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

Diff of /code/trunk/pcre_exec.c

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 150 by ph10, Tue Apr 17 08:22:40 2007 UTC revision 1298 by ph10, Fri Mar 22 16:13:13 2013 UTC
# Line 6  Line 6 
6  and semantics are as close as possible to those of the Perl 5 language.  and semantics are as close as possible to those of the Perl 5 language.
7    
8                         Written by Philip Hazel                         Written by Philip Hazel
9             Copyright (c) 1997-2007 University of Cambridge             Copyright (c) 1997-2013 University of Cambridge
10    
11  -----------------------------------------------------------------------------  -----------------------------------------------------------------------------
12  Redistribution and use in source and binary forms, with or without  Redistribution and use in source and binary forms, with or without
# Line 37  POSSIBILITY OF SUCH DAMAGE. Line 37  POSSIBILITY OF SUCH DAMAGE.
37  -----------------------------------------------------------------------------  -----------------------------------------------------------------------------
38  */  */
39    
   
40  /* This module contains pcre_exec(), the externally visible function that does  /* 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  pattern matching using an NFA algorithm, trying to mimic Perl as closely as
42  possible. There are also some static supporting functions. */  possible. There are also some static supporting functions. */
43    
44    #ifdef HAVE_CONFIG_H
45    #include "config.h"
46    #endif
47    
48  #define NLBLOCK md             /* Block containing newline information */  #define NLBLOCK md             /* Block containing newline information */
49  #define PSSTART start_subject  /* Field containing processed string start */  #define PSSTART start_subject  /* Field containing processed string start */
50  #define PSEND   end_subject    /* Field containing processed string end */  #define PSEND   end_subject    /* Field containing processed string end */
# Line 53  possible. There are also some static sup Line 56  possible. There are also some static sup
56  #undef min  #undef min
57  #undef max  #undef max
58    
59  /* The chain of eptrblocks for tail recursions uses memory in stack workspace,  /* The md->capture_last field uses the lower 16 bits for the last captured
60  obtained at top level, the size of which is defined by EPTR_WORK_SIZE. */  substring (which can never be greater than 65535) and a bit in the top half
61    to mean "capture vector overflowed". This odd way of doing things was
62  #define EPTR_WORK_SIZE (1000)  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    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    /* 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    as stack usage is to be discouraged. */
76    
77  /* Flag bits for the match() function */  #define MATCH_CONDASSERT     1  /* Called to check a condition assertion */
78    #define MATCH_CBEGROUP       2  /* Could-be-empty unlimited repeat group */
 #define match_condassert     0x01  /* Called to check a condition assertion */  
 #define match_cbegroup       0x02  /* Could-be-empty unlimited repeat group */  
 #define match_tail_recursed  0x04  /* Tail recursive call */  
79    
80  /* Non-error returns from the match() function. Error returns are externally  /* Non-error returns from the match() function. Error returns are externally
81  defined PCRE_ERROR_xxx codes, which are all negative. */  defined PCRE_ERROR_xxx codes, which are all negative. */
# Line 70  defined PCRE_ERROR_xxx codes, which are Line 83  defined PCRE_ERROR_xxx codes, which are
83  #define MATCH_MATCH        1  #define MATCH_MATCH        1
84  #define MATCH_NOMATCH      0  #define MATCH_NOMATCH      0
85    
86    /* Special internal returns from the match() function. Make them sufficiently
87    negative to avoid the external error codes. */
88    
89    #define MATCH_ACCEPT       (-999)
90    #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    #define MATCH_PRUNE        (-995)
96    #define MATCH_SKIP         (-994)
97    #define MATCH_SKIP_ARG     (-993)
98    #define MATCH_THEN         (-992)
99    #define MATCH_BACKTRACK_MAX MATCH_THEN
100    #define MATCH_BACKTRACK_MIN MATCH_COMMIT
101    
102  /* Maximum number of ints of offset to save on the stack for recursive calls.  /* 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,  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. */  because the offset vector is always a multiple of 3 long. */
# Line 81  because the offset vector is always a mu Line 110  because the offset vector is always a mu
110  static const char rep_min[] = { 0, 0, 1, 1, 0, 0 };  static const char rep_min[] = { 0, 0, 1, 1, 0, 0 };
111  static const char rep_max[] = { 0, 0, 0, 0, 1, 1 };  static const char rep_max[] = { 0, 0, 0, 0, 1, 1 };
112    
113    #ifdef PCRE_DEBUG
   
 #ifdef DEBUG  
114  /*************************************************  /*************************************************
115  *        Debugging function to print chars       *  *        Debugging function to print chars       *
116  *************************************************/  *************************************************/
# Line 101  Returns:     nothing Line 128  Returns:     nothing
128  */  */
129    
130  static void  static void
131  pchars(const uschar *p, int length, BOOL is_subject, match_data *md)  pchars(const pcre_uchar *p, int length, BOOL is_subject, match_data *md)
132  {  {
133  unsigned int c;  pcre_uint32 c;
134    BOOL utf = md->utf;
135  if (is_subject && length > md->end_subject - p) length = md->end_subject - p;  if (is_subject && length > md->end_subject - p) length = md->end_subject - p;
136  while (length-- > 0)  while (length-- > 0)
137    if (isprint(c = *(p++))) printf("%c", c); else printf("\\x%02x", c);    if (isprint(c = RAWUCHARINCTEST(p))) printf("%c", (char)c); else printf("\\x{%02x}", c);
138  }  }
139  #endif  #endif
140    
# Line 116  while (length-- > 0) Line 144  while (length-- > 0)
144  *          Match a back-reference                *  *          Match a back-reference                *
145  *************************************************/  *************************************************/
146    
147  /* If a back reference hasn't been set, the length that is passed is greater  /* Normally, if a back reference hasn't been set, the length that is passed is
148  than the number of characters left in the string, so the match fails.  negative, so the match always fails. However, in JavaScript compatibility mode,
149    the length passed is zero. Note that in caseless UTF-8 mode, the number of
150    subject bytes matched may be different to the number of reference bytes.
151    
152  Arguments:  Arguments:
153    offset      index into the offset vector    offset      index into the offset vector
154    eptr        points into the subject    eptr        pointer into the subject
155    length      length to be matched    length      length of reference to be matched (number of bytes)
156    md          points to match data block    md          points to match data block
157    ims         the ims flags    caseless    TRUE if caseless
158    
159  Returns:      TRUE if matched  Returns:      >= 0 the number of subject bytes matched
160                  -1 no match
161                  -2 partial match; always given if at end subject
162  */  */
163    
164  static BOOL  static int
165  match_ref(int offset, register USPTR eptr, int length, match_data *md,  match_ref(int offset, register PCRE_PUCHAR eptr, int length, match_data *md,
166    unsigned long int ims)    BOOL caseless)
167  {  {
168  USPTR p = md->start_subject + md->offset_vector[offset];  PCRE_PUCHAR eptr_start = eptr;
169    register PCRE_PUCHAR p = md->start_subject + md->offset_vector[offset];
170    #ifdef SUPPORT_UTF
171    BOOL utf = md->utf;
172    #endif
173    
174  #ifdef DEBUG  #ifdef PCRE_DEBUG
175  if (eptr >= md->end_subject)  if (eptr >= md->end_subject)
176    printf("matching subject <null>");    printf("matching subject <null>");
177  else  else
# Line 148  pchars(p, length, FALSE, md); Line 184  pchars(p, length, FALSE, md);
184  printf("\n");  printf("\n");
185  #endif  #endif
186    
187  /* Always fail if not enough characters left */  /* Always fail if reference not set (and not JavaScript compatible - in that
188    case the length is passed as zero). */
189    
190  if (length > md->end_subject - eptr) return FALSE;  if (length < 0) return -1;
191    
192  /* Separate the caselesss case for speed */  /* 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    
196  if ((ims & PCRE_CASELESS) != 0)  if (caseless)
197    {    {
198    while (length-- > 0)  #ifdef SUPPORT_UTF
199      if (md->lcc[*p++] != md->lcc[*eptr++]) return FALSE;  #ifdef SUPPORT_UCP
200      if (utf)
201        {
202        /* Match characters up to the end of the reference. NOTE: the number of
203        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    
211        PCRE_PUCHAR endptr = p + length;
212        while (p < endptr)
213          {
214          pcre_uint32 c, d;
215          const ucd_record *ur;
216          if (eptr >= md->end_subject) return -2;   /* Partial match */
217          GETCHARINC(c, eptr);
218          GETCHARINC(d, p);
219          ur = GET_UCD(d);
220          if (c != d && c != d + ur->other_case)
221            {
222            const pcre_uint32 *pp = PRIV(ucd_caseless_sets) + ur->caseset;
223            for (;;)
224              {
225              if (c < *pp) return -1;
226              if (c == *pp++) break;
227              }
228            }
229          }
230        }
231      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        {
238        while (length-- > 0)
239          {
240          pcre_uint32 cc, cp;
241          if (eptr >= md->end_subject) return -2;   /* Partial match */
242          cc = RAWUCHARTEST(eptr);
243          cp = RAWUCHARTEST(p);
244          if (TABLE_GET(cp, md->lcc, cp) != TABLE_GET(cc, md->lcc, cc)) return -1;
245          p++;
246          eptr++;
247          }
248        }
249    }    }
250    
251    /* In the caseful case, we can just compare the bytes, whether or not we
252    are in UTF-8 mode. */
253    
254  else  else
255    { while (length-- > 0) if (*p++ != *eptr++) return FALSE; }    {
256      while (length-- > 0)
257        {
258        if (eptr >= md->end_subject) return -2;   /* Partial match */
259        if (RAWUCHARINCTEST(p) != RAWUCHARINCTEST(eptr)) return -1;
260        }
261      }
262    
263  return TRUE;  return (int)(eptr - eptr_start);
264  }  }
265    
266    
# Line 188  calls by keeping local variables that ne Line 286  calls by keeping local variables that ne
286  obtained from malloc() instead instead of on the stack. Macros are used to  obtained from malloc() instead instead of on the stack. Macros are used to
287  achieve this so that the actual code doesn't look very different to what it  achieve this so that the actual code doesn't look very different to what it
288  always used to.  always used to.
289    
290    The original heap-recursive code used longjmp(). However, it seems that this
291    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    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    reduced because the result can be "passed back" by straight setting of the
302    variable instead of being passed in the frame.
303  ****************************************************************************  ****************************************************************************
304  ***************************************************************************/  ***************************************************************************/
305    
306    /* Numbers for RMATCH calls. When this list is changed, the code at HEAP_RETURN
307    below must be updated in sync.  */
308    
309    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           RM41,  RM42, RM43, RM44, RM45, RM46, RM47, RM48, RM49, RM50,
314           RM51,  RM52, RM53, RM54, RM55, RM56, RM57, RM58, RM59, RM60,
315           RM61,  RM62, RM63, RM64, RM65, RM66, RM67, RM68 };
316    
317  /* These versions of the macros use the stack, as normal. There are debugging  /* These versions of the macros use the stack, as normal. There are debugging
318  versions and production versions. */  versions and production versions. Note that the "rw" argument of RMATCH isn't
319    actually used in this definition. */
320    
321  #ifndef NO_RECURSE  #ifndef NO_RECURSE
322  #define REGISTER register  #define REGISTER register
323  #ifdef DEBUG  
324  #define RMATCH(rx,ra,rb,rc,rd,re,rf,rg) \  #ifdef PCRE_DEBUG
325    #define RMATCH(ra,rb,rc,rd,re,rw) \
326    { \    { \
327    printf("match() called in line %d\n", __LINE__); \    printf("match() called in line %d\n", __LINE__); \
328    rx = match(ra,rb,rc,rd,re,rf,rg,rdepth+1); \    rrc = match(ra,rb,mstart,rc,rd,re,rdepth+1); \
329    printf("to line %d\n", __LINE__); \    printf("to line %d\n", __LINE__); \
330    }    }
331  #define RRETURN(ra) \  #define RRETURN(ra) \
332    { \    { \
333    printf("match() returned %d from line %d ", ra, __LINE__); \    printf("match() returned %d from line %d\n", ra, __LINE__); \
334    return ra; \    return ra; \
335    }    }
336  #else  #else
337  #define RMATCH(rx,ra,rb,rc,rd,re,rf,rg) \  #define RMATCH(ra,rb,rc,rd,re,rw) \
338    rx = match(ra,rb,rc,rd,re,rf,rg,rdepth+1)    rrc = match(ra,rb,mstart,rc,rd,re,rdepth+1)
339  #define RRETURN(ra) return ra  #define RRETURN(ra) return ra
340  #endif  #endif
341    
342  #else  #else
343    
344    
345  /* These versions of the macros manage a private stack on the heap. Note  /* These versions of the macros manage a private stack on the heap. Note that
346  that the rd argument of RMATCH isn't actually used. It's the md argument of  the "rd" argument of RMATCH isn't actually used in this definition. It's the md
347  match(), which never changes. */  argument of match(), which never changes. */
348    
349  #define REGISTER  #define REGISTER
350    
351  #define RMATCH(rx,ra,rb,rc,rd,re,rf,rg)\  #define RMATCH(ra,rb,rc,rd,re,rw)\
352    {\    {\
353    heapframe *newframe = (pcre_stack_malloc)(sizeof(heapframe));\    heapframe *newframe = frame->Xnextframe;\
354    if (setjmp(frame->Xwhere) == 0)\    if (newframe == NULL)\
     {\  
     newframe->Xeptr = ra;\  
     newframe->Xecode = rb;\  
     newframe->Xoffset_top = rc;\  
     newframe->Xims = re;\  
     newframe->Xeptrb = rf;\  
     newframe->Xflags = rg;\  
     newframe->Xrdepth = frame->Xrdepth + 1;\  
     newframe->Xprevframe = frame;\  
     frame = newframe;\  
     DPRINTF(("restarting from line %d\n", __LINE__));\  
     goto HEAP_RECURSE;\  
     }\  
   else\  
355      {\      {\
356      DPRINTF(("longjumped back to line %d\n", __LINE__));\      newframe = (heapframe *)(PUBL(stack_malloc))(sizeof(heapframe));\
357      frame = md->thisframe;\      if (newframe == NULL) RRETURN(PCRE_ERROR_NOMEMORY);\
358      rx = frame->Xresult;\      newframe->Xnextframe = NULL;\
359        frame->Xnextframe = newframe;\
360      }\      }\
361      frame->Xwhere = rw;\
362      newframe->Xeptr = ra;\
363      newframe->Xecode = rb;\
364      newframe->Xmstart = mstart;\
365      newframe->Xoffset_top = rc;\
366      newframe->Xeptrb = re;\
367      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    }    }
375    
376  #define RRETURN(ra)\  #define RRETURN(ra)\
377    {\    {\
378    heapframe *newframe = frame;\    heapframe *oldframe = frame;\
379    frame = newframe->Xprevframe;\    frame = oldframe->Xprevframe;\
   (pcre_stack_free)(newframe);\  
380    if (frame != NULL)\    if (frame != NULL)\
381      {\      {\
382      frame->Xresult = ra;\      rrc = ra;\
383      md->thisframe = frame;\      goto HEAP_RETURN;\
     longjmp(frame->Xwhere, 1);\  
384      }\      }\
385    return ra;\    return ra;\
386    }    }
# Line 268  match(), which never changes. */ Line 390  match(), which never changes. */
390    
391  typedef struct heapframe {  typedef struct heapframe {
392    struct heapframe *Xprevframe;    struct heapframe *Xprevframe;
393      struct heapframe *Xnextframe;
394    
395    /* Function arguments that may change */    /* Function arguments that may change */
396    
397    const uschar *Xeptr;    PCRE_PUCHAR Xeptr;
398    const uschar *Xecode;    const pcre_uchar *Xecode;
399      PCRE_PUCHAR Xmstart;
400    int Xoffset_top;    int Xoffset_top;
   long int Xims;  
401    eptrblock *Xeptrb;    eptrblock *Xeptrb;
   int Xflags;  
402    unsigned int Xrdepth;    unsigned int Xrdepth;
403    
404    /* Function local variables */    /* Function local variables */
405    
406    const uschar *Xcallpat;    PCRE_PUCHAR Xcallpat;
407    const uschar *Xcharptr;  #ifdef SUPPORT_UTF
408    const uschar *Xdata;    PCRE_PUCHAR Xcharptr;
409    const uschar *Xnext;  #endif
410    const uschar *Xpp;    PCRE_PUCHAR Xdata;
411    const uschar *Xprev;    PCRE_PUCHAR Xnext;
412    const uschar *Xsaved_eptr;    PCRE_PUCHAR Xpp;
413      PCRE_PUCHAR Xprev;
414      PCRE_PUCHAR Xsaved_eptr;
415    
416    recursion_info Xnew_recursive;    recursion_info Xnew_recursive;
417    
# Line 295  typedef struct heapframe { Line 419  typedef struct heapframe {
419    BOOL Xcondition;    BOOL Xcondition;
420    BOOL Xprev_is_word;    BOOL Xprev_is_word;
421    
   unsigned long int Xoriginal_ims;  
   
422  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
423    int Xprop_type;    int Xprop_type;
424    int Xprop_value;    unsigned int Xprop_value;
425    int Xprop_fail_result;    int Xprop_fail_result;
   int Xprop_category;  
   int Xprop_chartype;  
   int Xprop_script;  
426    int Xoclength;    int Xoclength;
427    uschar Xocchars[8];    pcre_uchar Xocchars[6];
428  #endif  #endif
429    
430      int Xcodelink;
431    int Xctype;    int Xctype;
432    unsigned int Xfc;    unsigned int Xfc;
433    int Xfi;    int Xfi;
434    int Xlength;    int Xlength;
435    int Xmax;    int Xmax;
436    int Xmin;    int Xmin;
437    int Xnumber;    unsigned int Xnumber;
438    int Xoffset;    int Xoffset;
439    int Xop;    unsigned int Xop;
440    int Xsave_capture_last;    pcre_int32 Xsave_capture_last;
441    int Xsave_offset1, Xsave_offset2, Xsave_offset3;    int Xsave_offset1, Xsave_offset2, Xsave_offset3;
442    int Xstacksave[REC_STACK_SAVE_MAX];    int Xstacksave[REC_STACK_SAVE_MAX];
443    
444    eptrblock Xnewptrb;    eptrblock Xnewptrb;
445    
446    /* Place to pass back result, and where to jump back to */    /* Where to jump back to */
447    
448    int  Xresult;    int Xwhere;
   jmp_buf Xwhere;  
449    
450  } heapframe;  } heapframe;
451    
# Line 344  typedef struct heapframe { Line 463  typedef struct heapframe {
463    
464  /* This function is called recursively in many circumstances. Whenever it  /* This function is called recursively in many circumstances. Whenever it
465  returns a negative (error) response, the outer incarnation must also return the  returns a negative (error) response, the outer incarnation must also return the
466  same response.  same response. */
467    
468    /* These macros pack up tests that are used for partial matching, and which
469    appear several times in the code. We set the "hit end" flag if the pointer is
470    at the end of the subject and also past the start of the subject (i.e.
471    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    
475    #define CHECK_PARTIAL()\
476      if (md->partial != 0 && eptr >= md->end_subject && \
477          eptr > md->start_used_ptr) \
478        { \
479        md->hitend = TRUE; \
480        if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL); \
481        }
482    
483  Performance note: It might be tempting to extract commonly used fields from the  #define SCHECK_PARTIAL()\
484  md structure (e.g. utf8, end_subject) into individual variables to improve    if (md->partial != 0 && eptr > md->start_used_ptr) \
485        { \
486        md->hitend = TRUE; \
487        if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL); \
488        }
489    
490    
491    /* Performance note: It might be tempting to extract commonly used fields from
492    the md structure (e.g. utf, end_subject) into individual variables to improve
493  performance. Tests using gcc on a SPARC disproved this; in the first case, it  performance. Tests using gcc on a SPARC disproved this; in the first case, it
494  made performance worse.  made performance worse.
495    
496  Arguments:  Arguments:
497     eptr        pointer to current character in subject     eptr        pointer to current character in subject
498     ecode       pointer to current position in compiled code     ecode       pointer to current position in compiled code
499       mstart      pointer to the current match start position (can be modified
500                     by encountering \K)
501     offset_top  current top pointer     offset_top  current top pointer
502     md          pointer to "static" info for the match     md          pointer to "static" info for the match
    ims         current /i, /m, and /s options  
503     eptrb       pointer to chain of blocks containing eptr at start of     eptrb       pointer to chain of blocks containing eptr at start of
504                   brackets - for testing for empty matches                   brackets - for testing for empty matches
    flags       can contain  
                  match_condassert - this is an assertion condition  
                  match_cbegroup - this is the start of an unlimited repeat  
                    group that can match an empty string  
                  match_tail_recursed - this is a tail_recursed group  
505     rdepth      the recursion depth     rdepth      the recursion depth
506    
507  Returns:       MATCH_MATCH if matched            )  these values are >= 0  Returns:       MATCH_MATCH if matched            )  these values are >= 0
508                 MATCH_NOMATCH if failed to match  )                 MATCH_NOMATCH if failed to match  )
509                   a negative MATCH_xxx value for PRUNE, SKIP, etc
510                 a negative PCRE_ERROR_xxx value if aborted by an error condition                 a negative PCRE_ERROR_xxx value if aborted by an error condition
511                   (e.g. stopped by repeated call or recursion limit)                   (e.g. stopped by repeated call or recursion limit)
512  */  */
513    
514  static int  static int
515  match(REGISTER USPTR eptr, REGISTER const uschar *ecode,  match(REGISTER PCRE_PUCHAR eptr, REGISTER const pcre_uchar *ecode,
516    int offset_top, match_data *md, unsigned long int ims, eptrblock *eptrb,    PCRE_PUCHAR mstart, int offset_top, match_data *md, eptrblock *eptrb,
517    int flags, unsigned int rdepth)    unsigned int rdepth)
518  {  {
519  /* These variables do not need to be preserved over recursion in this function,  /* These variables do not need to be preserved over recursion in this function,
520  so they can be ordinary variables in all cases. Mark some of them with  so they can be ordinary variables in all cases. Mark some of them with
# Line 383  so they can be ordinary variables in all Line 522  so they can be ordinary variables in all
522    
523  register int  rrc;         /* Returns from recursive calls */  register int  rrc;         /* Returns from recursive calls */
524  register int  i;           /* Used for loops not involving calls to RMATCH() */  register int  i;           /* Used for loops not involving calls to RMATCH() */
525  register unsigned int c;   /* Character values not kept over RMATCH() calls */  register pcre_uint32 c;    /* Character values not kept over RMATCH() calls */
526  register BOOL utf8;        /* Local copy of UTF-8 flag for speed */  register BOOL utf;         /* Local copy of UTF flag for speed */
527    
528  BOOL minimize, possessive; /* Quantifier options */  BOOL minimize, possessive; /* Quantifier options */
529    BOOL caseless;
530    int condcode;
531    
532  /* When recursion is not being used, all "local" variables that have to be  /* When recursion is not being used, all "local" variables that have to be
533  preserved over calls to RMATCH() are part of a "frame" which is obtained from  preserved over calls to RMATCH() are part of a "frame". We set up the top-level
534  heap storage. Set up the top-level frame here; others are obtained from the  frame on the stack here; subsequent instantiations are obtained from the heap
535  heap whenever RMATCH() does a "recursion". See the macro definitions above. */  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    boost in many cases where there is not much "recursion". */
538    
539  #ifdef NO_RECURSE  #ifdef NO_RECURSE
540  heapframe *frame = (pcre_stack_malloc)(sizeof(heapframe));  heapframe *frame = (heapframe *)md->match_frames_base;
 frame->Xprevframe = NULL;            /* Marks the top level */  
541    
542  /* Copy in the original argument variables */  /* Copy in the original argument variables */
543    
544  frame->Xeptr = eptr;  frame->Xeptr = eptr;
545  frame->Xecode = ecode;  frame->Xecode = ecode;
546    frame->Xmstart = mstart;
547  frame->Xoffset_top = offset_top;  frame->Xoffset_top = offset_top;
 frame->Xims = ims;  
548  frame->Xeptrb = eptrb;  frame->Xeptrb = eptrb;
 frame->Xflags = flags;  
549  frame->Xrdepth = rdepth;  frame->Xrdepth = rdepth;
550    
551  /* This is where control jumps back to to effect "recursion" */  /* This is where control jumps back to to effect "recursion" */
# Line 415  HEAP_RECURSE: Line 556  HEAP_RECURSE:
556    
557  #define eptr               frame->Xeptr  #define eptr               frame->Xeptr
558  #define ecode              frame->Xecode  #define ecode              frame->Xecode
559    #define mstart             frame->Xmstart
560  #define offset_top         frame->Xoffset_top  #define offset_top         frame->Xoffset_top
 #define ims                frame->Xims  
561  #define eptrb              frame->Xeptrb  #define eptrb              frame->Xeptrb
 #define flags              frame->Xflags  
562  #define rdepth             frame->Xrdepth  #define rdepth             frame->Xrdepth
563    
564  /* Ditto for the local variables */  /* Ditto for the local variables */
565    
566  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
567  #define charptr            frame->Xcharptr  #define charptr            frame->Xcharptr
568  #endif  #endif
569  #define callpat            frame->Xcallpat  #define callpat            frame->Xcallpat
570    #define codelink           frame->Xcodelink
571  #define data               frame->Xdata  #define data               frame->Xdata
572  #define next               frame->Xnext  #define next               frame->Xnext
573  #define pp                 frame->Xpp  #define pp                 frame->Xpp
# Line 439  HEAP_RECURSE: Line 580  HEAP_RECURSE:
580  #define condition          frame->Xcondition  #define condition          frame->Xcondition
581  #define prev_is_word       frame->Xprev_is_word  #define prev_is_word       frame->Xprev_is_word
582    
 #define original_ims       frame->Xoriginal_ims  
   
583  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
584  #define prop_type          frame->Xprop_type  #define prop_type          frame->Xprop_type
585  #define prop_value         frame->Xprop_value  #define prop_value         frame->Xprop_value
586  #define prop_fail_result   frame->Xprop_fail_result  #define prop_fail_result   frame->Xprop_fail_result
 #define prop_category      frame->Xprop_category  
 #define prop_chartype      frame->Xprop_chartype  
 #define prop_script        frame->Xprop_script  
587  #define oclength           frame->Xoclength  #define oclength           frame->Xoclength
588  #define occhars            frame->Xocchars  #define occhars            frame->Xocchars
589  #endif  #endif
# Line 477  i, and fc and c, can be the same variabl Line 613  i, and fc and c, can be the same variabl
613  #define fi i  #define fi i
614  #define fc c  #define fc c
615    
616    /* 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    
624  #ifdef SUPPORT_UTF8                /* Many of these variables are used only  */  #ifdef SUPPORT_UTF
625  const uschar *charptr;             /* in small blocks of the code. My normal */  const pcre_uchar *charptr;
626  #endif                             /* style of coding would have declared    */  #endif
627  const uschar *callpat;             /* them within each of those blocks.      */  const pcre_uchar *callpat;
628  const uschar *data;                /* However, in order to accommodate the   */  const pcre_uchar *data;
629  const uschar *next;                /* version of this code that uses an      */  const pcre_uchar *next;
630  USPTR         pp;                  /* external "stack" implemented on the    */  PCRE_PUCHAR       pp;
631  const uschar *prev;                /* heap, it is easier to declare them all */  const pcre_uchar *prev;
632  USPTR         saved_eptr;          /* here, so the declarations can be cut   */  PCRE_PUCHAR       saved_eptr;
633                                     /* out in a block. The only declarations  */  
634  recursion_info new_recursive;      /* within blocks below are for variables  */  recursion_info new_recursive;
635                                     /* that do not have to be preserved over  */  
636  BOOL cur_is_word;                  /* a recursive call to RMATCH().          */  BOOL cur_is_word;
637  BOOL condition;  BOOL condition;
638  BOOL prev_is_word;  BOOL prev_is_word;
639    
 unsigned long int original_ims;  
   
640  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
641  int prop_type;  int prop_type;
642  int prop_value;  unsigned int prop_value;
643  int prop_fail_result;  int prop_fail_result;
 int prop_category;  
 int prop_chartype;  
 int prop_script;  
644  int oclength;  int oclength;
645  uschar occhars[8];  pcre_uchar occhars[6];
646  #endif  #endif
647    
648    int codelink;
649  int ctype;  int ctype;
650  int length;  int length;
651  int max;  int max;
652  int min;  int min;
653  int number;  unsigned int number;
654  int offset;  int offset;
655  int op;  unsigned int op;
656  int save_capture_last;  pcre_int32 save_capture_last;
657  int save_offset1, save_offset2, save_offset3;  int save_offset1, save_offset2, save_offset3;
658  int stacksave[REC_STACK_SAVE_MAX];  int stacksave[REC_STACK_SAVE_MAX];
659    
660  eptrblock newptrb;  eptrblock newptrb;
661    
662    /* There is a special fudge for calling match() in a way that causes it to
663    measure the size of its basic stack frame when the stack is being used for
664    recursion. The second argument (ecode) being NULL triggers this behaviour. It
665    cannot normally ever be NULL. The return is the negated value of the frame
666    size. */
667    
668    if (ecode == NULL)
669      {
670      if (rdepth == 0)
671        return match((PCRE_PUCHAR)&rdepth, NULL, NULL, 0, NULL, NULL, 1);
672      else
673        {
674        int len = (char *)&rdepth - (char *)eptr;
675        return (len > 0)? -len : len;
676        }
677      }
678  #endif     /* NO_RECURSE */  #endif     /* NO_RECURSE */
679    
680    /* 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    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    #define foc           number
691    #define save_mark     data
692    
693  /* These statements are here to stop the compiler complaining about unitialized  /* These statements are here to stop the compiler complaining about unitialized
694  variables. */  variables. */
695    
# Line 540  TAIL_RECURSE: Line 709  TAIL_RECURSE:
709  /* OK, now we can get on with the real code of the function. Recursive calls  /* 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  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()  NO_RECURSE is *not* defined, these just turn into a recursive call to match()
712  and a "return", respectively (possibly with some debugging if DEBUG is  and a "return", respectively (possibly with some debugging if PCRE_DEBUG is
713  defined). However, RMATCH isn't like a function call because it's quite a  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,  complicated macro. It has to be used in one particular way. This shouldn't,
715  however, impact performance when true recursion is being used. */  however, impact performance when true recursion is being used. */
716    
717    #ifdef SUPPORT_UTF
718    utf = md->utf;       /* Local copy of the flag */
719    #else
720    utf = FALSE;
721    #endif
722    
723  /* First check that we haven't called match() too many times, or that we  /* First check that we haven't called match() too many times, or that we
724  haven't exceeded the recursive call limit. */  haven't exceeded the recursive call limit. */
725    
726  if (md->match_call_count++ >= md->match_limit) RRETURN(PCRE_ERROR_MATCHLIMIT);  if (md->match_call_count++ >= md->match_limit) RRETURN(PCRE_ERROR_MATCHLIMIT);
727  if (rdepth >= md->match_limit_recursion) RRETURN(PCRE_ERROR_RECURSIONLIMIT);  if (rdepth >= md->match_limit_recursion) RRETURN(PCRE_ERROR_RECURSIONLIMIT);
728    
 original_ims = ims;    /* Save for resetting on ')' */  
   
 #ifdef SUPPORT_UTF8  
 utf8 = md->utf8;       /* Local copy of the flag */  
 #else  
 utf8 = FALSE;  
 #endif  
   
729  /* At the start of a group with an unlimited repeat that may match an empty  /* At the start of a group with an unlimited repeat that may match an empty
730  string, the match_cbegroup flag is set. When this is the case, add the current  string, the variable md->match_function_type is set to MATCH_CBEGROUP. It is
731  subject pointer to the chain of such remembered pointers, to be checked when we  done this way to save having to use another function argument, which would take
732  hit the closing ket, in order to break infinite loops that match no characters.  up space on the stack. See also MATCH_CONDASSERT below.
733  When match() is called in other circumstances, don't add to the chain. If this  
734  is a tail recursion, use a block from the workspace, as the one on the stack is  When MATCH_CBEGROUP is set, add the current subject pointer to the chain of
735  already used. */  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 ((flags & match_cbegroup) != 0)  if (md->match_function_type == MATCH_CBEGROUP)
742    {    {
743    eptrblock *p;    newptrb.epb_saved_eptr = eptr;
744    if ((flags & match_tail_recursed) != 0)    newptrb.epb_prev = eptrb;
745      {    eptrb = &newptrb;
746      if (md->eptrn >= EPTR_WORK_SIZE) RRETURN(PCRE_ERROR_NULLWSLIMIT);    md->match_function_type = 0;
     p = md->eptrchain + md->eptrn++;  
     }  
   else p = &newptrb;  
   p->epb_saved_eptr = eptr;  
   p->epb_prev = eptrb;  
   eptrb = p;  
747    }    }
748    
749  /* Now start processing the opcodes. */  /* Now start processing the opcodes. */
# Line 588  for (;;) Line 753  for (;;)
753    minimize = possessive = FALSE;    minimize = possessive = FALSE;
754    op = *ecode;    op = *ecode;
755    
   /* For partial matching, remember if we ever hit the end of the subject after  
   matching at least one subject character. */  
   
   if (md->partial &&  
       eptr >= md->end_subject &&  
       eptr > md->start_match)  
     md->hitend = TRUE;  
   
756    switch(op)    switch(op)
757      {      {
758      /* Handle a capturing bracket. If there is space in the offset vector, save      case OP_MARK:
759      the current subject position in the working slot at the top of the vector.      md->nomatch_mark = ecode + 2;
760      We mustn't change the current values of the data slot, because they may be      md->mark = NULL;    /* In case previously set by assertion */
761      set from a previous iteration of this group, and be referred to by a      RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
762      reference inside the group.        eptrb, RM55);
763        if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
764      If the bracket fails to match, we need to restore this value and also the           md->mark == NULL) md->mark = ecode + 2;
765      values of the final offsets, in case they were set by a previous iteration  
766      of the same bracket.      /* 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        unaltered. */
772    
773        else if (rrc == MATCH_SKIP_ARG &&
774            STRCMP_UC_UC_TEST(ecode + 2, md->start_match_ptr) == 0)
775          {
776          md->start_match_ptr = eptr;
777          RRETURN(MATCH_SKIP);
778          }
779        RRETURN(rrc);
780    
781        case OP_FAIL:
782        RRETURN(MATCH_NOMATCH);
783    
784        case OP_COMMIT:
785        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
786          eptrb, RM52);
787        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
788        RRETURN(MATCH_COMMIT);
789    
790        case OP_PRUNE:
791        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
792          eptrb, RM51);
793        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
794        RRETURN(MATCH_PRUNE);
795    
796        case OP_PRUNE_ARG:
797        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          eptrb, RM56);
801        if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
802             md->mark == NULL) md->mark = ecode + 2;
803        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
804        RRETURN(MATCH_PRUNE);
805    
806        case OP_SKIP:
807        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
808          eptrb, RM53);
809        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
810        md->start_match_ptr = eptr;   /* Pass back current position */
811        RRETURN(MATCH_SKIP);
812    
813        /* Note that, for Perl compatibility, SKIP with an argument does NOT set
814        nomatch_mark. When a pattern match ends with a SKIP_ARG for which there was
815        not a matching mark, we have to re-run the match, ignoring the SKIP_ARG
816        that failed and any that preceed it (either they also failed, or were not
817        triggered). To do this, we maintain a count of executed SKIP_ARGs. If a
818        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    
821        case OP_SKIP_ARG:
822        md->skip_arg_count++;
823        if (md->skip_arg_count <= md->ignore_skip_arg)
824          {
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          eptrb, RM57);
830        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
831    
832        /* 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        caught by a matching MARK, or get to the top, where it causes a rematch
835        with md->ignore_skip_arg set to the value of md->skip_arg_count. */
836    
837        md->start_match_ptr = ecode + 2;
838        RRETURN(MATCH_SKIP_ARG);
839    
840        /* 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    
844        case OP_THEN:
845        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
846          eptrb, RM54);
847        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
848        md->start_match_ptr = ecode;
849        RRETURN(MATCH_THEN);
850    
851        case OP_THEN_ARG:
852        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          md, eptrb, RM58);
856        if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
857             md->mark == NULL) md->mark = ecode + 2;
858        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
859        md->start_match_ptr = ecode;
860        RRETURN(MATCH_THEN);
861    
862        /* Handle an atomic group that does not contain any capturing parentheses.
863        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        less stack intensive way.
869    
870        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        save_mark = md->mark;
879        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            if (md->start_match_ptr < next &&
891                (*ecode == OP_ALT || *next == OP_ALT))
892              rrc = MATCH_NOMATCH;
893            }
894    
895          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
896          ecode += GET(ecode,1);
897          md->mark = save_mark;
898          }
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        /* 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        the group. A failure to match might occur after the group has succeeded,
950        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    
954      If there isn't enough space in the offset vector, treat this as if it were      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      a non-capturing bracket. Don't worry about setting the flag for the error
# Line 617  for (;;) Line 960  for (;;)
960      number = GET2(ecode, 1+LINK_SIZE);      number = GET2(ecode, 1+LINK_SIZE);
961      offset = number << 1;      offset = number << 1;
962    
963  #ifdef DEBUG  #ifdef PCRE_DEBUG
964      printf("start bracket %d\n", number);      printf("start bracket %d\n", number);
965      printf("subject=");      printf("subject=");
966      pchars(eptr, 16, TRUE, md);      pchars(eptr, 16, TRUE, md);
# Line 630  for (;;) Line 973  for (;;)
973        save_offset2 = md->offset_vector[offset+1];        save_offset2 = md->offset_vector[offset+1];
974        save_offset3 = md->offset_vector[md->offset_end - number];        save_offset3 = md->offset_vector[md->offset_end - number];
975        save_capture_last = md->capture_last;        save_capture_last = md->capture_last;
976          save_mark = md->mark;
977    
978        DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));        DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
979        md->offset_vector[md->offset_end - number] = eptr - md->start_subject;        md->offset_vector[md->offset_end - number] =
980            (int)(eptr - md->start_subject);
981    
982        flags = (op == OP_SCBRA)? match_cbegroup : 0;        for (;;)
       do  
983          {          {
984          RMATCH(rrc, eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,          if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
985            ims, eptrb, flags);          RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
986              eptrb, RM1);
987            if (rrc == MATCH_ONCE) break;  /* Backing up through an atomic group */
988    
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            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            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            happen from now on. Otherwise, THEN is passed back to an outer
996            alternative. This implements Perl's treatment of parenthesized groups,
997            where a group not containing | does not affect the current alternative,
998            that is, (X) is NOT the same as (X|(*F)). */
999    
1000            if (rrc == MATCH_THEN)
1001              {
1002              next = ecode + GET(ecode,1);
1003              if (md->start_match_ptr < next &&
1004                  (*ecode == OP_ALT || *next == OP_ALT))
1005                rrc = MATCH_NOMATCH;
1006              }
1007    
1008            /* Anything other than NOMATCH is passed back. */
1009    
1010          if (rrc != MATCH_NOMATCH) RRETURN(rrc);          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1011          md->capture_last = save_capture_last;          md->capture_last = save_capture_last;
1012          ecode += GET(ecode, 1);          ecode += GET(ecode, 1);
1013            md->mark = save_mark;
1014            if (*ecode != OP_ALT) break;
1015          }          }
       while (*ecode == OP_ALT);  
1016    
1017        DPRINTF(("bracket %d failed\n", number));        DPRINTF(("bracket %d failed\n", number));
   
1018        md->offset_vector[offset] = save_offset1;        md->offset_vector[offset] = save_offset1;
1019        md->offset_vector[offset+1] = save_offset2;        md->offset_vector[offset+1] = save_offset2;
1020        md->offset_vector[md->offset_end - number] = save_offset3;        md->offset_vector[md->offset_end - number] = save_offset3;
1021    
1022        RRETURN(MATCH_NOMATCH);        /* At this point, rrc will be one of MATCH_ONCE or MATCH_NOMATCH. */
1023    
1024          RRETURN(rrc);
1025        }        }
1026    
1027      /* Insufficient room for saving captured contents. Treat as a non-capturing      /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
1028      bracket. */      as a non-capturing bracket. */
1029    
1030        /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1031        /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1032    
1033      DPRINTF(("insufficient capture room: treat as non-capturing\n"));      DPRINTF(("insufficient capture room: treat as non-capturing\n"));
1034    
1035      /* Non-capturing bracket. Loop for all the alternatives. When we get to the      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1036      final alternative within the brackets, we would return the result of a      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1037      recursive call to match() whatever happened. We can reduce stack usage by  
1038      turning this into a tail recursion. */      /* Non-capturing or atomic group, except for possessive with unlimited
1039        repeat and ONCE group with no captures. Loop for all the alternatives.
1040    
1041        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    
1048        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        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        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    
1058        case OP_ONCE:
1059      case OP_BRA:      case OP_BRA:
1060      case OP_SBRA:      case OP_SBRA:
1061      DPRINTF(("start non-capturing bracket\n"));      DPRINTF(("start non-capturing bracket\n"));
1062      flags = (op >= OP_SBRA)? match_cbegroup : 0;  
1063      for (;;)      for (;;)
1064        {        {
1065        if (ecode[GET(ecode, 1)] != OP_ALT)        if (op >= OP_SBRA || op == OP_ONCE)
1066            md->match_function_type = MATCH_CBEGROUP;
1067    
1068          /* If this is not a possibly empty group, and there are no (*THEN)s in
1069          the pattern, and this is the final alternative, optimize as described
1070          above. */
1071    
1072          else if (!md->hasthen && ecode[GET(ecode, 1)] != OP_ALT)
1073          {          {
1074          ecode += _pcre_OP_lengths[*ecode];          ecode += PRIV(OP_lengths)[*ecode];
         flags |= match_tail_recursed;  
         DPRINTF(("bracket 0 tail recursion\n"));  
1075          goto TAIL_RECURSE;          goto TAIL_RECURSE;
1076          }          }
1077    
1078        /* For non-final alternatives, continue the loop for a NOMATCH result;        /* In all other cases, we have to make another call to match(). */
       otherwise return. */  
1079    
1080        RMATCH(rrc, eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md, ims,        save_mark = md->mark;
1081          eptrb, flags);        save_capture_last = md->capture_last;
1082        if (rrc != MATCH_NOMATCH) RRETURN(rrc);        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md, eptrb,
1083            RM2);
1084    
1085          /* See comment in the code for capturing groups above about handling
1086          THEN. */
1087    
1088          if (rrc == MATCH_THEN)
1089            {
1090            next = ecode + GET(ecode,1);
1091            if (md->start_match_ptr < next &&
1092                (*ecode == OP_ALT || *next == OP_ALT))
1093              rrc = MATCH_NOMATCH;
1094            }
1095    
1096          if (rrc != MATCH_NOMATCH)
1097            {
1098            if (rrc == MATCH_ONCE)
1099              {
1100              const pcre_uchar *scode = ecode;
1101              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                }
1106              if (md->once_target == scode) rrc = MATCH_NOMATCH;
1107              }
1108            RRETURN(rrc);
1109            }
1110        ecode += GET(ecode, 1);        ecode += GET(ecode, 1);
1111        }        md->mark = save_mark;
1112      /* Control never reaches here. */        if (*ecode != OP_ALT) break;
1113          md->capture_last = save_capture_last;
1114          }
1115    
1116        RRETURN(MATCH_NOMATCH);
1117    
1118        /* Handle possessive capturing brackets with an unlimited repeat. We come
1119        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    
1126        case OP_CBRAPOS:
1127        case OP_SCBRAPOS:
1128        allow_zero = FALSE;
1129    
1130      /* Conditional group: compilation checked that there are no more than      POSSESSIVE_CAPTURE:
1131      two branches. If the condition is false, skipping the first branch takes us      number = GET2(ecode, 1+LINK_SIZE);
1132      past the end if there is only one branch, but that's OK because that is      offset = number << 1;
     exactly what going to the ket would do. As there is only one branch to be  
     obeyed, we can use tail recursion to avoid using another stack frame. */  
1133    
1134      case OP_COND:  #ifdef PCRE_DEBUG
1135      case OP_SCOND:      printf("start possessive bracket %d\n", number);
1136      if (ecode[LINK_SIZE+1] == OP_RREF)         /* Recursion test */      printf("subject=");
1137        {      pchars(eptr, 16, TRUE, md);
1138        offset = GET2(ecode, LINK_SIZE + 2);     /* Recursion group number*/      printf("\n");
1139        condition = md->recursive != NULL &&  #endif
         (offset == RREF_ANY || offset == md->recursive->group_num);  
       ecode += condition? 3 : GET(ecode, 1);  
       }  
1140    
1141      else if (ecode[LINK_SIZE+1] == OP_CREF)    /* Group used test */      if (offset < md->offset_max)
1142        {        {
1143        offset = GET2(ecode, LINK_SIZE+2) << 1;  /* Doubled ref number */        matched_once = FALSE;
1144        condition = offset < offset_top && md->offset_vector[offset] >= 0;        code_offset = (int)(ecode - md->start_code);
       ecode += condition? 3 : GET(ecode, 1);  
       }  
1145    
1146      else if (ecode[LINK_SIZE+1] == OP_DEF)     /* DEFINE - always false */        save_offset1 = md->offset_vector[offset];
1147        {        save_offset2 = md->offset_vector[offset+1];
1148        condition = FALSE;        save_offset3 = md->offset_vector[md->offset_end - number];
1149        ecode += GET(ecode, 1);        save_capture_last = md->capture_last;
       }  
1150    
1151      /* The condition is an assertion. Call match() to evaluate it - setting        DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
     the final argument match_condassert causes it to stop at the end of an  
     assertion. */  
1152    
1153      else        /* 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        RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,        restart it with a new subject starting position, remembering that we had
1156            match_condassert);        at least one match. For MATCH_NOMATCH, carry on with the alternatives, as
1157        if (rrc == MATCH_MATCH)        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          condition = TRUE;        continue from afterwards. Otherwise it has failed; restore the previous
1160          ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE + 2);        capture values before returning NOMATCH. */
1161          while (*ecode == OP_ALT) ecode += GET(ecode, 1);  
1162          for (;;)
1163            {
1164            md->offset_vector[md->offset_end - number] =
1165              (int)(eptr - md->start_subject);
1166            if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
1167            RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
1168              eptrb, RM63);
1169            if (rrc == MATCH_KETRPOS)
1170              {
1171              offset_top = md->end_offset_top;
1172              eptr = md->end_match_ptr;
1173              ecode = md->start_code + code_offset;
1174              save_capture_last = md->capture_last;
1175              matched_once = TRUE;
1176              continue;
1177              }
1178    
1179            /* 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              if (md->start_match_ptr < next &&
1186                  (*ecode == OP_ALT || *next == OP_ALT))
1187                rrc = MATCH_NOMATCH;
1188              }
1189    
1190            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1191            md->capture_last = save_capture_last;
1192            ecode += GET(ecode, 1);
1193            if (*ecode != OP_ALT) break;
1194          }          }
1195        else if (rrc != MATCH_NOMATCH)  
1196          if (!matched_once)
1197          {          {
1198          RRETURN(rrc);         /* Need braces because of following else */          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        else  
1203          if (allow_zero || matched_once)
1204          {          {
1205          condition = FALSE;          ecode += 1 + LINK_SIZE;
1206          ecode += GET(ecode, 1);          break;
1207          }          }
       }  
1208    
1209      /* We are now at the branch that is to be obeyed. As there is only one,        RRETURN(MATCH_NOMATCH);
     we can use tail recursion to avoid using another stack frame. If the second  
     alternative doesn't exist, we can just plough on. */  
   
     if (condition || *ecode == OP_ALT)  
       {  
       ecode += 1 + LINK_SIZE;  
       flags = match_tail_recursed | ((op == OP_SCOND)? match_cbegroup : 0);  
       goto TAIL_RECURSE;  
       }  
     else  
       {  
       ecode += 1 + LINK_SIZE;  
1210        }        }
     break;  
   
1211    
1212      /* End of the pattern. If we are in a top-level recursion, we should      /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
1213      restore the offsets appropriately and continue from after the call. */      as a non-capturing bracket. */
1214    
1215      case OP_END:      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1216      if (md->recursive != NULL && md->recursive->group_num == 0)      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
       {  
       recursion_info *rec = md->recursive;  
       DPRINTF(("End of pattern in a (?0) recursion\n"));  
       md->recursive = rec->prevrec;  
       memmove(md->offset_vector, rec->offset_save,  
         rec->saved_max * sizeof(int));  
       md->start_match = rec->save_start;  
       ims = original_ims;  
       ecode = rec->after_call;  
       break;  
       }  
1217    
1218      /* Otherwise, if PCRE_NOTEMPTY is set, fail if we have matched an empty      DPRINTF(("insufficient capture room: treat as non-capturing\n"));
     string - backtracking will then try other alternatives, if any. */  
1219    
1220      if (md->notempty && eptr == md->start_match) RRETURN(MATCH_NOMATCH);      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1221      md->end_match_ptr = eptr;          /* Record where we ended */      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
     md->end_offset_top = offset_top;   /* and how many extracts were taken */  
     RRETURN(MATCH_MATCH);  
   
     /* Change option settings */  
   
     case OP_OPT:  
     ims = ecode[1];  
     ecode += 2;  
     DPRINTF(("ims set to %02lx\n", ims));  
     break;  
1222    
1223      /* Assertion brackets. Check the alternative branches in turn - the      /* Non-capturing possessive bracket with unlimited repeat. We come here
1224      matching won't pass the KET for an assertion. If any one branch matches,      from BRAZERO with allow_zero = TRUE. The code is similar to the above,
1225      the assertion is true. Lookbehind assertions have an OP_REVERSE item at the      without the capturing complication. It is written out separately for speed
1226      start of each branch to move the current point backwards, so the code at      and cleanliness. */
1227      this level is identical to the lookahead case. */  
1228        case OP_BRAPOS:
1229        case OP_SBRAPOS:
1230        allow_zero = FALSE;
1231    
1232        POSSESSIVE_NON_CAPTURE:
1233        matched_once = FALSE;
1234        code_offset = (int)(ecode - md->start_code);
1235        save_capture_last = md->capture_last;
1236    
1237      case OP_ASSERT:      for (;;)
     case OP_ASSERTBACK:  
     do  
1238        {        {
1239        RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL, 0);        if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
1240        if (rrc == MATCH_MATCH) break;        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
1241            eptrb, RM48);
1242          if (rrc == MATCH_KETRPOS)
1243            {
1244            offset_top = md->end_offset_top;
1245            eptr = md->end_match_ptr;
1246            ecode = md->start_code + code_offset;
1247            matched_once = TRUE;
1248            continue;
1249            }
1250    
1251          /* 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            if (md->start_match_ptr < next &&
1258                (*ecode == OP_ALT || *next == OP_ALT))
1259              rrc = MATCH_NOMATCH;
1260            }
1261    
1262        if (rrc != MATCH_NOMATCH) RRETURN(rrc);        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1263        ecode += GET(ecode, 1);        ecode += GET(ecode, 1);
1264          if (*ecode != OP_ALT) break;
1265          md->capture_last = save_capture_last;
1266        }        }
     while (*ecode == OP_ALT);  
     if (*ecode == OP_KET) RRETURN(MATCH_NOMATCH);  
1267    
1268      /* If checking an assertion for a condition, return MATCH_MATCH. */      if (matched_once || allow_zero)
1269          {
1270          ecode += 1 + LINK_SIZE;
1271          break;
1272          }
1273        RRETURN(MATCH_NOMATCH);
1274    
1275      if ((flags & match_condassert) != 0) RRETURN(MATCH_MATCH);      /* Control never reaches here. */
1276    
1277      /* Continue from after the assertion, updating the offsets high water      /* Conditional group: compilation checked that there are no more than
1278      mark, since extracts may have been taken during the assertion. */      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        exactly what going to the ket would do. */
1281    
1282      do ecode += GET(ecode,1); while (*ecode == OP_ALT);      case OP_COND:
1283      ecode += 1 + LINK_SIZE;      case OP_SCOND:
1284      offset_top = md->end_offset_top;      codelink = GET(ecode, 1);
     continue;  
1285    
1286      /* Negative assertion: all branches must fail to match */      /* Because of the way auto-callout works during compile, a callout item is
1287        inserted between OP_COND and an assertion condition. */
1288    
1289      case OP_ASSERT_NOT:      if (ecode[LINK_SIZE+1] == OP_CALLOUT)
     case OP_ASSERTBACK_NOT:  
     do  
1290        {        {
1291        RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL, 0);        if (PUBL(callout) != NULL)
1292        if (rrc == MATCH_MATCH) RRETURN(MATCH_NOMATCH);          {
1293        if (rrc != MATCH_NOMATCH) RRETURN(rrc);          PUBL(callout_block) cb;
1294        ecode += GET(ecode,1);          cb.version          = 2;   /* Version 1 of the callout block */
1295            cb.callout_number   = ecode[LINK_SIZE+2];
1296            cb.offset_vector    = md->offset_vector;
1297    #if defined COMPILE_PCRE8
1298            cb.subject          = (PCRE_SPTR)md->start_subject;
1299    #elif defined COMPILE_PCRE16
1300            cb.subject          = (PCRE_SPTR16)md->start_subject;
1301    #elif defined COMPILE_PCRE32
1302            cb.subject          = (PCRE_SPTR32)md->start_subject;
1303    #endif
1304            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            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            cb.capture_last     = md->capture_last & CAPLMASK;
1311            /* Internal change requires this for API compatibility. */
1312            if (cb.capture_last == 0) cb.capture_last = -1;
1313            cb.callout_data     = md->callout_data;
1314            cb.mark             = md->nomatch_mark;
1315            if ((rrc = (*PUBL(callout))(&cb)) > 0) RRETURN(MATCH_NOMATCH);
1316            if (rrc < 0) RRETURN(rrc);
1317            }
1318          ecode += PRIV(OP_lengths)[OP_CALLOUT];
1319          codelink -= PRIV(OP_lengths)[OP_CALLOUT];
1320          }
1321    
1322        condcode = ecode[LINK_SIZE+1];
1323    
1324        /* Now see what the actual condition is */
1325    
1326        if (condcode == OP_RREF || condcode == OP_NRREF)    /* Recursion test */
1327          {
1328          if (md->recursive == NULL)                /* Not recursing => FALSE */
1329            {
1330            condition = FALSE;
1331            ecode += GET(ecode, 1);
1332            }
1333          else
1334            {
1335            unsigned int recno = GET2(ecode, LINK_SIZE + 2);   /* Recursion group number*/
1336            condition = (recno == RREF_ANY || recno == md->recursive->group_num);
1337    
1338            /* 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    
1343            if (!condition && condcode == OP_NRREF)
1344              {
1345              pcre_uchar *slotA = md->name_table;
1346              for (i = 0; i < md->name_count; i++)
1347                {
1348                if (GET2(slotA, 0) == recno) break;
1349                slotA += md->name_entry_size;
1350                }
1351    
1352              /* 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    
1356              if (i < md->name_count)
1357                {
1358                pcre_uchar *slotB = slotA;
1359                while (slotB > md->name_table)
1360                  {
1361                  slotB -= md->name_entry_size;
1362                  if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
1363                    {
1364                    condition = GET2(slotB, 0) == md->recursive->group_num;
1365                    if (condition) break;
1366                    }
1367                  else break;
1368                  }
1369    
1370                /* Scan up for duplicates */
1371    
1372                if (!condition)
1373                  {
1374                  slotB = slotA;
1375                  for (i++; i < md->name_count; i++)
1376                    {
1377                    slotB += md->name_entry_size;
1378                    if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
1379                      {
1380                      condition = GET2(slotB, 0) == md->recursive->group_num;
1381                      if (condition) break;
1382                      }
1383                    else break;
1384                    }
1385                  }
1386                }
1387              }
1388    
1389            /* Chose branch according to the condition */
1390    
1391            ecode += condition? 1 + IMM2_SIZE : GET(ecode, 1);
1392            }
1393          }
1394    
1395        else if (condcode == OP_CREF || condcode == OP_NCREF)  /* Group used test */
1396          {
1397          offset = GET2(ecode, LINK_SIZE+2) << 1;  /* Doubled ref number */
1398          condition = offset < offset_top && md->offset_vector[offset] >= 0;
1399    
1400          /* If the numbered capture is unset, but the reference was by name,
1401          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          if (!condition && condcode == OP_NCREF)
1406            {
1407            unsigned int refno = offset >> 1;
1408            pcre_uchar *slotA = md->name_table;
1409    
1410            for (i = 0; i < md->name_count; i++)
1411              {
1412              if (GET2(slotA, 0) == refno) break;
1413              slotA += md->name_entry_size;
1414              }
1415    
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            for duplicates. */
1419    
1420            if (i < md->name_count)
1421              {
1422              pcre_uchar *slotB = slotA;
1423              while (slotB > md->name_table)
1424                {
1425                slotB -= md->name_entry_size;
1426                if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
1427                  {
1428                  offset = GET2(slotB, 0) << 1;
1429                  condition = offset < offset_top &&
1430                    md->offset_vector[offset] >= 0;
1431                  if (condition) break;
1432                  }
1433                else break;
1434                }
1435    
1436              /* Scan up for duplicates */
1437    
1438              if (!condition)
1439                {
1440                slotB = slotA;
1441                for (i++; i < md->name_count; i++)
1442                  {
1443                  slotB += md->name_entry_size;
1444                  if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
1445                    {
1446                    offset = GET2(slotB, 0) << 1;
1447                    condition = offset < offset_top &&
1448                      md->offset_vector[offset] >= 0;
1449                    if (condition) break;
1450                    }
1451                  else break;
1452                  }
1453                }
1454              }
1455            }
1456    
1457          /* Chose branch according to the condition */
1458    
1459          ecode += condition? 1 + IMM2_SIZE : GET(ecode, 1);
1460          }
1461    
1462        else if (condcode == OP_DEF)     /* DEFINE - always false */
1463          {
1464          condition = FALSE;
1465          ecode += GET(ecode, 1);
1466          }
1467    
1468        /* The condition is an assertion. Call match() to evaluate it - setting
1469        md->match_function_type to MATCH_CONDASSERT causes it to stop at the end of
1470        an assertion. */
1471    
1472        else
1473          {
1474          md->match_function_type = MATCH_CONDASSERT;
1475          RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM3);
1476          if (rrc == MATCH_MATCH)
1477            {
1478            if (md->end_offset_top > offset_top)
1479              offset_top = md->end_offset_top;  /* Captures may have happened */
1480            condition = TRUE;
1481            ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE + 2);
1482            while (*ecode == OP_ALT) ecode += GET(ecode, 1);
1483            }
1484    
1485          /* PCRE doesn't allow the effect of (*THEN) to escape beyond an
1486          assertion; it is therefore treated as NOMATCH. */
1487    
1488          else if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN)
1489            {
1490            RRETURN(rrc);         /* Need braces because of following else */
1491            }
1492          else
1493            {
1494            condition = FALSE;
1495            ecode += codelink;
1496            }
1497          }
1498    
1499        /* 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    
1509        if (condition || *ecode == OP_ALT)
1510          {
1511          if (op != OP_SCOND)
1512            {
1513            ecode += 1 + LINK_SIZE;
1514            goto TAIL_RECURSE;
1515            }
1516    
1517          md->match_function_type = MATCH_CBEGROUP;
1518          RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM49);
1519          RRETURN(rrc);
1520          }
1521    
1522         /* Condition false & no alternative; continue after the group. */
1523    
1524        else
1525          {
1526          ecode += 1 + LINK_SIZE;
1527          }
1528        break;
1529    
1530    
1531        /* Before OP_ACCEPT there may be any number of OP_CLOSE opcodes,
1532        to close any currently open capturing brackets. */
1533    
1534        case OP_CLOSE:
1535        number = GET2(ecode, 1);   /* Must be less than 65536 */
1536        offset = number << 1;
1537    
1538    #ifdef PCRE_DEBUG
1539          printf("end bracket %d at *ACCEPT", number);
1540          printf("\n");
1541    #endif
1542    
1543        md->capture_last = (md->capture_last & OVFLMASK) | number;
1544        if (offset >= md->offset_max) md->capture_last |= OVFLBIT; else
1545          {
1546          md->offset_vector[offset] =
1547            md->offset_vector[md->offset_end - number];
1548          md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
1549          if (offset_top <= offset) offset_top = offset + 2;
1550          }
1551        ecode += 1 + IMM2_SIZE;
1552        break;
1553    
1554    
1555        /* End of the pattern, either real or forced. */
1556    
1557        case OP_END:
1558        case OP_ACCEPT:
1559        case OP_ASSERT_ACCEPT:
1560    
1561        /* 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        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    
1566        if (eptr == mstart && op != OP_ASSERT_ACCEPT &&
1567             md->recursive == NULL &&
1568             (md->notempty ||
1569               (md->notempty_atstart &&
1570                 mstart == md->start_subject + md->start_offset)))
1571          RRETURN(MATCH_NOMATCH);
1572    
1573        /* Otherwise, we have a match. */
1574    
1575        md->end_match_ptr = eptr;           /* Record where we ended */
1576        md->end_offset_top = offset_top;    /* and how many extracts were taken */
1577        md->start_match_ptr = mstart;       /* and the start (\K can modify) */
1578    
1579        /* For some reason, the macros don't work properly if an expression is
1580        given as the argument to RRETURN when the heap is in use. */
1581    
1582        rrc = (op == OP_END)? MATCH_MATCH : MATCH_ACCEPT;
1583        RRETURN(rrc);
1584    
1585        /* 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        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        match() to remember this case. */
1595    
1596        case OP_ASSERT:
1597        case OP_ASSERTBACK:
1598        save_mark = md->mark;
1599        if (md->match_function_type == MATCH_CONDASSERT)
1600          {
1601          condassert = TRUE;
1602          md->match_function_type = 0;
1603          }
1604        else condassert = FALSE;
1605    
1606        /* Loop for each branch */
1607    
1608        do
1609          {
1610          RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM4);
1611          if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT)
1612            {
1613            mstart = md->start_match_ptr;   /* In case \K reset it */
1614            break;
1615            }
1616          md->mark = save_mark;
1617    
1618          /* See comment in the code for capturing groups above about handling
1619          THEN. */
1620    
1621          if (rrc == MATCH_THEN)
1622            {
1623            next = ecode + GET(ecode,1);
1624            if (md->start_match_ptr < next &&
1625                (*ecode == OP_ALT || *next == OP_ALT))
1626              rrc = MATCH_NOMATCH;
1627            }
1628    
1629          /* Anything other than NOMATCH causes the assertion to fail. This
1630          includes COMMIT, SKIP, and PRUNE. However, this consistent approach does
1631          not always have exactly the same effect as in Perl. */
1632    
1633          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1634          ecode += GET(ecode, 1);
1635        }        }
1636      while (*ecode == OP_ALT);      while (*ecode == OP_ALT);
1637    
1638        /* If we have tried all the alternative branches, the assertion has
1639        failed. */
1640    
1641        if (*ecode == OP_KET) RRETURN(MATCH_NOMATCH);
1642    
1643        /* If checking an assertion for a condition, return MATCH_MATCH. */
1644    
1645        if (condassert) RRETURN(MATCH_MATCH);
1646    
1647      if ((flags & match_condassert) != 0) RRETURN(MATCH_MATCH);      /* Continue from after a successful assertion, updating the offsets high
1648        water mark, since extracts may have been taken during the assertion. */
1649    
1650        do ecode += GET(ecode,1); while (*ecode == OP_ALT);
1651      ecode += 1 + LINK_SIZE;      ecode += 1 + LINK_SIZE;
1652        offset_top = md->end_offset_top;
1653        continue;
1654    
1655        /* Negative assertion: all branches must fail to match for the assertion to
1656        succeed. */
1657    
1658        case OP_ASSERT_NOT:
1659        case OP_ASSERTBACK_NOT:
1660        save_mark = md->mark;
1661        if (md->match_function_type == MATCH_CONDASSERT)
1662          {
1663          condassert = TRUE;
1664          md->match_function_type = 0;
1665          }
1666        else condassert = FALSE;
1667    
1668        /* Loop for each alternative branch. */
1669    
1670        do
1671          {
1672          RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM5);
1673          md->mark = save_mark;
1674    
1675          /* A successful match means the assertion has failed. */
1676    
1677          if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) RRETURN(MATCH_NOMATCH);
1678    
1679          /* See comment in the code for capturing groups above about handling
1680          THEN. */
1681    
1682          if (rrc == MATCH_THEN)
1683            {
1684            next = ecode + GET(ecode,1);
1685            if (md->start_match_ptr < next &&
1686                (*ecode == OP_ALT || *next == OP_ALT))
1687              rrc = MATCH_NOMATCH;
1688            }
1689    
1690          /* No match on a branch means we must carry on and try the next branch.
1691          Anything else, in particular, SKIP, PRUNE, etc. causes a failure in the
1692          enclosing branch. This is a consistent approach, but does not always have
1693          the same effect as in Perl. */
1694    
1695          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1696          ecode += GET(ecode,1);
1697          }
1698        while (*ecode == OP_ALT);
1699    
1700        /* All branches in the assertion failed to match. */
1701    
1702        if (condassert) RRETURN(MATCH_MATCH);  /* Condition assertion */
1703        ecode += 1 + LINK_SIZE;                /* Continue with current branch */
1704      continue;      continue;
1705    
1706      /* Move the subject pointer back. This occurs only at the start of      /* Move the subject pointer back. This occurs only at the start of
# Line 846  for (;;) Line 1709  for (;;)
1709      back a number of characters, not bytes. */      back a number of characters, not bytes. */
1710    
1711      case OP_REVERSE:      case OP_REVERSE:
1712  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
1713      if (utf8)      if (utf)
1714        {        {
1715        i = GET(ecode, 1);        i = GET(ecode, 1);
1716        while (i-- > 0)        while (i-- > 0)
1717          {          {
1718          eptr--;          eptr--;
1719          if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);          if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
1720          BACKCHAR(eptr)          BACKCHAR(eptr);
1721          }          }
1722        }        }
1723      else      else
# Line 867  for (;;) Line 1730  for (;;)
1730        if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);        if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
1731        }        }
1732    
1733      /* Skip to next op code */      /* Save the earliest consulted character, then skip to next op code */
1734    
1735        if (eptr < md->start_used_ptr) md->start_used_ptr = eptr;
1736      ecode += 1 + LINK_SIZE;      ecode += 1 + LINK_SIZE;
1737      break;      break;
1738    
# Line 877  for (;;) Line 1741  for (;;)
1741      function is able to force a failure. */      function is able to force a failure. */
1742    
1743      case OP_CALLOUT:      case OP_CALLOUT:
1744      if (pcre_callout != NULL)      if (PUBL(callout) != NULL)
1745        {        {
1746        pcre_callout_block cb;        PUBL(callout_block) cb;
1747        cb.version          = 1;   /* Version 1 of the callout block */        cb.version          = 2;   /* Version 1 of the callout block */
1748        cb.callout_number   = ecode[1];        cb.callout_number   = ecode[1];
1749        cb.offset_vector    = md->offset_vector;        cb.offset_vector    = md->offset_vector;
1750    #if defined COMPILE_PCRE8
1751        cb.subject          = (PCRE_SPTR)md->start_subject;        cb.subject          = (PCRE_SPTR)md->start_subject;
1752        cb.subject_length   = md->end_subject - md->start_subject;  #elif defined COMPILE_PCRE16
1753        cb.start_match      = md->start_match - md->start_subject;        cb.subject          = (PCRE_SPTR16)md->start_subject;
1754        cb.current_position = eptr - md->start_subject;  #elif defined COMPILE_PCRE32
1755          cb.subject          = (PCRE_SPTR32)md->start_subject;
1756    #endif
1757          cb.subject_length   = (int)(md->end_subject - md->start_subject);
1758          cb.start_match      = (int)(mstart - md->start_subject);
1759          cb.current_position = (int)(eptr - md->start_subject);
1760        cb.pattern_position = GET(ecode, 2);        cb.pattern_position = GET(ecode, 2);
1761        cb.next_item_length = GET(ecode, 2 + LINK_SIZE);        cb.next_item_length = GET(ecode, 2 + LINK_SIZE);
1762        cb.capture_top      = offset_top/2;        cb.capture_top      = offset_top/2;
1763        cb.capture_last     = md->capture_last;        cb.capture_last     = md->capture_last & CAPLMASK;
1764          /* Internal change requires this for API compatibility. */
1765          if (cb.capture_last == 0) cb.capture_last = -1;
1766        cb.callout_data     = md->callout_data;        cb.callout_data     = md->callout_data;
1767        if ((rrc = (*pcre_callout)(&cb)) > 0) RRETURN(MATCH_NOMATCH);        cb.mark             = md->nomatch_mark;
1768          if ((rrc = (*PUBL(callout))(&cb)) > 0) RRETURN(MATCH_NOMATCH);
1769        if (rrc < 0) RRETURN(rrc);        if (rrc < 0) RRETURN(rrc);
1770        }        }
1771      ecode += 2 + 2*LINK_SIZE;      ecode += 2 + 2*LINK_SIZE;
# Line 902  for (;;) Line 1775  for (;;)
1775      offset data is the offset to the starting bracket from the start of the      offset data is the offset to the starting bracket from the start of the
1776      whole pattern. (This is so that it works from duplicated subpatterns.)      whole pattern. (This is so that it works from duplicated subpatterns.)
1777    
1778      If there are any capturing brackets started but not finished, we have to      The state of the capturing groups is preserved over recursion, and
1779      save their starting points and reinstate them after the recursion. However,      re-instated afterwards. We don't know how many are started and not yet
1780      we don't know how many such there are (offset_top records the completed      finished (offset_top records the completed total) so we just have to save
1781      total) so we just have to save all the potential data. There may be up to      all the potential data. There may be up to 65535 such values, which is too
1782      65535 such values, which is too large to put on the stack, but using malloc      large to put on the stack, but using malloc for small numbers seems
1783      for small numbers seems expensive. As a compromise, the stack is used when      expensive. As a compromise, the stack is used when there are no more than
1784      there are no more than REC_STACK_SAVE_MAX values to store; otherwise malloc      REC_STACK_SAVE_MAX values to store; otherwise malloc is used.
     is used. A problem is what to do if the malloc fails ... there is no way of  
     returning to the top level with an error. Save the top REC_STACK_SAVE_MAX  
     values on the stack, and accept that the rest may be wrong.  
1785    
1786      There are also other values that have to be saved. We use a chained      There are also other values that have to be saved. We use a chained
1787      sequence of blocks that actually live on the stack. Thanks to Robin Houston      sequence of blocks that actually live on the stack. Thanks to Robin Houston
1788      for the original version of this logic. */      for the original version of this logic. It has, however, been hacked around
1789        a lot, so he is not to blame for the current way it works. */
1790    
1791      case OP_RECURSE:      case OP_RECURSE:
1792        {        {
1793          recursion_info *ri;
1794          unsigned int recno;
1795    
1796        callpat = md->start_code + GET(ecode, 1);        callpat = md->start_code + GET(ecode, 1);
1797        new_recursive.group_num = (callpat == md->start_code)? 0 :        recno = (callpat == md->start_code)? 0 :
1798          GET2(callpat, 1 + LINK_SIZE);          GET2(callpat, 1 + LINK_SIZE);
1799    
1800          /* Check for repeating a recursion without advancing the subject pointer.
1801          This should catch convoluted mutual recursions. (Some simple cases are
1802          caught at compile time.) */
1803    
1804          for (ri = md->recursive; ri != NULL; ri = ri->prevrec)
1805            if (recno == ri->group_num && eptr == ri->subject_position)
1806              RRETURN(PCRE_ERROR_RECURSELOOP);
1807    
1808        /* Add to "recursing stack" */        /* Add to "recursing stack" */
1809    
1810          new_recursive.group_num = recno;
1811          new_recursive.saved_capture_last = md->capture_last;
1812          new_recursive.subject_position = eptr;
1813        new_recursive.prevrec = md->recursive;        new_recursive.prevrec = md->recursive;
1814        md->recursive = &new_recursive;        md->recursive = &new_recursive;
1815    
1816        /* Find where to continue from afterwards */        /* Where to continue from afterwards */
1817    
1818        ecode += 1 + LINK_SIZE;        ecode += 1 + LINK_SIZE;
       new_recursive.after_call = ecode;  
1819    
1820        /* Now save the offset data. */        /* Now save the offset data */
1821    
1822        new_recursive.saved_max = md->offset_end;        new_recursive.saved_max = md->offset_end;
1823        if (new_recursive.saved_max <= REC_STACK_SAVE_MAX)        if (new_recursive.saved_max <= REC_STACK_SAVE_MAX)
# Line 941  for (;;) Line 1825  for (;;)
1825        else        else
1826          {          {
1827          new_recursive.offset_save =          new_recursive.offset_save =
1828            (int *)(pcre_malloc)(new_recursive.saved_max * sizeof(int));            (int *)(PUBL(malloc))(new_recursive.saved_max * sizeof(int));
1829          if (new_recursive.offset_save == NULL) RRETURN(PCRE_ERROR_NOMEMORY);          if (new_recursive.offset_save == NULL) RRETURN(PCRE_ERROR_NOMEMORY);
1830          }          }
   
1831        memcpy(new_recursive.offset_save, md->offset_vector,        memcpy(new_recursive.offset_save, md->offset_vector,
1832              new_recursive.saved_max * sizeof(int));              new_recursive.saved_max * sizeof(int));
       new_recursive.save_start = md->start_match;  
       md->start_match = eptr;  
1833    
1834        /* OK, now we can do the recursion. For each top-level alternative we        /* OK, now we can do the recursion. After processing each alternative,
1835        restore the offset and recursion data. */        restore the offset data and the last captured value. If there were nested
1836          recursions, md->recursive might be changed, so reset it before looping.
1837          */
1838    
1839        DPRINTF(("Recursing into group %d\n", new_recursive.group_num));        DPRINTF(("Recursing into group %d\n", new_recursive.group_num));
1840        flags = (*callpat >= OP_SBRA)? match_cbegroup : 0;        cbegroup = (*callpat >= OP_SBRA);
1841        do        do
1842          {          {
1843          RMATCH(rrc, eptr, callpat + _pcre_OP_lengths[*callpat], offset_top,          if (cbegroup) md->match_function_type = MATCH_CBEGROUP;
1844            md, ims, eptrb, flags);          RMATCH(eptr, callpat + PRIV(OP_lengths)[*callpat], offset_top,
1845          if (rrc == MATCH_MATCH)            md, eptrb, RM6);
1846            memcpy(md->offset_vector, new_recursive.offset_save,
1847                new_recursive.saved_max * sizeof(int));
1848            md->capture_last = new_recursive.saved_capture_last;
1849            md->recursive = new_recursive.prevrec;
1850            if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT)
1851            {            {
1852            DPRINTF(("Recursion matched\n"));            DPRINTF(("Recursion matched\n"));
           md->recursive = new_recursive.prevrec;  
1853            if (new_recursive.offset_save != stacksave)            if (new_recursive.offset_save != stacksave)
1854              (pcre_free)(new_recursive.offset_save);              (PUBL(free))(new_recursive.offset_save);
1855            RRETURN(MATCH_MATCH);  
1856            }            /* Set where we got to in the subject, and reset the start in case
1857          else if (rrc != MATCH_NOMATCH)            it was changed by \K. This *is* propagated back out of a recursion,
1858              for Perl compatibility. */
1859    
1860              eptr = md->end_match_ptr;
1861              mstart = md->start_match_ptr;
1862              goto RECURSION_MATCHED;        /* Exit loop; end processing */
1863              }
1864    
1865            /* PCRE does not allow THEN, SKIP, PRUNE or COMMIT to escape beyond a
1866            recursion; they cause a NOMATCH for the entire recursion. These codes
1867            are defined in a range that can be tested for. */
1868    
1869            if (rrc >= MATCH_BACKTRACK_MIN && rrc <= MATCH_BACKTRACK_MAX)
1870              RRETURN(MATCH_NOMATCH);
1871    
1872            /* Any return code other than NOMATCH is an error. */
1873    
1874            if (rrc != MATCH_NOMATCH)
1875            {            {
1876            DPRINTF(("Recursion gave error %d\n", rrc));            DPRINTF(("Recursion gave error %d\n", rrc));
1877              if (new_recursive.offset_save != stacksave)
1878                (PUBL(free))(new_recursive.offset_save);
1879            RRETURN(rrc);            RRETURN(rrc);
1880            }            }
1881    
1882          md->recursive = &new_recursive;          md->recursive = &new_recursive;
         memcpy(md->offset_vector, new_recursive.offset_save,  
             new_recursive.saved_max * sizeof(int));  
1883          callpat += GET(callpat, 1);          callpat += GET(callpat, 1);
1884          }          }
1885        while (*callpat == OP_ALT);        while (*callpat == OP_ALT);
# Line 983  for (;;) Line 1887  for (;;)
1887        DPRINTF(("Recursion didn't match\n"));        DPRINTF(("Recursion didn't match\n"));
1888        md->recursive = new_recursive.prevrec;        md->recursive = new_recursive.prevrec;
1889        if (new_recursive.offset_save != stacksave)        if (new_recursive.offset_save != stacksave)
1890          (pcre_free)(new_recursive.offset_save);          (PUBL(free))(new_recursive.offset_save);
1891        RRETURN(MATCH_NOMATCH);        RRETURN(MATCH_NOMATCH);
1892        }        }
     /* Control never reaches here */  
   
     /* "Once" brackets are like assertion brackets except that after a match,  
     the point in the subject string is not moved back. Thus there can never be  
     a move back into the brackets. Friedl calls these "atomic" subpatterns.  
     Check the alternative branches in turn - the matching won't pass the KET  
     for this kind of subpattern. If any one branch matches, we carry on as at  
     the end of a normal bracket, leaving the subject pointer. */  
   
     case OP_ONCE:  
     prev = ecode;  
     saved_eptr = eptr;  
1893    
1894      do      RECURSION_MATCHED:
1895        {      break;
       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims,  
         eptrb, 0);  
       if (rrc == MATCH_MATCH) break;  
       if (rrc != MATCH_NOMATCH) RRETURN(rrc);  
       ecode += GET(ecode,1);  
       }  
     while (*ecode == OP_ALT);  
   
     /* If hit the end of the group (which could be repeated), fail */  
   
     if (*ecode != OP_ONCE && *ecode != OP_ALT) RRETURN(MATCH_NOMATCH);  
   
     /* Continue as from after the assertion, updating the offsets high water  
     mark, since extracts may have been taken. */  
   
     do ecode += GET(ecode, 1); while (*ecode == OP_ALT);  
   
     offset_top = md->end_offset_top;  
     eptr = md->end_match_ptr;  
   
     /* For a non-repeating ket, just continue at this level. This also  
     happens for a repeating ket if no characters were matched in the group.  
     This is the forcible breaking of infinite loops as implemented in Perl  
     5.005. If there is an options reset, it will get obeyed in the normal  
     course of events. */  
   
     if (*ecode == OP_KET || eptr == saved_eptr)  
       {  
       ecode += 1+LINK_SIZE;  
       break;  
       }  
   
     /* The repeating kets try the rest of the pattern or restart from the  
     preceding bracket, in the appropriate order. The second "call" of match()  
     uses tail recursion, to avoid using another stack frame. We need to reset  
     any options that changed within the bracket before re-running it, so  
     check the next opcode. */  
   
     if (ecode[1+LINK_SIZE] == OP_OPT)  
       {  
       ims = (ims & ~PCRE_IMS) | ecode[4];  
       DPRINTF(("ims set to %02lx at group repeat\n", ims));  
       }  
   
     if (*ecode == OP_KETRMIN)  
       {  
       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb, 0);  
       if (rrc != MATCH_NOMATCH) RRETURN(rrc);  
       ecode = prev;  
       flags = match_tail_recursed;  
       goto TAIL_RECURSE;  
       }  
     else  /* OP_KETRMAX */  
       {  
       RMATCH(rrc, eptr, prev, offset_top, md, ims, eptrb, match_cbegroup);  
       if (rrc != MATCH_NOMATCH) RRETURN(rrc);  
       ecode += 1 + LINK_SIZE;  
       flags = match_tail_recursed;  
       goto TAIL_RECURSE;  
       }  
     /* Control never gets here */  
1896    
1897      /* An alternation is the end of a branch; scan along to find the end of the      /* An alternation is the end of a branch; scan along to find the end of the
1898      bracketed group and go to there. */      bracketed group and go to there. */
# Line 1070  for (;;) Line 1901  for (;;)
1901      do ecode += GET(ecode,1); while (*ecode == OP_ALT);      do ecode += GET(ecode,1); while (*ecode == OP_ALT);
1902      break;      break;
1903    
1904      /* BRAZERO and BRAMINZERO occur just before a bracket group, indicating      /* BRAZERO, BRAMINZERO and SKIPZERO occur just before a bracket group,
1905      that it may occur zero times. It may repeat infinitely, or not at all -      indicating that it may occur zero times. It may repeat infinitely, or not
1906      i.e. it could be ()* or ()? in the pattern. Brackets with fixed upper      at all - i.e. it could be ()* or ()? or even (){0} in the pattern. Brackets
1907      repeat limits are compiled as a number of copies, with the optional ones      with fixed upper repeat limits are compiled as a number of copies, with the
1908      preceded by BRAZERO or BRAMINZERO. */      optional ones preceded by BRAZERO or BRAMINZERO. */
1909    
1910      case OP_BRAZERO:      case OP_BRAZERO:
1911        {      next = ecode + 1;
1912        next = ecode+1;      RMATCH(eptr, next, offset_top, md, eptrb, RM10);
1913        RMATCH(rrc, eptr, next, offset_top, md, ims, eptrb, 0);      if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1914        if (rrc != MATCH_NOMATCH) RRETURN(rrc);      do next += GET(next, 1); while (*next == OP_ALT);
1915        do next += GET(next,1); while (*next == OP_ALT);      ecode = next + 1 + LINK_SIZE;
       ecode = next + 1 + LINK_SIZE;  
       }  
1916      break;      break;
1917    
1918      case OP_BRAMINZERO:      case OP_BRAMINZERO:
1919        {      next = ecode + 1;
1920        next = ecode+1;      do next += GET(next, 1); while (*next == OP_ALT);
1921        do next += GET(next, 1); while (*next == OP_ALT);      RMATCH(eptr, next + 1+LINK_SIZE, offset_top, md, eptrb, RM11);
1922        RMATCH(rrc, eptr, next + 1+LINK_SIZE, offset_top, md, ims, eptrb, 0);      if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1923        if (rrc != MATCH_NOMATCH) RRETURN(rrc);      ecode++;
1924        ecode++;      break;
1925        }  
1926        case OP_SKIPZERO:
1927        next = ecode+1;
1928        do next += GET(next,1); while (*next == OP_ALT);
1929        ecode = next + 1 + LINK_SIZE;
1930      break;      break;
1931    
1932        /* BRAPOSZERO occurs before a possessive bracket group. Don't do anything
1933        here; just jump to the group, with allow_zero set TRUE. */
1934    
1935        case OP_BRAPOSZERO:
1936        op = *(++ecode);
1937        allow_zero = TRUE;
1938        if (op == OP_CBRAPOS || op == OP_SCBRAPOS) goto POSSESSIVE_CAPTURE;
1939          goto POSSESSIVE_NON_CAPTURE;
1940    
1941      /* End of a group, repeated or non-repeating. */      /* End of a group, repeated or non-repeating. */
1942    
1943      case OP_KET:      case OP_KET:
1944      case OP_KETRMIN:      case OP_KETRMIN:
1945      case OP_KETRMAX:      case OP_KETRMAX:
1946        case OP_KETRPOS:
1947      prev = ecode - GET(ecode, 1);      prev = ecode - GET(ecode, 1);
1948    
1949      /* If this was a group that remembered the subject start, in order to break      /* If this was a group that remembered the subject start, in order to break
1950      infinite repeats of empty string matches, retrieve the subject start from      infinite repeats of empty string matches, retrieve the subject start from
1951      the chain. Otherwise, set it NULL. */      the chain. Otherwise, set it NULL. */
1952    
1953      if (*prev >= OP_SBRA)      if (*prev >= OP_SBRA || *prev == OP_ONCE)
1954        {        {
1955        saved_eptr = eptrb->epb_saved_eptr;   /* Value at start of group */        saved_eptr = eptrb->epb_saved_eptr;   /* Value at start of group */
1956        eptrb = eptrb->epb_prev;              /* Backup to previous group */        eptrb = eptrb->epb_prev;              /* Backup to previous group */
1957        }        }
1958      else saved_eptr = NULL;      else saved_eptr = NULL;
1959    
1960      /* If we are at the end of an assertion group, stop matching and return      /* If we are at the end of an assertion group or a non-capturing atomic
1961      MATCH_MATCH, but record the current high water mark for use by positive      group, stop matching and return MATCH_MATCH, but record the current high
1962      assertions. Do this also for the "once" (atomic) groups. */      water mark for use by positive assertions. We also need to record the match
1963        start in case it was changed by \K. */
1964      if (*prev == OP_ASSERT || *prev == OP_ASSERT_NOT ||  
1965          *prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT ||      if ((*prev >= OP_ASSERT && *prev <= OP_ASSERTBACK_NOT) ||
1966          *prev == OP_ONCE)           *prev == OP_ONCE_NC)
1967        {        {
1968        md->end_match_ptr = eptr;      /* For ONCE */        md->end_match_ptr = eptr;      /* For ONCE_NC */
1969        md->end_offset_top = offset_top;        md->end_offset_top = offset_top;
1970        RRETURN(MATCH_MATCH);        md->start_match_ptr = mstart;
1971          RRETURN(MATCH_MATCH);         /* Sets md->mark */
1972        }        }
1973    
1974      /* For capturing groups we have to check the group number back at the start      /* For capturing groups we have to check the group number back at the start
1975      and if necessary complete handling an extraction by setting the offsets and      and if necessary complete handling an extraction by setting the offsets and
1976      bumping the high water mark. Note that whole-pattern recursion is coded as      bumping the high water mark. Whole-pattern recursion is coded as a recurse
1977      a recurse into group 0, so it won't be picked up here. Instead, we catch it      into group 0, so it won't be picked up here. Instead, we catch it when the
1978      when the OP_END is reached. Other recursion is handled here. */      OP_END is reached. Other recursion is handled here. We just have to record
1979        the current subject position and start match pointer and give a MATCH
1980        return. */
1981    
1982      if (*prev == OP_CBRA || *prev == OP_SCBRA)      if (*prev == OP_CBRA || *prev == OP_SCBRA ||
1983            *prev == OP_CBRAPOS || *prev == OP_SCBRAPOS)
1984        {        {
1985        number = GET2(prev, 1+LINK_SIZE);        number = GET2(prev, 1+LINK_SIZE);
1986        offset = number << 1;        offset = number << 1;
1987    
1988  #ifdef DEBUG  #ifdef PCRE_DEBUG
1989        printf("end bracket %d", number);        printf("end bracket %d", number);
1990        printf("\n");        printf("\n");
1991  #endif  #endif
1992    
1993        md->capture_last = number;        /* Handle a recursively called group. */
1994        if (offset >= md->offset_max) md->offset_overflow = TRUE; else  
1995          if (md->recursive != NULL && md->recursive->group_num == number)
1996          {          {
1997          md->offset_vector[offset] =          md->end_match_ptr = eptr;
1998            md->offset_vector[md->offset_end - number];          md->start_match_ptr = mstart;
1999          md->offset_vector[offset+1] = eptr - md->start_subject;          RRETURN(MATCH_MATCH);
         if (offset_top <= offset) offset_top = offset + 2;  
2000          }          }
2001    
2002        /* Handle a recursively called group. Restore the offsets        /* Deal with capturing */
       appropriately and continue from after the call. */  
2003    
2004        if (md->recursive != NULL && md->recursive->group_num == number)        md->capture_last = (md->capture_last & OVFLMASK) | number;
2005          if (offset >= md->offset_max) md->capture_last |= OVFLBIT; else
2006          {          {
2007          recursion_info *rec = md->recursive;          /* If offset is greater than offset_top, it means that we are
2008          DPRINTF(("Recursion (%d) succeeded - continuing\n", number));          "skipping" a capturing group, and that group's offsets must be marked
2009          md->recursive = rec->prevrec;          unset. In earlier versions of PCRE, all the offsets were unset at the
2010          md->start_match = rec->save_start;          start of matching, but this doesn't work because atomic groups and
2011          memcpy(md->offset_vector, rec->offset_save,          assertions can cause a value to be set that should later be unset.
2012            rec->saved_max * sizeof(int));          Example: matching /(?>(a))b|(a)c/ against "ac". This sets group 1 as
2013          ecode = rec->after_call;          part of the atomic group, but this is not on the final matching path,
2014          ims = original_ims;          so must be unset when 2 is set. (If there is no group 2, there is no
2015          break;          problem, because offset_top will then be 2, indicating no capture.) */
         }  
       }  
2016    
2017      /* For both capturing and non-capturing groups, reset the value of the ims          if (offset > offset_top)
2018      flags, in case they got changed during the group. */            {
2019              register int *iptr = md->offset_vector + offset_top;
2020              register int *iend = md->offset_vector + offset;
2021              while (iptr < iend) *iptr++ = -1;
2022              }
2023    
2024      ims = original_ims;          /* Now make the extraction */
     DPRINTF(("ims reset to %02lx\n", ims));  
2025    
2026      /* For a non-repeating ket, just continue at this level. This also          md->offset_vector[offset] =
2027      happens for a repeating ket if no characters were matched in the group.            md->offset_vector[md->offset_end - number];
2028      This is the forcible breaking of infinite loops as implemented in Perl          md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
2029      5.005. If there is an options reset, it will get obeyed in the normal          if (offset_top <= offset) offset_top = offset + 2;
2030      course of events. */          }
2031          }
2032    
2033        /* For an ordinary non-repeating ket, just continue at this level. This
2034        also happens for a repeating ket if no characters were matched in the
2035        group. This is the forcible breaking of infinite loops as implemented in
2036        Perl 5.005. For a non-repeating atomic group that includes captures,
2037        establish a backup point by processing the rest of the pattern at a lower
2038        level. If this results in a NOMATCH return, pass MATCH_ONCE back to the
2039        original OP_ONCE level, thereby bypassing intermediate backup points, but
2040        resetting any captures that happened along the way. */
2041    
2042      if (*ecode == OP_KET || eptr == saved_eptr)      if (*ecode == OP_KET || eptr == saved_eptr)
2043        {        {
2044        ecode += 1 + LINK_SIZE;        if (*prev == OP_ONCE)
2045            {
2046            RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM12);
2047            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2048            md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
2049            RRETURN(MATCH_ONCE);
2050            }
2051          ecode += 1 + LINK_SIZE;    /* Carry on at this level */
2052        break;        break;
2053        }        }
2054    
2055      /* The repeating kets try the rest of the pattern or restart from the      /* OP_KETRPOS is a possessive repeating ket. Remember the current position,
2056      preceding bracket, in the appropriate order. In the second case, we can use      and return the MATCH_KETRPOS. This makes it possible to do the repeats one
2057      tail recursion to avoid using another stack frame. */      at a time from the outer level, thus saving stack. */
2058    
2059        if (*ecode == OP_KETRPOS)
2060          {
2061          md->end_match_ptr = eptr;
2062          md->end_offset_top = offset_top;
2063          RRETURN(MATCH_KETRPOS);
2064          }
2065    
2066      flags = (*prev >= OP_SBRA)? match_cbegroup : 0;      /* The normal repeating kets try the rest of the pattern or restart from
2067        the preceding bracket, in the appropriate order. In the second case, we can
2068        use tail recursion to avoid using another stack frame, unless we have an
2069        an atomic group or an unlimited repeat of a group that can match an empty
2070        string. */
2071    
2072      if (*ecode == OP_KETRMIN)      if (*ecode == OP_KETRMIN)
2073        {        {
2074        RMATCH(rrc, eptr, ecode + 1+LINK_SIZE, offset_top, md, ims, eptrb, 0);        RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM7);
2075        if (rrc != MATCH_NOMATCH) RRETURN(rrc);        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2076          if (*prev == OP_ONCE)
2077            {
2078            RMATCH(eptr, prev, offset_top, md, eptrb, RM8);
2079            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2080            md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
2081            RRETURN(MATCH_ONCE);
2082            }
2083          if (*prev >= OP_SBRA)    /* Could match an empty string */
2084            {
2085            RMATCH(eptr, prev, offset_top, md, eptrb, RM50);
2086            RRETURN(rrc);
2087            }
2088        ecode = prev;        ecode = prev;
       flags |= match_tail_recursed;  
2089        goto TAIL_RECURSE;        goto TAIL_RECURSE;
2090        }        }
2091      else  /* OP_KETRMAX */      else  /* OP_KETRMAX */
2092        {        {
2093        RMATCH(rrc, eptr, prev, offset_top, md, ims, eptrb, flags);        RMATCH(eptr, prev, offset_top, md, eptrb, RM13);
2094          if (rrc == MATCH_ONCE && md->once_target == prev) rrc = MATCH_NOMATCH;
2095        if (rrc != MATCH_NOMATCH) RRETURN(rrc);        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2096          if (*prev == OP_ONCE)
2097            {
2098            RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM9);
2099            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2100            md->once_target = prev;
2101            RRETURN(MATCH_ONCE);
2102            }
2103        ecode += 1 + LINK_SIZE;        ecode += 1 + LINK_SIZE;
       flags = match_tail_recursed;  
2104        goto TAIL_RECURSE;        goto TAIL_RECURSE;
2105        }        }
2106      /* Control never gets here */      /* Control never gets here */
2107    
2108      /* Start of subject unless notbol, or after internal newline if multiline */      /* Not multiline mode: start of subject assertion, unless notbol. */
2109    
2110      case OP_CIRC:      case OP_CIRC:
2111      if (md->notbol && eptr == md->start_subject) RRETURN(MATCH_NOMATCH);      if (md->notbol && eptr == md->start_subject) RRETURN(MATCH_NOMATCH);
     if ((ims & PCRE_MULTILINE) != 0)  
       {  
       if (eptr != md->start_subject &&  
           (eptr == md->end_subject || !WAS_NEWLINE(eptr)))  
         RRETURN(MATCH_NOMATCH);  
       ecode++;  
       break;  
       }  
     /* ... else fall through */  
2112    
2113      /* Start of subject assertion */      /* Start of subject assertion */
2114    
# Line 1232  for (;;) Line 2117  for (;;)
2117      ecode++;      ecode++;
2118      break;      break;
2119    
2120        /* Multiline mode: start of subject unless notbol, or after any newline. */
2121    
2122        case OP_CIRCM:
2123        if (md->notbol && eptr == md->start_subject) RRETURN(MATCH_NOMATCH);
2124        if (eptr != md->start_subject &&
2125            (eptr == md->end_subject || !WAS_NEWLINE(eptr)))
2126          RRETURN(MATCH_NOMATCH);
2127        ecode++;
2128        break;
2129    
2130      /* Start of match assertion */      /* Start of match assertion */
2131    
2132      case OP_SOM:      case OP_SOM:
# Line 1239  for (;;) Line 2134  for (;;)
2134      ecode++;      ecode++;
2135      break;      break;
2136    
2137      /* Assert before internal newline if multiline, or before a terminating      /* Reset the start of match point */
     newline unless endonly is set, else end of subject unless noteol is set. */  
2138    
2139      case OP_DOLL:      case OP_SET_SOM:
2140      if ((ims & PCRE_MULTILINE) != 0)      mstart = eptr;
2141        ecode++;
2142        break;
2143    
2144        /* Multiline mode: assert before any newline, or before end of subject
2145        unless noteol is set. */
2146    
2147        case OP_DOLLM:
2148        if (eptr < md->end_subject)
2149        {        {
2150        if (eptr < md->end_subject)        if (!IS_NEWLINE(eptr))
2151          { if (!IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH); }          {
2152        else          if (md->partial != 0 &&
2153          { if (md->noteol) RRETURN(MATCH_NOMATCH); }              eptr + 1 >= md->end_subject &&
2154        ecode++;              NLBLOCK->nltype == NLTYPE_FIXED &&
2155        break;              NLBLOCK->nllen == 2 &&
2156                RAWUCHARTEST(eptr) == NLBLOCK->nl[0])
2157              {
2158              md->hitend = TRUE;
2159              if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2160              }
2161            RRETURN(MATCH_NOMATCH);
2162            }
2163        }        }
2164      else      else
2165        {        {
2166        if (md->noteol) RRETURN(MATCH_NOMATCH);        if (md->noteol) RRETURN(MATCH_NOMATCH);
2167        if (!md->endonly)        SCHECK_PARTIAL();
         {  
         if (eptr != md->end_subject &&  
             (!IS_NEWLINE(eptr) || eptr != md->end_subject - md->nllen))  
           RRETURN(MATCH_NOMATCH);  
         ecode++;  
         break;  
         }  
2168        }        }
2169        ecode++;
2170        break;
2171    
2172        /* Not multiline mode: assert before a terminating newline or before end of
2173        subject unless noteol is set. */
2174    
2175        case OP_DOLL:
2176        if (md->noteol) RRETURN(MATCH_NOMATCH);
2177        if (!md->endonly) goto ASSERT_NL_OR_EOS;
2178    
2179      /* ... else fall through for endonly */      /* ... else fall through for endonly */
2180    
2181      /* End of subject assertion (\z) */      /* End of subject assertion (\z) */
2182    
2183      case OP_EOD:      case OP_EOD:
2184      if (eptr < md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr < md->end_subject) RRETURN(MATCH_NOMATCH);
2185        SCHECK_PARTIAL();
2186      ecode++;      ecode++;
2187      break;      break;
2188    
2189      /* End of subject or ending \n assertion (\Z) */      /* End of subject or ending \n assertion (\Z) */
2190    
2191      case OP_EODN:      case OP_EODN:
2192      if (eptr != md->end_subject &&      ASSERT_NL_OR_EOS:
2193        if (eptr < md->end_subject &&
2194          (!IS_NEWLINE(eptr) || eptr != md->end_subject - md->nllen))          (!IS_NEWLINE(eptr) || eptr != md->end_subject - md->nllen))
2195          {
2196          if (md->partial != 0 &&
2197              eptr + 1 >= md->end_subject &&
2198              NLBLOCK->nltype == NLTYPE_FIXED &&
2199              NLBLOCK->nllen == 2 &&
2200              RAWUCHARTEST(eptr) == NLBLOCK->nl[0])
2201            {
2202            md->hitend = TRUE;
2203            if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2204            }
2205        RRETURN(MATCH_NOMATCH);        RRETURN(MATCH_NOMATCH);
2206          }
2207    
2208        /* Either at end of string or \n before end. */
2209    
2210        SCHECK_PARTIAL();
2211      ecode++;      ecode++;
2212      break;      break;
2213    
# Line 1290  for (;;) Line 2219  for (;;)
2219    
2220        /* Find out if the previous and current characters are "word" characters.        /* Find out if the previous and current characters are "word" characters.
2221        It takes a bit more work in UTF-8 mode. Characters > 255 are assumed to        It takes a bit more work in UTF-8 mode. Characters > 255 are assumed to
2222        be "non-word" characters. */        be "non-word" characters. Remember the earliest consulted character for
2223          partial matching. */
2224    
2225  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
2226        if (utf8)        if (utf)
2227          {          {
2228            /* Get status of previous character */
2229    
2230          if (eptr == md->start_subject) prev_is_word = FALSE; else          if (eptr == md->start_subject) prev_is_word = FALSE; else
2231            {            {
2232            const uschar *lastptr = eptr - 1;            PCRE_PUCHAR lastptr = eptr - 1;
2233            while((*lastptr & 0xc0) == 0x80) lastptr--;            BACKCHAR(lastptr);
2234              if (lastptr < md->start_used_ptr) md->start_used_ptr = lastptr;
2235            GETCHAR(c, lastptr);            GETCHAR(c, lastptr);
2236    #ifdef SUPPORT_UCP
2237              if (md->use_ucp)
2238                {
2239                if (c == '_') prev_is_word = TRUE; else
2240                  {
2241                  int cat = UCD_CATEGORY(c);
2242                  prev_is_word = (cat == ucp_L || cat == ucp_N);
2243                  }
2244                }
2245              else
2246    #endif
2247            prev_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;            prev_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;
2248            }            }
2249          if (eptr >= md->end_subject) cur_is_word = FALSE; else  
2250            /* Get status of next character */
2251    
2252            if (eptr >= md->end_subject)
2253              {
2254              SCHECK_PARTIAL();
2255              cur_is_word = FALSE;
2256              }
2257            else
2258            {            {
2259            GETCHAR(c, eptr);            GETCHAR(c, eptr);
2260    #ifdef SUPPORT_UCP
2261              if (md->use_ucp)
2262                {
2263                if (c == '_') cur_is_word = TRUE; else
2264                  {
2265                  int cat = UCD_CATEGORY(c);
2266                  cur_is_word = (cat == ucp_L || cat == ucp_N);
2267                  }
2268                }
2269              else
2270    #endif
2271            cur_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;            cur_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;
2272            }            }
2273          }          }
2274        else        else
2275  #endif  #endif
2276    
2277        /* More streamlined when not in UTF-8 mode */        /* Not in UTF-8 mode, but we may still have PCRE_UCP set, and for
2278          consistency with the behaviour of \w we do use it in this case. */
2279    
2280          {          {
2281          prev_is_word = (eptr != md->start_subject) &&          /* Get status of previous character */
2282            ((md->ctypes[eptr[-1]] & ctype_word) != 0);  
2283          cur_is_word = (eptr < md->end_subject) &&          if (eptr == md->start_subject) prev_is_word = FALSE; else
2284            ((md->ctypes[*eptr] & ctype_word) != 0);            {
2285              if (eptr <= md->start_used_ptr) md->start_used_ptr = eptr - 1;
2286    #ifdef SUPPORT_UCP
2287              if (md->use_ucp)
2288                {
2289                c = eptr[-1];
2290                if (c == '_') prev_is_word = TRUE; else
2291                  {
2292                  int cat = UCD_CATEGORY(c);
2293                  prev_is_word = (cat == ucp_L || cat == ucp_N);
2294                  }
2295                }
2296              else
2297    #endif
2298              prev_is_word = MAX_255(eptr[-1])
2299                && ((md->ctypes[eptr[-1]] & ctype_word) != 0);
2300              }
2301    
2302            /* Get status of next character */
2303    
2304            if (eptr >= md->end_subject)
2305              {
2306              SCHECK_PARTIAL();
2307              cur_is_word = FALSE;
2308              }
2309            else
2310    #ifdef SUPPORT_UCP
2311            if (md->use_ucp)
2312              {
2313              c = *eptr;
2314              if (c == '_') cur_is_word = TRUE; else
2315                {
2316                int cat = UCD_CATEGORY(c);
2317                cur_is_word = (cat == ucp_L || cat == ucp_N);
2318                }
2319              }
2320            else
2321    #endif
2322            cur_is_word = MAX_255(*eptr)
2323              && ((md->ctypes[*eptr] & ctype_word) != 0);
2324          }          }
2325    
2326        /* Now see if the situation is what we want */        /* Now see if the situation is what we want */
# Line 1328  for (;;) Line 2331  for (;;)
2331        }        }
2332      break;      break;
2333    
2334      /* Match a single character type; inline for speed */      /* Match any single character type except newline; have to take care with
2335        CRLF newlines and partial matching. */
2336    
2337      case OP_ANY:      case OP_ANY:
2338      if ((ims & PCRE_DOTALL) == 0)      if (IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH);
2339        if (md->partial != 0 &&
2340            eptr + 1 >= md->end_subject &&
2341            NLBLOCK->nltype == NLTYPE_FIXED &&
2342            NLBLOCK->nllen == 2 &&
2343            RAWUCHARTEST(eptr) == NLBLOCK->nl[0])
2344        {        {
2345        if (IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH);        md->hitend = TRUE;
2346          if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2347        }        }
2348      if (eptr++ >= md->end_subject) RRETURN(MATCH_NOMATCH);  
2349      if (utf8)      /* Fall through */
2350        while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;  
2351        /* Match any single character whatsoever. */
2352    
2353        case OP_ALLANY:
2354        if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
2355          {                            /* not be updated before SCHECK_PARTIAL. */
2356          SCHECK_PARTIAL();
2357          RRETURN(MATCH_NOMATCH);
2358          }
2359        eptr++;
2360    #ifdef SUPPORT_UTF
2361        if (utf) ACROSSCHAR(eptr < md->end_subject, *eptr, eptr++);
2362    #endif
2363      ecode++;      ecode++;
2364      break;      break;
2365    
# Line 1345  for (;;) Line 2367  for (;;)
2367      any byte, even newline, independent of the setting of PCRE_DOTALL. */      any byte, even newline, independent of the setting of PCRE_DOTALL. */
2368    
2369      case OP_ANYBYTE:      case OP_ANYBYTE:
2370      if (eptr++ >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
2371          {                            /* not be updated before SCHECK_PARTIAL. */
2372          SCHECK_PARTIAL();
2373          RRETURN(MATCH_NOMATCH);
2374          }
2375        eptr++;
2376      ecode++;      ecode++;
2377      break;      break;
2378    
2379      case OP_NOT_DIGIT:      case OP_NOT_DIGIT:
2380      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2381          {
2382          SCHECK_PARTIAL();
2383          RRETURN(MATCH_NOMATCH);
2384          }
2385      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2386      if (      if (
2387  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2388         c < 256 &&         c < 256 &&
2389  #endif  #endif
2390         (md->ctypes[c] & ctype_digit) != 0         (md->ctypes[c] & ctype_digit) != 0
# Line 1363  for (;;) Line 2394  for (;;)
2394      break;      break;
2395    
2396      case OP_DIGIT:      case OP_DIGIT:
2397      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2398          {
2399          SCHECK_PARTIAL();
2400          RRETURN(MATCH_NOMATCH);
2401          }
2402      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2403      if (      if (
2404  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2405         c >= 256 ||         c > 255 ||
2406  #endif  #endif
2407         (md->ctypes[c] & ctype_digit) == 0         (md->ctypes[c] & ctype_digit) == 0
2408         )         )
# Line 1376  for (;;) Line 2411  for (;;)
2411      break;      break;
2412    
2413      case OP_NOT_WHITESPACE:      case OP_NOT_WHITESPACE:
2414      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2415          {
2416          SCHECK_PARTIAL();
2417          RRETURN(MATCH_NOMATCH);
2418          }
2419      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2420      if (      if (
2421  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2422         c < 256 &&         c < 256 &&
2423  #endif  #endif
2424         (md->ctypes[c] & ctype_space) != 0         (md->ctypes[c] & ctype_space) != 0
# Line 1389  for (;;) Line 2428  for (;;)
2428      break;      break;
2429    
2430      case OP_WHITESPACE:      case OP_WHITESPACE:
2431      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2432          {
2433          SCHECK_PARTIAL();
2434          RRETURN(MATCH_NOMATCH);
2435          }
2436      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2437      if (      if (
2438  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2439         c >= 256 ||         c > 255 ||
2440  #endif  #endif
2441         (md->ctypes[c] & ctype_space) == 0         (md->ctypes[c] & ctype_space) == 0
2442         )         )
# Line 1402  for (;;) Line 2445  for (;;)
2445      break;      break;
2446    
2447      case OP_NOT_WORDCHAR:      case OP_NOT_WORDCHAR:
2448      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2449          {
2450          SCHECK_PARTIAL();
2451          RRETURN(MATCH_NOMATCH);
2452          }
2453      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2454      if (      if (
2455  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2456         c < 256 &&         c < 256 &&
2457  #endif  #endif
2458         (md->ctypes[c] & ctype_word) != 0         (md->ctypes[c] & ctype_word) != 0
# Line 1415  for (;;) Line 2462  for (;;)
2462      break;      break;
2463    
2464      case OP_WORDCHAR:      case OP_WORDCHAR:
2465      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2466          {
2467          SCHECK_PARTIAL();
2468          RRETURN(MATCH_NOMATCH);
2469          }
2470      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2471      if (      if (
2472  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2473         c >= 256 ||         c > 255 ||
2474  #endif  #endif
2475         (md->ctypes[c] & ctype_word) == 0         (md->ctypes[c] & ctype_word) == 0
2476         )         )
# Line 1428  for (;;) Line 2479  for (;;)
2479      break;      break;
2480    
2481      case OP_ANYNL:      case OP_ANYNL:
2482      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2483          {
2484          SCHECK_PARTIAL();
2485          RRETURN(MATCH_NOMATCH);
2486          }
2487      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2488      switch(c)      switch(c)
2489        {        {
2490        default: RRETURN(MATCH_NOMATCH);        default: RRETURN(MATCH_NOMATCH);
2491        case 0x000d:  
2492        if (eptr < md->end_subject && *eptr == 0x0a) eptr++;        case CHAR_CR:
2493          if (eptr >= md->end_subject)
2494            {
2495            SCHECK_PARTIAL();
2496            }
2497          else if (RAWUCHARTEST(eptr) == CHAR_LF) eptr++;
2498          break;
2499    
2500          case CHAR_LF:
2501        break;        break;
2502        case 0x000a:  
2503        case 0x000b:        case CHAR_VT:
2504        case 0x000c:        case CHAR_FF:
2505        case 0x0085:        case CHAR_NEL:
2506    #ifndef EBCDIC
2507        case 0x2028:        case 0x2028:
2508        case 0x2029:        case 0x2029:
2509    #endif  /* Not EBCDIC */
2510          if (md->bsr_anycrlf) RRETURN(MATCH_NOMATCH);
2511        break;        break;
2512        }        }
2513      ecode++;      ecode++;
2514      break;      break;
2515    
2516        case OP_NOT_HSPACE:
2517        if (eptr >= md->end_subject)
2518          {
2519          SCHECK_PARTIAL();
2520          RRETURN(MATCH_NOMATCH);
2521          }
2522        GETCHARINCTEST(c, eptr);
2523        switch(c)
2524          {
2525          HSPACE_CASES: RRETURN(MATCH_NOMATCH);  /* Byte and multibyte cases */
2526          default: break;
2527          }
2528        ecode++;
2529        break;
2530    
2531        case OP_HSPACE:
2532        if (eptr >= md->end_subject)
2533          {
2534          SCHECK_PARTIAL();
2535          RRETURN(MATCH_NOMATCH);
2536          }
2537        GETCHARINCTEST(c, eptr);
2538        switch(c)
2539          {
2540          HSPACE_CASES: break;  /* Byte and multibyte cases */
2541          default: RRETURN(MATCH_NOMATCH);
2542          }
2543        ecode++;
2544        break;
2545    
2546        case OP_NOT_VSPACE:
2547        if (eptr >= md->end_subject)
2548          {
2549          SCHECK_PARTIAL();
2550          RRETURN(MATCH_NOMATCH);
2551          }
2552        GETCHARINCTEST(c, eptr);
2553        switch(c)
2554          {
2555          VSPACE_CASES: RRETURN(MATCH_NOMATCH);
2556          default: break;
2557          }
2558        ecode++;
2559        break;
2560    
2561        case OP_VSPACE:
2562        if (eptr >= md->end_subject)
2563          {
2564          SCHECK_PARTIAL();
2565          RRETURN(MATCH_NOMATCH);
2566          }
2567        GETCHARINCTEST(c, eptr);
2568        switch(c)
2569          {
2570          VSPACE_CASES: break;
2571          default: RRETURN(MATCH_NOMATCH);
2572          }
2573        ecode++;
2574        break;
2575    
2576  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
2577      /* Check the next character by Unicode property. We will get here only      /* Check the next character by Unicode property. We will get here only
2578      if the support is in the binary; otherwise a compile-time error occurs. */      if the support is in the binary; otherwise a compile-time error occurs. */
2579    
2580      case OP_PROP:      case OP_PROP:
2581      case OP_NOTPROP:      case OP_NOTPROP:
2582      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2583          {
2584          SCHECK_PARTIAL();
2585          RRETURN(MATCH_NOMATCH);
2586          }
2587      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2588        {        {
2589        int chartype, script;        const pcre_uint32 *cp;
2590        int category = _pcre_ucp_findprop(c, &chartype, &script);        const ucd_record *prop = GET_UCD(c);
2591    
2592        switch(ecode[1])        switch(ecode[1])
2593          {          {
# Line 1466  for (;;) Line 2596  for (;;)
2596          break;          break;
2597    
2598          case PT_LAMP:          case PT_LAMP:
2599          if ((chartype == ucp_Lu ||          if ((prop->chartype == ucp_Lu ||
2600               chartype == ucp_Ll ||               prop->chartype == ucp_Ll ||
2601               chartype == ucp_Lt) == (op == OP_NOTPROP))               prop->chartype == ucp_Lt) == (op == OP_NOTPROP))
2602            RRETURN(MATCH_NOMATCH);            RRETURN(MATCH_NOMATCH);
2603           break;          break;
2604    
2605          case PT_GC:          case PT_GC:
2606          if ((ecode[2] != category) == (op == OP_PROP))          if ((ecode[2] != PRIV(ucp_gentype)[prop->chartype]) == (op == OP_PROP))
2607            RRETURN(MATCH_NOMATCH);            RRETURN(MATCH_NOMATCH);
2608          break;          break;
2609    
2610          case PT_PC:          case PT_PC:
2611          if ((ecode[2] != chartype) == (op == OP_PROP))          if ((ecode[2] != prop->chartype) == (op == OP_PROP))
2612            RRETURN(MATCH_NOMATCH);            RRETURN(MATCH_NOMATCH);
2613          break;          break;
2614    
2615          case PT_SC:          case PT_SC:
2616          if ((ecode[2] != script) == (op == OP_PROP))          if ((ecode[2] != prop->script) == (op == OP_PROP))
2617            RRETURN(MATCH_NOMATCH);            RRETURN(MATCH_NOMATCH);
2618          break;          break;
2619    
2620            /* These are specials */
2621    
2622            case PT_ALNUM:
2623            if ((PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2624                 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == (op == OP_NOTPROP))
2625              RRETURN(MATCH_NOMATCH);
2626            break;
2627    
2628            case PT_SPACE:    /* Perl space */
2629            if ((PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2630                 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2631                   == (op == OP_NOTPROP))
2632              RRETURN(MATCH_NOMATCH);
2633            break;
2634    
2635            case PT_PXSPACE:  /* POSIX space */
2636            if ((PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2637                 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2638                 c == CHAR_FF || c == CHAR_CR)
2639                   == (op == OP_NOTPROP))
2640              RRETURN(MATCH_NOMATCH);
2641            break;
2642    
2643            case PT_WORD:
2644            if ((PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2645                 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2646                 c == CHAR_UNDERSCORE) == (op == OP_NOTPROP))
2647              RRETURN(MATCH_NOMATCH);
2648            break;
2649    
2650            case PT_CLIST:
2651            cp = PRIV(ucd_caseless_sets) + ecode[2];
2652            for (;;)
2653              {
2654              if (c < *cp)
2655                { if (op == OP_PROP) { RRETURN(MATCH_NOMATCH); } else break; }
2656              if (c == *cp++)
2657                { if (op == OP_PROP) break; else { RRETURN(MATCH_NOMATCH); } }
2658              }
2659            break;
2660    
2661            case PT_UCNC:
2662            if ((c == CHAR_DOLLAR_SIGN || c == CHAR_COMMERCIAL_AT ||
2663                 c == CHAR_GRAVE_ACCENT || (c >= 0xa0 && c <= 0xd7ff) ||
2664                 c >= 0xe000) == (op == OP_NOTPROP))
2665              RRETURN(MATCH_NOMATCH);
2666            break;
2667    
2668            /* This should never occur */
2669    
2670          default:          default:
2671          RRETURN(PCRE_ERROR_INTERNAL);          RRETURN(PCRE_ERROR_INTERNAL);
2672          }          }
# Line 1499  for (;;) Line 2679  for (;;)
2679      is in the binary; otherwise a compile-time error occurs. */      is in the binary; otherwise a compile-time error occurs. */
2680    
2681      case OP_EXTUNI:      case OP_EXTUNI:
2682      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2683      GETCHARINCTEST(c, eptr);        {
2684          SCHECK_PARTIAL();
2685          RRETURN(MATCH_NOMATCH);
2686          }
2687        else
2688        {        {
2689        int chartype, script;        int lgb, rgb;
2690        int category = _pcre_ucp_findprop(c, &chartype, &script);        GETCHARINCTEST(c, eptr);
2691        if (category == ucp_M) RRETURN(MATCH_NOMATCH);        lgb = UCD_GRAPHBREAK(c);
2692        while (eptr < md->end_subject)        while (eptr < md->end_subject)
2693          {          {
2694          int len = 1;          int len = 1;
2695          if (!utf8) c = *eptr; else          if (!utf) c = *eptr; else { GETCHARLEN(c, eptr, len); }
2696            {          rgb = UCD_GRAPHBREAK(c);
2697            GETCHARLEN(c, eptr, len);          if ((PRIV(ucp_gbtable)[lgb] & (1 << rgb)) == 0) break;
2698            }          lgb = rgb;
         category = _pcre_ucp_findprop(c, &chartype, &script);  
         if (category != ucp_M) break;  
2699          eptr += len;          eptr += len;
2700          }          }
2701        }        }
2702        CHECK_PARTIAL();
2703      ecode++;      ecode++;
2704      break;      break;
2705  #endif  #endif  /* SUPPORT_UCP */
2706    
2707    
2708      /* Match a back reference, possibly repeatedly. Look past the end of the      /* Match a back reference, possibly repeatedly. Look past the end of the
# Line 1530  for (;;) Line 2713  for (;;)
2713      it as matched, any number of times (otherwise there could be infinite      it as matched, any number of times (otherwise there could be infinite
2714      loops). */      loops). */
2715    
2716      case OP_REF:      case OP_REF:
2717        {      case OP_REFI:
2718        offset = GET2(ecode, 1) << 1;               /* Doubled ref number */      caseless = op == OP_REFI;
2719        ecode += 3;                                 /* Advance past item */      offset = GET2(ecode, 1) << 1;               /* Doubled ref number */
2720        ecode += 1 + IMM2_SIZE;
2721    
2722        /* If the reference is unset, there are two possibilities:
2723    
2724        (a) In the default, Perl-compatible state, set the length negative;
2725        this ensures that every attempt at a match fails. We can't just fail
2726        here, because of the possibility of quantifiers with zero minima.
2727    
2728        (b) If the JavaScript compatibility flag is set, set the length to zero
2729        so that the back reference matches an empty string.
2730    
2731        /* If the reference is unset, set the length to be longer than the amount      Otherwise, set the length to the length of what was matched by the
2732        of subject left; this ensures that every attempt at a match fails. We      referenced subpattern. */
       can't just fail here, because of the possibility of quantifiers with zero  
       minima. */  
   
       length = (offset >= offset_top || md->offset_vector[offset] < 0)?  
         md->end_subject - eptr + 1 :  
         md->offset_vector[offset+1] - md->offset_vector[offset];  
2733    
2734        /* Set up for repetition, or handle the non-repeated case */      if (offset >= offset_top || md->offset_vector[offset] < 0)
2735          length = (md->jscript_compat)? 0 : -1;
2736        else
2737          length = md->offset_vector[offset+1] - md->offset_vector[offset];
2738    
2739        switch (*ecode)      /* Set up for repetition, or handle the non-repeated case */
         {  
         case OP_CRSTAR:  
         case OP_CRMINSTAR:  
         case OP_CRPLUS:  
         case OP_CRMINPLUS:  
         case OP_CRQUERY:  
         case OP_CRMINQUERY:  
         c = *ecode++ - OP_CRSTAR;  
         minimize = (c & 1) != 0;  
         min = rep_min[c];                 /* Pick up values from tables; */  
         max = rep_max[c];                 /* zero for max => infinity */  
         if (max == 0) max = INT_MAX;  
         break;  
2740    
2741          case OP_CRRANGE:      switch (*ecode)
2742          case OP_CRMINRANGE:        {
2743          minimize = (*ecode == OP_CRMINRANGE);        case OP_CRSTAR:
2744          min = GET2(ecode, 1);        case OP_CRMINSTAR:
2745          max = GET2(ecode, 3);        case OP_CRPLUS:
2746          if (max == 0) max = INT_MAX;        case OP_CRMINPLUS:
2747          ecode += 5;        case OP_CRQUERY:
2748          break;        case OP_CRMINQUERY:
2749          c = *ecode++ - OP_CRSTAR;
2750          minimize = (c & 1) != 0;
2751          min = rep_min[c];                 /* Pick up values from tables; */
2752          max = rep_max[c];                 /* zero for max => infinity */
2753          if (max == 0) max = INT_MAX;
2754          break;
2755    
2756          default:               /* No repeat follows */        case OP_CRRANGE:
2757          if (!match_ref(offset, eptr, length, md, ims)) RRETURN(MATCH_NOMATCH);        case OP_CRMINRANGE:
2758          eptr += length;        minimize = (*ecode == OP_CRMINRANGE);
2759          continue;              /* With the main loop */        min = GET2(ecode, 1);
2760          max = GET2(ecode, 1 + IMM2_SIZE);
2761          if (max == 0) max = INT_MAX;
2762          ecode += 1 + 2 * IMM2_SIZE;
2763          break;
2764    
2765          default:               /* No repeat follows */
2766          if ((length = match_ref(offset, eptr, length, md, caseless)) < 0)
2767            {
2768            if (length == -2) eptr = md->end_subject;   /* Partial match */
2769            CHECK_PARTIAL();
2770            RRETURN(MATCH_NOMATCH);
2771          }          }
2772          eptr += length;
2773          continue;              /* With the main loop */
2774          }
2775    
2776        /* If the length of the reference is zero, just continue with the      /* Handle repeated back references. If the length of the reference is
2777        main loop. */      zero, just continue with the main loop. If the length is negative, it
2778        means the reference is unset in non-Java-compatible mode. If the minimum is
2779        zero, we can continue at the same level without recursion. For any other
2780        minimum, carrying on will result in NOMATCH. */
2781    
2782        if (length == 0) continue;      if (length == 0) continue;
2783        if (length < 0 && min == 0) continue;
2784    
2785        /* First, ensure the minimum number of matches are present. We get back      /* First, ensure the minimum number of matches are present. We get back
2786        the length of the reference string explicitly rather than passing the      the length of the reference string explicitly rather than passing the
2787        address of eptr, so that eptr can be a register variable. */      address of eptr, so that eptr can be a register variable. */
2788    
2789        for (i = 1; i <= min; i++)      for (i = 1; i <= min; i++)
2790          {
2791          int slength;
2792          if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
2793          {          {
2794          if (!match_ref(offset, eptr, length, md, ims)) RRETURN(MATCH_NOMATCH);          if (slength == -2) eptr = md->end_subject;   /* Partial match */
2795          eptr += length;          CHECK_PARTIAL();
2796            RRETURN(MATCH_NOMATCH);
2797          }          }
2798          eptr += slength;
2799          }
2800    
2801        /* If min = max, continue at the same level without recursion.      /* If min = max, continue at the same level without recursion.
2802        They are not both allowed to be zero. */      They are not both allowed to be zero. */
2803    
2804        if (min == max) continue;      if (min == max) continue;
2805    
2806        /* If minimizing, keep trying and advancing the pointer */      /* If minimizing, keep trying and advancing the pointer */
2807    
2808        if (minimize)      if (minimize)
2809          {
2810          for (fi = min;; fi++)
2811          {          {
2812          for (fi = min;; fi++)          int slength;
2813            RMATCH(eptr, ecode, offset_top, md, eptrb, RM14);
2814            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2815            if (fi >= max) RRETURN(MATCH_NOMATCH);
2816            if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
2817            {            {
2818            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);            if (slength == -2) eptr = md->end_subject;   /* Partial match */
2819            if (rrc != MATCH_NOMATCH) RRETURN(rrc);            CHECK_PARTIAL();
2820            if (fi >= max || !match_ref(offset, eptr, length, md, ims))            RRETURN(MATCH_NOMATCH);
             RRETURN(MATCH_NOMATCH);  
           eptr += length;  
2821            }            }
2822          /* Control never gets here */          eptr += slength;
2823          }          }
2824          /* Control never gets here */
2825          }
2826    
2827        /* If maximizing, find the longest string and work backwards */      /* If maximizing, find the longest string and work backwards */
2828    
2829        else      else
2830          {
2831          pp = eptr;
2832          for (i = min; i < max; i++)
2833          {          {
2834          pp = eptr;          int slength;
2835          for (i = min; i < max; i++)          if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
           {  
           if (!match_ref(offset, eptr, length, md, ims)) break;  
           eptr += length;  
           }  
         while (eptr >= pp)  
2836            {            {
2837            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);            /* Can't use CHECK_PARTIAL because we don't want to update eptr in
2838            if (rrc != MATCH_NOMATCH) RRETURN(rrc);            the soft partial matching case. */
2839            eptr -= length;  
2840              if (slength == -2 && md->partial != 0 &&
2841                  md->end_subject > md->start_used_ptr)
2842                {
2843                md->hitend = TRUE;
2844                if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2845                }
2846              break;
2847            }            }
2848          RRETURN(MATCH_NOMATCH);          eptr += slength;
2849            }
2850    
2851          while (eptr >= pp)
2852            {
2853            RMATCH(eptr, ecode, offset_top, md, eptrb, RM15);
2854            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2855            eptr -= length;
2856          }          }
2857          RRETURN(MATCH_NOMATCH);
2858        }        }
2859      /* Control never gets here */      /* Control never gets here */
2860    
   
   
2861      /* Match a bit-mapped character class, possibly repeatedly. This op code is      /* Match a bit-mapped character class, possibly repeatedly. This op code is
2862      used when all the characters in the class have values in the range 0-255,      used when all the characters in the class have values in the range 0-255,
2863      and either the matching is caseful, or the characters are in the range      and either the matching is caseful, or the characters are in the range
# Line 1648  for (;;) Line 2872  for (;;)
2872      case OP_NCLASS:      case OP_NCLASS:
2873      case OP_CLASS:      case OP_CLASS:
2874        {        {
2875          /* The data variable is saved across frames, so the byte map needs to
2876          be stored there. */
2877    #define BYTE_MAP ((pcre_uint8 *)data)
2878        data = ecode + 1;                /* Save for matching */        data = ecode + 1;                /* Save for matching */
2879        ecode += 33;                     /* Advance past the item */        ecode += 1 + (32 / sizeof(pcre_uchar)); /* Advance past the item */
2880    
2881        switch (*ecode)        switch (*ecode)
2882          {          {
# Line 1670  for (;;) Line 2897  for (;;)
2897          case OP_CRMINRANGE:          case OP_CRMINRANGE:
2898          minimize = (*ecode == OP_CRMINRANGE);          minimize = (*ecode == OP_CRMINRANGE);
2899          min = GET2(ecode, 1);          min = GET2(ecode, 1);
2900          max = GET2(ecode, 3);          max = GET2(ecode, 1 + IMM2_SIZE);
2901          if (max == 0) max = INT_MAX;          if (max == 0) max = INT_MAX;
2902          ecode += 5;          ecode += 1 + 2 * IMM2_SIZE;
2903          break;          break;
2904    
2905          default:               /* No repeat follows */          default:               /* No repeat follows */
# Line 1682  for (;;) Line 2909  for (;;)
2909    
2910        /* First, ensure the minimum number of matches are present. */        /* First, ensure the minimum number of matches are present. */
2911    
2912  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
2913        /* UTF-8 mode */        if (utf)
       if (utf8)  
2914          {          {
2915          for (i = 1; i <= min; i++)          for (i = 1; i <= min; i++)
2916            {            {
2917            if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);            if (eptr >= md->end_subject)
2918                {
2919                SCHECK_PARTIAL();
2920                RRETURN(MATCH_NOMATCH);
2921                }
2922            GETCHARINC(c, eptr);            GETCHARINC(c, eptr);
2923            if (c > 255)            if (c > 255)
2924              {              {
2925              if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);              if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
2926              }              }
2927            else            else
2928              {              if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
             if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);  
             }  
2929            }            }
2930          }          }
2931        else        else
2932  #endif  #endif
2933        /* Not UTF-8 mode */        /* Not UTF mode */
2934          {          {
2935          for (i = 1; i <= min; i++)          for (i = 1; i <= min; i++)
2936            {            {
2937            if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);            if (eptr >= md->end_subject)
2938                {
2939                SCHECK_PARTIAL();
2940                RRETURN(MATCH_NOMATCH);
2941                }
2942            c = *eptr++;            c = *eptr++;
2943            if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);  #ifndef COMPILE_PCRE8
2944              if (c > 255)
2945                {
2946                if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
2947                }
2948              else
2949    #endif
2950                if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
2951            }            }
2952          }          }
2953    
# Line 1722  for (;;) Line 2961  for (;;)
2961    
2962        if (minimize)        if (minimize)
2963          {          {
2964  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
2965          /* UTF-8 mode */          if (utf)
         if (utf8)  
2966            {            {
2967            for (fi = min;; fi++)            for (fi = min;; fi++)
2968              {              {
2969              RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              RMATCH(eptr, ecode, offset_top, md, eptrb, RM16);
2970              if (rrc != MATCH_NOMATCH) RRETURN(rrc);              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2971              if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);              if (fi >= max) RRETURN(MATCH_NOMATCH);
2972                if (eptr >= md->end_subject)
2973                  {
2974                  SCHECK_PARTIAL();
2975                  RRETURN(MATCH_NOMATCH);
2976                  }
2977              GETCHARINC(c, eptr);              GETCHARINC(c, eptr);
2978              if (c > 255)              if (c > 255)
2979                {                {
2980                if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);                if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
2981                }                }
2982              else              else
2983                {                if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
               if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);  
               }  
2984              }              }
2985            }            }
2986          else          else
2987  #endif  #endif
2988          /* Not UTF-8 mode */          /* Not UTF mode */
2989            {            {
2990            for (fi = min;; fi++)            for (fi = min;; fi++)
2991              {              {
2992              RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              RMATCH(eptr, ecode, offset_top, md, eptrb, RM17);
2993              if (rrc != MATCH_NOMATCH) RRETURN(rrc);              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2994              if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);              if (fi >= max) RRETURN(MATCH_NOMATCH);
2995                if (eptr >= md->end_subject)
2996                  {
2997                  SCHECK_PARTIAL();
2998                  RRETURN(MATCH_NOMATCH);
2999                  }
3000              c = *eptr++;              c = *eptr++;
3001              if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);  #ifndef COMPILE_PCRE8
3002                if (c > 255)
3003                  {
3004                  if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
3005                  }
3006                else
3007    #endif
3008                  if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
3009              }              }
3010            }            }
3011          /* Control never gets here */          /* Control never gets here */
# Line 1764  for (;;) Line 3017  for (;;)
3017          {          {
3018          pp = eptr;          pp = eptr;
3019    
3020  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
3021          /* UTF-8 mode */          if (utf)
         if (utf8)  
3022            {            {
3023            for (i = min; i < max; i++)            for (i = min; i < max; i++)
3024              {              {
3025              int len = 1;              int len = 1;
3026              if (eptr >= md->end_subject) break;              if (eptr >= md->end_subject)
3027                  {
3028                  SCHECK_PARTIAL();
3029                  break;
3030                  }
3031              GETCHARLEN(c, eptr, len);              GETCHARLEN(c, eptr, len);
3032              if (c > 255)              if (c > 255)
3033                {                {
3034                if (op == OP_CLASS) break;                if (op == OP_CLASS) break;
3035                }                }
3036              else              else
3037                {                if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) break;
               if ((data[c/8] & (1 << (c&7))) == 0) break;  
               }  
3038              eptr += len;              eptr += len;
3039              }              }
3040            for (;;)            for (;;)
3041              {              {
3042              RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              RMATCH(eptr, ecode, offset_top, md, eptrb, RM18);
3043              if (rrc != MATCH_NOMATCH) RRETURN(rrc);              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3044              if (eptr-- == pp) break;        /* Stop if tried at original pos */              if (eptr-- == pp) break;        /* Stop if tried at original pos */
3045              BACKCHAR(eptr);              BACKCHAR(eptr);
# Line 1793  for (;;) Line 3047  for (;;)
3047            }            }
3048          else          else
3049  #endif  #endif
3050            /* Not UTF-8 mode */            /* Not UTF mode */
3051            {            {
3052            for (i = min; i < max; i++)            for (i = min; i < max; i++)
3053              {              {
3054              if (eptr >= md->end_subject) break;              if (eptr >= md->end_subject)
3055                  {
3056                  SCHECK_PARTIAL();
3057                  break;
3058                  }
3059              c = *eptr;              c = *eptr;
3060              if ((data[c/8] & (1 << (c&7))) == 0) break;  #ifndef COMPILE_PCRE8
3061                if (c > 255)
3062                  {
3063                  if (op == OP_CLASS) break;
3064                  }
3065                else
3066    #endif
3067                  if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) break;
3068              eptr++;              eptr++;
3069              }              }
3070            while (eptr >= pp)            while (eptr >= pp)
3071              {              {
3072              RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              RMATCH(eptr, ecode, offset_top, md, eptrb, RM19);
3073              if (rrc != MATCH_NOMATCH) RRETURN(rrc);              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3074              eptr--;              eptr--;
3075              }              }
# Line 1812  for (;;) Line 3077  for (;;)
3077    
3078          RRETURN(MATCH_NOMATCH);          RRETURN(MATCH_NOMATCH);
3079          }          }
3080    #undef BYTE_MAP
3081        }        }
3082      /* Control never gets here */      /* Control never gets here */
3083    
3084    
3085      /* Match an extended character class. This opcode is encountered only      /* Match an extended character class. This opcode is encountered only
3086      in UTF-8 mode, because that's the only time it is compiled. */      when UTF-8 mode mode is supported. Nevertheless, we may not be in UTF-8
3087        mode, because Unicode properties are supported in non-UTF-8 mode. */
3088    
3089  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3090      case OP_XCLASS:      case OP_XCLASS:
3091        {        {
3092        data = ecode + 1 + LINK_SIZE;                /* Save for matching */        data = ecode + 1 + LINK_SIZE;                /* Save for matching */
# Line 1844  for (;;) Line 3111  for (;;)
3111          case OP_CRMINRANGE:          case OP_CRMINRANGE:
3112          minimize = (*ecode == OP_CRMINRANGE);          minimize = (*ecode == OP_CRMINRANGE);
3113          min = GET2(ecode, 1);          min = GET2(ecode, 1);
3114          max = GET2(ecode, 3);          max = GET2(ecode, 1 + IMM2_SIZE);
3115          if (max == 0) max = INT_MAX;          if (max == 0) max = INT_MAX;
3116          ecode += 5;          ecode += 1 + 2 * IMM2_SIZE;
3117          break;          break;
3118    
3119          default:               /* No repeat follows */          default:               /* No repeat follows */
# Line 1858  for (;;) Line 3125  for (;;)
3125    
3126        for (i = 1; i <= min; i++)        for (i = 1; i <= min; i++)
3127          {          {
3128          if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);          if (eptr >= md->end_subject)
3129          GETCHARINC(c, eptr);            {
3130          if (!_pcre_xclass(c, data)) RRETURN(MATCH_NOMATCH);            SCHECK_PARTIAL();
3131              RRETURN(MATCH_NOMATCH);
3132              }
3133            GETCHARINCTEST(c, eptr);
3134            if (!PRIV(xclass)(c, data, utf)) RRETURN(MATCH_NOMATCH);
3135          }          }
3136    
3137        /* If max == min we can continue with the main loop without the        /* If max == min we can continue with the main loop without the
# Line 1875  for (;;) Line 3146  for (;;)
3146          {          {
3147          for (fi = min;; fi++)          for (fi = min;; fi++)
3148            {            {
3149            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);            RMATCH(eptr, ecode, offset_top, md, eptrb, RM20);
3150            if (rrc != MATCH_NOMATCH) RRETURN(rrc);            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3151            if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);            if (fi >= max) RRETURN(MATCH_NOMATCH);
3152            GETCHARINC(c, eptr);            if (eptr >= md->end_subject)
3153            if (!_pcre_xclass(c, data)) RRETURN(MATCH_NOMATCH);              {
3154                SCHECK_PARTIAL();
3155                RRETURN(MATCH_NOMATCH);
3156                }
3157              GETCHARINCTEST(c, eptr);
3158              if (!PRIV(xclass)(c, data, utf)) RRETURN(MATCH_NOMATCH);
3159            }            }
3160          /* Control never gets here */          /* Control never gets here */
3161          }          }
# Line 1892  for (;;) Line 3168  for (;;)
3168          for (i = min; i < max; i++)          for (i = min; i < max; i++)
3169            {            {
3170            int len = 1;            int len = 1;
3171            if (eptr >= md->end_subject) break;            if (eptr >= md->end_subject)
3172            GETCHARLEN(c, eptr, len);              {
3173            if (!_pcre_xclass(c, data)) break;              SCHECK_PARTIAL();
3174                break;
3175                }
3176    #ifdef SUPPORT_UTF
3177              GETCHARLENTEST(c, eptr, len);
3178    #else
3179              c = *eptr;
3180    #endif
3181              if (!PRIV(xclass)(c, data, utf)) break;
3182            eptr += len;            eptr += len;
3183            }            }
3184          for(;;)          for(;;)
3185            {            {
3186            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);            RMATCH(eptr, ecode, offset_top, md, eptrb, RM21);
3187            if (rrc != MATCH_NOMATCH) RRETURN(rrc);            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3188            if (eptr-- == pp) break;        /* Stop if tried at original pos */            if (eptr-- == pp) break;        /* Stop if tried at original pos */
3189            BACKCHAR(eptr)  #ifdef SUPPORT_UTF
3190              if (utf) BACKCHAR(eptr);
3191    #endif
3192            }            }
3193          RRETURN(MATCH_NOMATCH);          RRETURN(MATCH_NOMATCH);
3194          }          }
# Line 1914  for (;;) Line 3200  for (;;)
3200      /* Match a single character, casefully */      /* Match a single character, casefully */
3201    
3202      case OP_CHAR:      case OP_CHAR:
3203  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
3204      if (utf8)      if (utf)
3205        {        {
3206        length = 1;        length = 1;
3207        ecode++;        ecode++;
3208        GETCHARLEN(fc, ecode, length);        GETCHARLEN(fc, ecode, length);
3209        if (length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);        if (length > md->end_subject - eptr)
3210        while (length-- > 0) if (*ecode++ != *eptr++) RRETURN(MATCH_NOMATCH);          {
3211            CHECK_PARTIAL();             /* Not SCHECK_PARTIAL() */
3212            RRETURN(MATCH_NOMATCH);
3213            }
3214          while (length-- > 0) if (*ecode++ != RAWUCHARINC(eptr)) RRETURN(MATCH_NOMATCH);
3215        }        }
3216      else      else
3217  #endif  #endif
3218        /* Not UTF mode */
     /* Non-UTF-8 mode */  
3219        {        {
3220        if (md->end_subject - eptr < 1) RRETURN(MATCH_NOMATCH);        if (md->end_subject - eptr < 1)
3221            {
3222            SCHECK_PARTIAL();            /* This one can use SCHECK_PARTIAL() */
3223            RRETURN(MATCH_NOMATCH);
3224            }
3225        if (ecode[1] != *eptr++) RRETURN(MATCH_NOMATCH);        if (ecode[1] != *eptr++) RRETURN(MATCH_NOMATCH);
3226        ecode += 2;        ecode += 2;
3227        }        }
3228      break;      break;
3229    
3230      /* Match a single character, caselessly */      /* Match a single character, caselessly. If we are at the end of the
3231        subject, give up immediately. */
3232    
3233        case OP_CHARI:
3234        if (eptr >= md->end_subject)
3235          {
3236          SCHECK_PARTIAL();
3237          RRETURN(MATCH_NOMATCH);
3238          }
3239    
3240      case OP_CHARNC:  #ifdef SUPPORT_UTF
3241  #ifdef SUPPORT_UTF8      if (utf)
     if (utf8)  
3242        {        {
3243        length = 1;        length = 1;
3244        ecode++;        ecode++;
3245        GETCHARLEN(fc, ecode, length);        GETCHARLEN(fc, ecode, length);
3246    
       if (length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);  
   
3247        /* If the pattern character's value is < 128, we have only one byte, and        /* If the pattern character's value is < 128, we have only one byte, and
3248        can use the fast lookup table. */        we know that its other case must also be one byte long, so we can use the
3249          fast lookup table. We know that there is at least one byte left in the
3250          subject. */
3251    
3252        if (fc < 128)        if (fc < 128)
3253          {          {
3254          if (md->lcc[*ecode++] != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);          pcre_uint32 cc = RAWUCHAR(eptr);
3255            if (md->lcc[fc] != TABLE_GET(cc, md->lcc, cc)) RRETURN(MATCH_NOMATCH);
3256            ecode++;
3257            eptr++;
3258          }          }
3259    
3260        /* Otherwise we must pick up the subject character */        /* Otherwise we must pick up the subject character. Note that we cannot
3261          use the value of "length" to check for sufficient bytes left, because the
3262          other case of the character may have more or fewer bytes.  */
3263    
3264        else        else
3265          {          {
3266          unsigned int dc;          pcre_uint32 dc;
3267          GETCHARINC(dc, eptr);          GETCHARINC(dc, eptr);
3268          ecode += length;          ecode += length;
3269    
# Line 1968  for (;;) Line 3273  for (;;)
3273          if (fc != dc)          if (fc != dc)
3274            {            {
3275  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
3276            if (dc != _pcre_ucp_othercase(fc))            if (dc != UCD_OTHERCASE(fc))
3277  #endif  #endif
3278              RRETURN(MATCH_NOMATCH);              RRETURN(MATCH_NOMATCH);
3279            }            }
3280          }          }
3281        }        }
3282      else      else
3283  #endif   /* SUPPORT_UTF8 */  #endif   /* SUPPORT_UTF */
3284    
3285      /* Non-UTF-8 mode */      /* Not UTF mode */
3286        {        {
3287        if (md->end_subject - eptr < 1) RRETURN(MATCH_NOMATCH);        if (TABLE_GET(ecode[1], md->lcc, ecode[1])
3288        if (md->lcc[ecode[1]] != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);            != TABLE_GET(*eptr, md->lcc, *eptr)) RRETURN(MATCH_NOMATCH);
3289          eptr++;
3290        ecode += 2;        ecode += 2;
3291        }        }
3292      break;      break;
# Line 1988  for (;;) Line 3294  for (;;)
3294      /* Match a single character repeatedly. */      /* Match a single character repeatedly. */
3295    
3296      case OP_EXACT:      case OP_EXACT:
3297        case OP_EXACTI:
3298      min = max = GET2(ecode, 1);      min = max = GET2(ecode, 1);
3299      ecode += 3;      ecode += 1 + IMM2_SIZE;
3300      goto REPEATCHAR;      goto REPEATCHAR;
3301    
3302      case OP_POSUPTO:      case OP_POSUPTO:
3303        case OP_POSUPTOI:
3304      possessive = TRUE;      possessive = TRUE;
3305      /* Fall through */      /* Fall through */
3306    
3307      case OP_UPTO:      case OP_UPTO:
3308        case OP_UPTOI:
3309      case OP_MINUPTO:      case OP_MINUPTO:
3310        case OP_MINUPTOI:
3311      min = 0;      min = 0;
3312      max = GET2(ecode, 1);      max = GET2(ecode, 1);
3313      minimize = *ecode == OP_MINUPTO;      minimize = *ecode == OP_MINUPTO || *ecode == OP_MINUPTOI;
3314      ecode += 3;      ecode += 1 + IMM2_SIZE;
3315      goto REPEATCHAR;      goto REPEATCHAR;
3316    
3317      case OP_POSSTAR:      case OP_POSSTAR:
3318        case OP_POSSTARI:
3319      possessive = TRUE;      possessive = TRUE;
3320      min = 0;      min = 0;
3321      max = INT_MAX;      max = INT_MAX;
# Line 2012  for (;;) Line 3323  for (;;)
3323      goto REPEATCHAR;      goto REPEATCHAR;
3324    
3325      case OP_POSPLUS:      case OP_POSPLUS:
3326        case OP_POSPLUSI:
3327      possessive = TRUE;      possessive = TRUE;
3328      min = 1;      min = 1;
3329      max = INT_MAX;      max = INT_MAX;
# Line 2019  for (;;) Line 3331  for (;;)
3331      goto REPEATCHAR;      goto REPEATCHAR;
3332    
3333      case OP_POSQUERY:      case OP_POSQUERY:
3334        case OP_POSQUERYI:
3335      possessive = TRUE;      possessive = TRUE;
3336      min = 0;      min = 0;
3337      max = 1;      max = 1;
# Line 2026  for (;;) Line 3339  for (;;)
3339      goto REPEATCHAR;      goto REPEATCHAR;
3340    
3341      case OP_STAR:      case OP_STAR:
3342        case OP_STARI:
3343      case OP_MINSTAR:      case OP_MINSTAR:
3344        case OP_MINSTARI:
3345      case OP_PLUS:      case OP_PLUS:
3346        case OP_PLUSI:
3347      case OP_MINPLUS:      case OP_MINPLUS:
3348        case OP_MINPLUSI:
3349      case OP_QUERY:      case OP_QUERY:
3350        case OP_QUERYI:
3351      case OP_MINQUERY:      case OP_MINQUERY:
3352      c = *ecode++ - OP_STAR;      case OP_MINQUERYI:
3353        c = *ecode++ - ((op < OP_STARI)? OP_STAR : OP_STARI);
3354      minimize = (c & 1) != 0;      minimize = (c & 1) != 0;
3355      min = rep_min[c];                 /* Pick up values from tables; */      min = rep_min[c];                 /* Pick up values from tables; */
3356      max = rep_max[c];                 /* zero for max => infinity */      max = rep_max[c];                 /* zero for max => i