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

Contents of /code/trunk/pcre_study.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 475 - (hide annotations) (download)
Sat Jan 2 18:21:30 2010 UTC (4 years, 11 months ago) by ph10
File MIME type: text/plain
File size: 27445 byte(s)
Tidies to allow easier embedded compilation; avoid (double) where possible.

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    
52 nigel 93 /* Returns from set_start_bits() */
53    
54     enum { SSB_FAIL, SSB_DONE, SSB_CONTINUE };
55    
56    
57 ph10 455
58 nigel 77 /*************************************************
59 ph10 455 * Find the minimum subject length for a group *
60     *************************************************/
61    
62     /* Scan a parenthesized group and compute the minimum length of subject that
63 ph10 461 is needed to match it. This is a lower bound; it does not mean there is a
64 ph10 455 string of that length that matches. In UTF8 mode, the result is in characters
65     rather than bytes.
66    
67     Arguments:
68     code pointer to start of group (the bracket)
69     startcode pointer to start of the whole pattern
70 ph10 461 options the compiling options
71 ph10 455
72     Returns: the minimum length
73     -1 if \C was encountered
74 ph10 461 -2 internal error (missing capturing bracket)
75 ph10 455 */
76    
77     static int
78     find_minlength(const uschar *code, const uschar *startcode, int options)
79     {
80     int length = -1;
81     BOOL utf8 = (options & PCRE_UTF8) != 0;
82     BOOL had_recurse = FALSE;
83     register int branchlength = 0;
84     register uschar *cc = (uschar *)code + 1 + LINK_SIZE;
85    
86     if (*code == OP_CBRA || *code == OP_SCBRA) cc += 2;
87    
88     /* Scan along the opcodes for this branch. If we get to the end of the
89     branch, check the length against that of the other branches. */
90    
91     for (;;)
92     {
93     int d, min;
94 ph10 461 uschar *cs, *ce;
95 ph10 455 register int op = *cc;
96 ph10 461
97 ph10 455 switch (op)
98     {
99 ph10 471 case OP_COND:
100     case OP_SCOND:
101    
102     /* If there is only one branch in a condition, the implied branch has zero
103     length, so we don't add anything. This covers the DEFINE "condition"
104     automatically. */
105    
106     cs = cc + GET(cc, 1);
107     if (*cs != OP_ALT)
108     {
109     cc = cs + 1 + LINK_SIZE;
110     break;
111     }
112    
113     /* Otherwise we can fall through and treat it the same as any other
114     subpattern. */
115    
116 ph10 455 case OP_CBRA:
117 ph10 461 case OP_SCBRA:
118 ph10 455 case OP_BRA:
119 ph10 461 case OP_SBRA:
120 ph10 455 case OP_ONCE:
121     d = find_minlength(cc, startcode, options);
122     if (d < 0) return d;
123     branchlength += d;
124     do cc += GET(cc, 1); while (*cc == OP_ALT);
125     cc += 1 + LINK_SIZE;
126     break;
127    
128     /* Reached end of a branch; if it's a ket it is the end of a nested
129     call. If it's ALT it is an alternation in a nested call. If it is
130     END it's the end of the outer call. All can be handled by the same code. */
131    
132     case OP_ALT:
133     case OP_KET:
134     case OP_KETRMAX:
135     case OP_KETRMIN:
136     case OP_END:
137 ph10 461 if (length < 0 || (!had_recurse && branchlength < length))
138 ph10 455 length = branchlength;
139     if (*cc != OP_ALT) return length;
140     cc += 1 + LINK_SIZE;
141     branchlength = 0;
142 ph10 461 had_recurse = FALSE;
143 ph10 455 break;
144    
145     /* Skip over assertive subpatterns */
146    
147     case OP_ASSERT:
148     case OP_ASSERT_NOT:
149     case OP_ASSERTBACK:
150     case OP_ASSERTBACK_NOT:
151     do cc += GET(cc, 1); while (*cc == OP_ALT);
152     /* Fall through */
153    
154     /* Skip over things that don't match chars */
155    
156     case OP_REVERSE:
157     case OP_CREF:
158 ph10 459 case OP_NCREF:
159 ph10 455 case OP_RREF:
160 ph10 459 case OP_NRREF:
161 ph10 455 case OP_DEF:
162     case OP_OPT:
163     case OP_CALLOUT:
164     case OP_SOD:
165     case OP_SOM:
166     case OP_EOD:
167     case OP_EODN:
168     case OP_CIRC:
169     case OP_DOLL:
170     case OP_NOT_WORD_BOUNDARY:
171     case OP_WORD_BOUNDARY:
172     cc += _pcre_OP_lengths[*cc];
173     break;
174 ph10 461
175 ph10 455 /* Skip over a subpattern that has a {0} or {0,x} quantifier */
176    
177     case OP_BRAZERO:
178 ph10 461 case OP_BRAMINZERO:
179 ph10 455 case OP_SKIPZERO:
180     cc += _pcre_OP_lengths[*cc];
181     do cc += GET(cc, 1); while (*cc == OP_ALT);
182     cc += 1 + LINK_SIZE;
183     break;
184    
185     /* Handle literal characters and + repetitions */
186    
187     case OP_CHAR:
188     case OP_CHARNC:
189     case OP_NOT:
190     case OP_PLUS:
191     case OP_MINPLUS:
192     case OP_POSPLUS:
193     case OP_NOTPLUS:
194     case OP_NOTMINPLUS:
195     case OP_NOTPOSPLUS:
196     branchlength++;
197     cc += 2;
198     #ifdef SUPPORT_UTF8
199     if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
200     #endif
201     break;
202 ph10 461
203 ph10 455 case OP_TYPEPLUS:
204     case OP_TYPEMINPLUS:
205 ph10 461 case OP_TYPEPOSPLUS:
206 ph10 455 branchlength++;
207     cc += (cc[1] == OP_PROP || cc[1] == OP_NOTPROP)? 4 : 2;
208     break;
209    
210     /* Handle exact repetitions. The count is already in characters, but we
211     need to skip over a multibyte character in UTF8 mode. */
212    
213     case OP_EXACT:
214 ph10 461 case OP_NOTEXACT:
215 ph10 455 branchlength += GET2(cc,1);
216     cc += 4;
217     #ifdef SUPPORT_UTF8
218     if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
219     #endif
220     break;
221    
222     case OP_TYPEEXACT:
223     branchlength += GET2(cc,1);
224     cc += (cc[3] == OP_PROP || cc[3] == OP_NOTPROP)? 6 : 4;
225     break;
226    
227     /* Handle single-char non-literal matchers */
228    
229     case OP_PROP:
230     case OP_NOTPROP:
231     cc += 2;
232     /* Fall through */
233    
234     case OP_NOT_DIGIT:
235     case OP_DIGIT:
236     case OP_NOT_WHITESPACE:
237     case OP_WHITESPACE:
238     case OP_NOT_WORDCHAR:
239     case OP_WORDCHAR:
240     case OP_ANY:
241     case OP_ALLANY:
242     case OP_EXTUNI:
243 ph10 461 case OP_HSPACE:
244 ph10 455 case OP_NOT_HSPACE:
245     case OP_VSPACE:
246 ph10 461 case OP_NOT_VSPACE:
247 ph10 455 branchlength++;
248     cc++;
249     break;
250 ph10 461
251 ph10 455 /* "Any newline" might match two characters */
252 ph10 461
253 ph10 455 case OP_ANYNL:
254     branchlength += 2;
255     cc++;
256 ph10 461 break;
257 ph10 455
258     /* The single-byte matcher means we can't proceed in UTF-8 mode */
259    
260     case OP_ANYBYTE:
261     #ifdef SUPPORT_UTF8
262     if (utf8) return -1;
263     #endif
264     branchlength++;
265     cc++;
266 ph10 461 break;
267 ph10 455
268     /* For repeated character types, we have to test for \p and \P, which have
269     an extra two bytes of parameters. */
270    
271     case OP_TYPESTAR:
272     case OP_TYPEMINSTAR:
273     case OP_TYPEQUERY:
274     case OP_TYPEMINQUERY:
275     case OP_TYPEPOSSTAR:
276     case OP_TYPEPOSQUERY:
277     if (cc[1] == OP_PROP || cc[1] == OP_NOTPROP) cc += 2;
278     cc += _pcre_OP_lengths[op];
279     break;
280    
281     case OP_TYPEUPTO:
282     case OP_TYPEMINUPTO:
283     case OP_TYPEPOSUPTO:
284     if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
285     cc += _pcre_OP_lengths[op];
286     break;
287    
288     /* Check a class for variable quantification */
289    
290     #ifdef SUPPORT_UTF8
291     case OP_XCLASS:
292     cc += GET(cc, 1) - 33;
293     /* Fall through */
294     #endif
295    
296     case OP_CLASS:
297     case OP_NCLASS:
298     cc += 33;
299    
300     switch (*cc)
301     {
302     case OP_CRPLUS:
303     case OP_CRMINPLUS:
304     branchlength++;
305 ph10 461 /* Fall through */
306 ph10 455
307     case OP_CRSTAR:
308     case OP_CRMINSTAR:
309     case OP_CRQUERY:
310     case OP_CRMINQUERY:
311 ph10 461 cc++;
312 ph10 455 break;
313 ph10 461
314 ph10 455 case OP_CRRANGE:
315     case OP_CRMINRANGE:
316     branchlength += GET2(cc,1);
317     cc += 5;
318     break;
319 ph10 461
320 ph10 455 default:
321     branchlength++;
322 ph10 461 break;
323 ph10 455 }
324     break;
325 ph10 461
326     /* Backreferences and subroutine calls are treated in the same way: we find
327     the minimum length for the subpattern. A recursion, however, causes an
328 ph10 455 a flag to be set that causes the length of this branch to be ignored. The
329     logic is that a recursion can only make sense if there is another
330     alternation that stops the recursing. That will provide the minimum length
331     (when no recursion happens). A backreference within the group that it is
332 ph10 469 referencing behaves in the same way.
333    
334 ph10 467 If PCRE_JAVASCRIPT_COMPAT is set, a backreference to an unset bracket
335     matches an empty string (by default it causes a matching failure), so in
336     that case we must set the minimum length to zero. */
337 ph10 461
338 ph10 455 case OP_REF:
339 ph10 467 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0)
340 ph10 469 {
341 ph10 467 ce = cs = (uschar *)_pcre_find_bracket(startcode, utf8, GET2(cc, 1));
342     if (cs == NULL) return -2;
343     do ce += GET(ce, 1); while (*ce == OP_ALT);
344     if (cc > cs && cc < ce)
345     {
346     d = 0;
347     had_recurse = TRUE;
348     }
349     else d = find_minlength(cs, startcode, options);
350 ph10 461 }
351 ph10 469 else d = 0;
352 ph10 461 cc += 3;
353 ph10 455
354     /* Handle repeated back references */
355 ph10 461
356 ph10 455 switch (*cc)
357     {
358     case OP_CRSTAR:
359     case OP_CRMINSTAR:
360     case OP_CRQUERY:
361     case OP_CRMINQUERY:
362     min = 0;
363     cc++;
364     break;
365 ph10 461
366 ph10 455 case OP_CRRANGE:
367     case OP_CRMINRANGE:
368     min = GET2(cc, 1);
369     cc += 5;
370     break;
371 ph10 461
372 ph10 455 default:
373     min = 1;
374     break;
375     }
376    
377     branchlength += min * d;
378 ph10 461 break;
379 ph10 455
380 ph10 461 case OP_RECURSE:
381 ph10 455 cs = ce = (uschar *)startcode + GET(cc, 1);
382     if (cs == NULL) return -2;
383     do ce += GET(ce, 1); while (*ce == OP_ALT);
384     if (cc > cs && cc < ce)
385 ph10 461 had_recurse = TRUE;
386     else
387 ph10 455 branchlength += find_minlength(cs, startcode, options);
388     cc += 1 + LINK_SIZE;
389     break;
390    
391     /* Anything else does not or need not match a character. We can get the
392 ph10 461 item's length from the table, but for those that can match zero occurrences
393     of a character, we must take special action for UTF-8 characters. */
394    
395 ph10 455 case OP_UPTO:
396 ph10 461 case OP_NOTUPTO:
397 ph10 455 case OP_MINUPTO:
398 ph10 461 case OP_NOTMINUPTO:
399 ph10 455 case OP_POSUPTO:
400     case OP_STAR:
401     case OP_MINSTAR:
402 ph10 461 case OP_NOTMINSTAR:
403 ph10 455 case OP_POSSTAR:
404 ph10 461 case OP_NOTPOSSTAR:
405 ph10 455 case OP_QUERY:
406     case OP_MINQUERY:
407     case OP_NOTMINQUERY:
408     case OP_POSQUERY:
409 ph10 461 case OP_NOTPOSQUERY:
410 ph10 455 cc += _pcre_OP_lengths[op];
411 ph10 461 #ifdef SUPPORT_UTF8
412 ph10 455 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
413 ph10 461 #endif
414 ph10 455 break;
415    
416     /* For the record, these are the opcodes that are matched by "default":
417     OP_ACCEPT, OP_CLOSE, OP_COMMIT, OP_FAIL, OP_PRUNE, OP_SET_SOM, OP_SKIP,
418     OP_THEN. */
419 ph10 461
420 ph10 455 default:
421     cc += _pcre_OP_lengths[op];
422     break;
423     }
424     }
425     /* Control never gets here */
426     }
427    
428    
429    
430     /*************************************************
431 nigel 77 * Set a bit and maybe its alternate case *
432     *************************************************/
433    
434     /* Given a character, set its bit in the table, and also the bit for the other
435     version of a letter if we are caseless.
436    
437     Arguments:
438     start_bits points to the bit map
439     c is the character
440     caseless the caseless flag
441     cd the block with char table pointers
442    
443     Returns: nothing
444     */
445    
446     static void
447 ph10 475 set_table_bit(uschar *start_bits, unsigned int c, BOOL caseless,
448     compile_data *cd)
449 nigel 77 {
450     start_bits[c/8] |= (1 << (c&7));
451     if (caseless && (cd->ctypes[c] & ctype_letter) != 0)
452     start_bits[cd->fcc[c]/8] |= (1 << (cd->fcc[c]&7));
453     }
454    
455    
456    
457     /*************************************************
458 nigel 93 * Create bitmap of starting bytes *
459 nigel 77 *************************************************/
460    
461 nigel 93 /* This function scans a compiled unanchored expression recursively and
462     attempts to build a bitmap of the set of possible starting bytes. As time goes
463     by, we may be able to get more clever at doing this. The SSB_CONTINUE return is
464     useful for parenthesized groups in patterns such as (a*)b where the group
465     provides some optional starting bytes but scanning must continue at the outer
466     level to find at least one mandatory byte. At the outermost level, this
467     function fails unless the result is SSB_DONE.
468 nigel 77
469     Arguments:
470     code points to an expression
471     start_bits points to a 32-byte table, initialized to 0
472     caseless the current state of the caseless flag
473     utf8 TRUE if in UTF-8 mode
474     cd the block with char table pointers
475    
476 nigel 93 Returns: SSB_FAIL => Failed to find any starting bytes
477     SSB_DONE => Found mandatory starting bytes
478     SSB_CONTINUE => Found optional starting bytes
479 nigel 77 */
480    
481 nigel 93 static int
482 nigel 77 set_start_bits(const uschar *code, uschar *start_bits, BOOL caseless,
483     BOOL utf8, compile_data *cd)
484     {
485     register int c;
486 nigel 93 int yield = SSB_DONE;
487 nigel 77
488 nigel 91 #if 0
489     /* ========================================================================= */
490     /* The following comment and code was inserted in January 1999. In May 2006,
491     when it was observed to cause compiler warnings about unused values, I took it
492     out again. If anybody is still using OS/2, they will have to put it back
493     manually. */
494    
495 nigel 77 /* This next statement and the later reference to dummy are here in order to
496     trick the optimizer of the IBM C compiler for OS/2 into generating correct
497     code. Apparently IBM isn't going to fix the problem, and we would rather not
498     disable optimization (in this module it actually makes a big difference, and
499     the pcre module can use all the optimization it can get). */
500    
501     volatile int dummy;
502 nigel 91 /* ========================================================================= */
503     #endif
504 nigel 77
505     do
506     {
507 nigel 93 const uschar *tcode = code + (((int)*code == OP_CBRA)? 3:1) + LINK_SIZE;
508 nigel 77 BOOL try_next = TRUE;
509    
510 nigel 93 while (try_next) /* Loop for items in this branch */
511 nigel 77 {
512 nigel 93 int rc;
513     switch(*tcode)
514 nigel 77 {
515 nigel 93 /* Fail if we reach something we don't understand */
516 nigel 77
517     default:
518 nigel 93 return SSB_FAIL;
519 nigel 77
520 nigel 93 /* If we hit a bracket or a positive lookahead assertion, recurse to set
521     bits from within the subpattern. If it can't find anything, we have to
522     give up. If it finds some mandatory character(s), we are done for this
523     branch. Otherwise, carry on scanning after the subpattern. */
524    
525     case OP_BRA:
526     case OP_SBRA:
527     case OP_CBRA:
528     case OP_SCBRA:
529     case OP_ONCE:
530     case OP_ASSERT:
531     rc = set_start_bits(tcode, start_bits, caseless, utf8, cd);
532     if (rc == SSB_FAIL) return SSB_FAIL;
533     if (rc == SSB_DONE) try_next = FALSE; else
534     {
535     do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
536     tcode += 1 + LINK_SIZE;
537     }
538     break;
539    
540     /* If we hit ALT or KET, it means we haven't found anything mandatory in
541     this branch, though we might have found something optional. For ALT, we
542     continue with the next alternative, but we have to arrange that the final
543     result from subpattern is SSB_CONTINUE rather than SSB_DONE. For KET,
544     return SSB_CONTINUE: if this is the top level, that indicates failure,
545     but after a nested subpattern, it causes scanning to continue. */
546    
547     case OP_ALT:
548     yield = SSB_CONTINUE;
549     try_next = FALSE;
550     break;
551    
552     case OP_KET:
553     case OP_KETRMAX:
554     case OP_KETRMIN:
555     return SSB_CONTINUE;
556    
557 nigel 77 /* Skip over callout */
558    
559     case OP_CALLOUT:
560     tcode += 2 + 2*LINK_SIZE;
561     break;
562    
563     /* Skip over lookbehind and negative lookahead assertions */
564    
565     case OP_ASSERT_NOT:
566     case OP_ASSERTBACK:
567     case OP_ASSERTBACK_NOT:
568     do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
569 nigel 93 tcode += 1 + LINK_SIZE;
570 nigel 77 break;
571    
572     /* Skip over an option setting, changing the caseless flag */
573    
574     case OP_OPT:
575     caseless = (tcode[1] & PCRE_CASELESS) != 0;
576     tcode += 2;
577     break;
578    
579     /* BRAZERO does the bracket, but carries on. */
580    
581     case OP_BRAZERO:
582     case OP_BRAMINZERO:
583 nigel 93 if (set_start_bits(++tcode, start_bits, caseless, utf8, cd) == SSB_FAIL)
584     return SSB_FAIL;
585 nigel 91 /* =========================================================================
586     See the comment at the head of this function concerning the next line,
587     which was an old fudge for the benefit of OS/2.
588 nigel 77 dummy = 1;
589 nigel 91 ========================================================================= */
590 nigel 77 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
591 nigel 93 tcode += 1 + LINK_SIZE;
592 nigel 77 break;
593    
594 ph10 335 /* SKIPZERO skips the bracket. */
595    
596     case OP_SKIPZERO:
597 ph10 358 tcode++;
598 ph10 335 do tcode += GET(tcode,1); while (*tcode == OP_ALT);
599     tcode += 1 + LINK_SIZE;
600     break;
601    
602 nigel 77 /* Single-char * or ? sets the bit and tries the next item */
603    
604     case OP_STAR:
605     case OP_MINSTAR:
606 nigel 93 case OP_POSSTAR:
607 nigel 77 case OP_QUERY:
608     case OP_MINQUERY:
609 nigel 93 case OP_POSQUERY:
610 ph10 475 set_table_bit(start_bits, tcode[1], caseless, cd);
611 nigel 77 tcode += 2;
612     #ifdef SUPPORT_UTF8
613 nigel 93 if (utf8 && tcode[-1] >= 0xc0)
614     tcode += _pcre_utf8_table4[tcode[-1] & 0x3f];
615 nigel 77 #endif
616     break;
617    
618     /* Single-char upto sets the bit and tries the next */
619    
620     case OP_UPTO:
621     case OP_MINUPTO:
622 nigel 93 case OP_POSUPTO:
623 ph10 475 set_table_bit(start_bits, tcode[3], caseless, cd);
624 nigel 77 tcode += 4;
625     #ifdef SUPPORT_UTF8
626 nigel 93 if (utf8 && tcode[-1] >= 0xc0)
627     tcode += _pcre_utf8_table4[tcode[-1] & 0x3f];
628 nigel 77 #endif
629     break;
630    
631     /* At least one single char sets the bit and stops */
632    
633     case OP_EXACT: /* Fall through */
634     tcode += 2;
635    
636     case OP_CHAR:
637     case OP_CHARNC:
638     case OP_PLUS:
639     case OP_MINPLUS:
640 nigel 93 case OP_POSPLUS:
641 ph10 475 set_table_bit(start_bits, tcode[1], caseless, cd);
642 nigel 77 try_next = FALSE;
643     break;
644    
645     /* Single character type sets the bits and stops */
646    
647     case OP_NOT_DIGIT:
648     for (c = 0; c < 32; c++)
649     start_bits[c] |= ~cd->cbits[c+cbit_digit];
650     try_next = FALSE;
651     break;
652    
653     case OP_DIGIT:
654     for (c = 0; c < 32; c++)
655     start_bits[c] |= cd->cbits[c+cbit_digit];
656     try_next = FALSE;
657     break;
658    
659 nigel 91 /* The cbit_space table has vertical tab as whitespace; we have to
660     discard it. */
661    
662 nigel 77 case OP_NOT_WHITESPACE:
663     for (c = 0; c < 32; c++)
664 nigel 91 {
665     int d = cd->cbits[c+cbit_space];
666     if (c == 1) d &= ~0x08;
667     start_bits[c] |= ~d;
668     }
669 nigel 77 try_next = FALSE;
670     break;
671    
672 nigel 91 /* The cbit_space table has vertical tab as whitespace; we have to
673     discard it. */
674    
675 nigel 77 case OP_WHITESPACE:
676     for (c = 0; c < 32; c++)
677 nigel 91 {
678     int d = cd->cbits[c+cbit_space];
679     if (c == 1) d &= ~0x08;
680     start_bits[c] |= d;
681     }
682 nigel 77 try_next = FALSE;
683     break;
684    
685     case OP_NOT_WORDCHAR:
686     for (c = 0; c < 32; c++)
687     start_bits[c] |= ~cd->cbits[c+cbit_word];
688     try_next = FALSE;
689     break;
690    
691     case OP_WORDCHAR:
692     for (c = 0; c < 32; c++)
693     start_bits[c] |= cd->cbits[c+cbit_word];
694     try_next = FALSE;
695     break;
696    
697     /* One or more character type fudges the pointer and restarts, knowing
698     it will hit a single character type and stop there. */
699    
700     case OP_TYPEPLUS:
701     case OP_TYPEMINPLUS:
702     tcode++;
703     break;
704    
705     case OP_TYPEEXACT:
706     tcode += 3;
707     break;
708    
709     /* Zero or more repeats of character types set the bits and then
710     try again. */
711    
712     case OP_TYPEUPTO:
713     case OP_TYPEMINUPTO:
714 nigel 93 case OP_TYPEPOSUPTO:
715 nigel 77 tcode += 2; /* Fall through */
716    
717     case OP_TYPESTAR:
718     case OP_TYPEMINSTAR:
719 nigel 93 case OP_TYPEPOSSTAR:
720 nigel 77 case OP_TYPEQUERY:
721     case OP_TYPEMINQUERY:
722 nigel 93 case OP_TYPEPOSQUERY:
723 nigel 77 switch(tcode[1])
724     {
725     case OP_ANY:
726 ph10 345 case OP_ALLANY:
727 nigel 93 return SSB_FAIL;
728 nigel 77
729     case OP_NOT_DIGIT:
730     for (c = 0; c < 32; c++)
731     start_bits[c] |= ~cd->cbits[c+cbit_digit];
732     break;
733    
734     case OP_DIGIT:
735     for (c = 0; c < 32; c++)
736     start_bits[c] |= cd->cbits[c+cbit_digit];
737     break;
738    
739 nigel 91 /* The cbit_space table has vertical tab as whitespace; we have to
740     discard it. */
741    
742 nigel 77 case OP_NOT_WHITESPACE:
743     for (c = 0; c < 32; c++)
744 nigel 91 {
745     int d = cd->cbits[c+cbit_space];
746     if (c == 1) d &= ~0x08;
747     start_bits[c] |= ~d;
748     }
749 nigel 77 break;
750    
751 nigel 91 /* The cbit_space table has vertical tab as whitespace; we have to
752     discard it. */
753    
754 nigel 77 case OP_WHITESPACE:
755     for (c = 0; c < 32; c++)
756 nigel 91 {
757     int d = cd->cbits[c+cbit_space];
758     if (c == 1) d &= ~0x08;
759     start_bits[c] |= d;
760     }
761 nigel 77 break;
762    
763     case OP_NOT_WORDCHAR:
764     for (c = 0; c < 32; c++)
765     start_bits[c] |= ~cd->cbits[c+cbit_word];
766     break;
767    
768     case OP_WORDCHAR:
769     for (c = 0; c < 32; c++)
770     start_bits[c] |= cd->cbits[c+cbit_word];
771     break;
772     }
773    
774     tcode += 2;
775     break;
776    
777     /* Character class where all the information is in a bit map: set the
778     bits and either carry on or not, according to the repeat count. If it was
779     a negative class, and we are operating with UTF-8 characters, any byte
780     with a value >= 0xc4 is a potentially valid starter because it starts a
781     character with a value > 255. */
782    
783     case OP_NCLASS:
784 ph10 111 #ifdef SUPPORT_UTF8
785 nigel 77 if (utf8)
786     {
787     start_bits[24] |= 0xf0; /* Bits for 0xc4 - 0xc8 */
788     memset(start_bits+25, 0xff, 7); /* Bits for 0xc9 - 0xff */
789     }
790 ph10 111 #endif
791 nigel 77 /* Fall through */
792    
793     case OP_CLASS:
794     {
795     tcode++;
796    
797     /* In UTF-8 mode, the bits in a bit map correspond to character
798     values, not to byte values. However, the bit map we are constructing is
799     for byte values. So we have to do a conversion for characters whose
800     value is > 127. In fact, there are only two possible starting bytes for
801     characters in the range 128 - 255. */
802    
803 ph10 107 #ifdef SUPPORT_UTF8
804 nigel 77 if (utf8)
805     {
806     for (c = 0; c < 16; c++) start_bits[c] |= tcode[c];
807     for (c = 128; c < 256; c++)
808     {
809     if ((tcode[c/8] && (1 << (c&7))) != 0)
810     {
811     int d = (c >> 6) | 0xc0; /* Set bit for this starter */
812     start_bits[d/8] |= (1 << (d&7)); /* and then skip on to the */
813     c = (c & 0xc0) + 0x40 - 1; /* next relevant character. */
814     }
815     }
816     }
817    
818     /* In non-UTF-8 mode, the two bit maps are completely compatible. */
819    
820     else
821 ph10 111 #endif
822 nigel 77 {
823     for (c = 0; c < 32; c++) start_bits[c] |= tcode[c];
824     }
825    
826     /* Advance past the bit map, and act on what follows */
827    
828     tcode += 32;
829     switch (*tcode)
830     {
831     case OP_CRSTAR:
832     case OP_CRMINSTAR:
833     case OP_CRQUERY:
834     case OP_CRMINQUERY:
835     tcode++;
836     break;
837    
838     case OP_CRRANGE:
839     case OP_CRMINRANGE:
840     if (((tcode[1] << 8) + tcode[2]) == 0) tcode += 5;
841     else try_next = FALSE;
842     break;
843    
844     default:
845     try_next = FALSE;
846     break;
847     }
848     }
849     break; /* End of bitmap class handling */
850    
851     } /* End of switch */
852     } /* End of try_next loop */
853    
854     code += GET(code, 1); /* Advance to next branch */
855     }
856     while (*code == OP_ALT);
857 nigel 93 return yield;
858 nigel 77 }
859    
860    
861    
862     /*************************************************
863     * Study a compiled expression *
864     *************************************************/
865    
866     /* This function is handed a compiled expression that it must study to produce
867     information that will speed up the matching. It returns a pcre_extra block
868     which then gets handed back to pcre_exec().
869    
870     Arguments:
871     re points to the compiled expression
872     options contains option bits
873     errorptr points to where to place error messages;
874     set NULL unless error
875    
876     Returns: pointer to a pcre_extra block, with study_data filled in and the
877 ph10 455 appropriate flags set;
878 nigel 77 NULL on error or if no optimization possible
879     */
880    
881 ph10 359 PCRE_EXP_DEFN pcre_extra * PCRE_CALL_CONVENTION
882 nigel 77 pcre_study(const pcre *external_re, int options, const char **errorptr)
883     {
884 ph10 455 int min;
885     BOOL bits_set = FALSE;
886 nigel 77 uschar start_bits[32];
887     pcre_extra *extra;
888     pcre_study_data *study;
889     const uschar *tables;
890 nigel 91 uschar *code;
891     compile_data compile_block;
892 nigel 77 const real_pcre *re = (const real_pcre *)external_re;
893    
894     *errorptr = NULL;
895    
896     if (re == NULL || re->magic_number != MAGIC_NUMBER)
897     {
898     *errorptr = "argument is not a compiled regular expression";
899     return NULL;
900     }
901    
902     if ((options & ~PUBLIC_STUDY_OPTIONS) != 0)
903     {
904     *errorptr = "unknown or incorrect option bit(s) set";
905     return NULL;
906     }
907    
908 nigel 91 code = (uschar *)re + re->name_table_offset +
909     (re->name_count * re->name_entry_size);
910    
911 nigel 77 /* For an anchored pattern, or an unanchored pattern that has a first char, or
912 ph10 461 a multiline pattern that matches only at "line starts", there is no point in
913 ph10 455 seeking a list of starting bytes. */
914 nigel 77
915 ph10 455 if ((re->options & PCRE_ANCHORED) == 0 &&
916     (re->flags & (PCRE_FIRSTSET|PCRE_STARTLINE)) == 0)
917     {
918     /* Set the character tables in the block that is passed around */
919 ph10 461
920 ph10 455 tables = re->tables;
921     if (tables == NULL)
922     (void)pcre_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES,
923     (void *)(&tables));
924 ph10 461
925 ph10 455 compile_block.lcc = tables + lcc_offset;
926     compile_block.fcc = tables + fcc_offset;
927     compile_block.cbits = tables + cbits_offset;
928     compile_block.ctypes = tables + ctypes_offset;
929 ph10 461
930 ph10 455 /* See if we can find a fixed set of initial characters for the pattern. */
931 ph10 461
932 ph10 455 memset(start_bits, 0, 32 * sizeof(uschar));
933 ph10 461 bits_set = set_start_bits(code, start_bits,
934     (re->options & PCRE_CASELESS) != 0, (re->options & PCRE_UTF8) != 0,
935 ph10 455 &compile_block) == SSB_DONE;
936     }
937 ph10 461
938 ph10 455 /* Find the minimum length of subject string. */
939 nigel 77
940 ph10 455 min = find_minlength(code, code, re->options);
941 nigel 77
942 ph10 455 /* Return NULL if no optimization is possible. */
943 nigel 77
944 ph10 455 if (!bits_set && min < 0) return NULL;
945 nigel 77
946     /* Get a pcre_extra block and a pcre_study_data block. The study data is put in
947     the latter, which is pointed to by the former, which may also get additional
948     data set later by the calling program. At the moment, the size of
949     pcre_study_data is fixed. We nevertheless save it in a field for returning via
950     the pcre_fullinfo() function so that if it becomes variable in the future, we
951     don't have to change that code. */
952    
953     extra = (pcre_extra *)(pcre_malloc)
954     (sizeof(pcre_extra) + sizeof(pcre_study_data));
955    
956     if (extra == NULL)
957     {
958     *errorptr = "failed to get memory";
959     return NULL;
960     }
961    
962     study = (pcre_study_data *)((char *)extra + sizeof(pcre_extra));
963     extra->flags = PCRE_EXTRA_STUDY_DATA;
964     extra->study_data = study;
965    
966     study->size = sizeof(pcre_study_data);
967 ph10 455 study->flags = 0;
968 nigel 77
969 ph10 455 if (bits_set)
970     {
971     study->flags |= PCRE_STUDY_MAPPED;
972     memcpy(study->start_bits, start_bits, sizeof(start_bits));
973     }
974 ph10 461
975 ph10 455 if (min >= 0)
976     {
977     study->flags |= PCRE_STUDY_MINLEN;
978     study->minlength = min;
979 ph10 461 }
980 ph10 455
981 nigel 77 return extra;
982     }
983    
984     /* 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