1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language (but see
7 below for why this module is different).
9 Written by Philip Hazel
10 Copyright (c) 1997-2010 University of Cambridge
12 -----------------------------------------------------------------------------
13 Redistribution and use in source and binary forms, with or without
14 modification, are permitted provided that the following conditions are met:
16 * Redistributions of source code must retain the above copyright notice,
17 this list of conditions and the following disclaimer.
19 * Redistributions in binary form must reproduce the above copyright
20 notice, this list of conditions and the following disclaimer in the
21 documentation and/or other materials provided with the distribution.
23 * Neither the name of the University of Cambridge nor the names of its
24 contributors may be used to endorse or promote products derived from
25 this software without specific prior written permission.
27 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
28 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
31 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37 POSSIBILITY OF SUCH DAMAGE.
38 -----------------------------------------------------------------------------
42 /* This module contains the external function pcre_dfa_exec(), which is an
43 alternative matching function that uses a sort of DFA algorithm (not a true
44 FSM). This is NOT Perl- compatible, but it has advantages in certain
48 /* NOTE ABOUT PERFORMANCE: A user of this function sent some code that improved
49 the performance of his patterns greatly. I could not use it as it stood, as it
50 was not thread safe, and made assumptions about pattern sizes. Also, it caused
51 test 7 to loop, and test 9 to crash with a segfault.
53 The issue is the check for duplicate states, which is done by a simple linear
54 search up the state list. (Grep for "duplicate" below to find the code.) For
55 many patterns, there will never be many states active at one time, so a simple
56 linear search is fine. In patterns that have many active states, it might be a
57 bottleneck. The suggested code used an indexing scheme to remember which states
58 had previously been used for each character, and avoided the linear search when
59 it knew there was no chance of a duplicate. This was implemented when adding
60 states to the state lists.
62 I wrote some thread-safe, not-limited code to try something similar at the time
63 of checking for duplicates (instead of when adding states), using index vectors
64 on the stack. It did give a 13% improvement with one specially constructed
65 pattern for certain subject strings, but on other strings and on many of the
66 simpler patterns in the test suite it did worse. The major problem, I think,
67 was the extra time to initialize the index. This had to be done for each call
68 of internal_dfa_exec(). (The supplied patch used a static vector, initialized
69 only once - I suspect this was the cause of the problems with the tests.)
71 Overall, I concluded that the gains in some cases did not outweigh the losses
72 in others, so I abandoned this code. */
80 #define NLBLOCK md /* Block containing newline information */
81 #define PSSTART start_subject /* Field containing processed string start */
82 #define PSEND end_subject /* Field containing processed string end */
84 #include "pcre_internal.h"
87 /* For use to indent debugging output */
92 /*************************************************
93 * Code parameters and static tables *
94 *************************************************/
96 /* These are offsets that are used to turn the OP_TYPESTAR and friends opcodes
97 into others, under special conditions. A gap of 20 between the blocks should be
98 enough. The resulting opcodes don't have to be less than 256 because they are
99 never stored, so we push them well clear of the normal opcodes. */
101 #define OP_PROP_EXTRA 300
102 #define OP_EXTUNI_EXTRA 320
103 #define OP_ANYNL_EXTRA 340
104 #define OP_HSPACE_EXTRA 360
105 #define OP_VSPACE_EXTRA 380
108 /* This table identifies those opcodes that are followed immediately by a
109 character that is to be tested in some way. This makes it possible to
110 centralize the loading of these characters. In the case of Type * etc, the
111 "character" is the opcode for \D, \d, \S, \s, \W, or \w, which will always be a
112 small value. Non-zero values in the table are the offsets from the opcode where
113 the character is to be found. ***NOTE*** If the start of this table is
114 modified, the three tables that follow must also be modified. */
116 static const uschar coptable[] = {
118 0, 0, 0, 0, 0, /* \A, \G, \K, \B, \b */
119 0, 0, 0, 0, 0, 0, /* \D, \d, \S, \s, \W, \w */
120 0, 0, 0, /* Any, AllAny, Anybyte */
122 0, 0, 0, 0, 0, /* \R, \H, \h, \V, \v */
124 0, 0, 0, 0, 0, /* \Z, \z, Opt, ^, $ */
128 /* Positive single-char repeats */
129 1, 1, 1, 1, 1, 1, /* *, *?, +, +?, ?, ?? */
130 3, 3, 3, /* upto, minupto, exact */
131 1, 1, 1, 3, /* *+, ++, ?+, upto+ */
132 /* Negative single-char repeats - only for chars < 256 */
133 1, 1, 1, 1, 1, 1, /* NOT *, *?, +, +?, ?, ?? */
134 3, 3, 3, /* NOT upto, minupto, exact */
135 1, 1, 1, 3, /* NOT *+, ++, ?+, updo+ */
136 /* Positive type repeats */
137 1, 1, 1, 1, 1, 1, /* Type *, *?, +, +?, ?, ?? */
138 3, 3, 3, /* Type upto, minupto, exact */
139 1, 1, 1, 3, /* Type *+, ++, ?+, upto+ */
140 /* Character class & ref repeats */
141 0, 0, 0, 0, 0, 0, /* *, *?, +, +?, ?, ?? */
142 0, 0, /* CRRANGE, CRMINRANGE */
145 0, /* XCLASS - variable length */
155 0, /* Assert behind */
156 0, /* Assert behind not */
158 0, 0, 0, 0, /* ONCE, BRA, CBRA, COND */
159 0, 0, 0, /* SBRA, SCBRA, SCOND */
160 0, 0, /* CREF, NCREF */
161 0, 0, /* RREF, NRREF */
163 0, 0, /* BRAZERO, BRAMINZERO */
164 0, 0, 0, /* MARK, PRUNE, PRUNE_ARG, */
165 0, 0, 0, 0, /* SKIP, SKIP_ARG, THEN, THEN_ARG, */
166 0, 0, 0, 0, 0 /* COMMIT, FAIL, ACCEPT, CLOSE, SKIPZERO */
169 /* This table identifies those opcodes that inspect a character. It is used to
170 remember the fact that a character could have been inspected when the end of
171 the subject is reached. ***NOTE*** If the start of this table is modified, the
172 two tables that follow must also be modified. */
174 static const uschar poptable[] = {
176 0, 0, 0, 1, 1, /* \A, \G, \K, \B, \b */
177 1, 1, 1, 1, 1, 1, /* \D, \d, \S, \s, \W, \w */
178 1, 1, 1, /* Any, AllAny, Anybyte */
180 1, 1, 1, 1, 1, /* \R, \H, \h, \V, \v */
182 0, 0, 0, 0, 0, /* \Z, \z, Opt, ^, $ */
186 /* Positive single-char repeats */
187 1, 1, 1, 1, 1, 1, /* *, *?, +, +?, ?, ?? */
188 1, 1, 1, /* upto, minupto, exact */
189 1, 1, 1, 1, /* *+, ++, ?+, upto+ */
190 /* Negative single-char repeats - only for chars < 256 */
191 1, 1, 1, 1, 1, 1, /* NOT *, *?, +, +?, ?, ?? */
192 1, 1, 1, /* NOT upto, minupto, exact */
193 1, 1, 1, 1, /* NOT *+, ++, ?+, upto+ */
194 /* Positive type repeats */
195 1, 1, 1, 1, 1, 1, /* Type *, *?, +, +?, ?, ?? */
196 1, 1, 1, /* Type upto, minupto, exact */
197 1, 1, 1, 1, /* Type *+, ++, ?+, upto+ */
198 /* Character class & ref repeats */
199 1, 1, 1, 1, 1, 1, /* *, *?, +, +?, ?, ?? */
200 1, 1, /* CRRANGE, CRMINRANGE */
203 1, /* XCLASS - variable length */
213 0, /* Assert behind */
214 0, /* Assert behind not */
216 0, 0, 0, 0, /* ONCE, BRA, CBRA, COND */
217 0, 0, 0, /* SBRA, SCBRA, SCOND */
218 0, 0, /* CREF, NCREF */
219 0, 0, /* RREF, NRREF */
221 0, 0, /* BRAZERO, BRAMINZERO */
222 0, 0, 0, /* MARK, PRUNE, PRUNE_ARG, */
223 0, 0, 0, 0, /* SKIP, SKIP_ARG, THEN, THEN_ARG, */
224 0, 0, 0, 0, 0 /* COMMIT, FAIL, ACCEPT, CLOSE, SKIPZERO */
227 /* These 2 tables allow for compact code for testing for \D, \d, \S, \s, \W,
230 static const uschar toptable1[] = {
232 ctype_digit, ctype_digit,
233 ctype_space, ctype_space,
234 ctype_word, ctype_word,
235 0, 0 /* OP_ANY, OP_ALLANY */
238 static const uschar toptable2[] = {
243 1, 1 /* OP_ANY, OP_ALLANY */
247 /* Structure for holding data about a particular state, which is in effect the
248 current data for an active path through the match tree. It must consist
249 entirely of ints because the working vector we are passed, and which we put
250 these structures in, is a vector of ints. */
252 typedef struct stateblock {
253 int offset; /* Offset to opcode */
254 int count; /* Count for repeats */
255 int ims; /* ims flag bits */
256 int data; /* Some use extra data */
259 #define INTS_PER_STATEBLOCK (sizeof(stateblock)/sizeof(int))
263 /*************************************************
264 * Print character string *
265 *************************************************/
267 /* Character string printing function for debugging.
271 length number of bytes
278 pchars(unsigned char *p, int length, FILE *f)
283 if (isprint(c = *(p++)))
286 fprintf(f, "\\x%02x", c);
293 /*************************************************
294 * Execute a Regular Expression - DFA engine *
295 *************************************************/
297 /* This internal function applies a compiled pattern to a subject string,
298 starting at a given point, using a DFA engine. This function is called from the
299 external one, possibly multiple times if the pattern is not anchored. The
300 function calls itself recursively for some kinds of subpattern.
303 md the match_data block with fixed information
304 this_start_code the opening bracket of this subexpression's code
305 current_subject where we currently are in the subject string
306 start_offset start offset in the subject string
307 offsets vector to contain the matching string offsets
308 offsetcount size of same
309 workspace vector of workspace
311 ims the current ims flags
312 rlevel function call recursion level
313 recursing regex recursive call level
315 Returns: > 0 => number of match offset pairs placed in offsets
316 = 0 => offsets overflowed; longest matches are present
317 -1 => failed to match
318 < -1 => some kind of unexpected problem
320 The following macros are used for adding states to the two state vectors (one
321 for the current character, one for the following character). */
323 #define ADD_ACTIVE(x,y) \
324 if (active_count++ < wscount) \
326 next_active_state->offset = (x); \
327 next_active_state->count = (y); \
328 next_active_state->ims = ims; \
329 next_active_state++; \
330 DPRINTF(("%.*sADD_ACTIVE(%d,%d)\n", rlevel*2-2, SP, (x), (y))); \
332 else return PCRE_ERROR_DFA_WSSIZE
334 #define ADD_ACTIVE_DATA(x,y,z) \
335 if (active_count++ < wscount) \
337 next_active_state->offset = (x); \
338 next_active_state->count = (y); \
339 next_active_state->ims = ims; \
340 next_active_state->data = (z); \
341 next_active_state++; \
342 DPRINTF(("%.*sADD_ACTIVE_DATA(%d,%d,%d)\n", rlevel*2-2, SP, (x), (y), (z))); \
344 else return PCRE_ERROR_DFA_WSSIZE
346 #define ADD_NEW(x,y) \
347 if (new_count++ < wscount) \
349 next_new_state->offset = (x); \
350 next_new_state->count = (y); \
351 next_new_state->ims = ims; \
353 DPRINTF(("%.*sADD_NEW(%d,%d)\n", rlevel*2-2, SP, (x), (y))); \
355 else return PCRE_ERROR_DFA_WSSIZE
357 #define ADD_NEW_DATA(x,y,z) \
358 if (new_count++ < wscount) \
360 next_new_state->offset = (x); \
361 next_new_state->count = (y); \
362 next_new_state->ims = ims; \
363 next_new_state->data = (z); \
365 DPRINTF(("%.*sADD_NEW_DATA(%d,%d,%d)\n", rlevel*2-2, SP, (x), (y), (z))); \
367 else return PCRE_ERROR_DFA_WSSIZE
369 /* And now, here is the code */
374 const uschar *this_start_code,
375 const uschar *current_subject,
385 stateblock *active_states, *new_states, *temp_states;
386 stateblock *next_active_state, *next_new_state;
388 const uschar *ctypes, *lcc, *fcc;
390 const uschar *end_code, *first_op;
392 int active_count, new_count, match_count;
394 /* Some fields in the md block are frequently referenced, so we load them into
395 independent variables in the hope that this will perform better. */
397 const uschar *start_subject = md->start_subject;
398 const uschar *end_subject = md->end_subject;
399 const uschar *start_code = md->start_code;
402 BOOL utf8 = (md->poptions & PCRE_UTF8) != 0;
411 wscount = (wscount - (wscount % (INTS_PER_STATEBLOCK * 2))) /
412 (2 * INTS_PER_STATEBLOCK);
414 DPRINTF(("\n%.*s---------------------\n"
415 "%.*sCall to internal_dfa_exec f=%d r=%d\n",
416 rlevel*2-2, SP, rlevel*2-2, SP, rlevel, recursing));
418 ctypes = md->tables + ctypes_offset;
419 lcc = md->tables + lcc_offset;
420 fcc = md->tables + fcc_offset;
422 match_count = PCRE_ERROR_NOMATCH; /* A negative number */
424 active_states = (stateblock *)(workspace + 2);
425 next_new_state = new_states = active_states + wscount;
428 first_op = this_start_code + 1 + LINK_SIZE +
429 ((*this_start_code == OP_CBRA || *this_start_code == OP_SCBRA)? 2:0);
431 /* The first thing in any (sub) pattern is a bracket of some sort. Push all
432 the alternative states onto the list, and find out where the end is. This
433 makes is possible to use this function recursively, when we want to stop at a
434 matching internal ket rather than at the end.
436 If the first opcode in the first alternative is OP_REVERSE, we are dealing with
437 a backward assertion. In that case, we have to find out the maximum amount to
438 move back, and set up each alternative appropriately. */
440 if (*first_op == OP_REVERSE)
445 end_code = this_start_code;
448 int back = GET(end_code, 2+LINK_SIZE);
449 if (back > max_back) max_back = back;
450 end_code += GET(end_code, 1);
452 while (*end_code == OP_ALT);
454 /* If we can't go back the amount required for the longest lookbehind
455 pattern, go back as far as we can; some alternatives may still be viable. */
458 /* In character mode we have to step back character by character */
462 for (gone_back = 0; gone_back < max_back; gone_back++)
464 if (current_subject <= start_subject) break;
466 while (current_subject > start_subject &&
467 (*current_subject & 0xc0) == 0x80)
474 /* In byte-mode we can do this quickly. */
477 gone_back = (current_subject - max_back < start_subject)?
478 (int)(current_subject - start_subject) : max_back;
479 current_subject -= gone_back;
482 /* Save the earliest consulted character */
484 if (current_subject < md->start_used_ptr)
485 md->start_used_ptr = current_subject;
487 /* Now we can process the individual branches. */
489 end_code = this_start_code;
492 int back = GET(end_code, 2+LINK_SIZE);
493 if (back <= gone_back)
495 int bstate = (int)(end_code - start_code + 2 + 2*LINK_SIZE);
496 ADD_NEW_DATA(-bstate, 0, gone_back - back);
498 end_code += GET(end_code, 1);
500 while (*end_code == OP_ALT);
503 /* This is the code for a "normal" subpattern (not a backward assertion). The
504 start of a whole pattern is always one of these. If we are at the top level,
505 we may be asked to restart matching from the same point that we reached for a
506 previous partial match. We still have to scan through the top-level branches to
507 find the end state. */
511 end_code = this_start_code;
515 if (rlevel == 1 && (md->moptions & PCRE_DFA_RESTART) != 0)
517 do { end_code += GET(end_code, 1); } while (*end_code == OP_ALT);
518 new_count = workspace[1];
520 memcpy(new_states, active_states, new_count * sizeof(stateblock));
527 int length = 1 + LINK_SIZE +
528 ((*this_start_code == OP_CBRA || *this_start_code == OP_SCBRA)? 2:0);
531 ADD_NEW((int)(end_code - start_code + length), 0);
532 end_code += GET(end_code, 1);
533 length = 1 + LINK_SIZE;
535 while (*end_code == OP_ALT);
539 workspace[0] = 0; /* Bit indicating which vector is current */
541 DPRINTF(("%.*sEnd state = %d\n", rlevel*2-2, SP, end_code - start_code));
543 /* Loop for scanning the subject */
545 ptr = current_subject;
552 BOOL could_continue = FALSE;
554 /* Make the new state list into the active state list and empty the
557 temp_states = active_states;
558 active_states = new_states;
559 new_states = temp_states;
560 active_count = new_count;
563 workspace[0] ^= 1; /* Remember for the restarting feature */
564 workspace[1] = active_count;
567 printf("%.*sNext character: rest of subject = \"", rlevel*2-2, SP);
568 pchars((uschar *)ptr, strlen((char *)ptr), stdout);
571 printf("%.*sActive states: ", rlevel*2-2, SP);
572 for (i = 0; i < active_count; i++)
573 printf("%d/%d ", active_states[i].offset, active_states[i].count);
577 /* Set the pointers for adding new states */
579 next_active_state = active_states + active_count;
580 next_new_state = new_states;
582 /* Load the current character from the subject outside the loop, as many
583 different states may want to look at it, and we assume that at least one
586 if (ptr < end_subject)
588 clen = 1; /* Number of bytes in the character */
590 if (utf8) { GETCHARLEN(c, ptr, clen); } else
591 #endif /* SUPPORT_UTF8 */
596 clen = 0; /* This indicates the end of the subject */
597 c = NOTACHAR; /* This value should never actually be used */
600 /* Scan up the active states and act on each one. The result of an action
601 may be to add more states to the currently active list (e.g. on hitting a
602 parenthesis) or it may be to put states on the new list, for considering
603 when we move the character pointer on. */
605 for (i = 0; i < active_count; i++)
607 stateblock *current_state = active_states + i;
609 int state_offset = current_state->offset;
610 int count, codevalue, rrc;
613 printf ("%.*sProcessing state %d c=", rlevel*2-2, SP, state_offset);
614 if (clen == 0) printf("EOL\n");
615 else if (c > 32 && c < 127) printf("'%c'\n", c);
616 else printf("0x%02x\n", c);
619 /* This variable is referred to implicity in the ADD_xxx macros. */
621 ims = current_state->ims;
623 /* A negative offset is a special case meaning "hold off going to this
624 (negated) state until the number of characters in the data field have
627 if (state_offset < 0)
629 if (current_state->data > 0)
631 DPRINTF(("%.*sSkipping this character\n", rlevel*2-2, SP));
632 ADD_NEW_DATA(state_offset, current_state->count,
633 current_state->data - 1);
638 current_state->offset = state_offset = -state_offset;
642 /* Check for a duplicate state with the same count, and skip if found.
643 See the note at the head of this module about the possibility of improving
646 for (j = 0; j < i; j++)
648 if (active_states[j].offset == state_offset &&
649 active_states[j].count == current_state->count)
651 DPRINTF(("%.*sDuplicate state: skipped\n", rlevel*2-2, SP));
652 goto NEXT_ACTIVE_STATE;
656 /* The state offset is the offset to the opcode */
658 code = start_code + state_offset;
661 /* If this opcode inspects a character, but we are at the end of the
662 subject, remember the fact for use when testing for a partial match. */
664 if (clen == 0 && poptable[codevalue] != 0)
665 could_continue = TRUE;
667 /* If this opcode is followed by an inline character, load it. It is
668 tempting to test for the presence of a subject character here, but that
669 is wrong, because sometimes zero repetitions of the subject are
672 We also use this mechanism for opcodes such as OP_TYPEPLUS that take an
673 argument that is not a data character - but is always one byte long. We
674 have to take special action to deal with \P, \p, \H, \h, \V, \v and \X in
675 this case. To keep the other cases fast, convert these ones to new opcodes.
678 if (coptable[codevalue] > 0)
682 if (utf8) { GETCHARLEN(d, (code + coptable[codevalue]), dlen); } else
683 #endif /* SUPPORT_UTF8 */
684 d = code[coptable[codevalue]];
685 if (codevalue >= OP_TYPESTAR)
689 case OP_ANYBYTE: return PCRE_ERROR_DFA_UITEM;
691 case OP_PROP: codevalue += OP_PROP_EXTRA; break;
692 case OP_ANYNL: codevalue += OP_ANYNL_EXTRA; break;
693 case OP_EXTUNI: codevalue += OP_EXTUNI_EXTRA; break;
695 case OP_HSPACE: codevalue += OP_HSPACE_EXTRA; break;
697 case OP_VSPACE: codevalue += OP_VSPACE_EXTRA; break;
704 dlen = 0; /* Not strictly necessary, but compilers moan */
705 d = NOTACHAR; /* if these variables are not set. */
709 /* Now process the individual opcodes */
713 /* ========================================================================== */
714 /* These cases are never obeyed. This is a fudge that causes a compile-
715 time error if the vectors coptable or poptable, which are indexed by
716 opcode, are not the correct length. It seems to be the only way to do
717 such a check at compile time, as the sizeof() operator does not work
718 in the C preprocessor. */
720 case OP_TABLE_LENGTH:
721 case OP_TABLE_LENGTH +
722 ((sizeof(coptable) == OP_TABLE_LENGTH) &&
723 (sizeof(poptable) == OP_TABLE_LENGTH)):
726 /* ========================================================================== */
727 /* Reached a closing bracket. If not at the end of the pattern, carry
728 on with the next opcode. Otherwise, unless we have an empty string and
729 PCRE_NOTEMPTY is set, or PCRE_NOTEMPTY_ATSTART is set and we are at the
730 start of the subject, save the match data, shifting up all previous
731 matches so we always have the longest first. */
736 if (code != end_code)
738 ADD_ACTIVE(state_offset + 1 + LINK_SIZE, 0);
739 if (codevalue != OP_KET)
741 ADD_ACTIVE(state_offset - GET(code, 1), 0);
746 if (ptr > current_subject ||
747 ((md->moptions & PCRE_NOTEMPTY) == 0 &&
748 ((md->moptions & PCRE_NOTEMPTY_ATSTART) == 0 ||
749 current_subject > start_subject + md->start_offset)))
751 if (match_count < 0) match_count = (offsetcount >= 2)? 1 : 0;
752 else if (match_count > 0 && ++match_count * 2 >= offsetcount)
754 count = ((match_count == 0)? offsetcount : match_count * 2) - 2;
755 if (count > 0) memmove(offsets + 2, offsets, count * sizeof(int));
756 if (offsetcount >= 2)
758 offsets[0] = (int)(current_subject - start_subject);
759 offsets[1] = (int)(ptr - start_subject);
760 DPRINTF(("%.*sSet matched string = \"%.*s\"\n", rlevel*2-2, SP,
761 offsets[1] - offsets[0], current_subject));
763 if ((md->moptions & PCRE_DFA_SHORTEST) != 0)
765 DPRINTF(("%.*sEnd of internal_dfa_exec %d: returning %d\n"
766 "%.*s---------------------\n\n", rlevel*2-2, SP, rlevel,
767 match_count, rlevel*2-2, SP));
774 /* ========================================================================== */
775 /* These opcodes add to the current list of states without looking
776 at the current character. */
778 /*-----------------------------------------------------------------*/
780 do { code += GET(code, 1); } while (*code == OP_ALT);
781 ADD_ACTIVE((int)(code - start_code), 0);
784 /*-----------------------------------------------------------------*/
789 ADD_ACTIVE((int)(code - start_code + 1 + LINK_SIZE), 0);
790 code += GET(code, 1);
792 while (*code == OP_ALT);
795 /*-----------------------------------------------------------------*/
798 ADD_ACTIVE((int)(code - start_code + 3 + LINK_SIZE), 0);
799 code += GET(code, 1);
800 while (*code == OP_ALT)
802 ADD_ACTIVE((int)(code - start_code + 1 + LINK_SIZE), 0);
803 code += GET(code, 1);
807 /*-----------------------------------------------------------------*/
810 ADD_ACTIVE(state_offset + 1, 0);
811 code += 1 + GET(code, 2);
812 while (*code == OP_ALT) code += GET(code, 1);
813 ADD_ACTIVE((int)(code - start_code + 1 + LINK_SIZE), 0);
816 /*-----------------------------------------------------------------*/
818 code += 1 + GET(code, 2);
819 while (*code == OP_ALT) code += GET(code, 1);
820 ADD_ACTIVE((int)(code - start_code + 1 + LINK_SIZE), 0);
823 /*-----------------------------------------------------------------*/
825 if ((ptr == start_subject && (md->moptions & PCRE_NOTBOL) == 0) ||
826 ((ims & PCRE_MULTILINE) != 0 &&
827 ptr != end_subject &&
829 { ADD_ACTIVE(state_offset + 1, 0); }
832 /*-----------------------------------------------------------------*/
834 if (ptr >= end_subject)
836 if ((md->moptions & PCRE_PARTIAL_HARD) != 0)
837 could_continue = TRUE;
838 else { ADD_ACTIVE(state_offset + 1, 0); }
842 /*-----------------------------------------------------------------*/
845 ADD_ACTIVE(state_offset + 2, 0);
848 /*-----------------------------------------------------------------*/
850 if (ptr == start_subject) { ADD_ACTIVE(state_offset + 1, 0); }
853 /*-----------------------------------------------------------------*/
855 if (ptr == start_subject + start_offset) { ADD_ACTIVE(state_offset + 1, 0); }
859 /* ========================================================================== */
860 /* These opcodes inspect the next subject character, and sometimes
861 the previous one as well, but do not have an argument. The variable
862 clen contains the length of the current character and is zero if we are
863 at the end of the subject. */
865 /*-----------------------------------------------------------------*/
867 if (clen > 0 && !IS_NEWLINE(ptr))
868 { ADD_NEW(state_offset + 1, 0); }
871 /*-----------------------------------------------------------------*/
874 { ADD_NEW(state_offset + 1, 0); }
877 /*-----------------------------------------------------------------*/
879 if (clen == 0 && (md->moptions & PCRE_PARTIAL_HARD) != 0)
880 could_continue = TRUE;
881 else if (clen == 0 || (IS_NEWLINE(ptr) && ptr == end_subject - md->nllen))
882 { ADD_ACTIVE(state_offset + 1, 0); }
885 /*-----------------------------------------------------------------*/
887 if ((md->moptions & PCRE_NOTEOL) == 0)
889 if (clen == 0 && (md->moptions & PCRE_PARTIAL_HARD) != 0)
890 could_continue = TRUE;
891 else if (clen == 0 ||
892 ((md->poptions & PCRE_DOLLAR_ENDONLY) == 0 && IS_NEWLINE(ptr) &&
893 ((ims & PCRE_MULTILINE) != 0 || ptr == end_subject - md->nllen)
895 { ADD_ACTIVE(state_offset + 1, 0); }
897 else if ((ims & PCRE_MULTILINE) != 0 && IS_NEWLINE(ptr))
898 { ADD_ACTIVE(state_offset + 1, 0); }
901 /*-----------------------------------------------------------------*/
906 if (clen > 0 && c < 256 &&
907 ((ctypes[c] & toptable1[codevalue]) ^ toptable2[codevalue]) != 0)
908 { ADD_NEW(state_offset + 1, 0); }
911 /*-----------------------------------------------------------------*/
913 case OP_NOT_WHITESPACE:
914 case OP_NOT_WORDCHAR:
915 if (clen > 0 && (c >= 256 ||
916 ((ctypes[c] & toptable1[codevalue]) ^ toptable2[codevalue]) != 0))
917 { ADD_NEW(state_offset + 1, 0); }
920 /*-----------------------------------------------------------------*/
921 case OP_WORD_BOUNDARY:
922 case OP_NOT_WORD_BOUNDARY:
924 int left_word, right_word;
926 if (ptr > start_subject)
928 const uschar *temp = ptr - 1;
929 if (temp < md->start_used_ptr) md->start_used_ptr = temp;
931 if (utf8) BACKCHAR(temp);
933 GETCHARTEST(d, temp);
935 if ((md->poptions & PCRE_UCP) != 0)
937 if (d == '_') left_word = TRUE; else
939 int cat = UCD_CATEGORY(d);
940 left_word = (cat == ucp_L || cat == ucp_N);
945 left_word = d < 256 && (ctypes[d] & ctype_word) != 0;
947 else left_word = FALSE;
952 if ((md->poptions & PCRE_UCP) != 0)
954 if (c == '_') right_word = TRUE; else
956 int cat = UCD_CATEGORY(c);
957 right_word = (cat == ucp_L || cat == ucp_N);
962 right_word = c < 256 && (ctypes[c] & ctype_word) != 0;
964 else right_word = FALSE;
966 if ((left_word == right_word) == (codevalue == OP_NOT_WORD_BOUNDARY))
967 { ADD_ACTIVE(state_offset + 1, 0); }
972 /*-----------------------------------------------------------------*/
973 /* Check the next character by Unicode property. We will get here only
974 if the support is in the binary; otherwise a compile-time error occurs.
983 int chartype = UCD_CHARTYPE(c);
991 OK = chartype == ucp_Lu || chartype == ucp_Ll ||
996 OK = _pcre_ucp_gentype[chartype] == code[2];
1000 OK = chartype == code[2];
1004 OK = UCD_SCRIPT(c) == code[2];
1007 /* These are specials for combination cases. */
1010 OK = _pcre_ucp_gentype[chartype] == ucp_L ||
1011 _pcre_ucp_gentype[chartype] == ucp_N;
1014 case PT_SPACE: /* Perl space */
1015 OK = _pcre_ucp_gentype[chartype] == ucp_Z ||
1016 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR;
1019 case PT_PXSPACE: /* POSIX space */
1020 OK = _pcre_ucp_gentype[chartype] == ucp_Z ||
1021 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
1022 c == CHAR_FF || c == CHAR_CR;
1026 OK = _pcre_ucp_gentype[chartype] == ucp_L ||
1027 _pcre_ucp_gentype[chartype] == ucp_N ||
1028 c == CHAR_UNDERSCORE;
1031 /* Should never occur, but keep compilers from grumbling. */
1034 OK = codevalue != OP_PROP;
1038 if (OK == (codevalue == OP_PROP)) { ADD_NEW(state_offset + 3, 0); }
1045 /* ========================================================================== */
1046 /* These opcodes likewise inspect the subject character, but have an
1047 argument that is not a data character. It is one of these opcodes:
1048 OP_ANY, OP_ALLANY, OP_DIGIT, OP_NOT_DIGIT, OP_WHITESPACE, OP_NOT_SPACE,
1049 OP_WORDCHAR, OP_NOT_WORDCHAR. The value is loaded into d. */
1052 case OP_TYPEMINPLUS:
1053 case OP_TYPEPOSPLUS:
1054 count = current_state->count; /* Already matched */
1055 if (count > 0) { ADD_ACTIVE(state_offset + 2, 0); }
1058 if ((c >= 256 && d != OP_DIGIT && d != OP_WHITESPACE && d != OP_WORDCHAR) ||
1060 (d != OP_ANY || !IS_NEWLINE(ptr)) &&
1061 ((ctypes[c] & toptable1[d]) ^ toptable2[d]) != 0))
1063 if (count > 0 && codevalue == OP_TYPEPOSPLUS)
1065 active_count--; /* Remove non-match possibility */
1066 next_active_state--;
1069 ADD_NEW(state_offset, count);
1074 /*-----------------------------------------------------------------*/
1076 case OP_TYPEMINQUERY:
1077 case OP_TYPEPOSQUERY:
1078 ADD_ACTIVE(state_offset + 2, 0);
1081 if ((c >= 256 && d != OP_DIGIT && d != OP_WHITESPACE && d != OP_WORDCHAR) ||
1083 (d != OP_ANY || !IS_NEWLINE(ptr)) &&
1084 ((ctypes[c] & toptable1[d]) ^ toptable2[d]) != 0))
1086 if (codevalue == OP_TYPEPOSQUERY)
1088 active_count--; /* Remove non-match possibility */
1089 next_active_state--;
1091 ADD_NEW(state_offset + 2, 0);
1096 /*-----------------------------------------------------------------*/
1098 case OP_TYPEMINSTAR:
1099 case OP_TYPEPOSSTAR:
1100 ADD_ACTIVE(state_offset + 2, 0);
1103 if ((c >= 256 && d != OP_DIGIT && d != OP_WHITESPACE && d != OP_WORDCHAR) ||
1105 (d != OP_ANY || !IS_NEWLINE(ptr)) &&
1106 ((ctypes[c] & toptable1[d]) ^ toptable2[d]) != 0))
1108 if (codevalue == OP_TYPEPOSSTAR)
1110 active_count--; /* Remove non-match possibility */
1111 next_active_state--;
1113 ADD_NEW(state_offset, 0);
1118 /*-----------------------------------------------------------------*/
1120 count = current_state->count; /* Number already matched */
1123 if ((c >= 256 && d != OP_DIGIT && d != OP_WHITESPACE && d != OP_WORDCHAR) ||
1125 (d != OP_ANY || !IS_NEWLINE(ptr)) &&
1126 ((ctypes[c] & toptable1[d]) ^ toptable2[d]) != 0))
1128 if (++count >= GET2(code, 1))
1129 { ADD_NEW(state_offset + 4, 0); }
1131 { ADD_NEW(state_offset, count); }
1136 /*-----------------------------------------------------------------*/
1138 case OP_TYPEMINUPTO:
1139 case OP_TYPEPOSUPTO:
1140 ADD_ACTIVE(state_offset + 4, 0);
1141 count = current_state->count; /* Number already matched */
1144 if ((c >= 256 && d != OP_DIGIT && d != OP_WHITESPACE && d != OP_WORDCHAR) ||
1146 (d != OP_ANY || !IS_NEWLINE(ptr)) &&
1147 ((ctypes[c] & toptable1[d]) ^ toptable2[d]) != 0))
1149 if (codevalue == OP_TYPEPOSUPTO)
1151 active_count--; /* Remove non-match possibility */
1152 next_active_state--;
1154 if (++count >= GET2(code, 1))
1155 { ADD_NEW(state_offset + 4, 0); }
1157 { ADD_NEW(state_offset, count); }
1162 /* ========================================================================== */
1163 /* These are virtual opcodes that are used when something like
1164 OP_TYPEPLUS has OP_PROP, OP_NOTPROP, OP_ANYNL, or OP_EXTUNI as its
1165 argument. It keeps the code above fast for the other cases. The argument
1166 is in the d variable. */
1169 case OP_PROP_EXTRA + OP_TYPEPLUS:
1170 case OP_PROP_EXTRA + OP_TYPEMINPLUS:
1171 case OP_PROP_EXTRA + OP_TYPEPOSPLUS:
1172 count = current_state->count; /* Already matched */
1173 if (count > 0) { ADD_ACTIVE(state_offset + 4, 0); }
1177 int chartype = UCD_CHARTYPE(c);
1185 OK = chartype == ucp_Lu || chartype == ucp_Ll ||
1190 OK = _pcre_ucp_gentype[chartype] == code[3];
1194 OK = chartype == code[3];
1198 OK = UCD_SCRIPT(c) == code[3];
1201 /* These are specials for combination cases. */
1204 OK = _pcre_ucp_gentype[chartype] == ucp_L ||
1205 _pcre_ucp_gentype[chartype] == ucp_N;
1208 case PT_SPACE: /* Perl space */
1209 OK = _pcre_ucp_gentype[chartype] == ucp_Z ||
1210 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR;
1213 case PT_PXSPACE: /* POSIX space */
1214 OK = _pcre_ucp_gentype[chartype] == ucp_Z ||
1215 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
1216 c == CHAR_FF || c == CHAR_CR;
1220 OK = _pcre_ucp_gentype[chartype] == ucp_L ||
1221 _pcre_ucp_gentype[chartype] == ucp_N ||
1222 c == CHAR_UNDERSCORE;
1225 /* Should never occur, but keep compilers from grumbling. */
1228 OK = codevalue != OP_PROP;
1232 if (OK == (d == OP_PROP))
1234 if (count > 0 && codevalue == OP_PROP_EXTRA + OP_TYPEPOSPLUS)
1236 active_count--; /* Remove non-match possibility */
1237 next_active_state--;
1240 ADD_NEW(state_offset, count);
1245 /*-----------------------------------------------------------------*/
1246 case OP_EXTUNI_EXTRA + OP_TYPEPLUS:
1247 case OP_EXTUNI_EXTRA + OP_TYPEMINPLUS:
1248 case OP_EXTUNI_EXTRA + OP_TYPEPOSPLUS:
1249 count = current_state->count; /* Already matched */
1250 if (count > 0) { ADD_ACTIVE(state_offset + 2, 0); }
1251 if (clen > 0 && UCD_CATEGORY(c) != ucp_M)
1253 const uschar *nptr = ptr + clen;
1255 if (count > 0 && codevalue == OP_EXTUNI_EXTRA + OP_TYPEPOSPLUS)
1257 active_count--; /* Remove non-match possibility */
1258 next_active_state--;
1260 while (nptr < end_subject)
1264 GETCHARLEN(nd, nptr, ndlen);
1265 if (UCD_CATEGORY(nd) != ucp_M) break;
1270 ADD_NEW_DATA(-state_offset, count, ncount);
1275 /*-----------------------------------------------------------------*/
1276 case OP_ANYNL_EXTRA + OP_TYPEPLUS:
1277 case OP_ANYNL_EXTRA + OP_TYPEMINPLUS:
1278 case OP_ANYNL_EXTRA + OP_TYPEPOSPLUS:
1279 count = current_state->count; /* Already matched */
1280 if (count > 0) { ADD_ACTIVE(state_offset + 2, 0); }
1291 if ((md->moptions & PCRE_BSR_ANYCRLF) != 0) break;
1295 if (ptr + 1 < end_subject && ptr[1] == 0x0a) ncount = 1;
1300 if (count > 0 && codevalue == OP_ANYNL_EXTRA + OP_TYPEPOSPLUS)
1302 active_count--; /* Remove non-match possibility */
1303 next_active_state--;
1306 ADD_NEW_DATA(-state_offset, count, ncount);
1315 /*-----------------------------------------------------------------*/
1316 case OP_VSPACE_EXTRA + OP_TYPEPLUS:
1317 case OP_VSPACE_EXTRA + OP_TYPEMINPLUS:
1318 case OP_VSPACE_EXTRA + OP_TYPEPOSPLUS:
1319 count = current_state->count; /* Already matched */
1320 if (count > 0) { ADD_ACTIVE(state_offset + 2, 0); }
1341 if (OK == (d == OP_VSPACE))
1343 if (count > 0 && codevalue == OP_VSPACE_EXTRA + OP_TYPEPOSPLUS)
1345 active_count--; /* Remove non-match possibility */
1346 next_active_state--;
1349 ADD_NEW_DATA(-state_offset, count, 0);
1354 /*-----------------------------------------------------------------*/
1355 case OP_HSPACE_EXTRA + OP_TYPEPLUS:
1356 case OP_HSPACE_EXTRA + OP_TYPEMINPLUS:
1357 case OP_HSPACE_EXTRA + OP_TYPEPOSPLUS:
1358 count = current_state->count; /* Already matched */
1359 if (count > 0) { ADD_ACTIVE(state_offset + 2, 0); }
1366 case 0x20: /* SPACE */
1367 case 0xa0: /* NBSP */
1368 case 0x1680: /* OGHAM SPACE MARK */
1369 case 0x180e: /* MONGOLIAN VOWEL SEPARATOR */
1370 case 0x2000: /* EN QUAD */
1371 case 0x2001: /* EM QUAD */
1372 case 0x2002: /* EN SPACE */
1373 case 0x2003: /* EM SPACE */
1374 case 0x2004: /* THREE-PER-EM SPACE */
1375 case 0x2005: /* FOUR-PER-EM SPACE */
1376 case 0x2006: /* SIX-PER-EM SPACE */
1377 case 0x2007: /* FIGURE SPACE */
1378 case 0x2008: /* PUNCTUATION SPACE */
1379 case 0x2009: /* THIN SPACE */
1380 case 0x200A: /* HAIR SPACE */
1381 case 0x202f: /* NARROW NO-BREAK SPACE */
1382 case 0x205f: /* MEDIUM MATHEMATICAL SPACE */
1383 case 0x3000: /* IDEOGRAPHIC SPACE */
1392 if (OK == (d == OP_HSPACE))
1394 if (count > 0 && codevalue == OP_HSPACE_EXTRA + OP_TYPEPOSPLUS)
1396 active_count--; /* Remove non-match possibility */
1397 next_active_state--;
1400 ADD_NEW_DATA(-state_offset, count, 0);
1405 /*-----------------------------------------------------------------*/
1407 case OP_PROP_EXTRA + OP_TYPEQUERY:
1408 case OP_PROP_EXTRA + OP_TYPEMINQUERY:
1409 case OP_PROP_EXTRA + OP_TYPEPOSQUERY:
1413 case OP_PROP_EXTRA + OP_TYPESTAR:
1414 case OP_PROP_EXTRA + OP_TYPEMINSTAR:
1415 case OP_PROP_EXTRA + OP_TYPEPOSSTAR:
1420 ADD_ACTIVE(state_offset + 4, 0);
1424 int chartype = UCD_CHARTYPE(c);
1432 OK = chartype == ucp_Lu || chartype == ucp_Ll ||
1437 OK = _pcre_ucp_gentype[chartype] == code[3];
1441 OK = chartype == code[3];
1445 OK = UCD_SCRIPT(c) == code[3];
1448 /* These are specials for combination cases. */
1451 OK = _pcre_ucp_gentype[chartype] == ucp_L ||
1452 _pcre_ucp_gentype[chartype] == ucp_N;
1455 case PT_SPACE: /* Perl space */
1456 OK = _pcre_ucp_gentype[chartype] == ucp_Z ||
1457 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR;
1460 case PT_PXSPACE: /* POSIX space */
1461 OK = _pcre_ucp_gentype[chartype] == ucp_Z ||
1462 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
1463 c == CHAR_FF || c == CHAR_CR;
1467 OK = _pcre_ucp_gentype[chartype] == ucp_L ||
1468 _pcre_ucp_gentype[chartype] == ucp_N ||
1469 c == CHAR_UNDERSCORE;
1472 /* Should never occur, but keep compilers from grumbling. */
1475 OK = codevalue != OP_PROP;
1479 if (OK == (d == OP_PROP))
1481 if (codevalue == OP_PROP_EXTRA + OP_TYPEPOSSTAR ||
1482 codevalue == OP_PROP_EXTRA + OP_TYPEPOSQUERY)
1484 active_count--; /* Remove non-match possibility */
1485 next_active_state--;
1487 ADD_NEW(state_offset + count, 0);
1492 /*-----------------------------------------------------------------*/
1493 case OP_EXTUNI_EXTRA + OP_TYPEQUERY:
1494 case OP_EXTUNI_EXTRA + OP_TYPEMINQUERY:
1495 case OP_EXTUNI_EXTRA + OP_TYPEPOSQUERY:
1499 case OP_EXTUNI_EXTRA + OP_TYPESTAR:
1500 case OP_EXTUNI_EXTRA + OP_TYPEMINSTAR:
1501 case OP_EXTUNI_EXTRA + OP_TYPEPOSSTAR:
1506 ADD_ACTIVE(state_offset + 2, 0);
1507 if (clen > 0 && UCD_CATEGORY(c) != ucp_M)
1509 const uschar *nptr = ptr + clen;
1511 if (codevalue == OP_EXTUNI_EXTRA + OP_TYPEPOSSTAR ||
1512 codevalue == OP_EXTUNI_EXTRA + OP_TYPEPOSQUERY)
1514 active_count--; /* Remove non-match possibility */
1515 next_active_state--;
1517 while (nptr < end_subject)
1521 GETCHARLEN(nd, nptr, ndlen);
1522 if (UCD_CATEGORY(nd) != ucp_M) break;
1526 ADD_NEW_DATA(-(state_offset + count), 0, ncount);
1531 /*-----------------------------------------------------------------*/
1532 case OP_ANYNL_EXTRA + OP_TYPEQUERY:
1533 case OP_ANYNL_EXTRA + OP_TYPEMINQUERY:
1534 case OP_ANYNL_EXTRA + OP_TYPEPOSQUERY:
1538 case OP_ANYNL_EXTRA + OP_TYPESTAR:
1539 case OP_ANYNL_EXTRA + OP_TYPEMINSTAR:
1540 case OP_ANYNL_EXTRA + OP_TYPEPOSSTAR:
1544 ADD_ACTIVE(state_offset + 2, 0);
1555 if ((md->moptions & PCRE_BSR_ANYCRLF) != 0) break;
1559 if (ptr + 1 < end_subject && ptr[1] == 0x0a) ncount = 1;
1564 if (codevalue == OP_ANYNL_EXTRA + OP_TYPEPOSSTAR ||
1565 codevalue == OP_ANYNL_EXTRA + OP_TYPEPOSQUERY)
1567 active_count--; /* Remove non-match possibility */
1568 next_active_state--;
1570 ADD_NEW_DATA(-(state_offset + count), 0, ncount);
1579 /*-----------------------------------------------------------------*/
1580 case OP_VSPACE_EXTRA + OP_TYPEQUERY:
1581 case OP_VSPACE_EXTRA + OP_TYPEMINQUERY:
1582 case OP_VSPACE_EXTRA + OP_TYPEPOSQUERY:
1586 case OP_VSPACE_EXTRA + OP_TYPESTAR:
1587 case OP_VSPACE_EXTRA + OP_TYPEMINSTAR:
1588 case OP_VSPACE_EXTRA + OP_TYPEPOSSTAR:
1592 ADD_ACTIVE(state_offset + 2, 0);
1612 if (OK == (d == OP_VSPACE))
1614 if (codevalue == OP_VSPACE_EXTRA + OP_TYPEPOSSTAR ||
1615 codevalue == OP_VSPACE_EXTRA + OP_TYPEPOSQUERY)
1617 active_count--; /* Remove non-match possibility */
1618 next_active_state--;
1620 ADD_NEW_DATA(-(state_offset + count), 0, 0);
1625 /*-----------------------------------------------------------------*/
1626 case OP_HSPACE_EXTRA + OP_TYPEQUERY:
1627 case OP_HSPACE_EXTRA + OP_TYPEMINQUERY:
1628 case OP_HSPACE_EXTRA + OP_TYPEPOSQUERY:
1632 case OP_HSPACE_EXTRA + OP_TYPESTAR:
1633 case OP_HSPACE_EXTRA + OP_TYPEMINSTAR:
1634 case OP_HSPACE_EXTRA + OP_TYPEPOSSTAR:
1638 ADD_ACTIVE(state_offset + 2, 0);
1645 case 0x20: /* SPACE */
1646 case 0xa0: /* NBSP */
1647 case 0x1680: /* OGHAM SPACE MARK */
1648 case 0x180e: /* MONGOLIAN VOWEL SEPARATOR */
1649 case 0x2000: /* EN QUAD */
1650 case 0x2001: /* EM QUAD */
1651 case 0x2002: /* EN SPACE */
1652 case 0x2003: /* EM SPACE */
1653 case 0x2004: /* THREE-PER-EM SPACE */
1654 case 0x2005: /* FOUR-PER-EM SPACE */
1655 case 0x2006: /* SIX-PER-EM SPACE */
1656 case 0x2007: /* FIGURE SPACE */
1657 case 0x2008: /* PUNCTUATION SPACE */
1658 case 0x2009: /* THIN SPACE */
1659 case 0x200A: /* HAIR SPACE */
1660 case 0x202f: /* NARROW NO-BREAK SPACE */
1661 case 0x205f: /* MEDIUM MATHEMATICAL SPACE */
1662 case 0x3000: /* IDEOGRAPHIC SPACE */
1671 if (OK == (d == OP_HSPACE))
1673 if (codevalue == OP_HSPACE_EXTRA + OP_TYPEPOSSTAR ||
1674 codevalue == OP_HSPACE_EXTRA + OP_TYPEPOSQUERY)
1676 active_count--; /* Remove non-match possibility */
1677 next_active_state--;
1679 ADD_NEW_DATA(-(state_offset + count), 0, 0);
1684 /*-----------------------------------------------------------------*/
1686 case OP_PROP_EXTRA + OP_TYPEEXACT:
1687 case OP_PROP_EXTRA + OP_TYPEUPTO:
1688 case OP_PROP_EXTRA + OP_TYPEMINUPTO:
1689 case OP_PROP_EXTRA + OP_TYPEPOSUPTO:
1690 if (codevalue != OP_PROP_EXTRA + OP_TYPEEXACT)
1691 { ADD_ACTIVE(state_offset + 6, 0); }
1692 count = current_state->count; /* Number already matched */
1696 int chartype = UCD_CHARTYPE(c);
1704 OK = chartype == ucp_Lu || chartype == ucp_Ll ||
1709 OK = _pcre_ucp_gentype[chartype] == code[5];
1713 OK = chartype == code[5];
1717 OK = UCD_SCRIPT(c) == code[5];
1720 /* These are specials for combination cases. */
1723 OK = _pcre_ucp_gentype[chartype] == ucp_L ||
1724 _pcre_ucp_gentype[chartype] == ucp_N;
1727 case PT_SPACE: /* Perl space */
1728 OK = _pcre_ucp_gentype[chartype] == ucp_Z ||
1729 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR;
1732 case PT_PXSPACE: /* POSIX space */
1733 OK = _pcre_ucp_gentype[chartype] == ucp_Z ||
1734 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
1735 c == CHAR_FF || c == CHAR_CR;
1739 OK = _pcre_ucp_gentype[chartype] == ucp_L ||
1740 _pcre_ucp_gentype[chartype] == ucp_N ||
1741 c == CHAR_UNDERSCORE;
1744 /* Should never occur, but keep compilers from grumbling. */
1747 OK = codevalue != OP_PROP;
1751 if (OK == (d == OP_PROP))
1753 if (codevalue == OP_PROP_EXTRA + OP_TYPEPOSUPTO)
1755 active_count--; /* Remove non-match possibility */
1756 next_active_state--;
1758 if (++count >= GET2(code, 1))
1759 { ADD_NEW(state_offset + 6, 0); }
1761 { ADD_NEW(state_offset, count); }
1766 /*-----------------------------------------------------------------*/
1767 case OP_EXTUNI_EXTRA + OP_TYPEEXACT:
1768 case OP_EXTUNI_EXTRA + OP_TYPEUPTO:
1769 case OP_EXTUNI_EXTRA + OP_TYPEMINUPTO:
1770 case OP_EXTUNI_EXTRA + OP_TYPEPOSUPTO:
1771 if (codevalue != OP_EXTUNI_EXTRA + OP_TYPEEXACT)
1772 { ADD_ACTIVE(state_offset + 4, 0); }
1773 count = current_state->count; /* Number already matched */
1774 if (clen > 0 && UCD_CATEGORY(c) != ucp_M)
1776 const uschar *nptr = ptr + clen;
1778 if (codevalue == OP_EXTUNI_EXTRA + OP_TYPEPOSUPTO)
1780 active_count--; /* Remove non-match possibility */
1781 next_active_state--;
1783 while (nptr < end_subject)
1787 GETCHARLEN(nd, nptr, ndlen);
1788 if (UCD_CATEGORY(nd) != ucp_M) break;
1792 if (++count >= GET2(code, 1))
1793 { ADD_NEW_DATA(-(state_offset + 4), 0, ncount); }
1795 { ADD_NEW_DATA(-state_offset, count, ncount); }
1800 /*-----------------------------------------------------------------*/
1801 case OP_ANYNL_EXTRA + OP_TYPEEXACT:
1802 case OP_ANYNL_EXTRA + OP_TYPEUPTO:
1803 case OP_ANYNL_EXTRA + OP_TYPEMINUPTO:
1804 case OP_ANYNL_EXTRA + OP_TYPEPOSUPTO:
1805 if (codevalue != OP_ANYNL_EXTRA + OP_TYPEEXACT)
1806 { ADD_ACTIVE(state_offset + 4, 0); }
1807 count = current_state->count; /* Number already matched */
1818 if ((md->moptions & PCRE_BSR_ANYCRLF) != 0) break;
1822 if (ptr + 1 < end_subject && ptr[1] == 0x0a) ncount = 1;
1827 if (codevalue == OP_ANYNL_EXTRA + OP_TYPEPOSUPTO)
1829 active_count--; /* Remove non-match possibility */
1830 next_active_state--;
1832 if (++count >= GET2(code, 1))
1833 { ADD_NEW_DATA(-(state_offset + 4), 0, ncount); }
1835 { ADD_NEW_DATA(-state_offset, count, ncount); }
1844 /*-----------------------------------------------------------------*/
1845 case OP_VSPACE_EXTRA + OP_TYPEEXACT:
1846 case OP_VSPACE_EXTRA + OP_TYPEUPTO:
1847 case OP_VSPACE_EXTRA + OP_TYPEMINUPTO:
1848 case OP_VSPACE_EXTRA + OP_TYPEPOSUPTO:
1849 if (codevalue != OP_VSPACE_EXTRA + OP_TYPEEXACT)
1850 { ADD_ACTIVE(state_offset + 4, 0); }
1851 count = current_state->count; /* Number already matched */
1871 if (OK == (d == OP_VSPACE))
1873 if (codevalue == OP_VSPACE_EXTRA + OP_TYPEPOSUPTO)
1875 active_count--; /* Remove non-match possibility */
1876 next_active_state--;
1878 if (++count >= GET2(code, 1))
1879 { ADD_NEW_DATA(-(state_offset + 4), 0, 0); }
1881 { ADD_NEW_DATA(-state_offset, count, 0); }
1886 /*-----------------------------------------------------------------*/
1887 case OP_HSPACE_EXTRA + OP_TYPEEXACT:
1888 case OP_HSPACE_EXTRA + OP_TYPEUPTO:
1889 case OP_HSPACE_EXTRA + OP_TYPEMINUPTO:
1890 case OP_HSPACE_EXTRA + OP_TYPEPOSUPTO:
1891 if (codevalue != OP_HSPACE_EXTRA + OP_TYPEEXACT)
1892 { ADD_ACTIVE(state_offset + 4, 0); }
1893 count = current_state->count; /* Number already matched */
1900 case 0x20: /* SPACE */
1901 case 0xa0: /* NBSP */
1902 case 0x1680: /* OGHAM SPACE MARK */
1903 case 0x180e: /* MONGOLIAN VOWEL SEPARATOR */
1904 case 0x2000: /* EN QUAD */
1905 case 0x2001: /* EM QUAD */
1906 case 0x2002: /* EN SPACE */
1907 case 0x2003: /* EM SPACE */
1908 case 0x2004: /* THREE-PER-EM SPACE */
1909 case 0x2005: /* FOUR-PER-EM SPACE */
1910 case 0x2006: /* SIX-PER-EM SPACE */
1911 case 0x2007: /* FIGURE SPACE */
1912 case 0x2008: /* PUNCTUATION SPACE */
1913 case 0x2009: /* THIN SPACE */
1914 case 0x200A: /* HAIR SPACE */
1915 case 0x202f: /* NARROW NO-BREAK SPACE */
1916 case 0x205f: /* MEDIUM MATHEMATICAL SPACE */
1917 case 0x3000: /* IDEOGRAPHIC SPACE */
1926 if (OK == (d == OP_HSPACE))
1928 if (codevalue == OP_HSPACE_EXTRA + OP_TYPEPOSUPTO)
1930 active_count--; /* Remove non-match possibility */
1931 next_active_state--;
1933 if (++count >= GET2(code, 1))
1934 { ADD_NEW_DATA(-(state_offset + 4), 0, 0); }
1936 { ADD_NEW_DATA(-state_offset, count, 0); }
1941 /* ========================================================================== */
1942 /* These opcodes are followed by a character that is usually compared
1943 to the current subject character; it is loaded into d. We still get
1944 here even if there is no subject character, because in some cases zero
1945 repetitions are permitted. */
1947 /*-----------------------------------------------------------------*/
1949 if (clen > 0 && c == d) { ADD_NEW(state_offset + dlen + 1, 0); }
1952 /*-----------------------------------------------------------------*/
1954 if (clen == 0) break;
1959 if (c == d) { ADD_NEW(state_offset + dlen + 1, 0); } else
1961 unsigned int othercase;
1962 if (c < 128) othercase = fcc[c]; else
1964 /* If we have Unicode property support, we can use it to test the
1965 other case of the character. */
1968 othercase = UCD_OTHERCASE(c);
1970 othercase = NOTACHAR;
1973 if (d == othercase) { ADD_NEW(state_offset + dlen + 1, 0); }
1977 #endif /* SUPPORT_UTF8 */
1979 /* Non-UTF-8 mode */
1981 if (lcc[c] == lcc[d]) { ADD_NEW(state_offset + 2, 0); }
1987 /*-----------------------------------------------------------------*/
1988 /* This is a tricky one because it can match more than one character.
1989 Find out how many characters to skip, and then set up a negative state
1990 to wait for them to pass before continuing. */
1993 if (clen > 0 && UCD_CATEGORY(c) != ucp_M)
1995 const uschar *nptr = ptr + clen;
1997 while (nptr < end_subject)
2000 GETCHARLEN(c, nptr, nclen);
2001 if (UCD_CATEGORY(c) != ucp_M) break;
2005 ADD_NEW_DATA(-(state_offset + 1), 0, ncount);
2010 /*-----------------------------------------------------------------*/
2011 /* This is a tricky like EXTUNI because it too can match more than one
2012 character (when CR is followed by LF). In this case, set up a negative
2013 state to wait for one character to pass before continuing. */
2016 if (clen > 0) switch(c)
2023 if ((md->moptions & PCRE_BSR_ANYCRLF) != 0) break;
2026 ADD_NEW(state_offset + 1, 0);
2030 if (ptr + 1 < end_subject && ptr[1] == 0x0a)
2032 ADD_NEW_DATA(-(state_offset + 1), 0, 1);
2036 ADD_NEW(state_offset + 1, 0);
2042 /*-----------------------------------------------------------------*/
2044 if (clen > 0) switch(c)
2056 ADD_NEW(state_offset + 1, 0);
2061 /*-----------------------------------------------------------------*/
2063 if (clen > 0) switch(c)
2072 ADD_NEW(state_offset + 1, 0);
2079 /*-----------------------------------------------------------------*/
2081 if (clen > 0) switch(c)
2084 case 0x20: /* SPACE */
2085 case 0xa0: /* NBSP */
2086 case 0x1680: /* OGHAM SPACE MARK */
2087 case 0x180e: /* MONGOLIAN VOWEL SEPARATOR */
2088 case 0x2000: /* EN QUAD */
2089 case 0x2001: /* EM QUAD */
2090 case 0x2002: /* EN SPACE */
2091 case 0x2003: /* EM SPACE */
2092 case 0x2004: /* THREE-PER-EM SPACE */
2093 case 0x2005: /* FOUR-PER-EM SPACE */
2094 case 0x2006: /* SIX-PER-EM SPACE */
2095 case 0x2007: /* FIGURE SPACE */
2096 case 0x2008: /* PUNCTUATION SPACE */
2097 case 0x2009: /* THIN SPACE */
2098 case 0x200A: /* HAIR SPACE */
2099 case 0x202f: /* NARROW NO-BREAK SPACE */
2100 case 0x205f: /* MEDIUM MATHEMATICAL SPACE */
2101 case 0x3000: /* IDEOGRAPHIC SPACE */
2105 ADD_NEW(state_offset + 1, 0);
2110 /*-----------------------------------------------------------------*/
2112 if (clen > 0) switch(c)
2115 case 0x20: /* SPACE */
2116 case 0xa0: /* NBSP */
2117 case 0x1680: /* OGHAM SPACE MARK */
2118 case 0x180e: /* MONGOLIAN VOWEL SEPARATOR */
2119 case 0x2000: /* EN QUAD */
2120 case 0x2001: /* EM QUAD */
2121 case 0x2002: /* EN SPACE */
2122 case 0x2003: /* EM SPACE */
2123 case 0x2004: /* THREE-PER-EM SPACE */
2124 case 0x2005: /* FOUR-PER-EM SPACE */
2125 case 0x2006: /* SIX-PER-EM SPACE */
2126 case 0x2007: /* FIGURE SPACE */
2127 case 0x2008: /* PUNCTUATION SPACE */
2128 case 0x2009: /* THIN SPACE */
2129 case 0x200A: /* HAIR SPACE */
2130 case 0x202f: /* NARROW NO-BREAK SPACE */
2131 case 0x205f: /* MEDIUM MATHEMATICAL SPACE */
2132 case 0x3000: /* IDEOGRAPHIC SPACE */
2133 ADD_NEW(state_offset + 1, 0);
2138 /*-----------------------------------------------------------------*/
2139 /* Match a negated single character. This is only used for one-byte
2140 characters, that is, we know that d < 256. The character we are
2141 checking (c) can be multibyte. */
2146 unsigned int otherd = ((ims & PCRE_CASELESS) != 0)? fcc[d] : d;
2147 if (c != d && c != otherd) { ADD_NEW(state_offset + dlen + 1, 0); }
2151 /*-----------------------------------------------------------------*/
2158 count = current_state->count; /* Already matched */
2159 if (count > 0) { ADD_ACTIVE(state_offset + dlen + 1, 0); }
2162 unsigned int otherd = NOTACHAR;
2163 if ((ims & PCRE_CASELESS) != 0)
2166 if (utf8 && d >= 128)
2169 otherd = UCD_OTHERCASE(d);
2170 #endif /* SUPPORT_UCP */
2173 #endif /* SUPPORT_UTF8 */
2176 if ((c == d || c == otherd) == (codevalue < OP_NOTSTAR))
2179 (codevalue == OP_POSPLUS || codevalue == OP_NOTPOSPLUS))
2181 active_count--; /* Remove non-match possibility */
2182 next_active_state--;
2185 ADD_NEW(state_offset, count);
2190 /*-----------------------------------------------------------------*/
2195 case OP_NOTMINQUERY:
2196 case OP_NOTPOSQUERY:
2197 ADD_ACTIVE(state_offset + dlen + 1, 0);
2200 unsigned int otherd = NOTACHAR;
2201 if ((ims & PCRE_CASELESS) != 0)
2204 if (utf8 && d >= 128)
2207 otherd = UCD_OTHERCASE(d);
2208 #endif /* SUPPORT_UCP */
2211 #endif /* SUPPORT_UTF8 */
2214 if ((c == d || c == otherd) == (codevalue < OP_NOTSTAR))
2216 if (codevalue == OP_POSQUERY || codevalue == OP_NOTPOSQUERY)
2218 active_count--; /* Remove non-match possibility */
2219 next_active_state--;
2221 ADD_NEW(state_offset + dlen + 1, 0);
2226 /*-----------------------------------------------------------------*/
2233 ADD_ACTIVE(state_offset + dlen + 1, 0);
2236 unsigned int otherd = NOTACHAR;
2237 if ((ims & PCRE_CASELESS) != 0)
2240 if (utf8 && d >= 128)
2243 otherd = UCD_OTHERCASE(d);
2244 #endif /* SUPPORT_UCP */
2247 #endif /* SUPPORT_UTF8 */
2250 if ((c == d || c == otherd) == (codevalue < OP_NOTSTAR))
2252 if (codevalue == OP_POSSTAR || codevalue == OP_NOTPOSSTAR)
2254 active_count--; /* Remove non-match possibility */
2255 next_active_state--;
2257 ADD_NEW(state_offset, 0);
2262 /*-----------------------------------------------------------------*/
2265 count = current_state->count; /* Number already matched */
2268 unsigned int otherd = NOTACHAR;
2269 if ((ims & PCRE_CASELESS) != 0)
2272 if (utf8 && d >= 128)
2275 otherd = UCD_OTHERCASE(d);
2276 #endif /* SUPPORT_UCP */
2279 #endif /* SUPPORT_UTF8 */
2282 if ((c == d || c == otherd) == (codevalue < OP_NOTSTAR))
2284 if (++count >= GET2(code, 1))
2285 { ADD_NEW(state_offset + dlen + 3, 0); }
2287 { ADD_NEW(state_offset, count); }
2292 /*-----------------------------------------------------------------*/
2299 ADD_ACTIVE(state_offset + dlen + 3, 0);
2300 count = current_state->count; /* Number already matched */
2303 unsigned int otherd = NOTACHAR;
2304 if ((ims & PCRE_CASELESS) != 0)
2307 if (utf8 && d >= 128)
2310 otherd = UCD_OTHERCASE(d);
2311 #endif /* SUPPORT_UCP */
2314 #endif /* SUPPORT_UTF8 */
2317 if ((c == d || c == otherd) == (codevalue < OP_NOTSTAR))
2319 if (codevalue == OP_POSUPTO || codevalue == OP_NOTPOSUPTO)
2321 active_count--; /* Remove non-match possibility */
2322 next_active_state--;
2324 if (++count >= GET2(code, 1))
2325 { ADD_NEW(state_offset + dlen + 3, 0); }
2327 { ADD_NEW(state_offset, count); }
2333 /* ========================================================================== */
2334 /* These are the class-handling opcodes */
2340 BOOL isinclass = FALSE;
2341 int next_state_offset;
2342 const uschar *ecode;
2344 /* For a simple class, there is always just a 32-byte table, and we
2345 can set isinclass from it. */
2347 if (codevalue != OP_XCLASS)
2352 isinclass = (c > 255)? (codevalue == OP_NCLASS) :
2353 ((code[1 + c/8] & (1 << (c&7))) != 0);
2357 /* An extended class may have a table or a list of single characters,
2358 ranges, or both, and it may be positive or negative. There's a
2359 function that sorts all this out. */
2363 ecode = code + GET(code, 1);
2364 if (clen > 0) isinclass = _pcre_xclass(c, code + 1 + LINK_SIZE);
2367 /* At this point, isinclass is set for all kinds of class, and ecode
2368 points to the byte after the end of the class. If there is a
2369 quantifier, this is where it will be. */
2371 next_state_offset = (int)(ecode - start_code);
2377 ADD_ACTIVE(next_state_offset + 1, 0);
2378 if (isinclass) { ADD_NEW(state_offset, 0); }
2383 count = current_state->count; /* Already matched */
2384 if (count > 0) { ADD_ACTIVE(next_state_offset + 1, 0); }
2385 if (isinclass) { count++; ADD_NEW(state_offset, count); }
2390 ADD_ACTIVE(next_state_offset + 1, 0);
2391 if (isinclass) { ADD_NEW(next_state_offset + 1, 0); }
2396 count = current_state->count; /* Already matched */
2397 if (count >= GET2(ecode, 1))
2398 { ADD_ACTIVE(next_state_offset + 5, 0); }
2401 int max = GET2(ecode, 3);
2402 if (++count >= max && max != 0) /* Max 0 => no limit */
2403 { ADD_NEW(next_state_offset + 5, 0); }
2405 { ADD_NEW(state_offset, count); }
2410 if (isinclass) { ADD_NEW(next_state_offset, 0); }
2416 /* ========================================================================== */
2417 /* These are the opcodes for fancy brackets of various kinds. We have
2418 to use recursion in order to handle them. The "always failing" assertion
2419 (?!) is optimised to OP_FAIL when compiling, so we have to support that,
2420 though the other "backtracking verbs" are not supported. */
2423 forced_fail++; /* Count FAILs for multiple states */
2429 case OP_ASSERTBACK_NOT:
2432 int local_offsets[2];
2433 int local_workspace[1000];
2434 const uschar *endasscode = code + GET(code, 1);
2436 while (*endasscode == OP_ALT) endasscode += GET(endasscode, 1);
2438 rc = internal_dfa_exec(
2439 md, /* static match data */
2440 code, /* this subexpression's code */
2441 ptr, /* where we currently are */
2442 (int)(ptr - start_subject), /* start offset */
2443 local_offsets, /* offset vector */
2444 sizeof(local_offsets)/sizeof(int), /* size of same */
2445 local_workspace, /* workspace vector */
2446 sizeof(local_workspace)/sizeof(int), /* size of same */
2447 ims, /* the current ims flags */
2448 rlevel, /* function recursion level */
2449 recursing); /* pass on regex recursion */
2451 if (rc == PCRE_ERROR_DFA_UITEM) return rc;
2452 if ((rc >= 0) == (codevalue == OP_ASSERT || codevalue == OP_ASSERTBACK))
2453 { ADD_ACTIVE((int)(endasscode + LINK_SIZE + 1 - start_code), 0); }
2457 /*-----------------------------------------------------------------*/
2461 int local_offsets[1000];
2462 int local_workspace[1000];
2463 int codelink = GET(code, 1);
2466 /* Because of the way auto-callout works during compile, a callout item
2467 is inserted between OP_COND and an assertion condition. This does not
2468 happen for the other conditions. */
2470 if (code[LINK_SIZE+1] == OP_CALLOUT)
2473 if (pcre_callout != NULL)
2475 pcre_callout_block cb;
2476 cb.version = 1; /* Version 1 of the callout block */
2477 cb.callout_number = code[LINK_SIZE+2];
2478 cb.offset_vector = offsets;
2479 cb.subject = (PCRE_SPTR)start_subject;
2480 cb.subject_length = (int)(end_subject - start_subject);
2481 cb.start_match = (int)(current_subject - start_subject);
2482 cb.current_position = (int)(ptr - start_subject);
2483 cb.pattern_position = GET(code, LINK_SIZE + 3);
2484 cb.next_item_length = GET(code, 3 + 2*LINK_SIZE);
2486 cb.capture_last = -1;
2487 cb.callout_data = md->callout_data;
2488 if ((rrc = (*pcre_callout)(&cb)) < 0) return rrc; /* Abandon */
2490 if (rrc > 0) break; /* Fail this thread */
2491 code += _pcre_OP_lengths[OP_CALLOUT]; /* Skip callout data */
2494 condcode = code[LINK_SIZE+1];
2496 /* Back reference conditions are not supported */
2498 if (condcode == OP_CREF || condcode == OP_NCREF)
2499 return PCRE_ERROR_DFA_UCOND;
2501 /* The DEFINE condition is always false */
2503 if (condcode == OP_DEF)
2504 { ADD_ACTIVE(state_offset + codelink + LINK_SIZE + 1, 0); }
2506 /* The only supported version of OP_RREF is for the value RREF_ANY,
2507 which means "test if in any recursion". We can't test for specifically
2510 else if (condcode == OP_RREF || condcode == OP_NRREF)
2512 int value = GET2(code, LINK_SIZE+2);
2513 if (value != RREF_ANY) return PCRE_ERROR_DFA_UCOND;
2515 { ADD_ACTIVE(state_offset + LINK_SIZE + 4, 0); }
2516 else { ADD_ACTIVE(state_offset + codelink + LINK_SIZE + 1, 0); }
2519 /* Otherwise, the condition is an assertion */
2524 const uschar *asscode = code + LINK_SIZE + 1;
2525 const uschar *endasscode = asscode + GET(asscode, 1);
2527 while (*endasscode == OP_ALT) endasscode += GET(endasscode, 1);
2529 rc = internal_dfa_exec(
2530 md, /* fixed match data */
2531 asscode, /* this subexpression's code */
2532 ptr, /* where we currently are */
2533 (int)(ptr - start_subject), /* start offset */
2534 local_offsets, /* offset vector */
2535 sizeof(local_offsets)/sizeof(int), /* size of same */
2536 local_workspace, /* workspace vector */
2537 sizeof(local_workspace)/sizeof(int), /* size of same */
2538 ims, /* the current ims flags */
2539 rlevel, /* function recursion level */
2540 recursing); /* pass on regex recursion */
2542 if (rc == PCRE_ERROR_DFA_UITEM) return rc;
2544 (condcode == OP_ASSERT || condcode == OP_ASSERTBACK))
2545 { ADD_ACTIVE((int)(endasscode + LINK_SIZE + 1 - start_code), 0); }
2547 { ADD_ACTIVE(state_offset + codelink + LINK_SIZE + 1, 0); }
2552 /*-----------------------------------------------------------------*/
2555 int local_offsets[1000];
2556 int local_workspace[1000];
2559 DPRINTF(("%.*sStarting regex recursion %d\n", rlevel*2-2, SP,
2562 rc = internal_dfa_exec(
2563 md, /* fixed match data */
2564 start_code + GET(code, 1), /* this subexpression's code */
2565 ptr, /* where we currently are */
2566 (int)(ptr - start_subject), /* start offset */
2567 local_offsets, /* offset vector */
2568 sizeof(local_offsets)/sizeof(int), /* size of same */
2569 local_workspace, /* workspace vector */
2570 sizeof(local_workspace)/sizeof(int), /* size of same */
2571 ims, /* the current ims flags */
2572 rlevel, /* function recursion level */
2573 recursing + 1); /* regex recurse level */
2575 DPRINTF(("%.*sReturn from regex recursion %d: rc=%d\n", rlevel*2-2, SP,
2576 recursing + 1, rc));
2578 /* Ran out of internal offsets */
2580 if (rc == 0) return PCRE_ERROR_DFA_RECURSE;
2582 /* For each successful matched substring, set up the next state with a
2583 count of characters to skip before trying it. Note that the count is in
2584 characters, not bytes. */
2588 for (rc = rc*2 - 2; rc >= 0; rc -= 2)
2590 const uschar *p = start_subject + local_offsets[rc];
2591 const uschar *pp = start_subject + local_offsets[rc+1];
2592 int charcount = local_offsets[rc+1] - local_offsets[rc];
2593 while (p < pp) if ((*p++ & 0xc0) == 0x80) charcount--;
2596 ADD_NEW_DATA(-(state_offset + LINK_SIZE + 1), 0, (charcount - 1));
2600 ADD_ACTIVE(state_offset + LINK_SIZE + 1, 0);
2604 else if (rc != PCRE_ERROR_NOMATCH) return rc;
2608 /*-----------------------------------------------------------------*/
2611 int local_offsets[2];
2612 int local_workspace[1000];
2614 int rc = internal_dfa_exec(
2615 md, /* fixed match data */
2616 code, /* this subexpression's code */
2617 ptr, /* where we currently are */
2618 (int)(ptr - start_subject), /* start offset */
2619 local_offsets, /* offset vector */
2620 sizeof(local_offsets)/sizeof(int), /* size of same */
2621 local_workspace, /* workspace vector */
2622 sizeof(local_workspace)/sizeof(int), /* size of same */
2623 ims, /* the current ims flags */
2624 rlevel, /* function recursion level */
2625 recursing); /* pass on regex recursion */
2629 const uschar *end_subpattern = code;
2630 int charcount = local_offsets[1] - local_offsets[0];
2631 int next_state_offset, repeat_state_offset;
2633 do { end_subpattern += GET(end_subpattern, 1); }
2634 while (*end_subpattern == OP_ALT);
2636 (int)(end_subpattern - start_code + LINK_SIZE + 1);
2638 /* If the end of this subpattern is KETRMAX or KETRMIN, we must
2639 arrange for the repeat state also to be added to the relevant list.
2640 Calculate the offset, or set -1 for no repeat. */
2642 repeat_state_offset = (*end_subpattern == OP_KETRMAX ||
2643 *end_subpattern == OP_KETRMIN)?
2644 (int)(end_subpattern - start_code - GET(end_subpattern, 1)) : -1;
2646 /* If we have matched an empty string, add the next state at the
2647 current character pointer. This is important so that the duplicate
2648 checking kicks in, which is what breaks infinite loops that match an
2653 ADD_ACTIVE(next_state_offset, 0);
2656 /* Optimization: if there are no more active states, and there
2657 are no new states yet set up, then skip over the subject string
2658 right here, to save looping. Otherwise, set up the new state to swing
2659 into action when the end of the substring is reached. */
2661 else if (i + 1 >= active_count && new_count == 0)
2665 ADD_NEW(next_state_offset, 0);
2667 /* If we are adding a repeat state at the new character position,
2668 we must fudge things so that it is the only current state.
2669 Otherwise, it might be a duplicate of one we processed before, and
2670 that would cause it to be skipped. */
2672 if (repeat_state_offset >= 0)
2674 next_active_state = active_states;
2677 ADD_ACTIVE(repeat_state_offset, 0);
2682 const uschar *p = start_subject + local_offsets[0];
2683 const uschar *pp = start_subject + local_offsets[1];
2684 while (p < pp) if ((*p++ & 0xc0) == 0x80) charcount--;
2685 ADD_NEW_DATA(-next_state_offset, 0, (charcount - 1));
2686 if (repeat_state_offset >= 0)
2687 { ADD_NEW_DATA(-repeat_state_offset, 0, (charcount - 1)); }
2691 else if (rc != PCRE_ERROR_NOMATCH) return rc;
2696 /* ========================================================================== */
2697 /* Handle callouts */
2701 if (pcre_callout != NULL)
2703 pcre_callout_block cb;
2704 cb.version = 1; /* Version 1 of the callout block */
2705 cb.callout_number = code[1];
2706 cb.offset_vector = offsets;
2707 cb.subject = (PCRE_SPTR)start_subject;
2708 cb.subject_length = (int)(end_subject - start_subject);
2709 cb.start_match = (int)(current_subject - start_subject);
2710 cb.current_position = (int)(ptr - start_subject);
2711 cb.pattern_position = GET(code, 2);
2712 cb.next_item_length = GET(code, 2 + LINK_SIZE);
2714 cb.capture_last = -1;
2715 cb.callout_data = md->callout_data;
2716 if ((rrc = (*pcre_callout)(&cb)) < 0) return rrc; /* Abandon */
2719 { ADD_ACTIVE(state_offset + _pcre_OP_lengths[OP_CALLOUT], 0); }
2723 /* ========================================================================== */
2724 default: /* Unsupported opcode */
2725 return PCRE_ERROR_DFA_UITEM;
2728 NEXT_ACTIVE_STATE: continue;
2730 } /* End of loop scanning active states */
2732 /* We have finished the processing at the current subject character. If no
2733 new states have been set for the next character, we have found all the
2734 matches that we are going to find. If we are at the top level and partial
2735 matching has been requested, check for appropriate conditions.
2737 The "forced_ fail" variable counts the number of (*F) encountered for the
2738 character. If it is equal to the original active_count (saved in
2739 workspace[1]) it means that (*F) was found on every active state. In this
2740 case we don't want to give a partial match.
2742 The "could_continue" variable is true if a state could have continued but
2743 for the fact that the end of the subject was reached. */
2747 if (rlevel == 1 && /* Top level, and */
2748 could_continue && /* Some could go on */
2749 forced_fail != workspace[1] && /* Not all forced fail & */
2751 (md->moptions & PCRE_PARTIAL_HARD) != 0 /* Hard partial */
2753 ((md->moptions & PCRE_PARTIAL_SOFT) != 0 && /* Soft partial and */
2754 match_count < 0) /* no matches */
2756 ptr >= end_subject && /* Reached end of subject */
2757 ptr > md->start_used_ptr) /* Inspected non-empty string */
2759 if (offsetcount >= 2)
2761 offsets[0] = (int)(md->start_used_ptr - start_subject);
2762 offsets[1] = (int)(end_subject - start_subject);
2764 match_count = PCRE_ERROR_PARTIAL;
2767 DPRINTF(("%.*sEnd of internal_dfa_exec %d: returning %d\n"
2768 "%.*s---------------------\n\n", rlevel*2-2, SP, rlevel, match_count,
2770 break; /* In effect, "return", but see the comment below */
2773 /* One or more states are active for the next character. */
2775 ptr += clen; /* Advance to next subject character */
2776 } /* Loop to move along the subject string */
2778 /* Control gets here from "break" a few lines above. We do it this way because
2779 if we use "return" above, we have compiler trouble. Some compilers warn if
2780 there's nothing here because they think the function doesn't return a value. On
2781 the other hand, if we put a dummy statement here, some more clever compilers
2782 complain that it can't be reached. Sigh. */
2790 /*************************************************
2791 * Execute a Regular Expression - DFA engine *
2792 *************************************************/
2794 /* This external function applies a compiled re to a subject string using a DFA
2795 engine. This function calls the internal function multiple times if the pattern
2799 argument_re points to the compiled expression
2800 extra_data points to extra data or is NULL
2801 subject points to the subject string
2802 length length of subject string (may contain binary zeros)
2803 start_offset where to start in the subject string
2805 offsets vector of match offsets
2806 offsetcount size of same
2807 workspace workspace vector
2808 wscount size of same
2810 Returns: > 0 => number of match offset pairs placed in offsets
2811 = 0 => offsets overflowed; longest matches are present
2812 -1 => failed to match
2813 < -1 => some kind of unexpected problem
2816 PCRE_EXP_DEFN int PCRE_CALL_CONVENTION
2817 pcre_dfa_exec(const pcre *argument_re, const pcre_extra *extra_data,
2818 const char *subject, int length, int start_offset, int options, int *offsets,
2819 int offsetcount, int *workspace, int wscount)
2821 real_pcre *re = (real_pcre *)argument_re;
2822 dfa_match_data match_block;
2823 dfa_match_data *md = &match_block;
2824 BOOL utf8, anchored, startline, firstline;
2825 const uschar *current_subject, *end_subject, *lcc;
2827 pcre_study_data internal_study;
2828 const pcre_study_data *study = NULL;
2829 real_pcre internal_re;
2831 const uschar *req_byte_ptr;
2832 const uschar *start_bits = NULL;
2833 BOOL first_byte_caseless = FALSE;
2834 BOOL req_byte_caseless = FALSE;
2835 int first_byte = -1;
2840 /* Plausibility checks */
2842 if ((options & ~PUBLIC_DFA_EXEC_OPTIONS) != 0) return PCRE_ERROR_BADOPTION;
2843 if (re == NULL || subject == NULL || workspace == NULL ||
2844 (offsets == NULL && offsetcount > 0)) return PCRE_ERROR_NULL;
2845 if (offsetcount < 0) return PCRE_ERROR_BADCOUNT;
2846 if (wscount < 20) return PCRE_ERROR_DFA_WSSIZE;
2847 if (start_offset < 0 || start_offset > length) return PCRE_ERROR_BADOFFSET;
2849 /* We need to find the pointer to any study data before we test for byte
2850 flipping, so we scan the extra_data block first. This may set two fields in the
2851 match block, so we must initialize them beforehand. However, the other fields
2852 in the match block must not be set until after the byte flipping. */
2854 md->tables = re->tables;
2855 md->callout_data = NULL;
2857 if (extra_data != NULL)
2859 unsigned int flags = extra_data->flags;
2860 if ((flags & PCRE_EXTRA_STUDY_DATA) != 0)
2861 study = (const pcre_study_data *)extra_data->study_data;
2862 if ((flags & PCRE_EXTRA_MATCH_LIMIT) != 0) return PCRE_ERROR_DFA_UMLIMIT;
2863 if ((flags & PCRE_EXTRA_MATCH_LIMIT_RECURSION) != 0)
2864 return PCRE_ERROR_DFA_UMLIMIT;
2865 if ((flags & PCRE_EXTRA_CALLOUT_DATA) != 0)
2866 md->callout_data = extra_data->callout_data;
2867 if ((flags & PCRE_EXTRA_TABLES) != 0)
2868 md->tables = extra_data->tables;
2871 /* Check that the first field in the block is the magic number. If it is not,
2872 test for a regex that was compiled on a host of opposite endianness. If this is
2873 the case, flipped values are put in internal_re and internal_study if there was
2876 if (re->magic_number != MAGIC_NUMBER)
2878 re = _pcre_try_flipped(re, &internal_re, study, &internal_study);
2879 if (re == NULL) return PCRE_ERROR_BADMAGIC;
2880 if (study != NULL) study = &internal_study;
2883 /* Set some local values */
2885 current_subject = (const unsigned char *)subject + start_offset;
2886 end_subject = (const unsigned char *)subject + length;
2887 req_byte_ptr = current_subject - 1;
2890 utf8 = (re->options & PCRE_UTF8) != 0;
2895 anchored = (options & (PCRE_ANCHORED|PCRE_DFA_RESTART)) != 0 ||
2896 (re->options & PCRE_ANCHORED) != 0;
2898 /* The remaining fixed data for passing around. */
2900 md->start_code = (const uschar *)argument_re +
2901 re->name_table_offset + re->name_count * re->name_entry_size;
2902 md->start_subject = (const unsigned char *)subject;
2903 md->end_subject = end_subject;
2904 md->start_offset = start_offset;
2905 md->moptions = options;
2906 md->poptions = re->options;
2908 /* If the BSR option is not set at match time, copy what was set
2911 if ((md->moptions & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) == 0)
2913 if ((re->options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) != 0)
2914 md->moptions |= re->options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE);
2916 else md->moptions |= PCRE_BSR_ANYCRLF;
2920 /* Handle different types of newline. The three bits give eight cases. If
2921 nothing is set at run time, whatever was used at compile time applies. */
2923 switch ((((options & PCRE_NEWLINE_BITS) == 0)? re->options : (pcre_uint32)options) &
2926 case 0: newline = NEWLINE; break; /* Compile-time default */
2927 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
2928 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
2929 case PCRE_NEWLINE_CR+
2930 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
2931 case PCRE_NEWLINE_ANY: newline = -1; break;
2932 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
2933 default: return PCRE_ERROR_BADNEWLINE;
2938 md->nltype = NLTYPE_ANYCRLF;
2940 else if (newline < 0)
2942 md->nltype = NLTYPE_ANY;
2946 md->nltype = NLTYPE_FIXED;
2950 md->nl[0] = (newline >> 8) & 255;
2951 md->nl[1] = newline & 255;
2956 md->nl[0] = newline;
2960 /* Check a UTF-8 string if required. Unfortunately there's no way of passing
2961 back the character offset. */
2964 if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0)
2967 if ((tb = _pcre_valid_utf8((uschar *)subject, length)) >= 0)
2968 return (tb == length && (options & PCRE_PARTIAL_HARD) != 0)?
2969 PCRE_ERROR_SHORTUTF8 : PCRE_ERROR_BADUTF8;
2970 if (start_offset > 0 && start_offset < length)
2972 tb = ((USPTR)subject)[start_offset] & 0xc0;
2973 if (tb == 0x80) return PCRE_ERROR_BADUTF8_OFFSET;
2978 /* If the exec call supplied NULL for tables, use the inbuilt ones. This
2979 is a feature that makes it possible to save compiled regex and re-use them
2980 in other programs later. */
2982 if (md->tables == NULL) md->tables = _pcre_default_tables;
2984 /* The lower casing table and the "must be at the start of a line" flag are
2985 used in a loop when finding where to start. */
2987 lcc = md->tables + lcc_offset;
2988 startline = (re->flags & PCRE_STARTLINE) != 0;
2989 firstline = (re->options & PCRE_FIRSTLINE) != 0;
2991 /* Set up the first character to match, if available. The first_byte value is
2992 never set for an anchored regular expression, but the anchoring may be forced
2993 at run time, so we have to test for anchoring. The first char may be unset for
2994 an unanchored pattern, of course. If there's no first char and the pattern was
2995 studied, there may be a bitmap of possible first characters. */
2999 if ((re->flags & PCRE_FIRSTSET) != 0)
3001 first_byte = re->first_byte & 255;
3002 if ((first_byte_caseless = ((re->first_byte & REQ_CASELESS) != 0)) == TRUE)
3003 first_byte = lcc[first_byte];
3007 if (!startline && study != NULL &&
3008 (study->flags & PCRE_STUDY_MAPPED) != 0)
3009 start_bits = study->start_bits;
3013 /* For anchored or unanchored matches, there may be a "last known required
3016 if ((re->flags & PCRE_REQCHSET) != 0)
3018 req_byte = re->req_byte & 255;
3019 req_byte_caseless = (re->req_byte & REQ_CASELESS) != 0;
3020 req_byte2 = (md->tables + fcc_offset)[req_byte]; /* case flipped */
3023 /* Call the main matching function, looping for a non-anchored regex after a
3024 failed match. If not restarting, perform certain optimizations at the start of
3031 if ((options & PCRE_DFA_RESTART) == 0)
3033 const uschar *save_end_subject = end_subject;
3035 /* If firstline is TRUE, the start of the match is constrained to the first
3036 line of a multiline string. Implement this by temporarily adjusting
3037 end_subject so that we stop scanning at a newline. If the match fails at
3038 the newline, later code breaks this loop. */
3042 USPTR t = current_subject;
3046 while (t < md->end_subject && !IS_NEWLINE(t))
3049 while (t < end_subject && (*t & 0xc0) == 0x80) t++;
3054 while (t < md->end_subject && !IS_NEWLINE(t)) t++;
3058 /* There are some optimizations that avoid running the match if a known
3059 starting point is not found. However, there is an option that disables
3060 these, for testing and for ensuring that all callouts do actually occur.
3061 The option can be set in the regex by (*NO_START_OPT) or passed in
3062 match-time options. */
3064 if (((options | re->options) & PCRE_NO_START_OPTIMIZE) == 0)
3066 /* Advance to a known first byte. */
3068 if (first_byte >= 0)
3070 if (first_byte_caseless)
3071 while (current_subject < end_subject &&
3072 lcc[*current_subject] != first_byte)
3075 while (current_subject < end_subject &&
3076 *current_subject != first_byte)
3080 /* Or to just after a linebreak for a multiline match if possible */
3084 if (current_subject > md->start_subject + start_offset)
3089 while (current_subject < end_subject &&
3090 !WAS_NEWLINE(current_subject))
3093 while(current_subject < end_subject &&
3094 (*current_subject & 0xc0) == 0x80)
3100 while (current_subject < end_subject && !WAS_NEWLINE(current_subject))
3103 /* If we have just passed a CR and the newline option is ANY or
3104 ANYCRLF, and we are now at a LF, advance the match position by one
3107 if (current_subject[-1] == CHAR_CR &&
3108 (md->nltype == NLTYPE_ANY || md->nltype == NLTYPE_ANYCRLF) &&
3109 current_subject < end_subject &&
3110 *current_subject == CHAR_NL)
3115 /* Or to a non-unique first char after study */
3117 else if (start_bits != NULL)
3119 while (current_subject < end_subject)
3121 register unsigned int c = *current_subject;
3122 if ((start_bits[c/8] & (1 << (c&7))) == 0)
3127 while(current_subject < end_subject &&
3128 (*current_subject & 0xc0) == 0x80) current_subject++;
3136 /* Restore fudged end_subject */
3138 end_subject = save_end_subject;
3140 /* The following two optimizations are disabled for partial matching or if
3141 disabling is explicitly requested (and of course, by the test above, this
3142 code is not obeyed when restarting after a partial match). */
3144 if ((options & PCRE_NO_START_OPTIMIZE) == 0 &&
3145 (options & (PCRE_PARTIAL_HARD|PCRE_PARTIAL_SOFT)) == 0)
3147 /* If the pattern was studied, a minimum subject length may be set. This
3148 is a lower bound; no actual string of that length may actually match the
3149 pattern. Although the value is, strictly, in characters, we treat it as
3150 bytes to avoid spending too much time in this optimization. */
3152 if (study != NULL && (study->flags & PCRE_STUDY_MINLEN) != 0 &&
3153 (pcre_uint32)(end_subject - current_subject) < study->minlength)
3154 return PCRE_ERROR_NOMATCH;
3156 /* If req_byte is set, we know that that character must appear in the
3157 subject for the match to succeed. If the first character is set, req_byte
3158 must be later in the subject; otherwise the test starts at the match
3159 point. This optimization can save a huge amount of work in patterns with
3160 nested unlimited repeats that aren't going to match. Writing separate
3161 code for cased/caseless versions makes it go faster, as does using an
3162 autoincrement and backing off on a match.
3164 HOWEVER: when the subject string is very, very long, searching to its end
3165 can take a long time, and give bad performance on quite ordinary
3166 patterns. This showed up when somebody was matching /^C/ on a 32-megabyte
3167 string... so we don't do this when the string is sufficiently long. */
3169 if (req_byte >= 0 && end_subject - current_subject < REQ_BYTE_MAX)
3171 register const uschar *p = current_subject + ((first_byte >= 0)? 1 : 0);
3173 /* We don't need to repeat the search if we haven't yet reached the
3174 place we found it at last time. */
3176 if (p > req_byte_ptr)
3178 if (req_byte_caseless)
3180 while (p < end_subject)
3182 register int pp = *p++;
3183 if (pp == req_byte || pp == req_byte2) { p--; break; }
3188 while (p < end_subject)
3190 if (*p++ == req_byte) { p--; break; }
3194 /* If we can't find the required character, break the matching loop,
3195 which will cause a return or PCRE_ERROR_NOMATCH. */
3197 if (p >= end_subject) break;
3199 /* If we have found the required character, save the point where we
3200 found it, so that we don't search again next time round the loop if
3201 the start hasn't passed this character yet. */
3207 } /* End of optimizations that are done when not restarting */
3209 /* OK, now we can do the business */
3211 md->start_used_ptr = current_subject;
3213 rc = internal_dfa_exec(
3214 md, /* fixed match data */
3215 md->start_code, /* this subexpression's code */
3216 current_subject, /* where we currently are */
3217 start_offset, /* start offset in subject */
3218 offsets, /* offset vector */
3219 offsetcount, /* size of same */
3220 workspace, /* workspace vector */
3221 wscount, /* size of same */
3222 re->options & (PCRE_CASELESS|PCRE_MULTILINE|PCRE_DOTALL), /* ims flags */
3223 0, /* function recurse level */
3224 0); /* regex recurse level */
3226 /* Anything other than "no match" means we are done, always; otherwise, carry
3227 on only if not anchored. */
3229 if (rc != PCRE_ERROR_NOMATCH || anchored) return rc;
3231 /* Advance to the next subject character unless we are at the end of a line
3232 and firstline is set. */
3234 if (firstline && IS_NEWLINE(current_subject)) break;
3238 while (current_subject < end_subject && (*current_subject & 0xc0) == 0x80)
3241 if (current_subject > end_subject) break;
3243 /* If we have just passed a CR and we are now at a LF, and the pattern does
3244 not contain any explicit matches for \r or \n, and the newline option is CRLF
3245 or ANY or ANYCRLF, advance the match position by one more character. */
3247 if (current_subject[-1] == CHAR_CR &&
3248 current_subject < end_subject &&
3249 *current_subject == CHAR_NL &&
3250 (re->flags & PCRE_HASCRORLF) == 0 &&
3251 (md->nltype == NLTYPE_ANY ||
3252 md->nltype == NLTYPE_ANYCRLF ||
3256 } /* "Bumpalong" loop */
3258 return PCRE_ERROR_NOMATCH;
3261 /* End of pcre_dfa_exec.c */