/[pcre]/code/trunk/pcre_study.c
ViewVC logotype

Contents of /code/trunk/pcre_study.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 538 - (show annotations) (download)
Wed Jun 9 19:30:57 2010 UTC (4 years, 4 months ago) by ph10
File MIME type: text/plain
File size: 30436 byte(s)
Fix pcre_study() problem with non-C-locale chartables in UTF-8 mode.

1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
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.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2010 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_study(), along with local
42 supporting functions. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #include "pcre_internal.h"
50
51 #define SET_BIT(c) start_bits[c/8] |= (1 << (c&7))
52
53 /* Returns from set_start_bits() */
54
55 enum { SSB_FAIL, SSB_DONE, SSB_CONTINUE };
56
57
58
59 /*************************************************
60 * Find the minimum subject length for a group *
61 *************************************************/
62
63 /* Scan a parenthesized group and compute the minimum length of subject that
64 is needed to match it. This is a lower bound; it does not mean there is a
65 string of that length that matches. In UTF8 mode, the result is in characters
66 rather than bytes.
67
68 Arguments:
69 code pointer to start of group (the bracket)
70 startcode pointer to start of the whole pattern
71 options the compiling options
72
73 Returns: the minimum length
74 -1 if \C was encountered
75 -2 internal error (missing capturing bracket)
76 */
77
78 static int
79 find_minlength(const uschar *code, const uschar *startcode, int options)
80 {
81 int length = -1;
82 BOOL utf8 = (options & PCRE_UTF8) != 0;
83 BOOL had_recurse = FALSE;
84 register int branchlength = 0;
85 register uschar *cc = (uschar *)code + 1 + LINK_SIZE;
86
87 if (*code == OP_CBRA || *code == OP_SCBRA) cc += 2;
88
89 /* Scan along the opcodes for this branch. If we get to the end of the
90 branch, check the length against that of the other branches. */
91
92 for (;;)
93 {
94 int d, min;
95 uschar *cs, *ce;
96 register int op = *cc;
97
98 switch (op)
99 {
100 case OP_COND:
101 case OP_SCOND:
102
103 /* If there is only one branch in a condition, the implied branch has zero
104 length, so we don't add anything. This covers the DEFINE "condition"
105 automatically. */
106
107 cs = cc + GET(cc, 1);
108 if (*cs != OP_ALT)
109 {
110 cc = cs + 1 + LINK_SIZE;
111 break;
112 }
113
114 /* Otherwise we can fall through and treat it the same as any other
115 subpattern. */
116
117 case OP_CBRA:
118 case OP_SCBRA:
119 case OP_BRA:
120 case OP_SBRA:
121 case OP_ONCE:
122 d = find_minlength(cc, startcode, options);
123 if (d < 0) return d;
124 branchlength += d;
125 do cc += GET(cc, 1); while (*cc == OP_ALT);
126 cc += 1 + LINK_SIZE;
127 break;
128
129 /* Reached end of a branch; if it's a ket it is the end of a nested
130 call. If it's ALT it is an alternation in a nested call. If it is
131 END it's the end of the outer call. All can be handled by the same code. */
132
133 case OP_ALT:
134 case OP_KET:
135 case OP_KETRMAX:
136 case OP_KETRMIN:
137 case OP_END:
138 if (length < 0 || (!had_recurse && branchlength < length))
139 length = branchlength;
140 if (*cc != OP_ALT) return length;
141 cc += 1 + LINK_SIZE;
142 branchlength = 0;
143 had_recurse = FALSE;
144 break;
145
146 /* Skip over assertive subpatterns */
147
148 case OP_ASSERT:
149 case OP_ASSERT_NOT:
150 case OP_ASSERTBACK:
151 case OP_ASSERTBACK_NOT:
152 do cc += GET(cc, 1); while (*cc == OP_ALT);
153 /* Fall through */
154
155 /* Skip over things that don't match chars */
156
157 case OP_REVERSE:
158 case OP_CREF:
159 case OP_NCREF:
160 case OP_RREF:
161 case OP_NRREF:
162 case OP_DEF:
163 case OP_OPT:
164 case OP_CALLOUT:
165 case OP_SOD:
166 case OP_SOM:
167 case OP_EOD:
168 case OP_EODN:
169 case OP_CIRC:
170 case OP_DOLL:
171 case OP_NOT_WORD_BOUNDARY:
172 case OP_WORD_BOUNDARY:
173 cc += _pcre_OP_lengths[*cc];
174 break;
175
176 /* Skip over a subpattern that has a {0} or {0,x} quantifier */
177
178 case OP_BRAZERO:
179 case OP_BRAMINZERO:
180 case OP_SKIPZERO:
181 cc += _pcre_OP_lengths[*cc];
182 do cc += GET(cc, 1); while (*cc == OP_ALT);
183 cc += 1 + LINK_SIZE;
184 break;
185
186 /* Handle literal characters and + repetitions */
187
188 case OP_CHAR:
189 case OP_CHARNC:
190 case OP_NOT:
191 case OP_PLUS:
192 case OP_MINPLUS:
193 case OP_POSPLUS:
194 case OP_NOTPLUS:
195 case OP_NOTMINPLUS:
196 case OP_NOTPOSPLUS:
197 branchlength++;
198 cc += 2;
199 #ifdef SUPPORT_UTF8
200 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
201 #endif
202 break;
203
204 case OP_TYPEPLUS:
205 case OP_TYPEMINPLUS:
206 case OP_TYPEPOSPLUS:
207 branchlength++;
208 cc += (cc[1] == OP_PROP || cc[1] == OP_NOTPROP)? 4 : 2;
209 break;
210
211 /* Handle exact repetitions. The count is already in characters, but we
212 need to skip over a multibyte character in UTF8 mode. */
213
214 case OP_EXACT:
215 case OP_NOTEXACT:
216 branchlength += GET2(cc,1);
217 cc += 4;
218 #ifdef SUPPORT_UTF8
219 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
220 #endif
221 break;
222
223 case OP_TYPEEXACT:
224 branchlength += GET2(cc,1);
225 cc += (cc[3] == OP_PROP || cc[3] == OP_NOTPROP)? 6 : 4;
226 break;
227
228 /* Handle single-char non-literal matchers */
229
230 case OP_PROP:
231 case OP_NOTPROP:
232 cc += 2;
233 /* Fall through */
234
235 case OP_NOT_DIGIT:
236 case OP_DIGIT:
237 case OP_NOT_WHITESPACE:
238 case OP_WHITESPACE:
239 case OP_NOT_WORDCHAR:
240 case OP_WORDCHAR:
241 case OP_ANY:
242 case OP_ALLANY:
243 case OP_EXTUNI:
244 case OP_HSPACE:
245 case OP_NOT_HSPACE:
246 case OP_VSPACE:
247 case OP_NOT_VSPACE:
248 branchlength++;
249 cc++;
250 break;
251
252 /* "Any newline" might match two characters */
253
254 case OP_ANYNL:
255 branchlength += 2;
256 cc++;
257 break;
258
259 /* The single-byte matcher means we can't proceed in UTF-8 mode */
260
261 case OP_ANYBYTE:
262 #ifdef SUPPORT_UTF8
263 if (utf8) return -1;
264 #endif
265 branchlength++;
266 cc++;
267 break;
268
269 /* For repeated character types, we have to test for \p and \P, which have
270 an extra two bytes of parameters. */
271
272 case OP_TYPESTAR:
273 case OP_TYPEMINSTAR:
274 case OP_TYPEQUERY:
275 case OP_TYPEMINQUERY:
276 case OP_TYPEPOSSTAR:
277 case OP_TYPEPOSQUERY:
278 if (cc[1] == OP_PROP || cc[1] == OP_NOTPROP) cc += 2;
279 cc += _pcre_OP_lengths[op];
280 break;
281
282 case OP_TYPEUPTO:
283 case OP_TYPEMINUPTO:
284 case OP_TYPEPOSUPTO:
285 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
286 cc += _pcre_OP_lengths[op];
287 break;
288
289 /* Check a class for variable quantification */
290
291 #ifdef SUPPORT_UTF8
292 case OP_XCLASS:
293 cc += GET(cc, 1) - 33;
294 /* Fall through */
295 #endif
296
297 case OP_CLASS:
298 case OP_NCLASS:
299 cc += 33;
300
301 switch (*cc)
302 {
303 case OP_CRPLUS:
304 case OP_CRMINPLUS:
305 branchlength++;
306 /* Fall through */
307
308 case OP_CRSTAR:
309 case OP_CRMINSTAR:
310 case OP_CRQUERY:
311 case OP_CRMINQUERY:
312 cc++;
313 break;
314
315 case OP_CRRANGE:
316 case OP_CRMINRANGE:
317 branchlength += GET2(cc,1);
318 cc += 5;
319 break;
320
321 default:
322 branchlength++;
323 break;
324 }
325 break;
326
327 /* Backreferences and subroutine calls are treated in the same way: we find
328 the minimum length for the subpattern. A recursion, however, causes an
329 a flag to be set that causes the length of this branch to be ignored. The
330 logic is that a recursion can only make sense if there is another
331 alternation that stops the recursing. That will provide the minimum length
332 (when no recursion happens). A backreference within the group that it is
333 referencing behaves in the same way.
334
335 If PCRE_JAVASCRIPT_COMPAT is set, a backreference to an unset bracket
336 matches an empty string (by default it causes a matching failure), so in
337 that case we must set the minimum length to zero. */
338
339 case OP_REF:
340 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0)
341 {
342 ce = cs = (uschar *)_pcre_find_bracket(startcode, utf8, GET2(cc, 1));
343 if (cs == NULL) return -2;
344 do ce += GET(ce, 1); while (*ce == OP_ALT);
345 if (cc > cs && cc < ce)
346 {
347 d = 0;
348 had_recurse = TRUE;
349 }
350 else d = find_minlength(cs, startcode, options);
351 }
352 else d = 0;
353 cc += 3;
354
355 /* Handle repeated back references */
356
357 switch (*cc)
358 {
359 case OP_CRSTAR:
360 case OP_CRMINSTAR:
361 case OP_CRQUERY:
362 case OP_CRMINQUERY:
363 min = 0;
364 cc++;
365 break;
366
367 case OP_CRRANGE:
368 case OP_CRMINRANGE:
369 min = GET2(cc, 1);
370 cc += 5;
371 break;
372
373 default:
374 min = 1;
375 break;
376 }
377
378 branchlength += min * d;
379 break;
380
381 case OP_RECURSE:
382 cs = ce = (uschar *)startcode + GET(cc, 1);
383 if (cs == NULL) return -2;
384 do ce += GET(ce, 1); while (*ce == OP_ALT);
385 if (cc > cs && cc < ce)
386 had_recurse = TRUE;
387 else
388 branchlength += find_minlength(cs, startcode, options);
389 cc += 1 + LINK_SIZE;
390 break;
391
392 /* Anything else does not or need not match a character. We can get the
393 item's length from the table, but for those that can match zero occurrences
394 of a character, we must take special action for UTF-8 characters. */
395
396 case OP_UPTO:
397 case OP_NOTUPTO:
398 case OP_MINUPTO:
399 case OP_NOTMINUPTO:
400 case OP_POSUPTO:
401 case OP_STAR:
402 case OP_MINSTAR:
403 case OP_NOTMINSTAR:
404 case OP_POSSTAR:
405 case OP_NOTPOSSTAR:
406 case OP_QUERY:
407 case OP_MINQUERY:
408 case OP_NOTMINQUERY:
409 case OP_POSQUERY:
410 case OP_NOTPOSQUERY:
411 cc += _pcre_OP_lengths[op];
412 #ifdef SUPPORT_UTF8
413 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
414 #endif
415 break;
416
417 /* Skip these, but we need to add in the name length. */
418
419 case OP_MARK:
420 case OP_PRUNE_ARG:
421 case OP_SKIP_ARG:
422 case OP_THEN_ARG:
423 cc += _pcre_OP_lengths[op] + cc[1];
424 break;
425
426 /* For the record, these are the opcodes that are matched by "default":
427 OP_ACCEPT, OP_CLOSE, OP_COMMIT, OP_FAIL, OP_PRUNE, OP_SET_SOM, OP_SKIP,
428 OP_THEN. */
429
430 default:
431 cc += _pcre_OP_lengths[op];
432 break;
433 }
434 }
435 /* Control never gets here */
436 }
437
438
439
440 /*************************************************
441 * Set a bit and maybe its alternate case *
442 *************************************************/
443
444 /* Given a character, set its first byte's bit in the table, and also the
445 corresponding bit for the other version of a letter if we are caseless. In
446 UTF-8 mode, for characters greater than 127, we can only do the caseless thing
447 when Unicode property support is available.
448
449 Arguments:
450 start_bits points to the bit map
451 p points to the character
452 caseless the caseless flag
453 cd the block with char table pointers
454 utf8 TRUE for UTF-8 mode
455
456 Returns: pointer after the character
457 */
458
459 static const uschar *
460 set_table_bit(uschar *start_bits, const uschar *p, BOOL caseless,
461 compile_data *cd, BOOL utf8)
462 {
463 unsigned int c = *p;
464
465 SET_BIT(c);
466
467 #ifdef SUPPORT_UTF8
468 if (utf8 && c > 127)
469 {
470 GETCHARINC(c, p);
471 #ifdef SUPPORT_UCP
472 if (caseless)
473 {
474 uschar buff[8];
475 c = UCD_OTHERCASE(c);
476 (void)_pcre_ord2utf8(c, buff);
477 SET_BIT(buff[0]);
478 }
479 #endif
480 return p;
481 }
482 #endif
483
484 /* Not UTF-8 mode, or character is less than 127. */
485
486 if (caseless && (cd->ctypes[c] & ctype_letter) != 0) SET_BIT(cd->fcc[c]);
487 return p + 1;
488 }
489
490
491
492 /*************************************************
493 * Create bitmap of starting bytes *
494 *************************************************/
495
496 /* This function scans a compiled unanchored expression recursively and
497 attempts to build a bitmap of the set of possible starting bytes. As time goes
498 by, we may be able to get more clever at doing this. The SSB_CONTINUE return is
499 useful for parenthesized groups in patterns such as (a*)b where the group
500 provides some optional starting bytes but scanning must continue at the outer
501 level to find at least one mandatory byte. At the outermost level, this
502 function fails unless the result is SSB_DONE.
503
504 Arguments:
505 code points to an expression
506 start_bits points to a 32-byte table, initialized to 0
507 caseless the current state of the caseless flag
508 utf8 TRUE if in UTF-8 mode
509 cd the block with char table pointers
510
511 Returns: SSB_FAIL => Failed to find any starting bytes
512 SSB_DONE => Found mandatory starting bytes
513 SSB_CONTINUE => Found optional starting bytes
514 */
515
516 static int
517 set_start_bits(const uschar *code, uschar *start_bits, BOOL caseless,
518 BOOL utf8, compile_data *cd)
519 {
520 register int c;
521 int yield = SSB_DONE;
522 int table_limit = utf8? 16:32;
523
524 #if 0
525 /* ========================================================================= */
526 /* The following comment and code was inserted in January 1999. In May 2006,
527 when it was observed to cause compiler warnings about unused values, I took it
528 out again. If anybody is still using OS/2, they will have to put it back
529 manually. */
530
531 /* This next statement and the later reference to dummy are here in order to
532 trick the optimizer of the IBM C compiler for OS/2 into generating correct
533 code. Apparently IBM isn't going to fix the problem, and we would rather not
534 disable optimization (in this module it actually makes a big difference, and
535 the pcre module can use all the optimization it can get). */
536
537 volatile int dummy;
538 /* ========================================================================= */
539 #endif
540
541 do
542 {
543 const uschar *tcode = code + (((int)*code == OP_CBRA)? 3:1) + LINK_SIZE;
544 BOOL try_next = TRUE;
545
546 while (try_next) /* Loop for items in this branch */
547 {
548 int rc;
549 switch(*tcode)
550 {
551 /* Fail if we reach something we don't understand */
552
553 default:
554 return SSB_FAIL;
555
556 /* If we hit a bracket or a positive lookahead assertion, recurse to set
557 bits from within the subpattern. If it can't find anything, we have to
558 give up. If it finds some mandatory character(s), we are done for this
559 branch. Otherwise, carry on scanning after the subpattern. */
560
561 case OP_BRA:
562 case OP_SBRA:
563 case OP_CBRA:
564 case OP_SCBRA:
565 case OP_ONCE:
566 case OP_ASSERT:
567 rc = set_start_bits(tcode, start_bits, caseless, utf8, cd);
568 if (rc == SSB_FAIL) return SSB_FAIL;
569 if (rc == SSB_DONE) try_next = FALSE; else
570 {
571 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
572 tcode += 1 + LINK_SIZE;
573 }
574 break;
575
576 /* If we hit ALT or KET, it means we haven't found anything mandatory in
577 this branch, though we might have found something optional. For ALT, we
578 continue with the next alternative, but we have to arrange that the final
579 result from subpattern is SSB_CONTINUE rather than SSB_DONE. For KET,
580 return SSB_CONTINUE: if this is the top level, that indicates failure,
581 but after a nested subpattern, it causes scanning to continue. */
582
583 case OP_ALT:
584 yield = SSB_CONTINUE;
585 try_next = FALSE;
586 break;
587
588 case OP_KET:
589 case OP_KETRMAX:
590 case OP_KETRMIN:
591 return SSB_CONTINUE;
592
593 /* Skip over callout */
594
595 case OP_CALLOUT:
596 tcode += 2 + 2*LINK_SIZE;
597 break;
598
599 /* Skip over lookbehind and negative lookahead assertions */
600
601 case OP_ASSERT_NOT:
602 case OP_ASSERTBACK:
603 case OP_ASSERTBACK_NOT:
604 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
605 tcode += 1 + LINK_SIZE;
606 break;
607
608 /* Skip over an option setting, changing the caseless flag */
609
610 case OP_OPT:
611 caseless = (tcode[1] & PCRE_CASELESS) != 0;
612 tcode += 2;
613 break;
614
615 /* BRAZERO does the bracket, but carries on. */
616
617 case OP_BRAZERO:
618 case OP_BRAMINZERO:
619 if (set_start_bits(++tcode, start_bits, caseless, utf8, cd) == SSB_FAIL)
620 return SSB_FAIL;
621 /* =========================================================================
622 See the comment at the head of this function concerning the next line,
623 which was an old fudge for the benefit of OS/2.
624 dummy = 1;
625 ========================================================================= */
626 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
627 tcode += 1 + LINK_SIZE;
628 break;
629
630 /* SKIPZERO skips the bracket. */
631
632 case OP_SKIPZERO:
633 tcode++;
634 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
635 tcode += 1 + LINK_SIZE;
636 break;
637
638 /* Single-char * or ? sets the bit and tries the next item */
639
640 case OP_STAR:
641 case OP_MINSTAR:
642 case OP_POSSTAR:
643 case OP_QUERY:
644 case OP_MINQUERY:
645 case OP_POSQUERY:
646 tcode = set_table_bit(start_bits, tcode + 1, caseless, cd, utf8);
647 break;
648
649 /* Single-char upto sets the bit and tries the next */
650
651 case OP_UPTO:
652 case OP_MINUPTO:
653 case OP_POSUPTO:
654 tcode = set_table_bit(start_bits, tcode + 3, caseless, cd, utf8);
655 break;
656
657 /* At least one single char sets the bit and stops */
658
659 case OP_EXACT: /* Fall through */
660 tcode += 2;
661
662 case OP_CHAR:
663 case OP_CHARNC:
664 case OP_PLUS:
665 case OP_MINPLUS:
666 case OP_POSPLUS:
667 (void)set_table_bit(start_bits, tcode + 1, caseless, cd, utf8);
668 try_next = FALSE;
669 break;
670
671 /* Special spacing and line-terminating items. These recognize specific
672 lists of characters. The difference between VSPACE and ANYNL is that the
673 latter can match the two-character CRLF sequence, but that is not
674 relevant for finding the first character, so their code here is
675 identical. */
676
677 case OP_HSPACE:
678 SET_BIT(0x09);
679 SET_BIT(0x20);
680 if (utf8)
681 {
682 SET_BIT(0xC2); /* For U+00A0 */
683 SET_BIT(0xE1); /* For U+1680, U+180E */
684 SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
685 SET_BIT(0xE3); /* For U+3000 */
686 }
687 else SET_BIT(0xA0);
688 try_next = FALSE;
689 break;
690
691 case OP_ANYNL:
692 case OP_VSPACE:
693 SET_BIT(0x0A);
694 SET_BIT(0x0B);
695 SET_BIT(0x0C);
696 SET_BIT(0x0D);
697 if (utf8)
698 {
699 SET_BIT(0xC2); /* For U+0085 */
700 SET_BIT(0xE2); /* For U+2028, U+2029 */
701 }
702 else SET_BIT(0x85);
703 try_next = FALSE;
704 break;
705
706 /* Single character types set the bits and stop. Note that if PCRE_UCP
707 is set, we do not see these op codes because \d etc are converted to
708 properties. Therefore, these apply in the case when only ASCII characters
709 are recognized to match the types. In UTF-8 mode, we must restrict
710 ourselves to bytes less than 128, as otherwise there can be confusion
711 with bytes in the middle of UTF-8 characters. (In a "traditional"
712 environment, the tables will only recognize ASCII characters anyway, but
713 in at least one Windows environment, some higher bytes bits were set in
714 the tables.) */
715
716 case OP_NOT_DIGIT:
717 for (c = 0; c < table_limit; c++)
718 start_bits[c] |= ~cd->cbits[c+cbit_digit];
719 try_next = FALSE;
720 break;
721
722 case OP_DIGIT:
723 for (c = 0; c < table_limit; c++)
724 start_bits[c] |= cd->cbits[c+cbit_digit];
725 try_next = FALSE;
726 break;
727
728 /* The cbit_space table has vertical tab as whitespace; we have to
729 discard it. */
730
731 case OP_NOT_WHITESPACE:
732 for (c = 0; c < table_limit; c++)
733 {
734 int d = cd->cbits[c+cbit_space];
735 if (c == 1) d &= ~0x08;
736 start_bits[c] |= ~d;
737 }
738 try_next = FALSE;
739 break;
740
741 /* The cbit_space table has vertical tab as whitespace; we have to
742 discard it. */
743
744 case OP_WHITESPACE:
745 for (c = 0; c < table_limit; c++)
746 {
747 int d = cd->cbits[c+cbit_space];
748 if (c == 1) d &= ~0x08;
749 start_bits[c] |= d;
750 }
751 try_next = FALSE;
752 break;
753
754 case OP_NOT_WORDCHAR:
755 for (c = 0; c < table_limit; c++)
756 start_bits[c] |= ~cd->cbits[c+cbit_word];
757 try_next = FALSE;
758 break;
759
760 case OP_WORDCHAR:
761 for (c = 0; c < table_limit; c++)
762 start_bits[c] |= cd->cbits[c+cbit_word];
763 try_next = FALSE;
764 break;
765
766 /* One or more character type fudges the pointer and restarts, knowing
767 it will hit a single character type and stop there. */
768
769 case OP_TYPEPLUS:
770 case OP_TYPEMINPLUS:
771 case OP_TYPEPOSPLUS:
772 tcode++;
773 break;
774
775 case OP_TYPEEXACT:
776 tcode += 3;
777 break;
778
779 /* Zero or more repeats of character types set the bits and then
780 try again. */
781
782 case OP_TYPEUPTO:
783 case OP_TYPEMINUPTO:
784 case OP_TYPEPOSUPTO:
785 tcode += 2; /* Fall through */
786
787 case OP_TYPESTAR:
788 case OP_TYPEMINSTAR:
789 case OP_TYPEPOSSTAR:
790 case OP_TYPEQUERY:
791 case OP_TYPEMINQUERY:
792 case OP_TYPEPOSQUERY:
793 switch(tcode[1])
794 {
795 default:
796 case OP_ANY:
797 case OP_ALLANY:
798 return SSB_FAIL;
799
800 case OP_HSPACE:
801 SET_BIT(0x09);
802 SET_BIT(0x20);
803 if (utf8)
804 {
805 SET_BIT(0xC2); /* For U+00A0 */
806 SET_BIT(0xE1); /* For U+1680, U+180E */
807 SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
808 SET_BIT(0xE3); /* For U+3000 */
809 }
810 else SET_BIT(0xA0);
811 break;
812
813 case OP_ANYNL:
814 case OP_VSPACE:
815 SET_BIT(0x0A);
816 SET_BIT(0x0B);
817 SET_BIT(0x0C);
818 SET_BIT(0x0D);
819 if (utf8)
820 {
821 SET_BIT(0xC2); /* For U+0085 */
822 SET_BIT(0xE2); /* For U+2028, U+2029 */
823 }
824 else SET_BIT(0x85);
825 break;
826
827 case OP_NOT_DIGIT:
828 for (c = 0; c < table_limit; c++)
829 start_bits[c] |= ~cd->cbits[c+cbit_digit];
830 break;
831
832 case OP_DIGIT:
833 for (c = 0; c < table_limit; c++)
834 start_bits[c] |= cd->cbits[c+cbit_digit];
835 break;
836
837 /* The cbit_space table has vertical tab as whitespace; we have to
838 discard it. */
839
840 case OP_NOT_WHITESPACE:
841 for (c = 0; c < table_limit; c++)
842 {
843 int d = cd->cbits[c+cbit_space];
844 if (c == 1) d &= ~0x08;
845 start_bits[c] |= ~d;
846 }
847 break;
848
849 /* The cbit_space table has vertical tab as whitespace; we have to
850 discard it. */
851
852 case OP_WHITESPACE:
853 for (c = 0; c < table_limit; c++)
854 {
855 int d = cd->cbits[c+cbit_space];
856 if (c == 1) d &= ~0x08;
857 start_bits[c] |= d;
858 }
859 break;
860
861 case OP_NOT_WORDCHAR:
862 for (c = 0; c < table_limit; c++)
863 start_bits[c] |= ~cd->cbits[c+cbit_word];
864 break;
865
866 case OP_WORDCHAR:
867 for (c = 0; c < table_limit; c++)
868 start_bits[c] |= cd->cbits[c+cbit_word];
869 break;
870 }
871
872 tcode += 2;
873 break;
874
875 /* Character class where all the information is in a bit map: set the
876 bits and either carry on or not, according to the repeat count. If it was
877 a negative class, and we are operating with UTF-8 characters, any byte
878 with a value >= 0xc4 is a potentially valid starter because it starts a
879 character with a value > 255. */
880
881 case OP_NCLASS:
882 #ifdef SUPPORT_UTF8
883 if (utf8)
884 {
885 start_bits[24] |= 0xf0; /* Bits for 0xc4 - 0xc8 */
886 memset(start_bits+25, 0xff, 7); /* Bits for 0xc9 - 0xff */
887 }
888 #endif
889 /* Fall through */
890
891 case OP_CLASS:
892 {
893 tcode++;
894
895 /* In UTF-8 mode, the bits in a bit map correspond to character
896 values, not to byte values. However, the bit map we are constructing is
897 for byte values. So we have to do a conversion for characters whose
898 value is > 127. In fact, there are only two possible starting bytes for
899 characters in the range 128 - 255. */
900
901 #ifdef SUPPORT_UTF8
902 if (utf8)
903 {
904 for (c = 0; c < 16; c++) start_bits[c] |= tcode[c];
905 for (c = 128; c < 256; c++)
906 {
907 if ((tcode[c/8] && (1 << (c&7))) != 0)
908 {
909 int d = (c >> 6) | 0xc0; /* Set bit for this starter */
910 start_bits[d/8] |= (1 << (d&7)); /* and then skip on to the */
911 c = (c & 0xc0) + 0x40 - 1; /* next relevant character. */
912 }
913 }
914 }
915
916 /* In non-UTF-8 mode, the two bit maps are completely compatible. */
917
918 else
919 #endif
920 {
921 for (c = 0; c < 32; c++) start_bits[c] |= tcode[c];
922 }
923
924 /* Advance past the bit map, and act on what follows */
925
926 tcode += 32;
927 switch (*tcode)
928 {
929 case OP_CRSTAR:
930 case OP_CRMINSTAR:
931 case OP_CRQUERY:
932 case OP_CRMINQUERY:
933 tcode++;
934 break;
935
936 case OP_CRRANGE:
937 case OP_CRMINRANGE:
938 if (((tcode[1] << 8) + tcode[2]) == 0) tcode += 5;
939 else try_next = FALSE;
940 break;
941
942 default:
943 try_next = FALSE;
944 break;
945 }
946 }
947 break; /* End of bitmap class handling */
948
949 } /* End of switch */
950 } /* End of try_next loop */
951
952 code += GET(code, 1); /* Advance to next branch */
953 }
954 while (*code == OP_ALT);
955 return yield;
956 }
957
958
959
960 /*************************************************
961 * Study a compiled expression *
962 *************************************************/
963
964 /* This function is handed a compiled expression that it must study to produce
965 information that will speed up the matching. It returns a pcre_extra block
966 which then gets handed back to pcre_exec().
967
968 Arguments:
969 re points to the compiled expression
970 options contains option bits
971 errorptr points to where to place error messages;
972 set NULL unless error
973
974 Returns: pointer to a pcre_extra block, with study_data filled in and the
975 appropriate flags set;
976 NULL on error or if no optimization possible
977 */
978
979 PCRE_EXP_DEFN pcre_extra * PCRE_CALL_CONVENTION
980 pcre_study(const pcre *external_re, int options, const char **errorptr)
981 {
982 int min;
983 BOOL bits_set = FALSE;
984 uschar start_bits[32];
985 pcre_extra *extra;
986 pcre_study_data *study;
987 const uschar *tables;
988 uschar *code;
989 compile_data compile_block;
990 const real_pcre *re = (const real_pcre *)external_re;
991
992 *errorptr = NULL;
993
994 if (re == NULL || re->magic_number != MAGIC_NUMBER)
995 {
996 *errorptr = "argument is not a compiled regular expression";
997 return NULL;
998 }
999
1000 if ((options & ~PUBLIC_STUDY_OPTIONS) != 0)
1001 {
1002 *errorptr = "unknown or incorrect option bit(s) set";
1003 return NULL;
1004 }
1005
1006 code = (uschar *)re + re->name_table_offset +
1007 (re->name_count * re->name_entry_size);
1008
1009 /* For an anchored pattern, or an unanchored pattern that has a first char, or
1010 a multiline pattern that matches only at "line starts", there is no point in
1011 seeking a list of starting bytes. */
1012
1013 if ((re->options & PCRE_ANCHORED) == 0 &&
1014 (re->flags & (PCRE_FIRSTSET|PCRE_STARTLINE)) == 0)
1015 {
1016 /* Set the character tables in the block that is passed around */
1017
1018 tables = re->tables;
1019 if (tables == NULL)
1020 (void)pcre_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES,
1021 (void *)(&tables));
1022
1023 compile_block.lcc = tables + lcc_offset;
1024 compile_block.fcc = tables + fcc_offset;
1025 compile_block.cbits = tables + cbits_offset;
1026 compile_block.ctypes = tables + ctypes_offset;
1027
1028 /* See if we can find a fixed set of initial characters for the pattern. */
1029
1030 memset(start_bits, 0, 32 * sizeof(uschar));
1031 bits_set = set_start_bits(code, start_bits,
1032 (re->options & PCRE_CASELESS) != 0, (re->options & PCRE_UTF8) != 0,
1033 &compile_block) == SSB_DONE;
1034 }
1035
1036 /* Find the minimum length of subject string. */
1037
1038 min = find_minlength(code, code, re->options);
1039
1040 /* Return NULL if no optimization is possible. */
1041
1042 if (!bits_set && min < 0) return NULL;
1043
1044 /* Get a pcre_extra block and a pcre_study_data block. The study data is put in
1045 the latter, which is pointed to by the former, which may also get additional
1046 data set later by the calling program. At the moment, the size of
1047 pcre_study_data is fixed. We nevertheless save it in a field for returning via
1048 the pcre_fullinfo() function so that if it becomes variable in the future, we
1049 don't have to change that code. */
1050
1051 extra = (pcre_extra *)(pcre_malloc)
1052 (sizeof(pcre_extra) + sizeof(pcre_study_data));
1053
1054 if (extra == NULL)
1055 {
1056 *errorptr = "failed to get memory";
1057 return NULL;
1058 }
1059
1060 study = (pcre_study_data *)((char *)extra + sizeof(pcre_extra));
1061 extra->flags = PCRE_EXTRA_STUDY_DATA;
1062 extra->study_data = study;
1063
1064 study->size = sizeof(pcre_study_data);
1065 study->flags = 0;
1066
1067 if (bits_set)
1068 {
1069 study->flags |= PCRE_STUDY_MAPPED;
1070 memcpy(study->start_bits, start_bits, sizeof(start_bits));
1071 }
1072
1073 if (min >= 0)
1074 {
1075 study->flags |= PCRE_STUDY_MINLEN;
1076 study->minlength = min;
1077 }
1078
1079 return extra;
1080 }
1081
1082 /* End of pcre_study.c */

Properties

Name Value
svn:eol-style native
svn:keywords "Author Date Id Revision Url"

webmaster@exim.org
ViewVC Help
Powered by ViewVC 1.1.12