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

Contents of /code/trunk/pcre_study.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 520 - (show annotations)
Sat May 22 18:54:05 2010 UTC (5 years ago) by ph10
File MIME type: text/plain
File size: 28252 byte(s)
Error occurred while calculating annotation data.
Fix caseless bug with pcre_study() for starting letter greater than 127.
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
52 /* Returns from set_start_bits() */
53
54 enum { SSB_FAIL, SSB_DONE, SSB_CONTINUE };
55
56
57
58 /*************************************************
59 * Find the minimum subject length for a group *
60 *************************************************/
61
62 /* Scan a parenthesized group and compute the minimum length of subject that
63 is needed to match it. This is a lower bound; it does not mean there is a
64 string of that length that matches. In UTF8 mode, the result is in characters
65 rather than bytes.
66
67 Arguments:
68 code pointer to start of group (the bracket)
69 startcode pointer to start of the whole pattern
70 options the compiling options
71
72 Returns: the minimum length
73 -1 if \C was encountered
74 -2 internal error (missing capturing bracket)
75 */
76
77 static int
78 find_minlength(const uschar *code, const uschar *startcode, int options)
79 {
80 int length = -1;
81 BOOL utf8 = (options & PCRE_UTF8) != 0;
82 BOOL had_recurse = FALSE;
83 register int branchlength = 0;
84 register uschar *cc = (uschar *)code + 1 + LINK_SIZE;
85
86 if (*code == OP_CBRA || *code == OP_SCBRA) cc += 2;
87
88 /* Scan along the opcodes for this branch. If we get to the end of the
89 branch, check the length against that of the other branches. */
90
91 for (;;)
92 {
93 int d, min;
94 uschar *cs, *ce;
95 register int op = *cc;
96
97 switch (op)
98 {
99 case OP_COND:
100 case OP_SCOND:
101
102 /* If there is only one branch in a condition, the implied branch has zero
103 length, so we don't add anything. This covers the DEFINE "condition"
104 automatically. */
105
106 cs = cc + GET(cc, 1);
107 if (*cs != OP_ALT)
108 {
109 cc = cs + 1 + LINK_SIZE;
110 break;
111 }
112
113 /* Otherwise we can fall through and treat it the same as any other
114 subpattern. */
115
116 case OP_CBRA:
117 case OP_SCBRA:
118 case OP_BRA:
119 case OP_SBRA:
120 case OP_ONCE:
121 d = find_minlength(cc, startcode, options);
122 if (d < 0) return d;
123 branchlength += d;
124 do cc += GET(cc, 1); while (*cc == OP_ALT);
125 cc += 1 + LINK_SIZE;
126 break;
127
128 /* Reached end of a branch; if it's a ket it is the end of a nested
129 call. If it's ALT it is an alternation in a nested call. If it is
130 END it's the end of the outer call. All can be handled by the same code. */
131
132 case OP_ALT:
133 case OP_KET:
134 case OP_KETRMAX:
135 case OP_KETRMIN:
136 case OP_END:
137 if (length < 0 || (!had_recurse && branchlength < length))
138 length = branchlength;
139 if (*cc != OP_ALT) return length;
140 cc += 1 + LINK_SIZE;
141 branchlength = 0;
142 had_recurse = FALSE;
143 break;
144
145 /* Skip over assertive subpatterns */
146
147 case OP_ASSERT:
148 case OP_ASSERT_NOT:
149 case OP_ASSERTBACK:
150 case OP_ASSERTBACK_NOT:
151 do cc += GET(cc, 1); while (*cc == OP_ALT);
152 /* Fall through */
153
154 /* Skip over things that don't match chars */
155
156 case OP_REVERSE:
157 case OP_CREF:
158 case OP_NCREF:
159 case OP_RREF:
160 case OP_NRREF:
161 case OP_DEF:
162 case OP_OPT:
163 case OP_CALLOUT:
164 case OP_SOD:
165 case OP_SOM:
166 case OP_EOD:
167 case OP_EODN:
168 case OP_CIRC:
169 case OP_DOLL:
170 case OP_NOT_WORD_BOUNDARY:
171 case OP_WORD_BOUNDARY:
172 cc += _pcre_OP_lengths[*cc];
173 break;
174
175 /* Skip over a subpattern that has a {0} or {0,x} quantifier */
176
177 case OP_BRAZERO:
178 case OP_BRAMINZERO:
179 case OP_SKIPZERO:
180 cc += _pcre_OP_lengths[*cc];
181 do cc += GET(cc, 1); while (*cc == OP_ALT);
182 cc += 1 + LINK_SIZE;
183 break;
184
185 /* Handle literal characters and + repetitions */
186
187 case OP_CHAR:
188 case OP_CHARNC:
189 case OP_NOT:
190 case OP_PLUS:
191 case OP_MINPLUS:
192 case OP_POSPLUS:
193 case OP_NOTPLUS:
194 case OP_NOTMINPLUS:
195 case OP_NOTPOSPLUS:
196 branchlength++;
197 cc += 2;
198 #ifdef SUPPORT_UTF8
199 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
200 #endif
201 break;
202
203 case OP_TYPEPLUS:
204 case OP_TYPEMINPLUS:
205 case OP_TYPEPOSPLUS:
206 branchlength++;
207 cc += (cc[1] == OP_PROP || cc[1] == OP_NOTPROP)? 4 : 2;
208 break;
209
210 /* Handle exact repetitions. The count is already in characters, but we
211 need to skip over a multibyte character in UTF8 mode. */
212
213 case OP_EXACT:
214 case OP_NOTEXACT:
215 branchlength += GET2(cc,1);
216 cc += 4;
217 #ifdef SUPPORT_UTF8
218 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
219 #endif
220 break;
221
222 case OP_TYPEEXACT:
223 branchlength += GET2(cc,1);
224 cc += (cc[3] == OP_PROP || cc[3] == OP_NOTPROP)? 6 : 4;
225 break;
226
227 /* Handle single-char non-literal matchers */
228
229 case OP_PROP:
230 case OP_NOTPROP:
231 cc += 2;
232 /* Fall through */
233
234 case OP_NOT_DIGIT:
235 case OP_DIGIT:
236 case OP_NOT_WHITESPACE:
237 case OP_WHITESPACE:
238 case OP_NOT_WORDCHAR:
239 case OP_WORDCHAR:
240 case OP_ANY:
241 case OP_ALLANY:
242 case OP_EXTUNI:
243 case OP_HSPACE:
244 case OP_NOT_HSPACE:
245 case OP_VSPACE:
246 case OP_NOT_VSPACE:
247 branchlength++;
248 cc++;
249 break;
250
251 /* "Any newline" might match two characters */
252
253 case OP_ANYNL:
254 branchlength += 2;
255 cc++;
256 break;
257
258 /* The single-byte matcher means we can't proceed in UTF-8 mode */
259
260 case OP_ANYBYTE:
261 #ifdef SUPPORT_UTF8
262 if (utf8) return -1;
263 #endif
264 branchlength++;
265 cc++;
266 break;
267
268 /* For repeated character types, we have to test for \p and \P, which have
269 an extra two bytes of parameters. */
270
271 case OP_TYPESTAR:
272 case OP_TYPEMINSTAR:
273 case OP_TYPEQUERY:
274 case OP_TYPEMINQUERY:
275 case OP_TYPEPOSSTAR:
276 case OP_TYPEPOSQUERY:
277 if (cc[1] == OP_PROP || cc[1] == OP_NOTPROP) cc += 2;
278 cc += _pcre_OP_lengths[op];
279 break;
280
281 case OP_TYPEUPTO:
282 case OP_TYPEMINUPTO:
283 case OP_TYPEPOSUPTO:
284 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
285 cc += _pcre_OP_lengths[op];
286 break;
287
288 /* Check a class for variable quantification */
289
290 #ifdef SUPPORT_UTF8
291 case OP_XCLASS:
292 cc += GET(cc, 1) - 33;
293 /* Fall through */
294 #endif
295
296 case OP_CLASS:
297 case OP_NCLASS:
298 cc += 33;
299
300 switch (*cc)
301 {
302 case OP_CRPLUS:
303 case OP_CRMINPLUS:
304 branchlength++;
305 /* Fall through */
306
307 case OP_CRSTAR:
308 case OP_CRMINSTAR:
309 case OP_CRQUERY:
310 case OP_CRMINQUERY:
311 cc++;
312 break;
313
314 case OP_CRRANGE:
315 case OP_CRMINRANGE:
316 branchlength += GET2(cc,1);
317 cc += 5;
318 break;
319
320 default:
321 branchlength++;
322 break;
323 }
324 break;
325
326 /* Backreferences and subroutine calls are treated in the same way: we find
327 the minimum length for the subpattern. A recursion, however, causes an
328 a flag to be set that causes the length of this branch to be ignored. The
329 logic is that a recursion can only make sense if there is another
330 alternation that stops the recursing. That will provide the minimum length
331 (when no recursion happens). A backreference within the group that it is
332 referencing behaves in the same way.
333
334 If PCRE_JAVASCRIPT_COMPAT is set, a backreference to an unset bracket
335 matches an empty string (by default it causes a matching failure), so in
336 that case we must set the minimum length to zero. */
337
338 case OP_REF:
339 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0)
340 {
341 ce = cs = (uschar *)_pcre_find_bracket(startcode, utf8, GET2(cc, 1));
342 if (cs == NULL) return -2;
343 do ce += GET(ce, 1); while (*ce == OP_ALT);
344 if (cc > cs && cc < ce)
345 {
346 d = 0;
347 had_recurse = TRUE;
348 }
349 else d = find_minlength(cs, startcode, options);
350 }
351 else d = 0;
352 cc += 3;
353
354 /* Handle repeated back references */
355
356 switch (*cc)
357 {
358 case OP_CRSTAR:
359 case OP_CRMINSTAR:
360 case OP_CRQUERY:
361 case OP_CRMINQUERY:
362 min = 0;
363 cc++;
364 break;
365
366 case OP_CRRANGE:
367 case OP_CRMINRANGE:
368 min = GET2(cc, 1);
369 cc += 5;
370 break;
371
372 default:
373 min = 1;
374 break;
375 }
376
377 branchlength += min * d;
378 break;
379
380 case OP_RECURSE:
381 cs = ce = (uschar *)startcode + GET(cc, 1);
382 if (cs == NULL) return -2;
383 do ce += GET(ce, 1); while (*ce == OP_ALT);
384 if (cc > cs && cc < ce)
385 had_recurse = TRUE;
386 else
387 branchlength += find_minlength(cs, startcode, options);
388 cc += 1 + LINK_SIZE;
389 break;
390
391 /* Anything else does not or need not match a character. We can get the
392 item's length from the table, but for those that can match zero occurrences
393 of a character, we must take special action for UTF-8 characters. */
394
395 case OP_UPTO:
396 case OP_NOTUPTO:
397 case OP_MINUPTO:
398 case OP_NOTMINUPTO:
399 case OP_POSUPTO:
400 case OP_STAR:
401 case OP_MINSTAR:
402 case OP_NOTMINSTAR:
403 case OP_POSSTAR:
404 case OP_NOTPOSSTAR:
405 case OP_QUERY:
406 case OP_MINQUERY:
407 case OP_NOTMINQUERY:
408 case OP_POSQUERY:
409 case OP_NOTPOSQUERY:
410 cc += _pcre_OP_lengths[op];
411 #ifdef SUPPORT_UTF8
412 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
413 #endif
414 break;
415
416 /* Skip these, but we need to add in the name length. */
417
418 case OP_MARK:
419 case OP_PRUNE_ARG:
420 case OP_SKIP_ARG:
421 case OP_THEN_ARG:
422 cc += _pcre_OP_lengths[op] + cc[1];
423 break;
424
425 /* For the record, these are the opcodes that are matched by "default":
426 OP_ACCEPT, OP_CLOSE, OP_COMMIT, OP_FAIL, OP_PRUNE, OP_SET_SOM, OP_SKIP,
427 OP_THEN. */
428
429 default:
430 cc += _pcre_OP_lengths[op];
431 break;
432 }
433 }
434 /* Control never gets here */
435 }
436
437
438
439 /*************************************************
440 * Set a bit and maybe its alternate case *
441 *************************************************/
442
443 /* Given a character, set its first byte's bit in the table, and also the
444 corresponding bit for the other version of a letter if we are caseless. In
445 UTF-8 mode, for characters greater than 127, we can only do the caseless thing
446 when Unicode property support is available.
447
448 Arguments:
449 start_bits points to the bit map
450 p points to the character
451 caseless the caseless flag
452 cd the block with char table pointers
453 utf8 TRUE for UTF-8 mode
454
455 Returns: pointer after the character
456 */
457
458 static const uschar *
459 set_table_bit(uschar *start_bits, const uschar *p, BOOL caseless,
460 compile_data *cd, BOOL utf8)
461 {
462 unsigned int c = *p;
463 start_bits[c/8] |= (1 << (c&7));
464
465 #ifdef SUPPORT_UTF8
466 if (utf8 && c > 127)
467 {
468 GETCHARINC(c, p);
469 #ifdef SUPPORT_UCP
470 if (caseless)
471 {
472 uschar buff[8];
473 c = UCD_OTHERCASE(c);
474 (void)_pcre_ord2utf8(c, buff);
475 c = buff[0];
476 start_bits[c/8] |= (1 << (c&7));
477 }
478 #endif
479 return p;
480 }
481 #endif
482
483 /* Not UTF-8 mode, or character is less than 127. */
484
485 if (caseless && (cd->ctypes[c] & ctype_letter) != 0)
486 start_bits[cd->fcc[c]/8] |= (1 << (cd->fcc[c]&7));
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
523 #if 0
524 /* ========================================================================= */
525 /* The following comment and code was inserted in January 1999. In May 2006,
526 when it was observed to cause compiler warnings about unused values, I took it
527 out again. If anybody is still using OS/2, they will have to put it back
528 manually. */
529
530 /* This next statement and the later reference to dummy are here in order to
531 trick the optimizer of the IBM C compiler for OS/2 into generating correct
532 code. Apparently IBM isn't going to fix the problem, and we would rather not
533 disable optimization (in this module it actually makes a big difference, and
534 the pcre module can use all the optimization it can get). */
535
536 volatile int dummy;
537 /* ========================================================================= */
538 #endif
539
540 do
541 {
542 const uschar *tcode = code + (((int)*code == OP_CBRA)? 3:1) + LINK_SIZE;
543 BOOL try_next = TRUE;
544
545 while (try_next) /* Loop for items in this branch */
546 {
547 int rc;
548 switch(*tcode)
549 {
550 /* Fail if we reach something we don't understand */
551
552 default:
553 return SSB_FAIL;
554
555 /* If we hit a bracket or a positive lookahead assertion, recurse to set
556 bits from within the subpattern. If it can't find anything, we have to
557 give up. If it finds some mandatory character(s), we are done for this
558 branch. Otherwise, carry on scanning after the subpattern. */
559
560 case OP_BRA:
561 case OP_SBRA:
562 case OP_CBRA:
563 case OP_SCBRA:
564 case OP_ONCE:
565 case OP_ASSERT:
566 rc = set_start_bits(tcode, start_bits, caseless, utf8, cd);
567 if (rc == SSB_FAIL) return SSB_FAIL;
568 if (rc == SSB_DONE) try_next = FALSE; else
569 {
570 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
571 tcode += 1 + LINK_SIZE;
572 }
573 break;
574
575 /* If we hit ALT or KET, it means we haven't found anything mandatory in
576 this branch, though we might have found something optional. For ALT, we
577 continue with the next alternative, but we have to arrange that the final
578 result from subpattern is SSB_CONTINUE rather than SSB_DONE. For KET,
579 return SSB_CONTINUE: if this is the top level, that indicates failure,
580 but after a nested subpattern, it causes scanning to continue. */
581
582 case OP_ALT:
583 yield = SSB_CONTINUE;
584 try_next = FALSE;
585 break;
586
587 case OP_KET:
588 case OP_KETRMAX:
589 case OP_KETRMIN:
590 return SSB_CONTINUE;
591
592 /* Skip over callout */
593
594 case OP_CALLOUT:
595 tcode += 2 + 2*LINK_SIZE;
596 break;
597
598 /* Skip over lookbehind and negative lookahead assertions */
599
600 case OP_ASSERT_NOT:
601 case OP_ASSERTBACK:
602 case OP_ASSERTBACK_NOT:
603 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
604 tcode += 1 + LINK_SIZE;
605 break;
606
607 /* Skip over an option setting, changing the caseless flag */
608
609 case OP_OPT:
610 caseless = (tcode[1] & PCRE_CASELESS) != 0;
611 tcode += 2;
612 break;
613
614 /* BRAZERO does the bracket, but carries on. */
615
616 case OP_BRAZERO:
617 case OP_BRAMINZERO:
618 if (set_start_bits(++tcode, start_bits, caseless, utf8, cd) == SSB_FAIL)
619 return SSB_FAIL;
620 /* =========================================================================
621 See the comment at the head of this function concerning the next line,
622 which was an old fudge for the benefit of OS/2.
623 dummy = 1;
624 ========================================================================= */
625 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
626 tcode += 1 + LINK_SIZE;
627 break;
628
629 /* SKIPZERO skips the bracket. */
630
631 case OP_SKIPZERO:
632 tcode++;
633 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
634 tcode += 1 + LINK_SIZE;
635 break;
636
637 /* Single-char * or ? sets the bit and tries the next item */
638
639 case OP_STAR:
640 case OP_MINSTAR:
641 case OP_POSSTAR:
642 case OP_QUERY:
643 case OP_MINQUERY:
644 case OP_POSQUERY:
645 tcode = set_table_bit(start_bits, tcode + 1, caseless, cd, utf8);
646 break;
647
648 /* Single-char upto sets the bit and tries the next */
649
650 case OP_UPTO:
651 case OP_MINUPTO:
652 case OP_POSUPTO:
653 tcode = set_table_bit(start_bits, tcode + 3, caseless, cd, utf8);
654 break;
655
656 /* At least one single char sets the bit and stops */
657
658 case OP_EXACT: /* Fall through */
659 tcode += 2;
660
661 case OP_CHAR:
662 case OP_CHARNC:
663 case OP_PLUS:
664 case OP_MINPLUS:
665 case OP_POSPLUS:
666 (void)set_table_bit(start_bits, tcode + 1, caseless, cd, utf8);
667 try_next = FALSE;
668 break;
669
670 /* Single character types set the bits and stop. Note that if PCRE_UCP
671 is set, we do not see these op codes because \d etc are converted to
672 properties. Therefore, these apply in the case when only ASCII characters
673 are recognized to match the types. */
674
675 case OP_NOT_DIGIT:
676 for (c = 0; c < 32; c++)
677 start_bits[c] |= ~cd->cbits[c+cbit_digit];
678 try_next = FALSE;
679 break;
680
681 case OP_DIGIT:
682 for (c = 0; c < 32; c++)
683 start_bits[c] |= cd->cbits[c+cbit_digit];
684 try_next = FALSE;
685 break;
686
687 /* The cbit_space table has vertical tab as whitespace; we have to
688 discard it. */
689
690 case OP_NOT_WHITESPACE:
691 for (c = 0; c < 32; c++)
692 {
693 int d = cd->cbits[c+cbit_space];
694 if (c == 1) d &= ~0x08;
695 start_bits[c] |= ~d;
696 }
697 try_next = FALSE;
698 break;
699
700 /* The cbit_space table has vertical tab as whitespace; we have to
701 discard it. */
702
703 case OP_WHITESPACE:
704 for (c = 0; c < 32; c++)
705 {
706 int d = cd->cbits[c+cbit_space];
707 if (c == 1) d &= ~0x08;
708 start_bits[c] |= d;
709 }
710 try_next = FALSE;
711 break;
712
713 case OP_NOT_WORDCHAR:
714 for (c = 0; c < 32; c++)
715 start_bits[c] |= ~cd->cbits[c+cbit_word];
716 try_next = FALSE;
717 break;
718
719 case OP_WORDCHAR:
720 for (c = 0; c < 32; c++)
721 start_bits[c] |= cd->cbits[c+cbit_word];
722 try_next = FALSE;
723 break;
724
725 /* One or more character type fudges the pointer and restarts, knowing
726 it will hit a single character type and stop there. */
727
728 case OP_TYPEPLUS:
729 case OP_TYPEMINPLUS:
730 tcode++;
731 break;
732
733 case OP_TYPEEXACT:
734 tcode += 3;
735 break;
736
737 /* Zero or more repeats of character types set the bits and then
738 try again. */
739
740 case OP_TYPEUPTO:
741 case OP_TYPEMINUPTO:
742 case OP_TYPEPOSUPTO:
743 tcode += 2; /* Fall through */
744
745 case OP_TYPESTAR:
746 case OP_TYPEMINSTAR:
747 case OP_TYPEPOSSTAR:
748 case OP_TYPEQUERY:
749 case OP_TYPEMINQUERY:
750 case OP_TYPEPOSQUERY:
751 switch(tcode[1])
752 {
753 case OP_ANY:
754 case OP_ALLANY:
755 return SSB_FAIL;
756
757 case OP_NOT_DIGIT:
758 for (c = 0; c < 32; c++)
759 start_bits[c] |= ~cd->cbits[c+cbit_digit];
760 break;
761
762 case OP_DIGIT:
763 for (c = 0; c < 32; c++)
764 start_bits[c] |= cd->cbits[c+cbit_digit];
765 break;
766
767 /* The cbit_space table has vertical tab as whitespace; we have to
768 discard it. */
769
770 case OP_NOT_WHITESPACE:
771 for (c = 0; c < 32; c++)
772 {
773 int d = cd->cbits[c+cbit_space];
774 if (c == 1) d &= ~0x08;
775 start_bits[c] |= ~d;
776 }
777 break;
778
779 /* The cbit_space table has vertical tab as whitespace; we have to
780 discard it. */
781
782 case OP_WHITESPACE:
783 for (c = 0; c < 32; c++)
784 {
785 int d = cd->cbits[c+cbit_space];
786 if (c == 1) d &= ~0x08;
787 start_bits[c] |= d;
788 }
789 break;
790
791 case OP_NOT_WORDCHAR:
792 for (c = 0; c < 32; c++)
793 start_bits[c] |= ~cd->cbits[c+cbit_word];
794 break;
795
796 case OP_WORDCHAR:
797 for (c = 0; c < 32; c++)
798 start_bits[c] |= cd->cbits[c+cbit_word];
799 break;
800 }
801
802 tcode += 2;
803 break;
804
805 /* Character class where all the information is in a bit map: set the
806 bits and either carry on or not, according to the repeat count. If it was
807 a negative class, and we are operating with UTF-8 characters, any byte
808 with a value >= 0xc4 is a potentially valid starter because it starts a
809 character with a value > 255. */
810
811 case OP_NCLASS:
812 #ifdef SUPPORT_UTF8
813 if (utf8)
814 {
815 start_bits[24] |= 0xf0; /* Bits for 0xc4 - 0xc8 */
816 memset(start_bits+25, 0xff, 7); /* Bits for 0xc9 - 0xff */
817 }
818 #endif
819 /* Fall through */
820
821 case OP_CLASS:
822 {
823 tcode++;
824
825 /* In UTF-8 mode, the bits in a bit map correspond to character
826 values, not to byte values. However, the bit map we are constructing is
827 for byte values. So we have to do a conversion for characters whose
828 value is > 127. In fact, there are only two possible starting bytes for
829 characters in the range 128 - 255. */
830
831 #ifdef SUPPORT_UTF8
832 if (utf8)
833 {
834 for (c = 0; c < 16; c++) start_bits[c] |= tcode[c];
835 for (c = 128; c < 256; c++)
836 {
837 if ((tcode[c/8] && (1 << (c&7))) != 0)
838 {
839 int d = (c >> 6) | 0xc0; /* Set bit for this starter */
840 start_bits[d/8] |= (1 << (d&7)); /* and then skip on to the */
841 c = (c & 0xc0) + 0x40 - 1; /* next relevant character. */
842 }
843 }
844 }
845
846 /* In non-UTF-8 mode, the two bit maps are completely compatible. */
847
848 else
849 #endif
850 {
851 for (c = 0; c < 32; c++) start_bits[c] |= tcode[c];
852 }
853
854 /* Advance past the bit map, and act on what follows */
855
856 tcode += 32;
857 switch (*tcode)
858 {
859 case OP_CRSTAR:
860 case OP_CRMINSTAR:
861 case OP_CRQUERY:
862 case OP_CRMINQUERY:
863 tcode++;
864 break;
865
866 case OP_CRRANGE:
867 case OP_CRMINRANGE:
868 if (((tcode[1] << 8) + tcode[2]) == 0) tcode += 5;
869 else try_next = FALSE;
870 break;
871
872 default:
873 try_next = FALSE;
874 break;
875 }
876 }
877 break; /* End of bitmap class handling */
878
879 } /* End of switch */
880 } /* End of try_next loop */
881
882 code += GET(code, 1); /* Advance to next branch */
883 }
884 while (*code == OP_ALT);
885 return yield;
886 }
887
888
889
890 /*************************************************
891 * Study a compiled expression *
892 *************************************************/
893
894 /* This function is handed a compiled expression that it must study to produce
895 information that will speed up the matching. It returns a pcre_extra block
896 which then gets handed back to pcre_exec().
897
898 Arguments:
899 re points to the compiled expression
900 options contains option bits
901 errorptr points to where to place error messages;
902 set NULL unless error
903
904 Returns: pointer to a pcre_extra block, with study_data filled in and the
905 appropriate flags set;
906 NULL on error or if no optimization possible
907 */
908
909 PCRE_EXP_DEFN pcre_extra * PCRE_CALL_CONVENTION
910 pcre_study(const pcre *external_re, int options, const char **errorptr)
911 {
912 int min;
913 BOOL bits_set = FALSE;
914 uschar start_bits[32];
915 pcre_extra *extra;
916 pcre_study_data *study;
917 const uschar *tables;
918 uschar *code;
919 compile_data compile_block;
920 const real_pcre *re = (const real_pcre *)external_re;
921
922 *errorptr = NULL;
923
924 if (re == NULL || re->magic_number != MAGIC_NUMBER)
925 {
926 *errorptr = "argument is not a compiled regular expression";
927 return NULL;
928 }
929
930 if ((options & ~PUBLIC_STUDY_OPTIONS) != 0)
931 {
932 *errorptr = "unknown or incorrect option bit(s) set";
933 return NULL;
934 }
935
936 code = (uschar *)re + re->name_table_offset +
937 (re->name_count * re->name_entry_size);
938
939 /* For an anchored pattern, or an unanchored pattern that has a first char, or
940 a multiline pattern that matches only at "line starts", there is no point in
941 seeking a list of starting bytes. */
942
943 if ((re->options & PCRE_ANCHORED) == 0 &&
944 (re->flags & (PCRE_FIRSTSET|PCRE_STARTLINE)) == 0)
945 {
946 /* Set the character tables in the block that is passed around */
947
948 tables = re->tables;
949 if (tables == NULL)
950 (void)pcre_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES,
951 (void *)(&tables));
952
953 compile_block.lcc = tables + lcc_offset;
954 compile_block.fcc = tables + fcc_offset;
955 compile_block.cbits = tables + cbits_offset;
956 compile_block.ctypes = tables + ctypes_offset;
957
958 /* See if we can find a fixed set of initial characters for the pattern. */
959
960 memset(start_bits, 0, 32 * sizeof(uschar));
961 bits_set = set_start_bits(code, start_bits,
962 (re->options & PCRE_CASELESS) != 0, (re->options & PCRE_UTF8) != 0,
963 &compile_block) == SSB_DONE;
964 }
965
966 /* Find the minimum length of subject string. */
967
968 min = find_minlength(code, code, re->options);
969
970 /* Return NULL if no optimization is possible. */
971
972 if (!bits_set && min < 0) return NULL;
973
974 /* Get a pcre_extra block and a pcre_study_data block. The study data is put in
975 the latter, which is pointed to by the former, which may also get additional
976 data set later by the calling program. At the moment, the size of
977 pcre_study_data is fixed. We nevertheless save it in a field for returning via
978 the pcre_fullinfo() function so that if it becomes variable in the future, we
979 don't have to change that code. */
980
981 extra = (pcre_extra *)(pcre_malloc)
982 (sizeof(pcre_extra) + sizeof(pcre_study_data));
983
984 if (extra == NULL)
985 {
986 *errorptr = "failed to get memory";
987 return NULL;
988 }
989
990 study = (pcre_study_data *)((char *)extra + sizeof(pcre_extra));
991 extra->flags = PCRE_EXTRA_STUDY_DATA;
992 extra->study_data = study;
993
994 study->size = sizeof(pcre_study_data);
995 study->flags = 0;
996
997 if (bits_set)
998 {
999 study->flags |= PCRE_STUDY_MAPPED;
1000 memcpy(study->start_bits, start_bits, sizeof(start_bits));
1001 }
1002
1003 if (min >= 0)
1004 {
1005 study->flags |= PCRE_STUDY_MINLEN;
1006 study->minlength = min;
1007 }
1008
1009 return extra;
1010 }
1011
1012 /* End of pcre_study.c */

Properties

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

  ViewVC Help
Powered by ViewVC 1.1.5