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

Contents of /code/trunk/pcre_study.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 471 - (show annotations) (download)
Fri Dec 11 15:11:55 2009 UTC (4 years, 4 months ago) by ph10
File MIME type: text/plain
File size: 27418 byte(s)
Fix study bug with single-branch conditions, including (DEFINE).

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) 1997-2009 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 /* For the record, these are the opcodes that are matched by "default":
417 OP_ACCEPT, OP_CLOSE, OP_COMMIT, OP_FAIL, OP_PRUNE, OP_SET_SOM, OP_SKIP,
418 OP_THEN. */
419
420 default:
421 cc += _pcre_OP_lengths[op];
422 break;
423 }
424 }
425 /* Control never gets here */
426 }
427
428
429
430 /*************************************************
431 * Set a bit and maybe its alternate case *
432 *************************************************/
433
434 /* Given a character, set its bit in the table, and also the bit for the other
435 version of a letter if we are caseless.
436
437 Arguments:
438 start_bits points to the bit map
439 c is the character
440 caseless the caseless flag
441 cd the block with char table pointers
442
443 Returns: nothing
444 */
445
446 static void
447 set_bit(uschar *start_bits, unsigned int c, BOOL caseless, compile_data *cd)
448 {
449 start_bits[c/8] |= (1 << (c&7));
450 if (caseless && (cd->ctypes[c] & ctype_letter) != 0)
451 start_bits[cd->fcc[c]/8] |= (1 << (cd->fcc[c]&7));
452 }
453
454
455
456 /*************************************************
457 * Create bitmap of starting bytes *
458 *************************************************/
459
460 /* This function scans a compiled unanchored expression recursively and
461 attempts to build a bitmap of the set of possible starting bytes. As time goes
462 by, we may be able to get more clever at doing this. The SSB_CONTINUE return is
463 useful for parenthesized groups in patterns such as (a*)b where the group
464 provides some optional starting bytes but scanning must continue at the outer
465 level to find at least one mandatory byte. At the outermost level, this
466 function fails unless the result is SSB_DONE.
467
468 Arguments:
469 code points to an expression
470 start_bits points to a 32-byte table, initialized to 0
471 caseless the current state of the caseless flag
472 utf8 TRUE if in UTF-8 mode
473 cd the block with char table pointers
474
475 Returns: SSB_FAIL => Failed to find any starting bytes
476 SSB_DONE => Found mandatory starting bytes
477 SSB_CONTINUE => Found optional starting bytes
478 */
479
480 static int
481 set_start_bits(const uschar *code, uschar *start_bits, BOOL caseless,
482 BOOL utf8, compile_data *cd)
483 {
484 register int c;
485 int yield = SSB_DONE;
486
487 #if 0
488 /* ========================================================================= */
489 /* The following comment and code was inserted in January 1999. In May 2006,
490 when it was observed to cause compiler warnings about unused values, I took it
491 out again. If anybody is still using OS/2, they will have to put it back
492 manually. */
493
494 /* This next statement and the later reference to dummy are here in order to
495 trick the optimizer of the IBM C compiler for OS/2 into generating correct
496 code. Apparently IBM isn't going to fix the problem, and we would rather not
497 disable optimization (in this module it actually makes a big difference, and
498 the pcre module can use all the optimization it can get). */
499
500 volatile int dummy;
501 /* ========================================================================= */
502 #endif
503
504 do
505 {
506 const uschar *tcode = code + (((int)*code == OP_CBRA)? 3:1) + LINK_SIZE;
507 BOOL try_next = TRUE;
508
509 while (try_next) /* Loop for items in this branch */
510 {
511 int rc;
512 switch(*tcode)
513 {
514 /* Fail if we reach something we don't understand */
515
516 default:
517 return SSB_FAIL;
518
519 /* If we hit a bracket or a positive lookahead assertion, recurse to set
520 bits from within the subpattern. If it can't find anything, we have to
521 give up. If it finds some mandatory character(s), we are done for this
522 branch. Otherwise, carry on scanning after the subpattern. */
523
524 case OP_BRA:
525 case OP_SBRA:
526 case OP_CBRA:
527 case OP_SCBRA:
528 case OP_ONCE:
529 case OP_ASSERT:
530 rc = set_start_bits(tcode, start_bits, caseless, utf8, cd);
531 if (rc == SSB_FAIL) return SSB_FAIL;
532 if (rc == SSB_DONE) try_next = FALSE; else
533 {
534 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
535 tcode += 1 + LINK_SIZE;
536 }
537 break;
538
539 /* If we hit ALT or KET, it means we haven't found anything mandatory in
540 this branch, though we might have found something optional. For ALT, we
541 continue with the next alternative, but we have to arrange that the final
542 result from subpattern is SSB_CONTINUE rather than SSB_DONE. For KET,
543 return SSB_CONTINUE: if this is the top level, that indicates failure,
544 but after a nested subpattern, it causes scanning to continue. */
545
546 case OP_ALT:
547 yield = SSB_CONTINUE;
548 try_next = FALSE;
549 break;
550
551 case OP_KET:
552 case OP_KETRMAX:
553 case OP_KETRMIN:
554 return SSB_CONTINUE;
555
556 /* Skip over callout */
557
558 case OP_CALLOUT:
559 tcode += 2 + 2*LINK_SIZE;
560 break;
561
562 /* Skip over lookbehind and negative lookahead assertions */
563
564 case OP_ASSERT_NOT:
565 case OP_ASSERTBACK:
566 case OP_ASSERTBACK_NOT:
567 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
568 tcode += 1 + LINK_SIZE;
569 break;
570
571 /* Skip over an option setting, changing the caseless flag */
572
573 case OP_OPT:
574 caseless = (tcode[1] & PCRE_CASELESS) != 0;
575 tcode += 2;
576 break;
577
578 /* BRAZERO does the bracket, but carries on. */
579
580 case OP_BRAZERO:
581 case OP_BRAMINZERO:
582 if (set_start_bits(++tcode, start_bits, caseless, utf8, cd) == SSB_FAIL)
583 return SSB_FAIL;
584 /* =========================================================================
585 See the comment at the head of this function concerning the next line,
586 which was an old fudge for the benefit of OS/2.
587 dummy = 1;
588 ========================================================================= */
589 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
590 tcode += 1 + LINK_SIZE;
591 break;
592
593 /* SKIPZERO skips the bracket. */
594
595 case OP_SKIPZERO:
596 tcode++;
597 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
598 tcode += 1 + LINK_SIZE;
599 break;
600
601 /* Single-char * or ? sets the bit and tries the next item */
602
603 case OP_STAR:
604 case OP_MINSTAR:
605 case OP_POSSTAR:
606 case OP_QUERY:
607 case OP_MINQUERY:
608 case OP_POSQUERY:
609 set_bit(start_bits, tcode[1], caseless, cd);
610 tcode += 2;
611 #ifdef SUPPORT_UTF8
612 if (utf8 && tcode[-1] >= 0xc0)
613 tcode += _pcre_utf8_table4[tcode[-1] & 0x3f];
614 #endif
615 break;
616
617 /* Single-char upto sets the bit and tries the next */
618
619 case OP_UPTO:
620 case OP_MINUPTO:
621 case OP_POSUPTO:
622 set_bit(start_bits, tcode[3], caseless, cd);
623 tcode += 4;
624 #ifdef SUPPORT_UTF8
625 if (utf8 && tcode[-1] >= 0xc0)
626 tcode += _pcre_utf8_table4[tcode[-1] & 0x3f];
627 #endif
628 break;
629
630 /* At least one single char sets the bit and stops */
631
632 case OP_EXACT: /* Fall through */
633 tcode += 2;
634
635 case OP_CHAR:
636 case OP_CHARNC:
637 case OP_PLUS:
638 case OP_MINPLUS:
639 case OP_POSPLUS:
640 set_bit(start_bits, tcode[1], caseless, cd);
641 try_next = FALSE;
642 break;
643
644 /* Single character type sets the bits and stops */
645
646 case OP_NOT_DIGIT:
647 for (c = 0; c < 32; c++)
648 start_bits[c] |= ~cd->cbits[c+cbit_digit];
649 try_next = FALSE;
650 break;
651
652 case OP_DIGIT:
653 for (c = 0; c < 32; c++)
654 start_bits[c] |= cd->cbits[c+cbit_digit];
655 try_next = FALSE;
656 break;
657
658 /* The cbit_space table has vertical tab as whitespace; we have to
659 discard it. */
660
661 case OP_NOT_WHITESPACE:
662 for (c = 0; c < 32; c++)
663 {
664 int d = cd->cbits[c+cbit_space];
665 if (c == 1) d &= ~0x08;
666 start_bits[c] |= ~d;
667 }
668 try_next = FALSE;
669 break;
670
671 /* The cbit_space table has vertical tab as whitespace; we have to
672 discard it. */
673
674 case OP_WHITESPACE:
675 for (c = 0; c < 32; c++)
676 {
677 int d = cd->cbits[c+cbit_space];
678 if (c == 1) d &= ~0x08;
679 start_bits[c] |= d;
680 }
681 try_next = FALSE;
682 break;
683
684 case OP_NOT_WORDCHAR:
685 for (c = 0; c < 32; c++)
686 start_bits[c] |= ~cd->cbits[c+cbit_word];
687 try_next = FALSE;
688 break;
689
690 case OP_WORDCHAR:
691 for (c = 0; c < 32; c++)
692 start_bits[c] |= cd->cbits[c+cbit_word];
693 try_next = FALSE;
694 break;
695
696 /* One or more character type fudges the pointer and restarts, knowing
697 it will hit a single character type and stop there. */
698
699 case OP_TYPEPLUS:
700 case OP_TYPEMINPLUS:
701 tcode++;
702 break;
703
704 case OP_TYPEEXACT:
705 tcode += 3;
706 break;
707
708 /* Zero or more repeats of character types set the bits and then
709 try again. */
710
711 case OP_TYPEUPTO:
712 case OP_TYPEMINUPTO:
713 case OP_TYPEPOSUPTO:
714 tcode += 2; /* Fall through */
715
716 case OP_TYPESTAR:
717 case OP_TYPEMINSTAR:
718 case OP_TYPEPOSSTAR:
719 case OP_TYPEQUERY:
720 case OP_TYPEMINQUERY:
721 case OP_TYPEPOSQUERY:
722 switch(tcode[1])
723 {
724 case OP_ANY:
725 case OP_ALLANY:
726 return SSB_FAIL;
727
728 case OP_NOT_DIGIT:
729 for (c = 0; c < 32; c++)
730 start_bits[c] |= ~cd->cbits[c+cbit_digit];
731 break;
732
733 case OP_DIGIT:
734 for (c = 0; c < 32; c++)
735 start_bits[c] |= cd->cbits[c+cbit_digit];
736 break;
737
738 /* The cbit_space table has vertical tab as whitespace; we have to
739 discard it. */
740
741 case OP_NOT_WHITESPACE:
742 for (c = 0; c < 32; c++)
743 {
744 int d = cd->cbits[c+cbit_space];
745 if (c == 1) d &= ~0x08;
746 start_bits[c] |= ~d;
747 }
748 break;
749
750 /* The cbit_space table has vertical tab as whitespace; we have to
751 discard it. */
752
753 case OP_WHITESPACE:
754 for (c = 0; c < 32; c++)
755 {
756 int d = cd->cbits[c+cbit_space];
757 if (c == 1) d &= ~0x08;
758 start_bits[c] |= d;
759 }
760 break;
761
762 case OP_NOT_WORDCHAR:
763 for (c = 0; c < 32; c++)
764 start_bits[c] |= ~cd->cbits[c+cbit_word];
765 break;
766
767 case OP_WORDCHAR:
768 for (c = 0; c < 32; c++)
769 start_bits[c] |= cd->cbits[c+cbit_word];
770 break;
771 }
772
773 tcode += 2;
774 break;
775
776 /* Character class where all the information is in a bit map: set the
777 bits and either carry on or not, according to the repeat count. If it was
778 a negative class, and we are operating with UTF-8 characters, any byte
779 with a value >= 0xc4 is a potentially valid starter because it starts a
780 character with a value > 255. */
781
782 case OP_NCLASS:
783 #ifdef SUPPORT_UTF8
784 if (utf8)
785 {
786 start_bits[24] |= 0xf0; /* Bits for 0xc4 - 0xc8 */
787 memset(start_bits+25, 0xff, 7); /* Bits for 0xc9 - 0xff */
788 }
789 #endif
790 /* Fall through */
791
792 case OP_CLASS:
793 {
794 tcode++;
795
796 /* In UTF-8 mode, the bits in a bit map correspond to character
797 values, not to byte values. However, the bit map we are constructing is
798 for byte values. So we have to do a conversion for characters whose
799 value is > 127. In fact, there are only two possible starting bytes for
800 characters in the range 128 - 255. */
801
802 #ifdef SUPPORT_UTF8
803 if (utf8)
804 {
805 for (c = 0; c < 16; c++) start_bits[c] |= tcode[c];
806 for (c = 128; c < 256; c++)
807 {
808 if ((tcode[c/8] && (1 << (c&7))) != 0)
809 {
810 int d = (c >> 6) | 0xc0; /* Set bit for this starter */
811 start_bits[d/8] |= (1 << (d&7)); /* and then skip on to the */
812 c = (c & 0xc0) + 0x40 - 1; /* next relevant character. */
813 }
814 }
815 }
816
817 /* In non-UTF-8 mode, the two bit maps are completely compatible. */
818
819 else
820 #endif
821 {
822 for (c = 0; c < 32; c++) start_bits[c] |= tcode[c];
823 }
824
825 /* Advance past the bit map, and act on what follows */
826
827 tcode += 32;
828 switch (*tcode)
829 {
830 case OP_CRSTAR:
831 case OP_CRMINSTAR:
832 case OP_CRQUERY:
833 case OP_CRMINQUERY:
834 tcode++;
835 break;
836
837 case OP_CRRANGE:
838 case OP_CRMINRANGE:
839 if (((tcode[1] << 8) + tcode[2]) == 0) tcode += 5;
840 else try_next = FALSE;
841 break;
842
843 default:
844 try_next = FALSE;
845 break;
846 }
847 }
848 break; /* End of bitmap class handling */
849
850 } /* End of switch */
851 } /* End of try_next loop */
852
853 code += GET(code, 1); /* Advance to next branch */
854 }
855 while (*code == OP_ALT);
856 return yield;
857 }
858
859
860
861 /*************************************************
862 * Study a compiled expression *
863 *************************************************/
864
865 /* This function is handed a compiled expression that it must study to produce
866 information that will speed up the matching. It returns a pcre_extra block
867 which then gets handed back to pcre_exec().
868
869 Arguments:
870 re points to the compiled expression
871 options contains option bits
872 errorptr points to where to place error messages;
873 set NULL unless error
874
875 Returns: pointer to a pcre_extra block, with study_data filled in and the
876 appropriate flags set;
877 NULL on error or if no optimization possible
878 */
879
880 PCRE_EXP_DEFN pcre_extra * PCRE_CALL_CONVENTION
881 pcre_study(const pcre *external_re, int options, const char **errorptr)
882 {
883 int min;
884 BOOL bits_set = FALSE;
885 uschar start_bits[32];
886 pcre_extra *extra;
887 pcre_study_data *study;
888 const uschar *tables;
889 uschar *code;
890 compile_data compile_block;
891 const real_pcre *re = (const real_pcre *)external_re;
892
893 *errorptr = NULL;
894
895 if (re == NULL || re->magic_number != MAGIC_NUMBER)
896 {
897 *errorptr = "argument is not a compiled regular expression";
898 return NULL;
899 }
900
901 if ((options & ~PUBLIC_STUDY_OPTIONS) != 0)
902 {
903 *errorptr = "unknown or incorrect option bit(s) set";
904 return NULL;
905 }
906
907 code = (uschar *)re + re->name_table_offset +
908 (re->name_count * re->name_entry_size);
909
910 /* For an anchored pattern, or an unanchored pattern that has a first char, or
911 a multiline pattern that matches only at "line starts", there is no point in
912 seeking a list of starting bytes. */
913
914 if ((re->options & PCRE_ANCHORED) == 0 &&
915 (re->flags & (PCRE_FIRSTSET|PCRE_STARTLINE)) == 0)
916 {
917 /* Set the character tables in the block that is passed around */
918
919 tables = re->tables;
920 if (tables == NULL)
921 (void)pcre_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES,
922 (void *)(&tables));
923
924 compile_block.lcc = tables + lcc_offset;
925 compile_block.fcc = tables + fcc_offset;
926 compile_block.cbits = tables + cbits_offset;
927 compile_block.ctypes = tables + ctypes_offset;
928
929 /* See if we can find a fixed set of initial characters for the pattern. */
930
931 memset(start_bits, 0, 32 * sizeof(uschar));
932 bits_set = set_start_bits(code, start_bits,
933 (re->options & PCRE_CASELESS) != 0, (re->options & PCRE_UTF8) != 0,
934 &compile_block) == SSB_DONE;
935 }
936
937 /* Find the minimum length of subject string. */
938
939 min = find_minlength(code, code, re->options);
940
941 /* Return NULL if no optimization is possible. */
942
943 if (!bits_set && min < 0) return NULL;
944
945 /* Get a pcre_extra block and a pcre_study_data block. The study data is put in
946 the latter, which is pointed to by the former, which may also get additional
947 data set later by the calling program. At the moment, the size of
948 pcre_study_data is fixed. We nevertheless save it in a field for returning via
949 the pcre_fullinfo() function so that if it becomes variable in the future, we
950 don't have to change that code. */
951
952 extra = (pcre_extra *)(pcre_malloc)
953 (sizeof(pcre_extra) + sizeof(pcre_study_data));
954
955 if (extra == NULL)
956 {
957 *errorptr = "failed to get memory";
958 return NULL;
959 }
960
961 study = (pcre_study_data *)((char *)extra + sizeof(pcre_extra));
962 extra->flags = PCRE_EXTRA_STUDY_DATA;
963 extra->study_data = study;
964
965 study->size = sizeof(pcre_study_data);
966 study->flags = 0;
967
968 if (bits_set)
969 {
970 study->flags |= PCRE_STUDY_MAPPED;
971 memcpy(study->start_bits, start_bits, sizeof(start_bits));
972 }
973
974 if (min >= 0)
975 {
976 study->flags |= PCRE_STUDY_MINLEN;
977 study->minlength = min;
978 }
979
980 return extra;
981 }
982
983 /* 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