1 /* Extended regular expression matching and search library.
2 Copyright (C) 2002-2013 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 General Public
8 License as published by the Free Software Foundation; either
9 version 3 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 General Public License for more details.
16 You should have received a copy of the GNU General Public
17 License along with the GNU C Library; if not, see
18 <http://www.gnu.org/licenses/>. */
22 static reg_errcode_t match_ctx_init (re_match_context_t *cache, int eflags,
23 Idx n) internal_function;
24 static void match_ctx_clean (re_match_context_t *mctx) internal_function;
25 static void match_ctx_free (re_match_context_t *cache) internal_function;
26 static reg_errcode_t match_ctx_add_entry (re_match_context_t *cache, Idx node,
27 Idx str_idx, Idx from, Idx to)
29 static Idx search_cur_bkref_entry (const re_match_context_t *mctx, Idx str_idx)
31 static reg_errcode_t match_ctx_add_subtop (re_match_context_t *mctx, Idx node,
32 Idx str_idx) internal_function;
33 static re_sub_match_last_t * match_ctx_add_sublast (re_sub_match_top_t *subtop,
34 Idx node, Idx str_idx)
36 static void sift_ctx_init (re_sift_context_t *sctx, re_dfastate_t **sifted_sts,
37 re_dfastate_t **limited_sts, Idx last_node,
40 static reg_errcode_t re_search_internal (const regex_t *preg,
41 const char *string, Idx length,
42 Idx start, Idx last_start, Idx stop,
43 size_t nmatch, regmatch_t pmatch[],
44 int eflags) internal_function;
45 static regoff_t re_search_2_stub (struct re_pattern_buffer *bufp,
46 const char *string1, Idx length1,
47 const char *string2, Idx length2,
48 Idx start, regoff_t range,
49 struct re_registers *regs,
50 Idx stop, bool ret_len) internal_function;
51 static regoff_t re_search_stub (struct re_pattern_buffer *bufp,
52 const char *string, Idx length, Idx start,
53 regoff_t range, Idx stop,
54 struct re_registers *regs,
55 bool ret_len) internal_function;
56 static unsigned re_copy_regs (struct re_registers *regs, regmatch_t *pmatch,
57 Idx nregs, int regs_allocated) internal_function;
58 static reg_errcode_t prune_impossible_nodes (re_match_context_t *mctx)
60 static Idx check_matching (re_match_context_t *mctx, bool fl_longest_match,
61 Idx *p_match_first) internal_function;
62 static Idx check_halt_state_context (const re_match_context_t *mctx,
63 const re_dfastate_t *state, Idx idx)
65 static void update_regs (const re_dfa_t *dfa, regmatch_t *pmatch,
66 regmatch_t *prev_idx_match, Idx cur_node,
67 Idx cur_idx, Idx nmatch) internal_function;
68 static reg_errcode_t push_fail_stack (struct re_fail_stack_t *fs,
69 Idx str_idx, Idx dest_node, Idx nregs,
71 re_node_set *eps_via_nodes)
73 static reg_errcode_t set_regs (const regex_t *preg,
74 const re_match_context_t *mctx,
75 size_t nmatch, regmatch_t *pmatch,
76 bool fl_backtrack) internal_function;
77 static reg_errcode_t free_fail_stack_return (struct re_fail_stack_t *fs)
81 static int sift_states_iter_mb (const re_match_context_t *mctx,
82 re_sift_context_t *sctx,
83 Idx node_idx, Idx str_idx, Idx max_str_idx)
85 #endif /* RE_ENABLE_I18N */
86 static reg_errcode_t sift_states_backward (const re_match_context_t *mctx,
87 re_sift_context_t *sctx)
89 static reg_errcode_t build_sifted_states (const re_match_context_t *mctx,
90 re_sift_context_t *sctx, Idx str_idx,
91 re_node_set *cur_dest)
93 static reg_errcode_t update_cur_sifted_state (const re_match_context_t *mctx,
94 re_sift_context_t *sctx,
96 re_node_set *dest_nodes)
98 static reg_errcode_t add_epsilon_src_nodes (const re_dfa_t *dfa,
99 re_node_set *dest_nodes,
100 const re_node_set *candidates)
102 static bool check_dst_limits (const re_match_context_t *mctx,
103 const re_node_set *limits,
104 Idx dst_node, Idx dst_idx, Idx src_node,
105 Idx src_idx) internal_function;
106 static int check_dst_limits_calc_pos_1 (const re_match_context_t *mctx,
107 int boundaries, Idx subexp_idx,
108 Idx from_node, Idx bkref_idx)
110 static int check_dst_limits_calc_pos (const re_match_context_t *mctx,
111 Idx limit, Idx subexp_idx,
112 Idx node, Idx str_idx,
113 Idx bkref_idx) internal_function;
114 static reg_errcode_t check_subexp_limits (const re_dfa_t *dfa,
115 re_node_set *dest_nodes,
116 const re_node_set *candidates,
118 struct re_backref_cache_entry *bkref_ents,
119 Idx str_idx) internal_function;
120 static reg_errcode_t sift_states_bkref (const re_match_context_t *mctx,
121 re_sift_context_t *sctx,
122 Idx str_idx, const re_node_set *candidates)
124 static reg_errcode_t merge_state_array (const re_dfa_t *dfa,
126 re_dfastate_t **src, Idx num)
128 static re_dfastate_t *find_recover_state (reg_errcode_t *err,
129 re_match_context_t *mctx) internal_function;
130 static re_dfastate_t *transit_state (reg_errcode_t *err,
131 re_match_context_t *mctx,
132 re_dfastate_t *state) internal_function;
133 static re_dfastate_t *merge_state_with_log (reg_errcode_t *err,
134 re_match_context_t *mctx,
135 re_dfastate_t *next_state)
137 static reg_errcode_t check_subexp_matching_top (re_match_context_t *mctx,
138 re_node_set *cur_nodes,
139 Idx str_idx) internal_function;
141 static re_dfastate_t *transit_state_sb (reg_errcode_t *err,
142 re_match_context_t *mctx,
143 re_dfastate_t *pstate)
146 #ifdef RE_ENABLE_I18N
147 static reg_errcode_t transit_state_mb (re_match_context_t *mctx,
148 re_dfastate_t *pstate)
150 #endif /* RE_ENABLE_I18N */
151 static reg_errcode_t transit_state_bkref (re_match_context_t *mctx,
152 const re_node_set *nodes)
154 static reg_errcode_t get_subexp (re_match_context_t *mctx,
155 Idx bkref_node, Idx bkref_str_idx)
157 static reg_errcode_t get_subexp_sub (re_match_context_t *mctx,
158 const re_sub_match_top_t *sub_top,
159 re_sub_match_last_t *sub_last,
160 Idx bkref_node, Idx bkref_str)
162 static Idx find_subexp_node (const re_dfa_t *dfa, const re_node_set *nodes,
163 Idx subexp_idx, int type) internal_function;
164 static reg_errcode_t check_arrival (re_match_context_t *mctx,
165 state_array_t *path, Idx top_node,
166 Idx top_str, Idx last_node, Idx last_str,
167 int type) internal_function;
168 static reg_errcode_t check_arrival_add_next_nodes (re_match_context_t *mctx,
170 re_node_set *cur_nodes,
171 re_node_set *next_nodes)
173 static reg_errcode_t check_arrival_expand_ecl (const re_dfa_t *dfa,
174 re_node_set *cur_nodes,
175 Idx ex_subexp, int type)
177 static reg_errcode_t check_arrival_expand_ecl_sub (const re_dfa_t *dfa,
178 re_node_set *dst_nodes,
179 Idx target, Idx ex_subexp,
180 int type) internal_function;
181 static reg_errcode_t expand_bkref_cache (re_match_context_t *mctx,
182 re_node_set *cur_nodes, Idx cur_str,
183 Idx subexp_num, int type)
185 static bool build_trtable (const re_dfa_t *dfa,
186 re_dfastate_t *state) internal_function;
187 #ifdef RE_ENABLE_I18N
188 static int check_node_accept_bytes (const re_dfa_t *dfa, Idx node_idx,
189 const re_string_t *input, Idx idx)
192 static unsigned int find_collation_sequence_value (const unsigned char *mbs,
196 #endif /* RE_ENABLE_I18N */
197 static Idx group_nodes_into_DFAstates (const re_dfa_t *dfa,
198 const re_dfastate_t *state,
199 re_node_set *states_node,
200 bitset_t *states_ch) internal_function;
201 static bool check_node_accept (const re_match_context_t *mctx,
202 const re_token_t *node, Idx idx)
204 static reg_errcode_t extend_buffers (re_match_context_t *mctx, int min_len)
207 /* Entry point for POSIX code. */
209 /* regexec searches for a given pattern, specified by PREG, in the
212 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
213 'regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
214 least NMATCH elements, and we set them to the offsets of the
215 corresponding matched substrings.
217 EFLAGS specifies "execution flags" which affect matching: if
218 REG_NOTBOL is set, then ^ does not match at the beginning of the
219 string; if REG_NOTEOL is set, then $ does not match at the end.
221 We return 0 if we find a match and REG_NOMATCH if not. */
224 regexec (preg, string, nmatch, pmatch, eflags)
225 const regex_t *_Restrict_ preg;
226 const char *_Restrict_ string;
228 regmatch_t pmatch[_Restrict_arr_];
234 re_dfa_t *dfa = preg->buffer;
237 if (eflags & ~(REG_NOTBOL | REG_NOTEOL | REG_STARTEND))
240 if (eflags & REG_STARTEND)
242 start = pmatch[0].rm_so;
243 length = pmatch[0].rm_eo;
248 length = strlen (string);
251 __libc_lock_lock (dfa->lock);
253 err = re_search_internal (preg, string, length, start, length,
254 length, 0, NULL, eflags);
256 err = re_search_internal (preg, string, length, start, length,
257 length, nmatch, pmatch, eflags);
258 __libc_lock_unlock (dfa->lock);
259 return err != REG_NOERROR;
263 # include <shlib-compat.h>
264 versioned_symbol (libc, __regexec, regexec, GLIBC_2_3_4);
266 # if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_3_4)
267 __typeof__ (__regexec) __compat_regexec;
270 attribute_compat_text_section
271 __compat_regexec (const regex_t *_Restrict_ preg,
272 const char *_Restrict_ string, size_t nmatch,
273 regmatch_t pmatch[], int eflags)
275 return regexec (preg, string, nmatch, pmatch,
276 eflags & (REG_NOTBOL | REG_NOTEOL));
278 compat_symbol (libc, __compat_regexec, regexec, GLIBC_2_0);
282 /* Entry points for GNU code. */
284 /* re_match, re_search, re_match_2, re_search_2
286 The former two functions operate on STRING with length LENGTH,
287 while the later two operate on concatenation of STRING1 and STRING2
288 with lengths LENGTH1 and LENGTH2, respectively.
290 re_match() matches the compiled pattern in BUFP against the string,
291 starting at index START.
293 re_search() first tries matching at index START, then it tries to match
294 starting from index START + 1, and so on. The last start position tried
295 is START + RANGE. (Thus RANGE = 0 forces re_search to operate the same
298 The parameter STOP of re_{match,search}_2 specifies that no match exceeding
299 the first STOP characters of the concatenation of the strings should be
302 If REGS is not NULL, and BUFP->no_sub is not set, the offsets of the match
303 and all groups is stored in REGS. (For the "_2" variants, the offsets are
304 computed relative to the concatenation, not relative to the individual
307 On success, re_match* functions return the length of the match, re_search*
308 return the position of the start of the match. Return value -1 means no
309 match was found and -2 indicates an internal error. */
312 re_match (bufp, string, length, start, regs)
313 struct re_pattern_buffer *bufp;
316 struct re_registers *regs;
318 return re_search_stub (bufp, string, length, start, 0, length, regs, true);
321 weak_alias (__re_match, re_match)
325 re_search (bufp, string, length, start, range, regs)
326 struct re_pattern_buffer *bufp;
330 struct re_registers *regs;
332 return re_search_stub (bufp, string, length, start, range, length, regs,
336 weak_alias (__re_search, re_search)
340 re_match_2 (bufp, string1, length1, string2, length2, start, regs, stop)
341 struct re_pattern_buffer *bufp;
342 const char *string1, *string2;
343 Idx length1, length2, start, stop;
344 struct re_registers *regs;
346 return re_search_2_stub (bufp, string1, length1, string2, length2,
347 start, 0, regs, stop, true);
350 weak_alias (__re_match_2, re_match_2)
354 re_search_2 (bufp, string1, length1, string2, length2, start, range, regs, stop)
355 struct re_pattern_buffer *bufp;
356 const char *string1, *string2;
357 Idx length1, length2, start, stop;
359 struct re_registers *regs;
361 return re_search_2_stub (bufp, string1, length1, string2, length2,
362 start, range, regs, stop, false);
365 weak_alias (__re_search_2, re_search_2)
369 re_search_2_stub (struct re_pattern_buffer *bufp,
370 const char *string1, Idx length1,
371 const char *string2, Idx length2,
372 Idx start, regoff_t range, struct re_registers *regs,
373 Idx stop, bool ret_len)
377 Idx len = length1 + length2;
380 verify (! TYPE_SIGNED (Idx));
381 if (BE (len < length1, 0))
383 /* if (BE (length1 < 0 || length2 < 0 || stop < 0, 0))
386 /* Concatenate the strings. */
390 s = re_malloc (char, len);
392 if (BE (s == NULL, 0))
395 memcpy (__mempcpy (s, string1, length1), string2, length2);
397 memcpy (s, string1, length1);
398 memcpy (s + length1, string2, length2);
407 rval = re_search_stub (bufp, str, len, start, range, stop, regs,
413 /* The parameters have the same meaning as those of re_search.
414 Additional parameters:
415 If RET_LEN is true the length of the match is returned (re_match style);
416 otherwise the position of the match is returned. */
419 re_search_stub (struct re_pattern_buffer *bufp,
420 const char *string, Idx length,
421 Idx start, regoff_t range, Idx stop, struct re_registers *regs,
424 reg_errcode_t result;
430 re_dfa_t *dfa = bufp->buffer;
432 Idx last_start = start + range;
434 /* Check for out-of-range. */
435 verify (! TYPE_SIGNED (Idx));
436 /* if (BE (start < 0, 0))
438 if (BE (start > length, 0))
440 if (BE (length < last_start || (0 <= range && last_start < start), 0))
442 else if (BE (/* last_start < 0 || */ (range < 0 && start <= last_start), 0))
445 __libc_lock_lock (dfa->lock);
447 eflags |= (bufp->not_bol) ? REG_NOTBOL : 0;
448 eflags |= (bufp->not_eol) ? REG_NOTEOL : 0;
450 /* Compile fastmap if we haven't yet. */
451 if (start < last_start && bufp->fastmap != NULL && !bufp->fastmap_accurate)
452 re_compile_fastmap (bufp);
454 if (BE (bufp->no_sub, 0))
457 /* We need at least 1 register. */
460 else if (BE (bufp->regs_allocated == REGS_FIXED
461 && regs->num_regs <= bufp->re_nsub, 0))
463 nregs = regs->num_regs;
464 if (BE (nregs < 1, 0))
466 /* Nothing can be copied to regs. */
472 nregs = bufp->re_nsub + 1;
473 pmatch = re_malloc (regmatch_t, nregs);
474 if (BE (pmatch == NULL, 0))
480 result = re_search_internal (bufp, string, length, start, last_start, stop,
481 nregs, pmatch, eflags);
485 /* I hope we needn't fill their regs with -1's when no match was found. */
486 if (result != REG_NOERROR)
487 rval = result == REG_NOMATCH ? -1 : -2;
488 else if (regs != NULL)
490 /* If caller wants register contents data back, copy them. */
491 bufp->regs_allocated = re_copy_regs (regs, pmatch, nregs,
492 bufp->regs_allocated);
493 if (BE (bufp->regs_allocated == REGS_UNALLOCATED, 0))
497 if (BE (rval == 0, 1))
501 assert (pmatch[0].rm_so == start);
502 rval = pmatch[0].rm_eo - start;
505 rval = pmatch[0].rm_so;
509 __libc_lock_unlock (dfa->lock);
514 re_copy_regs (struct re_registers *regs, regmatch_t *pmatch, Idx nregs,
517 int rval = REGS_REALLOCATE;
519 Idx need_regs = nregs + 1;
520 /* We need one extra element beyond 'num_regs' for the '-1' marker GNU code
523 /* Have the register data arrays been allocated? */
524 if (regs_allocated == REGS_UNALLOCATED)
525 { /* No. So allocate them with malloc. */
526 regs->start = re_malloc (regoff_t, need_regs);
527 if (BE (regs->start == NULL, 0))
528 return REGS_UNALLOCATED;
529 regs->end = re_malloc (regoff_t, need_regs);
530 if (BE (regs->end == NULL, 0))
532 re_free (regs->start);
533 return REGS_UNALLOCATED;
535 regs->num_regs = need_regs;
537 else if (regs_allocated == REGS_REALLOCATE)
538 { /* Yes. If we need more elements than were already
539 allocated, reallocate them. If we need fewer, just
541 if (BE (need_regs > regs->num_regs, 0))
543 regoff_t *new_start = re_realloc (regs->start, regoff_t, need_regs);
545 if (BE (new_start == NULL, 0))
546 return REGS_UNALLOCATED;
547 new_end = re_realloc (regs->end, regoff_t, need_regs);
548 if (BE (new_end == NULL, 0))
551 return REGS_UNALLOCATED;
553 regs->start = new_start;
555 regs->num_regs = need_regs;
560 assert (regs_allocated == REGS_FIXED);
561 /* This function may not be called with REGS_FIXED and nregs too big. */
562 assert (regs->num_regs >= nregs);
567 for (i = 0; i < nregs; ++i)
569 regs->start[i] = pmatch[i].rm_so;
570 regs->end[i] = pmatch[i].rm_eo;
572 for ( ; i < regs->num_regs; ++i)
573 regs->start[i] = regs->end[i] = -1;
578 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
579 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
580 this memory for recording register information. STARTS and ENDS
581 must be allocated using the malloc library routine, and must each
582 be at least NUM_REGS * sizeof (regoff_t) bytes long.
584 If NUM_REGS == 0, then subsequent matches should allocate their own
587 Unless this function is called, the first search or match using
588 PATTERN_BUFFER will allocate its own register data, without
589 freeing the old data. */
592 re_set_registers (bufp, regs, num_regs, starts, ends)
593 struct re_pattern_buffer *bufp;
594 struct re_registers *regs;
595 __re_size_t num_regs;
596 regoff_t *starts, *ends;
600 bufp->regs_allocated = REGS_REALLOCATE;
601 regs->num_regs = num_regs;
602 regs->start = starts;
607 bufp->regs_allocated = REGS_UNALLOCATED;
609 regs->start = regs->end = NULL;
613 weak_alias (__re_set_registers, re_set_registers)
616 /* Entry points compatible with 4.2 BSD regex library. We don't define
617 them unless specifically requested. */
619 #if defined _REGEX_RE_COMP || defined _LIBC
627 return 0 == regexec (&re_comp_buf, s, 0, NULL, 0);
629 #endif /* _REGEX_RE_COMP */
631 /* Internal entry point. */
633 /* Searches for a compiled pattern PREG in the string STRING, whose
634 length is LENGTH. NMATCH, PMATCH, and EFLAGS have the same
635 meaning as with regexec. LAST_START is START + RANGE, where
636 START and RANGE have the same meaning as with re_search.
637 Return REG_NOERROR if we find a match, and REG_NOMATCH if not,
638 otherwise return the error code.
639 Note: We assume front end functions already check ranges.
640 (0 <= LAST_START && LAST_START <= LENGTH) */
643 __attribute_warn_unused_result__
644 re_search_internal (const regex_t *preg,
645 const char *string, Idx length,
646 Idx start, Idx last_start, Idx stop,
647 size_t nmatch, regmatch_t pmatch[],
651 const re_dfa_t *dfa = preg->buffer;
652 Idx left_lim, right_lim;
654 bool fl_longest_match;
657 Idx match_last = REG_MISSING;
661 #if defined _LIBC || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L)
662 re_match_context_t mctx = { .dfa = dfa };
664 re_match_context_t mctx;
666 char *fastmap = ((preg->fastmap != NULL && preg->fastmap_accurate
667 && start != last_start && !preg->can_be_null)
668 ? preg->fastmap : NULL);
669 RE_TRANSLATE_TYPE t = preg->translate;
671 #if !(defined _LIBC || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L))
672 memset (&mctx, '\0', sizeof (re_match_context_t));
676 extra_nmatch = (nmatch > preg->re_nsub) ? nmatch - (preg->re_nsub + 1) : 0;
677 nmatch -= extra_nmatch;
679 /* Check if the DFA haven't been compiled. */
680 if (BE (preg->used == 0 || dfa->init_state == NULL
681 || dfa->init_state_word == NULL || dfa->init_state_nl == NULL
682 || dfa->init_state_begbuf == NULL, 0))
686 /* We assume front-end functions already check them. */
687 assert (0 <= last_start && last_start <= length);
690 /* If initial states with non-begbuf contexts have no elements,
691 the regex must be anchored. If preg->newline_anchor is set,
692 we'll never use init_state_nl, so do not check it. */
693 if (dfa->init_state->nodes.nelem == 0
694 && dfa->init_state_word->nodes.nelem == 0
695 && (dfa->init_state_nl->nodes.nelem == 0
696 || !preg->newline_anchor))
698 if (start != 0 && last_start != 0)
700 start = last_start = 0;
703 /* We must check the longest matching, if nmatch > 0. */
704 fl_longest_match = (nmatch != 0 || dfa->nbackref);
706 err = re_string_allocate (&mctx.input, string, length, dfa->nodes_len + 1,
707 preg->translate, (preg->syntax & RE_ICASE) != 0,
709 if (BE (err != REG_NOERROR, 0))
711 mctx.input.stop = stop;
712 mctx.input.raw_stop = stop;
713 mctx.input.newline_anchor = preg->newline_anchor;
715 err = match_ctx_init (&mctx, eflags, dfa->nbackref * 2);
716 if (BE (err != REG_NOERROR, 0))
719 /* We will log all the DFA states through which the dfa pass,
720 if nmatch > 1, or this dfa has "multibyte node", which is a
721 back-reference or a node which can accept multibyte character or
722 multi character collating element. */
723 if (nmatch > 1 || dfa->has_mb_node)
725 /* Avoid overflow. */
726 if (BE ((MIN (IDX_MAX, SIZE_MAX / sizeof (re_dfastate_t *))
727 <= mctx.input.bufs_len), 0))
733 mctx.state_log = re_malloc (re_dfastate_t *, mctx.input.bufs_len + 1);
734 if (BE (mctx.state_log == NULL, 0))
741 mctx.state_log = NULL;
744 mctx.input.tip_context = (eflags & REG_NOTBOL) ? CONTEXT_BEGBUF
745 : CONTEXT_NEWLINE | CONTEXT_BEGBUF;
747 /* Check incrementally whether the input string matches. */
748 incr = (last_start < start) ? -1 : 1;
749 left_lim = (last_start < start) ? last_start : start;
750 right_lim = (last_start < start) ? start : last_start;
751 sb = dfa->mb_cur_max == 1;
754 ? ((sb || !(preg->syntax & RE_ICASE || t) ? 4 : 0)
755 | (start <= last_start ? 2 : 0)
756 | (t != NULL ? 1 : 0))
759 for (;; match_first += incr)
762 if (match_first < left_lim || right_lim < match_first)
765 /* Advance as rapidly as possible through the string, until we
766 find a plausible place to start matching. This may be done
767 with varying efficiency, so there are various possibilities:
768 only the most common of them are specialized, in order to
769 save on code size. We use a switch statement for speed. */
777 /* Fastmap with single-byte translation, match forward. */
778 while (BE (match_first < right_lim, 1)
779 && !fastmap[t[(unsigned char) string[match_first]]])
781 goto forward_match_found_start_or_reached_end;
784 /* Fastmap without translation, match forward. */
785 while (BE (match_first < right_lim, 1)
786 && !fastmap[(unsigned char) string[match_first]])
789 forward_match_found_start_or_reached_end:
790 if (BE (match_first == right_lim, 0))
792 ch = match_first >= length
793 ? 0 : (unsigned char) string[match_first];
794 if (!fastmap[t ? t[ch] : ch])
801 /* Fastmap without multi-byte translation, match backwards. */
802 while (match_first >= left_lim)
804 ch = match_first >= length
805 ? 0 : (unsigned char) string[match_first];
806 if (fastmap[t ? t[ch] : ch])
810 if (match_first < left_lim)
815 /* In this case, we can't determine easily the current byte,
816 since it might be a component byte of a multibyte
817 character. Then we use the constructed buffer instead. */
820 /* If MATCH_FIRST is out of the valid range, reconstruct the
822 __re_size_t offset = match_first - mctx.input.raw_mbs_idx;
823 if (BE (offset >= (__re_size_t) mctx.input.valid_raw_len, 0))
825 err = re_string_reconstruct (&mctx.input, match_first,
827 if (BE (err != REG_NOERROR, 0))
830 offset = match_first - mctx.input.raw_mbs_idx;
832 /* If MATCH_FIRST is out of the buffer, leave it as '\0'.
833 Note that MATCH_FIRST must not be smaller than 0. */
834 ch = (match_first >= length
835 ? 0 : re_string_byte_at (&mctx.input, offset));
839 if (match_first < left_lim || match_first > right_lim)
848 /* Reconstruct the buffers so that the matcher can assume that
849 the matching starts from the beginning of the buffer. */
850 err = re_string_reconstruct (&mctx.input, match_first, eflags);
851 if (BE (err != REG_NOERROR, 0))
854 #ifdef RE_ENABLE_I18N
855 /* Don't consider this char as a possible match start if it part,
856 yet isn't the head, of a multibyte character. */
857 if (!sb && !re_string_first_byte (&mctx.input, 0))
861 /* It seems to be appropriate one, then use the matcher. */
862 /* We assume that the matching starts from 0. */
863 mctx.state_log_top = mctx.nbkref_ents = mctx.max_mb_elem_len = 0;
864 match_last = check_matching (&mctx, fl_longest_match,
865 start <= last_start ? &match_first : NULL);
866 if (match_last != REG_MISSING)
868 if (BE (match_last == REG_ERROR, 0))
875 mctx.match_last = match_last;
876 if ((!preg->no_sub && nmatch > 1) || dfa->nbackref)
878 re_dfastate_t *pstate = mctx.state_log[match_last];
879 mctx.last_node = check_halt_state_context (&mctx, pstate,
882 if ((!preg->no_sub && nmatch > 1 && dfa->has_plural_match)
885 err = prune_impossible_nodes (&mctx);
886 if (err == REG_NOERROR)
888 if (BE (err != REG_NOMATCH, 0))
890 match_last = REG_MISSING;
893 break; /* We found a match. */
897 match_ctx_clean (&mctx);
901 assert (match_last != REG_MISSING);
902 assert (err == REG_NOERROR);
905 /* Set pmatch[] if we need. */
910 /* Initialize registers. */
911 for (reg_idx = 1; reg_idx < nmatch; ++reg_idx)
912 pmatch[reg_idx].rm_so = pmatch[reg_idx].rm_eo = -1;
914 /* Set the points where matching start/end. */
916 pmatch[0].rm_eo = mctx.match_last;
917 /* FIXME: This function should fail if mctx.match_last exceeds
918 the maximum possible regoff_t value. We need a new error
919 code REG_OVERFLOW. */
921 if (!preg->no_sub && nmatch > 1)
923 err = set_regs (preg, &mctx, nmatch, pmatch,
924 dfa->has_plural_match && dfa->nbackref > 0);
925 if (BE (err != REG_NOERROR, 0))
929 /* At last, add the offset to each register, since we slid
930 the buffers so that we could assume that the matching starts
932 for (reg_idx = 0; reg_idx < nmatch; ++reg_idx)
933 if (pmatch[reg_idx].rm_so != -1)
935 #ifdef RE_ENABLE_I18N
936 if (BE (mctx.input.offsets_needed != 0, 0))
938 pmatch[reg_idx].rm_so =
939 (pmatch[reg_idx].rm_so == mctx.input.valid_len
940 ? mctx.input.valid_raw_len
941 : mctx.input.offsets[pmatch[reg_idx].rm_so]);
942 pmatch[reg_idx].rm_eo =
943 (pmatch[reg_idx].rm_eo == mctx.input.valid_len
944 ? mctx.input.valid_raw_len
945 : mctx.input.offsets[pmatch[reg_idx].rm_eo]);
948 assert (mctx.input.offsets_needed == 0);
950 pmatch[reg_idx].rm_so += match_first;
951 pmatch[reg_idx].rm_eo += match_first;
953 for (reg_idx = 0; reg_idx < extra_nmatch; ++reg_idx)
955 pmatch[nmatch + reg_idx].rm_so = -1;
956 pmatch[nmatch + reg_idx].rm_eo = -1;
960 for (reg_idx = 0; reg_idx + 1 < nmatch; reg_idx++)
961 if (dfa->subexp_map[reg_idx] != reg_idx)
963 pmatch[reg_idx + 1].rm_so
964 = pmatch[dfa->subexp_map[reg_idx] + 1].rm_so;
965 pmatch[reg_idx + 1].rm_eo
966 = pmatch[dfa->subexp_map[reg_idx] + 1].rm_eo;
971 re_free (mctx.state_log);
973 match_ctx_free (&mctx);
974 re_string_destruct (&mctx.input);
979 __attribute_warn_unused_result__
980 prune_impossible_nodes (re_match_context_t *mctx)
982 const re_dfa_t *const dfa = mctx->dfa;
983 Idx halt_node, match_last;
985 re_dfastate_t **sifted_states;
986 re_dfastate_t **lim_states = NULL;
987 re_sift_context_t sctx;
989 assert (mctx->state_log != NULL);
991 match_last = mctx->match_last;
992 halt_node = mctx->last_node;
994 /* Avoid overflow. */
995 if (BE (MIN (IDX_MAX, SIZE_MAX / sizeof (re_dfastate_t *)) <= match_last, 0))
998 sifted_states = re_malloc (re_dfastate_t *, match_last + 1);
999 if (BE (sifted_states == NULL, 0))
1006 lim_states = re_malloc (re_dfastate_t *, match_last + 1);
1007 if (BE (lim_states == NULL, 0))
1014 memset (lim_states, '\0',
1015 sizeof (re_dfastate_t *) * (match_last + 1));
1016 sift_ctx_init (&sctx, sifted_states, lim_states, halt_node,
1018 ret = sift_states_backward (mctx, &sctx);
1019 re_node_set_free (&sctx.limits);
1020 if (BE (ret != REG_NOERROR, 0))
1022 if (sifted_states[0] != NULL || lim_states[0] != NULL)
1027 if (! REG_VALID_INDEX (match_last))
1032 } while (mctx->state_log[match_last] == NULL
1033 || !mctx->state_log[match_last]->halt);
1034 halt_node = check_halt_state_context (mctx,
1035 mctx->state_log[match_last],
1038 ret = merge_state_array (dfa, sifted_states, lim_states,
1040 re_free (lim_states);
1042 if (BE (ret != REG_NOERROR, 0))
1047 sift_ctx_init (&sctx, sifted_states, lim_states, halt_node, match_last);
1048 ret = sift_states_backward (mctx, &sctx);
1049 re_node_set_free (&sctx.limits);
1050 if (BE (ret != REG_NOERROR, 0))
1052 if (sifted_states[0] == NULL)
1058 re_free (mctx->state_log);
1059 mctx->state_log = sifted_states;
1060 sifted_states = NULL;
1061 mctx->last_node = halt_node;
1062 mctx->match_last = match_last;
1065 re_free (sifted_states);
1066 re_free (lim_states);
1070 /* Acquire an initial state and return it.
1071 We must select appropriate initial state depending on the context,
1072 since initial states may have constraints like "\<", "^", etc.. */
1074 static inline re_dfastate_t *
1075 __attribute__ ((always_inline)) internal_function
1076 acquire_init_state_context (reg_errcode_t *err, const re_match_context_t *mctx,
1079 const re_dfa_t *const dfa = mctx->dfa;
1080 if (dfa->init_state->has_constraint)
1082 unsigned int context;
1083 context = re_string_context_at (&mctx->input, idx - 1, mctx->eflags);
1084 if (IS_WORD_CONTEXT (context))
1085 return dfa->init_state_word;
1086 else if (IS_ORDINARY_CONTEXT (context))
1087 return dfa->init_state;
1088 else if (IS_BEGBUF_CONTEXT (context) && IS_NEWLINE_CONTEXT (context))
1089 return dfa->init_state_begbuf;
1090 else if (IS_NEWLINE_CONTEXT (context))
1091 return dfa->init_state_nl;
1092 else if (IS_BEGBUF_CONTEXT (context))
1094 /* It is relatively rare case, then calculate on demand. */
1095 return re_acquire_state_context (err, dfa,
1096 dfa->init_state->entrance_nodes,
1100 /* Must not happen? */
1101 return dfa->init_state;
1104 return dfa->init_state;
1107 /* Check whether the regular expression match input string INPUT or not,
1108 and return the index where the matching end. Return REG_MISSING if
1109 there is no match, and return REG_ERROR in case of an error.
1110 FL_LONGEST_MATCH means we want the POSIX longest matching.
1111 If P_MATCH_FIRST is not NULL, and the match fails, it is set to the
1112 next place where we may want to try matching.
1113 Note that the matcher assumes that the matching starts from the current
1114 index of the buffer. */
1117 internal_function __attribute_warn_unused_result__
1118 check_matching (re_match_context_t *mctx, bool fl_longest_match,
1121 const re_dfa_t *const dfa = mctx->dfa;
1124 Idx match_last = REG_MISSING;
1125 Idx cur_str_idx = re_string_cur_idx (&mctx->input);
1126 re_dfastate_t *cur_state;
1127 bool at_init_state = p_match_first != NULL;
1128 Idx next_start_idx = cur_str_idx;
1131 cur_state = acquire_init_state_context (&err, mctx, cur_str_idx);
1132 /* An initial state must not be NULL (invalid). */
1133 if (BE (cur_state == NULL, 0))
1135 assert (err == REG_ESPACE);
1139 if (mctx->state_log != NULL)
1141 mctx->state_log[cur_str_idx] = cur_state;
1143 /* Check OP_OPEN_SUBEXP in the initial state in case that we use them
1144 later. E.g. Processing back references. */
1145 if (BE (dfa->nbackref, 0))
1147 at_init_state = false;
1148 err = check_subexp_matching_top (mctx, &cur_state->nodes, 0);
1149 if (BE (err != REG_NOERROR, 0))
1152 if (cur_state->has_backref)
1154 err = transit_state_bkref (mctx, &cur_state->nodes);
1155 if (BE (err != REG_NOERROR, 0))
1161 /* If the RE accepts NULL string. */
1162 if (BE (cur_state->halt, 0))
1164 if (!cur_state->has_constraint
1165 || check_halt_state_context (mctx, cur_state, cur_str_idx))
1167 if (!fl_longest_match)
1171 match_last = cur_str_idx;
1177 while (!re_string_eoi (&mctx->input))
1179 re_dfastate_t *old_state = cur_state;
1180 Idx next_char_idx = re_string_cur_idx (&mctx->input) + 1;
1182 if ((BE (next_char_idx >= mctx->input.bufs_len, 0)
1183 && mctx->input.bufs_len < mctx->input.len)
1184 || (BE (next_char_idx >= mctx->input.valid_len, 0)
1185 && mctx->input.valid_len < mctx->input.len))
1187 err = extend_buffers (mctx, next_char_idx + 1);
1188 if (BE (err != REG_NOERROR, 0))
1190 assert (err == REG_ESPACE);
1195 cur_state = transit_state (&err, mctx, cur_state);
1196 if (mctx->state_log != NULL)
1197 cur_state = merge_state_with_log (&err, mctx, cur_state);
1199 if (cur_state == NULL)
1201 /* Reached the invalid state or an error. Try to recover a valid
1202 state using the state log, if available and if we have not
1203 already found a valid (even if not the longest) match. */
1204 if (BE (err != REG_NOERROR, 0))
1207 if (mctx->state_log == NULL
1208 || (match && !fl_longest_match)
1209 || (cur_state = find_recover_state (&err, mctx)) == NULL)
1213 if (BE (at_init_state, 0))
1215 if (old_state == cur_state)
1216 next_start_idx = next_char_idx;
1218 at_init_state = false;
1221 if (cur_state->halt)
1223 /* Reached a halt state.
1224 Check the halt state can satisfy the current context. */
1225 if (!cur_state->has_constraint
1226 || check_halt_state_context (mctx, cur_state,
1227 re_string_cur_idx (&mctx->input)))
1229 /* We found an appropriate halt state. */
1230 match_last = re_string_cur_idx (&mctx->input);
1233 /* We found a match, do not modify match_first below. */
1234 p_match_first = NULL;
1235 if (!fl_longest_match)
1242 *p_match_first += next_start_idx;
1247 /* Check NODE match the current context. */
1251 check_halt_node_context (const re_dfa_t *dfa, Idx node, unsigned int context)
1253 re_token_type_t type = dfa->nodes[node].type;
1254 unsigned int constraint = dfa->nodes[node].constraint;
1255 if (type != END_OF_RE)
1259 if (NOT_SATISFY_NEXT_CONSTRAINT (constraint, context))
1264 /* Check the halt state STATE match the current context.
1265 Return 0 if not match, if the node, STATE has, is a halt node and
1266 match the context, return the node. */
1270 check_halt_state_context (const re_match_context_t *mctx,
1271 const re_dfastate_t *state, Idx idx)
1274 unsigned int context;
1276 assert (state->halt);
1278 context = re_string_context_at (&mctx->input, idx, mctx->eflags);
1279 for (i = 0; i < state->nodes.nelem; ++i)
1280 if (check_halt_node_context (mctx->dfa, state->nodes.elems[i], context))
1281 return state->nodes.elems[i];
1285 /* Compute the next node to which "NFA" transit from NODE("NFA" is a NFA
1286 corresponding to the DFA).
1287 Return the destination node, and update EPS_VIA_NODES;
1288 return REG_MISSING in case of errors. */
1292 proceed_next_node (const re_match_context_t *mctx, Idx nregs, regmatch_t *regs,
1293 Idx *pidx, Idx node, re_node_set *eps_via_nodes,
1294 struct re_fail_stack_t *fs)
1296 const re_dfa_t *const dfa = mctx->dfa;
1299 if (IS_EPSILON_NODE (dfa->nodes[node].type))
1301 re_node_set *cur_nodes = &mctx->state_log[*pidx]->nodes;
1302 re_node_set *edests = &dfa->edests[node];
1304 ok = re_node_set_insert (eps_via_nodes, node);
1307 /* Pick up a valid destination, or return REG_MISSING if none
1309 for (dest_node = REG_MISSING, i = 0; i < edests->nelem; ++i)
1311 Idx candidate = edests->elems[i];
1312 if (!re_node_set_contains (cur_nodes, candidate))
1314 if (dest_node == REG_MISSING)
1315 dest_node = candidate;
1319 /* In order to avoid infinite loop like "(a*)*", return the second
1320 epsilon-transition if the first was already considered. */
1321 if (re_node_set_contains (eps_via_nodes, dest_node))
1324 /* Otherwise, push the second epsilon-transition on the fail stack. */
1326 && push_fail_stack (fs, *pidx, candidate, nregs, regs,
1330 /* We know we are going to exit. */
1339 re_token_type_t type = dfa->nodes[node].type;
1341 #ifdef RE_ENABLE_I18N
1342 if (dfa->nodes[node].accept_mb)
1343 naccepted = check_node_accept_bytes (dfa, node, &mctx->input, *pidx);
1345 #endif /* RE_ENABLE_I18N */
1346 if (type == OP_BACK_REF)
1348 Idx subexp_idx = dfa->nodes[node].opr.idx + 1;
1349 naccepted = regs[subexp_idx].rm_eo - regs[subexp_idx].rm_so;
1352 if (regs[subexp_idx].rm_so == -1 || regs[subexp_idx].rm_eo == -1)
1356 char *buf = (char *) re_string_get_buffer (&mctx->input);
1357 if (memcmp (buf + regs[subexp_idx].rm_so, buf + *pidx,
1366 ok = re_node_set_insert (eps_via_nodes, node);
1369 dest_node = dfa->edests[node].elems[0];
1370 if (re_node_set_contains (&mctx->state_log[*pidx]->nodes,
1377 || check_node_accept (mctx, dfa->nodes + node, *pidx))
1379 Idx dest_node = dfa->nexts[node];
1380 *pidx = (naccepted == 0) ? *pidx + 1 : *pidx + naccepted;
1381 if (fs && (*pidx > mctx->match_last || mctx->state_log[*pidx] == NULL
1382 || !re_node_set_contains (&mctx->state_log[*pidx]->nodes,
1385 re_node_set_empty (eps_via_nodes);
1392 static reg_errcode_t
1393 internal_function __attribute_warn_unused_result__
1394 push_fail_stack (struct re_fail_stack_t *fs, Idx str_idx, Idx dest_node,
1395 Idx nregs, regmatch_t *regs, re_node_set *eps_via_nodes)
1398 Idx num = fs->num++;
1399 if (fs->num == fs->alloc)
1401 struct re_fail_stack_ent_t *new_array;
1402 new_array = realloc (fs->stack, (sizeof (struct re_fail_stack_ent_t)
1404 if (new_array == NULL)
1407 fs->stack = new_array;
1409 fs->stack[num].idx = str_idx;
1410 fs->stack[num].node = dest_node;
1411 fs->stack[num].regs = re_malloc (regmatch_t, nregs);
1412 if (fs->stack[num].regs == NULL)
1414 memcpy (fs->stack[num].regs, regs, sizeof (regmatch_t) * nregs);
1415 err = re_node_set_init_copy (&fs->stack[num].eps_via_nodes, eps_via_nodes);
1421 pop_fail_stack (struct re_fail_stack_t *fs, Idx *pidx, Idx nregs,
1422 regmatch_t *regs, re_node_set *eps_via_nodes)
1424 Idx num = --fs->num;
1425 assert (REG_VALID_INDEX (num));
1426 *pidx = fs->stack[num].idx;
1427 memcpy (regs, fs->stack[num].regs, sizeof (regmatch_t) * nregs);
1428 re_node_set_free (eps_via_nodes);
1429 re_free (fs->stack[num].regs);
1430 *eps_via_nodes = fs->stack[num].eps_via_nodes;
1431 return fs->stack[num].node;
1434 /* Set the positions where the subexpressions are starts/ends to registers
1436 Note: We assume that pmatch[0] is already set, and
1437 pmatch[i].rm_so == pmatch[i].rm_eo == -1 for 0 < i < nmatch. */
1439 static reg_errcode_t
1440 internal_function __attribute_warn_unused_result__
1441 set_regs (const regex_t *preg, const re_match_context_t *mctx, size_t nmatch,
1442 regmatch_t *pmatch, bool fl_backtrack)
1444 const re_dfa_t *dfa = preg->buffer;
1446 re_node_set eps_via_nodes;
1447 struct re_fail_stack_t *fs;
1448 struct re_fail_stack_t fs_body = { 0, 2, NULL };
1449 regmatch_t *prev_idx_match;
1450 bool prev_idx_match_malloced = false;
1453 assert (nmatch > 1);
1454 assert (mctx->state_log != NULL);
1459 fs->stack = re_malloc (struct re_fail_stack_ent_t, fs->alloc);
1460 if (fs->stack == NULL)
1466 cur_node = dfa->init_node;
1467 re_node_set_init_empty (&eps_via_nodes);
1469 if (__libc_use_alloca (nmatch * sizeof (regmatch_t)))
1470 prev_idx_match = (regmatch_t *) alloca (nmatch * sizeof (regmatch_t));
1473 prev_idx_match = re_malloc (regmatch_t, nmatch);
1474 if (prev_idx_match == NULL)
1476 free_fail_stack_return (fs);
1479 prev_idx_match_malloced = true;
1481 memcpy (prev_idx_match, pmatch, sizeof (regmatch_t) * nmatch);
1483 for (idx = pmatch[0].rm_so; idx <= pmatch[0].rm_eo ;)
1485 update_regs (dfa, pmatch, prev_idx_match, cur_node, idx, nmatch);
1487 if (idx == pmatch[0].rm_eo && cur_node == mctx->last_node)
1492 for (reg_idx = 0; reg_idx < nmatch; ++reg_idx)
1493 if (pmatch[reg_idx].rm_so > -1 && pmatch[reg_idx].rm_eo == -1)
1495 if (reg_idx == nmatch)
1497 re_node_set_free (&eps_via_nodes);
1498 if (prev_idx_match_malloced)
1499 re_free (prev_idx_match);
1500 return free_fail_stack_return (fs);
1502 cur_node = pop_fail_stack (fs, &idx, nmatch, pmatch,
1507 re_node_set_free (&eps_via_nodes);
1508 if (prev_idx_match_malloced)
1509 re_free (prev_idx_match);
1514 /* Proceed to next node. */
1515 cur_node = proceed_next_node (mctx, nmatch, pmatch, &idx, cur_node,
1516 &eps_via_nodes, fs);
1518 if (BE (! REG_VALID_INDEX (cur_node), 0))
1520 if (BE (cur_node == REG_ERROR, 0))
1522 re_node_set_free (&eps_via_nodes);
1523 if (prev_idx_match_malloced)
1524 re_free (prev_idx_match);
1525 free_fail_stack_return (fs);
1529 cur_node = pop_fail_stack (fs, &idx, nmatch, pmatch,
1533 re_node_set_free (&eps_via_nodes);
1534 if (prev_idx_match_malloced)
1535 re_free (prev_idx_match);
1540 re_node_set_free (&eps_via_nodes);
1541 if (prev_idx_match_malloced)
1542 re_free (prev_idx_match);
1543 return free_fail_stack_return (fs);
1546 static reg_errcode_t
1548 free_fail_stack_return (struct re_fail_stack_t *fs)
1553 for (fs_idx = 0; fs_idx < fs->num; ++fs_idx)
1555 re_node_set_free (&fs->stack[fs_idx].eps_via_nodes);
1556 re_free (fs->stack[fs_idx].regs);
1558 re_free (fs->stack);
1565 update_regs (const re_dfa_t *dfa, regmatch_t *pmatch,
1566 regmatch_t *prev_idx_match, Idx cur_node, Idx cur_idx, Idx nmatch)
1568 int type = dfa->nodes[cur_node].type;
1569 if (type == OP_OPEN_SUBEXP)
1571 Idx reg_num = dfa->nodes[cur_node].opr.idx + 1;
1573 /* We are at the first node of this sub expression. */
1574 if (reg_num < nmatch)
1576 pmatch[reg_num].rm_so = cur_idx;
1577 pmatch[reg_num].rm_eo = -1;
1580 else if (type == OP_CLOSE_SUBEXP)
1582 Idx reg_num = dfa->nodes[cur_node].opr.idx + 1;
1583 if (reg_num < nmatch)
1585 /* We are at the last node of this sub expression. */
1586 if (pmatch[reg_num].rm_so < cur_idx)
1588 pmatch[reg_num].rm_eo = cur_idx;
1589 /* This is a non-empty match or we are not inside an optional
1590 subexpression. Accept this right away. */
1591 memcpy (prev_idx_match, pmatch, sizeof (regmatch_t) * nmatch);
1595 if (dfa->nodes[cur_node].opt_subexp
1596 && prev_idx_match[reg_num].rm_so != -1)
1597 /* We transited through an empty match for an optional
1598 subexpression, like (a?)*, and this is not the subexp's
1599 first match. Copy back the old content of the registers
1600 so that matches of an inner subexpression are undone as
1601 well, like in ((a?))*. */
1602 memcpy (pmatch, prev_idx_match, sizeof (regmatch_t) * nmatch);
1604 /* We completed a subexpression, but it may be part of
1605 an optional one, so do not update PREV_IDX_MATCH. */
1606 pmatch[reg_num].rm_eo = cur_idx;
1612 /* This function checks the STATE_LOG from the SCTX->last_str_idx to 0
1613 and sift the nodes in each states according to the following rules.
1614 Updated state_log will be wrote to STATE_LOG.
1616 Rules: We throw away the Node 'a' in the STATE_LOG[STR_IDX] if...
1617 1. When STR_IDX == MATCH_LAST(the last index in the state_log):
1618 If 'a' isn't the LAST_NODE and 'a' can't epsilon transit to
1619 the LAST_NODE, we throw away the node 'a'.
1620 2. When 0 <= STR_IDX < MATCH_LAST and 'a' accepts
1621 string 's' and transit to 'b':
1622 i. If 'b' isn't in the STATE_LOG[STR_IDX+strlen('s')], we throw
1624 ii. If 'b' is in the STATE_LOG[STR_IDX+strlen('s')] but 'b' is
1625 thrown away, we throw away the node 'a'.
1626 3. When 0 <= STR_IDX < MATCH_LAST and 'a' epsilon transit to 'b':
1627 i. If 'b' isn't in the STATE_LOG[STR_IDX], we throw away the
1629 ii. If 'b' is in the STATE_LOG[STR_IDX] but 'b' is thrown away,
1630 we throw away the node 'a'. */
1632 #define STATE_NODE_CONTAINS(state,node) \
1633 ((state) != NULL && re_node_set_contains (&(state)->nodes, node))
1635 static reg_errcode_t
1637 sift_states_backward (const re_match_context_t *mctx, re_sift_context_t *sctx)
1641 Idx str_idx = sctx->last_str_idx;
1642 re_node_set cur_dest;
1645 assert (mctx->state_log != NULL && mctx->state_log[str_idx] != NULL);
1648 /* Build sifted state_log[str_idx]. It has the nodes which can epsilon
1649 transit to the last_node and the last_node itself. */
1650 err = re_node_set_init_1 (&cur_dest, sctx->last_node);
1651 if (BE (err != REG_NOERROR, 0))
1653 err = update_cur_sifted_state (mctx, sctx, str_idx, &cur_dest);
1654 if (BE (err != REG_NOERROR, 0))
1657 /* Then check each states in the state_log. */
1660 /* Update counters. */
1661 null_cnt = (sctx->sifted_states[str_idx] == NULL) ? null_cnt + 1 : 0;
1662 if (null_cnt > mctx->max_mb_elem_len)
1664 memset (sctx->sifted_states, '\0',
1665 sizeof (re_dfastate_t *) * str_idx);
1666 re_node_set_free (&cur_dest);
1669 re_node_set_empty (&cur_dest);
1672 if (mctx->state_log[str_idx])
1674 err = build_sifted_states (mctx, sctx, str_idx, &cur_dest);
1675 if (BE (err != REG_NOERROR, 0))
1679 /* Add all the nodes which satisfy the following conditions:
1680 - It can epsilon transit to a node in CUR_DEST.
1682 And update state_log. */
1683 err = update_cur_sifted_state (mctx, sctx, str_idx, &cur_dest);
1684 if (BE (err != REG_NOERROR, 0))
1689 re_node_set_free (&cur_dest);
1693 static reg_errcode_t
1694 internal_function __attribute_warn_unused_result__
1695 build_sifted_states (const re_match_context_t *mctx, re_sift_context_t *sctx,
1696 Idx str_idx, re_node_set *cur_dest)
1698 const re_dfa_t *const dfa = mctx->dfa;
1699 const re_node_set *cur_src = &mctx->state_log[str_idx]->non_eps_nodes;
1702 /* Then build the next sifted state.
1703 We build the next sifted state on 'cur_dest', and update
1704 'sifted_states[str_idx]' with 'cur_dest'.
1706 'cur_dest' is the sifted state from 'state_log[str_idx + 1]'.
1707 'cur_src' points the node_set of the old 'state_log[str_idx]'
1708 (with the epsilon nodes pre-filtered out). */
1709 for (i = 0; i < cur_src->nelem; i++)
1711 Idx prev_node = cur_src->elems[i];
1716 re_token_type_t type = dfa->nodes[prev_node].type;
1717 assert (!IS_EPSILON_NODE (type));
1719 #ifdef RE_ENABLE_I18N
1720 /* If the node may accept "multi byte". */
1721 if (dfa->nodes[prev_node].accept_mb)
1722 naccepted = sift_states_iter_mb (mctx, sctx, prev_node,
1723 str_idx, sctx->last_str_idx);
1724 #endif /* RE_ENABLE_I18N */
1726 /* We don't check backreferences here.
1727 See update_cur_sifted_state(). */
1729 && check_node_accept (mctx, dfa->nodes + prev_node, str_idx)
1730 && STATE_NODE_CONTAINS (sctx->sifted_states[str_idx + 1],
1731 dfa->nexts[prev_node]))
1737 if (sctx->limits.nelem)
1739 Idx to_idx = str_idx + naccepted;
1740 if (check_dst_limits (mctx, &sctx->limits,
1741 dfa->nexts[prev_node], to_idx,
1742 prev_node, str_idx))
1745 ok = re_node_set_insert (cur_dest, prev_node);
1753 /* Helper functions. */
1755 static reg_errcode_t
1757 clean_state_log_if_needed (re_match_context_t *mctx, Idx next_state_log_idx)
1759 Idx top = mctx->state_log_top;
1761 if ((next_state_log_idx >= mctx->input.bufs_len
1762 && mctx->input.bufs_len < mctx->input.len)
1763 || (next_state_log_idx >= mctx->input.valid_len
1764 && mctx->input.valid_len < mctx->input.len))
1767 err = extend_buffers (mctx, next_state_log_idx + 1);
1768 if (BE (err != REG_NOERROR, 0))
1772 if (top < next_state_log_idx)
1774 memset (mctx->state_log + top + 1, '\0',
1775 sizeof (re_dfastate_t *) * (next_state_log_idx - top));
1776 mctx->state_log_top = next_state_log_idx;
1781 static reg_errcode_t
1783 merge_state_array (const re_dfa_t *dfa, re_dfastate_t **dst,
1784 re_dfastate_t **src, Idx num)
1788 for (st_idx = 0; st_idx < num; ++st_idx)
1790 if (dst[st_idx] == NULL)
1791 dst[st_idx] = src[st_idx];
1792 else if (src[st_idx] != NULL)
1794 re_node_set merged_set;
1795 err = re_node_set_init_union (&merged_set, &dst[st_idx]->nodes,
1796 &src[st_idx]->nodes);
1797 if (BE (err != REG_NOERROR, 0))
1799 dst[st_idx] = re_acquire_state (&err, dfa, &merged_set);
1800 re_node_set_free (&merged_set);
1801 if (BE (err != REG_NOERROR, 0))
1808 static reg_errcode_t
1810 update_cur_sifted_state (const re_match_context_t *mctx,
1811 re_sift_context_t *sctx, Idx str_idx,
1812 re_node_set *dest_nodes)
1814 const re_dfa_t *const dfa = mctx->dfa;
1815 reg_errcode_t err = REG_NOERROR;
1816 const re_node_set *candidates;
1817 candidates = ((mctx->state_log[str_idx] == NULL) ? NULL
1818 : &mctx->state_log[str_idx]->nodes);
1820 if (dest_nodes->nelem == 0)
1821 sctx->sifted_states[str_idx] = NULL;
1826 /* At first, add the nodes which can epsilon transit to a node in
1828 err = add_epsilon_src_nodes (dfa, dest_nodes, candidates);
1829 if (BE (err != REG_NOERROR, 0))
1832 /* Then, check the limitations in the current sift_context. */
1833 if (sctx->limits.nelem)
1835 err = check_subexp_limits (dfa, dest_nodes, candidates, &sctx->limits,
1836 mctx->bkref_ents, str_idx);
1837 if (BE (err != REG_NOERROR, 0))
1842 sctx->sifted_states[str_idx] = re_acquire_state (&err, dfa, dest_nodes);
1843 if (BE (err != REG_NOERROR, 0))
1847 if (candidates && mctx->state_log[str_idx]->has_backref)
1849 err = sift_states_bkref (mctx, sctx, str_idx, candidates);
1850 if (BE (err != REG_NOERROR, 0))
1856 static reg_errcode_t
1857 internal_function __attribute_warn_unused_result__
1858 add_epsilon_src_nodes (const re_dfa_t *dfa, re_node_set *dest_nodes,
1859 const re_node_set *candidates)
1861 reg_errcode_t err = REG_NOERROR;
1864 re_dfastate_t *state = re_acquire_state (&err, dfa, dest_nodes);
1865 if (BE (err != REG_NOERROR, 0))
1868 if (!state->inveclosure.alloc)
1870 err = re_node_set_alloc (&state->inveclosure, dest_nodes->nelem);
1871 if (BE (err != REG_NOERROR, 0))
1873 for (i = 0; i < dest_nodes->nelem; i++)
1875 err = re_node_set_merge (&state->inveclosure,
1876 dfa->inveclosures + dest_nodes->elems[i]);
1877 if (BE (err != REG_NOERROR, 0))
1881 return re_node_set_add_intersect (dest_nodes, candidates,
1882 &state->inveclosure);
1885 static reg_errcode_t
1887 sub_epsilon_src_nodes (const re_dfa_t *dfa, Idx node, re_node_set *dest_nodes,
1888 const re_node_set *candidates)
1892 re_node_set *inv_eclosure = dfa->inveclosures + node;
1893 re_node_set except_nodes;
1894 re_node_set_init_empty (&except_nodes);
1895 for (ecl_idx = 0; ecl_idx < inv_eclosure->nelem; ++ecl_idx)
1897 Idx cur_node = inv_eclosure->elems[ecl_idx];
1898 if (cur_node == node)
1900 if (IS_EPSILON_NODE (dfa->nodes[cur_node].type))
1902 Idx edst1 = dfa->edests[cur_node].elems[0];
1903 Idx edst2 = ((dfa->edests[cur_node].nelem > 1)
1904 ? dfa->edests[cur_node].elems[1] : REG_MISSING);
1905 if ((!re_node_set_contains (inv_eclosure, edst1)
1906 && re_node_set_contains (dest_nodes, edst1))
1907 || (REG_VALID_NONZERO_INDEX (edst2)
1908 && !re_node_set_contains (inv_eclosure, edst2)
1909 && re_node_set_contains (dest_nodes, edst2)))
1911 err = re_node_set_add_intersect (&except_nodes, candidates,
1912 dfa->inveclosures + cur_node);
1913 if (BE (err != REG_NOERROR, 0))
1915 re_node_set_free (&except_nodes);
1921 for (ecl_idx = 0; ecl_idx < inv_eclosure->nelem; ++ecl_idx)
1923 Idx cur_node = inv_eclosure->elems[ecl_idx];
1924 if (!re_node_set_contains (&except_nodes, cur_node))
1926 Idx idx = re_node_set_contains (dest_nodes, cur_node) - 1;
1927 re_node_set_remove_at (dest_nodes, idx);
1930 re_node_set_free (&except_nodes);
1936 check_dst_limits (const re_match_context_t *mctx, const re_node_set *limits,
1937 Idx dst_node, Idx dst_idx, Idx src_node, Idx src_idx)
1939 const re_dfa_t *const dfa = mctx->dfa;
1940 Idx lim_idx, src_pos, dst_pos;
1942 Idx dst_bkref_idx = search_cur_bkref_entry (mctx, dst_idx);
1943 Idx src_bkref_idx = search_cur_bkref_entry (mctx, src_idx);
1944 for (lim_idx = 0; lim_idx < limits->nelem; ++lim_idx)
1947 struct re_backref_cache_entry *ent;
1948 ent = mctx->bkref_ents + limits->elems[lim_idx];
1949 subexp_idx = dfa->nodes[ent->node].opr.idx;
1951 dst_pos = check_dst_limits_calc_pos (mctx, limits->elems[lim_idx],
1952 subexp_idx, dst_node, dst_idx,
1954 src_pos = check_dst_limits_calc_pos (mctx, limits->elems[lim_idx],
1955 subexp_idx, src_node, src_idx,
1959 <src> <dst> ( <subexp> )
1960 ( <subexp> ) <src> <dst>
1961 ( <subexp1> <src> <subexp2> <dst> <subexp3> ) */
1962 if (src_pos == dst_pos)
1963 continue; /* This is unrelated limitation. */
1972 check_dst_limits_calc_pos_1 (const re_match_context_t *mctx, int boundaries,
1973 Idx subexp_idx, Idx from_node, Idx bkref_idx)
1975 const re_dfa_t *const dfa = mctx->dfa;
1976 const re_node_set *eclosures = dfa->eclosures + from_node;
1979 /* Else, we are on the boundary: examine the nodes on the epsilon
1981 for (node_idx = 0; node_idx < eclosures->nelem; ++node_idx)
1983 Idx node = eclosures->elems[node_idx];
1984 switch (dfa->nodes[node].type)
1987 if (bkref_idx != REG_MISSING)
1989 struct re_backref_cache_entry *ent = mctx->bkref_ents + bkref_idx;
1995 if (ent->node != node)
1998 if (subexp_idx < BITSET_WORD_BITS
1999 && !(ent->eps_reachable_subexps_map
2000 & ((bitset_word_t) 1 << subexp_idx)))
2003 /* Recurse trying to reach the OP_OPEN_SUBEXP and
2004 OP_CLOSE_SUBEXP cases below. But, if the
2005 destination node is the same node as the source
2006 node, don't recurse because it would cause an
2007 infinite loop: a regex that exhibits this behavior
2009 dst = dfa->edests[node].elems[0];
2010 if (dst == from_node)
2014 else /* if (boundaries & 2) */
2019 check_dst_limits_calc_pos_1 (mctx, boundaries, subexp_idx,
2021 if (cpos == -1 /* && (boundaries & 1) */)
2023 if (cpos == 0 && (boundaries & 2))
2026 if (subexp_idx < BITSET_WORD_BITS)
2027 ent->eps_reachable_subexps_map
2028 &= ~((bitset_word_t) 1 << subexp_idx);
2030 while (ent++->more);
2034 case OP_OPEN_SUBEXP:
2035 if ((boundaries & 1) && subexp_idx == dfa->nodes[node].opr.idx)
2039 case OP_CLOSE_SUBEXP:
2040 if ((boundaries & 2) && subexp_idx == dfa->nodes[node].opr.idx)
2049 return (boundaries & 2) ? 1 : 0;
2054 check_dst_limits_calc_pos (const re_match_context_t *mctx, Idx limit,
2055 Idx subexp_idx, Idx from_node, Idx str_idx,
2058 struct re_backref_cache_entry *lim = mctx->bkref_ents + limit;
2061 /* If we are outside the range of the subexpression, return -1 or 1. */
2062 if (str_idx < lim->subexp_from)
2065 if (lim->subexp_to < str_idx)
2068 /* If we are within the subexpression, return 0. */
2069 boundaries = (str_idx == lim->subexp_from);
2070 boundaries |= (str_idx == lim->subexp_to) << 1;
2071 if (boundaries == 0)
2074 /* Else, examine epsilon closure. */
2075 return check_dst_limits_calc_pos_1 (mctx, boundaries, subexp_idx,
2076 from_node, bkref_idx);
2079 /* Check the limitations of sub expressions LIMITS, and remove the nodes
2080 which are against limitations from DEST_NODES. */
2082 static reg_errcode_t
2084 check_subexp_limits (const re_dfa_t *dfa, re_node_set *dest_nodes,
2085 const re_node_set *candidates, re_node_set *limits,
2086 struct re_backref_cache_entry *bkref_ents, Idx str_idx)
2089 Idx node_idx, lim_idx;
2091 for (lim_idx = 0; lim_idx < limits->nelem; ++lim_idx)
2094 struct re_backref_cache_entry *ent;
2095 ent = bkref_ents + limits->elems[lim_idx];
2097 if (str_idx <= ent->subexp_from || ent->str_idx < str_idx)
2098 continue; /* This is unrelated limitation. */
2100 subexp_idx = dfa->nodes[ent->node].opr.idx;
2101 if (ent->subexp_to == str_idx)
2103 Idx ops_node = REG_MISSING;
2104 Idx cls_node = REG_MISSING;
2105 for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx)
2107 Idx node = dest_nodes->elems[node_idx];
2108 re_token_type_t type = dfa->nodes[node].type;
2109 if (type == OP_OPEN_SUBEXP
2110 && subexp_idx == dfa->nodes[node].opr.idx)
2112 else if (type == OP_CLOSE_SUBEXP
2113 && subexp_idx == dfa->nodes[node].opr.idx)
2117 /* Check the limitation of the open subexpression. */
2118 /* Note that (ent->subexp_to = str_idx != ent->subexp_from). */
2119 if (REG_VALID_INDEX (ops_node))
2121 err = sub_epsilon_src_nodes (dfa, ops_node, dest_nodes,
2123 if (BE (err != REG_NOERROR, 0))
2127 /* Check the limitation of the close subexpression. */
2128 if (REG_VALID_INDEX (cls_node))
2129 for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx)
2131 Idx node = dest_nodes->elems[node_idx];
2132 if (!re_node_set_contains (dfa->inveclosures + node,
2134 && !re_node_set_contains (dfa->eclosures + node,
2137 /* It is against this limitation.
2138 Remove it form the current sifted state. */
2139 err = sub_epsilon_src_nodes (dfa, node, dest_nodes,
2141 if (BE (err != REG_NOERROR, 0))
2147 else /* (ent->subexp_to != str_idx) */
2149 for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx)
2151 Idx node = dest_nodes->elems[node_idx];
2152 re_token_type_t type = dfa->nodes[node].type;
2153 if (type == OP_CLOSE_SUBEXP || type == OP_OPEN_SUBEXP)
2155 if (subexp_idx != dfa->nodes[node].opr.idx)
2157 /* It is against this limitation.
2158 Remove it form the current sifted state. */
2159 err = sub_epsilon_src_nodes (dfa, node, dest_nodes,
2161 if (BE (err != REG_NOERROR, 0))
2170 static reg_errcode_t
2171 internal_function __attribute_warn_unused_result__
2172 sift_states_bkref (const re_match_context_t *mctx, re_sift_context_t *sctx,
2173 Idx str_idx, const re_node_set *candidates)
2175 const re_dfa_t *const dfa = mctx->dfa;
2178 re_sift_context_t local_sctx;
2179 Idx first_idx = search_cur_bkref_entry (mctx, str_idx);
2181 if (first_idx == REG_MISSING)
2184 local_sctx.sifted_states = NULL; /* Mark that it hasn't been initialized. */
2186 for (node_idx = 0; node_idx < candidates->nelem; ++node_idx)
2189 re_token_type_t type;
2190 struct re_backref_cache_entry *entry;
2191 node = candidates->elems[node_idx];
2192 type = dfa->nodes[node].type;
2193 /* Avoid infinite loop for the REs like "()\1+". */
2194 if (node == sctx->last_node && str_idx == sctx->last_str_idx)
2196 if (type != OP_BACK_REF)
2199 entry = mctx->bkref_ents + first_idx;
2200 enabled_idx = first_idx;
2207 re_dfastate_t *cur_state;
2209 if (entry->node != node)
2211 subexp_len = entry->subexp_to - entry->subexp_from;
2212 to_idx = str_idx + subexp_len;
2213 dst_node = (subexp_len ? dfa->nexts[node]
2214 : dfa->edests[node].elems[0]);
2216 if (to_idx > sctx->last_str_idx
2217 || sctx->sifted_states[to_idx] == NULL
2218 || !STATE_NODE_CONTAINS (sctx->sifted_states[to_idx], dst_node)
2219 || check_dst_limits (mctx, &sctx->limits, node,
2220 str_idx, dst_node, to_idx))
2223 if (local_sctx.sifted_states == NULL)
2226 err = re_node_set_init_copy (&local_sctx.limits, &sctx->limits);
2227 if (BE (err != REG_NOERROR, 0))
2230 local_sctx.last_node = node;
2231 local_sctx.last_str_idx = str_idx;
2232 ok = re_node_set_insert (&local_sctx.limits, enabled_idx);
2238 cur_state = local_sctx.sifted_states[str_idx];
2239 err = sift_states_backward (mctx, &local_sctx);
2240 if (BE (err != REG_NOERROR, 0))
2242 if (sctx->limited_states != NULL)
2244 err = merge_state_array (dfa, sctx->limited_states,
2245 local_sctx.sifted_states,
2247 if (BE (err != REG_NOERROR, 0))
2250 local_sctx.sifted_states[str_idx] = cur_state;
2251 re_node_set_remove (&local_sctx.limits, enabled_idx);
2253 /* mctx->bkref_ents may have changed, reload the pointer. */
2254 entry = mctx->bkref_ents + enabled_idx;
2256 while (enabled_idx++, entry++->more);
2260 if (local_sctx.sifted_states != NULL)
2262 re_node_set_free (&local_sctx.limits);
2269 #ifdef RE_ENABLE_I18N
2272 sift_states_iter_mb (const re_match_context_t *mctx, re_sift_context_t *sctx,
2273 Idx node_idx, Idx str_idx, Idx max_str_idx)
2275 const re_dfa_t *const dfa = mctx->dfa;
2277 /* Check the node can accept "multi byte". */
2278 naccepted = check_node_accept_bytes (dfa, node_idx, &mctx->input, str_idx);
2279 if (naccepted > 0 && str_idx + naccepted <= max_str_idx &&
2280 !STATE_NODE_CONTAINS (sctx->sifted_states[str_idx + naccepted],
2281 dfa->nexts[node_idx]))
2282 /* The node can't accept the "multi byte", or the
2283 destination was already thrown away, then the node
2284 could't accept the current input "multi byte". */
2286 /* Otherwise, it is sure that the node could accept
2287 'naccepted' bytes input. */
2290 #endif /* RE_ENABLE_I18N */
2293 /* Functions for state transition. */
2295 /* Return the next state to which the current state STATE will transit by
2296 accepting the current input byte, and update STATE_LOG if necessary.
2297 If STATE can accept a multibyte char/collating element/back reference
2298 update the destination of STATE_LOG. */
2300 static re_dfastate_t *
2301 internal_function __attribute_warn_unused_result__
2302 transit_state (reg_errcode_t *err, re_match_context_t *mctx,
2303 re_dfastate_t *state)
2305 re_dfastate_t **trtable;
2308 #ifdef RE_ENABLE_I18N
2309 /* If the current state can accept multibyte. */
2310 if (BE (state->accept_mb, 0))
2312 *err = transit_state_mb (mctx, state);
2313 if (BE (*err != REG_NOERROR, 0))
2316 #endif /* RE_ENABLE_I18N */
2318 /* Then decide the next state with the single byte. */
2321 /* don't use transition table */
2322 return transit_state_sb (err, mctx, state);
2325 /* Use transition table */
2326 ch = re_string_fetch_byte (&mctx->input);
2329 trtable = state->trtable;
2330 if (BE (trtable != NULL, 1))
2333 trtable = state->word_trtable;
2334 if (BE (trtable != NULL, 1))
2336 unsigned int context;
2338 = re_string_context_at (&mctx->input,
2339 re_string_cur_idx (&mctx->input) - 1,
2341 if (IS_WORD_CONTEXT (context))
2342 return trtable[ch + SBC_MAX];
2347 if (!build_trtable (mctx->dfa, state))
2353 /* Retry, we now have a transition table. */
2357 /* Update the state_log if we need */
2358 static re_dfastate_t *
2360 merge_state_with_log (reg_errcode_t *err, re_match_context_t *mctx,
2361 re_dfastate_t *next_state)
2363 const re_dfa_t *const dfa = mctx->dfa;
2364 Idx cur_idx = re_string_cur_idx (&mctx->input);
2366 if (cur_idx > mctx->state_log_top)
2368 mctx->state_log[cur_idx] = next_state;
2369 mctx->state_log_top = cur_idx;
2371 else if (mctx->state_log[cur_idx] == 0)
2373 mctx->state_log[cur_idx] = next_state;
2377 re_dfastate_t *pstate;
2378 unsigned int context;
2379 re_node_set next_nodes, *log_nodes, *table_nodes = NULL;
2380 /* If (state_log[cur_idx] != 0), it implies that cur_idx is
2381 the destination of a multibyte char/collating element/
2382 back reference. Then the next state is the union set of
2383 these destinations and the results of the transition table. */
2384 pstate = mctx->state_log[cur_idx];
2385 log_nodes = pstate->entrance_nodes;
2386 if (next_state != NULL)
2388 table_nodes = next_state->entrance_nodes;
2389 *err = re_node_set_init_union (&next_nodes, table_nodes,
2391 if (BE (*err != REG_NOERROR, 0))
2395 next_nodes = *log_nodes;
2396 /* Note: We already add the nodes of the initial state,
2397 then we don't need to add them here. */
2399 context = re_string_context_at (&mctx->input,
2400 re_string_cur_idx (&mctx->input) - 1,
2402 next_state = mctx->state_log[cur_idx]
2403 = re_acquire_state_context (err, dfa, &next_nodes, context);
2404 /* We don't need to check errors here, since the return value of
2405 this function is next_state and ERR is already set. */
2407 if (table_nodes != NULL)
2408 re_node_set_free (&next_nodes);
2411 if (BE (dfa->nbackref, 0) && next_state != NULL)
2413 /* Check OP_OPEN_SUBEXP in the current state in case that we use them
2414 later. We must check them here, since the back references in the
2415 next state might use them. */
2416 *err = check_subexp_matching_top (mctx, &next_state->nodes,
2418 if (BE (*err != REG_NOERROR, 0))
2421 /* If the next state has back references. */
2422 if (next_state->has_backref)
2424 *err = transit_state_bkref (mctx, &next_state->nodes);
2425 if (BE (*err != REG_NOERROR, 0))
2427 next_state = mctx->state_log[cur_idx];
2434 /* Skip bytes in the input that correspond to part of a
2435 multi-byte match, then look in the log for a state
2436 from which to restart matching. */
2437 static re_dfastate_t *
2439 find_recover_state (reg_errcode_t *err, re_match_context_t *mctx)
2441 re_dfastate_t *cur_state;
2444 Idx max = mctx->state_log_top;
2445 Idx cur_str_idx = re_string_cur_idx (&mctx->input);
2449 if (++cur_str_idx > max)
2451 re_string_skip_bytes (&mctx->input, 1);
2453 while (mctx->state_log[cur_str_idx] == NULL);
2455 cur_state = merge_state_with_log (err, mctx, NULL);
2457 while (*err == REG_NOERROR && cur_state == NULL);
2461 /* Helper functions for transit_state. */
2463 /* From the node set CUR_NODES, pick up the nodes whose types are
2464 OP_OPEN_SUBEXP and which have corresponding back references in the regular
2465 expression. And register them to use them later for evaluating the
2466 corresponding back references. */
2468 static reg_errcode_t
2470 check_subexp_matching_top (re_match_context_t *mctx, re_node_set *cur_nodes,
2473 const re_dfa_t *const dfa = mctx->dfa;
2477 /* TODO: This isn't efficient.
2478 Because there might be more than one nodes whose types are
2479 OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all
2482 for (node_idx = 0; node_idx < cur_nodes->nelem; ++node_idx)
2484 Idx node = cur_nodes->elems[node_idx];
2485 if (dfa->nodes[node].type == OP_OPEN_SUBEXP
2486 && dfa->nodes[node].opr.idx < BITSET_WORD_BITS
2487 && (dfa->used_bkref_map
2488 & ((bitset_word_t) 1 << dfa->nodes[node].opr.idx)))
2490 err = match_ctx_add_subtop (mctx, node, str_idx);
2491 if (BE (err != REG_NOERROR, 0))
2499 /* Return the next state to which the current state STATE will transit by
2500 accepting the current input byte. */
2502 static re_dfastate_t *
2503 transit_state_sb (reg_errcode_t *err, re_match_context_t *mctx,
2504 re_dfastate_t *state)
2506 const re_dfa_t *const dfa = mctx->dfa;
2507 re_node_set next_nodes;
2508 re_dfastate_t *next_state;
2509 Idx node_cnt, cur_str_idx = re_string_cur_idx (&mctx->input);
2510 unsigned int context;
2512 *err = re_node_set_alloc (&next_nodes, state->nodes.nelem + 1);
2513 if (BE (*err != REG_NOERROR, 0))
2515 for (node_cnt = 0; node_cnt < state->nodes.nelem; ++node_cnt)
2517 Idx cur_node = state->nodes.elems[node_cnt];
2518 if (check_node_accept (mctx, dfa->nodes + cur_node, cur_str_idx))
2520 *err = re_node_set_merge (&next_nodes,
2521 dfa->eclosures + dfa->nexts[cur_node]);
2522 if (BE (*err != REG_NOERROR, 0))
2524 re_node_set_free (&next_nodes);
2529 context = re_string_context_at (&mctx->input, cur_str_idx, mctx->eflags);
2530 next_state = re_acquire_state_context (err, dfa, &next_nodes, context);
2531 /* We don't need to check errors here, since the return value of
2532 this function is next_state and ERR is already set. */
2534 re_node_set_free (&next_nodes);
2535 re_string_skip_bytes (&mctx->input, 1);
2540 #ifdef RE_ENABLE_I18N
2541 static reg_errcode_t
2543 transit_state_mb (re_match_context_t *mctx, re_dfastate_t *pstate)
2545 const re_dfa_t *const dfa = mctx->dfa;
2549 for (i = 0; i < pstate->nodes.nelem; ++i)
2551 re_node_set dest_nodes, *new_nodes;
2552 Idx cur_node_idx = pstate->nodes.elems[i];
2555 unsigned int context;
2556 re_dfastate_t *dest_state;
2558 if (!dfa->nodes[cur_node_idx].accept_mb)
2561 if (dfa->nodes[cur_node_idx].constraint)
2563 context = re_string_context_at (&mctx->input,
2564 re_string_cur_idx (&mctx->input),
2566 if (NOT_SATISFY_NEXT_CONSTRAINT (dfa->nodes[cur_node_idx].constraint,
2571 /* How many bytes the node can accept? */
2572 naccepted = check_node_accept_bytes (dfa, cur_node_idx, &mctx->input,
2573 re_string_cur_idx (&mctx->input));
2577 /* The node can accepts 'naccepted' bytes. */
2578 dest_idx = re_string_cur_idx (&mctx->input) + naccepted;
2579 mctx->max_mb_elem_len = ((mctx->max_mb_elem_len < naccepted) ? naccepted
2580 : mctx->max_mb_elem_len);
2581 err = clean_state_log_if_needed (mctx, dest_idx);
2582 if (BE (err != REG_NOERROR, 0))
2585 assert (dfa->nexts[cur_node_idx] != REG_MISSING);
2587 new_nodes = dfa->eclosures + dfa->nexts[cur_node_idx];
2589 dest_state = mctx->state_log[dest_idx];
2590 if (dest_state == NULL)
2591 dest_nodes = *new_nodes;
2594 err = re_node_set_init_union (&dest_nodes,
2595 dest_state->entrance_nodes, new_nodes);
2596 if (BE (err != REG_NOERROR, 0))
2599 context = re_string_context_at (&mctx->input, dest_idx - 1,
2601 mctx->state_log[dest_idx]
2602 = re_acquire_state_context (&err, dfa, &dest_nodes, context);
2603 if (dest_state != NULL)
2604 re_node_set_free (&dest_nodes);
2605 if (BE (mctx->state_log[dest_idx] == NULL && err != REG_NOERROR, 0))
2610 #endif /* RE_ENABLE_I18N */
2612 static reg_errcode_t
2614 transit_state_bkref (re_match_context_t *mctx, const re_node_set *nodes)
2616 const re_dfa_t *const dfa = mctx->dfa;
2619 Idx cur_str_idx = re_string_cur_idx (&mctx->input);
2621 for (i = 0; i < nodes->nelem; ++i)
2623 Idx dest_str_idx, prev_nelem, bkc_idx;
2624 Idx node_idx = nodes->elems[i];
2625 unsigned int context;
2626 const re_token_t *node = dfa->nodes + node_idx;
2627 re_node_set *new_dest_nodes;
2629 /* Check whether 'node' is a backreference or not. */
2630 if (node->type != OP_BACK_REF)
2633 if (node->constraint)
2635 context = re_string_context_at (&mctx->input, cur_str_idx,
2637 if (NOT_SATISFY_NEXT_CONSTRAINT (node->constraint, context))
2641 /* 'node' is a backreference.
2642 Check the substring which the substring matched. */
2643 bkc_idx = mctx->nbkref_ents;
2644 err = get_subexp (mctx, node_idx, cur_str_idx);
2645 if (BE (err != REG_NOERROR, 0))
2648 /* And add the epsilon closures (which is 'new_dest_nodes') of
2649 the backreference to appropriate state_log. */
2651 assert (dfa->nexts[node_idx] != REG_MISSING);
2653 for (; bkc_idx < mctx->nbkref_ents; ++bkc_idx)
2656 re_dfastate_t *dest_state;
2657 struct re_backref_cache_entry *bkref_ent;
2658 bkref_ent = mctx->bkref_ents + bkc_idx;
2659 if (bkref_ent->node != node_idx || bkref_ent->str_idx != cur_str_idx)
2661 subexp_len = bkref_ent->subexp_to - bkref_ent->subexp_from;
2662 new_dest_nodes = (subexp_len == 0
2663 ? dfa->eclosures + dfa->edests[node_idx].elems[0]
2664 : dfa->eclosures + dfa->nexts[node_idx]);
2665 dest_str_idx = (cur_str_idx + bkref_ent->subexp_to
2666 - bkref_ent->subexp_from);
2667 context = re_string_context_at (&mctx->input, dest_str_idx - 1,
2669 dest_state = mctx->state_log[dest_str_idx];
2670 prev_nelem = ((mctx->state_log[cur_str_idx] == NULL) ? 0
2671 : mctx->state_log[cur_str_idx]->nodes.nelem);
2672 /* Add 'new_dest_node' to state_log. */
2673 if (dest_state == NULL)
2675 mctx->state_log[dest_str_idx]
2676 = re_acquire_state_context (&err, dfa, new_dest_nodes,
2678 if (BE (mctx->state_log[dest_str_idx] == NULL
2679 && err != REG_NOERROR, 0))
2684 re_node_set dest_nodes;
2685 err = re_node_set_init_union (&dest_nodes,
2686 dest_state->entrance_nodes,
2688 if (BE (err != REG_NOERROR, 0))
2690 re_node_set_free (&dest_nodes);
2693 mctx->state_log[dest_str_idx]
2694 = re_acquire_state_context (&err, dfa, &dest_nodes, context);
2695 re_node_set_free (&dest_nodes);
2696 if (BE (mctx->state_log[dest_str_idx] == NULL
2697 && err != REG_NOERROR, 0))
2700 /* We need to check recursively if the backreference can epsilon
2703 && mctx->state_log[cur_str_idx]->nodes.nelem > prev_nelem)
2705 err = check_subexp_matching_top (mctx, new_dest_nodes,
2707 if (BE (err != REG_NOERROR, 0))
2709 err = transit_state_bkref (mctx, new_dest_nodes);
2710 if (BE (err != REG_NOERROR, 0))
2720 /* Enumerate all the candidates which the backreference BKREF_NODE can match
2721 at BKREF_STR_IDX, and register them by match_ctx_add_entry().
2722 Note that we might collect inappropriate candidates here.
2723 However, the cost of checking them strictly here is too high, then we
2724 delay these checking for prune_impossible_nodes(). */
2726 static reg_errcode_t
2727 internal_function __attribute_warn_unused_result__
2728 get_subexp (re_match_context_t *mctx, Idx bkref_node, Idx bkref_str_idx)
2730 const re_dfa_t *const dfa = mctx->dfa;
2731 Idx subexp_num, sub_top_idx;
2732 const char *buf = (const char *) re_string_get_buffer (&mctx->input);
2733 /* Return if we have already checked BKREF_NODE at BKREF_STR_IDX. */
2734 Idx cache_idx = search_cur_bkref_entry (mctx, bkref_str_idx);
2735 if (cache_idx != REG_MISSING)
2737 const struct re_backref_cache_entry *entry
2738 = mctx->bkref_ents + cache_idx;
2740 if (entry->node == bkref_node)
2741 return REG_NOERROR; /* We already checked it. */
2742 while (entry++->more);
2745 subexp_num = dfa->nodes[bkref_node].opr.idx;
2747 /* For each sub expression */
2748 for (sub_top_idx = 0; sub_top_idx < mctx->nsub_tops; ++sub_top_idx)
2751 re_sub_match_top_t *sub_top = mctx->sub_tops[sub_top_idx];
2752 re_sub_match_last_t *sub_last;
2753 Idx sub_last_idx, sl_str, bkref_str_off;
2755 if (dfa->nodes[sub_top->node].opr.idx != subexp_num)
2756 continue; /* It isn't related. */
2758 sl_str = sub_top->str_idx;
2759 bkref_str_off = bkref_str_idx;
2760 /* At first, check the last node of sub expressions we already
2762 for (sub_last_idx = 0; sub_last_idx < sub_top->nlasts; ++sub_last_idx)
2764 regoff_t sl_str_diff;
2765 sub_last = sub_top->lasts[sub_last_idx];
2766 sl_str_diff = sub_last->str_idx - sl_str;
2767 /* The matched string by the sub expression match with the substring
2768 at the back reference? */
2769 if (sl_str_diff > 0)
2771 if (BE (bkref_str_off + sl_str_diff > mctx->input.valid_len, 0))
2773 /* Not enough chars for a successful match. */
2774 if (bkref_str_off + sl_str_diff > mctx->input.len)
2777 err = clean_state_log_if_needed (mctx,
2780 if (BE (err != REG_NOERROR, 0))
2782 buf = (const char *) re_string_get_buffer (&mctx->input);
2784 if (memcmp (buf + bkref_str_off, buf + sl_str, sl_str_diff) != 0)
2785 /* We don't need to search this sub expression any more. */
2788 bkref_str_off += sl_str_diff;
2789 sl_str += sl_str_diff;
2790 err = get_subexp_sub (mctx, sub_top, sub_last, bkref_node,
2793 /* Reload buf, since the preceding call might have reallocated
2795 buf = (const char *) re_string_get_buffer (&mctx->input);
2797 if (err == REG_NOMATCH)
2799 if (BE (err != REG_NOERROR, 0))
2803 if (sub_last_idx < sub_top->nlasts)
2805 if (sub_last_idx > 0)
2807 /* Then, search for the other last nodes of the sub expression. */
2808 for (; sl_str <= bkref_str_idx; ++sl_str)
2811 regoff_t sl_str_off;
2812 const re_node_set *nodes;
2813 sl_str_off = sl_str - sub_top->str_idx;
2814 /* The matched string by the sub expression match with the substring
2815 at the back reference? */
2818 if (BE (bkref_str_off >= mctx->input.valid_len, 0))
2820 /* If we are at the end of the input, we cannot match. */
2821 if (bkref_str_off >= mctx->input.len)
2824 err = extend_buffers (mctx, bkref_str_off + 1);
2825 if (BE (err != REG_NOERROR, 0))
2828 buf = (const char *) re_string_get_buffer (&mctx->input);
2830 if (buf [bkref_str_off++] != buf[sl_str - 1])
2831 break; /* We don't need to search this sub expression
2834 if (mctx->state_log[sl_str] == NULL)
2836 /* Does this state have a ')' of the sub expression? */
2837 nodes = &mctx->state_log[sl_str]->nodes;
2838 cls_node = find_subexp_node (dfa, nodes, subexp_num,
2840 if (cls_node == REG_MISSING)
2842 if (sub_top->path == NULL)
2844 sub_top->path = calloc (sizeof (state_array_t),
2845 sl_str - sub_top->str_idx + 1);
2846 if (sub_top->path == NULL)
2849 /* Can the OP_OPEN_SUBEXP node arrive the OP_CLOSE_SUBEXP node
2850 in the current context? */
2851 err = check_arrival (mctx, sub_top->path, sub_top->node,
2852 sub_top->str_idx, cls_node, sl_str,
2854 if (err == REG_NOMATCH)
2856 if (BE (err != REG_NOERROR, 0))
2858 sub_last = match_ctx_add_sublast (sub_top, cls_node, sl_str);
2859 if (BE (sub_last == NULL, 0))
2861 err = get_subexp_sub (mctx, sub_top, sub_last, bkref_node,
2863 if (err == REG_NOMATCH)
2870 /* Helper functions for get_subexp(). */
2872 /* Check SUB_LAST can arrive to the back reference BKREF_NODE at BKREF_STR.
2873 If it can arrive, register the sub expression expressed with SUB_TOP
2876 static reg_errcode_t
2878 get_subexp_sub (re_match_context_t *mctx, const re_sub_match_top_t *sub_top,
2879 re_sub_match_last_t *sub_last, Idx bkref_node, Idx bkref_str)
2883 /* Can the subexpression arrive the back reference? */
2884 err = check_arrival (mctx, &sub_last->path, sub_last->node,
2885 sub_last->str_idx, bkref_node, bkref_str,
2887 if (err != REG_NOERROR)
2889 err = match_ctx_add_entry (mctx, bkref_node, bkref_str, sub_top->str_idx,
2891 if (BE (err != REG_NOERROR, 0))
2893 to_idx = bkref_str + sub_last->str_idx - sub_top->str_idx;
2894 return clean_state_log_if_needed (mctx, to_idx);
2897 /* Find the first node which is '(' or ')' and whose index is SUBEXP_IDX.
2898 Search '(' if FL_OPEN, or search ')' otherwise.
2899 TODO: This function isn't efficient...
2900 Because there might be more than one nodes whose types are
2901 OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all
2907 find_subexp_node (const re_dfa_t *dfa, const re_node_set *nodes,
2908 Idx subexp_idx, int type)
2911 for (cls_idx = 0; cls_idx < nodes->nelem; ++cls_idx)
2913 Idx cls_node = nodes->elems[cls_idx];
2914 const re_token_t *node = dfa->nodes + cls_node;
2915 if (node->type == type
2916 && node->opr.idx == subexp_idx)
2922 /* Check whether the node TOP_NODE at TOP_STR can arrive to the node
2923 LAST_NODE at LAST_STR. We record the path onto PATH since it will be
2925 Return REG_NOERROR if it can arrive, or REG_NOMATCH otherwise. */
2927 static reg_errcode_t
2928 internal_function __attribute_warn_unused_result__
2929 check_arrival (re_match_context_t *mctx, state_array_t *path, Idx top_node,
2930 Idx top_str, Idx last_node, Idx last_str, int type)
2932 const re_dfa_t *const dfa = mctx->dfa;
2933 reg_errcode_t err = REG_NOERROR;
2934 Idx subexp_num, backup_cur_idx, str_idx, null_cnt;
2935 re_dfastate_t *cur_state = NULL;
2936 re_node_set *cur_nodes, next_nodes;
2937 re_dfastate_t **backup_state_log;
2938 unsigned int context;
2940 subexp_num = dfa->nodes[top_node].opr.idx;
2941 /* Extend the buffer if we need. */
2942 if (BE (path->alloc < last_str + mctx->max_mb_elem_len + 1, 0))
2944 re_dfastate_t **new_array;
2945 Idx old_alloc = path->alloc;
2946 Idx incr_alloc = last_str + mctx->max_mb_elem_len + 1;
2948 if (BE (IDX_MAX - old_alloc < incr_alloc, 0))
2950 new_alloc = old_alloc + incr_alloc;
2951 if (BE (SIZE_MAX / sizeof (re_dfastate_t *) < new_alloc, 0))
2953 new_array = re_realloc (path->array, re_dfastate_t *, new_alloc);
2954 if (BE (new_array == NULL, 0))
2956 path->array = new_array;
2957 path->alloc = new_alloc;
2958 memset (new_array + old_alloc, '\0',
2959 sizeof (re_dfastate_t *) * (path->alloc - old_alloc));
2962 str_idx = path->next_idx ? path->next_idx : top_str;
2964 /* Temporary modify MCTX. */
2965 backup_state_log = mctx->state_log;
2966 backup_cur_idx = mctx->input.cur_idx;
2967 mctx->state_log = path->array;
2968 mctx->input.cur_idx = str_idx;
2970 /* Setup initial node set. */
2971 context = re_string_context_at (&mctx->input, str_idx - 1, mctx->eflags);
2972 if (str_idx == top_str)
2974 err = re_node_set_init_1 (&next_nodes, top_node);
2975 if (BE (err != REG_NOERROR, 0))
2977 err = check_arrival_expand_ecl (dfa, &next_nodes, subexp_num, type);
2978 if (BE (err != REG_NOERROR, 0))
2980 re_node_set_free (&next_nodes);
2986 cur_state = mctx->state_log[str_idx];
2987 if (cur_state && cur_state->has_backref)
2989 err = re_node_set_init_copy (&next_nodes, &cur_state->nodes);
2990 if (BE (err != REG_NOERROR, 0))
2994 re_node_set_init_empty (&next_nodes);
2996 if (str_idx == top_str || (cur_state && cur_state->has_backref))
2998 if (next_nodes.nelem)
3000 err = expand_bkref_cache (mctx, &next_nodes, str_idx,
3002 if (BE (err != REG_NOERROR, 0))
3004 re_node_set_free (&next_nodes);
3008 cur_state = re_acquire_state_context (&err, dfa, &next_nodes, context);
3009 if (BE (cur_state == NULL && err != REG_NOERROR, 0))
3011 re_node_set_free (&next_nodes);
3014 mctx->state_log[str_idx] = cur_state;
3017 for (null_cnt = 0; str_idx < last_str && null_cnt <= mctx->max_mb_elem_len;)
3019 re_node_set_empty (&next_nodes);
3020 if (mctx->state_log[str_idx + 1])
3022 err = re_node_set_merge (&next_nodes,
3023 &mctx->state_log[str_idx + 1]->nodes);
3024 if (BE (err != REG_NOERROR, 0))
3026 re_node_set_free (&next_nodes);
3032 err = check_arrival_add_next_nodes (mctx, str_idx,
3033 &cur_state->non_eps_nodes,
3035 if (BE (err != REG_NOERROR, 0))
3037 re_node_set_free (&next_nodes);
3042 if (next_nodes.nelem)
3044 err = check_arrival_expand_ecl (dfa, &next_nodes, subexp_num, type);
3045 if (BE (err != REG_NOERROR, 0))
3047 re_node_set_free (&next_nodes);
3050 err = expand_bkref_cache (mctx, &next_nodes, str_idx,
3052 if (BE (err != REG_NOERROR, 0))
3054 re_node_set_free (&next_nodes);
3058 context = re_string_context_at (&mctx->input, str_idx - 1, mctx->eflags);
3059 cur_state = re_acquire_state_context (&err, dfa, &next_nodes, context);
3060 if (BE (cur_state == NULL && err != REG_NOERROR, 0))
3062 re_node_set_free (&next_nodes);
3065 mctx->state_log[str_idx] = cur_state;
3066 null_cnt = cur_state == NULL ? null_cnt + 1 : 0;
3068 re_node_set_free (&next_nodes);
3069 cur_nodes = (mctx->state_log[last_str] == NULL ? NULL
3070 : &mctx->state_log[last_str]->nodes);
3071 path->next_idx = str_idx;
3074 mctx->state_log = backup_state_log;
3075 mctx->input.cur_idx = backup_cur_idx;
3077 /* Then check the current node set has the node LAST_NODE. */
3078 if (cur_nodes != NULL && re_node_set_contains (cur_nodes, last_node))
3084 /* Helper functions for check_arrival. */
3086 /* Calculate the destination nodes of CUR_NODES at STR_IDX, and append them
3088 TODO: This function is similar to the functions transit_state*(),
3089 however this function has many additional works.
3090 Can't we unify them? */
3092 static reg_errcode_t
3093 internal_function __attribute_warn_unused_result__
3094 check_arrival_add_next_nodes (re_match_context_t *mctx, Idx str_idx,
3095 re_node_set *cur_nodes, re_node_set *next_nodes)
3097 const re_dfa_t *const dfa = mctx->dfa;
3100 #ifdef RE_ENABLE_I18N
3101 reg_errcode_t err = REG_NOERROR;
3103 re_node_set union_set;
3104 re_node_set_init_empty (&union_set);
3105 for (cur_idx = 0; cur_idx < cur_nodes->nelem; ++cur_idx)
3108 Idx cur_node = cur_nodes->elems[cur_idx];
3110 re_token_type_t type = dfa->nodes[cur_node].type;
3111 assert (!IS_EPSILON_NODE (type));
3113 #ifdef RE_ENABLE_I18N
3114 /* If the node may accept "multi byte". */
3115 if (dfa->nodes[cur_node].accept_mb)
3117 naccepted = check_node_accept_bytes (dfa, cur_node, &mctx->input,
3121 re_dfastate_t *dest_state;
3122 Idx next_node = dfa->nexts[cur_node];
3123 Idx next_idx = str_idx + naccepted;
3124 dest_state = mctx->state_log[next_idx];
3125 re_node_set_empty (&union_set);
3128 err = re_node_set_merge (&union_set, &dest_state->nodes);
3129 if (BE (err != REG_NOERROR, 0))
3131 re_node_set_free (&union_set);
3135 ok = re_node_set_insert (&union_set, next_node);
3138 re_node_set_free (&union_set);
3141 mctx->state_log[next_idx] = re_acquire_state (&err, dfa,
3143 if (BE (mctx->state_log[next_idx] == NULL
3144 && err != REG_NOERROR, 0))
3146 re_node_set_free (&union_set);
3151 #endif /* RE_ENABLE_I18N */
3153 || check_node_accept (mctx, dfa->nodes + cur_node, str_idx))
3155 ok = re_node_set_insert (next_nodes, dfa->nexts[cur_node]);
3158 re_node_set_free (&union_set);
3163 re_node_set_free (&union_set);
3167 /* For all the nodes in CUR_NODES, add the epsilon closures of them to
3168 CUR_NODES, however exclude the nodes which are:
3169 - inside the sub expression whose number is EX_SUBEXP, if FL_OPEN.
3170 - out of the sub expression whose number is EX_SUBEXP, if !FL_OPEN.
3173 static reg_errcode_t
3175 check_arrival_expand_ecl (const re_dfa_t *dfa, re_node_set *cur_nodes,
3176 Idx ex_subexp, int type)
3179 Idx idx, outside_node;
3180 re_node_set new_nodes;
3182 assert (cur_nodes->nelem);
3184 err = re_node_set_alloc (&new_nodes, cur_nodes->nelem);
3185 if (BE (err != REG_NOERROR, 0))
3187 /* Create a new node set NEW_NODES with the nodes which are epsilon
3188 closures of the node in CUR_NODES. */
3190 for (idx = 0; idx < cur_nodes->nelem; ++idx)
3192 Idx cur_node = cur_nodes->elems[idx];
3193 const re_node_set *eclosure = dfa->eclosures + cur_node;
3194 outside_node = find_subexp_node (dfa, eclosure, ex_subexp, type);
3195 if (outside_node == REG_MISSING)
3197 /* There are no problematic nodes, just merge them. */
3198 err = re_node_set_merge (&new_nodes, eclosure);
3199 if (BE (err != REG_NOERROR, 0))
3201 re_node_set_free (&new_nodes);
3207 /* There are problematic nodes, re-calculate incrementally. */
3208 err = check_arrival_expand_ecl_sub (dfa, &new_nodes, cur_node,
3210 if (BE (err != REG_NOERROR, 0))
3212 re_node_set_free (&new_nodes);
3217 re_node_set_free (cur_nodes);
3218 *cur_nodes = new_nodes;
3222 /* Helper function for check_arrival_expand_ecl.
3223 Check incrementally the epsilon closure of TARGET, and if it isn't
3224 problematic append it to DST_NODES. */
3226 static reg_errcode_t
3227 internal_function __attribute_warn_unused_result__
3228 check_arrival_expand_ecl_sub (const re_dfa_t *dfa, re_node_set *dst_nodes,
3229 Idx target, Idx ex_subexp, int type)
3232 for (cur_node = target; !re_node_set_contains (dst_nodes, cur_node);)
3236 if (dfa->nodes[cur_node].type == type
3237 && dfa->nodes[cur_node].opr.idx == ex_subexp)
3239 if (type == OP_CLOSE_SUBEXP)
3241 ok = re_node_set_insert (dst_nodes, cur_node);
3247 ok = re_node_set_insert (dst_nodes, cur_node);
3250 if (dfa->edests[cur_node].nelem == 0)
3252 if (dfa->edests[cur_node].nelem == 2)
3255 err = check_arrival_expand_ecl_sub (dfa, dst_nodes,
3256 dfa->edests[cur_node].elems[1],
3258 if (BE (err != REG_NOERROR, 0))
3261 cur_node = dfa->edests[cur_node].elems[0];
3267 /* For all the back references in the current state, calculate the
3268 destination of the back references by the appropriate entry
3269 in MCTX->BKREF_ENTS. */
3271 static reg_errcode_t
3272 internal_function __attribute_warn_unused_result__
3273 expand_bkref_cache (re_match_context_t *mctx, re_node_set *cur_nodes,
3274 Idx cur_str, Idx subexp_num, int type)
3276 const re_dfa_t *const dfa = mctx->dfa;
3278 Idx cache_idx_start = search_cur_bkref_entry (mctx, cur_str);
3279 struct re_backref_cache_entry *ent;
3281 if (cache_idx_start == REG_MISSING)
3285 ent = mctx->bkref_ents + cache_idx_start;
3288 Idx to_idx, next_node;
3290 /* Is this entry ENT is appropriate? */
3291 if (!re_node_set_contains (cur_nodes, ent->node))
3294 to_idx = cur_str + ent->subexp_to - ent->subexp_from;
3295 /* Calculate the destination of the back reference, and append it
3296 to MCTX->STATE_LOG. */
3297 if (to_idx == cur_str)
3299 /* The backreference did epsilon transit, we must re-check all the
3300 node in the current state. */
3301 re_node_set new_dests;
3302 reg_errcode_t err2, err3;
3303 next_node = dfa->edests[ent->node].elems[0];
3304 if (re_node_set_contains (cur_nodes, next_node))
3306 err = re_node_set_init_1 (&new_dests, next_node);
3307 err2 = check_arrival_expand_ecl (dfa, &new_dests, subexp_num, type);
3308 err3 = re_node_set_merge (cur_nodes, &new_dests);
3309 re_node_set_free (&new_dests);
3310 if (BE (err != REG_NOERROR || err2 != REG_NOERROR
3311 || err3 != REG_NOERROR, 0))
3313 err = (err != REG_NOERROR ? err
3314 : (err2 != REG_NOERROR ? err2 : err3));
3317 /* TODO: It is still inefficient... */
3322 re_node_set union_set;
3323 next_node = dfa->nexts[ent->node];
3324 if (mctx->state_log[to_idx])
3327 if (re_node_set_contains (&mctx->state_log[to_idx]->nodes,
3330 err = re_node_set_init_copy (&union_set,
3331 &mctx->state_log[to_idx]->nodes);
3332 ok = re_node_set_insert (&union_set, next_node);
3333 if (BE (err != REG_NOERROR || ! ok, 0))
3335 re_node_set_free (&union_set);
3336 err = err != REG_NOERROR ? err : REG_ESPACE;
3342 err = re_node_set_init_1 (&union_set, next_node);
3343 if (BE (err != REG_NOERROR, 0))
3346 mctx->state_log[to_idx] = re_acquire_state (&err, dfa, &union_set);
3347 re_node_set_free (&union_set);
3348 if (BE (mctx->state_log[to_idx] == NULL
3349 && err != REG_NOERROR, 0))
3353 while (ent++->more);
3357 /* Build transition table for the state.
3358 Return true if successful. */
3362 build_trtable (const re_dfa_t *dfa, re_dfastate_t *state)
3367 bool need_word_trtable = false;
3368 bitset_word_t elem, mask;
3369 bool dests_node_malloced = false;
3370 bool dest_states_malloced = false;
3371 Idx ndests; /* Number of the destination states from 'state'. */
3372 re_dfastate_t **trtable;
3373 re_dfastate_t **dest_states = NULL, **dest_states_word, **dest_states_nl;
3374 re_node_set follows, *dests_node;
3376 bitset_t acceptable;
3380 re_node_set dests_node[SBC_MAX];
3381 bitset_t dests_ch[SBC_MAX];
3384 /* We build DFA states which corresponds to the destination nodes
3385 from 'state'. 'dests_node[i]' represents the nodes which i-th
3386 destination state contains, and 'dests_ch[i]' represents the
3387 characters which i-th destination state accepts. */
3388 if (__libc_use_alloca (sizeof (struct dests_alloc)))
3389 dests_alloc = (struct dests_alloc *) alloca (sizeof (struct dests_alloc));
3392 dests_alloc = re_malloc (struct dests_alloc, 1);
3393 if (BE (dests_alloc == NULL, 0))
3395 dests_node_malloced = true;
3397 dests_node = dests_alloc->dests_node;
3398 dests_ch = dests_alloc->dests_ch;
3400 /* Initialize transition table. */
3401 state->word_trtable = state->trtable = NULL;
3403 /* At first, group all nodes belonging to 'state' into several
3405 ndests = group_nodes_into_DFAstates (dfa, state, dests_node, dests_ch);
3406 if (BE (! REG_VALID_NONZERO_INDEX (ndests), 0))
3408 if (dests_node_malloced)
3410 /* Return false in case of an error, true otherwise. */
3413 state->trtable = (re_dfastate_t **)
3414 calloc (sizeof (re_dfastate_t *), SBC_MAX);
3415 if (BE (state->trtable == NULL, 0))
3422 err = re_node_set_alloc (&follows, ndests + 1);
3423 if (BE (err != REG_NOERROR, 0))
3426 /* Avoid arithmetic overflow in size calculation. */
3427 if (BE ((((SIZE_MAX - (sizeof (re_node_set) + sizeof (bitset_t)) * SBC_MAX)
3428 / (3 * sizeof (re_dfastate_t *)))
3433 if (__libc_use_alloca ((sizeof (re_node_set) + sizeof (bitset_t)) * SBC_MAX
3434 + ndests * 3 * sizeof (re_dfastate_t *)))
3435 dest_states = (re_dfastate_t **)
3436 alloca (ndests * 3 * sizeof (re_dfastate_t *));
3439 dest_states = (re_dfastate_t **)
3440 malloc (ndests * 3 * sizeof (re_dfastate_t *));
3441 if (BE (dest_states == NULL, 0))
3444 if (dest_states_malloced)
3446 re_node_set_free (&follows);
3447 for (i = 0; i < ndests; ++i)
3448 re_node_set_free (dests_node + i);
3449 if (dests_node_malloced)
3453 dest_states_malloced = true;
3455 dest_states_word = dest_states + ndests;
3456 dest_states_nl = dest_states_word + ndests;
3457 bitset_empty (acceptable);
3459 /* Then build the states for all destinations. */
3460 for (i = 0; i < ndests; ++i)
3463 re_node_set_empty (&follows);
3464 /* Merge the follows of this destination states. */
3465 for (j = 0; j < dests_node[i].nelem; ++j)
3467 next_node = dfa->nexts[dests_node[i].elems[j]];
3468 if (next_node != REG_MISSING)
3470 err = re_node_set_merge (&follows, dfa->eclosures + next_node);
3471 if (BE (err != REG_NOERROR, 0))
3475 dest_states[i] = re_acquire_state_context (&err, dfa, &follows, 0);
3476 if (BE (dest_states[i] == NULL && err != REG_NOERROR, 0))
3478 /* If the new state has context constraint,
3479 build appropriate states for these contexts. */
3480 if (dest_states[i]->has_constraint)
3482 dest_states_word[i] = re_acquire_state_context (&err, dfa, &follows,
3484 if (BE (dest_states_word[i] == NULL && err != REG_NOERROR, 0))
3487 if (dest_states[i] != dest_states_word[i] && dfa->mb_cur_max > 1)
3488 need_word_trtable = true;
3490 dest_states_nl[i] = re_acquire_state_context (&err, dfa, &follows,
3492 if (BE (dest_states_nl[i] == NULL && err != REG_NOERROR, 0))
3497 dest_states_word[i] = dest_states[i];
3498 dest_states_nl[i] = dest_states[i];
3500 bitset_merge (acceptable, dests_ch[i]);
3503 if (!BE (need_word_trtable, 0))
3505 /* We don't care about whether the following character is a word
3506 character, or we are in a single-byte character set so we can
3507 discern by looking at the character code: allocate a
3508 256-entry transition table. */
3509 trtable = state->trtable =
3510 (re_dfastate_t **) calloc (sizeof (re_dfastate_t *), SBC_MAX);
3511 if (BE (trtable == NULL, 0))
3514 /* For all characters ch...: */
3515 for (i = 0; i < BITSET_WORDS; ++i)
3516 for (ch = i * BITSET_WORD_BITS, elem = acceptable[i], mask = 1;
3518 mask <<= 1, elem >>= 1, ++ch)
3519 if (BE (elem & 1, 0))
3521 /* There must be exactly one destination which accepts
3522 character ch. See group_nodes_into_DFAstates. */
3523 for (j = 0; (dests_ch[j][i] & mask) == 0; ++j)
3526 /* j-th destination accepts the word character ch. */
3527 if (dfa->word_char[i] & mask)
3528 trtable[ch] = dest_states_word[j];
3530 trtable[ch] = dest_states[j];
3535 /* We care about whether the following character is a word
3536 character, and we are in a multi-byte character set: discern
3537 by looking at the character code: build two 256-entry
3538 transition tables, one starting at trtable[0] and one
3539 starting at trtable[SBC_MAX]. */
3540 trtable = state->word_trtable =
3541 (re_dfastate_t **) calloc (sizeof (re_dfastate_t *), 2 * SBC_MAX);
3542 if (BE (trtable == NULL, 0))
3545 /* For all characters ch...: */
3546 for (i = 0; i < BITSET_WORDS; ++i)
3547 for (ch = i * BITSET_WORD_BITS, elem = acceptable[i], mask = 1;
3549 mask <<= 1, elem >>= 1, ++ch)
3550 if (BE (elem & 1, 0))
3552 /* There must be exactly one destination which accepts
3553 character ch. See group_nodes_into_DFAstates. */
3554 for (j = 0; (dests_ch[j][i] & mask) == 0; ++j)
3557 /* j-th destination accepts the word character ch. */
3558 trtable[ch] = dest_states[j];
3559 trtable[ch + SBC_MAX] = dest_states_word[j];
3564 if (bitset_contain (acceptable, NEWLINE_CHAR))
3566 /* The current state accepts newline character. */
3567 for (j = 0; j < ndests; ++j)
3568 if (bitset_contain (dests_ch[j], NEWLINE_CHAR))
3570 /* k-th destination accepts newline character. */
3571 trtable[NEWLINE_CHAR] = dest_states_nl[j];
3572 if (need_word_trtable)
3573 trtable[NEWLINE_CHAR + SBC_MAX] = dest_states_nl[j];
3574 /* There must be only one destination which accepts
3575 newline. See group_nodes_into_DFAstates. */
3580 if (dest_states_malloced)
3583 re_node_set_free (&follows);
3584 for (i = 0; i < ndests; ++i)
3585 re_node_set_free (dests_node + i);
3587 if (dests_node_malloced)
3593 /* Group all nodes belonging to STATE into several destinations.
3594 Then for all destinations, set the nodes belonging to the destination
3595 to DESTS_NODE[i] and set the characters accepted by the destination
3596 to DEST_CH[i]. This function return the number of destinations. */
3600 group_nodes_into_DFAstates (const re_dfa_t *dfa, const re_dfastate_t *state,
3601 re_node_set *dests_node, bitset_t *dests_ch)
3606 Idx ndests; /* Number of the destinations from 'state'. */
3607 bitset_t accepts; /* Characters a node can accept. */
3608 const re_node_set *cur_nodes = &state->nodes;
3609 bitset_empty (accepts);
3612 /* For all the nodes belonging to 'state', */
3613 for (i = 0; i < cur_nodes->nelem; ++i)
3615 re_token_t *node = &dfa->nodes[cur_nodes->elems[i]];
3616 re_token_type_t type = node->type;
3617 unsigned int constraint = node->constraint;
3619 /* Enumerate all single byte character this node can accept. */
3620 if (type == CHARACTER)
3621 bitset_set (accepts, node->opr.c);
3622 else if (type == SIMPLE_BRACKET)
3624 bitset_merge (accepts, node->opr.sbcset);
3626 else if (type == OP_PERIOD)
3628 #ifdef RE_ENABLE_I18N
3629 if (dfa->mb_cur_max > 1)
3630 bitset_merge (accepts, dfa->sb_char);
3633 bitset_set_all (accepts);
3634 if (!(dfa->syntax & RE_DOT_NEWLINE))
3635 bitset_clear (accepts, '\n');
3636 if (dfa->syntax & RE_DOT_NOT_NULL)
3637 bitset_clear (accepts, '\0');
3639 #ifdef RE_ENABLE_I18N
3640 else if (type == OP_UTF8_PERIOD)
3642 if (ASCII_CHARS % BITSET_WORD_BITS == 0)
3643 memset (accepts, -1, ASCII_CHARS / CHAR_BIT);
3645 bitset_merge (accepts, utf8_sb_map);
3646 if (!(dfa->syntax & RE_DOT_NEWLINE))
3647 bitset_clear (accepts, '\n');
3648 if (dfa->syntax & RE_DOT_NOT_NULL)
3649 bitset_clear (accepts, '\0');
3655 /* Check the 'accepts' and sift the characters which are not
3656 match it the context. */
3659 if (constraint & NEXT_NEWLINE_CONSTRAINT)
3661 bool accepts_newline = bitset_contain (accepts, NEWLINE_CHAR);
3662 bitset_empty (accepts);
3663 if (accepts_newline)
3664 bitset_set (accepts, NEWLINE_CHAR);
3668 if (constraint & NEXT_ENDBUF_CONSTRAINT)
3670 bitset_empty (accepts);
3674 if (constraint & NEXT_WORD_CONSTRAINT)
3676 bitset_word_t any_set = 0;
3677 if (type == CHARACTER && !node->word_char)
3679 bitset_empty (accepts);
3682 #ifdef RE_ENABLE_I18N
3683 if (dfa->mb_cur_max > 1)
3684 for (j = 0; j < BITSET_WORDS; ++j)
3685 any_set |= (accepts[j] &= (dfa->word_char[j] | ~dfa->sb_char[j]));
3688 for (j = 0; j < BITSET_WORDS; ++j)
3689 any_set |= (accepts[j] &= dfa->word_char[j]);
3693 if (constraint & NEXT_NOTWORD_CONSTRAINT)
3695 bitset_word_t any_set = 0;
3696 if (type == CHARACTER && node->word_char)
3698 bitset_empty (accepts);
3701 #ifdef RE_ENABLE_I18N
3702 if (dfa->mb_cur_max > 1)
3703 for (j = 0; j < BITSET_WORDS; ++j)
3704 any_set |= (accepts[j] &= ~(dfa->word_char[j] & dfa->sb_char[j]));
3707 for (j = 0; j < BITSET_WORDS; ++j)
3708 any_set |= (accepts[j] &= ~dfa->word_char[j]);
3714 /* Then divide 'accepts' into DFA states, or create a new
3715 state. Above, we make sure that accepts is not empty. */
3716 for (j = 0; j < ndests; ++j)
3718 bitset_t intersec; /* Intersection sets, see below. */
3720 /* Flags, see below. */
3721 bitset_word_t has_intersec, not_subset, not_consumed;
3723 /* Optimization, skip if this state doesn't accept the character. */
3724 if (type == CHARACTER && !bitset_contain (dests_ch[j], node->opr.c))
3727 /* Enumerate the intersection set of this state and 'accepts'. */
3729 for (k = 0; k < BITSET_WORDS; ++k)
3730 has_intersec |= intersec[k] = accepts[k] & dests_ch[j][k];
3731 /* And skip if the intersection set is empty. */
3735 /* Then check if this state is a subset of 'accepts'. */
3736 not_subset = not_consumed = 0;
3737 for (k = 0; k < BITSET_WORDS; ++k)
3739 not_subset |= remains[k] = ~accepts[k] & dests_ch[j][k];
3740 not_consumed |= accepts[k] = accepts[k] & ~dests_ch[j][k];
3743 /* If this state isn't a subset of 'accepts', create a
3744 new group state, which has the 'remains'. */
3747 bitset_copy (dests_ch[ndests], remains);
3748 bitset_copy (dests_ch[j], intersec);
3749 err = re_node_set_init_copy (dests_node + ndests, &dests_node[j]);
3750 if (BE (err != REG_NOERROR, 0))
3755 /* Put the position in the current group. */
3756 ok = re_node_set_insert (&dests_node[j], cur_nodes->elems[i]);
3760 /* If all characters are consumed, go to next node. */
3764 /* Some characters remain, create a new group. */
3767 bitset_copy (dests_ch[ndests], accepts);
3768 err = re_node_set_init_1 (dests_node + ndests, cur_nodes->elems[i]);
3769 if (BE (err != REG_NOERROR, 0))
3772 bitset_empty (accepts);
3777 for (j = 0; j < ndests; ++j)
3778 re_node_set_free (dests_node + j);
3782 #ifdef RE_ENABLE_I18N
3783 /* Check how many bytes the node 'dfa->nodes[node_idx]' accepts.
3784 Return the number of the bytes the node accepts.
3785 STR_IDX is the current index of the input string.
3787 This function handles the nodes which can accept one character, or
3788 one collating element like '.', '[a-z]', opposite to the other nodes
3789 can only accept one byte. */
3793 check_node_accept_bytes (const re_dfa_t *dfa, Idx node_idx,
3794 const re_string_t *input, Idx str_idx)
3796 const re_token_t *node = dfa->nodes + node_idx;
3797 int char_len, elem_len;
3800 if (BE (node->type == OP_UTF8_PERIOD, 0))
3802 unsigned char c = re_string_byte_at (input, str_idx), d;
3803 if (BE (c < 0xc2, 1))
3806 if (str_idx + 2 > input->len)
3809 d = re_string_byte_at (input, str_idx + 1);
3811 return (d < 0x80 || d > 0xbf) ? 0 : 2;
3815 if (c == 0xe0 && d < 0xa0)
3821 if (c == 0xf0 && d < 0x90)
3827 if (c == 0xf8 && d < 0x88)
3833 if (c == 0xfc && d < 0x84)
3839 if (str_idx + char_len > input->len)
3842 for (i = 1; i < char_len; ++i)
3844 d = re_string_byte_at (input, str_idx + i);
3845 if (d < 0x80 || d > 0xbf)
3851 char_len = re_string_char_size_at (input, str_idx);
3852 if (node->type == OP_PERIOD)
3856 /* FIXME: I don't think this if is needed, as both '\n'
3857 and '\0' are char_len == 1. */
3858 /* '.' accepts any one character except the following two cases. */
3859 if ((!(dfa->syntax & RE_DOT_NEWLINE) &&
3860 re_string_byte_at (input, str_idx) == '\n') ||
3861 ((dfa->syntax & RE_DOT_NOT_NULL) &&
3862 re_string_byte_at (input, str_idx) == '\0'))
3867 elem_len = re_string_elem_size_at (input, str_idx);
3868 if ((elem_len <= 1 && char_len <= 1) || char_len == 0)
3871 if (node->type == COMPLEX_BRACKET)
3873 const re_charset_t *cset = node->opr.mbcset;
3875 const unsigned char *pin
3876 = ((const unsigned char *) re_string_get_buffer (input) + str_idx);
3881 wchar_t wc = ((cset->nranges || cset->nchar_classes || cset->nmbchars)
3882 ? re_string_wchar_at (input, str_idx) : 0);
3884 /* match with multibyte character? */
3885 for (i = 0; i < cset->nmbchars; ++i)
3886 if (wc == cset->mbchars[i])
3888 match_len = char_len;
3889 goto check_node_accept_bytes_match;
3891 /* match with character_class? */
3892 for (i = 0; i < cset->nchar_classes; ++i)
3894 wctype_t wt = cset->char_classes[i];
3895 if (__iswctype (wc, wt))
3897 match_len = char_len;
3898 goto check_node_accept_bytes_match;
3903 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3906 unsigned int in_collseq = 0;
3907 const int32_t *table, *indirect;
3908 const unsigned char *weights, *extra;
3909 const char *collseqwc;
3910 /* This #include defines a local function! */
3911 # include <locale/weight.h>
3913 /* match with collating_symbol? */
3914 if (cset->ncoll_syms)
3915 extra = (const unsigned char *)
3916 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
3917 for (i = 0; i < cset->ncoll_syms; ++i)
3919 const unsigned char *coll_sym = extra + cset->coll_syms[i];
3920 /* Compare the length of input collating element and
3921 the length of current collating element. */
3922 if (*coll_sym != elem_len)
3924 /* Compare each bytes. */
3925 for (j = 0; j < *coll_sym; j++)
3926 if (pin[j] != coll_sym[1 + j])
3930 /* Match if every bytes is equal. */
3932 goto check_node_accept_bytes_match;
3938 if (elem_len <= char_len)
3940 collseqwc = _NL_CURRENT (LC_COLLATE, _NL_COLLATE_COLLSEQWC);
3941 in_collseq = __collseq_table_lookup (collseqwc, wc);
3944 in_collseq = find_collation_sequence_value (pin, elem_len);
3946 /* match with range expression? */
3947 /* FIXME: Implement rational ranges here, too. */
3948 for (i = 0; i < cset->nranges; ++i)
3949 if (cset->range_starts[i] <= in_collseq
3950 && in_collseq <= cset->range_ends[i])
3952 match_len = elem_len;
3953 goto check_node_accept_bytes_match;
3956 /* match with equivalence_class? */
3957 if (cset->nequiv_classes)
3959 const unsigned char *cp = pin;
3960 table = (const int32_t *)
3961 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
3962 weights = (const unsigned char *)
3963 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
3964 extra = (const unsigned char *)
3965 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
3966 indirect = (const int32_t *)
3967 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
3968 int32_t idx = findidx (&cp, elem_len);
3970 for (i = 0; i < cset->nequiv_classes; ++i)
3972 int32_t equiv_class_idx = cset->equiv_classes[i];
3973 size_t weight_len = weights[idx & 0xffffff];
3974 if (weight_len == weights[equiv_class_idx & 0xffffff]
3975 && (idx >> 24) == (equiv_class_idx >> 24))
3980 equiv_class_idx &= 0xffffff;
3982 while (cnt <= weight_len
3983 && (weights[equiv_class_idx + 1 + cnt]
3984 == weights[idx + 1 + cnt]))
3986 if (cnt > weight_len)
3988 match_len = elem_len;
3989 goto check_node_accept_bytes_match;
3998 /* match with range expression? */
3999 for (i = 0; i < cset->nranges; ++i)
4001 if (cset->range_starts[i] <= wc && wc <= cset->range_ends[i])
4003 match_len = char_len;
4004 goto check_node_accept_bytes_match;
4008 check_node_accept_bytes_match:
4009 if (!cset->non_match)
4016 return (elem_len > char_len) ? elem_len : char_len;
4025 find_collation_sequence_value (const unsigned char *mbs, size_t mbs_len)
4027 uint32_t nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
4032 /* No valid character. Match it as a single byte character. */
4033 const unsigned char *collseq = (const unsigned char *)
4034 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_COLLSEQMB);
4035 return collseq[mbs[0]];
4042 const unsigned char *extra = (const unsigned char *)
4043 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
4044 int32_t extrasize = (const unsigned char *)
4045 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB + 1) - extra;
4047 for (idx = 0; idx < extrasize;)
4051 int32_t elem_mbs_len;
4052 /* Skip the name of collating element name. */
4053 idx = idx + extra[idx] + 1;
4054 elem_mbs_len = extra[idx++];
4055 if (mbs_len == elem_mbs_len)
4057 for (mbs_cnt = 0; mbs_cnt < elem_mbs_len; ++mbs_cnt)
4058 if (extra[idx + mbs_cnt] != mbs[mbs_cnt])
4060 if (mbs_cnt == elem_mbs_len)
4061 /* Found the entry. */
4064 /* Skip the byte sequence of the collating element. */
4065 idx += elem_mbs_len;
4066 /* Adjust for the alignment. */
4067 idx = (idx + 3) & ~3;
4068 /* Skip the collation sequence value. */
4069 idx += sizeof (uint32_t);
4070 /* Skip the wide char sequence of the collating element. */
4071 idx = idx + sizeof (uint32_t) * (*(int32_t *) (extra + idx) + 1);
4072 /* If we found the entry, return the sequence value. */
4074 return *(uint32_t *) (extra + idx);
4075 /* Skip the collation sequence value. */
4076 idx += sizeof (uint32_t);
4082 #endif /* RE_ENABLE_I18N */
4084 /* Check whether the node accepts the byte which is IDX-th
4085 byte of the INPUT. */
4089 check_node_accept (const re_match_context_t *mctx, const re_token_t *node,
4093 ch = re_string_byte_at (&mctx->input, idx);
4097 if (node->opr.c != ch)
4101 case SIMPLE_BRACKET:
4102 if (!bitset_contain (node->opr.sbcset, ch))
4106 #ifdef RE_ENABLE_I18N
4107 case OP_UTF8_PERIOD:
4108 if (ch >= ASCII_CHARS)
4113 if ((ch == '\n' && !(mctx->dfa->syntax & RE_DOT_NEWLINE))
4114 || (ch == '\0' && (mctx->dfa->syntax & RE_DOT_NOT_NULL)))
4122 if (node->constraint)
4124 /* The node has constraints. Check whether the current context
4125 satisfies the constraints. */
4126 unsigned int context = re_string_context_at (&mctx->input, idx,
4128 if (NOT_SATISFY_NEXT_CONSTRAINT (node->constraint, context))
4135 /* Extend the buffers, if the buffers have run out. */
4137 static reg_errcode_t
4138 internal_function __attribute_warn_unused_result__
4139 extend_buffers (re_match_context_t *mctx, int min_len)
4142 re_string_t *pstr = &mctx->input;
4144 /* Avoid overflow. */
4145 if (BE (MIN (IDX_MAX, SIZE_MAX / sizeof (re_dfastate_t *)) / 2
4146 <= pstr->bufs_len, 0))
4149 /* Double the lengths of the buffers, but allocate at least MIN_LEN. */
4150 ret = re_string_realloc_buffers (pstr,
4152 MIN (pstr->len, pstr->bufs_len * 2)));
4153 if (BE (ret != REG_NOERROR, 0))
4156 if (mctx->state_log != NULL)
4158 /* And double the length of state_log. */
4159 /* XXX We have no indication of the size of this buffer. If this
4160 allocation fail we have no indication that the state_log array
4161 does not have the right size. */
4162 re_dfastate_t **new_array = re_realloc (mctx->state_log, re_dfastate_t *,
4163 pstr->bufs_len + 1);
4164 if (BE (new_array == NULL, 0))
4166 mctx->state_log = new_array;
4169 /* Then reconstruct the buffers. */
4172 #ifdef RE_ENABLE_I18N
4173 if (pstr->mb_cur_max > 1)
4175 ret = build_wcs_upper_buffer (pstr);
4176 if (BE (ret != REG_NOERROR, 0))
4180 #endif /* RE_ENABLE_I18N */
4181 build_upper_buffer (pstr);
4185 #ifdef RE_ENABLE_I18N
4186 if (pstr->mb_cur_max > 1)
4187 build_wcs_buffer (pstr);
4189 #endif /* RE_ENABLE_I18N */
4191 if (pstr->trans != NULL)
4192 re_string_translate_buffer (pstr);
4199 /* Functions for matching context. */
4201 /* Initialize MCTX. */
4203 static reg_errcode_t
4204 internal_function __attribute_warn_unused_result__
4205 match_ctx_init (re_match_context_t *mctx, int eflags, Idx n)
4207 mctx->eflags = eflags;
4208 mctx->match_last = REG_MISSING;
4211 /* Avoid overflow. */
4212 size_t max_object_size =
4213 MAX (sizeof (struct re_backref_cache_entry),
4214 sizeof (re_sub_match_top_t *));
4215 if (BE (MIN (IDX_MAX, SIZE_MAX / max_object_size) < n, 0))
4218 mctx->bkref_ents = re_malloc (struct re_backref_cache_entry, n);
4219 mctx->sub_tops = re_malloc (re_sub_match_top_t *, n);
4220 if (BE (mctx->bkref_ents == NULL || mctx->sub_tops == NULL, 0))
4223 /* Already zero-ed by the caller.
4225 mctx->bkref_ents = NULL;
4226 mctx->nbkref_ents = 0;
4227 mctx->nsub_tops = 0; */
4228 mctx->abkref_ents = n;
4229 mctx->max_mb_elem_len = 1;
4230 mctx->asub_tops = n;
4234 /* Clean the entries which depend on the current input in MCTX.
4235 This function must be invoked when the matcher changes the start index
4236 of the input, or changes the input string. */
4240 match_ctx_clean (re_match_context_t *mctx)
4243 for (st_idx = 0; st_idx < mctx->nsub_tops; ++st_idx)
4246 re_sub_match_top_t *top = mctx->sub_tops[st_idx];
4247 for (sl_idx = 0; sl_idx < top->nlasts; ++sl_idx)
4249 re_sub_match_last_t *last = top->lasts[sl_idx];
4250 re_free (last->path.array);
4253 re_free (top->lasts);
4256 re_free (top->path->array);
4257 re_free (top->path);
4262 mctx->nsub_tops = 0;
4263 mctx->nbkref_ents = 0;
4266 /* Free all the memory associated with MCTX. */
4270 match_ctx_free (re_match_context_t *mctx)
4272 /* First, free all the memory associated with MCTX->SUB_TOPS. */
4273 match_ctx_clean (mctx);
4274 re_free (mctx->sub_tops);
4275 re_free (mctx->bkref_ents);
4278 /* Add a new backreference entry to MCTX.
4279 Note that we assume that caller never call this function with duplicate
4280 entry, and call with STR_IDX which isn't smaller than any existing entry.
4283 static reg_errcode_t
4284 internal_function __attribute_warn_unused_result__
4285 match_ctx_add_entry (re_match_context_t *mctx, Idx node, Idx str_idx, Idx from,
4288 if (mctx->nbkref_ents >= mctx->abkref_ents)
4290 struct re_backref_cache_entry* new_entry;
4291 new_entry = re_realloc (mctx->bkref_ents, struct re_backref_cache_entry,
4292 mctx->abkref_ents * 2);
4293 if (BE (new_entry == NULL, 0))
4295 re_free (mctx->bkref_ents);
4298 mctx->bkref_ents = new_entry;
4299 memset (mctx->bkref_ents + mctx->nbkref_ents, '\0',
4300 sizeof (struct re_backref_cache_entry) * mctx->abkref_ents);
4301 mctx->abkref_ents *= 2;
4303 if (mctx->nbkref_ents > 0
4304 && mctx->bkref_ents[mctx->nbkref_ents - 1].str_idx == str_idx)
4305 mctx->bkref_ents[mctx->nbkref_ents - 1].more = 1;
4307 mctx->bkref_ents[mctx->nbkref_ents].node = node;
4308 mctx->bkref_ents[mctx->nbkref_ents].str_idx = str_idx;
4309 mctx->bkref_ents[mctx->nbkref_ents].subexp_from = from;
4310 mctx->bkref_ents[mctx->nbkref_ents].subexp_to = to;
4312 /* This is a cache that saves negative results of check_dst_limits_calc_pos.
4313 If bit N is clear, means that this entry won't epsilon-transition to
4314 an OP_OPEN_SUBEXP or OP_CLOSE_SUBEXP for the N+1-th subexpression. If
4315 it is set, check_dst_limits_calc_pos_1 will recurse and try to find one
4318 A backreference does not epsilon-transition unless it is empty, so set
4319 to all zeros if FROM != TO. */
4320 mctx->bkref_ents[mctx->nbkref_ents].eps_reachable_subexps_map
4321 = (from == to ? -1 : 0);
4323 mctx->bkref_ents[mctx->nbkref_ents++].more = 0;
4324 if (mctx->max_mb_elem_len < to - from)
4325 mctx->max_mb_elem_len = to - from;
4329 /* Return the first entry with the same str_idx, or REG_MISSING if none is
4330 found. Note that MCTX->BKREF_ENTS is already sorted by MCTX->STR_IDX. */
4334 search_cur_bkref_entry (const re_match_context_t *mctx, Idx str_idx)
4336 Idx left, right, mid, last;
4337 last = right = mctx->nbkref_ents;
4338 for (left = 0; left < right;)
4340 mid = (left + right) / 2;
4341 if (mctx->bkref_ents[mid].str_idx < str_idx)
4346 if (left < last && mctx->bkref_ents[left].str_idx == str_idx)
4352 /* Register the node NODE, whose type is OP_OPEN_SUBEXP, and which matches
4355 static reg_errcode_t
4356 internal_function __attribute_warn_unused_result__
4357 match_ctx_add_subtop (re_match_context_t *mctx, Idx node, Idx str_idx)
4360 assert (mctx->sub_tops != NULL);
4361 assert (mctx->asub_tops > 0);
4363 if (BE (mctx->nsub_tops == mctx->asub_tops, 0))
4365 Idx new_asub_tops = mctx->asub_tops * 2;
4366 re_sub_match_top_t **new_array = re_realloc (mctx->sub_tops,
4367 re_sub_match_top_t *,
4369 if (BE (new_array == NULL, 0))
4371 mctx->sub_tops = new_array;
4372 mctx->asub_tops = new_asub_tops;
4374 mctx->sub_tops[mctx->nsub_tops] = calloc (1, sizeof (re_sub_match_top_t));
4375 if (BE (mctx->sub_tops[mctx->nsub_tops] == NULL, 0))
4377 mctx->sub_tops[mctx->nsub_tops]->node = node;
4378 mctx->sub_tops[mctx->nsub_tops++]->str_idx = str_idx;
4382 /* Register the node NODE, whose type is OP_CLOSE_SUBEXP, and which matches
4383 at STR_IDX, whose corresponding OP_OPEN_SUBEXP is SUB_TOP. */
4385 static re_sub_match_last_t *
4387 match_ctx_add_sublast (re_sub_match_top_t *subtop, Idx node, Idx str_idx)
4389 re_sub_match_last_t *new_entry;
4390 if (BE (subtop->nlasts == subtop->alasts, 0))
4392 Idx new_alasts = 2 * subtop->alasts + 1;
4393 re_sub_match_last_t **new_array = re_realloc (subtop->lasts,
4394 re_sub_match_last_t *,
4396 if (BE (new_array == NULL, 0))
4398 subtop->lasts = new_array;
4399 subtop->alasts = new_alasts;
4401 new_entry = calloc (1, sizeof (re_sub_match_last_t));
4402 if (BE (new_entry != NULL, 1))
4404 subtop->lasts[subtop->nlasts] = new_entry;
4405 new_entry->node = node;
4406 new_entry->str_idx = str_idx;
4414 sift_ctx_init (re_sift_context_t *sctx, re_dfastate_t **sifted_sts,
4415 re_dfastate_t **limited_sts, Idx last_node, Idx last_str_idx)
4417 sctx->sifted_states = sifted_sts;
4418 sctx->limited_states = limited_sts;
4419 sctx->last_node = last_node;
4420 sctx->last_str_idx = last_str_idx;
4421 re_node_set_init_empty (&sctx->limits);