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

Contents of /code/trunk/pcre_study.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1414 - (show annotations)
Sun Dec 22 16:27:35 2013 UTC (5 years, 10 months ago) by zherczeg
File MIME type: text/plain
File size: 46779 byte(s)
A new flag is set, when property checks are present in an XCLASS.
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) 1997-2012 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 re compiled pattern block
70 code pointer to start of group (the bracket)
71 startcode pointer to start of the whole pattern's code
72 options the compiling options
73 int RECURSE depth
74
75 Returns: the minimum length
76 -1 if \C in UTF-8 mode or (*ACCEPT) was encountered
77 -2 internal error (missing capturing bracket)
78 -3 internal error (opcode not listed)
79 */
80
81 static int
82 find_minlength(const REAL_PCRE *re, const pcre_uchar *code,
83 const pcre_uchar *startcode, int options, int recurse_depth)
84 {
85 int length = -1;
86 /* PCRE_UTF16 has the same value as PCRE_UTF8. */
87 BOOL utf = (options & PCRE_UTF8) != 0;
88 BOOL had_recurse = FALSE;
89 register int branchlength = 0;
90 register pcre_uchar *cc = (pcre_uchar *)code + 1 + LINK_SIZE;
91
92 if (*code == OP_CBRA || *code == OP_SCBRA ||
93 *code == OP_CBRAPOS || *code == OP_SCBRAPOS) cc += IMM2_SIZE;
94
95 /* Scan along the opcodes for this branch. If we get to the end of the
96 branch, check the length against that of the other branches. */
97
98 for (;;)
99 {
100 int d, min;
101 pcre_uchar *cs, *ce;
102 register pcre_uchar op = *cc;
103
104 switch (op)
105 {
106 case OP_COND:
107 case OP_SCOND:
108
109 /* If there is only one branch in a condition, the implied branch has zero
110 length, so we don't add anything. This covers the DEFINE "condition"
111 automatically. */
112
113 cs = cc + GET(cc, 1);
114 if (*cs != OP_ALT)
115 {
116 cc = cs + 1 + LINK_SIZE;
117 break;
118 }
119
120 /* Otherwise we can fall through and treat it the same as any other
121 subpattern. */
122
123 case OP_CBRA:
124 case OP_SCBRA:
125 case OP_BRA:
126 case OP_SBRA:
127 case OP_CBRAPOS:
128 case OP_SCBRAPOS:
129 case OP_BRAPOS:
130 case OP_SBRAPOS:
131 case OP_ONCE:
132 case OP_ONCE_NC:
133 d = find_minlength(re, cc, startcode, options, recurse_depth);
134 if (d < 0) return d;
135 branchlength += d;
136 do cc += GET(cc, 1); while (*cc == OP_ALT);
137 cc += 1 + LINK_SIZE;
138 break;
139
140 /* ACCEPT makes things far too complicated; we have to give up. */
141
142 case OP_ACCEPT:
143 case OP_ASSERT_ACCEPT:
144 return -1;
145
146 /* Reached end of a branch; if it's a ket it is the end of a nested
147 call. If it's ALT it is an alternation in a nested call. If it is END it's
148 the end of the outer call. All can be handled by the same code. If an
149 ACCEPT was previously encountered, use the length that was in force at that
150 time, and pass back the shortest ACCEPT length. */
151
152 case OP_ALT:
153 case OP_KET:
154 case OP_KETRMAX:
155 case OP_KETRMIN:
156 case OP_KETRPOS:
157 case OP_END:
158 if (length < 0 || (!had_recurse && branchlength < length))
159 length = branchlength;
160 if (op != OP_ALT) return length;
161 cc += 1 + LINK_SIZE;
162 branchlength = 0;
163 had_recurse = FALSE;
164 break;
165
166 /* Skip over assertive subpatterns */
167
168 case OP_ASSERT:
169 case OP_ASSERT_NOT:
170 case OP_ASSERTBACK:
171 case OP_ASSERTBACK_NOT:
172 do cc += GET(cc, 1); while (*cc == OP_ALT);
173 /* Fall through */
174
175 /* Skip over things that don't match chars */
176
177 case OP_REVERSE:
178 case OP_CREF:
179 case OP_DNCREF:
180 case OP_RREF:
181 case OP_DNRREF:
182 case OP_DEF:
183 case OP_CALLOUT:
184 case OP_SOD:
185 case OP_SOM:
186 case OP_EOD:
187 case OP_EODN:
188 case OP_CIRC:
189 case OP_CIRCM:
190 case OP_DOLL:
191 case OP_DOLLM:
192 case OP_NOT_WORD_BOUNDARY:
193 case OP_WORD_BOUNDARY:
194 cc += PRIV(OP_lengths)[*cc];
195 break;
196
197 /* Skip over a subpattern that has a {0} or {0,x} quantifier */
198
199 case OP_BRAZERO:
200 case OP_BRAMINZERO:
201 case OP_BRAPOSZERO:
202 case OP_SKIPZERO:
203 cc += PRIV(OP_lengths)[*cc];
204 do cc += GET(cc, 1); while (*cc == OP_ALT);
205 cc += 1 + LINK_SIZE;
206 break;
207
208 /* Handle literal characters and + repetitions */
209
210 case OP_CHAR:
211 case OP_CHARI:
212 case OP_NOT:
213 case OP_NOTI:
214 case OP_PLUS:
215 case OP_PLUSI:
216 case OP_MINPLUS:
217 case OP_MINPLUSI:
218 case OP_POSPLUS:
219 case OP_POSPLUSI:
220 case OP_NOTPLUS:
221 case OP_NOTPLUSI:
222 case OP_NOTMINPLUS:
223 case OP_NOTMINPLUSI:
224 case OP_NOTPOSPLUS:
225 case OP_NOTPOSPLUSI:
226 branchlength++;
227 cc += 2;
228 #ifdef SUPPORT_UTF
229 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
230 #endif
231 break;
232
233 case OP_TYPEPLUS:
234 case OP_TYPEMINPLUS:
235 case OP_TYPEPOSPLUS:
236 branchlength++;
237 cc += (cc[1] == OP_PROP || cc[1] == OP_NOTPROP)? 4 : 2;
238 break;
239
240 /* Handle exact repetitions. The count is already in characters, but we
241 need to skip over a multibyte character in UTF8 mode. */
242
243 case OP_EXACT:
244 case OP_EXACTI:
245 case OP_NOTEXACT:
246 case OP_NOTEXACTI:
247 branchlength += GET2(cc,1);
248 cc += 2 + IMM2_SIZE;
249 #ifdef SUPPORT_UTF
250 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
251 #endif
252 break;
253
254 case OP_TYPEEXACT:
255 branchlength += GET2(cc,1);
256 cc += 2 + IMM2_SIZE + ((cc[1 + IMM2_SIZE] == OP_PROP
257 || cc[1 + IMM2_SIZE] == OP_NOTPROP)? 2 : 0);
258 break;
259
260 /* Handle single-char non-literal matchers */
261
262 case OP_PROP:
263 case OP_NOTPROP:
264 cc += 2;
265 /* Fall through */
266
267 case OP_NOT_DIGIT:
268 case OP_DIGIT:
269 case OP_NOT_WHITESPACE:
270 case OP_WHITESPACE:
271 case OP_NOT_WORDCHAR:
272 case OP_WORDCHAR:
273 case OP_ANY:
274 case OP_ALLANY:
275 case OP_EXTUNI:
276 case OP_HSPACE:
277 case OP_NOT_HSPACE:
278 case OP_VSPACE:
279 case OP_NOT_VSPACE:
280 branchlength++;
281 cc++;
282 break;
283
284 /* "Any newline" might match two characters, but it also might match just
285 one. */
286
287 case OP_ANYNL:
288 branchlength += 1;
289 cc++;
290 break;
291
292 /* The single-byte matcher means we can't proceed in UTF-8 mode. (In
293 non-UTF-8 mode \C will actually be turned into OP_ALLANY, so won't ever
294 appear, but leave the code, just in case.) */
295
296 case OP_ANYBYTE:
297 #ifdef SUPPORT_UTF
298 if (utf) return -1;
299 #endif
300 branchlength++;
301 cc++;
302 break;
303
304 /* For repeated character types, we have to test for \p and \P, which have
305 an extra two bytes of parameters. */
306
307 case OP_TYPESTAR:
308 case OP_TYPEMINSTAR:
309 case OP_TYPEQUERY:
310 case OP_TYPEMINQUERY:
311 case OP_TYPEPOSSTAR:
312 case OP_TYPEPOSQUERY:
313 if (cc[1] == OP_PROP || cc[1] == OP_NOTPROP) cc += 2;
314 cc += PRIV(OP_lengths)[op];
315 break;
316
317 case OP_TYPEUPTO:
318 case OP_TYPEMINUPTO:
319 case OP_TYPEPOSUPTO:
320 if (cc[1 + IMM2_SIZE] == OP_PROP
321 || cc[1 + IMM2_SIZE] == OP_NOTPROP) cc += 2;
322 cc += PRIV(OP_lengths)[op];
323 break;
324
325 /* Check a class for variable quantification */
326
327 case OP_CLASS:
328 case OP_NCLASS:
329 #if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32
330 case OP_XCLASS:
331 /* The original code caused an unsigned overflow in 64 bit systems,
332 so now we use a conditional statement. */
333 if (op == OP_XCLASS)
334 cc += GET(cc, 1);
335 else
336 cc += PRIV(OP_lengths)[OP_CLASS];
337 #else
338 cc += PRIV(OP_lengths)[OP_CLASS];
339 #endif
340
341 switch (*cc)
342 {
343 case OP_CRPLUS:
344 case OP_CRMINPLUS:
345 case OP_CRPOSPLUS:
346 branchlength++;
347 /* Fall through */
348
349 case OP_CRSTAR:
350 case OP_CRMINSTAR:
351 case OP_CRQUERY:
352 case OP_CRMINQUERY:
353 case OP_CRPOSSTAR:
354 case OP_CRPOSQUERY:
355 cc++;
356 break;
357
358 case OP_CRRANGE:
359 case OP_CRMINRANGE:
360 case OP_CRPOSRANGE:
361 branchlength += GET2(cc,1);
362 cc += 1 + 2 * IMM2_SIZE;
363 break;
364
365 default:
366 branchlength++;
367 break;
368 }
369 break;
370
371 /* Backreferences and subroutine calls are treated in the same way: we find
372 the minimum length for the subpattern. A recursion, however, causes an
373 a flag to be set that causes the length of this branch to be ignored. The
374 logic is that a recursion can only make sense if there is another
375 alternation that stops the recursing. That will provide the minimum length
376 (when no recursion happens). A backreference within the group that it is
377 referencing behaves in the same way.
378
379 If PCRE_JAVASCRIPT_COMPAT is set, a backreference to an unset bracket
380 matches an empty string (by default it causes a matching failure), so in
381 that case we must set the minimum length to zero. */
382
383 case OP_DNREF: /* Duplicate named pattern back reference */
384 case OP_DNREFI:
385 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0)
386 {
387 int count = GET2(cc, 1+IMM2_SIZE);
388 pcre_uchar *slot = (pcre_uchar *)re +
389 re->name_table_offset + GET2(cc, 1) * re->name_entry_size;
390 d = INT_MAX;
391 while (count-- > 0)
392 {
393 ce = cs = (pcre_uchar *)PRIV(find_bracket)(startcode, utf, GET2(slot, 0));
394 if (cs == NULL) return -2;
395 do ce += GET(ce, 1); while (*ce == OP_ALT);
396 if (cc > cs && cc < ce)
397 {
398 d = 0;
399 had_recurse = TRUE;
400 break;
401 }
402 else
403 {
404 int dd = find_minlength(re, cs, startcode, options, recurse_depth);
405 if (dd < d) d = dd;
406 }
407 slot += re->name_entry_size;
408 }
409 }
410 else d = 0;
411 cc += 1 + 2*IMM2_SIZE;
412 goto REPEAT_BACK_REFERENCE;
413
414 case OP_REF: /* Single back reference */
415 case OP_REFI:
416 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0)
417 {
418 ce = cs = (pcre_uchar *)PRIV(find_bracket)(startcode, utf, GET2(cc, 1));
419 if (cs == NULL) return -2;
420 do ce += GET(ce, 1); while (*ce == OP_ALT);
421 if (cc > cs && cc < ce)
422 {
423 d = 0;
424 had_recurse = TRUE;
425 }
426 else
427 {
428 d = find_minlength(re, cs, startcode, options, recurse_depth);
429 }
430 }
431 else d = 0;
432 cc += 1 + IMM2_SIZE;
433
434 /* Handle repeated back references */
435
436 REPEAT_BACK_REFERENCE:
437 switch (*cc)
438 {
439 case OP_CRSTAR:
440 case OP_CRMINSTAR:
441 case OP_CRQUERY:
442 case OP_CRMINQUERY:
443 case OP_CRPOSSTAR:
444 case OP_CRPOSQUERY:
445 min = 0;
446 cc++;
447 break;
448
449 case OP_CRPLUS:
450 case OP_CRMINPLUS:
451 case OP_CRPOSPLUS:
452 min = 1;
453 cc++;
454 break;
455
456 case OP_CRRANGE:
457 case OP_CRMINRANGE:
458 case OP_CRPOSRANGE:
459 min = GET2(cc, 1);
460 cc += 1 + 2 * IMM2_SIZE;
461 break;
462
463 default:
464 min = 1;
465 break;
466 }
467
468 branchlength += min * d;
469 break;
470
471 /* We can easily detect direct recursion, but not mutual recursion. This is
472 caught by a recursion depth count. */
473
474 case OP_RECURSE:
475 cs = ce = (pcre_uchar *)startcode + GET(cc, 1);
476 do ce += GET(ce, 1); while (*ce == OP_ALT);
477 if ((cc > cs && cc < ce) || recurse_depth > 10)
478 had_recurse = TRUE;
479 else
480 {
481 branchlength += find_minlength(re, cs, startcode, options,
482 recurse_depth + 1);
483 }
484 cc += 1 + LINK_SIZE;
485 break;
486
487 /* Anything else does not or need not match a character. We can get the
488 item's length from the table, but for those that can match zero occurrences
489 of a character, we must take special action for UTF-8 characters. As it
490 happens, the "NOT" versions of these opcodes are used at present only for
491 ASCII characters, so they could be omitted from this list. However, in
492 future that may change, so we include them here so as not to leave a
493 gotcha for a future maintainer. */
494
495 case OP_UPTO:
496 case OP_UPTOI:
497 case OP_NOTUPTO:
498 case OP_NOTUPTOI:
499 case OP_MINUPTO:
500 case OP_MINUPTOI:
501 case OP_NOTMINUPTO:
502 case OP_NOTMINUPTOI:
503 case OP_POSUPTO:
504 case OP_POSUPTOI:
505 case OP_NOTPOSUPTO:
506 case OP_NOTPOSUPTOI:
507
508 case OP_STAR:
509 case OP_STARI:
510 case OP_NOTSTAR:
511 case OP_NOTSTARI:
512 case OP_MINSTAR:
513 case OP_MINSTARI:
514 case OP_NOTMINSTAR:
515 case OP_NOTMINSTARI:
516 case OP_POSSTAR:
517 case OP_POSSTARI:
518 case OP_NOTPOSSTAR:
519 case OP_NOTPOSSTARI:
520
521 case OP_QUERY:
522 case OP_QUERYI:
523 case OP_NOTQUERY:
524 case OP_NOTQUERYI:
525 case OP_MINQUERY:
526 case OP_MINQUERYI:
527 case OP_NOTMINQUERY:
528 case OP_NOTMINQUERYI:
529 case OP_POSQUERY:
530 case OP_POSQUERYI:
531 case OP_NOTPOSQUERY:
532 case OP_NOTPOSQUERYI:
533
534 cc += PRIV(OP_lengths)[op];
535 #ifdef SUPPORT_UTF
536 if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
537 #endif
538 break;
539
540 /* Skip these, but we need to add in the name length. */
541
542 case OP_MARK:
543 case OP_PRUNE_ARG:
544 case OP_SKIP_ARG:
545 case OP_THEN_ARG:
546 cc += PRIV(OP_lengths)[op] + cc[1];
547 break;
548
549 /* The remaining opcodes are just skipped over. */
550
551 case OP_CLOSE:
552 case OP_COMMIT:
553 case OP_FAIL:
554 case OP_PRUNE:
555 case OP_SET_SOM:
556 case OP_SKIP:
557 case OP_THEN:
558 cc += PRIV(OP_lengths)[op];
559 break;
560
561 /* This should not occur: we list all opcodes explicitly so that when
562 new ones get added they are properly considered. */
563
564 default:
565 return -3;
566 }
567 }
568 /* Control never gets here */
569 }
570
571
572
573 /*************************************************
574 * Set a bit and maybe its alternate case *
575 *************************************************/
576
577 /* Given a character, set its first byte's bit in the table, and also the
578 corresponding bit for the other version of a letter if we are caseless. In
579 UTF-8 mode, for characters greater than 127, we can only do the caseless thing
580 when Unicode property support is available.
581
582 Arguments:
583 start_bits points to the bit map
584 p points to the character
585 caseless the caseless flag
586 cd the block with char table pointers
587 utf TRUE for UTF-8 / UTF-16 / UTF-32 mode
588
589 Returns: pointer after the character
590 */
591
592 static const pcre_uchar *
593 set_table_bit(pcre_uint8 *start_bits, const pcre_uchar *p, BOOL caseless,
594 compile_data *cd, BOOL utf)
595 {
596 pcre_uint32 c = *p;
597
598 #ifdef COMPILE_PCRE8
599 SET_BIT(c);
600
601 #ifdef SUPPORT_UTF
602 if (utf && c > 127)
603 {
604 GETCHARINC(c, p);
605 #ifdef SUPPORT_UCP
606 if (caseless)
607 {
608 pcre_uchar buff[6];
609 c = UCD_OTHERCASE(c);
610 (void)PRIV(ord2utf)(c, buff);
611 SET_BIT(buff[0]);
612 }
613 #endif /* Not SUPPORT_UCP */
614 return p;
615 }
616 #else /* Not SUPPORT_UTF */
617 (void)(utf); /* Stops warning for unused parameter */
618 #endif /* SUPPORT_UTF */
619
620 /* Not UTF-8 mode, or character is less than 127. */
621
622 if (caseless && (cd->ctypes[c] & ctype_letter) != 0) SET_BIT(cd->fcc[c]);
623 return p + 1;
624 #endif /* COMPILE_PCRE8 */
625
626 #if defined COMPILE_PCRE16 || defined COMPILE_PCRE32
627 if (c > 0xff)
628 {
629 c = 0xff;
630 caseless = FALSE;
631 }
632 SET_BIT(c);
633
634 #ifdef SUPPORT_UTF
635 if (utf && c > 127)
636 {
637 GETCHARINC(c, p);
638 #ifdef SUPPORT_UCP
639 if (caseless)
640 {
641 c = UCD_OTHERCASE(c);
642 if (c > 0xff)
643 c = 0xff;
644 SET_BIT(c);
645 }
646 #endif /* SUPPORT_UCP */
647 return p;
648 }
649 #else /* Not SUPPORT_UTF */
650 (void)(utf); /* Stops warning for unused parameter */
651 #endif /* SUPPORT_UTF */
652
653 if (caseless && (cd->ctypes[c] & ctype_letter) != 0) SET_BIT(cd->fcc[c]);
654 return p + 1;
655 #endif
656 }
657
658
659
660 /*************************************************
661 * Set bits for a positive character type *
662 *************************************************/
663
664 /* This function sets starting bits for a character type. In UTF-8 mode, we can
665 only do a direct setting for bytes less than 128, as otherwise there can be
666 confusion with bytes in the middle of UTF-8 characters. In a "traditional"
667 environment, the tables will only recognize ASCII characters anyway, but in at
668 least one Windows environment, some higher bytes bits were set in the tables.
669 So we deal with that case by considering the UTF-8 encoding.
670
671 Arguments:
672 start_bits the starting bitmap
673 cbit type the type of character wanted
674 table_limit 32 for non-UTF-8; 16 for UTF-8
675 cd the block with char table pointers
676
677 Returns: nothing
678 */
679
680 static void
681 set_type_bits(pcre_uint8 *start_bits, int cbit_type, unsigned int table_limit,
682 compile_data *cd)
683 {
684 register pcre_uint32 c;
685 for (c = 0; c < table_limit; c++) start_bits[c] |= cd->cbits[c+cbit_type];
686 #if defined SUPPORT_UTF && defined COMPILE_PCRE8
687 if (table_limit == 32) return;
688 for (c = 128; c < 256; c++)
689 {
690 if ((cd->cbits[c/8] & (1 << (c&7))) != 0)
691 {
692 pcre_uchar buff[6];
693 (void)PRIV(ord2utf)(c, buff);
694 SET_BIT(buff[0]);
695 }
696 }
697 #endif
698 }
699
700
701 /*************************************************
702 * Set bits for a negative character type *
703 *************************************************/
704
705 /* This function sets starting bits for a negative character type such as \D.
706 In UTF-8 mode, we can only do a direct setting for bytes less than 128, as
707 otherwise there can be confusion with bytes in the middle of UTF-8 characters.
708 Unlike in the positive case, where we can set appropriate starting bits for
709 specific high-valued UTF-8 characters, in this case we have to set the bits for
710 all high-valued characters. The lowest is 0xc2, but we overkill by starting at
711 0xc0 (192) for simplicity.
712
713 Arguments:
714 start_bits the starting bitmap
715 cbit type the type of character wanted
716 table_limit 32 for non-UTF-8; 16 for UTF-8
717 cd the block with char table pointers
718
719 Returns: nothing
720 */
721
722 static void
723 set_nottype_bits(pcre_uint8 *start_bits, int cbit_type, unsigned int table_limit,
724 compile_data *cd)
725 {
726 register pcre_uint32 c;
727 for (c = 0; c < table_limit; c++) start_bits[c] |= ~cd->cbits[c+cbit_type];
728 #if defined SUPPORT_UTF && defined COMPILE_PCRE8
729 if (table_limit != 32) for (c = 24; c < 32; c++) start_bits[c] = 0xff;
730 #endif
731 }
732
733
734
735 /*************************************************
736 * Create bitmap of starting bytes *
737 *************************************************/
738
739 /* This function scans a compiled unanchored expression recursively and
740 attempts to build a bitmap of the set of possible starting bytes. As time goes
741 by, we may be able to get more clever at doing this. The SSB_CONTINUE return is
742 useful for parenthesized groups in patterns such as (a*)b where the group
743 provides some optional starting bytes but scanning must continue at the outer
744 level to find at least one mandatory byte. At the outermost level, this
745 function fails unless the result is SSB_DONE.
746
747 Arguments:
748 code points to an expression
749 start_bits points to a 32-byte table, initialized to 0
750 utf TRUE if in UTF-8 / UTF-16 / UTF-32 mode
751 cd the block with char table pointers
752
753 Returns: SSB_FAIL => Failed to find any starting bytes
754 SSB_DONE => Found mandatory starting bytes
755 SSB_CONTINUE => Found optional starting bytes
756 SSB_UNKNOWN => Hit an unrecognized opcode
757 */
758
759 static int
760 set_start_bits(const pcre_uchar *code, pcre_uint8 *start_bits, BOOL utf,
761 compile_data *cd)
762 {
763 register pcre_uint32 c;
764 int yield = SSB_DONE;
765 #if defined SUPPORT_UTF && defined COMPILE_PCRE8
766 int table_limit = utf? 16:32;
767 #else
768 int table_limit = 32;
769 #endif
770
771 #if 0
772 /* ========================================================================= */
773 /* The following comment and code was inserted in January 1999. In May 2006,
774 when it was observed to cause compiler warnings about unused values, I took it
775 out again. If anybody is still using OS/2, they will have to put it back
776 manually. */
777
778 /* This next statement and the later reference to dummy are here in order to
779 trick the optimizer of the IBM C compiler for OS/2 into generating correct
780 code. Apparently IBM isn't going to fix the problem, and we would rather not
781 disable optimization (in this module it actually makes a big difference, and
782 the pcre module can use all the optimization it can get). */
783
784 volatile int dummy;
785 /* ========================================================================= */
786 #endif
787
788 do
789 {
790 BOOL try_next = TRUE;
791 const pcre_uchar *tcode = code + 1 + LINK_SIZE;
792
793 if (*code == OP_CBRA || *code == OP_SCBRA ||
794 *code == OP_CBRAPOS || *code == OP_SCBRAPOS) tcode += IMM2_SIZE;
795
796 while (try_next) /* Loop for items in this branch */
797 {
798 int rc;
799
800 switch(*tcode)
801 {
802 /* If we reach something we don't understand, it means a new opcode has
803 been created that hasn't been added to this code. Hopefully this problem
804 will be discovered during testing. */
805
806 default:
807 return SSB_UNKNOWN;
808
809 /* Fail for a valid opcode that implies no starting bits. */
810
811 case OP_ACCEPT:
812 case OP_ASSERT_ACCEPT:
813 case OP_ALLANY:
814 case OP_ANY:
815 case OP_ANYBYTE:
816 case OP_CIRC:
817 case OP_CIRCM:
818 case OP_CLOSE:
819 case OP_COMMIT:
820 case OP_COND:
821 case OP_CREF:
822 case OP_DEF:
823 case OP_DNCREF:
824 case OP_DNREF:
825 case OP_DNREFI:
826 case OP_DNRREF:
827 case OP_DOLL:
828 case OP_DOLLM:
829 case OP_END:
830 case OP_EOD:
831 case OP_EODN:
832 case OP_EXTUNI:
833 case OP_FAIL:
834 case OP_MARK:
835 case OP_NOT:
836 case OP_NOTEXACT:
837 case OP_NOTEXACTI:
838 case OP_NOTI:
839 case OP_NOTMINPLUS:
840 case OP_NOTMINPLUSI:
841 case OP_NOTMINQUERY:
842 case OP_NOTMINQUERYI:
843 case OP_NOTMINSTAR:
844 case OP_NOTMINSTARI:
845 case OP_NOTMINUPTO:
846 case OP_NOTMINUPTOI:
847 case OP_NOTPLUS:
848 case OP_NOTPLUSI:
849 case OP_NOTPOSPLUS:
850 case OP_NOTPOSPLUSI:
851 case OP_NOTPOSQUERY:
852 case OP_NOTPOSQUERYI:
853 case OP_NOTPOSSTAR:
854 case OP_NOTPOSSTARI:
855 case OP_NOTPOSUPTO:
856 case OP_NOTPOSUPTOI:
857 case OP_NOTPROP:
858 case OP_NOTQUERY:
859 case OP_NOTQUERYI:
860 case OP_NOTSTAR:
861 case OP_NOTSTARI:
862 case OP_NOTUPTO:
863 case OP_NOTUPTOI:
864 case OP_NOT_HSPACE:
865 case OP_NOT_VSPACE:
866 case OP_PROP:
867 case OP_PRUNE:
868 case OP_PRUNE_ARG:
869 case OP_RECURSE:
870 case OP_REF:
871 case OP_REFI:
872 case OP_REVERSE:
873 case OP_RREF:
874 case OP_SCOND:
875 case OP_SET_SOM:
876 case OP_SKIP:
877 case OP_SKIP_ARG:
878 case OP_SOD:
879 case OP_SOM:
880 case OP_THEN:
881 case OP_THEN_ARG:
882 return SSB_FAIL;
883
884 /* We can ignore word boundary tests. */
885
886 case OP_WORD_BOUNDARY:
887 case OP_NOT_WORD_BOUNDARY:
888 tcode++;
889 break;
890
891 /* If we hit a bracket or a positive lookahead assertion, recurse to set
892 bits from within the subpattern. If it can't find anything, we have to
893 give up. If it finds some mandatory character(s), we are done for this
894 branch. Otherwise, carry on scanning after the subpattern. */
895
896 case OP_BRA:
897 case OP_SBRA:
898 case OP_CBRA:
899 case OP_SCBRA:
900 case OP_BRAPOS:
901 case OP_SBRAPOS:
902 case OP_CBRAPOS:
903 case OP_SCBRAPOS:
904 case OP_ONCE:
905 case OP_ONCE_NC:
906 case OP_ASSERT:
907 rc = set_start_bits(tcode, start_bits, utf, cd);
908 if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
909 if (rc == SSB_DONE) try_next = FALSE; else
910 {
911 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
912 tcode += 1 + LINK_SIZE;
913 }
914 break;
915
916 /* If we hit ALT or KET, it means we haven't found anything mandatory in
917 this branch, though we might have found something optional. For ALT, we
918 continue with the next alternative, but we have to arrange that the final
919 result from subpattern is SSB_CONTINUE rather than SSB_DONE. For KET,
920 return SSB_CONTINUE: if this is the top level, that indicates failure,
921 but after a nested subpattern, it causes scanning to continue. */
922
923 case OP_ALT:
924 yield = SSB_CONTINUE;
925 try_next = FALSE;
926 break;
927
928 case OP_KET:
929 case OP_KETRMAX:
930 case OP_KETRMIN:
931 case OP_KETRPOS:
932 return SSB_CONTINUE;
933
934 /* Skip over callout */
935
936 case OP_CALLOUT:
937 tcode += 2 + 2*LINK_SIZE;
938 break;
939
940 /* Skip over lookbehind and negative lookahead assertions */
941
942 case OP_ASSERT_NOT:
943 case OP_ASSERTBACK:
944 case OP_ASSERTBACK_NOT:
945 do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
946 tcode += 1 + LINK_SIZE;
947 break;
948
949 /* BRAZERO does the bracket, but carries on. */
950
951 case OP_BRAZERO:
952 case OP_BRAMINZERO:
953 case OP_BRAPOSZERO:
954 rc = set_start_bits(++tcode, start_bits, utf, cd);
955 if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
956 /* =========================================================================
957 See the comment at the head of this function concerning the next line,
958 which was an old fudge for the benefit of OS/2.
959 dummy = 1;
960 ========================================================================= */
961 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
962 tcode += 1 + LINK_SIZE;
963 break;
964
965 /* SKIPZERO skips the bracket. */
966
967 case OP_SKIPZERO:
968 tcode++;
969 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
970 tcode += 1 + LINK_SIZE;
971 break;
972
973 /* Single-char * or ? sets the bit and tries the next item */
974
975 case OP_STAR:
976 case OP_MINSTAR:
977 case OP_POSSTAR:
978 case OP_QUERY:
979 case OP_MINQUERY:
980 case OP_POSQUERY:
981 tcode = set_table_bit(start_bits, tcode + 1, FALSE, cd, utf);
982 break;
983
984 case OP_STARI:
985 case OP_MINSTARI:
986 case OP_POSSTARI:
987 case OP_QUERYI:
988 case OP_MINQUERYI:
989 case OP_POSQUERYI:
990 tcode = set_table_bit(start_bits, tcode + 1, TRUE, cd, utf);
991 break;
992
993 /* Single-char upto sets the bit and tries the next */
994
995 case OP_UPTO:
996 case OP_MINUPTO:
997 case OP_POSUPTO:
998 tcode = set_table_bit(start_bits, tcode + 1 + IMM2_SIZE, FALSE, cd, utf);
999 break;
1000
1001 case OP_UPTOI:
1002 case OP_MINUPTOI:
1003 case OP_POSUPTOI:
1004 tcode = set_table_bit(start_bits, tcode + 1 + IMM2_SIZE, TRUE, cd, utf);
1005 break;
1006
1007 /* At least one single char sets the bit and stops */
1008
1009 case OP_EXACT:
1010 tcode += IMM2_SIZE;
1011 /* Fall through */
1012 case OP_CHAR:
1013 case OP_PLUS:
1014 case OP_MINPLUS:
1015 case OP_POSPLUS:
1016 (void)set_table_bit(start_bits, tcode + 1, FALSE, cd, utf);
1017 try_next = FALSE;
1018 break;
1019
1020 case OP_EXACTI:
1021 tcode += IMM2_SIZE;
1022 /* Fall through */
1023 case OP_CHARI:
1024 case OP_PLUSI:
1025 case OP_MINPLUSI:
1026 case OP_POSPLUSI:
1027 (void)set_table_bit(start_bits, tcode + 1, TRUE, cd, utf);
1028 try_next = FALSE;
1029 break;
1030
1031 /* Special spacing and line-terminating items. These recognize specific
1032 lists of characters. The difference between VSPACE and ANYNL is that the
1033 latter can match the two-character CRLF sequence, but that is not
1034 relevant for finding the first character, so their code here is
1035 identical. */
1036
1037 case OP_HSPACE:
1038 SET_BIT(CHAR_HT);
1039 SET_BIT(CHAR_SPACE);
1040 #ifdef SUPPORT_UTF
1041 if (utf)
1042 {
1043 #ifdef COMPILE_PCRE8
1044 SET_BIT(0xC2); /* For U+00A0 */
1045 SET_BIT(0xE1); /* For U+1680, U+180E */
1046 SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
1047 SET_BIT(0xE3); /* For U+3000 */
1048 #elif defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1049 SET_BIT(0xA0);
1050 SET_BIT(0xFF); /* For characters > 255 */
1051 #endif /* COMPILE_PCRE[8|16|32] */
1052 }
1053 else
1054 #endif /* SUPPORT_UTF */
1055 {
1056 #ifndef EBCDIC
1057 SET_BIT(0xA0);
1058 #endif /* Not EBCDIC */
1059 #if defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1060 SET_BIT(0xFF); /* For characters > 255 */
1061 #endif /* COMPILE_PCRE[16|32] */
1062 }
1063 try_next = FALSE;
1064 break;
1065
1066 case OP_ANYNL:
1067 case OP_VSPACE:
1068 SET_BIT(CHAR_LF);
1069 SET_BIT(CHAR_VT);
1070 SET_BIT(CHAR_FF);
1071 SET_BIT(CHAR_CR);
1072 #ifdef SUPPORT_UTF
1073 if (utf)
1074 {
1075 #ifdef COMPILE_PCRE8
1076 SET_BIT(0xC2); /* For U+0085 */
1077 SET_BIT(0xE2); /* For U+2028, U+2029 */
1078 #elif defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1079 SET_BIT(CHAR_NEL);
1080 SET_BIT(0xFF); /* For characters > 255 */
1081 #endif /* COMPILE_PCRE[8|16|32] */
1082 }
1083 else
1084 #endif /* SUPPORT_UTF */
1085 {
1086 SET_BIT(CHAR_NEL);
1087 #if defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1088 SET_BIT(0xFF); /* For characters > 255 */
1089 #endif
1090 }
1091 try_next = FALSE;
1092 break;
1093
1094 /* Single character types set the bits and stop. Note that if PCRE_UCP
1095 is set, we do not see these op codes because \d etc are converted to
1096 properties. Therefore, these apply in the case when only characters less
1097 than 256 are recognized to match the types. */
1098
1099 case OP_NOT_DIGIT:
1100 set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
1101 try_next = FALSE;
1102 break;
1103
1104 case OP_DIGIT:
1105 set_type_bits(start_bits, cbit_digit, table_limit, cd);
1106 try_next = FALSE;
1107 break;
1108
1109 /* The cbit_space table has vertical tab as whitespace; we have to
1110 ensure it is set as not whitespace. Luckily, the code value is the same
1111 (0x0b) in ASCII and EBCDIC, so we can just adjust the appropriate bit. */
1112
1113 case OP_NOT_WHITESPACE:
1114 set_nottype_bits(start_bits, cbit_space, table_limit, cd);
1115 start_bits[1] |= 0x08;
1116 try_next = FALSE;
1117 break;
1118
1119 /* The cbit_space table has vertical tab as whitespace; we have to not
1120 set it from the table. Luckily, the code value is the same (0x0b) in
1121 ASCII and EBCDIC, so we can just adjust the appropriate bit. */
1122
1123 case OP_WHITESPACE:
1124 c = start_bits[1]; /* Save in case it was already set */
1125 set_type_bits(start_bits, cbit_space, table_limit, cd);
1126 start_bits[1] = (start_bits[1] & ~0x08) | c;
1127 try_next = FALSE;
1128 break;
1129
1130 case OP_NOT_WORDCHAR:
1131 set_nottype_bits(start_bits, cbit_word, table_limit, cd);
1132 try_next = FALSE;
1133 break;
1134
1135 case OP_WORDCHAR:
1136 set_type_bits(start_bits, cbit_word, table_limit, cd);
1137 try_next = FALSE;
1138 break;
1139
1140 /* One or more character type fudges the pointer and restarts, knowing
1141 it will hit a single character type and stop there. */
1142
1143 case OP_TYPEPLUS:
1144 case OP_TYPEMINPLUS:
1145 case OP_TYPEPOSPLUS:
1146 tcode++;
1147 break;
1148
1149 case OP_TYPEEXACT:
1150 tcode += 1 + IMM2_SIZE;
1151 break;
1152
1153 /* Zero or more repeats of character types set the bits and then
1154 try again. */
1155
1156 case OP_TYPEUPTO:
1157 case OP_TYPEMINUPTO:
1158 case OP_TYPEPOSUPTO:
1159 tcode += IMM2_SIZE; /* Fall through */
1160
1161 case OP_TYPESTAR:
1162 case OP_TYPEMINSTAR:
1163 case OP_TYPEPOSSTAR:
1164 case OP_TYPEQUERY:
1165 case OP_TYPEMINQUERY:
1166 case OP_TYPEPOSQUERY:
1167 switch(tcode[1])
1168 {
1169 default:
1170 case OP_ANY:
1171 case OP_ALLANY:
1172 return SSB_FAIL;
1173
1174 case OP_HSPACE:
1175 SET_BIT(CHAR_HT);
1176 SET_BIT(CHAR_SPACE);
1177 #ifdef SUPPORT_UTF
1178 if (utf)
1179 {
1180 #ifdef COMPILE_PCRE8
1181 SET_BIT(0xC2); /* For U+00A0 */
1182 SET_BIT(0xE1); /* For U+1680, U+180E */
1183 SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
1184 SET_BIT(0xE3); /* For U+3000 */
1185 #elif defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1186 SET_BIT(0xA0);
1187 SET_BIT(0xFF); /* For characters > 255 */
1188 #endif /* COMPILE_PCRE[8|16|32] */
1189 }
1190 else
1191 #endif /* SUPPORT_UTF */
1192 #ifndef EBCDIC
1193 SET_BIT(0xA0);
1194 #endif /* Not EBCDIC */
1195 break;
1196
1197 case OP_ANYNL:
1198 case OP_VSPACE:
1199 SET_BIT(CHAR_LF);
1200 SET_BIT(CHAR_VT);
1201 SET_BIT(CHAR_FF);
1202 SET_BIT(CHAR_CR);
1203 #ifdef SUPPORT_UTF
1204 if (utf)
1205 {
1206 #ifdef COMPILE_PCRE8
1207 SET_BIT(0xC2); /* For U+0085 */
1208 SET_BIT(0xE2); /* For U+2028, U+2029 */
1209 #elif defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1210 SET_BIT(CHAR_NEL);
1211 SET_BIT(0xFF); /* For characters > 255 */
1212 #endif /* COMPILE_PCRE16 */
1213 }
1214 else
1215 #endif /* SUPPORT_UTF */
1216 SET_BIT(CHAR_NEL);
1217 break;
1218
1219 case OP_NOT_DIGIT:
1220 set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
1221 break;
1222
1223 case OP_DIGIT:
1224 set_type_bits(start_bits, cbit_digit, table_limit, cd);
1225 break;
1226
1227 /* The cbit_space table has vertical tab as whitespace; we no longer
1228 have to play fancy tricks because Perl added VT to its whitespace at
1229 release 5.18. PCRE added it at release 8.34. */
1230
1231 case OP_NOT_WHITESPACE:
1232 set_nottype_bits(start_bits, cbit_space, table_limit, cd);
1233 break;
1234
1235 case OP_WHITESPACE:
1236 set_type_bits(start_bits, cbit_space, table_limit, cd);
1237 break;
1238
1239 case OP_NOT_WORDCHAR:
1240 set_nottype_bits(start_bits, cbit_word, table_limit, cd);
1241 break;
1242
1243 case OP_WORDCHAR:
1244 set_type_bits(start_bits, cbit_word, table_limit, cd);
1245 break;
1246 }
1247
1248 tcode += 2;
1249 break;
1250
1251 /* Character class where all the information is in a bit map: set the
1252 bits and either carry on or not, according to the repeat count. If it was
1253 a negative class, and we are operating with UTF-8 characters, any byte
1254 with a value >= 0xc4 is a potentially valid starter because it starts a
1255 character with a value > 255. */
1256
1257 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
1258 case OP_XCLASS:
1259 if ((tcode[1 + LINK_SIZE] & XCL_HASPROP) != 0)
1260 return SSB_FAIL;
1261 /* All bits are set. */
1262 if ((tcode[1 + LINK_SIZE] & XCL_MAP) == 0 && (tcode[1 + LINK_SIZE] & XCL_NOT) != 0)
1263 return SSB_FAIL;
1264 #endif
1265 /* Fall through */
1266
1267 case OP_NCLASS:
1268 #if defined SUPPORT_UTF && defined COMPILE_PCRE8
1269 if (utf)
1270 {
1271 start_bits[24] |= 0xf0; /* Bits for 0xc4 - 0xc8 */
1272 memset(start_bits+25, 0xff, 7); /* Bits for 0xc9 - 0xff */
1273 }
1274 #endif
1275 #if defined COMPILE_PCRE16 || defined COMPILE_PCRE32
1276 SET_BIT(0xFF); /* For characters > 255 */
1277 #endif
1278 /* Fall through */
1279
1280 case OP_CLASS:
1281 {
1282 pcre_uint8 *map;
1283 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
1284 map = NULL;
1285 if (*tcode == OP_XCLASS)
1286 {
1287 if ((tcode[1 + LINK_SIZE] & XCL_MAP) != 0)
1288 map = (pcre_uint8 *)(tcode + 1 + LINK_SIZE + 1);
1289 tcode += GET(tcode, 1);
1290 }
1291 else
1292 #endif
1293 {
1294 tcode++;
1295 map = (pcre_uint8 *)tcode;
1296 tcode += 32 / sizeof(pcre_uchar);
1297 }
1298
1299 /* In UTF-8 mode, the bits in a bit map correspond to character
1300 values, not to byte values. However, the bit map we are constructing is
1301 for byte values. So we have to do a conversion for characters whose
1302 value is > 127. In fact, there are only two possible starting bytes for
1303 characters in the range 128 - 255. */
1304
1305 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
1306 if (map != NULL)
1307 #endif
1308 {
1309 #if defined SUPPORT_UTF && defined COMPILE_PCRE8
1310 if (utf)
1311 {
1312 for (c = 0; c < 16; c++) start_bits[c] |= map[c];
1313 for (c = 128; c < 256; c++)
1314 {
1315 if ((map[c/8] && (1 << (c&7))) != 0)
1316 {
1317 int d = (c >> 6) | 0xc0; /* Set bit for this starter */
1318 start_bits[d/8] |= (1 << (d&7)); /* and then skip on to the */
1319 c = (c & 0xc0) + 0x40 - 1; /* next relevant character. */
1320 }
1321 }
1322 }
1323 else
1324 #endif
1325 {
1326 /* In non-UTF-8 mode, the two bit maps are completely compatible. */
1327 for (c = 0; c < 32; c++) start_bits[c] |= map[c];
1328 }
1329 }
1330
1331 /* Advance past the bit map, and act on what follows. For a zero
1332 minimum repeat, continue; otherwise stop processing. */
1333
1334 switch (*tcode)
1335 {
1336 case OP_CRSTAR:
1337 case OP_CRMINSTAR:
1338 case OP_CRQUERY:
1339 case OP_CRMINQUERY:
1340 case OP_CRPOSSTAR:
1341 case OP_CRPOSQUERY:
1342 tcode++;
1343 break;
1344
1345 case OP_CRRANGE:
1346 case OP_CRMINRANGE:
1347 case OP_CRPOSRANGE:
1348 if (GET2(tcode, 1) == 0) tcode += 1 + 2 * IMM2_SIZE;
1349 else try_next = FALSE;
1350 break;
1351
1352 default:
1353 try_next = FALSE;
1354 break;
1355 }
1356 }
1357 break; /* End of bitmap class handling */
1358
1359 } /* End of switch */
1360 } /* End of try_next loop */
1361
1362 code += GET(code, 1); /* Advance to next branch */
1363 }
1364 while (*code == OP_ALT);
1365 return yield;
1366 }
1367
1368
1369
1370
1371
1372 /*************************************************
1373 * Study a compiled expression *
1374 *************************************************/
1375
1376 /* This function is handed a compiled expression that it must study to produce
1377 information that will speed up the matching. It returns a pcre[16]_extra block
1378 which then gets handed back to pcre_exec().
1379
1380 Arguments:
1381 re points to the compiled expression
1382 options contains option bits
1383 errorptr points to where to place error messages;
1384 set NULL unless error
1385
1386 Returns: pointer to a pcre[16]_extra block, with study_data filled in and
1387 the appropriate flags set;
1388 NULL on error or if no optimization possible
1389 */
1390
1391 #if defined COMPILE_PCRE8
1392 PCRE_EXP_DEFN pcre_extra * PCRE_CALL_CONVENTION
1393 pcre_study(const pcre *external_re, int options, const char **errorptr)
1394 #elif defined COMPILE_PCRE16
1395 PCRE_EXP_DEFN pcre16_extra * PCRE_CALL_CONVENTION
1396 pcre16_study(const pcre16 *external_re, int options, const char **errorptr)
1397 #elif defined COMPILE_PCRE32
1398 PCRE_EXP_DEFN pcre32_extra * PCRE_CALL_CONVENTION
1399 pcre32_study(const pcre32 *external_re, int options, const char **errorptr)
1400 #endif
1401 {
1402 int min;
1403 BOOL bits_set = FALSE;
1404 pcre_uint8 start_bits[32];
1405 PUBL(extra) *extra = NULL;
1406 pcre_study_data *study;
1407 const pcre_uint8 *tables;
1408 pcre_uchar *code;
1409 compile_data compile_block;
1410 const REAL_PCRE *re = (const REAL_PCRE *)external_re;
1411
1412
1413 *errorptr = NULL;
1414
1415 if (re == NULL || re->magic_number != MAGIC_NUMBER)
1416 {
1417 *errorptr = "argument is not a compiled regular expression";
1418 return NULL;
1419 }
1420
1421 if ((re->flags & PCRE_MODE) == 0)
1422 {
1423 #if defined COMPILE_PCRE8
1424 *errorptr = "argument not compiled in 8 bit mode";
1425 #elif defined COMPILE_PCRE16
1426 *errorptr = "argument not compiled in 16 bit mode";
1427 #elif defined COMPILE_PCRE32
1428 *errorptr = "argument not compiled in 32 bit mode";
1429 #endif
1430 return NULL;
1431 }
1432
1433 if ((options & ~PUBLIC_STUDY_OPTIONS) != 0)
1434 {
1435 *errorptr = "unknown or incorrect option bit(s) set";
1436 return NULL;
1437 }
1438
1439 code = (pcre_uchar *)re + re->name_table_offset +
1440 (re->name_count * re->name_entry_size);
1441
1442 /* For an anchored pattern, or an unanchored pattern that has a first char, or
1443 a multiline pattern that matches only at "line starts", there is no point in
1444 seeking a list of starting bytes. */
1445
1446 if ((re->options & PCRE_ANCHORED) == 0 &&
1447 (re->flags & (PCRE_FIRSTSET|PCRE_STARTLINE)) == 0)
1448 {
1449 int rc;
1450
1451 /* Set the character tables in the block that is passed around */
1452
1453 tables = re->tables;
1454
1455 #if defined COMPILE_PCRE8
1456 if (tables == NULL)
1457 (void)pcre_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES,
1458 (void *)(&tables));
1459 #elif defined COMPILE_PCRE16
1460 if (tables == NULL)
1461 (void)pcre16_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES,
1462 (void *)(&tables));
1463 #elif defined COMPILE_PCRE32
1464 if (tables == NULL)
1465 (void)pcre32_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES,
1466 (void *)(&tables));
1467 #endif
1468
1469 compile_block.lcc = tables + lcc_offset;
1470 compile_block.fcc = tables + fcc_offset;
1471 compile_block.cbits = tables + cbits_offset;
1472 compile_block.ctypes = tables + ctypes_offset;
1473
1474 /* See if we can find a fixed set of initial characters for the pattern. */
1475
1476 memset(start_bits, 0, 32 * sizeof(pcre_uint8));
1477 rc = set_start_bits(code, start_bits, (re->options & PCRE_UTF8) != 0,
1478 &compile_block);
1479 bits_set = rc == SSB_DONE;
1480 if (rc == SSB_UNKNOWN)
1481 {
1482 *errorptr = "internal error: opcode not recognized";
1483 return NULL;
1484 }
1485 }
1486
1487 /* Find the minimum length of subject string. */
1488
1489 switch(min = find_minlength(re, code, code, re->options, 0))
1490 {
1491 case -2: *errorptr = "internal error: missing capturing bracket"; return NULL;
1492 case -3: *errorptr = "internal error: opcode not recognized"; return NULL;
1493 default: break;
1494 }
1495
1496 /* If a set of starting bytes has been identified, or if the minimum length is
1497 greater than zero, or if JIT optimization has been requested, or if
1498 PCRE_STUDY_EXTRA_NEEDED is set, get a pcre[16]_extra block and a
1499 pcre_study_data block. The study data is put in the latter, which is pointed to
1500 by the former, which may also get additional data set later by the calling
1501 program. At the moment, the size of pcre_study_data is fixed. We nevertheless
1502 save it in a field for returning via the pcre_fullinfo() function so that if it
1503 becomes variable in the future, we don't have to change that code. */
1504
1505 if (bits_set || min > 0 || (options & (
1506 #ifdef SUPPORT_JIT
1507 PCRE_STUDY_JIT_COMPILE | PCRE_STUDY_JIT_PARTIAL_SOFT_COMPILE |
1508 PCRE_STUDY_JIT_PARTIAL_HARD_COMPILE |
1509 #endif
1510 PCRE_STUDY_EXTRA_NEEDED)) != 0)
1511 {
1512 extra = (PUBL(extra) *)(PUBL(malloc))
1513 (sizeof(PUBL(extra)) + sizeof(pcre_study_data));
1514 if (extra == NULL)
1515 {
1516 *errorptr = "failed to get memory";
1517 return NULL;
1518 }
1519
1520 study = (pcre_study_data *)((char *)extra + sizeof(PUBL(extra)));
1521 extra->flags = PCRE_EXTRA_STUDY_DATA;
1522 extra->study_data = study;
1523
1524 study->size = sizeof(pcre_study_data);
1525 study->flags = 0;
1526
1527 /* Set the start bits always, to avoid unset memory errors if the
1528 study data is written to a file, but set the flag only if any of the bits
1529 are set, to save time looking when none are. */
1530
1531 if (bits_set)
1532 {
1533 study->flags |= PCRE_STUDY_MAPPED;
1534 memcpy(study->start_bits, start_bits, sizeof(start_bits));
1535 }
1536 else memset(study->start_bits, 0, 32 * sizeof(pcre_uint8));
1537
1538 #ifdef PCRE_DEBUG
1539 if (bits_set)
1540 {
1541 pcre_uint8 *ptr = start_bits;
1542 int i;
1543
1544 printf("Start bits:\n");
1545 for (i = 0; i < 32; i++)
1546 printf("%3d: %02x%s", i * 8, *ptr++, ((i + 1) & 0x7) != 0? " " : "\n");
1547 }
1548 #endif
1549
1550 /* Always set the minlength value in the block, because the JIT compiler
1551 makes use of it. However, don't set the bit unless the length is greater than
1552 zero - the interpretive pcre_exec() and pcre_dfa_exec() needn't waste time
1553 checking the zero case. */
1554
1555 if (min > 0)
1556 {
1557 study->flags |= PCRE_STUDY_MINLEN;
1558 study->minlength = min;
1559 }
1560 else study->minlength = 0;
1561
1562 /* If JIT support was compiled and requested, attempt the JIT compilation.
1563 If no starting bytes were found, and the minimum length is zero, and JIT
1564 compilation fails, abandon the extra block and return NULL, unless
1565 PCRE_STUDY_EXTRA_NEEDED is set. */
1566
1567 #ifdef SUPPORT_JIT
1568 extra->executable_jit = NULL;
1569 if ((options & PCRE_STUDY_JIT_COMPILE) != 0)
1570 PRIV(jit_compile)(re, extra, JIT_COMPILE);
1571 if ((options & PCRE_STUDY_JIT_PARTIAL_SOFT_COMPILE) != 0)
1572 PRIV(jit_compile)(re, extra, JIT_PARTIAL_SOFT_COMPILE);
1573 if ((options & PCRE_STUDY_JIT_PARTIAL_HARD_COMPILE) != 0)
1574 PRIV(jit_compile)(re, extra, JIT_PARTIAL_HARD_COMPILE);
1575
1576 if (study->flags == 0 && (extra->flags & PCRE_EXTRA_EXECUTABLE_JIT) == 0 &&
1577 (options & PCRE_STUDY_EXTRA_NEEDED) == 0)
1578 {
1579 #if defined COMPILE_PCRE8
1580 pcre_free_study(extra);
1581 #elif defined COMPILE_PCRE16
1582 pcre16_free_study(extra);
1583 #elif defined COMPILE_PCRE32
1584 pcre32_free_study(extra);
1585 #endif
1586 extra = NULL;
1587 }
1588 #endif
1589 }
1590
1591 return extra;
1592 }
1593
1594
1595 /*************************************************
1596 * Free the study data *
1597 *************************************************/
1598
1599 /* This function frees the memory that was obtained by pcre_study().
1600
1601 Argument: a pointer to the pcre[16]_extra block
1602 Returns: nothing
1603 */
1604
1605 #if defined COMPILE_PCRE8
1606 PCRE_EXP_DEFN void
1607 pcre_free_study(pcre_extra *extra)
1608 #elif defined COMPILE_PCRE16
1609 PCRE_EXP_DEFN void
1610 pcre16_free_study(pcre16_extra *extra)
1611 #elif defined COMPILE_PCRE32
1612 PCRE_EXP_DEFN void
1613 pcre32_free_study(pcre32_extra *extra)
1614 #endif
1615 {
1616 if (extra == NULL)
1617 return;
1618 #ifdef SUPPORT_JIT
1619 if ((extra->flags & PCRE_EXTRA_EXECUTABLE_JIT) != 0 &&
1620 extra->executable_jit != NULL)
1621 PRIV(jit_free)(extra->executable_jit);
1622 #endif
1623 PUBL(free)(extra);
1624 }
1625
1626 /* 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