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

Contents of /code/trunk/pcre_study.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 510 - (show annotations) (download)
Sat Mar 27 17:45:29 2010 UTC (4 years, 5 months ago) by ph10
File MIME type: text/plain
File size: 27652 byte(s)
Add support for *MARK and names for *PRUNE, *SKIP, *THEN.

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 bit in the table, and also the bit for the other
444 version of a letter if we are caseless.
445
446 Arguments:
447 start_bits points to the bit map
448 c is the character
449 caseless the caseless flag
450 cd the block with char table pointers
451
452 Returns: nothing
453 */
454
455 static void
456 set_table_bit(uschar *start_bits, unsigned int c, BOOL caseless,
457 compile_data *cd)
458 {
459 start_bits[c/8] |= (1 << (c&7));
460 if (caseless && (cd->ctypes[c] & ctype_letter) != 0)
461 start_bits[cd->fcc[c]/8] |= (1 << (cd->fcc[c]&7));
462 }
463
464
465
466 /*************************************************
467 * Create bitmap of starting bytes *
468 *************************************************/
469
470 /* This function scans a compiled unanchored expression recursively and
471 attempts to build a bitmap of the set of possible starting bytes. As time goes
472 by, we may be able to get more clever at doing this. The SSB_CONTINUE return is
473 useful for parenthesized groups in patterns such as (a*)b where the group
474 provides some optional starting bytes but scanning must continue at the outer
475 level to find at least one mandatory byte. At the outermost level, this
476 function fails unless the result is SSB_DONE.
477
478 Arguments:
479 code points to an expression
480 start_bits points to a 32-byte table, initialized to 0
481 caseless the current state of the caseless flag
482 utf8 TRUE if in UTF-8 mode
483 cd the block with char table pointers
484
485 Returns: SSB_FAIL => Failed to find any starting bytes
486 SSB_DONE => Found mandatory starting bytes
487 SSB_CONTINUE => Found optional starting bytes
488 */
489
490 static int
491 set_start_bits(const uschar *code, uschar *start_bits, BOOL caseless,
492 BOOL utf8, compile_data *cd)
493 {
494 register int c;
495 int yield = SSB_DONE;
496
497 #if 0
498 /* ========================================================================= */
499 /* The following comment and code was inserted in January 1999. In May 2006,
500 when it was observed to cause compiler warnings about unused values, I took it
501 out again. If anybody is still using OS/2, they will have to put it back
502 manually. */
503
504 /* This next statement and the later reference to dummy are here in order to
505 trick the optimizer of the IBM C compiler for OS/2 into generating correct
506 code. Apparently IBM isn't going to fix the problem, and we would rather not
507 disable optimization (in this module it actually makes a big difference, and
508 the pcre module can use all the optimization it can get). */
509
510 volatile int dummy;
511 /* ========================================================================= */
512 #endif
513
514 do
515 {
516 const uschar *tcode = code + (((int)*code == OP_CBRA)? 3:1) + LINK_SIZE;
517 BOOL try_next = TRUE;
518
519 while (try_next) /* Loop for items in this branch */
520 {
521 int rc;
522 switch(*tcode)
523 {
524 /* Fail if we reach something we don't understand */
525
526 default:
527 return SSB_FAIL;
528
529 /* If we hit a bracket or a positive lookahead assertion, recurse to set
530 bits from within the subpattern. If it can't find anything, we have to
531 give up. If it finds some mandatory character(s), we are done for this
532 branch. Otherwise, carry on scanning after the subpattern. */
533
534 case OP_BRA:
535 case OP_SBRA:
536 case OP_CBRA:
537 case OP_SCBRA:
538 case OP_ONCE:
539 case OP_ASSERT:
540 rc = set_start_bits(tcode, start_bits, caseless, utf8, cd);
541 if (rc == SSB_FAIL) return SSB_FAIL;
542 if (rc == SSB_DONE) try_next = FALSE; else
543 {
544 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
545 tcode += 1 + LINK_SIZE;
546 }
547 break;
548
549 /* If we hit ALT or KET, it means we haven't found anything mandatory in
550 this branch, though we might have found something optional. For ALT, we
551 continue with the next alternative, but we have to arrange that the final
552 result from subpattern is SSB_CONTINUE rather than SSB_DONE. For KET,
553 return SSB_CONTINUE: if this is the top level, that indicates failure,
554 but after a nested subpattern, it causes scanning to continue. */
555
556 case OP_ALT:
557 yield = SSB_CONTINUE;
558 try_next = FALSE;
559 break;
560
561 case OP_KET:
562 case OP_KETRMAX:
563 case OP_KETRMIN:
564 return SSB_CONTINUE;
565
566 /* Skip over callout */
567
568 case OP_CALLOUT:
569 tcode += 2 + 2*LINK_SIZE;
570 break;
571
572 /* Skip over lookbehind and negative lookahead assertions */
573
574 case OP_ASSERT_NOT:
575 case OP_ASSERTBACK:
576 case OP_ASSERTBACK_NOT:
577 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
578 tcode += 1 + LINK_SIZE;
579 break;
580
581 /* Skip over an option setting, changing the caseless flag */
582
583 case OP_OPT:
584 caseless = (tcode[1] & PCRE_CASELESS) != 0;
585 tcode += 2;
586 break;
587
588 /* BRAZERO does the bracket, but carries on. */
589
590 case OP_BRAZERO:
591 case OP_BRAMINZERO:
592 if (set_start_bits(++tcode, start_bits, caseless, utf8, cd) == SSB_FAIL)
593 return SSB_FAIL;
594 /* =========================================================================
595 See the comment at the head of this function concerning the next line,
596 which was an old fudge for the benefit of OS/2.
597 dummy = 1;
598 ========================================================================= */
599 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
600 tcode += 1 + LINK_SIZE;
601 break;
602
603 /* SKIPZERO skips the bracket. */
604
605 case OP_SKIPZERO:
606 tcode++;
607 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
608 tcode += 1 + LINK_SIZE;
609 break;
610
611 /* Single-char * or ? sets the bit and tries the next item */
612
613 case OP_STAR:
614 case OP_MINSTAR:
615 case OP_POSSTAR:
616 case OP_QUERY:
617 case OP_MINQUERY:
618 case OP_POSQUERY:
619 set_table_bit(start_bits, tcode[1], caseless, cd);
620 tcode += 2;
621 #ifdef SUPPORT_UTF8
622 if (utf8 && tcode[-1] >= 0xc0)
623 tcode += _pcre_utf8_table4[tcode[-1] & 0x3f];
624 #endif
625 break;
626
627 /* Single-char upto sets the bit and tries the next */
628
629 case OP_UPTO:
630 case OP_MINUPTO:
631 case OP_POSUPTO:
632 set_table_bit(start_bits, tcode[3], caseless, cd);
633 tcode += 4;
634 #ifdef SUPPORT_UTF8
635 if (utf8 && tcode[-1] >= 0xc0)
636 tcode += _pcre_utf8_table4[tcode[-1] & 0x3f];
637 #endif
638 break;
639
640 /* At least one single char sets the bit and stops */
641
642 case OP_EXACT: /* Fall through */
643 tcode += 2;
644
645 case OP_CHAR:
646 case OP_CHARNC:
647 case OP_PLUS:
648 case OP_MINPLUS:
649 case OP_POSPLUS:
650 set_table_bit(start_bits, tcode[1], caseless, cd);
651 try_next = FALSE;
652 break;
653
654 /* Single character type sets the bits and stops */
655
656 case OP_NOT_DIGIT:
657 for (c = 0; c < 32; c++)
658 start_bits[c] |= ~cd->cbits[c+cbit_digit];
659 try_next = FALSE;
660 break;
661
662 case OP_DIGIT:
663 for (c = 0; c < 32; c++)
664 start_bits[c] |= cd->cbits[c+cbit_digit];
665 try_next = FALSE;
666 break;
667
668 /* The cbit_space table has vertical tab as whitespace; we have to
669 discard it. */
670
671 case OP_NOT_WHITESPACE:
672 for (c = 0; c < 32; c++)
673 {
674 int d = cd->cbits[c+cbit_space];
675 if (c == 1) d &= ~0x08;
676 start_bits[c] |= ~d;
677 }
678 try_next = FALSE;
679 break;
680
681 /* The cbit_space table has vertical tab as whitespace; we have to
682 discard it. */
683
684 case OP_WHITESPACE:
685 for (c = 0; c < 32; c++)
686 {
687 int d = cd->cbits[c+cbit_space];
688 if (c == 1) d &= ~0x08;
689 start_bits[c] |= d;
690 }
691 try_next = FALSE;
692 break;
693
694 case OP_NOT_WORDCHAR:
695 for (c = 0; c < 32; c++)
696 start_bits[c] |= ~cd->cbits[c+cbit_word];
697 try_next = FALSE;
698 break;
699
700 case OP_WORDCHAR:
701 for (c = 0; c < 32; c++)
702 start_bits[c] |= cd->cbits[c+cbit_word];
703 try_next = FALSE;
704 break;
705
706 /* One or more character type fudges the pointer and restarts, knowing
707 it will hit a single character type and stop there. */
708
709 case OP_TYPEPLUS:
710 case OP_TYPEMINPLUS:
711 tcode++;
712 break;
713
714 case OP_TYPEEXACT:
715 tcode += 3;
716 break;
717
718 /* Zero or more repeats of character types set the bits and then
719 try again. */
720
721 case OP_TYPEUPTO:
722 case OP_TYPEMINUPTO:
723 case OP_TYPEPOSUPTO:
724 tcode += 2; /* Fall through */
725
726 case OP_TYPESTAR:
727 case OP_TYPEMINSTAR:
728 case OP_TYPEPOSSTAR:
729 case OP_TYPEQUERY:
730 case OP_TYPEMINQUERY:
731 case OP_TYPEPOSQUERY:
732 switch(tcode[1])
733 {
734 case OP_ANY:
735 case OP_ALLANY:
736 return SSB_FAIL;
737
738 case OP_NOT_DIGIT:
739 for (c = 0; c < 32; c++)
740 start_bits[c] |= ~cd->cbits[c+cbit_digit];
741 break;
742
743 case OP_DIGIT:
744 for (c = 0; c < 32; c++)
745 start_bits[c] |= cd->cbits[c+cbit_digit];
746 break;
747
748 /* The cbit_space table has vertical tab as whitespace; we have to
749 discard it. */
750
751 case OP_NOT_WHITESPACE:
752 for (c = 0; c < 32; c++)
753 {
754 int d = cd->cbits[c+cbit_space];
755 if (c == 1) d &= ~0x08;
756 start_bits[c] |= ~d;
757 }
758 break;
759
760 /* The cbit_space table has vertical tab as whitespace; we have to
761 discard it. */
762
763 case OP_WHITESPACE:
764 for (c = 0; c < 32; c++)
765 {
766 int d = cd->cbits[c+cbit_space];
767 if (c == 1) d &= ~0x08;
768 start_bits[c] |= d;
769 }
770 break;
771
772 case OP_NOT_WORDCHAR:
773 for (c = 0; c < 32; c++)
774 start_bits[c] |= ~cd->cbits[c+cbit_word];
775 break;
776
777 case OP_WORDCHAR:
778 for (c = 0; c < 32; c++)
779 start_bits[c] |= cd->cbits[c+cbit_word];
780 break;
781 }
782
783 tcode += 2;
784 break;
785
786 /* Character class where all the information is in a bit map: set the
787 bits and either carry on or not, according to the repeat count. If it was
788 a negative class, and we are operating with UTF-8 characters, any byte
789 with a value >= 0xc4 is a potentially valid starter because it starts a
790 character with a value > 255. */
791
792 case OP_NCLASS:
793 #ifdef SUPPORT_UTF8
794 if (utf8)
795 {
796 start_bits[24] |= 0xf0; /* Bits for 0xc4 - 0xc8 */
797 memset(start_bits+25, 0xff, 7); /* Bits for 0xc9 - 0xff */
798 }
799 #endif
800 /* Fall through */
801
802 case OP_CLASS:
803 {
804 tcode++;
805
806 /* In UTF-8 mode, the bits in a bit map correspond to character
807 values, not to byte values. However, the bit map we are constructing is
808 for byte values. So we have to do a conversion for characters whose
809 value is > 127. In fact, there are only two possible starting bytes for
810 characters in the range 128 - 255. */
811
812 #ifdef SUPPORT_UTF8
813 if (utf8)
814 {
815 for (c = 0; c < 16; c++) start_bits[c] |= tcode[c];
816 for (c = 128; c < 256; c++)
817 {
818 if ((tcode[c/8] && (1 << (c&7))) != 0)
819 {
820 int d = (c >> 6) | 0xc0; /* Set bit for this starter */
821 start_bits[d/8] |= (1 << (d&7)); /* and then skip on to the */
822 c = (c & 0xc0) + 0x40 - 1; /* next relevant character. */
823 }
824 }
825 }
826
827 /* In non-UTF-8 mode, the two bit maps are completely compatible. */
828
829 else
830 #endif
831 {
832 for (c = 0; c < 32; c++) start_bits[c] |= tcode[c];
833 }
834
835 /* Advance past the bit map, and act on what follows */
836
837 tcode += 32;
838 switch (*tcode)
839 {
840 case OP_CRSTAR:
841 case OP_CRMINSTAR:
842 case OP_CRQUERY:
843 case OP_CRMINQUERY:
844 tcode++;
845 break;
846
847 case OP_CRRANGE:
848 case OP_CRMINRANGE:
849 if (((tcode[1] << 8) + tcode[2]) == 0) tcode += 5;
850 else try_next = FALSE;
851 break;
852
853 default:
854 try_next = FALSE;
855 break;
856 }
857 }
858 break; /* End of bitmap class handling */
859
860 } /* End of switch */
861 } /* End of try_next loop */
862
863 code += GET(code, 1); /* Advance to next branch */
864 }
865 while (*code == OP_ALT);
866 return yield;
867 }
868
869
870
871 /*************************************************
872 * Study a compiled expression *
873 *************************************************/
874
875 /* This function is handed a compiled expression that it must study to produce
876 information that will speed up the matching. It returns a pcre_extra block
877 which then gets handed back to pcre_exec().
878
879 Arguments:
880 re points to the compiled expression
881 options contains option bits
882 errorptr points to where to place error messages;
883 set NULL unless error
884
885 Returns: pointer to a pcre_extra block, with study_data filled in and the
886 appropriate flags set;
887 NULL on error or if no optimization possible
888 */
889
890 PCRE_EXP_DEFN pcre_extra * PCRE_CALL_CONVENTION
891 pcre_study(const pcre *external_re, int options, const char **errorptr)
892 {
893 int min;
894 BOOL bits_set = FALSE;
895 uschar start_bits[32];
896 pcre_extra *extra;
897 pcre_study_data *study;
898 const uschar *tables;
899 uschar *code;
900 compile_data compile_block;
901 const real_pcre *re = (const real_pcre *)external_re;
902
903 *errorptr = NULL;
904
905 if (re == NULL || re->magic_number != MAGIC_NUMBER)
906 {
907 *errorptr = "argument is not a compiled regular expression";
908 return NULL;
909 }
910
911 if ((options & ~PUBLIC_STUDY_OPTIONS) != 0)
912 {
913 *errorptr = "unknown or incorrect option bit(s) set";
914 return NULL;
915 }
916
917 code = (uschar *)re + re->name_table_offset +
918 (re->name_count * re->name_entry_size);
919
920 /* For an anchored pattern, or an unanchored pattern that has a first char, or
921 a multiline pattern that matches only at "line starts", there is no point in
922 seeking a list of starting bytes. */
923
924 if ((re->options & PCRE_ANCHORED) == 0 &&
925 (re->flags & (PCRE_FIRSTSET|PCRE_STARTLINE)) == 0)
926 {
927 /* Set the character tables in the block that is passed around */
928
929 tables = re->tables;
930 if (tables == NULL)
931 (void)pcre_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES,
932 (void *)(&tables));
933
934 compile_block.lcc = tables + lcc_offset;
935 compile_block.fcc = tables + fcc_offset;
936 compile_block.cbits = tables + cbits_offset;
937 compile_block.ctypes = tables + ctypes_offset;
938
939 /* See if we can find a fixed set of initial characters for the pattern. */
940
941 memset(start_bits, 0, 32 * sizeof(uschar));
942 bits_set = set_start_bits(code, start_bits,
943 (re->options & PCRE_CASELESS) != 0, (re->options & PCRE_UTF8) != 0,
944 &compile_block) == SSB_DONE;
945 }
946
947 /* Find the minimum length of subject string. */
948
949 min = find_minlength(code, code, re->options);
950
951 /* Return NULL if no optimization is possible. */
952
953 if (!bits_set && min < 0) return NULL;
954
955 /* Get a pcre_extra block and a pcre_study_data block. The study data is put in
956 the latter, which is pointed to by the former, which may also get additional
957 data set later by the calling program. At the moment, the size of
958 pcre_study_data is fixed. We nevertheless save it in a field for returning via
959 the pcre_fullinfo() function so that if it becomes variable in the future, we
960 don't have to change that code. */
961
962 extra = (pcre_extra *)(pcre_malloc)
963 (sizeof(pcre_extra) + sizeof(pcre_study_data));
964
965 if (extra == NULL)
966 {
967 *errorptr = "failed to get memory";
968 return NULL;
969 }
970
971 study = (pcre_study_data *)((char *)extra + sizeof(pcre_extra));
972 extra->flags = PCRE_EXTRA_STUDY_DATA;
973 extra->study_data = study;
974
975 study->size = sizeof(pcre_study_data);
976 study->flags = 0;
977
978 if (bits_set)
979 {
980 study->flags |= PCRE_STUDY_MAPPED;
981 memcpy(study->start_bits, start_bits, sizeof(start_bits));
982 }
983
984 if (min >= 0)
985 {
986 study->flags |= PCRE_STUDY_MINLEN;
987 study->minlength = min;
988 }
989
990 return extra;
991 }
992
993 /* 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