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

Contents of /code/trunk/pcre_exec.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1325 - (hide annotations) (download)
Fri May 10 14:03:18 2013 UTC (11 days, 2 hours ago) by ph10
File MIME type: text/plain
File size: 218954 byte(s)
Fix spelling mistakes in comments.

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 1251 Copyright (c) 1997-2013 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     /* This module contains pcre_exec(), the externally visible function that does
41     pattern matching using an NFA algorithm, trying to mimic Perl as closely as
42     possible. There are also some static supporting functions. */
43    
44 ph10 200 #ifdef HAVE_CONFIG_H
45 ph10 236 #include "config.h"
46 ph10 200 #endif
47 ph10 199
48 nigel 93 #define NLBLOCK md /* Block containing newline information */
49     #define PSSTART start_subject /* Field containing processed string start */
50     #define PSEND end_subject /* Field containing processed string end */
51    
52 nigel 77 #include "pcre_internal.h"
53    
54 ph10 137 /* Undefine some potentially clashing cpp symbols */
55    
56     #undef min
57     #undef max
58    
59 ph10 1271 /* The md->capture_last field uses the lower 16 bits for the last captured
60 ph10 1248 substring (which can never be greater than 65535) and a bit in the top half
61 ph10 1271 to mean "capture vector overflowed". This odd way of doing things was
62     implemented when it was realized that preserving and restoring the overflow bit
63     whenever the last capture number was saved/restored made for a neater
64     interface, and doing it this way saved on (a) another variable, which would
65     have increased the stack frame size (a big NO-NO in PCRE) and (b) another
66     separate set of save/restore instructions. The following defines are used in
67 ph10 1248 implementing this. */
68    
69     #define CAPLMASK 0x0000ffff /* The bits used for last_capture */
70     #define OVFLMASK 0xffff0000 /* The bits used for the overflow flag */
71     #define OVFLBIT 0x00010000 /* The bit that is set for overflow */
72    
73 ph10 625 /* Values for setting in md->match_function_type to indicate two special types
74     of call to match(). We do it this way to save on using another stack variable,
75 ph10 604 as stack usage is to be discouraged. */
76 nigel 77
77 ph10 604 #define MATCH_CONDASSERT 1 /* Called to check a condition assertion */
78     #define MATCH_CBEGROUP 2 /* Could-be-empty unlimited repeat group */
79 nigel 77
80     /* Non-error returns from the match() function. Error returns are externally
81     defined PCRE_ERROR_xxx codes, which are all negative. */
82    
83     #define MATCH_MATCH 1
84     #define MATCH_NOMATCH 0
85    
86 ph10 211 /* Special internal returns from the match() function. Make them sufficiently
87 ph10 210 negative to avoid the external error codes. */
88    
89 ph10 511 #define MATCH_ACCEPT (-999)
90 ph10 1271 #define MATCH_KETRPOS (-998)
91     #define MATCH_ONCE (-997)
92     /* The next 5 must be kept together and in sequence so that a test that checks
93     for any one of them can use a range. */
94     #define MATCH_COMMIT (-996)
95 ph10 618 #define MATCH_PRUNE (-995)
96     #define MATCH_SKIP (-994)
97     #define MATCH_SKIP_ARG (-993)
98     #define MATCH_THEN (-992)
99 ph10 1271 #define MATCH_BACKTRACK_MAX MATCH_THEN
100     #define MATCH_BACKTRACK_MIN MATCH_COMMIT
101 ph10 210
102 nigel 77 /* Maximum number of ints of offset to save on the stack for recursive calls.
103     If the offset vector is bigger, malloc is used. This should be a multiple of 3,
104     because the offset vector is always a multiple of 3 long. */
105    
106     #define REC_STACK_SAVE_MAX 30
107    
108     /* Min and max values for the common repeats; for the maxima, 0 => infinity */
109    
110     static const char rep_min[] = { 0, 0, 1, 1, 0, 0 };
111     static const char rep_max[] = { 0, 0, 0, 0, 1, 1 };
112    
113 ph10 475 #ifdef PCRE_DEBUG
114 nigel 77 /*************************************************
115     * Debugging function to print chars *
116     *************************************************/
117    
118     /* Print a sequence of chars in printable format, stopping at the end of the
119     subject if the requested.
120    
121     Arguments:
122     p points to characters
123     length number to print
124     is_subject TRUE if printing from within md->start_subject
125     md pointer to matching data block, if is_subject is TRUE
126    
127     Returns: nothing
128     */
129    
130     static void
131 ph10 836 pchars(const pcre_uchar *p, int length, BOOL is_subject, match_data *md)
132 nigel 77 {
133 chpe 1100 pcre_uint32 c;
134     BOOL utf = md->utf;
135 nigel 77 if (is_subject && length > md->end_subject - p) length = md->end_subject - p;
136     while (length-- > 0)
137 chpe 1100 if (isprint(c = RAWUCHARINCTEST(p))) printf("%c", (char)c); else printf("\\x{%02x}", c);
138 nigel 77 }
139     #endif
140    
141    
142    
143     /*************************************************
144     * Match a back-reference *
145     *************************************************/
146    
147 ph10 595 /* Normally, if a back reference hasn't been set, the length that is passed is
148     negative, so the match always fails. However, in JavaScript compatibility mode,
149 ph10 625 the length passed is zero. Note that in caseless UTF-8 mode, the number of
150 ph10 595 subject bytes matched may be different to the number of reference bytes.
151 nigel 77
152     Arguments:
153     offset index into the offset vector
154 ph10 595 eptr pointer into the subject
155     length length of reference to be matched (number of bytes)
156 nigel 77 md points to match data block
157 ph10 602 caseless TRUE if caseless
158 nigel 77
159 ph10 916 Returns: >= 0 the number of subject bytes matched
160     -1 no match
161 ph10 933 -2 partial match; always given if at end subject
162 nigel 77 */
163    
164 ph10 595 static int
165 ph10 836 match_ref(int offset, register PCRE_PUCHAR eptr, int length, match_data *md,
166 ph10 602 BOOL caseless)
167 nigel 77 {
168 ph10 916 PCRE_PUCHAR eptr_start = eptr;
169 ph10 836 register PCRE_PUCHAR p = md->start_subject + md->offset_vector[offset];
170 ph10 1155 #ifdef SUPPORT_UTF
171 chpe 1100 BOOL utf = md->utf;
172 ph10 1155 #endif
173 nigel 77
174 ph10 475 #ifdef PCRE_DEBUG
175 nigel 77 if (eptr >= md->end_subject)
176     printf("matching subject <null>");
177     else
178     {
179     printf("matching subject ");
180     pchars(eptr, length, TRUE, md);
181     }
182     printf(" against backref ");
183     pchars(p, length, FALSE, md);
184     printf("\n");
185     #endif
186    
187 ph10 933 /* Always fail if reference not set (and not JavaScript compatible - in that
188 ph10 916 case the length is passed as zero). */
189 nigel 77
190 ph10 595 if (length < 0) return -1;
191 nigel 77
192 ph10 354 /* Separate the caseless case for speed. In UTF-8 mode we can only do this
193     properly if Unicode properties are supported. Otherwise, we can check only
194     ASCII characters. */
195 nigel 77
196 ph10 602 if (caseless)
197 nigel 77 {
198 ph10 836 #ifdef SUPPORT_UTF
199 ph10 354 #ifdef SUPPORT_UCP
200 chpe 1100 if (utf)
201 ph10 354 {
202 ph10 625 /* Match characters up to the end of the reference. NOTE: the number of
203 ph10 1046 data units matched may differ, because in UTF-8 there are some characters
204     whose upper and lower case versions code have different numbers of bytes.
205     For example, U+023A (2 bytes in UTF-8) is the upper case version of U+2C65
206     (3 bytes in UTF-8); a sequence of 3 of the former uses 6 bytes, as does a
207     sequence of two of the latter. It is important, therefore, to check the
208     length along the reference, not along the subject (earlier code did this
209     wrong). */
210 ph10 625
211 ph10 836 PCRE_PUCHAR endptr = p + length;
212 ph10 595 while (p < endptr)
213 ph10 354 {
214 chpe 1084 pcre_uint32 c, d;
215 ph10 1046 const ucd_record *ur;
216 ph10 916 if (eptr >= md->end_subject) return -2; /* Partial match */
217 ph10 354 GETCHARINC(c, eptr);
218     GETCHARINC(d, p);
219 ph10 1046 ur = GET_UCD(d);
220 ph10 1221 if (c != d && c != d + ur->other_case)
221 ph10 1046 {
222 ph10 1221 const pcre_uint32 *pp = PRIV(ucd_caseless_sets) + ur->caseset;
223 zherczeg 1047 for (;;)
224     {
225     if (c < *pp) return -1;
226     if (c == *pp++) break;
227 ph10 1046 }
228 zherczeg 1047 }
229 ph10 358 }
230     }
231 ph10 354 else
232     #endif
233     #endif
234    
235     /* The same code works when not in UTF-8 mode and in UTF-8 mode when there
236     is no UCP support. */
237 ph10 597 {
238     while (length-- > 0)
239 ph10 836 {
240 ph10 1257 pcre_uint32 cc, cp;
241 ph10 916 if (eptr >= md->end_subject) return -2; /* Partial match */
242 chpe 1100 cc = RAWUCHARTEST(eptr);
243     cp = RAWUCHARTEST(p);
244     if (TABLE_GET(cp, md->lcc, cp) != TABLE_GET(cc, md->lcc, cc)) return -1;
245 ph10 836 p++;
246     eptr++;
247     }
248 ph10 625 }
249 nigel 77 }
250 ph10 358
251 ph10 354 /* In the caseful case, we can just compare the bytes, whether or not we
252     are in UTF-8 mode. */
253 ph10 358
254 nigel 77 else
255 ph10 625 {
256 ph10 933 while (length-- > 0)
257 zherczeg 915 {
258 ph10 916 if (eptr >= md->end_subject) return -2; /* Partial match */
259 chpe 1100 if (RAWUCHARINCTEST(p) != RAWUCHARINCTEST(eptr)) return -1;
260 ph10 933 }
261 ph10 597 }
262 nigel 77
263 ph10 916 return (int)(eptr - eptr_start);
264 nigel 77 }
265    
266    
267    
268     /***************************************************************************
269     ****************************************************************************
270     RECURSION IN THE match() FUNCTION
271    
272 nigel 87 The match() function is highly recursive, though not every recursive call
273     increases the recursive depth. Nevertheless, some regular expressions can cause
274     it to recurse to a great depth. I was writing for Unix, so I just let it call
275     itself recursively. This uses the stack for saving everything that has to be
276     saved for a recursive call. On Unix, the stack can be large, and this works
277     fine.
278 nigel 77
279 nigel 87 It turns out that on some non-Unix-like systems there are problems with
280     programs that use a lot of stack. (This despite the fact that every last chip
281     has oodles of memory these days, and techniques for extending the stack have
282     been known for decades.) So....
283 nigel 77
284     There is a fudge, triggered by defining NO_RECURSE, which avoids recursive
285     calls by keeping local variables that need to be preserved in blocks of memory
286 nigel 87 obtained from malloc() instead instead of on the stack. Macros are used to
287 nigel 77 achieve this so that the actual code doesn't look very different to what it
288     always used to.
289 ph10 164
290 ph10 165 The original heap-recursive code used longjmp(). However, it seems that this
291 ph10 164 can be very slow on some operating systems. Following a suggestion from Stan
292     Switzer, the use of longjmp() has been abolished, at the cost of having to
293     provide a unique number for each call to RMATCH. There is no way of generating
294     a sequence of numbers at compile time in C. I have given them names, to make
295     them stand out more clearly.
296    
297     Crude tests on x86 Linux show a small speedup of around 5-8%. However, on
298     FreeBSD, avoiding longjmp() more than halves the time taken to run the standard
299 ph10 165 tests. Furthermore, not using longjmp() means that local dynamic variables
300     don't have indeterminate values; this has meant that the frame size can be
301 ph10 164 reduced because the result can be "passed back" by straight setting of the
302     variable instead of being passed in the frame.
303 nigel 77 ****************************************************************************
304     ***************************************************************************/
305    
306 ph10 212 /* Numbers for RMATCH calls. When this list is changed, the code at HEAP_RETURN
307     below must be updated in sync. */
308 nigel 77
309 ph10 164 enum { RM1=1, RM2, RM3, RM4, RM5, RM6, RM7, RM8, RM9, RM10,
310     RM11, RM12, RM13, RM14, RM15, RM16, RM17, RM18, RM19, RM20,
311     RM21, RM22, RM23, RM24, RM25, RM26, RM27, RM28, RM29, RM30,
312     RM31, RM32, RM33, RM34, RM35, RM36, RM37, RM38, RM39, RM40,
313 ph10 210 RM41, RM42, RM43, RM44, RM45, RM46, RM47, RM48, RM49, RM50,
314 ph10 527 RM51, RM52, RM53, RM54, RM55, RM56, RM57, RM58, RM59, RM60,
315 ph10 1260 RM61, RM62, RM63, RM64, RM65, RM66, RM67, RM68 };
316 ph10 164
317 nigel 87 /* These versions of the macros use the stack, as normal. There are debugging
318 ph10 165 versions and production versions. Note that the "rw" argument of RMATCH isn't
319 ph10 501 actually used in this definition. */
320 nigel 77
321     #ifndef NO_RECURSE
322     #define REGISTER register
323 ph10 164
324 ph10 475 #ifdef PCRE_DEBUG
325 ph10 604 #define RMATCH(ra,rb,rc,rd,re,rw) \
326 nigel 87 { \
327     printf("match() called in line %d\n", __LINE__); \
328 ph10 836 rrc = match(ra,rb,mstart,rc,rd,re,rdepth+1); \
329 nigel 87 printf("to line %d\n", __LINE__); \
330     }
331     #define RRETURN(ra) \
332     { \
333 chpe 1100 printf("match() returned %d from line %d\n", ra, __LINE__); \
334 nigel 87 return ra; \
335     }
336     #else
337 ph10 604 #define RMATCH(ra,rb,rc,rd,re,rw) \
338 ph10 836 rrc = match(ra,rb,mstart,rc,rd,re,rdepth+1)
339 nigel 77 #define RRETURN(ra) return ra
340 nigel 87 #endif
341    
342 nigel 77 #else
343    
344    
345 ph10 164 /* These versions of the macros manage a private stack on the heap. Note that
346     the "rd" argument of RMATCH isn't actually used in this definition. It's the md
347     argument of match(), which never changes. */
348 nigel 77
349     #define REGISTER
350    
351 ph10 604 #define RMATCH(ra,rb,rc,rd,re,rw)\
352 nigel 77 {\
353 ph10 933 heapframe *newframe = frame->Xnextframe;\
354     if (newframe == NULL)\
355     {\
356     newframe = (heapframe *)(PUBL(stack_malloc))(sizeof(heapframe));\
357     if (newframe == NULL) RRETURN(PCRE_ERROR_NOMEMORY);\
358     newframe->Xnextframe = NULL;\
359     frame->Xnextframe = newframe;\
360     }\
361     frame->Xwhere = rw;\
362 ph10 164 newframe->Xeptr = ra;\
363     newframe->Xecode = rb;\
364 ph10 168 newframe->Xmstart = mstart;\
365 ph10 164 newframe->Xoffset_top = rc;\
366 ph10 602 newframe->Xeptrb = re;\
367 ph10 164 newframe->Xrdepth = frame->Xrdepth + 1;\
368     newframe->Xprevframe = frame;\
369     frame = newframe;\
370     DPRINTF(("restarting from line %d\n", __LINE__));\
371     goto HEAP_RECURSE;\
372     L_##rw:\
373     DPRINTF(("jumped back to line %d\n", __LINE__));\
374 nigel 77 }
375    
376     #define RRETURN(ra)\
377     {\
378 ph10 527 heapframe *oldframe = frame;\
379     frame = oldframe->Xprevframe;\
380 nigel 77 if (frame != NULL)\
381     {\
382 ph10 164 rrc = ra;\
383     goto HEAP_RETURN;\
384 nigel 77 }\
385     return ra;\
386     }
387    
388    
389     /* Structure for remembering the local variables in a private frame */
390    
391     typedef struct heapframe {
392     struct heapframe *Xprevframe;
393 ph10 933 struct heapframe *Xnextframe;
394 nigel 77
395     /* Function arguments that may change */
396    
397 ph10 836 PCRE_PUCHAR Xeptr;
398     const pcre_uchar *Xecode;
399     PCRE_PUCHAR Xmstart;
400 nigel 77 int Xoffset_top;
401     eptrblock *Xeptrb;
402 nigel 91 unsigned int Xrdepth;
403 nigel 77
404     /* Function local variables */
405    
406 ph10 836 PCRE_PUCHAR Xcallpat;
407     #ifdef SUPPORT_UTF
408     PCRE_PUCHAR Xcharptr;
409 ph10 406 #endif
410 ph10 836 PCRE_PUCHAR Xdata;
411     PCRE_PUCHAR Xnext;
412     PCRE_PUCHAR Xpp;
413     PCRE_PUCHAR Xprev;
414     PCRE_PUCHAR Xsaved_eptr;
415 nigel 77
416     recursion_info Xnew_recursive;
417    
418     BOOL Xcur_is_word;
419     BOOL Xcondition;
420     BOOL Xprev_is_word;
421    
422     #ifdef SUPPORT_UCP
423     int Xprop_type;
424 chpe 1135 unsigned int Xprop_value;
425 nigel 77 int Xprop_fail_result;
426 ph10 123 int Xoclength;
427 ph10 836 pcre_uchar Xocchars[6];
428 nigel 77 #endif
429    
430 ph10 403 int Xcodelink;
431 nigel 77 int Xctype;
432 nigel 93 unsigned int Xfc;
433 nigel 77 int Xfi;
434     int Xlength;
435     int Xmax;
436     int Xmin;
437 ph10 1236 unsigned int Xnumber;
438 nigel 77 int Xoffset;
439 ph10 1236 unsigned int Xop;
440 ph10 1248 pcre_int32 Xsave_capture_last;
441 nigel 77 int Xsave_offset1, Xsave_offset2, Xsave_offset3;
442     int Xstacksave[REC_STACK_SAVE_MAX];
443    
444     eptrblock Xnewptrb;
445    
446 ph10 164 /* Where to jump back to */
447 nigel 77
448 ph10 164 int Xwhere;
449 ph10 165
450 nigel 77 } heapframe;
451    
452     #endif
453    
454    
455     /***************************************************************************
456     ***************************************************************************/
457    
458    
459    
460     /*************************************************
461     * Match from current position *
462     *************************************************/
463    
464 nigel 93 /* This function is called recursively in many circumstances. Whenever it
465 nigel 77 returns a negative (error) response, the outer incarnation must also return the
466 ph10 426 same response. */
467 nigel 77
468 ph10 426 /* These macros pack up tests that are used for partial matching, and which
469 ph10 836 appear several times in the code. We set the "hit end" flag if the pointer is
470 ph10 426 at the end of the subject and also past the start of the subject (i.e.
471 ph10 427 something has been matched). For hard partial matching, we then return
472     immediately. The second one is used when we already know we are past the end of
473     the subject. */
474 ph10 426
475     #define CHECK_PARTIAL()\
476 ph10 553 if (md->partial != 0 && eptr >= md->end_subject && \
477     eptr > md->start_used_ptr) \
478     { \
479     md->hitend = TRUE; \
480 ph10 836 if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL); \
481 ph10 427 }
482 ph10 426
483     #define SCHECK_PARTIAL()\
484 ph10 553 if (md->partial != 0 && eptr > md->start_used_ptr) \
485     { \
486     md->hitend = TRUE; \
487 ph10 836 if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL); \
488 ph10 427 }
489 ph10 426
490 ph10 427
491 ph10 426 /* Performance note: It might be tempting to extract commonly used fields from
492 ph10 836 the md structure (e.g. utf, end_subject) into individual variables to improve
493 nigel 77 performance. Tests using gcc on a SPARC disproved this; in the first case, it
494     made performance worse.
495    
496     Arguments:
497 nigel 93 eptr pointer to current character in subject
498     ecode pointer to current position in compiled code
499 ph10 168 mstart pointer to the current match start position (can be modified
500 ph10 172 by encountering \K)
501 nigel 77 offset_top current top pointer
502     md pointer to "static" info for the match
503     eptrb pointer to chain of blocks containing eptr at start of
504     brackets - for testing for empty matches
505 nigel 87 rdepth the recursion depth
506 nigel 77
507     Returns: MATCH_MATCH if matched ) these values are >= 0
508     MATCH_NOMATCH if failed to match )
509 ph10 510 a negative MATCH_xxx value for PRUNE, SKIP, etc
510 nigel 77 a negative PCRE_ERROR_xxx value if aborted by an error condition
511 nigel 87 (e.g. stopped by repeated call or recursion limit)
512 nigel 77 */
513    
514     static int
515 ph10 836 match(REGISTER PCRE_PUCHAR eptr, REGISTER const pcre_uchar *ecode,
516 ph10 842 PCRE_PUCHAR mstart, int offset_top, match_data *md, eptrblock *eptrb,
517 ph10 835 unsigned int rdepth)
518 nigel 77 {
519     /* These variables do not need to be preserved over recursion in this function,
520 nigel 93 so they can be ordinary variables in all cases. Mark some of them with
521     "register" because they are used a lot in loops. */
522 nigel 77
523 nigel 91 register int rrc; /* Returns from recursive calls */
524     register int i; /* Used for loops not involving calls to RMATCH() */
525 chpe 1084 register pcre_uint32 c; /* Character values not kept over RMATCH() calls */
526 ph10 836 register BOOL utf; /* Local copy of UTF flag for speed */
527 nigel 77
528 nigel 93 BOOL minimize, possessive; /* Quantifier options */
529 ph10 602 BOOL caseless;
530 ph10 403 int condcode;
531 nigel 93
532 nigel 77 /* When recursion is not being used, all "local" variables that have to be
533 ph10 892 preserved over calls to RMATCH() are part of a "frame". We set up the top-level
534     frame on the stack here; subsequent instantiations are obtained from the heap
535 ph10 903 whenever RMATCH() does a "recursion". See the macro definitions above. Putting
536     the top-level on the stack rather than malloc-ing them all gives a performance
537 ph10 892 boost in many cases where there is not much "recursion". */
538 nigel 77
539     #ifdef NO_RECURSE
540 ph10 933 heapframe *frame = (heapframe *)md->match_frames_base;
541 nigel 77
542     /* Copy in the original argument variables */
543    
544     frame->Xeptr = eptr;
545     frame->Xecode = ecode;
546 ph10 168 frame->Xmstart = mstart;
547 nigel 77 frame->Xoffset_top = offset_top;
548     frame->Xeptrb = eptrb;
549 nigel 87 frame->Xrdepth = rdepth;
550 nigel 77
551     /* This is where control jumps back to to effect "recursion" */
552    
553     HEAP_RECURSE:
554    
555     /* Macros make the argument variables come from the current frame */
556    
557     #define eptr frame->Xeptr
558     #define ecode frame->Xecode
559 ph10 168 #define mstart frame->Xmstart
560 nigel 77 #define offset_top frame->Xoffset_top
561     #define eptrb frame->Xeptrb
562 nigel 87 #define rdepth frame->Xrdepth
563 nigel 77
564     /* Ditto for the local variables */
565    
566 ph10 836 #ifdef SUPPORT_UTF
567 nigel 77 #define charptr frame->Xcharptr
568     #endif
569     #define callpat frame->Xcallpat
570 ph10 403 #define codelink frame->Xcodelink
571 nigel 77 #define data frame->Xdata
572     #define next frame->Xnext
573     #define pp frame->Xpp
574     #define prev frame->Xprev
575     #define saved_eptr frame->Xsaved_eptr
576    
577     #define new_recursive frame->Xnew_recursive
578    
579     #define cur_is_word frame->Xcur_is_word
580     #define condition frame->Xcondition
581     #define prev_is_word frame->Xprev_is_word
582    
583     #ifdef SUPPORT_UCP
584     #define prop_type frame->Xprop_type
585 nigel 87 #define prop_value frame->Xprop_value
586 nigel 77 #define prop_fail_result frame->Xprop_fail_result
587 ph10 115 #define oclength frame->Xoclength
588     #define occhars frame->Xocchars
589 nigel 77 #endif
590    
591     #define ctype frame->Xctype
592     #define fc frame->Xfc
593     #define fi frame->Xfi
594     #define length frame->Xlength
595     #define max frame->Xmax
596     #define min frame->Xmin
597     #define number frame->Xnumber
598     #define offset frame->Xoffset
599     #define op frame->Xop
600     #define save_capture_last frame->Xsave_capture_last
601     #define save_offset1 frame->Xsave_offset1
602     #define save_offset2 frame->Xsave_offset2
603     #define save_offset3 frame->Xsave_offset3
604     #define stacksave frame->Xstacksave
605    
606     #define newptrb frame->Xnewptrb
607    
608     /* When recursion is being used, local variables are allocated on the stack and
609     get preserved during recursion in the normal way. In this environment, fi and
610     i, and fc and c, can be the same variables. */
611    
612 nigel 93 #else /* NO_RECURSE not defined */
613 nigel 77 #define fi i
614     #define fc c
615    
616 ph10 604 /* Many of the following variables are used only in small blocks of the code.
617     My normal style of coding would have declared them within each of those blocks.
618     However, in order to accommodate the version of this code that uses an external
619     "stack" implemented on the heap, it is easier to declare them all here, so the
620     declarations can be cut out in a block. The only declarations within blocks
621     below are for variables that do not have to be preserved over a recursive call
622     to RMATCH(). */
623 nigel 77
624 ph10 836 #ifdef SUPPORT_UTF
625     const pcre_uchar *charptr;
626 ph10 625 #endif
627 ph10 836 const pcre_uchar *callpat;
628     const pcre_uchar *data;
629     const pcre_uchar *next;
630     PCRE_PUCHAR pp;
631     const pcre_uchar *prev;
632     PCRE_PUCHAR saved_eptr;
633 ph10 625
634     recursion_info new_recursive;
635    
636     BOOL cur_is_word;
637 nigel 87 BOOL condition;
638 nigel 77 BOOL prev_is_word;
639    
640     #ifdef SUPPORT_UCP
641     int prop_type;
642 chpe 1135 unsigned int prop_value;
643 nigel 77 int prop_fail_result;
644 ph10 115 int oclength;
645 ph10 836 pcre_uchar occhars[6];
646 nigel 77 #endif
647    
648 ph10 399 int codelink;
649 nigel 77 int ctype;
650     int length;
651     int max;
652     int min;
653 ph10 1145 unsigned int number;
654 nigel 77 int offset;
655 ph10 1236 unsigned int op;
656 ph10 1248 pcre_int32 save_capture_last;
657 nigel 77 int save_offset1, save_offset2, save_offset3;
658     int stacksave[REC_STACK_SAVE_MAX];
659    
660     eptrblock newptrb;
661 ph10 893
662 ph10 903 /* There is a special fudge for calling match() in a way that causes it to
663 ph10 893 measure the size of its basic stack frame when the stack is being used for
664 ph10 895 recursion. The second argument (ecode) being NULL triggers this behaviour. It
665 ph10 901 cannot normally ever be NULL. The return is the negated value of the frame
666 ph10 895 size. */
667 ph10 893
668     if (ecode == NULL)
669     {
670     if (rdepth == 0)
671 ph10 895 return match((PCRE_PUCHAR)&rdepth, NULL, NULL, 0, NULL, NULL, 1);
672 ph10 893 else
673     {
674 ph10 895 int len = (char *)&rdepth - (char *)eptr;
675 ph10 893 return (len > 0)? -len : len;
676     }
677 ph10 903 }
678 nigel 93 #endif /* NO_RECURSE */
679 nigel 77
680 ph10 625 /* To save space on the stack and in the heap frame, I have doubled up on some
681     of the local variables that are used only in localised parts of the code, but
682     still need to be preserved over recursive calls of match(). These macros define
683 ph10 604 the alternative names that are used. */
684    
685     #define allow_zero cur_is_word
686     #define cbegroup condition
687     #define code_offset codelink
688     #define condassert condition
689     #define matched_once prev_is_word
690 ph10 836 #define foc number
691 ph10 882 #define save_mark data
692 ph10 604
693 nigel 77 /* These statements are here to stop the compiler complaining about unitialized
694     variables. */
695    
696     #ifdef SUPPORT_UCP
697 nigel 87 prop_value = 0;
698 nigel 77 prop_fail_result = 0;
699     #endif
700    
701 nigel 93
702 nigel 91 /* This label is used for tail recursion, which is used in a few cases even
703     when NO_RECURSE is not defined, in order to reduce the amount of stack that is
704     used. Thanks to Ian Taylor for noticing this possibility and sending the
705     original patch. */
706    
707     TAIL_RECURSE:
708    
709 nigel 87 /* OK, now we can get on with the real code of the function. Recursive calls
710     are specified by the macro RMATCH and RRETURN is used to return. When
711     NO_RECURSE is *not* defined, these just turn into a recursive call to match()
712 ph10 475 and a "return", respectively (possibly with some debugging if PCRE_DEBUG is
713 nigel 87 defined). However, RMATCH isn't like a function call because it's quite a
714     complicated macro. It has to be used in one particular way. This shouldn't,
715     however, impact performance when true recursion is being used. */
716 nigel 77
717 ph10 836 #ifdef SUPPORT_UTF
718     utf = md->utf; /* Local copy of the flag */
719 ph10 164 #else
720 ph10 836 utf = FALSE;
721 ph10 164 #endif
722    
723 nigel 87 /* First check that we haven't called match() too many times, or that we
724     haven't exceeded the recursive call limit. */
725    
726 nigel 77 if (md->match_call_count++ >= md->match_limit) RRETURN(PCRE_ERROR_MATCHLIMIT);
727 nigel 87 if (rdepth >= md->match_limit_recursion) RRETURN(PCRE_ERROR_RECURSIONLIMIT);
728 nigel 77
729 nigel 93 /* At the start of a group with an unlimited repeat that may match an empty
730 ph10 625 string, the variable md->match_function_type is set to MATCH_CBEGROUP. It is
731     done this way to save having to use another function argument, which would take
732 ph10 604 up space on the stack. See also MATCH_CONDASSERT below.
733 nigel 77
734 ph10 604 When MATCH_CBEGROUP is set, add the current subject pointer to the chain of
735     such remembered pointers, to be checked when we hit the closing ket, in order
736     to break infinite loops that match no characters. When match() is called in
737     other circumstances, don't add to the chain. The MATCH_CBEGROUP feature must
738     NOT be used with tail recursion, because the memory block that is used is on
739     the stack, so a new one may be required for each match(). */
740    
741     if (md->match_function_type == MATCH_CBEGROUP)
742 nigel 77 {
743 ph10 197 newptrb.epb_saved_eptr = eptr;
744     newptrb.epb_prev = eptrb;
745     eptrb = &newptrb;
746 ph10 604 md->match_function_type = 0;
747 nigel 77 }
748    
749 nigel 93 /* Now start processing the opcodes. */
750 nigel 77
751     for (;;)
752     {
753 nigel 93 minimize = possessive = FALSE;
754 nigel 77 op = *ecode;
755 ph10 625
756 nigel 93 switch(op)
757     {
758 ph10 510 case OP_MARK:
759 ph10 836 md->nomatch_mark = ecode + 2;
760     md->mark = NULL; /* In case previously set by assertion */
761     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
762 ph10 604 eptrb, RM55);
763 ph10 836 if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
764     md->mark == NULL) md->mark = ecode + 2;
765 ph10 512
766     /* A return of MATCH_SKIP_ARG means that matching failed at SKIP with an
767     argument, and we must check whether that argument matches this MARK's
768     argument. It is passed back in md->start_match_ptr (an overloading of that
769     variable). If it does match, we reset that variable to the current subject
770     position and return MATCH_SKIP. Otherwise, pass back the return code
771 ph10 510 unaltered. */
772 ph10 512
773 ph10 836 else if (rrc == MATCH_SKIP_ARG &&
774 chpe 1100 STRCMP_UC_UC_TEST(ecode + 2, md->start_match_ptr) == 0)
775 ph10 510 {
776     md->start_match_ptr = eptr;
777     RRETURN(MATCH_SKIP);
778     }
779     RRETURN(rrc);
780    
781 ph10 210 case OP_FAIL:
782 ph10 836 RRETURN(MATCH_NOMATCH);
783 ph10 211
784 ph10 510 case OP_COMMIT:
785 ph10 836 RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
786 ph10 604 eptrb, RM52);
787 ph10 1284 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
788 ph10 836 RRETURN(MATCH_COMMIT);
789 ph10 510
790 ph10 210 case OP_PRUNE:
791 ph10 836 RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
792 ph10 604 eptrb, RM51);
793 ph10 1284 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
794 ph10 836 RRETURN(MATCH_PRUNE);
795 ph10 211
796 ph10 510 case OP_PRUNE_ARG:
797 ph10 836 md->nomatch_mark = ecode + 2;
798     md->mark = NULL; /* In case previously set by assertion */
799     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
800 ph10 604 eptrb, RM56);
801 ph10 836 if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
802     md->mark == NULL) md->mark = ecode + 2;
803 ph10 1284 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
804 ph10 510 RRETURN(MATCH_PRUNE);
805 ph10 211
806 ph10 210 case OP_SKIP:
807 ph10 836 RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
808 ph10 604 eptrb, RM53);
809 ph10 1284 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
810 ph10 211 md->start_match_ptr = eptr; /* Pass back current position */
811 ph10 836 RRETURN(MATCH_SKIP);
812 ph10 211
813 ph10 836 /* Note that, for Perl compatibility, SKIP with an argument does NOT set
814 ph10 1320 nomatch_mark. When a pattern match ends with a SKIP_ARG for which there was
815 ph10 1274 not a matching mark, we have to re-run the match, ignoring the SKIP_ARG
816 ph10 1325 that failed and any that precede it (either they also failed, or were not
817 ph10 1320 triggered). To do this, we maintain a count of executed SKIP_ARGs. If a
818 ph10 1274 SKIP_ARG gets to top level, the match is re-run with md->ignore_skip_arg
819     set to the count of the one that failed. */
820 ph10 836
821 ph10 510 case OP_SKIP_ARG:
822 ph10 1274 md->skip_arg_count++;
823     if (md->skip_arg_count <= md->ignore_skip_arg)
824 ph10 836 {
825     ecode += PRIV(OP_lengths)[*ecode] + ecode[1];
826     break;
827     }
828     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
829 ph10 604 eptrb, RM57);
830 ph10 1284 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
831 ph10 1320
832 ph10 512 /* Pass back the current skip name by overloading md->start_match_ptr and
833     returning the special MATCH_SKIP_ARG return code. This will either be
834 ph10 836 caught by a matching MARK, or get to the top, where it causes a rematch
835 ph10 1274 with md->ignore_skip_arg set to the value of md->skip_arg_count. */
836 ph10 512
837 ph10 510 md->start_match_ptr = ecode + 2;
838 ph10 512 RRETURN(MATCH_SKIP_ARG);
839 ph10 553
840 ph10 716 /* For THEN (and THEN_ARG) we pass back the address of the opcode, so that
841     the branch in which it occurs can be determined. Overload the start of
842     match pointer to do this. */
843 ph10 512
844 ph10 210 case OP_THEN:
845 ph10 836 RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
846 ph10 604 eptrb, RM54);
847 ph10 210 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
848 ph10 716 md->start_match_ptr = ecode;
849 ph10 836 RRETURN(MATCH_THEN);
850 ph10 510
851     case OP_THEN_ARG:
852 ph10 836 md->nomatch_mark = ecode + 2;
853     md->mark = NULL; /* In case previously set by assertion */
854     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top,
855 ph10 716 md, eptrb, RM58);
856 ph10 836 if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
857     md->mark == NULL) md->mark = ecode + 2;
858 ph10 510 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
859 ph10 733 md->start_match_ptr = ecode;
860 ph10 212 RRETURN(MATCH_THEN);
861 ph10 733
862 ph10 723 /* Handle an atomic group that does not contain any capturing parentheses.
863 ph10 733 This can be handled like an assertion. Prior to 8.13, all atomic groups
864     were handled this way. In 8.13, the code was changed as below for ONCE, so
865     that backups pass through the group and thereby reset captured values.
866     However, this uses a lot more stack, so in 8.20, atomic groups that do not
867     contain any captures generate OP_ONCE_NC, which can be handled in the old,
868 ph10 723 less stack intensive way.
869 ph10 211
870 ph10 723 Check the alternative branches in turn - the matching won't pass the KET
871     for this kind of subpattern. If any one branch matches, we carry on as at
872     the end of a normal bracket, leaving the subject pointer, but resetting
873     the start-of-match value in case it was changed by \K. */
874    
875     case OP_ONCE_NC:
876     prev = ecode;
877     saved_eptr = eptr;
878 ph10 903 save_mark = md->mark;
879 ph10 723 do
880     {
881     RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM64);
882     if (rrc == MATCH_MATCH) /* Note: _not_ MATCH_ACCEPT */
883     {
884     mstart = md->start_match_ptr;
885     break;
886     }
887     if (rrc == MATCH_THEN)
888     {
889     next = ecode + GET(ecode,1);
890 ph10 733 if (md->start_match_ptr < next &&
891 ph10 723 (*ecode == OP_ALT || *next == OP_ALT))
892     rrc = MATCH_NOMATCH;
893 ph10 733 }
894    
895 ph10 723 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
896     ecode += GET(ecode,1);
897 ph10 903 md->mark = save_mark;
898 ph10 723 }
899     while (*ecode == OP_ALT);
900    
901     /* If hit the end of the group (which could be repeated), fail */
902    
903     if (*ecode != OP_ONCE_NC && *ecode != OP_ALT) RRETURN(MATCH_NOMATCH);
904    
905     /* Continue as from after the group, updating the offsets high water
906     mark, since extracts may have been taken. */
907    
908     do ecode += GET(ecode, 1); while (*ecode == OP_ALT);
909    
910     offset_top = md->end_offset_top;
911     eptr = md->end_match_ptr;
912    
913     /* For a non-repeating ket, just continue at this level. This also
914     happens for a repeating ket if no characters were matched in the group.
915     This is the forcible breaking of infinite loops as implemented in Perl
916     5.005. */
917    
918     if (*ecode == OP_KET || eptr == saved_eptr)
919     {
920     ecode += 1+LINK_SIZE;
921     break;
922     }
923    
924     /* The repeating kets try the rest of the pattern or restart from the
925     preceding bracket, in the appropriate order. The second "call" of match()
926     uses tail recursion, to avoid using another stack frame. */
927    
928     if (*ecode == OP_KETRMIN)
929     {
930     RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM65);
931     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
932     ecode = prev;
933     goto TAIL_RECURSE;
934     }
935     else /* OP_KETRMAX */
936     {
937     RMATCH(eptr, prev, offset_top, md, eptrb, RM66);
938     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
939     ecode += 1 + LINK_SIZE;
940     goto TAIL_RECURSE;
941     }
942     /* Control never gets here */
943    
944 ph10 604 /* Handle a capturing bracket, other than those that are possessive with an
945     unlimited repeat. If there is space in the offset vector, save the current
946     subject position in the working slot at the top of the vector. We mustn't
947     change the current values of the data slot, because they may be set from a
948     previous iteration of this group, and be referred to by a reference inside
949 ph10 625 the group. A failure to match might occur after the group has succeeded,
950 ph10 617 if something later on doesn't match. For this reason, we need to restore
951     the working value and also the values of the final offsets, in case they
952     were set by a previous iteration of the same bracket.
953 nigel 77
954 nigel 93 If there isn't enough space in the offset vector, treat this as if it were
955     a non-capturing bracket. Don't worry about setting the flag for the error
956     case here; that is handled in the code for KET. */
957 nigel 77
958 nigel 93 case OP_CBRA:
959     case OP_SCBRA:
960     number = GET2(ecode, 1+LINK_SIZE);
961 nigel 77 offset = number << 1;
962 ph10 625
963 ph10 475 #ifdef PCRE_DEBUG
964 nigel 93 printf("start bracket %d\n", number);
965     printf("subject=");
966 nigel 77 pchars(eptr, 16, TRUE, md);
967     printf("\n");
968     #endif
969    
970     if (offset < md->offset_max)
971     {
972     save_offset1 = md->offset_vector[offset];
973     save_offset2 = md->offset_vector[offset+1];
974     save_offset3 = md->offset_vector[md->offset_end - number];
975     save_capture_last = md->capture_last;
976 ph10 903 save_mark = md->mark;
977 nigel 77
978     DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
979 ph10 531 md->offset_vector[md->offset_end - number] =
980 ph10 530 (int)(eptr - md->start_subject);
981 nigel 77
982 ph10 604 for (;;)
983 nigel 77 {
984 ph10 625 if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
985 ph10 836 RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
986 ph10 604 eptrb, RM1);
987 ph10 618 if (rrc == MATCH_ONCE) break; /* Backing up through an atomic group */
988 ph10 733
989     /* If we backed up to a THEN, check whether it is within the current
990     branch by comparing the address of the THEN that is passed back with
991 ph10 716 the end of the branch. If it is within the current branch, and the
992     branch is one of two or more alternatives (it either starts or ends
993 ph10 733 with OP_ALT), we have reached the limit of THEN's action, so convert
994     the return code to NOMATCH, which will cause normal backtracking to
995 ph10 716 happen from now on. Otherwise, THEN is passed back to an outer
996 ph10 733 alternative. This implements Perl's treatment of parenthesized groups,
997     where a group not containing | does not affect the current alternative,
998 ph10 716 that is, (X) is NOT the same as (X|(*F)). */
999    
1000     if (rrc == MATCH_THEN)
1001     {
1002     next = ecode + GET(ecode,1);
1003 ph10 733 if (md->start_match_ptr < next &&
1004 ph10 716 (*ecode == OP_ALT || *next == OP_ALT))
1005     rrc = MATCH_NOMATCH;
1006 ph10 733 }
1007    
1008 ph10 716 /* Anything other than NOMATCH is passed back. */
1009    
1010     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1011 nigel 77 md->capture_last = save_capture_last;
1012     ecode += GET(ecode, 1);
1013 ph10 882 md->mark = save_mark;
1014 ph10 625 if (*ecode != OP_ALT) break;
1015 nigel 77 }
1016    
1017     DPRINTF(("bracket %d failed\n", number));
1018     md->offset_vector[offset] = save_offset1;
1019     md->offset_vector[offset+1] = save_offset2;
1020     md->offset_vector[md->offset_end - number] = save_offset3;
1021 ph10 625
1022 ph10 716 /* At this point, rrc will be one of MATCH_ONCE or MATCH_NOMATCH. */
1023 nigel 77
1024 ph10 716 RRETURN(rrc);
1025 nigel 77 }
1026    
1027 ph10 197 /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
1028     as a non-capturing bracket. */
1029 nigel 77
1030 ph10 197 /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1031     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1032    
1033 nigel 93 DPRINTF(("insufficient capture room: treat as non-capturing\n"));
1034 nigel 77
1035 ph10 197 /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1036     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1037    
1038 ph10 618 /* Non-capturing or atomic group, except for possessive with unlimited
1039 ph10 723 repeat and ONCE group with no captures. Loop for all the alternatives.
1040 ph10 708
1041 ph10 702 When we get to the final alternative within the brackets, we used to return
1042     the result of a recursive call to match() whatever happened so it was
1043     possible to reduce stack usage by turning this into a tail recursion,
1044     except in the case of a possibly empty group. However, now that there is
1045     the possiblity of (*THEN) occurring in the final alternative, this
1046     optimization is no longer always possible.
1047 ph10 625
1048 ph10 708 We can optimize if we know there are no (*THEN)s in the pattern; at present
1049     this is the best that can be done.
1050    
1051 ph10 625 MATCH_ONCE is returned when the end of an atomic group is successfully
1052     reached, but subsequent matching fails. It passes back up the tree (causing
1053     captured values to be reset) until the original atomic group level is
1054 ph10 618 reached. This is tested by comparing md->once_target with the start of the
1055     group. At this point, the return is converted into MATCH_NOMATCH so that
1056     previous backup points can be taken. */
1057 nigel 77
1058 ph10 618 case OP_ONCE:
1059 nigel 93 case OP_BRA:
1060     case OP_SBRA:
1061     DPRINTF(("start non-capturing bracket\n"));
1062 ph10 618
1063 nigel 91 for (;;)
1064 nigel 77 {
1065 ph10 976 if (op >= OP_SBRA || op == OP_ONCE)
1066     md->match_function_type = MATCH_CBEGROUP;
1067 ph10 702
1068     /* If this is not a possibly empty group, and there are no (*THEN)s in
1069 ph10 708 the pattern, and this is the final alternative, optimize as described
1070 ph10 702 above. */
1071    
1072     else if (!md->hasthen && ecode[GET(ecode, 1)] != OP_ALT)
1073     {
1074 ph10 836 ecode += PRIV(OP_lengths)[*ecode];
1075 ph10 702 goto TAIL_RECURSE;
1076 ph10 708 }
1077 ph10 702
1078     /* In all other cases, we have to make another call to match(). */
1079    
1080 ph10 882 save_mark = md->mark;
1081 ph10 1248 save_capture_last = md->capture_last;
1082 ph10 836 RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md, eptrb,
1083 ph10 604 RM2);
1084 ph10 903
1085 ph10 716 /* See comment in the code for capturing groups above about handling
1086     THEN. */
1087    
1088     if (rrc == MATCH_THEN)
1089 ph10 625 {
1090 ph10 716 next = ecode + GET(ecode,1);
1091 ph10 733 if (md->start_match_ptr < next &&
1092 ph10 716 (*ecode == OP_ALT || *next == OP_ALT))
1093     rrc = MATCH_NOMATCH;
1094 ph10 733 }
1095    
1096     if (rrc != MATCH_NOMATCH)
1097 ph10 716 {
1098 ph10 618 if (rrc == MATCH_ONCE)
1099     {
1100 ph10 836 const pcre_uchar *scode = ecode;
1101 ph10 618 if (*scode != OP_ONCE) /* If not at start, find it */
1102     {
1103     while (*scode == OP_ALT) scode += GET(scode, 1);
1104     scode -= GET(scode, 1);
1105 ph10 625 }
1106 ph10 618 if (md->once_target == scode) rrc = MATCH_NOMATCH;
1107 ph10 625 }
1108 ph10 550 RRETURN(rrc);
1109 ph10 625 }
1110 nigel 77 ecode += GET(ecode, 1);
1111 ph10 903 md->mark = save_mark;
1112 ph10 625 if (*ecode != OP_ALT) break;
1113 ph10 1248 md->capture_last = save_capture_last;
1114 nigel 77 }
1115 ph10 733
1116 ph10 609 RRETURN(MATCH_NOMATCH);
1117    
1118 ph10 625 /* Handle possessive capturing brackets with an unlimited repeat. We come
1119 ph10 604 here from BRAZERO with allow_zero set TRUE. The offset_vector values are
1120     handled similarly to the normal case above. However, the matching is
1121     different. The end of these brackets will always be OP_KETRPOS, which
1122     returns MATCH_KETRPOS without going further in the pattern. By this means
1123     we can handle the group by iteration rather than recursion, thereby
1124     reducing the amount of stack needed. */
1125 ph10 625
1126 ph10 604 case OP_CBRAPOS:
1127     case OP_SCBRAPOS:
1128     allow_zero = FALSE;
1129 ph10 625
1130 ph10 604 POSSESSIVE_CAPTURE:
1131     number = GET2(ecode, 1+LINK_SIZE);
1132     offset = number << 1;
1133    
1134     #ifdef PCRE_DEBUG
1135     printf("start possessive bracket %d\n", number);
1136     printf("subject=");
1137     pchars(eptr, 16, TRUE, md);
1138     printf("\n");
1139     #endif
1140    
1141     if (offset < md->offset_max)
1142     {
1143     matched_once = FALSE;
1144 ph10 836 code_offset = (int)(ecode - md->start_code);
1145 ph10 604
1146     save_offset1 = md->offset_vector[offset];
1147     save_offset2 = md->offset_vector[offset+1];
1148     save_offset3 = md->offset_vector[md->offset_end - number];
1149     save_capture_last = md->capture_last;
1150    
1151     DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
1152 ph10 625
1153     /* Each time round the loop, save the current subject position for use
1154     when the group matches. For MATCH_MATCH, the group has matched, so we
1155     restart it with a new subject starting position, remembering that we had
1156     at least one match. For MATCH_NOMATCH, carry on with the alternatives, as
1157     usual. If we haven't matched any alternatives in any iteration, check to
1158     see if a previous iteration matched. If so, the group has matched;
1159     continue from afterwards. Otherwise it has failed; restore the previous
1160 ph10 604 capture values before returning NOMATCH. */
1161 ph10 625
1162 ph10 604 for (;;)
1163     {
1164     md->offset_vector[md->offset_end - number] =
1165     (int)(eptr - md->start_subject);
1166 ph10 625 if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
1167 ph10 836 RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
1168 ph10 604 eptrb, RM63);
1169     if (rrc == MATCH_KETRPOS)
1170     {
1171     offset_top = md->end_offset_top;
1172     eptr = md->end_match_ptr;
1173 ph10 625 ecode = md->start_code + code_offset;
1174 ph10 604 save_capture_last = md->capture_last;
1175 ph10 625 matched_once = TRUE;
1176     continue;
1177     }
1178 ph10 733
1179 ph10 716 /* See comment in the code for capturing groups above about handling
1180     THEN. */
1181    
1182     if (rrc == MATCH_THEN)
1183     {
1184     next = ecode + GET(ecode,1);
1185 ph10 733 if (md->start_match_ptr < next &&
1186 ph10 716 (*ecode == OP_ALT || *next == OP_ALT))
1187     rrc = MATCH_NOMATCH;
1188 ph10 733 }
1189 ph10 716
1190     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1191 ph10 604 md->capture_last = save_capture_last;
1192     ecode += GET(ecode, 1);
1193 ph10 625 if (*ecode != OP_ALT) break;
1194 ph10 604 }
1195 ph10 610
1196 ph10 604 if (!matched_once)
1197 ph10 625 {
1198 ph10 604 md->offset_vector[offset] = save_offset1;
1199     md->offset_vector[offset+1] = save_offset2;
1200     md->offset_vector[md->offset_end - number] = save_offset3;
1201     }
1202 ph10 625
1203 ph10 604 if (allow_zero || matched_once)
1204 ph10 625 {
1205 ph10 604 ecode += 1 + LINK_SIZE;
1206     break;
1207 ph10 625 }
1208    
1209 ph10 604 RRETURN(MATCH_NOMATCH);
1210     }
1211 ph10 625
1212 ph10 604 /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
1213     as a non-capturing bracket. */
1214    
1215     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1216     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1217    
1218     DPRINTF(("insufficient capture room: treat as non-capturing\n"));
1219    
1220     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1221     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1222    
1223 ph10 625 /* Non-capturing possessive bracket with unlimited repeat. We come here
1224 ph10 604 from BRAZERO with allow_zero = TRUE. The code is similar to the above,
1225     without the capturing complication. It is written out separately for speed
1226     and cleanliness. */
1227    
1228     case OP_BRAPOS:
1229     case OP_SBRAPOS:
1230 ph10 625 allow_zero = FALSE;
1231    
1232 ph10 604 POSSESSIVE_NON_CAPTURE:
1233     matched_once = FALSE;
1234 ph10 836 code_offset = (int)(ecode - md->start_code);
1235 ph10 1248 save_capture_last = md->capture_last;
1236 ph10 604
1237     for (;;)
1238     {
1239 ph10 625 if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
1240 ph10 836 RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
1241 ph10 609 eptrb, RM48);
1242 ph10 604 if (rrc == MATCH_KETRPOS)
1243     {
1244 ph10 610 offset_top = md->end_offset_top;
1245 ph10 604 eptr = md->end_match_ptr;
1246 ph10 625 ecode = md->start_code + code_offset;
1247     matched_once = TRUE;
1248     continue;
1249     }
1250 ph10 733
1251 ph10 716 /* See comment in the code for capturing groups above about handling
1252     THEN. */
1253    
1254     if (rrc == MATCH_THEN)
1255     {
1256     next = ecode + GET(ecode,1);
1257 ph10 733 if (md->start_match_ptr < next &&
1258 ph10 716 (*ecode == OP_ALT || *next == OP_ALT))
1259     rrc = MATCH_NOMATCH;
1260 ph10 733 }
1261 ph10 716
1262     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1263 ph10 604 ecode += GET(ecode, 1);
1264 ph10 625 if (*ecode != OP_ALT) break;
1265 ph10 1248 md->capture_last = save_capture_last;
1266 ph10 604 }
1267 ph10 625
1268     if (matched_once || allow_zero)
1269 ph10 604 {
1270     ecode += 1 + LINK_SIZE;
1271     break;
1272 ph10 625 }
1273 ph10 604 RRETURN(MATCH_NOMATCH);
1274    
1275     /* Control never reaches here. */
1276    
1277 nigel 77 /* Conditional group: compilation checked that there are no more than
1278     two branches. If the condition is false, skipping the first branch takes us
1279     past the end if there is only one branch, but that's OK because that is
1280 ph10 609 exactly what going to the ket would do. */
1281 nigel 77
1282     case OP_COND:
1283 nigel 93 case OP_SCOND:
1284 ph10 604 codelink = GET(ecode, 1);
1285 ph10 406
1286 ph10 381 /* Because of the way auto-callout works during compile, a callout item is
1287     inserted between OP_COND and an assertion condition. */
1288 ph10 392
1289 ph10 381 if (ecode[LINK_SIZE+1] == OP_CALLOUT)
1290     {
1291 ph10 836 if (PUBL(callout) != NULL)
1292 ph10 381 {
1293 zherczeg 850 PUBL(callout_block) cb;
1294 ph10 645 cb.version = 2; /* Version 1 of the callout block */
1295 ph10 381 cb.callout_number = ecode[LINK_SIZE+2];
1296     cb.offset_vector = md->offset_vector;
1297 chpe 1055 #if defined COMPILE_PCRE8
1298 ph10 381 cb.subject = (PCRE_SPTR)md->start_subject;
1299 chpe 1055 #elif defined COMPILE_PCRE16
1300 zherczeg 852 cb.subject = (PCRE_SPTR16)md->start_subject;
1301 chpe 1055 #elif defined COMPILE_PCRE32
1302     cb.subject = (PCRE_SPTR32)md->start_subject;
1303 zherczeg 852 #endif
1304 ph10 530 cb.subject_length = (int)(md->end_subject - md->start_subject);
1305     cb.start_match = (int)(mstart - md->start_subject);
1306     cb.current_position = (int)(eptr - md->start_subject);
1307 ph10 381 cb.pattern_position = GET(ecode, LINK_SIZE + 3);
1308     cb.next_item_length = GET(ecode, 3 + 2*LINK_SIZE);
1309     cb.capture_top = offset_top/2;
1310 ph10 1248 cb.capture_last = md->capture_last & CAPLMASK;
1311 ph10 1271 /* Internal change requires this for API compatibility. */
1312     if (cb.capture_last == 0) cb.capture_last = -1;
1313 ph10 381 cb.callout_data = md->callout_data;
1314 ph10 836 cb.mark = md->nomatch_mark;
1315     if ((rrc = (*PUBL(callout))(&cb)) > 0) RRETURN(MATCH_NOMATCH);
1316 ph10 381 if (rrc < 0) RRETURN(rrc);
1317     }
1318 ph10 836 ecode += PRIV(OP_lengths)[OP_CALLOUT];
1319 ph10 1271 codelink -= PRIV(OP_lengths)[OP_CALLOUT];
1320 ph10 381 }
1321 ph10 392
1322 ph10 399 condcode = ecode[LINK_SIZE+1];
1323 ph10 406
1324 ph10 381 /* Now see what the actual condition is */
1325 ph10 392
1326 ph10 459 if (condcode == OP_RREF || condcode == OP_NRREF) /* Recursion test */
1327 nigel 77 {
1328 ph10 459 if (md->recursive == NULL) /* Not recursing => FALSE */
1329     {
1330 ph10 461 condition = FALSE;
1331     ecode += GET(ecode, 1);
1332     }
1333 ph10 459 else
1334 ph10 461 {
1335 ph10 1145 unsigned int recno = GET2(ecode, LINK_SIZE + 2); /* Recursion group number*/
1336 ph10 751 condition = (recno == RREF_ANY || recno == md->recursive->group_num);
1337 ph10 461
1338 ph10 459 /* If the test is for recursion into a specific subpattern, and it is
1339     false, but the test was set up by name, scan the table to see if the
1340     name refers to any other numbers, and test them. The condition is true
1341     if any one is set. */
1342 ph10 461
1343 ph10 751 if (!condition && condcode == OP_NRREF)
1344 ph10 459 {
1345 ph10 836 pcre_uchar *slotA = md->name_table;
1346 ph10 459 for (i = 0; i < md->name_count; i++)
1347 ph10 461 {
1348     if (GET2(slotA, 0) == recno) break;
1349 ph10 459 slotA += md->name_entry_size;
1350     }
1351 ph10 461
1352 ph10 459 /* Found a name for the number - there can be only one; duplicate
1353     names for different numbers are allowed, but not vice versa. First
1354     scan down for duplicates. */
1355 ph10 461
1356 ph10 459 if (i < md->name_count)
1357 ph10 461 {
1358 ph10 836 pcre_uchar *slotB = slotA;
1359 ph10 459 while (slotB > md->name_table)
1360     {
1361     slotB -= md->name_entry_size;
1362 ph10 836 if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
1363 ph10 459 {
1364     condition = GET2(slotB, 0) == md->recursive->group_num;
1365 ph10 461 if (condition) break;
1366     }
1367 ph10 459 else break;
1368 ph10 461 }
1369    
1370 ph10 459 /* Scan up for duplicates */
1371 ph10 461
1372 ph10 459 if (!condition)
1373 ph10 461 {
1374 ph10 459 slotB = slotA;
1375     for (i++; i < md->name_count; i++)
1376     {
1377     slotB += md->name_entry_size;
1378 ph10 836 if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
1379 ph10 459 {
1380     condition = GET2(slotB, 0) == md->recursive->group_num;
1381     if (condition) break;
1382 ph10 461 }
1383 ph10 459 else break;
1384 ph10 461 }
1385     }
1386 ph10 459 }
1387 ph10 461 }
1388    
1389 ph10 459 /* Chose branch according to the condition */
1390 ph10 461
1391 ph10 836 ecode += condition? 1 + IMM2_SIZE : GET(ecode, 1);
1392 ph10 459 }
1393 ph10 461 }
1394 nigel 93
1395 ph10 459 else if (condcode == OP_CREF || condcode == OP_NCREF) /* Group used test */
1396 nigel 93 {
1397 nigel 77 offset = GET2(ecode, LINK_SIZE+2) << 1; /* Doubled ref number */
1398 nigel 93 condition = offset < offset_top && md->offset_vector[offset] >= 0;
1399 ph10 461
1400 ph10 459 /* If the numbered capture is unset, but the reference was by name,
1401 ph10 461 scan the table to see if the name refers to any other numbers, and test
1402     them. The condition is true if any one is set. This is tediously similar
1403     to the code above, but not close enough to try to amalgamate. */
1404    
1405 ph10 459 if (!condition && condcode == OP_NCREF)
1406     {
1407 ph10 1145 unsigned int refno = offset >> 1;
1408 ph10 836 pcre_uchar *slotA = md->name_table;
1409 ph10 461
1410 ph10 459 for (i = 0; i < md->name_count; i++)
1411 ph10 461 {
1412     if (GET2(slotA, 0) == refno) break;
1413 ph10 459 slotA += md->name_entry_size;
1414     }
1415 ph10 461
1416     /* Found a name for the number - there can be only one; duplicate names
1417     for different numbers are allowed, but not vice versa. First scan down
1418 ph10 459 for duplicates. */
1419 ph10 461
1420 ph10 459 if (i < md->name_count)
1421 ph10 461 {
1422 ph10 836 pcre_uchar *slotB = slotA;
1423 ph10 459 while (slotB > md->name_table)
1424     {
1425     slotB -= md->name_entry_size;
1426 ph10 836 if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
1427 ph10 459 {
1428     offset = GET2(slotB, 0) << 1;
1429 ph10 461 condition = offset < offset_top &&
1430 ph10 459 md->offset_vector[offset] >= 0;
1431 ph10 461 if (condition) break;
1432     }
1433 ph10 459 else break;
1434 ph10 461 }
1435    
1436 ph10 459 /* Scan up for duplicates */
1437 ph10 461
1438 ph10 459 if (!condition)
1439 ph10 461 {
1440 ph10 459 slotB = slotA;
1441     for (i++; i < md->name_count; i++)
1442     {
1443     slotB += md->name_entry_size;
1444 ph10 836 if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
1445 ph10 459 {
1446     offset = GET2(slotB, 0) << 1;
1447 ph10 461 condition = offset < offset_top &&
1448 ph10 459 md->offset_vector[offset] >= 0;
1449 ph10 461 if (condition) break;
1450     }
1451 ph10 459 else break;
1452 ph10 461 }
1453     }
1454 ph10 459 }
1455 ph10 461 }
1456    
1457 ph10 459 /* Chose branch according to the condition */
1458    
1459 ph10 836 ecode += condition? 1 + IMM2_SIZE : GET(ecode, 1);
1460 nigel 77 }
1461    
1462 ph10 399 else if (condcode == OP_DEF) /* DEFINE - always false */
1463 nigel 93 {
1464     condition = FALSE;
1465     ecode += GET(ecode, 1);
1466     }
1467    
1468 nigel 77 /* The condition is an assertion. Call match() to evaluate it - setting
1469 ph10 604 md->match_function_type to MATCH_CONDASSERT causes it to stop at the end of
1470     an assertion. */
1471 nigel 77
1472     else
1473     {
1474 ph10 625 md->match_function_type = MATCH_CONDASSERT;
1475 ph10 604 RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM3);
1476 nigel 77 if (rrc == MATCH_MATCH)
1477     {
1478 ph10 619 if (md->end_offset_top > offset_top)
1479     offset_top = md->end_offset_top; /* Captures may have happened */
1480 nigel 93 condition = TRUE;
1481     ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE + 2);
1482 nigel 77 while (*ecode == OP_ALT) ecode += GET(ecode, 1);
1483     }
1484 ph10 733
1485 ph10 716 /* PCRE doesn't allow the effect of (*THEN) to escape beyond an
1486 ph10 733 assertion; it is therefore treated as NOMATCH. */
1487 ph10 716
1488 ph10 733 else if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN)
1489 nigel 77 {
1490     RRETURN(rrc); /* Need braces because of following else */
1491     }
1492 nigel 93 else
1493     {
1494     condition = FALSE;
1495 ph10 399 ecode += codelink;
1496 nigel 93 }
1497     }
1498 nigel 91
1499 ph10 716 /* We are now at the branch that is to be obeyed. As there is only one, can
1500     use tail recursion to avoid using another stack frame, except when there is
1501     unlimited repeat of a possibly empty group. In the latter case, a recursive
1502     call to match() is always required, unless the second alternative doesn't
1503     exist, in which case we can just plough on. Note that, for compatibility
1504     with Perl, the | in a conditional group is NOT treated as creating two
1505     alternatives. If a THEN is encountered in the branch, it propagates out to
1506     the enclosing alternative (unless nested in a deeper set of alternatives,
1507     of course). */
1508 nigel 91
1509 nigel 93 if (condition || *ecode == OP_ALT)
1510     {
1511 ph10 716 if (op != OP_SCOND)
1512 ph10 702 {
1513     ecode += 1 + LINK_SIZE;
1514     goto TAIL_RECURSE;
1515 ph10 708 }
1516 ph10 733
1517 ph10 716 md->match_function_type = MATCH_CBEGROUP;
1518 ph10 609 RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM49);
1519     RRETURN(rrc);
1520 nigel 77 }
1521 ph10 708
1522 ph10 702 /* Condition false & no alternative; continue after the group. */
1523 ph10 708
1524 ph10 702 else
1525 nigel 93 {
1526     ecode += 1 + LINK_SIZE;
1527     }
1528     break;
1529 nigel 77
1530 ph10 461
1531 ph10 447 /* Before OP_ACCEPT there may be any number of OP_CLOSE opcodes,
1532     to close any currently open capturing brackets. */
1533 ph10 461
1534 ph10 447 case OP_CLOSE:
1535 ph10 1248 number = GET2(ecode, 1); /* Must be less than 65536 */
1536 ph10 447 offset = number << 1;
1537 ph10 461
1538 ph10 475 #ifdef PCRE_DEBUG
1539 ph10 447 printf("end bracket %d at *ACCEPT", number);
1540     printf("\n");
1541     #endif
1542 nigel 77
1543 ph10 1248 md->capture_last = (md->capture_last & OVFLMASK) | number;
1544     if (offset >= md->offset_max) md->capture_last |= OVFLBIT; else
1545 ph10 447 {
1546     md->offset_vector[offset] =
1547     md->offset_vector[md->offset_end - number];
1548 ph10 530 md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
1549 ph10 447 if (offset_top <= offset) offset_top = offset + 2;
1550     }
1551 ph10 836 ecode += 1 + IMM2_SIZE;
1552 ph10 461 break;
1553 ph10 447
1554    
1555 ph10 619 /* End of the pattern, either real or forced. */
1556 nigel 77
1557 ph10 619 case OP_END:
1558 ph10 210 case OP_ACCEPT:
1559 ph10 625 case OP_ASSERT_ACCEPT:
1560    
1561 ph10 619 /* If we have matched an empty string, fail if not in an assertion and not
1562     in a recursion if either PCRE_NOTEMPTY is set, or if PCRE_NOTEMPTY_ATSTART
1563 ph10 613 is set and we have matched at the start of the subject. In both cases,
1564     backtracking will then try other alternatives, if any. */
1565 ph10 443
1566 ph10 619 if (eptr == mstart && op != OP_ASSERT_ACCEPT &&
1567 ph10 618 md->recursive == NULL &&
1568 ph10 619 (md->notempty ||
1569     (md->notempty_atstart &&
1570     mstart == md->start_subject + md->start_offset)))
1571 ph10 836 RRETURN(MATCH_NOMATCH);
1572 ph10 443
1573 ph10 442 /* Otherwise, we have a match. */
1574 ph10 625
1575 ph10 168 md->end_match_ptr = eptr; /* Record where we ended */
1576     md->end_offset_top = offset_top; /* and how many extracts were taken */
1577 ph10 210 md->start_match_ptr = mstart; /* and the start (\K can modify) */
1578 nigel 77
1579 ph10 512 /* For some reason, the macros don't work properly if an expression is
1580 ph10 836 given as the argument to RRETURN when the heap is in use. */
1581 ph10 512
1582     rrc = (op == OP_END)? MATCH_MATCH : MATCH_ACCEPT;
1583 ph10 836 RRETURN(rrc);
1584 ph10 512
1585 nigel 77 /* Assertion brackets. Check the alternative branches in turn - the
1586     matching won't pass the KET for an assertion. If any one branch matches,
1587     the assertion is true. Lookbehind assertions have an OP_REVERSE item at the
1588     start of each branch to move the current point backwards, so the code at
1589 ph10 625 this level is identical to the lookahead case. When the assertion is part
1590     of a condition, we want to return immediately afterwards. The caller of
1591     this incarnation of the match() function will have set MATCH_CONDASSERT in
1592     md->match_function type, and one of these opcodes will be the first opcode
1593     that is processed. We use a local variable that is preserved over calls to
1594 ph10 604 match() to remember this case. */
1595 nigel 77
1596     case OP_ASSERT:
1597     case OP_ASSERTBACK:
1598 ph10 903 save_mark = md->mark;
1599 ph10 604 if (md->match_function_type == MATCH_CONDASSERT)
1600     {
1601     condassert = TRUE;
1602     md->match_function_type = 0;
1603     }
1604 ph10 625 else condassert = FALSE;
1605    
1606 ph10 1296 /* Loop for each branch */
1607 ph10 1320
1608 nigel 77 do
1609     {
1610 ph10 604 RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM4);
1611 ph10 1320
1612 ph10 1302 /* A match means that the assertion is true; break out of the loop
1613     that matches its alternatives. */
1614 ph10 1320
1615 ph10 511 if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT)
1616 ph10 500 {
1617     mstart = md->start_match_ptr; /* In case \K reset it */
1618     break;
1619 ph10 501 }
1620 ph10 1320
1621 ph10 1302 /* If not matched, restore the previous mark setting. */
1622 ph10 1320
1623 ph10 940 md->mark = save_mark;
1624 ph10 733
1625 ph10 1296 /* See comment in the code for capturing groups above about handling
1626     THEN. */
1627 ph10 975
1628 ph10 1296 if (rrc == MATCH_THEN)
1629     {
1630     next = ecode + GET(ecode,1);
1631     if (md->start_match_ptr < next &&
1632     (*ecode == OP_ALT || *next == OP_ALT))
1633     rrc = MATCH_NOMATCH;
1634     }
1635 ph10 1320
1636 ph10 1302 /* Anything other than NOMATCH causes the entire assertion to fail,
1637     passing back the return code. This includes COMMIT, SKIP, PRUNE and an
1638     uncaptured THEN, which means they take their normal effect. This
1639     consistent approach does not always have exactly the same effect as in
1640     Perl. */
1641 ph10 940
1642 ph10 1296 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1643 nigel 77 ecode += GET(ecode, 1);
1644     }
1645 ph10 1302 while (*ecode == OP_ALT); /* Continue for next alternative */
1646 ph10 1320
1647 ph10 1296 /* If we have tried all the alternative branches, the assertion has
1648 ph10 1320 failed. If not, we broke out after a match. */
1649 ph10 625
1650 ph10 836 if (*ecode == OP_KET) RRETURN(MATCH_NOMATCH);
1651 nigel 77
1652     /* If checking an assertion for a condition, return MATCH_MATCH. */
1653    
1654 ph10 604 if (condassert) RRETURN(MATCH_MATCH);
1655 nigel 77
1656 ph10 1296 /* Continue from after a successful assertion, updating the offsets high
1657     water mark, since extracts may have been taken during the assertion. */
1658 nigel 77
1659     do ecode += GET(ecode,1); while (*ecode == OP_ALT);
1660     ecode += 1 + LINK_SIZE;
1661     offset_top = md->end_offset_top;
1662     continue;
1663    
1664 ph10 1320 /* Negative assertion: all branches must fail to match for the assertion to
1665 ph10 1296 succeed. */
1666 nigel 77
1667     case OP_ASSERT_NOT:
1668     case OP_ASSERTBACK_NOT:
1669 ph10 903 save_mark = md->mark;
1670 ph10 604 if (md->match_function_type == MATCH_CONDASSERT)
1671     {
1672     condassert = TRUE;
1673     md->match_function_type = 0;
1674     }
1675 ph10 625 else condassert = FALSE;
1676 ph10 604
1677 ph10 1296 /* Loop for each alternative branch. */
1678 ph10 1320
1679 nigel 77 do
1680     {
1681 ph10 604 RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM5);
1682 ph10 1302 md->mark = save_mark; /* Always restore the mark setting */
1683 ph10 1320
1684 ph10 1302 switch(rrc)
1685     {
1686     case MATCH_MATCH: /* A successful match means */
1687     case MATCH_ACCEPT: /* the assertion has failed. */
1688     RRETURN(MATCH_NOMATCH);
1689 ph10 1320
1690 ph10 1302 case MATCH_NOMATCH: /* Carry on with next branch */
1691 ph10 1320 break;
1692 ph10 1296
1693 ph10 1302 /* See comment in the code for capturing groups above about handling
1694     THEN. */
1695 ph10 1296
1696 ph10 1302 case MATCH_THEN:
1697 ph10 1296 next = ecode + GET(ecode,1);
1698     if (md->start_match_ptr < next &&
1699     (*ecode == OP_ALT || *next == OP_ALT))
1700 ph10 1320 {
1701 ph10 1296 rrc = MATCH_NOMATCH;
1702 ph10 1302 break;
1703     }
1704 ph10 1320 /* Otherwise fall through. */
1705    
1706 ph10 1302 /* COMMIT, SKIP, PRUNE, and an uncaptured THEN cause the whole
1707     assertion to fail to match, without considering any more alternatives.
1708     Failing to match means the assertion is true. This is a consistent
1709     approach, but does not always have the same effect as in Perl. */
1710    
1711     case MATCH_COMMIT:
1712     case MATCH_SKIP:
1713 ph10 1320 case MATCH_SKIP_ARG:
1714 ph10 1302 case MATCH_PRUNE:
1715     do ecode += GET(ecode,1); while (*ecode == OP_ALT);
1716     goto NEG_ASSERT_TRUE; /* Break out of alternation loop */
1717 ph10 1320
1718 ph10 1302 /* Anything else is an error */
1719 ph10 1320
1720 ph10 1302 default:
1721 ph10 1320 RRETURN(rrc);
1722 ph10 482 }
1723 ph10 1320
1724 ph10 1302 /* Continue with next branch */
1725 ph10 1320
1726 nigel 77 ecode += GET(ecode,1);
1727     }
1728     while (*ecode == OP_ALT);
1729 ph10 1320
1730 ph10 1296 /* All branches in the assertion failed to match. */
1731 ph10 1320
1732 ph10 1302 NEG_ASSERT_TRUE:
1733 ph10 604 if (condassert) RRETURN(MATCH_MATCH); /* Condition assertion */
1734 ph10 1296 ecode += 1 + LINK_SIZE; /* Continue with current branch */
1735 nigel 77 continue;
1736    
1737     /* Move the subject pointer back. This occurs only at the start of
1738     each branch of a lookbehind assertion. If we are too close to the start to
1739     move back, this match function fails. When working with UTF-8 we move
1740     back a number of characters, not bytes. */
1741    
1742     case OP_REVERSE:
1743 ph10 836 #ifdef SUPPORT_UTF
1744     if (utf)
1745 nigel 77 {
1746 nigel 93 i = GET(ecode, 1);
1747     while (i-- > 0)
1748 nigel 77 {
1749     eptr--;
1750 ph10 836 if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
1751 ph10 207 BACKCHAR(eptr);
1752 nigel 77 }
1753     }
1754     else
1755     #endif
1756    
1757     /* No UTF-8 support, or not in UTF-8 mode: count is byte count */
1758    
1759     {
1760 nigel 93 eptr -= GET(ecode, 1);
1761 ph10 836 if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
1762 nigel 77 }
1763    
1764 ph10 435 /* Save the earliest consulted character, then skip to next op code */
1765 nigel 77
1766 ph10 435 if (eptr < md->start_used_ptr) md->start_used_ptr = eptr;
1767 nigel 77 ecode += 1 + LINK_SIZE;
1768     break;
1769    
1770     /* The callout item calls an external function, if one is provided, passing
1771     details of the match so far. This is mainly for debugging, though the
1772     function is able to force a failure. */
1773    
1774     case OP_CALLOUT:
1775 ph10 836 if (PUBL(callout) != NULL)
1776 nigel 77 {
1777 zherczeg 850 PUBL(callout_block) cb;
1778 ph10 645 cb.version = 2; /* Version 1 of the callout block */
1779 nigel 77 cb.callout_number = ecode[1];
1780     cb.offset_vector = md->offset_vector;
1781 chpe 1055 #if defined COMPILE_PCRE8
1782 nigel 87 cb.subject = (PCRE_SPTR)md->start_subject;
1783 chpe 1055 #elif defined COMPILE_PCRE16
1784 zherczeg 852 cb.subject = (PCRE_SPTR16)md->start_subject;
1785 chpe 1055 #elif defined COMPILE_PCRE32
1786     cb.subject = (PCRE_SPTR32)md->start_subject;
1787 zherczeg 852 #endif
1788 ph10 530 cb.subject_length = (int)(md->end_subject - md->start_subject);
1789     cb.start_match = (int)(mstart - md->start_subject);
1790     cb.current_position = (int)(eptr - md->start_subject);
1791 nigel 77 cb.pattern_position = GET(ecode, 2);
1792     cb.next_item_length = GET(ecode, 2 + LINK_SIZE);
1793     cb.capture_top = offset_top/2;
1794 ph10 1248 cb.capture_last = md->capture_last & CAPLMASK;
1795 ph10 1271 /* Internal change requires this for API compatibility. */
1796     if (cb.capture_last == 0) cb.capture_last = -1;
1797 nigel 77 cb.callout_data = md->callout_data;
1798 ph10 836 cb.mark = md->nomatch_mark;
1799     if ((rrc = (*PUBL(callout))(&cb)) > 0) RRETURN(MATCH_NOMATCH);
1800 nigel 77 if (rrc < 0) RRETURN(rrc);
1801     }
1802     ecode += 2 + 2*LINK_SIZE;
1803     break;
1804    
1805     /* Recursion either matches the current regex, or some subexpression. The
1806     offset data is the offset to the starting bracket from the start of the
1807     whole pattern. (This is so that it works from duplicated subpatterns.)
1808 ph10 625
1809 ph10 618 The state of the capturing groups is preserved over recursion, and
1810 ph10 625 re-instated afterwards. We don't know how many are started and not yet
1811 ph10 618 finished (offset_top records the completed total) so we just have to save
1812     all the potential data. There may be up to 65535 such values, which is too
1813     large to put on the stack, but using malloc for small numbers seems
1814     expensive. As a compromise, the stack is used when there are no more than
1815     REC_STACK_SAVE_MAX values to store; otherwise malloc is used.
1816 nigel 77
1817     There are also other values that have to be saved. We use a chained
1818     sequence of blocks that actually live on the stack. Thanks to Robin Houston
1819 ph10 625 for the original version of this logic. It has, however, been hacked around
1820 ph10 618 a lot, so he is not to blame for the current way it works. */
1821 nigel 77
1822     case OP_RECURSE:
1823     {
1824 ph10 642 recursion_info *ri;
1825 ph10 1145 unsigned int recno;
1826 ph10 654
1827 nigel 77 callpat = md->start_code + GET(ecode, 1);
1828 ph10 642 recno = (callpat == md->start_code)? 0 :
1829 ph10 654 GET2(callpat, 1 + LINK_SIZE);
1830    
1831     /* Check for repeating a recursion without advancing the subject pointer.
1832 ph10 642 This should catch convoluted mutual recursions. (Some simple cases are
1833 ph10 654 caught at compile time.) */
1834    
1835 ph10 642 for (ri = md->recursive; ri != NULL; ri = ri->prevrec)
1836 ph10 654 if (recno == ri->group_num && eptr == ri->subject_position)
1837 ph10 642 RRETURN(PCRE_ERROR_RECURSELOOP);
1838 nigel 77
1839     /* Add to "recursing stack" */
1840    
1841 ph10 642 new_recursive.group_num = recno;
1842 ph10 1271 new_recursive.saved_capture_last = md->capture_last;
1843 ph10 642 new_recursive.subject_position = eptr;
1844 nigel 77 new_recursive.prevrec = md->recursive;
1845     md->recursive = &new_recursive;
1846    
1847 ph10 618 /* Where to continue from afterwards */
1848 nigel 77
1849     ecode += 1 + LINK_SIZE;
1850    
1851 ph10 618 /* Now save the offset data */
1852 nigel 77
1853     new_recursive.saved_max = md->offset_end;
1854     if (new_recursive.saved_max <= REC_STACK_SAVE_MAX)
1855     new_recursive.offset_save = stacksave;
1856     else
1857     {
1858     new_recursive.offset_save =
1859 ph10 836 (int *)(PUBL(malloc))(new_recursive.saved_max * sizeof(int));
1860 nigel 77 if (new_recursive.offset_save == NULL) RRETURN(PCRE_ERROR_NOMEMORY);
1861     }
1862     memcpy(new_recursive.offset_save, md->offset_vector,
1863     new_recursive.saved_max * sizeof(int));
1864 ph10 625
1865 ph10 618 /* OK, now we can do the recursion. After processing each alternative,
1866 ph10 1248 restore the offset data and the last captured value. If there were nested
1867     recursions, md->recursive might be changed, so reset it before looping.
1868     */
1869 nigel 77
1870     DPRINTF(("Recursing into group %d\n", new_recursive.group_num));
1871 ph10 604 cbegroup = (*callpat >= OP_SBRA);
1872 nigel 77 do
1873     {
1874 ph10 604 if (cbegroup) md->match_function_type = MATCH_CBEGROUP;
1875 ph10 836 RMATCH(eptr, callpat + PRIV(OP_lengths)[*callpat], offset_top,
1876 ph10 604 md, eptrb, RM6);
1877 ph10 618 memcpy(md->offset_vector, new_recursive.offset_save,
1878     new_recursive.saved_max * sizeof(int));
1879 ph10 1271 md->capture_last = new_recursive.saved_capture_last;
1880 ph10 681 md->recursive = new_recursive.prevrec;
1881 ph10 511 if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT)
1882 nigel 77 {
1883 nigel 87 DPRINTF(("Recursion matched\n"));
1884 nigel 77 if (new_recursive.offset_save != stacksave)
1885 ph10 836 (PUBL(free))(new_recursive.offset_save);
1886 ph10 618
1887     /* Set where we got to in the subject, and reset the start in case
1888 ph10 625 it was changed by \K. This *is* propagated back out of a recursion,
1889     for Perl compatibility. */
1890    
1891 ph10 618 eptr = md->end_match_ptr;
1892     mstart = md->start_match_ptr;
1893     goto RECURSION_MATCHED; /* Exit loop; end processing */
1894 nigel 77 }
1895 ph10 716
1896 ph10 1271 /* PCRE does not allow THEN, SKIP, PRUNE or COMMIT to escape beyond a
1897 ph10 1298 recursion; they cause a NOMATCH for the entire recursion. These codes
1898     are defined in a range that can be tested for. */
1899 ph10 1320
1900 ph10 1298 if (rrc >= MATCH_BACKTRACK_MIN && rrc <= MATCH_BACKTRACK_MAX)
1901 ph10 1320 RRETURN(MATCH_NOMATCH);
1902    
1903 ph10 1298 /* Any return code other than NOMATCH is an error. */
1904 ph10 716
1905 ph10 1298 if (rrc != MATCH_NOMATCH)
1906 nigel 87 {
1907     DPRINTF(("Recursion gave error %d\n", rrc));
1908 ph10 400 if (new_recursive.offset_save != stacksave)
1909 ph10 836 (PUBL(free))(new_recursive.offset_save);
1910 nigel 87 RRETURN(rrc);
1911     }
1912 nigel 77
1913     md->recursive = &new_recursive;
1914     callpat += GET(callpat, 1);
1915     }
1916     while (*callpat == OP_ALT);
1917    
1918     DPRINTF(("Recursion didn't match\n"));
1919     md->recursive = new_recursive.prevrec;
1920     if (new_recursive.offset_save != stacksave)
1921 ph10 836 (PUBL(free))(new_recursive.offset_save);
1922     RRETURN(MATCH_NOMATCH);
1923 nigel 77 }
1924 ph10 625
1925 ph10 618 RECURSION_MATCHED:
1926     break;
1927 nigel 77
1928     /* An alternation is the end of a branch; scan along to find the end of the
1929     bracketed group and go to there. */
1930    
1931     case OP_ALT:
1932     do ecode += GET(ecode,1); while (*ecode == OP_ALT);
1933     break;
1934    
1935 ph10 335 /* BRAZERO, BRAMINZERO and SKIPZERO occur just before a bracket group,
1936     indicating that it may occur zero times. It may repeat infinitely, or not
1937     at all - i.e. it could be ()* or ()? or even (){0} in the pattern. Brackets
1938     with fixed upper repeat limits are compiled as a number of copies, with the
1939     optional ones preceded by BRAZERO or BRAMINZERO. */
1940 ph10 625
1941 nigel 77 case OP_BRAZERO:
1942 ph10 604 next = ecode + 1;
1943     RMATCH(eptr, next, offset_top, md, eptrb, RM10);
1944     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1945     do next += GET(next, 1); while (*next == OP_ALT);
1946     ecode = next + 1 + LINK_SIZE;
1947 nigel 77 break;
1948 ph10 625
1949 nigel 77 case OP_BRAMINZERO:
1950 ph10 604 next = ecode + 1;
1951     do next += GET(next, 1); while (*next == OP_ALT);
1952     RMATCH(eptr, next + 1+LINK_SIZE, offset_top, md, eptrb, RM11);
1953     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1954     ecode++;
1955 nigel 77 break;
1956    
1957 ph10 335 case OP_SKIPZERO:
1958 ph10 604 next = ecode+1;
1959     do next += GET(next,1); while (*next == OP_ALT);
1960     ecode = next + 1 + LINK_SIZE;
1961 ph10 335 break;
1962 ph10 625
1963 ph10 604 /* BRAPOSZERO occurs before a possessive bracket group. Don't do anything
1964     here; just jump to the group, with allow_zero set TRUE. */
1965 ph10 625
1966 ph10 604 case OP_BRAPOSZERO:
1967 ph10 625 op = *(++ecode);
1968 ph10 604 allow_zero = TRUE;
1969     if (op == OP_CBRAPOS || op == OP_SCBRAPOS) goto POSSESSIVE_CAPTURE;
1970     goto POSSESSIVE_NON_CAPTURE;
1971 ph10 335
1972 nigel 93 /* End of a group, repeated or non-repeating. */
1973 nigel 77
1974     case OP_KET:
1975     case OP_KETRMIN:
1976     case OP_KETRMAX:
1977 ph10 625 case OP_KETRPOS:
1978 nigel 91 prev = ecode - GET(ecode, 1);
1979 ph10 625
1980 nigel 93 /* If this was a group that remembered the subject start, in order to break
1981     infinite repeats of empty string matches, retrieve the subject start from
1982     the chain. Otherwise, set it NULL. */
1983 nigel 77
1984 ph10 618 if (*prev >= OP_SBRA || *prev == OP_ONCE)
1985 nigel 93 {
1986     saved_eptr = eptrb->epb_saved_eptr; /* Value at start of group */
1987     eptrb = eptrb->epb_prev; /* Backup to previous group */
1988     }
1989     else saved_eptr = NULL;
1990 nigel 77
1991 ph10 733 /* If we are at the end of an assertion group or a non-capturing atomic
1992 ph10 723 group, stop matching and return MATCH_MATCH, but record the current high
1993     water mark for use by positive assertions. We also need to record the match
1994     start in case it was changed by \K. */
1995 nigel 93
1996 ph10 723 if ((*prev >= OP_ASSERT && *prev <= OP_ASSERTBACK_NOT) ||
1997 ph10 733 *prev == OP_ONCE_NC)
1998 nigel 91 {
1999 ph10 723 md->end_match_ptr = eptr; /* For ONCE_NC */
2000 nigel 91 md->end_offset_top = offset_top;
2001 ph10 500 md->start_match_ptr = mstart;
2002 ph10 836 RRETURN(MATCH_MATCH); /* Sets md->mark */
2003 nigel 91 }
2004 nigel 77
2005 nigel 93 /* For capturing groups we have to check the group number back at the start
2006     and if necessary complete handling an extraction by setting the offsets and
2007 ph10 618 bumping the high water mark. Whole-pattern recursion is coded as a recurse
2008     into group 0, so it won't be picked up here. Instead, we catch it when the
2009     OP_END is reached. Other recursion is handled here. We just have to record
2010     the current subject position and start match pointer and give a MATCH
2011     return. */
2012 nigel 77
2013 ph10 604 if (*prev == OP_CBRA || *prev == OP_SCBRA ||
2014     *prev == OP_CBRAPOS || *prev == OP_SCBRAPOS)
2015 nigel 91 {
2016 nigel 93 number = GET2(prev, 1+LINK_SIZE);
2017 nigel 91 offset = number << 1;
2018 ph10 461
2019 ph10 475 #ifdef PCRE_DEBUG
2020 nigel 91 printf("end bracket %d", number);
2021     printf("\n");
2022 nigel 77 #endif
2023    
2024 ph10 618 /* Handle a recursively called group. */
2025    
2026     if (md->recursive != NULL && md->recursive->group_num == number)
2027     {
2028     md->end_match_ptr = eptr;
2029     md->start_match_ptr = mstart;
2030     RRETURN(MATCH_MATCH);
2031     }
2032    
2033     /* Deal with capturing */
2034    
2035 ph10 1248 md->capture_last = (md->capture_last & OVFLMASK) | number;
2036     if (offset >= md->offset_max) md->capture_last |= OVFLBIT; else
2037 nigel 91 {
2038 ph10 625 /* If offset is greater than offset_top, it means that we are
2039     "skipping" a capturing group, and that group's offsets must be marked
2040     unset. In earlier versions of PCRE, all the offsets were unset at the
2041     start of matching, but this doesn't work because atomic groups and
2042 ph10 615 assertions can cause a value to be set that should later be unset.
2043     Example: matching /(?>(a))b|(a)c/ against "ac". This sets group 1 as
2044 ph10 625 part of the atomic group, but this is not on the final matching path,
2045     so must be unset when 2 is set. (If there is no group 2, there is no
2046 ph10 615 problem, because offset_top will then be 2, indicating no capture.) */
2047 ph10 625
2048 ph10 615 if (offset > offset_top)
2049     {
2050     register int *iptr = md->offset_vector + offset_top;
2051     register int *iend = md->offset_vector + offset;
2052     while (iptr < iend) *iptr++ = -1;
2053 ph10 625 }
2054    
2055 ph10 615 /* Now make the extraction */
2056    
2057 nigel 93 md->offset_vector[offset] =
2058     md->offset_vector[md->offset_end - number];
2059 ph10 530 md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
2060 nigel 93 if (offset_top <= offset) offset_top = offset + 2;
2061     }
2062 nigel 91 }
2063 nigel 77
2064 ph10 618 /* For an ordinary non-repeating ket, just continue at this level. This
2065     also happens for a repeating ket if no characters were matched in the
2066     group. This is the forcible breaking of infinite loops as implemented in
2067 ph10 723 Perl 5.005. For a non-repeating atomic group that includes captures,
2068     establish a backup point by processing the rest of the pattern at a lower
2069     level. If this results in a NOMATCH return, pass MATCH_ONCE back to the
2070     original OP_ONCE level, thereby bypassing intermediate backup points, but
2071     resetting any captures that happened along the way. */
2072 nigel 77
2073 nigel 91 if (*ecode == OP_KET || eptr == saved_eptr)
2074     {
2075 ph10 618 if (*prev == OP_ONCE)
2076     {
2077     RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM12);
2078     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2079     md->once_target = prev; /* Level at which to change to MATCH_NOMATCH */
2080 ph10 625 RRETURN(MATCH_ONCE);
2081     }
2082 ph10 618 ecode += 1 + LINK_SIZE; /* Carry on at this level */
2083 nigel 91 break;
2084     }
2085 ph10 625
2086     /* OP_KETRPOS is a possessive repeating ket. Remember the current position,
2087 ph10 604 and return the MATCH_KETRPOS. This makes it possible to do the repeats one
2088     at a time from the outer level, thus saving stack. */
2089 ph10 625
2090 ph10 604 if (*ecode == OP_KETRPOS)
2091 ph10 625 {
2092 ph10 604 md->end_match_ptr = eptr;
2093 ph10 625 md->end_offset_top = offset_top;
2094 ph10 604 RRETURN(MATCH_KETRPOS);
2095 ph10 625 }
2096 nigel 77
2097 ph10 604 /* The normal repeating kets try the rest of the pattern or restart from
2098     the preceding bracket, in the appropriate order. In the second case, we can
2099     use tail recursion to avoid using another stack frame, unless we have an
2100 ph10 618 an atomic group or an unlimited repeat of a group that can match an empty
2101     string. */
2102 nigel 77
2103 nigel 91 if (*ecode == OP_KETRMIN)
2104     {
2105 ph10 623 RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM7);
2106 nigel 91 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2107 ph10 618 if (*prev == OP_ONCE)
2108     {
2109 ph10 623 RMATCH(eptr, prev, offset_top, md, eptrb, RM8);
2110 ph10 618 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2111     md->once_target = prev; /* Level at which to change to MATCH_NOMATCH */
2112 ph10 625 RRETURN(MATCH_ONCE);
2113     }
2114 ph10 604 if (*prev >= OP_SBRA) /* Could match an empty string */
2115 ph10 197 {
2116 ph10 604 RMATCH(eptr, prev, offset_top, md, eptrb, RM50);
2117 ph10 197 RRETURN(rrc);
2118     }
2119 nigel 91 ecode = prev;
2120     goto TAIL_RECURSE;
2121 nigel 77 }
2122 nigel 91 else /* OP_KETRMAX */
2123     {
2124 ph10 604 RMATCH(eptr, prev, offset_top, md, eptrb, RM13);
2125 ph10 618 if (rrc == MATCH_ONCE && md->once_target == prev) rrc = MATCH_NOMATCH;
2126 nigel 91 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2127 ph10 618 if (*prev == OP_ONCE)
2128     {
2129 ph10 623 RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM9);
2130 ph10 618 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2131     md->once_target = prev;
2132 ph10 625 RRETURN(MATCH_ONCE);
2133     }
2134 nigel 91 ecode += 1 + LINK_SIZE;
2135     goto TAIL_RECURSE;
2136     }
2137     /* Control never gets here */
2138 nigel 77
2139 ph10 602 /* Not multiline mode: start of subject assertion, unless notbol. */
2140 nigel 77
2141     case OP_CIRC:
2142 ph10 836 if (md->notbol && eptr == md->start_subject) RRETURN(MATCH_NOMATCH);
2143 ph10 625
2144 nigel 77 /* Start of subject assertion */
2145    
2146     case OP_SOD:
2147 ph10 836 if (eptr != md->start_subject) RRETURN(MATCH_NOMATCH);
2148 nigel 77 ecode++;
2149     break;
2150 ph10 625
2151 ph10 602 /* Multiline mode: start of subject unless notbol, or after any newline. */
2152 nigel 77
2153 ph10 602 case OP_CIRCM:
2154 ph10 836 if (md->notbol && eptr == md->start_subject) RRETURN(MATCH_NOMATCH);
2155 ph10 602 if (eptr != md->start_subject &&
2156     (eptr == md->end_subject || !WAS_NEWLINE(eptr)))
2157 ph10 836 RRETURN(MATCH_NOMATCH);
2158 ph10 602 ecode++;
2159     break;
2160    
2161 nigel 77 /* Start of match assertion */
2162    
2163     case OP_SOM:
2164 ph10 836 if (eptr != md->start_subject + md->start_offset) RRETURN(MATCH_NOMATCH);
2165 nigel 77 ecode++;
2166     break;
2167 ph10 172
2168 ph10 168 /* Reset the start of match point */
2169 ph10 172
2170 ph10 168 case OP_SET_SOM:
2171     mstart = eptr;
2172 ph10 172 ecode++;
2173     break;
2174 nigel 77
2175 ph10 602 /* Multiline mode: assert before any newline, or before end of subject
2176     unless noteol is set. */
2177 nigel 77
2178 ph10 602 case OP_DOLLM:
2179     if (eptr < md->end_subject)
2180 ph10 933 {
2181     if (!IS_NEWLINE(eptr))
2182 ph10 916 {
2183 ph10 919 if (md->partial != 0 &&
2184     eptr + 1 >= md->end_subject &&
2185 ph10 916 NLBLOCK->nltype == NLTYPE_FIXED &&
2186 ph10 933 NLBLOCK->nllen == 2 &&
2187 chpe 1100 RAWUCHARTEST(eptr) == NLBLOCK->nl[0])
2188 ph10 933 {
2189 ph10 916 md->hitend = TRUE;
2190     if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2191     }
2192 ph10 933 RRETURN(MATCH_NOMATCH);
2193     }
2194 ph10 916 }
2195 ph10 602 else
2196 nigel 77 {
2197 ph10 836 if (md->noteol) RRETURN(MATCH_NOMATCH);
2198 ph10 602 SCHECK_PARTIAL();
2199 nigel 77 }
2200 ph10 602 ecode++;
2201     break;
2202 ph10 579
2203 ph10 625 /* Not multiline mode: assert before a terminating newline or before end of
2204 ph10 602 subject unless noteol is set. */
2205    
2206     case OP_DOLL:
2207 ph10 836 if (md->noteol) RRETURN(MATCH_NOMATCH);
2208 ph10 602 if (!md->endonly) goto ASSERT_NL_OR_EOS;
2209    
2210 nigel 91 /* ... else fall through for endonly */
2211 nigel 77
2212     /* End of subject assertion (\z) */
2213    
2214     case OP_EOD:
2215 ph10 836 if (eptr < md->end_subject) RRETURN(MATCH_NOMATCH);
2216 ph10 553 SCHECK_PARTIAL();
2217 nigel 77 ecode++;
2218     break;
2219    
2220     /* End of subject or ending \n assertion (\Z) */
2221    
2222     case OP_EODN:
2223 ph10 553 ASSERT_NL_OR_EOS:
2224     if (eptr < md->end_subject &&
2225 nigel 93 (!IS_NEWLINE(eptr) || eptr != md->end_subject - md->nllen))
2226 ph10 916 {
2227 ph10 919 if (md->partial != 0 &&
2228     eptr + 1 >= md->end_subject &&
2229 ph10 916 NLBLOCK->nltype == NLTYPE_FIXED &&
2230 ph10 933 NLBLOCK->nllen == 2 &&
2231 chpe 1100 RAWUCHARTEST(eptr) == NLBLOCK->nl[0])
2232 ph10 933 {
2233 ph10 916 md->hitend = TRUE;
2234     if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2235     }
2236 ph10 836 RRETURN(MATCH_NOMATCH);
2237 ph10 933 }
2238 ph10 579
2239 ph10 553 /* Either at end of string or \n before end. */
2240 ph10 579
2241 ph10 553 SCHECK_PARTIAL();
2242 nigel 77 ecode++;
2243     break;
2244    
2245     /* Word boundary assertions */
2246    
2247     case OP_NOT_WORD_BOUNDARY:
2248     case OP_WORD_BOUNDARY:
2249     {
2250    
2251     /* Find out if the previous and current characters are "word" characters.
2252     It takes a bit more work in UTF-8 mode. Characters > 255 are assumed to
2253 ph10 443 be "non-word" characters. Remember the earliest consulted character for
2254 ph10 435 partial matching. */
2255 nigel 77
2256 ph10 836 #ifdef SUPPORT_UTF
2257     if (utf)
2258 nigel 77 {
2259 ph10 518 /* Get status of previous character */
2260 ph10 527
2261 nigel 77 if (eptr == md->start_subject) prev_is_word = FALSE; else
2262     {
2263 ph10 836 PCRE_PUCHAR lastptr = eptr - 1;
2264     BACKCHAR(lastptr);
2265 ph10 443 if (lastptr < md->start_used_ptr) md->start_used_ptr = lastptr;
2266 nigel 77 GETCHAR(c, lastptr);
2267 ph10 527 #ifdef SUPPORT_UCP
2268 ph10 518 if (md->use_ucp)
2269     {
2270     if (c == '_') prev_is_word = TRUE; else
2271 ph10 527 {
2272 ph10 518 int cat = UCD_CATEGORY(c);
2273     prev_is_word = (cat == ucp_L || cat == ucp_N);
2274 ph10 527 }
2275     }
2276     else
2277     #endif
2278 nigel 77 prev_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;
2279     }
2280 ph10 527
2281 ph10 518 /* Get status of next character */
2282 ph10 527
2283 ph10 443 if (eptr >= md->end_subject)
2284 nigel 77 {
2285 ph10 443 SCHECK_PARTIAL();
2286     cur_is_word = FALSE;
2287 ph10 428 }
2288     else
2289     {
2290 nigel 77 GETCHAR(c, eptr);
2291 ph10 527 #ifdef SUPPORT_UCP
2292 ph10 518 if (md->use_ucp)
2293     {
2294     if (c == '_') cur_is_word = TRUE; else
2295 ph10 527 {
2296 ph10 518 int cat = UCD_CATEGORY(c);
2297     cur_is_word = (cat == ucp_L || cat == ucp_N);
2298 ph10 527 }
2299     }
2300     else
2301     #endif
2302 nigel 77 cur_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;
2303     }
2304     }
2305     else
2306     #endif
2307    
2308 ph10 527 /* Not in UTF-8 mode, but we may still have PCRE_UCP set, and for
2309 ph10 518 consistency with the behaviour of \w we do use it in this case. */
2310 nigel 77
2311     {
2312 ph10 518 /* Get status of previous character */
2313 ph10 527
2314 ph10 435 if (eptr == md->start_subject) prev_is_word = FALSE; else
2315     {
2316 ph10 443 if (eptr <= md->start_used_ptr) md->start_used_ptr = eptr - 1;
2317 ph10 527 #ifdef SUPPORT_UCP
2318 ph10 518 if (md->use_ucp)
2319     {
2320 ph10 527 c = eptr[-1];
2321 ph10 518 if (c == '_') prev_is_word = TRUE; else
2322 ph10 527 {
2323 ph10 518 int cat = UCD_CATEGORY(c);
2324     prev_is_word = (cat == ucp_L || cat == ucp_N);
2325 ph10 527 }
2326     }
2327     else
2328     #endif
2329 ph10 836 prev_is_word = MAX_255(eptr[-1])
2330     && ((md->ctypes[eptr[-1]] & ctype_word) != 0);
2331 ph10 435 }
2332 ph10 527
2333 ph10 518 /* Get status of next character */
2334 ph10 527
2335 ph10 443 if (eptr >= md->end_subject)
2336 ph10 428 {
2337 ph10 443 SCHECK_PARTIAL();
2338     cur_is_word = FALSE;
2339 ph10 428 }
2340 ph10 527 else
2341     #ifdef SUPPORT_UCP
2342 ph10 518 if (md->use_ucp)
2343     {
2344 ph10 527 c = *eptr;
2345 ph10 518 if (c == '_') cur_is_word = TRUE; else
2346 ph10 527 {
2347 ph10 518 int cat = UCD_CATEGORY(c);
2348     cur_is_word = (cat == ucp_L || cat == ucp_N);
2349 ph10 527 }
2350     }
2351     else
2352     #endif
2353 ph10 836 cur_is_word = MAX_255(*eptr)
2354     && ((md->ctypes[*eptr] & ctype_word) != 0);
2355 nigel 77 }
2356    
2357     /* Now see if the situation is what we want */
2358    
2359     if ((*ecode++ == OP_WORD_BOUNDARY)?
2360     cur_is_word == prev_is_word : cur_is_word != prev_is_word)
2361 ph10 836 RRETURN(MATCH_NOMATCH);
2362 nigel 77 }
2363     break;
2364    
2365 ph10 919 /* Match any single character type except newline; have to take care with
2366     CRLF newlines and partial matching. */
2367 nigel 77
2368     case OP_ANY:
2369 ph10 836 if (IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH);
2370 ph10 919 if (md->partial != 0 &&
2371     eptr + 1 >= md->end_subject &&
2372     NLBLOCK->nltype == NLTYPE_FIXED &&
2373 ph10 933 NLBLOCK->nllen == 2 &&
2374 chpe 1100 RAWUCHARTEST(eptr) == NLBLOCK->nl[0])
2375 ph10 933 {
2376 ph10 919 md->hitend = TRUE;
2377     if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2378     }
2379 ph10 933
2380 ph10 345 /* Fall through */
2381    
2382 ph10 919 /* Match any single character whatsoever. */
2383 ph10 933
2384 ph10 341 case OP_ALLANY:
2385 ph10 648 if (eptr >= md->end_subject) /* DO NOT merge the eptr++ here; it must */
2386     { /* not be updated before SCHECK_PARTIAL. */
2387 ph10 443 SCHECK_PARTIAL();
2388 ph10 836 RRETURN(MATCH_NOMATCH);
2389 ph10 443 }
2390 ph10 648 eptr++;
2391 ph10 836 #ifdef SUPPORT_UTF
2392     if (utf) ACROSSCHAR(eptr < md->end_subject, *eptr, eptr++);
2393     #endif
2394 nigel 77 ecode++;
2395     break;
2396    
2397     /* Match a single byte, even in UTF-8 mode. This opcode really does match
2398     any byte, even newline, independent of the setting of PCRE_DOTALL. */
2399    
2400     case OP_ANYBYTE:
2401 ph10 648 if (eptr >= md->end_subject) /* DO NOT merge the eptr++ here; it must */
2402     { /* not be updated before SCHECK_PARTIAL. */
2403 ph10 443 SCHECK_PARTIAL();
2404 ph10 836 RRETURN(MATCH_NOMATCH);
2405 ph10 443 }
2406 ph10 654 eptr++;
2407 nigel 77 ecode++;
2408     break;
2409    
2410     case OP_NOT_DIGIT:
2411 ph10 443 if (eptr >= md->end_subject)
2412 ph10 428 {
2413 ph10 443 SCHECK_PARTIAL();
2414 ph10 836 RRETURN(MATCH_NOMATCH);
2415 ph10 443 }
2416 nigel 77 GETCHARINCTEST(c, eptr);
2417     if (
2418 ph10 836 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2419 nigel 77 c < 256 &&
2420     #endif
2421     (md->ctypes[c] & ctype_digit) != 0
2422     )
2423 ph10 836 RRETURN(MATCH_NOMATCH);
2424 nigel 77 ecode++;
2425     break;
2426    
2427     case OP_DIGIT:
2428 ph10 443 if (eptr >= md->end_subject)
2429 ph10 428 {
2430 ph10 443 SCHECK_PARTIAL();
2431 ph10 836 RRETURN(MATCH_NOMATCH);
2432 ph10 443 }
2433 nigel 77 GETCHARINCTEST(c, eptr);
2434     if (
2435 ph10 836 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2436     c > 255 ||
2437 nigel 77 #endif
2438     (md->ctypes[c] & ctype_digit) == 0
2439     )
2440 ph10 836 RRETURN(MATCH_NOMATCH);
2441 nigel 77 ecode++;
2442     break;
2443    
2444     case OP_NOT_WHITESPACE:
2445 ph10 443 if (eptr >= md->end_subject)
2446 ph10 428 {
2447 ph10 443 SCHECK_PARTIAL();
2448 ph10 836 RRETURN(MATCH_NOMATCH);
2449 ph10 443 }
2450 nigel 77 GETCHARINCTEST(c, eptr);
2451     if (
2452 ph10 836 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2453 nigel 77 c < 256 &&
2454     #endif
2455     (md->ctypes[c] & ctype_space) != 0
2456     )
2457 ph10 836 RRETURN(MATCH_NOMATCH);
2458 nigel 77 ecode++;
2459     break;
2460    
2461     case OP_WHITESPACE:
2462 ph10 443 if (eptr >= md->end_subject)
2463 ph10 428 {
2464 ph10 443 SCHECK_PARTIAL();
2465 ph10 836 RRETURN(MATCH_NOMATCH);
2466 ph10 443 }
2467 nigel 77 GETCHARINCTEST(c, eptr);
2468     if (
2469 ph10 836 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2470     c > 255 ||
2471 nigel 77 #endif
2472     (md->ctypes[c] & ctype_space) == 0
2473     )
2474 ph10 836 RRETURN(MATCH_NOMATCH);
2475 nigel 77 ecode++;
2476     break;
2477    
2478     case OP_NOT_WORDCHAR:
2479 ph10 443 if (eptr >= md->end_subject)
2480 ph10 428 {
2481 ph10 443 SCHECK_PARTIAL();
2482 ph10 836 RRETURN(MATCH_NOMATCH);
2483 ph10 443 }
2484 nigel 77 GETCHARINCTEST(c, eptr);
2485     if (
2486 ph10 836 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2487 nigel 77 c < 256 &&
2488     #endif
2489     (md->ctypes[c] & ctype_word) != 0
2490     )
2491 ph10 836 RRETURN(MATCH_NOMATCH);
2492 nigel 77 ecode++;
2493     break;
2494    
2495     case OP_WORDCHAR:
2496 ph10 443 if (eptr >= md->end_subject)
2497 ph10 428 {
2498 ph10 443 SCHECK_PARTIAL();
2499 ph10 836 RRETURN(MATCH_NOMATCH);
2500 ph10 443 }
2501 nigel 77 GETCHARINCTEST(c, eptr);
2502     if (
2503 ph10 836 #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2504     c > 255 ||
2505 nigel 77 #endif
2506     (md->ctypes[c] & ctype_word) == 0
2507     )
2508 ph10 836 RRETURN(MATCH_NOMATCH);
2509 nigel 77 ecode++;
2510     break;
2511    
2512 nigel 93 case OP_ANYNL:
2513 ph10 443 if (eptr >= md->end_subject)
2514 ph10 428 {
2515 ph10 443 SCHECK_PARTIAL();
2516 ph10 836 RRETURN(MATCH_NOMATCH);
2517 ph10 443 }
2518 nigel 93 GETCHARINCTEST(c, eptr);
2519     switch(c)
2520     {
2521 ph10 836 default: RRETURN(MATCH_NOMATCH);
2522 ph10 625
2523 ph10 1033 case CHAR_CR:
2524 ph10 916 if (eptr >= md->end_subject)
2525     {
2526     SCHECK_PARTIAL();
2527 ph10 933 }
2528 chpe 1100 else if (RAWUCHARTEST(eptr) == CHAR_LF) eptr++;
2529 nigel 93 break;
2530 ph10 231
2531 ph10 1033 case CHAR_LF:
2532 ph10 231 break;
2533    
2534 ph10 1033 case CHAR_VT:
2535     case CHAR_FF:
2536     case CHAR_NEL:
2537 ph10 1041 #ifndef EBCDIC
2538 nigel 93 case 0x2028:
2539     case 0x2029:
2540 ph10 1041 #endif /* Not EBCDIC */
2541 ph10 836 if (md->bsr_anycrlf) RRETURN(MATCH_NOMATCH);
2542 nigel 93 break;
2543     }
2544     ecode++;
2545     break;
2546    
2547 ph10 178 case OP_NOT_HSPACE:
2548 ph10 443 if (eptr >= md->end_subject)
2549 ph10 428 {
2550 ph10 443 SCHECK_PARTIAL();
2551 ph10 836 RRETURN(MATCH_NOMATCH);
2552 ph10 443 }
2553 ph10 178 GETCHARINCTEST(c, eptr);
2554     switch(c)
2555     {
2556 ph10 1041 HSPACE_CASES: RRETURN(MATCH_NOMATCH); /* Byte and multibyte cases */
2557 ph10 178 default: break;
2558     }
2559     ecode++;
2560     break;
2561    
2562     case OP_HSPACE:
2563 ph10 443 if (eptr >= md->end_subject)
2564 ph10 428 {
2565 ph10 443 SCHECK_PARTIAL();
2566 ph10 836 RRETURN(MATCH_NOMATCH);
2567 ph10 443 }
2568 ph10 178 GETCHARINCTEST(c, eptr);
2569     switch(c)
2570     {
2571 ph10 1041 HSPACE_CASES: break; /* Byte and multibyte cases */
2572 ph10 836 default: RRETURN(MATCH_NOMATCH);
2573 ph10 178 }
2574     ecode++;
2575     break;
2576    
2577     case OP_NOT_VSPACE:
2578 ph10 443 if (eptr >= md->end_subject)
2579 ph10 428 {
2580 ph10 443 SCHECK_PARTIAL();
2581 ph10 836 RRETURN(MATCH_NOMATCH);
2582 ph10 443 }
2583 ph10 178 GETCHARINCTEST(c, eptr);
2584     switch(c)
2585     {
2586 ph10 1041 VSPACE_CASES: RRETURN(MATCH_NOMATCH);
2587 ph10 178 default: break;
2588     }
2589     ecode++;
2590     break;
2591    
2592     case OP_VSPACE:
2593 ph10 443 if (eptr >= md->end_subject)
2594 ph10 428 {
2595 ph10 443 SCHECK_PARTIAL();
2596 ph10 836 RRETURN(MATCH_NOMATCH);
2597 ph10 443 }
2598 ph10 178 GETCHARINCTEST(c, eptr);
2599     switch(c)
2600     {
2601 ph10 1041 VSPACE_CASES: break;
2602 ph10 836 default: RRETURN(MATCH_NOMATCH);
2603 ph10 178 }
2604     ecode++;
2605     break;
2606    
2607 nigel 77 #ifdef SUPPORT_UCP
2608     /* Check the next character by Unicode property. We will get here only
2609     if the support is in the binary; otherwise a compile-time error occurs. */
2610    
2611     case OP_PROP:
2612     case OP_NOTPROP:
2613 ph10 443 if (eptr >= md->end_subject)
2614 ph10 428 {
2615 ph10 443 SCHECK_PARTIAL();
2616 ph10 836 RRETURN(MATCH_NOMATCH);
2617 ph10 443 }
2618 nigel 77 GETCHARINCTEST(c, eptr);
2619     {
2620 ph10 1221 const pcre_uint32 *cp;
2621 ph10 384 const ucd_record *prop = GET_UCD(c);
2622 nigel 77
2623 nigel 87 switch(ecode[1])
2624     {
2625     case PT_ANY:
2626 ph10 836 if (op == OP_NOTPROP) RRETURN(MATCH_NOMATCH);
2627 nigel 87 break;
2628 nigel 77
2629 nigel 87 case PT_LAMP:
2630 ph10 349 if ((prop->chartype == ucp_Lu ||
2631     prop->chartype == ucp_Ll ||
2632     prop->chartype == ucp_Lt) == (op == OP_NOTPROP))
2633 ph10 836 RRETURN(MATCH_NOMATCH);
2634 ph10 517 break;
2635 nigel 87
2636     case PT_GC:
2637 ph10 836 if ((ecode[2] != PRIV(ucp_gentype)[prop->chartype]) == (op == OP_PROP))
2638     RRETURN(MATCH_NOMATCH);
2639 nigel 87 break;
2640    
2641     case PT_PC:
2642 ph10 349 if ((ecode[2] != prop->chartype) == (op == OP_PROP))
2643 ph10 836 RRETURN(MATCH_NOMATCH);
2644 nigel 87 break;
2645    
2646     case PT_SC:
2647 ph10 349 if ((ecode[2] != prop->script) == (op == OP_PROP))
2648 ph10 836 RRETURN(MATCH_NOMATCH);
2649 nigel 87 break;
2650 ph10 527
2651 ph10 517 /* These are specials */
2652 ph10 527
2653 ph10 517 case PT_ALNUM:
2654 ph10 836 if ((PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2655     PRIV(ucp_gentype)[prop->chartype] == ucp_N) == (op == OP_NOTPROP))
2656     RRETURN(MATCH_NOMATCH);
2657 ph10 527 break;
2658    
2659 ph10 517 case PT_SPACE: /* Perl space */
2660 ph10 836 if ((PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2661 ph10 517 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2662     == (op == OP_NOTPROP))
2663 ph10 836 RRETURN(MATCH_NOMATCH);
2664 ph10 527 break;
2665    
2666 ph10 517 case PT_PXSPACE: /* POSIX space */
2667 ph10 836 if ((PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2668 ph10 527 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2669 ph10 517 c == CHAR_FF || c == CHAR_CR)
2670     == (op == OP_NOTPROP))
2671 ph10 836 RRETURN(MATCH_NOMATCH);
2672 ph10 527 break;
2673 nigel 87
2674 ph10 527 case PT_WORD:
2675 ph10 836 if ((PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2676     PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2677 ph10 517 c == CHAR_UNDERSCORE) == (op == OP_NOTPROP))
2678 ph10 836 RRETURN(MATCH_NOMATCH);
2679 ph10 527 break;
2680 zherczeg 1047
2681 ph10 1046 case PT_CLIST:
2682 ph10 1218 cp = PRIV(ucd_caseless_sets) + ecode[2];
2683 ph10 1046 for (;;)
2684     {
2685     if (c < *cp)
2686 zherczeg 1047 { if (op == OP_PROP) { RRETURN(MATCH_NOMATCH); } else break; }
2687     if (c == *cp++)
2688     { if (op == OP_PROP) break; else { RRETURN(MATCH_NOMATCH); } }
2689     }
2690     break;
2691 ph10 1271
2692 ph10 1260 case PT_UCNC:
2693     if ((c == CHAR_DOLLAR_SIGN || c == CHAR_COMMERCIAL_AT ||
2694     c == CHAR_GRAVE_ACCENT || (c >= 0xa0 && c <= 0xd7ff) ||
2695     c >= 0xe000) == (op == OP_NOTPROP))
2696 ph10 1271 RRETURN(MATCH_NOMATCH);
2697     break;
2698 ph10 527
2699 ph10 517 /* This should never occur */
2700    
2701 nigel 87 default:
2702     RRETURN(PCRE_ERROR_INTERNAL);
2703 nigel 77 }
2704 nigel 87
2705     ecode += 3;
2706 nigel 77 }
2707     break;
2708    
2709     /* Match an extended Unicode sequence. We will get here only if the support
2710     is in the binary; otherwise a compile-time error occurs. */
2711    
2712     case OP_EXTUNI:
2713 ph10 443 if (eptr >= md->end_subject)
2714 ph10 428 {
2715 ph10 443 SCHECK_PARTIAL();
2716 ph10 836 RRETURN(MATCH_NOMATCH);
2717 ph10 443 }
2718 ph10 1011 else
2719 ph10 1041 {
2720     int lgb, rgb;
2721 ph10 1011 GETCHARINCTEST(c, eptr);
2722 ph10 1041 lgb = UCD_GRAPHBREAK(c);
2723 ph10 1011 while (eptr < md->end_subject)
2724     {
2725     int len = 1;
2726     if (!utf) c = *eptr; else { GETCHARLEN(c, eptr, len); }
2727 ph10 1041 rgb = UCD_GRAPHBREAK(c);
2728 ph10 1015 if ((PRIV(ucp_gbtable)[lgb] & (1 << rgb)) == 0) break;
2729 ph10 1041 lgb = rgb;
2730 ph10 1011 eptr += len;
2731 ph10 1041 }
2732 nigel 77 }
2733 ph10 933 CHECK_PARTIAL();
2734 nigel 77 ecode++;
2735     break;
2736 ph10 1046 #endif /* SUPPORT_UCP */
2737 nigel 77
2738    
2739     /* Match a back reference, possibly repeatedly. Look past the end of the
2740     item to see if there is repeat information following. The code is similar
2741     to that for character classes, but repeated for efficiency. Then obey
2742     similar code to character type repeats - written out again for speed.
2743     However, if the referenced string is the empty string, always treat
2744     it as matched, any number of times (otherwise there could be infinite
2745     loops). */
2746    
2747     case OP_REF:
2748 ph10 625 case OP_REFI:
2749     caseless = op == OP_REFI;
2750 ph10 595 offset = GET2(ecode, 1) << 1; /* Doubled ref number */
2751 ph10 836 ecode += 1 + IMM2_SIZE;
2752 ph10 345
2753 ph10 595 /* If the reference is unset, there are two possibilities:
2754 ph10 345
2755 ph10 595 (a) In the default, Perl-compatible state, set the length negative;
2756     this ensures that every attempt at a match fails. We can't just fail
2757     here, because of the possibility of quantifiers with zero minima.
2758 ph10 345
2759 ph10 595 (b) If the JavaScript compatibility flag is set, set the length to zero
2760     so that the back reference matches an empty string.
2761 ph10 345
2762 ph10 595 Otherwise, set the length to the length of what was matched by the
2763     referenced subpattern. */
2764 ph10 345
2765 ph10 595 if (offset >= offset_top || md->offset_vector[offset] < 0)
2766     length = (md->jscript_compat)? 0 : -1;
2767     else
2768     length = md->offset_vector[offset+1] - md->offset_vector[offset];
2769 nigel 77
2770 ph10 595 /* Set up for repetition, or handle the non-repeated case */
2771 nigel 77
2772 ph10 595 switch (*ecode)
2773     {
2774     case OP_CRSTAR:
2775     case OP_CRMINSTAR:
2776     case OP_CRPLUS:
2777     case OP_CRMINPLUS:
2778     case OP_CRQUERY:
2779     case OP_CRMINQUERY:
2780     c = *ecode++ - OP_CRSTAR;
2781     minimize = (c & 1) != 0;
2782     min = rep_min[c]; /* Pick up values from tables; */
2783     max = rep_max[c]; /* zero for max => infinity */
2784     if (max == 0) max = INT_MAX;
2785     break;
2786 nigel 77
2787 ph10 595 case OP_CRRANGE:
2788     case OP_CRMINRANGE:
2789     minimize = (*ecode == OP_CRMINRANGE);
2790     min = GET2(ecode, 1);
2791 ph10 836 max = GET2(ecode, 1 + IMM2_SIZE);
2792 ph10 595 if (max == 0) max = INT_MAX;
2793 ph10 836 ecode += 1 + 2 * IMM2_SIZE;
2794 ph10 595 break;
2795 nigel 77
2796 ph10 595 default: /* No repeat follows */
2797 ph10 602 if ((length = match_ref(offset, eptr, length, md, caseless)) < 0)
2798 ph10 595 {
2799 ph10 933 if (length == -2) eptr = md->end_subject; /* Partial match */
2800 ph10 595 CHECK_PARTIAL();
2801 ph10 836 RRETURN(MATCH_NOMATCH);
2802 nigel 77 }
2803 ph10 595 eptr += length;
2804     continue; /* With the main loop */
2805     }
2806 nigel 77
2807 ph10 595 /* Handle repeated back references. If the length of the reference is
2808 ph10 836 zero, just continue with the main loop. If the length is negative, it
2809 ph10 842 means the reference is unset in non-Java-compatible mode. If the minimum is
2810     zero, we can continue at the same level without recursion. For any other
2811 ph10 836 minimum, carrying on will result in NOMATCH. */
2812 ph10 443
2813 ph10 595 if (length == 0) continue;
2814 ph10 836 if (length < 0 && min == 0) continue;
2815 nigel 77
2816 ph10 595 /* First, ensure the minimum number of matches are present. We get back
2817     the length of the reference string explicitly rather than passing the
2818     address of eptr, so that eptr can be a register variable. */
2819 nigel 77
2820 ph10 595 for (i = 1; i <= min; i++)
2821     {
2822 ph10 625 int slength;
2823 ph10 602 if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
2824 nigel 77 {
2825 ph10 933 if (slength == -2) eptr = md->end_subject; /* Partial match */
2826 ph10 595 CHECK_PARTIAL();
2827 ph10 836 RRETURN(MATCH_NOMATCH);
2828 nigel 77 }
2829 ph10 595 eptr += slength;
2830     }
2831 nigel 77
2832 ph10 595 /* If min = max, continue at the same level without recursion.
2833     They are not both allowed to be zero. */
2834 nigel 77
2835 ph10 595 if (min == max) continue;
2836 nigel 77
2837 ph10 595 /* If minimizing, keep trying and advancing the pointer */
2838 nigel 77
2839 ph10 595 if (minimize)
2840     {
2841     for (fi = min;; fi++)
2842 nigel 77 {
2843 ph10 625 int slength;
2844 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM14);
2845 ph10 595 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2846 ph10 836 if (fi >= max) RRETURN(MATCH_NOMATCH);
2847 ph10 602 if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
2848 nigel 77 {
2849 ph10 933 if (slength == -2) eptr = md->end_subject; /* Partial match */
2850 ph10 595 CHECK_PARTIAL();
2851 ph10 836 RRETURN(MATCH_NOMATCH);
2852 nigel 77 }
2853 ph10 595 eptr += slength;
2854 nigel 77 }
2855 ph10 595 /* Control never gets here */
2856     }
2857 nigel 77
2858 ph10 595 /* If maximizing, find the longest string and work backwards */
2859 nigel 77
2860 ph10 595 else
2861     {
2862     pp = eptr;
2863     for (i = min; i < max; i++)
2864 nigel 77 {
2865 ph10 625 int slength;
2866 ph10 602 if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
2867 nigel 77 {
2868 ph10 933 /* Can't use CHECK_PARTIAL because we don't want to update eptr in
2869     the soft partial matching case. */
2870    
2871     if (slength == -2 && md->partial != 0 &&
2872 ph10 916 md->end_subject > md->start_used_ptr)
2873 ph10 933 {
2874 ph10 916 md->hitend = TRUE;
2875     if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2876     }
2877 ph10 595 break;
2878 nigel 77 }
2879 ph10 595 eptr += slength;
2880 nigel 77 }
2881 ph10 933
2882 ph10 595 while (eptr >= pp)
2883     {
2884 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM15);
2885 ph10 595 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2886     eptr -= length;
2887     }
2888 ph10 836 RRETURN(MATCH_NOMATCH);
2889 nigel 77 }
2890     /* Control never gets here */
2891    
2892     /* Match a bit-mapped character class, possibly repeatedly. This op code is
2893     used when all the characters in the class have values in the range 0-255,
2894     and either the matching is caseful, or the characters are in the range
2895     0-127 when UTF-8 processing is enabled. The only difference between
2896     OP_CLASS and OP_NCLASS occurs when a data character outside the range is
2897     encountered.
2898    
2899     First, look past the end of the item to see if there is repeat information
2900     following. Then obey similar code to character type repeats - written out
2901     again for speed. */
2902    
2903     case OP_NCLASS:
2904     case OP_CLASS:
2905     {
2906 ph10 836 /* The data variable is saved across frames, so the byte map needs to
2907     be stored there. */
2908     #define BYTE_MAP ((pcre_uint8 *)data)
2909 nigel 77 data = ecode + 1; /* Save for matching */
2910 ph10 836 ecode += 1 + (32 / sizeof(pcre_uchar)); /* Advance past the item */
2911 nigel 77
2912     switch (*ecode)
2913     {
2914     case OP_CRSTAR:
2915     case OP_CRMINSTAR:
2916     case OP_CRPLUS:
2917     case OP_CRMINPLUS:
2918     case OP_CRQUERY:
2919     case OP_CRMINQUERY:
2920     c = *ecode++ - OP_CRSTAR;
2921     minimize = (c & 1) != 0;
2922     min = rep_min[c]; /* Pick up values from tables; */
2923     max = rep_max[c]; /* zero for max => infinity */
2924     if (max == 0) max = INT_MAX;
2925     break;
2926    
2927     case OP_CRRANGE:
2928     case OP_CRMINRANGE:
2929     minimize = (*ecode == OP_CRMINRANGE);
2930     min = GET2(ecode, 1);
2931 ph10 836 max = GET2(ecode, 1 + IMM2_SIZE);
2932 nigel 77 if (max == 0) max = INT_MAX;
2933 ph10 836 ecode += 1 + 2 * IMM2_SIZE;
2934 nigel 77 break;
2935    
2936     default: /* No repeat follows */
2937     min = max = 1;
2938     break;
2939     }
2940    
2941     /* First, ensure the minimum number of matches are present. */
2942    
2943 ph10 836 #ifdef SUPPORT_UTF
2944     if (utf)
2945 nigel 77 {
2946     for (i = 1; i <= min; i++)
2947     {
2948 ph10 427 if (eptr >= md->end_subject)
2949 ph10 426 {
2950 ph10 428 SCHECK_PARTIAL();
2951 ph10 836 RRETURN(MATCH_NOMATCH);
2952 ph10 427 }
2953 nigel 77 GETCHARINC(c, eptr);
2954     if (c > 255)
2955     {
2956 ph10 836 if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
2957 nigel 77 }
2958     else
2959 ph10 836 if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
2960 nigel 77 }
2961     }
2962     else
2963     #endif
2964 ph10 836 /* Not UTF mode */
2965 nigel 77 {
2966     for (i = 1; i <= min; i++)
2967     {
2968 ph10 427 if (eptr >= md->end_subject)
2969 ph10 426 {
2970 ph10 428 SCHECK_PARTIAL();
2971 ph10 836 RRETURN(MATCH_NOMATCH);
2972 ph10 427 }
2973 nigel 77 c = *eptr++;
2974 ph10 836 #ifndef COMPILE_PCRE8
2975     if (c > 255)
2976     {
2977     if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
2978     }
2979     else
2980     #endif
2981     if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
2982 nigel 77 }
2983     }
2984    
2985     /* If max == min we can continue with the main loop without the
2986     need to recurse. */
2987    
2988     if (min == max) continue;
2989    
2990     /* If minimizing, keep testing the rest of the expression and advancing
2991     the pointer while it matches the class. */
2992    
2993     if (minimize)
2994     {
2995 ph10 836 #ifdef SUPPORT_UTF
2996     if (utf)
2997 nigel 77 {
2998     for (fi = min;; fi++)
2999     {
3000 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM16);
3001 nigel 77 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3002 ph10 836 if (fi >= max) RRETURN(MATCH_NOMATCH);
3003 ph10 427 if (eptr >= md->end_subject)
3004 ph10 426 {
3005 ph10 427 SCHECK_PARTIAL();
3006 ph10 836 RRETURN(MATCH_NOMATCH);
3007 ph10 427 }
3008 nigel 77 GETCHARINC(c, eptr);
3009     if (c > 255)
3010     {
3011 ph10 836 if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
3012 nigel 77 }
3013     else
3014 ph10 836 if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
3015 nigel 77 }
3016     }
3017     else
3018     #endif
3019 ph10 836 /* Not UTF mode */
3020 nigel 77 {
3021     for (fi = min;; fi++)
3022     {
3023 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM17);
3024 nigel 77 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3025 ph10 836 if (fi >= max) RRETURN(MATCH_NOMATCH);
3026 ph10 427 if (eptr >= md->end_subject)
3027 ph10 426 {
3028 ph10 427 SCHECK_PARTIAL();
3029 ph10 836 RRETURN(MATCH_NOMATCH);
3030 ph10 427 }
3031 nigel 77 c = *eptr++;
3032 ph10 836 #ifndef COMPILE_PCRE8
3033     if (c > 255)
3034     {
3035     if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
3036     }
3037     else
3038     #endif
3039     if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
3040 nigel 77 }
3041     }
3042     /* Control never gets here */
3043     }
3044    
3045     /* If maximizing, find the longest possible run, then work backwards. */
3046    
3047     else
3048     {
3049     pp = eptr;
3050    
3051 ph10 836 #ifdef SUPPORT_UTF
3052     if (utf)
3053 nigel 77 {
3054     for (i = min; i < max; i++)
3055     {
3056     int len = 1;
3057 ph10 463 if (eptr >= md->end_subject)
3058 ph10 462 {
3059 ph10 463 SCHECK_PARTIAL();
3060 ph10 462 break;
3061 ph10 463 }
3062 nigel 77 GETCHARLEN(c, eptr, len);
3063     if (c > 255)
3064     {
3065     if (op == OP_CLASS) break;
3066     }
3067     else
3068 ph10 836 if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) break;
3069 nigel 77 eptr += len;
3070     }
3071     for (;;)
3072     {
3073 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM18);
3074 nigel 77 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3075     if (eptr-- == pp) break; /* Stop if tried at original pos */
3076     BACKCHAR(eptr);
3077     }
3078     }
3079     else
3080     #endif
3081 ph10 836 /* Not UTF mode */
3082 nigel 77 {
3083     for (i = min; i < max; i++)
3084     {
3085 ph10 463 if (eptr >= md->end_subject)
3086 ph10 462 {
3087 ph10 463 SCHECK_PARTIAL();
3088 ph10 462 break;
3089 ph10 463 }
3090 nigel 77 c = *eptr;
3091 ph10 836 #ifndef COMPILE_PCRE8
3092     if (c > 255)
3093     {
3094     if (op == OP_CLASS) break;
3095     }
3096     else
3097     #endif
3098     if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) break;
3099 nigel 77 eptr++;
3100     }
3101     while (eptr >= pp)
3102     {
3103 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM19);
3104 nigel 87 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3105 nigel 77 eptr--;
3106     }
3107     }
3108    
3109 ph10 836 RRETURN(MATCH_NOMATCH);
3110 nigel 77 }
3111 ph10 836 #undef BYTE_MAP
3112 nigel 77 }
3113     /* Control never gets here */
3114    
3115    
3116     /* Match an extended character class. This opcode is encountered only
3117 ph10 384 when UTF-8 mode mode is supported. Nevertheless, we may not be in UTF-8
3118     mode, because Unicode properties are supported in non-UTF-8 mode. */
3119 nigel 77
3120 ph10 836 #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3121 nigel 77 case OP_XCLASS:
3122     {
3123     data = ecode + 1 + LINK_SIZE; /* Save for matching */
3124     ecode += GET(ecode, 1); /* Advance past the item */
3125    
3126     switch (*ecode)
3127     {
3128     case OP_CRSTAR:
3129     case OP_CRMINSTAR:
3130     case OP_CRPLUS:
3131     case OP_CRMINPLUS:
3132     case OP_CRQUERY:
3133     case OP_CRMINQUERY:
3134     c = *ecode++ - OP_CRSTAR;
3135     minimize = (c & 1) != 0;
3136     min = rep_min[c]; /* Pick up values from tables; */
3137     max = rep_max[c]; /* zero for max => infinity */
3138     if (max == 0) max = INT_MAX;
3139     break;
3140    
3141     case OP_CRRANGE:
3142     case OP_CRMINRANGE:
3143     minimize = (*ecode == OP_CRMINRANGE);
3144     min = GET2(ecode, 1);
3145 ph10 836 max = GET2(ecode, 1 + IMM2_SIZE);
3146 nigel 77 if (max == 0) max = INT_MAX;
3147 ph10 836 ecode += 1 + 2 * IMM2_SIZE;
3148 nigel 77 break;
3149    
3150     default: /* No repeat follows */
3151     min = max = 1;
3152     break;
3153     }
3154    
3155     /* First, ensure the minimum number of matches are present. */
3156    
3157     for (i = 1; i <= min; i++)
3158     {
3159 ph10 427 if (eptr >= md->end_subject)
3160 ph10 426 {
3161     SCHECK_PARTIAL();
3162 ph10 836 RRETURN(MATCH_NOMATCH);
3163 ph10 427 }
3164 ph10 384 GETCHARINCTEST(c, eptr);
3165 ph10 836 if (!PRIV(xclass)(c, data, utf)) RRETURN(MATCH_NOMATCH);
3166 nigel 77 }
3167    
3168     /* If max == min we can continue with the main loop without the
3169     need to recurse. */
3170    
3171     if (min == max) continue;
3172    
3173     /* If minimizing, keep testing the rest of the expression and advancing
3174     the pointer while it matches the class. */
3175    
3176     if (minimize)
3177     {
3178     for (fi = min;; fi++)
3179     {
3180 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM20);
3181 nigel 77 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3182 ph10 836 if (fi >= max) RRETURN(MATCH_NOMATCH);
3183 ph10 427 if (eptr >= md->end_subject)
3184 ph10 426 {
3185 ph10 427 SCHECK_PARTIAL();
3186 ph10 836 RRETURN(MATCH_NOMATCH);
3187 ph10 427 }
3188 ph10 384 GETCHARINCTEST(c, eptr);
3189 ph10 836 if (!PRIV(xclass)(c, data, utf)) RRETURN(MATCH_NOMATCH);
3190 nigel 77 }
3191     /* Control never gets here */
3192     }
3193    
3194     /* If maximizing, find the longest possible run, then work backwards. */
3195    
3196     else
3197     {
3198     pp = eptr;
3199     for (i = min; i < max; i++)
3200     {
3201     int len = 1;
3202 ph10 463 if (eptr >= md->end_subject)
3203 ph10 462 {
3204 ph10 463 SCHECK_PARTIAL();
3205 ph10 462 break;
3206 ph10 463 }
3207 ph10 836 #ifdef SUPPORT_UTF
3208 ph10 384 GETCHARLENTEST(c, eptr, len);
3209 ph10 836 #else
3210     c = *eptr;
3211     #endif
3212     if (!PRIV(xclass)(c, data, utf)) break;
3213 nigel 77 eptr += len;
3214     }
3215     for(;;)
3216     {
3217 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM21);
3218 nigel 77 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3219     if (eptr-- == pp) break; /* Stop if tried at original pos */
3220 ph10 836 #ifdef SUPPORT_UTF
3221     if (utf) BACKCHAR(eptr);
3222     #endif
3223 nigel 77 }
3224 ph10 836 RRETURN(MATCH_NOMATCH);
3225 nigel 77 }
3226    
3227     /* Control never gets here */
3228     }
3229     #endif /* End of XCLASS */
3230    
3231     /* Match a single character, casefully */
3232    
3233     case OP_CHAR:
3234 ph10 836 #ifdef SUPPORT_UTF
3235     if (utf)
3236 nigel 77 {
3237     length = 1;
3238     ecode++;
3239     GETCHARLEN(fc, ecode, length);
3240 ph10 443 if (length > md->end_subject - eptr)
3241 ph10 428 {
3242     CHECK_PARTIAL(); /* Not SCHECK_PARTIAL() */
3243 ph10 836 RRETURN(MATCH_NOMATCH);
3244 ph10 443 }
3245 chpe 1100 while (length-- > 0) if (*ecode++ != RAWUCHARINC(eptr)) RRETURN(MATCH_NOMATCH);
3246 nigel 77 }
3247     else
3248     #endif
3249 ph10 836 /* Not UTF mode */
3250 nigel 77 {
3251 ph10 443 if (md->end_subject - eptr < 1)
3252 ph10 428 {
3253     SCHECK_PARTIAL(); /* This one can use SCHECK_PARTIAL() */
3254 ph10 836 RRETURN(MATCH_NOMATCH);
3255 ph10 443 }
3256 ph10 836 if (ecode[1] != *eptr++) RRETURN(MATCH_NOMATCH);
3257 nigel 77 ecode += 2;
3258     }
3259     break;
3260    
3261 ph10 836 /* Match a single character, caselessly. If we are at the end of the
3262     subject, give up immediately. */
3263 nigel 77
3264 ph10 602 case OP_CHARI:
3265 ph10 836 if (eptr >= md->end_subject)
3266 nigel 77 {
3267 ph10 836 SCHECK_PARTIAL();
3268     RRETURN(MATCH_NOMATCH);
3269     }
3270    
3271     #ifdef SUPPORT_UTF
3272     if (utf)
3273     {
3274 nigel 77 length = 1;
3275     ecode++;
3276     GETCHARLEN(fc, ecode, length);
3277 ph10 788
3278 nigel 77 /* If the pattern character's value is < 128, we have only one byte, and
3279 ph10 836 we know that its other case must also be one byte long, so we can use the
3280     fast lookup table. We know that there is at least one byte left in the
3281     subject. */
3282 nigel 77
3283     if (fc < 128)
3284     {
3285 ph10 1257 pcre_uint32 cc = RAWUCHAR(eptr);
3286 chpe 1100 if (md->lcc[fc] != TABLE_GET(cc, md->lcc, cc)) RRETURN(MATCH_NOMATCH);
3287 ph10 836 ecode++;
3288     eptr++;
3289 nigel 77 }
3290    
3291 ph10 836 /* Otherwise we must pick up the subject character. Note that we cannot
3292     use the value of "length" to check for sufficient bytes left, because the
3293     other case of the character may have more or fewer bytes. */
3294 nigel 77
3295     else
3296     {
3297 chpe 1084 pcre_uint32 dc;
3298 nigel 77 GETCHARINC(dc, eptr);
3299     ecode += length;
3300    
3301     /* If we have Unicode property support, we can use it to test the other
3302 nigel 87 case of the character, if there is one. */
3303 nigel 77
3304     if (fc != dc)
3305     {
3306     #ifdef SUPPORT_UCP
3307 ph10 349 if (dc != UCD_OTHERCASE(fc))
3308 nigel 77 #endif
3309 ph10 836 RRETURN(MATCH_NOMATCH);
3310 nigel 77 }
3311     }
3312     }
3313     else
3314 ph10 836 #endif /* SUPPORT_UTF */
3315 nigel 77
3316 ph10 836 /* Not UTF mode */
3317 nigel 77 {
3318 ph10 836 if (TABLE_GET(ecode[1], md->lcc, ecode[1])
3319     != TABLE_GET(*eptr, md->lcc, *eptr)) RRETURN(MATCH_NOMATCH);
3320     eptr++;
3321 nigel 77 ecode += 2;
3322     }
3323     break;
3324    
3325 nigel 93 /* Match a single character repeatedly. */
3326 nigel 77
3327     case OP_EXACT:
3328 ph10 602 case OP_EXACTI:
3329 nigel 77 min = max = GET2(ecode, 1);
3330 ph10 836 ecode += 1 + IMM2_SIZE;
3331 nigel 77 goto REPEATCHAR;
3332    
3333 nigel 93 case OP_POSUPTO:
3334 ph10 602 case OP_POSUPTOI:
3335 nigel 93 possessive = TRUE;
3336     /* Fall through */
3337    
3338 nigel 77 case OP_UPTO:
3339 ph10 602 case OP_UPTOI:
3340 nigel 77 case OP_MINUPTO:
3341 ph10 602 case OP_MINUPTOI:
3342 nigel 77 min = 0;
3343     max = GET2(ecode, 1);
3344 ph10 602 minimize = *ecode == OP_MINUPTO || *ecode == OP_MINUPTOI;
3345 ph10 836 ecode += 1 + IMM2_SIZE;
3346 nigel 77 goto REPEATCHAR;
3347    
3348 nigel 93 case OP_POSSTAR:
3349 ph10 602 case OP_POSSTARI:
3350 nigel 93 possessive = TRUE;
3351     min = 0;
3352     max = INT_MAX;
3353     ecode++;
3354     goto REPEATCHAR;
3355    
3356     case OP_POSPLUS:
3357 ph10 602 case OP_POSPLUSI:
3358 nigel 93 possessive = TRUE;
3359     min = 1;
3360     max = INT_MAX;
3361     ecode++;
3362     goto REPEATCHAR;
3363    
3364     case OP_POSQUERY:
3365 ph10 602 case OP_POSQUERYI:
3366 nigel 93 possessive = TRUE;
3367     min = 0;
3368     max = 1;
3369     ecode++;
3370     goto REPEATCHAR;
3371    
3372 nigel 77 case OP_STAR:
3373 ph10 602 case OP_STARI:
3374 nigel 77 case OP_MINSTAR:
3375 ph10 602 case OP_MINSTARI:
3376 nigel 77 case OP_PLUS:
3377 ph10 602 case OP_PLUSI:
3378 nigel 77 case OP_MINPLUS:
3379 ph10 602 case OP_MINPLUSI:
3380 nigel 77 case OP_QUERY:
3381 ph10 602 case OP_QUERYI:
3382 nigel 77 case OP_MINQUERY:
3383 ph10 602 case OP_MINQUERYI:
3384     c = *ecode++ - ((op < OP_STARI)? OP_STAR : OP_STARI);
3385 nigel 77 minimize = (c & 1) != 0;
3386     min = rep_min[c]; /* Pick up values from tables; */
3387     max = rep_max[c]; /* zero for max => infinity */
3388     if (max == 0) max = INT_MAX;
3389    
3390 ph10 1320 /* Common code for all repeated single-character matches. We first check
3391     for the minimum number of characters. If the minimum equals the maximum, we
3392     are done. Otherwise, if minimizing, check the rest of the pattern for a
3393     match; if there isn't one, advance up to the maximum, one character at a
3394 ph10 1311 time.
3395 ph10 1320
3396     If maximizing, advance up to the maximum number of matching characters,
3397 ph10 1311 until eptr is past the end of the maximum run. If possessive, we are
3398     then done (no backing up). Otherwise, match at this position; anything
3399     other than no match is immediately returned. For nomatch, back up one
3400     character, unless we are matching \R and the last thing matched was
3401 ph10 1320 \r\n, in which case, back up two bytes. When we reach the first optional
3402     character position, we can save stack by doing a tail recurse.
3403    
3404 ph10 1311 The various UTF/non-UTF and caseful/caseless cases are handled separately,
3405     for speed. */
3406 nigel 77
3407     REPEATCHAR:
3408 ph10 836 #ifdef SUPPORT_UTF
3409     if (utf)
3410 nigel 77 {
3411     length = 1;
3412     charptr = ecode;
3413     GETCHARLEN(fc, ecode, length);
3414     ecode += length;
3415    
3416     /* Handle multibyte character matching specially here. There is
3417     support for caseless matching if UCP support is present. */
3418    
3419     if (length > 1)
3420     {
3421     #ifdef SUPPORT_UCP
3422 chpe 1102 pcre_uint32 othercase;
3423 ph10 602 if (op >= OP_STARI && /* Caseless */
3424 ph10 349 (othercase = UCD_OTHERCASE(fc)) != fc)
3425 ph10 836 oclength = PRIV(ord2utf)(othercase, occhars);
3426 ph10 115 else oclength = 0;
3427 nigel 77 #endif /* SUPPORT_UCP */
3428    
3429     for (i = 1; i <= min; i++)
3430     {
3431 ph10 426 if (eptr <= md->end_subject - length &&
3432 ph10 836 memcmp(eptr, charptr, IN_UCHARS(length)) == 0) eptr += length;
3433 ph10 123 #ifdef SUPPORT_UCP
3434 ph10 426 else if (oclength > 0 &&
3435     eptr <= md->end_subject - oclength &&
3436 ph10 836 memcmp(eptr, occhars, IN_UCHARS(oclength)) == 0) eptr += oclength;
3437 ph10 426 #endif /* SUPPORT_UCP */
3438 nigel 77 else
3439     {
3440 ph10 426 CHECK_PARTIAL();
3441 ph10 836 RRETURN(MATCH_NOMATCH);
3442 nigel 77 }
3443     }
3444    
3445     if (min == max) continue;
3446    
3447     if (minimize)
3448     {
3449     for (fi = min;; fi++)
3450     {
3451 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM22);
3452 nigel 77 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3453 ph10 836 if (fi >= max) RRETURN(MATCH_NOMATCH);
3454 ph10 426 if (eptr <= md->end_subject - length &&
3455 ph10 836 memcmp(eptr, charptr, IN_UCHARS(length)) == 0) eptr += length;
3456 ph10 123 #ifdef SUPPORT_UCP
3457 ph10 426 else if (oclength > 0 &&
3458     eptr <= md->end_subject - oclength &&
3459 ph10 836 memcmp(eptr, occhars, IN_UCHARS(oclength)) == 0) eptr += oclength;
3460 ph10 426 #endif /* SUPPORT_UCP */
3461 nigel 77 else
3462     {
3463 ph10 426 CHECK_PARTIAL();
3464 ph10 836 RRETURN(MATCH_NOMATCH);
3465 nigel 77 }
3466     }
3467     /* Control never gets here */
3468     }
3469 nigel 93
3470     else /* Maximize */
3471 nigel 77 {
3472     pp = eptr;
3473     for (i = min; i < max; i++)
3474     {
3475 ph10 426 if (eptr <= md->end_subject - length &&
3476 ph10 836 memcmp(eptr, charptr, IN_UCHARS(length)) == 0) eptr += length;
3477 ph10 123 #ifdef SUPPORT_UCP
3478 ph10 426 else if (oclength > 0 &&
3479     eptr <= md->end_subject - oclength &&
3480 ph10 836 memcmp(eptr, occhars, IN_UCHARS(oclength)) == 0) eptr += oclength;
3481 ph10 426 #endif /* SUPPORT_UCP */
3482 ph10 463 else
3483 ph10 462 {
3484 ph10 463 CHECK_PARTIAL();
3485 ph10 462 break;
3486 ph10 463 }
3487 nigel 77 }
3488 nigel 93
3489 ph10 1311 if (possessive) continue; /* No backtracking */
3490 ph10 120 for(;;)
3491 ph10 426 {
3492 ph10 1311 if (eptr == pp) goto TAIL_RECURSE;
3493 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM23);
3494 ph10 426 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3495 ph10 115 #ifdef SUPPORT_UCP
3496 ph10 426 eptr--;
3497     BACKCHAR(eptr);
3498 ph10 123 #else /* without SUPPORT_UCP */
3499 ph10 426 eptr -= length;
3500 ph10 123 #endif /* SUPPORT_UCP */
3501 ph10 426 }
3502 nigel 77 }
3503     /* Control never gets here */
3504     }
3505    
3506     /* If the length of a UTF-8 character is 1, we fall through here, and
3507     obey the code as for non-UTF-8 characters below, though in this case the
3508     value of fc will always be < 128. */
3509     }
3510     else
3511 ph10 836 #endif /* SUPPORT_UTF */
3512     /* When not in UTF-8 mode, load a single-byte character. */
3513     fc = *ecode++;
3514 nigel 77
3515 ph10 836 /* The value of fc at this point is always one character, though we may
3516     or may not be in UTF mode. The code is duplicated for the caseless and
3517 nigel 77 caseful cases, for speed, since matching characters is likely to be quite
3518     common. First, ensure the minimum number of matches are present. If min =
3519     max, continue at the same level without recursing. Otherwise, if
3520     minimizing, keep trying the rest of the expression and advancing one
3521     matching character if failing, up to the maximum. Alternatively, if
3522     maximizing, find the maximum number of characters and work backwards. */
3523    
3524     DPRINTF(("matching %c{%d,%d} against subject %.*s\n", fc, min, max,
3525 ph10 976 max, (char *)eptr));
3526 nigel 77
3527 ph10 602 if (op >= OP_STARI) /* Caseless */
3528 nigel 77 {
3529 ph10 836 #ifdef COMPILE_PCRE8
3530     /* fc must be < 128 if UTF is enabled. */
3531     foc = md->fcc[fc];
3532     #else
3533     #ifdef SUPPORT_UTF
3534     #ifdef SUPPORT_UCP
3535     if (utf && fc > 127)
3536     foc = UCD_OTHERCASE(fc);
3537     #else
3538     if (utf && fc > 127)
3539     foc = fc;
3540     #endif /* SUPPORT_UCP */
3541     else
3542     #endif /* SUPPORT_UTF */
3543     foc = TABLE_GET(fc, md->fcc, fc);
3544     #endif /* COMPILE_PCRE8 */
3545    
3546 nigel 77 for (i = 1; i <= min; i++)
3547 ph10 426 {
3548 ph10 1238 pcre_uint32 cc; /* Faster than pcre_uchar */
3549 ph10 426 if (eptr >= md->end_subject)
3550     {
3551     SCHECK_PARTIAL();
3552 ph10 836 RRETURN(MATCH_NOMATCH);
3553 ph10 426 }
3554 chpe 1100 cc = RAWUCHARTEST(eptr);
3555     if (fc != cc && foc != cc) RRETURN(MATCH_NOMATCH);
3556 ph10 836 eptr++;
3557 ph10 426 }
3558 nigel 77 if (min == max) continue;
3559     if (minimize)
3560     {
3561     for (fi = min;; fi++)
3562     {
3563 ph10 1238 pcre_uint32 cc; /* Faster than pcre_uchar */
3564 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM24);
3565 nigel 77 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3566 ph10 836 if (fi >= max) RRETURN(MATCH_NOMATCH);
3567 ph10 426 if (eptr >= md->end_subject)
3568     {
3569 ph10 427 SCHECK_PARTIAL();
3570 ph10 836 RRETURN(MATCH_NOMATCH);
3571 ph10 426 }
3572 chpe 1100 cc = RAWUCHARTEST(eptr);
3573     if (fc != cc && foc != cc) RRETURN(MATCH_NOMATCH);
3574 ph10 836 eptr++;
3575 nigel 77 }
3576     /* Control never gets here */
3577     }
3578 nigel 93 else /* Maximize */
3579 nigel 77 {
3580     pp = eptr;
3581     for (i = min; i < max; i++)
3582     {
3583 ph10 1238 pcre_uint32 cc; /* Faster than pcre_uchar */
3584 ph10 463 if (eptr >= md->end_subject)
3585 ph10 462 {
3586     SCHECK_PARTIAL();
3587     break;
3588 ph10 463 }
3589 chpe 1100 cc = RAWUCHARTEST(eptr);
3590     if (fc != cc && foc != cc) break;
3591 nigel 77 eptr++;
3592     }
3593 ph10 427
3594 ph10 1311 if (possessive) continue; /* No backtracking */
3595     for (;;)
3596 nigel 77 {
3597 ph10 1320 if (eptr == pp) goto TAIL_RECURSE;
3598 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM25);
3599 nigel 77 eptr--;
3600     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3601     }
3602 ph10 836 RRETURN(MATCH_NOMATCH);
3603 nigel 77 }
3604     /* Control never gets here */
3605     }
3606    
3607     /* Caseful comparisons (includes all multi-byte characters) */
3608    
3609     else
3610     {
3611 ph10 427 for (i = 1; i <= min; i++)
3612 ph10 426 {
3613     if (eptr >= md->end_subject)
3614     {
3615     SCHECK_PARTIAL();
3616 ph10 836 RRETURN(MATCH_NOMATCH);
3617 ph10 426 }
3618 chpe 1100 if (fc != RAWUCHARINCTEST(eptr)) RRETURN(MATCH_NOMATCH);
3619 ph10 427 }
3620 ph10 443
3621 nigel 77 if (min == max) continue;
3622 ph10 443
3623 nigel 77 if (minimize)
3624     {
3625     for (fi = min;; fi++)
3626     {
3627 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM26);
3628 nigel 77 if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3629 ph10 836 if (fi >= max) RRETURN(MATCH_NOMATCH);
3630 ph10 426 if (eptr >= md->end_subject)
3631 ph10 427 {
3632 ph10 426 SCHECK_PARTIAL();
3633 ph10 836 RRETURN(MATCH_NOMATCH);
3634 ph10 427 }
3635 chpe 1100 if (fc != RAWUCHARINCTEST(eptr)) RRETURN(MATCH_NOMATCH);
3636 nigel 77 }
3637     /* Control never gets here */
3638     }
3639 nigel 93 else /* Maximize */
3640 nigel 77 {
3641     pp = eptr;
3642     for (i = min; i < max; i++)
3643     {
3644 ph10 463 if (eptr >= md->end_subject)
3645 ph10 462 {
3646 ph10 463 SCHECK_PARTIAL();
3647 ph10 462 break;
3648 ph10 463 }
3649 chpe 1100 if (fc != RAWUCHARTEST(eptr)) break;
3650 nigel 77 eptr++;
3651     }
3652 ph10 1311 if (possessive) continue; /* No backtracking */
3653     for (;;)
3654 nigel 77 {
3655 ph10 1311 if (eptr == pp) goto TAIL_RECURSE;
3656 ph10 604 RMATCH(eptr, ecode, offset_top, md, eptrb, RM27);
3657 nigel 77 eptr--;
3658     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3659     }
3660 ph10 836 RRETURN(MATCH_NOMATCH);
3661 nigel 77 }
3662     }
3663     /* Control never gets here */
3664    
3665     /* Match a negated single one-byte character. The character we are
3666     checking can be multibyte. */
3667    
3668     case OP_NOT:
3669 ph10 625 case OP_NOTI:
3670 ph10 443 if (eptr >= md->end_subject)
3671 ph10 428 {
3672 ph10 443 SCHECK_PARTIAL();
3673 ph10 836 RRETURN(MATCH_NOMATCH);
3674 ph10 443 }
3675 zherczeg 924 #ifdef SUPPORT_UTF
3676     if (utf)
3677 nigel 77 {
3678 chpe 1092 register pcre_uint32 ch, och;
3679 zherczeg 924
3680     ecode++;
3681     GETCHARINC(ch, ecode);
3682     GETCHARINC(c, eptr);
3683    
3684     if (op == OP_NOT)
3685     {
3686     if (ch == c) RRETURN(MATCH_NOMATCH);
3687     }
3688     else
3689     {
3690 ph10 836 #ifdef SUPPORT_UCP
3691 zherczeg 924 if (ch > 127)
3692     och = UCD_OTHERCASE(ch);
3693 ph10 836 #else
3694 zherczeg 924 if (ch > 127)
3695     och = ch;
3696 ph10 836 #endif /* SUPPORT_UCP */
3697 zherczeg 924 else
3698     och = TABLE_GET(ch, md->fcc, ch);
3699     if (ch == c || och == c) RRETURN(MATCH_NOMATCH);
3700     }
3701 nigel 77 }
3702 zherczeg 924 else
3703     #endif
3704 nigel 77 {
3705 chpe 1092 register pcre_uint32 ch = ecode[1];
3706 zherczeg 924 c = *eptr++;
3707     if (ch == c || (op == OP_NOTI && TABLE_GET(ch, md->fcc, ch) == c))
3708     RRETURN(MATCH_NOMATCH);
3709     ecode += 2;
3710 nigel 77 }
3711     break;
3712    
3713     /* Match a negated single one-byte character repeatedly. This is almost a
3714     repeat of the code for a repeated single character, but I haven't found a
3715     nice way of commoning these up that doesn't require a test of the
3716     positive/negative option for each character match. Maybe that wouldn't add
3717     very much to the time taken, but character matching *is* what this is all
3718     about... */
3719    
3720     case OP_NOTEXACT:
3721 ph10 602 case OP_NOTEXACTI:
3722 nigel 77 min = max = GET2(ecode, 1);
3723 ph10 836 ecode += 1 + IMM2_SIZE;
3724 nigel 77 goto REPEATNOTCHAR;
3725    
3726     case OP_NOTUPTO:
3727 ph10 602 case OP_NOTUPTOI:
3728 nigel 77 case OP_NOTMINUPTO:
3729 ph10 602 case OP_NOTMINUPTOI:
3730 nigel 77 min = 0;
3731     max = GET2(ecode, 1);
3732 ph10 602 minimize = *ecode == OP_NOTMINUPTO || *ecode == OP_NOTMINUPTOI;
3733 ph10 836 ecode += 1 + IMM2_SIZE;
3734 nigel 77 goto REPEATNOTCHAR;
3735    
3736 nigel 93 case OP_NOTPOSSTAR:
3737 ph10 602 case OP_NOTPOSSTARI:
3738 nigel 93 possessive = TRUE;
3739     min = 0;
3740     max = INT_MAX;
3741     ecode++;
3742     goto REPEATNOTCHAR;
3743    
3744     case OP_NOTPOSPLUS:
3745 ph10 602 case OP_NOTPOSPLUSI:
3746 nigel 93 possessive = TRUE;
3747     min = 1;
3748     max = INT_MAX;
3749     ecode++;
3750     goto REPEATNOTCHAR;
3751    
3752     case OP_NOTPOSQUERY:
3753 ph10 602 case OP_NOTPOSQUERYI:
3754 nigel 93 possessive = TRUE;
3755     min = 0;
3756     max = 1;
3757     ecode++;
3758     goto REPEATNOTCHAR;
3759    
3760     case OP_NOTPOSUPTO:
3761 ph10 602 case OP_NOTPOSUPTOI:
3762 nigel 93 possessive = TRUE;
3763     min = 0;
3764     max = GET2(ecode, 1);
3765 ph10 836 ecode += 1 + IMM2_SIZE;
3766 nigel 93 goto REPEATNOTCHAR;
3767    
3768 nigel 77 case OP_NOTSTAR:
3769 ph10 602 case OP_NOTSTARI:
3770 nigel 77 case OP_NOTMINSTAR:
3771 ph10 602 case OP_NOTMINSTARI:
3772 nigel 77 case OP_NOTPLUS:
3773 ph10 602 case OP_NOTPLUSI:
3774 nigel 77 case OP_NOTMINPLUS:
3775 ph10 602 case OP_NOTMINPLUSI:
3776 nigel 77 case OP_NOTQUERY:
3777 ph10 602 case OP_NOTQUERYI:
3778 nigel 77 case OP_NOTMINQUERY:
3779 ph10 602 case OP_NOTMINQUERYI:
3780     c = *ecode++ - ((op >= OP_NOTSTARI)? OP_NOTSTARI: OP_NOTSTAR);
3781 nigel 77 minimize = (c & 1) != 0;
3782     min = rep_min[c]; /* Pick up values from tables; */
3783     max = rep_max[c]; /* zero for max => infinity */
3784     if (max == 0) max = INT_MAX;
3785    
3786 ph10 426 /* Common code for all repeated single-byte matches. */
3787 nigel 77
3788     REPEATNOTCHAR:
3789 zherczeg 924 GETCHARINCTEST(fc, ecode);
3790 nigel 77
3791     /* The code is duplicated for the caseless and caseful cases, for speed,
3792     since matching characters is likely to be quite common. First, ensure the
3793     minimum number of matches are present. If min = max, continue at the same
3794     level without recursing. Otherwise, if minimizing, keep trying the rest of
3795     the expression and advancing one matching character if failing, up to the
3796     maximum. Alternatively, if maximizing, find the maximum number of
3797     characters and work backwards. */
3798    
3799     DPRINTF(("negative matching %c{%d,%d} against subject %.*s\n", fc, min, max,
3800 ph10 976 max, (char *)eptr));
3801 nigel 77
3802 ph10 602 if (op >= OP_NOTSTARI) /* Caseless */
3803 nigel 77 {
3804 ph10 836 #ifdef SUPPORT_UTF
3805     #ifdef SUPPORT_UCP
3806     if (utf && fc > 127)
3807     foc = UCD_OTHERCASE(fc);
3808     #else
3809     if (utf && fc > 127)
3810     foc = fc;
3811     #endif /* SUPPORT_UCP */
3812     else
3813     #endif /* SUPPORT_UTF */
3814     foc = TABLE_GET(fc, md->fcc, fc);
3815 nigel 77
3816 ph10 836 #ifdef SUPPORT_UTF
3817     if (utf)
3818 nigel 77 {
3819 chpe 1100 register pcre_uint32 d;
3820 nigel 77 for (i = 1; i <= min; i++)
3821     {
3822 ph10 426 if (eptr >= md->end_subject)
3823     {
3824     SCHECK_PARTIAL();
3825 ph10 836 RRETURN(MATCH_NOMATCH);
3826 ph10 427 }
3827 nigel 77 GETCHARINC(d, eptr);
3828 zherczeg 924 if (fc == d || (unsigned int)foc == d) RRETURN(MATCH_NOMATCH);
3829 nigel 77 }
3830     }
3831     else
3832 ph10 1311 #endif /* SUPPORT_UTF */
3833 ph10 836 /* Not UTF mode */