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

Contents of /code/trunk/pcre_study.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 659 - (hide annotations) (download)
Tue Aug 16 09:48:29 2011 UTC (2 years, 11 months ago) by ph10
File MIME type: text/plain
File size: 37548 byte(s)
Formatting tidies for 8.13 release.

1 nigel 77 /*************************************************
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 ph10 475 Copyright (c) 1997-2010 University of Cambridge
10 nigel 77
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 ph10 200 #ifdef HAVE_CONFIG_H
46 ph10 236 #include "config.h"
47 ph10 200 #endif
48 ph10 199
49 nigel 77 #include "pcre_internal.h"
50    
51 ph10 524 #define SET_BIT(c) start_bits[c/8] |= (1 << (c&7))
52 nigel 77
53 nigel 93 /* Returns from set_start_bits() */
54    
55 ph10 605 enum { SSB_FAIL, SSB_DONE, SSB_CONTINUE, SSB_UNKNOWN };
56 nigel 93
57    
58 ph10 455
59 nigel 77 /*************************************************
60 ph10 455 * Find the minimum subject length for a group *
61     *************************************************/
62    
63     /* Scan a parenthesized group and compute the minimum length of subject that
64 ph10 461 is needed to match it. This is a lower bound; it does not mean there is a
65 ph10 455 string of that length that matches. In UTF8 mode, the result is in characters
66     rather than bytes.
67    
68     Arguments:
69 ph10 612 code pointer to start of group (the bracket)
70     startcode pointer to start of the whole pattern
71     options the compiling options
72     had_accept pointer to flag for (*ACCEPT) encountered
73 ph10 654 int RECURSE depth
74 ph10 455
75     Returns: the minimum length
76     -1 if \C was encountered
77 ph10 461 -2 internal error (missing capturing bracket)
78 ph10 605 -3 internal error (opcode not listed)
79 ph10 455 */
80    
81     static int
82 ph10 612 find_minlength(const uschar *code, const uschar *startcode, int options,
83 ph10 620 BOOL *had_accept_ptr, int recurse_depth)
84 ph10 455 {
85     int length = -1;
86     BOOL utf8 = (options & PCRE_UTF8) != 0;
87     BOOL had_recurse = FALSE;
88     register int branchlength = 0;
89     register uschar *cc = (uschar *)code + 1 + LINK_SIZE;
90    
91 ph10 604 if (*code == OP_CBRA || *code == OP_SCBRA ||
92     *code == OP_CBRAPOS || *code == OP_SCBRAPOS) cc += 2;
93 ph10 455
94     /* Scan along the opcodes for this branch. If we get to the end of the
95     branch, check the length against that of the other branches. */
96    
97     for (;;)
98     {
99     int d, min;
100 ph10 461 uschar *cs, *ce;
101 ph10 455 register int op = *cc;
102 ph10 461
103 ph10 455 switch (op)
104     {
105 ph10 471 case OP_COND:
106     case OP_SCOND:
107    
108     /* If there is only one branch in a condition, the implied branch has zero
109     length, so we don't add anything. This covers the DEFINE "condition"
110     automatically. */
111    
112     cs = cc + GET(cc, 1);
113     if (*cs != OP_ALT)
114     {
115     cc = cs + 1 + LINK_SIZE;
116     break;
117     }
118    
119     /* Otherwise we can fall through and treat it the same as any other
120     subpattern. */
121    
122 ph10 455 case OP_CBRA:
123 ph10 461 case OP_SCBRA:
124 ph10 455 case OP_BRA:
125 ph10 461 case OP_SBRA:
126 ph10 604 case OP_CBRAPOS:
127     case OP_SCBRAPOS:
128     case OP_BRAPOS:
129     case OP_SBRAPOS:
130 ph10 455 case OP_ONCE:
131 ph10 620 d = find_minlength(cc, startcode, options, had_accept_ptr, recurse_depth);
132 ph10 455 if (d < 0) return d;
133     branchlength += d;
134 ph10 654 if (*had_accept_ptr) return branchlength;
135 ph10 455 do cc += GET(cc, 1); while (*cc == OP_ALT);
136     cc += 1 + LINK_SIZE;
137     break;
138    
139     /* Reached end of a branch; if it's a ket it is the end of a nested
140 ph10 612 call. If it's ALT it is an alternation in a nested call. If it is END it's
141     the end of the outer call. All can be handled by the same code. If it is
142     ACCEPT, it is essentially the same as END, but we set a flag so that
143     counting stops. */
144 ph10 455
145 ph10 654 case OP_ACCEPT:
146     case OP_ASSERT_ACCEPT:
147 ph10 612 *had_accept_ptr = TRUE;
148 ph10 654 /* Fall through */
149 ph10 455 case OP_ALT:
150     case OP_KET:
151     case OP_KETRMAX:
152     case OP_KETRMIN:
153 ph10 605 case OP_KETRPOS:
154 ph10 455 case OP_END:
155 ph10 461 if (length < 0 || (!had_recurse && branchlength < length))
156 ph10 455 length = branchlength;
157 ph10 612 if (op != OP_ALT) return length;
158 ph10 455 cc += 1 + LINK_SIZE;
159     branchlength = 0;
160 ph10 461 had_recurse = FALSE;
161 ph10 455 break;
162    
163     /* Skip over assertive subpatterns */
164    
165     case OP_ASSERT:
166     case OP_ASSERT_NOT:
167     case OP_ASSERTBACK:
168     case OP_ASSERTBACK_NOT:
169     do cc += GET(cc, 1); while (*cc == OP_ALT);
170     /* Fall through */
171    
172     /* Skip over things that don't match chars */
173    
174     case OP_REVERSE:
175     case OP_CREF:
176 ph10 459 case OP_NCREF:
177 ph10 455 case OP_RREF:
178 ph10 459 case OP_NRREF:
179 ph10 455 case OP_DEF:
180     case OP_CALLOUT:
181     case OP_SOD:
182     case OP_SOM:
183     case OP_EOD:
184     case OP_EODN:
185     case OP_CIRC:
186 ph10 602 case OP_CIRCM:
187 ph10 455 case OP_DOLL:
188 ph10 602 case OP_DOLLM:
189 ph10 455 case OP_NOT_WORD_BOUNDARY:
190     case OP_WORD_BOUNDARY:
191     cc += _pcre_OP_lengths[*cc];
192     break;
193 ph10 461
194 ph10 455 /* Skip over a subpattern that has a {0} or {0,x} quantifier */
195    
196     case OP_BRAZERO:
197 ph10 461 case OP_BRAMINZERO:
198 ph10 604 case OP_BRAPOSZERO:
199 ph10 455 case OP_SKIPZERO:
200     cc += _pcre_OP_lengths[*cc];
201     do cc += GET(cc, 1); while (*cc == OP_ALT);
202     cc += 1 + LINK_SIZE;
203     break;
204    
205     /* Handle literal characters and + repetitions */
206    
207     case OP_CHAR:
208 ph10 602 case OP_CHARI:
209 ph10 455 case OP_NOT:
210 ph10 602 case OP_NOTI:
211 ph10 455 case OP_PLUS:
212 ph10 603 case OP_PLUSI:
213 ph10 455 case OP_MINPLUS:
214 ph10 603 case OP_MINPLUSI:
215 ph10 455 case OP_POSPLUS:
216 ph10 603 case OP_POSPLUSI:
217 ph10 455 case OP_NOTPLUS:
218 ph10 603 case OP_NOTPLUSI:
219 ph10 455 case OP_NOTMINPLUS:
220 ph10 603 case OP_NOTMINPLUSI:
221 ph10 455 case OP_NOTPOSPLUS:
222 ph10 603 case OP_NOTPOSPLUSI:
223 ph10 455 branchlength++;
224     cc += 2;
225     #ifdef SUPPORT_UTF8
226     if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
227     #endif
228     break;
229 ph10 461
230 ph10 455 case OP_TYPEPLUS:
231     case OP_TYPEMINPLUS:
232 ph10 461 case OP_TYPEPOSPLUS:
233 ph10 455 branchlength++;
234     cc += (cc[1] == OP_PROP || cc[1] == OP_NOTPROP)? 4 : 2;
235     break;
236    
237     /* Handle exact repetitions. The count is already in characters, but we
238     need to skip over a multibyte character in UTF8 mode. */
239    
240     case OP_EXACT:
241 ph10 603 case OP_EXACTI:
242 ph10 461 case OP_NOTEXACT:
243 ph10 603 case OP_NOTEXACTI:
244 ph10 455 branchlength += GET2(cc,1);
245     cc += 4;
246     #ifdef SUPPORT_UTF8
247     if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
248     #endif
249     break;
250    
251     case OP_TYPEEXACT:
252     branchlength += GET2(cc,1);
253     cc += (cc[3] == OP_PROP || cc[3] == OP_NOTPROP)? 6 : 4;
254     break;
255    
256     /* Handle single-char non-literal matchers */
257    
258     case OP_PROP:
259     case OP_NOTPROP:
260     cc += 2;
261     /* Fall through */
262    
263     case OP_NOT_DIGIT:
264     case OP_DIGIT:
265     case OP_NOT_WHITESPACE:
266     case OP_WHITESPACE:
267     case OP_NOT_WORDCHAR:
268     case OP_WORDCHAR:
269     case OP_ANY:
270     case OP_ALLANY:
271     case OP_EXTUNI:
272 ph10 461 case OP_HSPACE:
273 ph10 455 case OP_NOT_HSPACE:
274     case OP_VSPACE:
275 ph10 461 case OP_NOT_VSPACE:
276 ph10 455 branchlength++;
277     cc++;
278     break;
279 ph10 461
280 ph10 654 /* "Any newline" might match two characters, but it also might match just
281 ph10 611 one. */
282 ph10 461
283 ph10 455 case OP_ANYNL:
284 ph10 611 branchlength += 1;
285 ph10 455 cc++;
286 ph10 461 break;
287 ph10 455
288     /* The single-byte matcher means we can't proceed in UTF-8 mode */
289    
290     case OP_ANYBYTE:
291     #ifdef SUPPORT_UTF8
292     if (utf8) return -1;
293     #endif
294     branchlength++;
295     cc++;
296 ph10 461 break;
297 ph10 455
298     /* For repeated character types, we have to test for \p and \P, which have
299     an extra two bytes of parameters. */
300    
301     case OP_TYPESTAR:
302     case OP_TYPEMINSTAR:
303     case OP_TYPEQUERY:
304     case OP_TYPEMINQUERY:
305     case OP_TYPEPOSSTAR:
306     case OP_TYPEPOSQUERY:
307     if (cc[1] == OP_PROP || cc[1] == OP_NOTPROP) cc += 2;
308     cc += _pcre_OP_lengths[op];
309     break;
310    
311     case OP_TYPEUPTO:
312     case OP_TYPEMINUPTO:
313     case OP_TYPEPOSUPTO:
314     if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
315     cc += _pcre_OP_lengths[op];
316     break;
317    
318     /* Check a class for variable quantification */
319    
320     #ifdef SUPPORT_UTF8
321     case OP_XCLASS:
322     cc += GET(cc, 1) - 33;
323     /* Fall through */
324     #endif
325    
326     case OP_CLASS:
327     case OP_NCLASS:
328     cc += 33;
329    
330     switch (*cc)
331     {
332     case OP_CRPLUS:
333     case OP_CRMINPLUS:
334     branchlength++;
335 ph10 461 /* Fall through */
336 ph10 455
337     case OP_CRSTAR:
338     case OP_CRMINSTAR:
339     case OP_CRQUERY:
340     case OP_CRMINQUERY:
341 ph10 461 cc++;
342 ph10 455 break;
343 ph10 461
344 ph10 455 case OP_CRRANGE:
345     case OP_CRMINRANGE:
346     branchlength += GET2(cc,1);
347     cc += 5;
348     break;
349 ph10 461
350 ph10 455 default:
351     branchlength++;
352 ph10 461 break;
353 ph10 455 }
354     break;
355 ph10 461
356     /* Backreferences and subroutine calls are treated in the same way: we find
357     the minimum length for the subpattern. A recursion, however, causes an
358 ph10 455 a flag to be set that causes the length of this branch to be ignored. The
359     logic is that a recursion can only make sense if there is another
360     alternation that stops the recursing. That will provide the minimum length
361     (when no recursion happens). A backreference within the group that it is
362 ph10 469 referencing behaves in the same way.
363    
364 ph10 467 If PCRE_JAVASCRIPT_COMPAT is set, a backreference to an unset bracket
365     matches an empty string (by default it causes a matching failure), so in
366     that case we must set the minimum length to zero. */
367 ph10 461
368 ph10 455 case OP_REF:
369 ph10 602 case OP_REFI:
370 ph10 467 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0)
371 ph10 469 {
372 ph10 467 ce = cs = (uschar *)_pcre_find_bracket(startcode, utf8, GET2(cc, 1));
373     if (cs == NULL) return -2;
374     do ce += GET(ce, 1); while (*ce == OP_ALT);
375     if (cc > cs && cc < ce)
376     {
377     d = 0;
378     had_recurse = TRUE;
379     }
380 ph10 654 else
381 ph10 612 {
382 ph10 654 d = find_minlength(cs, startcode, options, had_accept_ptr,
383 ph10 620 recurse_depth);
384 ph10 654 *had_accept_ptr = FALSE;
385     }
386 ph10 461 }
387 ph10 469 else d = 0;
388 ph10 461 cc += 3;
389 ph10 455
390     /* Handle repeated back references */
391 ph10 461
392 ph10 455 switch (*cc)
393     {
394     case OP_CRSTAR:
395     case OP_CRMINSTAR:
396     case OP_CRQUERY:
397     case OP_CRMINQUERY:
398     min = 0;
399     cc++;
400     break;
401 ph10 605
402     case OP_CRPLUS:
403     case OP_CRMINPLUS:
404     min = 1;
405     cc++;
406     break;
407    
408 ph10 455 case OP_CRRANGE:
409     case OP_CRMINRANGE:
410     min = GET2(cc, 1);
411     cc += 5;
412     break;
413 ph10 461
414 ph10 455 default:
415     min = 1;
416     break;
417     }
418    
419     branchlength += min * d;
420 ph10 461 break;
421 ph10 455
422 ph10 654 /* We can easily detect direct recursion, but not mutual recursion. This is
423     caught by a recursion depth count. */
424    
425 ph10 461 case OP_RECURSE:
426 ph10 455 cs = ce = (uschar *)startcode + GET(cc, 1);
427     if (cs == NULL) return -2;
428     do ce += GET(ce, 1); while (*ce == OP_ALT);
429 ph10 620 if ((cc > cs && cc < ce) || recurse_depth > 10)
430 ph10 461 had_recurse = TRUE;
431     else
432 ph10 654 {
433 ph10 620 branchlength += find_minlength(cs, startcode, options, had_accept_ptr,
434     recurse_depth + 1);
435 ph10 612 *had_accept_ptr = FALSE;
436 ph10 654 }
437 ph10 455 cc += 1 + LINK_SIZE;
438     break;
439    
440     /* Anything else does not or need not match a character. We can get the
441 ph10 461 item's length from the table, but for those that can match zero occurrences
442 ph10 602 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 ph10 605 future that may change, so we include them here so as not to leave a
446 ph10 603 gotcha for a future maintainer. */
447 ph10 461
448 ph10 455 case OP_UPTO:
449 ph10 602 case OP_UPTOI:
450 ph10 461 case OP_NOTUPTO:
451 ph10 602 case OP_NOTUPTOI:
452 ph10 455 case OP_MINUPTO:
453 ph10 602 case OP_MINUPTOI:
454 ph10 461 case OP_NOTMINUPTO:
455 ph10 602 case OP_NOTMINUPTOI:
456 ph10 455 case OP_POSUPTO:
457 ph10 602 case OP_POSUPTOI:
458     case OP_NOTPOSUPTO:
459     case OP_NOTPOSUPTOI:
460    
461 ph10 455 case OP_STAR:
462 ph10 602 case OP_STARI:
463     case OP_NOTSTAR:
464     case OP_NOTSTARI:
465 ph10 455 case OP_MINSTAR:
466 ph10 602 case OP_MINSTARI:
467 ph10 461 case OP_NOTMINSTAR:
468 ph10 602 case OP_NOTMINSTARI:
469 ph10 455 case OP_POSSTAR:
470 ph10 602 case OP_POSSTARI:
471 ph10 461 case OP_NOTPOSSTAR:
472 ph10 602 case OP_NOTPOSSTARI:
473    
474 ph10 455 case OP_QUERY:
475 ph10 602 case OP_QUERYI:
476     case OP_NOTQUERY:
477     case OP_NOTQUERYI:
478 ph10 455 case OP_MINQUERY:
479 ph10 602 case OP_MINQUERYI:
480 ph10 455 case OP_NOTMINQUERY:
481 ph10 602 case OP_NOTMINQUERYI:
482 ph10 455 case OP_POSQUERY:
483 ph10 602 case OP_POSQUERYI:
484 ph10 461 case OP_NOTPOSQUERY:
485 ph10 602 case OP_NOTPOSQUERYI:
486    
487 ph10 455 cc += _pcre_OP_lengths[op];
488 ph10 461 #ifdef SUPPORT_UTF8
489 ph10 455 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
490 ph10 461 #endif
491 ph10 455 break;
492 ph10 512
493 ph10 510 /* Skip these, but we need to add in the name length. */
494 ph10 512
495 ph10 510 case OP_MARK:
496     case OP_PRUNE_ARG:
497     case OP_SKIP_ARG:
498     cc += _pcre_OP_lengths[op] + cc[1];
499 ph10 512 break;
500 ph10 455
501 ph10 550 case OP_THEN_ARG:
502     cc += _pcre_OP_lengths[op] + cc[1+LINK_SIZE];
503     break;
504 ph10 654
505 ph10 603 /* The remaining opcodes are just skipped over. */
506 ph10 461
507 ph10 603 case OP_CLOSE:
508     case OP_COMMIT:
509     case OP_FAIL:
510     case OP_PRUNE:
511     case OP_SET_SOM:
512     case OP_SKIP:
513 ph10 605 case OP_THEN:
514 ph10 455 cc += _pcre_OP_lengths[op];
515     break;
516 ph10 605
517 ph10 603 /* This should not occur: we list all opcodes explicitly so that when
518     new ones get added they are properly considered. */
519    
520     default:
521     return -3;
522 ph10 455 }
523     }
524     /* Control never gets here */
525     }
526    
527    
528    
529     /*************************************************
530 nigel 77 * Set a bit and maybe its alternate case *
531     *************************************************/
532    
533 ph10 535 /* Given a character, set its first byte's bit in the table, and also the
534 ph10 520 corresponding bit for the other version of a letter if we are caseless. In
535     UTF-8 mode, for characters greater than 127, we can only do the caseless thing
536     when Unicode property support is available.
537 nigel 77
538     Arguments:
539     start_bits points to the bit map
540 ph10 520 p points to the character
541 nigel 77 caseless the caseless flag
542     cd the block with char table pointers
543 ph10 535 utf8 TRUE for UTF-8 mode
544 nigel 77
545 ph10 520 Returns: pointer after the character
546 nigel 77 */
547    
548 ph10 520 static const uschar *
549     set_table_bit(uschar *start_bits, const uschar *p, BOOL caseless,
550     compile_data *cd, BOOL utf8)
551 nigel 77 {
552 ph10 520 unsigned int c = *p;
553    
554 ph10 524 SET_BIT(c);
555    
556 ph10 520 #ifdef SUPPORT_UTF8
557     if (utf8 && c > 127)
558     {
559     GETCHARINC(c, p);
560     #ifdef SUPPORT_UCP
561     if (caseless)
562     {
563 ph10 535 uschar buff[8];
564 ph10 520 c = UCD_OTHERCASE(c);
565 ph10 535 (void)_pcre_ord2utf8(c, buff);
566     SET_BIT(buff[0]);
567     }
568     #endif
569 ph10 520 return p;
570     }
571 ph10 535 #endif
572 ph10 520
573     /* Not UTF-8 mode, or character is less than 127. */
574    
575 ph10 524 if (caseless && (cd->ctypes[c] & ctype_letter) != 0) SET_BIT(cd->fcc[c]);
576 ph10 520 return p + 1;
577 nigel 77 }
578    
579    
580    
581     /*************************************************
582 ph10 539 * Set bits for a positive character type *
583     *************************************************/
584    
585     /* This function sets starting bits for a character type. In UTF-8 mode, we can
586     only do a direct setting for bytes less than 128, as otherwise there can be
587     confusion with bytes in the middle of UTF-8 characters. In a "traditional"
588     environment, the tables will only recognize ASCII characters anyway, but in at
589     least one Windows environment, some higher bytes bits were set in the tables.
590     So we deal with that case by considering the UTF-8 encoding.
591    
592     Arguments:
593     start_bits the starting bitmap
594     cbit type the type of character wanted
595     table_limit 32 for non-UTF-8; 16 for UTF-8
596 ph10 545 cd the block with char table pointers
597 ph10 539
598     Returns: nothing
599     */
600    
601     static void
602     set_type_bits(uschar *start_bits, int cbit_type, int table_limit,
603     compile_data *cd)
604     {
605 ph10 545 register int c;
606 ph10 539 for (c = 0; c < table_limit; c++) start_bits[c] |= cd->cbits[c+cbit_type];
607     if (table_limit == 32) return;
608     for (c = 128; c < 256; c++)
609     {
610     if ((cd->cbits[c/8] & (1 << (c&7))) != 0)
611     {
612     uschar buff[8];
613     (void)_pcre_ord2utf8(c, buff);
614 ph10 545 SET_BIT(buff[0]);
615     }
616     }
617 ph10 539 }
618    
619    
620     /*************************************************
621     * Set bits for a negative character type *
622     *************************************************/
623    
624     /* This function sets starting bits for a negative character type such as \D.
625     In UTF-8 mode, we can only do a direct setting for bytes less than 128, as
626     otherwise there can be confusion with bytes in the middle of UTF-8 characters.
627 ph10 545 Unlike in the positive case, where we can set appropriate starting bits for
628 ph10 539 specific high-valued UTF-8 characters, in this case we have to set the bits for
629 ph10 545 all high-valued characters. The lowest is 0xc2, but we overkill by starting at
630 ph10 539 0xc0 (192) for simplicity.
631    
632     Arguments:
633     start_bits the starting bitmap
634     cbit type the type of character wanted
635     table_limit 32 for non-UTF-8; 16 for UTF-8
636 ph10 545 cd the block with char table pointers
637 ph10 539
638     Returns: nothing
639     */
640    
641     static void
642     set_nottype_bits(uschar *start_bits, int cbit_type, int table_limit,
643     compile_data *cd)
644     {
645 ph10 545 register int c;
646 ph10 539 for (c = 0; c < table_limit; c++) start_bits[c] |= ~cd->cbits[c+cbit_type];
647     if (table_limit != 32) for (c = 24; c < 32; c++) start_bits[c] = 0xff;
648     }
649    
650    
651    
652     /*************************************************
653 nigel 93 * Create bitmap of starting bytes *
654 nigel 77 *************************************************/
655    
656 nigel 93 /* This function scans a compiled unanchored expression recursively and
657     attempts to build a bitmap of the set of possible starting bytes. As time goes
658     by, we may be able to get more clever at doing this. The SSB_CONTINUE return is
659     useful for parenthesized groups in patterns such as (a*)b where the group
660     provides some optional starting bytes but scanning must continue at the outer
661     level to find at least one mandatory byte. At the outermost level, this
662     function fails unless the result is SSB_DONE.
663 nigel 77
664     Arguments:
665     code points to an expression
666     start_bits points to a 32-byte table, initialized to 0
667     utf8 TRUE if in UTF-8 mode
668     cd the block with char table pointers
669    
670 nigel 93 Returns: SSB_FAIL => Failed to find any starting bytes
671     SSB_DONE => Found mandatory starting bytes
672     SSB_CONTINUE => Found optional starting bytes
673 ph10 605 SSB_UNKNOWN => Hit an unrecognized opcode
674 nigel 77 */
675    
676 nigel 93 static int
677 ph10 605 set_start_bits(const uschar *code, uschar *start_bits, BOOL utf8,
678 ph10 603 compile_data *cd)
679 nigel 77 {
680     register int c;
681 nigel 93 int yield = SSB_DONE;
682 ph10 538 int table_limit = utf8? 16:32;
683 nigel 77
684 nigel 91 #if 0
685     /* ========================================================================= */
686     /* The following comment and code was inserted in January 1999. In May 2006,
687     when it was observed to cause compiler warnings about unused values, I took it
688     out again. If anybody is still using OS/2, they will have to put it back
689     manually. */
690    
691 nigel 77 /* This next statement and the later reference to dummy are here in order to
692     trick the optimizer of the IBM C compiler for OS/2 into generating correct
693     code. Apparently IBM isn't going to fix the problem, and we would rather not
694     disable optimization (in this module it actually makes a big difference, and
695     the pcre module can use all the optimization it can get). */
696    
697     volatile int dummy;
698 nigel 91 /* ========================================================================= */
699     #endif
700 nigel 77
701     do
702     {
703     BOOL try_next = TRUE;
704 ph10 604 const uschar *tcode = code + 1 + LINK_SIZE;
705 ph10 605
706 ph10 604 if (*code == OP_CBRA || *code == OP_SCBRA ||
707     *code == OP_CBRAPOS || *code == OP_SCBRAPOS) tcode += 2;
708 nigel 77
709 nigel 93 while (try_next) /* Loop for items in this branch */
710 nigel 77 {
711 nigel 93 int rc;
712 ph10 612
713 nigel 93 switch(*tcode)
714 nigel 77 {
715 ph10 605 /* If we reach something we don't understand, it means a new opcode has
716     been created that hasn't been added to this code. Hopefully this problem
717     will be discovered during testing. */
718 nigel 77
719     default:
720 ph10 605 return SSB_UNKNOWN;
721    
722     /* Fail for a valid opcode that implies no starting bits. */
723    
724     case OP_ACCEPT:
725 ph10 654 case OP_ASSERT_ACCEPT:
726 ph10 605 case OP_ALLANY:
727     case OP_ANY:
728     case OP_ANYBYTE:
729     case OP_CIRC:
730 ph10 654 case OP_CIRCM:
731 ph10 605 case OP_CLOSE:
732     case OP_COMMIT:
733     case OP_COND:
734 ph10 654 case OP_CREF:
735 ph10 605 case OP_DEF:
736     case OP_DOLL:
737 ph10 654 case OP_DOLLM:
738 ph10 605 case OP_END:
739     case OP_EOD:
740 ph10 654 case OP_EODN:
741 ph10 605 case OP_EXTUNI:
742     case OP_FAIL:
743     case OP_MARK:
744     case OP_NCREF:
745     case OP_NOT:
746     case OP_NOTEXACT:
747     case OP_NOTEXACTI:
748 ph10 654 case OP_NOTI:
749 ph10 605 case OP_NOTMINPLUS:
750     case OP_NOTMINPLUSI:
751     case OP_NOTMINQUERY:
752     case OP_NOTMINQUERYI:
753     case OP_NOTMINSTAR:
754     case OP_NOTMINSTARI:
755     case OP_NOTMINUPTO:
756     case OP_NOTMINUPTOI:
757     case OP_NOTPLUS:
758     case OP_NOTPLUSI:
759     case OP_NOTPOSPLUS:
760     case OP_NOTPOSPLUSI:
761     case OP_NOTPOSQUERY:
762     case OP_NOTPOSQUERYI:
763     case OP_NOTPOSSTAR:
764     case OP_NOTPOSSTARI:
765     case OP_NOTPOSUPTO:
766     case OP_NOTPOSUPTOI:
767     case OP_NOTPROP:
768     case OP_NOTQUERY:
769     case OP_NOTQUERYI:
770     case OP_NOTSTAR:
771     case OP_NOTSTARI:
772     case OP_NOTUPTO:
773     case OP_NOTUPTOI:
774     case OP_NOT_HSPACE:
775     case OP_NOT_VSPACE:
776     case OP_NRREF:
777     case OP_PROP:
778     case OP_PRUNE:
779     case OP_PRUNE_ARG:
780     case OP_RECURSE:
781     case OP_REF:
782     case OP_REFI:
783     case OP_REVERSE:
784     case OP_RREF:
785 ph10 654 case OP_SCOND:
786 ph10 605 case OP_SET_SOM:
787     case OP_SKIP:
788     case OP_SKIP_ARG:
789     case OP_SOD:
790     case OP_SOM:
791     case OP_THEN:
792     case OP_THEN_ARG:
793     case OP_XCLASS:
794 nigel 93 return SSB_FAIL;
795 nigel 77
796 ph10 657 /* We can ignore word boundary tests. */
797 ph10 659
798 ph10 657 case OP_WORD_BOUNDARY:
799     case OP_NOT_WORD_BOUNDARY:
800 ph10 659 tcode++;
801     break;
802 ph10 657
803 nigel 93 /* If we hit a bracket or a positive lookahead assertion, recurse to set
804     bits from within the subpattern. If it can't find anything, we have to
805     give up. If it finds some mandatory character(s), we are done for this
806     branch. Otherwise, carry on scanning after the subpattern. */
807    
808     case OP_BRA:
809     case OP_SBRA:
810     case OP_CBRA:
811     case OP_SCBRA:
812 ph10 604 case OP_BRAPOS:
813     case OP_SBRAPOS:
814     case OP_CBRAPOS:
815     case OP_SCBRAPOS:
816 nigel 93 case OP_ONCE:
817     case OP_ASSERT:
818 ph10 603 rc = set_start_bits(tcode, start_bits, utf8, cd);
819 ph10 605 if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
820 nigel 93 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 ph10 605 case OP_KETRPOS:
843 nigel 93 return SSB_CONTINUE;
844    
845 nigel 77 /* 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 nigel 93 tcode += 1 + LINK_SIZE;
858 nigel 77 break;
859    
860     /* BRAZERO does the bracket, but carries on. */
861    
862     case OP_BRAZERO:
863     case OP_BRAMINZERO:
864 ph10 604 case OP_BRAPOSZERO:
865 ph10 605 rc = set_start_bits(++tcode, start_bits, utf8, cd);
866     if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
867 nigel 91 /* =========================================================================
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 nigel 77 dummy = 1;
871 nigel 91 ========================================================================= */
872 nigel 77 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
873 nigel 93 tcode += 1 + LINK_SIZE;
874 nigel 77 break;
875    
876 ph10 335 /* SKIPZERO skips the bracket. */
877    
878     case OP_SKIPZERO:
879 ph10 358 tcode++;
880 ph10 335 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
881     tcode += 1 + LINK_SIZE;
882     break;
883    
884 nigel 77 /* Single-char * or ? sets the bit and tries the next item */
885    
886     case OP_STAR:
887     case OP_MINSTAR:
888 nigel 93 case OP_POSSTAR:
889 nigel 77 case OP_QUERY:
890     case OP_MINQUERY:
891 nigel 93 case OP_POSQUERY:
892 ph10 603 tcode = set_table_bit(start_bits, tcode + 1, FALSE, cd, utf8);
893 nigel 77 break;
894    
895 ph10 602 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 nigel 77 /* Single-char upto sets the bit and tries the next */
905    
906     case OP_UPTO:
907     case OP_MINUPTO:
908 nigel 93 case OP_POSUPTO:
909 ph10 603 tcode = set_table_bit(start_bits, tcode + 3, FALSE, cd, utf8);
910 nigel 77 break;
911    
912 ph10 602 case OP_UPTOI:
913     case OP_MINUPTOI:
914     case OP_POSUPTOI:
915     tcode = set_table_bit(start_bits, tcode + 3, TRUE, cd, utf8);
916     break;
917    
918 nigel 77 /* At least one single char sets the bit and stops */
919    
920 ph10 602 case OP_EXACT:
921 nigel 77 tcode += 2;
922 ph10 602 /* Fall through */
923 nigel 77 case OP_CHAR:
924     case OP_PLUS:
925     case OP_MINPLUS:
926 nigel 93 case OP_POSPLUS:
927 ph10 603 (void)set_table_bit(start_bits, tcode + 1, FALSE, cd, utf8);
928 nigel 77 try_next = FALSE;
929     break;
930 ph10 535
931 ph10 603 case OP_EXACTI:
932     tcode += 2;
933     /* Fall through */
934 ph10 602 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 ph10 535 /* 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 ph10 524 identical. */
947 ph10 535
948 ph10 524 case OP_HSPACE:
949     SET_BIT(0x09);
950     SET_BIT(0x20);
951     if (utf8)
952 ph10 535 {
953 ph10 545 SET_BIT(0xC2); /* For U+00A0 */
954 ph10 524 SET_BIT(0xE1); /* For U+1680, U+180E */
955     SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
956 ph10 535 SET_BIT(0xE3); /* For U+3000 */
957 ph10 524 }
958 ph10 538 else SET_BIT(0xA0);
959 ph10 524 try_next = FALSE;
960 ph10 535 break;
961 nigel 77
962 ph10 535 case OP_ANYNL:
963 ph10 524 case OP_VSPACE:
964 ph10 535 SET_BIT(0x0A);
965     SET_BIT(0x0B);
966     SET_BIT(0x0C);
967     SET_BIT(0x0D);
968 ph10 545 if (utf8)
969     {
970     SET_BIT(0xC2); /* For U+0085 */
971 ph10 538 SET_BIT(0xE2); /* For U+2028, U+2029 */
972 ph10 545 }
973 ph10 538 else SET_BIT(0x85);
974 ph10 524 try_next = FALSE;
975 ph10 535 break;
976 ph10 524
977 ph10 535 /* 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 ph10 545 properties. Therefore, these apply in the case when only characters less
980 ph10 539 than 256 are recognized to match the types. */
981 nigel 77
982     case OP_NOT_DIGIT:
983 ph10 539 set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
984 nigel 77 try_next = FALSE;
985     break;
986    
987     case OP_DIGIT:
988 ph10 539 set_type_bits(start_bits, cbit_digit, table_limit, cd);
989 nigel 77 try_next = FALSE;
990     break;
991    
992 nigel 91 /* The cbit_space table has vertical tab as whitespace; we have to
993 ph10 539 ensure it is set as not whitespace. */
994 nigel 91
995 nigel 77 case OP_NOT_WHITESPACE:
996 ph10 539 set_nottype_bits(start_bits, cbit_space, table_limit, cd);
997     start_bits[1] |= 0x08;
998 nigel 77 try_next = FALSE;
999     break;
1000    
1001 nigel 91 /* The cbit_space table has vertical tab as whitespace; we have to
1002 ph10 539 not set it from the table. */
1003 nigel 91
1004 nigel 77 case OP_WHITESPACE:
1005 ph10 539 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 nigel 77 try_next = FALSE;
1009     break;
1010    
1011     case OP_NOT_WORDCHAR:
1012 ph10 539 set_nottype_bits(start_bits, cbit_word, table_limit, cd);
1013 nigel 77 try_next = FALSE;
1014     break;
1015    
1016     case OP_WORDCHAR:
1017 ph10 539 set_type_bits(start_bits, cbit_word, table_limit, cd);
1018 nigel 77 try_next = FALSE;
1019     break;
1020 ph10 545
1021 nigel 77 /* 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 ph10 535 case OP_TYPEPOSPLUS:
1027 nigel 77 tcode++;
1028     break;
1029    
1030     case OP_TYPEEXACT:
1031     tcode += 3;
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 nigel 93 case OP_TYPEPOSUPTO:
1040 nigel 77 tcode += 2; /* Fall through */
1041    
1042     case OP_TYPESTAR:
1043     case OP_TYPEMINSTAR:
1044 nigel 93 case OP_TYPEPOSSTAR:
1045 nigel 77 case OP_TYPEQUERY:
1046     case OP_TYPEMINQUERY:
1047 nigel 93 case OP_TYPEPOSQUERY:
1048 nigel 77 switch(tcode[1])
1049     {
1050 ph10 523 default:
1051 nigel 77 case OP_ANY:
1052 ph10 345 case OP_ALLANY:
1053 nigel 93 return SSB_FAIL;
1054 ph10 535
1055 ph10 524 case OP_HSPACE:
1056     SET_BIT(0x09);
1057     SET_BIT(0x20);
1058     if (utf8)
1059 ph10 535 {
1060 ph10 545 SET_BIT(0xC2); /* For U+00A0 */
1061 ph10 524 SET_BIT(0xE1); /* For U+1680, U+180E */
1062     SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
1063 ph10 535 SET_BIT(0xE3); /* For U+3000 */
1064 ph10 524 }
1065 ph10 538 else SET_BIT(0xA0);
1066 ph10 535 break;
1067    
1068     case OP_ANYNL:
1069 ph10 524 case OP_VSPACE:
1070 ph10 535 SET_BIT(0x0A);
1071     SET_BIT(0x0B);
1072     SET_BIT(0x0C);
1073     SET_BIT(0x0D);
1074 ph10 545 if (utf8)
1075 ph10 538 {
1076 ph10 545 SET_BIT(0xC2); /* For U+0085 */
1077 ph10 538 SET_BIT(0xE2); /* For U+2028, U+2029 */
1078 ph10 545 }
1079 ph10 538 else SET_BIT(0x85);
1080 ph10 535 break;
1081    
1082 nigel 77 case OP_NOT_DIGIT:
1083 ph10 539 set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
1084 nigel 77 break;
1085    
1086     case OP_DIGIT:
1087 ph10 539 set_type_bits(start_bits, cbit_digit, table_limit, cd);
1088 nigel 77 break;
1089    
1090 nigel 91 /* The cbit_space table has vertical tab as whitespace; we have to
1091 ph10 539 ensure it gets set as not whitespace. */
1092 nigel 91
1093 nigel 77 case OP_NOT_WHITESPACE:
1094 ph10 539 set_nottype_bits(start_bits, cbit_space, table_limit, cd);
1095 ph10 545 start_bits[1] |= 0x08;
1096 nigel 77 break;
1097    
1098 nigel 91 /* The cbit_space table has vertical tab as whitespace; we have to
1099 ph10 539 avoid setting it. */
1100 nigel 91
1101 nigel 77 case OP_WHITESPACE:
1102 ph10 539 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 nigel 77 break;
1106    
1107     case OP_NOT_WORDCHAR:
1108 ph10 539 set_nottype_bits(start_bits, cbit_word, table_limit, cd);
1109 nigel 77 break;
1110    
1111     case OP_WORDCHAR:
1112 ph10 539 set_type_bits(start_bits, cbit_word, table_limit, cd);
1113 nigel 77 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 ph10 111 #ifdef SUPPORT_UTF8
1127 nigel 77 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 ph10 111 #endif
1133 nigel 77 /* 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 ph10 107 #ifdef SUPPORT_UTF8
1146 nigel 77 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 ph10 111 #endif
1164 nigel 77 {
1165     for (c = 0; c < 32; c++) start_bits[c] |= tcode[c];
1166     }
1167    
1168 ph10 654 /* Advance past the bit map, and act on what follows. For a zero
1169 ph10 605 minimum repeat, continue; otherwise stop processing. */
1170 nigel 77
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 (((tcode[1] << 8) + tcode[2]) == 0) tcode += 5;
1184     else try_next = FALSE;
1185     break;
1186 ph10 654
1187 nigel 77 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 nigel 93 return yield;
1201 nigel 77 }
1202    
1203    
1204    
1205 ph10 605
1206    
1207 nigel 77 /*************************************************
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 ph10 455 appropriate flags set;
1223 nigel 77 NULL on error or if no optimization possible
1224     */
1225    
1226 ph10 359 PCRE_EXP_DEFN pcre_extra * PCRE_CALL_CONVENTION
1227 nigel 77 pcre_study(const pcre *external_re, int options, const char **errorptr)
1228     {
1229 ph10 455 int min;
1230     BOOL bits_set = FALSE;
1231 ph10 612 BOOL had_accept = FALSE;
1232 nigel 77 uschar start_bits[32];
1233     pcre_extra *extra;
1234     pcre_study_data *study;
1235     const uschar *tables;
1236 nigel 91 uschar *code;
1237     compile_data compile_block;
1238 nigel 77 const real_pcre *re = (const real_pcre *)external_re;
1239    
1240     *errorptr = NULL;
1241    
1242     if (re == NULL || re->magic_number != MAGIC_NUMBER)
1243     {
1244     *errorptr = "argument is not a compiled regular expression";
1245     return NULL;
1246     }
1247    
1248     if ((options & ~PUBLIC_STUDY_OPTIONS) != 0)
1249     {
1250     *errorptr = "unknown or incorrect option bit(s) set";
1251     return NULL;
1252     }
1253    
1254 nigel 91 code = (uschar *)re + re->name_table_offset +
1255     (re->name_count * re->name_entry_size);
1256    
1257 nigel 77 /* For an anchored pattern, or an unanchored pattern that has a first char, or
1258 ph10 461 a multiline pattern that matches only at "line starts", there is no point in
1259 ph10 455 seeking a list of starting bytes. */
1260 nigel 77
1261 ph10 455 if ((re->options & PCRE_ANCHORED) == 0 &&
1262     (re->flags & (PCRE_FIRSTSET|PCRE_STARTLINE)) == 0)
1263     {
1264 ph10 605 int rc;
1265    
1266 ph10 455 /* Set the character tables in the block that is passed around */
1267 ph10 461
1268 ph10 455 tables = re->tables;
1269     if (tables == NULL)
1270     (void)pcre_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES,
1271     (void *)(&tables));
1272 ph10 461
1273 ph10 455 compile_block.lcc = tables + lcc_offset;
1274     compile_block.fcc = tables + fcc_offset;
1275     compile_block.cbits = tables + cbits_offset;
1276     compile_block.ctypes = tables + ctypes_offset;
1277 ph10 461
1278 ph10 455 /* See if we can find a fixed set of initial characters for the pattern. */
1279 ph10 461
1280 ph10 455 memset(start_bits, 0, 32 * sizeof(uschar));
1281 ph10 605 rc = set_start_bits(code, start_bits, (re->options & PCRE_UTF8) != 0,
1282     &compile_block);
1283     bits_set = rc == SSB_DONE;
1284     if (rc == SSB_UNKNOWN) *errorptr = "internal error: opcode not recognized";
1285 ph10 455 }
1286 ph10 461
1287 ph10 455 /* Find the minimum length of subject string. */
1288 nigel 77
1289 ph10 620 switch(min = find_minlength(code, code, re->options, &had_accept, 0))
1290 ph10 603 {
1291     case -2: *errorptr = "internal error: missing capturing bracket"; break;
1292 ph10 605 case -3: *errorptr = "internal error: opcode not recognized"; break;
1293 ph10 603 default: break;
1294 ph10 605 }
1295 nigel 77
1296 ph10 605 /* Return NULL if there's been an error or if no optimization is possible. */
1297 nigel 77
1298 ph10 605 if (*errorptr != NULL || (!bits_set && min < 0)) return NULL;
1299 nigel 77
1300     /* Get a pcre_extra block and a pcre_study_data block. The study data is put in
1301     the latter, which is pointed to by the former, which may also get additional
1302     data set later by the calling program. At the moment, the size of
1303     pcre_study_data is fixed. We nevertheless save it in a field for returning via
1304     the pcre_fullinfo() function so that if it becomes variable in the future, we
1305     don't have to change that code. */
1306    
1307     extra = (pcre_extra *)(pcre_malloc)
1308     (sizeof(pcre_extra) + sizeof(pcre_study_data));
1309    
1310     if (extra == NULL)
1311     {
1312     *errorptr = "failed to get memory";
1313     return NULL;
1314     }
1315    
1316     study = (pcre_study_data *)((char *)extra + sizeof(pcre_extra));
1317     extra->flags = PCRE_EXTRA_STUDY_DATA;
1318     extra->study_data = study;
1319    
1320     study->size = sizeof(pcre_study_data);
1321 ph10 455 study->flags = 0;
1322 nigel 77
1323 ph10 455 if (bits_set)
1324     {
1325     study->flags |= PCRE_STUDY_MAPPED;
1326     memcpy(study->start_bits, start_bits, sizeof(start_bits));
1327     }
1328 ph10 461
1329 ph10 455 if (min >= 0)
1330     {
1331     study->flags |= PCRE_STUDY_MINLEN;
1332     study->minlength = min;
1333 ph10 461 }
1334 ph10 455
1335 nigel 77 return extra;
1336     }
1337    
1338     /* 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