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

Diff of /code/trunk/pcre_exec.c

Parent Directory | Revision Log | View Patch Patch

-revision 85 by nigel,
Sat Feb 24 21:41:13 2007 UTC
+revision 933 by ph10,
Sat Feb 25 12:18:23 2012 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2005 University of Cambridge
+            Copyright (c) 1997-2012 University of Cambridge
  -----------------------------------------------------------------------------
  Redistribution and use in source and binary forms, with or without
 Line 42 
 POSSIBILITY OF SUCH DAMAGE.
  pattern matching using an NFA algorithm, trying to mimic Perl as closely as
  possible. There are also some static supporting functions. */
+ #ifdef HAVE_CONFIG_H
+ #include "config.h"
+ #endif
+ #define NLBLOCK md             /* Block containing newline information */
+ #define PSSTART start_subject  /* Field containing processed string start */
+ #define PSEND   end_subject    /* Field containing processed string end */
  #include "pcre_internal.h"
+ /* Undefine some potentially clashing cpp symbols */
- /* Structure for building a chain of data that actually lives on the
+ #undef min
- 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;
-   const uschar *epb_saved_eptr;
- } eptrblock;
- /* Flag bits for the match() function */
+ /* Values for setting in md->match_function_type to indicate two special types
+ of call to match(). We do it this way to save on using another stack variable,
+ as stack usage is to be discouraged. */
- #define match_condassert   0x01    /* Called to check a condition assertion */
+ #define MATCH_CONDASSERT     1  /* Called to check a condition assertion */
- #define match_isgroup      0x02    /* Set if start of bracketed group */
+ #define MATCH_CBEGROUP       2  /* Could-be-empty unlimited repeat group */
  /* Non-error returns from the match() function. Error returns are externally
  defined PCRE_ERROR_xxx codes, which are all negative. */
-Line 68 
 defined PCRE_ERROR_xxx codes, which are
+Line 70 
 defined PCRE_ERROR_xxx codes, which are
  #define MATCH_MATCH        1
  #define MATCH_NOMATCH      0
+ /* Special internal returns from the match() function. Make them sufficiently
+ negative to avoid the external error codes. */
+ #define MATCH_ACCEPT       (-999)
+ #define MATCH_COMMIT       (-998)
+ #define MATCH_KETRPOS      (-997)
+ #define MATCH_ONCE         (-996)
+ #define MATCH_PRUNE        (-995)
+ #define MATCH_SKIP         (-994)
+ #define MATCH_SKIP_ARG     (-993)
+ #define MATCH_THEN         (-992)
  /* Maximum number of ints of offset to save on the stack for recursive calls.
  If the offset vector is bigger, malloc is used. This should be a multiple of 3,
  because the offset vector is always a multiple of 3 long. */
-Line 81 
 static const char rep_max[] = { 0, 0, 0,
+Line 95 
 static const char rep_max[] = { 0, 0, 0,
- #ifdef DEBUG
+ #ifdef PCRE_DEBUG
  /*************************************************
  *        Debugging function to print chars       *
  *************************************************/
-Line 99 
 Returns:     nothing
+Line 113 
 Returns:     nothing
  */
  static void
- 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)
  {
- int c;
+ unsigned int c;
  if (is_subject && length > md->end_subject - p) length = md->end_subject - p;
  while (length-- > 0)
    if (isprint(c = *(p++))) printf("%c", c); else printf("\\x%02x", c);
-Line 114 
 while (length-- > 0)
+Line 128 
 while (length-- > 0)
  *          Match a back-reference                *
  *************************************************/
- /* 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
- than the number of characters left in the string, so the match fails.
+ negative, so the match always fails. However, in JavaScript compatibility mode,
+ the length passed is zero. Note that in caseless UTF-8 mode, the number of
+ subject bytes matched may be different to the number of reference bytes.
  Arguments:
    offset      index into the offset vector
-   eptr        points into the subject
+   eptr        pointer into the subject
-   length      length to be matched
+   length      length of reference to be matched (number of bytes)
    md          points to match data block
-   ims         the ims flags
+   caseless    TRUE if caseless
- Returns:      TRUE if matched
+ Returns:      >= 0 the number of subject bytes matched
+               -1 no match
+               -2 partial match; always given if at end subject
  */
- static BOOL
+ static int
- match_ref(int offset, register const uschar *eptr, int length, match_data *md,
+ match_ref(int offset, register PCRE_PUCHAR eptr, int length, match_data *md,
-   unsigned long int ims)
+   BOOL caseless)
  {
- const uschar *p = md->start_subject + md->offset_vector[offset];
+ PCRE_PUCHAR eptr_start = eptr;
+ register PCRE_PUCHAR p = md->start_subject + md->offset_vector[offset];
- #ifdef DEBUG
+ #ifdef PCRE_DEBUG
  if (eptr >= md->end_subject)
    printf("matching subject <null>");
  else
-Line 146 
 pchars(p, length, FALSE, md);
+Line 165 
 pchars(p, length, FALSE, md);
  printf("\n");
  #endif
- /* Always fail if not enough characters left */
+ /* Always fail if reference not set (and not JavaScript compatible - in that
+ case the length is passed as zero). */
- if (length > md->end_subject - eptr) return FALSE;
+ if (length < 0) return -1;
- /* Separate the caselesss case for speed */
+ /* Separate the caseless case for speed. In UTF-8 mode we can only do this
+ properly if Unicode properties are supported. Otherwise, we can check only
+ ASCII characters. */
- if ((ims & PCRE_CASELESS) != 0)
+ if (caseless)
    {
-   while (length-- > 0)
+ #ifdef SUPPORT_UTF
-     if (md->lcc[*p++] != md->lcc[*eptr++]) return FALSE;
+ #ifdef SUPPORT_UCP
+   if (md->utf)
+     {
+     /* Match characters up to the end of the reference. NOTE: the number of
+     bytes matched may differ, because there are some characters whose upper and
+     lower case versions code as different numbers of bytes. For example, U+023A
+     (2 bytes in UTF-8) is the upper case version of U+2C65 (3 bytes in UTF-8);
+     a sequence of 3 of the former uses 6 bytes, as does a sequence of two of
+     the latter. It is important, therefore, to check the length along the
+     reference, not along the subject (earlier code did this wrong). */
+     PCRE_PUCHAR endptr = p + length;
+     while (p < endptr)
+       {
+       int c, d;
+       if (eptr >= md->end_subject) return -2;   /* Partial match */
+       GETCHARINC(c, eptr);
+       GETCHARINC(d, p);
+       if (c != d && c != UCD_OTHERCASE(d)) return -1;
+       }
+     }
+   else
+ #endif
+ #endif
+   /* The same code works when not in UTF-8 mode and in UTF-8 mode when there
+   is no UCP support. */
+     {
+     while (length-- > 0)
+       {
+       if (eptr >= md->end_subject) return -2;   /* Partial match */
+       if (TABLE_GET(*p, md->lcc, *p) != TABLE_GET(*eptr, md->lcc, *eptr)) return -1;
+       p++;
+       eptr++;
+       }
+     }
    }
+ /* In the caseful case, we can just compare the bytes, whether or not we
+ are in UTF-8 mode. */
  else
-   { while (length-- > 0) if (*p++ != *eptr++) return FALSE; }
+   {
+   while (length-- > 0)
+     {
+     if (eptr >= md->end_subject) return -2;   /* Partial match */
+     if (*p++ != *eptr++) return -1;
+     }
+   }
- return TRUE;
+ return (int)(eptr - eptr_start);
  }
-Line 169 
 return TRUE;
+Line 236 
 return TRUE;
  ****************************************************************************
                     RECURSION IN THE match() FUNCTION
- The match() function is highly recursive. Some regular expressions can cause
+ The match() function is highly recursive, though not every recursive call
- it to recurse thousands of times. I was writing for Unix, so I just let it
+ increases the recursive depth. Nevertheless, some regular expressions can cause
- call itself recursively. This uses the stack for saving everything that has
+ it to recurse to a great depth. I was writing for Unix, so I just let it call
- to be saved for a recursive call. On Unix, the stack can be large, and this
+ itself recursively. This uses the stack for saving everything that has to be
- works fine.
+ saved for a recursive call. On Unix, the stack can be large, and this works
+ fine.
- It turns out that on non-Unix systems there are problems with programs that
- use a lot of stack. (This despite the fact that every last chip has oodles
+ It turns out that on some non-Unix-like systems there are problems with
- of memory these days, and techniques for extending the stack have been known
+ programs that use a lot of stack. (This despite the fact that every last chip
- for decades.) So....
+ has oodles of memory these days, and techniques for extending the stack have
+ been known for decades.) So....
  There is a fudge, triggered by defining NO_RECURSE, which avoids recursive
  calls by keeping local variables that need to be preserved in blocks of memory
- 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
  achieve this so that the actual code doesn't look very different to what it
  always used to.
+ The original heap-recursive code used longjmp(). However, it seems that this
+ can be very slow on some operating systems. Following a suggestion from Stan
+ Switzer, the use of longjmp() has been abolished, at the cost of having to
+ provide a unique number for each call to RMATCH. There is no way of generating
+ a sequence of numbers at compile time in C. I have given them names, to make
+ them stand out more clearly.
+ Crude tests on x86 Linux show a small speedup of around 5-8%. However, on
+ FreeBSD, avoiding longjmp() more than halves the time taken to run the standard
+ tests. Furthermore, not using longjmp() means that local dynamic variables
+ don't have indeterminate values; this has meant that the frame size can be
+ reduced because the result can be "passed back" by straight setting of the
+ variable instead of being passed in the frame.
  ****************************************************************************
  ***************************************************************************/
+ /* Numbers for RMATCH calls. When this list is changed, the code at HEAP_RETURN
+ below must be updated in sync.  */
- /* These versions of the macros use the stack, as normal */
+ enum { RM1=1, RM2,  RM3,  RM4,  RM5,  RM6,  RM7,  RM8,  RM9,  RM10,
+        RM11,  RM12, RM13, RM14, RM15, RM16, RM17, RM18, RM19, RM20,
+        RM21,  RM22, RM23, RM24, RM25, RM26, RM27, RM28, RM29, RM30,
+        RM31,  RM32, RM33, RM34, RM35, RM36, RM37, RM38, RM39, RM40,
+        RM41,  RM42, RM43, RM44, RM45, RM46, RM47, RM48, RM49, RM50,
+        RM51,  RM52, RM53, RM54, RM55, RM56, RM57, RM58, RM59, RM60,
+        RM61,  RM62, RM63, RM64, RM65, RM66 };
+ /* These versions of the macros use the stack, as normal. There are debugging
+ versions and production versions. Note that the "rw" argument of RMATCH isn't
+ actually used in this definition. */
  #ifndef NO_RECURSE
  #define REGISTER register
- #define RMATCH(rx,ra,rb,rc,rd,re,rf,rg) rx = match(ra,rb,rc,rd,re,rf,rg)
+ #ifdef PCRE_DEBUG
+ #define RMATCH(ra,rb,rc,rd,re,rw) \
+   { \
+   printf("match() called in line %d\n", __LINE__); \
+   rrc = match(ra,rb,mstart,rc,rd,re,rdepth+1); \
+   printf("to line %d\n", __LINE__); \
+   }
+ #define RRETURN(ra) \
+   { \
+   printf("match() returned %d from line %d ", ra, __LINE__); \
+   return ra; \
+   }
+ #else
+ #define RMATCH(ra,rb,rc,rd,re,rw) \
+   rrc = match(ra,rb,mstart,rc,rd,re,rdepth+1)
  #define RRETURN(ra) return ra
+ #endif
  #else
- /* 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
- 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
- match(), which never changes. */
+ argument of match(), which never changes. */
  #define REGISTER
- #define RMATCH(rx,ra,rb,rc,rd,re,rf,rg)\
+ #define RMATCH(ra,rb,rc,rd,re,rw)\
    {\
-   heapframe *newframe = (pcre_stack_malloc)(sizeof(heapframe));\
+   heapframe *newframe = frame->Xnextframe;\
-   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->Xprevframe = frame;\
-     frame = newframe;\
-     DPRINTF(("restarting from line %d\n", __LINE__));\
-     goto HEAP_RECURSE;\
-     }\
-   else\
      {\
-     DPRINTF(("longjumped back to line %d\n", __LINE__));\
+     newframe = (heapframe *)(PUBL(stack_malloc))(sizeof(heapframe));\
-     frame = md->thisframe;\
+     if (newframe == NULL) RRETURN(PCRE_ERROR_NOMEMORY);\
-     rx = frame->Xresult;\
+     newframe->Xnextframe = NULL;\
+     frame->Xnextframe = newframe;\
      }\
+   frame->Xwhere = rw;\
+   newframe->Xeptr = ra;\
+   newframe->Xecode = rb;\
+   newframe->Xmstart = mstart;\
+   newframe->Xoffset_top = rc;\
+   newframe->Xeptrb = re;\
+   newframe->Xrdepth = frame->Xrdepth + 1;\
+   newframe->Xprevframe = frame;\
+   frame = newframe;\
+   DPRINTF(("restarting from line %d\n", __LINE__));\
+   goto HEAP_RECURSE;\
+   L_##rw:\
+   DPRINTF(("jumped back to line %d\n", __LINE__));\
    }
  #define RRETURN(ra)\
    {\
-   heapframe *newframe = frame;\
+   heapframe *oldframe = frame;\
-   frame = newframe->Xprevframe;\
+   frame = oldframe->Xprevframe;\
-   (pcre_stack_free)(newframe);\
    if (frame != NULL)\
      {\
-     frame->Xresult = ra;\
+     rrc = ra;\
-     md->thisframe = frame;\
+     goto HEAP_RETURN;\
-     longjmp(frame->Xwhere, 1);\
      }\
    return ra;\
    }
-Line 247 
 match(), which never changes. */
+Line 357 
 match(), which never changes. */
  typedef struct heapframe {
    struct heapframe *Xprevframe;
+   struct heapframe *Xnextframe;
    /* Function arguments that may change */
-   const uschar *Xeptr;
+   PCRE_PUCHAR Xeptr;
-   const uschar *Xecode;
+   const pcre_uchar *Xecode;
+   PCRE_PUCHAR Xmstart;
    int Xoffset_top;
-   long int Xims;
    eptrblock *Xeptrb;
-   int Xflags;
+   unsigned int Xrdepth;
    /* Function local variables */
-   const uschar *Xcallpat;
+   PCRE_PUCHAR Xcallpat;
-   const uschar *Xcharptr;
+ #ifdef SUPPORT_UTF
-   const uschar *Xdata;
+   PCRE_PUCHAR Xcharptr;
-   const uschar *Xnext;
+ #endif
-   const uschar *Xpp;
+   PCRE_PUCHAR Xdata;
-   const uschar *Xprev;
+   PCRE_PUCHAR Xnext;
-   const uschar *Xsaved_eptr;
+   PCRE_PUCHAR Xpp;
+   PCRE_PUCHAR Xprev;
+   PCRE_PUCHAR Xsaved_eptr;
    recursion_info Xnew_recursive;
    BOOL Xcur_is_word;
    BOOL Xcondition;
-   BOOL Xminimize;
    BOOL Xprev_is_word;
-   unsigned long int Xoriginal_ims;
  #ifdef SUPPORT_UCP
    int Xprop_type;
+   int Xprop_value;
    int Xprop_fail_result;
-   int Xprop_category;
+   int Xoclength;
-   int Xprop_chartype;
+   pcre_uchar Xocchars[6];
-   int Xprop_othercase;
-   int Xprop_test_against;
-   int *Xprop_test_variable;
  #endif
+   int Xcodelink;
    int Xctype;
-   int Xfc;
+   unsigned int Xfc;
    int Xfi;
    int Xlength;
    int Xmax;
-Line 301 
 typedef struct heapframe {
+Line 410 
 typedef struct heapframe {
    eptrblock Xnewptrb;
-   /* Place to pass back result, and where to jump back to */
+   /* Where to jump back to */
-   int  Xresult;
+   int Xwhere;
-   jmp_buf Xwhere;
  } heapframe;
-Line 320 
 typedef struct heapframe {
+Line 428 
 typedef struct heapframe {
  *         Match from current position            *
  *************************************************/
- /* 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
  returns a negative (error) response, the outer incarnation must also return the
- same response.
+ same response. */
+ /* These macros pack up tests that are used for partial matching, and which
+ appear several times in the code. We set the "hit end" flag if the pointer is
+ at the end of the subject and also past the start of the subject (i.e.
+ something has been matched). For hard partial matching, we then return
+ immediately. The second one is used when we already know we are past the end of
+ the subject. */
+ #define CHECK_PARTIAL()\
+   if (md->partial != 0 && eptr >= md->end_subject && \
+       eptr > md->start_used_ptr) \
+     { \
+     md->hitend = TRUE; \
+     if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL); \
+     }
- Performance note: It might be tempting to extract commonly used fields from the
+ #define SCHECK_PARTIAL()\
- md structure (e.g. utf8, end_subject) into individual variables to improve
+   if (md->partial != 0 && eptr > md->start_used_ptr) \
+     { \
+     md->hitend = TRUE; \
+     if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL); \
+     }
+ /* Performance note: It might be tempting to extract commonly used fields from
+ the md structure (e.g. utf, end_subject) into individual variables to improve
  performance. Tests using gcc on a SPARC disproved this; in the first case, it
  made performance worse.
  Arguments:
-    eptr        pointer in subject
+    eptr        pointer to current character in subject
-    ecode       position in code
+    ecode       pointer to current position in compiled code
+    mstart      pointer to the current match start position (can be modified
+                  by encountering \K)
     offset_top  current top pointer
     md          pointer to "static" info for the match
-    ims         current /i, /m, and /s options
     eptrb       pointer to chain of blocks containing eptr at start of
                   brackets - for testing for empty matches
-    flags       can contain
+    rdepth      the recursion depth
-                  match_condassert - this is an assertion condition
-                  match_isgroup - this is the start of a bracketed group
  Returns:       MATCH_MATCH if matched            )  these values are >= 0
                 MATCH_NOMATCH if failed to match  )
+                a negative MATCH_xxx value for PRUNE, SKIP, etc
                 a negative PCRE_ERROR_xxx value if aborted by an error condition
-                  (e.g. stopped by recursion limit)
+                  (e.g. stopped by repeated call or recursion limit)
  */
  static int
- match(REGISTER const uschar *eptr, REGISTER const uschar *ecode,
+ match(REGISTER PCRE_PUCHAR eptr, REGISTER const pcre_uchar *ecode,
-   int offset_top, match_data *md, unsigned long int ims, eptrblock *eptrb,
+   PCRE_PUCHAR mstart, int offset_top, match_data *md, eptrblock *eptrb,
-   int flags)
+   unsigned int rdepth)
  {
  /* These variables do not need to be preserved over recursion in this function,
- 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
- because they are used a lot in loops. */
+ "register" because they are used a lot in loops. */
- register int  rrc;    /* Returns from recursive calls */
+ register int  rrc;         /* Returns from recursive calls */
- register int  i;      /* Used for loops not involving calls to RMATCH() */
+ register int  i;           /* Used for loops not involving calls to RMATCH() */
- register int  c;      /* Character values not kept over RMATCH() calls */
+ register unsigned int c;   /* Character values not kept over RMATCH() calls */
- register BOOL utf8;   /* Local copy of UTF-8 flag for speed */
+ register BOOL utf;         /* Local copy of UTF flag for speed */
+ BOOL minimize, possessive; /* Quantifier options */
+ BOOL caseless;
+ int condcode;
  /* When recursion is not being used, all "local" variables that have to be
- 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
- 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
- heap whenever RMATCH() does a "recursion". See the macro definitions above. */
+ whenever RMATCH() does a "recursion". See the macro definitions above. Putting
+ the top-level on the stack rather than malloc-ing them all gives a performance
+ boost in many cases where there is not much "recursion". */
  #ifdef NO_RECURSE
- heapframe *frame = (pcre_stack_malloc)(sizeof(heapframe));
+ heapframe *frame = (heapframe *)md->match_frames_base;
- frame->Xprevframe = NULL;            /* Marks the top level */
  /* Copy in the original argument variables */
  frame->Xeptr = eptr;
  frame->Xecode = ecode;
+ frame->Xmstart = mstart;
  frame->Xoffset_top = offset_top;
- frame->Xims = ims;
  frame->Xeptrb = eptrb;
- frame->Xflags = flags;
+ frame->Xrdepth = rdepth;
  /* This is where control jumps back to to effect "recursion" */
-Line 390 
 HEAP_RECURSE:
+Line 523 
 HEAP_RECURSE:
  #define eptr               frame->Xeptr
  #define ecode              frame->Xecode
+ #define mstart             frame->Xmstart
  #define offset_top         frame->Xoffset_top
- #define ims                frame->Xims
  #define eptrb              frame->Xeptrb
- #define flags              frame->Xflags
+ #define rdepth             frame->Xrdepth
  /* Ditto for the local variables */
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
  #define charptr            frame->Xcharptr
  #endif
  #define callpat            frame->Xcallpat
+ #define codelink           frame->Xcodelink
  #define data               frame->Xdata
  #define next               frame->Xnext
  #define pp                 frame->Xpp
-Line 411 
 HEAP_RECURSE:
+Line 545 
 HEAP_RECURSE:
  #define cur_is_word        frame->Xcur_is_word
  #define condition          frame->Xcondition
- #define minimize           frame->Xminimize
  #define prev_is_word       frame->Xprev_is_word
- #define original_ims       frame->Xoriginal_ims
  #ifdef SUPPORT_UCP
  #define prop_type          frame->Xprop_type
+ #define prop_value         frame->Xprop_value
  #define prop_fail_result   frame->Xprop_fail_result
- #define prop_category      frame->Xprop_category
+ #define oclength           frame->Xoclength
- #define prop_chartype      frame->Xprop_chartype
+ #define occhars            frame->Xocchars
- #define prop_othercase     frame->Xprop_othercase
- #define prop_test_against  frame->Xprop_test_against
- #define prop_test_variable frame->Xprop_test_variable
  #endif
  #define ctype              frame->Xctype
-Line 447 
 HEAP_RECURSE:
+Line 576 
 HEAP_RECURSE:
  get preserved during recursion in the normal way. In this environment, fi and
  i, and fc and c, can be the same variables. */
- #else
+ #else         /* NO_RECURSE not defined */
  #define fi i
  #define fc c
+ /* Many of the following variables are used only in small blocks of the code.
+ My normal style of coding would have declared them within each of those blocks.
+ However, in order to accommodate the version of this code that uses an external
+ "stack" implemented on the heap, it is easier to declare them all here, so the
+ declarations can be cut out in a block. The only declarations within blocks
+ below are for variables that do not have to be preserved over a recursive call
+ to RMATCH(). */
- #ifdef SUPPORT_UTF8                /* Many of these variables are used ony */
+ #ifdef SUPPORT_UTF
- const uschar *charptr;             /* small blocks of the code. My normal  */
+ const pcre_uchar *charptr;
- #endif                             /* style of coding would have declared  */
+ #endif
- const uschar *callpat;             /* them within each of those blocks.    */
+ const pcre_uchar *callpat;
- const uschar *data;                /* However, in order to accommodate the */
+ const pcre_uchar *data;
- const uschar *next;                /* version of this code that uses an    */
+ const pcre_uchar *next;
- const uschar *pp;                  /* external "stack" implemented on the  */
+ PCRE_PUCHAR       pp;
- const uschar *prev;                /* heap, it is easier to declare them   */
+ const pcre_uchar *prev;
- const uschar *saved_eptr;          /* all here, so the declarations can    */
+ PCRE_PUCHAR       saved_eptr;
-                                    /* be cut out in a block. The only      */
- recursion_info new_recursive;      /* declarations within blocks below are */
+ recursion_info new_recursive;
-                                    /* for variables that do not have to    */
- BOOL cur_is_word;                  /* be preserved over a recursive call   */
- BOOL condition;                    /* to RMATCH().                         */
- BOOL minimize;
- BOOL prev_is_word;
- unsigned long int original_ims;
+ BOOL cur_is_word;
+ BOOL condition;
+ BOOL prev_is_word;
  #ifdef SUPPORT_UCP
  int prop_type;
+ int prop_value;
  int prop_fail_result;
- int prop_category;
+ int oclength;
- int prop_chartype;
+ pcre_uchar occhars[6];
- int prop_othercase;
- int prop_test_against;
- int *prop_test_variable;
  #endif
+ int codelink;
  int ctype;
  int length;
  int max;
-Line 493 
 int save_offset1, save_offset2, save_off
+Line 625 
 int save_offset1, save_offset2, save_off
  int stacksave[REC_STACK_SAVE_MAX];
  eptrblock newptrb;
- #endif
+ /* There is a special fudge for calling match() in a way that causes it to
+ measure the size of its basic stack frame when the stack is being used for
+ recursion. The second argument (ecode) being NULL triggers this behaviour. It
+ cannot normally ever be NULL. The return is the negated value of the frame
+ size. */
+ if (ecode == NULL)
+   {
+   if (rdepth == 0)
+     return match((PCRE_PUCHAR)&rdepth, NULL, NULL, 0, NULL, NULL, 1);
+   else
+     {
+     int len = (char *)&rdepth - (char *)eptr;
+     return (len > 0)? -len : len;
+     }
+   }
+ #endif     /* NO_RECURSE */
+ /* To save space on the stack and in the heap frame, I have doubled up on some
+ of the local variables that are used only in localised parts of the code, but
+ still need to be preserved over recursive calls of match(). These macros define
+ the alternative names that are used. */
+ #define allow_zero    cur_is_word
+ #define cbegroup      condition
+ #define code_offset   codelink
+ #define condassert    condition
+ #define matched_once  prev_is_word
+ #define foc           number
+ #define save_mark     data
  /* These statements are here to stop the compiler complaining about unitialized
  variables. */
  #ifdef SUPPORT_UCP
+ prop_value = 0;
  prop_fail_result = 0;
- prop_test_against = 0;
- prop_test_variable = NULL;
  #endif
- /* OK, now we can get on with the real code of the function. Recursion is
- specified by the macros RMATCH and RRETURN. When NO_RECURSE is *not* defined,
- these just turn into a recursive call to match() and a "return", respectively.
- However, RMATCH isn't like a function call because it's quite a complicated
- macro. It has to be used in one particular way. This shouldn't, however, impact
- performance when true recursion is being used. */
- if (md->match_call_count++ >= md->match_limit) RRETURN(PCRE_ERROR_MATCHLIMIT);
+ /* This label is used for tail recursion, which is used in a few cases even
+ when NO_RECURSE is not defined, in order to reduce the amount of stack that is
+ used. Thanks to Ian Taylor for noticing this possibility and sending the
+ original patch. */
+ TAIL_RECURSE:
+ /* OK, now we can get on with the real code of the function. Recursive calls
+ are specified by the macro RMATCH and RRETURN is used to return. When
+ NO_RECURSE is *not* defined, these just turn into a recursive call to match()
+ and a "return", respectively (possibly with some debugging if PCRE_DEBUG is
+ defined). However, RMATCH isn't like a function call because it's quite a
+ complicated macro. It has to be used in one particular way. This shouldn't,
+ however, impact performance when true recursion is being used. */
+ #ifdef SUPPORT_UTF
+ utf = md->utf;       /* Local copy of the flag */
+ #else
+ utf = FALSE;
+ #endif
- original_ims = ims;    /* Save for resetting on ')' */
+ /* First check that we haven't called match() too many times, or that we
- utf8 = md->utf8;       /* Local copy of the flag */
+ haven't exceeded the recursive call limit. */
+ if (md->match_call_count++ >= md->match_limit) RRETURN(PCRE_ERROR_MATCHLIMIT);
+ if (rdepth >= md->match_limit_recursion) RRETURN(PCRE_ERROR_RECURSIONLIMIT);
- /* At the start of a bracketed group, add the current subject pointer to the
+ /* At the start of a group with an unlimited repeat that may match an empty
- stack of such pointers, to be re-instated at the end of the group when we hit
+ string, the variable md->match_function_type is set to MATCH_CBEGROUP. It is
- the closing ket. When match() is called in other circumstances, we don't add to
+ done this way to save having to use another function argument, which would take
- this stack. */
+ up space on the stack. See also MATCH_CONDASSERT below.
+ When MATCH_CBEGROUP is set, add the current subject pointer to the chain of
+ such remembered pointers, to be checked when we hit the closing ket, in order
+ to break infinite loops that match no characters. When match() is called in
+ other circumstances, don't add to the chain. The MATCH_CBEGROUP feature must
+ NOT be used with tail recursion, because the memory block that is used is on
+ the stack, so a new one may be required for each match(). */
- if ((flags & match_isgroup) != 0)
+ if (md->match_function_type == MATCH_CBEGROUP)
    {
-   newptrb.epb_prev = eptrb;
    newptrb.epb_saved_eptr = eptr;
+   newptrb.epb_prev = eptrb;
    eptrb = &newptrb;
+   md->match_function_type = 0;
    }
- /* Now start processing the operations. */
+ /* Now start processing the opcodes. */
  for (;;)
    {
+   minimize = possessive = FALSE;
    op = *ecode;
-   minimize = FALSE;
-   /* For partial matching, remember if we ever hit the end of the subject after
+   switch(op)
-   matching at least one subject character. */
+     {
+     case OP_MARK:
+     md->nomatch_mark = ecode + 2;
+     md->mark = NULL;    /* In case previously set by assertion */
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
+       eptrb, RM55);
+     if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
+          md->mark == NULL) md->mark = ecode + 2;
+     /* A return of MATCH_SKIP_ARG means that matching failed at SKIP with an
+     argument, and we must check whether that argument matches this MARK's
+     argument. It is passed back in md->start_match_ptr (an overloading of that
+     variable). If it does match, we reset that variable to the current subject
+     position and return MATCH_SKIP. Otherwise, pass back the return code
+     unaltered. */
+     else if (rrc == MATCH_SKIP_ARG &&
+         STRCMP_UC_UC(ecode + 2, md->start_match_ptr) == 0)
+       {
+       md->start_match_ptr = eptr;
+       RRETURN(MATCH_SKIP);
+       }
+     RRETURN(rrc);
-   if (md->partial &&
+     case OP_FAIL:
-       eptr >= md->end_subject &&
+     RRETURN(MATCH_NOMATCH);
-       eptr > md->start_match)
-     md->hitend = TRUE;
-   /* 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. */
-   if (op > OP_BRA)
+     /* COMMIT overrides PRUNE, SKIP, and THEN */
-     {
-     number = op - OP_BRA;
+     case OP_COMMIT:
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
+       eptrb, RM52);
+     if (rrc != MATCH_NOMATCH && rrc != MATCH_PRUNE &&
+         rrc != MATCH_SKIP && rrc != MATCH_SKIP_ARG &&
+         rrc != MATCH_THEN)
+       RRETURN(rrc);
+     RRETURN(MATCH_COMMIT);
+     /* PRUNE overrides THEN */
+     case OP_PRUNE:
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
+       eptrb, RM51);
+     if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
+     RRETURN(MATCH_PRUNE);
+     case OP_PRUNE_ARG:
+     md->nomatch_mark = ecode + 2;
+     md->mark = NULL;    /* In case previously set by assertion */
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
+       eptrb, RM56);
+     if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
+          md->mark == NULL) md->mark = ecode + 2;
+     if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
+     RRETURN(MATCH_PRUNE);
+     /* SKIP overrides PRUNE and THEN */
+     case OP_SKIP:
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
+       eptrb, RM53);
+     if (rrc != MATCH_NOMATCH && rrc != MATCH_PRUNE && rrc != MATCH_THEN)
+       RRETURN(rrc);
+     md->start_match_ptr = eptr;   /* Pass back current position */
+     RRETURN(MATCH_SKIP);
+     /* Note that, for Perl compatibility, SKIP with an argument does NOT set
+     nomatch_mark. There is a flag that disables this opcode when re-matching a
+     pattern that ended with a SKIP for which there was not a matching MARK. */
+     case OP_SKIP_ARG:
+     if (md->ignore_skip_arg)
+       {
+       ecode += PRIV(OP_lengths)[*ecode] + ecode[1];
+       break;
+       }
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
+       eptrb, RM57);
+     if (rrc != MATCH_NOMATCH && rrc != MATCH_PRUNE && rrc != MATCH_THEN)
+       RRETURN(rrc);
+     /* Pass back the current skip name by overloading md->start_match_ptr and
+     returning the special MATCH_SKIP_ARG return code. This will either be
+     caught by a matching MARK, or get to the top, where it causes a rematch
+     with the md->ignore_skip_arg flag set. */
+     md->start_match_ptr = ecode + 2;
+     RRETURN(MATCH_SKIP_ARG);
+     /* For THEN (and THEN_ARG) we pass back the address of the opcode, so that
+     the branch in which it occurs can be determined. Overload the start of
+     match pointer to do this. */
+     case OP_THEN:
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
+       eptrb, RM54);
+     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     md->start_match_ptr = ecode;
+     RRETURN(MATCH_THEN);
+     case OP_THEN_ARG:
+     md->nomatch_mark = ecode + 2;
+     md->mark = NULL;    /* In case previously set by assertion */
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top,
+       md, eptrb, RM58);
+     if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
+          md->mark == NULL) md->mark = ecode + 2;
+     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     md->start_match_ptr = ecode;
+     RRETURN(MATCH_THEN);
+     /* Handle an atomic group that does not contain any capturing parentheses.
+     This can be handled like an assertion. Prior to 8.13, all atomic groups
+     were handled this way. In 8.13, the code was changed as below for ONCE, so
+     that backups pass through the group and thereby reset captured values.
+     However, this uses a lot more stack, so in 8.20, atomic groups that do not
+     contain any captures generate OP_ONCE_NC, which can be handled in the old,
+     less stack intensive way.
+     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, but resetting
+     the start-of-match value in case it was changed by \K. */
+     case OP_ONCE_NC:
+     prev = ecode;
+     saved_eptr = eptr;
+     save_mark = md->mark;
+     do
+       {
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM64);
+       if (rrc == MATCH_MATCH)  /* Note: _not_ MATCH_ACCEPT */
+         {
+         mstart = md->start_match_ptr;
+         break;
+         }
+       if (rrc == MATCH_THEN)
+         {
+         next = ecode + GET(ecode,1);
+         if (md->start_match_ptr < next &&
+             (*ecode == OP_ALT || *next == OP_ALT))
+           rrc = MATCH_NOMATCH;
+         }
+       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       ecode += GET(ecode,1);
+       md->mark = save_mark;
+       }
+     while (*ecode == OP_ALT);
+     /* If hit the end of the group (which could be repeated), fail */
+     if (*ecode != OP_ONCE_NC && *ecode != OP_ALT) RRETURN(MATCH_NOMATCH);
+     /* Continue as from after the group, updating the offsets high water
+     mark, since extracts may have been taken. */
+     do ecode += GET(ecode, 1); while (*ecode == OP_ALT);
+     offset_top = md->end_offset_top;
+     eptr = md->end_match_ptr;
+     /* For a non-repeating ket, just continue at this level. This also
+     happens for a repeating ket if no characters were matched in the group.
+     This is the forcible breaking of infinite loops as implemented in Perl
+.005. */
+     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. */
-     /* For extended extraction brackets (large number), we have to fish out the
+     if (*ecode == OP_KETRMIN)
-     number from a dummy opcode at the start. */
+       {
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM65);
+       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       ecode = prev;
+       goto TAIL_RECURSE;
+       }
+     else  /* OP_KETRMAX */
+       {
+       md->match_function_type = MATCH_CBEGROUP;
+       RMATCH(eptr, prev, offset_top, md, eptrb, RM66);
+       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       ecode += 1 + LINK_SIZE;
+       goto TAIL_RECURSE;
+       }
+     /* Control never gets here */
-     if (number > EXTRACT_BASIC_MAX)
+     /* Handle a capturing bracket, other than those that are possessive with an
-       number = GET2(ecode, 2+LINK_SIZE);
+     unlimited repeat. 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. A failure to match might occur after the group has succeeded,
+     if something later on doesn't match. For this reason, we need to restore
+     the working 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. */
+     case OP_CBRA:
+     case OP_SCBRA:
+     number = GET2(ecode, 1+LINK_SIZE);
      offset = number << 1;
- #ifdef DEBUG
+ #ifdef PCRE_DEBUG
-     printf("start bracket %d subject=", number);
+     printf("start bracket %d\n", number);
+     printf("subject=");
      pchars(eptr, 16, TRUE, md);
      printf("\n");
  #endif
-Line 580 
 for (;;)
+Line 947 
 for (;;)
        save_offset2 = md->offset_vector[offset+1];
        save_offset3 = md->offset_vector[md->offset_end - number];
        save_capture_last = md->capture_last;
+       save_mark = md->mark;
        DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
-       md->offset_vector[md->offset_end - number] = eptr - md->start_subject;
+       md->offset_vector[md->offset_end - number] =
+         (int)(eptr - md->start_subject);
-       do
+       for (;;)
          {
-         RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb,
+         if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
-           match_isgroup);
+         RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
+           eptrb, RM1);
+         if (rrc == MATCH_ONCE) break;  /* Backing up through an atomic group */
+         /* If we backed up to a THEN, check whether it is within the current
+         branch by comparing the address of the THEN that is passed back with
+         the end of the branch. If it is within the current branch, and the
+         branch is one of two or more alternatives (it either starts or ends
+         with OP_ALT), we have reached the limit of THEN's action, so convert
+         the return code to NOMATCH, which will cause normal backtracking to
+         happen from now on. Otherwise, THEN is passed back to an outer
+         alternative. This implements Perl's treatment of parenthesized groups,
+         where a group not containing | does not affect the current alternative,
+         that is, (X) is NOT the same as (X|(*F)). */
+         if (rrc == MATCH_THEN)
+           {
+           next = ecode + GET(ecode,1);
+           if (md->start_match_ptr < next &&
+               (*ecode == OP_ALT || *next == OP_ALT))
+             rrc = MATCH_NOMATCH;
+           }
+         /* Anything other than NOMATCH is passed back. */
          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
          md->capture_last = save_capture_last;
          ecode += GET(ecode, 1);
+         md->mark = save_mark;
+         if (*ecode != OP_ALT) break;
          }
-       while (*ecode == OP_ALT);
        DPRINTF(("bracket %d failed\n", number));
        md->offset_vector[offset] = save_offset1;
        md->offset_vector[offset+1] = save_offset2;
        md->offset_vector[md->offset_end - number] = save_offset3;
-       RRETURN(MATCH_NOMATCH);
+       /* At this point, rrc will be one of MATCH_ONCE or MATCH_NOMATCH. */
+       RRETURN(rrc);
        }
-     /* Insufficient room for saving captured contents */
+     /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
+     as a non-capturing bracket. */
-     else op = OP_BRA;
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-     }
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-   /* Other types of node can be handled by a switch */
+     DPRINTF(("insufficient capture room: treat as non-capturing\n"));
-   switch(op)
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-     {
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-     case OP_BRA:     /* Non-capturing bracket: optimized */
-     DPRINTF(("start bracket 0\n"));
+     /* Non-capturing or atomic group, except for possessive with unlimited
-     do
+     repeat and ONCE group with no captures. Loop for all the alternatives.
+     When we get to the final alternative within the brackets, we used to return
+     the result of a recursive call to match() whatever happened so it was
+     possible to reduce stack usage by turning this into a tail recursion,
+     except in the case of a possibly empty group. However, now that there is
+     the possiblity of (*THEN) occurring in the final alternative, this
+     optimization is no longer always possible.
+     We can optimize if we know there are no (*THEN)s in the pattern; at present
+     this is the best that can be done.
+     MATCH_ONCE is returned when the end of an atomic group is successfully
+     reached, but subsequent matching fails. It passes back up the tree (causing
+     captured values to be reset) until the original atomic group level is
+     reached. This is tested by comparing md->once_target with the start of the
+     group. At this point, the return is converted into MATCH_NOMATCH so that
+     previous backup points can be taken. */
+     case OP_ONCE:
+     case OP_BRA:
+     case OP_SBRA:
+     DPRINTF(("start non-capturing bracket\n"));
+     for (;;)
        {
-       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb,
+       if (op >= OP_SBRA || op == OP_ONCE) md->match_function_type = MATCH_CBEGROUP;
-         match_isgroup);
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       /* If this is not a possibly empty group, and there are no (*THEN)s in
+       the pattern, and this is the final alternative, optimize as described
+       above. */
+       else if (!md->hasthen && ecode[GET(ecode, 1)] != OP_ALT)
+         {
+         ecode += PRIV(OP_lengths)[*ecode];
+         goto TAIL_RECURSE;
+         }
+       /* In all other cases, we have to make another call to match(). */
+       save_mark = md->mark;
+       RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md, eptrb,
+         RM2);
+       /* See comment in the code for capturing groups above about handling
+       THEN. */
+       if (rrc == MATCH_THEN)
+         {
+         next = ecode + GET(ecode,1);
+         if (md->start_match_ptr < next &&
+             (*ecode == OP_ALT || *next == OP_ALT))
+           rrc = MATCH_NOMATCH;
+         }
+       if (rrc != MATCH_NOMATCH)
+         {
+         if (rrc == MATCH_ONCE)
+           {
+           const pcre_uchar *scode = ecode;
+           if (*scode != OP_ONCE)           /* If not at start, find it */
+             {
+             while (*scode == OP_ALT) scode += GET(scode, 1);
+             scode -= GET(scode, 1);
+             }
+           if (md->once_target == scode) rrc = MATCH_NOMATCH;
+           }
+         RRETURN(rrc);
+         }
        ecode += GET(ecode, 1);
+       md->mark = save_mark;
+       if (*ecode != OP_ALT) break;
        }
-     while (*ecode == OP_ALT);
-     DPRINTF(("bracket 0 failed\n"));
      RRETURN(MATCH_NOMATCH);
-     /* Conditional group: compilation checked that there are no more than
+     /* Handle possessive capturing brackets with an unlimited repeat. We come
-     two branches. If the condition is false, skipping the first branch takes us
+     here from BRAZERO with allow_zero set TRUE. The offset_vector values are
-     past the end if there is only one branch, but that's OK because that is
+     handled similarly to the normal case above. However, the matching is
-     exactly what going to the ket would do. */
+     different. The end of these brackets will always be OP_KETRPOS, which
+     returns MATCH_KETRPOS without going further in the pattern. By this means
+     we can handle the group by iteration rather than recursion, thereby
+     reducing the amount of stack needed. */
+     case OP_CBRAPOS:
+     case OP_SCBRAPOS:
+     allow_zero = FALSE;
-     case OP_COND:
+     POSSESSIVE_CAPTURE:
-     if (ecode[LINK_SIZE+1] == OP_CREF) /* Condition extract or recurse test */
+     number = GET2(ecode, 1+LINK_SIZE);
-       {
+     offset = number << 1;
-       offset = GET2(ecode, LINK_SIZE+2) << 1;  /* Doubled ref number */
-       condition = (offset == CREF_RECURSE * 2)?
-         (md->recursive != NULL) :
-         (offset < offset_top && md->offset_vector[offset] >= 0);
-       RMATCH(rrc, eptr, ecode + (condition?
-         (LINK_SIZE + 4) : (LINK_SIZE + 1 + GET(ecode, 1))),
-         offset_top, md, ims, eptrb, match_isgroup);
-       RRETURN(rrc);
-       }
-     /* The condition is an assertion. Call match() to evaluate it - setting
+ #ifdef PCRE_DEBUG
-     the final argument TRUE causes it to stop at the end of an assertion. */
+     printf("start possessive bracket %d\n", number);
+     printf("subject=");
+     pchars(eptr, 16, TRUE, md);
+     printf("\n");
+ #endif
-     else
+     if (offset < md->offset_max)
        {
-       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,
+       matched_once = FALSE;
-           match_condassert | match_isgroup);
+       code_offset = (int)(ecode - md->start_code);
-       if (rrc == MATCH_MATCH)
+       save_offset1 = md->offset_vector[offset];
+       save_offset2 = md->offset_vector[offset+1];
+       save_offset3 = md->offset_vector[md->offset_end - number];
+       save_capture_last = md->capture_last;
+       DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
+       /* Each time round the loop, save the current subject position for use
+       when the group matches. For MATCH_MATCH, the group has matched, so we
+       restart it with a new subject starting position, remembering that we had
+       at least one match. For MATCH_NOMATCH, carry on with the alternatives, as
+       usual. If we haven't matched any alternatives in any iteration, check to
+       see if a previous iteration matched. If so, the group has matched;
+       continue from afterwards. Otherwise it has failed; restore the previous
+       capture values before returning NOMATCH. */
+       for (;;)
+         {
+         md->offset_vector[md->offset_end - number] =
+           (int)(eptr - md->start_subject);
+         if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
+         RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
+           eptrb, RM63);
+         if (rrc == MATCH_KETRPOS)
+           {
+           offset_top = md->end_offset_top;
+           eptr = md->end_match_ptr;
+           ecode = md->start_code + code_offset;
+           save_capture_last = md->capture_last;
+           matched_once = TRUE;
+           continue;
+           }
+         /* See comment in the code for capturing groups above about handling
+         THEN. */
+         if (rrc == MATCH_THEN)
+           {
+           next = ecode + GET(ecode,1);
+           if (md->start_match_ptr < next &&
+               (*ecode == OP_ALT || *next == OP_ALT))
+             rrc = MATCH_NOMATCH;
+           }
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         md->capture_last = save_capture_last;
+         ecode += GET(ecode, 1);
+         if (*ecode != OP_ALT) break;
+         }
+       if (!matched_once)
          {
-         ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE+2);
+         md->offset_vector[offset] = save_offset1;
-         while (*ecode == OP_ALT) ecode += GET(ecode, 1);
+         md->offset_vector[offset+1] = save_offset2;
+         md->offset_vector[md->offset_end - number] = save_offset3;
          }
-       else if (rrc != MATCH_NOMATCH)
+       if (allow_zero || matched_once)
          {
-         RRETURN(rrc);         /* Need braces because of following else */
+         ecode += 1 + LINK_SIZE;
+         break;
          }
-       else ecode += GET(ecode, 1);
-       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb,
+       RRETURN(MATCH_NOMATCH);
-         match_isgroup);
-       RRETURN(rrc);
        }
-     /* Control never reaches here */
-     /* Skip over conditional reference or large extraction number data if
+     /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
-     encountered. */
+     as a non-capturing bracket. */
-     case OP_CREF:
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-     case OP_BRANUMBER:
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-     ecode += 3;
-     break;
-     /* End of the pattern. If we are in a recursion, we should restore the
+     DPRINTF(("insufficient capture room: treat as non-capturing\n"));
-     offsets appropriately and continue from after the call. */
-     case OP_END:
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-     if (md->recursive != NULL && md->recursive->group_num == 0)
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-       {
-       recursion_info *rec = md->recursive;
-       DPRINTF(("Hit the end in a (?0) recursion\n"));
-       md->recursive = rec->prevrec;
-       memmove(md->offset_vector, rec->offset_save,
-         rec->saved_max * sizeof(int));
-       md->start_match = rec->save_start;
-       ims = original_ims;
-       ecode = rec->after_call;
-       break;
-       }
-     /* Otherwise, if PCRE_NOTEMPTY is set, fail if we have matched an empty
+     /* Non-capturing possessive bracket with unlimited repeat. We come here
-     string - backtracking will then try other alternatives, if any. */
+     from BRAZERO with allow_zero = TRUE. The code is similar to the above,
+     without the capturing complication. It is written out separately for speed
+     and cleanliness. */
-     if (md->notempty && eptr == md->start_match) RRETURN(MATCH_NOMATCH);
+     case OP_BRAPOS:
-     md->end_match_ptr = eptr;          /* Record where we ended */
+     case OP_SBRAPOS:
-     md->end_offset_top = offset_top;   /* and how many extracts were taken */
+     allow_zero = FALSE;
-     RRETURN(MATCH_MATCH);
-     /* Change option settings */
-     case OP_OPT:
-     ims = ecode[1];
-     ecode += 2;
-     DPRINTF(("ims set to %02lx\n", ims));
-     break;
-     /* Assertion brackets. Check the alternative branches in turn - the
+     POSSESSIVE_NON_CAPTURE:
-     matching won't pass the KET for an assertion. If any one branch matches,
+     matched_once = FALSE;
-     the assertion is true. Lookbehind assertions have an OP_REVERSE item at the
+     code_offset = (int)(ecode - md->start_code);
-     start of each branch to move the current point backwards, so the code at
-     this level is identical to the lookahead case. */
-     case OP_ASSERT:
+     for (;;)
-     case OP_ASSERTBACK:
-     do
        {
-       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,
+       if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
-         match_isgroup);
+       RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
-       if (rrc == MATCH_MATCH) break;
+         eptrb, RM48);
+       if (rrc == MATCH_KETRPOS)
+         {
+         offset_top = md->end_offset_top;
+         eptr = md->end_match_ptr;
+         ecode = md->start_code + code_offset;
+         matched_once = TRUE;
+         continue;
+         }
+       /* See comment in the code for capturing groups above about handling
+       THEN. */
+       if (rrc == MATCH_THEN)
+         {
+         next = ecode + GET(ecode,1);
+         if (md->start_match_ptr < next &&
+             (*ecode == OP_ALT || *next == OP_ALT))
+           rrc = MATCH_NOMATCH;
+         }
        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
        ecode += GET(ecode, 1);
+       if (*ecode != OP_ALT) break;
        }
-     while (*ecode == OP_ALT);
-     if (*ecode == OP_KET) RRETURN(MATCH_NOMATCH);
-     /* If checking an assertion for a condition, return MATCH_MATCH. */
+     if (matched_once || allow_zero)
+       {
+       ecode += 1 + LINK_SIZE;
+       break;
+       }
+     RRETURN(MATCH_NOMATCH);
-     if ((flags & match_condassert) != 0) RRETURN(MATCH_MATCH);
+     /* Control never reaches here. */
-     /* Continue from after the assertion, updating the offsets high water
+     /* Conditional group: compilation checked that there are no more than
-     mark, since extracts may have been taken during the assertion. */
+     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. */
-     do ecode += GET(ecode,1); while (*ecode == OP_ALT);
+     case OP_COND:
-     ecode += 1 + LINK_SIZE;
+     case OP_SCOND:
-     offset_top = md->end_offset_top;
+     codelink = GET(ecode, 1);
-     continue;
-     /* Negative assertion: all branches must fail to match */
+     /* Because of the way auto-callout works during compile, a callout item is
+     inserted between OP_COND and an assertion condition. */
-     case OP_ASSERT_NOT:
+     if (ecode[LINK_SIZE+1] == OP_CALLOUT)
-     case OP_ASSERTBACK_NOT:
-     do
        {
-       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,
+       if (PUBL(callout) != NULL)
-         match_isgroup);
+         {
-       if (rrc == MATCH_MATCH) RRETURN(MATCH_NOMATCH);
+         PUBL(callout_block) cb;
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         cb.version          = 2;   /* Version 1 of the callout block */
-       ecode += GET(ecode,1);
+         cb.callout_number   = ecode[LINK_SIZE+2];
+         cb.offset_vector    = md->offset_vector;
+ #ifdef COMPILE_PCRE8
+         cb.subject          = (PCRE_SPTR)md->start_subject;
+ #else
+         cb.subject          = (PCRE_SPTR16)md->start_subject;
+ #endif
+         cb.subject_length   = (int)(md->end_subject - md->start_subject);
+         cb.start_match      = (int)(mstart - md->start_subject);
+         cb.current_position = (int)(eptr - md->start_subject);
+         cb.pattern_position = GET(ecode, LINK_SIZE + 3);
+         cb.next_item_length = GET(ecode, 3 + 2*LINK_SIZE);
+         cb.capture_top      = offset_top/2;
+         cb.capture_last     = md->capture_last;
+         cb.callout_data     = md->callout_data;
+         cb.mark             = md->nomatch_mark;
+         if ((rrc = (*PUBL(callout))(&cb)) > 0) RRETURN(MATCH_NOMATCH);
+         if (rrc < 0) RRETURN(rrc);
+         }
+       ecode += PRIV(OP_lengths)[OP_CALLOUT];
        }
-     while (*ecode == OP_ALT);
-     if ((flags & match_condassert) != 0) RRETURN(MATCH_MATCH);
-     ecode += 1 + LINK_SIZE;
+     condcode = ecode[LINK_SIZE+1];
-     continue;
-     /* Move the subject pointer back. This occurs only at the start of
+     /* Now see what the actual condition is */
-     each branch of a lookbehind assertion. If we are too close to the start to
-     move back, this match function fails. When working with UTF-8 we move
-     back a number of characters, not bytes. */
-     case OP_REVERSE:
+     if (condcode == OP_RREF || condcode == OP_NRREF)    /* Recursion test */
- #ifdef SUPPORT_UTF8
-     if (utf8)
        {
-       c = GET(ecode,1);
+       if (md->recursive == NULL)                /* Not recursing => FALSE */
-       for (i = 0; i < c; i++)
          {
-         eptr--;
+         condition = FALSE;
-         if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
+         ecode += GET(ecode, 1);
-         BACKCHAR(eptr)
          }
-       }
+       else
-     else
+         {
+         int recno = GET2(ecode, LINK_SIZE + 2);   /* Recursion group number*/
+         condition = (recno == RREF_ANY || recno == md->recursive->group_num);
+         /* If the test is for recursion into a specific subpattern, and it is
+         false, but the test was set up by name, scan the table to see if the
+         name refers to any other numbers, and test them. The condition is true
+         if any one is set. */
+         if (!condition && condcode == OP_NRREF)
+           {
+           pcre_uchar *slotA = md->name_table;
+           for (i = 0; i < md->name_count; i++)
+             {
+             if (GET2(slotA, 0) == recno) break;
+             slotA += md->name_entry_size;
+             }
+           /* Found a name for the number - there can be only one; duplicate
+           names for different numbers are allowed, but not vice versa. First
+           scan down for duplicates. */
+           if (i < md->name_count)
+             {
+             pcre_uchar *slotB = slotA;
+             while (slotB > md->name_table)
+               {
+               slotB -= md->name_entry_size;
+               if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
+                 {
+                 condition = GET2(slotB, 0) == md->recursive->group_num;
+                 if (condition) break;
+                 }
+               else break;
+               }
+             /* Scan up for duplicates */
+             if (!condition)
+               {
+               slotB = slotA;
+               for (i++; i < md->name_count; i++)
+                 {
+                 slotB += md->name_entry_size;
+                 if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
+                   {
+                   condition = GET2(slotB, 0) == md->recursive->group_num;
+                   if (condition) break;
+                   }
+                 else break;
+                 }
+               }
+             }
+           }
+         /* Chose branch according to the condition */
+         ecode += condition? 1 + IMM2_SIZE : GET(ecode, 1);
+         }
+       }
+     else if (condcode == OP_CREF || condcode == OP_NCREF)  /* Group used test */
+       {
+       offset = GET2(ecode, LINK_SIZE+2) << 1;  /* Doubled ref number */
+       condition = offset < offset_top && md->offset_vector[offset] >= 0;
+       /* If the numbered capture is unset, but the reference was by name,
+       scan the table to see if the name refers to any other numbers, and test
+       them. The condition is true if any one is set. This is tediously similar
+       to the code above, but not close enough to try to amalgamate. */
+       if (!condition && condcode == OP_NCREF)
+         {
+         int refno = offset >> 1;
+         pcre_uchar *slotA = md->name_table;
+         for (i = 0; i < md->name_count; i++)
+           {
+           if (GET2(slotA, 0) == refno) break;
+           slotA += md->name_entry_size;
+           }
+         /* Found a name for the number - there can be only one; duplicate names
+         for different numbers are allowed, but not vice versa. First scan down
+         for duplicates. */
+         if (i < md->name_count)
+           {
+           pcre_uchar *slotB = slotA;
+           while (slotB > md->name_table)
+             {
+             slotB -= md->name_entry_size;
+             if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
+               {
+               offset = GET2(slotB, 0) << 1;
+               condition = offset < offset_top &&
+                 md->offset_vector[offset] >= 0;
+               if (condition) break;
+               }
+             else break;
+             }
+           /* Scan up for duplicates */
+           if (!condition)
+             {
+             slotB = slotA;
+             for (i++; i < md->name_count; i++)
+               {
+               slotB += md->name_entry_size;
+               if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
+                 {
+                 offset = GET2(slotB, 0) << 1;
+                 condition = offset < offset_top &&
+                   md->offset_vector[offset] >= 0;
+                 if (condition) break;
+                 }
+               else break;
+               }
+             }
+           }
+         }
+       /* Chose branch according to the condition */
+       ecode += condition? 1 + IMM2_SIZE : GET(ecode, 1);
+       }
+     else if (condcode == OP_DEF)     /* DEFINE - always false */
+       {
+       condition = FALSE;
+       ecode += GET(ecode, 1);
+       }
+     /* The condition is an assertion. Call match() to evaluate it - setting
+     md->match_function_type to MATCH_CONDASSERT causes it to stop at the end of
+     an assertion. */
+     else
+       {
+       md->match_function_type = MATCH_CONDASSERT;
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM3);
+       if (rrc == MATCH_MATCH)
+         {
+         if (md->end_offset_top > offset_top)
+           offset_top = md->end_offset_top;  /* Captures may have happened */
+         condition = TRUE;
+         ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE + 2);
+         while (*ecode == OP_ALT) ecode += GET(ecode, 1);
+         }
+       /* PCRE doesn't allow the effect of (*THEN) to escape beyond an
+       assertion; it is therefore treated as NOMATCH. */
+       else if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN)
+         {
+         RRETURN(rrc);         /* Need braces because of following else */
+         }
+       else
+         {
+         condition = FALSE;
+         ecode += codelink;
+         }
+       }
+     /* We are now at the branch that is to be obeyed. As there is only one, can
+     use tail recursion to avoid using another stack frame, except when there is
+     unlimited repeat of a possibly empty group. In the latter case, a recursive
+     call to match() is always required, unless the second alternative doesn't
+     exist, in which case we can just plough on. Note that, for compatibility
+     with Perl, the | in a conditional group is NOT treated as creating two
+     alternatives. If a THEN is encountered in the branch, it propagates out to
+     the enclosing alternative (unless nested in a deeper set of alternatives,
+     of course). */
+     if (condition || *ecode == OP_ALT)
+       {
+       if (op != OP_SCOND)
+         {
+         ecode += 1 + LINK_SIZE;
+         goto TAIL_RECURSE;
+         }
+       md->match_function_type = MATCH_CBEGROUP;
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM49);
+       RRETURN(rrc);
+       }
+      /* Condition false & no alternative; continue after the group. */
+     else
+       {
+       ecode += 1 + LINK_SIZE;
+       }
+     break;
+     /* Before OP_ACCEPT there may be any number of OP_CLOSE opcodes,
+     to close any currently open capturing brackets. */
+     case OP_CLOSE:
+     number = GET2(ecode, 1);
+     offset = number << 1;
+ #ifdef PCRE_DEBUG
+       printf("end bracket %d at *ACCEPT", number);
+       printf("\n");
+ #endif
+     md->capture_last = number;
+     if (offset >= md->offset_max) md->offset_overflow = TRUE; else
+       {
+       md->offset_vector[offset] =
+         md->offset_vector[md->offset_end - number];
+       md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
+       if (offset_top <= offset) offset_top = offset + 2;
+       }
+     ecode += 1 + IMM2_SIZE;
+     break;
+     /* End of the pattern, either real or forced. */
+     case OP_END:
+     case OP_ACCEPT:
+     case OP_ASSERT_ACCEPT:
+     /* If we have matched an empty string, fail if not in an assertion and not
+     in a recursion if either PCRE_NOTEMPTY is set, or if PCRE_NOTEMPTY_ATSTART
+     is set and we have matched at the start of the subject. In both cases,
+     backtracking will then try other alternatives, if any. */
+     if (eptr == mstart && op != OP_ASSERT_ACCEPT &&
+          md->recursive == NULL &&
+          (md->notempty ||
+            (md->notempty_atstart &&
+              mstart == md->start_subject + md->start_offset)))
+       RRETURN(MATCH_NOMATCH);
+     /* Otherwise, we have a match. */
+     md->end_match_ptr = eptr;           /* Record where we ended */
+     md->end_offset_top = offset_top;    /* and how many extracts were taken */
+     md->start_match_ptr = mstart;       /* and the start (\K can modify) */
+     /* For some reason, the macros don't work properly if an expression is
+     given as the argument to RRETURN when the heap is in use. */
+     rrc = (op == OP_END)? MATCH_MATCH : MATCH_ACCEPT;
+     RRETURN(rrc);
+     /* Assertion brackets. Check the alternative branches in turn - the
+     matching won't pass the KET for an assertion. If any one branch matches,
+     the assertion is true. Lookbehind assertions have an OP_REVERSE item at the
+     start of each branch to move the current point backwards, so the code at
+     this level is identical to the lookahead case. When the assertion is part
+     of a condition, we want to return immediately afterwards. The caller of
+     this incarnation of the match() function will have set MATCH_CONDASSERT in
+     md->match_function type, and one of these opcodes will be the first opcode
+     that is processed. We use a local variable that is preserved over calls to
+     match() to remember this case. */
+     case OP_ASSERT:
+     case OP_ASSERTBACK:
+     save_mark = md->mark;
+     if (md->match_function_type == MATCH_CONDASSERT)
+       {
+       condassert = TRUE;
+       md->match_function_type = 0;
+       }
+     else condassert = FALSE;
+     do
+       {
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM4);
+       if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT)
+         {
+         mstart = md->start_match_ptr;   /* In case \K reset it */
+         break;
+         }
+       /* PCRE does not allow THEN to escape beyond an assertion; it is treated
+       as NOMATCH. */
+       if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
+       ecode += GET(ecode, 1);
+       md->mark = save_mark;
+       }
+     while (*ecode == OP_ALT);
+     if (*ecode == OP_KET) RRETURN(MATCH_NOMATCH);
+     /* If checking an assertion for a condition, return MATCH_MATCH. */
+     if (condassert) RRETURN(MATCH_MATCH);
+     /* Continue from after the assertion, updating the offsets high water
+     mark, since extracts may have been taken during the assertion. */
+     do ecode += GET(ecode,1); while (*ecode == OP_ALT);
+     ecode += 1 + LINK_SIZE;
+     offset_top = md->end_offset_top;
+     continue;
+     /* Negative assertion: all branches must fail to match. Encountering SKIP,
+     PRUNE, or COMMIT means we must assume failure without checking subsequent
+     branches. */
+     case OP_ASSERT_NOT:
+     case OP_ASSERTBACK_NOT:
+     save_mark = md->mark;
+     if (md->match_function_type == MATCH_CONDASSERT)
+       {
+       condassert = TRUE;
+       md->match_function_type = 0;
+       }
+     else condassert = FALSE;
+     do
+       {
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM5);
+       md->mark = save_mark;
+       if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) RRETURN(MATCH_NOMATCH);
+       if (rrc == MATCH_SKIP || rrc == MATCH_PRUNE || rrc == MATCH_COMMIT)
+         {
+         do ecode += GET(ecode,1); while (*ecode == OP_ALT);
+         break;
+         }
+       /* PCRE does not allow THEN to escape beyond an assertion; it is treated
+       as NOMATCH. */
+       if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
+       ecode += GET(ecode,1);
+       }
+     while (*ecode == OP_ALT);
+     if (condassert) RRETURN(MATCH_MATCH);  /* Condition assertion */
+     ecode += 1 + LINK_SIZE;
+     continue;
+     /* Move the subject pointer back. This occurs only at the start of
+     each branch of a lookbehind assertion. If we are too close to the start to
+     move back, this match function fails. When working with UTF-8 we move
+     back a number of characters, not bytes. */
+     case OP_REVERSE:
+ #ifdef SUPPORT_UTF
+     if (utf)
+       {
+       i = GET(ecode, 1);
+       while (i-- > 0)
+         {
+         eptr--;
+         if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
+         BACKCHAR(eptr);
+         }
+       }
+     else
  #endif
      /* No UTF-8 support, or not in UTF-8 mode: count is byte count */
        {
-       eptr -= GET(ecode,1);
+       eptr -= GET(ecode, 1);
        if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
        }
-     /* Skip to next op code */
+     /* Save the earliest consulted character, then skip to next op code */
+     if (eptr < md->start_used_ptr) md->start_used_ptr = eptr;
      ecode += 1 + LINK_SIZE;
      break;
-Line 794 
 for (;;)
+Line 1675 
 for (;;)
      function is able to force a failure. */
      case OP_CALLOUT:
-     if (pcre_callout != NULL)
+     if (PUBL(callout) != NULL)
        {
-       pcre_callout_block cb;
+       PUBL(callout_block) cb;
-       cb.version          = 1;   /* Version 1 of the callout block */
+       cb.version          = 2;   /* Version 1 of the callout block */
        cb.callout_number   = ecode[1];
        cb.offset_vector    = md->offset_vector;
-       cb.subject          = (const char *)md->start_subject;
+ #ifdef COMPILE_PCRE8
-       cb.subject_length   = md->end_subject - md->start_subject;
+       cb.subject          = (PCRE_SPTR)md->start_subject;
-       cb.start_match      = md->start_match - md->start_subject;
+ #else
-       cb.current_position = eptr - md->start_subject;
+       cb.subject          = (PCRE_SPTR16)md->start_subject;
+ #endif
+       cb.subject_length   = (int)(md->end_subject - md->start_subject);
+       cb.start_match      = (int)(mstart - md->start_subject);
+       cb.current_position = (int)(eptr - md->start_subject);
        cb.pattern_position = GET(ecode, 2);
        cb.next_item_length = GET(ecode, 2 + LINK_SIZE);
        cb.capture_top      = offset_top/2;
        cb.capture_last     = md->capture_last;
        cb.callout_data     = md->callout_data;
-       if ((rrc = (*pcre_callout)(&cb)) > 0) RRETURN(MATCH_NOMATCH);
+       cb.mark             = md->nomatch_mark;
+       if ((rrc = (*PUBL(callout))(&cb)) > 0) RRETURN(MATCH_NOMATCH);
        if (rrc < 0) RRETURN(rrc);
        }
      ecode += 2 + 2*LINK_SIZE;
-Line 819 
 for (;;)
+Line 1705 
 for (;;)
      offset data is the offset to the starting bracket from the start of the
      whole pattern. (This is so that it works from duplicated subpatterns.)
-     If there are any capturing brackets started but not finished, we have to
+     The state of the capturing groups is preserved over recursion, and
-     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
-     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
-     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
-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
-     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
-     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.
      There are also other values that have to be saved. We use a chained
      sequence of blocks that actually live on the stack. Thanks to Robin Houston
-     for the original version of this logic. */
+     for the original version of this logic. It has, however, been hacked around
+     a lot, so he is not to blame for the current way it works. */
      case OP_RECURSE:
        {
-       callpat = md->start_code + GET(ecode, 1);
+       recursion_info *ri;
-       new_recursive.group_num = *callpat - OP_BRA;
+       int recno;
-       /* For extended extraction brackets (large number), we have to fish out
+       callpat = md->start_code + GET(ecode, 1);
-       the number from a dummy opcode at the start. */
+       recno = (callpat == md->start_code)? 0 :
+         GET2(callpat, 1 + LINK_SIZE);
-       if (new_recursive.group_num > EXTRACT_BASIC_MAX)
+       /* Check for repeating a recursion without advancing the subject pointer.
-         new_recursive.group_num = GET2(callpat, 2+LINK_SIZE);
+       This should catch convoluted mutual recursions. (Some simple cases are
+       caught at compile time.) */
+       for (ri = md->recursive; ri != NULL; ri = ri->prevrec)
+         if (recno == ri->group_num && eptr == ri->subject_position)
+           RRETURN(PCRE_ERROR_RECURSELOOP);
        /* Add to "recursing stack" */
+       new_recursive.group_num = recno;
+       new_recursive.subject_position = eptr;
        new_recursive.prevrec = md->recursive;
        md->recursive = &new_recursive;
-       /* Find where to continue from afterwards */
+       /* Where to continue from afterwards */
        ecode += 1 + LINK_SIZE;
-       new_recursive.after_call = ecode;
-       /* Now save the offset data. */
+       /* Now save the offset data */
        new_recursive.saved_max = md->offset_end;
        if (new_recursive.saved_max <= REC_STACK_SAVE_MAX)
-Line 863 
 for (;;)
+Line 1754 
 for (;;)
        else
          {
          new_recursive.offset_save =
-           (int *)(pcre_malloc)(new_recursive.saved_max * sizeof(int));
+           (int *)(PUBL(malloc))(new_recursive.saved_max * sizeof(int));
          if (new_recursive.offset_save == NULL) RRETURN(PCRE_ERROR_NOMEMORY);
          }
        memcpy(new_recursive.offset_save, md->offset_vector,
              new_recursive.saved_max * sizeof(int));
-       new_recursive.save_start = md->start_match;
-       md->start_match = eptr;
-       /* OK, now we can do the recursion. For each top-level alternative we
+       /* OK, now we can do the recursion. After processing each alternative,
-       restore the offset and recursion data. */
+       restore the offset data. If there were nested recursions, md->recursive
+       might be changed, so reset it before looping. */
        DPRINTF(("Recursing into group %d\n", new_recursive.group_num));
+       cbegroup = (*callpat >= OP_SBRA);
        do
          {
-         RMATCH(rrc, eptr, callpat + 1 + LINK_SIZE, offset_top, md, ims,
+         if (cbegroup) md->match_function_type = MATCH_CBEGROUP;
-             eptrb, match_isgroup);
+         RMATCH(eptr, callpat + PRIV(OP_lengths)[*callpat], offset_top,
-         if (rrc == MATCH_MATCH)
+           md, eptrb, RM6);
+         memcpy(md->offset_vector, new_recursive.offset_save,
+             new_recursive.saved_max * sizeof(int));
+         md->recursive = new_recursive.prevrec;
+         if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT)
            {
-           md->recursive = new_recursive.prevrec;
+           DPRINTF(("Recursion matched\n"));
            if (new_recursive.offset_save != stacksave)
-             (pcre_free)(new_recursive.offset_save);
+             (PUBL(free))(new_recursive.offset_save);
-           RRETURN(MATCH_MATCH);
+           /* Set where we got to in the subject, and reset the start in case
+           it was changed by \K. This *is* propagated back out of a recursion,
+           for Perl compatibility. */
+           eptr = md->end_match_ptr;
+           mstart = md->start_match_ptr;
+           goto RECURSION_MATCHED;        /* Exit loop; end processing */
+           }
+         /* PCRE does not allow THEN to escape beyond a recursion; it is treated
+         as NOMATCH. */
+         else if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN)
+           {
+           DPRINTF(("Recursion gave error %d\n", rrc));
+           if (new_recursive.offset_save != stacksave)
+             (PUBL(free))(new_recursive.offset_save);
+           RRETURN(rrc);
            }
-         else if (rrc != MATCH_NOMATCH) RRETURN(rrc);
          md->recursive = &new_recursive;
-         memcpy(md->offset_vector, new_recursive.offset_save,
-             new_recursive.saved_max * sizeof(int));
          callpat += GET(callpat, 1);
          }
        while (*callpat == OP_ALT);
-Line 899 
 for (;;)
+Line 1808 
 for (;;)
        DPRINTF(("Recursion didn't match\n"));
        md->recursive = new_recursive.prevrec;
        if (new_recursive.offset_save != stacksave)
-         (pcre_free)(new_recursive.offset_save);
+         (PUBL(free))(new_recursive.offset_save);
        RRETURN(MATCH_NOMATCH);
        }
-     /* Control never reaches here */
-     /* "Once" brackets are like assertion brackets except that after a match,
-     the point in the subject string is not moved back. Thus there can never be
-     a move back into the brackets. Friedl calls these "atomic" subpatterns.
-     Check the alternative branches in turn - the matching won't pass the KET
-     for this kind of subpattern. If any one branch matches, we carry on as at
-     the end of a normal bracket, leaving the subject pointer. */
-     case OP_ONCE:
-       {
-       prev = ecode;
-       saved_eptr = eptr;
-       do
-         {
-         RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims,
-           eptrb, match_isgroup);
-         if (rrc == MATCH_MATCH) break;
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-         ecode += GET(ecode,1);
-         }
-       while (*ecode == OP_ALT);
-       /* If hit the end of the group (which could be repeated), fail */
-       if (*ecode != OP_ONCE && *ecode != OP_ALT) RRETURN(MATCH_NOMATCH);
-       /* Continue as from after the assertion, updating the offsets high water
-       mark, since extracts may have been taken. */
-       do ecode += GET(ecode,1); while (*ecode == OP_ALT);
-       offset_top = md->end_offset_top;
-       eptr = md->end_match_ptr;
-       /* For a non-repeating ket, just continue at this level. This also
-       happens for a repeating ket if no characters were matched in the group.
-       This is the forcible breaking of infinite loops as implemented in Perl
-.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. 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)
+     RECURSION_MATCHED:
-         {
+     break;
-         RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb, 0);
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-         RMATCH(rrc, eptr, prev, offset_top, md, ims, eptrb, match_isgroup);
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-         }
-       else  /* OP_KETRMAX */
-         {
-         RMATCH(rrc, eptr, prev, offset_top, md, ims, eptrb, match_isgroup);
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-         RMATCH(rrc, eptr, ecode + 1+LINK_SIZE, offset_top, md, ims, eptrb, 0);
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-         }
-       }
-     RRETURN(MATCH_NOMATCH);
      /* An alternation is the end of a branch; scan along to find the end of the
      bracketed group and go to there. */
-Line 985 
 for (;;)
+Line 1822 
 for (;;)
      do ecode += GET(ecode,1); while (*ecode == OP_ALT);
      break;
-     /* BRAZERO and BRAMINZERO occur just before a bracket group, indicating
+     /* BRAZERO, BRAMINZERO and SKIPZERO occur just before a bracket group,
-     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
-     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
-     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
-     preceded by BRAZERO or BRAMINZERO. */
+     optional ones preceded by BRAZERO or BRAMINZERO. */
      case OP_BRAZERO:
-       {
+     next = ecode + 1;
-       next = ecode+1;
+     RMATCH(eptr, next, offset_top, md, eptrb, RM10);
-       RMATCH(rrc, eptr, next, offset_top, md, ims, eptrb, match_isgroup);
+     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     do next += GET(next, 1); while (*next == OP_ALT);
-       do next += GET(next,1); while (*next == OP_ALT);
+     ecode = next + 1 + LINK_SIZE;
-       ecode = next + 1+LINK_SIZE;
-       }
      break;
      case OP_BRAMINZERO:
-       {
+     next = ecode + 1;
-       next = ecode+1;
+     do next += GET(next, 1); while (*next == OP_ALT);
-       do next += GET(next,1); while (*next == OP_ALT);
+     RMATCH(eptr, next + 1+LINK_SIZE, offset_top, md, eptrb, RM11);
-       RMATCH(rrc, eptr, next + 1+LINK_SIZE, offset_top, md, ims, eptrb,
+     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-         match_isgroup);
+     ecode++;
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     break;
-       ecode++;
-       }
+     case OP_SKIPZERO:
+     next = ecode+1;
+     do next += GET(next,1); while (*next == OP_ALT);
+     ecode = next + 1 + LINK_SIZE;
      break;
-     /* End of a group, repeated or non-repeating. If we are at the end of
+     /* BRAPOSZERO occurs before a possessive bracket group. Don't do anything
-     an assertion "group", stop matching and return MATCH_MATCH, but record the
+     here; just jump to the group, with allow_zero set TRUE. */
-     current high water mark for use by positive assertions. Do this also
-     for the "once" (not-backup up) groups. */
+     case OP_BRAPOSZERO:
+     op = *(++ecode);
+     allow_zero = TRUE;
+     if (op == OP_CBRAPOS || op == OP_SCBRAPOS) goto POSSESSIVE_CAPTURE;
+       goto POSSESSIVE_NON_CAPTURE;
+     /* End of a group, repeated or non-repeating. */
      case OP_KET:
      case OP_KETRMIN:
      case OP_KETRMAX:
-       {
+     case OP_KETRPOS:
-       prev = ecode - GET(ecode, 1);
+     prev = ecode - GET(ecode, 1);
-       saved_eptr = eptrb->epb_saved_eptr;
-       /* Back up the stack of bracket start pointers. */
+     /* If this was a group that remembered the subject start, in order to break
+     infinite repeats of empty string matches, retrieve the subject start from
+     the chain. Otherwise, set it NULL. */
-       eptrb = eptrb->epb_prev;
+     if (*prev >= OP_SBRA || *prev == OP_ONCE)
+       {
-       if (*prev == OP_ASSERT || *prev == OP_ASSERT_NOT ||
+       saved_eptr = eptrb->epb_saved_eptr;   /* Value at start of group */
-           *prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT ||
+       eptrb = eptrb->epb_prev;              /* Backup to previous group */
-           *prev == OP_ONCE)
+       }
-         {
+     else saved_eptr = NULL;
-         md->end_match_ptr = eptr;      /* For ONCE */
-         md->end_offset_top = offset_top;
-         RRETURN(MATCH_MATCH);
-         }
-       /* 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
-       group number back at the start and if necessary complete handling an
+     group, stop matching and return MATCH_MATCH, but record the current high
-       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
+     start in case it was changed by \K. */
-       if (*prev != OP_COND)
+     if ((*prev >= OP_ASSERT && *prev <= OP_ASSERTBACK_NOT) ||
-         {
+          *prev == OP_ONCE_NC)
-         number = *prev - OP_BRA;
+       {
+       md->end_match_ptr = eptr;      /* For ONCE_NC */
+       md->end_offset_top = offset_top;
+       md->start_match_ptr = mstart;
+       RRETURN(MATCH_MATCH);         /* Sets md->mark */
+       }
-         /* 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
-         the number from a dummy opcode at the start. */
+     and if necessary complete handling an extraction by setting the offsets and
+     bumping the high water mark. 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. Other recursion is handled here. We just have to record
+     the current subject position and start match pointer and give a MATCH
+     return. */
-         if (number > EXTRACT_BASIC_MAX) number = GET2(prev, 2+LINK_SIZE);
+     if (*prev == OP_CBRA || *prev == OP_SCBRA ||
-         offset = number << 1;
+         *prev == OP_CBRAPOS || *prev == OP_SCBRAPOS)
+       {
+       number = GET2(prev, 1+LINK_SIZE);
+       offset = number << 1;
- #ifdef DEBUG
+ #ifdef PCRE_DEBUG
-         printf("end bracket %d", number);
+       printf("end bracket %d", number);
-         printf("\n");
+       printf("\n");
  #endif
-         /* 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. */
-         if (number > 0)
+       if (md->recursive != NULL && md->recursive->group_num == number)
-           {
+         {
-           md->capture_last = number;
+         md->end_match_ptr = eptr;
-           if (offset >= md->offset_max) md->offset_overflow = TRUE; else
+         md->start_match_ptr = mstart;
-             {
+         RRETURN(MATCH_MATCH);
-             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;
-             }
-           /* Handle a recursively called group. Restore the offsets
+       /* Deal with capturing */
-           appropriately and continue from after the call. */
-           if (md->recursive != NULL && md->recursive->group_num == number)
+       md->capture_last = number;
-             {
+       if (offset >= md->offset_max) md->offset_overflow = TRUE; else
-             recursion_info *rec = md->recursive;
+         {
-             DPRINTF(("Recursion (%d) succeeded - continuing\n", number));
+         /* If offset is greater than offset_top, it means that we are
-             md->recursive = rec->prevrec;
+         "skipping" a capturing group, and that group's offsets must be marked
-             md->start_match = rec->save_start;
+         unset. In earlier versions of PCRE, all the offsets were unset at the
-             memcpy(md->offset_vector, rec->offset_save,
+         start of matching, but this doesn't work because atomic groups and
-               rec->saved_max * sizeof(int));
+         assertions can cause a value to be set that should later be unset.
-             ecode = rec->after_call;
+         Example: matching /(?>(a))b|(a)c/ against "ac". This sets group 1 as
-             ims = original_ims;
+         part of the atomic group, but this is not on the final matching path,
-             break;
+         so must be unset when 2 is set. (If there is no group 2, there is no
-             }
+         problem, because offset_top will then be 2, indicating no capture.) */
+         if (offset > offset_top)
+           {
+           register int *iptr = md->offset_vector + offset_top;
+           register int *iend = md->offset_vector + offset;
+           while (iptr < iend) *iptr++ = -1;
            }
-         }
-       /* Reset the value of the ims flags, in case they got changed during
+         /* Now make the extraction */
-       the group. */
-       ims = original_ims;
+         md->offset_vector[offset] =
-       DPRINTF(("ims reset to %02lx\n", ims));
+           md->offset_vector[md->offset_end - number];
+         md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
+         if (offset_top <= offset) offset_top = offset + 2;
+         }
+       }
-       /* For a non-repeating ket, just continue at this level. This also
+     /* For an ordinary non-repeating ket, just continue at this level. This
-       happens for a repeating ket if no characters were matched in the group.
+     also happens for a repeating ket if no characters were matched in the
-       This is the forcible breaking of infinite loops as implemented in Perl
+     group. This is the forcible breaking of infinite loops as implemented in
-.005. If there is an options reset, it will get obeyed in the normal
+     Perl 5.005. For a non-repeating atomic group that includes captures,
-       course of events. */
+     establish a backup point by processing the rest of the pattern at a lower
+     level. If this results in a NOMATCH return, pass MATCH_ONCE back to the
+     original OP_ONCE level, thereby bypassing intermediate backup points, but
+     resetting any captures that happened along the way. */
-       if (*ecode == OP_KET || eptr == saved_eptr)
+     if (*ecode == OP_KET || eptr == saved_eptr)
+       {
+       if (*prev == OP_ONCE)
          {
-         ecode += 1 + LINK_SIZE;
+         RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM12);
-         break;
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
+         RRETURN(MATCH_ONCE);
          }
+       ecode += 1 + LINK_SIZE;    /* Carry on at this level */
+       break;
+       }
+     /* OP_KETRPOS is a possessive repeating ket. Remember the current position,
+     and return the MATCH_KETRPOS. This makes it possible to do the repeats one
+     at a time from the outer level, thus saving stack. */
+     if (*ecode == OP_KETRPOS)
+       {
+       md->end_match_ptr = eptr;
+       md->end_offset_top = offset_top;
+       RRETURN(MATCH_KETRPOS);
+       }
-       /* The repeating kets try the rest of the pattern or restart from the
+     /* The normal repeating kets try the rest of the pattern or restart from
-       preceding bracket, in the appropriate order. */
+     the preceding bracket, in the appropriate order. In the second case, we can
+     use tail recursion to avoid using another stack frame, unless we have an
+     an atomic group or an unlimited repeat of a group that can match an empty
+     string. */
-       if (*ecode == OP_KETRMIN)
+     if (*ecode == OP_KETRMIN)
+       {
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM7);
+       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       if (*prev == OP_ONCE)
          {
-         RMATCH(rrc, eptr, ecode + 1+LINK_SIZE, offset_top, md, ims, eptrb, 0);
+         RMATCH(eptr, prev, offset_top, md, eptrb, RM8);
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-         RMATCH(rrc, eptr, prev, offset_top, md, ims, eptrb, match_isgroup);
          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
+         RRETURN(MATCH_ONCE);
          }
-       else  /* OP_KETRMAX */
+       if (*prev >= OP_SBRA)    /* Could match an empty string */
          {
-         RMATCH(rrc, eptr, prev, offset_top, md, ims, eptrb, match_isgroup);
+         md->match_function_type = MATCH_CBEGROUP;
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         RMATCH(eptr, prev, offset_top, md, eptrb, RM50);
-         RMATCH(rrc, eptr, ecode + 1+LINK_SIZE, offset_top, md, ims, eptrb, 0);
+         RRETURN(rrc);
+         }
+       ecode = prev;
+       goto TAIL_RECURSE;
+       }
+     else  /* OP_KETRMAX */
+       {
+       if (*prev >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
+       RMATCH(eptr, prev, offset_top, md, eptrb, RM13);
+       if (rrc == MATCH_ONCE && md->once_target == prev) rrc = MATCH_NOMATCH;
+       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       if (*prev == OP_ONCE)
+         {
+         RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM9);
          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         md->once_target = prev;
+         RRETURN(MATCH_ONCE);
          }
+       ecode += 1 + LINK_SIZE;
+       goto TAIL_RECURSE;
        }
+     /* Control never gets here */
-     RRETURN(MATCH_NOMATCH);
+     /* Not multiline mode: start of subject assertion, unless notbol. */
-     /* Start of subject unless notbol, or after internal newline if multiline */
      case OP_CIRC:
      if (md->notbol && eptr == md->start_subject) RRETURN(MATCH_NOMATCH);
-     if ((ims & PCRE_MULTILINE) != 0)
-       {
-       if (eptr != md->start_subject && eptr[-1] != NEWLINE)
-         RRETURN(MATCH_NOMATCH);
-       ecode++;
-       break;
-       }
-     /* ... else fall through */
      /* Start of subject assertion */
-Line 1149 
 for (;;)
+Line 2040 
 for (;;)
      ecode++;
      break;
+     /* Multiline mode: start of subject unless notbol, or after any newline. */
+     case OP_CIRCM:
+     if (md->notbol && eptr == md->start_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr != md->start_subject &&
+         (eptr == md->end_subject || !WAS_NEWLINE(eptr)))
+       RRETURN(MATCH_NOMATCH);
+     ecode++;
+     break;
      /* Start of match assertion */
      case OP_SOM:
-Line 1156 
 for (;;)
+Line 2057 
 for (;;)
      ecode++;
      break;
-     /* 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. */
-     case OP_DOLL:
+     case OP_SET_SOM:
-     if ((ims & PCRE_MULTILINE) != 0)
+     mstart = eptr;
+     ecode++;
+     break;
+     /* Multiline mode: assert before any newline, or before end of subject
+     unless noteol is set. */
+     case OP_DOLLM:
+     if (eptr < md->end_subject)
        {
-       if (eptr < md->end_subject)
+       if (!IS_NEWLINE(eptr))
-         { if (*eptr != NEWLINE) RRETURN(MATCH_NOMATCH); }
+         {
-       else
+         if (md->partial != 0 &&
-         { if (md->noteol) RRETURN(MATCH_NOMATCH); }
+             eptr + 1 >= md->end_subject &&
-       ecode++;
+             NLBLOCK->nltype == NLTYPE_FIXED &&
-       break;
+             NLBLOCK->nllen == 2 &&
+             *eptr == NLBLOCK->nl[0])
+           {
+           md->hitend = TRUE;
+           if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
+           }
+         RRETURN(MATCH_NOMATCH);
+         }
        }
      else
        {
        if (md->noteol) RRETURN(MATCH_NOMATCH);
-       if (!md->endonly)
+       SCHECK_PARTIAL();
-         {
-         if (eptr < md->end_subject - 1 ||
-            (eptr == md->end_subject - 1 && *eptr != NEWLINE))
-           RRETURN(MATCH_NOMATCH);
-         ecode++;
-         break;
-         }
        }
-     /* ... else fall through */
+     ecode++;
+     break;
+     /* Not multiline mode: assert before a terminating newline or before end of
+     subject unless noteol is set. */
+     case OP_DOLL:
+     if (md->noteol) RRETURN(MATCH_NOMATCH);
+     if (!md->endonly) goto ASSERT_NL_OR_EOS;
+     /* ... else fall through for endonly */
      /* End of subject assertion (\z) */
      case OP_EOD:
      if (eptr < md->end_subject) RRETURN(MATCH_NOMATCH);
+     SCHECK_PARTIAL();
      ecode++;
      break;
      /* End of subject or ending \n assertion (\Z) */
      case OP_EODN:
-     if (eptr < md->end_subject - 1 ||
+     ASSERT_NL_OR_EOS:
-        (eptr == md->end_subject - 1 && *eptr != NEWLINE)) RRETURN(MATCH_NOMATCH);
+     if (eptr < md->end_subject &&
+         (!IS_NEWLINE(eptr) || eptr != md->end_subject - md->nllen))
+       {
+       if (md->partial != 0 &&
+           eptr + 1 >= md->end_subject &&
+           NLBLOCK->nltype == NLTYPE_FIXED &&
+           NLBLOCK->nllen == 2 &&
+           *eptr == NLBLOCK->nl[0])
+         {
+         md->hitend = TRUE;
+         if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
+         }
+       RRETURN(MATCH_NOMATCH);
+       }
+     /* Either at end of string or \n before end. */
+     SCHECK_PARTIAL();
      ecode++;
      break;
-Line 1206 
 for (;;)
+Line 2142 
 for (;;)
        /* Find out if the previous and current characters are "word" characters.
        It takes a bit more work in UTF-8 mode. Characters > 255 are assumed to
-       be "non-word" characters. */
+       be "non-word" characters. Remember the earliest consulted character for
+       partial matching. */
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-       if (utf8)
+       if (utf)
          {
+         /* Get status of previous character */
          if (eptr == md->start_subject) prev_is_word = FALSE; else
            {
-           const uschar *lastptr = eptr - 1;
+           PCRE_PUCHAR lastptr = eptr - 1;
-           while((*lastptr & 0xc0) == 0x80) lastptr--;
+           BACKCHAR(lastptr);
+           if (lastptr < md->start_used_ptr) md->start_used_ptr = lastptr;
            GETCHAR(c, lastptr);
+ #ifdef SUPPORT_UCP
+           if (md->use_ucp)
+             {
+             if (c == '_') prev_is_word = TRUE; else
+               {
+               int cat = UCD_CATEGORY(c);
+               prev_is_word = (cat == ucp_L || cat == ucp_N);
+               }
+             }
+           else
+ #endif
            prev_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;
            }
-         if (eptr >= md->end_subject) cur_is_word = FALSE; else
+         /* Get status of next character */
+         if (eptr >= md->end_subject)
+           {
+           SCHECK_PARTIAL();
+           cur_is_word = FALSE;
+           }
+         else
            {
            GETCHAR(c, eptr);
+ #ifdef SUPPORT_UCP
+           if (md->use_ucp)
+             {
+             if (c == '_') cur_is_word = TRUE; else
+               {
+               int cat = UCD_CATEGORY(c);
+               cur_is_word = (cat == ucp_L || cat == ucp_N);
+               }
+             }
+           else
+ #endif
            cur_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;
            }
          }
        else
  #endif
-       /* More streamlined when not in UTF-8 mode */
+       /* Not in UTF-8 mode, but we may still have PCRE_UCP set, and for
+       consistency with the behaviour of \w we do use it in this case. */
          {
-         prev_is_word = (eptr != md->start_subject) &&
+         /* Get status of previous character */
-           ((md->ctypes[eptr[-1]] & ctype_word) != 0);
-         cur_is_word = (eptr < md->end_subject) &&
+         if (eptr == md->start_subject) prev_is_word = FALSE; else
-           ((md->ctypes[*eptr] & ctype_word) != 0);
+           {
+           if (eptr <= md->start_used_ptr) md->start_used_ptr = eptr - 1;
+ #ifdef SUPPORT_UCP
+           if (md->use_ucp)
+             {
+             c = eptr[-1];
+             if (c == '_') prev_is_word = TRUE; else
+               {
+               int cat = UCD_CATEGORY(c);
+               prev_is_word = (cat == ucp_L || cat == ucp_N);
+               }
+             }
+           else
+ #endif
+           prev_is_word = MAX_255(eptr[-1])
+             && ((md->ctypes[eptr[-1]] & ctype_word) != 0);
+           }
+         /* Get status of next character */
+         if (eptr >= md->end_subject)
+           {
+           SCHECK_PARTIAL();
+           cur_is_word = FALSE;
+           }
+         else
+ #ifdef SUPPORT_UCP
+         if (md->use_ucp)
+           {
+           c = *eptr;
+           if (c == '_') cur_is_word = TRUE; else
+             {
+             int cat = UCD_CATEGORY(c);
+             cur_is_word = (cat == ucp_L || cat == ucp_N);
+             }
+           }
+         else
+ #endif
+         cur_is_word = MAX_255(*eptr)
+           && ((md->ctypes[*eptr] & ctype_word) != 0);
          }
        /* Now see if the situation is what we want */
-Line 1242 
 for (;;)
+Line 2252 
 for (;;)
             cur_is_word == prev_is_word : cur_is_word != prev_is_word)
          RRETURN(MATCH_NOMATCH);
        }
      break;
+     /* Match any single character type except newline; have to take care with
+     CRLF newlines and partial matching. */
+     case OP_ANY:
+     if (IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH);
+     if (md->partial != 0 &&
+         eptr + 1 >= md->end_subject &&
+         NLBLOCK->nltype == NLTYPE_FIXED &&
+         NLBLOCK->nllen == 2 &&
+         *eptr == NLBLOCK->nl[0])
+       {
+       md->hitend = TRUE;
+       if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
+       }
-     /* Match a single character type; inline for speed */
+     /* Fall through */
-     case OP_ANY:
+     /* Match any single character whatsoever. */
-     if ((ims & PCRE_DOTALL) == 0 && eptr < md->end_subject && *eptr == NEWLINE)
+     case OP_ALLANY:
+     if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
+       {                            /* not be updated before SCHECK_PARTIAL. */
+       SCHECK_PARTIAL();
        RRETURN(MATCH_NOMATCH);
-     if (eptr++ >= md->end_subject) RRETURN(MATCH_NOMATCH);
+       }
- #ifdef SUPPORT_UTF8
+     eptr++;
-     if (utf8)
+ #ifdef SUPPORT_UTF
-       while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
+     if (utf) ACROSSCHAR(eptr < md->end_subject, *eptr, eptr++);
  #endif
      ecode++;
      break;
-Line 1261 
 for (;;)
+Line 2290 
 for (;;)
      any byte, even newline, independent of the setting of PCRE_DOTALL. */
      case OP_ANYBYTE:
-     if (eptr++ >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
+       {                            /* not be updated before SCHECK_PARTIAL. */
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
+     eptr++;
      ecode++;
      break;
      case OP_NOT_DIGIT:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
         c < 256 &&
  #endif
         (md->ctypes[c] & ctype_digit) != 0
-Line 1279 
 for (;;)
+Line 2317 
 for (;;)
      break;
      case OP_DIGIT:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
-        c >= 256 ||
+        c > 255 ||
  #endif
         (md->ctypes[c] & ctype_digit) == 0
         )
-Line 1292 
 for (;;)
+Line 2334 
 for (;;)
      break;
      case OP_NOT_WHITESPACE:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
         c < 256 &&
  #endif
         (md->ctypes[c] & ctype_space) != 0
-Line 1305 
 for (;;)
+Line 2351 
 for (;;)
      break;
      case OP_WHITESPACE:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
-        c >= 256 ||
+        c > 255 ||
  #endif
         (md->ctypes[c] & ctype_space) == 0
         )
-Line 1318 
 for (;;)
+Line 2368 
 for (;;)
      break;
      case OP_NOT_WORDCHAR:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
         c < 256 &&
  #endif
         (md->ctypes[c] & ctype_word) != 0
-Line 1331 
 for (;;)
+Line 2385 
 for (;;)
      break;
      case OP_WORDCHAR:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
-        c >= 256 ||
+        c > 255 ||
  #endif
         (md->ctypes[c] & ctype_word) == 0
         )
-Line 1343 
 for (;;)
+Line 2401 
 for (;;)
      ecode++;
      break;
+     case OP_ANYNL:
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
+     GETCHARINCTEST(c, eptr);
+     switch(c)
+       {
+       default: RRETURN(MATCH_NOMATCH);
+       case 0x000d:
+       if (eptr >= md->end_subject)
+         {
+         SCHECK_PARTIAL();
+         }
+       else if (*eptr == 0x0a) eptr++;
+       break;
+       case 0x000a:
+       break;
+       case 0x000b:
+       case 0x000c:
+       case 0x0085:
+       case 0x2028:
+       case 0x2029:
+       if (md->bsr_anycrlf) RRETURN(MATCH_NOMATCH);
+       break;
+       }
+     ecode++;
+     break;
+     case OP_NOT_HSPACE:
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
+     GETCHARINCTEST(c, eptr);
+     switch(c)
+       {
+       default: break;
+       case 0x09:      /* HT */
+       case 0x20:      /* SPACE */
+       case 0xa0:      /* NBSP */
+       case 0x1680:    /* OGHAM SPACE MARK */
+       case 0x180e:    /* MONGOLIAN VOWEL SEPARATOR */
+       case 0x2000:    /* EN QUAD */
+       case 0x2001:    /* EM QUAD */
+       case 0x2002:    /* EN SPACE */
+       case 0x2003:    /* EM SPACE */
+       case 0x2004:    /* THREE-PER-EM SPACE */
+       case 0x2005:    /* FOUR-PER-EM SPACE */
+       case 0x2006:    /* SIX-PER-EM SPACE */
+       case 0x2007:    /* FIGURE SPACE */
+       case 0x2008:    /* PUNCTUATION SPACE */
+       case 0x2009:    /* THIN SPACE */
+       case 0x200A:    /* HAIR SPACE */
+       case 0x202f:    /* NARROW NO-BREAK SPACE */
+       case 0x205f:    /* MEDIUM MATHEMATICAL SPACE */
+       case 0x3000:    /* IDEOGRAPHIC SPACE */
+       RRETURN(MATCH_NOMATCH);
+       }
+     ecode++;
+     break;
+     case OP_HSPACE:
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
+     GETCHARINCTEST(c, eptr);
+     switch(c)
+       {
+       default: RRETURN(MATCH_NOMATCH);
+       case 0x09:      /* HT */
+       case 0x20:      /* SPACE */
+       case 0xa0:      /* NBSP */
+       case 0x1680:    /* OGHAM SPACE MARK */
+       case 0x180e:    /* MONGOLIAN VOWEL SEPARATOR */
+       case 0x2000:    /* EN QUAD */
+       case 0x2001:    /* EM QUAD */
+       case 0x2002:    /* EN SPACE */
+       case 0x2003:    /* EM SPACE */
+       case 0x2004:    /* THREE-PER-EM SPACE */
+       case 0x2005:    /* FOUR-PER-EM SPACE */
+       case 0x2006:    /* SIX-PER-EM SPACE */
+       case 0x2007:    /* FIGURE SPACE */
+       case 0x2008:    /* PUNCTUATION SPACE */
+       case 0x2009:    /* THIN SPACE */
+       case 0x200A:    /* HAIR SPACE */
+       case 0x202f:    /* NARROW NO-BREAK SPACE */
+       case 0x205f:    /* MEDIUM MATHEMATICAL SPACE */
+       case 0x3000:    /* IDEOGRAPHIC SPACE */
+       break;
+       }
+     ecode++;
+     break;
+     case OP_NOT_VSPACE:
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
+     GETCHARINCTEST(c, eptr);
+     switch(c)
+       {
+       default: break;
+       case 0x0a:      /* LF */
+       case 0x0b:      /* VT */
+       case 0x0c:      /* FF */
+       case 0x0d:      /* CR */
+       case 0x85:      /* NEL */
+       case 0x2028:    /* LINE SEPARATOR */
+       case 0x2029:    /* PARAGRAPH SEPARATOR */
+       RRETURN(MATCH_NOMATCH);
+       }
+     ecode++;
+     break;
+     case OP_VSPACE:
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
+     GETCHARINCTEST(c, eptr);
+     switch(c)
+       {
+       default: RRETURN(MATCH_NOMATCH);
+       case 0x0a:      /* LF */
+       case 0x0b:      /* VT */
+       case 0x0c:      /* FF */
+       case 0x0d:      /* CR */
+       case 0x85:      /* NEL */
+       case 0x2028:    /* LINE SEPARATOR */
+       case 0x2029:    /* PARAGRAPH SEPARATOR */
+       break;
+       }
+     ecode++;
+     break;
  #ifdef SUPPORT_UCP
      /* Check the next character by Unicode property. We will get here only
      if the support is in the binary; otherwise a compile-time error occurs. */
      case OP_PROP:
      case OP_NOTPROP:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
        {
-       int chartype, rqdtype;
+       const ucd_record *prop = GET_UCD(c);
-       int othercase;
-       int category = _pcre_ucp_findchar(c, &chartype, &othercase);
-       rqdtype = *(++ecode);
+       switch(ecode[1])
-       ecode++;
-       if (rqdtype >= 128)
          {
-         if ((rqdtype - 128 != category) == (op == OP_PROP))
+         case PT_ANY:
+         if (op == OP_NOTPROP) RRETURN(MATCH_NOMATCH);
+         break;
+         case PT_LAMP:
+         if ((prop->chartype == ucp_Lu ||
+              prop->chartype == ucp_Ll ||
+              prop->chartype == ucp_Lt) == (op == OP_NOTPROP))
            RRETURN(MATCH_NOMATCH);
-         }
+         break;
-       else
-         {
+         case PT_GC:
-         if ((rqdtype != chartype) == (op == OP_PROP))
+         if ((ecode[2] != PRIV(ucp_gentype)[prop->chartype]) == (op == OP_PROP))
+           RRETURN(MATCH_NOMATCH);
+         break;
+         case PT_PC:
+         if ((ecode[2] != prop->chartype) == (op == OP_PROP))
            RRETURN(MATCH_NOMATCH);
+         break;
+         case PT_SC:
+         if ((ecode[2] != prop->script) == (op == OP_PROP))
+           RRETURN(MATCH_NOMATCH);
+         break;
+         /* These are specials */
+         case PT_ALNUM:
+         if ((PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
+              PRIV(ucp_gentype)[prop->chartype] == ucp_N) == (op == OP_NOTPROP))
+           RRETURN(MATCH_NOMATCH);
+         break;
+         case PT_SPACE:    /* Perl space */
+         if ((PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
+              c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
+                == (op == OP_NOTPROP))
+           RRETURN(MATCH_NOMATCH);
+         break;
+         case PT_PXSPACE:  /* POSIX space */
+         if ((PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
+              c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
+              c == CHAR_FF || c == CHAR_CR)
+                == (op == OP_NOTPROP))
+           RRETURN(MATCH_NOMATCH);
+         break;
+         case PT_WORD:
+         if ((PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
+              PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
+              c == CHAR_UNDERSCORE) == (op == OP_NOTPROP))
+           RRETURN(MATCH_NOMATCH);
+         break;
+         /* This should never occur */
+         default:
+         RRETURN(PCRE_ERROR_INTERNAL);
          }
+       ecode += 3;
        }
      break;
-Line 1376 
 for (;;)
+Line 2633 
 for (;;)
      is in the binary; otherwise a compile-time error occurs. */
      case OP_EXTUNI:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
+     if (UCD_CATEGORY(c) == ucp_M) RRETURN(MATCH_NOMATCH);
+     while (eptr < md->end_subject)
        {
-       int chartype;
+       int len = 1;
-       int othercase;
+       if (!utf) c = *eptr; else { GETCHARLEN(c, eptr, len); }
-       int category = _pcre_ucp_findchar(c, &chartype, &othercase);
+       if (UCD_CATEGORY(c) != ucp_M) break;
-       if (category == ucp_M) RRETURN(MATCH_NOMATCH);
+       eptr += len;
-       while (eptr < md->end_subject)
-         {
-         int len = 1;
-         if (!utf8) c = *eptr; else
-           {
-           GETCHARLEN(c, eptr, len);
-           }
-         category = _pcre_ucp_findchar(c, &chartype, &othercase);
-         if (category != ucp_M) break;
-         eptr += len;
-         }
        }
+     CHECK_PARTIAL();
      ecode++;
      break;
  #endif
-Line 1409 
 for (;;)
+Line 2662 
 for (;;)
      loops). */
      case OP_REF:
-       {
+     case OP_REFI:
-       offset = GET2(ecode, 1) << 1;               /* Doubled ref number */
+     caseless = op == OP_REFI;
-       ecode += 3;                                 /* Advance past item */
+     offset = GET2(ecode, 1) << 1;               /* Doubled ref number */
+     ecode += 1 + IMM2_SIZE;
-       /* 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];
-       /* Set up for repetition, or handle the non-repeated case */
+     (a) In the default, Perl-compatible state, set the length negative;
+     this ensures that every attempt at a match fails. We can't just fail
+     here, because of the possibility of quantifiers with zero minima.
-       switch (*ecode)
+     (b) If the JavaScript compatibility flag is set, set the length to zero
-         {
+     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;
-         case OP_CRRANGE:
+     Otherwise, set the length to the length of what was matched by the
-         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;
-         default:               /* No repeat follows */
+     if (offset >= offset_top || md->offset_vector[offset] < 0)
-         if (!match_ref(offset, eptr, length, md, ims)) RRETURN(MATCH_NOMATCH);
+       length = (md->jscript_compat)? 0 : -1;
-         eptr += length;
+     else
-         continue;              /* With the main loop */
+       length = md->offset_vector[offset+1] - md->offset_vector[offset];
+     /* Set up for repetition, or handle the non-repeated case */
+     switch (*ecode)
+       {
+       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;
+       case OP_CRRANGE:
+       case OP_CRMINRANGE:
+       minimize = (*ecode == OP_CRMINRANGE);
+       min = GET2(ecode, 1);
+       max = GET2(ecode, 1 + IMM2_SIZE);
+       if (max == 0) max = INT_MAX;
+       ecode += 1 + 2 * IMM2_SIZE;
+       break;
+       default:               /* No repeat follows */
+       if ((length = match_ref(offset, eptr, length, md, caseless)) < 0)
+         {
+         if (length == -2) eptr = md->end_subject;   /* Partial match */
+         CHECK_PARTIAL();
+         RRETURN(MATCH_NOMATCH);
          }
+       eptr += length;
+       continue;              /* With the main loop */
+       }
-       /* If the length of the reference is zero, just continue with the
+     /* Handle repeated back references. If the length of the reference is
-       main loop. */
+     zero, just continue with the main loop. If the length is negative, it
+     means the reference is unset in non-Java-compatible mode. If the minimum is
+     zero, we can continue at the same level without recursion. For any other
+     minimum, carrying on will result in NOMATCH. */
-       if (length == 0) continue;
+     if (length == 0) continue;
+     if (length < 0 && min == 0) continue;
-       /* First, ensure the minimum number of matches are present. We get back
+     /* First, ensure the minimum number of matches are present. We get back
-       the length of the reference string explicitly rather than passing the
+     the length of the reference string explicitly rather than passing the
-       address of eptr, so that eptr can be a register variable. */
+     address of eptr, so that eptr can be a register variable. */
-       for (i = 1; i <= min; i++)
+     for (i = 1; i <= min; i++)
+       {
+       int slength;
+       if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
          {
-         if (!match_ref(offset, eptr, length, md, ims)) RRETURN(MATCH_NOMATCH);
+         if (slength == -2) eptr = md->end_subject;   /* Partial match */
-         eptr += length;
+         CHECK_PARTIAL();
+         RRETURN(MATCH_NOMATCH);
          }
+       eptr += slength;
+       }
-       /* If min = max, continue at the same level without recursion.
+     /* If min = max, continue at the same level without recursion.
-       They are not both allowed to be zero. */
+     They are not both allowed to be zero. */
-       if (min == max) continue;
+     if (min == max) continue;
-       /* If minimizing, keep trying and advancing the pointer */
+     /* If minimizing, keep trying and advancing the pointer */
-       if (minimize)
+     if (minimize)
+       {
+       for (fi = min;; fi++)
          {
-         for (fi = min;; fi++)
+         int slength;
+         RMATCH(eptr, ecode, offset_top, md, eptrb, RM14);
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         if (fi >= max) RRETURN(MATCH_NOMATCH);
+         if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           if (slength == -2) eptr = md->end_subject;   /* Partial match */
-           if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+           CHECK_PARTIAL();
-           if (fi >= max || !match_ref(offset, eptr, length, md, ims))
+           RRETURN(MATCH_NOMATCH);
-             RRETURN(MATCH_NOMATCH);
-           eptr += length;
            }
-         /* Control never gets here */
+         eptr += slength;
          }
+       /* Control never gets here */
+       }
-       /* If maximizing, find the longest string and work backwards */
+     /* If maximizing, find the longest string and work backwards */
-       else
+     else
+       {
+       pp = eptr;
+       for (i = min; i < max; i++)
          {
-         pp = eptr;
+         int slength;
-         for (i = min; i < max; i++)
+         if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
-           {
-           if (!match_ref(offset, eptr, length, md, ims)) break;
-           eptr += length;
-           }
-         while (eptr >= pp)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           /* Can't use CHECK_PARTIAL because we don't want to update eptr in
-           if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+           the soft partial matching case. */
-           eptr -= length;
+           if (slength == -2 && md->partial != 0 &&
+               md->end_subject > md->start_used_ptr)
+             {
+             md->hitend = TRUE;
+             if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);
+             }
+           break;
            }
-         RRETURN(MATCH_NOMATCH);
+         eptr += slength;
+         }
+       while (eptr >= pp)
+         {
+         RMATCH(eptr, ecode, offset_top, md, eptrb, RM15);
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         eptr -= length;
          }
+       RRETURN(MATCH_NOMATCH);
        }
      /* Control never gets here */
      /* Match a bit-mapped character class, possibly repeatedly. This op code is
      used when all the characters in the class have values in the range 0-255,
      and either the matching is caseful, or the characters are in the range
-Line 1526 
 for (;;)
+Line 2820 
 for (;;)
      case OP_NCLASS:
      case OP_CLASS:
        {
+       /* The data variable is saved across frames, so the byte map needs to
+       be stored there. */
+ #define BYTE_MAP ((pcre_uint8 *)data)
        data = ecode + 1;                /* Save for matching */
-       ecode += 33;                     /* Advance past the item */
+       ecode += 1 + (32 / sizeof(pcre_uchar)); /* Advance past the item */
        switch (*ecode)
          {
-Line 1548 
 for (;;)
+Line 2845 
 for (;;)
          case OP_CRMINRANGE:
          minimize = (*ecode == OP_CRMINRANGE);
          min = GET2(ecode, 1);
-         max = GET2(ecode, 3);
+         max = GET2(ecode, 1 + IMM2_SIZE);
          if (max == 0) max = INT_MAX;
-         ecode += 5;
+         ecode += 1 + 2 * IMM2_SIZE;
          break;
          default:               /* No repeat follows */
-Line 1560 
 for (;;)
+Line 2857 
 for (;;)
        /* First, ensure the minimum number of matches are present. */
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-       /* UTF-8 mode */
+       if (utf)
-       if (utf8)
          {
          for (i = 1; i <= min; i++)
            {
-           if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            GETCHARINC(c, eptr);
            if (c > 255)
              {
              if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
              }
            else
-             {
+             if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
-             if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
-             }
            }
          }
        else
  #endif
-       /* Not UTF-8 mode */
+       /* Not UTF mode */
          {
          for (i = 1; i <= min; i++)
            {
-           if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            c = *eptr++;
-           if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
+ #ifndef COMPILE_PCRE8
+           if (c > 255)
+             {
+             if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
+             }
+           else
+ #endif
+             if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
            }
          }
-Line 1600 
 for (;;)
+Line 2909 
 for (;;)
        if (minimize)
          {
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-         /* UTF-8 mode */
+         if (utf)
-         if (utf8)
            {
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM16);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
              GETCHARINC(c, eptr);
              if (c > 255)
                {
                if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
                }
              else
-               {
+               if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
-               if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
-               }
              }
            }
          else
  #endif
-         /* Not UTF-8 mode */
+         /* Not UTF mode */
            {
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM17);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
              c = *eptr++;
-             if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
+ #ifndef COMPILE_PCRE8
+             if (c > 255)
+               {
+               if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
+               }
+             else
+ #endif
+               if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
              }
            }
          /* Control never gets here */
-Line 1642 
 for (;;)
+Line 2965 
 for (;;)
          {
          pp = eptr;
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-         /* UTF-8 mode */
+         if (utf)
-         if (utf8)
            {
            for (i = min; i < max; i++)
              {
              int len = 1;
-             if (eptr >= md->end_subject) break;
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               break;
+               }
              GETCHARLEN(c, eptr, len);
              if (c > 255)
                {
                if (op == OP_CLASS) break;
                }
              else
-               {
+               if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) break;
-               if ((data[c/8] & (1 << (c&7))) == 0) break;
-               }
              eptr += len;
              }
            for (;;)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM18);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (eptr-- == pp) break;        /* Stop if tried at original pos */
              BACKCHAR(eptr);
-Line 1671 
 for (;;)
+Line 2995 
 for (;;)
            }
          else
  #endif
-           /* Not UTF-8 mode */
+           /* Not UTF mode */
            {
            for (i = min; i < max; i++)
              {
-             if (eptr >= md->end_subject) break;
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               break;
+               }
              c = *eptr;
-             if ((data[c/8] & (1 << (c&7))) == 0) break;
+ #ifndef COMPILE_PCRE8
+             if (c > 255)
+               {
+               if (op == OP_CLASS) break;
+               }
+             else
+ #endif
+               if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) break;
              eptr++;
              }
            while (eptr >= pp)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM19);
-             eptr--;
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+             eptr--;
              }
            }
          RRETURN(MATCH_NOMATCH);
          }
+ #undef BYTE_MAP
        }
      /* Control never gets here */
      /* Match an extended character class. This opcode is encountered only
-     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
+     mode, because Unicode properties are supported in non-UTF-8 mode. */
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
      case OP_XCLASS:
        {
        data = ecode + 1 + LINK_SIZE;                /* Save for matching */
-Line 1722 
 for (;;)
+Line 3059 
 for (;;)
          case OP_CRMINRANGE:
          minimize = (*ecode == OP_CRMINRANGE);
          min = GET2(ecode, 1);
-         max = GET2(ecode, 3);
+         max = GET2(ecode, 1 + IMM2_SIZE);
          if (max == 0) max = INT_MAX;
-         ecode += 5;
+         ecode += 1 + 2 * IMM2_SIZE;
          break;
          default:               /* No repeat follows */
-Line 1736 
 for (;;)
+Line 3073 
 for (;;)
        for (i = 1; i <= min; i++)
          {
-         if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+         if (eptr >= md->end_subject)
-         GETCHARINC(c, eptr);
+           {
-         if (!_pcre_xclass(c, data)) RRETURN(MATCH_NOMATCH);
+           SCHECK_PARTIAL();
+           RRETURN(MATCH_NOMATCH);
+           }
+         GETCHARINCTEST(c, eptr);
+         if (!PRIV(xclass)(c, data, utf)) RRETURN(MATCH_NOMATCH);
          }
        /* If max == min we can continue with the main loop without the
-Line 1753 
 for (;;)
+Line 3094 
 for (;;)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM20);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-           if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+           if (fi >= max) RRETURN(MATCH_NOMATCH);
-           GETCHARINC(c, eptr);
+           if (eptr >= md->end_subject)
-           if (!_pcre_xclass(c, data)) RRETURN(MATCH_NOMATCH);
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           GETCHARINCTEST(c, eptr);
+           if (!PRIV(xclass)(c, data, utf)) RRETURN(MATCH_NOMATCH);
            }
          /* Control never gets here */
          }
-Line 1770 
 for (;;)
+Line 3116 
 for (;;)
          for (i = min; i < max; i++)
            {
            int len = 1;
-           if (eptr >= md->end_subject) break;
+           if (eptr >= md->end_subject)
-           GETCHARLEN(c, eptr, len);
+             {
-           if (!_pcre_xclass(c, data)) break;
+             SCHECK_PARTIAL();
+             break;
+             }
+ #ifdef SUPPORT_UTF
+           GETCHARLENTEST(c, eptr, len);
+ #else
+           c = *eptr;
+ #endif
+           if (!PRIV(xclass)(c, data, utf)) break;
            eptr += len;
            }
          for(;;)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM21);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (eptr-- == pp) break;        /* Stop if tried at original pos */
-           BACKCHAR(eptr)
+ #ifdef SUPPORT_UTF
+           if (utf) BACKCHAR(eptr);
+ #endif
            }
          RRETURN(MATCH_NOMATCH);
          }
-Line 1792 
 for (;;)
+Line 3148 
 for (;;)
      /* Match a single character, casefully */
      case OP_CHAR:
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-     if (utf8)
+     if (utf)
        {
        length = 1;
        ecode++;
        GETCHARLEN(fc, ecode, length);
-       if (length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
+       if (length > md->end_subject - eptr)
+         {
+         CHECK_PARTIAL();             /* Not SCHECK_PARTIAL() */
+         RRETURN(MATCH_NOMATCH);
+         }
        while (length-- > 0) if (*ecode++ != *eptr++) RRETURN(MATCH_NOMATCH);
        }
      else
  #endif
+     /* Not UTF mode */
-     /* Non-UTF-8 mode */
        {
-       if (md->end_subject - eptr < 1) RRETURN(MATCH_NOMATCH);
+       if (md->end_subject - eptr < 1)
+         {
+         SCHECK_PARTIAL();            /* This one can use SCHECK_PARTIAL() */
+         RRETURN(MATCH_NOMATCH);
+         }
        if (ecode[1] != *eptr++) RRETURN(MATCH_NOMATCH);
        ecode += 2;
        }
      break;
-     /* Match a single character, caselessly */
+     /* Match a single character, caselessly. If we are at the end of the
+     subject, give up immediately. */
+     case OP_CHARI:
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
-     case OP_CHARNC:
+ #ifdef SUPPORT_UTF
- #ifdef SUPPORT_UTF8
+     if (utf)
-     if (utf8)
        {
        length = 1;
        ecode++;
        GETCHARLEN(fc, ecode, length);
-       if (length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
        /* If the pattern character's value is < 128, we have only one byte, and
-       can use the fast lookup table. */
+       we know that its other case must also be one byte long, so we can use the
+       fast lookup table. We know that there is at least one byte left in the
+       subject. */
        if (fc < 128)
          {
-         if (md->lcc[*ecode++] != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);
+         if (md->lcc[fc]
+             != TABLE_GET(*eptr, md->lcc, *eptr)) RRETURN(MATCH_NOMATCH);
+         ecode++;
+         eptr++;
          }
-       /* Otherwise we must pick up the subject character */
+       /* Otherwise we must pick up the subject character. Note that we cannot
+       use the value of "length" to check for sufficient bytes left, because the
+       other case of the character may have more or fewer bytes.  */
        else
          {
-         int dc;
+         unsigned int dc;
          GETCHARINC(dc, eptr);
          ecode += length;
          /* If we have Unicode property support, we can use it to test the other
-         case of the character, if there is one. The result of _pcre_ucp_findchar() is
+         case of the character, if there is one. */
-         < 0 if the char isn't found, and othercase is returned as zero if there
-         isn't one. */
          if (fc != dc)
            {
  #ifdef SUPPORT_UCP
-           int chartype;
+           if (dc != UCD_OTHERCASE(fc))
-           int othercase;
-           if (_pcre_ucp_findchar(fc, &chartype, &othercase) < 0 || dc != othercase)
  #endif
              RRETURN(MATCH_NOMATCH);
            }
          }
        }
      else
- #endif   /* SUPPORT_UTF8 */
+ #endif   /* SUPPORT_UTF */
-     /* Non-UTF-8 mode */
+     /* Not UTF mode */
        {
-       if (md->end_subject - eptr < 1) RRETURN(MATCH_NOMATCH);
+       if (TABLE_GET(ecode[1], md->lcc, ecode[1])
-       if (md->lcc[ecode[1]] != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);
+           != TABLE_GET(*eptr, md->lcc, *eptr)) RRETURN(MATCH_NOMATCH);
+       eptr++;
        ecode += 2;
        }
      break;
-     /* Match a single character repeatedly; different opcodes share code. */
+     /* Match a single character repeatedly. */
      case OP_EXACT:
+     case OP_EXACTI:
      min = max = GET2(ecode, 1);
-     ecode += 3;
+     ecode += 1 + IMM2_SIZE;
      goto REPEATCHAR;
+     case OP_POSUPTO:
+     case OP_POSUPTOI:
+     possessive = TRUE;
+     /* Fall through */
      case OP_UPTO:
+     case OP_UPTOI:
      case OP_MINUPTO:
+     case OP_MINUPTOI:
      min = 0;
      max = GET2(ecode, 1);
-     minimize = *ecode == OP_MINUPTO;
+     minimize = *ecode == OP_MINUPTO || *ecode == OP_MINUPTOI;
-     ecode += 3;
+     ecode += 1 + IMM2_SIZE;
+     goto REPEATCHAR;
+     case OP_POSSTAR:
+     case OP_POSSTARI:
+     possessive = TRUE;
+     min = 0;
+     max = INT_MAX;
+     ecode++;
+     goto REPEATCHAR;
+     case OP_POSPLUS:
+     case OP_POSPLUSI:
+     possessive = TRUE;
+     min = 1;
+     max = INT_MAX;
+     ecode++;
+     goto REPEATCHAR;
+     case OP_POSQUERY:
+     case OP_POSQUERYI:
+     possessive = TRUE;
+     min = 0;
+     max = 1;
+     ecode++;
      goto REPEATCHAR;
      case OP_STAR:
+     case OP_STARI:
      case OP_MINSTAR:
+     case OP_MINSTARI:
      case OP_PLUS:
+     case OP_PLUSI:
      case OP_MINPLUS:
+     case OP_MINPLUSI:
      case OP_QUERY:
+     case OP_QUERYI:
      case OP_MINQUERY:
-     c = *ecode++ - OP_STAR;
+     case OP_MINQUERYI:
+     c = *ecode++ - ((op < OP_STARI)? OP_STAR : OP_STARI);
      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;
-     /* Common code for all repeated single-character matches. We can give
+     /* Common code for all repeated single-character matches. */
-     up quickly if there are fewer than the minimum number of characters left in
-     the subject. */
      REPEATCHAR:
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-     if (utf8)
+     if (utf)
        {
        length = 1;
        charptr = ecode;
        GETCHARLEN(fc, ecode, length);
-       if (min * length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
        ecode += length;
        /* Handle multibyte character matching specially here. There is
-Line 1913 
 for (;;)
+Line 3320 
 for (;;)
        if (length > 1)
          {
-         int oclength = 0;
-         uschar occhars[8];
  #ifdef SUPPORT_UCP
-         int othercase;
+         unsigned int othercase;
-         int chartype;
+         if (op >= OP_STARI &&     /* Caseless */
-         if ((ims & PCRE_CASELESS) != 0 &&
+             (othercase = UCD_OTHERCASE(fc)) != fc)
-              _pcre_ucp_findchar(fc, &chartype, &othercase) >= 0 &&
+           oclength = PRIV(ord2utf)(othercase, occhars);
-              othercase > 0)
+         else oclength = 0;
-           oclength = _pcre_ord2utf8(othercase, occhars);
  #endif  /* SUPPORT_UCP */
          for (i = 1; i <= min; i++)
            {
-           if (memcmp(eptr, charptr, length) == 0) eptr += length;
+           if (eptr <= md->end_subject - length &&
-           /* Need braces because of following else */
+             memcmp(eptr, charptr, IN_UCHARS(length)) == 0) eptr += length;
-           else if (oclength == 0) { RRETURN(MATCH_NOMATCH); }
+ #ifdef SUPPORT_UCP
+           else if (oclength > 0 &&
+                    eptr <= md->end_subject - oclength &&
+                    memcmp(eptr, occhars, IN_UCHARS(oclength)) == 0) eptr += oclength;
+ #endif  /* SUPPORT_UCP */
            else
              {
-             if (memcmp(eptr, occhars, oclength) != 0) RRETURN(MATCH_NOMATCH);
+             CHECK_PARTIAL();
-             eptr += oclength;
+             RRETURN(MATCH_NOMATCH);
              }
            }
-Line 1943 
 for (;;)
+Line 3350 
 for (;;)
            {
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM22);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
-             if (memcmp(eptr, charptr, length) == 0) eptr += length;
+             if (eptr <= md->end_subject - length &&
-             /* Need braces because of following else */
+               memcmp(eptr, charptr, IN_UCHARS(length)) == 0) eptr += length;
-             else if (oclength == 0) { RRETURN(MATCH_NOMATCH); }
+ #ifdef SUPPORT_UCP
+             else if (oclength > 0 &&
+                      eptr <= md->end_subject - oclength &&
+                      memcmp(eptr, occhars, IN_UCHARS(oclength)) == 0) eptr += oclength;
+ #endif  /* SUPPORT_UCP */
              else
                {
-               if (memcmp(eptr, occhars, oclength) != 0) RRETURN(MATCH_NOMATCH);
+               CHECK_PARTIAL();
-               eptr += oclength;
+               RRETURN(MATCH_NOMATCH);
                }
              }
            /* Control never gets here */
            }
-         else
+         else  /* Maximize */
            {
            pp = eptr;
            for (i = min; i < max; i++)
              {
-             if (eptr > md->end_subject - length) break;
+             if (eptr <= md->end_subject - length &&
-             if (memcmp(eptr, charptr, length) == 0) eptr += length;
+                 memcmp(eptr, charptr, IN_UCHARS(length)) == 0) eptr += length;
-             else if (oclength == 0) break;
+ #ifdef SUPPORT_UCP
+             else if (oclength > 0 &&
+                      eptr <= md->end_subject - oclength &&
+                      memcmp(eptr, occhars, IN_UCHARS(oclength)) == 0) eptr += oclength;
+ #endif  /* SUPPORT_UCP */
              else
                {
-               if (memcmp(eptr, occhars, oclength) != 0) break;
+               CHECK_PARTIAL();
-               eptr += oclength;
+               break;
                }
              }
-           while (eptr >= pp)
-            {
+           if (possessive) continue;
-            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
-            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+           for(;;)
-            eptr -= length;
+             {
-            }
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM23);
-           RRETURN(MATCH_NOMATCH);
+             if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+             if (eptr == pp) { RRETURN(MATCH_NOMATCH); }
+ #ifdef SUPPORT_UCP
+             eptr--;
+             BACKCHAR(eptr);
+ #else   /* without SUPPORT_UCP */
+             eptr -= length;
+ #endif  /* SUPPORT_UCP */
+             }
            }
          /* Control never gets here */
          }
-Line 1987 
 for (;;)
+Line 3411 
 for (;;)
        value of fc will always be < 128. */
        }
      else
- #endif  /* SUPPORT_UTF8 */
+ #endif  /* SUPPORT_UTF */
+       /* When not in UTF-8 mode, load a single-byte character. */
-     /* When not in UTF-8 mode, load a single-byte character. */
-       {
-       if (min > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
        fc = *ecode++;
-       }
-     /* The value of fc at this point is always less than 256, though we may or
+     /* The value of fc at this point is always one character, though we may
-     may not be in UTF-8 mode. The code is duplicated for the caseless and
+     or may not be in UTF mode. The code is duplicated for the caseless and
      caseful cases, for speed, since matching characters is likely to be quite
      common. First, ensure the minimum number of matches are present. If min =
      max, continue at the same level without recursing. Otherwise, if
-Line 2007 
 for (;;)
+Line 3427 
 for (;;)
      DPRINTF(("matching %c{%d,%d} against subject %.*s\n", fc, min, max,
        max, eptr));
-     if ((ims & PCRE_CASELESS) != 0)
+     if (op >= OP_STARI)  /* Caseless */
        {
-       fc = md->lcc[fc];
+ #ifdef COMPILE_PCRE8
+       /* fc must be < 128 if UTF is enabled. */
+       foc = md->fcc[fc];
+ #else
+ #ifdef SUPPORT_UTF
+ #ifdef SUPPORT_UCP
+       if (utf && fc > 127)
+         foc = UCD_OTHERCASE(fc);
+ #else
+       if (utf && fc > 127)
+         foc = fc;
+ #endif /* SUPPORT_UCP */
+       else
+ #endif /* SUPPORT_UTF */
+         foc = TABLE_GET(fc, md->fcc, fc);
+ #endif /* COMPILE_PCRE8 */
        for (i = 1; i <= min; i++)
-         if (fc != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);
+         {
+         if (eptr >= md->end_subject)
+           {
+           SCHECK_PARTIAL();
+           RRETURN(MATCH_NOMATCH);
+           }
+         if (fc != *eptr && foc != *eptr) RRETURN(MATCH_NOMATCH);
+         eptr++;
+         }
        if (min == max) continue;
        if (minimize)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM24);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-           if (fi >= max || eptr >= md->end_subject ||
+           if (fi >= max) RRETURN(MATCH_NOMATCH);
-               fc != md->lcc[*eptr++])
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
              RRETURN(MATCH_NOMATCH);
+             }
+           if (fc != *eptr && foc != *eptr) RRETURN(MATCH_NOMATCH);
+           eptr++;
            }
          /* Control never gets here */
          }
-       else
+       else  /* Maximize */
          {
          pp = eptr;
          for (i = min; i < max; i++)
            {
-           if (eptr >= md->end_subject || fc != md->lcc[*eptr]) break;
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             break;