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

Diff of /code/trunk/pcre_exec.c

Parent Directory | Revision Log | View Patch Patch

-revision 511 by ph10,
Mon Mar 29 09:25:38 2010 UTC
+revision 716 by ph10,
Tue Oct  4 16:38:05 2011 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2010 University of Cambridge
+            Copyright (c) 1997-2011 University of Cambridge
  -----------------------------------------------------------------------------
  Redistribution and use in source and binary forms, with or without
 Line 57 
 possible. There are also some static sup
  #undef min
  #undef max
- /* 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_cbegroup       0x02  /* Could-be-empty unlimited repeat 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 73 
 negative to avoid the external error cod
+Line 75 
 negative to avoid the external error cod
  #define MATCH_ACCEPT       (-999)
  #define MATCH_COMMIT       (-998)
- #define MATCH_PRUNE        (-997)
+ #define MATCH_KETRPOS      (-997)
- #define MATCH_SKIP         (-996)
+ #define MATCH_ONCE         (-996)
- #define MATCH_SKIP_ARG     (-995)
+ #define MATCH_PRUNE        (-995)
- #define MATCH_THEN         (-994)
+ #define MATCH_SKIP         (-994)
+ #define MATCH_SKIP_ARG     (-993)
+ #define MATCH_THEN         (-992)
  /* This is a convenience macro for code that occurs many times. */
-Line 132 
 while (length-- > 0)
+Line 136 
 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 if not matched, otherwise the number of subject bytes matched
  */
- static BOOL
+ static int
  match_ref(int offset, register USPTR eptr, int length, match_data *md,
-   unsigned long int ims)
+   BOOL caseless)
  {
- USPTR p = md->start_subject + md->offset_vector[offset];
+ USPTR eptr_start = eptr;
+ register USPTR p = md->start_subject + md->offset_vector[offset];
  #ifdef PCRE_DEBUG
  if (eptr >= md->end_subject)
-Line 164 
 pchars(p, length, FALSE, md);
+Line 171 
 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). */
- if (length > md->end_subject - eptr) return FALSE;
+ if (length < 0) return -1;
  /* 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)
    {
  #ifdef SUPPORT_UTF8
  #ifdef SUPPORT_UCP
    if (md->utf8)
      {
-     USPTR endptr = eptr + length;
+     /* Match characters up to the end of the reference. NOTE: the number of
-     while (eptr < endptr)
+     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). */
+     USPTR endptr = p + length;
+     while (p < endptr)
        {
        int c, d;
+       if (eptr >= md->end_subject) return -1;
        GETCHARINC(c, eptr);
        GETCHARINC(d, p);
-       if (c != d && c != UCD_OTHERCASE(d)) return FALSE;
+       if (c != d && c != UCD_OTHERCASE(d)) return -1;
        }
      }
    else
-Line 193 
 if ((ims & PCRE_CASELESS) != 0)
+Line 209 
 if ((ims & PCRE_CASELESS) != 0)
    /* 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 + length > md->end_subject) return -1;
-     { if (md->lcc[*p++] != md->lcc[*eptr++]) return FALSE; }
+     while (length-- > 0)
+       { if (md->lcc[*p++] != md->lcc[*eptr++]) return -1; }
+     }
    }
  /* 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; }
+   {
+   if (eptr + length > md->end_subject) return -1;
+   while (length-- > 0) if (*p++ != *eptr++) return -1;
+   }
- return TRUE;
+ return eptr - eptr_start;
  }
-Line 255 
 enum { RM1=1, RM2,  RM3,  RM4,  RM5,  RM
+Line 276 
 enum { RM1=1, RM2,  RM3,  RM4,  RM5,  RM
         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 };
+        RM51,  RM52, RM53, RM54, RM55, RM56, RM57, RM58, RM59, RM60,
+        RM61,  RM62, RM63 };
  /* 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
-Line 265 
 actually used in this definition. */
+Line 287 
 actually used in this definition. */
  #define REGISTER register
  #ifdef PCRE_DEBUG
- #define RMATCH(ra,rb,rc,rd,re,rf,rg,rw) \
+ #define RMATCH(ra,rb,rc,rd,re,rw) \
    { \
    printf("match() called in line %d\n", __LINE__); \
-   rrc = match(ra,rb,mstart,markptr,rc,rd,re,rf,rg,rdepth+1); \
+   rrc = match(ra,rb,mstart,markptr,rc,rd,re,rdepth+1); \
    printf("to line %d\n", __LINE__); \
    }
  #define RRETURN(ra) \
-Line 277 
 actually used in this definition. */
+Line 299 
 actually used in this definition. */
    return ra; \
    }
  #else
- #define RMATCH(ra,rb,rc,rd,re,rf,rg,rw) \
+ #define RMATCH(ra,rb,rc,rd,re,rw) \
-   rrc = match(ra,rb,mstart,markptr,rc,rd,re,rf,rg,rdepth+1)
+   rrc = match(ra,rb,mstart,markptr,rc,rd,re,rdepth+1)
  #define RRETURN(ra) return ra
  #endif
-Line 291 
 argument of match(), which never changes
+Line 313 
 argument of match(), which never changes
  #define REGISTER
- #define RMATCH(ra,rb,rc,rd,re,rf,rg,rw)\
+ #define RMATCH(ra,rb,rc,rd,re,rw)\
    {\
-   heapframe *newframe = (pcre_stack_malloc)(sizeof(heapframe));\
+   heapframe *newframe = (heapframe *)(pcre_stack_malloc)(sizeof(heapframe));\
+   if (newframe == NULL) RRETURN(PCRE_ERROR_NOMEMORY);\
    frame->Xwhere = rw; \
    newframe->Xeptr = ra;\
    newframe->Xecode = rb;\
    newframe->Xmstart = mstart;\
    newframe->Xmarkptr = markptr;\
    newframe->Xoffset_top = rc;\
-   newframe->Xims = re;\
+   newframe->Xeptrb = re;\
-   newframe->Xeptrb = rf;\
-   newframe->Xflags = rg;\
    newframe->Xrdepth = frame->Xrdepth + 1;\
    newframe->Xprevframe = frame;\
    frame = newframe;\
-Line 314 
 argument of match(), which never changes
+Line 335 
 argument of match(), which never changes
  #define RRETURN(ra)\
    {\
-   heapframe *newframe = frame;\
+   heapframe *oldframe = frame;\
-   frame = newframe->Xprevframe;\
+   frame = oldframe->Xprevframe;\
-   (pcre_stack_free)(newframe);\
+   (pcre_stack_free)(oldframe);\
    if (frame != NULL)\
      {\
      rrc = ra;\
-Line 338 
 typedef struct heapframe {
+Line 359 
 typedef struct heapframe {
    USPTR Xmstart;
    USPTR Xmarkptr;
    int Xoffset_top;
-   long int Xims;
    eptrblock *Xeptrb;
-   int Xflags;
    unsigned int Xrdepth;
    /* Function local variables */
-Line 361 
 typedef struct heapframe {
+Line 380 
 typedef struct heapframe {
    BOOL Xcondition;
    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 Xprop_chartype;
-   int Xprop_script;
    int Xoclength;
    uschar Xocchars[8];
  #endif
-Line 420 
 immediately. The second one is used when
+Line 434 
 immediately. The second one is used when
  the subject. */
  #define CHECK_PARTIAL()\
-   if (md->partial != 0 && eptr >= md->end_subject && eptr > mstart)\
+   if (md->partial != 0 && eptr >= md->end_subject && \
-     {\
+       eptr > md->start_used_ptr) \
-     md->hitend = TRUE;\
+     { \
-     if (md->partial > 1) MRRETURN(PCRE_ERROR_PARTIAL);\
+     md->hitend = TRUE; \
+     if (md->partial > 1) MRRETURN(PCRE_ERROR_PARTIAL); \
      }
  #define SCHECK_PARTIAL()\
-   if (md->partial != 0 && eptr > mstart)\
+   if (md->partial != 0 && eptr > md->start_used_ptr) \
-     {\
+     { \
-     md->hitend = TRUE;\
+     md->hitend = TRUE; \
-     if (md->partial > 1) MRRETURN(PCRE_ERROR_PARTIAL);\
+     if (md->partial > 1) MRRETURN(PCRE_ERROR_PARTIAL); \
      }
-Line 447 
 Arguments:
+Line 462 
 Arguments:
     markptr     pointer to the most recent MARK name, or NULL
     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
-                  match_condassert - this is an assertion condition
-                  match_cbegroup - this is the start of an unlimited repeat
-                    group that can match an empty string
     rdepth      the recursion depth
  Returns:       MATCH_MATCH if matched            )  these values are >= 0
-Line 465 
 Returns:       MATCH_MATCH if matched
+Line 475 
 Returns:       MATCH_MATCH if matched
  static int
  match(REGISTER USPTR eptr, REGISTER const uschar *ecode, USPTR mstart,
-   const uschar *markptr, int offset_top, match_data *md, unsigned long int ims,
+   const uschar *markptr, int offset_top, match_data *md, eptrblock *eptrb,
-   eptrblock *eptrb, int flags, unsigned int rdepth)
+   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 some of them with
-Line 478 
 register unsigned int c;   /* Character
+Line 488 
 register unsigned int c;   /* Character
  register BOOL utf8;        /* Local copy of UTF-8 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
-Line 486 
 heap storage. Set up the top-level frame
+Line 497 
 heap storage. Set up the top-level frame
  heap whenever RMATCH() does a "recursion". See the macro definitions above. */
  #ifdef NO_RECURSE
- heapframe *frame = (pcre_stack_malloc)(sizeof(heapframe));
+ heapframe *frame = (heapframe *)(pcre_stack_malloc)(sizeof(heapframe));
+ if (frame == NULL) RRETURN(PCRE_ERROR_NOMEMORY);
  frame->Xprevframe = NULL;            /* Marks the top level */
  /* Copy in the original argument variables */
-Line 496 
 frame->Xecode = ecode;
+Line 508 
 frame->Xecode = ecode;
  frame->Xmstart = mstart;
  frame->Xmarkptr = markptr;
  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 512 
 HEAP_RECURSE:
+Line 522 
 HEAP_RECURSE:
  #define mstart             frame->Xmstart
  #define markptr            frame->Xmarkptr
  #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 */
-Line 536 
 HEAP_RECURSE:
+Line 544 
 HEAP_RECURSE:
  #define condition          frame->Xcondition
  #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 prop_chartype      frame->Xprop_chartype
- #define prop_script        frame->Xprop_script
  #define oclength           frame->Xoclength
  #define occhars            frame->Xocchars
  #endif
-Line 574 
 i, and fc and c, can be the same variabl
+Line 577 
 i, and fc and c, can be the same variabl
  #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
+ const uschar *charptr;
+ #endif
+ const uschar *callpat;
+ const uschar *data;
+ const uschar *next;
+ USPTR         pp;
+ const uschar *prev;
+ USPTR         saved_eptr;
+ recursion_info new_recursive;
- #ifdef SUPPORT_UTF8                /* Many of these variables are used only  */
+ BOOL cur_is_word;
- const uschar *charptr;             /* in small blocks of the code. My normal */
- #endif                             /* style of coding would have declared    */
- const uschar *callpat;             /* them within each of those blocks.      */
- const uschar *data;                /* However, in order to accommodate the   */
- const uschar *next;                /* version of this code that uses an      */
- USPTR         pp;                  /* external "stack" implemented on the    */
- const uschar *prev;                /* heap, it is easier to declare them all */
- USPTR         saved_eptr;          /* here, so the declarations can be cut   */
-                                    /* out in a block. The only declarations  */
- recursion_info new_recursive;      /* within blocks below are for variables  */
-                                    /* that do not have to be preserved over  */
- BOOL cur_is_word;                  /* a recursive call to RMATCH().          */
  BOOL condition;
  BOOL prev_is_word;
- unsigned long int original_ims;
  #ifdef SUPPORT_UCP
  int prop_type;
  int prop_value;
  int prop_fail_result;
- int prop_category;
- int prop_chartype;
- int prop_script;
  int oclength;
  uschar occhars[8];
  #endif
-Line 619 
 int stacksave[REC_STACK_SAVE_MAX];
+Line 624 
 int stacksave[REC_STACK_SAVE_MAX];
  eptrblock newptrb;
  #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
  /* These statements are here to stop the compiler complaining about unitialized
  variables. */
-Line 655 
 haven't exceeded the recursive call limi
+Line 671 
 haven't exceeded the recursive call limi
  if (md->match_call_count++ >= md->match_limit) RRETURN(PCRE_ERROR_MATCHLIMIT);
  if (rdepth >= md->match_limit_recursion) RRETURN(PCRE_ERROR_RECURSIONLIMIT);
- original_ims = ims;    /* Save for resetting on ')' */
  /* At the start of a group with an unlimited repeat that may match an empty
- string, the match_cbegroup flag is set. When this is the case, add the current
+ string, the variable md->match_function_type is set to MATCH_CBEGROUP. It is
- subject pointer to the chain of such remembered pointers, to be checked when we
+ done this way to save having to use another function argument, which would take
- hit the closing ket, in order to break infinite loops that match no characters.
+ up space on the stack. See also MATCH_CONDASSERT below.
- When match() is called in other circumstances, don't add to the chain. The
- match_cbegroup flag must NOT be used with tail recursion, because the memory
+ When MATCH_CBEGROUP is set, add the current subject pointer to the chain of
- block that is used is on the stack, so a new one may be required for each
+ such remembered pointers, to be checked when we hit the closing ket, in order
- match(). */
+ 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_cbegroup) != 0)
+ if (md->match_function_type == MATCH_CBEGROUP)
    {
    newptrb.epb_saved_eptr = eptr;
    newptrb.epb_prev = eptrb;
    eptrb = &newptrb;
+   md->match_function_type = 0;
    }
  /* Now start processing the opcodes. */
-Line 685 
 for (;;)
+Line 703 
 for (;;)
      case OP_MARK:
      markptr = ecode + 2;
      RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode] + ecode[1], offset_top, md,
-       ims, eptrb, flags, RM51);
+       eptrb, RM55);
      /* 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. */
      if (rrc == MATCH_SKIP_ARG &&
          strcmp((char *)markptr, (char *)(md->start_match_ptr)) == 0)
        {
        md->start_match_ptr = eptr;
        RRETURN(MATCH_SKIP);
        }
      if (md->mark == NULL) md->mark = markptr;
      RRETURN(rrc);
      case OP_FAIL:
      MRRETURN(MATCH_NOMATCH);
+     /* COMMIT overrides PRUNE, SKIP, and THEN */
      case OP_COMMIT:
      RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
-       ims, eptrb, flags, RM52);
+       eptrb, RM52);
-     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     if (rrc != MATCH_NOMATCH && rrc != MATCH_PRUNE &&
+         rrc != MATCH_SKIP && rrc != MATCH_SKIP_ARG &&
+         rrc != MATCH_THEN)
+       RRETURN(rrc);
      MRRETURN(MATCH_COMMIT);
+     /* PRUNE overrides THEN */
      case OP_PRUNE:
      RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
-       ims, eptrb, flags, RM51);
+       eptrb, RM51);
-     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
      MRRETURN(MATCH_PRUNE);
      case OP_PRUNE_ARG:
      RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode] + ecode[1], offset_top, md,
-       ims, eptrb, flags, RM51);
+       eptrb, RM56);
-     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
      md->mark = ecode + 2;
      RRETURN(MATCH_PRUNE);
+     /* SKIP overrides PRUNE and THEN */
      case OP_SKIP:
      RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
-       ims, eptrb, flags, RM53);
+       eptrb, RM53);
-     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     if (rrc != MATCH_NOMATCH && rrc != MATCH_PRUNE && rrc != MATCH_THEN)
+       RRETURN(rrc);
      md->start_match_ptr = eptr;   /* Pass back current position */
      MRRETURN(MATCH_SKIP);
      case OP_SKIP_ARG:
      RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode] + ecode[1], offset_top, md,
-       ims, eptrb, flags, RM53);
+       eptrb, RM57);
-     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     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
+     /* Pass back the current skip name by overloading md->start_match_ptr and
-     caught by a matching MARK, or get to the top, where it is treated the same
+     returning the special MATCH_SKIP_ARG return code. This will either be
+     caught by a matching MARK, or get to the top, where it is treated the same
      as PRUNE. */
      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 + _pcre_OP_lengths[*ecode], offset_top, md,
-       ims, eptrb, flags, RM54);
+       eptrb, RM54);
      if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     md->start_match_ptr = ecode;
      MRRETURN(MATCH_THEN);
      case OP_THEN_ARG:
-     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode] + ecode[1], offset_top, md,
+     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode] + ecode[1], offset_top,
-       ims, eptrb, flags, RM54);
+       md, eptrb, RM58);
      if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     md->start_match_ptr = ecode;
      md->mark = ecode + 2;
      RRETURN(MATCH_THEN);
-     /* Handle a capturing bracket. If there is space in the offset vector, save
+     /* Handle a capturing bracket, other than those that are possessive with an
-     the current subject position in the working slot at the top of the vector.
+     unlimited repeat. If there is space in the offset vector, save the current
-     We mustn't change the current values of the data slot, because they may be
+     subject position in the working slot at the top of the vector. We mustn't
-     set from a previous iteration of this group, and be referred to by a
+     change the current values of the data slot, because they may be set from a
-     reference inside the group.
+     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 the bracket fails to match, we need to restore this value and also the
+     if something later on doesn't match. For this reason, we need to restore
-     values of the final offsets, in case they were set by a previous iteration
+     the working value and also the values of the final offsets, in case they
-     of the same bracket.
+     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
-Line 793 
 for (;;)
+Line 828 
 for (;;)
        save_capture_last = md->capture_last;
        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);
-       flags = (op == OP_SCBRA)? match_cbegroup : 0;
+       for (;;)
-       do
          {
+         if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
          RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
-           ims, eptrb, flags, RM1);
+           eptrb, RM1);
-         if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
+         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);
+         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;
-       if (rrc != MATCH_THEN) md->mark = markptr;
+       /* At this point, rrc will be one of MATCH_ONCE or MATCH_NOMATCH. */
-       RRETURN(MATCH_NOMATCH);
+       if (md->mark == NULL) md->mark = markptr;
+       RRETURN(rrc);
        }
      /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
-Line 827 
 for (;;)
+Line 887 
 for (;;)
      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-     /* Non-capturing bracket. Loop for all the alternatives. When we get to the
+     /* Non-capturing or atomic group, except for possessive with unlimited
-     final alternative within the brackets, we would return the result of a
+     repeat. Loop for all the alternatives.
-     recursive call to match() whatever happened. We can reduce stack usage by
-     turning this into a tail recursion, except in the case when match_cbegroup
+     When we get to the final alternative within the brackets, we used to return
-     is set.*/
+     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"));
-     flags = (op >= OP_SBRA)? match_cbegroup : 0;
      for (;;)
        {
-       if (ecode[GET(ecode, 1)] != OP_ALT)   /* Final alternative */
+       if (op >= OP_SBRA || op == OP_ONCE) md->match_function_type = MATCH_CBEGROUP;
+       /* 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 += _pcre_OP_lengths[*ecode];
+         goto TAIL_RECURSE;
+         }
+       /* In all other cases, we have to make another call to match(). */
+       RMATCH(eptr, ecode + _pcre_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 (flags == 0)    /* Not a possibly empty group */
+         if (rrc == MATCH_ONCE)
            {
-           ecode += _pcre_OP_lengths[*ecode];
+           const uschar *scode = ecode;
-           DPRINTF(("bracket 0 tail recursion\n"));
+           if (*scode != OP_ONCE)           /* If not at start, find it */
-           goto TAIL_RECURSE;
+             {
+             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);
+       if (*ecode != OP_ALT) break;
+       }
+     if (md->mark == NULL) md->mark = markptr;
+     RRETURN(MATCH_NOMATCH);
-         /* Possibly empty group; can't use tail recursion. */
+     /* Handle possessive capturing brackets with an unlimited repeat. We come
+     here from BRAZERO with allow_zero set TRUE. The offset_vector values are
+     handled similarly to the normal case above. However, the matching is
+     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;
-         RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md, ims,
+     POSSESSIVE_CAPTURE:
-           eptrb, flags, RM48);
+     number = GET2(ecode, 1+LINK_SIZE);
-         if (rrc == MATCH_NOMATCH) md->mark = markptr;
+     offset = number << 1;
-         RRETURN(rrc);
+ #ifdef PCRE_DEBUG
+     printf("start possessive bracket %d\n", number);
+     printf("subject=");
+     pchars(eptr, 16, TRUE, md);
+     printf("\n");
+ #endif
+     if (offset < md->offset_max)
+       {
+       matched_once = FALSE;
+       code_offset = ecode - md->start_code;
+       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 + _pcre_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;
          }
-       /* For non-final alternatives, continue the loop for a NOMATCH result;
+       if (!matched_once)
-       otherwise return. */
+         {
+         md->offset_vector[offset] = save_offset1;
+         md->offset_vector[offset+1] = save_offset2;
+         md->offset_vector[md->offset_end - number] = save_offset3;
+         }
-       RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md, ims,
+       if (md->mark == NULL) md->mark = markptr;
-         eptrb, flags, RM2);
+       if (allow_zero || matched_once)
-       if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
+         {
+         ecode += 1 + LINK_SIZE;
+         break;
+         }
+       RRETURN(MATCH_NOMATCH);
+       }
+     /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
+     as a non-capturing bracket. */
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
+     DPRINTF(("insufficient capture room: treat as non-capturing\n"));
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
+     /* Non-capturing possessive bracket with unlimited repeat. We come here
+     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. */
+     case OP_BRAPOS:
+     case OP_SBRAPOS:
+     allow_zero = FALSE;
+     POSSESSIVE_NON_CAPTURE:
+     matched_once = FALSE;
+     code_offset = ecode - md->start_code;
+     for (;;)
+       {
+       if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
+       RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
+         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;
+       }
+     if (matched_once || allow_zero)
+       {
+       ecode += 1 + LINK_SIZE;
+       break;
        }
+     RRETURN(MATCH_NOMATCH);
      /* Control never reaches here. */
      /* Conditional group: compilation checked that there are no more than
      two branches. If the condition is false, skipping the first branch takes us
      past the end if there is only one branch, but that's OK because that is
-     exactly what going to the ket would do. As there is only one branch to be
+     exactly what going to the ket would do. */
-     obeyed, we can use tail recursion to avoid using another stack frame. */
      case OP_COND:
      case OP_SCOND:
-     codelink= GET(ecode, 1);
+     codelink = GET(ecode, 1);
      /* Because of the way auto-callout works during compile, a callout item is
      inserted between OP_COND and an assertion condition. */
-Line 884 
 for (;;)
+Line 1138 
 for (;;)
        if (pcre_callout != NULL)
          {
          pcre_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[LINK_SIZE+2];
          cb.offset_vector    = md->offset_vector;
          cb.subject          = (PCRE_SPTR)md->start_subject;
-         cb.subject_length   = md->end_subject - md->start_subject;
+         cb.subject_length   = (int)(md->end_subject - md->start_subject);
-         cb.start_match      = mstart - md->start_subject;
+         cb.start_match      = (int)(mstart - md->start_subject);
-         cb.current_position = eptr - 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             = markptr;
          if ((rrc = (*pcre_callout)(&cb)) > 0) MRRETURN(MATCH_NOMATCH);
          if (rrc < 0) RRETURN(rrc);
          }
-Line 1049 
 for (;;)
+Line 1304 
 for (;;)
        }
      /* The condition is an assertion. Call match() to evaluate it - setting
-     the final argument match_condassert causes it to stop at the end of an
+     md->match_function_type to MATCH_CONDASSERT causes it to stop at the end of
-     assertion. */
+     an assertion. */
      else
        {
-       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,
+       md->match_function_type = MATCH_CONDASSERT;
-           match_condassert, RM3);
+       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);
          }
-       else if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN)
+       /* 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 */
          }
-Line 1073 
 for (;;)
+Line 1334 
 for (;;)
          }
        }
-     /* We are now at the branch that is to be obeyed. As there is only one,
+     /* We are now at the branch that is to be obeyed. As there is only one, can
-     we can use tail recursion to avoid using another stack frame, except when
+     use tail recursion to avoid using another stack frame, except when there is
-     match_cbegroup is required for an unlimited repeat of a possibly empty
+     unlimited repeat of a possibly empty group. In the latter case, a recursive
-     group. If the second alternative doesn't exist, we can just plough on. */
+     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)
        {
-       ecode += 1 + LINK_SIZE;
+       if (op != OP_SCOND)
-       if (op == OP_SCOND)        /* Possibly empty group */
          {
-         RMATCH(eptr, ecode, offset_top, md, ims, eptrb, match_cbegroup, RM49);
+         ecode += 1 + LINK_SIZE;
-         RRETURN(rrc);
-         }
-       else                       /* Group must match something */
-         {
-         flags = 0;
          goto TAIL_RECURSE;
          }
+       md->match_function_type = MATCH_CBEGROUP;
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM49);
+       RRETURN(rrc);
        }
-     else                         /* Condition false & no alternative */
+      /* Condition false & no alternative; continue after the group. */
+     else
        {
        ecode += 1 + LINK_SIZE;
        }
-Line 1116 
 for (;;)
+Line 1383 
 for (;;)
        {
        md->offset_vector[offset] =
          md->offset_vector[md->offset_end - number];
-       md->offset_vector[offset+1] = eptr - md->start_subject;
+       md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
        if (offset_top <= offset) offset_top = offset + 2;
        }
      ecode += 3;
      break;
-     /* End of the pattern, either real or forced. If we are in a top-level
+     /* End of the pattern, either real or forced. */
-     recursion, we should restore the offsets appropriately and continue from
-     after the call. */
-     case OP_ACCEPT:
      case OP_END:
-     if (md->recursive != NULL && md->recursive->group_num == 0)
+     case OP_ACCEPT:
-       {
+     case OP_ASSERT_ACCEPT:
-       recursion_info *rec = md->recursive;
-       DPRINTF(("End of pattern in a (?0) recursion\n"));
-       md->recursive = rec->prevrec;
-       memmove(md->offset_vector, rec->offset_save,
-         rec->saved_max * sizeof(int));
-       offset_top = rec->save_offset_top;
-       ims = original_ims;
-       ecode = rec->after_call;
-       break;
-       }
-     /* Otherwise, if we have matched an empty string, fail if PCRE_NOTEMPTY is
+     /* If we have matched an empty string, fail if not in an assertion and not
-     set, or if PCRE_NOTEMPTY_ATSTART is set and we have matched at the start of
+     in a recursion if either PCRE_NOTEMPTY is set, or if PCRE_NOTEMPTY_ATSTART
-     the subject. In both cases, backtracking will then try other alternatives,
+     is set and we have matched at the start of the subject. In both cases,
-     if any. */
+     backtracking will then try other alternatives, if any. */
-     if (eptr == mstart &&
+     if (eptr == mstart && op != OP_ASSERT_ACCEPT &&
-         (md->notempty ||
+          md->recursive == NULL &&
-           (md->notempty_atstart &&
+          (md->notempty ||
-             mstart == md->start_subject + md->start_offset)))
+            (md->notempty_atstart &&
+              mstart == md->start_subject + md->start_offset)))
        MRRETURN(MATCH_NOMATCH);
      /* Otherwise, we have a match. */
-Line 1158 
 for (;;)
+Line 1413 
 for (;;)
      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) */
-     MRRETURN(((op == OP_END)? MATCH_MATCH : MATCH_ACCEPT));
-     /* Change option settings */
+     /* For some reason, the macros don't work properly if an expression is
+     given as the argument to MRRETURN when the heap is in use. */
-     case OP_OPT:
+     rrc = (op == OP_END)? MATCH_MATCH : MATCH_ACCEPT;
-     ims = ecode[1];
+     MRRETURN(rrc);
-     ecode += 2;
-     DPRINTF(("ims set to %02lx\n", ims));
-     break;
      /* 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. */
+     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:
+     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, ims, NULL, 0,
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM4);
-         RM4);
        if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT)
          {
          mstart = md->start_match_ptr;   /* In case \K reset it */
+         markptr = md->mark;
          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 (*ecode == OP_KET) MRRETURN(MATCH_NOMATCH);
      /* If checking an assertion for a condition, return MATCH_MATCH. */
-     if ((flags & match_condassert) != 0) RRETURN(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. */
-Line 1209 
 for (;;)
+Line 1478 
 for (;;)
      case OP_ASSERT_NOT:
      case OP_ASSERTBACK_NOT:
+     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, ims, NULL, 0,
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM5);
-         RM5);
        if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) MRRETURN(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 ((flags & match_condassert) != 0) RRETURN(MATCH_MATCH);
+     if (condassert) RRETURN(MATCH_MATCH);  /* Condition assertion */
      ecode += 1 + LINK_SIZE;
      continue;
-Line 1270 
 for (;;)
+Line 1549 
 for (;;)
      if (pcre_callout != NULL)
        {
        pcre_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          = (PCRE_SPTR)md->start_subject;
-       cb.subject_length   = md->end_subject - md->start_subject;
+       cb.subject_length   = (int)(md->end_subject - md->start_subject);
-       cb.start_match      = mstart - md->start_subject;
+       cb.start_match      = (int)(mstart - md->start_subject);
-       cb.current_position = eptr - 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;
+       cb.mark             = markptr;
        if ((rrc = (*pcre_callout)(&cb)) > 0) MRRETURN(MATCH_NOMATCH);
        if (rrc < 0) RRETURN(rrc);
        }
-Line 1292 
 for (;;)
+Line 1572 
 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:
        {
+       recursion_info *ri;
+       int recno;
        callpat = md->start_code + GET(ecode, 1);
-       new_recursive.group_num = (callpat == md->start_code)? 0 :
+       recno = (callpat == md->start_code)? 0 :
          GET2(callpat, 1 + LINK_SIZE);
+       /* Check for repeating a recursion without advancing the subject pointer.
+       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 1334 
 for (;;)
+Line 1624 
 for (;;)
            (int *)(pcre_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_offset_top = offset_top;
-       /* 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));
-       flags = (*callpat >= OP_SBRA)? match_cbegroup : 0;
+       cbegroup = (*callpat >= OP_SBRA);
        do
          {
+         if (cbegroup) md->match_function_type = MATCH_CBEGROUP;
          RMATCH(eptr, callpat + _pcre_OP_lengths[*callpat], offset_top,
-           md, ims, eptrb, flags, RM6);
+           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)
            {
            DPRINTF(("Recursion matched\n"));
-           md->recursive = new_recursive.prevrec;
            if (new_recursive.offset_save != stacksave)
              (pcre_free)(new_recursive.offset_save);
-           MRRETURN(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 */
            }
-         else if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN)
+         /* 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)
-Line 1365 
 for (;;)
+Line 1668 
 for (;;)
            }
          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 1377 
 for (;;)
+Line 1678 
 for (;;)
          (pcre_free)(new_recursive.offset_save);
        MRRETURN(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, but resetting
-     the start-of-match value in case it was changed by \K. */
-     case OP_ONCE:
-     prev = ecode;
-     saved_eptr = eptr;
-     do
-       {
-       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb, 0, RM7);
-       if (rrc == MATCH_MATCH)  /* Note: _not_ MATCH_ACCEPT */
-         {
-         mstart = md->start_match_ptr;
-         break;
-         }
-       if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) 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
+     RECURSION_MATCHED:
-     preceding bracket, in the appropriate order. The second "call" of match()
+     break;
-     uses tail recursion, to avoid using another stack frame. We need to reset
-     any options that changed within the bracket before re-running it, so
-     check the next opcode. */
-     if (ecode[1+LINK_SIZE] == OP_OPT)
-       {
-       ims = (ims & ~PCRE_IMS) | ecode[4];
-       DPRINTF(("ims set to %02lx at group repeat\n", ims));
-       }
-     if (*ecode == OP_KETRMIN)
-       {
-       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb, 0, RM8);
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-       ecode = prev;
-       flags = 0;
-       goto TAIL_RECURSE;
-       }
-     else  /* OP_KETRMAX */
-       {
-       RMATCH(eptr, prev, offset_top, md, ims, eptrb, match_cbegroup, RM9);
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-       ecode += 1 + LINK_SIZE;
-       flags = 0;
-       goto TAIL_RECURSE;
-       }
-     /* Control never gets here */
      /* An alternation is the end of a branch; scan along to find the end of the
      bracketed group and go to there. */
-Line 1472 
 for (;;)
+Line 1696 
 for (;;)
      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(eptr, next, offset_top, md, ims, eptrb, 0, RM10);
+     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(eptr, next + 1+LINK_SIZE, offset_top, md, ims, eptrb, 0, RM11);
+     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     ecode++;
-       ecode++;
-       }
      break;
      case OP_SKIPZERO:
-       {
+     next = ecode+1;
-       next = ecode+1;
+     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;
+     /* BRAPOSZERO occurs before a possessive bracket group. Don't do anything
+     here; just jump to the group, with allow_zero set TRUE. */
+     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);
      /* 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. */
-     if (*prev >= OP_SBRA)
+     if (*prev >= OP_SBRA || *prev == OP_ONCE)
        {
        saved_eptr = eptrb->epb_saved_eptr;   /* Value at start of group */
        eptrb = eptrb->epb_prev;              /* Backup to previous group */
        }
      else saved_eptr = NULL;
-     /* If we are at the end of an assertion group or an atomic group, stop
+     /* If we are at the end of an assertion group, stop matching and return
-     matching and return MATCH_MATCH, but record the current high water mark for
+     MATCH_MATCH, but record the current high water mark for use by positive
-     use by positive assertions. We also need to record the match start in case
+     assertions. We also need to record the match start in case it was changed
-     it was changed by \K. */
+     by \K. */
      if (*prev == OP_ASSERT || *prev == OP_ASSERT_NOT ||
-         *prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT ||
+         *prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT)
-         *prev == OP_ONCE)
        {
        md->end_match_ptr = eptr;      /* For ONCE */
        md->end_offset_top = offset_top;
        md->start_match_ptr = mstart;
-       MRRETURN(MATCH_MATCH);
+       MRRETURN(MATCH_MATCH);         /* Sets md->mark */
        }
      /* For capturing groups we have to check the group number back at the start
      and if necessary complete handling an extraction by setting the offsets and
-     bumping the high water mark. Note that whole-pattern recursion is coded as
+     bumping the high water mark. Whole-pattern recursion is coded as a recurse
-     a recurse into group 0, so it won't be picked up here. Instead, we catch it
+     into group 0, so it won't be picked up here. Instead, we catch it when the
-     when the OP_END is reached. Other recursion is handled here. */
+     OP_END is reached. Other recursion is handled here. We just have to record
+     the current subject position and start match pointer and give a MATCH
+     return. */
-     if (*prev == OP_CBRA || *prev == OP_SCBRA)
+     if (*prev == OP_CBRA || *prev == OP_SCBRA ||
+         *prev == OP_CBRAPOS || *prev == OP_SCBRAPOS)
        {
        number = GET2(prev, 1+LINK_SIZE);
        offset = number << 1;
-Line 1548 
 for (;;)
+Line 1778 
 for (;;)
        printf("\n");
  #endif
+       /* Handle a recursively called group. */
+       if (md->recursive != NULL && md->recursive->group_num == number)
+         {
+         md->end_match_ptr = eptr;
+         md->start_match_ptr = mstart;
+         RRETURN(MATCH_MATCH);
+         }
+       /* Deal with capturing */
        md->capture_last = number;
        if (offset >= md->offset_max) md->offset_overflow = TRUE; else
          {
+         /* If offset is greater than offset_top, it means that we are
+         "skipping" a capturing group, and that group's offsets must be marked
+         unset. In earlier versions of PCRE, all the offsets were unset at the
+         start of matching, but this doesn't work because atomic groups and
+         assertions can cause a value to be set that should later be unset.
+         Example: matching /(?>(a))b|(a)c/ against "ac". This sets group 1 as
+         part of the atomic group, but this is not on the final matching path,
+         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;
+           }
+         /* Now make the extraction */
          md->offset_vector[offset] =
            md->offset_vector[md->offset_end - number];
-         md->offset_vector[offset+1] = eptr - md->start_subject;
+         md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
          if (offset_top <= offset) offset_top = offset + 2;
          }
+       }
-       /* Handle a recursively called group. Restore the offsets
+     /* For an ordinary non-repeating ket, just continue at this level. This
-       appropriately and continue from after the call. */
+     also happens for a repeating ket if no characters were matched in the
+     group. This is the forcible breaking of infinite loops as implemented in
+     Perl 5.005. For a non-repeating atomic group, 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 (md->recursive != NULL && md->recursive->group_num == number)
+     if (*ecode == OP_KET || eptr == saved_eptr)
+       {
+       if (*prev == OP_ONCE)
          {
-         recursion_info *rec = md->recursive;
+         RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM12);
-         DPRINTF(("Recursion (%d) succeeded - continuing\n", number));
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-         md->recursive = rec->prevrec;
+         md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
-         memcpy(md->offset_vector, rec->offset_save,
+         RRETURN(MATCH_ONCE);
-           rec->saved_max * sizeof(int));
-         offset_top = rec->save_offset_top;
-         ecode = rec->after_call;
-         ims = original_ims;
-         break;
          }
+       ecode += 1 + LINK_SIZE;    /* Carry on at this level */
+       break;
        }
-     /* For both capturing and non-capturing groups, reset the value of the ims
+     /* OP_KETRPOS is a possessive repeating ket. Remember the current position,
-     flags, in case they got changed during the group. */
+     and return the MATCH_KETRPOS. This makes it possible to do the repeats one
+     at a time from the outer level, thus saving stack. */
-     ims = original_ims;
-     DPRINTF(("ims reset to %02lx\n", ims));
-     /* 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)
+     if (*ecode == OP_KETRPOS)
        {
-       ecode += 1 + LINK_SIZE;
+       md->end_match_ptr = eptr;
-       break;
+       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. In the second case, we can use
+     the preceding bracket, in the appropriate order. In the second case, we can
-     tail recursion to avoid using another stack frame, unless we have an
+     use tail recursion to avoid using another stack frame, unless we have an
-     unlimited repeat of a group that can match an empty string. */
+     an atomic group or an unlimited repeat of a group that can match an empty
+     string. */
-     flags = (*prev >= OP_SBRA)? match_cbegroup : 0;
      if (*ecode == OP_KETRMIN)
        {
-       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb, 0, RM12);
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM7);
        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-       if (flags != 0)    /* Could match an empty string */
+       if (*prev == OP_ONCE)
+         {
+         RMATCH(eptr, prev, offset_top, md, eptrb, RM8);
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
+         RRETURN(MATCH_ONCE);
+         }
+       if (*prev >= OP_SBRA)    /* Could match an empty string */
          {
-         RMATCH(eptr, prev, offset_top, md, ims, eptrb, flags, RM50);
+         md->match_function_type = MATCH_CBEGROUP;
+         RMATCH(eptr, prev, offset_top, md, eptrb, RM50);
          RRETURN(rrc);
          }
        ecode = prev;
-Line 1613 
 for (;;)
+Line 1879 
 for (;;)
        }
      else  /* OP_KETRMAX */
        {
-       RMATCH(eptr, prev, offset_top, md, ims, eptrb, flags, RM13);
+       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;
-       flags = 0;
        goto TAIL_RECURSE;
        }
      /* Control never gets here */
-     /* Start of subject unless notbol, or after internal newline if multiline */
+     /* Not multiline mode: start of subject assertion, unless notbol. */
      case OP_CIRC:
      if (md->notbol && eptr == md->start_subject) MRRETURN(MATCH_NOMATCH);
-     if ((ims & PCRE_MULTILINE) != 0)
-       {
-       if (eptr != md->start_subject &&
-           (eptr == md->end_subject || !WAS_NEWLINE(eptr)))
-         MRRETURN(MATCH_NOMATCH);
-       ecode++;
-       break;
-       }
-     /* ... else fall through */
      /* Start of subject assertion */
-Line 1642 
 for (;;)
+Line 1907 
 for (;;)
      ecode++;
      break;
+     /* Multiline mode: start of subject unless notbol, or after any newline. */
+     case OP_CIRCM:
+     if (md->notbol && eptr == md->start_subject) MRRETURN(MATCH_NOMATCH);
+     if (eptr != md->start_subject &&
+         (eptr == md->end_subject || !WAS_NEWLINE(eptr)))
+       MRRETURN(MATCH_NOMATCH);
+     ecode++;
+     break;
      /* Start of match assertion */
      case OP_SOM:
-Line 1656 
 for (;;)
+Line 1931 
 for (;;)
      ecode++;
      break;
-     /* Assert before internal newline if multiline, or before a terminating
+     /* Multiline mode: assert before any newline, or before end of subject
-     newline unless endonly is set, else end of subject unless noteol is set. */
+     unless noteol is set. */
-     case OP_DOLL:
+     case OP_DOLLM:
-     if ((ims & PCRE_MULTILINE) != 0)
+     if (eptr < md->end_subject)
-       {
+       { if (!IS_NEWLINE(eptr)) MRRETURN(MATCH_NOMATCH); }
-       if (eptr < md->end_subject)
-         { if (!IS_NEWLINE(eptr)) MRRETURN(MATCH_NOMATCH); }
-       else
-         { if (md->noteol) MRRETURN(MATCH_NOMATCH); }
-       ecode++;
-       break;
-       }
      else
        {
        if (md->noteol) MRRETURN(MATCH_NOMATCH);
-       if (!md->endonly)
+       SCHECK_PARTIAL();
-         {
-         if (eptr != md->end_subject &&
-             (!IS_NEWLINE(eptr) || eptr != md->end_subject - md->nllen))
-           MRRETURN(MATCH_NOMATCH);
-         ecode++;
-         break;
-         }
        }
+     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) MRRETURN(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) MRRETURN(MATCH_NOMATCH);
+     SCHECK_PARTIAL();
      ecode++;
      break;
      /* End of subject or ending \n assertion (\Z) */
      case OP_EODN:
-     if (eptr != md->end_subject &&
+     ASSERT_NL_OR_EOS:
+     if (eptr < md->end_subject &&
          (!IS_NEWLINE(eptr) || eptr != md->end_subject - md->nllen))
        MRRETURN(MATCH_NOMATCH);
+     /* Either at end of string or \n before end. */
+     SCHECK_PARTIAL();
      ecode++;
      break;
-Line 1713 
 for (;;)
+Line 1990 
 for (;;)
  #ifdef SUPPORT_UTF8
        if (utf8)
          {
+         /* Get status of previous character */
          if (eptr == md->start_subject) prev_is_word = FALSE; else
            {
            USPTR lastptr = eptr - 1;
            while((*lastptr & 0xc0) == 0x80) 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;
            }
+         /* Get status of next character */
          if (eptr >= md->end_subject)
            {
            SCHECK_PARTIAL();
-Line 1729 
 for (;;)
+Line 2022 
 for (;;)
          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
-       /* 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. */
          {
+         /* Get status of previous character */
          if (eptr == md->start_subject) prev_is_word = FALSE; else
            {
            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 = ((md->ctypes[eptr[-1]] & ctype_word) != 0);
            }
+         /* Get status of next character */
          if (eptr >= md->end_subject)
            {
            SCHECK_PARTIAL();
            cur_is_word = FALSE;
            }
-         else cur_is_word = ((md->ctypes[*eptr] & ctype_word) != 0);
+         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 = ((md->ctypes[*eptr] & ctype_word) != 0);
          }
        /* Now see if the situation is what we want */
-Line 1766 
 for (;;)
+Line 2101 
 for (;;)
      /* Fall through */
      case OP_ALLANY:
-     if (eptr++ >= md->end_subject)
+     if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
-       {
+       {                            /* not be updated before SCHECK_PARTIAL. */
        SCHECK_PARTIAL();
        MRRETURN(MATCH_NOMATCH);
        }
+     eptr++;
      if (utf8) while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
      ecode++;
      break;
-Line 1779 
 for (;;)
+Line 2115 
 for (;;)
      any byte, even newline, independent of the setting of PCRE_DOTALL. */
      case OP_ANYBYTE:
-     if (eptr++ >= md->end_subject)
+     if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
-       {
+       {                            /* not be updated before SCHECK_PARTIAL. */
        SCHECK_PARTIAL();
        MRRETURN(MATCH_NOMATCH);
        }
+     eptr++;
      ecode++;
      break;
-Line 1899 
 for (;;)
+Line 2236 
 for (;;)
      switch(c)
        {
        default: MRRETURN(MATCH_NOMATCH);
        case 0x000d:
        if (eptr < md->end_subject && *eptr == 0x0a) eptr++;
        break;
-Line 2055 
 for (;;)
+Line 2393 
 for (;;)
               prop->chartype == ucp_Ll ||
               prop->chartype == ucp_Lt) == (op == OP_NOTPROP))
            MRRETURN(MATCH_NOMATCH);
-          break;
+         break;
          case PT_GC:
          if ((ecode[2] != _pcre_ucp_gentype[prop->chartype]) == (op == OP_PROP))
-Line 2072 
 for (;;)
+Line 2410 
 for (;;)
            MRRETURN(MATCH_NOMATCH);
          break;
+         /* These are specials */
+         case PT_ALNUM:
+         if ((_pcre_ucp_gentype[prop->chartype] == ucp_L ||
+              _pcre_ucp_gentype[prop->chartype] == ucp_N) == (op == OP_NOTPROP))
+           MRRETURN(MATCH_NOMATCH);
+         break;
+         case PT_SPACE:    /* Perl space */
+         if ((_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
+              c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
+                == (op == OP_NOTPROP))
+           MRRETURN(MATCH_NOMATCH);
+         break;
+         case PT_PXSPACE:  /* POSIX space */
+         if ((_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
+              c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
+              c == CHAR_FF || c == CHAR_CR)
+                == (op == OP_NOTPROP))
+           MRRETURN(MATCH_NOMATCH);
+         break;
+         case PT_WORD:
+         if ((_pcre_ucp_gentype[prop->chartype] == ucp_L ||
+              _pcre_ucp_gentype[prop->chartype] == ucp_N ||
+              c == CHAR_UNDERSCORE) == (op == OP_NOTPROP))
+           MRRETURN(MATCH_NOMATCH);
+         break;
+         /* This should never occur */
          default:
          RRETURN(PCRE_ERROR_INTERNAL);
          }
-Line 2090 
 for (;;)
+Line 2460 
 for (;;)
        MRRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
+     if (UCD_CATEGORY(c) == ucp_M) MRRETURN(MATCH_NOMATCH);
+     while (eptr < md->end_subject)
        {
-       int category = UCD_CATEGORY(c);
+       int len = 1;
-       if (category == ucp_M) MRRETURN(MATCH_NOMATCH);
+       if (!utf8) c = *eptr; else { GETCHARLEN(c, eptr, len); }
-       while (eptr < md->end_subject)
+       if (UCD_CATEGORY(c) != ucp_M) break;
-         {
+       eptr += len;
-         int len = 1;
-         if (!utf8) c = *eptr; else
-           {
-           GETCHARLEN(c, eptr, len);
-           }
-         category = UCD_CATEGORY(c);
-         if (category != ucp_M) break;
-         eptr += len;
-         }
        }
      ecode++;
      break;
-Line 2119 
 for (;;)
+Line 2482 
 for (;;)
      loops). */
      case OP_REF:
-       {
+     case OP_REFI:
-       offset = GET2(ecode, 1) << 1;               /* Doubled ref number */
+     caseless = op == OP_REFI;
-       ecode += 3;
+     offset = GET2(ecode, 1) << 1;               /* Doubled ref number */
+     ecode += 3;
-       /* If the reference is unset, there are two possibilities:
+     /* If the reference is unset, there are two possibilities:
-       (a) In the default, Perl-compatible state, set the length to be longer
+     (a) In the default, Perl-compatible state, set the length negative;
-       than the amount of subject left; this ensures that every attempt at a
+     this ensures that every attempt at a match fails. We can't just fail
-       match fails. We can't just fail here, because of the possibility of
+     here, because of the possibility of quantifiers with zero minima.
-       quantifiers with zero minima.
-       (b) If the JavaScript compatibility flag is set, set the length to zero
+     (b) If the JavaScript compatibility flag is set, set the length to zero
-       so that the back reference matches an empty string.
+     so that the back reference matches an empty string.
-       Otherwise, set the length to the length of what was matched by the
+     Otherwise, set the length to the length of what was matched by the
-       referenced subpattern. */
+     referenced subpattern. */
-       if (offset >= offset_top || md->offset_vector[offset] < 0)
+     if (offset >= offset_top || md->offset_vector[offset] < 0)
-         length = (md->jscript_compat)? 0 : md->end_subject - eptr + 1;
+       length = (md->jscript_compat)? 0 : -1;
-       else
+     else
-         length = md->offset_vector[offset+1] - md->offset_vector[offset];
+       length = md->offset_vector[offset+1] - md->offset_vector[offset];
-       /* Set up for repetition, or handle the non-repeated case */
+     /* Set up for repetition, or handle the non-repeated case */
-       switch (*ecode)
+     switch (*ecode)
-         {
+       {
-         case OP_CRSTAR:
+       case OP_CRSTAR:
-         case OP_CRMINSTAR:
+       case OP_CRMINSTAR:
-         case OP_CRPLUS:
+       case OP_CRPLUS:
-         case OP_CRMINPLUS:
+       case OP_CRMINPLUS:
-         case OP_CRQUERY:
+       case OP_CRQUERY:
-         case OP_CRMINQUERY:
+       case OP_CRMINQUERY:
-         c = *ecode++ - OP_CRSTAR;
+       c = *ecode++ - OP_CRSTAR;
-         minimize = (c & 1) != 0;
+       minimize = (c & 1) != 0;
-         min = rep_min[c];                 /* Pick up values from tables; */
+       min = rep_min[c];                 /* Pick up values from tables; */
-         max = rep_max[c];                 /* zero for max => infinity */
+       max = rep_max[c];                 /* zero for max => infinity */
-         if (max == 0) max = INT_MAX;
+       if (max == 0) max = INT_MAX;
-         break;
+       break;
-         case OP_CRRANGE:
+       case OP_CRRANGE:
-         case OP_CRMINRANGE:
+       case OP_CRMINRANGE:
-         minimize = (*ecode == OP_CRMINRANGE);
+       minimize = (*ecode == OP_CRMINRANGE);
-         min = GET2(ecode, 1);
+       min = GET2(ecode, 1);
-         max = GET2(ecode, 3);
+       max = GET2(ecode, 3);
-         if (max == 0) max = INT_MAX;
+       if (max == 0) max = INT_MAX;
-         ecode += 5;
+       ecode += 5;
-         break;
+       break;
-         default:               /* No repeat follows */
+       default:               /* No repeat follows */
-         if (!match_ref(offset, eptr, length, md, ims))
+       if ((length = match_ref(offset, eptr, length, md, caseless)) < 0)
-           {
+         {
-           CHECK_PARTIAL();
+         CHECK_PARTIAL();
-           MRRETURN(MATCH_NOMATCH);
+         MRRETURN(MATCH_NOMATCH);
-           }
-         eptr += length;
-         continue;              /* With the main loop */
          }
+       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 (length == 0) continue;
+     if (length == 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))
+         CHECK_PARTIAL();
-           {
+         MRRETURN(MATCH_NOMATCH);
-           CHECK_PARTIAL();
-           MRRETURN(MATCH_NOMATCH);
-           }
-         eptr += length;
          }
+       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) MRRETURN(MATCH_NOMATCH);
+         if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM14);
+           CHECK_PARTIAL();
-           if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+           MRRETURN(MATCH_NOMATCH);
-           if (fi >= max) MRRETURN(MATCH_NOMATCH);
-           if (!match_ref(offset, eptr, length, md, ims))
-             {
-             CHECK_PARTIAL();
-             MRRETURN(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))
-             {
-             CHECK_PARTIAL();
-             break;
-             }
-           eptr += length;
-           }
-         while (eptr >= pp)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM15);
+           CHECK_PARTIAL();
-           if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+           break;
-           eptr -= length;
            }
-         MRRETURN(MATCH_NOMATCH);
+         eptr += slength;
+         }
+       while (eptr >= pp)
+         {
+         RMATCH(eptr, ecode, offset_top, md, eptrb, RM15);
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         eptr -= length;
          }
+       MRRETURN(MATCH_NOMATCH);
        }
      /* Control never gets here */
-Line 2347 
 for (;;)
+Line 2712 
 for (;;)
            {
            for (fi = min;; fi++)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM16);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM16);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
-Line 2372 
 for (;;)
+Line 2737 
 for (;;)
            {
            for (fi = min;; fi++)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM17);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM17);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
-Line 2418 
 for (;;)
+Line 2783 
 for (;;)
              }
            for (;;)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM18);
+             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 2441 
 for (;;)
+Line 2806 
 for (;;)
              }
            while (eptr >= pp)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM19);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM19);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              eptr--;
              }
-Line 2517 
 for (;;)
+Line 2882 
 for (;;)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM20);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM20);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (fi >= max) MRRETURN(MATCH_NOMATCH);
            if (eptr >= md->end_subject)
-Line 2550 
 for (;;)
+Line 2915 
 for (;;)
            }
          for(;;)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM21);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM21);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (eptr-- == pp) break;        /* Stop if tried at original pos */
            if (utf8) BACKCHAR(eptr);
-Line 2595 
 for (;;)
+Line 2960 
 for (;;)
      /* Match a single character, caselessly */
-     case OP_CHARNC:
+     case OP_CHARI:
  #ifdef SUPPORT_UTF8
      if (utf8)
        {
-Line 2655 
 for (;;)
+Line 3020 
 for (;;)
      /* Match a single character repeatedly. */
      case OP_EXACT:
+     case OP_EXACTI:
      min = max = GET2(ecode, 1);
      ecode += 3;
      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;
      goto REPEATCHAR;
      case OP_POSSTAR:
+     case OP_POSSTARI:
      possessive = TRUE;
      min = 0;
      max = INT_MAX;
-Line 2679 
 for (;;)
+Line 3049 
 for (;;)
      goto REPEATCHAR;
      case OP_POSPLUS:
+     case OP_POSPLUSI:
      possessive = TRUE;
      min = 1;
      max = INT_MAX;
-Line 2686 
 for (;;)
+Line 3057 
 for (;;)
      goto REPEATCHAR;
      case OP_POSQUERY:
+     case OP_POSQUERYI:
      possessive = TRUE;
      min = 0;
      max = 1;
-Line 2693 
 for (;;)
+Line 3065 
 for (;;)
      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;
-Line 2723 
 for (;;)
+Line 3100 
 for (;;)
          {
  #ifdef SUPPORT_UCP
          unsigned int othercase;
-         if ((ims & PCRE_CASELESS) != 0 &&
+         if (op >= OP_STARI &&     /* Caseless */
              (othercase = UCD_OTHERCASE(fc)) != fc)
            oclength = _pcre_ord2utf8(othercase, occhars);
          else oclength = 0;
-Line 2751 
 for (;;)
+Line 3128 
 for (;;)
            {
            for (fi = min;; fi++)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM22);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM22);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
              if (eptr <= md->end_subject - length &&
-Line 2793 
 for (;;)
+Line 3170 
 for (;;)
            for(;;)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM23);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM23);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (eptr == pp) { MRRETURN(MATCH_NOMATCH); }
  #ifdef SUPPORT_UCP
-Line 2830 
 for (;;)
+Line 3207 
 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];
        for (i = 1; i <= min; i++)
-Line 2847 
 for (;;)
+Line 3224 
 for (;;)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM24);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM24);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (fi >= max) MRRETURN(MATCH_NOMATCH);
            if (eptr >= md->end_subject)
-Line 2877 
 for (;;)
+Line 3254 
 for (;;)
          while (eptr >= pp)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM25);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM25);
            eptr--;
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            }
-Line 2906 
 for (;;)
+Line 3283 
 for (;;)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM26);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM26);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (fi >= max) MRRETURN(MATCH_NOMATCH);
            if (eptr >= md->end_subject)
-Line 2935 
 for (;;)
+Line 3312 
 for (;;)
          while (eptr >= pp)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM27);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM27);
            eptr--;
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            }
-Line 2948 
 for (;;)
+Line 3325 
 for (;;)
      checking can be multibyte. */
      case OP_NOT:
+     case OP_NOTI:
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-Line 2955 
 for (;;)
+Line 3333 
 for (;;)
        }
      ecode++;
      GETCHARINCTEST(c, eptr);
-     if ((ims & PCRE_CASELESS) != 0)
+     if (op == OP_NOTI)         /* The caseless case */
        {
  #ifdef SUPPORT_UTF8
        if (c < 256)
-Line 2963 
 for (;;)
+Line 3341 
 for (;;)
        c = md->lcc[c];
        if (md->lcc[*ecode++] == c) MRRETURN(MATCH_NOMATCH);
        }
-     else
+     else    /* Caseful */
        {
        if (*ecode++ == c) MRRETURN(MATCH_NOMATCH);
        }
-Line 2977 
 for (;;)
+Line 3355 
 for (;;)
      about... */
      case OP_NOTEXACT:
+     case OP_NOTEXACTI:
      min = max = GET2(ecode, 1);
      ecode += 3;
      goto REPEATNOTCHAR;
      case OP_NOTUPTO:
+     case OP_NOTUPTOI:
      case OP_NOTMINUPTO:
+     case OP_NOTMINUPTOI:
      min = 0;
      max = GET2(ecode, 1);
-     minimize = *ecode == OP_NOTMINUPTO;
+     minimize = *ecode == OP_NOTMINUPTO || *ecode == OP_NOTMINUPTOI;
      ecode += 3;
      goto REPEATNOTCHAR;
      case OP_NOTPOSSTAR:
+     case OP_NOTPOSSTARI:
      possessive = TRUE;
      min = 0;
      max = INT_MAX;
-Line 2997 
 for (;;)
+Line 3379 
 for (;;)
      goto REPEATNOTCHAR;
      case OP_NOTPOSPLUS:
+     case OP_NOTPOSPLUSI:
      possessive = TRUE;
      min = 1;
      max = INT_MAX;
-Line 3004 
 for (;;)
+Line 3387 
 for (;;)
      goto REPEATNOTCHAR;
      case OP_NOTPOSQUERY:
+     case OP_NOTPOSQUERYI:
      possessive = TRUE;
      min = 0;
      max = 1;
-Line 3011 
 for (;;)
+Line 3395 
 for (;;)
      goto REPEATNOTCHAR;
      case OP_NOTPOSUPTO:
+     case OP_NOTPOSUPTOI:
      possessive = TRUE;
      min = 0;
      max = GET2(ecode, 1);
-Line 3018 
 for (;;)
+Line 3403 
 for (;;)
      goto REPEATNOTCHAR;
      case OP_NOTSTAR:
+     case OP_NOTSTARI:
      case OP_NOTMINSTAR:
+     case OP_NOTMINSTARI:
      case OP_NOTPLUS:
+     case OP_NOTPLUSI:
      case OP_NOTMINPLUS:
+     case OP_NOTMINPLUSI:
      case OP_NOTQUERY:
+     case OP_NOTQUERYI:
      case OP_NOTMINQUERY:
-     c = *ecode++ - OP_NOTSTAR;
+     case OP_NOTMINQUERYI:
+     c = *ecode++ - ((op >= OP_NOTSTARI)? OP_NOTSTARI: OP_NOTSTAR);
      minimize = (c & 1) != 0;
      min = rep_min[c];                 /* Pick up values from tables; */
      max = rep_max[c];                 /* zero for max => infinity */
-Line 3045 
 for (;;)
+Line 3436 
 for (;;)
      DPRINTF(("negative matching %c{%d,%d} against subject %.*s\n", fc, min, max,
        max, eptr));
-     if ((ims & PCRE_CASELESS) != 0)
+     if (op >= OP_NOTSTARI)     /* Caseless */
        {
        fc = md->lcc[fc];
-Line 3093 
 for (;;)
+Line 3484 
 for (;;)
            register unsigned int d;
            for (fi = min;; fi++)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM28);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM28);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
-Line 3112 
 for (;;)
+Line 3503 
 for (;;)
            {
            for (fi = min;; fi++)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM29);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM29);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
-Line 3153 
 for (;;)
+Line 3544 
 for (;;)
          if (possessive) continue;
          for(;;)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM30);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM30);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (eptr-- == pp) break;        /* Stop if tried at original pos */
              BACKCHAR(eptr);
-Line 3176 
 for (;;)
+Line 3567 
 for (;;)
            if (possessive) continue;
            while (eptr >= pp)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM31);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM31);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              eptr--;
              }
-Line 3233 
 for (;;)
+Line 3624 
 for (;;)
            register unsigned int d;
            for (fi = min;; fi++)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM32);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM32);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
-Line 3251 
 for (;;)
+Line 3642 
 for (;;)
            {
            for (fi = min;; fi++)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM33);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM33);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
-Line 3291 
 for (;;)
+Line 3682 
 for (;;)
            if (possessive) continue;
            for(;;)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM34);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM34);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (eptr-- == pp) break;        /* Stop if tried at original pos */
              BACKCHAR(eptr);
-Line 3314 
 for (;;)
+Line 3705 
 for (;;)
            if (possessive) continue;
            while (eptr >= pp)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM35);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM35);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              eptr--;
              }
-Line 3406 
 for (;;)
+Line 3797 
 for (;;)
      is tidier. Also separate the UCP code, which can be the same for both UTF-8
      and single-bytes. */
      if (min > 0)
        {
  #ifdef SUPPORT_UCP
        if (prop_type >= 0)
          {
          switch(prop_type)
            {
            case PT_ANY:
            if (prop_fail_result) MRRETURN(MATCH_NOMATCH);
+           for (i = 1; i <= min; i++)
+             {
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             }
+           break;
+           case PT_LAMP:
+           for (i = 1; i <= min; i++)
+             {
+             int chartype;
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             chartype = UCD_CHARTYPE(c);
+             if ((chartype == ucp_Lu ||
+                  chartype == ucp_Ll ||
+                  chartype == ucp_Lt) == prop_fail_result)
+               MRRETURN(MATCH_NOMATCH);
+             }
+           break;
+           case PT_GC:
+           for (i = 1; i <= min; i++)
+             {
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             if ((UCD_CATEGORY(c) == prop_value) == prop_fail_result)
+               MRRETURN(MATCH_NOMATCH);
+             }
+           break;
+           case PT_PC:
+           for (i = 1; i <= min; i++)
+             {
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             if ((UCD_CHARTYPE(c) == prop_value) == prop_fail_result)
+               MRRETURN(MATCH_NOMATCH);
+             }
+           break;
+           case PT_SC:
            for (i = 1; i <= min; i++)
              {
              if (eptr >= md->end_subject)
-Line 3423 
 for (;;)
+Line 3872 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
+             if ((UCD_SCRIPT(c) == prop_value) == prop_fail_result)
+               MRRETURN(MATCH_NOMATCH);
              }
            break;
-           case PT_LAMP:
+           case PT_ALNUM:
            for (i = 1; i <= min; i++)
              {
+             int category;
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_chartype = UCD_CHARTYPE(c);
+             category = UCD_CATEGORY(c);
-             if ((prop_chartype == ucp_Lu ||
+             if ((category == ucp_L || category == ucp_N) == prop_fail_result)
-                  prop_chartype == ucp_Ll ||
-                  prop_chartype == ucp_Lt) == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            break;
-           case PT_GC:
+           case PT_SPACE:    /* Perl space */
            for (i = 1; i <= min; i++)
              {
              if (eptr >= md->end_subject)
-Line 3452 
 for (;;)
+Line 3902 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
-             if ((prop_category == prop_value) == prop_fail_result)
+                  c == CHAR_FF || c == CHAR_CR)
+                    == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            break;
-           case PT_PC:
+           case PT_PXSPACE:  /* POSIX space */
            for (i = 1; i <= min; i++)
              {
              if (eptr >= md->end_subject)
-Line 3467 
 for (;;)
+Line 3918 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_chartype = UCD_CHARTYPE(c);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
-             if ((prop_chartype == prop_value) == prop_fail_result)
+                  c == CHAR_VT || c == CHAR_FF || c == CHAR_CR)
+                    == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            break;
-           case PT_SC:
+           case PT_WORD:
            for (i = 1; i <= min; i++)
              {
+             int category;
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_script = UCD_SCRIPT(c);
+             category = UCD_CATEGORY(c);
-             if ((prop_script == prop_value) == prop_fail_result)
+             if ((category == ucp_L || category == ucp_N || c == CHAR_UNDERSCORE)
+                    == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            break;
+           /* This should not occur */
            default:
            RRETURN(PCRE_ERROR_INTERNAL);
            }
-Line 3506 
 for (;;)
+Line 3962 
 for (;;)
              MRRETURN(MATCH_NOMATCH);
              }
            GETCHARINCTEST(c, eptr);
-           prop_category = UCD_CATEGORY(c);
+           if (UCD_CATEGORY(c) == ucp_M) MRRETURN(MATCH_NOMATCH);
-           if (prop_category == ucp_M) MRRETURN(MATCH_NOMATCH);
            while (eptr < md->end_subject)
              {
              int len = 1;
-             if (!utf8) c = *eptr;
+             if (!utf8) c = *eptr; else { GETCHARLEN(c, eptr, len); }
-               else { GETCHARLEN(c, eptr, len); }
+             if (UCD_CATEGORY(c) != ucp_M) break;
-             prop_category = UCD_CATEGORY(c);
-             if (prop_category != ucp_M) break;
              eptr += len;
              }
            }
-Line 3572 
 for (;;)
+Line 4025 
 for (;;)
            switch(c)
              {
              default: MRRETURN(MATCH_NOMATCH);
              case 0x000d:
              if (eptr < md->end_subject && *eptr == 0x0a) eptr++;
              break;
-Line 3848 
 for (;;)
+Line 4302 
 for (;;)
            switch(*eptr++)
              {
              default: MRRETURN(MATCH_NOMATCH);
              case 0x000d:
              if (eptr < md->end_subject && *eptr == 0x0a) eptr++;
              break;
              case 0x000a:
              break;
-Line 4040 
 for (;;)
+Line 4496 
 for (;;)
            case PT_ANY:
            for (fi = min;; fi++)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM36);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM36);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
-Line 4048 
 for (;;)
+Line 4504 
 for (;;)
                SCHECK_PARTIAL();
                MRRETURN(MATCH_NOMATCH);
                }
-             GETCHARINC(c, eptr);
+             GETCHARINCTEST(c, eptr);
              if (prop_fail_result) MRRETURN(MATCH_NOMATCH);
              }
            /* Control never gets here */
-Line 4056 
 for (;;)
+Line 4512 
 for (;;)
            case PT_LAMP:
            for (fi = min;; fi++)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM37);
+             int chartype;
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM37);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
-Line 4064 
 for (;;)
+Line 4521 
 for (;;)
                SCHECK_PARTIAL();
                MRRETURN(MATCH_NOMATCH);
                }
-             GETCHARINC(c, eptr);
+             GETCHARINCTEST(c, eptr);
-             prop_chartype = UCD_CHARTYPE(c);
+             chartype = UCD_CHARTYPE(c);
-             if ((prop_chartype == ucp_Lu ||
+             if ((chartype == ucp_Lu ||
-                  prop_chartype == ucp_Ll ||
+                  chartype == ucp_Ll ||
-                  prop_chartype == ucp_Lt) == prop_fail_result)
+                  chartype == ucp_Lt) == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            /* Control never gets here */
-Line 4076 
 for (;;)
+Line 4533 
 for (;;)
            case PT_GC:
            for (fi = min;; fi++)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM38);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM38);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
-Line 4084 
 for (;;)
+Line 4541 
 for (;;)
                SCHECK_PARTIAL();
                MRRETURN(MATCH_NOMATCH);
                }
-             GETCHARINC(c, eptr);
+             GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == prop_value) == prop_fail_result)
-             if ((prop_category == prop_value) == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            /* Control never gets here */
-Line 4094 
 for (;;)
+Line 4550 
 for (;;)
            case PT_PC:
            for (fi = min;; fi++)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM39);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM39);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
-Line 4102 
 for (;;)
+Line 4558 
 for (;;)
                SCHECK_PARTIAL();
                MRRETURN(MATCH_NOMATCH);
                }
-             GETCHARINC(c, eptr);
+             GETCHARINCTEST(c, eptr);
-             prop_chartype = UCD_CHARTYPE(c);
+             if ((UCD_CHARTYPE(c) == prop_value) == prop_fail_result)
-             if ((prop_chartype == prop_value) == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            /* Control never gets here */
-Line 4112 
 for (;;)
+Line 4567 
 for (;;)
            case PT_SC:
            for (fi = min;; fi++)
              {
-             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM40);
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM40);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
-Line 4120 
 for (;;)
+Line 4575 
 for (;;)
                SCHECK_PARTIAL();
                MRRETURN(MATCH_NOMATCH);
                }
-             GETCHARINC(c, eptr);
+             GETCHARINCTEST(c, eptr);
-             prop_script = UCD_SCRIPT(c);
+             if ((UCD_SCRIPT(c) == prop_value) == prop_fail_result)
-             if ((prop_script == prop_value) == prop_fail_result)
+               MRRETURN(MATCH_NOMATCH);
+             }
+           /* Control never gets here */
+           case PT_ALNUM:
+           for (fi = min;; fi++)
+             {
+             int category;
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM59);
+             if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+             if (fi >= max) MRRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             category = UCD_CATEGORY(c);
+             if ((category == ucp_L || category == ucp_N) == prop_fail_result)
+               MRRETURN(MATCH_NOMATCH);
+             }
+           /* Control never gets here */
+           case PT_SPACE:    /* Perl space */
+           for (fi = min;; fi++)
+             {
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM60);
+             if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+             if (fi >= max) MRRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
+                  c == CHAR_FF || c == CHAR_CR)
+                    == prop_fail_result)
+               MRRETURN(MATCH_NOMATCH);
+             }
+           /* Control never gets here */
+           case PT_PXSPACE:  /* POSIX space */
+           for (fi = min;; fi++)
+             {
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM61);
+             if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+             if (fi >= max) MRRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
+                  c == CHAR_VT || c == CHAR_FF || c == CHAR_CR)
+                    == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            /* Control never gets here */
+           case PT_WORD:
+           for (fi = min;; fi++)
+             {
+             int category;
+             RMATCH(eptr, ecode, offset_top, md, eptrb, RM62);
+             if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+             if (fi >= max) MRRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             category = UCD_CATEGORY(c);
+             if ((category == ucp_L ||
+                  category == ucp_N ||
+                  c == CHAR_UNDERSCORE)
+                    == prop_fail_result)
+               MRRETURN(MATCH_NOMATCH);
+             }
+           /* Control never gets here */
+           /* This should never occur */
            default:
            RRETURN(PCRE_ERROR_INTERNAL);
            }
-Line 4139 
 for (;;)
+Line 4674 
 for (;;)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM41);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM41);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (fi >= max) MRRETURN(MATCH_NOMATCH);
            if (eptr >= md->end_subject)
-Line 4148 
 for (;;)
+Line 4683 
 for (;;)
              MRRETURN(MATCH_NOMATCH);
              }
            GETCHARINCTEST(c, eptr);
-           prop_category = UCD_CATEGORY(c);
+           if (UCD_CATEGORY(c) == ucp_M) MRRETURN(MATCH_NOMATCH);
-           if (prop_category == ucp_M) MRRETURN(MATCH_NOMATCH);
            while (eptr < md->end_subject)
              {
              int len = 1;
-             if (!utf8) c = *eptr;
+             if (!utf8) c = *eptr; else { GETCHARLEN(c, eptr, len); }
-               else { GETCHARLEN(c, eptr, len); }
+             if (UCD_CATEGORY(c) != ucp_M) break;
-             prop_category = UCD_CATEGORY(c);
-             if (prop_category != ucp_M) break;
              eptr += len;
              }
            }
          }
        else
  #endif     /* SUPPORT_UCP */
-Line 4171 
 for (;;)
+Line 4702 
 for (;;)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM42);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM42);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (fi >= max) MRRETURN(MATCH_NOMATCH);
            if (eptr >= md->end_subject)
-Line 4334 
 for (;;)
+Line 4865 
 for (;;)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM43);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM43);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (fi >= max) MRRETURN(MATCH_NOMATCH);
            if (eptr >= md->end_subject)
-Line 4473 
 for (;;)
+Line 5004 
 for (;;)
                SCHECK_PARTIAL();
                break;
                }
-             GETCHARLEN(c, eptr, len);
+             GETCHARLENTEST(c, eptr, len);
              if (prop_fail_result) break;
              eptr+= len;
              }
-Line 4482 
 for (;;)
+Line 5013 
 for (;;)
            case PT_LAMP:
            for (i = min; i < max; i++)
              {
+             int chartype;
              int len = 1;
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
                break;
                }
-             GETCHARLEN(c, eptr, len);
+             GETCHARLENTEST(c, eptr, len);
-             prop_chartype = UCD_CHARTYPE(c);
+             chartype = UCD_CHARTYPE(c);
-             if ((prop_chartype == ucp_Lu ||
+             if ((chartype == ucp_Lu ||
-                  prop_chartype == ucp_Ll ||
+                  chartype == ucp_Ll ||
-                  prop_chartype == ucp_Lt) == prop_fail_result)
+                  chartype == ucp_Lt) == prop_fail_result)
                break;
              eptr+= len;
              }
-Line 4507 
 for (;;)
+Line 5039 
 for (;;)
                SCHECK_PARTIAL();
                break;
                }
-             GETCHARLEN(c, eptr, len);
+             GETCHARLENTEST(c, eptr, len);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == prop_value) == prop_fail_result) break;
-             if ((prop_category == prop_value) == prop_fail_result)
-               break;
              eptr+= len;
              }
            break;
-Line 4524 
 for (;;)
+Line 5054 
 for (;;)
                SCHECK_PARTIAL();
                break;
                }
-             GETCHARLEN(c, eptr, len);
+             GETCHARLENTEST(c, eptr, len);
-             prop_chartype = UCD_CHARTYPE(c);
+             if ((UCD_CHARTYPE(c) == prop_value) == prop_fail_result) break;
-             if ((prop_chartype == prop_value) == prop_fail_result)
-               break;
              eptr+= len;
              }
            break;
-Line 4541 
 for (;;)
+Line 5069 
 for (;;)
                SCHECK_PARTIAL();
                break;
                }
-             GETCHARLEN(c, eptr, len);
+             GETCHARLENTEST(c, eptr, len);
-             prop_script = UCD_SCRIPT(c);
+             if ((UCD_SCRIPT(c) == prop_value) == prop_fail_result) break;
-             if ((prop_script == prop_value) == prop_fail_result)
+             eptr+= len;
+             }
+           break;
+           case PT_ALNUM:
+           for (i = min; i < max; i++)
+             {
+             int category;
+             int len = 1;
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               break;
+               }
+             GETCHARLENTEST(c, eptr, len);
+             category = UCD_CATEGORY(c);
+             if ((category == ucp_L || category == ucp_N) == prop_fail_result)
+               break;
+             eptr+= len;
+             }
+           break;
+           case PT_SPACE:    /* Perl space */
+           for (i = min; i < max; i++)
+             {
+             int len = 1;
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               break;
+               }
+             GETCHARLENTEST(c, eptr, len);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
+                  c == CHAR_FF || c == CHAR_CR)
+                  == prop_fail_result)
                break;
              eptr+= len;
              }
            break;
+           case PT_PXSPACE:  /* POSIX space */
+           for (i = min; i < max; i++)
+             {
+             int len = 1;
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               break;
+               }
+             GETCHARLENTEST(c, eptr, len);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
+                  c == CHAR_VT || c == CHAR_FF || c == CHAR_CR)
+                  == prop_fail_result)
+               break;
+             eptr+= len;
+             }
+           break;
+           case PT_WORD:
+           for (i = min; i < max; i++)
+             {
+             int category;
+             int len = 1;
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               break;
+               }
+             GETCHARLENTEST(c, eptr, len);
+             category = UCD_CATEGORY(c);
+             if ((category == ucp_L || category == ucp_N ||
+                  c == CHAR_UNDERSCORE) == prop_fail_result)
+               break;
+             eptr+= len;
+             }
+           break;
+           default:
+           RRETURN(PCRE_ERROR_INTERNAL);
            }
          /* eptr is now past the end of the maximum run */
-Line 4555 
 for (;;)
+Line 5157 
 for (;;)
          if (possessive) continue;
          for(;;)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM44);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM44);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (eptr-- == pp) break;        /* Stop if tried at original pos */
            if (utf8) BACKCHAR(eptr);
-Line 4569 
 for (;;)
+Line 5171 
 for (;;)
          {
          for (i = min; i < max; i++)
            {
+           int len = 1;
            if (eptr >= md->end_subject)
              {
              SCHECK_PARTIAL();
              break;
              }
-           GETCHARINCTEST(c, eptr);
+           if (!utf8) c = *eptr; else { GETCHARLEN(c, eptr, len); }
-           prop_category = UCD_CATEGORY(c);
+           if (UCD_CATEGORY(c) == ucp_M) break;
-           if (prop_category == ucp_M) break;
+           eptr += len;
            while (eptr < md->end_subject)
              {
-             int len = 1;
+             len = 1;
-             if (!utf8) c = *eptr; else
+             if (!utf8) c = *eptr; else { GETCHARLEN(c, eptr, len); }
-               {
+             if (UCD_CATEGORY(c) != ucp_M) break;
-               GETCHARLEN(c, eptr, len);
-               }
-             prop_category = UCD_CATEGORY(c);
-             if (prop_category != ucp_M) break;
              eptr += len;
              }
            }
-Line 4596 
 for (;;)
+Line 5195 
 for (;;)
          for(;;)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM45);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM45);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (eptr-- == pp) break;        /* Stop if tried at original pos */
            for (;;)                        /* Move back over one extended */
              {
-             int len = 1;
              if (!utf8) c = *eptr; else
                {
                BACKCHAR(eptr);
-               GETCHARLEN(c, eptr, len);
+               GETCHAR(c, eptr);
                }
-             prop_category = UCD_CATEGORY(c);
+             if (UCD_CATEGORY(c) != ucp_M) break;
-             if (prop_category != ucp_M) break;
              eptr--;
              }
            }
-Line 4672 
 for (;;)
+Line 5269 
 for (;;)
                while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
                }
              }
-           else eptr = md->end_subject;   /* Unlimited UTF-8 repeat */
+           else
+             {
+             eptr = md->end_subject;   /* Unlimited UTF-8 repeat */
+             SCHECK_PARTIAL();
+             }
            break;
            /* The byte case is the same as non-UTF8 */
-Line 4880 
 for (;;)
+Line 5481 
 for (;;)
            RRETURN(PCRE_ERROR_INTERNAL);
            }
-         /* eptr is now past the end of the maximum run */
+         /* eptr is now past the end of the maximum run. If possessive, we are
+         done (no backing up). Otherwise, match at this position; anything other
+         than no match is immediately returned. For nomatch, back up one
+         character, unless we are matching \R and the last thing matched was
+         \r\n, in which case, back up two bytes. */
          if (possessive) continue;
          for(;;)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM46);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM46);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (eptr-- == pp) break;        /* Stop if tried at original pos */
            BACKCHAR(eptr);
+           if (ctype == OP_ANYNL && eptr > pp  && *eptr == '\n' &&
+               eptr[-1] == '\r') eptr--;
            }
          }
        else
-Line 5087 
 for (;;)
+Line 5694 
 for (;;)
            RRETURN(PCRE_ERROR_INTERNAL);
            }
-         /* eptr is now past the end of the maximum run */
+         /* eptr is now past the end of the maximum run. If possessive, we are
+         done (no backing up). Otherwise, match at this position; anything other
+         than no match is immediately returned. For nomatch, back up one
+         character (byte), unless we are matching \R and the last thing matched
+         was \r\n, in which case, back up two bytes. */
          if (possessive) continue;
          while (eptr >= pp)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM47);
+           RMATCH(eptr, ecode, offset_top, md, eptrb, RM47);
-           eptr--;
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+           eptr--;
+           if (ctype == OP_ANYNL && eptr > pp  && *eptr == '\n' &&
+               eptr[-1] == '\r') eptr--;
            }
          }
-Line 5133 
 switch (frame->Xwhere)
+Line 5746 
 switch (frame->Xwhere)
    LBL( 9) LBL(10) LBL(11) LBL(12) LBL(13) LBL(14) LBL(15) LBL(17)
    LBL(19) LBL(24) LBL(25) LBL(26) LBL(27) LBL(29) LBL(31) LBL(33)
    LBL(35) LBL(43) LBL(47) LBL(48) LBL(49) LBL(50) LBL(51) LBL(52)
-   LBL(53) LBL(54)
+   LBL(53) LBL(54) LBL(55) LBL(56) LBL(57) LBL(58) LBL(63)
  #ifdef SUPPORT_UTF8
    LBL(16) LBL(18) LBL(20) LBL(21) LBL(22) LBL(23) LBL(28) LBL(30)
    LBL(32) LBL(34) LBL(42) LBL(46)
  #ifdef SUPPORT_UCP
    LBL(36) LBL(37) LBL(38) LBL(39) LBL(40) LBL(41) LBL(44) LBL(45)
+   LBL(59) LBL(60) LBL(61) LBL(62)
  #endif  /* SUPPORT_UCP */
  #endif  /* SUPPORT_UTF8 */
    default:
-Line 5161 
 Undefine all the macros that were define
+Line 5775 
 Undefine all the macros that were define
  #undef ecode
  #undef mstart
  #undef offset_top
- #undef ims
  #undef eptrb
  #undef flags
-Line 5179 
 Undefine all the macros that were define
+Line 5792 
 Undefine all the macros that were define
  #undef condition
  #undef prev_is_word
- #undef original_ims
  #undef ctype
  #undef length
  #undef max
-Line 5237 
 pcre_exec(const pcre *argument_re, const
+Line 5848 
 pcre_exec(const pcre *argument_re, const
    PCRE_SPTR subject, int length, int start_offset, int options, int *offsets,
    int offsetcount)
  {
- int rc, resetcount, ocount;
+ int rc, ocount, arg_offset_max;
  int first_byte = -1;
  int req_byte = -1;
  int req_byte2 = -1;
  int newline;
- unsigned long int ims;
  BOOL using_temporary_offsets = FALSE;
  BOOL anchored;
  BOOL startline;
-Line 5272 
 if ((options & ~PUBLIC_EXEC_OPTIONS) !=
+Line 5882 
 if ((options & ~PUBLIC_EXEC_OPTIONS) !=
  if (re == NULL || subject == NULL ||
     (offsets == NULL && offsetcount > 0)) return PCRE_ERROR_NULL;
  if (offsetcount < 0) return PCRE_ERROR_BADCOUNT;
+ if (start_offset < 0 || start_offset > length) return PCRE_ERROR_BADOFFSET;
+ /* These two settings are used in the code for checking a UTF-8 string that
+ follows immediately afterwards. Other values in the md block are used only
+ during "normal" pcre_exec() processing, not when the JIT support is in use,
+ so they are set up later. */
+ utf8 = md->utf8 = (re->options & PCRE_UTF8) != 0;
+ md->partial = ((options & PCRE_PARTIAL_HARD) != 0)? 2 :
+               ((options & PCRE_PARTIAL_SOFT) != 0)? 1 : 0;
+ /* Check a UTF-8 string if required. Pass back the character offset and error
+ code for an invalid string if a results vector is available. */
+ #ifdef SUPPORT_UTF8
+ if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0)
+   {
+   int erroroffset;
+   int errorcode = _pcre_valid_utf8((USPTR)subject, length, &erroroffset);
+   if (errorcode != 0)
+     {
+     if (offsetcount >= 2)
+       {
+       offsets[0] = erroroffset;
+       offsets[1] = errorcode;
+       }
+     return (errorcode <= PCRE_UTF8_ERR5 && md->partial > 1)?
+       PCRE_ERROR_SHORTUTF8 : PCRE_ERROR_BADUTF8;
+     }
+   /* Check that a start_offset points to the start of a UTF-8 character. */
+   if (start_offset > 0 && start_offset < length &&
+       (((USPTR)subject)[start_offset] & 0xc0) == 0x80)
+     return PCRE_ERROR_BADUTF8_OFFSET;
+   }
+ #endif
+ /* If the pattern was successfully studied with JIT support, run the JIT
+ executable instead of the rest of this function. Most options must be set at
+ compile time for the JIT code to be usable. Fallback to the normal code path if
+ an unsupported flag is set. In particular, JIT does not support partial
+ matching. */
+ #ifdef SUPPORT_JIT
+ if (extra_data != NULL
+     && (extra_data->flags & PCRE_EXTRA_EXECUTABLE_JIT) != 0
+     && extra_data->executable_jit != NULL
+     && (options & ~(PCRE_NO_UTF8_CHECK | PCRE_NOTBOL | PCRE_NOTEOL |
+                     PCRE_NOTEMPTY | PCRE_NOTEMPTY_ATSTART)) == 0)
+   return _pcre_jit_exec(re, extra_data->executable_jit, subject, length,
+     start_offset, options, ((extra_data->flags & PCRE_EXTRA_MATCH_LIMIT) == 0)
+     ? MATCH_LIMIT : extra_data->match_limit, offsets, offsetcount);
+ #endif
- /* This information is for finding all the numbers associated with a given
+ /* Carry on with non-JIT matching. This information is for finding all the
- name, for condition testing. */
+ numbers associated with a given name, for condition testing. */
  md->name_table = (uschar *)re + re->name_table_offset;
  md->name_count = re->name_count;
-Line 5341 
 md->end_subject = md->start_subject + le
+Line 6004 
 md->end_subject = md->start_subject + le
  end_subject = md->end_subject;
  md->endonly = (re->options & PCRE_DOLLAR_ENDONLY) != 0;
- utf8 = md->utf8 = (re->options & PCRE_UTF8) != 0;
+ md->use_ucp = (re->options & PCRE_UCP) != 0;
  md->jscript_compat = (re->options & PCRE_JAVASCRIPT_COMPAT) != 0;
+ /* Some options are unpacked into BOOL variables in the hope that testing
+ them will be faster than individual option bits. */
  md->notbol = (options & PCRE_NOTBOL) != 0;
  md->noteol = (options & PCRE_NOTEOL) != 0;
  md->notempty = (options & PCRE_NOTEMPTY) != 0;
  md->notempty_atstart = (options & PCRE_NOTEMPTY_ATSTART) != 0;
- md->partial = ((options & PCRE_PARTIAL_HARD) != 0)? 2 :
-               ((options & PCRE_PARTIAL_SOFT) != 0)? 1 : 0;
  md->hitend = FALSE;
  md->mark = NULL;                        /* In case never set */
  md->recursive = NULL;                   /* No recursion at top level */
+ md->hasthen = (re->flags & PCRE_HASTHEN) != 0;
  md->lcc = tables + lcc_offset;
  md->ctypes = tables + ctypes_offset;
-Line 5431 
 defined (though never set). So there's n
+Line 6097 
 defined (though never set). So there's n
  if (md->partial && (re->flags & PCRE_NOPARTIAL) != 0)
    return PCRE_ERROR_BADPARTIAL;
- /* Check a UTF-8 string if required. Unfortunately there's no way of passing
- back the character offset. */
- #ifdef SUPPORT_UTF8
- if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0)
-   {
-   if (_pcre_valid_utf8((USPTR)subject, length) >= 0)
-     return PCRE_ERROR_BADUTF8;
-   if (start_offset > 0 && start_offset < length)
-     {
-     int tb = ((USPTR)subject)[start_offset];
-     if (tb > 127)
-       {
-       tb &= 0xc0;
-       if (tb != 0 && tb != 0xc0) return PCRE_ERROR_BADUTF8_OFFSET;
-       }
-     }
-   }
- #endif
- /* The ims options can vary during the matching as a result of the presence
- of (?ims) items in the pattern. They are kept in a local variable so that
- restoring at the exit of a group is easy. */
- ims = re->options & (PCRE_CASELESS|PCRE_MULTILINE|PCRE_DOTALL);
  /* If the expression has got more back references than the offsets supplied can
  hold, we get a temporary chunk of working store to use during the matching.
  Otherwise, we can use the vector supplied, rounding down its size to a multiple
  of 3. */
  ocount = offsetcount - (offsetcount % 3);
+ arg_offset_max = (2*ocount)/3;
  if (re->top_backref > 0 && re->top_backref >= ocount/3)
    {
-Line 5479 
 md->offset_max = (2*ocount)/3;
+Line 6120 
 md->offset_max = (2*ocount)/3;
  md->offset_overflow = FALSE;
  md->capture_last = -1;
- /* Compute the minimum number of offsets that we need to reset each time. Doing
- this makes a huge difference to execution time when there aren't many brackets
- in the pattern. */
- resetcount = 2 + re->top_bracket * 2;
- if (resetcount > offsetcount) resetcount = ocount;
  /* Reset the working variable associated with each extraction. These should
  never be used unless previously set, but they get saved and restored, and so we
- initialize them to avoid reading uninitialized locations. */
+ initialize them to avoid reading uninitialized locations. Also, unset the
+ offsets for the matched string. This is really just for tidiness with callouts,
+ in case they inspect these fields. */
  if (md->offset_vector != NULL)
    {
    register int *iptr = md->offset_vector + ocount;
-   register int *iend = iptr - resetcount/2 + 1;
+   register int *iend = iptr - re->top_bracket;
+   if (iend < md->offset_vector + 2) iend = md->offset_vector + 2;
    while (--iptr >= iend) *iptr = -1;
+   md->offset_vector[0] = md->offset_vector[1] = -1;
    }
  /* Set up the first character to match, if available. The first_byte value is
-Line 5528 
 if ((re->flags & PCRE_REQCHSET) != 0)
+Line 6166 
 if ((re->flags & PCRE_REQCHSET) != 0)
    }
  /* ==========================================================================*/
  /* Loop for handling unanchored repeated matching attempts; for anchored regexs
-Line 5538 
 for(;;)
+Line 6178 
 for(;;)
    USPTR save_end_subject = end_subject;
    USPTR new_start_match;
-   /* Reset the maximum number of extractions we might see. */
-   if (md->offset_vector != NULL)
-     {
-     register int *iptr = md->offset_vector;
-     register int *iend = iptr + resetcount;
-     while (iptr < iend) *iptr++ = -1;
-     }
    /* If firstline is TRUE, the start of the match is constrained to the first
    line of a multiline string. That is, the match must be before or at the first
    newline. Implement this by temporarily adjusting end_subject so that we stop
-Line 5574 
 for(;;)
+Line 6205 
 for(;;)
    /* There are some optimizations that avoid running the match if a known
    starting point is not found, or if a known later character is not present.
    However, there is an option that disables these, for testing and for ensuring
-   that all callouts do actually occur. */
+   that all callouts do actually occur. The option can be set in the regex by
+   (*NO_START_OPT) or passed in match-time options. */
-   if ((options & PCRE_NO_START_OPTIMIZE) == 0)
+   if (((options | re->options) & PCRE_NO_START_OPTIMIZE) == 0)
      {
      /* Advance to a unique first byte if there is one. */
-Line 5630 
 for(;;)
+Line 6262 
 for(;;)
        while (start_match < end_subject)
          {
          register unsigned int c = *start_match;
-         if ((start_bits[c/8] & (1 << (c&7))) == 0) start_match++;
+         if ((start_bits[c/8] & (1 << (c&7))) == 0)
-           else break;
+           {
+           start_match++;
+ #ifdef SUPPORT_UTF8
+           if (utf8)
+             while(start_match < end_subject && (*start_match & 0xc0) == 0x80)
+               start_match++;
+ #endif
+           }
+         else break;
          }
        }
      }   /* Starting optimizations */
-Line 5722 
 for(;;)
+Line 6362 
 for(;;)
    /* OK, we can now run the match. If "hitend" is set afterwards, remember the
    first starting point for which a partial match was found. */
    md->start_match_ptr = start_match;
    md->start_used_ptr = start_match;
    md->match_call_count = 0;
-   rc = match(start_match, md->start_code, start_match, NULL, 2, md, ims, NULL,
+   md->match_function_type = 0;
-, 0);
+   md->end_offset_top = 0;
+   rc = match(start_match, md->start_code, start_match, NULL, 2, md, NULL, 0);
    if (md->hitend && start_partial == NULL) start_partial = md->start_used_ptr;
    switch(rc)
      {
-     /* NOMATCH and PRUNE advance by one character. If MATCH_SKIP_ARG reaches
+     /* SKIP passes back the next starting point explicitly, but if it is the
-     this level it means that a MARK that matched the SKIP's arg was not found.
+     same as the match we have just done, treat it as NOMATCH. */
-     We treat this as NOMATCH. THEN at this level acts exactly like PRUNE. */
+     case MATCH_SKIP:
+     if (md->start_match_ptr != start_match)
+       {
+       new_start_match = md->start_match_ptr;
+       break;
+       }
+     /* Fall through */
+     /* If MATCH_SKIP_ARG reaches this level it means that a MARK that matched
+     the SKIP's arg was not found. We also treat this as NOMATCH. */
+     case MATCH_SKIP_ARG:
+     /* Fall through */
+     /* NOMATCH and PRUNE advance by one character. THEN at this level acts
+     exactly like PRUNE. */
      case MATCH_NOMATCH:
      case MATCH_PRUNE:
-     case MATCH_SKIP_ARG:
      case MATCH_THEN:
      new_start_match = start_match + 1;
  #ifdef SUPPORT_UTF8
-Line 5748 
 for(;;)
+Line 6404 
 for(;;)
  #endif
      break;
-     /* SKIP passes back the next starting point explicitly. */
-     case MATCH_SKIP:
-     new_start_match = md->start_match_ptr;
-     break;
      /* COMMIT disables the bumpalong, but otherwise behaves as NOMATCH. */
      case MATCH_COMMIT:
-Line 5828 
 if (rc == MATCH_MATCH || rc == MATCH_ACC
+Line 6478 
 if (rc == MATCH_MATCH || rc == MATCH_ACC
    {
    if (using_temporary_offsets)
      {
-     if (offsetcount >= 4)
+     if (arg_offset_max >= 4)
        {
        memcpy(offsets + 2, md->offset_vector + 2,
-         (offsetcount - 2) * sizeof(int));
+         (arg_offset_max - 2) * sizeof(int));
        DPRINTF(("Copied offsets from temporary memory\n"));
        }
-     if (md->end_offset_top > offsetcount) md->offset_overflow = TRUE;
+     if (md->end_offset_top > arg_offset_max) md->offset_overflow = TRUE;
      DPRINTF(("Freeing temporary memory\n"));