/[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 1274 by ph10, Fri Mar 8 11:35:41 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        /* COMMIT overrides PRUNE, SKIP, and THEN */
785    
786        case OP_COMMIT:
787        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
788          eptrb, RM52);
789        if (rrc != MATCH_NOMATCH && rrc != MATCH_PRUNE &&
790            rrc != MATCH_SKIP && rrc != MATCH_SKIP_ARG &&
791            rrc != MATCH_THEN)
792          RRETURN(rrc);
793        RRETURN(MATCH_COMMIT);
794    
795        /* PRUNE overrides THEN */
796    
797        case OP_PRUNE:
798        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
799          eptrb, RM51);
800        if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
801        RRETURN(MATCH_PRUNE);
802    
803        case OP_PRUNE_ARG:
804        md->nomatch_mark = ecode + 2;
805        md->mark = NULL;    /* In case previously set by assertion */
806        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
807          eptrb, RM56);
808        if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
809             md->mark == NULL) md->mark = ecode + 2;
810        if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
811        RRETURN(MATCH_PRUNE);
812    
813        /* SKIP overrides PRUNE and THEN */
814    
815        case OP_SKIP:
816        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
817          eptrb, RM53);
818        if (rrc != MATCH_NOMATCH && rrc != MATCH_PRUNE && rrc != MATCH_THEN)
819          RRETURN(rrc);
820        md->start_match_ptr = eptr;   /* Pass back current position */
821        RRETURN(MATCH_SKIP);
822    
823        /* Note that, for Perl compatibility, SKIP with an argument does NOT set
824        nomatch_mark. When a pattern match ends with a SKIP_ARG for which there was
825        not a matching mark, we have to re-run the match, ignoring the SKIP_ARG
826        that failed and any that preceed it (either they also failed, or were not
827        triggered). To do this, we maintain a count of executed SKIP_ARGs. If a
828        SKIP_ARG gets to top level, the match is re-run with md->ignore_skip_arg
829        set to the count of the one that failed. */
830    
831        case OP_SKIP_ARG:
832        md->skip_arg_count++;
833        if (md->skip_arg_count <= md->ignore_skip_arg)
834          {
835          ecode += PRIV(OP_lengths)[*ecode] + ecode[1];
836          break;
837          }
838        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
839          eptrb, RM57);
840        if (rrc != MATCH_NOMATCH && rrc != MATCH_PRUNE && rrc != MATCH_THEN)
841          RRETURN(rrc);
842    
843        /* Pass back the current skip name by overloading md->start_match_ptr and
844        returning the special MATCH_SKIP_ARG return code. This will either be
845        caught by a matching MARK, or get to the top, where it causes a rematch
846        with md->ignore_skip_arg set to the value of md->skip_arg_count. */
847    
848        md->start_match_ptr = ecode + 2;
849        RRETURN(MATCH_SKIP_ARG);
850    
851        /* For THEN (and THEN_ARG) we pass back the address of the opcode, so that
852        the branch in which it occurs can be determined. Overload the start of
853        match pointer to do this. */
854    
855        case OP_THEN:
856        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
857          eptrb, RM54);
858        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
859        md->start_match_ptr = ecode;
860        RRETURN(MATCH_THEN);
861    
862        case OP_THEN_ARG:
863        md->nomatch_mark = ecode + 2;
864        md->mark = NULL;    /* In case previously set by assertion */
865        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top,
866          md, eptrb, RM58);
867        if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
868             md->mark == NULL) md->mark = ecode + 2;
869        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
870        md->start_match_ptr = ecode;
871        RRETURN(MATCH_THEN);
872    
873        /* Handle an atomic group that does not contain any capturing parentheses.
874        This can be handled like an assertion. Prior to 8.13, all atomic groups
875        were handled this way. In 8.13, the code was changed as below for ONCE, so
876        that backups pass through the group and thereby reset captured values.
877        However, this uses a lot more stack, so in 8.20, atomic groups that do not
878        contain any captures generate OP_ONCE_NC, which can be handled in the old,
879        less stack intensive way.
880    
881    if (md->partial &&      Check the alternative branches in turn - the matching won't pass the KET
882        eptr >= md->end_subject &&      for this kind of subpattern. If any one branch matches, we carry on as at
883        eptr > md->start_match)      the end of a normal bracket, leaving the subject pointer, but resetting
884      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. */  
885    
886    if (op > OP_BRA)      case OP_ONCE_NC:
887      {      prev = ecode;
888      number = op - OP_BRA;      saved_eptr = eptr;
889        save_mark = md->mark;
890        do
891          {
892          RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM64);
893          if (rrc == MATCH_MATCH)  /* Note: _not_ MATCH_ACCEPT */
894            {
895            mstart = md->start_match_ptr;
896            break;
897            }
898          if (rrc == MATCH_THEN)
899            {
900            next = ecode + GET(ecode,1);
901            if (md->start_match_ptr < next &&
902                (*ecode == OP_ALT || *next == OP_ALT))
903              rrc = MATCH_NOMATCH;
904            }
905    
906          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
907          ecode += GET(ecode,1);
908          md->mark = save_mark;
909          }
910        while (*ecode == OP_ALT);
911    
912        /* If hit the end of the group (which could be repeated), fail */
913    
914        if (*ecode != OP_ONCE_NC && *ecode != OP_ALT) RRETURN(MATCH_NOMATCH);
915    
916      /* For extended extraction brackets (large number), we have to fish out the      /* Continue as from after the group, updating the offsets high water
917      number from a dummy opcode at the start. */      mark, since extracts may have been taken. */
918    
919        do ecode += GET(ecode, 1); while (*ecode == OP_ALT);
920    
921        offset_top = md->end_offset_top;
922        eptr = md->end_match_ptr;
923    
924        /* For a non-repeating ket, just continue at this level. This also
925        happens for a repeating ket if no characters were matched in the group.
926        This is the forcible breaking of infinite loops as implemented in Perl
927        5.005. */
928    
929        if (*ecode == OP_KET || eptr == saved_eptr)
930          {
931          ecode += 1+LINK_SIZE;
932          break;
933          }
934    
935        /* The repeating kets try the rest of the pattern or restart from the
936        preceding bracket, in the appropriate order. The second "call" of match()
937        uses tail recursion, to avoid using another stack frame. */
938    
939        if (*ecode == OP_KETRMIN)
940          {
941          RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM65);
942          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
943          ecode = prev;
944          goto TAIL_RECURSE;
945          }
946        else  /* OP_KETRMAX */
947          {
948          RMATCH(eptr, prev, offset_top, md, eptrb, RM66);
949          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
950          ecode += 1 + LINK_SIZE;
951          goto TAIL_RECURSE;
952          }
953        /* Control never gets here */
954    
955      if (number > EXTRACT_BASIC_MAX)      /* Handle a capturing bracket, other than those that are possessive with an
956        number = GET2(ecode, 2+LINK_SIZE);      unlimited repeat. If there is space in the offset vector, save the current
957        subject position in the working slot at the top of the vector. We mustn't
958        change the current values of the data slot, because they may be set from a
959        previous iteration of this group, and be referred to by a reference inside
960        the group. A failure to match might occur after the group has succeeded,
961        if something later on doesn't match. For this reason, we need to restore
962        the working value and also the values of the final offsets, in case they
963        were set by a previous iteration of the same bracket.
964    
965        If there isn't enough space in the offset vector, treat this as if it were
966        a non-capturing bracket. Don't worry about setting the flag for the error
967        case here; that is handled in the code for KET. */
968    
969        case OP_CBRA:
970        case OP_SCBRA:
971        number = GET2(ecode, 1+LINK_SIZE);
972      offset = number << 1;      offset = number << 1;
973    
974  #ifdef DEBUG  #ifdef PCRE_DEBUG
975      printf("start bracket %d subject=", number);      printf("start bracket %d\n", number);
976        printf("subject=");
977      pchars(eptr, 16, TRUE, md);      pchars(eptr, 16, TRUE, md);
978      printf("\n");      printf("\n");
979  #endif  #endif
# Line 620  for (;;) Line 984  for (;;)
984        save_offset2 = md->offset_vector[offset+1];        save_offset2 = md->offset_vector[offset+1];
985        save_offset3 = md->offset_vector[md->offset_end - number];        save_offset3 = md->offset_vector[md->offset_end - number];
986        save_capture_last = md->capture_last;        save_capture_last = md->capture_last;
987          save_mark = md->mark;
988    
989        DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));        DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
990        md->offset_vector[md->offset_end - number] = eptr - md->start_subject;        md->offset_vector[md->offset_end - number] =
991            (int)(eptr - md->start_subject);
992    
993        do        for (;;)
994          {          {
995          RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb,          if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
996            match_isgroup);          RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
997              eptrb, RM1);
998            if (rrc == MATCH_ONCE) break;  /* Backing up through an atomic group */
999    
1000            /* If we backed up to a THEN, check whether it is within the current
1001            branch by comparing the address of the THEN that is passed back with
1002            the end of the branch. If it is within the current branch, and the
1003            branch is one of two or more alternatives (it either starts or ends
1004            with OP_ALT), we have reached the limit of THEN's action, so convert
1005            the return code to NOMATCH, which will cause normal backtracking to
1006            happen from now on. Otherwise, THEN is passed back to an outer
1007            alternative. This implements Perl's treatment of parenthesized groups,
1008            where a group not containing | does not affect the current alternative,
1009            that is, (X) is NOT the same as (X|(*F)). */
1010    
1011            if (rrc == MATCH_THEN)
1012              {
1013              next = ecode + GET(ecode,1);
1014              if (md->start_match_ptr < next &&
1015                  (*ecode == OP_ALT || *next == OP_ALT))
1016                rrc = MATCH_NOMATCH;
1017              }
1018    
1019            /* Anything other than NOMATCH is passed back. */
1020    
1021          if (rrc != MATCH_NOMATCH) RRETURN(rrc);          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1022          md->capture_last = save_capture_last;          md->capture_last = save_capture_last;
1023          ecode += GET(ecode, 1);          ecode += GET(ecode, 1);
1024            md->mark = save_mark;
1025            if (*ecode != OP_ALT) break;
1026          }          }
       while (*ecode == OP_ALT);  
1027    
1028        DPRINTF(("bracket %d failed\n", number));        DPRINTF(("bracket %d failed\n", number));
   
1029        md->offset_vector[offset] = save_offset1;        md->offset_vector[offset] = save_offset1;
1030        md->offset_vector[offset+1] = save_offset2;        md->offset_vector[offset+1] = save_offset2;
1031        md->offset_vector[md->offset_end - number] = save_offset3;        md->offset_vector[md->offset_end - number] = save_offset3;
1032    
1033        RRETURN(MATCH_NOMATCH);        /* At this point, rrc will be one of MATCH_ONCE or MATCH_NOMATCH. */
1034    
1035          RRETURN(rrc);
1036        }        }
1037    
1038      /* Insufficient room for saving captured contents */      /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
1039        as a non-capturing bracket. */
1040    
1041      else op = OP_BRA;      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1042      }      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1043    
1044    /* Other types of node can be handled by a switch */      DPRINTF(("insufficient capture room: treat as non-capturing\n"));
1045    
1046    switch(op)      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1047      {      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1048      case OP_BRA:     /* Non-capturing bracket: optimized */  
1049      DPRINTF(("start bracket 0\n"));      /* Non-capturing or atomic group, except for possessive with unlimited
1050        repeat and ONCE group with no captures. Loop for all the alternatives.
1051    
1052        When we get to the final alternative within the brackets, we used to return
1053        the result of a recursive call to match() whatever happened so it was
1054        possible to reduce stack usage by turning this into a tail recursion,
1055        except in the case of a possibly empty group. However, now that there is
1056        the possiblity of (*THEN) occurring in the final alternative, this
1057        optimization is no longer always possible.
1058    
1059      /* Loop for all the alternatives */      We can optimize if we know there are no (*THEN)s in the pattern; at present
1060        this is the best that can be done.
1061    
1062        MATCH_ONCE is returned when the end of an atomic group is successfully
1063        reached, but subsequent matching fails. It passes back up the tree (causing
1064        captured values to be reset) until the original atomic group level is
1065        reached. This is tested by comparing md->once_target with the start of the
1066        group. At this point, the return is converted into MATCH_NOMATCH so that
1067        previous backup points can be taken. */
1068    
1069        case OP_ONCE:
1070        case OP_BRA:
1071        case OP_SBRA:
1072        DPRINTF(("start non-capturing bracket\n"));
1073    
1074      for (;;)      for (;;)
1075        {        {
1076        /* When we get to the final alternative within the brackets, we would        if (op >= OP_SBRA || op == OP_ONCE)
1077        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;  
        }  
1078    
1079        /* 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
1080        otherwise return. */        the pattern, and this is the final alternative, optimize as described
1081          above. */
1082    
1083        RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb,        else if (!md->hasthen && ecode[GET(ecode, 1)] != OP_ALT)
1084          match_isgroup);          {
1085        if (rrc != MATCH_NOMATCH) RRETURN(rrc);          ecode += PRIV(OP_lengths)[*ecode];
1086        ecode += GET(ecode, 1);          goto TAIL_RECURSE;
1087        }          }
     /* Control never reaches here. */  
1088    
1089      /* 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. */  
1090    
1091      case OP_COND:        save_mark = md->mark;
1092      if (ecode[LINK_SIZE+1] == OP_CREF) /* Condition extract or recurse test */        save_capture_last = md->capture_last;
1093        {        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md, eptrb,
1094        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;  
       }  
1095    
1096      /* The condition is an assertion. Call match() to evaluate it - setting        /* See comment in the code for capturing groups above about handling
1097      the final argument TRUE causes it to stop at the end of an assertion. */        THEN. */
1098    
1099      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)  
1100          {          {
1101          ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE+2);          next = ecode + GET(ecode,1);
1102          while (*ecode == OP_ALT) ecode += GET(ecode, 1);          if (md->start_match_ptr < next &&
1103                (*ecode == OP_ALT || *next == OP_ALT))
1104              rrc = MATCH_NOMATCH;
1105          }          }
1106        else if (rrc != MATCH_NOMATCH)  
1107          if (rrc != MATCH_NOMATCH)
1108          {          {
1109          RRETURN(rrc);         /* Need braces because of following else */          if (rrc == MATCH_ONCE)
1110              {
1111              const pcre_uchar *scode = ecode;
1112              if (*scode != OP_ONCE)           /* If not at start, find it */
1113                {
1114                while (*scode == OP_ALT) scode += GET(scode, 1);
1115                scode -= GET(scode, 1);
1116                }
1117              if (md->once_target == scode) rrc = MATCH_NOMATCH;
1118              }
1119            RRETURN(rrc);
1120          }          }
1121        else ecode += GET(ecode, 1);        ecode += GET(ecode, 1);
1122          md->mark = save_mark;
1123          if (*ecode != OP_ALT) break;
1124          md->capture_last = save_capture_last;
1125          }
1126    
1127        RRETURN(MATCH_NOMATCH);
1128    
1129        /* Handle possessive capturing brackets with an unlimited repeat. We come
1130        here from BRAZERO with allow_zero set TRUE. The offset_vector values are
1131        handled similarly to the normal case above. However, the matching is
1132        different. The end of these brackets will always be OP_KETRPOS, which
1133        returns MATCH_KETRPOS without going further in the pattern. By this means
1134        we can handle the group by iteration rather than recursion, thereby
1135        reducing the amount of stack needed. */
1136    
1137        case OP_CBRAPOS:
1138        case OP_SCBRAPOS:
1139        allow_zero = FALSE;
1140    
1141        /* We are now at the branch that is to be obeyed. As there is only one,      POSSESSIVE_CAPTURE:
1142        we can use tail recursion to avoid using another stack frame. */      number = GET2(ecode, 1+LINK_SIZE);
1143        offset = number << 1;
1144    
1145        ecode += 1 + LINK_SIZE;  #ifdef PCRE_DEBUG
1146        flags = match_isgroup;      printf("start possessive bracket %d\n", number);
1147        goto TAIL_RECURSE;      printf("subject=");
1148        }      pchars(eptr, 16, TRUE, md);
1149      /* Control never reaches here */      printf("\n");
1150    #endif
1151    
1152      /* Skip over conditional reference or large extraction number data if      if (offset < md->offset_max)
1153      encountered. */        {
1154          matched_once = FALSE;
1155          code_offset = (int)(ecode - md->start_code);
1156    
1157      case OP_CREF:        save_offset1 = md->offset_vector[offset];
1158      case OP_BRANUMBER:        save_offset2 = md->offset_vector[offset+1];
1159      ecode += 3;        save_offset3 = md->offset_vector[md->offset_end - number];
1160      break;        save_capture_last = md->capture_last;
1161    
1162      /* 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. */  
1163    
1164      case OP_END:        /* Each time round the loop, save the current subject position for use
1165      if (md->recursive != NULL && md->recursive->group_num == 0)        when the group matches. For MATCH_MATCH, the group has matched, so we
1166        {        restart it with a new subject starting position, remembering that we had
1167        recursion_info *rec = md->recursive;        at least one match. For MATCH_NOMATCH, carry on with the alternatives, as
1168        DPRINTF(("End of pattern in a (?0) recursion\n"));        usual. If we haven't matched any alternatives in any iteration, check to
1169        md->recursive = rec->prevrec;        see if a previous iteration matched. If so, the group has matched;
1170        memmove(md->offset_vector, rec->offset_save,        continue from afterwards. Otherwise it has failed; restore the previous
1171          rec->saved_max * sizeof(int));        capture values before returning NOMATCH. */
1172        md->start_match = rec->save_start;  
1173        ims = original_ims;        for (;;)
1174        ecode = rec->after_call;          {
1175        break;          md->offset_vector[md->offset_end - number] =
1176        }            (int)(eptr - md->start_subject);
1177            if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
1178            RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
1179              eptrb, RM63);
1180            if (rrc == MATCH_KETRPOS)
1181              {
1182              offset_top = md->end_offset_top;
1183              eptr = md->end_match_ptr;
1184              ecode = md->start_code + code_offset;
1185              save_capture_last = md->capture_last;
1186              matched_once = TRUE;
1187              continue;
1188              }
1189    
1190            /* See comment in the code for capturing groups above about handling
1191            THEN. */
1192    
1193            if (rrc == MATCH_THEN)
1194              {
1195              next = ecode + GET(ecode,1);
1196              if (md->start_match_ptr < next &&
1197                  (*ecode == OP_ALT || *next == OP_ALT))
1198                rrc = MATCH_NOMATCH;
1199              }
1200    
1201      /* Otherwise, if PCRE_NOTEMPTY is set, fail if we have matched an empty          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1202      string - backtracking will then try other alternatives, if any. */          md->capture_last = save_capture_last;
1203            ecode += GET(ecode, 1);
1204            if (*ecode != OP_ALT) break;
1205            }
1206    
1207      if (md->notempty && eptr == md->start_match) RRETURN(MATCH_NOMATCH);        if (!matched_once)
1208      md->end_match_ptr = eptr;          /* Record where we ended */          {
1209      md->end_offset_top = offset_top;   /* and how many extracts were taken */          md->offset_vector[offset] = save_offset1;
1210      RRETURN(MATCH_MATCH);          md->offset_vector[offset+1] = save_offset2;
1211            md->offset_vector[md->offset_end - number] = save_offset3;
1212      /* Change option settings */          }
   
     case OP_OPT:  
     ims = ecode[1];  
     ecode += 2;  
     DPRINTF(("ims set to %02lx\n", ims));  
     break;  
1213    
1214      /* Assertion brackets. Check the alternative branches in turn - the        if (allow_zero || matched_once)
1215      matching won't pass the KET for an assertion. If any one branch matches,          {
1216      the assertion is true. Lookbehind assertions have an OP_REVERSE item at the          ecode += 1 + LINK_SIZE;
1217      start of each branch to move the current point backwards, so the code at          break;
1218      this level is identical to the lookahead case. */          }
1219    
1220      case OP_ASSERT:        RRETURN(MATCH_NOMATCH);
     case OP_ASSERTBACK:  
     do  
       {  
       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,  
         match_isgroup);  
       if (rrc == MATCH_MATCH) break;  
       if (rrc != MATCH_NOMATCH) RRETURN(rrc);  
       ecode += GET(ecode, 1);  
1221        }        }
     while (*ecode == OP_ALT);  
     if (*ecode == OP_KET) RRETURN(MATCH_NOMATCH);  
   
     /* If checking an assertion for a condition, return MATCH_MATCH. */  
1222    
1223      if ((flags & match_condassert) != 0) RRETURN(MATCH_MATCH);      /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
1224        as a non-capturing bracket. */
1225    
1226      /* Continue from after the assertion, updating the offsets high water      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1227      mark, since extracts may have been taken during the assertion. */      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1228    
1229      do ecode += GET(ecode,1); while (*ecode == OP_ALT);      DPRINTF(("insufficient capture room: treat as non-capturing\n"));
     ecode += 1 + LINK_SIZE;  
     offset_top = md->end_offset_top;  
     continue;  
1230    
1231      /* Negative assertion: all branches must fail to match */      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1232        /* VVVVVVVVVVVVVVVVVVVVVVVVV */
1233    
1234        /* Non-capturing possessive bracket with unlimited repeat. We come here
1235        from BRAZERO with allow_zero = TRUE. The code is similar to the above,
1236        without the capturing complication. It is written out separately for speed
1237        and cleanliness. */
1238    
1239        case OP_BRAPOS:
1240        case OP_SBRAPOS:
1241        allow_zero = FALSE;
1242    
1243        POSSESSIVE_NON_CAPTURE:
1244        matched_once = FALSE;
1245        code_offset = (int)(ecode - md->start_code);
1246        save_capture_last = md->capture_last;
1247    
1248      case OP_ASSERT_NOT:      for (;;)
     case OP_ASSERTBACK_NOT:  
     do  
1249        {        {
1250        RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,        if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
1251          match_isgroup);        RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
1252        if (rrc == MATCH_MATCH) RRETURN(MATCH_NOMATCH);          eptrb, RM48);
1253          if (rrc == MATCH_KETRPOS)
1254            {
1255            offset_top = md->end_offset_top;
1256            eptr = md->end_match_ptr;
1257            ecode = md->start_code + code_offset;
1258            matched_once = TRUE;
1259            continue;
1260            }
1261    
1262          /* See comment in the code for capturing groups above about handling
1263          THEN. */
1264    
1265          if (rrc == MATCH_THEN)
1266            {
1267            next = ecode + GET(ecode,1);
1268            if (md->start_match_ptr < next &&
1269                (*ecode == OP_ALT || *next == OP_ALT))
1270              rrc = MATCH_NOMATCH;
1271            }
1272    
1273        if (rrc != MATCH_NOMATCH) RRETURN(rrc);        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1274        ecode += GET(ecode,1);        ecode += GET(ecode, 1);
1275          if (*ecode != OP_ALT) break;
1276          md->capture_last = save_capture_last;
1277        }        }
     while (*ecode == OP_ALT);  
1278    
1279      if ((flags & match_condassert) != 0) RRETURN(MATCH_MATCH);      if (matched_once || allow_zero)
1280          {
1281          ecode += 1 + LINK_SIZE;
1282          break;
1283          }
1284        RRETURN(MATCH_NOMATCH);
1285    
1286      ecode += 1 + LINK_SIZE;      /* Control never reaches here. */
     continue;  
1287    
1288      /* Move the subject pointer back. This occurs only at the start of      /* Conditional group: compilation checked that there are no more than
1289      each branch of a lookbehind assertion. If we are too close to the start to      two branches. If the condition is false, skipping the first branch takes us
1290      move back, this match function fails. When working with UTF-8 we move      past the end if there is only one branch, but that's OK because that is
1291      back a number of characters, not bytes. */      exactly what going to the ket would do. */
1292    
1293      case OP_REVERSE:      case OP_COND:
1294  #ifdef SUPPORT_UTF8      case OP_SCOND:
1295      if (utf8)      codelink = GET(ecode, 1);
1296    
1297        /* Because of the way auto-callout works during compile, a callout item is
1298        inserted between OP_COND and an assertion condition. */
1299    
1300        if (ecode[LINK_SIZE+1] == OP_CALLOUT)
1301        {        {
1302        c = GET(ecode,1);        if (PUBL(callout) != NULL)
1303        for (i = 0; i < c; i++)          {
1304          {          PUBL(callout_block) cb;
1305          eptr--;          cb.version          = 2;   /* Version 1 of the callout block */
1306          if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);          cb.callout_number   = ecode[LINK_SIZE+2];
1307          BACKCHAR(eptr)          cb.offset_vector    = md->offset_vector;
1308          }  #if defined COMPILE_PCRE8
1309        }          cb.subject          = (PCRE_SPTR)md->start_subject;
1310      else  #elif defined COMPILE_PCRE16
1311  #endif          cb.subject          = (PCRE_SPTR16)md->start_subject;
1312    #elif defined COMPILE_PCRE32
1313            cb.subject          = (PCRE_SPTR32)md->start_subject;
1314    #endif
1315            cb.subject_length   = (int)(md->end_subject - md->start_subject);
1316            cb.start_match      = (int)(mstart - md->start_subject);
1317            cb.current_position = (int)(eptr - md->start_subject);
1318            cb.pattern_position = GET(ecode, LINK_SIZE + 3);
1319            cb.next_item_length = GET(ecode, 3 + 2*LINK_SIZE);
1320            cb.capture_top      = offset_top/2;
1321            cb.capture_last     = md->capture_last & CAPLMASK;
1322            /* Internal change requires this for API compatibility. */
1323            if (cb.capture_last == 0) cb.capture_last = -1;
1324            cb.callout_data     = md->callout_data;
1325            cb.mark             = md->nomatch_mark;
1326            if ((rrc = (*PUBL(callout))(&cb)) > 0) RRETURN(MATCH_NOMATCH);
1327            if (rrc < 0) RRETURN(rrc);
1328            }
1329          ecode += PRIV(OP_lengths)[OP_CALLOUT];
1330          codelink -= PRIV(OP_lengths)[OP_CALLOUT];
1331          }
1332    
1333        condcode = ecode[LINK_SIZE+1];
1334    
1335      /* No UTF-8 support, or not in UTF-8 mode: count is byte count */      /* Now see what the actual condition is */
1336    
1337        if (condcode == OP_RREF || condcode == OP_NRREF)    /* Recursion test */
1338        {        {
1339        eptr -= GET(ecode,1);        if (md->recursive == NULL)                /* Not recursing => FALSE */
1340            {
1341            condition = FALSE;
1342            ecode += GET(ecode, 1);
1343            }
1344          else
1345            {
1346            unsigned int recno = GET2(ecode, LINK_SIZE + 2);   /* Recursion group number*/
1347            condition = (recno == RREF_ANY || recno == md->recursive->group_num);
1348    
1349            /* If the test is for recursion into a specific subpattern, and it is
1350            false, but the test was set up by name, scan the table to see if the
1351            name refers to any other numbers, and test them. The condition is true
1352            if any one is set. */
1353    
1354            if (!condition && condcode == OP_NRREF)
1355              {
1356              pcre_uchar *slotA = md->name_table;
1357              for (i = 0; i < md->name_count; i++)
1358                {
1359                if (GET2(slotA, 0) == recno) break;
1360                slotA += md->name_entry_size;
1361                }
1362    
1363              /* Found a name for the number - there can be only one; duplicate
1364              names for different numbers are allowed, but not vice versa. First
1365              scan down for duplicates. */
1366    
1367              if (i < md->name_count)
1368                {
1369                pcre_uchar *slotB = slotA;
1370                while (slotB > md->name_table)
1371                  {
1372                  slotB -= md->name_entry_size;
1373                  if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
1374                    {
1375                    condition = GET2(slotB, 0) == md->recursive->group_num;
1376                    if (condition) break;
1377                    }
1378                  else break;
1379                  }
1380    
1381                /* Scan up for duplicates */
1382    
1383                if (!condition)
1384                  {
1385                  slotB = slotA;
1386                  for (i++; i < md->name_count; i++)
1387                    {
1388                    slotB += md->name_entry_size;
1389                    if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
1390                      {
1391                      condition = GET2(slotB, 0) == md->recursive->group_num;
1392                      if (condition) break;
1393                      }
1394                    else break;
1395                    }
1396                  }
1397                }
1398              }
1399    
1400            /* Chose branch according to the condition */
1401    
1402            ecode += condition? 1 + IMM2_SIZE : GET(ecode, 1);
1403            }
1404          }
1405    
1406        else if (condcode == OP_CREF || condcode == OP_NCREF)  /* Group used test */
1407          {
1408          offset = GET2(ecode, LINK_SIZE+2) << 1;  /* Doubled ref number */
1409          condition = offset < offset_top && md->offset_vector[offset] >= 0;
1410    
1411          /* If the numbered capture is unset, but the reference was by name,
1412          scan the table to see if the name refers to any other numbers, and test
1413          them. The condition is true if any one is set. This is tediously similar
1414          to the code above, but not close enough to try to amalgamate. */
1415    
1416          if (!condition && condcode == OP_NCREF)
1417            {
1418            unsigned int refno = offset >> 1;
1419            pcre_uchar *slotA = md->name_table;
1420    
1421            for (i = 0; i < md->name_count; i++)
1422              {
1423              if (GET2(slotA, 0) == refno) break;
1424              slotA += md->name_entry_size;
1425              }
1426    
1427            /* Found a name for the number - there can be only one; duplicate names
1428            for different numbers are allowed, but not vice versa. First scan down
1429            for duplicates. */
1430    
1431            if (i < md->name_count)
1432              {
1433              pcre_uchar *slotB = slotA;
1434              while (slotB > md->name_table)
1435                {
1436                slotB -= md->name_entry_size;
1437                if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
1438                  {
1439                  offset = GET2(slotB, 0) << 1;
1440                  condition = offset < offset_top &&
1441                    md->offset_vector[offset] >= 0;
1442                  if (condition) break;
1443                  }
1444                else break;
1445                }
1446    
1447              /* Scan up for duplicates */
1448    
1449              if (!condition)
1450                {
1451                slotB = slotA;
1452                for (i++; i < md->name_count; i++)
1453                  {
1454                  slotB += md->name_entry_size;
1455                  if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
1456                    {
1457                    offset = GET2(slotB, 0) << 1;
1458                    condition = offset < offset_top &&
1459                      md->offset_vector[offset] >= 0;
1460                    if (condition) break;
1461                    }
1462                  else break;
1463                  }
1464                }
1465              }
1466            }
1467    
1468          /* Chose branch according to the condition */
1469    
1470          ecode += condition? 1 + IMM2_SIZE : GET(ecode, 1);
1471          }
1472    
1473        else if (condcode == OP_DEF)     /* DEFINE - always false */
1474          {
1475          condition = FALSE;
1476          ecode += GET(ecode, 1);
1477          }
1478    
1479        /* The condition is an assertion. Call match() to evaluate it - setting
1480        md->match_function_type to MATCH_CONDASSERT causes it to stop at the end of
1481        an assertion. */
1482    
1483        else
1484          {
1485          md->match_function_type = MATCH_CONDASSERT;
1486          RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM3);
1487          if (rrc == MATCH_MATCH)
1488            {
1489            if (md->end_offset_top > offset_top)
1490              offset_top = md->end_offset_top;  /* Captures may have happened */
1491            condition = TRUE;
1492            ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE + 2);
1493            while (*ecode == OP_ALT) ecode += GET(ecode, 1);
1494            }
1495    
1496          /* PCRE doesn't allow the effect of (*THEN) to escape beyond an
1497          assertion; it is therefore treated as NOMATCH. */
1498    
1499          else if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN)
1500            {
1501            RRETURN(rrc);         /* Need braces because of following else */
1502            }
1503          else
1504            {
1505            condition = FALSE;
1506            ecode += codelink;
1507            }
1508          }
1509    
1510        /* We are now at the branch that is to be obeyed. As there is only one, can
1511        use tail recursion to avoid using another stack frame, except when there is
1512        unlimited repeat of a possibly empty group. In the latter case, a recursive
1513        call to match() is always required, unless the second alternative doesn't
1514        exist, in which case we can just plough on. Note that, for compatibility
1515        with Perl, the | in a conditional group is NOT treated as creating two
1516        alternatives. If a THEN is encountered in the branch, it propagates out to
1517        the enclosing alternative (unless nested in a deeper set of alternatives,
1518        of course). */
1519    
1520        if (condition || *ecode == OP_ALT)
1521          {
1522          if (op != OP_SCOND)
1523            {
1524            ecode += 1 + LINK_SIZE;
1525            goto TAIL_RECURSE;
1526            }
1527    
1528          md->match_function_type = MATCH_CBEGROUP;
1529          RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM49);
1530          RRETURN(rrc);
1531          }
1532    
1533         /* Condition false & no alternative; continue after the group. */
1534    
1535        else
1536          {
1537          ecode += 1 + LINK_SIZE;
1538          }
1539        break;
1540    
1541    
1542        /* Before OP_ACCEPT there may be any number of OP_CLOSE opcodes,
1543        to close any currently open capturing brackets. */
1544    
1545        case OP_CLOSE:
1546        number = GET2(ecode, 1);   /* Must be less than 65536 */
1547        offset = number << 1;
1548    
1549    #ifdef PCRE_DEBUG
1550          printf("end bracket %d at *ACCEPT", number);
1551          printf("\n");
1552    #endif
1553    
1554        md->capture_last = (md->capture_last & OVFLMASK) | number;
1555        if (offset >= md->offset_max) md->capture_last |= OVFLBIT; else
1556          {
1557          md->offset_vector[offset] =
1558            md->offset_vector[md->offset_end - number];
1559          md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
1560          if (offset_top <= offset) offset_top = offset + 2;
1561          }
1562        ecode += 1 + IMM2_SIZE;
1563        break;
1564    
1565    
1566        /* End of the pattern, either real or forced. */
1567    
1568        case OP_END:
1569        case OP_ACCEPT:
1570        case OP_ASSERT_ACCEPT:
1571    
1572        /* If we have matched an empty string, fail if not in an assertion and not
1573        in a recursion if either PCRE_NOTEMPTY is set, or if PCRE_NOTEMPTY_ATSTART
1574        is set and we have matched at the start of the subject. In both cases,
1575        backtracking will then try other alternatives, if any. */
1576    
1577        if (eptr == mstart && op != OP_ASSERT_ACCEPT &&
1578             md->recursive == NULL &&
1579             (md->notempty ||
1580               (md->notempty_atstart &&
1581                 mstart == md->start_subject + md->start_offset)))
1582          RRETURN(MATCH_NOMATCH);
1583    
1584        /* Otherwise, we have a match. */
1585    
1586        md->end_match_ptr = eptr;           /* Record where we ended */
1587        md->end_offset_top = offset_top;    /* and how many extracts were taken */
1588        md->start_match_ptr = mstart;       /* and the start (\K can modify) */
1589    
1590        /* For some reason, the macros don't work properly if an expression is
1591        given as the argument to RRETURN when the heap is in use. */
1592    
1593        rrc = (op == OP_END)? MATCH_MATCH : MATCH_ACCEPT;
1594        RRETURN(rrc);
1595    
1596        /* Assertion brackets. Check the alternative branches in turn - the
1597        matching won't pass the KET for an assertion. If any one branch matches,
1598        the assertion is true. Lookbehind assertions have an OP_REVERSE item at the
1599        start of each branch to move the current point backwards, so the code at
1600        this level is identical to the lookahead case. When the assertion is part
1601        of a condition, we want to return immediately afterwards. The caller of
1602        this incarnation of the match() function will have set MATCH_CONDASSERT in
1603        md->match_function type, and one of these opcodes will be the first opcode
1604        that is processed. We use a local variable that is preserved over calls to
1605        match() to remember this case. */
1606    
1607        case OP_ASSERT:
1608        case OP_ASSERTBACK:
1609        save_mark = md->mark;
1610        if (md->match_function_type == MATCH_CONDASSERT)
1611          {
1612          condassert = TRUE;
1613          md->match_function_type = 0;
1614          }
1615        else condassert = FALSE;
1616    
1617        do
1618          {
1619          RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM4);
1620          if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT)
1621            {
1622            mstart = md->start_match_ptr;   /* In case \K reset it */
1623            break;
1624            }
1625          md->mark = save_mark;
1626    
1627          /* A COMMIT failure must fail the entire assertion, without trying any
1628          subsequent branches. */
1629    
1630          if (rrc == MATCH_COMMIT) RRETURN(MATCH_NOMATCH);
1631    
1632          /* PCRE does not allow THEN to escape beyond an assertion; it
1633          is treated as NOMATCH. */
1634    
1635          if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
1636          ecode += GET(ecode, 1);
1637          }
1638        while (*ecode == OP_ALT);
1639    
1640        if (*ecode == OP_KET) RRETURN(MATCH_NOMATCH);
1641    
1642        /* If checking an assertion for a condition, return MATCH_MATCH. */
1643    
1644        if (condassert) RRETURN(MATCH_MATCH);
1645    
1646        /* Continue from after the assertion, updating the offsets high water
1647        mark, since extracts may have been taken during the assertion. */
1648    
1649        do ecode += GET(ecode,1); while (*ecode == OP_ALT);
1650        ecode += 1 + LINK_SIZE;
1651        offset_top = md->end_offset_top;
1652        continue;
1653    
1654        /* Negative assertion: all branches must fail to match. Encountering SKIP,
1655        PRUNE, or COMMIT means we must assume failure without checking subsequent
1656        branches. */
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        do
1669          {
1670          RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM5);
1671          md->mark = save_mark;
1672          if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) RRETURN(MATCH_NOMATCH);
1673          if (rrc == MATCH_SKIP || rrc == MATCH_PRUNE || rrc == MATCH_COMMIT)
1674            {
1675            do ecode += GET(ecode,1); while (*ecode == OP_ALT);
1676            break;
1677            }
1678    
1679          /* PCRE does not allow THEN to escape beyond an assertion; it is treated
1680          as NOMATCH. */
1681    
1682          if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
1683          ecode += GET(ecode,1);
1684          }
1685        while (*ecode == OP_ALT);
1686    
1687        if (condassert) RRETURN(MATCH_MATCH);  /* Condition assertion */
1688    
1689        ecode += 1 + LINK_SIZE;
1690        continue;
1691    
1692        /* Move the subject pointer back. This occurs only at the start of
1693        each branch of a lookbehind assertion. If we are too close to the start to
1694        move back, this match function fails. When working with UTF-8 we move
1695        back a number of characters, not bytes. */
1696    
1697        case OP_REVERSE:
1698    #ifdef SUPPORT_UTF
1699        if (utf)
1700          {
1701          i = GET(ecode, 1);
1702          while (i-- > 0)
1703            {
1704            eptr--;
1705            if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
1706            BACKCHAR(eptr);
1707            }
1708          }
1709        else
1710    #endif
1711    
1712        /* No UTF-8 support, or not in UTF-8 mode: count is byte count */
1713    
1714          {
1715          eptr -= GET(ecode, 1);
1716        if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);        if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
1717        }        }
1718    
1719      /* Skip to next op code */      /* Save the earliest consulted character, then skip to next op code */
1720    
1721        if (eptr < md->start_used_ptr) md->start_used_ptr = eptr;
1722      ecode += 1 + LINK_SIZE;      ecode += 1 + LINK_SIZE;
1723      break;      break;
1724    
# Line 854  for (;;) Line 1727  for (;;)
1727      function is able to force a failure. */      function is able to force a failure. */
1728    
1729      case OP_CALLOUT:      case OP_CALLOUT:
1730      if (pcre_callout != NULL)      if (PUBL(callout) != NULL)
1731        {        {
1732        pcre_callout_block cb;        PUBL(callout_block) cb;
1733        cb.version          = 1;   /* Version 1 of the callout block */        cb.version          = 2;   /* Version 1 of the callout block */
1734        cb.callout_number   = ecode[1];        cb.callout_number   = ecode[1];
1735        cb.offset_vector    = md->offset_vector;        cb.offset_vector    = md->offset_vector;
1736    #if defined COMPILE_PCRE8
1737        cb.subject          = (PCRE_SPTR)md->start_subject;        cb.subject          = (PCRE_SPTR)md->start_subject;
1738        cb.subject_length   = md->end_subject - md->start_subject;  #elif defined COMPILE_PCRE16
1739        cb.start_match      = md->start_match - md->start_subject;        cb.subject          = (PCRE_SPTR16)md->start_subject;
1740        cb.current_position = eptr - md->start_subject;  #elif defined COMPILE_PCRE32
1741          cb.subject          = (PCRE_SPTR32)md->start_subject;
1742    #endif
1743          cb.subject_length   = (int)(md->end_subject - md->start_subject);
1744          cb.start_match      = (int)(mstart - md->start_subject);
1745          cb.current_position = (int)(eptr - md->start_subject);
1746        cb.pattern_position = GET(ecode, 2);        cb.pattern_position = GET(ecode, 2);
1747        cb.next_item_length = GET(ecode, 2 + LINK_SIZE);        cb.next_item_length = GET(ecode, 2 + LINK_SIZE);
1748        cb.capture_top      = offset_top/2;        cb.capture_top      = offset_top/2;
1749        cb.capture_last     = md->capture_last;        cb.capture_last     = md->capture_last & CAPLMASK;
1750          /* Internal change requires this for API compatibility. */
1751          if (cb.capture_last == 0) cb.capture_last = -1;
1752        cb.callout_data     = md->callout_data;        cb.callout_data     = md->callout_data;
1753        if ((rrc = (*pcre_callout)(&cb)) > 0) RRETURN(MATCH_NOMATCH);        cb.mark             = md->nomatch_mark;
1754          if ((rrc = (*PUBL(callout))(&cb)) > 0) RRETURN(MATCH_NOMATCH);
1755        if (rrc < 0) RRETURN(rrc);        if (rrc < 0) RRETURN(rrc);
1756        }        }
1757      ecode += 2 + 2*LINK_SIZE;      ecode += 2 + 2*LINK_SIZE;
# Line 879  for (;;) Line 1761  for (;;)
1761      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
1762      whole pattern. (This is so that it works from duplicated subpatterns.)      whole pattern. (This is so that it works from duplicated subpatterns.)
1763    
1764      If there are any capturing brackets started but not finished, we have to      The state of the capturing groups is preserved over recursion, and
1765      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
1766      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
1767      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
1768      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
1769      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
1770      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.  
1771    
1772      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
1773      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
1774      for the original version of this logic. */      for the original version of this logic. It has, however, been hacked around
1775        a lot, so he is not to blame for the current way it works. */
1776    
1777      case OP_RECURSE:      case OP_RECURSE:
1778        {        {
1779        callpat = md->start_code + GET(ecode, 1);        recursion_info *ri;
1780        new_recursive.group_num = *callpat - OP_BRA;        unsigned int recno;
1781    
1782        /* For extended extraction brackets (large number), we have to fish out        callpat = md->start_code + GET(ecode, 1);
1783        the number from a dummy opcode at the start. */        recno = (callpat == md->start_code)? 0 :
1784            GET2(callpat, 1 + LINK_SIZE);
1785    
1786        if (new_recursive.group_num > EXTRACT_BASIC_MAX)        /* Check for repeating a recursion without advancing the subject pointer.
1787          new_recursive.group_num = GET2(callpat, 2+LINK_SIZE);        This should catch convoluted mutual recursions. (Some simple cases are
1788          caught at compile time.) */
1789    
1790          for (ri = md->recursive; ri != NULL; ri = ri->prevrec)
1791            if (recno == ri->group_num && eptr == ri->subject_position)
1792              RRETURN(PCRE_ERROR_RECURSELOOP);
1793    
1794        /* Add to "recursing stack" */        /* Add to "recursing stack" */
1795    
1796          new_recursive.group_num = recno;
1797          new_recursive.saved_capture_last = md->capture_last;
1798          new_recursive.subject_position = eptr;
1799        new_recursive.prevrec = md->recursive;        new_recursive.prevrec = md->recursive;
1800        md->recursive = &new_recursive;        md->recursive = &new_recursive;
1801    
1802        /* Find where to continue from afterwards */        /* Where to continue from afterwards */
1803    
1804        ecode += 1 + LINK_SIZE;        ecode += 1 + LINK_SIZE;
       new_recursive.after_call = ecode;  
1805    
1806        /* Now save the offset data. */        /* Now save the offset data */
1807    
1808        new_recursive.saved_max = md->offset_end;        new_recursive.saved_max = md->offset_end;
1809        if (new_recursive.saved_max <= REC_STACK_SAVE_MAX)        if (new_recursive.saved_max <= REC_STACK_SAVE_MAX)
# Line 923  for (;;) Line 1811  for (;;)
1811        else        else
1812          {          {
1813          new_recursive.offset_save =          new_recursive.offset_save =
1814            (int *)(pcre_malloc)(new_recursive.saved_max * sizeof(int));            (int *)(PUBL(malloc))(new_recursive.saved_max * sizeof(int));
1815          if (new_recursive.offset_save == NULL) RRETURN(PCRE_ERROR_NOMEMORY);          if (new_recursive.offset_save == NULL) RRETURN(PCRE_ERROR_NOMEMORY);
1816          }          }
   
1817        memcpy(new_recursive.offset_save, md->offset_vector,        memcpy(new_recursive.offset_save, md->offset_vector,
1818              new_recursive.saved_max * sizeof(int));              new_recursive.saved_max * sizeof(int));
       new_recursive.save_start = md->start_match;  
       md->start_match = eptr;  
1819    
1820        /* OK, now we can do the recursion. For each top-level alternative we        /* OK, now we can do the recursion. After processing each alternative,
1821        restore the offset and recursion data. */        restore the offset data and the last captured value. If there were nested
1822          recursions, md->recursive might be changed, so reset it before looping.
1823          */
1824    
1825        DPRINTF(("Recursing into group %d\n", new_recursive.group_num));        DPRINTF(("Recursing into group %d\n", new_recursive.group_num));
1826          cbegroup = (*callpat >= OP_SBRA);
1827        do        do
1828          {          {
1829          RMATCH(rrc, eptr, callpat + 1 + LINK_SIZE, offset_top, md, ims,          if (cbegroup) md->match_function_type = MATCH_CBEGROUP;
1830              eptrb, match_isgroup);          RMATCH(eptr, callpat + PRIV(OP_lengths)[*callpat], offset_top,
1831          if (rrc == MATCH_MATCH)            md, eptrb, RM6);
1832            memcpy(md->offset_vector, new_recursive.offset_save,
1833                new_recursive.saved_max * sizeof(int));
1834            md->capture_last = new_recursive.saved_capture_last;
1835            md->recursive = new_recursive.prevrec;
1836            if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT)
1837            {            {
1838            DPRINTF(("Recursion matched\n"));            DPRINTF(("Recursion matched\n"));
           md->recursive = new_recursive.prevrec;  
1839            if (new_recursive.offset_save != stacksave)            if (new_recursive.offset_save != stacksave)
1840              (pcre_free)(new_recursive.offset_save);              (PUBL(free))(new_recursive.offset_save);
1841            RRETURN(MATCH_MATCH);  
1842              /* Set where we got to in the subject, and reset the start in case
1843              it was changed by \K. This *is* propagated back out of a recursion,
1844              for Perl compatibility. */
1845    
1846              eptr = md->end_match_ptr;
1847              mstart = md->start_match_ptr;
1848              goto RECURSION_MATCHED;        /* Exit loop; end processing */
1849            }            }
1850          else if (rrc != MATCH_NOMATCH)  
1851            /* PCRE does not allow THEN, SKIP, PRUNE or COMMIT to escape beyond a
1852            recursion; they are treated as NOMATCH. These codes are defined in a
1853            range that can be tested for. Any other return code is an error. */
1854    
1855            else if (rrc != MATCH_NOMATCH &&
1856                     (rrc < MATCH_BACKTRACK_MIN || rrc > MATCH_BACKTRACK_MAX))
1857            {            {
1858            DPRINTF(("Recursion gave error %d\n", rrc));            DPRINTF(("Recursion gave error %d\n", rrc));
1859              if (new_recursive.offset_save != stacksave)
1860                (PUBL(free))(new_recursive.offset_save);
1861            RRETURN(rrc);            RRETURN(rrc);
1862            }            }
1863    
1864          md->recursive = &new_recursive;          md->recursive = &new_recursive;
         memcpy(md->offset_vector, new_recursive.offset_save,  
             new_recursive.saved_max * sizeof(int));  
1865          callpat += GET(callpat, 1);          callpat += GET(callpat, 1);
1866          }          }
1867        while (*callpat == OP_ALT);        while (*callpat == OP_ALT);
# Line 964  for (;;) Line 1869  for (;;)
1869        DPRINTF(("Recursion didn't match\n"));        DPRINTF(("Recursion didn't match\n"));
1870        md->recursive = new_recursive.prevrec;        md->recursive = new_recursive.prevrec;
1871        if (new_recursive.offset_save != stacksave)        if (new_recursive.offset_save != stacksave)
1872          (pcre_free)(new_recursive.offset_save);          (PUBL(free))(new_recursive.offset_save);
1873        RRETURN(MATCH_NOMATCH);        RRETURN(MATCH_NOMATCH);
1874        }        }
     /* 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);  
1875    
1876      offset_top = md->end_offset_top;      RECURSION_MATCHED:
1877      eptr = md->end_match_ptr;      break;
   
     /* For a non-repeating ket, just continue at this level. This also  
     happens for a repeating ket if no characters were matched in the group.  
     This is the forcible breaking of infinite loops as implemented in Perl  
     5.005. If there is an options reset, it will get obeyed in the normal  
     course of events. */  
   
     if (*ecode == OP_KET || eptr == saved_eptr)  
       {  
       ecode += 1+LINK_SIZE;  
       break;  
       }  
   
     /* The repeating kets try the rest of the pattern or restart from the  
     preceding bracket, in the appropriate order. The second "call" of match()  
     uses tail recursion, to avoid using another stack frame. We need to reset  
     any options that changed within the bracket before re-running it, so  
     check the next opcode. */  
   
     if (ecode[1+LINK_SIZE] == OP_OPT)  
       {  
       ims = (ims & ~PCRE_IMS) | ecode[4];  
       DPRINTF(("ims set to %02lx at group repeat\n", ims));  
       }  
   
     if (*ecode == OP_KETRMIN)  
       {  
       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb, 0);  
       if (rrc != MATCH_NOMATCH) RRETURN(rrc);  
       ecode = prev;  
       flags = match_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 */  
1878    
1879      /* 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
1880      bracketed group and go to there. */      bracketed group and go to there. */
# Line 1051  for (;;) Line 1883  for (;;)
1883      do ecode += GET(ecode,1); while (*ecode == OP_ALT);      do ecode += GET(ecode,1); while (*ecode == OP_ALT);
1884      break;      break;
1885    
1886      /* BRAZERO and BRAMINZERO occur just before a bracket group, indicating      /* BRAZERO, BRAMINZERO and SKIPZERO occur just before a bracket group,
1887      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
1888      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
1889      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
1890      preceded by BRAZERO or BRAMINZERO. */      optional ones preceded by BRAZERO or BRAMINZERO. */
1891    
1892      case OP_BRAZERO:      case OP_BRAZERO:
1893        {      next = ecode + 1;
1894        next = ecode+1;      RMATCH(eptr, next, offset_top, md, eptrb, RM10);
1895        RMATCH(rrc, eptr, next, offset_top, md, ims, eptrb, match_isgroup);      if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1896        if (rrc != MATCH_NOMATCH) RRETURN(rrc);      do next += GET(next, 1); while (*next == OP_ALT);
1897        do next += GET(next,1); while (*next == OP_ALT);      ecode = next + 1 + LINK_SIZE;
       ecode = next + 1+LINK_SIZE;  
       }  
1898      break;      break;
1899    
1900      case OP_BRAMINZERO:      case OP_BRAMINZERO:
1901        {      next = ecode + 1;
1902        next = ecode+1;      do next += GET(next, 1); while (*next == OP_ALT);
1903        do next += GET(next,1); while (*next == OP_ALT);      RMATCH(eptr, next + 1+LINK_SIZE, offset_top, md, eptrb, RM11);
1904        RMATCH(rrc, eptr, next + 1+LINK_SIZE, offset_top, md, ims, eptrb,      if (rrc != MATCH_NOMATCH) RRETURN(rrc);
1905          match_isgroup);      ecode++;
       if (rrc != MATCH_NOMATCH) RRETURN(rrc);  
       ecode++;  
       }  
1906      break;      break;
1907    
1908      /* End of a group, repeated or non-repeating. If we are at the end of      case OP_SKIPZERO:
1909      an assertion "group", stop matching and return MATCH_MATCH, but record the      next = ecode+1;
1910      current high water mark for use by positive assertions. Do this also      do next += GET(next,1); while (*next == OP_ALT);
1911      for the "once" (not-backup up) groups. */      ecode = next + 1 + LINK_SIZE;
1912        break;
1913    
1914        /* BRAPOSZERO occurs before a possessive bracket group. Don't do anything
1915        here; just jump to the group, with allow_zero set TRUE. */
1916    
1917        case OP_BRAPOSZERO:
1918        op = *(++ecode);
1919        allow_zero = TRUE;
1920        if (op == OP_CBRAPOS || op == OP_SCBRAPOS) goto POSSESSIVE_CAPTURE;
1921          goto POSSESSIVE_NON_CAPTURE;
1922    
1923        /* End of a group, repeated or non-repeating. */
1924    
1925      case OP_KET:      case OP_KET:
1926      case OP_KETRMIN:      case OP_KETRMIN:
1927      case OP_KETRMAX:      case OP_KETRMAX:
1928        case OP_KETRPOS:
1929      prev = ecode - GET(ecode, 1);      prev = ecode - GET(ecode, 1);
     saved_eptr = eptrb->epb_saved_eptr;  
   
     /* Back up the stack of bracket start pointers. */  
1930    
1931      eptrb = eptrb->epb_prev;      /* If this was a group that remembered the subject start, in order to break
1932        infinite repeats of empty string matches, retrieve the subject start from
1933        the chain. Otherwise, set it NULL. */
1934    
1935      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)  
1936        {        {
1937        md->end_match_ptr = eptr;      /* For ONCE */        saved_eptr = eptrb->epb_saved_eptr;   /* Value at start of group */
1938        md->end_offset_top = offset_top;        eptrb = eptrb->epb_prev;              /* Backup to previous group */
       RRETURN(MATCH_MATCH);  
1939        }        }
1940        else saved_eptr = NULL;
1941    
1942      /* 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
1943      group number back at the start and if necessary complete handling an      group, stop matching and return MATCH_MATCH, but record the current high
1944      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
1945        start in case it was changed by \K. */
1946    
1947      if (*prev != OP_COND)      if ((*prev >= OP_ASSERT && *prev <= OP_ASSERTBACK_NOT) ||
1948             *prev == OP_ONCE_NC)
1949        {        {
1950        number = *prev - OP_BRA;        md->end_match_ptr = eptr;      /* For ONCE_NC */
1951          md->end_offset_top = offset_top;
1952          md->start_match_ptr = mstart;
1953          RRETURN(MATCH_MATCH);         /* Sets md->mark */
1954          }
1955    
1956        /* 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
1957        the number from a dummy opcode at the start. */      and if necessary complete handling an extraction by setting the offsets and
1958        bumping the high water mark. Whole-pattern recursion is coded as a recurse
1959        into group 0, so it won't be picked up here. Instead, we catch it when the
1960        OP_END is reached. Other recursion is handled here. We just have to record
1961        the current subject position and start match pointer and give a MATCH
1962        return. */
1963    
1964        if (number > EXTRACT_BASIC_MAX) number = GET2(prev, 2+LINK_SIZE);      if (*prev == OP_CBRA || *prev == OP_SCBRA ||
1965            *prev == OP_CBRAPOS || *prev == OP_SCBRAPOS)
1966          {
1967          number = GET2(prev, 1+LINK_SIZE);
1968        offset = number << 1;        offset = number << 1;
1969    
1970  #ifdef DEBUG  #ifdef PCRE_DEBUG
1971        printf("end bracket %d", number);        printf("end bracket %d", number);
1972        printf("\n");        printf("\n");
1973  #endif  #endif
1974    
1975        /* 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. */  
1976    
1977        if (number > 0)        if (md->recursive != NULL && md->recursive->group_num == number)
1978          {          {
1979          md->capture_last = number;          md->end_match_ptr = eptr;
1980          if (offset >= md->offset_max) md->offset_overflow = TRUE; else          md->start_match_ptr = mstart;
1981            {          RRETURN(MATCH_MATCH);
1982            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;  
           }  
1983    
1984          /* Handle a recursively called group. Restore the offsets        /* Deal with capturing */
         appropriately and continue from after the call. */  
1985    
1986          if (md->recursive != NULL && md->recursive->group_num == number)        md->capture_last = (md->capture_last & OVFLMASK) | number;
1987          if (offset >= md->offset_max) md->capture_last |= OVFLBIT; else
1988            {
1989            /* If offset is greater than offset_top, it means that we are
1990            "skipping" a capturing group, and that group's offsets must be marked
1991            unset. In earlier versions of PCRE, all the offsets were unset at the
1992            start of matching, but this doesn't work because atomic groups and
1993            assertions can cause a value to be set that should later be unset.
1994            Example: matching /(?>(a))b|(a)c/ against "ac". This sets group 1 as
1995            part of the atomic group, but this is not on the final matching path,
1996            so must be unset when 2 is set. (If there is no group 2, there is no
1997            problem, because offset_top will then be 2, indicating no capture.) */
1998    
1999            if (offset > offset_top)
2000            {            {
2001            recursion_info *rec = md->recursive;            register int *iptr = md->offset_vector + offset_top;
2002            DPRINTF(("Recursion (%d) succeeded - continuing\n", number));            register int *iend = md->offset_vector + offset;
2003            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;  
2004            }            }
2005    
2006            /* Now make the extraction */
2007    
2008            md->offset_vector[offset] =
2009              md->offset_vector[md->offset_end - number];
2010            md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
2011            if (offset_top <= offset) offset_top = offset + 2;
2012          }          }
2013        }        }
2014    
2015      /* Reset the value of the ims flags, in case they got changed during      /* For an ordinary non-repeating ket, just continue at this level. This
2016      the group. */      also happens for a repeating ket if no characters were matched in the
2017        group. This is the forcible breaking of infinite loops as implemented in
2018      ims = original_ims;      Perl 5.005. For a non-repeating atomic group that includes captures,
2019      DPRINTF(("ims reset to %02lx\n", ims));      establish a backup point by processing the rest of the pattern at a lower
2020        level. If this results in a NOMATCH return, pass MATCH_ONCE back to the
2021      /* For a non-repeating ket, just continue at this level. This also      original OP_ONCE level, thereby bypassing intermediate backup points, but
2022      happens for a repeating ket if no characters were matched in the group.      resetting any captures that happened along the way. */
     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. */  
2023    
2024      if (*ecode == OP_KET || eptr == saved_eptr)      if (*ecode == OP_KET || eptr == saved_eptr)
2025        {        {
2026        ecode += 1 + LINK_SIZE;        if (*prev == OP_ONCE)
2027            {
2028            RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM12);
2029            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2030            md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
2031            RRETURN(MATCH_ONCE);
2032            }
2033          ecode += 1 + LINK_SIZE;    /* Carry on at this level */
2034        break;        break;
2035        }        }
2036    
2037      /* The repeating kets try the rest of the pattern or restart from the      /* OP_KETRPOS is a possessive repeating ket. Remember the current position,
2038      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
2039      tail recursion to avoid using another stack frame. */      at a time from the outer level, thus saving stack. */
2040    
2041        if (*ecode == OP_KETRPOS)
2042          {
2043          md->end_match_ptr = eptr;
2044          md->end_offset_top = offset_top;
2045          RRETURN(MATCH_KETRPOS);
2046          }
2047    
2048        /* The normal repeating kets try the rest of the pattern or restart from
2049        the preceding bracket, in the appropriate order. In the second case, we can
2050        use tail recursion to avoid using another stack frame, unless we have an
2051        an atomic group or an unlimited repeat of a group that can match an empty
2052        string. */
2053    
2054      if (*ecode == OP_KETRMIN)      if (*ecode == OP_KETRMIN)
2055        {        {
2056        RMATCH(rrc, eptr, ecode + 1+LINK_SIZE, offset_top, md, ims, eptrb, 0);        RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM7);
2057        if (rrc != MATCH_NOMATCH) RRETURN(rrc);        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2058          if (*prev == OP_ONCE)
2059            {
2060            RMATCH(eptr, prev, offset_top, md, eptrb, RM8);
2061            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2062            md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
2063            RRETURN(MATCH_ONCE);
2064            }
2065          if (*prev >= OP_SBRA)    /* Could match an empty string */
2066            {
2067            RMATCH(eptr, prev, offset_top, md, eptrb, RM50);
2068            RRETURN(rrc);
2069            }
2070        ecode = prev;        ecode = prev;
       flags = match_isgroup;  
2071        goto TAIL_RECURSE;        goto TAIL_RECURSE;
2072        }        }
2073      else  /* OP_KETRMAX */      else  /* OP_KETRMAX */
2074        {        {
2075        RMATCH(rrc, eptr, prev, offset_top, md, ims, eptrb, match_isgroup);        RMATCH(eptr, prev, offset_top, md, eptrb, RM13);
2076          if (rrc == MATCH_ONCE && md->once_target == prev) rrc = MATCH_NOMATCH;
2077        if (rrc != MATCH_NOMATCH) RRETURN(rrc);        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2078          if (*prev == OP_ONCE)
2079            {
2080            RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM9);
2081            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2082            md->once_target = prev;
2083            RRETURN(MATCH_ONCE);
2084            }
2085        ecode += 1 + LINK_SIZE;        ecode += 1 + LINK_SIZE;
       flags = 0;  
2086        goto TAIL_RECURSE;        goto TAIL_RECURSE;
2087        }        }
2088      /* Control never gets here */      /* Control never gets here */
2089    
2090      /* Start of subject unless notbol, or after internal newline if multiline */      /* Not multiline mode: start of subject assertion, unless notbol. */
2091    
2092      case OP_CIRC:      case OP_CIRC:
2093      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 */  
2094    
2095      /* Start of subject assertion */      /* Start of subject assertion */
2096    
# Line 1218  for (;;) Line 2099  for (;;)
2099      ecode++;      ecode++;
2100      break;      break;
2101    
2102        /* Multiline mode: start of subject unless notbol, or after any newline. */
2103    
2104        case OP_CIRCM:
2105        if (md->notbol && eptr == md->start_subject) RRETURN(MATCH_NOMATCH);
2106        if (eptr != md->start_subject &&
2107            (eptr == md->end_subject || !WAS_NEWLINE(eptr)))
2108          RRETURN(MATCH_NOMATCH);
2109        ecode++;
2110        break;
2111    
2112      /* Start of match assertion */      /* Start of match assertion */
2113    
2114      case OP_SOM:      case OP_SOM:
# Line 1225  for (;;) Line 2116  for (;;)
2116      ecode++;      ecode++;
2117      break;      break;
2118    
2119      /* 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. */  
2120    
2121      case OP_DOLL:      case OP_SET_SOM:
2122      if ((ims & PCRE_MULTILINE) != 0)      mstart = eptr;
2123        ecode++;
2124        break;
2125    
2126        /* Multiline mode: assert before any newline, or before end of subject
2127        unless noteol is set. */
2128    
2129        case OP_DOLLM:
2130        if (eptr < md->end_subject)
2131        {        {
2132        if (eptr < md->end_subject)        if (!IS_NEWLINE(eptr))
2133          { if (!IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH); }          {
2134        else          if (md->partial != 0 &&
2135          { if (md->noteol) RRETURN(MATCH_NOMATCH); }              eptr + 1 >= md->end_subject &&
2136        ecode++;              NLBLOCK->nltype == NLTYPE_FIXED &&
2137        break;              NLBLOCK->nllen == 2 &&
2138                RAWUCHARTEST(eptr) == NLBLOCK->nl[0])
2139              {
2140              md->hitend = TRUE;
2141              if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2142              }
2143            RRETURN(MATCH_NOMATCH);
2144            }
2145        }        }
2146      else      else
2147        {        {
2148        if (md->noteol) RRETURN(MATCH_NOMATCH);        if (md->noteol) RRETURN(MATCH_NOMATCH);
2149        if (!md->endonly)        SCHECK_PARTIAL();
         {  
         if (eptr != md->end_subject &&  
             (eptr != md->end_subject - md->nllen || !IS_NEWLINE(eptr)))  
           RRETURN(MATCH_NOMATCH);  
         ecode++;  
         break;  
         }  
2150        }        }
2151        ecode++;
2152        break;
2153    
2154        /* Not multiline mode: assert before a terminating newline or before end of
2155        subject unless noteol is set. */
2156    
2157        case OP_DOLL:
2158        if (md->noteol) RRETURN(MATCH_NOMATCH);
2159        if (!md->endonly) goto ASSERT_NL_OR_EOS;
2160    
2161      /* ... else fall through for endonly */      /* ... else fall through for endonly */
2162    
2163      /* End of subject assertion (\z) */      /* End of subject assertion (\z) */
2164    
2165      case OP_EOD:      case OP_EOD:
2166      if (eptr < md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr < md->end_subject) RRETURN(MATCH_NOMATCH);
2167        SCHECK_PARTIAL();
2168      ecode++;      ecode++;
2169      break;      break;
2170    
2171      /* End of subject or ending \n assertion (\Z) */      /* End of subject or ending \n assertion (\Z) */
2172    
2173      case OP_EODN:      case OP_EODN:
2174      if (eptr != md->end_subject &&      ASSERT_NL_OR_EOS:
2175          (eptr != md->end_subject - md->nllen || !IS_NEWLINE(eptr)))      if (eptr < md->end_subject &&
2176            (!IS_NEWLINE(eptr) || eptr != md->end_subject - md->nllen))
2177          {
2178          if (md->partial != 0 &&
2179              eptr + 1 >= md->end_subject &&
2180              NLBLOCK->nltype == NLTYPE_FIXED &&
2181              NLBLOCK->nllen == 2 &&
2182              RAWUCHARTEST(eptr) == NLBLOCK->nl[0])
2183            {
2184            md->hitend = TRUE;
2185            if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2186            }
2187        RRETURN(MATCH_NOMATCH);        RRETURN(MATCH_NOMATCH);
2188          }
2189    
2190        /* Either at end of string or \n before end. */
2191    
2192        SCHECK_PARTIAL();
2193      ecode++;      ecode++;
2194      break;      break;
2195    
# Line 1276  for (;;) Line 2201  for (;;)
2201    
2202        /* Find out if the previous and current characters are "word" characters.        /* Find out if the previous and current characters are "word" characters.
2203        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
2204        be "non-word" characters. */        be "non-word" characters. Remember the earliest consulted character for
2205          partial matching. */
2206    
2207  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
2208        if (utf8)        if (utf)
2209          {          {
2210            /* Get status of previous character */
2211    
2212          if (eptr == md->start_subject) prev_is_word = FALSE; else          if (eptr == md->start_subject) prev_is_word = FALSE; else
2213            {            {
2214            const uschar *lastptr = eptr - 1;            PCRE_PUCHAR lastptr = eptr - 1;
2215            while((*lastptr & 0xc0) == 0x80) lastptr--;            BACKCHAR(lastptr);
2216              if (lastptr < md->start_used_ptr) md->start_used_ptr = lastptr;
2217            GETCHAR(c, lastptr);            GETCHAR(c, lastptr);
2218    #ifdef SUPPORT_UCP
2219              if (md->use_ucp)
2220                {
2221                if (c == '_') prev_is_word = TRUE; else
2222                  {
2223                  int cat = UCD_CATEGORY(c);
2224                  prev_is_word = (cat == ucp_L || cat == ucp_N);
2225                  }
2226                }
2227              else
2228    #endif
2229            prev_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;            prev_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;
2230            }            }
2231          if (eptr >= md->end_subject) cur_is_word = FALSE; else  
2232            /* Get status of next character */
2233    
2234            if (eptr >= md->end_subject)
2235              {
2236              SCHECK_PARTIAL();
2237              cur_is_word = FALSE;
2238              }
2239            else
2240            {            {
2241            GETCHAR(c, eptr);            GETCHAR(c, eptr);
2242    #ifdef SUPPORT_UCP
2243              if (md->use_ucp)
2244                {
2245                if (c == '_') cur_is_word = TRUE; else
2246                  {
2247                  int cat = UCD_CATEGORY(c);
2248                  cur_is_word = (cat == ucp_L || cat == ucp_N);
2249                  }
2250                }
2251              else
2252    #endif
2253            cur_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;            cur_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;
2254            }            }
2255          }          }
2256        else        else
2257  #endif  #endif
2258    
2259        /* More streamlined when not in UTF-8 mode */        /* Not in UTF-8 mode, but we may still have PCRE_UCP set, and for
2260          consistency with the behaviour of \w we do use it in this case. */
2261    
2262          {          {
2263          prev_is_word = (eptr != md->start_subject) &&          /* Get status of previous character */
2264            ((md->ctypes[eptr[-1]] & ctype_word) != 0);  
2265          cur_is_word = (eptr < md->end_subject) &&          if (eptr == md->start_subject) prev_is_word = FALSE; else
2266            ((md->ctypes[*eptr] & ctype_word) != 0);            {
2267              if (eptr <= md->start_used_ptr) md->start_used_ptr = eptr - 1;
2268    #ifdef SUPPORT_UCP
2269              if (md->use_ucp)
2270                {
2271                c = eptr[-1];
2272                if (c == '_') prev_is_word = TRUE; else
2273                  {
2274                  int cat = UCD_CATEGORY(c);
2275                  prev_is_word = (cat == ucp_L || cat == ucp_N);
2276                  }
2277                }
2278              else
2279    #endif
2280              prev_is_word = MAX_255(eptr[-1])
2281                && ((md->ctypes[eptr[-1]] & ctype_word) != 0);
2282              }
2283    
2284            /* Get status of next character */
2285    
2286            if (eptr >= md->end_subject)
2287              {
2288              SCHECK_PARTIAL();
2289              cur_is_word = FALSE;
2290              }
2291            else
2292    #ifdef SUPPORT_UCP
2293            if (md->use_ucp)
2294              {
2295              c = *eptr;
2296              if (c == '_') cur_is_word = TRUE; else
2297                {
2298                int cat = UCD_CATEGORY(c);
2299                cur_is_word = (cat == ucp_L || cat == ucp_N);
2300                }
2301              }
2302            else
2303    #endif
2304            cur_is_word = MAX_255(*eptr)
2305              && ((md->ctypes[*eptr] & ctype_word) != 0);
2306          }          }
2307    
2308        /* Now see if the situation is what we want */        /* Now see if the situation is what we want */
# Line 1314  for (;;) Line 2313  for (;;)
2313        }        }
2314      break;      break;
2315    
2316      /* Match a single character type; inline for speed */      /* Match any single character type except newline; have to take care with
2317        CRLF newlines and partial matching. */
2318    
2319      case OP_ANY:      case OP_ANY:
2320      if ((ims & PCRE_DOTALL) == 0)      if (IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH);
2321        if (md->partial != 0 &&
2322            eptr + 1 >= md->end_subject &&
2323            NLBLOCK->nltype == NLTYPE_FIXED &&
2324            NLBLOCK->nllen == 2 &&
2325            RAWUCHARTEST(eptr) == NLBLOCK->nl[0])
2326        {        {
2327        if (eptr <= md->end_subject - md->nllen && IS_NEWLINE(eptr))        md->hitend = TRUE;
2328          RRETURN(MATCH_NOMATCH);        if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2329        }        }
2330      if (eptr++ >= md->end_subject) RRETURN(MATCH_NOMATCH);  
2331      if (utf8)      /* Fall through */
2332        while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;  
2333        /* Match any single character whatsoever. */
2334    
2335        case OP_ALLANY:
2336        if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
2337          {                            /* not be updated before SCHECK_PARTIAL. */
2338          SCHECK_PARTIAL();
2339          RRETURN(MATCH_NOMATCH);
2340          }
2341        eptr++;
2342    #ifdef SUPPORT_UTF
2343        if (utf) ACROSSCHAR(eptr < md->end_subject, *eptr, eptr++);
2344    #endif
2345      ecode++;      ecode++;
2346      break;      break;
2347    
# Line 1332  for (;;) Line 2349  for (;;)
2349      any byte, even newline, independent of the setting of PCRE_DOTALL. */      any byte, even newline, independent of the setting of PCRE_DOTALL. */
2350    
2351      case OP_ANYBYTE:      case OP_ANYBYTE:
2352      if (eptr++ >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
2353          {                            /* not be updated before SCHECK_PARTIAL. */
2354          SCHECK_PARTIAL();
2355          RRETURN(MATCH_NOMATCH);
2356          }
2357        eptr++;
2358      ecode++;      ecode++;
2359      break;      break;
2360    
2361      case OP_NOT_DIGIT:      case OP_NOT_DIGIT:
2362      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2363          {
2364          SCHECK_PARTIAL();
2365          RRETURN(MATCH_NOMATCH);
2366          }
2367      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2368      if (      if (
2369  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2370         c < 256 &&         c < 256 &&
2371  #endif  #endif
2372         (md->ctypes[c] & ctype_digit) != 0         (md->ctypes[c] & ctype_digit) != 0
# Line 1350  for (;;) Line 2376  for (;;)
2376      break;      break;
2377    
2378      case OP_DIGIT:      case OP_DIGIT:
2379      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2380          {
2381          SCHECK_PARTIAL();
2382          RRETURN(MATCH_NOMATCH);
2383          }
2384      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2385      if (      if (
2386  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2387         c >= 256 ||         c > 255 ||
2388  #endif  #endif
2389         (md->ctypes[c] & ctype_digit) == 0         (md->ctypes[c] & ctype_digit) == 0
2390         )         )
# Line 1363  for (;;) Line 2393  for (;;)
2393      break;      break;
2394    
2395      case OP_NOT_WHITESPACE:      case OP_NOT_WHITESPACE:
2396      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2397          {
2398          SCHECK_PARTIAL();
2399          RRETURN(MATCH_NOMATCH);
2400          }
2401      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2402      if (      if (
2403  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2404         c < 256 &&         c < 256 &&
2405  #endif  #endif
2406         (md->ctypes[c] & ctype_space) != 0         (md->ctypes[c] & ctype_space) != 0
# Line 1376  for (;;) Line 2410  for (;;)
2410      break;      break;
2411    
2412      case OP_WHITESPACE:      case OP_WHITESPACE:
2413      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2414          {
2415          SCHECK_PARTIAL();
2416          RRETURN(MATCH_NOMATCH);
2417          }
2418      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2419      if (      if (
2420  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2421         c >= 256 ||         c > 255 ||
2422  #endif  #endif
2423         (md->ctypes[c] & ctype_space) == 0         (md->ctypes[c] & ctype_space) == 0
2424         )         )
# Line 1389  for (;;) Line 2427  for (;;)
2427      break;      break;
2428    
2429      case OP_NOT_WORDCHAR:      case OP_NOT_WORDCHAR:
2430      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2431          {
2432          SCHECK_PARTIAL();
2433          RRETURN(MATCH_NOMATCH);
2434          }
2435      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2436      if (      if (
2437  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2438         c < 256 &&         c < 256 &&
2439  #endif  #endif
2440         (md->ctypes[c] & ctype_word) != 0         (md->ctypes[c] & ctype_word) != 0
# Line 1402  for (;;) Line 2444  for (;;)
2444      break;      break;
2445    
2446      case OP_WORDCHAR:      case OP_WORDCHAR:
2447      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2448          {
2449          SCHECK_PARTIAL();
2450          RRETURN(MATCH_NOMATCH);
2451          }
2452      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2453      if (      if (
2454  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
2455         c >= 256 ||         c > 255 ||
2456  #endif  #endif
2457         (md->ctypes[c] & ctype_word) == 0         (md->ctypes[c] & ctype_word) == 0
2458         )         )
# Line 1414  for (;;) Line 2460  for (;;)
2460      ecode++;      ecode++;
2461      break;      break;
2462    
2463        case OP_ANYNL:
2464        if (eptr >= md->end_subject)
2465          {
2466          SCHECK_PARTIAL();
2467          RRETURN(MATCH_NOMATCH);
2468          }
2469        GETCHARINCTEST(c, eptr);
2470        switch(c)
2471          {
2472          default: RRETURN(MATCH_NOMATCH);
2473    
2474          case CHAR_CR:
2475          if (eptr >= md->end_subject)
2476            {
2477            SCHECK_PARTIAL();
2478            }
2479          else if (RAWUCHARTEST(eptr) == CHAR_LF) eptr++;
2480          break;
2481    
2482          case CHAR_LF:
2483          break;
2484    
2485          case CHAR_VT:
2486          case CHAR_FF:
2487          case CHAR_NEL:
2488    #ifndef EBCDIC
2489          case 0x2028:
2490          case 0x2029:
2491    #endif  /* Not EBCDIC */
2492          if (md->bsr_anycrlf) RRETURN(MATCH_NOMATCH);
2493          break;
2494          }
2495        ecode++;
2496        break;
2497    
2498        case OP_NOT_HSPACE:
2499        if (eptr >= md->end_subject)
2500          {
2501          SCHECK_PARTIAL();
2502          RRETURN(MATCH_NOMATCH);
2503          }
2504        GETCHARINCTEST(c, eptr);
2505        switch(c)
2506          {
2507          HSPACE_CASES: RRETURN(MATCH_NOMATCH);  /* Byte and multibyte cases */
2508          default: break;
2509          }
2510        ecode++;
2511        break;
2512    
2513        case OP_HSPACE:
2514        if (eptr >= md->end_subject)
2515          {
2516          SCHECK_PARTIAL();
2517          RRETURN(MATCH_NOMATCH);
2518          }
2519        GETCHARINCTEST(c, eptr);
2520        switch(c)
2521          {
2522          HSPACE_CASES: break;  /* Byte and multibyte cases */
2523          default: RRETURN(MATCH_NOMATCH);
2524          }
2525        ecode++;
2526        break;
2527    
2528        case OP_NOT_VSPACE:
2529        if (eptr >= md->end_subject)
2530          {
2531          SCHECK_PARTIAL();
2532          RRETURN(MATCH_NOMATCH);
2533          }
2534        GETCHARINCTEST(c, eptr);
2535        switch(c)
2536          {
2537          VSPACE_CASES: RRETURN(MATCH_NOMATCH);
2538          default: break;
2539          }
2540        ecode++;
2541        break;
2542    
2543        case OP_VSPACE:
2544        if (eptr >= md->end_subject)
2545          {
2546          SCHECK_PARTIAL();
2547          RRETURN(MATCH_NOMATCH);
2548          }
2549        GETCHARINCTEST(c, eptr);
2550        switch(c)
2551          {
2552          VSPACE_CASES: break;
2553          default: RRETURN(MATCH_NOMATCH);
2554          }
2555        ecode++;
2556        break;
2557    
2558  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
2559      /* Check the next character by Unicode property. We will get here only      /* Check the next character by Unicode property. We will get here only
2560      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. */
2561    
2562      case OP_PROP:      case OP_PROP:
2563      case OP_NOTPROP:      case OP_NOTPROP:
2564      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2565          {
2566          SCHECK_PARTIAL();
2567          RRETURN(MATCH_NOMATCH);
2568          }
2569      GETCHARINCTEST(c, eptr);      GETCHARINCTEST(c, eptr);
2570        {        {
2571        int chartype, script;        const pcre_uint32 *cp;
2572        int category = _pcre_ucp_findprop(c, &chartype, &script);        const ucd_record *prop = GET_UCD(c);
2573    
2574        switch(ecode[1])        switch(ecode[1])
2575          {          {
# Line 1433  for (;;) Line 2578  for (;;)
2578          break;          break;
2579    
2580          case PT_LAMP:          case PT_LAMP:
2581          if ((chartype == ucp_Lu ||          if ((prop->chartype == ucp_Lu ||
2582               chartype == ucp_Ll ||               prop->chartype == ucp_Ll ||
2583               chartype == ucp_Lt) == (op == OP_NOTPROP))               prop->chartype == ucp_Lt) == (op == OP_NOTPROP))
2584            RRETURN(MATCH_NOMATCH);            RRETURN(MATCH_NOMATCH);
2585           break;          break;
2586    
2587          case PT_GC:          case PT_GC:
2588          if ((ecode[2] != category) == (op == OP_PROP))          if ((ecode[2] != PRIV(ucp_gentype)[prop->chartype]) == (op == OP_PROP))
2589            RRETURN(MATCH_NOMATCH);            RRETURN(MATCH_NOMATCH);
2590          break;          break;
2591    
2592          case PT_PC:          case PT_PC:
2593          if ((ecode[2] != chartype) == (op == OP_PROP))          if ((ecode[2] != prop->chartype) == (op == OP_PROP))
2594            RRETURN(MATCH_NOMATCH);            RRETURN(MATCH_NOMATCH);
2595          break;          break;
2596    
2597          case PT_SC:          case PT_SC:
2598          if ((ecode[2] != script) == (op == OP_PROP))          if ((ecode[2] != prop->script) == (op == OP_PROP))
2599              RRETURN(MATCH_NOMATCH);
2600            break;
2601    
2602            /* These are specials */
2603    
2604            case PT_ALNUM:
2605            if ((PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2606                 PRIV(ucp_gentype)[prop->chartype] == ucp_N) == (op == OP_NOTPROP))
2607              RRETURN(MATCH_NOMATCH);
2608            break;
2609    
2610            case PT_SPACE:    /* Perl space */
2611            if ((PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2612                 c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2613                   == (op == OP_NOTPROP))
2614              RRETURN(MATCH_NOMATCH);
2615            break;
2616    
2617            case PT_PXSPACE:  /* POSIX space */
2618            if ((PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
2619                 c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2620                 c == CHAR_FF || c == CHAR_CR)
2621                   == (op == OP_NOTPROP))
2622              RRETURN(MATCH_NOMATCH);
2623            break;
2624    
2625            case PT_WORD:
2626            if ((PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
2627                 PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
2628                 c == CHAR_UNDERSCORE) == (op == OP_NOTPROP))
2629            RRETURN(MATCH_NOMATCH);            RRETURN(MATCH_NOMATCH);
2630          break;          break;
2631    
2632            case PT_CLIST:
2633            cp = PRIV(ucd_caseless_sets) + ecode[2];
2634            for (;;)
2635              {
2636              if (c < *cp)
2637                { if (op == OP_PROP) { RRETURN(MATCH_NOMATCH); } else break; }
2638              if (c == *cp++)
2639                { if (op == OP_PROP) break; else { RRETURN(MATCH_NOMATCH); } }
2640              }
2641            break;
2642    
2643            case PT_UCNC:
2644            if ((c == CHAR_DOLLAR_SIGN || c == CHAR_COMMERCIAL_AT ||
2645                 c == CHAR_GRAVE_ACCENT || (c >= 0xa0 && c <= 0xd7ff) ||
2646                 c >= 0xe000) == (op == OP_NOTPROP))
2647              RRETURN(MATCH_NOMATCH);
2648            break;
2649    
2650            /* This should never occur */
2651    
2652          default:          default:
2653          RRETURN(PCRE_ERROR_INTERNAL);          RRETURN(PCRE_ERROR_INTERNAL);
         break;  
2654          }          }
2655    
2656        ecode += 3;        ecode += 3;
# Line 1467  for (;;) Line 2661  for (;;)
2661      is in the binary; otherwise a compile-time error occurs. */      is in the binary; otherwise a compile-time error occurs. */
2662    
2663      case OP_EXTUNI:      case OP_EXTUNI:
2664      if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);      if (eptr >= md->end_subject)
2665      GETCHARINCTEST(c, eptr);        {
2666          SCHECK_PARTIAL();
2667          RRETURN(MATCH_NOMATCH);
2668          }
2669        else
2670        {        {
2671        int chartype, script;        int lgb, rgb;
2672        int category = _pcre_ucp_findprop(c, &chartype, &script);        GETCHARINCTEST(c, eptr);
2673        if (category == ucp_M) RRETURN(MATCH_NOMATCH);        lgb = UCD_GRAPHBREAK(c);
2674        while (eptr < md->end_subject)        while (eptr < md->end_subject)
2675          {          {
2676          int len = 1;          int len = 1;
2677          if (!utf8) c = *eptr; else          if (!utf) c = *eptr; else { GETCHARLEN(c, eptr, len); }
2678            {          rgb = UCD_GRAPHBREAK(c);
2679            GETCHARLEN(c, eptr, len);          if ((PRIV(ucp_gbtable)[lgb] & (1 << rgb)) == 0) break;
2680            }          lgb = rgb;
         category = _pcre_ucp_findprop(c, &chartype, &script);  
         if (category != ucp_M) break;  
2681          eptr += len;          eptr += len;
2682          }          }
2683        }        }
2684        CHECK_PARTIAL();
2685      ecode++;      ecode++;
2686      break;      break;
2687  #endif  #endif  /* SUPPORT_UCP */
2688    
2689    
2690      /* 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 2696  for (;;)
2696      loops). */      loops). */
2697    
2698      case OP_REF:      case OP_REF:
2699        {      case OP_REFI:
2700        offset = GET2(ecode, 1) << 1;               /* Doubled ref number */      caseless = op == OP_REFI;
2701        ecode += 3;                                 /* Advance past item */      offset = GET2(ecode, 1) << 1;               /* Doubled ref number */
2702        ecode += 1 + IMM2_SIZE;
2703    
2704        /* 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];  
2705    
2706        /* Set up for repetition, or handle the non-repeated case */      (a) In the default, Perl-compatible state, set the length negative;
2707        this ensures that every attempt at a match fails. We can't just fail
2708        here, because of the possibility of quantifiers with zero minima.
2709    
2710        switch (*ecode)      (b) If the JavaScript compatibility flag is set, set the length to zero
2711          {      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;  
2712    
2713          case OP_CRRANGE:      Otherwise, set the length to the length of what was matched by the
2714          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;  
2715    
2716          default:               /* No repeat follows */      if (offset >= offset_top || md->offset_vector[offset] < 0)
2717          if (!match_ref(offset, eptr, length, md, ims)) RRETURN(MATCH_NOMATCH);        length = (md->jscript_compat)? 0 : -1;
2718          eptr += length;      else
2719          continue;              /* With the main loop */        length = md->offset_vector[offset+1] - md->offset_vector[offset];
2720    
2721        /* Set up for repetition, or handle the non-repeated case */
2722    
2723        switch (*ecode)
2724          {
2725          case OP_CRSTAR:
2726          case OP_CRMINSTAR:
2727          case OP_CRPLUS:
2728          case OP_CRMINPLUS:
2729          case OP_CRQUERY:
2730          case OP_CRMINQUERY:
2731          c = *ecode++ - OP_CRSTAR;
2732          minimize = (c & 1) != 0;
2733          min = rep_min[c];                 /* Pick up values from tables; */
2734          max = rep_max[c];                 /* zero for max => infinity */
2735          if (max == 0) max = INT_MAX;
2736          break;
2737    
2738          case OP_CRRANGE:
2739          case OP_CRMINRANGE:
2740          minimize = (*ecode == OP_CRMINRANGE);
2741          min = GET2(ecode, 1);
2742          max = GET2(ecode, 1 + IMM2_SIZE);
2743          if (max == 0) max = INT_MAX;
2744          ecode += 1 + 2 * IMM2_SIZE;
2745          break;
2746    
2747          default:               /* No repeat follows */
2748          if ((length = match_ref(offset, eptr, length, md, caseless)) < 0)
2749            {
2750            if (length == -2) eptr = md->end_subject;   /* Partial match */
2751            CHECK_PARTIAL();
2752            RRETURN(MATCH_NOMATCH);
2753          }          }
2754          eptr += length;
2755          continue;              /* With the main loop */
2756          }
2757    
2758        /* If the length of the reference is zero, just continue with the      /* Handle repeated back references. If the length of the reference is
2759        main loop. */      zero, just continue with the main loop. If the length is negative, it
2760        means the reference is unset in non-Java-compatible mode. If the minimum is
2761        zero, we can continue at the same level without recursion. For any other
2762        minimum, carrying on will result in NOMATCH. */
2763    
2764        if (length == 0) continue;      if (length == 0) continue;
2765        if (length < 0 && min == 0) continue;
2766    
2767        /* First, ensure the minimum number of matches are present. We get back      /* First, ensure the minimum number of matches are present. We get back
2768        the length of the reference string explicitly rather than passing the      the length of the reference string explicitly rather than passing the
2769        address of eptr, so that eptr can be a register variable. */      address of eptr, so that eptr can be a register variable. */
2770    
2771        for (i = 1; i <= min; i++)      for (i = 1; i <= min; i++)
2772          {
2773          int slength;
2774          if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
2775          {          {
2776          if (!match_ref(offset, eptr, length, md, ims)) RRETURN(MATCH_NOMATCH);          if (slength == -2) eptr = md->end_subject;   /* Partial match */
2777          eptr += length;          CHECK_PARTIAL();
2778            RRETURN(MATCH_NOMATCH);
2779          }          }
2780          eptr += slength;
2781          }
2782    
2783        /* If min = max, continue at the same level without recursion.      /* If min = max, continue at the same level without recursion.
2784        They are not both allowed to be zero. */      They are not both allowed to be zero. */
2785    
2786        if (min == max) continue;      if (min == max) continue;
2787    
2788        /* If minimizing, keep trying and advancing the pointer */      /* If minimizing, keep trying and advancing the pointer */
2789    
2790        if (minimize)      if (minimize)
2791          {
2792          for (fi = min;; fi++)
2793          {          {
2794          for (fi = min;; fi++)          int slength;
2795            RMATCH(eptr, ecode, offset_top, md, eptrb, RM14);
2796            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2797            if (fi >= max) RRETURN(MATCH_NOMATCH);
2798            if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
2799            {            {
2800            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);            if (slength == -2) eptr = md->end_subject;   /* Partial match */
2801            if (rrc != MATCH_NOMATCH) RRETURN(rrc);            CHECK_PARTIAL();
2802            if (fi >= max || !match_ref(offset, eptr, length, md, ims))            RRETURN(MATCH_NOMATCH);
             RRETURN(MATCH_NOMATCH);  
           eptr += length;  
2803            }            }
2804          /* Control never gets here */          eptr += slength;
2805          }          }
2806          /* Control never gets here */
2807          }
2808    
2809        /* If maximizing, find the longest string and work backwards */      /* If maximizing, find the longest string and work backwards */
2810    
2811        else      else
2812          {
2813          pp = eptr;
2814          for (i = min; i < max; i++)
2815          {          {
2816          pp = eptr;          int slength;
2817          for (i = min; i < max; i++)          if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
2818            {            {
2819            if (!match_ref(offset, eptr, length, md, ims)) break;            /* Can't use CHECK_PARTIAL because we don't want to update eptr in
2820            eptr += length;            the soft partial matching case. */
2821            }  
2822          while (eptr >= pp)            if (slength == -2 && md->partial != 0 &&
2823            {                md->end_subject > md->start_used_ptr)
2824            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              {
2825            if (rrc != MATCH_NOMATCH) RRETURN(rrc);              md->hitend = TRUE;
2826            eptr -= length;              if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
2827                }
2828              break;
2829            }            }
2830          RRETURN(MATCH_NOMATCH);          eptr += slength;
2831            }
2832    
2833          while (eptr >= pp)
2834            {
2835            RMATCH(eptr, ecode, offset_top, md, eptrb, RM15);
2836            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2837            eptr -= length;
2838          }          }
2839          RRETURN(MATCH_NOMATCH);
2840        }        }
2841      /* Control never gets here */      /* Control never gets here */
2842    
   
   
2843      /* Match a bit-mapped character class, possibly repeatedly. This op code is      /* Match a bit-mapped character class, possibly repeatedly. This op code is
2844      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,
2845      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 2854  for (;;)
2854      case OP_NCLASS:      case OP_NCLASS:
2855      case OP_CLASS:      case OP_CLASS:
2856        {        {
2857          /* The data variable is saved across frames, so the byte map needs to
2858          be stored there. */
2859    #define BYTE_MAP ((pcre_uint8 *)data)
2860        data = ecode + 1;                /* Save for matching */        data = ecode + 1;                /* Save for matching */
2861        ecode += 33;                     /* Advance past the item */        ecode += 1 + (32 / sizeof(pcre_uchar)); /* Advance past the item */
2862    
2863        switch (*ecode)        switch (*ecode)
2864          {          {
# Line 1638  for (;;) Line 2879  for (;;)
2879          case OP_CRMINRANGE:          case OP_CRMINRANGE:
2880          minimize = (*ecode == OP_CRMINRANGE);          minimize = (*ecode == OP_CRMINRANGE);
2881          min = GET2(ecode, 1);          min = GET2(ecode, 1);
2882          max = GET2(ecode, 3);          max = GET2(ecode, 1 + IMM2_SIZE);
2883          if (max == 0) max = INT_MAX;          if (max == 0) max = INT_MAX;
2884          ecode += 5;          ecode += 1 + 2 * IMM2_SIZE;
2885          break;          break;
2886    
2887          default:               /* No repeat follows */          default:               /* No repeat follows */
# Line 1650  for (;;) Line 2891  for (;;)
2891    
2892        /* First, ensure the minimum number of matches are present. */        /* First, ensure the minimum number of matches are present. */
2893    
2894  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
2895        /* UTF-8 mode */        if (utf)
       if (utf8)  
2896          {          {
2897          for (i = 1; i <= min; i++)          for (i = 1; i <= min; i++)
2898            {            {
2899            if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);            if (eptr >= md->end_subject)
2900                {
2901                SCHECK_PARTIAL();
2902                RRETURN(MATCH_NOMATCH);
2903                }
2904            GETCHARINC(c, eptr);            GETCHARINC(c, eptr);
2905            if (c > 255)            if (c > 255)
2906              {              {
2907              if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);              if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
2908              }              }
2909            else            else
2910              {              if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
             if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);  
             }  
2911            }            }
2912          }          }
2913        else        else
2914  #endif  #endif
2915        /* Not UTF-8 mode */        /* Not UTF mode */
2916          {          {
2917          for (i = 1; i <= min; i++)          for (i = 1; i <= min; i++)
2918            {            {
2919            if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);            if (eptr >= md->end_subject)
2920                {
2921                SCHECK_PARTIAL();
2922                RRETURN(MATCH_NOMATCH);
2923                }
2924            c = *eptr++;            c = *eptr++;
2925            if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);  #ifndef COMPILE_PCRE8
2926              if (c > 255)
2927                {
2928                if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
2929                }
2930              else
2931    #endif
2932                if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
2933            }            }
2934          }          }
2935    
# Line 1690  for (;;) Line 2943  for (;;)
2943    
2944        if (minimize)        if (minimize)
2945          {          {
2946  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
2947          /* UTF-8 mode */          if (utf)
         if (utf8)  
2948            {            {
2949            for (fi = min;; fi++)            for (fi = min;; fi++)
2950              {              {
2951              RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              RMATCH(eptr, ecode, offset_top, md, eptrb, RM16);
2952              if (rrc != MATCH_NOMATCH) RRETURN(rrc);              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2953              if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);              if (fi >= max) RRETURN(MATCH_NOMATCH);
2954                if (eptr >= md->end_subject)
2955                  {
2956                  SCHECK_PARTIAL();
2957                  RRETURN(MATCH_NOMATCH);
2958                  }
2959              GETCHARINC(c, eptr);              GETCHARINC(c, eptr);
2960              if (c > 255)              if (c > 255)
2961                {                {
2962                if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);                if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
2963                }                }
2964              else              else
2965                {                if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
               if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);  
               }  
2966              }              }
2967            }            }
2968          else          else
2969  #endif  #endif
2970          /* Not UTF-8 mode */          /* Not UTF mode */
2971            {            {
2972            for (fi = min;; fi++)            for (fi = min;; fi++)
2973              {              {
2974              RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              RMATCH(eptr, ecode, offset_top, md, eptrb, RM17);
2975              if (rrc != MATCH_NOMATCH) RRETURN(rrc);              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
2976              if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);              if (fi >= max) RRETURN(MATCH_NOMATCH);
2977                if (eptr >= md->end_subject)
2978                  {
2979                  SCHECK_PARTIAL();
2980                  RRETURN(MATCH_NOMATCH);
2981                  }
2982              c = *eptr++;              c = *eptr++;
2983              if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);  #ifndef COMPILE_PCRE8
2984                if (c > 255)
2985                  {
2986                  if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
2987                  }
2988                else
2989    #endif
2990                  if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
2991              }              }
2992            }            }
2993          /* Control never gets here */          /* Control never gets here */
# Line 1732  for (;;) Line 2999  for (;;)
2999          {          {
3000          pp = eptr;          pp = eptr;
3001    
3002  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
3003          /* UTF-8 mode */          if (utf)
         if (utf8)  
3004            {            {
3005            for (i = min; i < max; i++)            for (i = min; i < max; i++)
3006              {              {
3007              int len = 1;              int len = 1;
3008              if (eptr >= md->end_subject) break;              if (eptr >= md->end_subject)
3009                  {
3010                  SCHECK_PARTIAL();
3011                  break;
3012                  }
3013              GETCHARLEN(c, eptr, len);              GETCHARLEN(c, eptr, len);
3014              if (c > 255)              if (c > 255)
3015                {                {
3016                if (op == OP_CLASS) break;                if (op == OP_CLASS) break;
3017                }                }
3018              else              else
3019                {                if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) break;
               if ((data[c/8] & (1 << (c&7))) == 0) break;  
               }  
3020              eptr += len;              eptr += len;
3021              }              }
3022            for (;;)            for (;;)
3023              {              {
3024              RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              RMATCH(eptr, ecode, offset_top, md, eptrb, RM18);
3025              if (rrc != MATCH_NOMATCH) RRETURN(rrc);              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3026              if (eptr-- == pp) break;        /* Stop if tried at original pos */              if (eptr-- == pp) break;        /* Stop if tried at original pos */
3027              BACKCHAR(eptr);              BACKCHAR(eptr);
# Line 1761  for (;;) Line 3029  for (;;)
3029            }            }
3030          else          else
3031  #endif  #endif
3032            /* Not UTF-8 mode */            /* Not UTF mode */
3033            {            {
3034            for (i = min; i < max; i++)            for (i = min; i < max; i++)
3035              {              {
3036              if (eptr >= md->end_subject) break;              if (eptr >= md->end_subject)
3037                  {
3038                  SCHECK_PARTIAL();
3039                  break;
3040                  }
3041              c = *eptr;              c = *eptr;
3042              if ((data[c/8] & (1 << (c&7))) == 0) break;  #ifndef COMPILE_PCRE8
3043                if (c > 255)
3044                  {
3045                  if (op == OP_CLASS) break;
3046                  }
3047                else
3048    #endif
3049                  if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) break;
3050              eptr++;              eptr++;
3051              }              }
3052            while (eptr >= pp)            while (eptr >= pp)
3053              {              {
3054              RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);              RMATCH(eptr, ecode, offset_top, md, eptrb, RM19);
3055              if (rrc != MATCH_NOMATCH) RRETURN(rrc);              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3056              eptr--;              eptr--;
3057              }              }
# Line 1780  for (;;) Line 3059  for (;;)
3059    
3060          RRETURN(MATCH_NOMATCH);          RRETURN(MATCH_NOMATCH);
3061          }          }
3062    #undef BYTE_MAP
3063        }        }
3064      /* Control never gets here */      /* Control never gets here */
3065    
3066    
3067      /* Match an extended character class. This opcode is encountered only      /* Match an extended character class. This opcode is encountered only
3068      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
3069        mode, because Unicode properties are supported in non-UTF-8 mode. */
3070    
3071  #ifdef SUPPORT_UTF8  #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
3072      case OP_XCLASS:      case OP_XCLASS:
3073        {        {
3074        data = ecode + 1 + LINK_SIZE;                /* Save for matching */        data = ecode + 1 + LINK_SIZE;                /* Save for matching */
# Line 1812  for (;;) Line 3093  for (;;)
3093          case OP_CRMINRANGE:          case OP_CRMINRANGE:
3094          minimize = (*ecode == OP_CRMINRANGE);          minimize = (*ecode == OP_CRMINRANGE);
3095          min = GET2(ecode, 1);          min = GET2(ecode, 1);
3096          max = GET2(ecode, 3);          max = GET2(ecode, 1 + IMM2_SIZE);
3097          if (max == 0) max = INT_MAX;          if (max == 0) max = INT_MAX;
3098          ecode += 5;          ecode += 1 + 2 * IMM2_SIZE;
3099          break;          break;
3100    
3101          default:               /* No repeat follows */          default:               /* No repeat follows */
# Line 1826  for (;;) Line 3107  for (;;)
3107    
3108        for (i = 1; i <= min; i++)        for (i = 1; i <= min; i++)
3109          {          {
3110          if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);          if (eptr >= md->end_subject)
3111          GETCHARINC(c, eptr);            {
3112          if (!_pcre_xclass(c, data)) RRETURN(MATCH_NOMATCH);            SCHECK_PARTIAL();
3113              RRETURN(MATCH_NOMATCH);
3114              }
3115            GETCHARINCTEST(c, eptr);
3116            if (!PRIV(xclass)(c, data, utf)) RRETURN(MATCH_NOMATCH);
3117          }          }
3118    
3119        /* 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 3128  for (;;)
3128          {          {
3129          for (fi = min;; fi++)          for (fi = min;; fi++)
3130            {            {
3131            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);            RMATCH(eptr, ecode, offset_top, md, eptrb, RM20);
3132            if (rrc != MATCH_NOMATCH) RRETURN(rrc);            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3133            if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);            if (fi >= max) RRETURN(MATCH_NOMATCH);
3134            GETCHARINC(c, eptr);            if (eptr >= md->end_subject)
3135            if (!_pcre_xclass(c, data)) RRETURN(MATCH_NOMATCH);              {
3136                SCHECK_PARTIAL();
3137                RRETURN(MATCH_NOMATCH);
3138                }
3139              GETCHARINCTEST(c, eptr);
3140              if (!PRIV(xclass)(c, data, utf)) RRETURN(MATCH_NOMATCH);
3141            }            }
3142          /* Control never gets here */          /* Control never gets here */
3143          }          }
# Line 1860  for (;;) Line 3150  for (;;)
3150          for (i = min; i < max; i++)          for (i = min; i < max; i++)
3151            {            {
3152            int len = 1;            int len = 1;
3153            if (eptr >= md->end_subject) break;            if (eptr >= md->end_subject)
3154            GETCHARLEN(c, eptr, len);              {
3155            if (!_pcre_xclass(c, data)) break;              SCHECK_PARTIAL();
3156                break;
3157                }
3158    #ifdef SUPPORT_UTF
3159              GETCHARLENTEST(c, eptr, len);
3160    #else
3161              c = *eptr;
3162    #endif
3163              if (!PRIV(xclass)(c, data, utf)) break;
3164            eptr += len;            eptr += len;
3165            }            }
3166          for(;;)          for(;;)
3167            {            {
3168            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);            RMATCH(eptr, ecode, offset_top, md, eptrb, RM21);
3169            if (rrc != MATCH_NOMATCH) RRETURN(rrc);            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
3170            if (eptr-- == pp) break;        /* Stop if tried at original pos */            if (eptr-- == pp) break;        /* Stop if tried at original pos */
3171            BACKCHAR(eptr)  #ifdef SUPPORT_UTF
3172              if (utf) BACKCHAR(eptr);
3173    #endif
3174            }            }
3175          RRETURN(MATCH_NOMATCH);          RRETURN(MATCH_NOMATCH);
3176          }          }
# Line 1882  for (;;) Line 3182  for (;;)
3182      /* Match a single character, casefully */      /* Match a single character, casefully */
3183    
3184      case OP_CHAR:      case OP_CHAR:
3185  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF
3186      if (utf8)      if (utf)
3187        {        {
3188        length = 1;        length = 1;
3189        ecode++;        ecode++;
3190        GETCHARLEN(fc, ecode, length);        GETCHARLEN(fc, ecode, length);
3191        if (length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);        if (length > md->end_subject - eptr)
3192        while (length-- > 0) if (*ecode++ != *eptr++) RRETURN(MATCH_NOMATCH);          {
3193            CHECK_PARTIAL();             /* Not SCHECK_PARTIAL() */
3194            RRETURN(MATCH_NOMATCH);
3195            }
3196          while (length-- > 0) if (*ecode++ != RAWUCHARINC(eptr)) RRETURN(MATCH_NOMATCH);
3197        }        }
3198      else      else
3199  #endif  #endif
3200        /* Not UTF mode */
     /* Non-UTF-8 mode */  
3201        {        {
3202        if (md->end_subject - eptr < 1) RRETURN(MATCH_NOMATCH);        if (md->end_subject - eptr < 1)
3203            {
3204            SCHECK_PARTIAL();            /* This one can use SCHECK_PARTIAL() */
3205            RRETURN(MATCH_NOMATCH);
3206            }
3207        if (ecode[1] != *eptr++) RRETURN(MATCH_NOMATCH);        if (ecode[1] != *eptr++) RRETURN(MATCH_NOMATCH);
3208        ecode += 2;        ecode += 2;
3209        }        }
3210      break;      break;
3211    
3212      /* Match a single character, caselessly */      /* Match a single character, caselessly. If we are at the end of the
3213        subject, give up immediately. */
3214    
3215        case OP_CHARI:
3216        if (eptr >= md->end_subject)
3217          {
3218          SCHECK_PARTIAL();
3219          RRETURN(MATCH_NOMATCH);
3220          }
3221    
3222      case OP_CHARNC:  #ifdef SUPPORT_UTF
3223  #ifdef SUPPORT_UTF8      if (utf)
     if (utf8)  
3224        {        {
3225        length = 1;        length = 1;
3226        ecode++;        ecode++;
3227        GETCHARLEN(fc, ecode, length);        GETCHARLEN(fc, ecode, length);
3228    
       if (length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);  
   
3229        /* 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
3230        can use the fast lookup table. */        we know that its other case must also be one byte long, so we can use the
3231          fast lookup table. We know that there is at least one byte left in the
3232          subject. */
3233    
3234        if (fc < 128)        if (fc < 128)
3235          {          {
3236          if (md->lcc[*ecode++] != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);          pcre_uint32 cc = RAWUCHAR(eptr);
3237            if (md->lcc[fc] != TABLE_GET(cc, md->lcc, cc)) RRETURN(MATCH_NOMATCH);
3238            ecode++;
3239            eptr++;
3240          }          }
3241    
3242        /* Otherwise we must pick up the subject character */        /* Otherwise we must pick up the subject character. Note that we cannot
3243          use the value of "length" to check for sufficient bytes left, because the
3244          other case of the character may have more or fewer bytes.  */
3245    
3246        else        else
3247          {          {
3248          int dc;          pcre_uint32 dc;
3249          GETCHARINC(dc, eptr);          GETCHARINC(dc, eptr);
3250          ecode += length;          ecode += length;
3251    
# Line 1936  for (;;) Line 3255  for (;;)
3255          if (fc != dc)          if (fc != dc)
3256            {            {
3257  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
3258            if (dc != _pcre_ucp_othercase(fc))            if (dc != UCD_OTHERCASE(fc))
3259  #endif  #endif
3260              RRETURN(MATCH_NOMATCH);              RRETURN(MATCH_NOMATCH);
3261            }            }
3262          }          }
3263        }        }
3264      else      else
3265  #endif   /* SUPPORT_UTF8 */  #endif   /* SUPPORT_UTF */
3266    
3267      /* Non-UTF-8 mode */      /* Not UTF mode */
3268        {        {
3269        if (md->end_subject - eptr < 1) RRETURN(MATCH_NOMATCH);        if (TABLE_GET(ecode[1], md->lcc, ecode[1])
3270        if (md->lcc[ecode[1]] != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);            != TABLE_GET(*eptr, md->lcc, *eptr)) RRETURN(MATCH_NOMATCH);
3271          eptr++;
3272        ecode += 2;        ecode += 2;
3273        }        }
3274      break;      break;
3275    
3276      /* Match a single character repeatedly; different opcodes share code. */      /* Match a single character repeatedly. */
3277    
3278      case OP_EXACT:      case OP_EXACT:
3279        case OP_EXACTI:
3280      min = max = GET2(ecode, 1);      min = max = GET2(ecode, 1);
3281      ecode += 3;      ecode += 1 + IMM2_SIZE;
3282      goto REPEATCHAR;      goto REPEATCHAR;
3283    
3284        case OP_POSUPTO:
3285        case OP_POSUPTOI:
3286        possessive = TRUE;
3287        /* Fall through */
3288    
3289      case OP_UPTO:      case OP_UPTO:
3290        case OP_UPTOI:
3291      case OP_MINUPTO:      case OP_MINUPTO:
3292        case OP_MINUPTOI:
3293      min = 0;      min = 0;
3294      max = GET2(ecode, 1);      max = GET2(ecode, 1);
3295      minimize = *ecode == OP_MINUPTO;      minimize = *ecode == OP_MINUPTO || *ecode == OP_MINUPTOI;
3296      ecode += 3;      ecode += 1 + IMM2_SIZE;
3297        goto REPEATCHAR;
3298    
3299        case OP_POSSTAR:
3300        case OP_POSSTARI:
3301        possessive = TRUE;
3302        min = 0;
3303        max = INT_MAX;
3304        ecode++;
3305        goto REPEATCHAR;