/[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 767 - (show annotations) (download)
Sat Nov 26 12:48:56 2011 UTC (2 years, 10 months ago) by zherczeg
File MIME type: text/plain
File size: 39322 byte(s)
Make simple patterns work in PCRE16
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 += 2;
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 += 4;
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 += (cc[3] == OP_PROP || cc[3] == OP_NOTPROP)? 6 : 4;
255 break;
256
257 /* Handle single-char non-literal matchers */
258
259 case OP_PROP:
260 case OP_NOTPROP:
261 cc += 2;
262 /* Fall through */
263
264 case OP_NOT_DIGIT:
265 case OP_DIGIT:
266 case OP_NOT_WHITESPACE:
267 case OP_WHITESPACE:
268 case OP_NOT_WORDCHAR:
269 case OP_WORDCHAR:
270 case OP_ANY:
271 case OP_ALLANY:
272 case OP_EXTUNI:
273 case OP_HSPACE:
274 case OP_NOT_HSPACE:
275 case OP_VSPACE:
276 case OP_NOT_VSPACE:
277 branchlength++;
278 cc++;
279 break;
280
281 /* "Any newline" might match two characters, but it also might match just
282 one. */
283
284 case OP_ANYNL:
285 branchlength += 1;
286 cc++;
287 break;
288
289 /* The single-byte matcher means we can't proceed in UTF-8 mode. (In
290 non-UTF-8 mode \C will actually be turned into OP_ALLANY, so won't ever
291 appear, but leave the code, just in case.) */
292
293 case OP_ANYBYTE:
294 #ifdef SUPPORT_UTF8
295 if (utf8) return -1;
296 #endif
297 branchlength++;
298 cc++;
299 break;
300
301 /* For repeated character types, we have to test for \p and \P, which have
302 an extra two bytes of parameters. */
303
304 case OP_TYPESTAR:
305 case OP_TYPEMINSTAR:
306 case OP_TYPEQUERY:
307 case OP_TYPEMINQUERY:
308 case OP_TYPEPOSSTAR:
309 case OP_TYPEPOSQUERY:
310 if (cc[1] == OP_PROP || cc[1] == OP_NOTPROP) cc += 2;
311 cc += PRIV(OP_lengths)[op];
312 break;
313
314 case OP_TYPEUPTO:
315 case OP_TYPEMINUPTO:
316 case OP_TYPEPOSUPTO:
317 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
318 cc += PRIV(OP_lengths)[op];
319 break;
320
321 /* Check a class for variable quantification */
322
323 #ifdef SUPPORT_UTF8
324 case OP_XCLASS:
325 cc += GET(cc, 1) - 33;
326 /* Fall through */
327 #endif
328
329 case OP_CLASS:
330 case OP_NCLASS:
331 cc += 33;
332
333 switch (*cc)
334 {
335 case OP_CRPLUS:
336 case OP_CRMINPLUS:
337 branchlength++;
338 /* Fall through */
339
340 case OP_CRSTAR:
341 case OP_CRMINSTAR:
342 case OP_CRQUERY:
343 case OP_CRMINQUERY:
344 cc++;
345 break;
346
347 case OP_CRRANGE:
348 case OP_CRMINRANGE:
349 branchlength += GET2(cc,1);
350 cc += 5;
351 break;
352
353 default:
354 branchlength++;
355 break;
356 }
357 break;
358
359 /* Backreferences and subroutine calls are treated in the same way: we find
360 the minimum length for the subpattern. A recursion, however, causes an
361 a flag to be set that causes the length of this branch to be ignored. The
362 logic is that a recursion can only make sense if there is another
363 alternation that stops the recursing. That will provide the minimum length
364 (when no recursion happens). A backreference within the group that it is
365 referencing behaves in the same way.
366
367 If PCRE_JAVASCRIPT_COMPAT is set, a backreference to an unset bracket
368 matches an empty string (by default it causes a matching failure), so in
369 that case we must set the minimum length to zero. */
370
371 case OP_REF:
372 case OP_REFI:
373 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0)
374 {
375 ce = cs = (pcre_uchar *)PRIV(find_bracket)(startcode, utf8, GET2(cc, 1));
376 if (cs == NULL) return -2;
377 do ce += GET(ce, 1); while (*ce == OP_ALT);
378 if (cc > cs && cc < ce)
379 {
380 d = 0;
381 had_recurse = TRUE;
382 }
383 else
384 {
385 d = find_minlength(cs, startcode, options, recurse_depth);
386 }
387 }
388 else d = 0;
389 cc += 3;
390
391 /* Handle repeated back references */
392
393 switch (*cc)
394 {
395 case OP_CRSTAR:
396 case OP_CRMINSTAR:
397 case OP_CRQUERY:
398 case OP_CRMINQUERY:
399 min = 0;
400 cc++;
401 break;
402
403 case OP_CRPLUS:
404 case OP_CRMINPLUS:
405 min = 1;
406 cc++;
407 break;
408
409 case OP_CRRANGE:
410 case OP_CRMINRANGE:
411 min = GET2(cc, 1);
412 cc += 5;
413 break;
414
415 default:
416 min = 1;
417 break;
418 }
419
420 branchlength += min * d;
421 break;
422
423 /* We can easily detect direct recursion, but not mutual recursion. This is
424 caught by a recursion depth count. */
425
426 case OP_RECURSE:
427 cs = ce = (pcre_uchar *)startcode + GET(cc, 1);
428 do ce += GET(ce, 1); while (*ce == OP_ALT);
429 if ((cc > cs && cc < ce) || recurse_depth > 10)
430 had_recurse = TRUE;
431 else
432 {
433 branchlength += find_minlength(cs, startcode, options, recurse_depth + 1);
434 }
435 cc += 1 + LINK_SIZE;
436 break;
437
438 /* Anything else does not or need not match a character. We can get the
439 item's length from the table, but for those that can match zero occurrences
440 of a character, we must take special action for UTF-8 characters. As it
441 happens, the "NOT" versions of these opcodes are used at present only for
442 ASCII characters, so they could be omitted from this list. However, in
443 future that may change, so we include them here so as not to leave a
444 gotcha for a future maintainer. */
445
446 case OP_UPTO:
447 case OP_UPTOI:
448 case OP_NOTUPTO:
449 case OP_NOTUPTOI:
450 case OP_MINUPTO:
451 case OP_MINUPTOI:
452 case OP_NOTMINUPTO:
453 case OP_NOTMINUPTOI:
454 case OP_POSUPTO:
455 case OP_POSUPTOI:
456 case OP_NOTPOSUPTO:
457 case OP_NOTPOSUPTOI:
458
459 case OP_STAR:
460 case OP_STARI:
461 case OP_NOTSTAR:
462 case OP_NOTSTARI:
463 case OP_MINSTAR:
464 case OP_MINSTARI:
465 case OP_NOTMINSTAR:
466 case OP_NOTMINSTARI:
467 case OP_POSSTAR:
468 case OP_POSSTARI:
469 case OP_NOTPOSSTAR:
470 case OP_NOTPOSSTARI:
471
472 case OP_QUERY:
473 case OP_QUERYI:
474 case OP_NOTQUERY:
475 case OP_NOTQUERYI:
476 case OP_MINQUERY:
477 case OP_MINQUERYI:
478 case OP_NOTMINQUERY:
479 case OP_NOTMINQUERYI:
480 case OP_POSQUERY:
481 case OP_POSQUERYI:
482 case OP_NOTPOSQUERY:
483 case OP_NOTPOSQUERYI:
484
485 cc += PRIV(OP_lengths)[op];
486 #ifdef SUPPORT_UTF8
487 if (utf8 && cc[-1] >= 0xc0) cc += PRIV(utf8_table4)[cc[-1] & 0x3f];
488 #endif
489 break;
490
491 /* Skip these, but we need to add in the name length. */
492
493 case OP_MARK:
494 case OP_PRUNE_ARG:
495 case OP_SKIP_ARG:
496 case OP_THEN_ARG:
497 cc += PRIV(OP_lengths)[op] + cc[1];
498 break;
499
500 /* The remaining opcodes are just skipped over. */
501
502 case OP_CLOSE:
503 case OP_COMMIT:
504 case OP_FAIL:
505 case OP_PRUNE:
506 case OP_SET_SOM:
507 case OP_SKIP:
508 case OP_THEN:
509 cc += PRIV(OP_lengths)[op];
510 break;
511
512 /* This should not occur: we list all opcodes explicitly so that when
513 new ones get added they are properly considered. */
514
515 default:
516 return -3;
517 }
518 }
519 /* Control never gets here */
520 }
521
522
523
524 /*************************************************
525 * Set a bit and maybe its alternate case *
526 *************************************************/
527
528 /* Given a character, set its first byte's bit in the table, and also the
529 corresponding bit for the other version of a letter if we are caseless. In
530 UTF-8 mode, for characters greater than 127, we can only do the caseless thing
531 when Unicode property support is available.
532
533 Arguments:
534 start_bits points to the bit map
535 p points to the character
536 caseless the caseless flag
537 cd the block with char table pointers
538 utf8 TRUE for UTF-8 mode
539
540 Returns: pointer after the character
541 */
542
543 static const pcre_uchar *
544 set_table_bit(pcre_uint8 *start_bits, const pcre_uchar *p, BOOL caseless,
545 compile_data *cd, BOOL utf8)
546 {
547 unsigned int c = *p;
548
549 SET_BIT(c);
550
551 #ifdef SUPPORT_UTF8
552 if (utf8 && c > 127)
553 {
554 GETCHARINC(c, p);
555 #ifdef SUPPORT_UCP
556 if (caseless)
557 {
558 pcre_uint8 buff[8];
559 c = UCD_OTHERCASE(c);
560 (void)PRIV(ord2utf8)(c, buff);
561 SET_BIT(buff[0]);
562 }
563 #endif
564 return p;
565 }
566 #endif
567
568 /* Not UTF-8 mode, or character is less than 127. */
569
570 if (caseless && (cd->ctypes[c] & ctype_letter) != 0) SET_BIT(cd->fcc[c]);
571 return p + 1;
572 }
573
574
575
576 /*************************************************
577 * Set bits for a positive character type *
578 *************************************************/
579
580 /* This function sets starting bits for a character type. In UTF-8 mode, we can
581 only do a direct setting for bytes less than 128, as otherwise there can be
582 confusion with bytes in the middle of UTF-8 characters. In a "traditional"
583 environment, the tables will only recognize ASCII characters anyway, but in at
584 least one Windows environment, some higher bytes bits were set in the tables.
585 So we deal with that case by considering the UTF-8 encoding.
586
587 Arguments:
588 start_bits the starting bitmap
589 cbit type the type of character wanted
590 table_limit 32 for non-UTF-8; 16 for UTF-8
591 cd the block with char table pointers
592
593 Returns: nothing
594 */
595
596 static void
597 set_type_bits(pcre_uint8 *start_bits, int cbit_type, int table_limit,
598 compile_data *cd)
599 {
600 register int c;
601 for (c = 0; c < table_limit; c++) start_bits[c] |= cd->cbits[c+cbit_type];
602 #ifdef SUPPORT_UTF8
603 if (table_limit == 32) return;
604 for (c = 128; c < 256; c++)
605 {
606 if ((cd->cbits[c/8] & (1 << (c&7))) != 0)
607 {
608 pcre_uint8 buff[8];
609 (void)PRIV(ord2utf8)(c, buff);
610 SET_BIT(buff[0]);
611 }
612 }
613 #endif
614 }
615
616
617 /*************************************************
618 * Set bits for a negative character type *
619 *************************************************/
620
621 /* This function sets starting bits for a negative character type such as \D.
622 In UTF-8 mode, we can only do a direct setting for bytes less than 128, as
623 otherwise there can be confusion with bytes in the middle of UTF-8 characters.
624 Unlike in the positive case, where we can set appropriate starting bits for
625 specific high-valued UTF-8 characters, in this case we have to set the bits for
626 all high-valued characters. The lowest is 0xc2, but we overkill by starting at
627 0xc0 (192) for simplicity.
628
629 Arguments:
630 start_bits the starting bitmap
631 cbit type the type of character wanted
632 table_limit 32 for non-UTF-8; 16 for UTF-8
633 cd the block with char table pointers
634
635 Returns: nothing
636 */
637
638 static void
639 set_nottype_bits(pcre_uint8 *start_bits, int cbit_type, int table_limit,
640 compile_data *cd)
641 {
642 register int c;
643 for (c = 0; c < table_limit; c++) start_bits[c] |= ~cd->cbits[c+cbit_type];
644 if (table_limit != 32) for (c = 24; c < 32; c++) start_bits[c] = 0xff;
645 }
646
647
648
649 /*************************************************
650 * Create bitmap of starting bytes *
651 *************************************************/
652
653 /* This function scans a compiled unanchored expression recursively and
654 attempts to build a bitmap of the set of possible starting bytes. As time goes
655 by, we may be able to get more clever at doing this. The SSB_CONTINUE return is
656 useful for parenthesized groups in patterns such as (a*)b where the group
657 provides some optional starting bytes but scanning must continue at the outer
658 level to find at least one mandatory byte. At the outermost level, this
659 function fails unless the result is SSB_DONE.
660
661 Arguments:
662 code points to an expression
663 start_bits points to a 32-byte table, initialized to 0
664 utf8 TRUE if in UTF-8 mode
665 cd the block with char table pointers
666
667 Returns: SSB_FAIL => Failed to find any starting bytes
668 SSB_DONE => Found mandatory starting bytes
669 SSB_CONTINUE => Found optional starting bytes
670 SSB_UNKNOWN => Hit an unrecognized opcode
671 */
672
673 static int
674 set_start_bits(const pcre_uchar *code, pcre_uint8 *start_bits, BOOL utf8,
675 compile_data *cd)
676 {
677 register int c;
678 int yield = SSB_DONE;
679 int table_limit = utf8? 16:32;
680
681 #if 0
682 /* ========================================================================= */
683 /* The following comment and code was inserted in January 1999. In May 2006,
684 when it was observed to cause compiler warnings about unused values, I took it
685 out again. If anybody is still using OS/2, they will have to put it back
686 manually. */
687
688 /* This next statement and the later reference to dummy are here in order to
689 trick the optimizer of the IBM C compiler for OS/2 into generating correct
690 code. Apparently IBM isn't going to fix the problem, and we would rather not
691 disable optimization (in this module it actually makes a big difference, and
692 the pcre module can use all the optimization it can get). */
693
694 volatile int dummy;
695 /* ========================================================================= */
696 #endif
697
698 do
699 {
700 BOOL try_next = TRUE;
701 const pcre_uchar *tcode = code + 1 + LINK_SIZE;
702
703 if (*code == OP_CBRA || *code == OP_SCBRA ||
704 *code == OP_CBRAPOS || *code == OP_SCBRAPOS) tcode += 2;
705
706 while (try_next) /* Loop for items in this branch */
707 {
708 int rc;
709
710 switch(*tcode)
711 {
712 /* If we reach something we don't understand, it means a new opcode has
713 been created that hasn't been added to this code. Hopefully this problem
714 will be discovered during testing. */
715
716 default:
717 return SSB_UNKNOWN;
718
719 /* Fail for a valid opcode that implies no starting bits. */
720
721 case OP_ACCEPT:
722 case OP_ASSERT_ACCEPT:
723 case OP_ALLANY:
724 case OP_ANY:
725 case OP_ANYBYTE:
726 case OP_CIRC:
727 case OP_CIRCM:
728 case OP_CLOSE:
729 case OP_COMMIT:
730 case OP_COND:
731 case OP_CREF:
732 case OP_DEF:
733 case OP_DOLL:
734 case OP_DOLLM:
735 case OP_END:
736 case OP_EOD:
737 case OP_EODN:
738 case OP_EXTUNI:
739 case OP_FAIL:
740 case OP_MARK:
741 case OP_NCREF:
742 case OP_NOT:
743 case OP_NOTEXACT:
744 case OP_NOTEXACTI:
745 case OP_NOTI:
746 case OP_NOTMINPLUS:
747 case OP_NOTMINPLUSI:
748 case OP_NOTMINQUERY:
749 case OP_NOTMINQUERYI:
750 case OP_NOTMINSTAR:
751 case OP_NOTMINSTARI:
752 case OP_NOTMINUPTO:
753 case OP_NOTMINUPTOI:
754 case OP_NOTPLUS:
755 case OP_NOTPLUSI:
756 case OP_NOTPOSPLUS:
757 case OP_NOTPOSPLUSI:
758 case OP_NOTPOSQUERY:
759 case OP_NOTPOSQUERYI:
760 case OP_NOTPOSSTAR:
761 case OP_NOTPOSSTARI:
762 case OP_NOTPOSUPTO:
763 case OP_NOTPOSUPTOI:
764 case OP_NOTPROP:
765 case OP_NOTQUERY:
766 case OP_NOTQUERYI:
767 case OP_NOTSTAR:
768 case OP_NOTSTARI:
769 case OP_NOTUPTO:
770 case OP_NOTUPTOI:
771 case OP_NOT_HSPACE:
772 case OP_NOT_VSPACE:
773 case OP_NRREF:
774 case OP_PROP:
775 case OP_PRUNE:
776 case OP_PRUNE_ARG:
777 case OP_RECURSE:
778 case OP_REF:
779 case OP_REFI:
780 case OP_REVERSE:
781 case OP_RREF:
782 case OP_SCOND:
783 case OP_SET_SOM:
784 case OP_SKIP:
785 case OP_SKIP_ARG:
786 case OP_SOD:
787 case OP_SOM:
788 case OP_THEN:
789 case OP_THEN_ARG:
790 case OP_XCLASS:
791 return SSB_FAIL;
792
793 /* We can ignore word boundary tests. */
794
795 case OP_WORD_BOUNDARY:
796 case OP_NOT_WORD_BOUNDARY:
797 tcode++;
798 break;
799
800 /* If we hit a bracket or a positive lookahead assertion, recurse to set
801 bits from within the subpattern. If it can't find anything, we have to
802 give up. If it finds some mandatory character(s), we are done for this
803 branch. Otherwise, carry on scanning after the subpattern. */
804
805 case OP_BRA:
806 case OP_SBRA:
807 case OP_CBRA:
808 case OP_SCBRA:
809 case OP_BRAPOS:
810 case OP_SBRAPOS:
811 case OP_CBRAPOS:
812 case OP_SCBRAPOS:
813 case OP_ONCE:
814 case OP_ONCE_NC:
815 case OP_ASSERT:
816 rc = set_start_bits(tcode, start_bits, utf8, cd);
817 if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
818 if (rc == SSB_DONE) try_next = FALSE; else
819 {
820 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
821 tcode += 1 + LINK_SIZE;
822 }
823 break;
824
825 /* If we hit ALT or KET, it means we haven't found anything mandatory in
826 this branch, though we might have found something optional. For ALT, we
827 continue with the next alternative, but we have to arrange that the final
828 result from subpattern is SSB_CONTINUE rather than SSB_DONE. For KET,
829 return SSB_CONTINUE: if this is the top level, that indicates failure,
830 but after a nested subpattern, it causes scanning to continue. */
831
832 case OP_ALT:
833 yield = SSB_CONTINUE;
834 try_next = FALSE;
835 break;
836
837 case OP_KET:
838 case OP_KETRMAX:
839 case OP_KETRMIN:
840 case OP_KETRPOS:
841 return SSB_CONTINUE;
842
843 /* Skip over callout */
844
845 case OP_CALLOUT:
846 tcode += 2 + 2*LINK_SIZE;
847 break;
848
849 /* Skip over lookbehind and negative lookahead assertions */
850
851 case OP_ASSERT_NOT:
852 case OP_ASSERTBACK:
853 case OP_ASSERTBACK_NOT:
854 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
855 tcode += 1 + LINK_SIZE;
856 break;
857
858 /* BRAZERO does the bracket, but carries on. */
859
860 case OP_BRAZERO:
861 case OP_BRAMINZERO:
862 case OP_BRAPOSZERO:
863 rc = set_start_bits(++tcode, start_bits, utf8, cd);
864 if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
865 /* =========================================================================
866 See the comment at the head of this function concerning the next line,
867 which was an old fudge for the benefit of OS/2.
868 dummy = 1;
869 ========================================================================= */
870 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
871 tcode += 1 + LINK_SIZE;
872 break;
873
874 /* SKIPZERO skips the bracket. */
875
876 case OP_SKIPZERO:
877 tcode++;
878 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
879 tcode += 1 + LINK_SIZE;
880 break;
881
882 /* Single-char * or ? sets the bit and tries the next item */
883
884 case OP_STAR:
885 case OP_MINSTAR:
886 case OP_POSSTAR:
887 case OP_QUERY:
888 case OP_MINQUERY:
889 case OP_POSQUERY:
890 tcode = set_table_bit(start_bits, tcode + 1, FALSE, cd, utf8);
891 break;
892
893 case OP_STARI:
894 case OP_MINSTARI:
895 case OP_POSSTARI:
896 case OP_QUERYI:
897 case OP_MINQUERYI:
898 case OP_POSQUERYI:
899 tcode = set_table_bit(start_bits, tcode + 1, TRUE, cd, utf8);
900 break;
901
902 /* Single-char upto sets the bit and tries the next */
903
904 case OP_UPTO:
905 case OP_MINUPTO:
906 case OP_POSUPTO:
907 tcode = set_table_bit(start_bits, tcode + 3, FALSE, cd, utf8);
908 break;
909
910 case OP_UPTOI:
911 case OP_MINUPTOI:
912 case OP_POSUPTOI:
913 tcode = set_table_bit(start_bits, tcode + 3, TRUE, cd, utf8);
914 break;
915
916 /* At least one single char sets the bit and stops */
917
918 case OP_EXACT:
919 tcode += 2;
920 /* Fall through */
921 case OP_CHAR:
922 case OP_PLUS:
923 case OP_MINPLUS:
924 case OP_POSPLUS:
925 (void)set_table_bit(start_bits, tcode + 1, FALSE, cd, utf8);
926 try_next = FALSE;
927 break;
928
929 case OP_EXACTI:
930 tcode += 2;
931 /* Fall through */
932 case OP_CHARI:
933 case OP_PLUSI:
934 case OP_MINPLUSI:
935 case OP_POSPLUSI:
936 (void)set_table_bit(start_bits, tcode + 1, TRUE, cd, utf8);
937 try_next = FALSE;
938 break;
939
940 /* Special spacing and line-terminating items. These recognize specific
941 lists of characters. The difference between VSPACE and ANYNL is that the
942 latter can match the two-character CRLF sequence, but that is not
943 relevant for finding the first character, so their code here is
944 identical. */
945
946 case OP_HSPACE:
947 SET_BIT(0x09);
948 SET_BIT(0x20);
949 if (utf8)
950 {
951 SET_BIT(0xC2); /* For U+00A0 */
952 SET_BIT(0xE1); /* For U+1680, U+180E */
953 SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
954 SET_BIT(0xE3); /* For U+3000 */
955 }
956 else SET_BIT(0xA0);
957 try_next = FALSE;
958 break;
959
960 case OP_ANYNL:
961 case OP_VSPACE:
962 SET_BIT(0x0A);
963 SET_BIT(0x0B);
964 SET_BIT(0x0C);
965 SET_BIT(0x0D);
966 if (utf8)
967 {
968 SET_BIT(0xC2); /* For U+0085 */
969 SET_BIT(0xE2); /* For U+2028, U+2029 */
970 }
971 else SET_BIT(0x85);
972 try_next = FALSE;
973 break;
974
975 /* Single character types set the bits and stop. Note that if PCRE_UCP
976 is set, we do not see these op codes because \d etc are converted to
977 properties. Therefore, these apply in the case when only characters less
978 than 256 are recognized to match the types. */
979
980 case OP_NOT_DIGIT:
981 set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
982 try_next = FALSE;
983 break;
984
985 case OP_DIGIT:
986 set_type_bits(start_bits, cbit_digit, table_limit, cd);
987 try_next = FALSE;
988 break;
989
990 /* The cbit_space table has vertical tab as whitespace; we have to
991 ensure it is set as not whitespace. */
992
993 case OP_NOT_WHITESPACE:
994 set_nottype_bits(start_bits, cbit_space, table_limit, cd);
995 start_bits[1] |= 0x08;
996 try_next = FALSE;
997 break;
998
999 /* The cbit_space table has vertical tab as whitespace; we have to
1000 not set it from the table. */
1001
1002 case OP_WHITESPACE:
1003 c = start_bits[1]; /* Save in case it was already set */
1004 set_type_bits(start_bits, cbit_space, table_limit, cd);
1005 start_bits[1] = (start_bits[1] & ~0x08) | c;
1006 try_next = FALSE;
1007 break;
1008
1009 case OP_NOT_WORDCHAR:
1010 set_nottype_bits(start_bits, cbit_word, table_limit, cd);
1011 try_next = FALSE;
1012 break;
1013
1014 case OP_WORDCHAR:
1015 set_type_bits(start_bits, cbit_word, table_limit, cd);
1016 try_next = FALSE;
1017 break;
1018
1019 /* One or more character type fudges the pointer and restarts, knowing
1020 it will hit a single character type and stop there. */
1021
1022 case OP_TYPEPLUS:
1023 case OP_TYPEMINPLUS:
1024 case OP_TYPEPOSPLUS:
1025 tcode++;
1026 break;
1027
1028 case OP_TYPEEXACT:
1029 tcode += 3;
1030 break;
1031
1032 /* Zero or more repeats of character types set the bits and then
1033 try again. */
1034
1035 case OP_TYPEUPTO:
1036 case OP_TYPEMINUPTO:
1037 case OP_TYPEPOSUPTO:
1038 tcode += 2; /* Fall through */
1039
1040 case OP_TYPESTAR:
1041 case OP_TYPEMINSTAR:
1042 case OP_TYPEPOSSTAR:
1043 case OP_TYPEQUERY:
1044 case OP_TYPEMINQUERY:
1045 case OP_TYPEPOSQUERY:
1046 switch(tcode[1])
1047 {
1048 default:
1049 case OP_ANY:
1050 case OP_ALLANY:
1051 return SSB_FAIL;
1052
1053 case OP_HSPACE:
1054 SET_BIT(0x09);
1055 SET_BIT(0x20);
1056 if (utf8)
1057 {
1058 SET_BIT(0xC2); /* For U+00A0 */
1059 SET_BIT(0xE1); /* For U+1680, U+180E */
1060 SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
1061 SET_BIT(0xE3); /* For U+3000 */
1062 }
1063 else SET_BIT(0xA0);
1064 break;
1065
1066 case OP_ANYNL:
1067 case OP_VSPACE:
1068 SET_BIT(0x0A);
1069 SET_BIT(0x0B);
1070 SET_BIT(0x0C);
1071 SET_BIT(0x0D);
1072 if (utf8)
1073 {
1074 SET_BIT(0xC2); /* For U+0085 */
1075 SET_BIT(0xE2); /* For U+2028, U+2029 */
1076 }
1077 else SET_BIT(0x85);
1078 break;
1079
1080 case OP_NOT_DIGIT:
1081 set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
1082 break;
1083
1084 case OP_DIGIT:
1085 set_type_bits(start_bits, cbit_digit, table_limit, cd);
1086 break;
1087
1088 /* The cbit_space table has vertical tab as whitespace; we have to
1089 ensure it gets set as not whitespace. */
1090
1091 case OP_NOT_WHITESPACE:
1092 set_nottype_bits(start_bits, cbit_space, table_limit, cd);
1093 start_bits[1] |= 0x08;
1094 break;
1095
1096 /* The cbit_space table has vertical tab as whitespace; we have to
1097 avoid setting it. */
1098
1099 case OP_WHITESPACE:
1100 c = start_bits[1]; /* Save in case it was already set */
1101 set_type_bits(start_bits, cbit_space, table_limit, cd);
1102 start_bits[1] = (start_bits[1] & ~0x08) | c;
1103 break;
1104
1105 case OP_NOT_WORDCHAR:
1106 set_nottype_bits(start_bits, cbit_word, table_limit, cd);
1107 break;
1108
1109 case OP_WORDCHAR:
1110 set_type_bits(start_bits, cbit_word, table_limit, cd);
1111 break;
1112 }
1113
1114 tcode += 2;
1115 break;
1116
1117 /* Character class where all the information is in a bit map: set the
1118 bits and either carry on or not, according to the repeat count. If it was
1119 a negative class, and we are operating with UTF-8 characters, any byte
1120 with a value >= 0xc4 is a potentially valid starter because it starts a
1121 character with a value > 255. */
1122
1123 case OP_NCLASS:
1124 #ifdef SUPPORT_UTF8
1125 if (utf8)
1126 {
1127 start_bits[24] |= 0xf0; /* Bits for 0xc4 - 0xc8 */
1128 memset(start_bits+25, 0xff, 7); /* Bits for 0xc9 - 0xff */
1129 }
1130 #endif
1131 /* Fall through */
1132
1133 case OP_CLASS:
1134 {
1135 tcode++;
1136
1137 /* In UTF-8 mode, the bits in a bit map correspond to character
1138 values, not to byte values. However, the bit map we are constructing is
1139 for byte values. So we have to do a conversion for characters whose
1140 value is > 127. In fact, there are only two possible starting bytes for
1141 characters in the range 128 - 255. */
1142
1143 #ifdef SUPPORT_UTF8
1144 if (utf8)
1145 {
1146 for (c = 0; c < 16; c++) start_bits[c] |= tcode[c];
1147 for (c = 128; c < 256; c++)
1148 {
1149 if ((tcode[c/8] && (1 << (c&7))) != 0)
1150 {
1151 int d = (c >> 6) | 0xc0; /* Set bit for this starter */
1152 start_bits[d/8] |= (1 << (d&7)); /* and then skip on to the */
1153 c = (c & 0xc0) + 0x40 - 1; /* next relevant character. */
1154 }
1155 }
1156 }
1157
1158 /* In non-UTF-8 mode, the two bit maps are completely compatible. */
1159
1160 else
1161 #endif
1162 {
1163 for (c = 0; c < 32; c++) start_bits[c] |= tcode[c];
1164 }
1165
1166 /* Advance past the bit map, and act on what follows. For a zero
1167 minimum repeat, continue; otherwise stop processing. */
1168
1169 tcode += 32;
1170 switch (*tcode)
1171 {
1172 case OP_CRSTAR:
1173 case OP_CRMINSTAR:
1174 case OP_CRQUERY:
1175 case OP_CRMINQUERY:
1176 tcode++;
1177 break;
1178
1179 case OP_CRRANGE:
1180 case OP_CRMINRANGE:
1181 if (((tcode[1] << 8) + tcode[2]) == 0) tcode += 5;
1182 else try_next = FALSE;
1183 break;
1184
1185 default:
1186 try_next = FALSE;
1187 break;
1188 }
1189 }
1190 break; /* End of bitmap class handling */
1191
1192 } /* End of switch */
1193 } /* End of try_next loop */
1194
1195 code += GET(code, 1); /* Advance to next branch */
1196 }
1197 while (*code == OP_ALT);
1198 return yield;
1199 }
1200
1201
1202
1203
1204
1205 /*************************************************
1206 * Study a compiled expression *
1207 *************************************************/
1208
1209 /* This function is handed a compiled expression that it must study to produce
1210 information that will speed up the matching. It returns a pcre_extra block
1211 which then gets handed back to pcre_exec().
1212
1213 Arguments:
1214 re points to the compiled expression
1215 options contains option bits
1216 errorptr points to where to place error messages;
1217 set NULL unless error
1218
1219 Returns: pointer to a pcre_extra block, with study_data filled in and the
1220 appropriate flags set;
1221 NULL on error or if no optimization possible
1222 */
1223
1224 #ifdef COMPILE_PCRE8
1225 PCRE_EXP_DEFN pcre_extra * PCRE_CALL_CONVENTION
1226 pcre_study(const pcre *external_re, int options, const char **errorptr)
1227 #else
1228 PCRE_EXP_DEFN pcre_extra * PCRE_CALL_CONVENTION
1229 pcre16_study(const pcre *external_re, int options, const char **errorptr)
1230 #endif
1231 {
1232 int min;
1233 BOOL bits_set = FALSE;
1234 pcre_uint8 start_bits[32];
1235 pcre_extra *extra = NULL;
1236 pcre_study_data *study;
1237 const pcre_uint8 *tables;
1238 pcre_uchar *code;
1239 compile_data compile_block;
1240 const real_pcre *re = (const real_pcre *)external_re;
1241
1242 *errorptr = NULL;
1243
1244 if (re == NULL || re->magic_number != MAGIC_NUMBER)
1245 {
1246 *errorptr = "argument is not a compiled regular expression";
1247 return NULL;
1248 }
1249
1250 if ((options & ~PUBLIC_STUDY_OPTIONS) != 0)
1251 {
1252 *errorptr = "unknown or incorrect option bit(s) set";
1253 return NULL;
1254 }
1255
1256 code = (pcre_uchar *)re + re->name_table_offset +
1257 (re->name_count * re->name_entry_size);
1258
1259 /* For an anchored pattern, or an unanchored pattern that has a first char, or
1260 a multiline pattern that matches only at "line starts", there is no point in
1261 seeking a list of starting bytes. */
1262
1263 if ((re->options & PCRE_ANCHORED) == 0 &&
1264 (re->flags & (PCRE_FIRSTSET|PCRE_STARTLINE)) == 0)
1265 {
1266 int rc;
1267
1268 /* Set the character tables in the block that is passed around */
1269
1270 tables = re->tables;
1271 if (tables == NULL)
1272 (void)pcre_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES,
1273 (void *)(&tables));
1274
1275 compile_block.lcc = tables + lcc_offset;
1276 compile_block.fcc = tables + fcc_offset;
1277 compile_block.cbits = tables + cbits_offset;
1278 compile_block.ctypes = tables + ctypes_offset;
1279
1280 /* See if we can find a fixed set of initial characters for the pattern. */
1281
1282 memset(start_bits, 0, 32 * sizeof(pcre_uint8));
1283 rc = set_start_bits(code, start_bits, (re->options & PCRE_UTF8) != 0,
1284 &compile_block);
1285 bits_set = rc == SSB_DONE;
1286 if (rc == SSB_UNKNOWN)
1287 {
1288 *errorptr = "internal error: opcode not recognized";
1289 return NULL;
1290 }
1291 }
1292
1293 /* Find the minimum length of subject string. */
1294
1295 switch(min = find_minlength(code, code, re->options, 0))
1296 {
1297 case -2: *errorptr = "internal error: missing capturing bracket"; return NULL;
1298 case -3: *errorptr = "internal error: opcode not recognized"; return NULL;
1299 default: break;
1300 }
1301
1302 /* If a set of starting bytes has been identified, or if the minimum length is
1303 greater than zero, or if JIT optimization has been requested, get a pcre_extra
1304 block and a pcre_study_data block. The study data is put in the latter, which
1305 is pointed to by the former, which may also get additional data set later by
1306 the calling program. At the moment, the size of pcre_study_data is fixed. We
1307 nevertheless save it in a field for returning via the pcre_fullinfo() function
1308 so that if it becomes variable in the future, we don't have to change that
1309 code. */
1310
1311 if (bits_set || min > 0
1312 #ifdef SUPPORT_JIT
1313 || (options & PCRE_STUDY_JIT_COMPILE) != 0
1314 #endif
1315 )
1316 {
1317 extra = (pcre_extra *)(pcre_malloc)
1318 (sizeof(pcre_extra) + sizeof(pcre_study_data));
1319 if (extra == NULL)
1320 {
1321 *errorptr = "failed to get memory";
1322 return NULL;
1323 }
1324
1325 study = (pcre_study_data *)((char *)extra + sizeof(pcre_extra));
1326 extra->flags = PCRE_EXTRA_STUDY_DATA;
1327 extra->study_data = study;
1328
1329 study->size = sizeof(pcre_study_data);
1330 study->flags = 0;
1331
1332 if (bits_set)
1333 {
1334 study->flags |= PCRE_STUDY_MAPPED;
1335 memcpy(study->start_bits, start_bits, sizeof(start_bits));
1336 }
1337
1338 /* Always set the minlength value in the block, because the JIT compiler
1339 makes use of it. However, don't set the bit unless the length is greater than
1340 zero - the interpretive pcre_exec() and pcre_dfa_exec() needn't waste time
1341 checking the zero case. */
1342
1343 if (min > 0)
1344 {
1345 study->flags |= PCRE_STUDY_MINLEN;
1346 study->minlength = min;
1347 }
1348 else study->minlength = 0;
1349
1350 /* If JIT support was compiled and requested, attempt the JIT compilation.
1351 If no starting bytes were found, and the minimum length is zero, and JIT
1352 compilation fails, abandon the extra block and return NULL. */
1353
1354 #ifdef SUPPORT_JIT
1355 extra->executable_jit = NULL;
1356 if ((options & PCRE_STUDY_JIT_COMPILE) != 0) PRIV(jit_compile)(re, extra);
1357 if (study->flags == 0 && (extra->flags & PCRE_EXTRA_EXECUTABLE_JIT) == 0)
1358 {
1359 pcre_free_study(extra);
1360 extra = NULL;
1361 }
1362 #endif
1363 }
1364
1365 return extra;
1366 }
1367
1368
1369 /*************************************************
1370 * Free the study data *
1371 *************************************************/
1372
1373 /* This function frees the memory that was obtained by pcre_study().
1374
1375 Argument: a pointer to the pcre_extra block
1376 Returns: nothing
1377 */
1378
1379 #ifdef COMPILE_PCRE8
1380 PCRE_EXP_DEFN void
1381 pcre_free_study(pcre_extra *extra)
1382 #else
1383 PCRE_EXP_DEFN void
1384 pcre16_free_study(pcre_extra *extra)
1385 #endif
1386 {
1387 #ifdef SUPPORT_JIT
1388 if ((extra->flags & PCRE_EXTRA_EXECUTABLE_JIT) != 0 &&
1389 extra->executable_jit != NULL)
1390 PRIV(jit_free)(extra->executable_jit);
1391 #endif
1392 pcre_free(extra);
1393 }
1394
1395 /* 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