1 /* Extended regular expression matching and search library.
2 Copyright (C) 2002, 2003, 2004, 2005, 2007, 2009 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, write to the Free
18 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
21 static reg_errcode_t
match_ctx_init (re_match_context_t
*cache
, int eflags
,
22 int n
) internal_function
;
23 static void match_ctx_clean (re_match_context_t
*mctx
) internal_function
;
24 static void match_ctx_free (re_match_context_t
*cache
) internal_function
;
25 static reg_errcode_t
match_ctx_add_entry (re_match_context_t
*cache
, int node
,
26 int str_idx
, int from
, int to
)
28 static int search_cur_bkref_entry (const re_match_context_t
*mctx
, int str_idx
)
30 static reg_errcode_t
match_ctx_add_subtop (re_match_context_t
*mctx
, int node
,
31 int str_idx
) internal_function
;
32 static re_sub_match_last_t
* match_ctx_add_sublast (re_sub_match_top_t
*subtop
,
33 int node
, int str_idx
)
35 static void sift_ctx_init (re_sift_context_t
*sctx
, re_dfastate_t
**sifted_sts
,
36 re_dfastate_t
**limited_sts
, int last_node
,
39 static reg_errcode_t
re_search_internal (const regex_t
*preg
,
40 const char *string
, int length
,
41 int start
, int range
, int stop
,
42 size_t nmatch
, regmatch_t pmatch
[],
43 int eflags
) internal_function
;
44 static int re_search_2_stub (struct re_pattern_buffer
*bufp
,
45 const char *string1
, int length1
,
46 const char *string2
, int length2
,
47 int start
, int range
, struct re_registers
*regs
,
48 int stop
, int ret_len
) internal_function
;
49 static int re_search_stub (struct re_pattern_buffer
*bufp
,
50 const char *string
, int length
, int start
,
51 int range
, int stop
, struct re_registers
*regs
,
52 int ret_len
) internal_function
;
53 static unsigned re_copy_regs (struct re_registers
*regs
, regmatch_t
*pmatch
,
54 int nregs
, int regs_allocated
) internal_function
;
55 static reg_errcode_t
prune_impossible_nodes (re_match_context_t
*mctx
)
57 static int check_matching (re_match_context_t
*mctx
, int fl_longest_match
,
58 int *p_match_first
) internal_function
;
59 static int check_halt_state_context (const re_match_context_t
*mctx
,
60 const re_dfastate_t
*state
, int idx
)
62 static void update_regs (const re_dfa_t
*dfa
, regmatch_t
*pmatch
,
63 regmatch_t
*prev_idx_match
, int cur_node
,
64 int cur_idx
, int nmatch
) internal_function
;
65 static reg_errcode_t
push_fail_stack (struct re_fail_stack_t
*fs
,
66 int str_idx
, int dest_node
, int nregs
,
68 re_node_set
*eps_via_nodes
)
70 static reg_errcode_t
set_regs (const regex_t
*preg
,
71 const re_match_context_t
*mctx
,
72 size_t nmatch
, regmatch_t
*pmatch
,
73 int fl_backtrack
) internal_function
;
74 static reg_errcode_t
free_fail_stack_return (struct re_fail_stack_t
*fs
)
78 static int sift_states_iter_mb (const re_match_context_t
*mctx
,
79 re_sift_context_t
*sctx
,
80 int node_idx
, int str_idx
, int max_str_idx
)
82 #endif /* RE_ENABLE_I18N */
83 static reg_errcode_t
sift_states_backward (const re_match_context_t
*mctx
,
84 re_sift_context_t
*sctx
)
86 static reg_errcode_t
build_sifted_states (const re_match_context_t
*mctx
,
87 re_sift_context_t
*sctx
, int str_idx
,
88 re_node_set
*cur_dest
)
90 static reg_errcode_t
update_cur_sifted_state (const re_match_context_t
*mctx
,
91 re_sift_context_t
*sctx
,
93 re_node_set
*dest_nodes
)
95 static reg_errcode_t
add_epsilon_src_nodes (const re_dfa_t
*dfa
,
96 re_node_set
*dest_nodes
,
97 const re_node_set
*candidates
)
99 static int check_dst_limits (const re_match_context_t
*mctx
,
101 int dst_node
, int dst_idx
, int src_node
,
102 int src_idx
) internal_function
;
103 static int check_dst_limits_calc_pos_1 (const re_match_context_t
*mctx
,
104 int boundaries
, int subexp_idx
,
105 int from_node
, int bkref_idx
)
107 static int check_dst_limits_calc_pos (const re_match_context_t
*mctx
,
108 int limit
, int subexp_idx
,
109 int node
, int str_idx
,
110 int bkref_idx
) internal_function
;
111 static reg_errcode_t
check_subexp_limits (const re_dfa_t
*dfa
,
112 re_node_set
*dest_nodes
,
113 const re_node_set
*candidates
,
115 struct re_backref_cache_entry
*bkref_ents
,
116 int str_idx
) internal_function
;
117 static reg_errcode_t
sift_states_bkref (const re_match_context_t
*mctx
,
118 re_sift_context_t
*sctx
,
119 int str_idx
, const re_node_set
*candidates
)
121 static reg_errcode_t
merge_state_array (const re_dfa_t
*dfa
,
123 re_dfastate_t
**src
, int num
)
125 static re_dfastate_t
*find_recover_state (reg_errcode_t
*err
,
126 re_match_context_t
*mctx
) internal_function
;
127 static re_dfastate_t
*transit_state (reg_errcode_t
*err
,
128 re_match_context_t
*mctx
,
129 re_dfastate_t
*state
) internal_function
;
130 static re_dfastate_t
*merge_state_with_log (reg_errcode_t
*err
,
131 re_match_context_t
*mctx
,
132 re_dfastate_t
*next_state
)
134 static reg_errcode_t
check_subexp_matching_top (re_match_context_t
*mctx
,
135 re_node_set
*cur_nodes
,
136 int str_idx
) internal_function
;
138 static re_dfastate_t
*transit_state_sb (reg_errcode_t
*err
,
139 re_match_context_t
*mctx
,
140 re_dfastate_t
*pstate
)
143 #ifdef RE_ENABLE_I18N
144 static reg_errcode_t
transit_state_mb (re_match_context_t
*mctx
,
145 re_dfastate_t
*pstate
)
147 #endif /* RE_ENABLE_I18N */
148 static reg_errcode_t
transit_state_bkref (re_match_context_t
*mctx
,
149 const re_node_set
*nodes
)
151 static reg_errcode_t
get_subexp (re_match_context_t
*mctx
,
152 int bkref_node
, int bkref_str_idx
)
154 static reg_errcode_t
get_subexp_sub (re_match_context_t
*mctx
,
155 const re_sub_match_top_t
*sub_top
,
156 re_sub_match_last_t
*sub_last
,
157 int bkref_node
, int bkref_str
)
159 static int find_subexp_node (const re_dfa_t
*dfa
, const re_node_set
*nodes
,
160 int subexp_idx
, int type
) internal_function
;
161 static reg_errcode_t
check_arrival (re_match_context_t
*mctx
,
162 state_array_t
*path
, int top_node
,
163 int top_str
, int last_node
, int last_str
,
164 int type
) internal_function
;
165 static reg_errcode_t
check_arrival_add_next_nodes (re_match_context_t
*mctx
,
167 re_node_set
*cur_nodes
,
168 re_node_set
*next_nodes
)
170 static reg_errcode_t
check_arrival_expand_ecl (const re_dfa_t
*dfa
,
171 re_node_set
*cur_nodes
,
172 int ex_subexp
, int type
)
174 static reg_errcode_t
check_arrival_expand_ecl_sub (const re_dfa_t
*dfa
,
175 re_node_set
*dst_nodes
,
176 int target
, int ex_subexp
,
177 int type
) internal_function
;
178 static reg_errcode_t
expand_bkref_cache (re_match_context_t
*mctx
,
179 re_node_set
*cur_nodes
, int cur_str
,
180 int subexp_num
, int type
)
182 static int build_trtable (const re_dfa_t
*dfa
,
183 re_dfastate_t
*state
) internal_function
;
184 #ifdef RE_ENABLE_I18N
185 static int check_node_accept_bytes (const re_dfa_t
*dfa
, int node_idx
,
186 const re_string_t
*input
, int idx
)
189 static unsigned int find_collation_sequence_value (const unsigned char *mbs
,
193 #endif /* RE_ENABLE_I18N */
194 static int group_nodes_into_DFAstates (const re_dfa_t
*dfa
,
195 const re_dfastate_t
*state
,
196 re_node_set
*states_node
,
197 bitset_t
*states_ch
) internal_function
;
198 static int check_node_accept (const re_match_context_t
*mctx
,
199 const re_token_t
*node
, int idx
)
201 static reg_errcode_t
extend_buffers (re_match_context_t
*mctx
)
204 /* Entry point for POSIX code. */
206 /* regexec searches for a given pattern, specified by PREG, in the
209 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
210 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
211 least NMATCH elements, and we set them to the offsets of the
212 corresponding matched substrings.
214 EFLAGS specifies `execution flags' which affect matching: if
215 REG_NOTBOL is set, then ^ does not match at the beginning of the
216 string; if REG_NOTEOL is set, then $ does not match at the end.
218 We return 0 if we find a match and REG_NOMATCH if not. */
221 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
222 const regex_t
*__restrict preg
;
223 const char *__restrict string
;
230 re_dfa_t
*dfa
= (re_dfa_t
*) preg
->buffer
;
232 if (eflags
& ~(REG_NOTBOL
| REG_NOTEOL
| REG_STARTEND
))
235 if (eflags
& REG_STARTEND
)
237 start
= pmatch
[0].rm_so
;
238 length
= pmatch
[0].rm_eo
;
243 length
= strlen (string
);
246 __libc_lock_lock (dfa
->lock
);
248 err
= re_search_internal (preg
, string
, length
, start
, length
- start
,
249 length
, 0, NULL
, eflags
);
251 err
= re_search_internal (preg
, string
, length
, start
, length
- start
,
252 length
, nmatch
, pmatch
, eflags
);
253 __libc_lock_unlock (dfa
->lock
);
254 return err
!= REG_NOERROR
;
258 # include <shlib-compat.h>
259 versioned_symbol (libc
, __regexec
, regexec
, GLIBC_2_3_4
);
261 # if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_3_4)
262 __typeof__ (__regexec
) __compat_regexec
;
265 attribute_compat_text_section
266 __compat_regexec (const regex_t
*__restrict preg
,
267 const char *__restrict string
, size_t nmatch
,
268 regmatch_t pmatch
[], int eflags
)
270 return regexec (preg
, string
, nmatch
, pmatch
,
271 eflags
& (REG_NOTBOL
| REG_NOTEOL
));
273 compat_symbol (libc
, __compat_regexec
, regexec
, GLIBC_2_0
);
277 /* Entry points for GNU code. */
279 /* re_match, re_search, re_match_2, re_search_2
281 The former two functions operate on STRING with length LENGTH,
282 while the later two operate on concatenation of STRING1 and STRING2
283 with lengths LENGTH1 and LENGTH2, respectively.
285 re_match() matches the compiled pattern in BUFP against the string,
286 starting at index START.
288 re_search() first tries matching at index START, then it tries to match
289 starting from index START + 1, and so on. The last start position tried
290 is START + RANGE. (Thus RANGE = 0 forces re_search to operate the same
293 The parameter STOP of re_{match,search}_2 specifies that no match exceeding
294 the first STOP characters of the concatenation of the strings should be
297 If REGS is not NULL, and BUFP->no_sub is not set, the offsets of the match
298 and all groups is stroed in REGS. (For the "_2" variants, the offsets are
299 computed relative to the concatenation, not relative to the individual
302 On success, re_match* functions return the length of the match, re_search*
303 return the position of the start of the match. Return value -1 means no
304 match was found and -2 indicates an internal error. */
307 re_match (bufp
, string
, length
, start
, regs
)
308 struct re_pattern_buffer
*bufp
;
311 struct re_registers
*regs
;
313 return re_search_stub (bufp
, string
, length
, start
, 0, length
, regs
, 1);
316 weak_alias (__re_match
, re_match
)
320 re_search (bufp
, string
, length
, start
, range
, regs
)
321 struct re_pattern_buffer
*bufp
;
323 int length
, start
, range
;
324 struct re_registers
*regs
;
326 return re_search_stub (bufp
, string
, length
, start
, range
, length
, regs
, 0);
329 weak_alias (__re_search
, re_search
)
333 re_match_2 (bufp
, string1
, length1
, string2
, length2
, start
, regs
, stop
)
334 struct re_pattern_buffer
*bufp
;
335 const char *string1
, *string2
;
336 int length1
, length2
, start
, stop
;
337 struct re_registers
*regs
;
339 return re_search_2_stub (bufp
, string1
, length1
, string2
, length2
,
340 start
, 0, regs
, stop
, 1);
343 weak_alias (__re_match_2
, re_match_2
)
347 re_search_2 (bufp
, string1
, length1
, string2
, length2
, start
, range
, regs
, stop
)
348 struct re_pattern_buffer
*bufp
;
349 const char *string1
, *string2
;
350 int length1
, length2
, start
, range
, stop
;
351 struct re_registers
*regs
;
353 return re_search_2_stub (bufp
, string1
, length1
, string2
, length2
,
354 start
, range
, regs
, stop
, 0);
357 weak_alias (__re_search_2
, re_search_2
)
361 re_search_2_stub (bufp
, string1
, length1
, string2
, length2
, start
, range
, regs
,
363 struct re_pattern_buffer
*bufp
;
364 const char *string1
, *string2
;
365 int length1
, length2
, start
, range
, stop
, ret_len
;
366 struct re_registers
*regs
;
370 int len
= length1
+ length2
;
373 if (BE (length1
< 0 || length2
< 0 || stop
< 0, 0))
376 /* Concatenate the strings. */
380 char *s
= re_malloc (char, len
);
382 if (BE (s
== NULL
, 0))
385 memcpy (__mempcpy (s
, string1
, length1
), string2
, length2
);
387 memcpy (s
, string1
, length1
);
388 memcpy (s
+ length1
, string2
, length2
);
398 rval
= re_search_stub (bufp
, str
, len
, start
, range
, stop
, regs
,
401 re_free ((char *) str
);
405 /* The parameters have the same meaning as those of re_search.
406 Additional parameters:
407 If RET_LEN is nonzero the length of the match is returned (re_match style);
408 otherwise the position of the match is returned. */
411 re_search_stub (bufp
, string
, length
, start
, range
, stop
, regs
, ret_len
)
412 struct re_pattern_buffer
*bufp
;
414 int length
, start
, range
, stop
, ret_len
;
415 struct re_registers
*regs
;
417 reg_errcode_t result
;
421 re_dfa_t
*dfa
= (re_dfa_t
*) bufp
->buffer
;
423 /* Check for out-of-range. */
424 if (BE (start
< 0 || start
> length
, 0))
426 if (BE (start
+ range
> length
, 0))
427 range
= length
- start
;
428 else if (BE (start
+ range
< 0, 0))
431 __libc_lock_lock (dfa
->lock
);
433 eflags
|= (bufp
->not_bol
) ? REG_NOTBOL
: 0;
434 eflags
|= (bufp
->not_eol
) ? REG_NOTEOL
: 0;
436 /* Compile fastmap if we haven't yet. */
437 if (range
> 0 && bufp
->fastmap
!= NULL
&& !bufp
->fastmap_accurate
)
438 re_compile_fastmap (bufp
);
440 if (BE (bufp
->no_sub
, 0))
443 /* We need at least 1 register. */
446 else if (BE (bufp
->regs_allocated
== REGS_FIXED
&&
447 regs
->num_regs
< bufp
->re_nsub
+ 1, 0))
449 nregs
= regs
->num_regs
;
450 if (BE (nregs
< 1, 0))
452 /* Nothing can be copied to regs. */
458 nregs
= bufp
->re_nsub
+ 1;
459 pmatch
= re_malloc (regmatch_t
, nregs
);
460 if (BE (pmatch
== NULL
, 0))
466 result
= re_search_internal (bufp
, string
, length
, start
, range
, stop
,
467 nregs
, pmatch
, eflags
);
471 /* I hope we needn't fill ther regs with -1's when no match was found. */
472 if (result
!= REG_NOERROR
)
474 else if (regs
!= NULL
)
476 /* If caller wants register contents data back, copy them. */
477 bufp
->regs_allocated
= re_copy_regs (regs
, pmatch
, nregs
,
478 bufp
->regs_allocated
);
479 if (BE (bufp
->regs_allocated
== REGS_UNALLOCATED
, 0))
483 if (BE (rval
== 0, 1))
487 assert (pmatch
[0].rm_so
== start
);
488 rval
= pmatch
[0].rm_eo
- start
;
491 rval
= pmatch
[0].rm_so
;
495 __libc_lock_unlock (dfa
->lock
);
500 re_copy_regs (regs
, pmatch
, nregs
, regs_allocated
)
501 struct re_registers
*regs
;
503 int nregs
, regs_allocated
;
505 int rval
= REGS_REALLOCATE
;
507 int need_regs
= nregs
+ 1;
508 /* We need one extra element beyond `num_regs' for the `-1' marker GNU code
511 /* Have the register data arrays been allocated? */
512 if (regs_allocated
== REGS_UNALLOCATED
)
513 { /* No. So allocate them with malloc. */
514 regs
->start
= re_malloc (regoff_t
, need_regs
);
515 regs
->end
= re_malloc (regoff_t
, need_regs
);
516 if (BE (regs
->start
== NULL
, 0) || BE (regs
->end
== NULL
, 0))
517 return REGS_UNALLOCATED
;
518 regs
->num_regs
= need_regs
;
520 else if (regs_allocated
== REGS_REALLOCATE
)
521 { /* Yes. If we need more elements than were already
522 allocated, reallocate them. If we need fewer, just
524 if (BE (need_regs
> regs
->num_regs
, 0))
526 regoff_t
*new_start
= re_realloc (regs
->start
, regoff_t
, need_regs
);
527 regoff_t
*new_end
= re_realloc (regs
->end
, regoff_t
, need_regs
);
528 if (BE (new_start
== NULL
, 0) || BE (new_end
== NULL
, 0))
529 return REGS_UNALLOCATED
;
530 regs
->start
= new_start
;
532 regs
->num_regs
= need_regs
;
537 assert (regs_allocated
== REGS_FIXED
);
538 /* This function may not be called with REGS_FIXED and nregs too big. */
539 assert (regs
->num_regs
>= nregs
);
544 for (i
= 0; i
< nregs
; ++i
)
546 regs
->start
[i
] = pmatch
[i
].rm_so
;
547 regs
->end
[i
] = pmatch
[i
].rm_eo
;
549 for ( ; i
< regs
->num_regs
; ++i
)
550 regs
->start
[i
] = regs
->end
[i
] = -1;
555 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
556 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
557 this memory for recording register information. STARTS and ENDS
558 must be allocated using the malloc library routine, and must each
559 be at least NUM_REGS * sizeof (regoff_t) bytes long.
561 If NUM_REGS == 0, then subsequent matches should allocate their own
564 Unless this function is called, the first search or match using
565 PATTERN_BUFFER will allocate its own register data, without
566 freeing the old data. */
569 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
570 struct re_pattern_buffer
*bufp
;
571 struct re_registers
*regs
;
573 regoff_t
*starts
, *ends
;
577 bufp
->regs_allocated
= REGS_REALLOCATE
;
578 regs
->num_regs
= num_regs
;
579 regs
->start
= starts
;
584 bufp
->regs_allocated
= REGS_UNALLOCATED
;
586 regs
->start
= regs
->end
= (regoff_t
*) 0;
590 weak_alias (__re_set_registers
, re_set_registers
)
593 /* Entry points compatible with 4.2 BSD regex library. We don't define
594 them unless specifically requested. */
596 #if defined _REGEX_RE_COMP || defined _LIBC
604 return 0 == regexec (&re_comp_buf
, s
, 0, NULL
, 0);
606 #endif /* _REGEX_RE_COMP */
608 /* Internal entry point. */
610 /* Searches for a compiled pattern PREG in the string STRING, whose
611 length is LENGTH. NMATCH, PMATCH, and EFLAGS have the same
612 mingings with regexec. START, and RANGE have the same meanings
614 Return REG_NOERROR if we find a match, and REG_NOMATCH if not,
615 otherwise return the error code.
616 Note: We assume front end functions already check ranges.
617 (START + RANGE >= 0 && START + RANGE <= LENGTH) */
620 re_search_internal (preg
, string
, length
, start
, range
, stop
, nmatch
, pmatch
,
624 int length
, start
, range
, stop
, eflags
;
629 const re_dfa_t
*dfa
= (const re_dfa_t
*) preg
->buffer
;
630 int left_lim
, right_lim
, incr
;
631 int fl_longest_match
, match_first
, match_kind
, match_last
= -1;
634 #if defined _LIBC || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L)
635 re_match_context_t mctx
= { .dfa
= dfa
};
637 re_match_context_t mctx
;
639 char *fastmap
= (preg
->fastmap
!= NULL
&& preg
->fastmap_accurate
640 && range
&& !preg
->can_be_null
) ? preg
->fastmap
: NULL
;
641 RE_TRANSLATE_TYPE t
= preg
->translate
;
643 #if !(defined _LIBC || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L))
644 memset (&mctx
, '\0', sizeof (re_match_context_t
));
648 extra_nmatch
= (nmatch
> preg
->re_nsub
) ? nmatch
- (preg
->re_nsub
+ 1) : 0;
649 nmatch
-= extra_nmatch
;
651 /* Check if the DFA haven't been compiled. */
652 if (BE (preg
->used
== 0 || dfa
->init_state
== NULL
653 || dfa
->init_state_word
== NULL
|| dfa
->init_state_nl
== NULL
654 || dfa
->init_state_begbuf
== NULL
, 0))
658 /* We assume front-end functions already check them. */
659 assert (start
+ range
>= 0 && start
+ range
<= length
);
662 /* If initial states with non-begbuf contexts have no elements,
663 the regex must be anchored. If preg->newline_anchor is set,
664 we'll never use init_state_nl, so do not check it. */
665 if (dfa
->init_state
->nodes
.nelem
== 0
666 && dfa
->init_state_word
->nodes
.nelem
== 0
667 && (dfa
->init_state_nl
->nodes
.nelem
== 0
668 || !preg
->newline_anchor
))
670 if (start
!= 0 && start
+ range
!= 0)
675 /* We must check the longest matching, if nmatch > 0. */
676 fl_longest_match
= (nmatch
!= 0 || dfa
->nbackref
);
678 err
= re_string_allocate (&mctx
.input
, string
, length
, dfa
->nodes_len
+ 1,
679 preg
->translate
, preg
->syntax
& RE_ICASE
, dfa
);
680 if (BE (err
!= REG_NOERROR
, 0))
682 mctx
.input
.stop
= stop
;
683 mctx
.input
.raw_stop
= stop
;
684 mctx
.input
.newline_anchor
= preg
->newline_anchor
;
686 err
= match_ctx_init (&mctx
, eflags
, dfa
->nbackref
* 2);
687 if (BE (err
!= REG_NOERROR
, 0))
690 /* We will log all the DFA states through which the dfa pass,
691 if nmatch > 1, or this dfa has "multibyte node", which is a
692 back-reference or a node which can accept multibyte character or
693 multi character collating element. */
694 if (nmatch
> 1 || dfa
->has_mb_node
)
696 mctx
.state_log
= re_malloc (re_dfastate_t
*, mctx
.input
.bufs_len
+ 1);
697 if (BE (mctx
.state_log
== NULL
, 0))
704 mctx
.state_log
= NULL
;
707 mctx
.input
.tip_context
= (eflags
& REG_NOTBOL
) ? CONTEXT_BEGBUF
708 : CONTEXT_NEWLINE
| CONTEXT_BEGBUF
;
710 /* Check incrementally whether of not the input string match. */
711 incr
= (range
< 0) ? -1 : 1;
712 left_lim
= (range
< 0) ? start
+ range
: start
;
713 right_lim
= (range
< 0) ? start
: start
+ range
;
714 sb
= dfa
->mb_cur_max
== 1;
717 ? ((sb
|| !(preg
->syntax
& RE_ICASE
|| t
) ? 4 : 0)
718 | (range
>= 0 ? 2 : 0)
719 | (t
!= NULL
? 1 : 0))
722 for (;; match_first
+= incr
)
725 if (match_first
< left_lim
|| right_lim
< match_first
)
728 /* Advance as rapidly as possible through the string, until we
729 find a plausible place to start matching. This may be done
730 with varying efficiency, so there are various possibilities:
731 only the most common of them are specialized, in order to
732 save on code size. We use a switch statement for speed. */
740 /* Fastmap with single-byte translation, match forward. */
741 while (BE (match_first
< right_lim
, 1)
742 && !fastmap
[t
[(unsigned char) string
[match_first
]]])
744 goto forward_match_found_start_or_reached_end
;
747 /* Fastmap without translation, match forward. */
748 while (BE (match_first
< right_lim
, 1)
749 && !fastmap
[(unsigned char) string
[match_first
]])
752 forward_match_found_start_or_reached_end
:
753 if (BE (match_first
== right_lim
, 0))
755 ch
= match_first
>= length
756 ? 0 : (unsigned char) string
[match_first
];
757 if (!fastmap
[t
? t
[ch
] : ch
])
764 /* Fastmap without multi-byte translation, match backwards. */
765 while (match_first
>= left_lim
)
767 ch
= match_first
>= length
768 ? 0 : (unsigned char) string
[match_first
];
769 if (fastmap
[t
? t
[ch
] : ch
])
773 if (match_first
< left_lim
)
778 /* In this case, we can't determine easily the current byte,
779 since it might be a component byte of a multibyte
780 character. Then we use the constructed buffer instead. */
783 /* If MATCH_FIRST is out of the valid range, reconstruct the
785 unsigned int offset
= match_first
- mctx
.input
.raw_mbs_idx
;
786 if (BE (offset
>= (unsigned int) mctx
.input
.valid_raw_len
, 0))
788 err
= re_string_reconstruct (&mctx
.input
, match_first
,
790 if (BE (err
!= REG_NOERROR
, 0))
793 offset
= match_first
- mctx
.input
.raw_mbs_idx
;
795 /* If MATCH_FIRST is out of the buffer, leave it as '\0'.
796 Note that MATCH_FIRST must not be smaller than 0. */
797 ch
= (match_first
>= length
798 ? 0 : re_string_byte_at (&mctx
.input
, offset
));
802 if (match_first
< left_lim
|| match_first
> right_lim
)
811 /* Reconstruct the buffers so that the matcher can assume that
812 the matching starts from the beginning of the buffer. */
813 err
= re_string_reconstruct (&mctx
.input
, match_first
, eflags
);
814 if (BE (err
!= REG_NOERROR
, 0))
817 #ifdef RE_ENABLE_I18N
818 /* Don't consider this char as a possible match start if it part,
819 yet isn't the head, of a multibyte character. */
820 if (!sb
&& !re_string_first_byte (&mctx
.input
, 0))
824 /* It seems to be appropriate one, then use the matcher. */
825 /* We assume that the matching starts from 0. */
826 mctx
.state_log_top
= mctx
.nbkref_ents
= mctx
.max_mb_elem_len
= 0;
827 match_last
= check_matching (&mctx
, fl_longest_match
,
828 range
>= 0 ? &match_first
: NULL
);
829 if (match_last
!= -1)
831 if (BE (match_last
== -2, 0))
838 mctx
.match_last
= match_last
;
839 if ((!preg
->no_sub
&& nmatch
> 1) || dfa
->nbackref
)
841 re_dfastate_t
*pstate
= mctx
.state_log
[match_last
];
842 mctx
.last_node
= check_halt_state_context (&mctx
, pstate
,
845 if ((!preg
->no_sub
&& nmatch
> 1 && dfa
->has_plural_match
)
848 err
= prune_impossible_nodes (&mctx
);
849 if (err
== REG_NOERROR
)
851 if (BE (err
!= REG_NOMATCH
, 0))
856 break; /* We found a match. */
860 match_ctx_clean (&mctx
);
864 assert (match_last
!= -1);
865 assert (err
== REG_NOERROR
);
868 /* Set pmatch[] if we need. */
873 /* Initialize registers. */
874 for (reg_idx
= 1; reg_idx
< nmatch
; ++reg_idx
)
875 pmatch
[reg_idx
].rm_so
= pmatch
[reg_idx
].rm_eo
= -1;
877 /* Set the points where matching start/end. */
879 pmatch
[0].rm_eo
= mctx
.match_last
;
881 if (!preg
->no_sub
&& nmatch
> 1)
883 err
= set_regs (preg
, &mctx
, nmatch
, pmatch
,
884 dfa
->has_plural_match
&& dfa
->nbackref
> 0);
885 if (BE (err
!= REG_NOERROR
, 0))
889 /* At last, add the offset to the each registers, since we slided
890 the buffers so that we could assume that the matching starts
892 for (reg_idx
= 0; reg_idx
< nmatch
; ++reg_idx
)
893 if (pmatch
[reg_idx
].rm_so
!= -1)
895 #ifdef RE_ENABLE_I18N
896 if (BE (mctx
.input
.offsets_needed
!= 0, 0))
898 pmatch
[reg_idx
].rm_so
=
899 (pmatch
[reg_idx
].rm_so
== mctx
.input
.valid_len
900 ? mctx
.input
.valid_raw_len
901 : mctx
.input
.offsets
[pmatch
[reg_idx
].rm_so
]);
902 pmatch
[reg_idx
].rm_eo
=
903 (pmatch
[reg_idx
].rm_eo
== mctx
.input
.valid_len
904 ? mctx
.input
.valid_raw_len
905 : mctx
.input
.offsets
[pmatch
[reg_idx
].rm_eo
]);
908 assert (mctx
.input
.offsets_needed
== 0);
910 pmatch
[reg_idx
].rm_so
+= match_first
;
911 pmatch
[reg_idx
].rm_eo
+= match_first
;
913 for (reg_idx
= 0; reg_idx
< extra_nmatch
; ++reg_idx
)
915 pmatch
[nmatch
+ reg_idx
].rm_so
= -1;
916 pmatch
[nmatch
+ reg_idx
].rm_eo
= -1;
920 for (reg_idx
= 0; reg_idx
+ 1 < nmatch
; reg_idx
++)
921 if (dfa
->subexp_map
[reg_idx
] != reg_idx
)
923 pmatch
[reg_idx
+ 1].rm_so
924 = pmatch
[dfa
->subexp_map
[reg_idx
] + 1].rm_so
;
925 pmatch
[reg_idx
+ 1].rm_eo
926 = pmatch
[dfa
->subexp_map
[reg_idx
] + 1].rm_eo
;
931 re_free (mctx
.state_log
);
933 match_ctx_free (&mctx
);
934 re_string_destruct (&mctx
.input
);
939 prune_impossible_nodes (mctx
)
940 re_match_context_t
*mctx
;
942 const re_dfa_t
*const dfa
= mctx
->dfa
;
943 int halt_node
, match_last
;
945 re_dfastate_t
**sifted_states
;
946 re_dfastate_t
**lim_states
= NULL
;
947 re_sift_context_t sctx
;
949 assert (mctx
->state_log
!= NULL
);
951 match_last
= mctx
->match_last
;
952 halt_node
= mctx
->last_node
;
953 sifted_states
= re_malloc (re_dfastate_t
*, match_last
+ 1);
954 if (BE (sifted_states
== NULL
, 0))
961 lim_states
= re_malloc (re_dfastate_t
*, match_last
+ 1);
962 if (BE (lim_states
== NULL
, 0))
969 memset (lim_states
, '\0',
970 sizeof (re_dfastate_t
*) * (match_last
+ 1));
971 sift_ctx_init (&sctx
, sifted_states
, lim_states
, halt_node
,
973 ret
= sift_states_backward (mctx
, &sctx
);
974 re_node_set_free (&sctx
.limits
);
975 if (BE (ret
!= REG_NOERROR
, 0))
977 if (sifted_states
[0] != NULL
|| lim_states
[0] != NULL
)
987 } while (mctx
->state_log
[match_last
] == NULL
988 || !mctx
->state_log
[match_last
]->halt
);
989 halt_node
= check_halt_state_context (mctx
,
990 mctx
->state_log
[match_last
],
993 ret
= merge_state_array (dfa
, sifted_states
, lim_states
,
995 re_free (lim_states
);
997 if (BE (ret
!= REG_NOERROR
, 0))
1002 sift_ctx_init (&sctx
, sifted_states
, lim_states
, halt_node
, match_last
);
1003 ret
= sift_states_backward (mctx
, &sctx
);
1004 re_node_set_free (&sctx
.limits
);
1005 if (BE (ret
!= REG_NOERROR
, 0))
1007 if (sifted_states
[0] == NULL
)
1013 re_free (mctx
->state_log
);
1014 mctx
->state_log
= sifted_states
;
1015 sifted_states
= NULL
;
1016 mctx
->last_node
= halt_node
;
1017 mctx
->match_last
= match_last
;
1020 re_free (sifted_states
);
1021 re_free (lim_states
);
1025 /* Acquire an initial state and return it.
1026 We must select appropriate initial state depending on the context,
1027 since initial states may have constraints like "\<", "^", etc.. */
1029 static inline re_dfastate_t
*
1030 __attribute ((always_inline
)) internal_function
1031 acquire_init_state_context (reg_errcode_t
*err
, const re_match_context_t
*mctx
,
1034 const re_dfa_t
*const dfa
= mctx
->dfa
;
1035 if (dfa
->init_state
->has_constraint
)
1037 unsigned int context
;
1038 context
= re_string_context_at (&mctx
->input
, idx
- 1, mctx
->eflags
);
1039 if (IS_WORD_CONTEXT (context
))
1040 return dfa
->init_state_word
;
1041 else if (IS_ORDINARY_CONTEXT (context
))
1042 return dfa
->init_state
;
1043 else if (IS_BEGBUF_CONTEXT (context
) && IS_NEWLINE_CONTEXT (context
))
1044 return dfa
->init_state_begbuf
;
1045 else if (IS_NEWLINE_CONTEXT (context
))
1046 return dfa
->init_state_nl
;
1047 else if (IS_BEGBUF_CONTEXT (context
))
1049 /* It is relatively rare case, then calculate on demand. */
1050 return re_acquire_state_context (err
, dfa
,
1051 dfa
->init_state
->entrance_nodes
,
1055 /* Must not happen? */
1056 return dfa
->init_state
;
1059 return dfa
->init_state
;
1062 /* Check whether the regular expression match input string INPUT or not,
1063 and return the index where the matching end, return -1 if not match,
1064 or return -2 in case of an error.
1065 FL_LONGEST_MATCH means we want the POSIX longest matching.
1066 If P_MATCH_FIRST is not NULL, and the match fails, it is set to the
1067 next place where we may want to try matching.
1068 Note that the matcher assume that the maching starts from the current
1069 index of the buffer. */
1073 check_matching (re_match_context_t
*mctx
, int fl_longest_match
,
1076 const re_dfa_t
*const dfa
= mctx
->dfa
;
1079 int match_last
= -1;
1080 int cur_str_idx
= re_string_cur_idx (&mctx
->input
);
1081 re_dfastate_t
*cur_state
;
1082 int at_init_state
= p_match_first
!= NULL
;
1083 int next_start_idx
= cur_str_idx
;
1086 cur_state
= acquire_init_state_context (&err
, mctx
, cur_str_idx
);
1087 /* An initial state must not be NULL (invalid). */
1088 if (BE (cur_state
== NULL
, 0))
1090 assert (err
== REG_ESPACE
);
1094 if (mctx
->state_log
!= NULL
)
1096 mctx
->state_log
[cur_str_idx
] = cur_state
;
1098 /* Check OP_OPEN_SUBEXP in the initial state in case that we use them
1099 later. E.g. Processing back references. */
1100 if (BE (dfa
->nbackref
, 0))
1103 err
= check_subexp_matching_top (mctx
, &cur_state
->nodes
, 0);
1104 if (BE (err
!= REG_NOERROR
, 0))
1107 if (cur_state
->has_backref
)
1109 err
= transit_state_bkref (mctx
, &cur_state
->nodes
);
1110 if (BE (err
!= REG_NOERROR
, 0))
1116 /* If the RE accepts NULL string. */
1117 if (BE (cur_state
->halt
, 0))
1119 if (!cur_state
->has_constraint
1120 || check_halt_state_context (mctx
, cur_state
, cur_str_idx
))
1122 if (!fl_longest_match
)
1126 match_last
= cur_str_idx
;
1132 while (!re_string_eoi (&mctx
->input
))
1134 re_dfastate_t
*old_state
= cur_state
;
1135 int next_char_idx
= re_string_cur_idx (&mctx
->input
) + 1;
1137 if (BE (next_char_idx
>= mctx
->input
.bufs_len
, 0)
1138 || (BE (next_char_idx
>= mctx
->input
.valid_len
, 0)
1139 && mctx
->input
.valid_len
< mctx
->input
.len
))
1141 err
= extend_buffers (mctx
);
1142 if (BE (err
!= REG_NOERROR
, 0))
1144 assert (err
== REG_ESPACE
);
1149 cur_state
= transit_state (&err
, mctx
, cur_state
);
1150 if (mctx
->state_log
!= NULL
)
1151 cur_state
= merge_state_with_log (&err
, mctx
, cur_state
);
1153 if (cur_state
== NULL
)
1155 /* Reached the invalid state or an error. Try to recover a valid
1156 state using the state log, if available and if we have not
1157 already found a valid (even if not the longest) match. */
1158 if (BE (err
!= REG_NOERROR
, 0))
1161 if (mctx
->state_log
== NULL
1162 || (match
&& !fl_longest_match
)
1163 || (cur_state
= find_recover_state (&err
, mctx
)) == NULL
)
1167 if (BE (at_init_state
, 0))
1169 if (old_state
== cur_state
)
1170 next_start_idx
= next_char_idx
;
1175 if (cur_state
->halt
)
1177 /* Reached a halt state.
1178 Check the halt state can satisfy the current context. */
1179 if (!cur_state
->has_constraint
1180 || check_halt_state_context (mctx
, cur_state
,
1181 re_string_cur_idx (&mctx
->input
)))
1183 /* We found an appropriate halt state. */
1184 match_last
= re_string_cur_idx (&mctx
->input
);
1187 /* We found a match, do not modify match_first below. */
1188 p_match_first
= NULL
;
1189 if (!fl_longest_match
)
1196 *p_match_first
+= next_start_idx
;
1201 /* Check NODE match the current context. */
1205 check_halt_node_context (const re_dfa_t
*dfa
, int node
, unsigned int context
)
1207 re_token_type_t type
= dfa
->nodes
[node
].type
;
1208 unsigned int constraint
= dfa
->nodes
[node
].constraint
;
1209 if (type
!= END_OF_RE
)
1213 if (NOT_SATISFY_NEXT_CONSTRAINT (constraint
, context
))
1218 /* Check the halt state STATE match the current context.
1219 Return 0 if not match, if the node, STATE has, is a halt node and
1220 match the context, return the node. */
1224 check_halt_state_context (const re_match_context_t
*mctx
,
1225 const re_dfastate_t
*state
, int idx
)
1228 unsigned int context
;
1230 assert (state
->halt
);
1232 context
= re_string_context_at (&mctx
->input
, idx
, mctx
->eflags
);
1233 for (i
= 0; i
< state
->nodes
.nelem
; ++i
)
1234 if (check_halt_node_context (mctx
->dfa
, state
->nodes
.elems
[i
], context
))
1235 return state
->nodes
.elems
[i
];
1239 /* Compute the next node to which "NFA" transit from NODE("NFA" is a NFA
1240 corresponding to the DFA).
1241 Return the destination node, and update EPS_VIA_NODES, return -1 in case
1246 proceed_next_node (const re_match_context_t
*mctx
, int nregs
, regmatch_t
*regs
,
1247 int *pidx
, int node
, re_node_set
*eps_via_nodes
,
1248 struct re_fail_stack_t
*fs
)
1250 const re_dfa_t
*const dfa
= mctx
->dfa
;
1252 if (IS_EPSILON_NODE (dfa
->nodes
[node
].type
))
1254 re_node_set
*cur_nodes
= &mctx
->state_log
[*pidx
]->nodes
;
1255 re_node_set
*edests
= &dfa
->edests
[node
];
1257 err
= re_node_set_insert (eps_via_nodes
, node
);
1258 if (BE (err
< 0, 0))
1260 /* Pick up a valid destination, or return -1 if none is found. */
1261 for (dest_node
= -1, i
= 0; i
< edests
->nelem
; ++i
)
1263 int candidate
= edests
->elems
[i
];
1264 if (!re_node_set_contains (cur_nodes
, candidate
))
1266 if (dest_node
== -1)
1267 dest_node
= candidate
;
1271 /* In order to avoid infinite loop like "(a*)*", return the second
1272 epsilon-transition if the first was already considered. */
1273 if (re_node_set_contains (eps_via_nodes
, dest_node
))
1276 /* Otherwise, push the second epsilon-transition on the fail stack. */
1278 && push_fail_stack (fs
, *pidx
, candidate
, nregs
, regs
,
1282 /* We know we are going to exit. */
1291 re_token_type_t type
= dfa
->nodes
[node
].type
;
1293 #ifdef RE_ENABLE_I18N
1294 if (dfa
->nodes
[node
].accept_mb
)
1295 naccepted
= check_node_accept_bytes (dfa
, node
, &mctx
->input
, *pidx
);
1297 #endif /* RE_ENABLE_I18N */
1298 if (type
== OP_BACK_REF
)
1300 int subexp_idx
= dfa
->nodes
[node
].opr
.idx
+ 1;
1301 naccepted
= regs
[subexp_idx
].rm_eo
- regs
[subexp_idx
].rm_so
;
1304 if (regs
[subexp_idx
].rm_so
== -1 || regs
[subexp_idx
].rm_eo
== -1)
1308 char *buf
= (char *) re_string_get_buffer (&mctx
->input
);
1309 if (memcmp (buf
+ regs
[subexp_idx
].rm_so
, buf
+ *pidx
,
1318 err
= re_node_set_insert (eps_via_nodes
, node
);
1319 if (BE (err
< 0, 0))
1321 dest_node
= dfa
->edests
[node
].elems
[0];
1322 if (re_node_set_contains (&mctx
->state_log
[*pidx
]->nodes
,
1329 || check_node_accept (mctx
, dfa
->nodes
+ node
, *pidx
))
1331 int dest_node
= dfa
->nexts
[node
];
1332 *pidx
= (naccepted
== 0) ? *pidx
+ 1 : *pidx
+ naccepted
;
1333 if (fs
&& (*pidx
> mctx
->match_last
|| mctx
->state_log
[*pidx
] == NULL
1334 || !re_node_set_contains (&mctx
->state_log
[*pidx
]->nodes
,
1337 re_node_set_empty (eps_via_nodes
);
1344 static reg_errcode_t
1346 push_fail_stack (struct re_fail_stack_t
*fs
, int str_idx
, int dest_node
,
1347 int nregs
, regmatch_t
*regs
, re_node_set
*eps_via_nodes
)
1350 int num
= fs
->num
++;
1351 if (fs
->num
== fs
->alloc
)
1353 struct re_fail_stack_ent_t
*new_array
;
1354 new_array
= realloc (fs
->stack
, (sizeof (struct re_fail_stack_ent_t
)
1356 if (new_array
== NULL
)
1359 fs
->stack
= new_array
;
1361 fs
->stack
[num
].idx
= str_idx
;
1362 fs
->stack
[num
].node
= dest_node
;
1363 fs
->stack
[num
].regs
= re_malloc (regmatch_t
, nregs
);
1364 if (fs
->stack
[num
].regs
== NULL
)
1366 memcpy (fs
->stack
[num
].regs
, regs
, sizeof (regmatch_t
) * nregs
);
1367 err
= re_node_set_init_copy (&fs
->stack
[num
].eps_via_nodes
, eps_via_nodes
);
1373 pop_fail_stack (struct re_fail_stack_t
*fs
, int *pidx
, int nregs
,
1374 regmatch_t
*regs
, re_node_set
*eps_via_nodes
)
1376 int num
= --fs
->num
;
1378 *pidx
= fs
->stack
[num
].idx
;
1379 memcpy (regs
, fs
->stack
[num
].regs
, sizeof (regmatch_t
) * nregs
);
1380 re_node_set_free (eps_via_nodes
);
1381 re_free (fs
->stack
[num
].regs
);
1382 *eps_via_nodes
= fs
->stack
[num
].eps_via_nodes
;
1383 return fs
->stack
[num
].node
;
1386 /* Set the positions where the subexpressions are starts/ends to registers
1388 Note: We assume that pmatch[0] is already set, and
1389 pmatch[i].rm_so == pmatch[i].rm_eo == -1 for 0 < i < nmatch. */
1391 static reg_errcode_t
1393 set_regs (const regex_t
*preg
, const re_match_context_t
*mctx
, size_t nmatch
,
1394 regmatch_t
*pmatch
, int fl_backtrack
)
1396 const re_dfa_t
*dfa
= (const re_dfa_t
*) preg
->buffer
;
1398 re_node_set eps_via_nodes
;
1399 struct re_fail_stack_t
*fs
;
1400 struct re_fail_stack_t fs_body
= { 0, 2, NULL
};
1401 regmatch_t
*prev_idx_match
;
1402 int prev_idx_match_malloced
= 0;
1405 assert (nmatch
> 1);
1406 assert (mctx
->state_log
!= NULL
);
1411 fs
->stack
= re_malloc (struct re_fail_stack_ent_t
, fs
->alloc
);
1412 if (fs
->stack
== NULL
)
1418 cur_node
= dfa
->init_node
;
1419 re_node_set_init_empty (&eps_via_nodes
);
1421 if (__libc_use_alloca (nmatch
* sizeof (regmatch_t
)))
1422 prev_idx_match
= (regmatch_t
*) alloca (nmatch
* sizeof (regmatch_t
));
1425 prev_idx_match
= re_malloc (regmatch_t
, nmatch
);
1426 if (prev_idx_match
== NULL
)
1428 free_fail_stack_return (fs
);
1431 prev_idx_match_malloced
= 1;
1433 memcpy (prev_idx_match
, pmatch
, sizeof (regmatch_t
) * nmatch
);
1435 for (idx
= pmatch
[0].rm_so
; idx
<= pmatch
[0].rm_eo
;)
1437 update_regs (dfa
, pmatch
, prev_idx_match
, cur_node
, idx
, nmatch
);
1439 if (idx
== pmatch
[0].rm_eo
&& cur_node
== mctx
->last_node
)
1444 for (reg_idx
= 0; reg_idx
< nmatch
; ++reg_idx
)
1445 if (pmatch
[reg_idx
].rm_so
> -1 && pmatch
[reg_idx
].rm_eo
== -1)
1447 if (reg_idx
== nmatch
)
1449 re_node_set_free (&eps_via_nodes
);
1450 if (prev_idx_match_malloced
)
1451 re_free (prev_idx_match
);
1452 return free_fail_stack_return (fs
);
1454 cur_node
= pop_fail_stack (fs
, &idx
, nmatch
, pmatch
,
1459 re_node_set_free (&eps_via_nodes
);
1460 if (prev_idx_match_malloced
)
1461 re_free (prev_idx_match
);
1466 /* Proceed to next node. */
1467 cur_node
= proceed_next_node (mctx
, nmatch
, pmatch
, &idx
, cur_node
,
1468 &eps_via_nodes
, fs
);
1470 if (BE (cur_node
< 0, 0))
1472 if (BE (cur_node
== -2, 0))
1474 re_node_set_free (&eps_via_nodes
);
1475 if (prev_idx_match_malloced
)
1476 re_free (prev_idx_match
);
1477 free_fail_stack_return (fs
);
1481 cur_node
= pop_fail_stack (fs
, &idx
, nmatch
, pmatch
,
1485 re_node_set_free (&eps_via_nodes
);
1486 if (prev_idx_match_malloced
)
1487 re_free (prev_idx_match
);
1492 re_node_set_free (&eps_via_nodes
);
1493 if (prev_idx_match_malloced
)
1494 re_free (prev_idx_match
);
1495 return free_fail_stack_return (fs
);
1498 static reg_errcode_t
1500 free_fail_stack_return (struct re_fail_stack_t
*fs
)
1505 for (fs_idx
= 0; fs_idx
< fs
->num
; ++fs_idx
)
1507 re_node_set_free (&fs
->stack
[fs_idx
].eps_via_nodes
);
1508 re_free (fs
->stack
[fs_idx
].regs
);
1510 re_free (fs
->stack
);
1517 update_regs (const re_dfa_t
*dfa
, regmatch_t
*pmatch
,
1518 regmatch_t
*prev_idx_match
, int cur_node
, int cur_idx
, int nmatch
)
1520 int type
= dfa
->nodes
[cur_node
].type
;
1521 if (type
== OP_OPEN_SUBEXP
)
1523 int reg_num
= dfa
->nodes
[cur_node
].opr
.idx
+ 1;
1525 /* We are at the first node of this sub expression. */
1526 if (reg_num
< nmatch
)
1528 pmatch
[reg_num
].rm_so
= cur_idx
;
1529 pmatch
[reg_num
].rm_eo
= -1;
1532 else if (type
== OP_CLOSE_SUBEXP
)
1534 int reg_num
= dfa
->nodes
[cur_node
].opr
.idx
+ 1;
1535 if (reg_num
< nmatch
)
1537 /* We are at the last node of this sub expression. */
1538 if (pmatch
[reg_num
].rm_so
< cur_idx
)
1540 pmatch
[reg_num
].rm_eo
= cur_idx
;
1541 /* This is a non-empty match or we are not inside an optional
1542 subexpression. Accept this right away. */
1543 memcpy (prev_idx_match
, pmatch
, sizeof (regmatch_t
) * nmatch
);
1547 if (dfa
->nodes
[cur_node
].opt_subexp
1548 && prev_idx_match
[reg_num
].rm_so
!= -1)
1549 /* We transited through an empty match for an optional
1550 subexpression, like (a?)*, and this is not the subexp's
1551 first match. Copy back the old content of the registers
1552 so that matches of an inner subexpression are undone as
1553 well, like in ((a?))*. */
1554 memcpy (pmatch
, prev_idx_match
, sizeof (regmatch_t
) * nmatch
);
1556 /* We completed a subexpression, but it may be part of
1557 an optional one, so do not update PREV_IDX_MATCH. */
1558 pmatch
[reg_num
].rm_eo
= cur_idx
;
1564 /* This function checks the STATE_LOG from the SCTX->last_str_idx to 0
1565 and sift the nodes in each states according to the following rules.
1566 Updated state_log will be wrote to STATE_LOG.
1568 Rules: We throw away the Node `a' in the STATE_LOG[STR_IDX] if...
1569 1. When STR_IDX == MATCH_LAST(the last index in the state_log):
1570 If `a' isn't the LAST_NODE and `a' can't epsilon transit to
1571 the LAST_NODE, we throw away the node `a'.
1572 2. When 0 <= STR_IDX < MATCH_LAST and `a' accepts
1573 string `s' and transit to `b':
1574 i. If 'b' isn't in the STATE_LOG[STR_IDX+strlen('s')], we throw
1576 ii. If 'b' is in the STATE_LOG[STR_IDX+strlen('s')] but 'b' is
1577 thrown away, we throw away the node `a'.
1578 3. When 0 <= STR_IDX < MATCH_LAST and 'a' epsilon transit to 'b':
1579 i. If 'b' isn't in the STATE_LOG[STR_IDX], we throw away the
1581 ii. If 'b' is in the STATE_LOG[STR_IDX] but 'b' is thrown away,
1582 we throw away the node `a'. */
1584 #define STATE_NODE_CONTAINS(state,node) \
1585 ((state) != NULL && re_node_set_contains (&(state)->nodes, node))
1587 static reg_errcode_t
1589 sift_states_backward (const re_match_context_t
*mctx
, re_sift_context_t
*sctx
)
1593 int str_idx
= sctx
->last_str_idx
;
1594 re_node_set cur_dest
;
1597 assert (mctx
->state_log
!= NULL
&& mctx
->state_log
[str_idx
] != NULL
);
1600 /* Build sifted state_log[str_idx]. It has the nodes which can epsilon
1601 transit to the last_node and the last_node itself. */
1602 err
= re_node_set_init_1 (&cur_dest
, sctx
->last_node
);
1603 if (BE (err
!= REG_NOERROR
, 0))
1605 err
= update_cur_sifted_state (mctx
, sctx
, str_idx
, &cur_dest
);
1606 if (BE (err
!= REG_NOERROR
, 0))
1609 /* Then check each states in the state_log. */
1612 /* Update counters. */
1613 null_cnt
= (sctx
->sifted_states
[str_idx
] == NULL
) ? null_cnt
+ 1 : 0;
1614 if (null_cnt
> mctx
->max_mb_elem_len
)
1616 memset (sctx
->sifted_states
, '\0',
1617 sizeof (re_dfastate_t
*) * str_idx
);
1618 re_node_set_free (&cur_dest
);
1621 re_node_set_empty (&cur_dest
);
1624 if (mctx
->state_log
[str_idx
])
1626 err
= build_sifted_states (mctx
, sctx
, str_idx
, &cur_dest
);
1627 if (BE (err
!= REG_NOERROR
, 0))
1631 /* Add all the nodes which satisfy the following conditions:
1632 - It can epsilon transit to a node in CUR_DEST.
1634 And update state_log. */
1635 err
= update_cur_sifted_state (mctx
, sctx
, str_idx
, &cur_dest
);
1636 if (BE (err
!= REG_NOERROR
, 0))
1641 re_node_set_free (&cur_dest
);
1645 static reg_errcode_t
1647 build_sifted_states (const re_match_context_t
*mctx
, re_sift_context_t
*sctx
,
1648 int str_idx
, re_node_set
*cur_dest
)
1650 const re_dfa_t
*const dfa
= mctx
->dfa
;
1651 const re_node_set
*cur_src
= &mctx
->state_log
[str_idx
]->non_eps_nodes
;
1654 /* Then build the next sifted state.
1655 We build the next sifted state on `cur_dest', and update
1656 `sifted_states[str_idx]' with `cur_dest'.
1658 `cur_dest' is the sifted state from `state_log[str_idx + 1]'.
1659 `cur_src' points the node_set of the old `state_log[str_idx]'
1660 (with the epsilon nodes pre-filtered out). */
1661 for (i
= 0; i
< cur_src
->nelem
; i
++)
1663 int prev_node
= cur_src
->elems
[i
];
1668 re_token_type_t type
= dfa
->nodes
[prev_node
].type
;
1669 assert (!IS_EPSILON_NODE (type
));
1671 #ifdef RE_ENABLE_I18N
1672 /* If the node may accept `multi byte'. */
1673 if (dfa
->nodes
[prev_node
].accept_mb
)
1674 naccepted
= sift_states_iter_mb (mctx
, sctx
, prev_node
,
1675 str_idx
, sctx
->last_str_idx
);
1676 #endif /* RE_ENABLE_I18N */
1678 /* We don't check backreferences here.
1679 See update_cur_sifted_state(). */
1681 && check_node_accept (mctx
, dfa
->nodes
+ prev_node
, str_idx
)
1682 && STATE_NODE_CONTAINS (sctx
->sifted_states
[str_idx
+ 1],
1683 dfa
->nexts
[prev_node
]))
1689 if (sctx
->limits
.nelem
)
1691 int to_idx
= str_idx
+ naccepted
;
1692 if (check_dst_limits (mctx
, &sctx
->limits
,
1693 dfa
->nexts
[prev_node
], to_idx
,
1694 prev_node
, str_idx
))
1697 ret
= re_node_set_insert (cur_dest
, prev_node
);
1698 if (BE (ret
== -1, 0))
1705 /* Helper functions. */
1707 static reg_errcode_t
1709 clean_state_log_if_needed (re_match_context_t
*mctx
, int next_state_log_idx
)
1711 int top
= mctx
->state_log_top
;
1713 if (next_state_log_idx
>= mctx
->input
.bufs_len
1714 || (next_state_log_idx
>= mctx
->input
.valid_len
1715 && mctx
->input
.valid_len
< mctx
->input
.len
))
1718 err
= extend_buffers (mctx
);
1719 if (BE (err
!= REG_NOERROR
, 0))
1723 if (top
< next_state_log_idx
)
1725 memset (mctx
->state_log
+ top
+ 1, '\0',
1726 sizeof (re_dfastate_t
*) * (next_state_log_idx
- top
));
1727 mctx
->state_log_top
= next_state_log_idx
;
1732 static reg_errcode_t
1734 merge_state_array (const re_dfa_t
*dfa
, re_dfastate_t
**dst
,
1735 re_dfastate_t
**src
, int num
)
1739 for (st_idx
= 0; st_idx
< num
; ++st_idx
)
1741 if (dst
[st_idx
] == NULL
)
1742 dst
[st_idx
] = src
[st_idx
];
1743 else if (src
[st_idx
] != NULL
)
1745 re_node_set merged_set
;
1746 err
= re_node_set_init_union (&merged_set
, &dst
[st_idx
]->nodes
,
1747 &src
[st_idx
]->nodes
);
1748 if (BE (err
!= REG_NOERROR
, 0))
1750 dst
[st_idx
] = re_acquire_state (&err
, dfa
, &merged_set
);
1751 re_node_set_free (&merged_set
);
1752 if (BE (err
!= REG_NOERROR
, 0))
1759 static reg_errcode_t
1761 update_cur_sifted_state (const re_match_context_t
*mctx
,
1762 re_sift_context_t
*sctx
, int str_idx
,
1763 re_node_set
*dest_nodes
)
1765 const re_dfa_t
*const dfa
= mctx
->dfa
;
1766 reg_errcode_t err
= REG_NOERROR
;
1767 const re_node_set
*candidates
;
1768 candidates
= ((mctx
->state_log
[str_idx
] == NULL
) ? NULL
1769 : &mctx
->state_log
[str_idx
]->nodes
);
1771 if (dest_nodes
->nelem
== 0)
1772 sctx
->sifted_states
[str_idx
] = NULL
;
1777 /* At first, add the nodes which can epsilon transit to a node in
1779 err
= add_epsilon_src_nodes (dfa
, dest_nodes
, candidates
);
1780 if (BE (err
!= REG_NOERROR
, 0))
1783 /* Then, check the limitations in the current sift_context. */
1784 if (sctx
->limits
.nelem
)
1786 err
= check_subexp_limits (dfa
, dest_nodes
, candidates
, &sctx
->limits
,
1787 mctx
->bkref_ents
, str_idx
);
1788 if (BE (err
!= REG_NOERROR
, 0))
1793 sctx
->sifted_states
[str_idx
] = re_acquire_state (&err
, dfa
, dest_nodes
);
1794 if (BE (err
!= REG_NOERROR
, 0))
1798 if (candidates
&& mctx
->state_log
[str_idx
]->has_backref
)
1800 err
= sift_states_bkref (mctx
, sctx
, str_idx
, candidates
);
1801 if (BE (err
!= REG_NOERROR
, 0))
1807 static reg_errcode_t
1809 add_epsilon_src_nodes (const re_dfa_t
*dfa
, re_node_set
*dest_nodes
,
1810 const re_node_set
*candidates
)
1812 reg_errcode_t err
= REG_NOERROR
;
1815 re_dfastate_t
*state
= re_acquire_state (&err
, dfa
, dest_nodes
);
1816 if (BE (err
!= REG_NOERROR
, 0))
1819 if (!state
->inveclosure
.alloc
)
1821 err
= re_node_set_alloc (&state
->inveclosure
, dest_nodes
->nelem
);
1822 if (BE (err
!= REG_NOERROR
, 0))
1824 for (i
= 0; i
< dest_nodes
->nelem
; i
++)
1825 re_node_set_merge (&state
->inveclosure
,
1826 dfa
->inveclosures
+ dest_nodes
->elems
[i
]);
1828 return re_node_set_add_intersect (dest_nodes
, candidates
,
1829 &state
->inveclosure
);
1832 static reg_errcode_t
1834 sub_epsilon_src_nodes (const re_dfa_t
*dfa
, int node
, re_node_set
*dest_nodes
,
1835 const re_node_set
*candidates
)
1839 re_node_set
*inv_eclosure
= dfa
->inveclosures
+ node
;
1840 re_node_set except_nodes
;
1841 re_node_set_init_empty (&except_nodes
);
1842 for (ecl_idx
= 0; ecl_idx
< inv_eclosure
->nelem
; ++ecl_idx
)
1844 int cur_node
= inv_eclosure
->elems
[ecl_idx
];
1845 if (cur_node
== node
)
1847 if (IS_EPSILON_NODE (dfa
->nodes
[cur_node
].type
))
1849 int edst1
= dfa
->edests
[cur_node
].elems
[0];
1850 int edst2
= ((dfa
->edests
[cur_node
].nelem
> 1)
1851 ? dfa
->edests
[cur_node
].elems
[1] : -1);
1852 if ((!re_node_set_contains (inv_eclosure
, edst1
)
1853 && re_node_set_contains (dest_nodes
, edst1
))
1855 && !re_node_set_contains (inv_eclosure
, edst2
)
1856 && re_node_set_contains (dest_nodes
, edst2
)))
1858 err
= re_node_set_add_intersect (&except_nodes
, candidates
,
1859 dfa
->inveclosures
+ cur_node
);
1860 if (BE (err
!= REG_NOERROR
, 0))
1862 re_node_set_free (&except_nodes
);
1868 for (ecl_idx
= 0; ecl_idx
< inv_eclosure
->nelem
; ++ecl_idx
)
1870 int cur_node
= inv_eclosure
->elems
[ecl_idx
];
1871 if (!re_node_set_contains (&except_nodes
, cur_node
))
1873 int idx
= re_node_set_contains (dest_nodes
, cur_node
) - 1;
1874 re_node_set_remove_at (dest_nodes
, idx
);
1877 re_node_set_free (&except_nodes
);
1883 check_dst_limits (const re_match_context_t
*mctx
, re_node_set
*limits
,
1884 int dst_node
, int dst_idx
, int src_node
, int src_idx
)
1886 const re_dfa_t
*const dfa
= mctx
->dfa
;
1887 int lim_idx
, src_pos
, dst_pos
;
1889 int dst_bkref_idx
= search_cur_bkref_entry (mctx
, dst_idx
);
1890 int src_bkref_idx
= search_cur_bkref_entry (mctx
, src_idx
);
1891 for (lim_idx
= 0; lim_idx
< limits
->nelem
; ++lim_idx
)
1894 struct re_backref_cache_entry
*ent
;
1895 ent
= mctx
->bkref_ents
+ limits
->elems
[lim_idx
];
1896 subexp_idx
= dfa
->nodes
[ent
->node
].opr
.idx
;
1898 dst_pos
= check_dst_limits_calc_pos (mctx
, limits
->elems
[lim_idx
],
1899 subexp_idx
, dst_node
, dst_idx
,
1901 src_pos
= check_dst_limits_calc_pos (mctx
, limits
->elems
[lim_idx
],
1902 subexp_idx
, src_node
, src_idx
,
1906 <src> <dst> ( <subexp> )
1907 ( <subexp> ) <src> <dst>
1908 ( <subexp1> <src> <subexp2> <dst> <subexp3> ) */
1909 if (src_pos
== dst_pos
)
1910 continue; /* This is unrelated limitation. */
1919 check_dst_limits_calc_pos_1 (const re_match_context_t
*mctx
, int boundaries
,
1920 int subexp_idx
, int from_node
, int bkref_idx
)
1922 const re_dfa_t
*const dfa
= mctx
->dfa
;
1923 const re_node_set
*eclosures
= dfa
->eclosures
+ from_node
;
1926 /* Else, we are on the boundary: examine the nodes on the epsilon
1928 for (node_idx
= 0; node_idx
< eclosures
->nelem
; ++node_idx
)
1930 int node
= eclosures
->elems
[node_idx
];
1931 switch (dfa
->nodes
[node
].type
)
1934 if (bkref_idx
!= -1)
1936 struct re_backref_cache_entry
*ent
= mctx
->bkref_ents
+ bkref_idx
;
1941 if (ent
->node
!= node
)
1944 if (subexp_idx
< BITSET_WORD_BITS
1945 && !(ent
->eps_reachable_subexps_map
1946 & ((bitset_word_t
) 1 << subexp_idx
)))
1949 /* Recurse trying to reach the OP_OPEN_SUBEXP and
1950 OP_CLOSE_SUBEXP cases below. But, if the
1951 destination node is the same node as the source
1952 node, don't recurse because it would cause an
1953 infinite loop: a regex that exhibits this behavior
1955 dst
= dfa
->edests
[node
].elems
[0];
1956 if (dst
== from_node
)
1960 else /* if (boundaries & 2) */
1965 check_dst_limits_calc_pos_1 (mctx
, boundaries
, subexp_idx
,
1967 if (cpos
== -1 /* && (boundaries & 1) */)
1969 if (cpos
== 0 && (boundaries
& 2))
1972 if (subexp_idx
< BITSET_WORD_BITS
)
1973 ent
->eps_reachable_subexps_map
1974 &= ~((bitset_word_t
) 1 << subexp_idx
);
1976 while (ent
++->more
);
1980 case OP_OPEN_SUBEXP
:
1981 if ((boundaries
& 1) && subexp_idx
== dfa
->nodes
[node
].opr
.idx
)
1985 case OP_CLOSE_SUBEXP
:
1986 if ((boundaries
& 2) && subexp_idx
== dfa
->nodes
[node
].opr
.idx
)
1995 return (boundaries
& 2) ? 1 : 0;
2000 check_dst_limits_calc_pos (const re_match_context_t
*mctx
, int limit
,
2001 int subexp_idx
, int from_node
, int str_idx
,
2004 struct re_backref_cache_entry
*lim
= mctx
->bkref_ents
+ limit
;
2007 /* If we are outside the range of the subexpression, return -1 or 1. */
2008 if (str_idx
< lim
->subexp_from
)
2011 if (lim
->subexp_to
< str_idx
)
2014 /* If we are within the subexpression, return 0. */
2015 boundaries
= (str_idx
== lim
->subexp_from
);
2016 boundaries
|= (str_idx
== lim
->subexp_to
) << 1;
2017 if (boundaries
== 0)
2020 /* Else, examine epsilon closure. */
2021 return check_dst_limits_calc_pos_1 (mctx
, boundaries
, subexp_idx
,
2022 from_node
, bkref_idx
);
2025 /* Check the limitations of sub expressions LIMITS, and remove the nodes
2026 which are against limitations from DEST_NODES. */
2028 static reg_errcode_t
2030 check_subexp_limits (const re_dfa_t
*dfa
, re_node_set
*dest_nodes
,
2031 const re_node_set
*candidates
, re_node_set
*limits
,
2032 struct re_backref_cache_entry
*bkref_ents
, int str_idx
)
2035 int node_idx
, lim_idx
;
2037 for (lim_idx
= 0; lim_idx
< limits
->nelem
; ++lim_idx
)
2040 struct re_backref_cache_entry
*ent
;
2041 ent
= bkref_ents
+ limits
->elems
[lim_idx
];
2043 if (str_idx
<= ent
->subexp_from
|| ent
->str_idx
< str_idx
)
2044 continue; /* This is unrelated limitation. */
2046 subexp_idx
= dfa
->nodes
[ent
->node
].opr
.idx
;
2047 if (ent
->subexp_to
== str_idx
)
2051 for (node_idx
= 0; node_idx
< dest_nodes
->nelem
; ++node_idx
)
2053 int node
= dest_nodes
->elems
[node_idx
];
2054 re_token_type_t type
= dfa
->nodes
[node
].type
;
2055 if (type
== OP_OPEN_SUBEXP
2056 && subexp_idx
== dfa
->nodes
[node
].opr
.idx
)
2058 else if (type
== OP_CLOSE_SUBEXP
2059 && subexp_idx
== dfa
->nodes
[node
].opr
.idx
)
2063 /* Check the limitation of the open subexpression. */
2064 /* Note that (ent->subexp_to = str_idx != ent->subexp_from). */
2067 err
= sub_epsilon_src_nodes (dfa
, ops_node
, dest_nodes
,
2069 if (BE (err
!= REG_NOERROR
, 0))
2073 /* Check the limitation of the close subexpression. */
2075 for (node_idx
= 0; node_idx
< dest_nodes
->nelem
; ++node_idx
)
2077 int node
= dest_nodes
->elems
[node_idx
];
2078 if (!re_node_set_contains (dfa
->inveclosures
+ node
,
2080 && !re_node_set_contains (dfa
->eclosures
+ node
,
2083 /* It is against this limitation.
2084 Remove it form the current sifted state. */
2085 err
= sub_epsilon_src_nodes (dfa
, node
, dest_nodes
,
2087 if (BE (err
!= REG_NOERROR
, 0))
2093 else /* (ent->subexp_to != str_idx) */
2095 for (node_idx
= 0; node_idx
< dest_nodes
->nelem
; ++node_idx
)
2097 int node
= dest_nodes
->elems
[node_idx
];
2098 re_token_type_t type
= dfa
->nodes
[node
].type
;
2099 if (type
== OP_CLOSE_SUBEXP
|| type
== OP_OPEN_SUBEXP
)
2101 if (subexp_idx
!= dfa
->nodes
[node
].opr
.idx
)
2103 /* It is against this limitation.
2104 Remove it form the current sifted state. */
2105 err
= sub_epsilon_src_nodes (dfa
, node
, dest_nodes
,
2107 if (BE (err
!= REG_NOERROR
, 0))
2116 static reg_errcode_t
2118 sift_states_bkref (const re_match_context_t
*mctx
, re_sift_context_t
*sctx
,
2119 int str_idx
, const re_node_set
*candidates
)
2121 const re_dfa_t
*const dfa
= mctx
->dfa
;
2124 re_sift_context_t local_sctx
;
2125 int first_idx
= search_cur_bkref_entry (mctx
, str_idx
);
2127 if (first_idx
== -1)
2130 local_sctx
.sifted_states
= NULL
; /* Mark that it hasn't been initialized. */
2132 for (node_idx
= 0; node_idx
< candidates
->nelem
; ++node_idx
)
2135 re_token_type_t type
;
2136 struct re_backref_cache_entry
*entry
;
2137 node
= candidates
->elems
[node_idx
];
2138 type
= dfa
->nodes
[node
].type
;
2139 /* Avoid infinite loop for the REs like "()\1+". */
2140 if (node
== sctx
->last_node
&& str_idx
== sctx
->last_str_idx
)
2142 if (type
!= OP_BACK_REF
)
2145 entry
= mctx
->bkref_ents
+ first_idx
;
2146 enabled_idx
= first_idx
;
2153 re_dfastate_t
*cur_state
;
2155 if (entry
->node
!= node
)
2157 subexp_len
= entry
->subexp_to
- entry
->subexp_from
;
2158 to_idx
= str_idx
+ subexp_len
;
2159 dst_node
= (subexp_len
? dfa
->nexts
[node
]
2160 : dfa
->edests
[node
].elems
[0]);
2162 if (to_idx
> sctx
->last_str_idx
2163 || sctx
->sifted_states
[to_idx
] == NULL
2164 || !STATE_NODE_CONTAINS (sctx
->sifted_states
[to_idx
], dst_node
)
2165 || check_dst_limits (mctx
, &sctx
->limits
, node
,
2166 str_idx
, dst_node
, to_idx
))
2169 if (local_sctx
.sifted_states
== NULL
)
2172 err
= re_node_set_init_copy (&local_sctx
.limits
, &sctx
->limits
);
2173 if (BE (err
!= REG_NOERROR
, 0))
2176 local_sctx
.last_node
= node
;
2177 local_sctx
.last_str_idx
= str_idx
;
2178 ret
= re_node_set_insert (&local_sctx
.limits
, enabled_idx
);
2179 if (BE (ret
< 0, 0))
2184 cur_state
= local_sctx
.sifted_states
[str_idx
];
2185 err
= sift_states_backward (mctx
, &local_sctx
);
2186 if (BE (err
!= REG_NOERROR
, 0))
2188 if (sctx
->limited_states
!= NULL
)
2190 err
= merge_state_array (dfa
, sctx
->limited_states
,
2191 local_sctx
.sifted_states
,
2193 if (BE (err
!= REG_NOERROR
, 0))
2196 local_sctx
.sifted_states
[str_idx
] = cur_state
;
2197 re_node_set_remove (&local_sctx
.limits
, enabled_idx
);
2199 /* mctx->bkref_ents may have changed, reload the pointer. */
2200 entry
= mctx
->bkref_ents
+ enabled_idx
;
2202 while (enabled_idx
++, entry
++->more
);
2206 if (local_sctx
.sifted_states
!= NULL
)
2208 re_node_set_free (&local_sctx
.limits
);
2215 #ifdef RE_ENABLE_I18N
2218 sift_states_iter_mb (const re_match_context_t
*mctx
, re_sift_context_t
*sctx
,
2219 int node_idx
, int str_idx
, int max_str_idx
)
2221 const re_dfa_t
*const dfa
= mctx
->dfa
;
2223 /* Check the node can accept `multi byte'. */
2224 naccepted
= check_node_accept_bytes (dfa
, node_idx
, &mctx
->input
, str_idx
);
2225 if (naccepted
> 0 && str_idx
+ naccepted
<= max_str_idx
&&
2226 !STATE_NODE_CONTAINS (sctx
->sifted_states
[str_idx
+ naccepted
],
2227 dfa
->nexts
[node_idx
]))
2228 /* The node can't accept the `multi byte', or the
2229 destination was already thrown away, then the node
2230 could't accept the current input `multi byte'. */
2232 /* Otherwise, it is sure that the node could accept
2233 `naccepted' bytes input. */
2236 #endif /* RE_ENABLE_I18N */
2239 /* Functions for state transition. */
2241 /* Return the next state to which the current state STATE will transit by
2242 accepting the current input byte, and update STATE_LOG if necessary.
2243 If STATE can accept a multibyte char/collating element/back reference
2244 update the destination of STATE_LOG. */
2246 static re_dfastate_t
*
2248 transit_state (reg_errcode_t
*err
, re_match_context_t
*mctx
,
2249 re_dfastate_t
*state
)
2251 re_dfastate_t
**trtable
;
2254 #ifdef RE_ENABLE_I18N
2255 /* If the current state can accept multibyte. */
2256 if (BE (state
->accept_mb
, 0))
2258 *err
= transit_state_mb (mctx
, state
);
2259 if (BE (*err
!= REG_NOERROR
, 0))
2262 #endif /* RE_ENABLE_I18N */
2264 /* Then decide the next state with the single byte. */
2267 /* don't use transition table */
2268 return transit_state_sb (err
, mctx
, state
);
2271 /* Use transition table */
2272 ch
= re_string_fetch_byte (&mctx
->input
);
2275 trtable
= state
->trtable
;
2276 if (BE (trtable
!= NULL
, 1))
2279 trtable
= state
->word_trtable
;
2280 if (BE (trtable
!= NULL
, 1))
2282 unsigned int context
;
2284 = re_string_context_at (&mctx
->input
,
2285 re_string_cur_idx (&mctx
->input
) - 1,
2287 if (IS_WORD_CONTEXT (context
))
2288 return trtable
[ch
+ SBC_MAX
];
2293 if (!build_trtable (mctx
->dfa
, state
))
2299 /* Retry, we now have a transition table. */
2303 /* Update the state_log if we need */
2306 merge_state_with_log (reg_errcode_t
*err
, re_match_context_t
*mctx
,
2307 re_dfastate_t
*next_state
)
2309 const re_dfa_t
*const dfa
= mctx
->dfa
;
2310 int cur_idx
= re_string_cur_idx (&mctx
->input
);
2312 if (cur_idx
> mctx
->state_log_top
)
2314 mctx
->state_log
[cur_idx
] = next_state
;
2315 mctx
->state_log_top
= cur_idx
;
2317 else if (mctx
->state_log
[cur_idx
] == 0)
2319 mctx
->state_log
[cur_idx
] = next_state
;
2323 re_dfastate_t
*pstate
;
2324 unsigned int context
;
2325 re_node_set next_nodes
, *log_nodes
, *table_nodes
= NULL
;
2326 /* If (state_log[cur_idx] != 0), it implies that cur_idx is
2327 the destination of a multibyte char/collating element/
2328 back reference. Then the next state is the union set of
2329 these destinations and the results of the transition table. */
2330 pstate
= mctx
->state_log
[cur_idx
];
2331 log_nodes
= pstate
->entrance_nodes
;
2332 if (next_state
!= NULL
)
2334 table_nodes
= next_state
->entrance_nodes
;
2335 *err
= re_node_set_init_union (&next_nodes
, table_nodes
,
2337 if (BE (*err
!= REG_NOERROR
, 0))
2341 next_nodes
= *log_nodes
;
2342 /* Note: We already add the nodes of the initial state,
2343 then we don't need to add them here. */
2345 context
= re_string_context_at (&mctx
->input
,
2346 re_string_cur_idx (&mctx
->input
) - 1,
2348 next_state
= mctx
->state_log
[cur_idx
]
2349 = re_acquire_state_context (err
, dfa
, &next_nodes
, context
);
2350 /* We don't need to check errors here, since the return value of
2351 this function is next_state and ERR is already set. */
2353 if (table_nodes
!= NULL
)
2354 re_node_set_free (&next_nodes
);
2357 if (BE (dfa
->nbackref
, 0) && next_state
!= NULL
)
2359 /* Check OP_OPEN_SUBEXP in the current state in case that we use them
2360 later. We must check them here, since the back references in the
2361 next state might use them. */
2362 *err
= check_subexp_matching_top (mctx
, &next_state
->nodes
,
2364 if (BE (*err
!= REG_NOERROR
, 0))
2367 /* If the next state has back references. */
2368 if (next_state
->has_backref
)
2370 *err
= transit_state_bkref (mctx
, &next_state
->nodes
);
2371 if (BE (*err
!= REG_NOERROR
, 0))
2373 next_state
= mctx
->state_log
[cur_idx
];
2380 /* Skip bytes in the input that correspond to part of a
2381 multi-byte match, then look in the log for a state
2382 from which to restart matching. */
2385 find_recover_state (reg_errcode_t
*err
, re_match_context_t
*mctx
)
2387 re_dfastate_t
*cur_state
;
2390 int max
= mctx
->state_log_top
;
2391 int cur_str_idx
= re_string_cur_idx (&mctx
->input
);
2395 if (++cur_str_idx
> max
)
2397 re_string_skip_bytes (&mctx
->input
, 1);
2399 while (mctx
->state_log
[cur_str_idx
] == NULL
);
2401 cur_state
= merge_state_with_log (err
, mctx
, NULL
);
2403 while (*err
== REG_NOERROR
&& cur_state
== NULL
);
2407 /* Helper functions for transit_state. */
2409 /* From the node set CUR_NODES, pick up the nodes whose types are
2410 OP_OPEN_SUBEXP and which have corresponding back references in the regular
2411 expression. And register them to use them later for evaluating the
2412 correspoding back references. */
2414 static reg_errcode_t
2416 check_subexp_matching_top (re_match_context_t
*mctx
, re_node_set
*cur_nodes
,
2419 const re_dfa_t
*const dfa
= mctx
->dfa
;
2423 /* TODO: This isn't efficient.
2424 Because there might be more than one nodes whose types are
2425 OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all
2428 for (node_idx
= 0; node_idx
< cur_nodes
->nelem
; ++node_idx
)
2430 int node
= cur_nodes
->elems
[node_idx
];
2431 if (dfa
->nodes
[node
].type
== OP_OPEN_SUBEXP
2432 && dfa
->nodes
[node
].opr
.idx
< BITSET_WORD_BITS
2433 && (dfa
->used_bkref_map
2434 & ((bitset_word_t
) 1 << dfa
->nodes
[node
].opr
.idx
)))
2436 err
= match_ctx_add_subtop (mctx
, node
, str_idx
);
2437 if (BE (err
!= REG_NOERROR
, 0))
2445 /* Return the next state to which the current state STATE will transit by
2446 accepting the current input byte. */
2448 static re_dfastate_t
*
2449 transit_state_sb (reg_errcode_t
*err
, re_match_context_t
*mctx
,
2450 re_dfastate_t
*state
)
2452 const re_dfa_t
*const dfa
= mctx
->dfa
;
2453 re_node_set next_nodes
;
2454 re_dfastate_t
*next_state
;
2455 int node_cnt
, cur_str_idx
= re_string_cur_idx (&mctx
->input
);
2456 unsigned int context
;
2458 *err
= re_node_set_alloc (&next_nodes
, state
->nodes
.nelem
+ 1);
2459 if (BE (*err
!= REG_NOERROR
, 0))
2461 for (node_cnt
= 0; node_cnt
< state
->nodes
.nelem
; ++node_cnt
)
2463 int cur_node
= state
->nodes
.elems
[node_cnt
];
2464 if (check_node_accept (mctx
, dfa
->nodes
+ cur_node
, cur_str_idx
))
2466 *err
= re_node_set_merge (&next_nodes
,
2467 dfa
->eclosures
+ dfa
->nexts
[cur_node
]);
2468 if (BE (*err
!= REG_NOERROR
, 0))
2470 re_node_set_free (&next_nodes
);
2475 context
= re_string_context_at (&mctx
->input
, cur_str_idx
, mctx
->eflags
);
2476 next_state
= re_acquire_state_context (err
, dfa
, &next_nodes
, context
);
2477 /* We don't need to check errors here, since the return value of
2478 this function is next_state and ERR is already set. */
2480 re_node_set_free (&next_nodes
);
2481 re_string_skip_bytes (&mctx
->input
, 1);
2486 #ifdef RE_ENABLE_I18N
2487 static reg_errcode_t
2489 transit_state_mb (re_match_context_t
*mctx
, re_dfastate_t
*pstate
)
2491 const re_dfa_t
*const dfa
= mctx
->dfa
;
2495 for (i
= 0; i
< pstate
->nodes
.nelem
; ++i
)
2497 re_node_set dest_nodes
, *new_nodes
;
2498 int cur_node_idx
= pstate
->nodes
.elems
[i
];
2499 int naccepted
, dest_idx
;
2500 unsigned int context
;
2501 re_dfastate_t
*dest_state
;
2503 if (!dfa
->nodes
[cur_node_idx
].accept_mb
)
2506 if (dfa
->nodes
[cur_node_idx
].constraint
)
2508 context
= re_string_context_at (&mctx
->input
,
2509 re_string_cur_idx (&mctx
->input
),
2511 if (NOT_SATISFY_NEXT_CONSTRAINT (dfa
->nodes
[cur_node_idx
].constraint
,
2516 /* How many bytes the node can accept? */
2517 naccepted
= check_node_accept_bytes (dfa
, cur_node_idx
, &mctx
->input
,
2518 re_string_cur_idx (&mctx
->input
));
2522 /* The node can accepts `naccepted' bytes. */
2523 dest_idx
= re_string_cur_idx (&mctx
->input
) + naccepted
;
2524 mctx
->max_mb_elem_len
= ((mctx
->max_mb_elem_len
< naccepted
) ? naccepted
2525 : mctx
->max_mb_elem_len
);
2526 err
= clean_state_log_if_needed (mctx
, dest_idx
);
2527 if (BE (err
!= REG_NOERROR
, 0))
2530 assert (dfa
->nexts
[cur_node_idx
] != -1);
2532 new_nodes
= dfa
->eclosures
+ dfa
->nexts
[cur_node_idx
];
2534 dest_state
= mctx
->state_log
[dest_idx
];
2535 if (dest_state
== NULL
)
2536 dest_nodes
= *new_nodes
;
2539 err
= re_node_set_init_union (&dest_nodes
,
2540 dest_state
->entrance_nodes
, new_nodes
);
2541 if (BE (err
!= REG_NOERROR
, 0))
2544 context
= re_string_context_at (&mctx
->input
, dest_idx
- 1,
2546 mctx
->state_log
[dest_idx
]
2547 = re_acquire_state_context (&err
, dfa
, &dest_nodes
, context
);
2548 if (dest_state
!= NULL
)
2549 re_node_set_free (&dest_nodes
);
2550 if (BE (mctx
->state_log
[dest_idx
] == NULL
&& err
!= REG_NOERROR
, 0))
2555 #endif /* RE_ENABLE_I18N */
2557 static reg_errcode_t
2559 transit_state_bkref (re_match_context_t
*mctx
, const re_node_set
*nodes
)
2561 const re_dfa_t
*const dfa
= mctx
->dfa
;
2564 int cur_str_idx
= re_string_cur_idx (&mctx
->input
);
2566 for (i
= 0; i
< nodes
->nelem
; ++i
)
2568 int dest_str_idx
, prev_nelem
, bkc_idx
;
2569 int node_idx
= nodes
->elems
[i
];
2570 unsigned int context
;
2571 const re_token_t
*node
= dfa
->nodes
+ node_idx
;
2572 re_node_set
*new_dest_nodes
;
2574 /* Check whether `node' is a backreference or not. */
2575 if (node
->type
!= OP_BACK_REF
)
2578 if (node
->constraint
)
2580 context
= re_string_context_at (&mctx
->input
, cur_str_idx
,
2582 if (NOT_SATISFY_NEXT_CONSTRAINT (node
->constraint
, context
))
2586 /* `node' is a backreference.
2587 Check the substring which the substring matched. */
2588 bkc_idx
= mctx
->nbkref_ents
;
2589 err
= get_subexp (mctx
, node_idx
, cur_str_idx
);
2590 if (BE (err
!= REG_NOERROR
, 0))
2593 /* And add the epsilon closures (which is `new_dest_nodes') of
2594 the backreference to appropriate state_log. */
2596 assert (dfa
->nexts
[node_idx
] != -1);
2598 for (; bkc_idx
< mctx
->nbkref_ents
; ++bkc_idx
)
2601 re_dfastate_t
*dest_state
;
2602 struct re_backref_cache_entry
*bkref_ent
;
2603 bkref_ent
= mctx
->bkref_ents
+ bkc_idx
;
2604 if (bkref_ent
->node
!= node_idx
|| bkref_ent
->str_idx
!= cur_str_idx
)
2606 subexp_len
= bkref_ent
->subexp_to
- bkref_ent
->subexp_from
;
2607 new_dest_nodes
= (subexp_len
== 0
2608 ? dfa
->eclosures
+ dfa
->edests
[node_idx
].elems
[0]
2609 : dfa
->eclosures
+ dfa
->nexts
[node_idx
]);
2610 dest_str_idx
= (cur_str_idx
+ bkref_ent
->subexp_to
2611 - bkref_ent
->subexp_from
);
2612 context
= re_string_context_at (&mctx
->input
, dest_str_idx
- 1,
2614 dest_state
= mctx
->state_log
[dest_str_idx
];
2615 prev_nelem
= ((mctx
->state_log
[cur_str_idx
] == NULL
) ? 0
2616 : mctx
->state_log
[cur_str_idx
]->nodes
.nelem
);
2617 /* Add `new_dest_node' to state_log. */
2618 if (dest_state
== NULL
)
2620 mctx
->state_log
[dest_str_idx
]
2621 = re_acquire_state_context (&err
, dfa
, new_dest_nodes
,
2623 if (BE (mctx
->state_log
[dest_str_idx
] == NULL
2624 && err
!= REG_NOERROR
, 0))
2629 re_node_set dest_nodes
;
2630 err
= re_node_set_init_union (&dest_nodes
,
2631 dest_state
->entrance_nodes
,
2633 if (BE (err
!= REG_NOERROR
, 0))
2635 re_node_set_free (&dest_nodes
);
2638 mctx
->state_log
[dest_str_idx
]
2639 = re_acquire_state_context (&err
, dfa
, &dest_nodes
, context
);
2640 re_node_set_free (&dest_nodes
);
2641 if (BE (mctx
->state_log
[dest_str_idx
] == NULL
2642 && err
!= REG_NOERROR
, 0))
2645 /* We need to check recursively if the backreference can epsilon
2648 && mctx
->state_log
[cur_str_idx
]->nodes
.nelem
> prev_nelem
)
2650 err
= check_subexp_matching_top (mctx
, new_dest_nodes
,
2652 if (BE (err
!= REG_NOERROR
, 0))
2654 err
= transit_state_bkref (mctx
, new_dest_nodes
);
2655 if (BE (err
!= REG_NOERROR
, 0))
2665 /* Enumerate all the candidates which the backreference BKREF_NODE can match
2666 at BKREF_STR_IDX, and register them by match_ctx_add_entry().
2667 Note that we might collect inappropriate candidates here.
2668 However, the cost of checking them strictly here is too high, then we
2669 delay these checking for prune_impossible_nodes(). */
2671 static reg_errcode_t
2673 get_subexp (re_match_context_t
*mctx
, int bkref_node
, int bkref_str_idx
)
2675 const re_dfa_t
*const dfa
= mctx
->dfa
;
2676 int subexp_num
, sub_top_idx
;
2677 const char *buf
= (const char *) re_string_get_buffer (&mctx
->input
);
2678 /* Return if we have already checked BKREF_NODE at BKREF_STR_IDX. */
2679 int cache_idx
= search_cur_bkref_entry (mctx
, bkref_str_idx
);
2680 if (cache_idx
!= -1)
2682 const struct re_backref_cache_entry
*entry
2683 = mctx
->bkref_ents
+ cache_idx
;
2685 if (entry
->node
== bkref_node
)
2686 return REG_NOERROR
; /* We already checked it. */
2687 while (entry
++->more
);
2690 subexp_num
= dfa
->nodes
[bkref_node
].opr
.idx
;
2692 /* For each sub expression */
2693 for (sub_top_idx
= 0; sub_top_idx
< mctx
->nsub_tops
; ++sub_top_idx
)
2696 re_sub_match_top_t
*sub_top
= mctx
->sub_tops
[sub_top_idx
];
2697 re_sub_match_last_t
*sub_last
;
2698 int sub_last_idx
, sl_str
, bkref_str_off
;
2700 if (dfa
->nodes
[sub_top
->node
].opr
.idx
!= subexp_num
)
2701 continue; /* It isn't related. */
2703 sl_str
= sub_top
->str_idx
;
2704 bkref_str_off
= bkref_str_idx
;
2705 /* At first, check the last node of sub expressions we already
2707 for (sub_last_idx
= 0; sub_last_idx
< sub_top
->nlasts
; ++sub_last_idx
)
2710 sub_last
= sub_top
->lasts
[sub_last_idx
];
2711 sl_str_diff
= sub_last
->str_idx
- sl_str
;
2712 /* The matched string by the sub expression match with the substring
2713 at the back reference? */
2714 if (sl_str_diff
> 0)
2716 if (BE (bkref_str_off
+ sl_str_diff
> mctx
->input
.valid_len
, 0))
2718 /* Not enough chars for a successful match. */
2719 if (bkref_str_off
+ sl_str_diff
> mctx
->input
.len
)
2722 err
= clean_state_log_if_needed (mctx
,
2725 if (BE (err
!= REG_NOERROR
, 0))
2727 buf
= (const char *) re_string_get_buffer (&mctx
->input
);
2729 if (memcmp (buf
+ bkref_str_off
, buf
+ sl_str
, sl_str_diff
) != 0)
2730 /* We don't need to search this sub expression any more. */
2733 bkref_str_off
+= sl_str_diff
;
2734 sl_str
+= sl_str_diff
;
2735 err
= get_subexp_sub (mctx
, sub_top
, sub_last
, bkref_node
,
2738 /* Reload buf, since the preceding call might have reallocated
2740 buf
= (const char *) re_string_get_buffer (&mctx
->input
);
2742 if (err
== REG_NOMATCH
)
2744 if (BE (err
!= REG_NOERROR
, 0))
2748 if (sub_last_idx
< sub_top
->nlasts
)
2750 if (sub_last_idx
> 0)
2752 /* Then, search for the other last nodes of the sub expression. */
2753 for (; sl_str
<= bkref_str_idx
; ++sl_str
)
2755 int cls_node
, sl_str_off
;
2756 const re_node_set
*nodes
;
2757 sl_str_off
= sl_str
- sub_top
->str_idx
;
2758 /* The matched string by the sub expression match with the substring
2759 at the back reference? */
2762 if (BE (bkref_str_off
>= mctx
->input
.valid_len
, 0))
2764 /* If we are at the end of the input, we cannot match. */
2765 if (bkref_str_off
>= mctx
->input
.len
)
2768 err
= extend_buffers (mctx
);
2769 if (BE (err
!= REG_NOERROR
, 0))
2772 buf
= (const char *) re_string_get_buffer (&mctx
->input
);
2774 if (buf
[bkref_str_off
++] != buf
[sl_str
- 1])
2775 break; /* We don't need to search this sub expression
2778 if (mctx
->state_log
[sl_str
] == NULL
)
2780 /* Does this state have a ')' of the sub expression? */
2781 nodes
= &mctx
->state_log
[sl_str
]->nodes
;
2782 cls_node
= find_subexp_node (dfa
, nodes
, subexp_num
,
2786 if (sub_top
->path
== NULL
)
2788 sub_top
->path
= calloc (sizeof (state_array_t
),
2789 sl_str
- sub_top
->str_idx
+ 1);
2790 if (sub_top
->path
== NULL
)
2793 /* Can the OP_OPEN_SUBEXP node arrive the OP_CLOSE_SUBEXP node
2794 in the current context? */
2795 err
= check_arrival (mctx
, sub_top
->path
, sub_top
->node
,
2796 sub_top
->str_idx
, cls_node
, sl_str
,
2798 if (err
== REG_NOMATCH
)
2800 if (BE (err
!= REG_NOERROR
, 0))
2802 sub_last
= match_ctx_add_sublast (sub_top
, cls_node
, sl_str
);
2803 if (BE (sub_last
== NULL
, 0))
2805 err
= get_subexp_sub (mctx
, sub_top
, sub_last
, bkref_node
,
2807 if (err
== REG_NOMATCH
)
2814 /* Helper functions for get_subexp(). */
2816 /* Check SUB_LAST can arrive to the back reference BKREF_NODE at BKREF_STR.
2817 If it can arrive, register the sub expression expressed with SUB_TOP
2820 static reg_errcode_t
2822 get_subexp_sub (re_match_context_t
*mctx
, const re_sub_match_top_t
*sub_top
,
2823 re_sub_match_last_t
*sub_last
, int bkref_node
, int bkref_str
)
2827 /* Can the subexpression arrive the back reference? */
2828 err
= check_arrival (mctx
, &sub_last
->path
, sub_last
->node
,
2829 sub_last
->str_idx
, bkref_node
, bkref_str
,
2831 if (err
!= REG_NOERROR
)
2833 err
= match_ctx_add_entry (mctx
, bkref_node
, bkref_str
, sub_top
->str_idx
,
2835 if (BE (err
!= REG_NOERROR
, 0))
2837 to_idx
= bkref_str
+ sub_last
->str_idx
- sub_top
->str_idx
;
2838 return clean_state_log_if_needed (mctx
, to_idx
);
2841 /* Find the first node which is '(' or ')' and whose index is SUBEXP_IDX.
2842 Search '(' if FL_OPEN, or search ')' otherwise.
2843 TODO: This function isn't efficient...
2844 Because there might be more than one nodes whose types are
2845 OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all
2851 find_subexp_node (const re_dfa_t
*dfa
, const re_node_set
*nodes
,
2852 int subexp_idx
, int type
)
2855 for (cls_idx
= 0; cls_idx
< nodes
->nelem
; ++cls_idx
)
2857 int cls_node
= nodes
->elems
[cls_idx
];
2858 const re_token_t
*node
= dfa
->nodes
+ cls_node
;
2859 if (node
->type
== type
2860 && node
->opr
.idx
== subexp_idx
)
2866 /* Check whether the node TOP_NODE at TOP_STR can arrive to the node
2867 LAST_NODE at LAST_STR. We record the path onto PATH since it will be
2869 Return REG_NOERROR if it can arrive, or REG_NOMATCH otherwise. */
2871 static reg_errcode_t
2873 check_arrival (re_match_context_t
*mctx
, state_array_t
*path
, int top_node
,
2874 int top_str
, int last_node
, int last_str
, int type
)
2876 const re_dfa_t
*const dfa
= mctx
->dfa
;
2877 reg_errcode_t err
= REG_NOERROR
;
2878 int subexp_num
, backup_cur_idx
, str_idx
, null_cnt
;
2879 re_dfastate_t
*cur_state
= NULL
;
2880 re_node_set
*cur_nodes
, next_nodes
;
2881 re_dfastate_t
**backup_state_log
;
2882 unsigned int context
;
2884 subexp_num
= dfa
->nodes
[top_node
].opr
.idx
;
2885 /* Extend the buffer if we need. */
2886 if (BE (path
->alloc
< last_str
+ mctx
->max_mb_elem_len
+ 1, 0))
2888 re_dfastate_t
**new_array
;
2889 int old_alloc
= path
->alloc
;
2890 path
->alloc
+= last_str
+ mctx
->max_mb_elem_len
+ 1;
2891 new_array
= re_realloc (path
->array
, re_dfastate_t
*, path
->alloc
);
2892 if (BE (new_array
== NULL
, 0))
2894 path
->alloc
= old_alloc
;
2897 path
->array
= new_array
;
2898 memset (new_array
+ old_alloc
, '\0',
2899 sizeof (re_dfastate_t
*) * (path
->alloc
- old_alloc
));
2902 str_idx
= path
->next_idx
? 0 : top_str
;
2904 /* Temporary modify MCTX. */
2905 backup_state_log
= mctx
->state_log
;
2906 backup_cur_idx
= mctx
->input
.cur_idx
;
2907 mctx
->state_log
= path
->array
;
2908 mctx
->input
.cur_idx
= str_idx
;
2910 /* Setup initial node set. */
2911 context
= re_string_context_at (&mctx
->input
, str_idx
- 1, mctx
->eflags
);
2912 if (str_idx
== top_str
)
2914 err
= re_node_set_init_1 (&next_nodes
, top_node
);
2915 if (BE (err
!= REG_NOERROR
, 0))
2917 err
= check_arrival_expand_ecl (dfa
, &next_nodes
, subexp_num
, type
);
2918 if (BE (err
!= REG_NOERROR
, 0))
2920 re_node_set_free (&next_nodes
);
2926 cur_state
= mctx
->state_log
[str_idx
];
2927 if (cur_state
&& cur_state
->has_backref
)
2929 err
= re_node_set_init_copy (&next_nodes
, &cur_state
->nodes
);
2930 if (BE (err
!= REG_NOERROR
, 0))
2934 re_node_set_init_empty (&next_nodes
);
2936 if (str_idx
== top_str
|| (cur_state
&& cur_state
->has_backref
))
2938 if (next_nodes
.nelem
)
2940 err
= expand_bkref_cache (mctx
, &next_nodes
, str_idx
,
2942 if (BE (err
!= REG_NOERROR
, 0))
2944 re_node_set_free (&next_nodes
);
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
;
2957 for (null_cnt
= 0; str_idx
< last_str
&& null_cnt
<= mctx
->max_mb_elem_len
;)
2959 re_node_set_empty (&next_nodes
);
2960 if (mctx
->state_log
[str_idx
+ 1])
2962 err
= re_node_set_merge (&next_nodes
,
2963 &mctx
->state_log
[str_idx
+ 1]->nodes
);
2964 if (BE (err
!= REG_NOERROR
, 0))
2966 re_node_set_free (&next_nodes
);
2972 err
= check_arrival_add_next_nodes (mctx
, str_idx
,
2973 &cur_state
->non_eps_nodes
,
2975 if (BE (err
!= REG_NOERROR
, 0))
2977 re_node_set_free (&next_nodes
);
2982 if (next_nodes
.nelem
)
2984 err
= check_arrival_expand_ecl (dfa
, &next_nodes
, subexp_num
, type
);
2985 if (BE (err
!= REG_NOERROR
, 0))
2987 re_node_set_free (&next_nodes
);
2990 err
= expand_bkref_cache (mctx
, &next_nodes
, str_idx
,
2992 if (BE (err
!= REG_NOERROR
, 0))
2994 re_node_set_free (&next_nodes
);
2998 context
= re_string_context_at (&mctx
->input
, str_idx
- 1, mctx
->eflags
);
2999 cur_state
= re_acquire_state_context (&err
, dfa
, &next_nodes
, context
);
3000 if (BE (cur_state
== NULL
&& err
!= REG_NOERROR
, 0))
3002 re_node_set_free (&next_nodes
);
3005 mctx
->state_log
[str_idx
] = cur_state
;
3006 null_cnt
= cur_state
== NULL
? null_cnt
+ 1 : 0;
3008 re_node_set_free (&next_nodes
);
3009 cur_nodes
= (mctx
->state_log
[last_str
] == NULL
? NULL
3010 : &mctx
->state_log
[last_str
]->nodes
);
3011 path
->next_idx
= str_idx
;
3014 mctx
->state_log
= backup_state_log
;
3015 mctx
->input
.cur_idx
= backup_cur_idx
;
3017 /* Then check the current node set has the node LAST_NODE. */
3018 if (cur_nodes
!= NULL
&& re_node_set_contains (cur_nodes
, last_node
))
3024 /* Helper functions for check_arrival. */
3026 /* Calculate the destination nodes of CUR_NODES at STR_IDX, and append them
3028 TODO: This function is similar to the functions transit_state*(),
3029 however this function has many additional works.
3030 Can't we unify them? */
3032 static reg_errcode_t
3034 check_arrival_add_next_nodes (re_match_context_t
*mctx
, int str_idx
,
3035 re_node_set
*cur_nodes
, re_node_set
*next_nodes
)
3037 const re_dfa_t
*const dfa
= mctx
->dfa
;
3040 reg_errcode_t err
= REG_NOERROR
;
3041 re_node_set union_set
;
3042 re_node_set_init_empty (&union_set
);
3043 for (cur_idx
= 0; cur_idx
< cur_nodes
->nelem
; ++cur_idx
)
3046 int cur_node
= cur_nodes
->elems
[cur_idx
];
3048 re_token_type_t type
= dfa
->nodes
[cur_node
].type
;
3049 assert (!IS_EPSILON_NODE (type
));
3051 #ifdef RE_ENABLE_I18N
3052 /* If the node may accept `multi byte'. */
3053 if (dfa
->nodes
[cur_node
].accept_mb
)
3055 naccepted
= check_node_accept_bytes (dfa
, cur_node
, &mctx
->input
,
3059 re_dfastate_t
*dest_state
;
3060 int next_node
= dfa
->nexts
[cur_node
];
3061 int next_idx
= str_idx
+ naccepted
;
3062 dest_state
= mctx
->state_log
[next_idx
];
3063 re_node_set_empty (&union_set
);
3066 err
= re_node_set_merge (&union_set
, &dest_state
->nodes
);
3067 if (BE (err
!= REG_NOERROR
, 0))
3069 re_node_set_free (&union_set
);
3073 result
= re_node_set_insert (&union_set
, next_node
);
3074 if (BE (result
< 0, 0))
3076 re_node_set_free (&union_set
);
3079 mctx
->state_log
[next_idx
] = re_acquire_state (&err
, dfa
,
3081 if (BE (mctx
->state_log
[next_idx
] == NULL
3082 && err
!= REG_NOERROR
, 0))
3084 re_node_set_free (&union_set
);
3089 #endif /* RE_ENABLE_I18N */
3091 || check_node_accept (mctx
, dfa
->nodes
+ cur_node
, str_idx
))
3093 result
= re_node_set_insert (next_nodes
, dfa
->nexts
[cur_node
]);
3094 if (BE (result
< 0, 0))
3096 re_node_set_free (&union_set
);
3101 re_node_set_free (&union_set
);
3105 /* For all the nodes in CUR_NODES, add the epsilon closures of them to
3106 CUR_NODES, however exclude the nodes which are:
3107 - inside the sub expression whose number is EX_SUBEXP, if FL_OPEN.
3108 - out of the sub expression whose number is EX_SUBEXP, if !FL_OPEN.
3111 static reg_errcode_t
3113 check_arrival_expand_ecl (const re_dfa_t
*dfa
, re_node_set
*cur_nodes
,
3114 int ex_subexp
, int type
)
3117 int idx
, outside_node
;
3118 re_node_set new_nodes
;
3120 assert (cur_nodes
->nelem
);
3122 err
= re_node_set_alloc (&new_nodes
, cur_nodes
->nelem
);
3123 if (BE (err
!= REG_NOERROR
, 0))
3125 /* Create a new node set NEW_NODES with the nodes which are epsilon
3126 closures of the node in CUR_NODES. */
3128 for (idx
= 0; idx
< cur_nodes
->nelem
; ++idx
)
3130 int cur_node
= cur_nodes
->elems
[idx
];
3131 const re_node_set
*eclosure
= dfa
->eclosures
+ cur_node
;
3132 outside_node
= find_subexp_node (dfa
, eclosure
, ex_subexp
, type
);
3133 if (outside_node
== -1)
3135 /* There are no problematic nodes, just merge them. */
3136 err
= re_node_set_merge (&new_nodes
, eclosure
);
3137 if (BE (err
!= REG_NOERROR
, 0))
3139 re_node_set_free (&new_nodes
);
3145 /* There are problematic nodes, re-calculate incrementally. */
3146 err
= check_arrival_expand_ecl_sub (dfa
, &new_nodes
, cur_node
,
3148 if (BE (err
!= REG_NOERROR
, 0))
3150 re_node_set_free (&new_nodes
);
3155 re_node_set_free (cur_nodes
);
3156 *cur_nodes
= new_nodes
;
3160 /* Helper function for check_arrival_expand_ecl.
3161 Check incrementally the epsilon closure of TARGET, and if it isn't
3162 problematic append it to DST_NODES. */
3164 static reg_errcode_t
3166 check_arrival_expand_ecl_sub (const re_dfa_t
*dfa
, re_node_set
*dst_nodes
,
3167 int target
, int ex_subexp
, int type
)
3170 for (cur_node
= target
; !re_node_set_contains (dst_nodes
, cur_node
);)
3174 if (dfa
->nodes
[cur_node
].type
== type
3175 && dfa
->nodes
[cur_node
].opr
.idx
== ex_subexp
)
3177 if (type
== OP_CLOSE_SUBEXP
)
3179 err
= re_node_set_insert (dst_nodes
, cur_node
);
3180 if (BE (err
== -1, 0))
3185 err
= re_node_set_insert (dst_nodes
, cur_node
);
3186 if (BE (err
== -1, 0))
3188 if (dfa
->edests
[cur_node
].nelem
== 0)
3190 if (dfa
->edests
[cur_node
].nelem
== 2)
3192 err
= check_arrival_expand_ecl_sub (dfa
, dst_nodes
,
3193 dfa
->edests
[cur_node
].elems
[1],
3195 if (BE (err
!= REG_NOERROR
, 0))
3198 cur_node
= dfa
->edests
[cur_node
].elems
[0];
3204 /* For all the back references in the current state, calculate the
3205 destination of the back references by the appropriate entry
3206 in MCTX->BKREF_ENTS. */
3208 static reg_errcode_t
3210 expand_bkref_cache (re_match_context_t
*mctx
, re_node_set
*cur_nodes
,
3211 int cur_str
, int subexp_num
, int type
)
3213 const re_dfa_t
*const dfa
= mctx
->dfa
;
3215 int cache_idx_start
= search_cur_bkref_entry (mctx
, cur_str
);
3216 struct re_backref_cache_entry
*ent
;
3218 if (cache_idx_start
== -1)
3222 ent
= mctx
->bkref_ents
+ cache_idx_start
;
3225 int to_idx
, next_node
;
3227 /* Is this entry ENT is appropriate? */
3228 if (!re_node_set_contains (cur_nodes
, ent
->node
))
3231 to_idx
= cur_str
+ ent
->subexp_to
- ent
->subexp_from
;
3232 /* Calculate the destination of the back reference, and append it
3233 to MCTX->STATE_LOG. */
3234 if (to_idx
== cur_str
)
3236 /* The backreference did epsilon transit, we must re-check all the
3237 node in the current state. */
3238 re_node_set new_dests
;
3239 reg_errcode_t err2
, err3
;
3240 next_node
= dfa
->edests
[ent
->node
].elems
[0];
3241 if (re_node_set_contains (cur_nodes
, next_node
))
3243 err
= re_node_set_init_1 (&new_dests
, next_node
);
3244 err2
= check_arrival_expand_ecl (dfa
, &new_dests
, subexp_num
, type
);
3245 err3
= re_node_set_merge (cur_nodes
, &new_dests
);
3246 re_node_set_free (&new_dests
);
3247 if (BE (err
!= REG_NOERROR
|| err2
!= REG_NOERROR
3248 || err3
!= REG_NOERROR
, 0))
3250 err
= (err
!= REG_NOERROR
? err
3251 : (err2
!= REG_NOERROR
? err2
: err3
));
3254 /* TODO: It is still inefficient... */
3259 re_node_set union_set
;
3260 next_node
= dfa
->nexts
[ent
->node
];
3261 if (mctx
->state_log
[to_idx
])
3264 if (re_node_set_contains (&mctx
->state_log
[to_idx
]->nodes
,
3267 err
= re_node_set_init_copy (&union_set
,
3268 &mctx
->state_log
[to_idx
]->nodes
);
3269 ret
= re_node_set_insert (&union_set
, next_node
);
3270 if (BE (err
!= REG_NOERROR
|| ret
< 0, 0))
3272 re_node_set_free (&union_set
);
3273 err
= err
!= REG_NOERROR
? err
: REG_ESPACE
;
3279 err
= re_node_set_init_1 (&union_set
, next_node
);
3280 if (BE (err
!= REG_NOERROR
, 0))
3283 mctx
->state_log
[to_idx
] = re_acquire_state (&err
, dfa
, &union_set
);
3284 re_node_set_free (&union_set
);
3285 if (BE (mctx
->state_log
[to_idx
] == NULL
3286 && err
!= REG_NOERROR
, 0))
3290 while (ent
++->more
);
3294 /* Build transition table for the state.
3295 Return 1 if succeeded, otherwise return NULL. */
3299 build_trtable (const re_dfa_t
*dfa
, re_dfastate_t
*state
)
3302 int i
, j
, ch
, need_word_trtable
= 0;
3303 bitset_word_t elem
, mask
;
3304 bool dests_node_malloced
= false;
3305 bool dest_states_malloced
= false;
3306 int ndests
; /* Number of the destination states from `state'. */
3307 re_dfastate_t
**trtable
;
3308 re_dfastate_t
**dest_states
= NULL
, **dest_states_word
, **dest_states_nl
;
3309 re_node_set follows
, *dests_node
;
3311 bitset_t acceptable
;
3315 re_node_set dests_node
[SBC_MAX
];
3316 bitset_t dests_ch
[SBC_MAX
];
3319 /* We build DFA states which corresponds to the destination nodes
3320 from `state'. `dests_node[i]' represents the nodes which i-th
3321 destination state contains, and `dests_ch[i]' represents the
3322 characters which i-th destination state accepts. */
3323 if (__libc_use_alloca (sizeof (struct dests_alloc
)))
3324 dests_alloc
= (struct dests_alloc
*) alloca (sizeof (struct dests_alloc
));
3327 dests_alloc
= re_malloc (struct dests_alloc
, 1);
3328 if (BE (dests_alloc
== NULL
, 0))
3330 dests_node_malloced
= true;
3332 dests_node
= dests_alloc
->dests_node
;
3333 dests_ch
= dests_alloc
->dests_ch
;
3335 /* Initialize transiton table. */
3336 state
->word_trtable
= state
->trtable
= NULL
;
3338 /* At first, group all nodes belonging to `state' into several
3340 ndests
= group_nodes_into_DFAstates (dfa
, state
, dests_node
, dests_ch
);
3341 if (BE (ndests
<= 0, 0))
3343 if (dests_node_malloced
)
3345 /* Return 0 in case of an error, 1 otherwise. */
3348 state
->trtable
= (re_dfastate_t
**)
3349 calloc (sizeof (re_dfastate_t
*), SBC_MAX
);
3355 err
= re_node_set_alloc (&follows
, ndests
+ 1);
3356 if (BE (err
!= REG_NOERROR
, 0))
3359 if (__libc_use_alloca ((sizeof (re_node_set
) + sizeof (bitset_t
)) * SBC_MAX
3360 + ndests
* 3 * sizeof (re_dfastate_t
*)))
3361 dest_states
= (re_dfastate_t
**)
3362 alloca (ndests
* 3 * sizeof (re_dfastate_t
*));
3365 dest_states
= (re_dfastate_t
**)
3366 malloc (ndests
* 3 * sizeof (re_dfastate_t
*));
3367 if (BE (dest_states
== NULL
, 0))
3370 if (dest_states_malloced
)
3372 re_node_set_free (&follows
);
3373 for (i
= 0; i
< ndests
; ++i
)
3374 re_node_set_free (dests_node
+ i
);
3375 if (dests_node_malloced
)
3379 dest_states_malloced
= true;
3381 dest_states_word
= dest_states
+ ndests
;
3382 dest_states_nl
= dest_states_word
+ ndests
;
3383 bitset_empty (acceptable
);
3385 /* Then build the states for all destinations. */
3386 for (i
= 0; i
< ndests
; ++i
)
3389 re_node_set_empty (&follows
);
3390 /* Merge the follows of this destination states. */
3391 for (j
= 0; j
< dests_node
[i
].nelem
; ++j
)
3393 next_node
= dfa
->nexts
[dests_node
[i
].elems
[j
]];
3394 if (next_node
!= -1)
3396 err
= re_node_set_merge (&follows
, dfa
->eclosures
+ next_node
);
3397 if (BE (err
!= REG_NOERROR
, 0))
3401 dest_states
[i
] = re_acquire_state_context (&err
, dfa
, &follows
, 0);
3402 if (BE (dest_states
[i
] == NULL
&& err
!= REG_NOERROR
, 0))
3404 /* If the new state has context constraint,
3405 build appropriate states for these contexts. */
3406 if (dest_states
[i
]->has_constraint
)
3408 dest_states_word
[i
] = re_acquire_state_context (&err
, dfa
, &follows
,
3410 if (BE (dest_states_word
[i
] == NULL
&& err
!= REG_NOERROR
, 0))
3413 if (dest_states
[i
] != dest_states_word
[i
] && dfa
->mb_cur_max
> 1)
3414 need_word_trtable
= 1;
3416 dest_states_nl
[i
] = re_acquire_state_context (&err
, dfa
, &follows
,
3418 if (BE (dest_states_nl
[i
] == NULL
&& err
!= REG_NOERROR
, 0))
3423 dest_states_word
[i
] = dest_states
[i
];
3424 dest_states_nl
[i
] = dest_states
[i
];
3426 bitset_merge (acceptable
, dests_ch
[i
]);
3429 if (!BE (need_word_trtable
, 0))
3431 /* We don't care about whether the following character is a word
3432 character, or we are in a single-byte character set so we can
3433 discern by looking at the character code: allocate a
3434 256-entry transition table. */
3435 trtable
= state
->trtable
=
3436 (re_dfastate_t
**) calloc (sizeof (re_dfastate_t
*), SBC_MAX
);
3437 if (BE (trtable
== NULL
, 0))
3440 /* For all characters ch...: */
3441 for (i
= 0; i
< BITSET_WORDS
; ++i
)
3442 for (ch
= i
* BITSET_WORD_BITS
, elem
= acceptable
[i
], mask
= 1;
3444 mask
<<= 1, elem
>>= 1, ++ch
)
3445 if (BE (elem
& 1, 0))
3447 /* There must be exactly one destination which accepts
3448 character ch. See group_nodes_into_DFAstates. */
3449 for (j
= 0; (dests_ch
[j
][i
] & mask
) == 0; ++j
)
3452 /* j-th destination accepts the word character ch. */
3453 if (dfa
->word_char
[i
] & mask
)
3454 trtable
[ch
] = dest_states_word
[j
];
3456 trtable
[ch
] = dest_states
[j
];
3461 /* We care about whether the following character is a word
3462 character, and we are in a multi-byte character set: discern
3463 by looking at the character code: build two 256-entry
3464 transition tables, one starting at trtable[0] and one
3465 starting at trtable[SBC_MAX]. */
3466 trtable
= state
->word_trtable
=
3467 (re_dfastate_t
**) calloc (sizeof (re_dfastate_t
*), 2 * SBC_MAX
);
3468 if (BE (trtable
== NULL
, 0))
3471 /* For all characters ch...: */
3472 for (i
= 0; i
< BITSET_WORDS
; ++i
)
3473 for (ch
= i
* BITSET_WORD_BITS
, elem
= acceptable
[i
], mask
= 1;
3475 mask
<<= 1, elem
>>= 1, ++ch
)
3476 if (BE (elem
& 1, 0))
3478 /* There must be exactly one destination which accepts
3479 character ch. See group_nodes_into_DFAstates. */
3480 for (j
= 0; (dests_ch
[j
][i
] & mask
) == 0; ++j
)
3483 /* j-th destination accepts the word character ch. */
3484 trtable
[ch
] = dest_states
[j
];
3485 trtable
[ch
+ SBC_MAX
] = dest_states_word
[j
];
3490 if (bitset_contain (acceptable
, NEWLINE_CHAR
))
3492 /* The current state accepts newline character. */
3493 for (j
= 0; j
< ndests
; ++j
)
3494 if (bitset_contain (dests_ch
[j
], NEWLINE_CHAR
))
3496 /* k-th destination accepts newline character. */
3497 trtable
[NEWLINE_CHAR
] = dest_states_nl
[j
];
3498 if (need_word_trtable
)
3499 trtable
[NEWLINE_CHAR
+ SBC_MAX
] = dest_states_nl
[j
];
3500 /* There must be only one destination which accepts
3501 newline. See group_nodes_into_DFAstates. */
3506 if (dest_states_malloced
)
3509 re_node_set_free (&follows
);
3510 for (i
= 0; i
< ndests
; ++i
)
3511 re_node_set_free (dests_node
+ i
);
3513 if (dests_node_malloced
)
3519 /* Group all nodes belonging to STATE into several destinations.
3520 Then for all destinations, set the nodes belonging to the destination
3521 to DESTS_NODE[i] and set the characters accepted by the destination
3522 to DEST_CH[i]. This function return the number of destinations. */
3526 group_nodes_into_DFAstates (const re_dfa_t
*dfa
, const re_dfastate_t
*state
,
3527 re_node_set
*dests_node
, bitset_t
*dests_ch
)
3532 int ndests
; /* Number of the destinations from `state'. */
3533 bitset_t accepts
; /* Characters a node can accept. */
3534 const re_node_set
*cur_nodes
= &state
->nodes
;
3535 bitset_empty (accepts
);
3538 /* For all the nodes belonging to `state', */
3539 for (i
= 0; i
< cur_nodes
->nelem
; ++i
)
3541 re_token_t
*node
= &dfa
->nodes
[cur_nodes
->elems
[i
]];
3542 re_token_type_t type
= node
->type
;
3543 unsigned int constraint
= node
->constraint
;
3545 /* Enumerate all single byte character this node can accept. */
3546 if (type
== CHARACTER
)
3547 bitset_set (accepts
, node
->opr
.c
);
3548 else if (type
== SIMPLE_BRACKET
)
3550 bitset_merge (accepts
, node
->opr
.sbcset
);
3552 else if (type
== OP_PERIOD
)
3554 #ifdef RE_ENABLE_I18N
3555 if (dfa
->mb_cur_max
> 1)
3556 bitset_merge (accepts
, dfa
->sb_char
);
3559 bitset_set_all (accepts
);
3560 if (!(dfa
->syntax
& RE_DOT_NEWLINE
))
3561 bitset_clear (accepts
, '\n');
3562 if (dfa
->syntax
& RE_DOT_NOT_NULL
)
3563 bitset_clear (accepts
, '\0');
3565 #ifdef RE_ENABLE_I18N
3566 else if (type
== OP_UTF8_PERIOD
)
3568 memset (accepts
, '\xff', sizeof (bitset_t
) / 2);
3569 if (!(dfa
->syntax
& RE_DOT_NEWLINE
))
3570 bitset_clear (accepts
, '\n');
3571 if (dfa
->syntax
& RE_DOT_NOT_NULL
)
3572 bitset_clear (accepts
, '\0');
3578 /* Check the `accepts' and sift the characters which are not
3579 match it the context. */
3582 if (constraint
& NEXT_NEWLINE_CONSTRAINT
)
3584 bool accepts_newline
= bitset_contain (accepts
, NEWLINE_CHAR
);
3585 bitset_empty (accepts
);
3586 if (accepts_newline
)
3587 bitset_set (accepts
, NEWLINE_CHAR
);
3591 if (constraint
& NEXT_ENDBUF_CONSTRAINT
)
3593 bitset_empty (accepts
);
3597 if (constraint
& NEXT_WORD_CONSTRAINT
)
3599 bitset_word_t any_set
= 0;
3600 if (type
== CHARACTER
&& !node
->word_char
)
3602 bitset_empty (accepts
);
3605 #ifdef RE_ENABLE_I18N
3606 if (dfa
->mb_cur_max
> 1)
3607 for (j
= 0; j
< BITSET_WORDS
; ++j
)
3608 any_set
|= (accepts
[j
] &= (dfa
->word_char
[j
] | ~dfa
->sb_char
[j
]));
3611 for (j
= 0; j
< BITSET_WORDS
; ++j
)
3612 any_set
|= (accepts
[j
] &= dfa
->word_char
[j
]);
3616 if (constraint
& NEXT_NOTWORD_CONSTRAINT
)
3618 bitset_word_t any_set
= 0;
3619 if (type
== CHARACTER
&& node
->word_char
)
3621 bitset_empty (accepts
);
3624 #ifdef RE_ENABLE_I18N
3625 if (dfa
->mb_cur_max
> 1)
3626 for (j
= 0; j
< BITSET_WORDS
; ++j
)
3627 any_set
|= (accepts
[j
] &= ~(dfa
->word_char
[j
] & dfa
->sb_char
[j
]));
3630 for (j
= 0; j
< BITSET_WORDS
; ++j
)
3631 any_set
|= (accepts
[j
] &= ~dfa
->word_char
[j
]);
3637 /* Then divide `accepts' into DFA states, or create a new
3638 state. Above, we make sure that accepts is not empty. */
3639 for (j
= 0; j
< ndests
; ++j
)
3641 bitset_t intersec
; /* Intersection sets, see below. */
3643 /* Flags, see below. */
3644 bitset_word_t has_intersec
, not_subset
, not_consumed
;
3646 /* Optimization, skip if this state doesn't accept the character. */
3647 if (type
== CHARACTER
&& !bitset_contain (dests_ch
[j
], node
->opr
.c
))
3650 /* Enumerate the intersection set of this state and `accepts'. */
3652 for (k
= 0; k
< BITSET_WORDS
; ++k
)
3653 has_intersec
|= intersec
[k
] = accepts
[k
] & dests_ch
[j
][k
];
3654 /* And skip if the intersection set is empty. */
3658 /* Then check if this state is a subset of `accepts'. */
3659 not_subset
= not_consumed
= 0;
3660 for (k
= 0; k
< BITSET_WORDS
; ++k
)
3662 not_subset
|= remains
[k
] = ~accepts
[k
] & dests_ch
[j
][k
];
3663 not_consumed
|= accepts
[k
] = accepts
[k
] & ~dests_ch
[j
][k
];
3666 /* If this state isn't a subset of `accepts', create a
3667 new group state, which has the `remains'. */
3670 bitset_copy (dests_ch
[ndests
], remains
);
3671 bitset_copy (dests_ch
[j
], intersec
);
3672 err
= re_node_set_init_copy (dests_node
+ ndests
, &dests_node
[j
]);
3673 if (BE (err
!= REG_NOERROR
, 0))
3678 /* Put the position in the current group. */
3679 result
= re_node_set_insert (&dests_node
[j
], cur_nodes
->elems
[i
]);
3680 if (BE (result
< 0, 0))
3683 /* If all characters are consumed, go to next node. */
3687 /* Some characters remain, create a new group. */
3690 bitset_copy (dests_ch
[ndests
], accepts
);
3691 err
= re_node_set_init_1 (dests_node
+ ndests
, cur_nodes
->elems
[i
]);
3692 if (BE (err
!= REG_NOERROR
, 0))
3695 bitset_empty (accepts
);
3700 for (j
= 0; j
< ndests
; ++j
)
3701 re_node_set_free (dests_node
+ j
);
3705 #ifdef RE_ENABLE_I18N
3706 /* Check how many bytes the node `dfa->nodes[node_idx]' accepts.
3707 Return the number of the bytes the node accepts.
3708 STR_IDX is the current index of the input string.
3710 This function handles the nodes which can accept one character, or
3711 one collating element like '.', '[a-z]', opposite to the other nodes
3712 can only accept one byte. */
3716 check_node_accept_bytes (const re_dfa_t
*dfa
, int node_idx
,
3717 const re_string_t
*input
, int str_idx
)
3719 const re_token_t
*node
= dfa
->nodes
+ node_idx
;
3720 int char_len
, elem_len
;
3723 if (BE (node
->type
== OP_UTF8_PERIOD
, 0))
3725 unsigned char c
= re_string_byte_at (input
, str_idx
), d
;
3726 if (BE (c
< 0xc2, 1))
3729 if (str_idx
+ 2 > input
->len
)
3732 d
= re_string_byte_at (input
, str_idx
+ 1);
3734 return (d
< 0x80 || d
> 0xbf) ? 0 : 2;
3738 if (c
== 0xe0 && d
< 0xa0)
3744 if (c
== 0xf0 && d
< 0x90)
3750 if (c
== 0xf8 && d
< 0x88)
3756 if (c
== 0xfc && d
< 0x84)
3762 if (str_idx
+ char_len
> input
->len
)
3765 for (i
= 1; i
< char_len
; ++i
)
3767 d
= re_string_byte_at (input
, str_idx
+ i
);
3768 if (d
< 0x80 || d
> 0xbf)
3774 char_len
= re_string_char_size_at (input
, str_idx
);
3775 if (node
->type
== OP_PERIOD
)
3779 /* FIXME: I don't think this if is needed, as both '\n'
3780 and '\0' are char_len == 1. */
3781 /* '.' accepts any one character except the following two cases. */
3782 if ((!(dfa
->syntax
& RE_DOT_NEWLINE
) &&
3783 re_string_byte_at (input
, str_idx
) == '\n') ||
3784 ((dfa
->syntax
& RE_DOT_NOT_NULL
) &&
3785 re_string_byte_at (input
, str_idx
) == '\0'))
3790 elem_len
= re_string_elem_size_at (input
, str_idx
);
3791 if ((elem_len
<= 1 && char_len
<= 1) || char_len
== 0)
3794 if (node
->type
== COMPLEX_BRACKET
)
3796 const re_charset_t
*cset
= node
->opr
.mbcset
;
3798 const unsigned char *pin
3799 = ((const unsigned char *) re_string_get_buffer (input
) + str_idx
);
3804 wchar_t wc
= ((cset
->nranges
|| cset
->nchar_classes
|| cset
->nmbchars
)
3805 ? re_string_wchar_at (input
, str_idx
) : 0);
3807 /* match with multibyte character? */
3808 for (i
= 0; i
< cset
->nmbchars
; ++i
)
3809 if (wc
== cset
->mbchars
[i
])
3811 match_len
= char_len
;
3812 goto check_node_accept_bytes_match
;
3814 /* match with character_class? */
3815 for (i
= 0; i
< cset
->nchar_classes
; ++i
)
3817 wctype_t wt
= cset
->char_classes
[i
];
3818 if (__iswctype (wc
, wt
))
3820 match_len
= char_len
;
3821 goto check_node_accept_bytes_match
;
3826 nrules
= _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
3829 unsigned int in_collseq
= 0;
3830 const int32_t *table
, *indirect
;
3831 const unsigned char *weights
, *extra
;
3832 const char *collseqwc
;
3833 /* This #include defines a local function! */
3834 # include <locale/weight.h>
3836 /* match with collating_symbol? */
3837 if (cset
->ncoll_syms
)
3838 extra
= (const unsigned char *)
3839 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
);
3840 for (i
= 0; i
< cset
->ncoll_syms
; ++i
)
3842 const unsigned char *coll_sym
= extra
+ cset
->coll_syms
[i
];
3843 /* Compare the length of input collating element and
3844 the length of current collating element. */
3845 if (*coll_sym
!= elem_len
)
3847 /* Compare each bytes. */
3848 for (j
= 0; j
< *coll_sym
; j
++)
3849 if (pin
[j
] != coll_sym
[1 + j
])
3853 /* Match if every bytes is equal. */
3855 goto check_node_accept_bytes_match
;
3861 if (elem_len
<= char_len
)
3863 collseqwc
= _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_COLLSEQWC
);
3864 in_collseq
= __collseq_table_lookup (collseqwc
, wc
);
3867 in_collseq
= find_collation_sequence_value (pin
, elem_len
);
3869 /* match with range expression? */
3870 for (i
= 0; i
< cset
->nranges
; ++i
)
3871 if (cset
->range_starts
[i
] <= in_collseq
3872 && in_collseq
<= cset
->range_ends
[i
])
3874 match_len
= elem_len
;
3875 goto check_node_accept_bytes_match
;
3878 /* match with equivalence_class? */
3879 if (cset
->nequiv_classes
)
3881 const unsigned char *cp
= pin
;
3882 table
= (const int32_t *)
3883 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_TABLEMB
);
3884 weights
= (const unsigned char *)
3885 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_WEIGHTMB
);
3886 extra
= (const unsigned char *)
3887 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_EXTRAMB
);
3888 indirect
= (const int32_t *)
3889 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_INDIRECTMB
);
3890 int32_t idx
= findidx (&cp
);
3892 for (i
= 0; i
< cset
->nequiv_classes
; ++i
)
3894 int32_t equiv_class_idx
= cset
->equiv_classes
[i
];
3895 size_t weight_len
= weights
[idx
& 0xffffff];
3896 if (weight_len
== weights
[equiv_class_idx
& 0xffffff]
3897 && (idx
>> 24) == (equiv_class_idx
>> 24))
3902 equiv_class_idx
&= 0xffffff;
3904 while (cnt
<= weight_len
3905 && (weights
[equiv_class_idx
+ 1 + cnt
]
3906 == weights
[idx
+ 1 + cnt
]))
3908 if (cnt
> weight_len
)
3910 match_len
= elem_len
;
3911 goto check_node_accept_bytes_match
;
3920 /* match with range expression? */
3922 wchar_t cmp_buf
[] = {L
'\0', L
'\0', wc
, L
'\0', L
'\0', L
'\0'};
3924 wchar_t cmp_buf
[] = {L
'\0', L
'\0', L
'\0', L
'\0', L
'\0', L
'\0'};
3927 for (i
= 0; i
< cset
->nranges
; ++i
)
3929 cmp_buf
[0] = cset
->range_starts
[i
];
3930 cmp_buf
[4] = cset
->range_ends
[i
];
3931 if (wcscoll (cmp_buf
, cmp_buf
+ 2) <= 0
3932 && wcscoll (cmp_buf
+ 2, cmp_buf
+ 4) <= 0)
3934 match_len
= char_len
;
3935 goto check_node_accept_bytes_match
;
3939 check_node_accept_bytes_match
:
3940 if (!cset
->non_match
)
3947 return (elem_len
> char_len
) ? elem_len
: char_len
;
3956 find_collation_sequence_value (const unsigned char *mbs
, size_t mbs_len
)
3958 uint32_t nrules
= _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
3963 /* No valid character. Match it as a single byte character. */
3964 const unsigned char *collseq
= (const unsigned char *)
3965 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_COLLSEQMB
);
3966 return collseq
[mbs
[0]];
3973 const unsigned char *extra
= (const unsigned char *)
3974 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
);
3975 int32_t extrasize
= (const unsigned char *)
3976 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
+ 1) - extra
;
3978 for (idx
= 0; idx
< extrasize
;)
3980 int mbs_cnt
, found
= 0;
3981 int32_t elem_mbs_len
;
3982 /* Skip the name of collating element name. */
3983 idx
= idx
+ extra
[idx
] + 1;
3984 elem_mbs_len
= extra
[idx
++];
3985 if (mbs_len
== elem_mbs_len
)
3987 for (mbs_cnt
= 0; mbs_cnt
< elem_mbs_len
; ++mbs_cnt
)
3988 if (extra
[idx
+ mbs_cnt
] != mbs
[mbs_cnt
])
3990 if (mbs_cnt
== elem_mbs_len
)
3991 /* Found the entry. */
3994 /* Skip the byte sequence of the collating element. */
3995 idx
+= elem_mbs_len
;
3996 /* Adjust for the alignment. */
3997 idx
= (idx
+ 3) & ~3;
3998 /* Skip the collation sequence value. */
3999 idx
+= sizeof (uint32_t);
4000 /* Skip the wide char sequence of the collating element. */
4001 idx
= idx
+ sizeof (uint32_t) * (extra
[idx
] + 1);
4002 /* If we found the entry, return the sequence value. */
4004 return *(uint32_t *) (extra
+ idx
);
4005 /* Skip the collation sequence value. */
4006 idx
+= sizeof (uint32_t);
4012 #endif /* RE_ENABLE_I18N */
4014 /* Check whether the node accepts the byte which is IDX-th
4015 byte of the INPUT. */
4019 check_node_accept (const re_match_context_t
*mctx
, const re_token_t
*node
,
4023 ch
= re_string_byte_at (&mctx
->input
, idx
);
4027 if (node
->opr
.c
!= ch
)
4031 case SIMPLE_BRACKET
:
4032 if (!bitset_contain (node
->opr
.sbcset
, ch
))
4036 #ifdef RE_ENABLE_I18N
4037 case OP_UTF8_PERIOD
:
4043 if ((ch
== '\n' && !(mctx
->dfa
->syntax
& RE_DOT_NEWLINE
))
4044 || (ch
== '\0' && (mctx
->dfa
->syntax
& RE_DOT_NOT_NULL
)))
4052 if (node
->constraint
)
4054 /* The node has constraints. Check whether the current context
4055 satisfies the constraints. */
4056 unsigned int context
= re_string_context_at (&mctx
->input
, idx
,
4058 if (NOT_SATISFY_NEXT_CONSTRAINT (node
->constraint
, context
))
4065 /* Extend the buffers, if the buffers have run out. */
4067 static reg_errcode_t
4069 extend_buffers (re_match_context_t
*mctx
)
4072 re_string_t
*pstr
= &mctx
->input
;
4074 /* Double the lengthes of the buffers. */
4075 ret
= re_string_realloc_buffers (pstr
, pstr
->bufs_len
* 2);
4076 if (BE (ret
!= REG_NOERROR
, 0))
4079 if (mctx
->state_log
!= NULL
)
4081 /* And double the length of state_log. */
4082 /* XXX We have no indication of the size of this buffer. If this
4083 allocation fail we have no indication that the state_log array
4084 does not have the right size. */
4085 re_dfastate_t
**new_array
= re_realloc (mctx
->state_log
, re_dfastate_t
*,
4086 pstr
->bufs_len
+ 1);
4087 if (BE (new_array
== NULL
, 0))
4089 mctx
->state_log
= new_array
;
4092 /* Then reconstruct the buffers. */
4095 #ifdef RE_ENABLE_I18N
4096 if (pstr
->mb_cur_max
> 1)
4098 ret
= build_wcs_upper_buffer (pstr
);
4099 if (BE (ret
!= REG_NOERROR
, 0))
4103 #endif /* RE_ENABLE_I18N */
4104 build_upper_buffer (pstr
);
4108 #ifdef RE_ENABLE_I18N
4109 if (pstr
->mb_cur_max
> 1)
4110 build_wcs_buffer (pstr
);
4112 #endif /* RE_ENABLE_I18N */
4114 if (pstr
->trans
!= NULL
)
4115 re_string_translate_buffer (pstr
);
4122 /* Functions for matching context. */
4124 /* Initialize MCTX. */
4126 static reg_errcode_t
4128 match_ctx_init (re_match_context_t
*mctx
, int eflags
, int n
)
4130 mctx
->eflags
= eflags
;
4131 mctx
->match_last
= -1;
4134 mctx
->bkref_ents
= re_malloc (struct re_backref_cache_entry
, n
);
4135 mctx
->sub_tops
= re_malloc (re_sub_match_top_t
*, n
);
4136 if (BE (mctx
->bkref_ents
== NULL
|| mctx
->sub_tops
== NULL
, 0))
4139 /* Already zero-ed by the caller.
4141 mctx->bkref_ents = NULL;
4142 mctx->nbkref_ents = 0;
4143 mctx->nsub_tops = 0; */
4144 mctx
->abkref_ents
= n
;
4145 mctx
->max_mb_elem_len
= 1;
4146 mctx
->asub_tops
= n
;
4150 /* Clean the entries which depend on the current input in MCTX.
4151 This function must be invoked when the matcher changes the start index
4152 of the input, or changes the input string. */
4156 match_ctx_clean (re_match_context_t
*mctx
)
4159 for (st_idx
= 0; st_idx
< mctx
->nsub_tops
; ++st_idx
)
4162 re_sub_match_top_t
*top
= mctx
->sub_tops
[st_idx
];
4163 for (sl_idx
= 0; sl_idx
< top
->nlasts
; ++sl_idx
)
4165 re_sub_match_last_t
*last
= top
->lasts
[sl_idx
];
4166 re_free (last
->path
.array
);
4169 re_free (top
->lasts
);
4172 re_free (top
->path
->array
);
4173 re_free (top
->path
);
4178 mctx
->nsub_tops
= 0;
4179 mctx
->nbkref_ents
= 0;
4182 /* Free all the memory associated with MCTX. */
4186 match_ctx_free (re_match_context_t
*mctx
)
4188 /* First, free all the memory associated with MCTX->SUB_TOPS. */
4189 match_ctx_clean (mctx
);
4190 re_free (mctx
->sub_tops
);
4191 re_free (mctx
->bkref_ents
);
4194 /* Add a new backreference entry to MCTX.
4195 Note that we assume that caller never call this function with duplicate
4196 entry, and call with STR_IDX which isn't smaller than any existing entry.
4199 static reg_errcode_t
4201 match_ctx_add_entry (re_match_context_t
*mctx
, int node
, int str_idx
, int from
,
4204 if (mctx
->nbkref_ents
>= mctx
->abkref_ents
)
4206 struct re_backref_cache_entry
* new_entry
;
4207 new_entry
= re_realloc (mctx
->bkref_ents
, struct re_backref_cache_entry
,
4208 mctx
->abkref_ents
* 2);
4209 if (BE (new_entry
== NULL
, 0))
4211 re_free (mctx
->bkref_ents
);
4214 mctx
->bkref_ents
= new_entry
;
4215 memset (mctx
->bkref_ents
+ mctx
->nbkref_ents
, '\0',
4216 sizeof (struct re_backref_cache_entry
) * mctx
->abkref_ents
);
4217 mctx
->abkref_ents
*= 2;
4219 if (mctx
->nbkref_ents
> 0
4220 && mctx
->bkref_ents
[mctx
->nbkref_ents
- 1].str_idx
== str_idx
)
4221 mctx
->bkref_ents
[mctx
->nbkref_ents
- 1].more
= 1;
4223 mctx
->bkref_ents
[mctx
->nbkref_ents
].node
= node
;
4224 mctx
->bkref_ents
[mctx
->nbkref_ents
].str_idx
= str_idx
;
4225 mctx
->bkref_ents
[mctx
->nbkref_ents
].subexp_from
= from
;
4226 mctx
->bkref_ents
[mctx
->nbkref_ents
].subexp_to
= to
;
4228 /* This is a cache that saves negative results of check_dst_limits_calc_pos.
4229 If bit N is clear, means that this entry won't epsilon-transition to
4230 an OP_OPEN_SUBEXP or OP_CLOSE_SUBEXP for the N+1-th subexpression. If
4231 it is set, check_dst_limits_calc_pos_1 will recurse and try to find one
4234 A backreference does not epsilon-transition unless it is empty, so set
4235 to all zeros if FROM != TO. */
4236 mctx
->bkref_ents
[mctx
->nbkref_ents
].eps_reachable_subexps_map
4237 = (from
== to
? ~0 : 0);
4239 mctx
->bkref_ents
[mctx
->nbkref_ents
++].more
= 0;
4240 if (mctx
->max_mb_elem_len
< to
- from
)
4241 mctx
->max_mb_elem_len
= to
- from
;
4245 /* Search for the first entry which has the same str_idx, or -1 if none is
4246 found. Note that MCTX->BKREF_ENTS is already sorted by MCTX->STR_IDX. */
4250 search_cur_bkref_entry (const re_match_context_t
*mctx
, int str_idx
)
4252 int left
, right
, mid
, last
;
4253 last
= right
= mctx
->nbkref_ents
;
4254 for (left
= 0; left
< right
;)
4256 mid
= (left
+ right
) / 2;
4257 if (mctx
->bkref_ents
[mid
].str_idx
< str_idx
)
4262 if (left
< last
&& mctx
->bkref_ents
[left
].str_idx
== str_idx
)
4268 /* Register the node NODE, whose type is OP_OPEN_SUBEXP, and which matches
4271 static reg_errcode_t
4273 match_ctx_add_subtop (re_match_context_t
*mctx
, int node
, int str_idx
)
4276 assert (mctx
->sub_tops
!= NULL
);
4277 assert (mctx
->asub_tops
> 0);
4279 if (BE (mctx
->nsub_tops
== mctx
->asub_tops
, 0))
4281 int new_asub_tops
= mctx
->asub_tops
* 2;
4282 re_sub_match_top_t
**new_array
= re_realloc (mctx
->sub_tops
,
4283 re_sub_match_top_t
*,
4285 if (BE (new_array
== NULL
, 0))
4287 mctx
->sub_tops
= new_array
;
4288 mctx
->asub_tops
= new_asub_tops
;
4290 mctx
->sub_tops
[mctx
->nsub_tops
] = calloc (1, sizeof (re_sub_match_top_t
));
4291 if (BE (mctx
->sub_tops
[mctx
->nsub_tops
] == NULL
, 0))
4293 mctx
->sub_tops
[mctx
->nsub_tops
]->node
= node
;
4294 mctx
->sub_tops
[mctx
->nsub_tops
++]->str_idx
= str_idx
;
4298 /* Register the node NODE, whose type is OP_CLOSE_SUBEXP, and which matches
4299 at STR_IDX, whose corresponding OP_OPEN_SUBEXP is SUB_TOP. */
4301 static re_sub_match_last_t
*
4303 match_ctx_add_sublast (re_sub_match_top_t
*subtop
, int node
, int str_idx
)
4305 re_sub_match_last_t
*new_entry
;
4306 if (BE (subtop
->nlasts
== subtop
->alasts
, 0))
4308 int new_alasts
= 2 * subtop
->alasts
+ 1;
4309 re_sub_match_last_t
**new_array
= re_realloc (subtop
->lasts
,
4310 re_sub_match_last_t
*,
4312 if (BE (new_array
== NULL
, 0))
4314 subtop
->lasts
= new_array
;
4315 subtop
->alasts
= new_alasts
;
4317 new_entry
= calloc (1, sizeof (re_sub_match_last_t
));
4318 if (BE (new_entry
!= NULL
, 1))
4320 subtop
->lasts
[subtop
->nlasts
] = new_entry
;
4321 new_entry
->node
= node
;
4322 new_entry
->str_idx
= str_idx
;
4330 sift_ctx_init (re_sift_context_t
*sctx
, re_dfastate_t
**sifted_sts
,
4331 re_dfastate_t
**limited_sts
, int last_node
, int last_str_idx
)
4333 sctx
->sifted_states
= sifted_sts
;
4334 sctx
->limited_states
= limited_sts
;
4335 sctx
->last_node
= last_node
;
4336 sctx
->last_str_idx
= last_str_idx
;
4337 re_node_set_init_empty (&sctx
->limits
);