeing the old data. */
512 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs,
513 unsigned num_regs, regoff_t *starts, regoff_t *ends)
517 bufp->regs_allocated = REGS_REALLOCATE;
518 regs->num_regs = num_regs;
519 regs->start = starts;
524 bufp->regs_allocated = REGS_UNALLOCATED;
526 regs->start = regs->end = (regoff_t *) 0;
530 /* Entry points compatible with 4.2 BSD regex library. We don't define
531 them unless specifically requested. */
533 #if defined _REGEX_RE_COMP || defined __UCLIBC__
536 re_exec (const char *s)
538 return 0 == regexec (re_comp_buf, s, 0, NULL, 0);
542 /* Internal entry point. */
544 /* Searches for a compiled pattern PREG in the string STRING, whose
545 length is LENGTH. NMATCH, PMATCH, and EFLAGS have the same
546 mingings with regexec. START, and RANGE have the same meanings
548 Return REG_NOERROR if we find a match, and REG_NOMATCH if not,
549 otherwise return the error code.
550 Note: We assume front end functions already check ranges.
551 (START + RANGE >= 0 && START + RANGE <= LENGTH) */
552 static reg_errcode_t internal_function
553 re_search_internal (const regex_t *preg, const char *string, int length,
554 int start, int range, int stop, size_t nmatch,
555 regmatch_t pmatch[], int eflags)
558 const re_dfa_t *dfa = (const re_dfa_t *) preg->buffer;
559 int left_lim, right_lim, incr;
560 int fl_longest_match, match_first, match_kind, match_last = -1;
563 re_match_context_t mctx;
564 char *fastmap = (preg->fastmap != NULL && preg->fastmap_accurate
565 && range && !preg->can_be_null) ? preg->fastmap : NULL;
566 __RE_TRANSLATE_TYPE t = preg->translate;
568 memset (&mctx, '\0', sizeof (re_match_context_t));
571 extra_nmatch = (nmatch > preg->re_nsub) ? nmatch - (preg->re_nsub + 1) : 0;
572 nmatch -= extra_nmatch;
574 /* Check if the DFA haven't been compiled. */
575 if (BE (preg->used == 0 || dfa->init_state == NULL
576 || dfa->init_state_word == NULL || dfa->init_state_nl == NULL
577 || dfa->init_state_begbuf == NULL, 0))
581 /* We assume front-end functions already check them. */
582 assert (start + range >= 0 && start + range <= length);
585 /* If initial states with non-begbuf contexts have no elements,
586 the regex must be anchored. If preg->newline_anchor is set,
587 we'll never use init_state_nl, so do not check it. */
588 if (dfa->init_state->nodes.nelem == 0
589 && dfa->init_state_word->nodes.nelem == 0
590 && (dfa->init_state_nl->nodes.nelem == 0
591 || !preg->newline_anchor))
593 if (start != 0 && start + range != 0)
598 /* We must check the longest matching, if nmatch > 0. */
599 fl_longest_match = (nmatch != 0 || dfa->nbackref);
601 err = re_string_allocate (&mctx.input, string, length, dfa->nodes_len + 1,
602 preg->translate, preg->syntax & RE_ICASE, dfa);
603 if (BE (err != REG_NOERROR, 0))
605 mctx.input.stop = stop;
606 mctx.input.raw_stop = stop;
607 mctx.input.newline_anchor = preg->newline_anchor;
609 err = match_ctx_init (&mctx, eflags, dfa->nbackref * 2);
610 if (BE (err != REG_NOERROR, 0))
613 /* We will log all the DFA states through which the dfa pass,
614 if nmatch > 1, or this dfa has "multibyte node", which is a
615 back-reference or a node which can accept multibyte character or
616 multi character collating element. */
617 if (nmatch > 1 || dfa->has_mb_node)
619 mctx.state_log = re_malloc (re_dfastate_t *, mctx.input.bufs_len + 1);
620 if (BE (mctx.state_log == NULL, 0))
627 mctx.state_log = NULL;
630 mctx.input.tip_context = (eflags & REG_NOTBOL) ? CONTEXT_BEGBUF
631 : CONTEXT_NEWLINE | CONTEXT_BEGBUF;
633 /* Check incrementally whether of not the input string match. */
634 incr = (range < 0) ? -1 : 1;
635 left_lim = (range < 0) ? start + range : start;
636 right_lim = (range < 0) ? start : start + range;
637 sb = dfa->mb_cur_max == 1;
640 ? ((sb || !(preg->syntax & RE_ICASE || t) ? 4 : 0)
641 | (range >= 0 ? 2 : 0)
642 | (t != NULL ? 1 : 0))
645 for (;; match_first += incr)
648 if (match_first < left_lim || right_lim < match_first)
651 /* Advance as rapidly as possible through the string, until we
652 find a plausible place to start matching. This may be done
653 with varying efficiency, so there are various possibilities:
654 only the most common of them are specialized, in order to
655 save on code size. We use a switch statement for speed. */
663 /* Fastmap with single-byte translation, match forward. */
664 while (BE (match_first < right_lim, 1)
665 && !fastmap[t[(unsigned char) string[match_first]]])
667 goto forward_match_found_start_or_reached_end;
670 /* Fastmap without translation, match forward. */
671 while (BE (match_first < right_lim, 1)
672 && !fastmap[(unsigned char) string[match_first]])
675 forward_match_found_start_or_reached_end:
676 if (BE (match_first == right_lim, 0))
678 ch = match_first >= length
679 ? 0 : (unsigned char) string[match_first];
680 if (!fastmap[t ? t[ch] : ch])
687 /* Fastmap without multi-byte translation, match backwards. */
688 while (match_first >= left_lim)
690 ch = match_first >= length
691 ? 0 : (unsigned char) string[match_first];
692 if (fastmap[t ? t[ch] : ch])
696 if (match_first < left_lim)
701 /* In this case, we can't determine easily the current byte,
702 since it might be a component byte of a multibyte
703 character. Then we use the constructed buffer instead. */
706 /* If MATCH_FIRST is out of the valid range, reconstruct the
708 unsigned int offset = match_first - mctx.input.raw_mbs_idx;
709 if (BE (offset >= (unsigned int) mctx.input.valid_raw_len, 0))
711 err = re_string_reconstruct (&mctx.input, match_first,
713 if (BE (err != REG_NOERROR, 0))
716 offset = match_first - mctx.input.raw_mbs_idx;
718 /* If MATCH_FIRST is out of the buffer, leave it as '\0'.
719 Note that MATCH_FIRST must not be smaller than 0. */
720 ch = (match_first >= length
721 ? 0 : re_string_byte_at (&mctx.input, offset));
725 if (match_first < left_lim || match_first > right_lim)
734 /* Reconstruct the buffers so that the matcher can assume that
735 the matching starts from the beginning of the buffer. */
736 err = re_string_reconstruct (&mctx.input, match_first, eflags);
737 if (BE (err != REG_NOERROR, 0))
740 #ifdef RE_ENABLE_I18N
741 /* Don't consider this char as a possible match start if it part,
742 yet isn't the head, of a multibyte character. */
743 if (!sb && !re_string_first_byte (&mctx.input, 0))
747 /* It seems to be appropriate one, then use the matcher. */
748 /* We assume that the matching starts from 0. */
749 mctx.state_log_top = mctx.nbkref_ents = mctx.max_mb_elem_len = 0;
750 match_last = check_matching (&mctx, fl_longest_match,
751 range >= 0 ? &match_first : NULL);
752 if (match_last != -1)
754 if (BE (match_last == -2, 0))
761 mctx.match_last = match_last;
762 if ((!preg->no_sub && nmatch > 1) || dfa->nbackref)
764 re_dfastate_t *pstate = mctx.state_log[match_last];
765 mctx.last_node = check_halt_state_context (&mctx, pstate,
768 if ((!preg->no_sub && nmatch > 1 && dfa->has_plural_match)
771 err = prune_impossible_nodes (&mctx);
772 if (err == REG_NOERROR)
774 if (BE (err != REG_NOMATCH, 0))
779 break; /* We found a match. */
783 match_ctx_clean (&mctx);
787 assert (match_last != -1);
788 assert (err == REG_NOERROR);
791 /* Set pmatch[] if we need. */
796 /* Initialize registers. */
797 for (reg_idx = 1; reg_idx < nmatch; ++reg_idx)
798 pmatch[reg_idx].rm_so = pmatch[reg_idx].rm_eo = -1;
800 /* Set the points where matching start/end. */
802 pmatch[0].rm_eo = mctx.match_last;
804 if (!preg->no_sub && nmatch > 1)
806 err = set_regs (preg, &mctx, nmatch, pmatch,
807 dfa->has_plural_match && dfa->nbackref > 0);
808 if (BE (err != REG_NOERROR, 0))
812 /* At last, add the offset to the each registers, since we slided
813 the buffers so that we could assume that the matching starts
815 for (reg_idx = 0; reg_idx < nmatch; ++reg_idx)
816 if (pmatch[reg_idx].rm_so != -1)
818 #ifdef RE_ENABLE_I18N
819 if (BE (mctx.input.offsets_needed != 0, 0))
821 pmatch[reg_idx].rm_so =
822 (pmatch[reg_idx].rm_so == mctx.input.valid_len
823 ? mctx.input.valid_raw_len
824 : mctx.input.offsets[pmatch[reg_idx].rm_so]);
825 pmatch[reg_idx].rm_eo =
826 (pmatch[reg_idx].rm_eo == mctx.input.valid_len
827 ? mctx.input.valid_raw_len
828 : mctx.input.offsets[pmatch[reg_idx].rm_eo]);
831 assert (mctx.input.offsets_needed == 0);
833 pmatch[reg_idx].rm_so += match_first;
834 pmatch[reg_idx].rm_eo += match_first;
836 for (reg_idx = 0; reg_idx < extra_nmatch; ++reg_idx)
838 pmatch[nmatch + reg_idx].rm_so = -1;
839 pmatch[nmatch + reg_idx].rm_eo = -1;
843 for (reg_idx = 0; reg_idx + 1 < nmatch; reg_idx++)
844 if (dfa->subexp_map[reg_idx] != reg_idx)
846 pmatch[reg_idx + 1].rm_so
847 = pmatch[dfa->subexp_map[reg_idx] + 1].rm_so;
848 pmatch[reg_idx + 1].rm_eo
849 = pmatch[dfa->subexp_map[reg_idx] + 1].rm_eo;
854 re_free (mctx.state_log);
856 match_ctx_free (&mctx);
857 re_string_destruct (&mctx.input);
861 static reg_errcode_t internal_function
862 prune_impossible_nodes (re_match_context_t *mctx)
864 const re_dfa_t *const dfa = mctx->dfa;
865 int halt_node, match_last;
867 re_dfastate_t **sifted_states;
868 re_dfastate_t **lim_states = NULL;
869 re_sift_context_t sctx;
871 assert (mctx->state_log != NULL);
873 match_last = mctx->match_last;
874 halt_node = mctx->last_node;
875 sifted_states = re_malloc (re_dfastate_t *, match_last + 1);
876 if (BE (sifted_states == NULL, 0))
883 lim_states = re_malloc (re_dfastate_t *, match_last + 1);
884 if (BE (lim_states == NULL, 0))
891 memset (lim_states, '\0',
892 sizeof (re_dfastate_t *) * (match_last + 1));
893 sift_ctx_init (&sctx, sifted_states, lim_states, halt_node,
895 ret = sift_states_backward (mctx, &sctx);
896 re_node_set_free (&sctx.limits);
897 if (BE (ret != REG_NOERROR, 0))
899 if (sifted_states[0] != NULL || lim_states[0] != NULL)
909 } while (mctx->state_log[match_last] == NULL
910 || !mctx->state_log[match_last]->halt);
911 halt_node = check_halt_state_context (mctx,
912 mctx->state_log[match_last],
915 ret = merge_state_array (dfa, sifted_states, lim_states,
917 re_free (lim_states);
919 if (BE (ret != REG_NOERROR, 0))
924 sift_ctx_init (&sctx, sifted_states, lim_states, halt_node, match_last);
925 ret = sift_states_backward (mctx, &sctx);
926 re_node_set_free (&sctx.limits);
927 if (BE (ret != REG_NOERROR, 0))
930 re_free (mctx->state_log);
931 mctx->state_log = sifted_states;
932 sifted_states = NULL;
933 mctx->last_node = halt_node;
934 mctx->match_last = match_last;
937 re_free (sifted_states);
938 re_free (lim_states);
942 /* Acquire an initial state and return it.
943 We must select appropriate initial state depending on the context,
944 since initial states may have constraints like "\<", "^", etc.. */
946 static __inline__ re_dfastate_t *
947 __attribute ((always_inline)) internal_function
948 acquire_init_state_context (reg_errcode_t *err, const re_match_context_t *mctx,
951 const re_dfa_t *const dfa = mctx->dfa;
952 if (dfa->init_state->has_constraint)
954 unsigned int context;
955 context = re_string_context_at (&mctx->input, idx - 1, mctx->eflags);
956 if (IS_WORD_CONTEXT (context))
957 return dfa->init_state_word;
958 else if (IS_ORDINARY_CONTEXT (context))
959 return dfa->init_state;
960 else if (IS_BEGBUF_CONTEXT (context) && IS_NEWLINE_CONTEXT (context))
961 return dfa->init_state_begbuf;
962 else if (IS_NEWLINE_CONTEXT (context))
963 return dfa->init_state_nl;
964 else if (IS_BEGBUF_CONTEXT (context))
966 /* It is relatively rare case, then calculate on demand. */
967 return re_acquire_state_context (err, dfa,
968 dfa->init_state->entrance_nodes,
972 /* Must not happen? */
973 return dfa->init_state;
976 return dfa->init_state;
979 /* Check whether the regular expression match input string INPUT or not,
980 and return the index where the matching end, return -1 if not match,
981 or return -2 in case of an error.
982 FL_LONGEST_MATCH means we want the POSIX longest matching.
983 If P_MATCH_FIRST is not NULL, and the match fails, it is set to the
984 next place where we may want to try matching.
985 Note that the matcher assume that the maching starts from the current
986 index of the buffer. */
990 check_matching (re_match_context_t *mctx, int fl_longest_match,
993 const re_dfa_t *const dfa = mctx->dfa;
997 int cur_str_idx = re_string_cur_idx (&mctx->input);
998 re_dfastate_t *cur_state;
999 int at_init_state = p_match_first != NULL;
1000 int next_start_idx = cur_str_idx;
1003 cur_state = acquire_init_state_context (&err, mctx, cur_str_idx);
1004 /* An initial state must not be NULL (invalid). */
1005 if (BE (cur_state == NULL, 0))
1007 assert (err == REG_ESPACE);
1011 if (mctx->state_log != NULL)
1013 mctx->state_log[cur_str_idx] = cur_state;
1015 /* Check OP_OPEN_SUBEXP in the initial state in case that we use them
1016 later. E.g. Processing back references. */
1017 if (BE (dfa->nbackref, 0))
1020 err = check_subexp_matching_top (mctx, &cur_state->nodes, 0);
1021 if (BE (err != REG_NOERROR, 0))
1024 if (cur_state->has_backref)
1026 err = transit_state_bkref (mctx, &cur_state->nodes);
1027 if (BE (err != REG_NOERROR, 0))
1033 /* If the RE accepts NULL string. */
1034 if (BE (cur_state->halt, 0))
1036 if (!cur_state->has_constraint
1037 || check_halt_state_context (mctx, cur_state, cur_str_idx))
1039 if (!fl_longest_match)
1043 match_last = cur_str_idx;
1049 while (!re_string_eoi (&mctx->input))
1051 re_dfastate_t *old_state = cur_state;
1052 int next_char_idx = re_string_cur_idx (&mctx->input) + 1;
1054 if (BE (next_char_idx >= mctx->input.bufs_len, 0)
1055 || (BE (next_char_idx >= mctx->input.valid_len, 0)
1056 && mctx->input.valid_len < mctx->input.len))
1058 err = extend_buffers (mctx);
1059 if (BE (err != REG_NOERROR, 0))
1061 assert (err == REG_ESPACE);
1066 cur_state = transit_state (&err, mctx, cur_state);
1067 if (mctx->state_log != NULL)
1068 cur_state = merge_state_with_log (&err, mctx, cur_state);
1070 if (cur_state == NULL)
1072 /* Reached the invalid state or an error. Try to recover a valid
1073 state using the state log, if available and if we have not
1074 already found a valid (even if not the longest) match. */
1075 if (BE (err != REG_NOERROR, 0))
1078 if (mctx->state_log == NULL
1079 || (match && !fl_longest_match)
1080 || (cur_state = find_recover_state (&err, mctx)) == NULL)
1084 if (BE (at_init_state, 0))
1086 if (old_state == cur_state)
1087 next_start_idx = next_char_idx;
1092 if (cur_state->halt)
1094 /* Reached a halt state.
1095 Check the halt state can satisfy the current context. */
1096 if (!cur_state->has_constraint
1097 || check_halt_state_context (mctx, cur_state,
1098 re_string_cur_idx (&mctx->input)))
1100 /* We found an appropriate halt state. */
1101 match_last = re_string_cur_idx (&mctx->input);
1104 /* We found a match, do not modify match_first below. */
1105 p_match_first = NULL;
1106 if (!fl_longest_match)
1113 *p_match_first += next_start_idx;
1118 /* Check NODE match the current context. */
1122 check_halt_node_context (const re_dfa_t *dfa, int node, unsigned int context)
1124 re_token_type_t type = dfa->nodes[node].type;
1125 unsigned int constraint = dfa->nodes[node].constraint;
1126 if (type != END_OF_RE)
1130 if (NOT_SATISFY_NEXT_CONSTRAINT (constraint, context))
1135 /* Check the halt state STATE match the current context.
1136 Return 0 if not match, if the node, STATE has, is a halt node and
1137 match the context, return the node. */
1141 check_halt_state_context (const re_match_context_t *mctx,
1142 const re_dfastate_t *state, int idx)
1145 unsigned int context;
1147 assert (state->halt);
1149 context = re_string_context_at (&mctx->input, idx, mctx->eflags);
1150 for (i = 0; i < state->nodes.nelem; ++i)
1151 if (check_halt_node_context (mctx->dfa, state->nodes.elems[i], context))
1152 return state->nodes.elems[i];
1156 /* Compute the next node to which "NFA" transit from NODE("NFA" is a NFA
1157 corresponding to the DFA).
1158 Return the destination node, and update EPS_VIA_NODES, return -1 in case
1163 proceed_next_node (const re_match_context_t *mctx, int nregs, regmatch_t *regs,
1164 int *pidx, int node, re_node_set *eps_via_nodes,
1165 struct re_fail_stack_t *fs)
1167 const re_dfa_t *const dfa = mctx->dfa;
1169 if (IS_EPSILON_NODE (dfa->nodes[node].type))
1171 re_node_set *cur_nodes = &mctx->state_log[*pidx]->nodes;
1172 re_node_set *edests = &dfa->edests[node];
1174 err = re_node_set_insert (eps_via_nodes, node);
1175 if (BE (err < 0, 0))
1177 /* Pick up a valid destination, or return -1 if none is found. */
1178 for (dest_node = -1, i = 0; i < edests->nelem; ++i)
1180 int candidate = edests->elems[i];
1181 if (!re_node_set_contains (cur_nodes, candidate))
1183 if (dest_node == -1)
1184 dest_node = candidate;
1188 /* In order to avoid infinite loop like "(a*)*", return the second
1189 epsilon-transition if the first was already considered. */
1190 if (re_node_set_contains (eps_via_nodes, dest_node))
1193 /* Otherwise, push the second epsilon-transition on the fail stack. */
1195 && push_fail_stack (fs, *pidx, candidate, nregs, regs,
1199 /* We know we are going to exit. */
1208 re_token_type_t type = dfa->nodes[node].type;
1210 #ifdef RE_ENABLE_I18N
1211 if (dfa->nodes[node].accept_mb)
1212 naccepted = check_node_accept_bytes (dfa, node, &mctx->input, *pidx);
1214 #endif /* RE_ENABLE_I18N */
1215 if (type == OP_BACK_REF)
1217 int subexp_idx = dfa->nodes[node].opr.idx + 1;
1218 naccepted = regs[subexp_idx].rm_eo - regs[subexp_idx].rm_so;
1221 if (regs[subexp_idx].rm_so == -1 || regs[subexp_idx].rm_eo == -1)
1225 char *buf = (char *) re_string_get_buffer (&mctx->input);
1226 if (memcmp (buf + regs[subexp_idx].rm_so, buf + *pidx,
1235 err = re_node_set_insert (eps_via_nodes, node);
1236 if (BE (err < 0, 0))
1238 dest_node = dfa->edests[node].elems[0];
1239 if (re_node_set_contains (&mctx->state_log[*pidx]->nodes,
1246 || check_node_accept (mctx, dfa->nodes + node, *pidx))
1248 int dest_node = dfa->nexts[node];
1249 *pidx = (naccepted == 0) ? *pidx + 1 : *pidx + naccepted;
1250 if (fs && (*pidx > mctx->match_last || mctx->state_log[*pidx] == NULL
1251 || !re_node_set_contains (&mctx->state_log[*pidx]->nodes,
1254 re_node_set_empty (eps_via_nodes);
1261 static reg_errcode_t
1263 push_fail_stack (struct re_fail_stack_t *fs, int str_idx, int dest_node,
1264 int nregs, regmatch_t *regs, re_node_set *eps_via_nodes)
1267 int num = fs->num++;
1268 if (fs->num == fs->alloc)
1270 struct re_fail_stack_ent_t *new_array;
1271 new_array = realloc (fs->stack, (sizeof (struct re_fail_stack_ent_t)
1273 if (new_array == NULL)
1276 fs->stack = new_array;
1278 fs->stack[num].idx = str_idx;
1279 fs->stack[num].node = dest_node;
1280 fs->stack[num].regs = re_malloc (regmatch_t, nregs);
1281 if (fs->stack[num].regs == NULL)
1283 memcpy (fs->stack[num].regs, regs, sizeof (regmatch_t) * nregs);
1284 err = re_node_set_init_copy (&fs->stack[num].eps_via_nodes, eps_via_nodes);
1290 pop_fail_stack (struct re_fail_stack_t *fs, int *pidx, int nregs,
1291 regmatch_t *regs, re_node_set *eps_via_nodes)
1293 int num = --fs->num;
1295 *pidx = fs->stack[num].idx;
1296 memcpy (regs, fs->stack[num].regs, sizeof (regmatch_t) * nregs);
1297 re_node_set_free (eps_via_nodes);
1298 re_free (fs->stack[num].regs);
1299 *eps_via_nodes = fs->stack[num].eps_via_nodes;
1300 return fs->stack[num].node;
1303 /* Set the positions where the subexpressions are starts/ends to registers
1305 Note: We assume that pmatch[0] is already set, and
1306 pmatch[i].rm_so == pmatch[i].rm_eo == -1 for 0 < i < nmatch. */
1308 static reg_errcode_t
1310 set_regs (const regex_t *preg, const re_match_context_t *mctx, size_t nmatch,
1311 regmatch_t *pmatch, int fl_backtrack)
1313 const re_dfa_t *dfa = (const re_dfa_t *) preg->buffer;
1315 re_node_set eps_via_nodes;
1316 struct re_fail_stack_t *fs;
1317 struct re_fail_stack_t fs_body = { 0, 2, NULL };
1318 regmatch_t *prev_idx_match;
1319 int prev_idx_match_malloced = 0;
1322 assert (nmatch > 1);
1323 assert (mctx->state_log != NULL);
1328 fs->stack = re_malloc (struct re_fail_stack_ent_t, fs->alloc);
1329 if (fs->stack == NULL)
1335 cur_node = dfa->init_node;
1336 re_node_set_init_empty (&eps_via_nodes);
1338 if (__libc_use_alloca (nmatch * sizeof (regmatch_t)))
1339 prev_idx_match = (regmatch_t *) alloca (nmatch * sizeof (regmatch_t));
1342 prev_idx_match = re_malloc (regmatch_t, nmatch);
1343 if (prev_idx_match == NULL)
1345 free_fail_stack_return (fs);
1348 prev_idx_match_malloced = 1;
1350 memcpy (prev_idx_match, pmatch, sizeof (regmatch_t) * nmatch);
1352 for (idx = pmatch[0].rm_so; idx <= pmatch[0].rm_eo ;)
1354 update_regs (dfa, pmatch, prev_idx_match, cur_node, idx, nmatch);
1356 if (idx == pmatch[0].rm_eo && cur_node == mctx->last_node)
1361 for (reg_idx = 0; reg_idx < nmatch; ++reg_idx)
1362 if (pmatch[reg_idx].rm_so > -1 && pmatch[reg_idx].rm_eo == -1)
1364 if (reg_idx == nmatch)
1366 re_node_set_free (&eps_via_nodes);
1367 if (prev_idx_match_malloced)
1368 re_free (prev_idx_match);
1369 return free_fail_stack_return (fs);
1371 cur_node = pop_fail_stack (fs, &idx, nmatch, pmatch,
1376 re_node_set_free (&eps_via_nodes);
1377 if (prev_idx_match_malloced)
1378 re_free (prev_idx_match);
1383 /* Proceed to next node. */
1384 cur_node = proceed_next_node (mctx, nmatch, pmatch, &idx, cur_node,
1385 &eps_via_nodes, fs);
1387 if (BE (cur_node < 0, 0))
1389 if (BE (cur_node == -2, 0))
1391 re_node_set_free (&eps_via_nodes);
1392 if (prev_idx_match_malloced)
1393 re_free (prev_idx_match);
1394 free_fail_stack_return (fs);
1398 cur_node = pop_fail_stack (fs, &idx, nmatch, pmatch,
1402 re_node_set_free (&eps_via_nodes);
1403 if (prev_idx_match_malloced)
1404 re_free (prev_idx_match);
1409 re_node_set_free (&eps_via_nodes);
1410 if (prev_idx_match_malloced)
1411 re_free (prev_idx_match);
1412 return free_fail_stack_return (fs);
1415 static reg_errcode_t
1417 free_fail_stack_return (struct re_fail_stack_t *fs)
1422 for (fs_idx = 0; fs_idx < fs->num; ++fs_idx)
1424 re_node_set_free (&fs->stack[fs_idx].eps_via_nodes);
1425 re_free (fs->stack[fs_idx].regs);
1427 re_free (fs->stack);
1434 update_regs (const re_dfa_t *dfa, regmatch_t *pmatch,
1435 regmatch_t *prev_idx_match, int cur_node, int cur_idx, int nmatch)
1437 int type = dfa->nodes[cur_node].type;
1438 if (type == OP_OPEN_SUBEXP)
1440 int reg_num = dfa->nodes[cur_node].opr.idx + 1;
1442 /* We are at the first node of this sub expression. */
1443 if (reg_num < nmatch)
1445 pmatch[reg_num].rm_so = cur_idx;
1446 pmatch[reg_num].rm_eo = -1;
1449 else if (type == OP_CLOSE_SUBEXP)
1451 int reg_num = dfa->nodes[cur_node].opr.idx + 1;
1452 if (reg_num < nmatch)
1454 /* We are at the last node of this sub expression. */
1455 if (pmatch[reg_num].rm_so < cur_idx)
1457 pmatch[reg_num].rm_eo = cur_idx;
1458 /* This is a non-empty match or we are not inside an optional
1459 subexpression. Accept this right away. */
1460 memcpy (prev_idx_match, pmatch, sizeof (regmatch_t) * nmatch);
1464 if (dfa->nodes[cur_node].opt_subexp
1465 && prev_idx_match[reg_num].rm_so != -1)
1466 /* We transited through an empty match for an optional
1467 subexpression, like (a?)*, and this is not the subexp's
1468 first match. Copy back the old content of the registers
1469 so that matches of an inner subexpression are undone as
1470 well, like in ((a?))*. */
1471 memcpy (pmatch, prev_idx_match, sizeof (regmatch_t) * nmatch);
1473 /* We completed a subexpression, but it may be part of
1474 an optional one, so do not update PREV_IDX_MATCH. */
1475 pmatch[reg_num].rm_eo = cur_idx;
1481 /* This function checks the STATE_LOG from the SCTX->last_str_idx to 0
1482 and sift the nodes in each states according to the following rules.
1483 Updated state_log will be wrote to STATE_LOG.
1485 Rules: We throw away the Node `a' in the STATE_LOG[STR_IDX] if...
1486 1. When STR_IDX == MATCH_LAST(the last index in the state_log):
1487 If `a' isn't the LAST_NODE and `a' can't epsilon transit to
1488 the LAST_NODE, we throw away the node `a'.
1489 2. When 0 <= STR_IDX < MATCH_LAST and `a' accepts
1490 string `s' and transit to `b':
1491 i. If 'b' isn't in the STATE_LOG[STR_IDX+strlen('s')], we throw
1493 ii. If 'b' is in the STATE_LOG[STR_IDX+strlen('s')] but 'b' is
1494 thrown away, we throw away the node `a'.
1495 3. When 0 <= STR_IDX < MATCH_LAST and 'a' epsilon transit to 'b':
1496 i. If 'b' isn't in the STATE_LOG[STR_IDX], we throw away the
1498 ii. If 'b' is in the STATE_LOG[STR_IDX] but 'b' is thrown away,
1499 we throw away the node `a'. */
1501 #define STATE_NODE_CONTAINS(state,node) \
1502 ((state) != NULL && re_node_set_contains (&(state)->nodes, node))
1504 static reg_errcode_t
1506 sift_states_backward (const re_match_context_t *mctx, re_sift_context_t *sctx)
1510 int str_idx = sctx->last_str_idx;
1511 re_node_set cur_dest;
1514 assert (mctx->state_log != NULL && mctx->state_log[str_idx] != NULL);
1517 /* Build sifted state_log[str_idx]. It has the nodes which can epsilon
1518 transit to the last_node and the last_node itself. */
1519 err = re_node_set_init_1 (&cur_dest, sctx->last_node);
1520 if (BE (err != REG_NOERROR, 0))
1522 err = update_cur_sifted_state (mctx, sctx, str_idx, &cur_dest);
1523 if (BE (err != REG_NOERROR, 0))
1526 /* Then check each states in the state_log. */
1529 /* Update counters. */
1530 null_cnt = (sctx->sifted_states[str_idx] == NULL) ? null_cnt + 1 : 0;
1531 if (null_cnt > mctx->max_mb_elem_len)
1533 memset (sctx->sifted_states, '\0',
1534 sizeof (re_dfastate_t *) * str_idx);
1535 re_node_set_free (&cur_dest);
1538 re_node_set_empty (&cur_dest);
1541 if (mctx->state_log[str_idx])
1543 err = build_sifted_states (mctx, sctx, str_idx, &cur_dest);
1544 if (BE (err != REG_NOERROR, 0))
1548 /* Add all the nodes which satisfy the following conditions:
1549 - It can epsilon transit to a node in CUR_DEST.
1551 And update state_log. */
1552 err = update_cur_sifted_state (mctx, sctx, str_idx, &cur_dest);
1553 if (BE (err != REG_NOERROR, 0))
1558 re_node_set_free (&cur_dest);
1562 static reg_errcode_t
1564 build_sifted_states (const re_match_context_t *mctx, re_sift_context_t *sctx,
1565 int str_idx, re_node_set *cur_dest)
1567 const re_dfa_t *const dfa = mctx->dfa;
1568 const re_node_set *cur_src = &mctx->state_log[str_idx]->non_eps_nodes;
1571 /* Then build the next sifted state.
1572 We build the next sifted state on `cur_dest', and update
1573 `sifted_states[str_idx]' with `cur_dest'.
1575 `cur_dest' is the sifted state from `state_log[str_idx + 1]'.
1576 `cur_src' points the node_set of the old `state_log[str_idx]'
1577 (with the epsilon nodes pre-filtered out). */
1578 for (i = 0; i < cur_src->nelem; i++)
1580 int prev_node = cur_src->elems[i];
1585 re_token_type_t type = dfa->nodes[prev_node].type;
1586 assert (!IS_EPSILON_NODE (type));
1588 #ifdef RE_ENABLE_I18N
1589 /* If the node may accept `multi byte'. */
1590 if (dfa->nodes[prev_node].accept_mb)
1591 naccepted = sift_states_iter_mb (mctx, sctx, prev_node,
1592 str_idx, sctx->last_str_idx);
1593 #endif /* RE_ENABLE_I18N */
1595 /* We don't check backreferences here.
1596 See update_cur_sifted_state(). */
1598 && check_node_accept (mctx, dfa->nodes + prev_node, str_idx)
1599 && STATE_NODE_CONTAINS (sctx->sifted_states[str_idx + 1],
1600 dfa->nexts[prev_node]))
1606 if (sctx->limits.nelem)
1608 int to_idx = str_idx + naccepted;
1609 if (check_dst_limits (mctx, &sctx->limits,
1610 dfa->nexts[prev_node], to_idx,
1611 prev_node, str_idx))
1614 ret = re_node_set_insert (cur_dest, prev_node);
1615 if (BE (ret == -1, 0))
1622 /* Helper functions. */
1624 static reg_errcode_t
1626 clean_state_log_if_needed (re_match_context_t *mctx, int next_state_log_idx)
1628 int top = mctx->state_log_top;
1630 if (next_state_log_idx >= mctx->input.bufs_len
1631 || (next_state_log_idx >= mctx->input.valid_len
1632 && mctx->input.valid_len < mctx->input.len))
1635 err = extend_buffers (mctx);
1636 if (BE (err != REG_NOERROR, 0))
1640 if (top < next_state_log_idx)
1642 memset (mctx->state_log + top + 1, '\0',
1643 sizeof (re_dfastate_t *) * (next_state_log_idx - top));
1644 mctx->state_log_top = next_state_log_idx;
1649 static reg_errcode_t
1651 merge_state_array (const re_dfa_t *dfa, re_dfastate_t **dst,
1652 re_dfastate_t **src, int num)
1656 for (st_idx = 0; st_idx < num; ++st_idx)
1658 if (dst[st_idx] == NULL)
1659 dst[st_idx] = src[st_idx];
1660 else if (src[st_idx] != NULL)
1662 re_node_set merged_set;
1663 err = re_node_set_init_union (&merged_set, &dst[st_idx]->nodes,
1664 &src[st_idx]->nodes);
1665 if (BE (err != REG_NOERROR, 0))
1667 dst[st_idx] = re_acquire_state (&err, dfa, &merged_set);
1668 re_node_set_free (&merged_set);
1669 if (BE (err != REG_NOERROR, 0))
1676 static reg_errcode_t
1678 update_cur_sifted_state (const re_match_context_t *mctx,
1679 re_sift_context_t *sctx, int str_idx,
1680 re_node_set *dest_nodes)
1682 const re_dfa_t *const dfa = mctx->dfa;
1683 reg_errcode_t err = REG_NOERROR;
1684 const re_node_set *candidates;
1685 candidates = ((mctx->state_log[str_idx] == NULL) ? NULL
1686 : &mctx->state_log[str_idx]->nodes);
1688 if (dest_nodes->nelem == 0)
1689 sctx->sifted_states[str_idx] = NULL;
1694 /* At first, add the nodes which can epsilon transit to a node in
1696 err = add_epsilon_src_nodes (dfa, dest_nodes, candidates);
1697 if (BE (err != REG_NOERROR, 0))
1700 /* Then, check the limitations in the current sift_context. */
1701 if (sctx->limits.nelem)
1703 err = check_subexp_limits (dfa, dest_nodes, candidates, &sctx->limits,
1704 mctx->bkref_ents, str_idx);
1705 if (BE (err != REG_NOERROR, 0))
1710 sctx->sifted_states[str_idx] = re_acquire_state (&err, dfa, dest_nodes);
1711 if (BE (err != REG_NOERROR, 0))
1715 if (candidates && mctx->state_log[str_idx]->has_backref)
1717 err = sift_states_bkref (mctx, sctx, str_idx, candidates);
1718 if (BE (err != REG_NOERROR, 0))
1724 static reg_errcode_t
1726 add_epsilon_src_nodes (const re_dfa_t *dfa, re_node_set *dest_nodes,
1727 const re_node_set *candidates)
1729 reg_errcode_t err = REG_NOERROR;
1732 re_dfastate_t *state = re_acquire_state (&err, dfa, dest_nodes);
1733 if (BE (err != REG_NOERROR, 0))
1736 if (!state->inveclosure.alloc)
1738 err = re_node_set_alloc (&state->inveclosure, dest_nodes->nelem);
1739 if (BE (err != REG_NOERROR, 0))
1741 for (i = 0; i < dest_nodes->nelem; i++)
1742 re_node_set_merge (&state->inveclosure,
1743 dfa->inveclosures + dest_nodes->elems[i]);
1745 return re_node_set_add_intersect (dest_nodes, candidates,
1746 &state->inveclosure);
1749 static reg_errcode_t
1751 sub_epsilon_src_nodes (const re_dfa_t *dfa, int node, re_node_set *dest_nodes,
1752 const re_node_set *candidates)
1756 re_node_set *inv_eclosure = dfa->inveclosures + node;
1757 re_node_set except_nodes;
1758 re_node_set_init_empty (&except_nodes);
1759 for (ecl_idx = 0; ecl_idx < inv_eclosure->nelem; ++ecl_idx)
1761 int cur_node = inv_eclosure->elems[ecl_idx];
1762 if (cur_node == node)
1764 if (IS_EPSILON_NODE (dfa->nodes[cur_node].type))
1766 int edst1 = dfa->edests[cur_node].elems[0];
1767 int edst2 = ((dfa->edests[cur_node].nelem > 1)
1768 ? dfa->edests[cur_node].elems[1] : -1);
1769 if ((!re_node_set_contains (inv_eclosure, edst1)
1770 && re_node_set_contains (dest_nodes, edst1))
1772 && !re_node_set_contains (inv_eclosure, edst2)
1773 && re_node_set_contains (dest_nodes, edst2)))
1775 err = re_node_set_add_intersect (&except_nodes, candidates,
1776 dfa->inveclosures + cur_node);
1777 if (BE (err != REG_NOERROR, 0))
1779 re_node_set_free (&except_nodes);
1785 for (ecl_idx = 0; ecl_idx < inv_eclosure->nelem; ++ecl_idx)
1787 int cur_node = inv_eclosure->elems[ecl_idx];
1788 if (!re_node_set_contains (&except_nodes, cur_node))
1790 int idx = re_node_set_contains (dest_nodes, cur_node) - 1;
1791 re_node_set_remove_at (dest_nodes, idx);
1794 re_node_set_free (&except_nodes);
1800 check_dst_limits (const re_match_context_t *mctx, re_node_set *limits,
1801 int dst_node, int dst_idx, int src_node, int src_idx)
1803 const re_dfa_t *const dfa = mctx->dfa;
1804 int lim_idx, src_pos, dst_pos;
1806 int dst_bkref_idx = search_cur_bkref_entry (mctx, dst_idx);
1807 int src_bkref_idx = search_cur_bkref_entry (mctx, src_idx);
1808 for (lim_idx = 0; lim_idx < limits->nelem; ++lim_idx)
1811 struct re_backref_cache_entry *ent;
1812 ent = mctx->bkref_ents + limits->elems[lim_idx];
1813 subexp_idx = dfa->nodes[ent->node].opr.idx;
1815 dst_pos = check_dst_limits_calc_pos (mctx, limits->elems[lim_idx],
1816 subexp_idx, dst_node, dst_idx,
1818 src_pos = check_dst_limits_calc_pos (mctx, limits->elems[lim_idx],
1819 subexp_idx, src_node, src_idx,
1823 <src> <dst> ( <subexp> )
1824 ( <subexp> ) <src> <dst>
1825 ( <subexp1> <src> <subexp2> <dst> <subexp3> ) */
1826 if (src_pos == dst_pos)
1827 continue; /* This is unrelated limitation. */
1836 check_dst_limits_calc_pos_1 (const re_match_context_t *mctx, int boundaries,
1837 int subexp_idx, int from_node, int bkref_idx)
1839 const re_dfa_t *const dfa = mctx->dfa;
1840 const re_node_set *eclosures = dfa->eclosures + from_node;
1843 /* Else, we are on the boundary: examine the nodes on the epsilon
1845 for (node_idx = 0; node_idx < eclosures->nelem; ++node_idx)
1847 int node = eclosures->elems[node_idx];
1848 switch (dfa->nodes[node].type)
1851 if (bkref_idx != -1)
1853 struct re_backref_cache_entry *ent = mctx->bkref_ents + bkref_idx;
1858 if (ent->node != node)
1861 if (subexp_idx < BITSET_WORD_BITS
1862 && !(ent->eps_reachable_subexps_map
1863 & ((bitset_word_t) 1 << subexp_idx)))
1866 /* Recurse trying to reach the OP_OPEN_SUBEXP and
1867 OP_CLOSE_SUBEXP cases below. But, if the
1868 destination node is the same node as the source
1869 node, don't recurse because it would cause an
1870 infinite loop: a regex that exhibits this behavior
1872 dst = dfa->edests[node].elems[0];
1873 if (dst == from_node)
1877 else /* if (boundaries & 2) */
1882 check_dst_limits_calc_pos_1 (mctx, boundaries, subexp_idx,
1884 if (cpos == -1 /* && (boundaries & 1) */)
1886 if (cpos == 0 && (boundaries & 2))
1889 if (subexp_idx < BITSET_WORD_BITS)
1890 ent->eps_reachable_subexps_map
1891 &= ~((bitset_word_t) 1 << subexp_idx);
1893 while (ent++->more);
1897 case OP_OPEN_SUBEXP:
1898 if ((boundaries & 1) && subexp_idx == dfa->nodes[node].opr.idx)
1902 case OP_CLOSE_SUBEXP:
1903 if ((boundaries & 2) && subexp_idx == dfa->nodes[node].opr.idx)
1912 return (boundaries & 2) ? 1 : 0;
1917 check_dst_limits_calc_pos (const re_match_context_t *mctx, int limit,
1918 int subexp_idx, int from_node, int str_idx,
1921 struct re_backref_cache_entry *lim = mctx->bkref_ents + limit;
1924 /* If we are outside the range of the subexpression, return -1 or 1. */
1925 if (str_idx < lim->subexp_from)
1928 if (lim->subexp_to < str_idx)
1931 /* If we are within the subexpression, return 0. */
1932 boundaries = (str_idx == lim->subexp_from);
1933 boundaries |= (str_idx == lim->subexp_to) << 1;
1934 if (boundaries == 0)
1937 /* Else, examine epsilon closure. */
1938 return check_dst_limits_calc_pos_1 (mctx, boundaries, subexp_idx,
1939 from_node, bkref_idx);
1942 /* Check the limitations of sub expressions LIMITS, and remove the nodes
1943 which are against limitations from DEST_NODES. */
1945 static reg_errcode_t
1947 check_subexp_limits (const re_dfa_t *dfa, re_node_set *dest_nodes,
1948 const re_node_set *candidates, re_node_set *limits,
1949 struct re_backref_cache_entry *bkref_ents, int str_idx)
1952 int node_idx, lim_idx;
1954 for (lim_idx = 0; lim_idx < limits->nelem; ++lim_idx)
1957 struct re_backref_cache_entry *ent;
1958 ent = bkref_ents + limits->elems[lim_idx];
1960 if (str_idx <= ent->subexp_from || ent->str_idx < str_idx)
1961 continue; /* This is unrelated limitation. */
1963 subexp_idx = dfa->nodes[ent->node].opr.idx;
1964 if (ent->subexp_to == str_idx)
1968 for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx)
1970 int node = dest_nodes->elems[node_idx];
1971 re_token_type_t type = dfa->nodes[node].type;
1972 if (type == OP_OPEN_SUBEXP
1973 && subexp_idx == dfa->nodes[node].opr.idx)
1975 else if (type == OP_CLOSE_SUBEXP
1976 && subexp_idx == dfa->nodes[node].opr.idx)
1980 /* Check the limitation of the open subexpression. */
1981 /* Note that (ent->subexp_to = str_idx != ent->subexp_from). */
1984 err = sub_epsilon_src_nodes (dfa, ops_node, dest_nodes,
1986 if (BE (err != REG_NOERROR, 0))
1990 /* Check the limitation of the close subexpression. */
1992 for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx)
1994 int node = dest_nodes->elems[node_idx];
1995 if (!re_node_set_contains (dfa->inveclosures + node,
1997 && !re_node_set_contains (dfa->eclosures + node,
2000 /* It is against this limitation.
2001 Remove it form the current sifted state. */
2002 err = sub_epsilon_src_nodes (dfa, node, dest_nodes,
2004 if (BE (err != REG_NOERROR, 0))
2010 else /* (ent->subexp_to != str_idx) */
2012 for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx)
2014 int node = dest_nodes->elems[node_idx];
2015 re_token_type_t type = dfa->nodes[node].type;
2016 if (type == OP_CLOSE_SUBEXP || type == OP_OPEN_SUBEXP)
2018 if (subexp_idx != dfa->nodes[node].opr.idx)
2020 /* It is against this limitation.
2021 Remove it form the current sifted state. */
2022 err = sub_epsilon_src_nodes (dfa, node, dest_nodes,
2024 if (BE (err != REG_NOERROR, 0))
2033 static reg_errcode_t
2035 sift_states_bkref (const re_match_context_t *mctx, re_sift_context_t *sctx,
2036 int str_idx, const re_node_set *candidates)
2038 const re_dfa_t *const dfa = mctx->dfa;
2041 re_sift_context_t local_sctx;
2042 int first_idx = search_cur_bkref_entry (mctx, str_idx);
2044 if (first_idx == -1)
2047 local_sctx.sifted_states = NULL; /* Mark that it hasn't been initialized. */
2049 for (node_idx = 0; node_idx < candidates->nelem; ++node_idx)
2052 re_token_type_t type;
2053 struct re_backref_cache_entry *entry;
2054 node = candidates->elems[node_idx];
2055 type = dfa->nodes[node].type;
2056 /* Avoid infinite loop for the REs like "()\1+". */
2057 if (node == sctx->last_node && str_idx == sctx->last_str_idx)
2059 if (type != OP_BACK_REF)
2062 entry = mctx->bkref_ents + first_idx;
2063 enabled_idx = first_idx;
2070 re_dfastate_t *cur_state;
2072 if (entry->node != node)
2074 subexp_len = entry->subexp_to - entry->subexp_from;
2075 to_idx = str_idx + subexp_len;
2076 dst_node = (subexp_len ? dfa->nexts[node]
2077 : dfa->edests[node].elems[0]);
2079 if (to_idx > sctx->last_str_idx
2080 || sctx->sifted_states[to_idx] == NULL
2081 || !STATE_NODE_CONTAINS (sctx->sifted_states[to_idx], dst_node)
2082 || check_dst_limits (mctx, &sctx->limits, node,
2083 str_idx, dst_node, to_idx))
2086 if (local_sctx.sifted_states == NULL)
2089 err = re_node_set_init_copy (&local_sctx.limits, &sctx->limits);
2090 if (BE (err != REG_NOERROR, 0))
2093 local_sctx.last_node = node;
2094 local_sctx.last_str_idx = str_idx;
2095 ret = re_node_set_insert (&local_sctx.limits, enabled_idx);
2096 if (BE (ret < 0, 0))
2101 cur_state = local_sctx.sifted_states[str_idx];
2102 err = sift_states_backward (mctx, &local_sctx);
2103 if (BE (err != REG_NOERROR, 0))
2105 if (sctx->limited_states != NULL)
2107 err = merge_state_array (dfa, sctx->limited_states,
2108 local_sctx.sifted_states,
2110 if (BE (err != REG_NOERROR, 0))
2113 local_sctx.sifted_states[str_idx] = cur_state;
2114 re_node_set_remove (&local_sctx.limits, enabled_idx);
2116 /* mctx->bkref_ents may have changed, reload the pointer. */
2117 entry = mctx->bkref_ents + enabled_idx;
2119 while (enabled_idx++, entry++->more);
2123 if (local_sctx.sifted_states != NULL)
2125 re_node_set_free (&local_sctx.limits);
2132 #ifdef RE_ENABLE_I18N
2135 sift_states_iter_mb (const re_match_context_t *mctx, re_sift_context_t *sctx,
2136 int node_idx, int str_idx, int max_str_idx)
2138 const re_dfa_t *const dfa = mctx->dfa;
2140 /* Check the node can accept `multi byte'. */
2141 naccepted = check_node_accept_bytes (dfa, node_idx, &mctx->input, str_idx);
2142 if (naccepted > 0 && str_idx + naccepted <= max_str_idx &&
2143 !STATE_NODE_CONTAINS (sctx->sifted_states[str_idx + naccepted],
2144 dfa->nexts[node_idx]))
2145 /* The node can't accept the `multi byte', or the
2146 destination was already thrown away, then the node
2147 could't accept the current input `multi byte'. */
2149 /* Otherwise, it is sure that the node could accept
2150 `naccepted' bytes input. */
2153 #endif /* RE_ENABLE_I18N */
2156 /* Functions for state transition. */
2158 /* Return the next state to which the current state STATE will transit by
2159 accepting the current input byte, and update STATE_LOG if necessary.
2160 If STATE can accept a multibyte char/collating element/back reference
2161 update the destination of STATE_LOG. */
2163 static re_dfastate_t *
2165 transit_state (reg_errcode_t *err, re_match_context_t *mctx,
2166 re_dfastate_t *state)
2168 re_dfastate_t **trtable;
2171 #ifdef RE_ENABLE_I18N
2172 /* If the current state can accept multibyte. */
2173 if (BE (state->accept_mb, 0))
2175 *err = transit_state_mb (mctx, state);
2176 if (BE (*err != REG_NOERROR, 0))
2179 #endif /* RE_ENABLE_I18N */
2181 /* Then decide the next state with the single byte. */
2184 /* don't use transition table */
2185 return transit_state_sb (err, mctx, state);
2188 /* Use transition table */
2189 ch = re_string_fetch_byte (&mctx->input);
2192 trtable = state->trtable;
2193 if (BE (trtable != NULL, 1))
2196 trtable = state->word_trtable;
2197 if (BE (trtable != NULL, 1))
2199 unsigned int context;
2201 = re_string_context_at (&mctx->input,
2202 re_string_cur_idx (&mctx->input) - 1,
2204 if (IS_WORD_CONTEXT (context))
2205 return trtable[ch + SBC_MAX];
2210 if (!build_trtable (mctx->dfa, state))
2216 /* Retry, we now have a transition table. */
2220 /* Update the state_log if we need */
2223 merge_state_with_log (reg_errcode_t *err, re_match_context_t *mctx,
2224 re_dfastate_t *next_state)
2226 const re_dfa_t *const dfa = mctx->dfa;
2227 int cur_idx = re_string_cur_idx (&mctx->input);
2229 if (cur_idx > mctx->state_log_top)
2231 mctx->state_log[cur_idx] = next_state;
2232 mctx->state_log_top = cur_idx;
2234 else if (mctx->state_log[cur_idx] == 0)
2236 mctx->state_log[cur_idx] = next_state;
2240 re_dfastate_t *pstate;
2241 unsigned int context;
2242 re_node_set next_nodes, *log_nodes, *table_nodes = NULL;
2243 /* If (state_log[cur_idx] != 0), it implies that cur_idx is
2244 the destination of a multibyte char/collating element/
2245 back reference. Then the next state is the union set of
2246 these destinations and the results of the transition table. */
2247 pstate = mctx->state_log[cur_idx];
2248 log_nodes = pstate->entrance_nodes;
2249 if (next_state != NULL)
2251 table_nodes = next_state->entrance_nodes;
2252 *err = re_node_set_init_union (&next_nodes, table_nodes,
2254 if (BE (*err != REG_NOERROR, 0))
2258 next_nodes = *log_nodes;
2259 /* Note: We already add the nodes of the initial state,
2260 then we don't need to add them here. */
2262 context = re_string_context_at (&mctx->input,
2263 re_string_cur_idx (&mctx->input) - 1,
2265 next_state = mctx->state_log[cur_idx]
2266 = re_acquire_state_context (err, dfa, &next_nodes, context);
2267 /* We don't need to check errors here, since the return value of
2268 this function is next_state and ERR is already set. */
2270 if (table_nodes != NULL)
2271 re_node_set_free (&next_nodes);
2274 if (BE (dfa->nbackref, 0) && next_state != NULL)
2276 /* Check OP_OPEN_SUBEXP in the current state in case that we use them
2277 later. We must check them here, since the back references in the
2278 next state might use them. */
2279 *err = check_subexp_matching_top (mctx, &next_state->nodes,
2281 if (BE (*err != REG_NOERROR, 0))
2284 /* If the next state has back references. */
2285 if (next_state->has_backref)
2287 *err = transit_state_bkref (mctx, &next_state->nodes);
2288 if (BE (*err != REG_NOERROR, 0))
2290 next_state = mctx->state_log[cur_idx];
2297 /* Skip bytes in the input that correspond to part of a
2298 multi-byte match, then look in the log for a state
2299 from which to restart matching. */
2302 find_recover_state (reg_errcode_t *err, re_match_context_t *mctx)
2304 re_dfastate_t *cur_state;
2307 int max = mctx->state_log_top;
2308 int cur_str_idx = re_string_cur_idx (&mctx->input);
2312 if (++cur_str_idx > max)
2314 re_string_skip_bytes (&mctx->input, 1);
2316 while (mctx->state_log[cur_str_idx] == NULL);
2318 cur_state = merge_state_with_log (err, mctx, NULL);
2320 while (*err == REG_NOERROR && cur_state == NULL);
2324 /* Helper functions for transit_state. */
2326 /* From the node set CUR_NODES, pick up the nodes whose types are
2327 OP_OPEN_SUBEXP and which have corresponding back references in the regular
2328 expression. And register them to use them later for evaluating the
2329 correspoding back references. */
2331 static reg_errcode_t
2333 check_subexp_matching_top (re_match_context_t *mctx, re_node_set *cur_nodes,
2336 const re_dfa_t *const dfa = mctx->dfa;
2340 /* TODO: This isn't efficient.
2341 Because there might be more than one nodes whose types are
2342 OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all
2345 for (node_idx = 0; node_idx < cur_nodes->nelem; ++node_idx)
2347 int node = cur_nodes->elems[node_idx];
2348 if (dfa->nodes[node].type == OP_OPEN_SUBEXP
2349 && dfa->nodes[node].opr.idx < BITSET_WORD_BITS
2350 && (dfa->used_bkref_map
2351 & ((bitset_word_t) 1 << dfa->nodes[node].opr.idx)))
2353 err = match_ctx_add_subtop (mctx, node, str_idx);
2354 if (BE (err != REG_NOERROR, 0))
2362 /* Return the next state to which the current state STATE will transit by
2363 accepting the current input byte. */
2365 static re_dfastate_t *
2366 transit_state_sb (reg_errcode_t *err, re_match_context_t *mctx,
2367 re_dfastate_t *state)
2369 const re_dfa_t *const dfa = mctx->dfa;
2370 re_node_set next_nodes;
2371 re_dfastate_t *next_state;
2372 int node_cnt, cur_str_idx = re_string_cur_idx (&mctx->input);
2373 unsigned int context;
2375 *err = re_node_set_alloc (&next_nodes, state->nodes.nelem + 1);
2376 if (BE (*err != REG_NOERROR, 0))
2378 for (node_cnt = 0; node_cnt < state->nodes.nelem; ++node_cnt)
2380 int cur_node = state->nodes.elems[node_cnt];
2381 if (check_node_accept (mctx, dfa->nodes + cur_node, cur_str_idx))
2383 *err = re_node_set_merge (&next_nodes,
2384 dfa->eclosures + dfa->nexts[cur_node]);
2385 if (BE (*err != REG_NOERROR, 0))
2387 re_node_set_free (&next_nodes);
2392 context = re_string_context_at (&mctx->input, cur_str_idx, mctx->eflags);
2393 next_state = re_acquire_state_context (err, dfa, &next_nodes, context);
2394 /* We don't need to check errors here, since the return value of
2395 this function is next_state and ERR is already set. */
2397 re_node_set_free (&next_nodes);
2398 re_string_skip_bytes (&mctx->input, 1);
2403 #ifdef RE_ENABLE_I18N
2404 static reg_errcode_t
2406 transit_state_mb (re_match_context_t *mctx, re_dfastate_t *pstate)
2408 const re_dfa_t *const dfa = mctx->dfa;
2412 for (i = 0; i < pstate->nodes.nelem; ++i)
2414 re_node_set dest_nodes, *new_nodes;
2415 int cur_node_idx = pstate->nodes.elems[i];
2416 int naccepted, dest_idx;
2417 unsigned int context;
2418 re_dfastate_t *dest_state;
2420 if (!dfa->nodes[cur_node_idx].accept_mb)
2423 if (dfa->nodes[cur_node_idx].constraint)
2425 context = re_string_context_at (&mctx->input,
2426 re_string_cur_idx (&mctx->input),
2428 if (NOT_SATISFY_NEXT_CONSTRAINT (dfa->nodes[cur_node_idx].constraint,
2433 /* How many bytes the node can accept? */
2434 naccepted = check_node_accept_bytes (dfa, cur_node_idx, &mctx->input,
2435 re_string_cur_idx (&mctx->input));
2439 /* The node can accepts `naccepted' bytes. */
2440 dest_idx = re_string_cur_idx (&mctx->input) + naccepted;
2441 mctx->max_mb_elem_len = ((mctx->max_mb_elem_len < naccepted) ? naccepted
2442 : mctx->max_mb_elem_len);
2443 err = clean_state_log_if_needed (mctx, dest_idx);
2444 if (BE (err != REG_NOERROR, 0))
2447 assert (dfa->nexts[cur_node_idx] != -1);
2449 new_nodes = dfa->eclosures + dfa->nexts[cur_node_idx];
2451 dest_state = mctx->state_log[dest_idx];
2452 if (dest_state == NULL)
2453 dest_nodes = *new_nodes;
2456 err = re_node_set_init_union (&dest_nodes,
2457 dest_state->entrance_nodes, new_nodes);
2458 if (BE (err != REG_NOERROR, 0))
2461 context = re_string_context_at (&mctx->input, dest_idx - 1,
2463 mctx->state_log[dest_idx]
2464 = re_acquire_state_context (&err, dfa, &dest_nodes, context);
2465 if (dest_state != NULL)
2466 re_node_set_free (&dest_nodes);
2467 if (BE (mctx->state_log[dest_idx] == NULL && err != REG_NOERROR, 0))
2472 #endif /* RE_ENABLE_I18N */
2474 static reg_errcode_t
2476 transit_state_bkref (re_match_context_t *mctx, const re_node_set *nodes)
2478 const re_dfa_t *const dfa = mctx->dfa;
2481 int cur_str_idx = re_string_cur_idx (&mctx->input);
2483 for (i = 0; i < nodes->nelem; ++i)
2485 int dest_str_idx, prev_nelem, bkc_idx;
2486 int node_idx = nodes->elems[i];
2487 unsigned int context;
2488 const re_token_t *node = dfa->nodes + node_idx;
2489 re_node_set *new_dest_nodes;
2491 /* Check whether `node' is a backreference or not. */
2492 if (node->type != OP_BACK_REF)
2495 if (node->constraint)
2497 context = re_string_context_at (&mctx->input, cur_str_idx,
2499 if (NOT_SATISFY_NEXT_CONSTRAINT (node->constraint, context))
2503 /* `node' is a backreference.
2504 Check the substring which the substring matched. */
2505 bkc_idx = mctx->nbkref_ents;
2506 err = get_subexp (mctx, node_idx, cur_str_idx);
2507 if (BE (err != REG_NOERROR, 0))
2510 /* And add the epsilon closures (which is `new_dest_nodes') of
2511 the backreference to appropriate state_log. */
2513 assert (dfa->nexts[node_idx] != -1);
2515 for (; bkc_idx < mctx->nbkref_ents; ++bkc_idx)
2518 re_dfastate_t *dest_state;
2519 struct re_backref_cache_entry *bkref_ent;
2520 bkref_ent = mctx->bkref_ents + bkc_idx;
2521 if (bkref_ent->node != node_idx || bkref_ent->str_idx != cur_str_idx)
2523 subexp_len = bkref_ent->subexp_to - bkref_ent->subexp_from;
2524 new_dest_nodes = (subexp_len == 0
2525 ? dfa->eclosures + dfa->edests[node_idx].elems[0]
2526 : dfa->eclosures + dfa->nexts[node_idx]);
2527 dest_str_idx = (cur_str_idx + bkref_ent->subexp_to
2528 - bkref_ent->subexp_from);
2529 context = re_string_context_at (&mctx->input, dest_str_idx - 1,
2531 dest_state = mctx->state_log[dest_str_idx];
2532 prev_nelem = ((mctx->state_log[cur_str_idx] == NULL) ? 0
2533 : mctx->state_log[cur_str_idx]->nodes.nelem);
2534 /* Add `new_dest_node' to state_log. */
2535 if (dest_state == NULL)
2537 mctx->state_log[dest_str_idx]
2538 = re_acquire_state_context (&err, dfa, new_dest_nodes,
2540 if (BE (mctx->state_log[dest_str_idx] == NULL
2541 && err != REG_NOERROR, 0))
2546 re_node_set dest_nodes;
2547 err = re_node_set_init_union (&dest_nodes,
2548 dest_state->entrance_nodes,
2550 if (BE (err != REG_NOERROR, 0))
2552 re_node_set_free (&dest_nodes);
2555 mctx->state_log[dest_str_idx]
2556 = re_acquire_state_context (&err, dfa, &dest_nodes, context);
2557 re_node_set_free (&dest_nodes);
2558 if (BE (mctx->state_log[dest_str_idx] == NULL
2559 && err != REG_NOERROR, 0))
2562 /* We need to check recursively if the backreference can epsilon
2565 && mctx->state_log[cur_str_idx]->nodes.nelem > prev_nelem)
2567 err = check_subexp_matching_top (mctx, new_dest_nodes,
2569 if (BE (err != REG_NOERROR, 0))
2571 err = transit_state_bkref (mctx, new_dest_nodes);
2572 if (BE (err != REG_NOERROR, 0))
2582 /* Enumerate all the candidates which the backreference BKREF_NODE can match
2583 at BKREF_STR_IDX, and register them by match_ctx_add_entry().
2584 Note that we might collect inappropriate candidates here.
2585 However, the cost of checking them strictly here is too high, then we
2586 delay these checking for prune_impossible_nodes(). */
2588 static reg_errcode_t
2590 get_subexp (re_match_context_t *mctx, int bkref_node, int bkref_str_idx)
2592 const re_dfa_t *const dfa = mctx->dfa;
2593 int subexp_num, sub_top_idx;
2594 const char *buf = (const char *) re_string_get_buffer (&mctx->input);
2595 /* Return if we have already checked BKREF_NODE at BKREF_STR_IDX. */
2596 int cache_idx = search_cur_bkref_entry (mctx, bkref_str_idx);
2597 if (cache_idx != -1)
2599 const struct re_backref_cache_entry *entry
2600 = mctx->bkref_ents + cache_idx;
2602 if (entry->node == bkref_node)
2603 return REG_NOERROR; /* We already checked it. */
2604 while (entry++->more);
2607 subexp_num = dfa->nodes[bkref_node].opr.idx;
2609 /* For each sub expression */
2610 for (sub_top_idx = 0; sub_top_idx < mctx->nsub_tops; ++sub_top_idx)
2613 re_sub_match_top_t *sub_top = mctx->sub_tops[sub_top_idx];
2614 re_sub_match_last_t *sub_last;
2615 int sub_last_idx, sl_str, bkref_str_off;
2617 if (dfa->nodes[sub_top->node].opr.idx != subexp_num)
2618 continue; /* It isn't related. */
2620 sl_str = sub_top->str_idx;
2621 bkref_str_off = bkref_str_idx;
2622 /* At first, check the last node of sub expressions we already
2624 for (sub_last_idx = 0; sub_last_idx < sub_top->nlasts; ++sub_last_idx)
2627 sub_last = sub_top->lasts[sub_last_idx];
2628 sl_str_diff = sub_last->str_idx - sl_str;
2629 /* The matched string by the sub expression match with the substring
2630 at the back reference? */
2631 if (sl_str_diff > 0)
2633 if (BE (bkref_str_off + sl_str_diff > mctx->input.valid_len, 0))
2635 /* Not enough chars for a successful match. */
2636 if (bkref_str_off + sl_str_diff > mctx->input.len)
2639 err = clean_state_log_if_needed (mctx,
2642 if (BE (err != REG_NOERROR, 0))
2644 buf = (const char *) re_string_get_buffer (&mctx->input);
2646 if (memcmp (buf + bkref_str_off, buf + sl_str, sl_str_diff) != 0)
2647 /* We don't need to search this sub expression any more. */
2650 bkref_str_off += sl_str_diff;
2651 sl_str += sl_str_diff;
2652 err = get_subexp_sub (mctx, sub_top, sub_last, bkref_node,
2655 /* Reload buf, since the preceding call might have reallocated
2657 buf = (const char *) re_string_get_buffer (&mctx->input);
2659 if (err == REG_NOMATCH)
2661 if (BE (err != REG_NOERROR, 0))
2665 if (sub_last_idx < sub_top->nlasts)
2667 if (sub_last_idx > 0)
2669 /* Then, search for the other last nodes of the sub expression. */
2670 for (; sl_str <= bkref_str_idx; ++sl_str)
2672 int cls_node, sl_str_off;
2673 const re_node_set *nodes;
2674 sl_str_off = sl_str - sub_top->str_idx;
2675 /* The matched string by the sub expression match with the substring
2676 at the back reference? */
2679 if (BE (bkref_str_off >= mctx->input.valid_len, 0))
2681 /* If we are at the end of the input, we cannot match. */
2682 if (bkref_str_off >= mctx->input.len)
2685 err = extend_buffers (mctx);
2686 if (BE (err != REG_NOERROR, 0))
2689 buf = (const char *) re_string_get_buffer (&mctx->input);
2691 if (buf [bkref_str_off++] != buf[sl_str - 1])
2692 break; /* We don't need to search this sub expression
2695 if (mctx->state_log[sl_str] == NULL)
2697 /* Does this state have a ')' of the sub expression? */
2698 nodes = &mctx->state_log[sl_str]->nodes;
2699 cls_node = find_subexp_node (dfa, nodes, subexp_num,
2703 if (sub_top->path == NULL)
2705 sub_top->path = calloc (sizeof (state_array_t),
2706 sl_str - sub_top->str_idx + 1);
2707 if (sub_top->path == NULL)
2710 /* Can the OP_OPEN_SUBEXP node arrive the OP_CLOSE_SUBEXP node
2711 in the current context? */
2712 err = check_arrival (mctx, sub_top->path, sub_top->node,
2713 sub_top->str_idx, cls_node, sl_str,
2715 if (err == REG_NOMATCH)
2717 if (BE (err != REG_NOERROR, 0))
2719 sub_last = match_ctx_add_sublast (sub_top, cls_node, sl_str);
2720 if (BE (sub_last == NULL, 0))
2722 err = get_subexp_sub (mctx, sub_top, sub_last, bkref_node,
2724 if (err == REG_NOMATCH)
2731 /* Helper functions for get_subexp(). */
2733 /* Check SUB_LAST can arrive to the back reference BKREF_NODE at BKREF_STR.
2734 If it can arrive, register the sub expression expressed with SUB_TOP
2737 static reg_errcode_t
2739 get_subexp_sub (re_match_context_t *mctx, const re_sub_match_top_t *sub_top,
2740 re_sub_match_last_t *sub_last, int bkref_node, int bkref_str)
2744 /* Can the subexpression arrive the back reference? */
2745 err = check_arrival (mctx, &sub_last->path, sub_last->node,
2746 sub_last->str_idx, bkref_node, bkref_str,
2748 if (err != REG_NOERROR)
2750 err = match_ctx_add_entry (mctx, bkref_node, bkref_str, sub_top->str_idx,
2752 if (BE (err != REG_NOERROR, 0))
2754 to_idx = bkref_str + sub_last->str_idx - sub_top->str_idx;
2755 return clean_state_log_if_needed (mctx, to_idx);
2758 /* Find the first node which is '(' or ')' and whose index is SUBEXP_IDX.
2759 Search '(' if FL_OPEN, or search ')' otherwise.
2760 TODO: This function isn't efficient...
2761 Because there might be more than one nodes whose types are
2762 OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all
2768 find_subexp_node (const re_dfa_t *dfa, const re_node_set *nodes,
2769 int subexp_idx, int type)
2772 for (cls_idx = 0; cls_idx < nodes->nelem; ++cls_idx)
2774 int cls_node = nodes->elems[cls_idx];
2775 const re_token_t *node = dfa->nodes + cls_node;
2776 if (node->type == type
2777 && node->opr.idx == subexp_idx)
2783 /* Check whether the node TOP_NODE at TOP_STR can arrive to the node
2784 LAST_NODE at LAST_STR. We record the path onto PATH since it will be
2786 Return REG_NOERROR if it can arrive, or REG_NOMATCH otherwise. */
2788 static reg_errcode_t
2790 check_arrival (re_match_context_t *mctx, state_array_t *path, int top_node,
2791 int top_str, int last_node, int last_str, int type)
2793 const re_dfa_t *const dfa = mctx->dfa;
2794 reg_errcode_t err = REG_NOERROR;
2795 int subexp_num, backup_cur_idx, str_idx, null_cnt;
2796 re_dfastate_t *cur_state = NULL;
2797 re_node_set *cur_nodes, next_nodes;
2798 re_dfastate_t **backup_state_log;
2799 unsigned int context;
2801 subexp_num = dfa->nodes[top_node].opr.idx;
2802 /* Extend the buffer if we need. */
2803 if (BE (path->alloc < last_str + mctx->max_mb_elem_len + 1, 0))
2805 re_dfastate_t **new_array;
2806 int old_alloc = path->alloc;
2807 path->alloc += last_str + mctx->max_mb_elem_len + 1;
2808 new_array = re_realloc (path->array, re_dfastate_t *, path->alloc);
2809 if (BE (new_array == NULL, 0))
2811 path->alloc = old_alloc;
2814 path->array = new_array;
2815 memset (new_array + old_alloc, '\0',
2816 sizeof (re_dfastate_t *) * (path->alloc - old_alloc));
2819 str_idx = path->next_idx ?: top_str;
2821 /* Temporary modify MCTX. */
2822 backup_state_log = mctx->state_log;
2823 backup_cur_idx = mctx->input.cur_idx;
2824 mctx->state_log = path->array;
2825 mctx->input.cur_idx = str_idx;
2827 /* Setup initial node set. */
2828 context = re_string_context_at (&mctx->input, str_idx - 1, mctx->eflags);
2829 if (str_idx == top_str)
2831 err = re_node_set_init_1 (&next_nodes, top_node);
2832 if (BE (err != REG_NOERROR, 0))
2834 err = check_arrival_expand_ecl (dfa, &next_nodes, subexp_num, type);
2835 if (BE (err != REG_NOERROR, 0))
2837 re_node_set_free (&next_nodes);
2843 cur_state = mctx->state_log[str_idx];
2844 if (cur_state && cur_state->has_backref)
2846 err = re_node_set_init_copy (&next_nodes, &cur_state->nodes);
2847 if (BE (err != REG_NOERROR, 0))
2851 re_node_set_init_empty (&next_nodes);
2853 if (str_idx == top_str || (cur_state && cur_state->has_backref))
2855 if (next_nodes.nelem)
2857 err = expand_bkref_cache (mctx, &next_nodes, str_idx,
2859 if (BE (err != REG_NOERROR, 0))
2861 re_node_set_free (&next_nodes);
2865 cur_state = re_acquire_state_context (&err, dfa, &next_nodes, context);
2866 if (BE (cur_state == NULL && err != REG_NOERROR, 0))
2868 re_node_set_free (&next_nodes);
2871 mctx->state_log[str_idx] = cur_state;
2874 for (null_cnt = 0; str_idx < last_str && null_cnt <= mctx->max_mb_elem_len;)
2876 re_node_set_empty (&next_nodes);
2877 if (mctx->state_log[str_idx + 1])
2879 err = re_node_set_merge (&next_nodes,
2880 &mctx->state_log[str_idx + 1]->nodes);
2881 if (BE (err != REG_NOERROR, 0))
2883 re_node_set_free (&next_nodes);
2889 err = check_arrival_add_next_nodes (mctx, str_idx,
2890 &cur_state->non_eps_nodes,
2892 if (BE (err != REG_NOERROR, 0))
2894 re_node_set_free (&next_nodes);
2899 if (next_nodes.nelem)
2901 err = check_arrival_expand_ecl (dfa, &next_nodes, subexp_num, type);
2902 if (BE (err != REG_NOERROR, 0))
2904 re_node_set_free (&next_nodes);
2907 err = expand_bkref_cache (mctx, &next_nodes, str_idx,
2909 if (BE (err != REG_NOERROR, 0))
2911 re_node_set_free (&next_nodes);
2915 context = re_string_context_at (&mctx->input, str_idx - 1, mctx->eflags);
2916 cur_state = re_acquire_state_context (&err, dfa, &next_nodes, context);
2917 if (BE (cur_state == NULL && err != REG_NOERROR, 0))
2919 re_node_set_free (&next_nodes);
2922 mctx->state_log[str_idx] = cur_state;
2923 null_cnt = cur_state == NULL ? null_cnt + 1 : 0;
2925 re_node_set_free (&next_nodes);
2926 cur_nodes = (mctx->state_log[last_str] == NULL ? NULL
2927 : &mctx->state_log[last_str]->nodes);
2928 path->next_idx = str_idx;
2931 mctx->state_log = backup_state_log;
2932 mctx->input.cur_idx = backup_cur_idx;
2934 /* Then check the current node set has the node LAST_NODE. */
2935 if (cur_nodes != NULL && re_node_set_contains (cur_nodes, last_node))
2941 /* Helper functions for check_arrival. */
2943 /* Calculate the destination nodes of CUR_NODES at STR_IDX, and append them
2945 TODO: This function is similar to the functions transit_state*(),
2946 however this function has many additional works.
2947 Can't we unify them? */
2949 static reg_errcode_t
2951 check_arrival_add_next_nodes (re_match_context_t *mctx, int str_idx,
2952 re_node_set *cur_nodes, re_node_set *next_nodes)
2954 const re_dfa_t *const dfa = mctx->dfa;
2957 #ifdef RE_ENABLE_I18N
2958 reg_errcode_t err = REG_NOERROR;
2960 re_node_set union_set;
2961 re_node_set_init_empty (&union_set);
2962 for (cur_idx = 0; cur_idx < cur_nodes->nelem; ++cur_idx)
2965 int cur_node = cur_nodes->elems[cur_idx];
2967 re_token_type_t type = dfa->nodes[cur_node].type;
2968 assert (!IS_EPSILON_NODE (type));
2970 #ifdef RE_ENABLE_I18N
2971 /* If the node may accept `multi byte'. */
2972 if (dfa->nodes[cur_node].accept_mb)
2974 naccepted = check_node_accept_bytes (dfa, cur_node, &mctx->input,
2978 re_dfastate_t *dest_state;
2979 int next_node = dfa->nexts[cur_node];
2980 int next_idx = str_idx + naccepted;
2981 dest_state = mctx->state_log[next_idx];
2982 re_node_set_empty (&union_set);
2985 err = re_node_set_merge (&union_set, &dest_state->nodes);
2986 if (BE (err != REG_NOERROR, 0))
2988 re_node_set_free (&union_set);
2992 result = re_node_set_insert (&union_set, next_node);
2993 if (BE (result < 0, 0))
2995 re_node_set_free (&union_set);
2998 mctx->state_log[next_idx] = re_acquire_state (&err, dfa,
3000 if (BE (mctx->state_log[next_idx] == NULL
3001 && err != REG_NOERROR, 0))
3003 re_node_set_free (&union_set);
3008 #endif /* RE_ENABLE_I18N */
3010 || check_node_accept (mctx, dfa->nodes + cur_node, str_idx))
3012 result = re_node_set_insert (next_nodes, dfa->nexts[cur_node]);
3013 if (BE (result < 0, 0))
3015 re_node_set_free (&union_set);
3020 re_node_set_free (&union_set);
3024 /* For all the nodes in CUR_NODES, add the epsilon closures of them to
3025 CUR_NODES, however exclude the nodes which are:
3026 - inside the sub expression whose number is EX_SUBEXP, if FL_OPEN.
3027 - out of the sub expression whose number is EX_SUBEXP, if !FL_OPEN.
3030 static reg_errcode_t
3032 check_arrival_expand_ecl (const re_dfa_t *dfa, re_node_set *cur_nodes,
3033 int ex_subexp, int type)
3036 int idx, outside_node;
3037 re_node_set new_nodes;
3039 assert (cur_nodes->nelem);
3041 err = re_node_set_alloc (&new_nodes, cur_nodes->nelem);
3042 if (BE (err != REG_NOERROR, 0))
3044 /* Create a new node set NEW_NODES with the nodes which are epsilon
3045 closures of the node in CUR_NODES. */
3047 for (idx = 0; idx < cur_nodes->nelem; ++idx)
3049 int cur_node = cur_nodes->elems[idx];
3050 const re_node_set *eclosure = dfa->eclosures + cur_node;
3051 outside_node = find_subexp_node (dfa, eclosure, ex_subexp, type);
3052 if (outside_node == -1)
3054 /* There are no problematic nodes, just merge them. */
3055 err = re_node_set_merge (&new_nodes, eclosure);
3056 if (BE (err != REG_NOERROR, 0))
3058 re_node_set_free (&new_nodes);
3064 /* There are problematic nodes, re-calculate incrementally. */
3065 err = check_arrival_expand_ecl_sub (dfa, &new_nodes, cur_node,
3067 if (BE (err != REG_NOERROR, 0))
3069 re_node_set_free (&new_nodes);
3074 re_node_set_free (cur_nodes);
3075 *cur_nodes = new_nodes;
3079 /* Helper function for check_arrival_expand_ecl.
3080 Check incrementally the epsilon closure of TARGET, and if it isn't
3081 problematic append it to DST_NODES. */
3083 static reg_errcode_t
3085 check_arrival_expand_ecl_sub (const re_dfa_t *dfa, re_node_set *dst_nodes,
3086 int target, int ex_subexp, int type)
3089 for (cur_node = target; !re_node_set_contains (dst_nodes, cur_node);)
3093 if (dfa->nodes[cur_node].type == type
3094 && dfa->nodes[cur_node].opr.idx == ex_subexp)
3096 if (type == OP_CLOSE_SUBEXP)
3098 err = re_node_set_insert (dst_nodes, cur_node);
3099 if (BE (err == -1, 0))
3104 err = re_node_set_insert (dst_nodes, cur_node);
3105 if (BE (err == -1, 0))
3107 if (dfa->edests[cur_node].nelem == 0)
3109 if (dfa->edests[cur_node].nelem == 2)
3111 err = check_arrival_expand_ecl_sub (dfa, dst_nodes,
3112 dfa->edests[cur_node].elems[1],
3114 if (BE (err != REG_NOERROR, 0))
3117 cur_node = dfa->edests[cur_node].elems[0];
3123 /* For all the back references in the current state, calculate the
3124 destination of the back references by the appropriate entry
3125 in MCTX->BKREF_ENTS. */
3127 static reg_errcode_t
3129 expand_bkref_cache (re_match_context_t *mctx, re_node_set *cur_nodes,
3130 int cur_str, int subexp_num, int type)
3132 const re_dfa_t *const dfa = mctx->dfa;
3134 int cache_idx_start = search_cur_bkref_entry (mctx, cur_str);
3135 struct re_backref_cache_entry *ent;
3137 if (cache_idx_start == -1)
3141 ent = mctx->bkref_ents + cache_idx_start;
3144 int to_idx, next_node;
3146 /* Is this entry ENT is appropriate? */
3147 if (!re_node_set_contains (cur_nodes, ent->node))
3150 to_idx = cur_str + ent->subexp_to - ent->subexp_from;
3151 /* Calculate the destination of the back reference, and append it
3152 to MCTX->STATE_LOG. */
3153 if (to_idx == cur_str)
3155 /* The backreference did epsilon transit, we must re-check all the
3156 node in the current state. */
3157 re_node_set new_dests;
3158 reg_errcode_t err2, err3;
3159 next_node = dfa->edests[ent->node].elems[0];
3160 if (re_node_set_contains (cur_nodes, next_node))
3162 err = re_node_set_init_1 (&new_dests, next_node);
3163 err2 = check_arrival_expand_ecl (dfa, &new_dests, subexp_num, type);
3164 err3 = re_node_set_merge (cur_nodes, &new_dests);
3165 re_node_set_free (&new_dests);
3166 if (BE (err != REG_NOERROR || err2 != REG_NOERROR
3167 || err3 != REG_NOERROR, 0))
3169 err = (err != REG_NOERROR ? err
3170 : (err2 != REG_NOERROR ? err2 : err3));
3173 /* TODO: It is still inefficient... */
3178 re_node_set union_set;
3179 next_node = dfa->nexts[ent->node];
3180 if (mctx->state_log[to_idx])
3183 if (re_node_set_contains (&mctx->state_log[to_idx]->nodes,
3186 err = re_node_set_init_copy (&union_set,
3187 &mctx->state_log[to_idx]->nodes);
3188 ret = re_node_set_insert (&union_set, next_node);
3189 if (BE (err != REG_NOERROR || ret < 0, 0))
3191 re_node_set_free (&union_set);
3192 err = err != REG_NOERROR ? err : REG_ESPACE;
3198 err = re_node_set_init_1 (&union_set, next_node);
3199 if (BE (err != REG_NOERROR, 0))
3202 mctx->state_log[to_idx] = re_acquire_state (&err, dfa, &union_set);
3203 re_node_set_free (&union_set);
3204 if (BE (mctx->state_log[to_idx] == NULL
3205 && err != REG_NOERROR, 0))
3209 while (ent++->more);
3213 /* Build transition table for the state.
3214 Return 1 if succeeded, otherwise return NULL. */
3218 build_trtable (const re_dfa_t *dfa, re_dfastate_t *state)
3221 int i, j, ch, need_word_trtable = 0;
3222 bitset_word_t elem, mask;
3223 bool dests_node_malloced = false;
3224 bool dest_states_malloced = false;
3225 int ndests; /* Number of the destination states from `state'. */
3226 re_dfastate_t **trtable;
3227 re_dfastate_t **dest_states = NULL, **dest_states_word, **dest_states_nl;
3228 re_node_set follows, *dests_node;
3230 bitset_t acceptable;
3234 re_node_set dests_node[SBC_MAX];
3235 bitset_t dests_ch[SBC_MAX];
3238 /* We build DFA states which corresponds to the destination nodes
3239 from `state'. `dests_node[i]' represents the nodes which i-th
3240 destination state contains, and `dests_ch[i]' represents the
3241 characters which i-th destination state accepts. */
3242 if (__libc_use_alloca (sizeof (struct dests_alloc)))
3243 dests_alloc = (struct dests_alloc *) alloca (sizeof (struct dests_alloc));
3246 dests_alloc = re_malloc (struct dests_alloc, 1);
3247 if (BE (dests_alloc == NULL, 0))
3249 dests_node_malloced = true;
3251 dests_node = dests_alloc->dests_node;
3252 dests_ch = dests_alloc->dests_ch;
3254 /* Initialize transiton table. */
3255 state->word_trtable = state->trtable = NULL;
3257 /* At first, group all nodes belonging to `state' into several
3259 ndests = group_nodes_into_DFAstates (dfa, state, dests_node, dests_ch);
3260 if (BE (ndests <= 0, 0))
3262 if (dests_node_malloced)
3264 /* Return 0 in case of an error, 1 otherwise. */
3267 state->trtable = (re_dfastate_t **)
3268 calloc (sizeof (re_dfastate_t *), SBC_MAX);
3274 err = re_node_set_alloc (&follows, ndests + 1);
3275 if (BE (err != REG_NOERROR, 0))
3278 if (__libc_use_alloca ((sizeof (re_node_set) + sizeof (bitset_t)) * SBC_MAX
3279 + ndests * 3 * sizeof (re_dfastate_t *)))
3280 dest_states = (re_dfastate_t **)
3281 alloca (ndests * 3 * sizeof (re_dfastate_t *));
3284 dest_states = (re_dfastate_t **)
3285 malloc (ndests * 3 * sizeof (re_dfastate_t *));
3286 if (BE (dest_states == NULL, 0))
3289 if (dest_states_malloced)
3291 re_node_set_free (&follows);
3292 for (i = 0; i < ndests; ++i)
3293 re_node_set_free (dests_node + i);
3294 if (dests_node_malloced)
3298 dest_states_malloced = true;
3300 dest_states_word = dest_states + ndests;
3301 dest_states_nl = dest_states_word + ndests;
3302 bitset_empty (acceptable);
3304 /* Then build the states for all destinations. */
3305 for (i = 0; i < ndests; ++i)
3308 re_node_set_empty (&follows);
3309 /* Merge the follows of this destination states. */
3310 for (j = 0; j < dests_node[i].nelem; ++j)
3312 next_node = dfa->nexts[dests_node[i].elems[j]];
3313 if (next_node != -1)
3315 err = re_node_set_merge (&follows, dfa->eclosures + next_node);
3316 if (BE (err != REG_NOERROR, 0))
3320 dest_states[i] = re_acquire_state_context (&err, dfa, &follows, 0);
3321 if (BE (dest_states[i] == NULL && err != REG_NOERROR, 0))
3323 /* If the new state has context constraint,
3324 build appropriate states for these contexts. */
3325 if (dest_states[i]->has_constraint)
3327 dest_states_word[i] = re_acquire_state_context (&err, dfa, &follows,
3329 if (BE (dest_states_word[i] == NULL && err != REG_NOERROR, 0))
3332 if (dest_states[i] != dest_states_word[i] && dfa->mb_cur_max > 1)
3333 need_word_trtable = 1;
3335 dest_states_nl[i] = re_acquire_state_context (&err, dfa, &follows,
3337 if (BE (dest_states_nl[i] == NULL && err != REG_NOERROR, 0))
3342 dest_states_word[i] = dest_states[i];
3343 dest_states_nl[i] = dest_states[i];
3345 bitset_merge (acceptable, dests_ch[i]);
3348 if (!BE (need_word_trtable, 0))
3350 /* We don't care about whether the following character is a word
3351 character, or we are in a single-byte character set so we can
3352 discern by looking at the character code: allocate a
3353 256-entry transition table. */
3354 trtable = state->trtable = calloc (sizeof (re_dfastate_t *), SBC_MAX);
3355 if (BE (trtable == NULL, 0))
3358 /* For all characters ch...: */
3359 for (i = 0; i < BITSET_WORDS; ++i)
3360 for (ch = i * BITSET_WORD_BITS, elem = acceptable[i], mask = 1;
3362 mask <<= 1, elem >>= 1, ++ch)
3363 if (BE (elem & 1, 0))
3365 /* There must be exactly one destination which accepts
3366 character ch. See group_nodes_into_DFAstates. */
3367 for (j = 0; (dests_ch[j][i] & mask) == 0; ++j)
3370 /* j-th destination accepts the word character ch. */
3371 if (dfa->word_char[i] & mask)
3372 trtable[ch] = dest_states_word[j];
3374 trtable[ch] = dest_states[j];
3379 /* We care about whether the following character is a word
3380 character, and we are in a multi-byte character set: discern
3381 by looking at the character code: build two 256-entry
3382 transition tables, one starting at trtable[0] and one
3383 starting at trtable[SBC_MAX]. */
3384 trtable = state->word_trtable = calloc (sizeof (re_dfastate_t *), 2 * SBC_MAX);
3385 if (BE (trtable == NULL, 0))
3388 /* For all characters ch...: */
3389 for (i = 0; i < BITSET_WORDS; ++i)
3390 for (ch = i * BITSET_WORD_BITS, elem = acceptable[i], mask = 1;
3392 mask <<= 1, elem >>= 1, ++ch)
3393 if (BE (elem & 1, 0))
3395 /* There must be exactly one destination which accepts
3396 character ch. See group_nodes_into_DFAstates. */
3397 for (j = 0; (dests_ch[j][i] & mask) == 0; ++j)
3400 /* j-th destination accepts the word character ch. */
3401 trtable[ch] = dest_states[j];
3402 trtable[ch + SBC_MAX] = dest_states_word[j];
3407 if (bitset_contain (acceptable, NEWLINE_CHAR))
3409 /* The current state accepts newline character. */
3410 for (j = 0; j < ndests; ++j)
3411 if (bitset_contain (dests_ch[j], NEWLINE_CHAR))
3413 /* k-th destination accepts newline character. */
3414 trtable[NEWLINE_CHAR] = dest_states_nl[j];
3415 if (need_word_trtable)
3416 trtable[NEWLINE_CHAR + SBC_MAX] = dest_states_nl[j];
3417 /* There must be only one destination which accepts
3418 newline. See group_nodes_into_DFAstates. */
3423 if (dest_states_malloced)
3426 re_node_set_free (&follows);
3427 for (i = 0; i < ndests; ++i)
3428 re_node_set_free (dests_node + i);
3430 if (dests_node_malloced)
3436 /* Group all nodes belonging to STATE into several destinations.
3437 Then for all destinations, set the nodes belonging to the destination
3438 to DESTS_NODE[i] and set the characters accepted by the destination
3439 to DEST_CH[i]. This function return the number of destinations. */
3443 group_nodes_into_DFAstates (const re_dfa_t *dfa, const re_dfastate_t *state,
3444 re_node_set *dests_node, bitset_t *dests_ch)
3449 int ndests; /* Number of the destinations from `state'. */
3450 bitset_t accepts; /* Characters a node can accept. */
3451 const re_node_set *cur_nodes = &state->nodes;
3452 bitset_empty (accepts);
3455 /* For all the nodes belonging to `state', */
3456 for (i = 0; i < cur_nodes->nelem; ++i)
3458 re_token_t *node = &dfa->nodes[cur_nodes->elems[i]];
3459 re_token_type_t type = node->type;
3460 unsigned int constraint = node->constraint;
3462 /* Enumerate all single byte character this node can accept. */
3463 if (type == CHARACTER)
3464 bitset_set (accepts, node->opr.c);
3465 else if (type == SIMPLE_BRACKET)
3467 bitset_merge (accepts, node->opr.sbcset);
3469 else if (type == OP_PERIOD)
3471 #ifdef RE_ENABLE_I18N
3472 if (dfa->mb_cur_max > 1)
3473 bitset_merge (accepts, dfa->sb_char);
3476 bitset_set_all (accepts);
3477 if (!(dfa->syntax & RE_DOT_NEWLINE))
3478 bitset_clear (accepts, '\n');
3479 if (dfa->syntax & RE_DOT_NOT_NULL)
3480 bitset_clear (accepts, '\0');
3482 #ifdef RE_ENABLE_I18N
3483 else if (type == OP_UTF8_PERIOD)
3485 memset (accepts, '\xff', sizeof (bitset_t) / 2);
3486 if (!(dfa->syntax & RE_DOT_NEWLINE))
3487 bitset_clear (accepts, '\n');
3488 if (dfa->syntax & RE_DOT_NOT_NULL)
3489 bitset_clear (accepts, '\0');
3495 /* Check the `accepts' and sift the characters which are not
3496 match it the context. */
3499 if (constraint & NEXT_NEWLINE_CONSTRAINT)
3501 bool accepts_newline = bitset_contain (accepts, NEWLINE_CHAR);
3502 bitset_empty (accepts);
3503 if (accepts_newline)
3504 bitset_set (accepts, NEWLINE_CHAR);
3508 if (constraint & NEXT_ENDBUF_CONSTRAINT)
3510 bitset_empty (accepts);
3514 if (constraint & NEXT_WORD_CONSTRAINT)
3516 bitset_word_t any_set = 0;
3517 if (type == CHARACTER && !node->word_char)
3519 bitset_empty (accepts);
3522 #ifdef RE_ENABLE_I18N
3523 if (dfa->mb_cur_max > 1)
3524 for (j = 0; j < BITSET_WORDS; ++j)
3525 any_set |= (accepts[j] &= (dfa->word_char[j] | ~dfa->sb_char[j]));
3528 for (j = 0; j < BITSET_WORDS; ++j)
3529 any_set |= (accepts[j] &= dfa->word_char[j]);
3533 if (constraint & NEXT_NOTWORD_CONSTRAINT)
3535 bitset_word_t any_set = 0;
3536 if (type == CHARACTER && node->word_char)
3538 bitset_empty (accepts);
3541 #ifdef RE_ENABLE_I18N
3542 if (dfa->mb_cur_max > 1)
3543 for (j = 0; j < BITSET_WORDS; ++j)
3544 any_set |= (accepts[j] &= ~(dfa->word_char[j] & dfa->sb_char[j]));
3547 for (j = 0; j < BITSET_WORDS; ++j)
3548 any_set |= (accepts[j] &= ~dfa->word_char[j]);
3554 /* Then divide `accepts' into DFA states, or create a new
3555 state. Above, we make sure that accepts is not empty. */
3556 for (j = 0; j < ndests; ++j)
3558 bitset_t intersec; /* Intersection sets, see below. */
3560 /* Flags, see below. */
3561 bitset_word_t has_intersec, not_subset, not_consumed;
3563 /* Optimization, skip if this state doesn't accept the character. */
3564 if (type == CHARACTER && !bitset_contain (dests_ch[j], node->opr.c))
3567 /* Enumerate the intersection set of this state and `accepts'. */
3569 for (k = 0; k < BITSET_WORDS; ++k)
3570 has_intersec |= intersec[k] = accepts[k] & dests_ch[j][k];
3571 /* And skip if the intersection set is empty. */
3575 /* Then check if this state is a subset of `accepts'. */
3576 not_subset = not_consumed = 0;
3577 for (k = 0; k < BITSET_WORDS; ++k)
3579 not_subset |= remains[k] = ~accepts[k] & dests_ch[j][k];
3580 not_consumed |= accepts[k] = accepts[k] & ~dests_ch[j][k];
3583 /* If this state isn't a subset of `accepts', create a
3584 new group state, which has the `remains'. */
3587 bitset_copy (dests_ch[ndests], remains);
3588 bitset_copy (dests_ch[j], intersec);
3589 err = re_node_set_init_copy (dests_node + ndests, &dests_node[j]);
3590 if (BE (err != REG_NOERROR, 0))
3595 /* Put the position in the current group. */
3596 result = re_node_set_insert (&dests_node[j], cur_nodes->elems[i]);
3597 if (BE (result < 0, 0))
3600 /* If all characters are consumed, go to next node. */
3604 /* Some characters remain, create a new group. */
3607 bitset_copy (dests_ch[ndests], accepts);
3608 err = re_node_set_init_1 (dests_node + ndests, cur_nodes->elems[i]);
3609 if (BE (err != REG_NOERROR, 0))
3612 bitset_empty (accepts);
3617 for (j = 0; j < ndests; ++j)
3618 re_node_set_free (dests_node + j);
3622 #ifdef RE_ENABLE_I18N
3623 /* Check how many bytes the node `dfa->nodes[node_idx]' accepts.
3624 Return the number of the bytes the node accepts.
3625 STR_IDX is the current index of the input string.
3627 This function handles the nodes which can accept one character, or
3628 one collating element like '.', '[a-z]', opposite to the other nodes
3629 can only accept one byte. */
3633 check_node_accept_bytes (const re_dfa_t *dfa, int node_idx,
3634 const re_string_t *input, int str_idx)
3636 const re_token_t *node = dfa->nodes + node_idx;
3637 int char_len, elem_len;
3640 if (BE (node->type == OP_UTF8_PERIOD, 0))
3642 unsigned char c = re_string_byte_at (input, str_idx), d;
3643 if (BE (c < 0xc2, 1))
3646 if (str_idx + 2 > input->len)
3649 d = re_string_byte_at (input, str_idx + 1);
3651 return (d < 0x80 || d > 0xbf) ? 0 : 2;
3655 if (c == 0xe0 && d < 0xa0)
3661 if (c == 0xf0 && d < 0x90)
3667 if (c == 0xf8 && d < 0x88)
3673 if (c == 0xfc && d < 0x84)
3679 if (str_idx + char_len > input->len)
3682 for (i = 1; i < char_len; ++i)
3684 d = re_string_byte_at (input, str_idx + i);
3685 if (d < 0x80 || d > 0xbf)
3691 char_len = re_string_char_size_at (input, str_idx);
3692 if (node->type == OP_PERIOD)
3696 /* FIXME: I don't think this if is needed, as both '\n'
3697 and '\0' are char_len == 1. */
3698 /* '.' accepts any one character except the following two cases. */
3699 if ((!(dfa->syntax & RE_DOT_NEWLINE) &&
3700 re_string_byte_at (input, str_idx) == '\n') ||
3701 ((dfa->syntax & RE_DOT_NOT_NULL) &&
3702 re_string_byte_at (input, str_idx) == '\0'))
3707 elem_len = re_string_elem_size_at (input, str_idx);
3708 if ((elem_len <= 1 && char_len <= 1) || char_len == 0)
3711 if (node->type == COMPLEX_BRACKET)
3713 const re_charset_t *cset = node->opr.mbcset;
3715 const unsigned char *pin
3716 = ((const unsigned char *) re_string_get_buffer (input) + str_idx);
3721 wchar_t wc = ((cset->nranges || cset->nchar_classes || cset->nmbchars)
3722 ? re_string_wchar_at (input, str_idx) : 0);
3724 /* match with multibyte character? */
3725 for (i = 0; i < cset->nmbchars; ++i)
3726 if (wc == cset->mbchars[i])
3728 match_len = char_len;
3729 goto check_node_accept_bytes_match;
3731 /* match with character_class? */
3732 for (i = 0; i < cset->nchar_classes; ++i)
3734 wctype_t wt = cset->char_classes[i];
3735 if (__iswctype (wc, wt))
3737 match_len = char_len;
3738 goto check_node_accept_bytes_match;
3743 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3746 unsigned int in_collseq = 0;
3747 const int32_t *table, *indirect;
3748 const unsigned char *weights, *extra;
3749 const char *collseqwc;
3751 /* This #include defines a local function! */
3752 # include <locale/weight.h>
3754 /* match with collating_symbol? */
3755 if (cset->ncoll_syms)
3756 extra = (const unsigned char *)
3757 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
3758 for (i = 0; i < cset->ncoll_syms; ++i)
3760 const unsigned char *coll_sym = extra + cset->coll_syms[i];
3761 /* Compare the length of input collating element and
3762 the length of current collating element. */
3763 if (*coll_sym != elem_len)
3765 /* Compare each bytes. */
3766 for (j = 0; j < *coll_sym; j++)
3767 if (pin[j] != coll_sym[1 + j])
3771 /* Match if every bytes is equal. */
3773 goto check_node_accept_bytes_match;
3779 if (elem_len <= char_len)
3781 collseqwc = _NL_CURRENT (LC_COLLATE, _NL_COLLATE_COLLSEQWC);
3782 in_collseq = __collseq_table_lookup (collseqwc, wc);
3785 in_collseq = find_collation_sequence_value (pin, elem_len);
3787 /* match with range expression? */
3788 for (i = 0; i < cset->nranges; ++i)
3789 if (cset->range_starts[i] <= in_collseq
3790 && in_collseq <= cset->range_ends[i])
3792 match_len = elem_len;
3793 goto check_node_accept_bytes_match;
3796 /* match with equivalence_class? */
3797 if (cset->nequiv_classes)
3799 const unsigned char *cp = pin;
3800 table = (const int32_t *)
3801 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
3802 weights = (const unsigned char *)
3803 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
3804 extra = (const unsigned char *)
3805 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
3806 indirect = (const int32_t *)
3807 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
3808 idx = findidx (&cp);
3810 for (i = 0; i < cset->nequiv_classes; ++i)
3812 int32_t equiv_class_idx = cset->equiv_classes[i];
3813 size_t weight_len = weights[idx];
3814 if (weight_len == weights[equiv_class_idx])
3817 while (cnt <= weight_len
3818 && (weights[equiv_class_idx + 1 + cnt]
3819 == weights[idx + 1 + cnt]))
3821 if (cnt > weight_len)
3823 match_len = elem_len;
3824 goto check_node_accept_bytes_match;
3833 /* match with range expression? */
3836 memset (cmp_buf, 0, sizeof(cmp_buf));
3838 for (i = 0; i < cset->nranges; ++i)
3840 cmp_buf[0] = cset->range_starts[i];
3841 cmp_buf[4] = cset->range_ends[i];
3842 if (wcscoll (cmp_buf, cmp_buf + 2) <= 0
3843 && wcscoll (cmp_buf + 2, cmp_buf + 4) <= 0)
3845 match_len = char_len;
3846 goto check_node_accept_bytes_match;
3851 check_node_accept_bytes_match:
3852 if (!cset->non_match)
3856 return (elem_len > char_len) ? elem_len : char_len;
3864 find_collation_sequence_value (const unsigned char *mbs, size_t mbs_len)
3866 uint32_t nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3871 /* No valid character. Match it as a single byte character. */
3872 const unsigned char *collseq = (const unsigned char *)
3873 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_COLLSEQMB);
3874 return collseq[mbs[0]];
3881 const unsigned char *extra = (const unsigned char *)
3882 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
3883 int32_t extrasize = (const unsigned char *)
3884 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB + 1) - extra;
3886 for (idx = 0; idx < extrasize;)
3888 int mbs_cnt, found = 0;
3889 int32_t elem_mbs_len;
3890 /* Skip the name of collating element name. */
3891 idx = idx + extra[idx] + 1;
3892 elem_mbs_len = extra[idx++];
3893 if (mbs_len == elem_mbs_len)
3895 for (mbs_cnt = 0; mbs_cnt < elem_mbs_len; ++mbs_cnt)
3896 if (extra[idx + mbs_cnt] != mbs[mbs_cnt])
3898 if (mbs_cnt == elem_mbs_len)
3899 /* Found the entry. */
3902 /* Skip the byte sequence of the collating element. */
3903 idx += elem_mbs_len;
3904 /* Adjust for the alignment. */
3905 idx = (idx + 3) & ~3;
3906 /* Skip the collation sequence value. */
3907 idx += sizeof (uint32_t);
3908 /* Skip the wide char sequence of the collating element. */
3909 idx = idx + sizeof (uint32_t) * (extra[idx] + 1);
3910 /* If we found the entry, return the sequence value. */
3912 return *(uint32_t *) (extra + idx);
3913 /* Skip the collation sequence value. */
3914 idx += sizeof (uint32_t);
3920 #endif /* RE_ENABLE_I18N */
3922 /* Check whether the node accepts the byte which is IDX-th
3923 byte of the INPUT. */
3927 check_node_accept (const re_match_context_t *mctx, const re_token_t *node,
3931 ch = re_string_byte_at (&mctx->input, idx);
3935 if (node->opr.c != ch)
3939 case SIMPLE_BRACKET:
3940 if (!bitset_contain (node->opr.sbcset, ch))
3944 #ifdef RE_ENABLE_I18N
3945 case OP_UTF8_PERIOD:
3951 if ((ch == '\n' && !(mctx->dfa->syntax & RE_DOT_NEWLINE))
3952 || (ch == '\0' && (mctx->dfa->syntax & RE_DOT_NOT_NULL)))
3960 if (node->constraint)
3962 /* The node has constraints. Check whether the current context
3963 satisfies the constraints. */
3964 unsigned int context = re_string_context_at (&mctx->input, idx,
3966 if (NOT_SATISFY_NEXT_CONSTRAINT (node->constraint, context))
3973 /* Extend the buffers, if the buffers have run out. */
3975 static reg_errcode_t
3977 extend_buffers (re_match_context_t *mctx)
3980 re_string_t *pstr = &mctx->input;
3982 /* Double the lengthes of the buffers. */
3983 ret = re_string_realloc_buffers (pstr, pstr->bufs_len * 2);
3984 if (BE (ret != REG_NOERROR, 0))
3987 if (mctx->state_log != NULL)
3989 /* And double the length of state_log. */
3990 /* XXX We have no indication of the size of this buffer. If this
3991 allocation fail we have no indication that the state_log array
3992 does not have the right size. */
3993 re_dfastate_t **new_array = re_realloc (mctx->state_log, re_dfastate_t *,
3994 pstr->bufs_len + 1);
3995 if (BE (new_array == NULL, 0))
3997 mctx->state_log = new_array;
4000 /* Then reconstruct the buffers. */
4003 #ifdef RE_ENABLE_I18N
4004 if (pstr->mb_cur_max > 1)
4006 ret = build_wcs_upper_buffer (pstr);
4007 if (BE (ret != REG_NOERROR, 0))
4011 #endif /* RE_ENABLE_I18N */
4012 build_upper_buffer (pstr);
4016 #ifdef RE_ENABLE_I18N
4017 if (pstr->mb_cur_max > 1)
4018 build_wcs_buffer (pstr);
4020 #endif /* RE_ENABLE_I18N */
4022 if (pstr->trans != NULL)
4023 re_string_translate_buffer (pstr);
4030 /* Functions for matching context. */
4032 /* Initialize MCTX. */
4034 static reg_errcode_t
4036 match_ctx_init (re_match_context_t *mctx, int eflags, int n)
4038 mctx->eflags = eflags;
4039 mctx->match_last = -1;
4042 mctx->bkref_ents = re_malloc (struct re_backref_cache_entry, n);
4043 mctx->sub_tops = re_malloc (re_sub_match_top_t *, n);
4044 if (BE (mctx->bkref_ents == NULL || mctx->sub_tops == NULL, 0))
4047 /* Already zero-ed by the caller.
4049 mctx->bkref_ents = NULL;
4050 mctx->nbkref_ents = 0;
4051 mctx->nsub_tops = 0; */
4052 mctx->abkref_ents = n;
4053 mctx->max_mb_elem_len = 1;
4054 mctx->asub_tops = n;
4058 /* Clean the entries which depend on the current input in MCTX.
4059 This function must be invoked when the matcher changes the start index
4060 of the input, or changes the input string. */
4064 match_ctx_clean (re_match_context_t *mctx)
4067 for (st_idx = 0; st_idx < mctx->nsub_tops; ++st_idx)
4070 re_sub_match_top_t *top = mctx->sub_tops[st_idx];
4071 for (sl_idx = 0; sl_idx < top->nlasts; ++sl_idx)
4073 re_sub_match_last_t *last = top->lasts[sl_idx];
4074 re_free (last->path.array);
4077 re_free (top->lasts);
4080 re_free (top->path->array);
4081 re_free (top->path);
4086 mctx->nsub_tops = 0;
4087 mctx->nbkref_ents = 0;
4090 /* Free all the memory associated with MCTX. */
4094 match_ctx_free (re_match_context_t *mctx)
4096 /* First, free all the memory associated with MCTX->SUB_TOPS. */
4097 match_ctx_clean (mctx);
4098 re_free (mctx->sub_tops);
4099 re_free (mctx->bkref_ents);
4102 /* Add a new backreference entry to MCTX.
4103 Note that we assume that caller never call this function with duplicate
4104 entry, and call with STR_IDX which isn't smaller than any existing entry.
4107 static reg_errcode_t
4109 match_ctx_add_entry (re_match_context_t *mctx, int node, int str_idx, int from,
4112 if (mctx->nbkref_ents >= mctx->abkref_ents)
4114 struct re_backref_cache_entry* new_entry;
4115 new_entry = re_realloc (mctx->bkref_ents, struct re_backref_cache_entry,
4116 mctx->abkref_ents * 2);
4117 if (BE (new_entry == NULL, 0))
4119 re_free (mctx->bkref_ents);
4122 mctx->bkref_ents = new_entry;
4123 memset (mctx->bkref_ents + mctx->nbkref_ents, '\0',
4124 sizeof (struct re_backref_cache_entry) * mctx->abkref_ents);
4125 mctx->abkref_ents *= 2;
4127 if (mctx->nbkref_ents > 0
4128 && mctx->bkref_ents[mctx->nbkref_ents - 1].str_idx == str_idx)
4129 mctx->bkref_ents[mctx->nbkref_ents - 1].more = 1;
4131 mctx->bkref_ents[mctx->nbkref_ents].node = node;
4132 mctx->bkref_ents[mctx->nbkref_ents].str_idx = str_idx;
4133 mctx->bkref_ents[mctx->nbkref_ents].subexp_from = from;
4134 mctx->bkref_ents[mctx->nbkref_ents].subexp_to = to;
4136 /* This is a cache that saves negative results of check_dst_limits_calc_pos.
4137 If bit N is clear, means that this entry won't epsilon-transition to
4138 an OP_OPEN_SUBEXP or OP_CLOSE_SUBEXP for the N+1-th subexpression. If
4139 it is set, check_dst_limits_calc_pos_1 will recurse and try to find one
4142 A backreference does not epsilon-transition unless it is empty, so set
4143 to all zeros if FROM != TO. */
4144 mctx->bkref_ents[mctx->nbkref_ents].eps_reachable_subexps_map
4145 = (from == to ? ~0 : 0);
4147 mctx->bkref_ents[mctx->nbkref_ents++].more = 0;
4148 if (mctx->max_mb_elem_len < to - from)
4149 mctx->max_mb_elem_len = to - from;
4153 /* Search for the first entry which has the same str_idx, or -1 if none is
4154 found. Note that MCTX->BKREF_ENTS is already sorted by MCTX->STR_IDX. */
4158 search_cur_bkref_entry (const re_match_context_t *mctx, int str_idx)
4160 int left, right, mid, last;
4161 last = right = mctx->nbkref_ents;
4162 for (left = 0; left < right;)
4164 mid = (left + right) / 2;
4165 if (mctx->bkref_ents[mid].str_idx < str_idx)
4170 if (left < last && mctx->bkref_ents[left].str_idx == str_idx)
4176 /* Register the node NODE, whose type is OP_OPEN_SUBEXP, and which matches
4179 static reg_errcode_t
4181 match_ctx_add_subtop (re_match_context_t *mctx, int node, int str_idx)
4184 assert (mctx->sub_tops != NULL);
4185 assert (mctx->asub_tops > 0);
4187 if (BE (mctx->nsub_tops == mctx->asub_tops, 0))
4189 int new_asub_tops = mctx->asub_tops * 2;
4190 re_sub_match_top_t **new_array = re_realloc (mctx->sub_tops,
4191 re_sub_match_top_t *,
4193 if (BE (new_array == NULL, 0))
4195 mctx->sub_tops = new_array;
4196 mctx->asub_tops = new_asub_tops;
4198 mctx->sub_tops[mctx->nsub_tops] = calloc (1, sizeof (re_sub_match_top_t));
4199 if (BE (mctx->sub_tops[mctx->nsub_tops] == NULL, 0))
4201 mctx->sub_tops[mctx->nsub_tops]->node = node;
4202 mctx->sub_tops[mctx->nsub_tops++]->str_idx = str_idx;
4206 /* Register the node NODE, whose type is OP_CLOSE_SUBEXP, and which matches
4207 at STR_IDX, whose corresponding OP_OPEN_SUBEXP is SUB_TOP. */
4209 static re_sub_match_last_t *
4211 match_ctx_add_sublast (re_sub_match_top_t *subtop, int node, int str_idx)
4213 re_sub_match_last_t *new_entry;
4214 if (BE (subtop->nlasts == subtop->alasts, 0))
4216 int new_alasts = 2 * subtop->alasts + 1;
4217 re_sub_match_last_t **new_array = re_realloc (subtop->lasts,
4218 re_sub_match_last_t *,
4220 if (BE (new_array == NULL, 0))
4222 subtop->lasts = new_array;
4223 subtop->alasts = new_alasts;
4225 new_entry = calloc (1, sizeof (re_sub_match_last_t));
4226 if (BE (new_entry != NULL, 1))
4228 subtop->lasts[subtop->nlasts] = new_entry;
4229 new_entry->node = node;
4230 new_entry->str_idx = str_idx;
4238 sift_ctx_init (re_sift_context_t *sctx, re_dfastate_t **sifted_sts,
4239 re_dfastate_t **limited_sts, int last_node, int last_str_idx)
4241 sctx->sifted_states = sifted_sts;
4242 sctx->limited_states = limited_sts;
4243 sctx->last_node = last_node;
4244 sctx->last_str_idx = last_str_idx;
4245 re_node_set_init_empty (&sctx->limits);