/[pcre]/code/branches/pcre16/pcre_study.c
ViewVC logotype

Contents of /code/branches/pcre16/pcre_study.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 769 - (show annotations)
Sun Nov 27 17:40:33 2011 UTC (9 years, 4 months ago) by zherczeg
File MIME type: text/plain
File size: 39521 byte(s)
Update those opcode sizes whose depend on IMM2_SIZE
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, SSB_UNKNOWN };
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 int RECURSE depth
73
74 Returns: the minimum length
75 -1 if \C in UTF-8 mode or (*ACCEPT) was encountered
76 -2 internal error (missing capturing bracket)
77 -3 internal error (opcode not listed)
78 */
79
80 static int
81 find_minlength(const pcre_uchar *code, const pcre_uchar *startcode, int options,
82 int recurse_depth)
83 {
84 int length = -1;
85 BOOL utf8 = (options & PCRE_UTF8) != 0;
86 BOOL had_recurse = FALSE;
87 register int branchlength = 0;
88 register pcre_uchar *cc = (pcre_uchar *)code + 1 + LINK_SIZE;
89
90 if (*code == OP_CBRA || *code == OP_SCBRA ||
91 *code == OP_CBRAPOS || *code == OP_SCBRAPOS) cc += IMM2_SIZE;
92
93 /* Scan along the opcodes for this branch. If we get to the end of the
94 branch, check the length against that of the other branches. */
95
96 for (;;)
97 {
98 int d, min;
99 pcre_uchar *cs, *ce;
100 register int op = *cc;
101
102 switch (op)
103 {
104 case OP_COND:
105 case OP_SCOND:
106
107 /* If there is only one branch in a condition, the implied branch has zero
108 length, so we don't add anything. This covers the DEFINE "condition"
109 automatically. */
110
111 cs = cc + GET(cc, 1);
112 if (*cs != OP_ALT)
113 {
114 cc = cs + 1 + LINK_SIZE;
115 break;
116 }
117
118 /* Otherwise we can fall through and treat it the same as any other
119 subpattern. */
120
121 case OP_CBRA:
122 case OP_SCBRA:
123 case OP_BRA:
124 case OP_SBRA:
125 case OP_CBRAPOS:
126 case OP_SCBRAPOS:
127 case OP_BRAPOS:
128 case OP_SBRAPOS:
129 case OP_ONCE:
130 case OP_ONCE_NC:
131 d = find_minlength(cc, startcode, options, recurse_depth);
132 if (d < 0) return d;
133 branchlength += d;
134 do cc += GET(cc, 1); while (*cc == OP_ALT);
135 cc += 1 + LINK_SIZE;
136 break;
137
138 /* ACCEPT makes things far too complicated; we have to give up. */
139
140 case OP_ACCEPT:
141 case OP_ASSERT_ACCEPT:
142 return -1;
143
144 /* Reached end of a branch; if it's a ket it is the end of a nested
145 call. If it's ALT it is an alternation in a nested call. If it is END it's
146 the end of the outer call. All can be handled by the same code. If an
147 ACCEPT was previously encountered, use the length that was in force at that
148 time, and pass back the shortest ACCEPT length. */
149
150 case OP_ALT:
151 case OP_KET:
152 case OP_KETRMAX:
153 case OP_KETRMIN:
154 case OP_KETRPOS:
155 case OP_END:
156 if (length < 0 || (!had_recurse && branchlength < length))
157 length = branchlength;
158 if (op != OP_ALT) return length;
159 cc += 1 + LINK_SIZE;
160 branchlength = 0;
161 had_recurse = FALSE;
162 break;
163
164 /* Skip over assertive subpatterns */
165
166 case OP_ASSERT:
167 case OP_ASSERT_NOT:
168 case OP_ASSERTBACK:
169 case OP_ASSERTBACK_NOT:
170 do cc += GET(cc, 1); while (*cc == OP_ALT);
171 /* Fall through */
172
173 /* Skip over things that don't match chars */
174
175 case OP_REVERSE:
176 case OP_CREF:
177 case OP_NCREF:
178 case OP_RREF:
179 case OP_NRREF:
180 case OP_DEF:
181 case OP_CALLOUT:
182 case OP_SOD:
183 case OP_SOM:
184 case OP_EOD:
185 case OP_EODN:
186 case OP_CIRC:
187 case OP_CIRCM:
188 case OP_DOLL:
189 case OP_DOLLM:
190 case OP_NOT_WORD_BOUNDARY:
191 case OP_WORD_BOUNDARY:
192 cc += PRIV(OP_lengths)[*cc];
193 break;
194
195 /* Skip over a subpattern that has a {0} or {0,x} quantifier */
196
197 case OP_BRAZERO:
198 case OP_BRAMINZERO:
199 case OP_BRAPOSZERO:
200 case OP_SKIPZERO:
201 cc += PRIV(OP_lengths)[*cc];
202 do cc += GET(cc, 1); while (*cc == OP_ALT);
203 cc += 1 + LINK_SIZE;
204 break;
205
206 /* Handle literal characters and + repetitions */
207
208 case OP_CHAR:
209 case OP_CHARI:
210 case OP_NOT:
211 case OP_NOTI:
212 case OP_PLUS:
213 case OP_PLUSI:
214 case OP_MINPLUS:
215 case OP_MINPLUSI:
216 case OP_POSPLUS:
217 case OP_POSPLUSI:
218 case OP_NOTPLUS:
219 case OP_NOTPLUSI:
220 case OP_NOTMINPLUS:
221 case OP_NOTMINPLUSI:
222 case OP_NOTPOSPLUS:
223 case OP_NOTPOSPLUSI:
224 branchlength++;
225 cc += 2;
226 #ifdef SUPPORT_UTF8
227 if (utf8 && cc[-1] >= 0xc0) cc += PRIV(utf8_table4)[cc[-1] & 0x3f];
228 #endif
229 break;
230
231 case OP_TYPEPLUS:
232 case OP_TYPEMINPLUS:
233 case OP_TYPEPOSPLUS:
234 branchlength++;
235 cc += (cc[1] == OP_PROP || cc[1] == OP_NOTPROP)? 4 : 2;
236 break;
237
238 /* Handle exact repetitions. The count is already in characters, but we
239 need to skip over a multibyte character in UTF8 mode. */
240
241 case OP_EXACT:
242 case OP_EXACTI:
243 case OP_NOTEXACT:
244 case OP_NOTEXACTI:
245 branchlength += GET2(cc,1);
246 cc += 2 + IMM2_SIZE;
247 #ifdef SUPPORT_UTF8
248 if (utf8 && cc[-1] >= 0xc0) cc += PRIV(utf8_table4)[cc[-1] & 0x3f];
249 #endif
250 break;
251
252 case OP_TYPEEXACT:
253 branchlength += GET2(cc,1);
254 cc += 2 + IMM2_SIZE + ((cc[1 + IMM2_SIZE] == OP_PROP
255 || cc[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
256 break;
257
258 /* Handle single-char non-literal matchers */
259
260 case OP_PROP:
261 case OP_NOTPROP:
262 cc += 2;
263 /* Fall through */
264
265 case OP_NOT_DIGIT:
266 case OP_DIGIT:
267 case OP_NOT_WHITESPACE:
268 case OP_WHITESPACE:
269 case OP_NOT_WORDCHAR:
270 case OP_WORDCHAR:
271 case OP_ANY:
272 case OP_ALLANY:
273 case OP_EXTUNI:
274 case OP_HSPACE:
275 case OP_NOT_HSPACE:
276 case OP_VSPACE:
277 case OP_NOT_VSPACE:
278 branchlength++;
279 cc++;
280 break;
281
282 /* "Any newline" might match two characters, but it also might match just
283 one. */
284
285 case OP_ANYNL:
286 branchlength += 1;
287 cc++;
288 break;
289
290 /* The single-byte matcher means we can't proceed in UTF-8 mode. (In
291 non-UTF-8 mode \C will actually be turned into OP_ALLANY, so won't ever
292 appear, but leave the code, just in case.) */
293
294 case OP_ANYBYTE:
295 #ifdef SUPPORT_UTF8
296 if (utf8) return -1;
297 #endif
298 branchlength++;
299 cc++;
300 break;
301
302 /* For repeated character types, we have to test for \p and \P, which have
303 an extra two bytes of parameters. */
304
305 case OP_TYPESTAR:
306 case OP_TYPEMINSTAR:
307 case OP_TYPEQUERY:
308 case OP_TYPEMINQUERY:
309 case OP_TYPEPOSSTAR:
310 case OP_TYPEPOSQUERY:
311 if (cc[1] == OP_PROP || cc[1] == OP_NOTPROP) cc += 2;
312 cc += PRIV(OP_lengths)[op];
313 break;
314
315 case OP_TYPEUPTO:
316 case OP_TYPEMINUPTO:
317 case OP_TYPEPOSUPTO:
318 if (cc[1 + IMM2_SIZE] == OP_PROP
319 || cc[1 + IMM2_SIZE] == OP_NOTPROP) cc += 2;
320 cc += PRIV(OP_lengths)[op];
321 break;
322
323 /* Check a class for variable quantification */
324
325 #ifdef SUPPORT_UTF8
326 case OP_XCLASS:
327 cc += GET(cc, 1) - 33;
328 /* Fall through */
329 #endif
330
331 case OP_CLASS:
332 case OP_NCLASS:
333 cc += 33;
334
335 switch (*cc)
336 {
337 case OP_CRPLUS:
338 case OP_CRMINPLUS:
339 branchlength++;
340 /* Fall through */
341
342 case OP_CRSTAR:
343 case OP_CRMINSTAR:
344 case OP_CRQUERY:
345 case OP_CRMINQUERY:
346 cc++;
347 break;
348
349 case OP_CRRANGE:
350 case OP_CRMINRANGE:
351 branchlength += GET2(cc,1);
352 cc += 1 + 2 * IMM2_SIZE;
353 break;
354
355 default:
356 branchlength++;
357 break;
358 }
359 break;
360
361 /* Backreferences and subroutine calls are treated in the same way: we find
362 the minimum length for the subpattern. A recursion, however, causes an
363 a flag to be set that causes the length of this branch to be ignored. The
364 logic is that a recursion can only make sense if there is another
365 alternation that stops the recursing. That will provide the minimum length
366 (when no recursion happens). A backreference within the group that it is
367 referencing behaves in the same way.
368
369 If PCRE_JAVASCRIPT_COMPAT is set, a backreference to an unset bracket
370 matches an empty string (by default it causes a matching failure), so in
371 that case we must set the minimum length to zero. */
372
373 case OP_REF:
374 case OP_REFI:
375 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0)
376 {
377 ce = cs = (pcre_uchar *)PRIV(find_bracket)(startcode, utf8, GET2(cc, 1));
378 if (cs == NULL) return -2;
379 do ce += GET(ce, 1); while (*ce == OP_ALT);
380 if (cc > cs && cc < ce)
381 {
382 d = 0;
383 had_recurse = TRUE;
384 }
385 else
386 {
387 d = find_minlength(cs, startcode, options, recurse_depth);
388 }
389 }
390 else d = 0;
391 cc += 1 + IMM2_SIZE;
392
393 /* Handle repeated back references */
394
395 switch (*cc)
396 {
397 case OP_CRSTAR:
398 case OP_CRMINSTAR:
399 case OP_CRQUERY:
400 case OP_CRMINQUERY:
401 min = 0;
402 cc++;
403 break;
404
405 case OP_CRPLUS:
406 case OP_CRMINPLUS:
407 min = 1;
408 cc++;
409 break;
410
411 case OP_CRRANGE:
412 case OP_CRMINRANGE:
413 min = GET2(cc, 1);
414 cc += 1 + 2 * IMM2_SIZE;
415 break;
416
417 default:
418 min = 1;
419 break;
420 }
421
422 branchlength += min * d;
423 break;
424
425 /* We can easily detect direct recursion, but not mutual recursion. This is
426 caught by a recursion depth count. */
427
428 case OP_RECURSE:
429 cs = ce = (pcre_uchar *)startcode + GET(cc, 1);
430 do ce += GET(ce, 1); while (*ce == OP_ALT);
431 if ((cc > cs && cc < ce) || recurse_depth > 10)
432 had_recurse = TRUE;
433 else
434 {
435 branchlength += find_minlength(cs, startcode, options, recurse_depth + 1);
436 }
437 cc += 1 + LINK_SIZE;
438 break;
439
440 /* Anything else does not or need not match a character. We can get the
441 item's length from the table, but for those that can match zero occurrences
442 of a character, we must take special action for UTF-8 characters. As it
443 happens, the "NOT" versions of these opcodes are used at present only for
444 ASCII characters, so they could be omitted from this list. However, in
445 future that may change, so we include them here so as not to leave a
446 gotcha for a future maintainer. */
447
448 case OP_UPTO:
449 case OP_UPTOI:
450 case OP_NOTUPTO:
451 case OP_NOTUPTOI:
452 case OP_MINUPTO:
453 case OP_MINUPTOI:
454 case OP_NOTMINUPTO:
455 case OP_NOTMINUPTOI:
456 case OP_POSUPTO:
457 case OP_POSUPTOI:
458 case OP_NOTPOSUPTO:
459 case OP_NOTPOSUPTOI:
460
461 case OP_STAR:
462 case OP_STARI:
463 case OP_NOTSTAR:
464 case OP_NOTSTARI:
465 case OP_MINSTAR:
466 case OP_MINSTARI:
467 case OP_NOTMINSTAR:
468 case OP_NOTMINSTARI:
469 case OP_POSSTAR:
470 case OP_POSSTARI:
471 case OP_NOTPOSSTAR:
472 case OP_NOTPOSSTARI:
473
474 case OP_QUERY:
475 case OP_QUERYI:
476 case OP_NOTQUERY:
477 case OP_NOTQUERYI:
478 case OP_MINQUERY:
479 case OP_MINQUERYI:
480 case OP_NOTMINQUERY:
481 case OP_NOTMINQUERYI:
482 case OP_POSQUERY:
483 case OP_POSQUERYI:
484 case OP_NOTPOSQUERY:
485 case OP_NOTPOSQUERYI:
486
487 cc += PRIV(OP_lengths)[op];
488 #ifdef SUPPORT_UTF8
489 if (utf8 && cc[-1] >= 0xc0) cc += PRIV(utf8_table4)[cc[-1] & 0x3f];
490 #endif
491 break;
492
493 /* Skip these, but we need to add in the name length. */
494
495 case OP_MARK:
496 case OP_PRUNE_ARG:
497 case OP_SKIP_ARG:
498 case OP_THEN_ARG:
499 cc += PRIV(OP_lengths)[op] + cc[1];
500 break;
501
502 /* The remaining opcodes are just skipped over. */
503
504 case OP_CLOSE:
505 case OP_COMMIT:
506 case OP_FAIL:
507 case OP_PRUNE:
508 case OP_SET_SOM:
509 case OP_SKIP:
510 case OP_THEN:
511 cc += PRIV(OP_lengths)[op];
512 break;
513
514 /* This should not occur: we list all opcodes explicitly so that when
515 new ones get added they are properly considered. */
516
517 default:
518 return -3;
519 }
520 }
521 /* Control never gets here */
522 }
523
524
525
526 /*************************************************
527 * Set a bit and maybe its alternate case *
528 *************************************************/
529
530 /* Given a character, set its first byte's bit in the table, and also the
531 corresponding bit for the other version of a letter if we are caseless. In
532 UTF-8 mode, for characters greater than 127, we can only do the caseless thing
533 when Unicode property support is available.
534
535 Arguments:
536 start_bits points to the bit map
537 p points to the character
538 caseless the caseless flag
539 cd the block with char table pointers
540 utf8 TRUE for UTF-8 mode
541
542 Returns: pointer after the character
543 */
544
545 static const pcre_uchar *
546 set_table_bit(pcre_uint8 *start_bits, const pcre_uchar *p, BOOL caseless,
547 compile_data *cd, BOOL utf8)
548 {
549 unsigned int c = *p;
550
551 SET_BIT(c);
552
553 #ifdef SUPPORT_UTF8
554 if (utf8 && c > 127)
555 {
556 GETCHARINC(c, p);
557 #ifdef SUPPORT_UCP
558 if (caseless)
559 {
560 pcre_uint8 buff[8];
561 c = UCD_OTHERCASE(c);
562 (void)PRIV(ord2utf8)(c, buff);
563 SET_BIT(buff[0]);
564 }
565 #endif
566 return p;
567 }
568 #endif
569
570 /* Not UTF-8 mode, or character is less than 127. */
571
572 if (caseless && (cd->ctypes[c] & ctype_letter) != 0) SET_BIT(cd->fcc[c]);
573 return p + 1;
574 }
575
576
577
578 /*************************************************
579 * Set bits for a positive character type *
580 *************************************************/
581
582 /* This function sets starting bits for a character type. In UTF-8 mode, we can
583 only do a direct setting for bytes less than 128, as otherwise there can be
584 confusion with bytes in the middle of UTF-8 characters. In a "traditional"
585 environment, the tables will only recognize ASCII characters anyway, but in at
586 least one Windows environment, some higher bytes bits were set in the tables.
587 So we deal with that case by considering the UTF-8 encoding.
588
589 Arguments:
590 start_bits the starting bitmap
591 cbit type the type of character wanted
592 table_limit 32 for non-UTF-8; 16 for UTF-8
593 cd the block with char table pointers
594
595 Returns: nothing
596 */
597
598 static void
599 set_type_bits(pcre_uint8 *start_bits, int cbit_type, int table_limit,
600 compile_data *cd)
601 {
602 register int c;
603 for (c = 0; c < table_limit; c++) start_bits[c] |= cd->cbits[c+cbit_type];
604 #ifdef SUPPORT_UTF8
605 if (table_limit == 32) return;
606 for (c = 128; c < 256; c++)
607 {
608 if ((cd->cbits[c/8] & (1 << (c&7))) != 0)
609 {
610 pcre_uint8 buff[8];
611 (void)PRIV(ord2utf8)(c, buff);
612 SET_BIT(buff[0]);
613 }
614 }
615 #endif
616 }
617
618
619 /*************************************************
620 * Set bits for a negative character type *
621 *************************************************/
622
623 /* This function sets starting bits for a negative character type such as \D.
624 In UTF-8 mode, we can only do a direct setting for bytes less than 128, as
625 otherwise there can be confusion with bytes in the middle of UTF-8 characters.
626 Unlike in the positive case, where we can set appropriate starting bits for
627 specific high-valued UTF-8 characters, in this case we have to set the bits for
628 all high-valued characters. The lowest is 0xc2, but we overkill by starting at
629 0xc0 (192) for simplicity.
630
631 Arguments:
632 start_bits the starting bitmap
633 cbit type the type of character wanted
634 table_limit 32 for non-UTF-8; 16 for UTF-8
635 cd the block with char table pointers
636
637 Returns: nothing
638 */
639
640 static void
641 set_nottype_bits(pcre_uint8 *start_bits, int cbit_type, int table_limit,
642 compile_data *cd)
643 {
644 register int c;
645 for (c = 0; c < table_limit; c++) start_bits[c] |= ~cd->cbits[c+cbit_type];
646 if (table_limit != 32) for (c = 24; c < 32; c++) start_bits[c] = 0xff;
647 }
648
649
650
651 /*************************************************
652 * Create bitmap of starting bytes *
653 *************************************************/
654
655 /* This function scans a compiled unanchored expression recursively and
656 attempts to build a bitmap of the set of possible starting bytes. As time goes
657 by, we may be able to get more clever at doing this. The SSB_CONTINUE return is
658 useful for parenthesized groups in patterns such as (a*)b where the group
659 provides some optional starting bytes but scanning must continue at the outer
660 level to find at least one mandatory byte. At the outermost level, this
661 function fails unless the result is SSB_DONE.
662
663 Arguments:
664 code points to an expression
665 start_bits points to a 32-byte table, initialized to 0
666 utf8 TRUE if in UTF-8 mode
667 cd the block with char table pointers
668
669 Returns: SSB_FAIL => Failed to find any starting bytes
670 SSB_DONE => Found mandatory starting bytes
671 SSB_CONTINUE => Found optional starting bytes
672 SSB_UNKNOWN => Hit an unrecognized opcode
673 */
674
675 static int
676 set_start_bits(const pcre_uchar *code, pcre_uint8 *start_bits, BOOL utf8,
677 compile_data *cd)
678 {
679 register int c;
680 int yield = SSB_DONE;
681 int table_limit = utf8? 16:32;
682
683 #if 0
684 /* ========================================================================= */
685 /* The following comment and code was inserted in January 1999. In May 2006,
686 when it was observed to cause compiler warnings about unused values, I took it
687 out again. If anybody is still using OS/2, they will have to put it back
688 manually. */
689
690 /* This next statement and the later reference to dummy are here in order to
691 trick the optimizer of the IBM C compiler for OS/2 into generating correct
692 code. Apparently IBM isn't going to fix the problem, and we would rather not
693 disable optimization (in this module it actually makes a big difference, and
694 the pcre module can use all the optimization it can get). */
695
696 volatile int dummy;
697 /* ========================================================================= */
698 #endif
699
700 do
701 {
702 BOOL try_next = TRUE;
703 const pcre_uchar *tcode = code + 1 + LINK_SIZE;
704
705 if (*code == OP_CBRA || *code == OP_SCBRA ||
706 *code == OP_CBRAPOS || *code == OP_SCBRAPOS) tcode += IMM2_SIZE;
707
708 while (try_next) /* Loop for items in this branch */
709 {
710 int rc;
711
712 switch(*tcode)
713 {
714 /* If we reach something we don't understand, it means a new opcode has
715 been created that hasn't been added to this code. Hopefully this problem
716 will be discovered during testing. */
717
718 default:
719 return SSB_UNKNOWN;
720
721 /* Fail for a valid opcode that implies no starting bits. */
722
723 case OP_ACCEPT:
724 case OP_ASSERT_ACCEPT:
725 case OP_ALLANY:
726 case OP_ANY:
727 case OP_ANYBYTE:
728 case OP_CIRC:
729 case OP_CIRCM:
730 case OP_CLOSE:
731 case OP_COMMIT:
732 case OP_COND:
733 case OP_CREF:
734 case OP_DEF:
735 case OP_DOLL:
736 case OP_DOLLM:
737 case OP_END:
738 case OP_EOD:
739 case OP_EODN:
740 case OP_EXTUNI:
741 case OP_FAIL:
742 case OP_MARK:
743 case OP_NCREF:
744 case OP_NOT:
745 case OP_NOTEXACT:
746 case OP_NOTEXACTI:
747 case OP_NOTI:
748 case OP_NOTMINPLUS:
749 case OP_NOTMINPLUSI:
750 case OP_NOTMINQUERY:
751 case OP_NOTMINQUERYI:
752 case OP_NOTMINSTAR:
753 case OP_NOTMINSTARI:
754 case OP_NOTMINUPTO:
755 case OP_NOTMINUPTOI:
756 case OP_NOTPLUS:
757 case OP_NOTPLUSI:
758 case OP_NOTPOSPLUS:
759 case OP_NOTPOSPLUSI:
760 case OP_NOTPOSQUERY:
761 case OP_NOTPOSQUERYI:
762 case OP_NOTPOSSTAR:
763 case OP_NOTPOSSTARI:
764 case OP_NOTPOSUPTO:
765 case OP_NOTPOSUPTOI:
766 case OP_NOTPROP:
767 case OP_NOTQUERY:
768 case OP_NOTQUERYI:
769 case OP_NOTSTAR:
770 case OP_NOTSTARI:
771 case OP_NOTUPTO:
772 case OP_NOTUPTOI:
773 case OP_NOT_HSPACE:
774 case OP_NOT_VSPACE:
775 case OP_NRREF:
776 case OP_PROP:
777 case OP_PRUNE:
778 case OP_PRUNE_ARG:
779 case OP_RECURSE:
780 case OP_REF:
781 case OP_REFI:
782 case OP_REVERSE:
783 case OP_RREF:
784 case OP_SCOND:
785 case OP_SET_SOM:
786 case OP_SKIP:
787 case OP_SKIP_ARG:
788 case OP_SOD:
789 case OP_SOM:
790 case OP_THEN:
791 case OP_THEN_ARG:
792 case OP_XCLASS:
793 return SSB_FAIL;
794
795 /* We can ignore word boundary tests. */
796
797 case OP_WORD_BOUNDARY:
798 case OP_NOT_WORD_BOUNDARY:
799 tcode++;
800 break;
801
802 /* If we hit a bracket or a positive lookahead assertion, recurse to set
803 bits from within the subpattern. If it can't find anything, we have to
804 give up. If it finds some mandatory character(s), we are done for this
805 branch. Otherwise, carry on scanning after the subpattern. */
806
807 case OP_BRA:
808 case OP_SBRA:
809 case OP_CBRA:
810 case OP_SCBRA:
811 case OP_BRAPOS:
812 case OP_SBRAPOS:
813 case OP_CBRAPOS:
814 case OP_SCBRAPOS:
815 case OP_ONCE:
816 case OP_ONCE_NC:
817 case OP_ASSERT:
818 rc = set_start_bits(tcode, start_bits, utf8, cd);
819 if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
820 if (rc == SSB_DONE) try_next = FALSE; else
821 {
822 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
823 tcode += 1 + LINK_SIZE;
824 }
825 break;
826
827 /* If we hit ALT or KET, it means we haven't found anything mandatory in
828 this branch, though we might have found something optional. For ALT, we
829 continue with the next alternative, but we have to arrange that the final
830 result from subpattern is SSB_CONTINUE rather than SSB_DONE. For KET,
831 return SSB_CONTINUE: if this is the top level, that indicates failure,
832 but after a nested subpattern, it causes scanning to continue. */
833
834 case OP_ALT:
835 yield = SSB_CONTINUE;
836 try_next = FALSE;
837 break;
838
839 case OP_KET:
840 case OP_KETRMAX:
841 case OP_KETRMIN:
842 case OP_KETRPOS:
843 return SSB_CONTINUE;
844
845 /* Skip over callout */
846
847 case OP_CALLOUT:
848 tcode += 2 + 2*LINK_SIZE;
849 break;
850
851 /* Skip over lookbehind and negative lookahead assertions */
852
853 case OP_ASSERT_NOT:
854 case OP_ASSERTBACK:
855 case OP_ASSERTBACK_NOT:
856 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
857 tcode += 1 + LINK_SIZE;
858 break;
859
860 /* BRAZERO does the bracket, but carries on. */
861
862 case OP_BRAZERO:
863 case OP_BRAMINZERO:
864 case OP_BRAPOSZERO:
865 rc = set_start_bits(++tcode, start_bits, utf8, cd);
866 if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
867 /* =========================================================================
868 See the comment at the head of this function concerning the next line,
869 which was an old fudge for the benefit of OS/2.
870 dummy = 1;
871 ========================================================================= */
872 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
873 tcode += 1 + LINK_SIZE;
874 break;
875
876 /* SKIPZERO skips the bracket. */
877
878 case OP_SKIPZERO:
879 tcode++;
880 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
881 tcode += 1 + LINK_SIZE;
882 break;
883
884 /* Single-char * or ? sets the bit and tries the next item */
885
886 case OP_STAR:
887 case OP_MINSTAR:
888 case OP_POSSTAR:
889 case OP_QUERY:
890 case OP_MINQUERY:
891 case OP_POSQUERY:
892 tcode = set_table_bit(start_bits, tcode + 1, FALSE, cd, utf8);
893 break;
894
895 case OP_STARI:
896 case OP_MINSTARI:
897 case OP_POSSTARI:
898 case OP_QUERYI:
899 case OP_MINQUERYI:
900 case OP_POSQUERYI:
901 tcode = set_table_bit(start_bits, tcode + 1, TRUE, cd, utf8);
902 break;
903
904 /* Single-char upto sets the bit and tries the next */
905
906 case OP_UPTO:
907 case OP_MINUPTO:
908 case OP_POSUPTO:
909 tcode = set_table_bit(start_bits, tcode + 1 + IMM2_SIZE, FALSE, cd, utf8);
910 break;
911
912 case OP_UPTOI:
913 case OP_MINUPTOI:
914 case OP_POSUPTOI:
915 tcode = set_table_bit(start_bits, tcode + 1 + IMM2_SIZE, TRUE, cd, utf8);
916 break;
917
918 /* At least one single char sets the bit and stops */
919
920 case OP_EXACT:
921 tcode += IMM2_SIZE;
922 /* Fall through */
923 case OP_CHAR:
924 case OP_PLUS:
925 case OP_MINPLUS:
926 case OP_POSPLUS:
927 (void)set_table_bit(start_bits, tcode + 1, FALSE, cd, utf8);
928 try_next = FALSE;
929 break;
930
931 case OP_EXACTI:
932 tcode += IMM2_SIZE;
933 /* Fall through */
934 case OP_CHARI:
935 case OP_PLUSI:
936 case OP_MINPLUSI:
937 case OP_POSPLUSI:
938 (void)set_table_bit(start_bits, tcode + 1, TRUE, cd, utf8);
939 try_next = FALSE;
940 break;
941
942 /* Special spacing and line-terminating items. These recognize specific
943 lists of characters. The difference between VSPACE and ANYNL is that the
944 latter can match the two-character CRLF sequence, but that is not
945 relevant for finding the first character, so their code here is
946 identical. */
947
948 case OP_HSPACE:
949 SET_BIT(0x09);
950 SET_BIT(0x20);
951 if (utf8)
952 {
953 SET_BIT(0xC2); /* For U+00A0 */
954 SET_BIT(0xE1); /* For U+1680, U+180E */
955 SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
956 SET_BIT(0xE3); /* For U+3000 */
957 }
958 else SET_BIT(0xA0);
959 try_next = FALSE;
960 break;
961
962 case OP_ANYNL:
963 case OP_VSPACE:
964 SET_BIT(0x0A);
965 SET_BIT(0x0B);
966 SET_BIT(0x0C);
967 SET_BIT(0x0D);
968 if (utf8)
969 {
970 SET_BIT(0xC2); /* For U+0085 */
971 SET_BIT(0xE2); /* For U+2028, U+2029 */
972 }
973 else SET_BIT(0x85);
974 try_next = FALSE;
975 break;
976
977 /* Single character types set the bits and stop. Note that if PCRE_UCP
978 is set, we do not see these op codes because \d etc are converted to
979 properties. Therefore, these apply in the case when only characters less
980 than 256 are recognized to match the types. */
981
982 case OP_NOT_DIGIT:
983 set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
984 try_next = FALSE;
985 break;
986
987 case OP_DIGIT:
988 set_type_bits(start_bits, cbit_digit, table_limit, cd);
989 try_next = FALSE;
990 break;
991
992 /* The cbit_space table has vertical tab as whitespace; we have to
993 ensure it is set as not whitespace. */
994
995 case OP_NOT_WHITESPACE:
996 set_nottype_bits(start_bits, cbit_space, table_limit, cd);
997 start_bits[1] |= 0x08;
998 try_next = FALSE;
999 break;
1000
1001 /* The cbit_space table has vertical tab as whitespace; we have to
1002 not set it from the table. */
1003
1004 case OP_WHITESPACE:
1005 c = start_bits[1]; /* Save in case it was already set */
1006 set_type_bits(start_bits, cbit_space, table_limit, cd);
1007 start_bits[1] = (start_bits[1] & ~0x08) | c;
1008 try_next = FALSE;
1009 break;
1010
1011 case OP_NOT_WORDCHAR:
1012 set_nottype_bits(start_bits, cbit_word, table_limit, cd);
1013 try_next = FALSE;
1014 break;
1015
1016 case OP_WORDCHAR:
1017 set_type_bits(start_bits, cbit_word, table_limit, cd);
1018 try_next = FALSE;
1019 break;
1020
1021 /* One or more character type fudges the pointer and restarts, knowing
1022 it will hit a single character type and stop there. */
1023
1024 case OP_TYPEPLUS:
1025 case OP_TYPEMINPLUS:
1026 case OP_TYPEPOSPLUS:
1027 tcode++;
1028 break;
1029
1030 case OP_TYPEEXACT:
1031 tcode += 1 + IMM2_SIZE;
1032 break;
1033
1034 /* Zero or more repeats of character types set the bits and then
1035 try again. */
1036
1037 case OP_TYPEUPTO:
1038 case OP_TYPEMINUPTO:
1039 case OP_TYPEPOSUPTO:
1040 tcode += IMM2_SIZE; /* Fall through */
1041
1042 case OP_TYPESTAR:
1043 case OP_TYPEMINSTAR:
1044 case OP_TYPEPOSSTAR:
1045 case OP_TYPEQUERY:
1046 case OP_TYPEMINQUERY:
1047 case OP_TYPEPOSQUERY:
1048 switch(tcode[1])
1049 {
1050 default:
1051 case OP_ANY:
1052 case OP_ALLANY:
1053 return SSB_FAIL;
1054
1055 case OP_HSPACE:
1056 SET_BIT(0x09);
1057 SET_BIT(0x20);
1058 if (utf8)
1059 {
1060 SET_BIT(0xC2); /* For U+00A0 */
1061 SET_BIT(0xE1); /* For U+1680, U+180E */
1062 SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
1063 SET_BIT(0xE3); /* For U+3000 */
1064 }
1065 else SET_BIT(0xA0);
1066 break;
1067
1068 case OP_ANYNL:
1069 case OP_VSPACE:
1070 SET_BIT(0x0A);
1071 SET_BIT(0x0B);
1072 SET_BIT(0x0C);
1073 SET_BIT(0x0D);
1074 if (utf8)
1075 {
1076 SET_BIT(0xC2); /* For U+0085 */
1077 SET_BIT(0xE2); /* For U+2028, U+2029 */
1078 }
1079 else SET_BIT(0x85);
1080 break;
1081
1082 case OP_NOT_DIGIT:
1083 set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
1084 break;
1085
1086 case OP_DIGIT:
1087 set_type_bits(start_bits, cbit_digit, table_limit, cd);
1088 break;
1089
1090 /* The cbit_space table has vertical tab as whitespace; we have to
1091 ensure it gets set as not whitespace. */
1092
1093 case OP_NOT_WHITESPACE:
1094 set_nottype_bits(start_bits, cbit_space, table_limit, cd);
1095 start_bits[1] |= 0x08;
1096 break;
1097
1098 /* The cbit_space table has vertical tab as whitespace; we have to
1099 avoid setting it. */
1100
1101 case OP_WHITESPACE:
1102 c = start_bits[1]; /* Save in case it was already set */
1103 set_type_bits(start_bits, cbit_space, table_limit, cd);
1104 start_bits[1] = (start_bits[1] & ~0x08) | c;
1105 break;
1106
1107 case OP_NOT_WORDCHAR:
1108 set_nottype_bits(start_bits, cbit_word, table_limit, cd);
1109 break;
1110
1111 case OP_WORDCHAR:
1112 set_type_bits(start_bits, cbit_word, table_limit, cd);
1113 break;
1114 }
1115
1116 tcode += 2;
1117 break;
1118
1119 /* Character class where all the information is in a bit map: set the
1120 bits and either carry on or not, according to the repeat count. If it was
1121 a negative class, and we are operating with UTF-8 characters, any byte
1122 with a value >= 0xc4 is a potentially valid starter because it starts a
1123 character with a value > 255. */
1124
1125 case OP_NCLASS:
1126 #ifdef SUPPORT_UTF8
1127 if (utf8)
1128 {
1129 start_bits[24] |= 0xf0; /* Bits for 0xc4 - 0xc8 */
1130 memset(start_bits+25, 0xff, 7); /* Bits for 0xc9 - 0xff */
1131 }
1132 #endif
1133 /* Fall through */
1134
1135 case OP_CLASS:
1136 {
1137 tcode++;
1138
1139 /* In UTF-8 mode, the bits in a bit map correspond to character
1140 values, not to byte values. However, the bit map we are constructing is
1141 for byte values. So we have to do a conversion for characters whose
1142 value is > 127. In fact, there are only two possible starting bytes for
1143 characters in the range 128 - 255. */
1144
1145 #ifdef SUPPORT_UTF8
1146 if (utf8)
1147 {
1148 for (c = 0; c < 16; c++) start_bits[c] |= tcode[c];
1149 for (c = 128; c < 256; c++)
1150 {
1151 if ((tcode[c/8] && (1 << (c&7))) != 0)
1152 {
1153 int d = (c >> 6) | 0xc0; /* Set bit for this starter */
1154 start_bits[d/8] |= (1 << (d&7)); /* and then skip on to the */
1155 c = (c & 0xc0) + 0x40 - 1; /* next relevant character. */
1156 }
1157 }
1158 }
1159
1160 /* In non-UTF-8 mode, the two bit maps are completely compatible. */
1161
1162 else
1163 #endif
1164 {
1165 for (c = 0; c < 32; c++) start_bits[c] |= tcode[c];
1166 }
1167
1168 /* Advance past the bit map, and act on what follows. For a zero
1169 minimum repeat, continue; otherwise stop processing. */
1170
1171 tcode += 32;
1172 switch (*tcode)
1173 {
1174 case OP_CRSTAR:
1175 case OP_CRMINSTAR:
1176 case OP_CRQUERY:
1177 case OP_CRMINQUERY:
1178 tcode++;
1179 break;
1180
1181 case OP_CRRANGE:
1182 case OP_CRMINRANGE:
1183 if (GET2(tcode, 1) == 0) tcode += 1 + 2 * IMM2_SIZE;
1184 else try_next = FALSE;
1185 break;
1186
1187 default:
1188 try_next = FALSE;
1189 break;
1190 }
1191 }
1192 break; /* End of bitmap class handling */
1193
1194 } /* End of switch */
1195 } /* End of try_next loop */
1196
1197 code += GET(code, 1); /* Advance to next branch */
1198 }
1199 while (*code == OP_ALT);
1200 return yield;
1201 }
1202
1203
1204
1205
1206
1207 /*************************************************
1208 * Study a compiled expression *
1209 *************************************************/
1210
1211 /* This function is handed a compiled expression that it must study to produce
1212 information that will speed up the matching. It returns a pcre_extra block
1213 which then gets handed back to pcre_exec().
1214
1215 Arguments:
1216 re points to the compiled expression
1217 options contains option bits
1218 errorptr points to where to place error messages;
1219 set NULL unless error
1220
1221 Returns: pointer to a pcre_extra block, with study_data filled in and the
1222 appropriate flags set;
1223 NULL on error or if no optimization possible
1224 */
1225
1226 #ifdef COMPILE_PCRE8
1227 PCRE_EXP_DEFN pcre_extra * PCRE_CALL_CONVENTION
1228 pcre_study(const pcre *external_re, int options, const char **errorptr)
1229 #else
1230 PCRE_EXP_DEFN pcre_extra * PCRE_CALL_CONVENTION
1231 pcre16_study(const pcre *external_re, int options, const char **errorptr)
1232 #endif
1233 {
1234 int min;
1235 BOOL bits_set = FALSE;
1236 pcre_uint8 start_bits[32];
1237 pcre_extra *extra = NULL;
1238 pcre_study_data *study;
1239 const pcre_uint8 *tables;
1240 pcre_uchar *code;
1241 compile_data compile_block;
1242 const real_pcre *re = (const real_pcre *)external_re;
1243
1244 *errorptr = NULL;
1245
1246 if (re == NULL || re->magic_number != MAGIC_NUMBER)
1247 {
1248 *errorptr = "argument is not a compiled regular expression";
1249 return NULL;
1250 }
1251
1252 if ((options & ~PUBLIC_STUDY_OPTIONS) != 0)
1253 {
1254 *errorptr = "unknown or incorrect option bit(s) set";
1255 return NULL;
1256 }
1257
1258 code = (pcre_uchar *)re + re->name_table_offset +
1259 (re->name_count * re->name_entry_size);
1260
1261 /* For an anchored pattern, or an unanchored pattern that has a first char, or
1262 a multiline pattern that matches only at "line starts", there is no point in
1263 seeking a list of starting bytes. */
1264
1265 if ((re->options & PCRE_ANCHORED) == 0 &&
1266 (re->flags & (PCRE_FIRSTSET|PCRE_STARTLINE)) == 0)
1267 {
1268 int rc;
1269
1270 /* Set the character tables in the block that is passed around */
1271
1272 tables = re->tables;
1273 if (tables == NULL)
1274 (void)pcre_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES,
1275 (void *)(&tables));
1276
1277 compile_block.lcc = tables + lcc_offset;
1278 compile_block.fcc = tables + fcc_offset;
1279 compile_block.cbits = tables + cbits_offset;
1280 compile_block.ctypes = tables + ctypes_offset;
1281
1282 /* See if we can find a fixed set of initial characters for the pattern. */
1283
1284 memset(start_bits, 0, 32 * sizeof(pcre_uint8));
1285 rc = set_start_bits(code, start_bits, (re->options & PCRE_UTF8) != 0,
1286 &compile_block);
1287 bits_set = rc == SSB_DONE;
1288 if (rc == SSB_UNKNOWN)
1289 {
1290 *errorptr = "internal error: opcode not recognized";
1291 return NULL;
1292 }
1293 }
1294
1295 /* Find the minimum length of subject string. */
1296
1297 switch(min = find_minlength(code, code, re->options, 0))
1298 {
1299 case -2: *errorptr = "internal error: missing capturing bracket"; return NULL;
1300 case -3: *errorptr = "internal error: opcode not recognized"; return NULL;
1301 default: break;
1302 }
1303
1304 /* If a set of starting bytes has been identified, or if the minimum length is
1305 greater than zero, or if JIT optimization has been requested, get a pcre_extra
1306 block and a pcre_study_data block. The study data is put in the latter, which
1307 is pointed to by the former, which may also get additional data set later by
1308 the calling program. At the moment, the size of pcre_study_data is fixed. We
1309 nevertheless save it in a field for returning via the pcre_fullinfo() function
1310 so that if it becomes variable in the future, we don't have to change that
1311 code. */
1312
1313 if (bits_set || min > 0
1314 #ifdef SUPPORT_JIT
1315 || (options & PCRE_STUDY_JIT_COMPILE) != 0
1316 #endif
1317 )
1318 {
1319 extra = (pcre_extra *)(pcre_malloc)
1320 (sizeof(pcre_extra) + sizeof(pcre_study_data));
1321 if (extra == NULL)
1322 {
1323 *errorptr = "failed to get memory";
1324 return NULL;
1325 }
1326
1327 study = (pcre_study_data *)((char *)extra + sizeof(pcre_extra));
1328 extra->flags = PCRE_EXTRA_STUDY_DATA;
1329 extra->study_data = study;
1330
1331 study->size = sizeof(pcre_study_data);
1332 study->flags = 0;
1333
1334 if (bits_set)
1335 {
1336 study->flags |= PCRE_STUDY_MAPPED;
1337 memcpy(study->start_bits, start_bits, sizeof(start_bits));
1338 }
1339
1340 /* Always set the minlength value in the block, because the JIT compiler
1341 makes use of it. However, don't set the bit unless the length is greater than
1342 zero - the interpretive pcre_exec() and pcre_dfa_exec() needn't waste time
1343 checking the zero case. */
1344
1345 if (min > 0)
1346 {
1347 study->flags |= PCRE_STUDY_MINLEN;
1348 study->minlength = min;
1349 }
1350 else study->minlength = 0;
1351
1352 /* If JIT support was compiled and requested, attempt the JIT compilation.
1353 If no starting bytes were found, and the minimum length is zero, and JIT
1354 compilation fails, abandon the extra block and return NULL. */
1355
1356 #ifdef SUPPORT_JIT
1357 extra->executable_jit = NULL;
1358 if ((options & PCRE_STUDY_JIT_COMPILE) != 0) PRIV(jit_compile)(re, extra);
1359 if (study->flags == 0 && (extra->flags & PCRE_EXTRA_EXECUTABLE_JIT) == 0)
1360 {
1361 pcre_free_study(extra);
1362 extra = NULL;
1363 }
1364 #endif
1365 }
1366
1367 return extra;
1368 }
1369
1370
1371 /*************************************************
1372 * Free the study data *
1373 *************************************************/
1374
1375 /* This function frees the memory that was obtained by pcre_study().
1376
1377 Argument: a pointer to the pcre_extra block
1378 Returns: nothing
1379 */
1380
1381 #ifdef COMPILE_PCRE8
1382 PCRE_EXP_DEFN void
1383 pcre_free_study(pcre_extra *extra)
1384 #else
1385 PCRE_EXP_DEFN void
1386 pcre16_free_study(pcre_extra *extra)
1387 #endif
1388 {
1389 #ifdef SUPPORT_JIT
1390 if ((extra->flags & PCRE_EXTRA_EXECUTABLE_JIT) != 0 &&
1391 extra->executable_jit != NULL)
1392 PRIV(jit_free)(extra->executable_jit);
1393 #endif
1394 pcre_free(extra);
1395 }
1396
1397 /* 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