1 /* Extended regular expression matching and search library.
2 Copyright (C) 2002-2017 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4 Contributed by Isamu Hasegawa <isamu@yamato.ibm.com>.
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2.1 of the License, or (at your option) any later version.
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; if not, see
18 <http://www.gnu.org/licenses/>. */
22 static reg_errcode_t
match_ctx_init (re_match_context_t
*cache
, int eflags
,
24 static void match_ctx_clean (re_match_context_t
*mctx
);
25 static void match_ctx_free (re_match_context_t
*cache
);
26 static reg_errcode_t
match_ctx_add_entry (re_match_context_t
*cache
, int node
,
27 int str_idx
, int from
, int to
);
28 static int search_cur_bkref_entry (const re_match_context_t
*mctx
,
30 static reg_errcode_t
match_ctx_add_subtop (re_match_context_t
*mctx
, int node
,
32 static re_sub_match_last_t
* match_ctx_add_sublast (re_sub_match_top_t
*subtop
,
33 int node
, int str_idx
);
34 static void sift_ctx_init (re_sift_context_t
*sctx
, re_dfastate_t
**sifted_sts
,
35 re_dfastate_t
**limited_sts
, int last_node
,
37 static reg_errcode_t
re_search_internal (const regex_t
*preg
,
38 const char *string
, int length
,
39 int start
, int range
, int stop
,
40 size_t nmatch
, regmatch_t pmatch
[],
42 static int re_search_2_stub (struct re_pattern_buffer
*bufp
,
43 const char *string1
, int length1
,
44 const char *string2
, int length2
,
45 int start
, int range
, struct re_registers
*regs
,
46 int stop
, int ret_len
);
47 static int re_search_stub (struct re_pattern_buffer
*bufp
,
48 const char *string
, int length
, int start
,
49 int range
, int stop
, struct re_registers
*regs
,
51 static unsigned re_copy_regs (struct re_registers
*regs
, regmatch_t
*pmatch
,
52 int nregs
, int regs_allocated
);
53 static reg_errcode_t
prune_impossible_nodes (re_match_context_t
*mctx
);
54 static int check_matching (re_match_context_t
*mctx
, int fl_longest_match
,
56 static int check_halt_state_context (const re_match_context_t
*mctx
,
57 const re_dfastate_t
*state
, int idx
);
58 static void update_regs (const re_dfa_t
*dfa
, regmatch_t
*pmatch
,
59 regmatch_t
*prev_idx_match
, int cur_node
,
60 int cur_idx
, int nmatch
);
61 static reg_errcode_t
push_fail_stack (struct re_fail_stack_t
*fs
,
62 int str_idx
, int dest_node
, int nregs
,
64 re_node_set
*eps_via_nodes
);
65 static reg_errcode_t
set_regs (const regex_t
*preg
,
66 const re_match_context_t
*mctx
,
67 size_t nmatch
, regmatch_t
*pmatch
,
69 static reg_errcode_t
free_fail_stack_return (struct re_fail_stack_t
*fs
);
72 static int sift_states_iter_mb (const re_match_context_t
*mctx
,
73 re_sift_context_t
*sctx
,
74 int node_idx
, int str_idx
, int max_str_idx
);
75 #endif /* RE_ENABLE_I18N */
76 static reg_errcode_t
sift_states_backward (const re_match_context_t
*mctx
,
77 re_sift_context_t
*sctx
);
78 static reg_errcode_t
build_sifted_states (const re_match_context_t
*mctx
,
79 re_sift_context_t
*sctx
, int str_idx
,
80 re_node_set
*cur_dest
);
81 static reg_errcode_t
update_cur_sifted_state (const re_match_context_t
*mctx
,
82 re_sift_context_t
*sctx
,
84 re_node_set
*dest_nodes
);
85 static reg_errcode_t
add_epsilon_src_nodes (const re_dfa_t
*dfa
,
86 re_node_set
*dest_nodes
,
87 const re_node_set
*candidates
);
88 static int check_dst_limits (const re_match_context_t
*mctx
,
90 int dst_node
, int dst_idx
, int src_node
,
92 static int check_dst_limits_calc_pos_1 (const re_match_context_t
*mctx
,
93 int boundaries
, int subexp_idx
,
94 int from_node
, int bkref_idx
);
95 static int check_dst_limits_calc_pos (const re_match_context_t
*mctx
,
96 int limit
, int subexp_idx
,
97 int node
, int str_idx
,
99 static reg_errcode_t
check_subexp_limits (const re_dfa_t
*dfa
,
100 re_node_set
*dest_nodes
,
101 const re_node_set
*candidates
,
103 struct re_backref_cache_entry
*bkref_ents
,
105 static reg_errcode_t
sift_states_bkref (const re_match_context_t
*mctx
,
106 re_sift_context_t
*sctx
,
108 const re_node_set
*candidates
);
109 static reg_errcode_t
merge_state_array (const re_dfa_t
*dfa
,
111 re_dfastate_t
**src
, int num
);
112 static re_dfastate_t
*find_recover_state (reg_errcode_t
*err
,
113 re_match_context_t
*mctx
);
114 static re_dfastate_t
*transit_state (reg_errcode_t
*err
,
115 re_match_context_t
*mctx
,
116 re_dfastate_t
*state
);
117 static re_dfastate_t
*merge_state_with_log (reg_errcode_t
*err
,
118 re_match_context_t
*mctx
,
119 re_dfastate_t
*next_state
);
120 static reg_errcode_t
check_subexp_matching_top (re_match_context_t
*mctx
,
121 re_node_set
*cur_nodes
,
124 static re_dfastate_t
*transit_state_sb (reg_errcode_t
*err
,
125 re_match_context_t
*mctx
,
126 re_dfastate_t
*pstate
);
128 #ifdef RE_ENABLE_I18N
129 static reg_errcode_t
transit_state_mb (re_match_context_t
*mctx
,
130 re_dfastate_t
*pstate
);
131 #endif /* RE_ENABLE_I18N */
132 static reg_errcode_t
transit_state_bkref (re_match_context_t
*mctx
,
133 const re_node_set
*nodes
);
134 static reg_errcode_t
get_subexp (re_match_context_t
*mctx
,
135 int bkref_node
, int bkref_str_idx
);
136 static reg_errcode_t
get_subexp_sub (re_match_context_t
*mctx
,
137 const re_sub_match_top_t
*sub_top
,
138 re_sub_match_last_t
*sub_last
,
139 int bkref_node
, int bkref_str
);
140 static int find_subexp_node (const re_dfa_t
*dfa
, const re_node_set
*nodes
,
141 int subexp_idx
, int type
);
142 static reg_errcode_t
check_arrival (re_match_context_t
*mctx
,
143 state_array_t
*path
, int top_node
,
144 int top_str
, int last_node
, int last_str
,
146 static reg_errcode_t
check_arrival_add_next_nodes (re_match_context_t
*mctx
,
148 re_node_set
*cur_nodes
,
149 re_node_set
*next_nodes
);
150 static reg_errcode_t
check_arrival_expand_ecl (const re_dfa_t
*dfa
,
151 re_node_set
*cur_nodes
,
152 int ex_subexp
, int type
);
153 static reg_errcode_t
check_arrival_expand_ecl_sub (const re_dfa_t
*dfa
,
154 re_node_set
*dst_nodes
,
155 int target
, int ex_subexp
,
157 static reg_errcode_t
expand_bkref_cache (re_match_context_t
*mctx
,
158 re_node_set
*cur_nodes
, int cur_str
,
159 int subexp_num
, int type
);
160 static int build_trtable (const re_dfa_t
*dfa
, re_dfastate_t
*state
);
161 #ifdef RE_ENABLE_I18N
162 static int check_node_accept_bytes (const re_dfa_t
*dfa
, int node_idx
,
163 const re_string_t
*input
, int idx
);
165 static unsigned int find_collation_sequence_value (const unsigned char *mbs
,
168 #endif /* RE_ENABLE_I18N */
169 static int group_nodes_into_DFAstates (const re_dfa_t
*dfa
,
170 const re_dfastate_t
*state
,
171 re_node_set
*states_node
,
172 bitset_t
*states_ch
);
173 static int check_node_accept (const re_match_context_t
*mctx
,
174 const re_token_t
*node
, int idx
);
175 static reg_errcode_t
extend_buffers (re_match_context_t
*mctx
, int min_len
);
177 /* Entry point for POSIX code. */
179 /* regexec searches for a given pattern, specified by PREG, in the
182 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
183 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
184 least NMATCH elements, and we set them to the offsets of the
185 corresponding matched substrings.
187 EFLAGS specifies `execution flags' which affect matching: if
188 REG_NOTBOL is set, then ^ does not match at the beginning of the
189 string; if REG_NOTEOL is set, then $ does not match at the end.
191 We return 0 if we find a match and REG_NOMATCH if not. */
194 regexec (const regex_t
*__restrict preg
, const char *__restrict string
,
195 size_t nmatch
, regmatch_t pmatch
[], int eflags
)
199 re_dfa_t
*dfa
= (re_dfa_t
*) preg
->buffer
;
201 if (eflags
& ~(REG_NOTBOL
| REG_NOTEOL
| REG_STARTEND
))
204 if (eflags
& REG_STARTEND
)
206 start
= pmatch
[0].rm_so
;
207 length
= pmatch
[0].rm_eo
;
212 length
= strlen (string
);
215 __libc_lock_lock (dfa
->lock
);
217 err
= re_search_internal (preg
, string
, length
, start
, length
- start
,
218 length
, 0, NULL
, eflags
);
220 err
= re_search_internal (preg
, string
, length
, start
, length
- start
,
221 length
, nmatch
, pmatch
, eflags
);
222 __libc_lock_unlock (dfa
->lock
);
223 return err
!= REG_NOERROR
;
227 # include <shlib-compat.h>
228 versioned_symbol (libc
, __regexec
, regexec
, GLIBC_2_3_4
);
230 # if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_3_4)
231 __typeof__ (__regexec
) __compat_regexec
;
234 attribute_compat_text_section
235 __compat_regexec (const regex_t
*__restrict preg
,
236 const char *__restrict string
, size_t nmatch
,
237 regmatch_t pmatch
[], int eflags
)
239 return regexec (preg
, string
, nmatch
, pmatch
,
240 eflags
& (REG_NOTBOL
| REG_NOTEOL
));
242 compat_symbol (libc
, __compat_regexec
, regexec
, GLIBC_2_0
);
246 /* Entry points for GNU code. */
248 /* re_match, re_search, re_match_2, re_search_2
250 The former two functions operate on STRING with length LENGTH,
251 while the later two operate on concatenation of STRING1 and STRING2
252 with lengths LENGTH1 and LENGTH2, respectively.
254 re_match() matches the compiled pattern in BUFP against the string,
255 starting at index START.
257 re_search() first tries matching at index START, then it tries to match
258 starting from index START + 1, and so on. The last start position tried
259 is START + RANGE. (Thus RANGE = 0 forces re_search to operate the same
262 The parameter STOP of re_{match,search}_2 specifies that no match exceeding
263 the first STOP characters of the concatenation of the strings should be
266 If REGS is not NULL, and BUFP->no_sub is not set, the offsets of the match
267 and all groups is stroed in REGS. (For the "_2" variants, the offsets are
268 computed relative to the concatenation, not relative to the individual
271 On success, re_match* functions return the length of the match, re_search*
272 return the position of the start of the match. Return value -1 means no
273 match was found and -2 indicates an internal error. */
276 re_match (struct re_pattern_buffer
*bufp
, const char *string
, int length
,
277 int start
, struct re_registers
*regs
)
279 return re_search_stub (bufp
, string
, length
, start
, 0, length
, regs
, 1);
282 weak_alias (__re_match
, re_match
)
286 re_search (struct re_pattern_buffer
*bufp
, const char *string
, int length
,
287 int start
, int range
, struct re_registers
*regs
)
289 return re_search_stub (bufp
, string
, length
, start
, range
, length
, regs
, 0);
292 weak_alias (__re_search
, re_search
)
296 re_match_2 (struct re_pattern_buffer
*bufp
, const char *string1
, int length1
,
297 const char *string2
, int length2
, int start
,
298 struct re_registers
*regs
, int stop
)
300 return re_search_2_stub (bufp
, string1
, length1
, string2
, length2
,
301 start
, 0, regs
, stop
, 1);
304 weak_alias (__re_match_2
, re_match_2
)
308 re_search_2 (struct re_pattern_buffer
*bufp
, const char *string1
, int length1
,
309 const char *string2
, int length2
, int start
, int range
,
310 struct re_registers
*regs
, int stop
)
312 return re_search_2_stub (bufp
, string1
, length1
, string2
, length2
,
313 start
, range
, regs
, stop
, 0);
316 weak_alias (__re_search_2
, re_search_2
)
320 re_search_2_stub (struct re_pattern_buffer
*bufp
, const char *string1
,
321 int length1
, const char *string2
, int length2
, int start
,
322 int range
, struct re_registers
*regs
,
323 int stop
, int ret_len
)
327 int len
= length1
+ length2
;
330 if (BE (length1
< 0 || length2
< 0 || stop
< 0 || len
< length1
, 0))
333 /* Concatenate the strings. */
337 s
= re_malloc (char, len
);
339 if (BE (s
== NULL
, 0))
342 memcpy (__mempcpy (s
, string1
, length1
), string2
, length2
);
344 memcpy (s
, string1
, length1
);
345 memcpy (s
+ length1
, string2
, length2
);
354 rval
= re_search_stub (bufp
, str
, len
, start
, range
, stop
, regs
, ret_len
);
359 /* The parameters have the same meaning as those of re_search.
360 Additional parameters:
361 If RET_LEN is nonzero the length of the match is returned (re_match style);
362 otherwise the position of the match is returned. */
365 re_search_stub (struct re_pattern_buffer
*bufp
, const char *string
, int length
,
366 int start
, int range
, int stop
, struct re_registers
*regs
,
369 reg_errcode_t result
;
373 re_dfa_t
*dfa
= (re_dfa_t
*) bufp
->buffer
;
375 /* Check for out-of-range. */
376 if (BE (start
< 0 || start
> length
, 0))
378 if (BE (start
+ range
> length
, 0))
379 range
= length
- start
;
380 else if (BE (start
+ range
< 0, 0))
383 __libc_lock_lock (dfa
->lock
);
385 eflags
|= (bufp
->not_bol
) ? REG_NOTBOL
: 0;
386 eflags
|= (bufp
->not_eol
) ? REG_NOTEOL
: 0;
388 /* Compile fastmap if we haven't yet. */
389 if (range
> 0 && bufp
->fastmap
!= NULL
&& !bufp
->fastmap_accurate
)
390 re_compile_fastmap (bufp
);
392 if (BE (bufp
->no_sub
, 0))
395 /* We need at least 1 register. */
398 else if (BE (bufp
->regs_allocated
== REGS_FIXED
&&
399 regs
->num_regs
< bufp
->re_nsub
+ 1, 0))
401 nregs
= regs
->num_regs
;
402 if (BE (nregs
< 1, 0))
404 /* Nothing can be copied to regs. */
410 nregs
= bufp
->re_nsub
+ 1;
411 pmatch
= re_malloc (regmatch_t
, nregs
);
412 if (BE (pmatch
== NULL
, 0))
418 result
= re_search_internal (bufp
, string
, length
, start
, range
, stop
,
419 nregs
, pmatch
, eflags
);
423 /* I hope we needn't fill ther regs with -1's when no match was found. */
424 if (result
!= REG_NOERROR
)
426 else if (regs
!= NULL
)
428 /* If caller wants register contents data back, copy them. */
429 bufp
->regs_allocated
= re_copy_regs (regs
, pmatch
, nregs
,
430 bufp
->regs_allocated
);
431 if (BE (bufp
->regs_allocated
== REGS_UNALLOCATED
, 0))
435 if (BE (rval
== 0, 1))
439 assert (pmatch
[0].rm_so
== start
);
440 rval
= pmatch
[0].rm_eo
- start
;
443 rval
= pmatch
[0].rm_so
;
447 __libc_lock_unlock (dfa
->lock
);
452 re_copy_regs (struct re_registers
*regs
, regmatch_t
*pmatch
, int nregs
,
455 int rval
= REGS_REALLOCATE
;
457 int need_regs
= nregs
+ 1;
458 /* We need one extra element beyond `num_regs' for the `-1' marker GNU code
461 /* Have the register data arrays been allocated? */
462 if (regs_allocated
== REGS_UNALLOCATED
)
463 { /* No. So allocate them with malloc. */
464 regs
->start
= re_malloc (regoff_t
, need_regs
);
465 if (BE (regs
->start
== NULL
, 0))
466 return REGS_UNALLOCATED
;
467 regs
->end
= re_malloc (regoff_t
, need_regs
);
468 if (BE (regs
->end
== NULL
, 0))
470 re_free (regs
->start
);
471 return REGS_UNALLOCATED
;
473 regs
->num_regs
= need_regs
;
475 else if (regs_allocated
== REGS_REALLOCATE
)
476 { /* Yes. If we need more elements than were already
477 allocated, reallocate them. If we need fewer, just
479 if (BE (need_regs
> regs
->num_regs
, 0))
481 regoff_t
*new_start
= re_realloc (regs
->start
, regoff_t
, need_regs
);
483 if (BE (new_start
== NULL
, 0))
484 return REGS_UNALLOCATED
;
485 new_end
= re_realloc (regs
->end
, regoff_t
, need_regs
);
486 if (BE (new_end
== NULL
, 0))
489 return REGS_UNALLOCATED
;
491 regs
->start
= new_start
;
493 regs
->num_regs
= need_regs
;
498 assert (regs_allocated
== REGS_FIXED
);
499 /* This function may not be called with REGS_FIXED and nregs too big. */
500 assert (regs
->num_regs
>= nregs
);
505 for (i
= 0; i
< nregs
; ++i
)
507 regs
->start
[i
] = pmatch
[i
].rm_so
;
508 regs
->end
[i
] = pmatch
[i
].rm_eo
;
510 for ( ; i
< regs
->num_regs
; ++i
)
511 regs
->start
[i
] = regs
->end
[i
] = -1;
516 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
517 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
518 this memory for recording register information. STARTS and ENDS
519 must be allocated using the malloc library routine, and must each
520 be at least NUM_REGS * sizeof (regoff_t) bytes long.
522 If NUM_REGS == 0, then subsequent matches should allocate their own
525 Unless this function is called, the first search or match using
526 PATTERN_BUFFER will allocate its own register data, without
527 freeing the old data. */
530 re_set_registers (struct re_pattern_buffer
*bufp
, struct re_registers
*regs
,
531 unsigned num_regs
, regoff_t
*starts
, regoff_t
*ends
)
535 bufp
->regs_allocated
= REGS_REALLOCATE
;
536 regs
->num_regs
= num_regs
;
537 regs
->start
= starts
;
542 bufp
->regs_allocated
= REGS_UNALLOCATED
;
544 regs
->start
= regs
->end
= (regoff_t
*) 0;
548 weak_alias (__re_set_registers
, re_set_registers
)
551 /* Entry points compatible with 4.2 BSD regex library. We don't define
552 them unless specifically requested. */
554 #if defined _REGEX_RE_COMP || defined _LIBC
559 re_exec (const char *s
)
561 return 0 == regexec (&re_comp_buf
, s
, 0, NULL
, 0);
563 #endif /* _REGEX_RE_COMP */
565 /* Internal entry point. */
567 /* Searches for a compiled pattern PREG in the string STRING, whose
568 length is LENGTH. NMATCH, PMATCH, and EFLAGS have the same
569 mingings with regexec. START, and RANGE have the same meanings
571 Return REG_NOERROR if we find a match, and REG_NOMATCH if not,
572 otherwise return the error code.
573 Note: We assume front end functions already check ranges.
574 (START + RANGE >= 0 && START + RANGE <= LENGTH) */
577 __attribute_warn_unused_result__
578 re_search_internal (const regex_t
*preg
, const char *string
, int length
,
579 int start
, int range
, int stop
, size_t nmatch
,
580 regmatch_t pmatch
[], int eflags
)
583 const re_dfa_t
*dfa
= (const re_dfa_t
*) preg
->buffer
;
584 int left_lim
, right_lim
, incr
;
585 int fl_longest_match
, match_first
, match_kind
, match_last
= -1;
588 #if defined _LIBC || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L)
589 re_match_context_t mctx
= { .dfa
= dfa
};
591 re_match_context_t mctx
;
593 char *fastmap
= (preg
->fastmap
!= NULL
&& preg
->fastmap_accurate
594 && range
&& !preg
->can_be_null
) ? preg
->fastmap
: NULL
;
595 RE_TRANSLATE_TYPE t
= preg
->translate
;
597 #if !(defined _LIBC || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L))
598 memset (&mctx
, '\0', sizeof (re_match_context_t
));
602 extra_nmatch
= (nmatch
> preg
->re_nsub
) ? nmatch
- (preg
->re_nsub
+ 1) : 0;
603 nmatch
-= extra_nmatch
;
605 /* Check if the DFA haven't been compiled. */
606 if (BE (preg
->used
== 0 || dfa
->init_state
== NULL
607 || dfa
->init_state_word
== NULL
|| dfa
->init_state_nl
== NULL
608 || dfa
->init_state_begbuf
== NULL
, 0))
612 /* We assume front-end functions already check them. */
613 assert (start
+ range
>= 0 && start
+ range
<= length
);
616 /* If initial states with non-begbuf contexts have no elements,
617 the regex must be anchored. If preg->newline_anchor is set,
618 we'll never use init_state_nl, so do not check it. */
619 if (dfa
->init_state
->nodes
.nelem
== 0
620 && dfa
->init_state_word
->nodes
.nelem
== 0
621 && (dfa
->init_state_nl
->nodes
.nelem
== 0
622 || !preg
->newline_anchor
))
624 if (start
!= 0 && start
+ range
!= 0)
629 /* We must check the longest matching, if nmatch > 0. */
630 fl_longest_match
= (nmatch
!= 0 || dfa
->nbackref
);
632 err
= re_string_allocate (&mctx
.input
, string
, length
, dfa
->nodes_len
+ 1,
633 preg
->translate
, preg
->syntax
& RE_ICASE
, dfa
);
634 if (BE (err
!= REG_NOERROR
, 0))
636 mctx
.input
.stop
= stop
;
637 mctx
.input
.raw_stop
= stop
;
638 mctx
.input
.newline_anchor
= preg
->newline_anchor
;
640 err
= match_ctx_init (&mctx
, eflags
, dfa
->nbackref
* 2);
641 if (BE (err
!= REG_NOERROR
, 0))
644 /* We will log all the DFA states through which the dfa pass,
645 if nmatch > 1, or this dfa has "multibyte node", which is a
646 back-reference or a node which can accept multibyte character or
647 multi character collating element. */
648 if (nmatch
> 1 || dfa
->has_mb_node
)
650 /* Avoid overflow. */
651 if (BE (SIZE_MAX
/ sizeof (re_dfastate_t
*) <= mctx
.input
.bufs_len
, 0))
657 mctx
.state_log
= re_malloc (re_dfastate_t
*, mctx
.input
.bufs_len
+ 1);
658 if (BE (mctx
.state_log
== NULL
, 0))
665 mctx
.state_log
= NULL
;
668 mctx
.input
.tip_context
= (eflags
& REG_NOTBOL
) ? CONTEXT_BEGBUF
669 : CONTEXT_NEWLINE
| CONTEXT_BEGBUF
;
671 /* Check incrementally whether of not the input string match. */
672 incr
= (range
< 0) ? -1 : 1;
673 left_lim
= (range
< 0) ? start
+ range
: start
;
674 right_lim
= (range
< 0) ? start
: start
+ range
;
675 sb
= dfa
->mb_cur_max
== 1;
678 ? ((sb
|| !(preg
->syntax
& RE_ICASE
|| t
) ? 4 : 0)
679 | (range
>= 0 ? 2 : 0)
680 | (t
!= NULL
? 1 : 0))
683 for (;; match_first
+= incr
)
686 if (match_first
< left_lim
|| right_lim
< match_first
)
689 /* Advance as rapidly as possible through the string, until we
690 find a plausible place to start matching. This may be done
691 with varying efficiency, so there are various possibilities:
692 only the most common of them are specialized, in order to
693 save on code size. We use a switch statement for speed. */
701 /* Fastmap with single-byte translation, match forward. */
702 while (BE (match_first
< right_lim
, 1)
703 && !fastmap
[t
[(unsigned char) string
[match_first
]]])
705 goto forward_match_found_start_or_reached_end
;
708 /* Fastmap without translation, match forward. */
709 while (BE (match_first
< right_lim
, 1)
710 && !fastmap
[(unsigned char) string
[match_first
]])
713 forward_match_found_start_or_reached_end
:
714 if (BE (match_first
== right_lim
, 0))
716 ch
= match_first
>= length
717 ? 0 : (unsigned char) string
[match_first
];
718 if (!fastmap
[t
? t
[ch
] : ch
])
725 /* Fastmap without multi-byte translation, match backwards. */
726 while (match_first
>= left_lim
)
728 ch
= match_first
>= length
729 ? 0 : (unsigned char) string
[match_first
];
730 if (fastmap
[t
? t
[ch
] : ch
])
734 if (match_first
< left_lim
)
739 /* In this case, we can't determine easily the current byte,
740 since it might be a component byte of a multibyte
741 character. Then we use the constructed buffer instead. */
744 /* If MATCH_FIRST is out of the valid range, reconstruct the
746 unsigned int offset
= match_first
- mctx
.input
.raw_mbs_idx
;
747 if (BE (offset
>= (unsigned int) mctx
.input
.valid_raw_len
, 0))
749 err
= re_string_reconstruct (&mctx
.input
, match_first
,
751 if (BE (err
!= REG_NOERROR
, 0))
754 offset
= match_first
- mctx
.input
.raw_mbs_idx
;
756 /* If MATCH_FIRST is out of the buffer, leave it as '\0'.
757 Note that MATCH_FIRST must not be smaller than 0. */
758 ch
= (match_first
>= length
759 ? 0 : re_string_byte_at (&mctx
.input
, offset
));
763 if (match_first
< left_lim
|| match_first
> right_lim
)
772 /* Reconstruct the buffers so that the matcher can assume that
773 the matching starts from the beginning of the buffer. */
774 err
= re_string_reconstruct (&mctx
.input
, match_first
, eflags
);
775 if (BE (err
!= REG_NOERROR
, 0))
778 #ifdef RE_ENABLE_I18N
779 /* Don't consider this char as a possible match start if it part,
780 yet isn't the head, of a multibyte character. */
781 if (!sb
&& !re_string_first_byte (&mctx
.input
, 0))
785 /* It seems to be appropriate one, then use the matcher. */
786 /* We assume that the matching starts from 0. */
787 mctx
.state_log_top
= mctx
.nbkref_ents
= mctx
.max_mb_elem_len
= 0;
788 match_last
= check_matching (&mctx
, fl_longest_match
,
789 range
>= 0 ? &match_first
: NULL
);
790 if (match_last
!= -1)
792 if (BE (match_last
== -2, 0))
799 mctx
.match_last
= match_last
;
800 if ((!preg
->no_sub
&& nmatch
> 1) || dfa
->nbackref
)
802 re_dfastate_t
*pstate
= mctx
.state_log
[match_last
];
803 mctx
.last_node
= check_halt_state_context (&mctx
, pstate
,
806 if ((!preg
->no_sub
&& nmatch
> 1 && dfa
->has_plural_match
)
809 err
= prune_impossible_nodes (&mctx
);
810 if (err
== REG_NOERROR
)
812 if (BE (err
!= REG_NOMATCH
, 0))
817 break; /* We found a match. */
821 match_ctx_clean (&mctx
);
825 assert (match_last
!= -1);
826 assert (err
== REG_NOERROR
);
829 /* Set pmatch[] if we need. */
834 /* Initialize registers. */
835 for (reg_idx
= 1; reg_idx
< nmatch
; ++reg_idx
)
836 pmatch
[reg_idx
].rm_so
= pmatch
[reg_idx
].rm_eo
= -1;
838 /* Set the points where matching start/end. */
840 pmatch
[0].rm_eo
= mctx
.match_last
;
842 if (!preg
->no_sub
&& nmatch
> 1)
844 err
= set_regs (preg
, &mctx
, nmatch
, pmatch
,
845 dfa
->has_plural_match
&& dfa
->nbackref
> 0);
846 if (BE (err
!= REG_NOERROR
, 0))
850 /* At last, add the offset to the each registers, since we slided
851 the buffers so that we could assume that the matching starts
853 for (reg_idx
= 0; reg_idx
< nmatch
; ++reg_idx
)
854 if (pmatch
[reg_idx
].rm_so
!= -1)
856 #ifdef RE_ENABLE_I18N
857 if (BE (mctx
.input
.offsets_needed
!= 0, 0))
859 pmatch
[reg_idx
].rm_so
=
860 (pmatch
[reg_idx
].rm_so
== mctx
.input
.valid_len
861 ? mctx
.input
.valid_raw_len
862 : mctx
.input
.offsets
[pmatch
[reg_idx
].rm_so
]);
863 pmatch
[reg_idx
].rm_eo
=
864 (pmatch
[reg_idx
].rm_eo
== mctx
.input
.valid_len
865 ? mctx
.input
.valid_raw_len
866 : mctx
.input
.offsets
[pmatch
[reg_idx
].rm_eo
]);
869 assert (mctx
.input
.offsets_needed
== 0);
871 pmatch
[reg_idx
].rm_so
+= match_first
;
872 pmatch
[reg_idx
].rm_eo
+= match_first
;
874 for (reg_idx
= 0; reg_idx
< extra_nmatch
; ++reg_idx
)
876 pmatch
[nmatch
+ reg_idx
].rm_so
= -1;
877 pmatch
[nmatch
+ reg_idx
].rm_eo
= -1;
881 for (reg_idx
= 0; reg_idx
+ 1 < nmatch
; reg_idx
++)
882 if (dfa
->subexp_map
[reg_idx
] != reg_idx
)
884 pmatch
[reg_idx
+ 1].rm_so
885 = pmatch
[dfa
->subexp_map
[reg_idx
] + 1].rm_so
;
886 pmatch
[reg_idx
+ 1].rm_eo
887 = pmatch
[dfa
->subexp_map
[reg_idx
] + 1].rm_eo
;
892 re_free (mctx
.state_log
);
894 match_ctx_free (&mctx
);
895 re_string_destruct (&mctx
.input
);
900 __attribute_warn_unused_result__
901 prune_impossible_nodes (re_match_context_t
*mctx
)
903 const re_dfa_t
*const dfa
= mctx
->dfa
;
904 int halt_node
, match_last
;
906 re_dfastate_t
**sifted_states
;
907 re_dfastate_t
**lim_states
= NULL
;
908 re_sift_context_t sctx
;
910 assert (mctx
->state_log
!= NULL
);
912 match_last
= mctx
->match_last
;
913 halt_node
= mctx
->last_node
;
915 /* Avoid overflow. */
916 if (BE (SIZE_MAX
/ sizeof (re_dfastate_t
*) <= match_last
, 0))
919 sifted_states
= re_malloc (re_dfastate_t
*, match_last
+ 1);
920 if (BE (sifted_states
== NULL
, 0))
927 lim_states
= re_malloc (re_dfastate_t
*, match_last
+ 1);
928 if (BE (lim_states
== NULL
, 0))
935 memset (lim_states
, '\0',
936 sizeof (re_dfastate_t
*) * (match_last
+ 1));
937 sift_ctx_init (&sctx
, sifted_states
, lim_states
, halt_node
,
939 ret
= sift_states_backward (mctx
, &sctx
);
940 re_node_set_free (&sctx
.limits
);
941 if (BE (ret
!= REG_NOERROR
, 0))
943 if (sifted_states
[0] != NULL
|| lim_states
[0] != NULL
)
953 } while (mctx
->state_log
[match_last
] == NULL
954 || !mctx
->state_log
[match_last
]->halt
);
955 halt_node
= check_halt_state_context (mctx
,
956 mctx
->state_log
[match_last
],
959 ret
= merge_state_array (dfa
, sifted_states
, lim_states
,
961 re_free (lim_states
);
963 if (BE (ret
!= REG_NOERROR
, 0))
968 sift_ctx_init (&sctx
, sifted_states
, lim_states
, halt_node
, match_last
);
969 ret
= sift_states_backward (mctx
, &sctx
);
970 re_node_set_free (&sctx
.limits
);
971 if (BE (ret
!= REG_NOERROR
, 0))
973 if (sifted_states
[0] == NULL
)
979 re_free (mctx
->state_log
);
980 mctx
->state_log
= sifted_states
;
981 sifted_states
= NULL
;
982 mctx
->last_node
= halt_node
;
983 mctx
->match_last
= match_last
;
986 re_free (sifted_states
);
987 re_free (lim_states
);
991 /* Acquire an initial state and return it.
992 We must select appropriate initial state depending on the context,
993 since initial states may have constraints like "\<", "^", etc.. */
995 static inline re_dfastate_t
*
996 __attribute ((always_inline
))
997 acquire_init_state_context (reg_errcode_t
*err
, const re_match_context_t
*mctx
,
1000 const re_dfa_t
*const dfa
= mctx
->dfa
;
1001 if (dfa
->init_state
->has_constraint
)
1003 unsigned int context
;
1004 context
= re_string_context_at (&mctx
->input
, idx
- 1, mctx
->eflags
);
1005 if (IS_WORD_CONTEXT (context
))
1006 return dfa
->init_state_word
;
1007 else if (IS_ORDINARY_CONTEXT (context
))
1008 return dfa
->init_state
;
1009 else if (IS_BEGBUF_CONTEXT (context
) && IS_NEWLINE_CONTEXT (context
))
1010 return dfa
->init_state_begbuf
;
1011 else if (IS_NEWLINE_CONTEXT (context
))
1012 return dfa
->init_state_nl
;
1013 else if (IS_BEGBUF_CONTEXT (context
))
1015 /* It is relatively rare case, then calculate on demand. */
1016 return re_acquire_state_context (err
, dfa
,
1017 dfa
->init_state
->entrance_nodes
,
1021 /* Must not happen? */
1022 return dfa
->init_state
;
1025 return dfa
->init_state
;
1028 /* Check whether the regular expression match input string INPUT or not,
1029 and return the index where the matching end, return -1 if not match,
1030 or return -2 in case of an error.
1031 FL_LONGEST_MATCH means we want the POSIX longest matching.
1032 If P_MATCH_FIRST is not NULL, and the match fails, it is set to the
1033 next place where we may want to try matching.
1034 Note that the matcher assume that the maching starts from the current
1035 index of the buffer. */
1038 __attribute_warn_unused_result__
1039 check_matching (re_match_context_t
*mctx
, int fl_longest_match
,
1042 const re_dfa_t
*const dfa
= mctx
->dfa
;
1045 int match_last
= -1;
1046 int cur_str_idx
= re_string_cur_idx (&mctx
->input
);
1047 re_dfastate_t
*cur_state
;
1048 int at_init_state
= p_match_first
!= NULL
;
1049 int next_start_idx
= cur_str_idx
;
1052 cur_state
= acquire_init_state_context (&err
, mctx
, cur_str_idx
);
1053 /* An initial state must not be NULL (invalid). */
1054 if (BE (cur_state
== NULL
, 0))
1056 assert (err
== REG_ESPACE
);
1060 if (mctx
->state_log
!= NULL
)
1062 mctx
->state_log
[cur_str_idx
] = cur_state
;
1064 /* Check OP_OPEN_SUBEXP in the initial state in case that we use them
1065 later. E.g. Processing back references. */
1066 if (BE (dfa
->nbackref
, 0))
1069 err
= check_subexp_matching_top (mctx
, &cur_state
->nodes
, 0);
1070 if (BE (err
!= REG_NOERROR
, 0))
1073 if (cur_state
->has_backref
)
1075 err
= transit_state_bkref (mctx
, &cur_state
->nodes
);
1076 if (BE (err
!= REG_NOERROR
, 0))
1082 /* If the RE accepts NULL string. */
1083 if (BE (cur_state
->halt
, 0))
1085 if (!cur_state
->has_constraint
1086 || check_halt_state_context (mctx
, cur_state
, cur_str_idx
))
1088 if (!fl_longest_match
)
1092 match_last
= cur_str_idx
;
1098 while (!re_string_eoi (&mctx
->input
))
1100 re_dfastate_t
*old_state
= cur_state
;
1101 int next_char_idx
= re_string_cur_idx (&mctx
->input
) + 1;
1103 if ((BE (next_char_idx
>= mctx
->input
.bufs_len
, 0)
1104 && mctx
->input
.bufs_len
< mctx
->input
.len
)
1105 || (BE (next_char_idx
>= mctx
->input
.valid_len
, 0)
1106 && mctx
->input
.valid_len
< mctx
->input
.len
))
1108 err
= extend_buffers (mctx
, next_char_idx
+ 1);
1109 if (BE (err
!= REG_NOERROR
, 0))
1111 assert (err
== REG_ESPACE
);
1116 cur_state
= transit_state (&err
, mctx
, cur_state
);
1117 if (mctx
->state_log
!= NULL
)
1118 cur_state
= merge_state_with_log (&err
, mctx
, cur_state
);
1120 if (cur_state
== NULL
)
1122 /* Reached the invalid state or an error. Try to recover a valid
1123 state using the state log, if available and if we have not
1124 already found a valid (even if not the longest) match. */
1125 if (BE (err
!= REG_NOERROR
, 0))
1128 if (mctx
->state_log
== NULL
1129 || (match
&& !fl_longest_match
)
1130 || (cur_state
= find_recover_state (&err
, mctx
)) == NULL
)
1134 if (BE (at_init_state
, 0))
1136 if (old_state
== cur_state
)
1137 next_start_idx
= next_char_idx
;
1142 if (cur_state
->halt
)
1144 /* Reached a halt state.
1145 Check the halt state can satisfy the current context. */
1146 if (!cur_state
->has_constraint
1147 || check_halt_state_context (mctx
, cur_state
,
1148 re_string_cur_idx (&mctx
->input
)))
1150 /* We found an appropriate halt state. */
1151 match_last
= re_string_cur_idx (&mctx
->input
);
1154 /* We found a match, do not modify match_first below. */
1155 p_match_first
= NULL
;
1156 if (!fl_longest_match
)
1163 *p_match_first
+= next_start_idx
;
1168 /* Check NODE match the current context. */
1171 check_halt_node_context (const re_dfa_t
*dfa
, int node
, unsigned int context
)
1173 re_token_type_t type
= dfa
->nodes
[node
].type
;
1174 unsigned int constraint
= dfa
->nodes
[node
].constraint
;
1175 if (type
!= END_OF_RE
)
1179 if (NOT_SATISFY_NEXT_CONSTRAINT (constraint
, context
))
1184 /* Check the halt state STATE match the current context.
1185 Return 0 if not match, if the node, STATE has, is a halt node and
1186 match the context, return the node. */
1189 check_halt_state_context (const re_match_context_t
*mctx
,
1190 const re_dfastate_t
*state
, int idx
)
1193 unsigned int context
;
1195 assert (state
->halt
);
1197 context
= re_string_context_at (&mctx
->input
, idx
, mctx
->eflags
);
1198 for (i
= 0; i
< state
->nodes
.nelem
; ++i
)
1199 if (check_halt_node_context (mctx
->dfa
, state
->nodes
.elems
[i
], context
))
1200 return state
->nodes
.elems
[i
];
1204 /* Compute the next node to which "NFA" transit from NODE("NFA" is a NFA
1205 corresponding to the DFA).
1206 Return the destination node, and update EPS_VIA_NODES, return -1 in case
1210 proceed_next_node (const re_match_context_t
*mctx
, int nregs
, regmatch_t
*regs
,
1211 int *pidx
, int node
, re_node_set
*eps_via_nodes
,
1212 struct re_fail_stack_t
*fs
)
1214 const re_dfa_t
*const dfa
= mctx
->dfa
;
1216 if (IS_EPSILON_NODE (dfa
->nodes
[node
].type
))
1218 re_node_set
*cur_nodes
= &mctx
->state_log
[*pidx
]->nodes
;
1219 re_node_set
*edests
= &dfa
->edests
[node
];
1221 err
= re_node_set_insert (eps_via_nodes
, node
);
1222 if (BE (err
< 0, 0))
1224 /* Pick up a valid destination, or return -1 if none is found. */
1225 for (dest_node
= -1, i
= 0; i
< edests
->nelem
; ++i
)
1227 int candidate
= edests
->elems
[i
];
1228 if (!re_node_set_contains (cur_nodes
, candidate
))
1230 if (dest_node
== -1)
1231 dest_node
= candidate
;
1235 /* In order to avoid infinite loop like "(a*)*", return the second
1236 epsilon-transition if the first was already considered. */
1237 if (re_node_set_contains (eps_via_nodes
, dest_node
))
1240 /* Otherwise, push the second epsilon-transition on the fail stack. */
1242 && push_fail_stack (fs
, *pidx
, candidate
, nregs
, regs
,
1246 /* We know we are going to exit. */
1255 re_token_type_t type
= dfa
->nodes
[node
].type
;
1257 #ifdef RE_ENABLE_I18N
1258 if (dfa
->nodes
[node
].accept_mb
)
1259 naccepted
= check_node_accept_bytes (dfa
, node
, &mctx
->input
, *pidx
);
1261 #endif /* RE_ENABLE_I18N */
1262 if (type
== OP_BACK_REF
)
1264 int subexp_idx
= dfa
->nodes
[node
].opr
.idx
+ 1;
1265 naccepted
= regs
[subexp_idx
].rm_eo
- regs
[subexp_idx
].rm_so
;
1268 if (regs
[subexp_idx
].rm_so
== -1 || regs
[subexp_idx
].rm_eo
== -1)
1272 char *buf
= (char *) re_string_get_buffer (&mctx
->input
);
1273 if (memcmp (buf
+ regs
[subexp_idx
].rm_so
, buf
+ *pidx
,
1282 err
= re_node_set_insert (eps_via_nodes
, node
);
1283 if (BE (err
< 0, 0))
1285 dest_node
= dfa
->edests
[node
].elems
[0];
1286 if (re_node_set_contains (&mctx
->state_log
[*pidx
]->nodes
,
1293 || check_node_accept (mctx
, dfa
->nodes
+ node
, *pidx
))
1295 int dest_node
= dfa
->nexts
[node
];
1296 *pidx
= (naccepted
== 0) ? *pidx
+ 1 : *pidx
+ naccepted
;
1297 if (fs
&& (*pidx
> mctx
->match_last
|| mctx
->state_log
[*pidx
] == NULL
1298 || !re_node_set_contains (&mctx
->state_log
[*pidx
]->nodes
,
1301 re_node_set_empty (eps_via_nodes
);
1308 static reg_errcode_t
1309 __attribute_warn_unused_result__
1310 push_fail_stack (struct re_fail_stack_t
*fs
, int str_idx
, int dest_node
,
1311 int nregs
, regmatch_t
*regs
, re_node_set
*eps_via_nodes
)
1314 int num
= fs
->num
++;
1315 if (fs
->num
== fs
->alloc
)
1317 struct re_fail_stack_ent_t
*new_array
;
1318 new_array
= realloc (fs
->stack
, (sizeof (struct re_fail_stack_ent_t
)
1320 if (new_array
== NULL
)
1323 fs
->stack
= new_array
;
1325 fs
->stack
[num
].idx
= str_idx
;
1326 fs
->stack
[num
].node
= dest_node
;
1327 fs
->stack
[num
].regs
= re_malloc (regmatch_t
, nregs
);
1328 if (fs
->stack
[num
].regs
== NULL
)
1330 memcpy (fs
->stack
[num
].regs
, regs
, sizeof (regmatch_t
) * nregs
);
1331 err
= re_node_set_init_copy (&fs
->stack
[num
].eps_via_nodes
, eps_via_nodes
);
1336 pop_fail_stack (struct re_fail_stack_t
*fs
, int *pidx
, int nregs
,
1337 regmatch_t
*regs
, re_node_set
*eps_via_nodes
)
1339 int num
= --fs
->num
;
1341 *pidx
= fs
->stack
[num
].idx
;
1342 memcpy (regs
, fs
->stack
[num
].regs
, sizeof (regmatch_t
) * nregs
);
1343 re_node_set_free (eps_via_nodes
);
1344 re_free (fs
->stack
[num
].regs
);
1345 *eps_via_nodes
= fs
->stack
[num
].eps_via_nodes
;
1346 return fs
->stack
[num
].node
;
1349 /* Set the positions where the subexpressions are starts/ends to registers
1351 Note: We assume that pmatch[0] is already set, and
1352 pmatch[i].rm_so == pmatch[i].rm_eo == -1 for 0 < i < nmatch. */
1354 static reg_errcode_t
1355 __attribute_warn_unused_result__
1356 set_regs (const regex_t
*preg
, const re_match_context_t
*mctx
, size_t nmatch
,
1357 regmatch_t
*pmatch
, int fl_backtrack
)
1359 const re_dfa_t
*dfa
= (const re_dfa_t
*) preg
->buffer
;
1361 re_node_set eps_via_nodes
;
1362 struct re_fail_stack_t
*fs
;
1363 struct re_fail_stack_t fs_body
= { 0, 2, NULL
};
1364 regmatch_t
*prev_idx_match
;
1365 int prev_idx_match_malloced
= 0;
1368 assert (nmatch
> 1);
1369 assert (mctx
->state_log
!= NULL
);
1374 fs
->stack
= re_malloc (struct re_fail_stack_ent_t
, fs
->alloc
);
1375 if (fs
->stack
== NULL
)
1381 cur_node
= dfa
->init_node
;
1382 re_node_set_init_empty (&eps_via_nodes
);
1384 if (__libc_use_alloca (nmatch
* sizeof (regmatch_t
)))
1385 prev_idx_match
= (regmatch_t
*) alloca (nmatch
* sizeof (regmatch_t
));
1388 prev_idx_match
= re_malloc (regmatch_t
, nmatch
);
1389 if (prev_idx_match
== NULL
)
1391 free_fail_stack_return (fs
);
1394 prev_idx_match_malloced
= 1;
1396 memcpy (prev_idx_match
, pmatch
, sizeof (regmatch_t
) * nmatch
);
1398 for (idx
= pmatch
[0].rm_so
; idx
<= pmatch
[0].rm_eo
;)
1400 update_regs (dfa
, pmatch
, prev_idx_match
, cur_node
, idx
, nmatch
);
1402 if (idx
== pmatch
[0].rm_eo
&& cur_node
== mctx
->last_node
)
1407 for (reg_idx
= 0; reg_idx
< nmatch
; ++reg_idx
)
1408 if (pmatch
[reg_idx
].rm_so
> -1 && pmatch
[reg_idx
].rm_eo
== -1)
1410 if (reg_idx
== nmatch
)
1412 re_node_set_free (&eps_via_nodes
);
1413 if (prev_idx_match_malloced
)
1414 re_free (prev_idx_match
);
1415 return free_fail_stack_return (fs
);
1417 cur_node
= pop_fail_stack (fs
, &idx
, nmatch
, pmatch
,
1422 re_node_set_free (&eps_via_nodes
);
1423 if (prev_idx_match_malloced
)
1424 re_free (prev_idx_match
);
1429 /* Proceed to next node. */
1430 cur_node
= proceed_next_node (mctx
, nmatch
, pmatch
, &idx
, cur_node
,
1431 &eps_via_nodes
, fs
);
1433 if (BE (cur_node
< 0, 0))
1435 if (BE (cur_node
== -2, 0))
1437 re_node_set_free (&eps_via_nodes
);
1438 if (prev_idx_match_malloced
)
1439 re_free (prev_idx_match
);
1440 free_fail_stack_return (fs
);
1444 cur_node
= pop_fail_stack (fs
, &idx
, nmatch
, pmatch
,
1448 re_node_set_free (&eps_via_nodes
);
1449 if (prev_idx_match_malloced
)
1450 re_free (prev_idx_match
);
1455 re_node_set_free (&eps_via_nodes
);
1456 if (prev_idx_match_malloced
)
1457 re_free (prev_idx_match
);
1458 return free_fail_stack_return (fs
);
1461 static reg_errcode_t
1462 free_fail_stack_return (struct re_fail_stack_t
*fs
)
1467 for (fs_idx
= 0; fs_idx
< fs
->num
; ++fs_idx
)
1469 re_node_set_free (&fs
->stack
[fs_idx
].eps_via_nodes
);
1470 re_free (fs
->stack
[fs_idx
].regs
);
1472 re_free (fs
->stack
);
1478 update_regs (const re_dfa_t
*dfa
, regmatch_t
*pmatch
,
1479 regmatch_t
*prev_idx_match
, int cur_node
, int cur_idx
, int nmatch
)
1481 int type
= dfa
->nodes
[cur_node
].type
;
1482 if (type
== OP_OPEN_SUBEXP
)
1484 int reg_num
= dfa
->nodes
[cur_node
].opr
.idx
+ 1;
1486 /* We are at the first node of this sub expression. */
1487 if (reg_num
< nmatch
)
1489 pmatch
[reg_num
].rm_so
= cur_idx
;
1490 pmatch
[reg_num
].rm_eo
= -1;
1493 else if (type
== OP_CLOSE_SUBEXP
)
1495 int reg_num
= dfa
->nodes
[cur_node
].opr
.idx
+ 1;
1496 if (reg_num
< nmatch
)
1498 /* We are at the last node of this sub expression. */
1499 if (pmatch
[reg_num
].rm_so
< cur_idx
)
1501 pmatch
[reg_num
].rm_eo
= cur_idx
;
1502 /* This is a non-empty match or we are not inside an optional
1503 subexpression. Accept this right away. */
1504 memcpy (prev_idx_match
, pmatch
, sizeof (regmatch_t
) * nmatch
);
1508 if (dfa
->nodes
[cur_node
].opt_subexp
1509 && prev_idx_match
[reg_num
].rm_so
!= -1)
1510 /* We transited through an empty match for an optional
1511 subexpression, like (a?)*, and this is not the subexp's
1512 first match. Copy back the old content of the registers
1513 so that matches of an inner subexpression are undone as
1514 well, like in ((a?))*. */
1515 memcpy (pmatch
, prev_idx_match
, sizeof (regmatch_t
) * nmatch
);
1517 /* We completed a subexpression, but it may be part of
1518 an optional one, so do not update PREV_IDX_MATCH. */
1519 pmatch
[reg_num
].rm_eo
= cur_idx
;
1525 /* This function checks the STATE_LOG from the SCTX->last_str_idx to 0
1526 and sift the nodes in each states according to the following rules.
1527 Updated state_log will be wrote to STATE_LOG.
1529 Rules: We throw away the Node `a' in the STATE_LOG[STR_IDX] if...
1530 1. When STR_IDX == MATCH_LAST(the last index in the state_log):
1531 If `a' isn't the LAST_NODE and `a' can't epsilon transit to
1532 the LAST_NODE, we throw away the node `a'.
1533 2. When 0 <= STR_IDX < MATCH_LAST and `a' accepts
1534 string `s' and transit to `b':
1535 i. If 'b' isn't in the STATE_LOG[STR_IDX+strlen('s')], we throw
1537 ii. If 'b' is in the STATE_LOG[STR_IDX+strlen('s')] but 'b' is
1538 thrown away, we throw away the node `a'.
1539 3. When 0 <= STR_IDX < MATCH_LAST and 'a' epsilon transit to 'b':
1540 i. If 'b' isn't in the STATE_LOG[STR_IDX], we throw away the
1542 ii. If 'b' is in the STATE_LOG[STR_IDX] but 'b' is thrown away,
1543 we throw away the node `a'. */
1545 #define STATE_NODE_CONTAINS(state,node) \
1546 ((state) != NULL && re_node_set_contains (&(state)->nodes, node))
1548 static reg_errcode_t
1549 sift_states_backward (const re_match_context_t
*mctx
, re_sift_context_t
*sctx
)
1553 int str_idx
= sctx
->last_str_idx
;
1554 re_node_set cur_dest
;
1557 assert (mctx
->state_log
!= NULL
&& mctx
->state_log
[str_idx
] != NULL
);
1560 /* Build sifted state_log[str_idx]. It has the nodes which can epsilon
1561 transit to the last_node and the last_node itself. */
1562 err
= re_node_set_init_1 (&cur_dest
, sctx
->last_node
);
1563 if (BE (err
!= REG_NOERROR
, 0))
1565 err
= update_cur_sifted_state (mctx
, sctx
, str_idx
, &cur_dest
);
1566 if (BE (err
!= REG_NOERROR
, 0))
1569 /* Then check each states in the state_log. */
1572 /* Update counters. */
1573 null_cnt
= (sctx
->sifted_states
[str_idx
] == NULL
) ? null_cnt
+ 1 : 0;
1574 if (null_cnt
> mctx
->max_mb_elem_len
)
1576 memset (sctx
->sifted_states
, '\0',
1577 sizeof (re_dfastate_t
*) * str_idx
);
1578 re_node_set_free (&cur_dest
);
1581 re_node_set_empty (&cur_dest
);
1584 if (mctx
->state_log
[str_idx
])
1586 err
= build_sifted_states (mctx
, sctx
, str_idx
, &cur_dest
);
1587 if (BE (err
!= REG_NOERROR
, 0))
1591 /* Add all the nodes which satisfy the following conditions:
1592 - It can epsilon transit to a node in CUR_DEST.
1594 And update state_log. */
1595 err
= update_cur_sifted_state (mctx
, sctx
, str_idx
, &cur_dest
);
1596 if (BE (err
!= REG_NOERROR
, 0))
1601 re_node_set_free (&cur_dest
);
1605 static reg_errcode_t
1606 __attribute_warn_unused_result__
1607 build_sifted_states (const re_match_context_t
*mctx
, re_sift_context_t
*sctx
,
1608 int str_idx
, re_node_set
*cur_dest
)
1610 const re_dfa_t
*const dfa
= mctx
->dfa
;
1611 const re_node_set
*cur_src
= &mctx
->state_log
[str_idx
]->non_eps_nodes
;
1614 /* Then build the next sifted state.
1615 We build the next sifted state on `cur_dest', and update
1616 `sifted_states[str_idx]' with `cur_dest'.
1618 `cur_dest' is the sifted state from `state_log[str_idx + 1]'.
1619 `cur_src' points the node_set of the old `state_log[str_idx]'
1620 (with the epsilon nodes pre-filtered out). */
1621 for (i
= 0; i
< cur_src
->nelem
; i
++)
1623 int prev_node
= cur_src
->elems
[i
];
1628 re_token_type_t type
= dfa
->nodes
[prev_node
].type
;
1629 assert (!IS_EPSILON_NODE (type
));
1631 #ifdef RE_ENABLE_I18N
1632 /* If the node may accept `multi byte'. */
1633 if (dfa
->nodes
[prev_node
].accept_mb
)
1634 naccepted
= sift_states_iter_mb (mctx
, sctx
, prev_node
,
1635 str_idx
, sctx
->last_str_idx
);
1636 #endif /* RE_ENABLE_I18N */
1638 /* We don't check backreferences here.
1639 See update_cur_sifted_state(). */
1641 && check_node_accept (mctx
, dfa
->nodes
+ prev_node
, str_idx
)
1642 && STATE_NODE_CONTAINS (sctx
->sifted_states
[str_idx
+ 1],
1643 dfa
->nexts
[prev_node
]))
1649 if (sctx
->limits
.nelem
)
1651 int to_idx
= str_idx
+ naccepted
;
1652 if (check_dst_limits (mctx
, &sctx
->limits
,
1653 dfa
->nexts
[prev_node
], to_idx
,
1654 prev_node
, str_idx
))
1657 ret
= re_node_set_insert (cur_dest
, prev_node
);
1658 if (BE (ret
== -1, 0))
1665 /* Helper functions. */
1667 static reg_errcode_t
1668 clean_state_log_if_needed (re_match_context_t
*mctx
, int next_state_log_idx
)
1670 int top
= mctx
->state_log_top
;
1672 if ((next_state_log_idx
>= mctx
->input
.bufs_len
1673 && mctx
->input
.bufs_len
< mctx
->input
.len
)
1674 || (next_state_log_idx
>= mctx
->input
.valid_len
1675 && mctx
->input
.valid_len
< mctx
->input
.len
))
1678 err
= extend_buffers (mctx
, next_state_log_idx
+ 1);
1679 if (BE (err
!= REG_NOERROR
, 0))
1683 if (top
< next_state_log_idx
)
1685 memset (mctx
->state_log
+ top
+ 1, '\0',
1686 sizeof (re_dfastate_t
*) * (next_state_log_idx
- top
));
1687 mctx
->state_log_top
= next_state_log_idx
;
1692 static reg_errcode_t
1693 merge_state_array (const re_dfa_t
*dfa
, re_dfastate_t
**dst
,
1694 re_dfastate_t
**src
, int num
)
1698 for (st_idx
= 0; st_idx
< num
; ++st_idx
)
1700 if (dst
[st_idx
] == NULL
)
1701 dst
[st_idx
] = src
[st_idx
];
1702 else if (src
[st_idx
] != NULL
)
1704 re_node_set merged_set
;
1705 err
= re_node_set_init_union (&merged_set
, &dst
[st_idx
]->nodes
,
1706 &src
[st_idx
]->nodes
);
1707 if (BE (err
!= REG_NOERROR
, 0))
1709 dst
[st_idx
] = re_acquire_state (&err
, dfa
, &merged_set
);
1710 re_node_set_free (&merged_set
);
1711 if (BE (err
!= REG_NOERROR
, 0))
1718 static reg_errcode_t
1719 update_cur_sifted_state (const re_match_context_t
*mctx
,
1720 re_sift_context_t
*sctx
, int str_idx
,
1721 re_node_set
*dest_nodes
)
1723 const re_dfa_t
*const dfa
= mctx
->dfa
;
1724 reg_errcode_t err
= REG_NOERROR
;
1725 const re_node_set
*candidates
;
1726 candidates
= ((mctx
->state_log
[str_idx
] == NULL
) ? NULL
1727 : &mctx
->state_log
[str_idx
]->nodes
);
1729 if (dest_nodes
->nelem
== 0)
1730 sctx
->sifted_states
[str_idx
] = NULL
;
1735 /* At first, add the nodes which can epsilon transit to a node in
1737 err
= add_epsilon_src_nodes (dfa
, dest_nodes
, candidates
);
1738 if (BE (err
!= REG_NOERROR
, 0))
1741 /* Then, check the limitations in the current sift_context. */
1742 if (sctx
->limits
.nelem
)
1744 err
= check_subexp_limits (dfa
, dest_nodes
, candidates
, &sctx
->limits
,
1745 mctx
->bkref_ents
, str_idx
);
1746 if (BE (err
!= REG_NOERROR
, 0))
1751 sctx
->sifted_states
[str_idx
] = re_acquire_state (&err
, dfa
, dest_nodes
);
1752 if (BE (err
!= REG_NOERROR
, 0))
1756 if (candidates
&& mctx
->state_log
[str_idx
]->has_backref
)
1758 err
= sift_states_bkref (mctx
, sctx
, str_idx
, candidates
);
1759 if (BE (err
!= REG_NOERROR
, 0))
1765 static reg_errcode_t
1766 __attribute_warn_unused_result__
1767 add_epsilon_src_nodes (const re_dfa_t
*dfa
, re_node_set
*dest_nodes
,
1768 const re_node_set
*candidates
)
1770 reg_errcode_t err
= REG_NOERROR
;
1773 re_dfastate_t
*state
= re_acquire_state (&err
, dfa
, dest_nodes
);
1774 if (BE (err
!= REG_NOERROR
, 0))
1777 if (!state
->inveclosure
.alloc
)
1779 err
= re_node_set_alloc (&state
->inveclosure
, dest_nodes
->nelem
);
1780 if (BE (err
!= REG_NOERROR
, 0))
1782 for (i
= 0; i
< dest_nodes
->nelem
; i
++)
1784 err
= re_node_set_merge (&state
->inveclosure
,
1785 dfa
->inveclosures
+ dest_nodes
->elems
[i
]);
1786 if (BE (err
!= REG_NOERROR
, 0))
1790 return re_node_set_add_intersect (dest_nodes
, candidates
,
1791 &state
->inveclosure
);
1794 static reg_errcode_t
1795 sub_epsilon_src_nodes (const re_dfa_t
*dfa
, int node
, re_node_set
*dest_nodes
,
1796 const re_node_set
*candidates
)
1800 re_node_set
*inv_eclosure
= dfa
->inveclosures
+ node
;
1801 re_node_set except_nodes
;
1802 re_node_set_init_empty (&except_nodes
);
1803 for (ecl_idx
= 0; ecl_idx
< inv_eclosure
->nelem
; ++ecl_idx
)
1805 int cur_node
= inv_eclosure
->elems
[ecl_idx
];
1806 if (cur_node
== node
)
1808 if (IS_EPSILON_NODE (dfa
->nodes
[cur_node
].type
))
1810 int edst1
= dfa
->edests
[cur_node
].elems
[0];
1811 int edst2
= ((dfa
->edests
[cur_node
].nelem
> 1)
1812 ? dfa
->edests
[cur_node
].elems
[1] : -1);
1813 if ((!re_node_set_contains (inv_eclosure
, edst1
)
1814 && re_node_set_contains (dest_nodes
, edst1
))
1816 && !re_node_set_contains (inv_eclosure
, edst2
)
1817 && re_node_set_contains (dest_nodes
, edst2
)))
1819 err
= re_node_set_add_intersect (&except_nodes
, candidates
,
1820 dfa
->inveclosures
+ cur_node
);
1821 if (BE (err
!= REG_NOERROR
, 0))
1823 re_node_set_free (&except_nodes
);
1829 for (ecl_idx
= 0; ecl_idx
< inv_eclosure
->nelem
; ++ecl_idx
)
1831 int cur_node
= inv_eclosure
->elems
[ecl_idx
];
1832 if (!re_node_set_contains (&except_nodes
, cur_node
))
1834 int idx
= re_node_set_contains (dest_nodes
, cur_node
) - 1;
1835 re_node_set_remove_at (dest_nodes
, idx
);
1838 re_node_set_free (&except_nodes
);
1843 check_dst_limits (const re_match_context_t
*mctx
, re_node_set
*limits
,
1844 int dst_node
, int dst_idx
, int src_node
, int src_idx
)
1846 const re_dfa_t
*const dfa
= mctx
->dfa
;
1847 int lim_idx
, src_pos
, dst_pos
;
1849 int dst_bkref_idx
= search_cur_bkref_entry (mctx
, dst_idx
);
1850 int src_bkref_idx
= search_cur_bkref_entry (mctx
, src_idx
);
1851 for (lim_idx
= 0; lim_idx
< limits
->nelem
; ++lim_idx
)
1854 struct re_backref_cache_entry
*ent
;
1855 ent
= mctx
->bkref_ents
+ limits
->elems
[lim_idx
];
1856 subexp_idx
= dfa
->nodes
[ent
->node
].opr
.idx
;
1858 dst_pos
= check_dst_limits_calc_pos (mctx
, limits
->elems
[lim_idx
],
1859 subexp_idx
, dst_node
, dst_idx
,
1861 src_pos
= check_dst_limits_calc_pos (mctx
, limits
->elems
[lim_idx
],
1862 subexp_idx
, src_node
, src_idx
,
1866 <src> <dst> ( <subexp> )
1867 ( <subexp> ) <src> <dst>
1868 ( <subexp1> <src> <subexp2> <dst> <subexp3> ) */
1869 if (src_pos
== dst_pos
)
1870 continue; /* This is unrelated limitation. */
1878 check_dst_limits_calc_pos_1 (const re_match_context_t
*mctx
, int boundaries
,
1879 int subexp_idx
, int from_node
, int bkref_idx
)
1881 const re_dfa_t
*const dfa
= mctx
->dfa
;
1882 const re_node_set
*eclosures
= dfa
->eclosures
+ from_node
;
1885 /* Else, we are on the boundary: examine the nodes on the epsilon
1887 for (node_idx
= 0; node_idx
< eclosures
->nelem
; ++node_idx
)
1889 int node
= eclosures
->elems
[node_idx
];
1890 switch (dfa
->nodes
[node
].type
)
1893 if (bkref_idx
!= -1)
1895 struct re_backref_cache_entry
*ent
= mctx
->bkref_ents
+ bkref_idx
;
1900 if (ent
->node
!= node
)
1903 if (subexp_idx
< BITSET_WORD_BITS
1904 && !(ent
->eps_reachable_subexps_map
1905 & ((bitset_word_t
) 1 << subexp_idx
)))
1908 /* Recurse trying to reach the OP_OPEN_SUBEXP and
1909 OP_CLOSE_SUBEXP cases below. But, if the
1910 destination node is the same node as the source
1911 node, don't recurse because it would cause an
1912 infinite loop: a regex that exhibits this behavior
1914 dst
= dfa
->edests
[node
].elems
[0];
1915 if (dst
== from_node
)
1919 else /* if (boundaries & 2) */
1924 check_dst_limits_calc_pos_1 (mctx
, boundaries
, subexp_idx
,
1926 if (cpos
== -1 /* && (boundaries & 1) */)
1928 if (cpos
== 0 && (boundaries
& 2))
1931 if (subexp_idx
< BITSET_WORD_BITS
)
1932 ent
->eps_reachable_subexps_map
1933 &= ~((bitset_word_t
) 1 << subexp_idx
);
1935 while (ent
++->more
);
1939 case OP_OPEN_SUBEXP
:
1940 if ((boundaries
& 1) && subexp_idx
== dfa
->nodes
[node
].opr
.idx
)
1944 case OP_CLOSE_SUBEXP
:
1945 if ((boundaries
& 2) && subexp_idx
== dfa
->nodes
[node
].opr
.idx
)
1954 return (boundaries
& 2) ? 1 : 0;
1958 check_dst_limits_calc_pos (const re_match_context_t
*mctx
, int limit
,
1959 int subexp_idx
, int from_node
, int str_idx
,
1962 struct re_backref_cache_entry
*lim
= mctx
->bkref_ents
+ limit
;
1965 /* If we are outside the range of the subexpression, return -1 or 1. */
1966 if (str_idx
< lim
->subexp_from
)
1969 if (lim
->subexp_to
< str_idx
)
1972 /* If we are within the subexpression, return 0. */
1973 boundaries
= (str_idx
== lim
->subexp_from
);
1974 boundaries
|= (str_idx
== lim
->subexp_to
) << 1;
1975 if (boundaries
== 0)
1978 /* Else, examine epsilon closure. */
1979 return check_dst_limits_calc_pos_1 (mctx
, boundaries
, subexp_idx
,
1980 from_node
, bkref_idx
);
1983 /* Check the limitations of sub expressions LIMITS, and remove the nodes
1984 which are against limitations from DEST_NODES. */
1986 static reg_errcode_t
1987 check_subexp_limits (const re_dfa_t
*dfa
, re_node_set
*dest_nodes
,
1988 const re_node_set
*candidates
, re_node_set
*limits
,
1989 struct re_backref_cache_entry
*bkref_ents
, int str_idx
)
1992 int node_idx
, lim_idx
;
1994 for (lim_idx
= 0; lim_idx
< limits
->nelem
; ++lim_idx
)
1997 struct re_backref_cache_entry
*ent
;
1998 ent
= bkref_ents
+ limits
->elems
[lim_idx
];
2000 if (str_idx
<= ent
->subexp_from
|| ent
->str_idx
< str_idx
)
2001 continue; /* This is unrelated limitation. */
2003 subexp_idx
= dfa
->nodes
[ent
->node
].opr
.idx
;
2004 if (ent
->subexp_to
== str_idx
)
2008 for (node_idx
= 0; node_idx
< dest_nodes
->nelem
; ++node_idx
)
2010 int node
= dest_nodes
->elems
[node_idx
];
2011 re_token_type_t type
= dfa
->nodes
[node
].type
;
2012 if (type
== OP_OPEN_SUBEXP
2013 && subexp_idx
== dfa
->nodes
[node
].opr
.idx
)
2015 else if (type
== OP_CLOSE_SUBEXP
2016 && subexp_idx
== dfa
->nodes
[node
].opr
.idx
)
2020 /* Check the limitation of the open subexpression. */
2021 /* Note that (ent->subexp_to = str_idx != ent->subexp_from). */
2024 err
= sub_epsilon_src_nodes (dfa
, ops_node
, dest_nodes
,
2026 if (BE (err
!= REG_NOERROR
, 0))
2030 /* Check the limitation of the close subexpression. */
2032 for (node_idx
= 0; node_idx
< dest_nodes
->nelem
; ++node_idx
)
2034 int node
= dest_nodes
->elems
[node_idx
];
2035 if (!re_node_set_contains (dfa
->inveclosures
+ node
,
2037 && !re_node_set_contains (dfa
->eclosures
+ node
,
2040 /* It is against this limitation.
2041 Remove it form the current sifted state. */
2042 err
= sub_epsilon_src_nodes (dfa
, node
, dest_nodes
,
2044 if (BE (err
!= REG_NOERROR
, 0))
2050 else /* (ent->subexp_to != str_idx) */
2052 for (node_idx
= 0; node_idx
< dest_nodes
->nelem
; ++node_idx
)
2054 int node
= dest_nodes
->elems
[node_idx
];
2055 re_token_type_t type
= dfa
->nodes
[node
].type
;
2056 if (type
== OP_CLOSE_SUBEXP
|| type
== OP_OPEN_SUBEXP
)
2058 if (subexp_idx
!= dfa
->nodes
[node
].opr
.idx
)
2060 /* It is against this limitation.
2061 Remove it form the current sifted state. */
2062 err
= sub_epsilon_src_nodes (dfa
, node
, dest_nodes
,
2064 if (BE (err
!= REG_NOERROR
, 0))
2073 static reg_errcode_t
2074 __attribute_warn_unused_result__
2075 sift_states_bkref (const re_match_context_t
*mctx
, re_sift_context_t
*sctx
,
2076 int str_idx
, const re_node_set
*candidates
)
2078 const re_dfa_t
*const dfa
= mctx
->dfa
;
2081 re_sift_context_t local_sctx
;
2082 int first_idx
= search_cur_bkref_entry (mctx
, str_idx
);
2084 if (first_idx
== -1)
2087 local_sctx
.sifted_states
= NULL
; /* Mark that it hasn't been initialized. */
2089 for (node_idx
= 0; node_idx
< candidates
->nelem
; ++node_idx
)
2092 re_token_type_t type
;
2093 struct re_backref_cache_entry
*entry
;
2094 node
= candidates
->elems
[node_idx
];
2095 type
= dfa
->nodes
[node
].type
;
2096 /* Avoid infinite loop for the REs like "()\1+". */
2097 if (node
== sctx
->last_node
&& str_idx
== sctx
->last_str_idx
)
2099 if (type
!= OP_BACK_REF
)
2102 entry
= mctx
->bkref_ents
+ first_idx
;
2103 enabled_idx
= first_idx
;
2110 re_dfastate_t
*cur_state
;
2112 if (entry
->node
!= node
)
2114 subexp_len
= entry
->subexp_to
- entry
->subexp_from
;
2115 to_idx
= str_idx
+ subexp_len
;
2116 dst_node
= (subexp_len
? dfa
->nexts
[node
]
2117 : dfa
->edests
[node
].elems
[0]);
2119 if (to_idx
> sctx
->last_str_idx
2120 || sctx
->sifted_states
[to_idx
] == NULL
2121 || !STATE_NODE_CONTAINS (sctx
->sifted_states
[to_idx
], dst_node
)
2122 || check_dst_limits (mctx
, &sctx
->limits
, node
,
2123 str_idx
, dst_node
, to_idx
))
2126 if (local_sctx
.sifted_states
== NULL
)
2129 err
= re_node_set_init_copy (&local_sctx
.limits
, &sctx
->limits
);
2130 if (BE (err
!= REG_NOERROR
, 0))
2133 local_sctx
.last_node
= node
;
2134 local_sctx
.last_str_idx
= str_idx
;
2135 ret
= re_node_set_insert (&local_sctx
.limits
, enabled_idx
);
2136 if (BE (ret
< 0, 0))
2141 cur_state
= local_sctx
.sifted_states
[str_idx
];
2142 err
= sift_states_backward (mctx
, &local_sctx
);
2143 if (BE (err
!= REG_NOERROR
, 0))
2145 if (sctx
->limited_states
!= NULL
)
2147 err
= merge_state_array (dfa
, sctx
->limited_states
,
2148 local_sctx
.sifted_states
,
2150 if (BE (err
!= REG_NOERROR
, 0))
2153 local_sctx
.sifted_states
[str_idx
] = cur_state
;
2154 re_node_set_remove (&local_sctx
.limits
, enabled_idx
);
2156 /* mctx->bkref_ents may have changed, reload the pointer. */
2157 entry
= mctx
->bkref_ents
+ enabled_idx
;
2159 while (enabled_idx
++, entry
++->more
);
2163 if (local_sctx
.sifted_states
!= NULL
)
2165 re_node_set_free (&local_sctx
.limits
);
2172 #ifdef RE_ENABLE_I18N
2174 sift_states_iter_mb (const re_match_context_t
*mctx
, re_sift_context_t
*sctx
,
2175 int node_idx
, int str_idx
, int max_str_idx
)
2177 const re_dfa_t
*const dfa
= mctx
->dfa
;
2179 /* Check the node can accept `multi byte'. */
2180 naccepted
= check_node_accept_bytes (dfa
, node_idx
, &mctx
->input
, str_idx
);
2181 if (naccepted
> 0 && str_idx
+ naccepted
<= max_str_idx
&&
2182 !STATE_NODE_CONTAINS (sctx
->sifted_states
[str_idx
+ naccepted
],
2183 dfa
->nexts
[node_idx
]))
2184 /* The node can't accept the `multi byte', or the
2185 destination was already thrown away, then the node
2186 could't accept the current input `multi byte'. */
2188 /* Otherwise, it is sure that the node could accept
2189 `naccepted' bytes input. */
2192 #endif /* RE_ENABLE_I18N */
2195 /* Functions for state transition. */
2197 /* Return the next state to which the current state STATE will transit by
2198 accepting the current input byte, and update STATE_LOG if necessary.
2199 If STATE can accept a multibyte char/collating element/back reference
2200 update the destination of STATE_LOG. */
2202 static re_dfastate_t
*
2203 __attribute_warn_unused_result__
2204 transit_state (reg_errcode_t
*err
, re_match_context_t
*mctx
,
2205 re_dfastate_t
*state
)
2207 re_dfastate_t
**trtable
;
2210 #ifdef RE_ENABLE_I18N
2211 /* If the current state can accept multibyte. */
2212 if (BE (state
->accept_mb
, 0))
2214 *err
= transit_state_mb (mctx
, state
);
2215 if (BE (*err
!= REG_NOERROR
, 0))
2218 #endif /* RE_ENABLE_I18N */
2220 /* Then decide the next state with the single byte. */
2223 /* don't use transition table */
2224 return transit_state_sb (err
, mctx
, state
);
2227 /* Use transition table */
2228 ch
= re_string_fetch_byte (&mctx
->input
);
2231 trtable
= state
->trtable
;
2232 if (BE (trtable
!= NULL
, 1))
2235 trtable
= state
->word_trtable
;
2236 if (BE (trtable
!= NULL
, 1))
2238 unsigned int context
;
2240 = re_string_context_at (&mctx
->input
,
2241 re_string_cur_idx (&mctx
->input
) - 1,
2243 if (IS_WORD_CONTEXT (context
))
2244 return trtable
[ch
+ SBC_MAX
];
2249 if (!build_trtable (mctx
->dfa
, state
))
2255 /* Retry, we now have a transition table. */
2259 /* Update the state_log if we need */
2261 merge_state_with_log (reg_errcode_t
*err
, re_match_context_t
*mctx
,
2262 re_dfastate_t
*next_state
)
2264 const re_dfa_t
*const dfa
= mctx
->dfa
;
2265 int cur_idx
= re_string_cur_idx (&mctx
->input
);
2267 if (cur_idx
> mctx
->state_log_top
)
2269 mctx
->state_log
[cur_idx
] = next_state
;
2270 mctx
->state_log_top
= cur_idx
;
2272 else if (mctx
->state_log
[cur_idx
] == 0)
2274 mctx
->state_log
[cur_idx
] = next_state
;
2278 re_dfastate_t
*pstate
;
2279 unsigned int context
;
2280 re_node_set next_nodes
, *log_nodes
, *table_nodes
= NULL
;
2281 /* If (state_log[cur_idx] != 0), it implies that cur_idx is
2282 the destination of a multibyte char/collating element/
2283 back reference. Then the next state is the union set of
2284 these destinations and the results of the transition table. */
2285 pstate
= mctx
->state_log
[cur_idx
];
2286 log_nodes
= pstate
->entrance_nodes
;
2287 if (next_state
!= NULL
)
2289 table_nodes
= next_state
->entrance_nodes
;
2290 *err
= re_node_set_init_union (&next_nodes
, table_nodes
,
2292 if (BE (*err
!= REG_NOERROR
, 0))
2296 next_nodes
= *log_nodes
;
2297 /* Note: We already add the nodes of the initial state,
2298 then we don't need to add them here. */
2300 context
= re_string_context_at (&mctx
->input
,
2301 re_string_cur_idx (&mctx
->input
) - 1,
2303 next_state
= mctx
->state_log
[cur_idx
]
2304 = re_acquire_state_context (err
, dfa
, &next_nodes
, context
);
2305 /* We don't need to check errors here, since the return value of
2306 this function is next_state and ERR is already set. */
2308 if (table_nodes
!= NULL
)
2309 re_node_set_free (&next_nodes
);
2312 if (BE (dfa
->nbackref
, 0) && next_state
!= NULL
)
2314 /* Check OP_OPEN_SUBEXP in the current state in case that we use them
2315 later. We must check them here, since the back references in the
2316 next state might use them. */
2317 *err
= check_subexp_matching_top (mctx
, &next_state
->nodes
,
2319 if (BE (*err
!= REG_NOERROR
, 0))
2322 /* If the next state has back references. */
2323 if (next_state
->has_backref
)
2325 *err
= transit_state_bkref (mctx
, &next_state
->nodes
);
2326 if (BE (*err
!= REG_NOERROR
, 0))
2328 next_state
= mctx
->state_log
[cur_idx
];
2335 /* Skip bytes in the input that correspond to part of a
2336 multi-byte match, then look in the log for a state
2337 from which to restart matching. */
2339 find_recover_state (reg_errcode_t
*err
, re_match_context_t
*mctx
)
2341 re_dfastate_t
*cur_state
;
2344 int max
= mctx
->state_log_top
;
2345 int cur_str_idx
= re_string_cur_idx (&mctx
->input
);
2349 if (++cur_str_idx
> max
)
2351 re_string_skip_bytes (&mctx
->input
, 1);
2353 while (mctx
->state_log
[cur_str_idx
] == NULL
);
2355 cur_state
= merge_state_with_log (err
, mctx
, NULL
);
2357 while (*err
== REG_NOERROR
&& cur_state
== NULL
);
2361 /* Helper functions for transit_state. */
2363 /* From the node set CUR_NODES, pick up the nodes whose types are
2364 OP_OPEN_SUBEXP and which have corresponding back references in the regular
2365 expression. And register them to use them later for evaluating the
2366 correspoding back references. */
2368 static reg_errcode_t
2369 check_subexp_matching_top (re_match_context_t
*mctx
, re_node_set
*cur_nodes
,
2372 const re_dfa_t
*const dfa
= mctx
->dfa
;
2376 /* TODO: This isn't efficient.
2377 Because there might be more than one nodes whose types are
2378 OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all
2381 for (node_idx
= 0; node_idx
< cur_nodes
->nelem
; ++node_idx
)
2383 int node
= cur_nodes
->elems
[node_idx
];
2384 if (dfa
->nodes
[node
].type
== OP_OPEN_SUBEXP
2385 && dfa
->nodes
[node
].opr
.idx
< BITSET_WORD_BITS
2386 && (dfa
->used_bkref_map
2387 & ((bitset_word_t
) 1 << dfa
->nodes
[node
].opr
.idx
)))
2389 err
= match_ctx_add_subtop (mctx
, node
, str_idx
);
2390 if (BE (err
!= REG_NOERROR
, 0))
2398 /* Return the next state to which the current state STATE will transit by
2399 accepting the current input byte. */
2401 static re_dfastate_t
*
2402 transit_state_sb (reg_errcode_t
*err
, re_match_context_t
*mctx
,
2403 re_dfastate_t
*state
)
2405 const re_dfa_t
*const dfa
= mctx
->dfa
;
2406 re_node_set next_nodes
;
2407 re_dfastate_t
*next_state
;
2408 int node_cnt
, cur_str_idx
= re_string_cur_idx (&mctx
->input
);
2409 unsigned int context
;
2411 *err
= re_node_set_alloc (&next_nodes
, state
->nodes
.nelem
+ 1);
2412 if (BE (*err
!= REG_NOERROR
, 0))
2414 for (node_cnt
= 0; node_cnt
< state
->nodes
.nelem
; ++node_cnt
)
2416 int cur_node
= state
->nodes
.elems
[node_cnt
];
2417 if (check_node_accept (mctx
, dfa
->nodes
+ cur_node
, cur_str_idx
))
2419 *err
= re_node_set_merge (&next_nodes
,
2420 dfa
->eclosures
+ dfa
->nexts
[cur_node
]);
2421 if (BE (*err
!= REG_NOERROR
, 0))
2423 re_node_set_free (&next_nodes
);
2428 context
= re_string_context_at (&mctx
->input
, cur_str_idx
, mctx
->eflags
);
2429 next_state
= re_acquire_state_context (err
, dfa
, &next_nodes
, context
);
2430 /* We don't need to check errors here, since the return value of
2431 this function is next_state and ERR is already set. */
2433 re_node_set_free (&next_nodes
);
2434 re_string_skip_bytes (&mctx
->input
, 1);
2439 #ifdef RE_ENABLE_I18N
2440 static reg_errcode_t
2441 transit_state_mb (re_match_context_t
*mctx
, re_dfastate_t
*pstate
)
2443 const re_dfa_t
*const dfa
= mctx
->dfa
;
2447 for (i
= 0; i
< pstate
->nodes
.nelem
; ++i
)
2449 re_node_set dest_nodes
, *new_nodes
;
2450 int cur_node_idx
= pstate
->nodes
.elems
[i
];
2451 int naccepted
, dest_idx
;
2452 unsigned int context
;
2453 re_dfastate_t
*dest_state
;
2455 if (!dfa
->nodes
[cur_node_idx
].accept_mb
)
2458 if (dfa
->nodes
[cur_node_idx
].constraint
)
2460 context
= re_string_context_at (&mctx
->input
,
2461 re_string_cur_idx (&mctx
->input
),
2463 if (NOT_SATISFY_NEXT_CONSTRAINT (dfa
->nodes
[cur_node_idx
].constraint
,
2468 /* How many bytes the node can accept? */
2469 naccepted
= check_node_accept_bytes (dfa
, cur_node_idx
, &mctx
->input
,
2470 re_string_cur_idx (&mctx
->input
));
2474 /* The node can accepts `naccepted' bytes. */
2475 dest_idx
= re_string_cur_idx (&mctx
->input
) + naccepted
;
2476 mctx
->max_mb_elem_len
= ((mctx
->max_mb_elem_len
< naccepted
) ? naccepted
2477 : mctx
->max_mb_elem_len
);
2478 err
= clean_state_log_if_needed (mctx
, dest_idx
);
2479 if (BE (err
!= REG_NOERROR
, 0))
2482 assert (dfa
->nexts
[cur_node_idx
] != -1);
2484 new_nodes
= dfa
->eclosures
+ dfa
->nexts
[cur_node_idx
];
2486 dest_state
= mctx
->state_log
[dest_idx
];
2487 if (dest_state
== NULL
)
2488 dest_nodes
= *new_nodes
;
2491 err
= re_node_set_init_union (&dest_nodes
,
2492 dest_state
->entrance_nodes
, new_nodes
);
2493 if (BE (err
!= REG_NOERROR
, 0))
2496 context
= re_string_context_at (&mctx
->input
, dest_idx
- 1,
2498 mctx
->state_log
[dest_idx
]
2499 = re_acquire_state_context (&err
, dfa
, &dest_nodes
, context
);
2500 if (dest_state
!= NULL
)
2501 re_node_set_free (&dest_nodes
);
2502 if (BE (mctx
->state_log
[dest_idx
] == NULL
&& err
!= REG_NOERROR
, 0))
2507 #endif /* RE_ENABLE_I18N */
2509 static reg_errcode_t
2510 transit_state_bkref (re_match_context_t
*mctx
, const re_node_set
*nodes
)
2512 const re_dfa_t
*const dfa
= mctx
->dfa
;
2515 int cur_str_idx
= re_string_cur_idx (&mctx
->input
);
2517 for (i
= 0; i
< nodes
->nelem
; ++i
)
2519 int dest_str_idx
, prev_nelem
, bkc_idx
;
2520 int node_idx
= nodes
->elems
[i
];
2521 unsigned int context
;
2522 const re_token_t
*node
= dfa
->nodes
+ node_idx
;
2523 re_node_set
*new_dest_nodes
;
2525 /* Check whether `node' is a backreference or not. */
2526 if (node
->type
!= OP_BACK_REF
)
2529 if (node
->constraint
)
2531 context
= re_string_context_at (&mctx
->input
, cur_str_idx
,
2533 if (NOT_SATISFY_NEXT_CONSTRAINT (node
->constraint
, context
))
2537 /* `node' is a backreference.
2538 Check the substring which the substring matched. */
2539 bkc_idx
= mctx
->nbkref_ents
;
2540 err
= get_subexp (mctx
, node_idx
, cur_str_idx
);
2541 if (BE (err
!= REG_NOERROR
, 0))
2544 /* And add the epsilon closures (which is `new_dest_nodes') of
2545 the backreference to appropriate state_log. */
2547 assert (dfa
->nexts
[node_idx
] != -1);
2549 for (; bkc_idx
< mctx
->nbkref_ents
; ++bkc_idx
)
2552 re_dfastate_t
*dest_state
;
2553 struct re_backref_cache_entry
*bkref_ent
;
2554 bkref_ent
= mctx
->bkref_ents
+ bkc_idx
;
2555 if (bkref_ent
->node
!= node_idx
|| bkref_ent
->str_idx
!= cur_str_idx
)
2557 subexp_len
= bkref_ent
->subexp_to
- bkref_ent
->subexp_from
;
2558 new_dest_nodes
= (subexp_len
== 0
2559 ? dfa
->eclosures
+ dfa
->edests
[node_idx
].elems
[0]
2560 : dfa
->eclosures
+ dfa
->nexts
[node_idx
]);
2561 dest_str_idx
= (cur_str_idx
+ bkref_ent
->subexp_to
2562 - bkref_ent
->subexp_from
);
2563 context
= re_string_context_at (&mctx
->input
, dest_str_idx
- 1,
2565 dest_state
= mctx
->state_log
[dest_str_idx
];
2566 prev_nelem
= ((mctx
->state_log
[cur_str_idx
] == NULL
) ? 0
2567 : mctx
->state_log
[cur_str_idx
]->nodes
.nelem
);
2568 /* Add `new_dest_node' to state_log. */
2569 if (dest_state
== NULL
)
2571 mctx
->state_log
[dest_str_idx
]
2572 = re_acquire_state_context (&err
, dfa
, new_dest_nodes
,
2574 if (BE (mctx
->state_log
[dest_str_idx
] == NULL
2575 && err
!= REG_NOERROR
, 0))
2580 re_node_set dest_nodes
;
2581 err
= re_node_set_init_union (&dest_nodes
,
2582 dest_state
->entrance_nodes
,
2584 if (BE (err
!= REG_NOERROR
, 0))
2586 re_node_set_free (&dest_nodes
);
2589 mctx
->state_log
[dest_str_idx
]
2590 = re_acquire_state_context (&err
, dfa
, &dest_nodes
, context
);
2591 re_node_set_free (&dest_nodes
);
2592 if (BE (mctx
->state_log
[dest_str_idx
] == NULL
2593 && err
!= REG_NOERROR
, 0))
2596 /* We need to check recursively if the backreference can epsilon
2599 && mctx
->state_log
[cur_str_idx
]->nodes
.nelem
> prev_nelem
)
2601 err
= check_subexp_matching_top (mctx
, new_dest_nodes
,
2603 if (BE (err
!= REG_NOERROR
, 0))
2605 err
= transit_state_bkref (mctx
, new_dest_nodes
);
2606 if (BE (err
!= REG_NOERROR
, 0))
2616 /* Enumerate all the candidates which the backreference BKREF_NODE can match
2617 at BKREF_STR_IDX, and register them by match_ctx_add_entry().
2618 Note that we might collect inappropriate candidates here.
2619 However, the cost of checking them strictly here is too high, then we
2620 delay these checking for prune_impossible_nodes(). */
2622 static reg_errcode_t
2623 __attribute_warn_unused_result__
2624 get_subexp (re_match_context_t
*mctx
, int bkref_node
, int bkref_str_idx
)
2626 const re_dfa_t
*const dfa
= mctx
->dfa
;
2627 int subexp_num
, sub_top_idx
;
2628 const char *buf
= (const char *) re_string_get_buffer (&mctx
->input
);
2629 /* Return if we have already checked BKREF_NODE at BKREF_STR_IDX. */
2630 int cache_idx
= search_cur_bkref_entry (mctx
, bkref_str_idx
);
2631 if (cache_idx
!= -1)
2633 const struct re_backref_cache_entry
*entry
2634 = mctx
->bkref_ents
+ cache_idx
;
2636 if (entry
->node
== bkref_node
)
2637 return REG_NOERROR
; /* We already checked it. */
2638 while (entry
++->more
);
2641 subexp_num
= dfa
->nodes
[bkref_node
].opr
.idx
;
2643 /* For each sub expression */
2644 for (sub_top_idx
= 0; sub_top_idx
< mctx
->nsub_tops
; ++sub_top_idx
)
2647 re_sub_match_top_t
*sub_top
= mctx
->sub_tops
[sub_top_idx
];
2648 re_sub_match_last_t
*sub_last
;
2649 int sub_last_idx
, sl_str
, bkref_str_off
;
2651 if (dfa
->nodes
[sub_top
->node
].opr
.idx
!= subexp_num
)
2652 continue; /* It isn't related. */
2654 sl_str
= sub_top
->str_idx
;
2655 bkref_str_off
= bkref_str_idx
;
2656 /* At first, check the last node of sub expressions we already
2658 for (sub_last_idx
= 0; sub_last_idx
< sub_top
->nlasts
; ++sub_last_idx
)
2661 sub_last
= sub_top
->lasts
[sub_last_idx
];
2662 sl_str_diff
= sub_last
->str_idx
- sl_str
;
2663 /* The matched string by the sub expression match with the substring
2664 at the back reference? */
2665 if (sl_str_diff
> 0)
2667 if (BE (bkref_str_off
+ sl_str_diff
> mctx
->input
.valid_len
, 0))
2669 /* Not enough chars for a successful match. */
2670 if (bkref_str_off
+ sl_str_diff
> mctx
->input
.len
)
2673 err
= clean_state_log_if_needed (mctx
,
2676 if (BE (err
!= REG_NOERROR
, 0))
2678 buf
= (const char *) re_string_get_buffer (&mctx
->input
);
2680 if (memcmp (buf
+ bkref_str_off
, buf
+ sl_str
, sl_str_diff
) != 0)
2681 /* We don't need to search this sub expression any more. */
2684 bkref_str_off
+= sl_str_diff
;
2685 sl_str
+= sl_str_diff
;
2686 err
= get_subexp_sub (mctx
, sub_top
, sub_last
, bkref_node
,
2689 /* Reload buf, since the preceding call might have reallocated
2691 buf
= (const char *) re_string_get_buffer (&mctx
->input
);
2693 if (err
== REG_NOMATCH
)
2695 if (BE (err
!= REG_NOERROR
, 0))
2699 if (sub_last_idx
< sub_top
->nlasts
)
2701 if (sub_last_idx
> 0)
2703 /* Then, search for the other last nodes of the sub expression. */
2704 for (; sl_str
<= bkref_str_idx
; ++sl_str
)
2706 int cls_node
, sl_str_off
;
2707 const re_node_set
*nodes
;
2708 sl_str_off
= sl_str
- sub_top
->str_idx
;
2709 /* The matched string by the sub expression match with the substring
2710 at the back reference? */
2713 if (BE (bkref_str_off
>= mctx
->input
.valid_len
, 0))
2715 /* If we are at the end of the input, we cannot match. */
2716 if (bkref_str_off
>= mctx
->input
.len
)
2719 err
= extend_buffers (mctx
, bkref_str_off
+ 1);
2720 if (BE (err
!= REG_NOERROR
, 0))
2723 buf
= (const char *) re_string_get_buffer (&mctx
->input
);
2725 if (buf
[bkref_str_off
++] != buf
[sl_str
- 1])
2726 break; /* We don't need to search this sub expression
2729 if (mctx
->state_log
[sl_str
] == NULL
)
2731 /* Does this state have a ')' of the sub expression? */
2732 nodes
= &mctx
->state_log
[sl_str
]->nodes
;
2733 cls_node
= find_subexp_node (dfa
, nodes
, subexp_num
,
2737 if (sub_top
->path
== NULL
)
2739 sub_top
->path
= calloc (sizeof (state_array_t
),
2740 sl_str
- sub_top
->str_idx
+ 1);
2741 if (sub_top
->path
== NULL
)
2744 /* Can the OP_OPEN_SUBEXP node arrive the OP_CLOSE_SUBEXP node
2745 in the current context? */
2746 err
= check_arrival (mctx
, sub_top
->path
, sub_top
->node
,
2747 sub_top
->str_idx
, cls_node
, sl_str
,
2749 if (err
== REG_NOMATCH
)
2751 if (BE (err
!= REG_NOERROR
, 0))
2753 sub_last
= match_ctx_add_sublast (sub_top
, cls_node
, sl_str
);
2754 if (BE (sub_last
== NULL
, 0))
2756 err
= get_subexp_sub (mctx
, sub_top
, sub_last
, bkref_node
,
2758 if (err
== REG_NOMATCH
)
2765 /* Helper functions for get_subexp(). */
2767 /* Check SUB_LAST can arrive to the back reference BKREF_NODE at BKREF_STR.
2768 If it can arrive, register the sub expression expressed with SUB_TOP
2771 static reg_errcode_t
2772 get_subexp_sub (re_match_context_t
*mctx
, const re_sub_match_top_t
*sub_top
,
2773 re_sub_match_last_t
*sub_last
, int bkref_node
, int bkref_str
)
2777 /* Can the subexpression arrive the back reference? */
2778 err
= check_arrival (mctx
, &sub_last
->path
, sub_last
->node
,
2779 sub_last
->str_idx
, bkref_node
, bkref_str
,
2781 if (err
!= REG_NOERROR
)
2783 err
= match_ctx_add_entry (mctx
, bkref_node
, bkref_str
, sub_top
->str_idx
,
2785 if (BE (err
!= REG_NOERROR
, 0))
2787 to_idx
= bkref_str
+ sub_last
->str_idx
- sub_top
->str_idx
;
2788 return clean_state_log_if_needed (mctx
, to_idx
);
2791 /* Find the first node which is '(' or ')' and whose index is SUBEXP_IDX.
2792 Search '(' if FL_OPEN, or search ')' otherwise.
2793 TODO: This function isn't efficient...
2794 Because there might be more than one nodes whose types are
2795 OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all
2800 find_subexp_node (const re_dfa_t
*dfa
, const re_node_set
*nodes
,
2801 int subexp_idx
, int type
)
2804 for (cls_idx
= 0; cls_idx
< nodes
->nelem
; ++cls_idx
)
2806 int cls_node
= nodes
->elems
[cls_idx
];
2807 const re_token_t
*node
= dfa
->nodes
+ cls_node
;
2808 if (node
->type
== type
2809 && node
->opr
.idx
== subexp_idx
)
2815 /* Check whether the node TOP_NODE at TOP_STR can arrive to the node
2816 LAST_NODE at LAST_STR. We record the path onto PATH since it will be
2818 Return REG_NOERROR if it can arrive, or REG_NOMATCH otherwise. */
2820 static reg_errcode_t
2821 __attribute_warn_unused_result__
2822 check_arrival (re_match_context_t
*mctx
, state_array_t
*path
, int top_node
,
2823 int top_str
, int last_node
, int last_str
, int type
)
2825 const re_dfa_t
*const dfa
= mctx
->dfa
;
2826 reg_errcode_t err
= REG_NOERROR
;
2827 int subexp_num
, backup_cur_idx
, str_idx
, null_cnt
;
2828 re_dfastate_t
*cur_state
= NULL
;
2829 re_node_set
*cur_nodes
, next_nodes
;
2830 re_dfastate_t
**backup_state_log
;
2831 unsigned int context
;
2833 subexp_num
= dfa
->nodes
[top_node
].opr
.idx
;
2834 /* Extend the buffer if we need. */
2835 if (BE (path
->alloc
< last_str
+ mctx
->max_mb_elem_len
+ 1, 0))
2837 re_dfastate_t
**new_array
;
2838 int old_alloc
= path
->alloc
;
2839 path
->alloc
+= last_str
+ mctx
->max_mb_elem_len
+ 1;
2840 new_array
= re_realloc (path
->array
, re_dfastate_t
*, path
->alloc
);
2841 if (BE (new_array
== NULL
, 0))
2843 path
->alloc
= old_alloc
;
2846 path
->array
= new_array
;
2847 memset (new_array
+ old_alloc
, '\0',
2848 sizeof (re_dfastate_t
*) * (path
->alloc
- old_alloc
));
2851 str_idx
= path
->next_idx
?: top_str
;
2853 /* Temporary modify MCTX. */
2854 backup_state_log
= mctx
->state_log
;
2855 backup_cur_idx
= mctx
->input
.cur_idx
;
2856 mctx
->state_log
= path
->array
;
2857 mctx
->input
.cur_idx
= str_idx
;
2859 /* Setup initial node set. */
2860 context
= re_string_context_at (&mctx
->input
, str_idx
- 1, mctx
->eflags
);
2861 if (str_idx
== top_str
)
2863 err
= re_node_set_init_1 (&next_nodes
, top_node
);
2864 if (BE (err
!= REG_NOERROR
, 0))
2866 err
= check_arrival_expand_ecl (dfa
, &next_nodes
, subexp_num
, type
);
2867 if (BE (err
!= REG_NOERROR
, 0))
2869 re_node_set_free (&next_nodes
);
2875 cur_state
= mctx
->state_log
[str_idx
];
2876 if (cur_state
&& cur_state
->has_backref
)
2878 err
= re_node_set_init_copy (&next_nodes
, &cur_state
->nodes
);
2879 if (BE (err
!= REG_NOERROR
, 0))
2883 re_node_set_init_empty (&next_nodes
);
2885 if (str_idx
== top_str
|| (cur_state
&& cur_state
->has_backref
))
2887 if (next_nodes
.nelem
)
2889 err
= expand_bkref_cache (mctx
, &next_nodes
, str_idx
,
2891 if (BE (err
!= REG_NOERROR
, 0))
2893 re_node_set_free (&next_nodes
);
2897 cur_state
= re_acquire_state_context (&err
, dfa
, &next_nodes
, context
);
2898 if (BE (cur_state
== NULL
&& err
!= REG_NOERROR
, 0))
2900 re_node_set_free (&next_nodes
);
2903 mctx
->state_log
[str_idx
] = cur_state
;
2906 for (null_cnt
= 0; str_idx
< last_str
&& null_cnt
<= mctx
->max_mb_elem_len
;)
2908 re_node_set_empty (&next_nodes
);
2909 if (mctx
->state_log
[str_idx
+ 1])
2911 err
= re_node_set_merge (&next_nodes
,
2912 &mctx
->state_log
[str_idx
+ 1]->nodes
);
2913 if (BE (err
!= REG_NOERROR
, 0))
2915 re_node_set_free (&next_nodes
);
2921 err
= check_arrival_add_next_nodes (mctx
, str_idx
,
2922 &cur_state
->non_eps_nodes
,
2924 if (BE (err
!= REG_NOERROR
, 0))
2926 re_node_set_free (&next_nodes
);
2931 if (next_nodes
.nelem
)
2933 err
= check_arrival_expand_ecl (dfa
, &next_nodes
, subexp_num
, type
);
2934 if (BE (err
!= REG_NOERROR
, 0))
2936 re_node_set_free (&next_nodes
);
2939 err
= expand_bkref_cache (mctx
, &next_nodes
, str_idx
,
2941 if (BE (err
!= REG_NOERROR
, 0))
2943 re_node_set_free (&next_nodes
);
2947 context
= re_string_context_at (&mctx
->input
, str_idx
- 1, mctx
->eflags
);
2948 cur_state
= re_acquire_state_context (&err
, dfa
, &next_nodes
, context
);
2949 if (BE (cur_state
== NULL
&& err
!= REG_NOERROR
, 0))
2951 re_node_set_free (&next_nodes
);
2954 mctx
->state_log
[str_idx
] = cur_state
;
2955 null_cnt
= cur_state
== NULL
? null_cnt
+ 1 : 0;
2957 re_node_set_free (&next_nodes
);
2958 cur_nodes
= (mctx
->state_log
[last_str
] == NULL
? NULL
2959 : &mctx
->state_log
[last_str
]->nodes
);
2960 path
->next_idx
= str_idx
;
2963 mctx
->state_log
= backup_state_log
;
2964 mctx
->input
.cur_idx
= backup_cur_idx
;
2966 /* Then check the current node set has the node LAST_NODE. */
2967 if (cur_nodes
!= NULL
&& re_node_set_contains (cur_nodes
, last_node
))
2973 /* Helper functions for check_arrival. */
2975 /* Calculate the destination nodes of CUR_NODES at STR_IDX, and append them
2977 TODO: This function is similar to the functions transit_state*(),
2978 however this function has many additional works.
2979 Can't we unify them? */
2981 static reg_errcode_t
2982 __attribute_warn_unused_result__
2983 check_arrival_add_next_nodes (re_match_context_t
*mctx
, int str_idx
,
2984 re_node_set
*cur_nodes
, re_node_set
*next_nodes
)
2986 const re_dfa_t
*const dfa
= mctx
->dfa
;
2989 reg_errcode_t err
= REG_NOERROR
;
2990 re_node_set union_set
;
2991 re_node_set_init_empty (&union_set
);
2992 for (cur_idx
= 0; cur_idx
< cur_nodes
->nelem
; ++cur_idx
)
2995 int cur_node
= cur_nodes
->elems
[cur_idx
];
2997 re_token_type_t type
= dfa
->nodes
[cur_node
].type
;
2998 assert (!IS_EPSILON_NODE (type
));
3000 #ifdef RE_ENABLE_I18N
3001 /* If the node may accept `multi byte'. */
3002 if (dfa
->nodes
[cur_node
].accept_mb
)
3004 naccepted
= check_node_accept_bytes (dfa
, cur_node
, &mctx
->input
,
3008 re_dfastate_t
*dest_state
;
3009 int next_node
= dfa
->nexts
[cur_node
];
3010 int next_idx
= str_idx
+ naccepted
;
3011 dest_state
= mctx
->state_log
[next_idx
];
3012 re_node_set_empty (&union_set
);
3015 err
= re_node_set_merge (&union_set
, &dest_state
->nodes
);
3016 if (BE (err
!= REG_NOERROR
, 0))
3018 re_node_set_free (&union_set
);
3022 result
= re_node_set_insert (&union_set
, next_node
);
3023 if (BE (result
< 0, 0))
3025 re_node_set_free (&union_set
);
3028 mctx
->state_log
[next_idx
] = re_acquire_state (&err
, dfa
,
3030 if (BE (mctx
->state_log
[next_idx
] == NULL
3031 && err
!= REG_NOERROR
, 0))
3033 re_node_set_free (&union_set
);
3038 #endif /* RE_ENABLE_I18N */
3040 || check_node_accept (mctx
, dfa
->nodes
+ cur_node
, str_idx
))
3042 result
= re_node_set_insert (next_nodes
, dfa
->nexts
[cur_node
]);
3043 if (BE (result
< 0, 0))
3045 re_node_set_free (&union_set
);
3050 re_node_set_free (&union_set
);
3054 /* For all the nodes in CUR_NODES, add the epsilon closures of them to
3055 CUR_NODES, however exclude the nodes which are:
3056 - inside the sub expression whose number is EX_SUBEXP, if FL_OPEN.
3057 - out of the sub expression whose number is EX_SUBEXP, if !FL_OPEN.
3060 static reg_errcode_t
3061 check_arrival_expand_ecl (const re_dfa_t
*dfa
, re_node_set
*cur_nodes
,
3062 int ex_subexp
, int type
)
3065 int idx
, outside_node
;
3066 re_node_set new_nodes
;
3068 assert (cur_nodes
->nelem
);
3070 err
= re_node_set_alloc (&new_nodes
, cur_nodes
->nelem
);
3071 if (BE (err
!= REG_NOERROR
, 0))
3073 /* Create a new node set NEW_NODES with the nodes which are epsilon
3074 closures of the node in CUR_NODES. */
3076 for (idx
= 0; idx
< cur_nodes
->nelem
; ++idx
)
3078 int cur_node
= cur_nodes
->elems
[idx
];
3079 const re_node_set
*eclosure
= dfa
->eclosures
+ cur_node
;
3080 outside_node
= find_subexp_node (dfa
, eclosure
, ex_subexp
, type
);
3081 if (outside_node
== -1)
3083 /* There are no problematic nodes, just merge them. */
3084 err
= re_node_set_merge (&new_nodes
, eclosure
);
3085 if (BE (err
!= REG_NOERROR
, 0))
3087 re_node_set_free (&new_nodes
);
3093 /* There are problematic nodes, re-calculate incrementally. */
3094 err
= check_arrival_expand_ecl_sub (dfa
, &new_nodes
, cur_node
,
3096 if (BE (err
!= REG_NOERROR
, 0))
3098 re_node_set_free (&new_nodes
);
3103 re_node_set_free (cur_nodes
);
3104 *cur_nodes
= new_nodes
;
3108 /* Helper function for check_arrival_expand_ecl.
3109 Check incrementally the epsilon closure of TARGET, and if it isn't
3110 problematic append it to DST_NODES. */
3112 static reg_errcode_t
3113 __attribute_warn_unused_result__
3114 check_arrival_expand_ecl_sub (const re_dfa_t
*dfa
, re_node_set
*dst_nodes
,
3115 int target
, int ex_subexp
, int type
)
3118 for (cur_node
= target
; !re_node_set_contains (dst_nodes
, cur_node
);)
3122 if (dfa
->nodes
[cur_node
].type
== type
3123 && dfa
->nodes
[cur_node
].opr
.idx
== ex_subexp
)
3125 if (type
== OP_CLOSE_SUBEXP
)
3127 err
= re_node_set_insert (dst_nodes
, cur_node
);
3128 if (BE (err
== -1, 0))
3133 err
= re_node_set_insert (dst_nodes
, cur_node
);
3134 if (BE (err
== -1, 0))
3136 if (dfa
->edests
[cur_node
].nelem
== 0)
3138 if (dfa
->edests
[cur_node
].nelem
== 2)
3140 err
= check_arrival_expand_ecl_sub (dfa
, dst_nodes
,
3141 dfa
->edests
[cur_node
].elems
[1],
3143 if (BE (err
!= REG_NOERROR
, 0))
3146 cur_node
= dfa
->edests
[cur_node
].elems
[0];
3152 /* For all the back references in the current state, calculate the
3153 destination of the back references by the appropriate entry
3154 in MCTX->BKREF_ENTS. */
3156 static reg_errcode_t
3157 __attribute_warn_unused_result__
3158 expand_bkref_cache (re_match_context_t
*mctx
, re_node_set
*cur_nodes
,
3159 int cur_str
, int subexp_num
, int type
)
3161 const re_dfa_t
*const dfa
= mctx
->dfa
;
3163 int cache_idx_start
= search_cur_bkref_entry (mctx
, cur_str
);
3164 struct re_backref_cache_entry
*ent
;
3166 if (cache_idx_start
== -1)
3170 ent
= mctx
->bkref_ents
+ cache_idx_start
;
3173 int to_idx
, next_node
;
3175 /* Is this entry ENT is appropriate? */
3176 if (!re_node_set_contains (cur_nodes
, ent
->node
))
3179 to_idx
= cur_str
+ ent
->subexp_to
- ent
->subexp_from
;
3180 /* Calculate the destination of the back reference, and append it
3181 to MCTX->STATE_LOG. */
3182 if (to_idx
== cur_str
)
3184 /* The backreference did epsilon transit, we must re-check all the
3185 node in the current state. */
3186 re_node_set new_dests
;
3187 reg_errcode_t err2
, err3
;
3188 next_node
= dfa
->edests
[ent
->node
].elems
[0];
3189 if (re_node_set_contains (cur_nodes
, next_node
))
3191 err
= re_node_set_init_1 (&new_dests
, next_node
);
3192 err2
= check_arrival_expand_ecl (dfa
, &new_dests
, subexp_num
, type
);
3193 err3
= re_node_set_merge (cur_nodes
, &new_dests
);
3194 re_node_set_free (&new_dests
);
3195 if (BE (err
!= REG_NOERROR
|| err2
!= REG_NOERROR
3196 || err3
!= REG_NOERROR
, 0))
3198 err
= (err
!= REG_NOERROR
? err
3199 : (err2
!= REG_NOERROR
? err2
: err3
));
3202 /* TODO: It is still inefficient... */
3207 re_node_set union_set
;
3208 next_node
= dfa
->nexts
[ent
->node
];
3209 if (mctx
->state_log
[to_idx
])
3212 if (re_node_set_contains (&mctx
->state_log
[to_idx
]->nodes
,
3215 err
= re_node_set_init_copy (&union_set
,
3216 &mctx
->state_log
[to_idx
]->nodes
);
3217 ret
= re_node_set_insert (&union_set
, next_node
);
3218 if (BE (err
!= REG_NOERROR
|| ret
< 0, 0))
3220 re_node_set_free (&union_set
);
3221 err
= err
!= REG_NOERROR
? err
: REG_ESPACE
;
3227 err
= re_node_set_init_1 (&union_set
, next_node
);
3228 if (BE (err
!= REG_NOERROR
, 0))
3231 mctx
->state_log
[to_idx
] = re_acquire_state (&err
, dfa
, &union_set
);
3232 re_node_set_free (&union_set
);
3233 if (BE (mctx
->state_log
[to_idx
] == NULL
3234 && err
!= REG_NOERROR
, 0))
3238 while (ent
++->more
);
3242 /* Build transition table for the state.
3243 Return 1 if succeeded, otherwise return NULL. */
3246 build_trtable (const re_dfa_t
*dfa
, re_dfastate_t
*state
)
3249 int i
, j
, ch
, need_word_trtable
= 0;
3250 bitset_word_t elem
, mask
;
3251 bool dests_node_malloced
= false;
3252 bool dest_states_malloced
= false;
3253 int ndests
; /* Number of the destination states from `state'. */
3254 re_dfastate_t
**trtable
;
3255 re_dfastate_t
**dest_states
= NULL
, **dest_states_word
, **dest_states_nl
;
3256 re_node_set follows
, *dests_node
;
3258 bitset_t acceptable
;
3262 re_node_set dests_node
[SBC_MAX
];
3263 bitset_t dests_ch
[SBC_MAX
];
3266 /* We build DFA states which corresponds to the destination nodes
3267 from `state'. `dests_node[i]' represents the nodes which i-th
3268 destination state contains, and `dests_ch[i]' represents the
3269 characters which i-th destination state accepts. */
3270 if (__libc_use_alloca (sizeof (struct dests_alloc
)))
3271 dests_alloc
= (struct dests_alloc
*) alloca (sizeof (struct dests_alloc
));
3274 dests_alloc
= re_malloc (struct dests_alloc
, 1);
3275 if (BE (dests_alloc
== NULL
, 0))
3277 dests_node_malloced
= true;
3279 dests_node
= dests_alloc
->dests_node
;
3280 dests_ch
= dests_alloc
->dests_ch
;
3282 /* Initialize transiton table. */
3283 state
->word_trtable
= state
->trtable
= NULL
;
3285 /* At first, group all nodes belonging to `state' into several
3287 ndests
= group_nodes_into_DFAstates (dfa
, state
, dests_node
, dests_ch
);
3288 if (BE (ndests
<= 0, 0))
3290 if (dests_node_malloced
)
3292 /* Return 0 in case of an error, 1 otherwise. */
3295 state
->trtable
= (re_dfastate_t
**)
3296 calloc (sizeof (re_dfastate_t
*), SBC_MAX
);
3297 if (BE (state
->trtable
== NULL
, 0))
3304 err
= re_node_set_alloc (&follows
, ndests
+ 1);
3305 if (BE (err
!= REG_NOERROR
, 0))
3308 /* Avoid arithmetic overflow in size calculation. */
3309 if (BE ((((SIZE_MAX
- (sizeof (re_node_set
) + sizeof (bitset_t
)) * SBC_MAX
)
3310 / (3 * sizeof (re_dfastate_t
*)))
3315 if (__libc_use_alloca ((sizeof (re_node_set
) + sizeof (bitset_t
)) * SBC_MAX
3316 + ndests
* 3 * sizeof (re_dfastate_t
*)))
3317 dest_states
= (re_dfastate_t
**)
3318 alloca (ndests
* 3 * sizeof (re_dfastate_t
*));
3321 dest_states
= (re_dfastate_t
**)
3322 malloc (ndests
* 3 * sizeof (re_dfastate_t
*));
3323 if (BE (dest_states
== NULL
, 0))
3326 if (dest_states_malloced
)
3328 re_node_set_free (&follows
);
3329 for (i
= 0; i
< ndests
; ++i
)
3330 re_node_set_free (dests_node
+ i
);
3331 if (dests_node_malloced
)
3335 dest_states_malloced
= true;
3337 dest_states_word
= dest_states
+ ndests
;
3338 dest_states_nl
= dest_states_word
+ ndests
;
3339 bitset_empty (acceptable
);
3341 /* Then build the states for all destinations. */
3342 for (i
= 0; i
< ndests
; ++i
)
3345 re_node_set_empty (&follows
);
3346 /* Merge the follows of this destination states. */
3347 for (j
= 0; j
< dests_node
[i
].nelem
; ++j
)
3349 next_node
= dfa
->nexts
[dests_node
[i
].elems
[j
]];
3350 if (next_node
!= -1)
3352 err
= re_node_set_merge (&follows
, dfa
->eclosures
+ next_node
);
3353 if (BE (err
!= REG_NOERROR
, 0))
3357 dest_states
[i
] = re_acquire_state_context (&err
, dfa
, &follows
, 0);
3358 if (BE (dest_states
[i
] == NULL
&& err
!= REG_NOERROR
, 0))
3360 /* If the new state has context constraint,
3361 build appropriate states for these contexts. */
3362 if (dest_states
[i
]->has_constraint
)
3364 dest_states_word
[i
] = re_acquire_state_context (&err
, dfa
, &follows
,
3366 if (BE (dest_states_word
[i
] == NULL
&& err
!= REG_NOERROR
, 0))
3369 if (dest_states
[i
] != dest_states_word
[i
] && dfa
->mb_cur_max
> 1)
3370 need_word_trtable
= 1;
3372 dest_states_nl
[i
] = re_acquire_state_context (&err
, dfa
, &follows
,
3374 if (BE (dest_states_nl
[i
] == NULL
&& err
!= REG_NOERROR
, 0))
3379 dest_states_word
[i
] = dest_states
[i
];
3380 dest_states_nl
[i
] = dest_states
[i
];
3382 bitset_merge (acceptable
, dests_ch
[i
]);
3385 if (!BE (need_word_trtable
, 0))
3387 /* We don't care about whether the following character is a word
3388 character, or we are in a single-byte character set so we can
3389 discern by looking at the character code: allocate a
3390 256-entry transition table. */
3391 trtable
= state
->trtable
=
3392 (re_dfastate_t
**) calloc (sizeof (re_dfastate_t
*), SBC_MAX
);
3393 if (BE (trtable
== NULL
, 0))
3396 /* For all characters ch...: */
3397 for (i
= 0; i
< BITSET_WORDS
; ++i
)
3398 for (ch
= i
* BITSET_WORD_BITS
, elem
= acceptable
[i
], mask
= 1;
3400 mask
<<= 1, elem
>>= 1, ++ch
)
3401 if (BE (elem
& 1, 0))
3403 /* There must be exactly one destination which accepts
3404 character ch. See group_nodes_into_DFAstates. */
3405 for (j
= 0; (dests_ch
[j
][i
] & mask
) == 0; ++j
)
3408 /* j-th destination accepts the word character ch. */
3409 if (dfa
->word_char
[i
] & mask
)
3410 trtable
[ch
] = dest_states_word
[j
];
3412 trtable
[ch
] = dest_states
[j
];
3417 /* We care about whether the following character is a word
3418 character, and we are in a multi-byte character set: discern
3419 by looking at the character code: build two 256-entry
3420 transition tables, one starting at trtable[0] and one
3421 starting at trtable[SBC_MAX]. */
3422 trtable
= state
->word_trtable
=
3423 (re_dfastate_t
**) calloc (sizeof (re_dfastate_t
*), 2 * SBC_MAX
);
3424 if (BE (trtable
== NULL
, 0))
3427 /* For all characters ch...: */
3428 for (i
= 0; i
< BITSET_WORDS
; ++i
)
3429 for (ch
= i
* BITSET_WORD_BITS
, elem
= acceptable
[i
], mask
= 1;
3431 mask
<<= 1, elem
>>= 1, ++ch
)
3432 if (BE (elem
& 1, 0))
3434 /* There must be exactly one destination which accepts
3435 character ch. See group_nodes_into_DFAstates. */
3436 for (j
= 0; (dests_ch
[j
][i
] & mask
) == 0; ++j
)
3439 /* j-th destination accepts the word character ch. */
3440 trtable
[ch
] = dest_states
[j
];
3441 trtable
[ch
+ SBC_MAX
] = dest_states_word
[j
];
3446 if (bitset_contain (acceptable
, NEWLINE_CHAR
))
3448 /* The current state accepts newline character. */
3449 for (j
= 0; j
< ndests
; ++j
)
3450 if (bitset_contain (dests_ch
[j
], NEWLINE_CHAR
))
3452 /* k-th destination accepts newline character. */
3453 trtable
[NEWLINE_CHAR
] = dest_states_nl
[j
];
3454 if (need_word_trtable
)
3455 trtable
[NEWLINE_CHAR
+ SBC_MAX
] = dest_states_nl
[j
];
3456 /* There must be only one destination which accepts
3457 newline. See group_nodes_into_DFAstates. */
3462 if (dest_states_malloced
)
3465 re_node_set_free (&follows
);
3466 for (i
= 0; i
< ndests
; ++i
)
3467 re_node_set_free (dests_node
+ i
);
3469 if (dests_node_malloced
)
3475 /* Group all nodes belonging to STATE into several destinations.
3476 Then for all destinations, set the nodes belonging to the destination
3477 to DESTS_NODE[i] and set the characters accepted by the destination
3478 to DEST_CH[i]. This function return the number of destinations. */
3481 group_nodes_into_DFAstates (const re_dfa_t
*dfa
, const re_dfastate_t
*state
,
3482 re_node_set
*dests_node
, bitset_t
*dests_ch
)
3487 int ndests
; /* Number of the destinations from `state'. */
3488 bitset_t accepts
; /* Characters a node can accept. */
3489 const re_node_set
*cur_nodes
= &state
->nodes
;
3490 bitset_empty (accepts
);
3493 /* For all the nodes belonging to `state', */
3494 for (i
= 0; i
< cur_nodes
->nelem
; ++i
)
3496 re_token_t
*node
= &dfa
->nodes
[cur_nodes
->elems
[i
]];
3497 re_token_type_t type
= node
->type
;
3498 unsigned int constraint
= node
->constraint
;
3500 /* Enumerate all single byte character this node can accept. */
3501 if (type
== CHARACTER
)
3502 bitset_set (accepts
, node
->opr
.c
);
3503 else if (type
== SIMPLE_BRACKET
)
3505 bitset_merge (accepts
, node
->opr
.sbcset
);
3507 else if (type
== OP_PERIOD
)
3509 #ifdef RE_ENABLE_I18N
3510 if (dfa
->mb_cur_max
> 1)
3511 bitset_merge (accepts
, dfa
->sb_char
);
3514 bitset_set_all (accepts
);
3515 if (!(dfa
->syntax
& RE_DOT_NEWLINE
))
3516 bitset_clear (accepts
, '\n');
3517 if (dfa
->syntax
& RE_DOT_NOT_NULL
)
3518 bitset_clear (accepts
, '\0');
3520 #ifdef RE_ENABLE_I18N
3521 else if (type
== OP_UTF8_PERIOD
)
3523 memset (accepts
, '\xff', sizeof (bitset_t
) / 2);
3524 if (!(dfa
->syntax
& RE_DOT_NEWLINE
))
3525 bitset_clear (accepts
, '\n');
3526 if (dfa
->syntax
& RE_DOT_NOT_NULL
)
3527 bitset_clear (accepts
, '\0');
3533 /* Check the `accepts' and sift the characters which are not
3534 match it the context. */
3537 if (constraint
& NEXT_NEWLINE_CONSTRAINT
)
3539 bool accepts_newline
= bitset_contain (accepts
, NEWLINE_CHAR
);
3540 bitset_empty (accepts
);
3541 if (accepts_newline
)
3542 bitset_set (accepts
, NEWLINE_CHAR
);
3546 if (constraint
& NEXT_ENDBUF_CONSTRAINT
)
3548 bitset_empty (accepts
);
3552 if (constraint
& NEXT_WORD_CONSTRAINT
)
3554 bitset_word_t any_set
= 0;
3555 if (type
== CHARACTER
&& !node
->word_char
)
3557 bitset_empty (accepts
);
3560 #ifdef RE_ENABLE_I18N
3561 if (dfa
->mb_cur_max
> 1)
3562 for (j
= 0; j
< BITSET_WORDS
; ++j
)
3563 any_set
|= (accepts
[j
] &= (dfa
->word_char
[j
] | ~dfa
->sb_char
[j
]));
3566 for (j
= 0; j
< BITSET_WORDS
; ++j
)
3567 any_set
|= (accepts
[j
] &= dfa
->word_char
[j
]);
3571 if (constraint
& NEXT_NOTWORD_CONSTRAINT
)
3573 bitset_word_t any_set
= 0;
3574 if (type
== CHARACTER
&& node
->word_char
)
3576 bitset_empty (accepts
);
3579 #ifdef RE_ENABLE_I18N
3580 if (dfa
->mb_cur_max
> 1)
3581 for (j
= 0; j
< BITSET_WORDS
; ++j
)
3582 any_set
|= (accepts
[j
] &= ~(dfa
->word_char
[j
] & dfa
->sb_char
[j
]));
3585 for (j
= 0; j
< BITSET_WORDS
; ++j
)
3586 any_set
|= (accepts
[j
] &= ~dfa
->word_char
[j
]);
3592 /* Then divide `accepts' into DFA states, or create a new
3593 state. Above, we make sure that accepts is not empty. */
3594 for (j
= 0; j
< ndests
; ++j
)
3596 bitset_t intersec
; /* Intersection sets, see below. */
3598 /* Flags, see below. */
3599 bitset_word_t has_intersec
, not_subset
, not_consumed
;
3601 /* Optimization, skip if this state doesn't accept the character. */
3602 if (type
== CHARACTER
&& !bitset_contain (dests_ch
[j
], node
->opr
.c
))
3605 /* Enumerate the intersection set of this state and `accepts'. */
3607 for (k
= 0; k
< BITSET_WORDS
; ++k
)
3608 has_intersec
|= intersec
[k
] = accepts
[k
] & dests_ch
[j
][k
];
3609 /* And skip if the intersection set is empty. */
3613 /* Then check if this state is a subset of `accepts'. */
3614 not_subset
= not_consumed
= 0;
3615 for (k
= 0; k
< BITSET_WORDS
; ++k
)
3617 not_subset
|= remains
[k
] = ~accepts
[k
] & dests_ch
[j
][k
];
3618 not_consumed
|= accepts
[k
] = accepts
[k
] & ~dests_ch
[j
][k
];
3621 /* If this state isn't a subset of `accepts', create a
3622 new group state, which has the `remains'. */
3625 bitset_copy (dests_ch
[ndests
], remains
);
3626 bitset_copy (dests_ch
[j
], intersec
);
3627 err
= re_node_set_init_copy (dests_node
+ ndests
, &dests_node
[j
]);
3628 if (BE (err
!= REG_NOERROR
, 0))
3633 /* Put the position in the current group. */
3634 result
= re_node_set_insert (&dests_node
[j
], cur_nodes
->elems
[i
]);
3635 if (BE (result
< 0, 0))
3638 /* If all characters are consumed, go to next node. */
3642 /* Some characters remain, create a new group. */
3645 bitset_copy (dests_ch
[ndests
], accepts
);
3646 err
= re_node_set_init_1 (dests_node
+ ndests
, cur_nodes
->elems
[i
]);
3647 if (BE (err
!= REG_NOERROR
, 0))
3650 bitset_empty (accepts
);
3655 for (j
= 0; j
< ndests
; ++j
)
3656 re_node_set_free (dests_node
+ j
);
3660 #ifdef RE_ENABLE_I18N
3661 /* Check how many bytes the node `dfa->nodes[node_idx]' accepts.
3662 Return the number of the bytes the node accepts.
3663 STR_IDX is the current index of the input string.
3665 This function handles the nodes which can accept one character, or
3666 one collating element like '.', '[a-z]', opposite to the other nodes
3667 can only accept one byte. */
3670 # include <locale/weight.h>
3674 check_node_accept_bytes (const re_dfa_t
*dfa
, int node_idx
,
3675 const re_string_t
*input
, int str_idx
)
3677 const re_token_t
*node
= dfa
->nodes
+ node_idx
;
3678 int char_len
, elem_len
;
3681 if (BE (node
->type
== OP_UTF8_PERIOD
, 0))
3683 unsigned char c
= re_string_byte_at (input
, str_idx
), d
;
3684 if (BE (c
< 0xc2, 1))
3687 if (str_idx
+ 2 > input
->len
)
3690 d
= re_string_byte_at (input
, str_idx
+ 1);
3692 return (d
< 0x80 || d
> 0xbf) ? 0 : 2;
3696 if (c
== 0xe0 && d
< 0xa0)
3702 if (c
== 0xf0 && d
< 0x90)
3708 if (c
== 0xf8 && d
< 0x88)
3714 if (c
== 0xfc && d
< 0x84)
3720 if (str_idx
+ char_len
> input
->len
)
3723 for (i
= 1; i
< char_len
; ++i
)
3725 d
= re_string_byte_at (input
, str_idx
+ i
);
3726 if (d
< 0x80 || d
> 0xbf)
3732 char_len
= re_string_char_size_at (input
, str_idx
);
3733 if (node
->type
== OP_PERIOD
)
3737 /* FIXME: I don't think this if is needed, as both '\n'
3738 and '\0' are char_len == 1. */
3739 /* '.' accepts any one character except the following two cases. */
3740 if ((!(dfa
->syntax
& RE_DOT_NEWLINE
) &&
3741 re_string_byte_at (input
, str_idx
) == '\n') ||
3742 ((dfa
->syntax
& RE_DOT_NOT_NULL
) &&
3743 re_string_byte_at (input
, str_idx
) == '\0'))
3748 elem_len
= re_string_elem_size_at (input
, str_idx
);
3749 if ((elem_len
<= 1 && char_len
<= 1) || char_len
== 0)
3752 if (node
->type
== COMPLEX_BRACKET
)
3754 const re_charset_t
*cset
= node
->opr
.mbcset
;
3756 const unsigned char *pin
3757 = ((const unsigned char *) re_string_get_buffer (input
) + str_idx
);
3762 wchar_t wc
= ((cset
->nranges
|| cset
->nchar_classes
|| cset
->nmbchars
)
3763 ? re_string_wchar_at (input
, str_idx
) : 0);
3765 /* match with multibyte character? */
3766 for (i
= 0; i
< cset
->nmbchars
; ++i
)
3767 if (wc
== cset
->mbchars
[i
])
3769 match_len
= char_len
;
3770 goto check_node_accept_bytes_match
;
3772 /* match with character_class? */
3773 for (i
= 0; i
< cset
->nchar_classes
; ++i
)
3775 wctype_t wt
= cset
->char_classes
[i
];
3776 if (__iswctype (wc
, wt
))
3778 match_len
= char_len
;
3779 goto check_node_accept_bytes_match
;
3784 nrules
= _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
3787 unsigned int in_collseq
= 0;
3788 const int32_t *table
, *indirect
;
3789 const unsigned char *weights
, *extra
;
3790 const char *collseqwc
;
3792 /* match with collating_symbol? */
3793 if (cset
->ncoll_syms
)
3794 extra
= (const unsigned char *)
3795 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
);
3796 for (i
= 0; i
< cset
->ncoll_syms
; ++i
)
3798 const unsigned char *coll_sym
= extra
+ cset
->coll_syms
[i
];
3799 /* Compare the length of input collating element and
3800 the length of current collating element. */
3801 if (*coll_sym
!= elem_len
)
3803 /* Compare each bytes. */
3804 for (j
= 0; j
< *coll_sym
; j
++)
3805 if (pin
[j
] != coll_sym
[1 + j
])
3809 /* Match if every bytes is equal. */
3811 goto check_node_accept_bytes_match
;
3817 if (elem_len
<= char_len
)
3819 collseqwc
= _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_COLLSEQWC
);
3820 in_collseq
= __collseq_table_lookup (collseqwc
, wc
);
3823 in_collseq
= find_collation_sequence_value (pin
, elem_len
);
3825 /* match with range expression? */
3826 for (i
= 0; i
< cset
->nranges
; ++i
)
3827 if (cset
->range_starts
[i
] <= in_collseq
3828 && in_collseq
<= cset
->range_ends
[i
])
3830 match_len
= elem_len
;
3831 goto check_node_accept_bytes_match
;
3834 /* match with equivalence_class? */
3835 if (cset
->nequiv_classes
)
3837 const unsigned char *cp
= pin
;
3838 table
= (const int32_t *)
3839 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_TABLEMB
);
3840 weights
= (const unsigned char *)
3841 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_WEIGHTMB
);
3842 extra
= (const unsigned char *)
3843 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_EXTRAMB
);
3844 indirect
= (const int32_t *)
3845 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_INDIRECTMB
);
3846 int32_t idx
= findidx (table
, indirect
, extra
, &cp
, elem_len
);
3848 for (i
= 0; i
< cset
->nequiv_classes
; ++i
)
3850 int32_t equiv_class_idx
= cset
->equiv_classes
[i
];
3851 size_t weight_len
= weights
[idx
& 0xffffff];
3852 if (weight_len
== weights
[equiv_class_idx
& 0xffffff]
3853 && (idx
>> 24) == (equiv_class_idx
>> 24))
3858 equiv_class_idx
&= 0xffffff;
3860 while (cnt
<= weight_len
3861 && (weights
[equiv_class_idx
+ 1 + cnt
]
3862 == weights
[idx
+ 1 + cnt
]))
3864 if (cnt
> weight_len
)
3866 match_len
= elem_len
;
3867 goto check_node_accept_bytes_match
;
3876 /* match with range expression? */
3878 wchar_t cmp_buf
[] = {L
'\0', L
'\0', wc
, L
'\0', L
'\0', L
'\0'};
3880 wchar_t cmp_buf
[] = {L
'\0', L
'\0', L
'\0', L
'\0', L
'\0', L
'\0'};
3883 for (i
= 0; i
< cset
->nranges
; ++i
)
3885 cmp_buf
[0] = cset
->range_starts
[i
];
3886 cmp_buf
[4] = cset
->range_ends
[i
];
3887 if (__wcscoll (cmp_buf
, cmp_buf
+ 2) <= 0
3888 && __wcscoll (cmp_buf
+ 2, cmp_buf
+ 4) <= 0)
3890 match_len
= char_len
;
3891 goto check_node_accept_bytes_match
;
3895 check_node_accept_bytes_match
:
3896 if (!cset
->non_match
)
3903 return (elem_len
> char_len
) ? elem_len
: char_len
;
3911 find_collation_sequence_value (const unsigned char *mbs
, size_t mbs_len
)
3913 uint32_t nrules
= _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
3918 /* No valid character. Match it as a single byte character. */
3919 const unsigned char *collseq
= (const unsigned char *)
3920 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_COLLSEQMB
);
3921 return collseq
[mbs
[0]];
3928 const unsigned char *extra
= (const unsigned char *)
3929 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
);
3930 int32_t extrasize
= (const unsigned char *)
3931 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
+ 1) - extra
;
3933 for (idx
= 0; idx
< extrasize
;)
3935 int mbs_cnt
, found
= 0;
3936 int32_t elem_mbs_len
;
3937 /* Skip the name of collating element name. */
3938 idx
= idx
+ extra
[idx
] + 1;
3939 elem_mbs_len
= extra
[idx
++];
3940 if (mbs_len
== elem_mbs_len
)
3942 for (mbs_cnt
= 0; mbs_cnt
< elem_mbs_len
; ++mbs_cnt
)
3943 if (extra
[idx
+ mbs_cnt
] != mbs
[mbs_cnt
])
3945 if (mbs_cnt
== elem_mbs_len
)
3946 /* Found the entry. */
3949 /* Skip the byte sequence of the collating element. */
3950 idx
+= elem_mbs_len
;
3951 /* Adjust for the alignment. */
3952 idx
= (idx
+ 3) & ~3;
3953 /* Skip the collation sequence value. */
3954 idx
+= sizeof (uint32_t);
3955 /* Skip the wide char sequence of the collating element. */
3956 idx
= idx
+ sizeof (uint32_t) * (*(int32_t *) (extra
+ idx
) + 1);
3957 /* If we found the entry, return the sequence value. */
3959 return *(uint32_t *) (extra
+ idx
);
3960 /* Skip the collation sequence value. */
3961 idx
+= sizeof (uint32_t);
3967 #endif /* RE_ENABLE_I18N */
3969 /* Check whether the node accepts the byte which is IDX-th
3970 byte of the INPUT. */
3973 check_node_accept (const re_match_context_t
*mctx
, const re_token_t
*node
,
3977 ch
= re_string_byte_at (&mctx
->input
, idx
);
3981 if (node
->opr
.c
!= ch
)
3985 case SIMPLE_BRACKET
:
3986 if (!bitset_contain (node
->opr
.sbcset
, ch
))
3990 #ifdef RE_ENABLE_I18N
3991 case OP_UTF8_PERIOD
:
3997 if ((ch
== '\n' && !(mctx
->dfa
->syntax
& RE_DOT_NEWLINE
))
3998 || (ch
== '\0' && (mctx
->dfa
->syntax
& RE_DOT_NOT_NULL
)))
4006 if (node
->constraint
)
4008 /* The node has constraints. Check whether the current context
4009 satisfies the constraints. */
4010 unsigned int context
= re_string_context_at (&mctx
->input
, idx
,
4012 if (NOT_SATISFY_NEXT_CONSTRAINT (node
->constraint
, context
))
4019 /* Extend the buffers, if the buffers have run out. */
4021 static reg_errcode_t
4022 __attribute_warn_unused_result__
4023 extend_buffers (re_match_context_t
*mctx
, int min_len
)
4026 re_string_t
*pstr
= &mctx
->input
;
4028 /* Avoid overflow. */
4029 if (BE (INT_MAX
/ 2 / sizeof (re_dfastate_t
*) <= pstr
->bufs_len
, 0))
4032 /* Double the lengthes of the buffers, but allocate at least MIN_LEN. */
4033 ret
= re_string_realloc_buffers (pstr
,
4035 MIN (pstr
->len
, pstr
->bufs_len
* 2)));
4036 if (BE (ret
!= REG_NOERROR
, 0))
4039 if (mctx
->state_log
!= NULL
)
4041 /* And double the length of state_log. */
4042 /* XXX We have no indication of the size of this buffer. If this
4043 allocation fail we have no indication that the state_log array
4044 does not have the right size. */
4045 re_dfastate_t
**new_array
= re_realloc (mctx
->state_log
, re_dfastate_t
*,
4046 pstr
->bufs_len
+ 1);
4047 if (BE (new_array
== NULL
, 0))
4049 mctx
->state_log
= new_array
;
4052 /* Then reconstruct the buffers. */
4055 #ifdef RE_ENABLE_I18N
4056 if (pstr
->mb_cur_max
> 1)
4058 ret
= build_wcs_upper_buffer (pstr
);
4059 if (BE (ret
!= REG_NOERROR
, 0))
4063 #endif /* RE_ENABLE_I18N */
4064 build_upper_buffer (pstr
);
4068 #ifdef RE_ENABLE_I18N
4069 if (pstr
->mb_cur_max
> 1)
4070 build_wcs_buffer (pstr
);
4072 #endif /* RE_ENABLE_I18N */
4074 if (pstr
->trans
!= NULL
)
4075 re_string_translate_buffer (pstr
);
4082 /* Functions for matching context. */
4084 /* Initialize MCTX. */
4086 static reg_errcode_t
4087 __attribute_warn_unused_result__
4088 match_ctx_init (re_match_context_t
*mctx
, int eflags
, int n
)
4090 mctx
->eflags
= eflags
;
4091 mctx
->match_last
= -1;
4094 mctx
->bkref_ents
= re_malloc (struct re_backref_cache_entry
, n
);
4095 mctx
->sub_tops
= re_malloc (re_sub_match_top_t
*, n
);
4096 if (BE (mctx
->bkref_ents
== NULL
|| mctx
->sub_tops
== NULL
, 0))
4099 /* Already zero-ed by the caller.
4101 mctx->bkref_ents = NULL;
4102 mctx->nbkref_ents = 0;
4103 mctx->nsub_tops = 0; */
4104 mctx
->abkref_ents
= n
;
4105 mctx
->max_mb_elem_len
= 1;
4106 mctx
->asub_tops
= n
;
4110 /* Clean the entries which depend on the current input in MCTX.
4111 This function must be invoked when the matcher changes the start index
4112 of the input, or changes the input string. */
4115 match_ctx_clean (re_match_context_t
*mctx
)
4118 for (st_idx
= 0; st_idx
< mctx
->nsub_tops
; ++st_idx
)
4121 re_sub_match_top_t
*top
= mctx
->sub_tops
[st_idx
];
4122 for (sl_idx
= 0; sl_idx
< top
->nlasts
; ++sl_idx
)
4124 re_sub_match_last_t
*last
= top
->lasts
[sl_idx
];
4125 re_free (last
->path
.array
);
4128 re_free (top
->lasts
);
4131 re_free (top
->path
->array
);
4132 re_free (top
->path
);
4137 mctx
->nsub_tops
= 0;
4138 mctx
->nbkref_ents
= 0;
4141 /* Free all the memory associated with MCTX. */
4144 match_ctx_free (re_match_context_t
*mctx
)
4146 /* First, free all the memory associated with MCTX->SUB_TOPS. */
4147 match_ctx_clean (mctx
);
4148 re_free (mctx
->sub_tops
);
4149 re_free (mctx
->bkref_ents
);
4152 /* Add a new backreference entry to MCTX.
4153 Note that we assume that caller never call this function with duplicate
4154 entry, and call with STR_IDX which isn't smaller than any existing entry.
4157 static reg_errcode_t
4158 __attribute_warn_unused_result__
4159 match_ctx_add_entry (re_match_context_t
*mctx
, int node
, int str_idx
, int from
,
4162 if (mctx
->nbkref_ents
>= mctx
->abkref_ents
)
4164 struct re_backref_cache_entry
* new_entry
;
4165 new_entry
= re_realloc (mctx
->bkref_ents
, struct re_backref_cache_entry
,
4166 mctx
->abkref_ents
* 2);
4167 if (BE (new_entry
== NULL
, 0))
4169 re_free (mctx
->bkref_ents
);
4172 mctx
->bkref_ents
= new_entry
;
4173 memset (mctx
->bkref_ents
+ mctx
->nbkref_ents
, '\0',
4174 sizeof (struct re_backref_cache_entry
) * mctx
->abkref_ents
);
4175 mctx
->abkref_ents
*= 2;
4177 if (mctx
->nbkref_ents
> 0
4178 && mctx
->bkref_ents
[mctx
->nbkref_ents
- 1].str_idx
== str_idx
)
4179 mctx
->bkref_ents
[mctx
->nbkref_ents
- 1].more
= 1;
4181 mctx
->bkref_ents
[mctx
->nbkref_ents
].node
= node
;
4182 mctx
->bkref_ents
[mctx
->nbkref_ents
].str_idx
= str_idx
;
4183 mctx
->bkref_ents
[mctx
->nbkref_ents
].subexp_from
= from
;
4184 mctx
->bkref_ents
[mctx
->nbkref_ents
].subexp_to
= to
;
4186 /* This is a cache that saves negative results of check_dst_limits_calc_pos.
4187 If bit N is clear, means that this entry won't epsilon-transition to
4188 an OP_OPEN_SUBEXP or OP_CLOSE_SUBEXP for the N+1-th subexpression. If
4189 it is set, check_dst_limits_calc_pos_1 will recurse and try to find one
4192 A backreference does not epsilon-transition unless it is empty, so set
4193 to all zeros if FROM != TO. */
4194 mctx
->bkref_ents
[mctx
->nbkref_ents
].eps_reachable_subexps_map
4195 = (from
== to
? ~0 : 0);
4197 mctx
->bkref_ents
[mctx
->nbkref_ents
++].more
= 0;
4198 if (mctx
->max_mb_elem_len
< to
- from
)
4199 mctx
->max_mb_elem_len
= to
- from
;
4203 /* Search for the first entry which has the same str_idx, or -1 if none is
4204 found. Note that MCTX->BKREF_ENTS is already sorted by MCTX->STR_IDX. */
4207 search_cur_bkref_entry (const re_match_context_t
*mctx
, int str_idx
)
4209 int left
, right
, mid
, last
;
4210 last
= right
= mctx
->nbkref_ents
;
4211 for (left
= 0; left
< right
;)
4213 mid
= (left
+ right
) / 2;
4214 if (mctx
->bkref_ents
[mid
].str_idx
< str_idx
)
4219 if (left
< last
&& mctx
->bkref_ents
[left
].str_idx
== str_idx
)
4225 /* Register the node NODE, whose type is OP_OPEN_SUBEXP, and which matches
4228 static reg_errcode_t
4229 __attribute_warn_unused_result__
4230 match_ctx_add_subtop (re_match_context_t
*mctx
, int node
, int str_idx
)
4233 assert (mctx
->sub_tops
!= NULL
);
4234 assert (mctx
->asub_tops
> 0);
4236 if (BE (mctx
->nsub_tops
== mctx
->asub_tops
, 0))
4238 int new_asub_tops
= mctx
->asub_tops
* 2;
4239 re_sub_match_top_t
**new_array
= re_realloc (mctx
->sub_tops
,
4240 re_sub_match_top_t
*,
4242 if (BE (new_array
== NULL
, 0))
4244 mctx
->sub_tops
= new_array
;
4245 mctx
->asub_tops
= new_asub_tops
;
4247 mctx
->sub_tops
[mctx
->nsub_tops
] = calloc (1, sizeof (re_sub_match_top_t
));
4248 if (BE (mctx
->sub_tops
[mctx
->nsub_tops
] == NULL
, 0))
4250 mctx
->sub_tops
[mctx
->nsub_tops
]->node
= node
;
4251 mctx
->sub_tops
[mctx
->nsub_tops
++]->str_idx
= str_idx
;
4255 /* Register the node NODE, whose type is OP_CLOSE_SUBEXP, and which matches
4256 at STR_IDX, whose corresponding OP_OPEN_SUBEXP is SUB_TOP. */
4258 static re_sub_match_last_t
*
4259 match_ctx_add_sublast (re_sub_match_top_t
*subtop
, int node
, int str_idx
)
4261 re_sub_match_last_t
*new_entry
;
4262 if (BE (subtop
->nlasts
== subtop
->alasts
, 0))
4264 int new_alasts
= 2 * subtop
->alasts
+ 1;
4265 re_sub_match_last_t
**new_array
= re_realloc (subtop
->lasts
,
4266 re_sub_match_last_t
*,
4268 if (BE (new_array
== NULL
, 0))
4270 subtop
->lasts
= new_array
;
4271 subtop
->alasts
= new_alasts
;
4273 new_entry
= calloc (1, sizeof (re_sub_match_last_t
));
4274 if (BE (new_entry
!= NULL
, 1))
4276 subtop
->lasts
[subtop
->nlasts
] = new_entry
;
4277 new_entry
->node
= node
;
4278 new_entry
->str_idx
= str_idx
;
4285 sift_ctx_init (re_sift_context_t
*sctx
, re_dfastate_t
**sifted_sts
,
4286 re_dfastate_t
**limited_sts
, int last_node
, int last_str_idx
)
4288 sctx
->sifted_states
= sifted_sts
;
4289 sctx
->limited_states
= limited_sts
;
4290 sctx
->last_node
= last_node
;
4291 sctx
->last_str_idx
= last_str_idx
;
4292 re_node_set_init_empty (&sctx
->limits
);