/[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 91 by nigel, Sat Feb 24 21:41:34 2007 UTC revision 1296 by ph10, Tue Mar 19 16:29:12 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) 1997-2006 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  #define NLBLOCK md           /* The block containing newline information */  #ifdef HAVE_CONFIG_H
45    #include "config.h"
46    #endif
47    
48    #define NLBLOCK md             /* Block containing newline information */
49    #define PSSTART start_subject  /* Field containing processed string start */
50    #define PSEND   end_subject    /* Field containing processed string end */
51    
52  #include "pcre_internal.h"  #include "pcre_internal.h"
53    
54    /* Undefine some potentially clashing cpp symbols */
55    
56  /* Structure for building a chain of data that actually lives on the  #undef min
57  stack, for holding the values of the subject pointer at the start of each  #undef max
 subpattern, so as to detect when an empty string has been matched by a  
 subpattern - to break infinite loops. When NO_RECURSE is set, these blocks  
 are on the heap, not on the stack. */  
   
 typedef struct eptrblock {  
   struct eptrblock *epb_prev;  
   USPTR epb_saved_eptr;  
 } eptrblock;  
58    
59  /* Flag bits for the match() function */  /* The md->capture_last field uses the lower 16 bits for the last captured
60    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    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  #define match_condassert   0x01    /* Called to check a condition assertion */  #define MATCH_CONDASSERT     1  /* Called to check a condition assertion */
78  #define match_isgroup      0x02    /* Set if start of bracketed group */  #define MATCH_CBEGROUP       2  /* Could-be-empty unlimited repeat group */
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 68  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 79  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 99  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  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 114  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 146  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 186  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 266  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    
418    BOOL Xcur_is_word;    BOOL Xcur_is_word;
419    BOOL Xcondition;    BOOL Xcondition;
   BOOL Xminimize;  
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;
426    int Xprop_category;    int Xoclength;
427    int Xprop_chartype;    pcre_uchar Xocchars[6];
   int Xprop_script;  
   int *Xprop_test_variable;  
428  #endif  #endif
429    
430      int Xcodelink;
431    int Xctype;    int Xctype;
432    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 340  typedef struct heapframe { Line 461  typedef struct heapframe {
461  *         Match from current position            *  *         Match from current position            *
462  *************************************************/  *************************************************/
463    
464  /* On entry ecode points to the first opcode, and eptr to the first character  /* This function is called recursively in many circumstances. Whenever it
 in the subject string, while eptrb holds the value of eptr at the start of the  
 last bracketed group - used for breaking infinite loops matching zero-length  
 strings. 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 in subject     eptr        pointer to current character in subject
498     ecode       position in 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_isgroup - this is the start of a bracketed 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 them with "register"  so they can be ordinary variables in all cases. Mark some of them with
521  because they are used a lot in loops. */  "register" because they are used a lot in loops. */
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 */
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 412  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 434  HEAP_RECURSE: Line 578  HEAP_RECURSE:
578    
579  #define cur_is_word        frame->Xcur_is_word  #define cur_is_word        frame->Xcur_is_word
580  #define condition          frame->Xcondition  #define condition          frame->Xcondition
 #define minimize           frame->Xminimize  
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
587  #define prop_category      frame->Xprop_category  #define oclength           frame->Xoclength
588  #define prop_chartype      frame->Xprop_chartype  #define occhars            frame->Xocchars
 #define prop_script        frame->Xprop_script  
 #define prop_test_variable frame->Xprop_test_variable  
589  #endif  #endif
590    
591  #define ctype              frame->Xctype  #define ctype              frame->Xctype
# Line 470  HEAP_RECURSE: Line 609  HEAP_RECURSE:
609  get preserved during recursion in the normal way. In this environment, fi and  get preserved during recursion in the normal way. In this environment, fi and
610  i, and fc and c, can be the same variables. */  i, and fc and c, can be the same variables. */
611    
612  #else  #else         /* NO_RECURSE not defined */
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_UTF
625    const pcre_uchar *charptr;
626    #endif
627    const pcre_uchar *callpat;
628    const pcre_uchar *data;
629    const pcre_uchar *next;
630    PCRE_PUCHAR       pp;
631    const pcre_uchar *prev;
632    PCRE_PUCHAR       saved_eptr;
633    
634  #ifdef SUPPORT_UTF8                /* Many of these variables are used only  */  recursion_info new_recursive;
635  const uschar *charptr;             /* in small blocks of the code. My normal */  
636  #endif                             /* style of coding would have declared    */  BOOL cur_is_word;
 const uschar *callpat;             /* them within each of those blocks.      */  
 const uschar *data;                /* However, in order to accommodate the   */  
 const uschar *next;                /* version of this code that uses an      */  
 USPTR         pp;                  /* external "stack" implemented on the    */  
 const uschar *prev;                /* heap, it is easier to declare them all */  
 USPTR         saved_eptr;          /* here, so the declarations can be cut   */  
                                    /* out in a block. The only declarations  */  
 recursion_info new_recursive;      /* within blocks below are for variables  */  
                                    /* that do not have to be preserved over  */  
 BOOL cur_is_word;                  /* a recursive call to RMATCH().          */  
637  BOOL condition;  BOOL condition;
 BOOL minimize;  
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;
644  int prop_category;  int oclength;
645  int prop_chartype;  pcre_uchar occhars[6];
 int prop_script;  
 int *prop_test_variable;  
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  #endif  
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 */
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. */
# Line 524  variables. */ Line 696  variables. */
696  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
697  prop_value = 0;  prop_value = 0;
698  prop_fail_result = 0;  prop_fail_result = 0;
 prop_test_variable = NULL;  
699  #endif  #endif
700    
701    
702  /* This label is used for tail recursion, which is used in a few cases even  /* This label is used for tail recursion, which is used in a few cases even
703  when NO_RECURSE is not defined, in order to reduce the amount of stack that is  when NO_RECURSE is not defined, in order to reduce the amount of stack that is
704  used. Thanks to Ian Taylor for noticing this possibility and sending the  used. Thanks to Ian Taylor for noticing this possibility and sending the
# Line 537  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    
729  original_ims = ims;    /* Save for resetting on ')' */  /* At the start of a group with an unlimited repeat that may match an empty
730    string, the variable md->match_function_type is set to MATCH_CBEGROUP. It is
731  #ifdef SUPPORT_UTF8  done this way to save having to use another function argument, which would take
732  utf8 = md->utf8;       /* Local copy of the flag */  up space on the stack. See also MATCH_CONDASSERT below.
733  #else  
734  utf8 = FALSE;  When MATCH_CBEGROUP is set, add the current subject pointer to the chain of
735  #endif  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  /* At the start of a bracketed group, add the current subject pointer to the  other circumstances, don't add to the chain. The MATCH_CBEGROUP feature must
738  stack of such pointers, to be re-instated at the end of the group when we hit  NOT be used with tail recursion, because the memory block that is used is on
739  the closing ket. When match() is called in other circumstances, we don't add to  the stack, so a new one may be required for each match(). */
 this stack. */  
740    
741  if ((flags & match_isgroup) != 0)  if (md->match_function_type == MATCH_CBEGROUP)
742    {    {
   newptrb.epb_prev = eptrb;  
743    newptrb.epb_saved_eptr = eptr;    newptrb.epb_saved_eptr = eptr;
744      newptrb.epb_prev = eptrb;
745    eptrb = &newptrb;    eptrb = &newptrb;
746      md->match_function_type = 0;
747    }    }
748    
749  /* Now start processing the operations. */  /* Now start processing the opcodes. */
750    
751  for (;;)  for (;;)
752    {    {
753      minimize = possessive = FALSE;
754    op = *ecode;    op = *ecode;
   minimize = FALSE;  
755    
756    /* For partial matching, remember if we ever hit the end of the subject after    switch(op)
757    matching at least one subject character. */      {
758        case OP_MARK:
759        md->nomatch_mark = ecode + 2;
760        md->mark = NULL;    /* In case previously set by assertion */
761        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
762          eptrb, RM55);
763        if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
764             md->mark == NULL) md->mark = ecode + 2;
765    
766        /* A return of MATCH_SKIP_ARG means that matching failed at SKIP with an
767        argument, and we must check whether that argument matches this MARK's
768        argument. It is passed back in md->start_match_ptr (an overloading of that
769        variable). If it does match, we reset that variable to the current subject
770        position and return MATCH_SKIP. Otherwise, pass back the return code
771        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    if (md->partial &&      Check the alternative branches in turn - the matching won't pass the KET
871        eptr >= md->end_subject &&      for this kind of subpattern. If any one branch matches, we carry on as at
872        eptr > md->start_match)      the end of a normal bracket, leaving the subject pointer, but resetting
873      md->hitend = TRUE;      the start-of-match value in case it was changed by \K. */
   
   /* Opening capturing bracket. If there is space in the offset vector, save  
   the current subject position in the working slot at the top of the vector. We  
   mustn't change the current values of the data slot, because they may be set  
   from a previous iteration of this group, and be referred to by a reference  
   inside the group.  
   
   If the bracket fails to match, we need to restore this value and also the  
   values of the final offsets, in case they were set by a previous iteration of  
   the same bracket.  
   
   If there isn't enough space in the offset vector, treat this as if it were a  
   non-capturing bracket. Don't worry about setting the flag for the error case  
   here; that is handled in the code for KET. */  
874    
875    if (op > OP_BRA)      case OP_ONCE_NC:
876      {      prev = ecode;
877      number = op - OP_BRA;      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      /* For extended extraction brackets (large number), we have to fish out the      /* Continue as from after the group, updating the offsets high water
906      number from a dummy opcode at the start. */      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      if (number > EXTRACT_BASIC_MAX)      /* Handle a capturing bracket, other than those that are possessive with an
945        number = GET2(ecode, 2+LINK_SIZE);      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
955        a non-capturing bracket. Don't worry about setting the flag for the error
956        case here; that is handled in the code for KET. */
957    
958        case OP_CBRA:
959        case OP_SCBRA:
960        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 subject=", number);      printf("start bracket %d\n", number);
965        printf("subject=");
966      pchars(eptr, 16, TRUE, md);      pchars(eptr, 16, TRUE, md);
967      printf("\n");      printf("\n");
968  #endif  #endif
# Line 620  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        do        for (;;)
983          {          {
984          RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb,          if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
985            match_isgroup);          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 */      /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
1028        as a non-capturing bracket. */
1029    
1030      else op = OP_BRA;      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1031      }      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1032    
1033    /* Other types of node can be handled by a switch */      DPRINTF(("insufficient capture room: treat as non-capturing\n"));
1034    
1035    switch(op)      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1036      {      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1037      case OP_BRA:     /* Non-capturing bracket: optimized */  
1038      DPRINTF(("start bracket 0\n"));      /* 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      /* Loop for all the alternatives */      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:
1060        case OP_SBRA:
1061        DPRINTF(("start non-capturing bracket\n"));
1062    
1063      for (;;)      for (;;)
1064        {        {
1065        /* When we get to the final alternative within the brackets, we would        if (op >= OP_SBRA || op == OP_ONCE)
1066        return the result of a recursive call to match() whatever happened. We          md->match_function_type = MATCH_CBEGROUP;
       can reduce stack usage by turning this into a tail recursion. */  
   
       if (ecode[GET(ecode, 1)] != OP_ALT)  
        {  
        ecode += 1 + LINK_SIZE;  
        flags = match_isgroup;  
        DPRINTF(("bracket 0 tail recursion\n"));  
        goto TAIL_RECURSE;  
        }  
1067    
1068        /* For non-final alternatives, continue the loop for a NOMATCH result;        /* If this is not a possibly empty group, and there are no (*THEN)s in
1069        otherwise return. */        the pattern, and this is the final alternative, optimize as described
1070          above. */
1071    
1072        RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb,        else if (!md->hasthen && ecode[GET(ecode, 1)] != OP_ALT)
1073          match_isgroup);          {
1074        if (rrc != MATCH_NOMATCH) RRETURN(rrc);          ecode += PRIV(OP_lengths)[*ecode];
1075        ecode += GET(ecode, 1);          goto TAIL_RECURSE;
1076        }          }
     /* Control never reaches here. */  
1077    
1078      /* Conditional group: compilation checked that there are no more than        /* In all other cases, we have to make another call to match(). */
     two branches. If the condition is false, skipping the first branch takes us  
     past the end if there is only one branch, but that's OK because that is  
     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. */  
1079    
1080      case OP_COND:        save_mark = md->mark;
1081      if (ecode[LINK_SIZE+1] == OP_CREF) /* Condition extract or recurse test */        save_capture_last = md->capture_last;
1082        {        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md, eptrb,
1083        offset = GET2(ecode, LINK_SIZE+2) << 1;  /* Doubled ref number */          RM2);
       condition = (offset == CREF_RECURSE * 2)?  
         (md->recursive != NULL) :  
         (offset < offset_top && md->offset_vector[offset] >= 0);  
       ecode += condition? (LINK_SIZE + 4) : (LINK_SIZE + 1 + GET(ecode, 1));  
       flags = match_isgroup;  
       goto TAIL_RECURSE;  
       }  
1084    
1085      /* The condition is an assertion. Call match() to evaluate it - setting        /* See comment in the code for capturing groups above about handling
1086      the final argument TRUE causes it to stop at the end of an assertion. */        THEN. */
1087    
1088      else        if (rrc == MATCH_THEN)
       {  
       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,  
           match_condassert | match_isgroup);  
       if (rrc == MATCH_MATCH)  
1089          {          {
1090          ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE+2);          next = ecode + GET(ecode,1);
1091          while (*ecode == OP_ALT) ecode += GET(ecode, 1);          if (md->start_match_ptr < next &&
1092                (*ecode == OP_ALT || *next == OP_ALT))
1093              rrc = MATCH_NOMATCH;
1094          }          }
1095        else if (rrc != MATCH_NOMATCH)  
1096          if (rrc != MATCH_NOMATCH)
1097          {          {
1098          RRETURN(rrc);         /* Need braces because of following else */          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        else ecode += GET(ecode, 1);        ecode += GET(ecode, 1);
1111          md->mark = save_mark;
1112          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        /* We are now at the branch that is to be obeyed. As there is only one,      POSSESSIVE_CAPTURE:
1131        we can use tail recursion to avoid using another stack frame. */      number = GET2(ecode, 1+LINK_SIZE);
1132        offset = number << 1;
1133    
1134        ecode += 1 + LINK_SIZE;  #ifdef PCRE_DEBUG
1135        flags = match_isgroup;      printf("start possessive bracket %d\n", number);
1136        goto TAIL_RECURSE;      printf("subject=");
1137        }      pchars(eptr, 16, TRUE, md);
1138      /* Control never reaches here */      printf("\n");
1139    #endif
1140    
1141      /* Skip over conditional reference or large extraction number data if      if (offset < md->offset_max)
1142      encountered. */        {
1143          matched_once = FALSE;
1144          code_offset = (int)(ecode - md->start_code);
1145    
1146      case OP_CREF:        save_offset1 = md->offset_vector[offset];
1147      case OP_BRANUMBER:        save_offset2 = md->offset_vector[offset+1];
1148      ecode += 3;        save_offset3 = md->offset_vector[md->offset_end - number];
1149      break;        save_capture_last = md->capture_last;
1150    
1151      /* End of the pattern. If we are in a recursion, we should restore the        DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
     offsets appropriately and continue from after the call. */  
1152    
1153      case OP_END:        /* Each time round the loop, save the current subject position for use
1154      if (md->recursive != NULL && md->recursive->group_num == 0)        when the group matches. For MATCH_MATCH, the group has matched, so we
1155        {        restart it with a new subject starting position, remembering that we had
1156        recursion_info *rec = md->recursive;        at least one match. For MATCH_NOMATCH, carry on with the alternatives, as
1157        DPRINTF(("End of pattern in a (?0) recursion\n"));        usual. If we haven't matched any alternatives in any iteration, check to
1158        md->recursive = rec->prevrec;        see if a previous iteration matched. If so, the group has matched;
1159        memmove(md->offset_vector, rec->offset_save,        continue from afterwards. Otherwise it has failed; restore the previous
1160          rec->saved_max * sizeof(int));        capture values before returning NOMATCH. */
1161        md->start_match = rec->save_start;  
1162        ims = original_ims;        for (;;)
1163        ecode = rec->after_call;          {
1164        break;          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    
1196          if (!matched_once)
1197            {
1198            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    
1203          if (allow_zero || matched_once)
1204            {
1205            ecode += 1 + LINK_SIZE;
1206            break;
1207            }
1208    
1209          RRETURN(MATCH_NOMATCH);
1210        }        }
1211    
1212      /* Otherwise, if PCRE_NOTEMPTY is set, fail if we have matched an empty      /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
1213      string - backtracking will then try other alternatives, if any. */      as a non-capturing bracket. */
1214    
1215      if (md->notempty && eptr == md->start_match) RRETURN(MATCH_NOMATCH);      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1216      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;  
1217    
1218      /* Assertion brackets. Check the alternative branches in turn - the      DPRINTF(("insufficient capture room: treat as non-capturing\n"));
     matching won't pass the KET for an assertion. If any one branch matches,  
     the assertion is true. Lookbehind assertions have an OP_REVERSE item at the  
     start of each branch to move the current point backwards, so the code at  
     this level is identical to the lookahead case. */  
1219    
1220      case OP_ASSERT:      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1221      case OP_ASSERTBACK:      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1222      do  
1223        /* Non-capturing possessive bracket with unlimited repeat. We come here
1224        from BRAZERO with allow_zero = TRUE. The code is similar to the above,
1225        without the capturing complication. It is written out separately for speed
1226        and cleanliness. */
1227    
1228        case OP_BRAPOS:
1229        case OP_SBRAPOS:
1230        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        for (;;)
1238        {        {
1239        RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,        if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
1240          match_isgroup);        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
1241        if (rrc == MATCH_MATCH) break;          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,        if (PUBL(callout) != NULL)
1292          match_isgroup);          {
1293        if (rrc == MATCH_MATCH) RRETURN(MATCH_NOMATCH);          PUBL(callout_block) cb;
1294        if (rrc != MATCH_NOMATCH) RRETURN(rrc);          cb.version          = 2;   /* Version 1 of the callout block */
1295        ecode += GET(ecode,1);          cb.callout_number   = ecode[LINK_SIZE+2];
1296        }          cb.offset_vector    = md->offset_vector;
1297      while (*ecode == OP_ALT);  #if defined COMPILE_PCRE8
1298            cb.subject          = (PCRE_SPTR)md->start_subject;
1299      if ((flags & match_condassert) != 0) RRETURN(MATCH_MATCH);  #elif defined COMPILE_PCRE16
1300            cb.subject          = (PCRE_SPTR16)md->start_subject;
1301      ecode += 1 + LINK_SIZE;  #elif defined COMPILE_PCRE32
1302      continue;          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      /* Move the subject pointer back. This occurs only at the start of      /* Now see what the actual condition is */
     each branch of a lookbehind assertion. If we are too close to the start to  
     move back, this match function fails. When working with UTF-8 we move  
     back a number of characters, not bytes. */  
1325    
1326      case OP_REVERSE:      if (condcode == OP_RREF || condcode == OP_NRREF)    /* Recursion test */
 #ifdef SUPPORT_UTF8  
     if (utf8)  
1327        {        {
1328        c = GET(ecode,1);        if (md->recursive == NULL)                /* Not recursing => FALSE */
       for (i = 0; i < c; i++)  
1329          {          {
1330          eptr--;          condition = FALSE;
1331          if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);          ecode += GET(ecode, 1);
         BACKCHAR(eptr)  
1332          }          }
1333        }        else
1334      else          {
1335  #endif          unsigned int recno = GET2(ecode, LINK_SIZE + 2);   /* Recursion group number*/
1336            condition = (recno == RREF_ANY || recno == md->recursive->group_num);
1337    
1338      /* No UTF-8 support, or not in UTF-8 mode: count is byte count */          /* 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);
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        /* 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;
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        eptr -= GET(ecode,1);        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;
1705    
1706        /* Move the subject pointer back. This occurs only at the start of
1707        each branch of a lookbehind assertion. If we are too close to the start to
1708        move back, this match function fails. When working with UTF-8 we move
1709        back a number of characters, not bytes. */
1710    
1711        case OP_REVERSE:
1712    #ifdef SUPPORT_UTF
1713        if (utf)
1714          {
1715          i = GET(ecode, 1);
1716          while (i-- > 0)
1717            {
1718            eptr--;
1719            if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
1720            BACKCHAR(eptr);
1721            }
1722          }
1723        else
1724    #endif
1725    
1726        /* No UTF-8 support, or not in UTF-8 mode: count is byte count */
1727    
1728          {
1729          eptr -= GET(ecode, 1);
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 854  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 879  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        callpat = md->start_code + GET(ecode, 1);        recursion_info *ri;
1794        new_recursive.group_num = *callpat - OP_BRA;        unsigned int recno;
1795    
1796        /* For extended extraction brackets (large number), we have to fish out        callpat = md->start_code + GET(ecode, 1);
1797        the number from a dummy opcode at the start. */        recno = (callpat == md->start_code)? 0 :
1798            GET2(callpat, 1 + LINK_SIZE);
1799    
1800        if (new_recursive.group_num > EXTRACT_BASIC_MAX)        /* Check for repeating a recursion without advancing the subject pointer.
1801          new_recursive.group_num = GET2(callpat, 2+LINK_SIZE);        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 923  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          cbegroup = (*callpat >= OP_SBRA);
1841        do        do
1842          {          {
1843          RMATCH(rrc, eptr, callpat + 1 + LINK_SIZE, offset_top, md, ims,          if (cbegroup) md->match_function_type = MATCH_CBEGROUP;
1844              eptrb, match_isgroup);          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              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          else if (rrc != MATCH_NOMATCH)  
1865            /* PCRE does not allow THEN, SKIP, PRUNE or COMMIT to escape beyond a
1866            recursion; they are treated as NOMATCH. These codes are defined in a
1867            range that can be tested for. Any other return code is an error. */
1868    
1869            else if (rrc != MATCH_NOMATCH &&
1870                     (rrc < MATCH_BACKTRACK_MIN || rrc > MATCH_BACKTRACK_MAX))
1871            {            {
1872            DPRINTF(("Recursion gave error %d\n", rrc));            DPRINTF(("Recursion gave error %d\n", rrc));
1873              if (new_recursive.offset_save != stacksave)
1874                (PUBL(free))(new_recursive.offset_save);
1875            RRETURN(rrc);            RRETURN(rrc);
1876            }            }
1877    
1878          md->recursive = &new_recursive;          md->recursive = &new_recursive;
         memcpy(md->offset_vector, new_recursive.offset_save,  
             new_recursive.saved_max * sizeof(int));  
1879          callpat += GET(callpat, 1);          callpat += GET(callpat, 1);
1880          }          }
1881        while (*callpat == OP_ALT);        while (*callpat == OP_ALT);
# Line 964  for (;;) Line 1883  for (;;)
1883        DPRINTF(("Recursion didn't match\n"));        DPRINTF(("Recursion didn't match\n"));
1884        md->recursive = new_recursive.prevrec;        md->recursive = new_recursive.prevrec;
1885        if (new_recursive.offset_save != stacksave)        if (new_recursive.offset_save != stacksave)
1886          (pcre_free)(new_recursive.offset_save);          (PUBL(free))(new_recursive.offset_save);
1887        RRETURN(MATCH_NOMATCH);        RRETURN(MATCH_NOMATCH);
1888        }        }
     /* 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;  
   
     do  
       {  
       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims,  
         eptrb, match_isgroup);  
       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));  
       }  
1889    
1890      if (*ecode == OP_KETRMIN)      RECURSION_MATCHED:
1891        {      break;
       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb, 0);  
       if (rrc != MATCH_NOMATCH) RRETURN(rrc);  
       ecode = prev;  
       flags = match_isgroup;  
       goto TAIL_RECURSE;  
       }  
     else  /* OP_KETRMAX */  
       {  
       RMATCH(rrc, eptr, prev, offset_top, md, ims, eptrb, match_isgroup);  
       if (rrc != MATCH_NOMATCH) RRETURN(rrc);  
       ecode += 1 + LINK_SIZE;  
       flags = 0;  
       goto TAIL_RECURSE;  
       }  
     /* Control never gets here */  
1892    
1893      /* 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
1894      bracketed group and go to there. */      bracketed group and go to there. */
# Line 1051  for (;;) Line 1897  for (;;)
1897      do ecode += GET(ecode,1); while (*ecode == OP_ALT);      do ecode += GET(ecode,1); while (*ecode == OP_ALT);
1898      break;      break;
1899    
1900      /* BRAZERO and BRAMINZERO occur just before a bracket group, indicating      /* BRAZERO, BRAMINZERO and SKIPZERO occur just before a bracket group,
1901      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
1902      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
1903      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
1904      preceded by BRAZERO or BRAMINZERO. */      optional ones preceded by BRAZERO or BRAMINZERO. */
1905    
1906      case OP_BRAZERO:      case OP_BRAZERO:
1907        {      next = ecode + 1;
1908        next = ecode+1;      RMATCH(eptr, next, offset_top, md, eptrb, RM10);
1909        RMATCH(rrc, eptr, next, offset_top, md, ims, eptrb, match_isgroup);      if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1910        if (rrc != MATCH_NOMATCH) RRETURN(rrc);      do next += GET(next, 1); while (*next == OP_ALT);
1911        do next += GET(next,1); while (*next == OP_ALT);      ecode = next + 1 + LINK_SIZE;
       ecode = next + 1+LINK_SIZE;  
       }  
1912      break;      break;
1913    
1914      case OP_BRAMINZERO:      case OP_BRAMINZERO:
1915        {      next = ecode + 1;
1916        next = ecode+1;      do next += GET(next, 1); while (*next == OP_ALT);
1917        do next += GET(next,1); while (*next == OP_ALT);      RMATCH(eptr, next + 1+LINK_SIZE, offset_top, md, eptrb, RM11);
1918        RMATCH(rrc, eptr, next + 1+LINK_SIZE, offset_top, md, ims, eptrb,      if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1919          match_isgroup);      ecode++;
1920        if (rrc != MATCH_NOMATCH) RRETURN(rrc);      break;
1921        ecode++;  
1922        }      case OP_SKIPZERO:
1923        next = ecode+1;
1924        do next += GET(next,1); while (*next == OP_ALT);
1925        ecode = next + 1 + LINK_SIZE;
1926      break;      break;
1927    
1928      /* End of a group, repeated or non-repeating. If we are at the end of      /* BRAPOSZERO occurs before a possessive bracket group. Don't do anything
1929      an assertion "group", stop matching and return MATCH_MATCH, but record the      here; just jump to the group, with allow_zero set TRUE. */
1930      current high water mark for use by positive assertions. Do this also  
1931      for the "once" (not-backup up) groups. */      case OP_BRAPOSZERO:
1932        op = *(++ecode);
1933        allow_zero = TRUE;
1934        if (op == OP_CBRAPOS || op == OP_SCBRAPOS) goto POSSESSIVE_CAPTURE;
1935          goto POSSESSIVE_NON_CAPTURE;
1936    
1937        /* End of a group, repeated or non-repeating. */
1938    
1939      case OP_KET:      case OP_KET:
1940      case OP_KETRMIN:      case OP_KETRMIN:
1941      case OP_KETRMAX:      case OP_KETRMAX:
1942        case OP_KETRPOS:
1943      prev = ecode - GET(ecode, 1);      prev = ecode - GET(ecode, 1);
     saved_eptr = eptrb->epb_saved_eptr;  
   
     /* Back up the stack of bracket start pointers. */  
1944    
1945      eptrb = eptrb->epb_prev;      /* If this was a group that remembered the subject start, in order to break
1946        infinite repeats of empty string matches, retrieve the subject start from
1947        the chain. Otherwise, set it NULL. */
1948    
1949      if (*prev == OP_ASSERT || *prev == OP_ASSERT_NOT ||      if (*prev >= OP_SBRA || *prev == OP_ONCE)
         *prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT ||  
         *prev == OP_ONCE)  
1950        {        {
1951        md->end_match_ptr = eptr;      /* For ONCE */        saved_eptr = eptrb->epb_saved_eptr;   /* Value at start of group */
1952        md->end_offset_top = offset_top;        eptrb = eptrb->epb_prev;              /* Backup to previous group */
       RRETURN(MATCH_MATCH);  
1953        }        }
1954        else saved_eptr = NULL;
1955    
1956      /* In all other cases except a conditional group we have to check the      /* If we are at the end of an assertion group or a non-capturing atomic
1957      group number back at the start and if necessary complete handling an      group, stop matching and return MATCH_MATCH, but record the current high
1958      extraction by setting the offsets and bumping the high water mark. */      water mark for use by positive assertions. We also need to record the match
1959        start in case it was changed by \K. */
1960    
1961      if (*prev != OP_COND)      if ((*prev >= OP_ASSERT && *prev <= OP_ASSERTBACK_NOT) ||
1962             *prev == OP_ONCE_NC)
1963        {        {
1964        number = *prev - OP_BRA;        md->end_match_ptr = eptr;      /* For ONCE_NC */
1965          md->end_offset_top = offset_top;
1966          md->start_match_ptr = mstart;
1967          RRETURN(MATCH_MATCH);         /* Sets md->mark */
1968          }
1969    
1970        /* For extended extraction brackets (large number), we have to fish out      /* For capturing groups we have to check the group number back at the start
1971        the number from a dummy opcode at the start. */      and if necessary complete handling an extraction by setting the offsets and
1972        bumping the high water mark. Whole-pattern recursion is coded as a recurse
1973        into group 0, so it won't be picked up here. Instead, we catch it when the
1974        OP_END is reached. Other recursion is handled here. We just have to record
1975        the current subject position and start match pointer and give a MATCH
1976        return. */
1977    
1978        if (number > EXTRACT_BASIC_MAX) number = GET2(prev, 2+LINK_SIZE);      if (*prev == OP_CBRA || *prev == OP_SCBRA ||
1979            *prev == OP_CBRAPOS || *prev == OP_SCBRAPOS)
1980          {
1981          number = GET2(prev, 1+LINK_SIZE);
1982        offset = number << 1;        offset = number << 1;
1983    
1984  #ifdef DEBUG  #ifdef PCRE_DEBUG
1985        printf("end bracket %d", number);        printf("end bracket %d", number);
1986        printf("\n");        printf("\n");
1987  #endif  #endif
1988    
1989        /* Test for a numbered group. This includes groups called as a result        /* Handle a recursively called group. */
       of recursion. Note that whole-pattern recursion is coded as a recurse  
       into group 0, so it won't be picked up here. Instead, we catch it when  
       the OP_END is reached. */  
1990    
1991        if (number > 0)        if (md->recursive != NULL && md->recursive->group_num == number)
1992          {          {
1993          md->capture_last = number;          md->end_match_ptr = eptr;
1994          if (offset >= md->offset_max) md->offset_overflow = TRUE; else          md->start_match_ptr = mstart;
1995            {          RRETURN(MATCH_MATCH);
1996            md->offset_vector[offset] =          }
             md->offset_vector[md->offset_end - number];  
           md->offset_vector[offset+1] = eptr - md->start_subject;  
           if (offset_top <= offset) offset_top = offset + 2;  
           }  
1997    
1998          /* Handle a recursively called group. Restore the offsets        /* Deal with capturing */
         appropriately and continue from after the call. */  
1999    
2000          if (md->recursive != NULL && md->recursive->group_num == number)        md->capture_last = (md->capture_last & OVFLMASK) | number;
2001          if (offset >= md->offset_max) md->capture_last |= OVFLBIT; else
2002            {
2003            /* If offset is greater than offset_top, it means that we are
2004            "skipping" a capturing group, and that group's offsets must be marked
2005            unset. In earlier versions of PCRE, all the offsets were unset at the
2006            start of matching, but this doesn't work because atomic groups and
2007            assertions can cause a value to be set that should later be unset.
2008            Example: matching /(?>(a))b|(a)c/ against "ac". This sets group 1 as
2009            part of the atomic group, but this is not on the final matching path,
2010            so must be unset when 2 is set. (If there is no group 2, there is no
2011            problem, because offset_top will then be 2, indicating no capture.) */
2012    
2013            if (offset > offset_top)
2014            {            {
2015            recursion_info *rec = md->recursive;            register int *iptr = md->offset_vector + offset_top;
2016            DPRINTF(("Recursion (%d) succeeded - continuing\n", number));            register int *iend = md->offset_vector + offset;
2017            md->recursive = rec->prevrec;            while (iptr < iend) *iptr++ = -1;
           md->start_match = rec->save_start;  
           memcpy(md->offset_vector, rec->offset_save,  
             rec->saved_max * sizeof(int));  
           ecode = rec->after_call;  
           ims = original_ims;  
           break;  
2018            }            }
         }  
       }  
2019    
2020      /* Reset the value of the ims flags, in case they got changed during          /* Now make the extraction */
     the group. */  
2021    
2022      ims = original_ims;          md->offset_vector[offset] =
2023      DPRINTF(("ims reset to %02lx\n", ims));            md->offset_vector[md->offset_end - number];
2024            md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
2025            if (offset_top <= offset) offset_top = offset + 2;
2026            }
2027          }
2028    
2029      /* For a non-repeating ket, just continue at this level. This also      /* For an ordinary non-repeating ket, just continue at this level. This
2030      happens for a repeating ket if no characters were matched in the group.      also happens for a repeating ket if no characters were matched in the
2031      This is the forcible breaking of infinite loops as implemented in Perl      group. This is the forcible breaking of infinite loops as implemented in
2032      5.005. If there is an options reset, it will get obeyed in the normal      Perl 5.005. For a non-repeating atomic group that includes captures,
2033      course of events. */      establish a backup point by processing the rest of the pattern at a lower
2034        level. If this results in a NOMATCH return, pass MATCH_ONCE back to the
2035        original OP_ONCE level, thereby bypassing intermediate backup points, but
2036        resetting any captures that happened along the way. */
2037    
2038      if (*ecode == OP_KET || eptr == saved_eptr)      if (*ecode == OP_KET || eptr == saved_eptr)
2039        {        {
2040        ecode += 1 + LINK_SIZE;        if (*prev == OP_ONCE)
2041            {
2042            RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM12);
2043            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2044            md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
2045            RRETURN(MATCH_ONCE);
2046            }
2047          ecode += 1 + LINK_SIZE;    /* Carry on at this level */
2048        break;        break;
2049        }        }
2050    
2051      /* The repeating kets try the rest of the pattern or restart from the      /* OP_KETRPOS is a possessive repeating ket. Remember the current position,
2052      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
2053      tail recursion to avoid using another stack frame. */      at a time from the outer level, thus saving stack. */
2054    
2055        if (*ecode == OP_KETRPOS)
2056          {
2057          md->end_match_ptr = eptr;
2058          md->end_offset_top = offset_top;
2059          RRETURN(MATCH_KETRPOS);
2060          }
2061    
2062        /* The normal repeating kets try the rest of the pattern or restart from
2063        the preceding bracket, in the appropriate order. In the second case, we can
2064        use tail recursion to avoid using another stack frame, unless we have an
2065        an atomic group or an unlimited repeat of a group that can match an empty
2066        string. */
2067    
2068      if (*ecode == OP_KETRMIN)      if (*ecode == OP_KETRMIN)
2069        {        {
2070        RMATCH(rrc, eptr, ecode + 1+LINK_SIZE, offset_top, md, ims, eptrb, 0);        RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM7);
2071        if (rrc != MATCH_NOMATCH) RRETURN(rrc);        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2072          if (*prev == OP_ONCE)
2073            {
2074            RMATCH(eptr, prev, offset_top, md, eptrb, RM8);
2075            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2076            md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
2077            RRETURN(MATCH_ONCE);
2078            }
2079          if (*prev >= OP_SBRA)    /* Could match an empty string */
2080            {
2081            RMATCH(eptr, prev, offset_top, md, eptrb, RM50);
2082            RRETURN(rrc);
2083            }
2084        ecode = prev;        ecode = prev;
       flags = match_isgroup;  
2085        goto TAIL_RECURSE;        goto TAIL_RECURSE;
2086        }        }
2087      else  /* OP_KETRMAX */      else  /* OP_KETRMAX */
2088        {        {
2089        RMATCH(rrc, eptr, prev, offset_top, md, ims, eptrb, match_isgroup);        RMATCH(eptr, prev, offset_top, md, eptrb, RM13);
2090          if (rrc == MATCH_ONCE && md->once_target == prev) rrc = MATCH_NOMATCH;
2091        if (rrc != MATCH_NOMATCH) RRETURN(rrc);        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2092          if (*prev == OP_ONCE)
2093            {
2094            RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM9);
2095            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2096            md->once_target = prev;
2097            RRETURN(MATCH_ONCE);
2098            }
2099        ecode += 1 + LINK_SIZE;        ecode += 1 + LINK_SIZE;
       flags = 0;  
2100        goto TAIL_RECURSE;        goto TAIL_RECURSE;
2101        }        }
2102      /* Control never gets here */      /* Control never gets here */
2103    
2104      /* Start of subject unless notbol, or after internal newline if multiline */      /* Not multiline mode: start of subject assertion, unless notbol. */
2105    
2106      case OP_CIRC:      case OP_CIRC:
2107      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 ||  
            eptr < md->start_subject + md->nllen ||  
            !IS_NEWLINE(eptr - md->nllen)))  
         RRETURN(MATCH_NOMATCH);  
       ecode++;  
       break;  
       }  
     /* ... else fall through */  
2108    
2109      /* Start of subject assertion */      /* Start of subject assertion */
2110    
# Line 1218  for (;;) Line 2113  for (;;)
2113      ecode++;      ecode++;
2114      break;      break;
2115    
2116        /* Multiline mode: start of subject unless notbol, or after any newline. */
2117    
2118        case OP_CIRCM:
2119        if (md->notbol && eptr == md->start_subject) RRETURN(MATCH_NOMATCH);
2120        if (eptr != md->start_subject &&
2121            (eptr == md->end_subject || !WAS_NEWLINE(eptr)))
2122          RRETURN(MATCH_NOMATCH);
2123        ecode++;
2124        break;
2125    
2126      /* Start of match assertion */      /* Start of match assertion */
2127    
2128      case OP_SOM:      case OP_SOM:
# Line 1225  for (;;) Line 2130  for (;;)
2130      ecode++;      ecode++;
2131      break;      break;
2132    
2133      /* 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. */  
2134    
2135      case OP_DOLL:      case OP_SET_SOM:
2136      if ((ims & PCRE_MULTILINE) != 0)      mstart = eptr;
2137        ecode++;
2138        break;
2139    
2140        /* Multiline mode: assert before any newline, or before end of subject
2141        unless noteol is set. */
2142    
2143        case OP_DOLLM:
2144        if (eptr < md->end_subject)
2145        {        {
2146        if (eptr < md->end_subject)        if (!IS_NEWLINE(eptr))
2147          { if (!IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH); }          {
2148        else          if (md->partial != 0 &&
2149          { if (md->noteol) RRETURN(MATCH_NOMATCH); }              eptr + 1 >= md->end_subject &&
2150        ecode++;              NLBLOCK->nltype == NLTYPE_FIXED &&
2151        break;              NLBLOCK->nllen == 2 &&
2152                RAWUCHARTEST(eptr) == NLBLOCK->nl[0])
2153              {
2154              md->hitend = TRUE;
2155              if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2156              }
2157            RRETURN(MATCH_NOMATCH);
2158            }
2159        }        }
2160      else      else
2161        {        {
2162        if (md->noteol) RRETURN(MATCH_NOMATCH);        if (md->noteol) RRETURN(MATCH_NOMATCH);
2163        if (!md->endonly)        SCHECK_PARTIAL();
         {  
         if (eptr != md->end_subject &&  
             (eptr != md->end_subject - md->nllen || !IS_NEWLINE(eptr)))  
           RRETURN(MATCH_NOMATCH);  
         ecode++;  
         break;  
         }  
2164        }        }
2165        ecode++;
2166        break;
2167    
2168        /* Not multiline mode: assert before a terminating newline or before end of
2169        subject unless noteol is set. */
2170    
2171        case OP_DOLL:
2172        if (md->noteol) RRETURN(MATCH_NOMATCH);
2173        if (!md->endonly) goto ASSERT_NL_OR_EOS;
2174    
2175      /* ... else fall through for endonly */      /* ... else fall through for endonly */
2176    
2177      /* End of subject assertion (\z) */      /* End of subject assertion (\z) */
2178    
2179      case OP_EOD:      case OP_EOD:
2180      if (eptr < md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr < md->end_subject) RRETURN(MATCH_NOMATCH);
2181        SCHECK_PARTIAL();
2182      ecode++;      ecode++;
2183      break;      break;
2184    
2185      /* End of subject or ending \n assertion (\Z) */      /* End of subject or ending \n assertion (\Z) */
2186    
2187      case OP_EODN:      case OP_EODN:
2188      if (eptr != md->end_subject &&      ASSERT_NL_OR_EOS:
2189          (eptr != md->end_subject - md->nllen || !IS_NEWLINE(eptr)))      if (eptr < md->end_subject &&
2190            (!IS_NEWLINE(eptr) || eptr != md->end_subject - md->nllen))
2191          {
2192          if (md->partial != 0 &&
2193              eptr + 1 >= md->end_subject &&
2194              NLBLOCK->nltype == NLTYPE_FIXED &&
2195              NLBLOCK->nllen == 2 &&
2196              RAWUCHARTEST(eptr) == NLBLOCK->nl[0])
2197            {
2198            md->hitend = TRUE;
2199            if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2200            }
2201        RRETURN(MATCH_NOMATCH);        RRETURN(MATCH_NOMATCH);
2202          }
2203    
2204        /* Either at end of string or \n before end. */
2205    
2206        SCHECK_PARTIAL();
2207      ecode++;      ecode++;
2208      break;      break;
2209    
# Line 1276  for (;;) Line 2215  for (;;)
2215    
2216        /* Find out if the previous and current characters are "word" characters.        /* Find out if the previous and current characters are "word" characters.
2217        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
2218        be "non-word" characters. */        be "non-word" characters. Remember the earliest consulted character for
2219          partial matching. */
2220    
2221  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
2222        if (utf8)        if (utf)
2223          {          {
2224            /* Get status of previous character */
2225    
2226          if (eptr == md->start_subject) prev_is_word = FALSE; else          if (eptr == md->start_subject) prev_is_word = FALSE; else
2227            {            {
2228            const uschar *lastptr = eptr - 1;            PCRE_PUCHAR lastptr = eptr - 1;
2229            while((*lastptr & 0xc0) == 0x80) lastptr--;            BACKCHAR(lastptr);
2230              if (lastptr < md->start_used_ptr) md->start_used_ptr = lastptr;
2231            GETCHAR(c, lastptr);            GETCHAR(c, lastptr);
2232    #ifdef SUPPORT_UCP
2233              if (md->use_ucp)
2234                {
2235                if (c == '_') prev_is_word = TRUE; else
2236                  {
2237                  int cat = UCD_CATEGORY(c);
2238                  prev_is_word = (cat == ucp_L || cat == ucp_N);
2239                  }
2240                }
2241              else
2242    #endif
2243            prev_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;            prev_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;
2244            }            }
2245          if (eptr >= md->end_subject) cur_is_word = FALSE; else  
2246            /* Get status of next character */
2247    
2248            if (eptr >= md->end_subject)
2249              {
2250              SCHECK_PARTIAL();
2251              cur_is_word = FALSE;
2252              }
2253            else
2254            {            {
2255            GETCHAR(c, eptr);            GETCHAR(c, eptr);
2256    #ifdef SUPPORT_UCP
2257              if (md->use_ucp)
2258                {
2259                if (c == '_') cur_is_word = TRUE; else
2260                  {
2261                  int cat = UCD_CATEGORY(c);
2262                  cur_is_word = (cat == ucp_L || cat == ucp_N);
2263                  }
2264                }
2265              else
2266    #endif
2267            cur_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;            cur_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;
2268            }            }
2269          }          }
2270        else        else
2271  #endif  #endif
2272    
2273        /* More streamlined when not in UTF-8 mode */        /* Not in UTF-8 mode, but we may still have PCRE_UCP set, and for
2274          consistency with the behaviour of \w we do use it in this case. */
2275    
2276          {          {
2277          prev_is_word = (eptr != md->start_subject) &&          /* Get status of previous character */
2278            ((md->ctypes[eptr[-1]] & ctype_word) != 0);  
2279          cur_is_word = (eptr < md->end_subject) &&          if (eptr == md->start_subject) prev_is_word = FALSE; else
2280            ((md->ctypes[*eptr] & ctype_word) != 0);            {
2281              if (eptr <= md->start_used_ptr) md->start_used_ptr = eptr - 1;
2282    #ifdef SUPPORT_UCP
2283              if (md->use_ucp)
2284                {
2285                c = eptr[-1];
2286                if (c == '_') prev_is_word = TRUE; else
2287                  {
2288                  int cat = UCD_CATEGORY(c);
2289                  prev_is_word = (cat == ucp_L || cat == ucp_N);
2290                  }
2291                }
2292              else
2293    #endif
2294              prev_is_word = MAX_255(eptr[-1])
2295                && ((md->ctypes[eptr[-1]] & ctype_word) != 0);
2296              }
2297    
2298            /* Get status of next character */
2299    
2300            if (eptr >= md->end_subject)
2301              {
2302              SCHECK_PARTIAL();
2303              cur_is_word = FALSE;
2304              }
2305            else
2306    #ifdef SUPPORT_UCP
2307            if (md->use_ucp)
2308              {
2309              c = *eptr;
2310              if (c == '_') cur_is_word = TRUE; else
2311                {
2312                int cat = UCD_CATEGORY(c);
2313                cur_is_word = (cat == ucp_L || cat == ucp_N);
2314                }
2315              }
2316            else
2317    #endif
2318            cur_is_word = MAX_255(*eptr)
2319              && ((md->ctypes[*eptr] & ctype_word) != 0);
2320          }          }
2321    
2322        /* Now see if the situation is what we want */        /* Now see if the situation is what we want */
# Line 1314  for (;;) Line 2327  for (;;)
2327        }        }
2328      break;      break;
2329    
2330      /* Match a single character type; inline for speed */      /* Match any single character type except newline; have to take care with
2331        CRLF newlines and partial matching. */
2332    
2333      case OP_ANY:      case OP_ANY:
2334      if ((ims & PCRE_DOTALL) == 0)      if (IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH);
2335        if (md->partial != 0 &&
2336            eptr + 1 >= md->end_subject &&
2337            NLBLOCK->nltype == NLTYPE_FIXED &&
2338            NLBLOCK->nllen == 2 &&
2339            RAWUCHARTEST(eptr) == NLBLOCK->nl[0])
2340        {        {
2341        if (eptr <= md->end_subject - md->nllen && IS_NEWLINE(eptr))        md->hitend = TRUE;
2342          RRETURN(MATCH_NOMATCH);        if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2343        }        }
2344      if (eptr++ >= md->end_subject) RRETURN(MATCH_NOMATCH);  
2345      if (utf8)      /* Fall through */
2346        while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;  
2347        /* Match any single character whatsoever. */
2348    
2349        case OP_ALLANY:
2350        if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
2351          {                            /* not be updated before SCHECK_PARTIAL. */
2352          SCHECK_PARTIAL();
2353          RRETURN(MATCH_NOMATCH);
2354          }
2355        eptr++;
2356    #ifdef SUPPORT_UTF
2357        if (utf) ACROSSCHAR(eptr < md->end_subject, *eptr, eptr++);
2358    #endif
2359      ecode++;      ecode++;
2360      break;      break;
2361    
# Line 1332  for (;;) Line 2363  for (;;)
2363      any byte, even newline, independent of the setting of PCRE_DOTALL. */      any byte, even newline, independent of the setting of PCRE_DOTALL. */
2364    
2365      case OP_ANYBYTE:      case OP_ANYBYTE:
2366      if (eptr++ >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
2367          {                            /* not be updated before SCHECK_PARTIAL. */
2368          SCHECK_PARTIAL();
2369          RRETURN(MATCH_NOMATCH);
2370          }
2371        eptr++;
2372      ecode++;      ecode++;
2373      break;      break;
2374    
2375      case OP_NOT_DIGIT:      case OP_NOT_DIGIT:
2376      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2377          {
2378          SCHECK_PARTIAL();
2379          RRETURN(MATCH_NOMATCH);
2380          }
2381      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2382      if (      if (
2383  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2384         c < 256 &&         c < 256 &&
2385  #endif  #endif
2386         (md->ctypes[c] & ctype_digit) != 0         (md->ctypes[c] & ctype_digit) != 0
# Line 1350  for (;;) Line 2390  for (;;)
2390      break;      break;
2391    
2392      case OP_DIGIT:      case OP_DIGIT:
2393      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2394          {
2395          SCHECK_PARTIAL();
2396          RRETURN(MATCH_NOMATCH);
2397          }
2398      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2399      if (      if (
2400  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2401         c >= 256 ||         c > 255 ||
2402  #endif  #endif
2403         (md->ctypes[c] & ctype_digit) == 0         (md->ctypes[c] & ctype_digit) == 0
2404         )         )
# Line 1363  for (;;) Line 2407  for (;;)
2407      break;      break;
2408    
2409      case OP_NOT_WHITESPACE:      case OP_NOT_WHITESPACE:
2410      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2411          {
2412          SCHECK_PARTIAL();
2413          RRETURN(MATCH_NOMATCH);
2414          }
2415      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2416      if (      if (
2417  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2418         c < 256 &&         c < 256 &&
2419  #endif  #endif
2420         (md->ctypes[c] & ctype_space) != 0         (md->ctypes[c] & ctype_space) != 0
# Line 1376  for (;;) Line 2424  for (;;)
2424      break;      break;
2425    
2426      case OP_WHITESPACE:      case OP_WHITESPACE:
2427      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2428          {
2429          SCHECK_PARTIAL();
2430          RRETURN(MATCH_NOMATCH);
2431          }
2432      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2433      if (      if (
2434  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2435         c >= 256 ||         c > 255 ||
2436  #endif  #endif
2437         (md->ctypes[c] & ctype_space) == 0         (md->ctypes[c] & ctype_space) == 0
2438         )         )
# Line 1389  for (;;) Line 2441  for (;;)
2441      break;      break;
2442    
2443      case OP_NOT_WORDCHAR:      case OP_NOT_WORDCHAR:
2444      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2445          {
2446          SCHECK_PARTIAL();
2447          RRETURN(MATCH_NOMATCH);
2448          }
2449      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2450      if (      if (
2451  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2452         c < 256 &&         c < 256 &&
2453  #endif  #endif
2454         (md->ctypes[c] & ctype_word) != 0         (md->ctypes[c] & ctype_word) != 0
# Line 1402  for (;;) Line 2458  for (;;)
2458      break;      break;
2459    
2460      case OP_WORDCHAR:      case OP_WORDCHAR:
2461      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2462          {
2463          SCHECK_PARTIAL();
2464          RRETURN(MATCH_NOMATCH);
2465          }
2466      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2467      if (      if (
2468  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2469         c >= 256 ||         c > 255 ||
2470  #endif  #endif
2471         (md->ctypes[c] & ctype_word) == 0         (md->ctypes[c] & ctype_word) == 0
2472         )         )
# Line 1414  for (;;) Line 2474  for (;;)
2474      ecode++;      ecode++;
2475      break;      break;
2476    
2477        case OP_ANYNL:
2478        if (eptr >= md->end_subject)
2479          {
2480          SCHECK_PARTIAL();
2481          RRETURN(MATCH_NOMATCH);
2482          }
2483        GETCHARINCTEST(c, eptr);
2484        switch(c)
2485          {
2486          default: RRETURN(MATCH_NOMATCH);
2487    
2488          case CHAR_CR:
2489          if (eptr >= md->end_subject)
2490            {
2491            SCHECK_PARTIAL();
2492            }
2493          else if (RAWUCHARTEST(eptr) == CHAR_LF) eptr++;
2494          break;
2495    
2496          case CHAR_LF:
2497          break;
2498    
2499          case CHAR_VT:
2500          case CHAR_FF:
2501          case CHAR_NEL:
2502    #ifndef EBCDIC
2503          case 0x2028:
2504          case 0x2029:
2505    #endif  /* Not EBCDIC */
2506          if (md->bsr_anycrlf) RRETURN(MATCH_NOMATCH);
2507          break;
2508          }
2509        ecode++;
2510        break;
2511    
2512        case OP_NOT_HSPACE:
2513        if (eptr >= md->end_subject)
2514          {
2515          SCHECK_PARTIAL();
2516          RRETURN(MATCH_NOMATCH);
2517          }
2518        GETCHARINCTEST(c, eptr);
2519        switch(c)
2520          {
2521          HSPACE_CASES: RRETURN(MATCH_NOMATCH);  /* Byte and multibyte cases */
2522          default: break;
2523          }
2524        ecode++;
2525        break;
2526    
2527        case OP_HSPACE:
2528        if (eptr >= md->end_subject)
2529          {
2530          SCHECK_PARTIAL();
2531          RRETURN(MATCH_NOMATCH);
2532          }
2533        GETCHARINCTEST(c, eptr);
2534        switch(c)
2535          {
2536          HSPACE_CASES: break;  /* Byte and multibyte cases */
2537          default: RRETURN(MATCH_NOMATCH);
2538          }
2539        ecode++;
2540        break;
2541    
2542        case OP_NOT_VSPACE:
2543        if (eptr >= md->end_subject)
2544          {
2545          SCHECK_PARTIAL();
2546          RRETURN(MATCH_NOMATCH);
2547          }
2548        GETCHARINCTEST(c, eptr);
2549        switch(c)
2550          {
2551          VSPACE_CASES: RRETURN(MATCH_NOMATCH);
2552          default: break;
2553          }
2554        ecode++;
2555        break;
2556    
2557        case OP_VSPACE:
2558        if (eptr >= md->end_subject)
2559          {
2560          SCHECK_PARTIAL();
2561          RRETURN(MATCH_NOMATCH);
2562          }
2563        GETCHARINCTEST(c, eptr);
2564        switch(c)
2565          {
2566          VSPACE_CASES: break;
2567          default: RRETURN(MATCH_NOMATCH);
2568          }
2569        ecode++;
2570        break;
2571    
2572  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
2573      /* Check the next character by Unicode property. We will get here only      /* Check the next character by Unicode property. We will get here only
2574      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. */
2575    
2576      case OP_PROP:      case OP_PROP:
2577      case OP_NOTPROP:      case OP_NOTPROP:
2578      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2579          {
2580          SCHECK_PARTIAL();
2581          RRETURN(MATCH_NOMATCH);
2582          }
2583      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2584        {        {
2585        int chartype, script;        const pcre_uint32 *cp;
2586        int category = _pcre_ucp_findprop(c, &chartype, &script);        const ucd_record *prop = GET_UCD(c);
2587    
2588        switch(ecode[1])        switch(ecode[1])
2589          {          {
# Line 1433  for (;;) Line 2592  for (;;)
2592          break;          break;
2593    
2594          case PT_LAMP:          case PT_LAMP:
2595          if ((chartype == ucp_Lu ||          if ((prop->chartype == ucp_Lu ||
2596               chartype == ucp_Ll ||               prop->chartype == ucp_Ll ||
2597               chartype == ucp_Lt) == (op == OP_NOTPROP))               prop->chartype == ucp_Lt) == (op == OP_NOTPROP))
2598            RRETURN(MATCH_NOMATCH);            RRETURN(MATCH_NOMATCH);
2599           break;          break;
2600    
2601          case PT_GC:          case PT_GC:
2602          if ((ecode[2] != category) == (op == OP_PROP))          if ((ecode[2] != PRIV(ucp_gentype)[prop->chartype]) == (op == OP_PROP))
2603            RRETURN(MATCH_NOMATCH);            RRETURN(MATCH_NOMATCH);
2604          break;          break;
2605    
2606          case PT_PC:          case PT_PC:
2607          if ((ecode[2] != chartype) == (op == OP_PROP))          if ((ecode[2] != prop->chartype) == (op == OP_PROP))
2608            RRETURN(MATCH_NOMATCH);            RRETURN(MATCH_NOMATCH);
2609          break;          break;
2610    
2611          case PT_SC:          case PT_SC:
2612          if ((ecode[2] != script) == (op == OP_PROP))          if ((ecode[2] != prop->script) == (op == OP_PROP))
2613              RRETURN(MATCH_NOMATCH);
2614            break;
2615    
2616            /* These are specials */
2617    
2618            case PT_ALNUM:
2619            if ((PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2620                 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == (op == OP_NOTPROP))
2621              RRETURN(MATCH_NOMATCH);
2622            break;
2623    
2624            case PT_SPACE:    /* Perl space */
2625            if ((PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2626                 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2627                   == (op == OP_NOTPROP))
2628              RRETURN(MATCH_NOMATCH);
2629            break;
2630    
2631            case PT_PXSPACE:  /* POSIX space */
2632            if ((PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2633                 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2634                 c == CHAR_FF || c == CHAR_CR)
2635                   == (op == OP_NOTPROP))
2636              RRETURN(MATCH_NOMATCH);
2637            break;
2638    
2639            case PT_WORD:
2640            if ((PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2641                 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2642                 c == CHAR_UNDERSCORE) == (op == OP_NOTPROP))
2643            RRETURN(MATCH_NOMATCH);            RRETURN(MATCH_NOMATCH);
2644          break;          break;
2645    
2646            case PT_CLIST:
2647            cp = PRIV(ucd_caseless_sets) + ecode[2];
2648            for (;;)
2649              {
2650              if (c < *cp)
2651                { if (op == OP_PROP) { RRETURN(MATCH_NOMATCH); } else break; }
2652              if (c == *cp++)
2653                { if (op == OP_PROP) break; else { RRETURN(MATCH_NOMATCH); } }
2654              }
2655            break;
2656    
2657            case PT_UCNC:
2658            if ((c == CHAR_DOLLAR_SIGN || c == CHAR_COMMERCIAL_AT ||
2659                 c == CHAR_GRAVE_ACCENT || (c >= 0xa0 && c <= 0xd7ff) ||
2660                 c >= 0xe000) == (op == OP_NOTPROP))
2661              RRETURN(MATCH_NOMATCH);
2662            break;
2663    
2664            /* This should never occur */
2665    
2666          default:          default:
2667          RRETURN(PCRE_ERROR_INTERNAL);          RRETURN(PCRE_ERROR_INTERNAL);
         break;  
2668          }          }
2669    
2670        ecode += 3;        ecode += 3;
# Line 1467  for (;;) Line 2675  for (;;)
2675      is in the binary; otherwise a compile-time error occurs. */      is in the binary; otherwise a compile-time error occurs. */
2676    
2677      case OP_EXTUNI:      case OP_EXTUNI:
2678      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2679      GETCHARINCTEST(c, eptr);        {
2680          SCHECK_PARTIAL();
2681          RRETURN(MATCH_NOMATCH);
2682          }
2683        else
2684        {        {
2685        int chartype, script;        int lgb, rgb;
2686        int category = _pcre_ucp_findprop(c, &chartype, &script);        GETCHARINCTEST(c, eptr);
2687        if (category == ucp_M) RRETURN(MATCH_NOMATCH);        lgb = UCD_GRAPHBREAK(c);
2688        while (eptr < md->end_subject)        while (eptr < md->end_subject)
2689          {          {
2690          int len = 1;          int len = 1;
2691          if (!utf8) c = *eptr; else          if (!utf) c = *eptr; else { GETCHARLEN(c, eptr, len); }
2692            {          rgb = UCD_GRAPHBREAK(c);
2693            GETCHARLEN(c, eptr, len);          if ((PRIV(ucp_gbtable)[lgb] & (1 << rgb)) == 0) break;
2694            }          lgb = rgb;
         category = _pcre_ucp_findprop(c, &chartype, &script);  
         if (category != ucp_M) break;  
2695          eptr += len;          eptr += len;
2696          }          }
2697        }        }
2698        CHECK_PARTIAL();
2699      ecode++;      ecode++;
2700      break;      break;
2701  #endif  #endif  /* SUPPORT_UCP */
2702    
2703    
2704      /* 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 1499  for (;;) Line 2710  for (;;)
2710      loops). */      loops). */
2711    
2712      case OP_REF:      case OP_REF:
2713        {      case OP_REFI:
2714        offset = GET2(ecode, 1) << 1;               /* Doubled ref number */      caseless = op == OP_REFI;
2715        ecode += 3;                                 /* Advance past item */      offset = GET2(ecode, 1) << 1;               /* Doubled ref number */
2716        ecode += 1 + IMM2_SIZE;
2717    
2718        /* If the reference is unset, set the length to be longer than the amount      /* If the reference is unset, there are two possibilities:
       of subject left; this ensures that every attempt at a match fails. We  
       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];  
2719    
2720        /* Set up for repetition, or handle the non-repeated case */      (a) In the default, Perl-compatible state, set the length negative;
2721        this ensures that every attempt at a match fails. We can't just fail
2722        here, because of the possibility of quantifiers with zero minima.
2723    
2724        switch (*ecode)      (b) If the JavaScript compatibility flag is set, set the length to zero
2725          {      so that the back reference matches an empty string.
         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;  
2726    
2727          case OP_CRRANGE:      Otherwise, set the length to the length of what was matched by the
2728          case OP_CRMINRANGE:      referenced subpattern. */
         minimize = (*ecode == OP_CRMINRANGE);  
         min = GET2(ecode, 1);  
         max = GET2(ecode, 3);  
         if (max == 0) max = INT_MAX;  
         ecode += 5;  
         break;  
2729    
2730          default:               /* No repeat follows */      if (offset >= offset_top || md->offset_vector[offset] < 0)
2731          if (!match_ref(offset, eptr, length, md, ims)) RRETURN(MATCH_NOMATCH);        length = (md->jscript_compat)? 0 : -1;
2732          eptr += length;      else
2733          continue;              /* With the main loop */        length = md->offset_vector[offset+1] - md->offset_vector[offset];
2734    
2735        /* Set up for repetition, or handle the non-repeated case */
2736    
2737        switch (*ecode)
2738          {
2739          case OP_CRSTAR:
2740          case OP_CRMINSTAR:
2741          case OP_CRPLUS:
2742          case OP_CRMINPLUS:
2743          case OP_CRQUERY:
2744          case OP_CRMINQUERY:
2745          c = *ecode++ - OP_CRSTAR;
2746          minimize = (c & 1) != 0;
2747          min = rep_min[c];                 /* Pick up values from tables; */
2748          max = rep_max[c];                 /* zero for max => infinity */
2749          if (max == 0) max = INT_MAX;
2750          break;
2751    
2752          case OP_CRRANGE:
2753          case OP_CRMINRANGE:
2754          minimize = (*ecode == OP_CRMINRANGE);
2755          min = GET2(ecode, 1);
2756          max = GET2(ecode, 1 + IMM2_SIZE);
2757          if (max == 0) max = INT_MAX;
2758          ecode += 1 + 2 * IMM2_SIZE;
2759          break;
2760    
2761          default:               /* No repeat follows */
2762          if ((length = match_ref(offset, eptr, length, md, caseless)) < 0)
2763            {
2764            if (length == -2) eptr = md->end_subject;   /* Partial match */
2765            CHECK_PARTIAL();
2766            RRETURN(MATCH_NOMATCH);
2767          }          }
2768          eptr += length;
2769          continue;              /* With the main loop */
2770          }
2771    
2772        /* If the length of the reference is zero, just continue with the      /* Handle repeated back references. If the length of the reference is
2773        main loop. */      zero, just continue with the main loop. If the length is negative, it
2774        means the reference is unset in non-Java-compatible mode. If the minimum is
2775        zero, we can continue at the same level without recursion. For any other
2776        minimum, carrying on will result in NOMATCH. */
2777    
2778        if (length == 0) continue;      if (length == 0) continue;
2779        if (length < 0 && min == 0) continue;
2780    
2781        /* First, ensure the minimum number of matches are present. We get back      /* First, ensure the minimum number of matches are present. We get back
2782        the length of the reference string explicitly rather than passing the      the length of the reference string explicitly rather than passing the
2783        address of eptr, so that eptr can be a register variable. */      address of eptr, so that eptr can be a register variable. */
2784    
2785        for (i = 1; i <= min; i++)      for (i = 1; i <= min; i++)
2786          {
2787          int slength;
2788          if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
2789          {          {
2790          if (!match_ref(offset, eptr, length, md, ims)) RRETURN(MATCH_NOMATCH);          if (slength == -2) eptr = md->end_subject;   /* Partial match */
2791          eptr += length;          CHECK_PARTIAL();
2792            RRETURN(MATCH_NOMATCH);
2793          }          }
2794          eptr += slength;
2795          }
2796    
2797        /* If min = max, continue at the same level without recursion.      /* If min = max, continue at the same level without recursion.
2798        They are not both allowed to be zero. */      They are not both allowed to be zero. */
2799    
2800        if (min == max) continue;      if (min == max) continue;
2801    
2802        /* If minimizing, keep trying and advancing the pointer */      /* If minimizing, keep trying and advancing the pointer */
2803    
2804        if (minimize)      if (minimize)
2805          {
2806          for (fi = min;; fi++)
2807          {          {
2808          for (fi = min;; fi++)          int slength;
2809            RMATCH(eptr, ecode, offset_top, md, eptrb, RM14);
2810            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2811            if (fi >= max) RRETURN(MATCH_NOMATCH);
2812            if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
2813            {            {
2814            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);            if (slength == -2) eptr = md->end_subject;   /* Partial match */
2815            if (rrc != MATCH_NOMATCH) RRETURN(rrc);            CHECK_PARTIAL();
2816            if (fi >= max || !match_ref(offset, eptr, length, md, ims))            RRETURN(MATCH_NOMATCH);
             RRETURN(MATCH_NOMATCH);  
           eptr += length;  
2817            }            }
2818          /* Control never gets here */          eptr += slength;
2819          }          }
2820          /* Control never gets here */
2821          }
2822    
2823        /* If maximizing, find the longest string and work backwards */      /* If maximizing, find the longest string and work backwards */
2824    
2825        else      else
2826          {
2827          pp = eptr;
2828          for (i = min; i < max; i++)
2829          {          {
2830          pp = eptr;          int slength;
2831          for (i = min; i < max; i++)          if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
2832            {            {
2833            if (!match_ref(offset, eptr, length, md, ims)) break;            /* Can't use CHECK_PARTIAL because we don't want to update eptr in
2834            eptr += length;            the soft partial matching case. */
2835            }  
2836          while (eptr >= pp)            if (slength == -2 && md->partial != 0 &&
2837            {                md->end_subject > md->start_used_ptr)
2838            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              {
2839            if (rrc != MATCH_NOMATCH) RRETURN(rrc);              md->hitend = TRUE;
2840            eptr -= length;              if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2841                }
2842              break;
2843            }            }
2844          RRETURN(MATCH_NOMATCH);          eptr += slength;
2845            }
2846    
2847          while (eptr >= pp)
2848            {
2849            RMATCH(eptr, ecode, offset_top, md, eptrb, RM15);
2850            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2851            eptr -= length;
2852          }          }
2853          RRETURN(MATCH_NOMATCH);
2854        }        }
2855      /* Control never gets here */      /* Control never gets here */
2856    
   
   
2857      /* Match a bit-mapped character class, possibly repeatedly. This op code is      /* Match a bit-mapped character class, possibly repeatedly. This op code is
2858      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,
2859      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 1616  for (;;) Line 2868  for (;;)
2868      case OP_NCLASS:      case OP_NCLASS:
2869      case OP_CLASS:      case OP_CLASS:
2870        {        {
2871          /* The data variable is saved across frames, so the byte map needs to
2872          be stored there. */
2873    #define BYTE_MAP ((pcre_uint8 *)data)
2874        data = ecode + 1;                /* Save for matching */        data = ecode + 1;                /* Save for matching */
2875        ecode += 33;                     /* Advance past the item */        ecode += 1 + (32 / sizeof(pcre_uchar)); /* Advance past the item */
2876    
2877        switch (*ecode)        switch (*ecode)
2878          {          {
# Line 1638  for (;;) Line 2893  for (;;)
2893          case OP_CRMINRANGE:          case OP_CRMINRANGE:
2894          minimize = (*ecode == OP_CRMINRANGE);          minimize = (*ecode == OP_CRMINRANGE);
2895          min = GET2(ecode, 1);          min = GET2(ecode, 1);
2896          max = GET2(ecode, 3);          max = GET2(ecode, 1 + IMM2_SIZE);
2897          if (max == 0) max = INT_MAX;          if (max == 0) max = INT_MAX;
2898          ecode += 5;          ecode += 1 + 2 * IMM2_SIZE;
2899          break;          break;
2900    
2901          default:               /* No repeat follows */          default:               /* No repeat follows */
# Line 1650  for (;;) Line 2905  for (;;)
2905    
2906        /* First, ensure the minimum number of matches are present. */        /* First, ensure the minimum number of matches are present. */
2907    
2908  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
2909        /* UTF-8 mode */        if (utf)
       if (utf8)  
2910          {          {
2911          for (i = 1; i <= min; i++)          for (i = 1; i <= min; i++)
2912            {            {
2913            if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);            if (eptr >= md->end_subject)
2914                {
2915                SCHECK_PARTIAL();
2916                RRETURN(MATCH_NOMATCH);
2917                }
2918            GETCHARINC(c, eptr);            GETCHARINC(c, eptr);
2919            if (c > 255)            if (c > 255)
2920              {              {
2921              if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);              if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
2922              }              }
2923            else            else
2924              {              if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
             if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);  
             }  
2925            }            }
2926          }          }
2927        else        else
2928  #endif  #endif
2929        /* Not UTF-8 mode */        /* Not UTF mode */
2930          {          {
2931          for (i = 1; i <= min; i++)          for (i = 1; i <= min; i++)
2932            {            {
2933            if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);            if (eptr >= md->end_subject)
2934                {
2935                SCHECK_PARTIAL();
2936                RRETURN(MATCH_NOMATCH);
2937                }
2938            c = *eptr++;            c = *eptr++;
2939            if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);  #ifndef COMPILE_PCRE8
2940              if (c > 255)
2941                {
2942                if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
2943                }
2944              else
2945    #endif
2946                if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
2947            }            }
2948          }          }
2949    
# Line 1690  for (;;) Line 2957  for (;;)
2957    
2958        if (minimize)        if (minimize)
2959          {          {
2960  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
2961          /* UTF-8 mode */          if (utf)
         if (utf8)  
2962            {            {
2963            for (fi = min;; fi++)            for (fi = min;; fi++)
2964              {              {
2965              RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              RMATCH(eptr, ecode, offset_top, md, eptrb, RM16);
2966              if (rrc != MATCH_NOMATCH) RRETURN(rrc);              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2967              if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);              if (fi >= max) RRETURN(MATCH_NOMATCH);
2968                if (eptr >= md->end_subject)
2969                  {
2970                  SCHECK_PARTIAL();
2971                  RRETURN(MATCH_NOMATCH);
2972                  }
2973              GETCHARINC(c, eptr);              GETCHARINC(c, eptr);
2974              if (c > 255)              if (c > 255)
2975                {                {
2976                if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);                if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
2977                }                }
2978              else              else
2979                {                if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
               if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);  
               }  
2980              }              }
2981            }            }
2982          else          else
2983  #endif  #endif
2984          /* Not UTF-8 mode */          /* Not UTF mode */
2985            {            {
2986            for (fi = min;; fi++)            for (fi = min;; fi++)
2987              {              {
2988              RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              RMATCH(eptr, ecode, offset_top, md, eptrb, RM17);
2989              if (rrc != MATCH_NOMATCH) RRETURN(rrc);              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2990              if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);              if (fi >= max) RRETURN(MATCH_NOMATCH);
2991                if (eptr >= md->end_subject)
2992                  {
2993                  SCHECK_PARTIAL();
2994                  RRETURN(MATCH_NOMATCH);
2995                  }
2996              c = *eptr++;              c = *eptr++;
2997              if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);  #ifndef COMPILE_PCRE8
2998                if (c > 255)
2999                  {
3000                  if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
3001                  }
3002                else
3003    #endif
3004                  if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
3005              }              }
3006            }            }
3007          /* Control never gets here */          /* Control never gets here */
# Line 1732  for (;;) Line 3013  for (;;)
3013          {          {
3014          pp = eptr;          pp = eptr;
3015    
3016  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
3017          /* UTF-8 mode */          if (utf)
         if (utf8)  
3018            {            {
3019            for (i = min; i < max; i++)            for (i = min; i < max; i++)
3020              {              {
3021              int len = 1;              int len = 1;
3022              if (eptr >= md->end_subject) break;              if (eptr >= md->end_subject)
3023                  {
3024                  SCHECK_PARTIAL();
3025                  break;
3026                  }
3027              GETCHARLEN(c, eptr, len);              GETCHARLEN(c, eptr, len);
3028              if (c > 255)              if (c > 255)
3029                {                {
3030                if (op == OP_CLASS) break;                if (op == OP_CLASS) break;
3031                }                }
3032              else              else
3033                {                if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) break;
               if ((data[c/8] & (1 << (c&7))) == 0) break;  
               }  
3034              eptr += len;              eptr += len;
3035              }              }
3036            for (;;)            for (;;)
3037              {              {
3038              RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              RMATCH(eptr, ecode, offset_top, md, eptrb, RM18);
3039              if (rrc != MATCH_NOMATCH) RRETURN(rrc);              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3040              if (eptr-- == pp) break;        /* Stop if tried at original pos */              if (eptr-- == pp) break;        /* Stop if tried at original pos */
3041              BACKCHAR(eptr);              BACKCHAR(eptr);
# Line 1761  for (;;) Line 3043  for (;;)
3043            }            }
3044          else          else
3045  #endif  #endif
3046            /* Not UTF-8 mode */            /* Not UTF mode */
3047            {            {
3048            for (i = min; i < max; i++)            for (i = min; i < max; i++)
3049              {              {
3050              if (eptr >= md->end_subject) break;              if (eptr >= md->end_subject)
3051                  {
3052                  SCHECK_PARTIAL();
3053                  break;
3054                  }
3055              c = *eptr;              c = *eptr;
3056              if ((data[c/8] & (1 << (c&7))) == 0) break;  #ifndef COMPILE_PCRE8
3057                if (c > 255)
3058                  {
3059                  if (op == OP_CLASS) break;
3060                  }
3061                else
3062    #endif
3063                  if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) break;
3064              eptr++;              eptr++;
3065              }              }
3066            while (eptr >= pp)            while (eptr >= pp)
3067              {              {
3068              RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              RMATCH(eptr, ecode, offset_top, md, eptrb, RM19);
3069              if (rrc != MATCH_NOMATCH) RRETURN(rrc);              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3070              eptr--;              eptr--;
3071              }              }
# Line 1780  for (;;) Line 3073  for (;;)
3073    
3074          RRETURN(MATCH_NOMATCH);          RRETURN(MATCH_NOMATCH);
3075          }          }
3076    #undef BYTE_MAP
3077        }        }
3078      /* Control never gets here */      /* Control never gets here */
3079    
3080    
3081      /* Match an extended character class. This opcode is encountered only      /* Match an extended character class. This opcode is encountered only
3082      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
3083        mode, because Unicode properties are supported in non-UTF-8 mode. */
3084    
3085  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3086      case OP_XCLASS:      case OP_XCLASS:
3087        {        {
3088        data = ecode + 1 + LINK_SIZE;                /* Save for matching */        data = ecode + 1 + LINK_SIZE;                /* Save for matching */
# Line 1812  for (;;) Line 3107  for (;;)
3107          case OP_CRMINRANGE:          case OP_CRMINRANGE:
3108          minimize = (*ecode == OP_CRMINRANGE);          minimize = (*ecode == OP_CRMINRANGE);
3109          min = GET2(ecode, 1);          min = GET2(ecode, 1);
3110          max = GET2(ecode, 3);          max = GET2(ecode, 1 + IMM2_SIZE);
3111          if (max == 0) max = INT_MAX;          if (max == 0) max = INT_MAX;
3112          ecode += 5;          ecode += 1 + 2 * IMM2_SIZE;
3113          break;          break;
3114    
3115          default:               /* No repeat follows */          default:               /* No repeat follows */
# Line 1826  for (;;) Line 3121  for (;;)
3121    
3122        for (i = 1; i <= min; i++)        for (i = 1; i <= min; i++)
3123          {          {
3124          if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);          if (eptr >= md->end_subject)
3125          GETCHARINC(c, eptr);            {
3126          if (!_pcre_xclass(c, data)) RRETURN(MATCH_NOMATCH);            SCHECK_PARTIAL();
3127              RRETURN(MATCH_NOMATCH);
3128              }
3129            GETCHARINCTEST(c, eptr);
3130            if (!PRIV(xclass)(c, data, utf)) RRETURN(MATCH_NOMATCH);
3131          }          }
3132    
3133        /* 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 1843  for (;;) Line 3142  for (;;)
3142          {          {
3143          for (fi = min;; fi++)          for (fi = min;; fi++)
3144            {            {
3145            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);            RMATCH(eptr, ecode, offset_top, md, eptrb, RM20);
3146            if (rrc != MATCH_NOMATCH) RRETURN(rrc);            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3147            if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);            if (fi >= max) RRETURN(MATCH_NOMATCH);
3148            GETCHARINC(c, eptr);            if (eptr >= md->end_subject)
3149            if (!_pcre_xclass(c, data)) RRETURN(MATCH_NOMATCH);              {
3150                SCHECK_PARTIAL();
3151                RRETURN(MATCH_NOMATCH);
3152                }
3153              GETCHARINCTEST(c, eptr);
3154              if (!PRIV(xclass)(c, data, utf)) RRETURN(MATCH_NOMATCH);
3155            }            }
3156          /* Control never gets here */          /* Control never gets here */
3157          }          }
# Line 1860  for (;;) Line 3164  for (;;)
3164          for (i = min; i < max; i++)          for (i = min; i < max; i++)
3165            {            {
3166            int len = 1;            int len = 1;
3167            if (eptr >= md->end_subject) break;            if (eptr >= md->end_subject)
3168            GETCHARLEN(c, eptr, len);              {
3169            if (!_pcre_xclass(c, data)) break;              SCHECK_PARTIAL();
3170                break;
3171                }
3172    #ifdef SUPPORT_UTF
3173              GETCHARLENTEST(c, eptr, len);
3174    #else
3175              c = *eptr;
3176    #endif
3177              if (!PRIV(xclass)(c, data, utf)) break;
3178            eptr += len;            eptr += len;
3179            }            }
3180          for(;;)          for(;;)
3181            {            {
3182            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);            RMATCH(eptr, ecode, offset_top, md, eptrb, RM21);
3183            if (rrc != MATCH_NOMATCH) RRETURN(rrc);            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3184            if (eptr-- == pp) break;        /* Stop if tried at original pos */            if (eptr-- == pp) break;        /* Stop if tried at original pos */
3185            BACKCHAR(eptr)  #ifdef SUPPORT_UTF
3186              if (utf) BACKCHAR(eptr);
3187    #endif
3188            }            }
3189          RRETURN(MATCH_NOMATCH);          RRETURN(MATCH_NOMATCH);
3190          }          }
# Line 1882  for (;;) Line 3196  for (;;)
3196      /* Match a single character, casefully */      /* Match a single character, casefully */
3197    
3198      case OP_CHAR:      case OP_CHAR:
3199  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
3200      if (utf8)      if (utf)
3201        {        {
3202        length = 1;        length = 1;
3203        ecode++;        ecode++;
3204        GETCHARLEN(fc, ecode, length);        GETCHARLEN(fc, ecode, length);
3205        if (length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);        if (length > md->end_subject - eptr)
3206        while (length-- > 0) if (*ecode++ != *eptr++) RRETURN(MATCH_NOMATCH);          {
3207            CHECK_PARTIAL();             /* Not SCHECK_PARTIAL() */
3208            RRETURN(MATCH_NOMATCH);
3209            }
3210          while (length-- > 0) if (*ecode++ != RAWUCHARINC(eptr)) RRETURN(MATCH_NOMATCH);
3211        }        }
3212      else      else
3213  #endif  #endif
3214        /* Not UTF mode */
     /* Non-UTF-8 mode */  
3215        {        {
3216        if (md->end_subject - eptr < 1) RRETURN(MATCH_NOMATCH);        if (md->end_subject - eptr < 1)
3217            {
3218            SCHECK_PARTIAL();            /* This one can use SCHECK_PARTIAL() */
3219            RRETURN(MATCH_NOMATCH);
3220            }
3221        if (ecode[1] != *eptr++) RRETURN(MATCH_NOMATCH);        if (ecode[1] != *eptr++) RRETURN(MATCH_NOMATCH);
3222        ecode += 2;        ecode += 2;
3223        }        }
3224      break;      break;
3225    
3226      /* Match a single character, caselessly */      /* Match a single character, caselessly. If we are at the end of the
3227        subject, give up immediately. */
3228    
3229        case OP_CHARI:
3230        if (eptr >= md->end_subject)
3231          {
3232          SCHECK_PARTIAL();
3233          RRETURN(MATCH_NOMATCH);
3234          }
3235    
3236      case OP_CHARNC:  #ifdef SUPPORT_UTF
3237  #ifdef SUPPORT_UTF8      if (utf)
     if (utf8)  
3238        {        {
3239        length = 1;        length = 1;
3240        ecode++;        ecode++;
3241        GETCHARLEN(fc, ecode, length);        GETCHARLEN(fc, ecode, length);
3242    
       if (length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);  
   
3243        /* 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
3244        can use the fast lookup table. */        we know that its other case must also be one byte long, so we can use the
3245          fast lookup table. We know that there is at least one byte left in the
3246          subject. */
3247    
3248        if (fc < 128)        if (fc < 128)
3249          {          {
3250          if (md->lcc[*ecode++] != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);          pcre_uint32 cc = RAWUCHAR(eptr);
3251            if (md->lcc[fc] != TABLE_GET(cc, md->lcc, cc)) RRETURN(MATCH_NOMATCH);
3252            ecode++;
3253            eptr++;
3254          }          }
3255    
3256        /* Otherwise we must pick up the subject character */        /* Otherwise we must pick up the subject character. Note that we cannot
3257          use the value of "length" to check for sufficient bytes left, because the
3258          other case of the character may have more or fewer bytes.  */
3259    
3260        else        else
3261          {          {
3262          int dc;          pcre_uint32 dc;
3263          GETCHARINC(dc, eptr);          GETCHARINC(dc, eptr);
3264          ecode += length;          ecode += length;
3265    
# Line 1936  for (;;) Line 3269  for (;;)
3269          if (fc != dc)          if (fc != dc)
3270            {            {
3271  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
3272            if (dc != _pcre_ucp_othercase(fc))            if (dc != UCD_OTHERCASE(fc))
3273  #endif  #endif
3274              RRETURN(MATCH_NOMATCH);              RRETURN(MATCH_NOMATCH);
3275            }            }
3276          }          }
3277        }        }
3278      else      else
3279  #endif   /* SUPPORT_UTF8 */  #endif   /* SUPPORT_UTF */
3280    
3281      /* Non-UTF-8 mode */      /* Not UTF mode */
3282        {        {
3283        if (md->end_subject - eptr < 1) RRETURN(MATCH_NOMATCH);        if (TABLE_GET(ecode[1], md->lcc, ecode[1])
3284        if (md->lcc[ecode[1]] != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);            != TABLE_GET(*eptr, md->lcc, *eptr)) RRETURN(MATCH_NOMATCH);
3285          eptr++;
3286        ecode += 2;        ecode += 2;
3287        }        }
3288      break;      break;
3289    
3290      /* Match a single character repeatedly; different opcodes share code. */      /* Match a single character repeatedly. */
3291    
3292      case OP_EXACT:      case OP_EXACT:
3293        case OP_EXACTI:
3294      min = max = GET2(ecode, 1);      min = max = GET2(ecode, 1);
3295      ecode += 3;      ecode += 1 + IMM2_SIZE;
3296      goto REPEATCHAR;      goto REPEATCHAR;
3297    
3298        case OP_POSUPTO:
3299        case OP_POSUPTOI:
3300        possessive = TRUE;
3301        /* Fall through */
3302    
3303      case OP_UPTO:      case OP_UPTO:
3304        case OP_UPTOI:
3305      case OP_MINUPTO:      case OP_MINUPTO:
3306        case OP_MINUPTOI:
3307      min = 0;      min = 0;
3308      max = GET2(ecode, 1);      max = GET2(ecode, 1);
3309      minimize = *ecode == OP_MINUPTO;      minimize = *ecode == OP_MINUPTO || *ecode == OP_MINUPTOI;
3310      ecode += 3;      ecode += 1 + IMM2_SIZE;
3311        goto REPEATCHAR;
3312    
3313        case OP_POSSTAR:
3314        case OP_POSSTARI:
3315        possessive = TRUE;
3316        min = 0;
3317        max = INT_MAX;
3318        ecode++;
3319        goto REPEATCHAR;
3320    
3321        case OP_POSPLUS:
3322        case OP_POSPLUSI:
3323        possessive = TRUE;
3324        min = 1;
3325        max = INT_MAX;
3326        ecode++;
3327        goto REPEATCHAR;
3328    
3329        case OP_POSQUERY: