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

Diff of /code/trunk/pcre_exec.c

Parent Directory | Revision Log | View Patch Patch

-revision 199 by ph10,
Tue Jul 31 14:39:09 2007 UTC
+revision 614 by ph10,
Sat Jul  9 10:48:16 2011 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2007 University of Cambridge
+            Copyright (c) 1997-2011 University of Cambridge
  -----------------------------------------------------------------------------
  Redistribution and use in source and binary forms, with or without
 Line 42 
 POSSIBILITY OF SUCH DAMAGE.
  pattern matching using an NFA algorithm, trying to mimic Perl as closely as
  possible. There are also some static supporting functions. */
  #ifdef HAVE_CONFIG_H
- #include <config.h>
+ #include "config.h"
  #endif
  #define NLBLOCK md             /* Block containing newline information */
  #define PSSTART start_subject  /* Field containing processed string start */
 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 68 
 defined PCRE_ERROR_xxx codes, which are
+Line 70 
 defined PCRE_ERROR_xxx codes, which are
  #define MATCH_MATCH        1
  #define MATCH_NOMATCH      0
+ /* Special internal returns from the match() function. Make them sufficiently
+ negative to avoid the external error codes. */
+ #define MATCH_ACCEPT       (-999)
+ #define MATCH_COMMIT       (-998)
+ #define MATCH_KETRPOS      (-997)
+ #define MATCH_PRUNE        (-996)
+ #define MATCH_SKIP         (-995)
+ #define MATCH_SKIP_ARG     (-994)
+ #define MATCH_THEN         (-993)
+ /* This is a convenience macro for code that occurs many times. */
+ #define MRRETURN(ra) \
+   { \
+   md->mark = markptr; \
+   RRETURN(ra); \
+   }
  /* Maximum number of ints of offset to save on the stack for recursive calls.
  If the offset vector is bigger, malloc is used. This should be a multiple of 3,
  because the offset vector is always a multiple of 3 long. */
-Line 81 
 static const char rep_max[] = { 0, 0, 0,
+Line 102 
 static const char rep_max[] = { 0, 0, 0,
- #ifdef DEBUG
+ #ifdef PCRE_DEBUG
  /*************************************************
  *        Debugging function to print chars       *
  *************************************************/
-Line 114 
 while (length-- > 0)
+Line 135 
 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 DEBUG
+ #ifdef PCRE_DEBUG
  if (eptr >= md->end_subject)
    printf("matching subject <null>");
  else
-Line 146 
 pchars(p, length, FALSE, md);
+Line 170 
 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 caselesss case for speed */
+ /* Separate the caseless case for speed. In UTF-8 mode we can only do this
+ properly if Unicode properties are supported. Otherwise, we can check only
+ ASCII characters. */
- if ((ims & PCRE_CASELESS) != 0)
+ if (caseless)
    {
-   while (length-- > 0)
+ #ifdef SUPPORT_UTF8
-     if (md->lcc[*p++] != md->lcc[*eptr++]) return FALSE;
+ #ifdef SUPPORT_UCP
+   if (md->utf8)
+     {
+     /* Match characters up to the end of the reference. NOTE: the number of
+     bytes matched may differ, because there are some characters whose upper and
+     lower case versions code as different numbers of bytes. For example, U+023A
+     (2 bytes in UTF-8) is the upper case version of U+2C65 (3 bytes in UTF-8);
+     a sequence of 3 of the former uses 6 bytes, as does a sequence of two of
+     the latter. It is important, therefore, to check the length along the
+     reference, not along the subject (earlier code did this wrong). */
+     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 -1;
+       }
+     }
+   else
+ #endif
+ #endif
+   /* The same code works when not in UTF-8 mode and in UTF-8 mode when there
+   is no UCP support. */
+     {
+     if (eptr + length > md->end_subject) return -1;
+     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 203 
 variable instead of being passed in the
+Line 267 
 variable instead of being passed in the
  ****************************************************************************
  ***************************************************************************/
+ /* Numbers for RMATCH calls. When this list is changed, the code at HEAP_RETURN
- /* Numbers for RMATCH calls */
+ below must be updated in sync.  */
  enum { RM1=1, RM2,  RM3,  RM4,  RM5,  RM6,  RM7,  RM8,  RM9,  RM10,
         RM11,  RM12, RM13, RM14, RM15, RM16, RM17, RM18, RM19, RM20,
         RM21,  RM22, RM23, RM24, RM25, RM26, RM27, RM28, RM29, RM30,
         RM31,  RM32, RM33, RM34, RM35, RM36, RM37, RM38, RM39, RM40,
-        RM41,  RM42, RM43, RM44, RM45, RM46, RM47, RM48, RM49, RM50 };
+        RM41,  RM42, RM43, RM44, RM45, RM46, RM47, RM48, RM49, RM50,
+        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
- actuall used in this definition. */
+ actually used in this definition. */
  #ifndef NO_RECURSE
  #define REGISTER register
- #ifdef DEBUG
+ #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,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 233 
 actuall used in this definition. */
+Line 298 
 actuall 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,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 247 
 argument of match(), which never changes
+Line 312 
 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 269 
 argument of match(), which never changes
+Line 334 
 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 288 
 typedef struct heapframe {
+Line 353 
 typedef struct heapframe {
    /* Function arguments that may change */
-   const uschar *Xeptr;
+   USPTR Xeptr;
    const uschar *Xecode;
-   const uschar *Xmstart;
+   USPTR Xmstart;
+   USPTR Xmarkptr;
    int Xoffset_top;
-   long int Xims;
    eptrblock *Xeptrb;
-   int Xflags;
    unsigned int Xrdepth;
    /* Function local variables */
-   const uschar *Xcallpat;
+   USPTR Xcallpat;
-   const uschar *Xcharptr;
+ #ifdef SUPPORT_UTF8
-   const uschar *Xdata;
+   USPTR Xcharptr;
-   const uschar *Xnext;
+ #endif
-   const uschar *Xpp;
+   USPTR Xdata;
-   const uschar *Xprev;
+   USPTR Xnext;
-   const uschar *Xsaved_eptr;
+   USPTR Xpp;
+   USPTR Xprev;
+   USPTR Xsaved_eptr;
    recursion_info Xnew_recursive;
-Line 313 
 typedef struct heapframe {
+Line 379 
 typedef struct heapframe {
    BOOL Xcondition;
    BOOL Xprev_is_word;
-   unsigned long int Xoriginal_ims;
  #ifdef SUPPORT_UCP
    int Xprop_type;
    int Xprop_value;
-Line 326 
 typedef struct heapframe {
+Line 390 
 typedef struct heapframe {
    uschar Xocchars[8];
  #endif
+   int Xcodelink;
    int Xctype;
    unsigned int Xfc;
    int Xfi;
-Line 361 
 typedef struct heapframe {
+Line 426 
 typedef struct heapframe {
  /* This function is called recursively in many circumstances. Whenever it
  returns a negative (error) response, the outer incarnation must also return the
- same response.
+ same response. */
+ /* These macros pack up tests that are used for partial matching, and which
+ appears several times in the code. We set the "hit end" flag if the pointer is
+ at the end of the subject and also past the start of the subject (i.e.
+ something has been matched). For hard partial matching, we then return
+ immediately. The second one is used when we already know we are past the end of
+ the subject. */
+ #define CHECK_PARTIAL()\
+   if (md->partial != 0 && eptr >= md->end_subject && \
+       eptr > md->start_used_ptr) \
+     { \
+     md->hitend = TRUE; \
+     if (md->partial > 1) MRRETURN(PCRE_ERROR_PARTIAL); \
+     }
- Performance note: It might be tempting to extract commonly used fields from the
+ #define SCHECK_PARTIAL()\
- md structure (e.g. utf8, end_subject) into individual variables to improve
+   if (md->partial != 0 && eptr > md->start_used_ptr) \
+     { \
+     md->hitend = TRUE; \
+     if (md->partial > 1) MRRETURN(PCRE_ERROR_PARTIAL); \
+     }
+ /* Performance note: It might be tempting to extract commonly used fields from
+ the md structure (e.g. utf8, end_subject) into individual variables to improve
  performance. Tests using gcc on a SPARC disproved this; in the first case, it
  made performance worse.
-Line 373 
 Arguments:
+Line 461 
 Arguments:
     ecode       pointer to current position in compiled code
     mstart      pointer to the current match start position (can be modified
                   by encountering \K)
+    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
                 MATCH_NOMATCH if failed to match  )
+                a negative MATCH_xxx value for PRUNE, SKIP, etc
                 a negative PCRE_ERROR_xxx value if aborted by an error condition
                   (e.g. stopped by repeated call or recursion limit)
  */
  static int
- match(REGISTER USPTR eptr, REGISTER const uschar *ecode, const uschar *mstart,
+ match(REGISTER USPTR eptr, REGISTER const uschar *ecode, USPTR mstart,
-   int offset_top, match_data *md, unsigned long int ims, eptrblock *eptrb,
+   const uschar *markptr, int offset_top, match_data *md, 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 405 
 register unsigned int c;   /* Character
+Line 490 
 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
  preserved over calls to RMATCH() are part of a "frame" which is obtained from
-Line 412 
 heap storage. Set up the top-level frame
+Line 499 
 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 420 
 frame->Xprevframe = NULL;            /*
+Line 508 
 frame->Xprevframe = NULL;            /*
  frame->Xeptr = eptr;
  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 435 
 HEAP_RECURSE:
+Line 522 
 HEAP_RECURSE:
  #define eptr               frame->Xeptr
  #define ecode              frame->Xecode
  #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 447 
 HEAP_RECURSE:
+Line 533 
 HEAP_RECURSE:
  #define charptr            frame->Xcharptr
  #endif
  #define callpat            frame->Xcallpat
+ #define codelink           frame->Xcodelink
  #define data               frame->Xdata
  #define next               frame->Xnext
  #define pp                 frame->Xpp
-Line 459 
 HEAP_RECURSE:
+Line 546 
 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
-Line 497 
 i, and fc and c, can be the same variabl
+Line 582 
 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.
- #ifdef SUPPORT_UTF8                /* Many of these variables are used only  */
+ My normal style of coding would have declared them within each of those blocks.
- const uschar *charptr;             /* in small blocks of the code. My normal */
+ However, in order to accommodate the version of this code that uses an external
- #endif                             /* style of coding would have declared    */
+ "stack" implemented on the heap, it is easier to declare them all here, so the
- const uschar *callpat;             /* them within each of those blocks.      */
+ declarations can be cut out in a block. The only declarations within blocks
- const uschar *data;                /* However, in order to accommodate the   */
+ below are for variables that do not have to be preserved over a recursive call
- const uschar *next;                /* version of this code that uses an      */
+ to RMATCH(). */
- USPTR         pp;                  /* external "stack" implemented on the    */
- const uschar *prev;                /* heap, it is easier to declare them all */
+ #ifdef SUPPORT_UTF8
- USPTR         saved_eptr;          /* here, so the declarations can be cut   */
+ const uschar *charptr;
-                                    /* out in a block. The only declarations  */
+ #endif
- recursion_info new_recursive;      /* within blocks below are for variables  */
+ const uschar *callpat;
-                                    /* that do not have to be preserved over  */
+ const uschar *data;
- BOOL cur_is_word;                  /* a recursive call to RMATCH().          */
+ const uschar *next;
+ USPTR         pp;
+ const uschar *prev;
+ USPTR         saved_eptr;
+ recursion_info new_recursive;
+ BOOL cur_is_word;
  BOOL condition;
  BOOL prev_is_word;
- unsigned long int original_ims;
  #ifdef SUPPORT_UCP
  int prop_type;
  int prop_value;
-Line 527 
 int oclength;
+Line 617 
 int oclength;
  uschar occhars[8];
  #endif
+ int codelink;
  int ctype;
  int length;
  int max;
-Line 541 
 int stacksave[REC_STACK_SAVE_MAX];
+Line 632 
 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 560 
 TAIL_RECURSE:
+Line 662 
 TAIL_RECURSE:
  /* OK, now we can get on with the real code of the function. Recursive calls
  are specified by the macro RMATCH and RRETURN is used to return. When
  NO_RECURSE is *not* defined, these just turn into a recursive call to match()
- and a "return", respectively (possibly with some debugging if DEBUG is
+ and a "return", respectively (possibly with some debugging if PCRE_DEBUG is
  defined). However, RMATCH isn't like a function call because it's quite a
  complicated macro. It has to be used in one particular way. This shouldn't,
  however, impact performance when true recursion is being used. */
-Line 577 
 haven't exceeded the recursive call limi
+Line 679 
 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 601 
 for (;;)
+Line 705 
 for (;;)
    {
    minimize = possessive = FALSE;
    op = *ecode;
-   /* For partial matching, remember if we ever hit the end of the subject after
-   matching at least one subject character. */
-   if (md->partial &&
-       eptr >= md->end_subject &&
-       eptr > mstart)
-     md->hitend = TRUE;
    switch(op)
      {
-     /* Handle a capturing bracket. If there is space in the offset vector, save
+     case OP_MARK:
-     the current subject position in the working slot at the top of the vector.
+     markptr = ecode + 2;
-     We mustn't change the current values of the data slot, because they may be
+     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode] + ecode[1], offset_top, md,
-     set from a previous iteration of this group, and be referred to by a
+       eptrb, RM55);
-     reference inside the group.
+     /* A return of MATCH_SKIP_ARG means that matching failed at SKIP with an
-     If the bracket fails to match, we need to restore this value and also the
+     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,
+       eptrb, RM52);
+     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,
+       eptrb, RM51);
+     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,
+       eptrb, RM56);
+     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,
+       eptrb, RM53);
+     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,
+       eptrb, RM57);
+     if (rrc != MATCH_NOMATCH && rrc != MATCH_PRUNE && rrc != MATCH_THEN)
+       RRETURN(rrc);
+     /* Pass back the current skip name by overloading md->start_match_ptr and
+     returning the special MATCH_SKIP_ARG return code. This will either be
+     caught by a matching MARK, or get to the top, where it 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 bracket or
+     the alt that is at the start of the current branch. This makes it possible
+     to skip back past alternatives that precede the THEN within the current
+     branch. */
+     case OP_THEN:
+     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
+       eptrb, RM54);
+     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     md->start_match_ptr = ecode - GET(ecode, 1);
+     MRRETURN(MATCH_THEN);
+     case OP_THEN_ARG:
+     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode] + ecode[1+LINK_SIZE],
+       offset_top, md, eptrb, RM58);
+     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     md->start_match_ptr = ecode - GET(ecode, 1);
+     md->mark = ecode + LINK_SIZE + 2;
+     RRETURN(MATCH_THEN);
+     /* Handle a capturing bracket, other than those that are possessive with an
+     unlimited repeat. If there is space in the offset vector, save the current
+     subject position in the working slot at the top of the vector. We mustn't
+     change the current values of the data slot, because they may be set from a
+     previous iteration of this group, and be referred to by a reference inside
+     the group. If we fail to match, we need to restore this value and also the
      values of the final offsets, in case they were set by a previous iteration
      of the same bracket.
-Line 630 
 for (;;)
+Line 820 
 for (;;)
      case OP_SCBRA:
      number = GET2(ecode, 1+LINK_SIZE);
      offset = number << 1;
- #ifdef DEBUG
+ #ifdef PCRE_DEBUG
      printf("start bracket %d\n", number);
      printf("subject=");
      pchars(eptr, 16, TRUE, md);
-Line 646 
 for (;;)
+Line 836 
 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
          {
-         RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
+         if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
-           ims, eptrb, flags, RM1);
+         RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+           eptrb, RM1);
+         if (rrc != MATCH_NOMATCH &&
+             (rrc != MATCH_THEN || md->start_match_ptr != ecode))
+           RRETURN(rrc);
+         /* If md->end_offset_top is greater than offset_top, it means that the
+         branch we have just failed to match did manage to match some capturing
+         parentheses within an atomic group or an assertion. Although offset_top
+         reverts to its original value at this level, we must unset the captured
+         values in case a later match sets a higher capturing number. Example:
+         matching /((?>(a))b|(a)c)/ against "ac". This captures 3, but we need
+         to ensure that 2 - which was captured in the atomic matching - is
+         unset. */
+         if (md->end_offset_top > offset_top)
+           {
+           register int *iptr = md->offset_vector + offset_top;
+           register int *iend = md->offset_vector + md->end_offset_top;
+           while (iptr < iend) *iptr++ = -1;
+           }
          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));
-Line 665 
 for (;;)
+Line 875 
 for (;;)
        md->offset_vector[offset+1] = save_offset2;
        md->offset_vector[md->offset_end - number] = save_offset3;
+       if (rrc != MATCH_THEN && md->mark == NULL) md->mark = markptr;
        RRETURN(MATCH_NOMATCH);
        }
-Line 679 
 for (;;)
+Line 890 
 for (;;)
      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-     /* Non-capturing bracket. Loop for all the alternatives. When we get to the
+     /* Non-capturing bracket, except for possessive with unlimited repeat. Loop
-     final alternative within the brackets, we would return the result of a
+     for all the alternatives. When we get to the final alternative within the
-     recursive call to match() whatever happened. We can reduce stack usage by
+     brackets, we used to return the result of a recursive call to match()
-     turning this into a tail recursion, except in the case when match_cbegroup
+     whatever happened so it was possible to reduce stack usage by turning this
-     is set.*/
+     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 possible. */
      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) md->match_function_type = MATCH_CBEGROUP;
-         {
+       RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md, eptrb,
-         if (flags == 0)    /* Not a possibly empty group */
+         RM2);
+       if (rrc != MATCH_NOMATCH &&
+           (rrc != MATCH_THEN || md->start_match_ptr != ecode))
+         RRETURN(rrc);
+       /* See explanatory comment above under OP_CBRA. */
+       if (md->end_offset_top > offset_top)
+         {
+         register int *iptr = md->offset_vector + offset_top;
+         register int *iend = md->offset_vector + md->end_offset_top;
+         while (iptr < iend) *iptr++ = -1;
+         }
+       ecode += GET(ecode, 1);
+       if (*ecode != OP_ALT) break;
+       }
+     if (rrc != MATCH_THEN && md->mark == NULL) md->mark = markptr;
+     RRETURN(MATCH_NOMATCH);
+     /* 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;
+     POSSESSIVE_CAPTURE:
+     number = GET2(ecode, 1+LINK_SIZE);
+     offset = number << 1;
+ #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)
            {
-           ecode += _pcre_OP_lengths[*ecode];
+           offset_top = md->end_offset_top;
-           DPRINTF(("bracket 0 tail recursion\n"));
+           eptr = md->end_match_ptr;
-           goto TAIL_RECURSE;
+           ecode = md->start_code + code_offset;
+           save_capture_last = md->capture_last;
+           matched_once = TRUE;
+           continue;
+           }
+         if (rrc != MATCH_NOMATCH &&
+             (rrc != MATCH_THEN || md->start_match_ptr != ecode))
+           RRETURN(rrc);
+         /* See explanatory comment above under OP_CBRA. */
+         if (md->end_offset_top > offset_top)
+           {
+           register int *iptr = md->offset_vector + offset_top;
+           register int *iend = md->offset_vector + md->end_offset_top;
+           while (iptr < iend) *iptr++ = -1;
            }
-         /* Possibly empty group; can't use tail recursion. */
+         md->capture_last = save_capture_last;
+         ecode += GET(ecode, 1);
+         if (*ecode != OP_ALT) break;
+         }
+       if (!matched_once)
+         {
+         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 == NULL) md->mark = markptr;
+       if (allow_zero || matched_once)
+         {
+         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 */
-         RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md, ims,
+     /* Non-capturing possessive bracket with unlimited repeat. We come here
-           eptrb, flags, RM48);
+     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;
+         }
+       if (rrc != MATCH_NOMATCH &&
+           (rrc != MATCH_THEN || md->start_match_ptr != ecode))
          RRETURN(rrc);
+       /* See explanatory comment above under OP_CBRA. */
+       if (md->end_offset_top > offset_top)
+         {
+         register int *iptr = md->offset_vector + offset_top;
+         register int *iend = md->offset_vector + md->end_offset_top;
+         while (iptr < iend) *iptr++ = -1;
          }
-       /* For non-final alternatives, continue the loop for a NOMATCH result;
-       otherwise return. */
-       RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md, ims,
-         eptrb, flags, RM2);
-       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:
-     if (ecode[LINK_SIZE+1] == OP_RREF)         /* Recursion test */
+     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. */
+     if (ecode[LINK_SIZE+1] == OP_CALLOUT)
        {
-       offset = GET2(ecode, LINK_SIZE + 2);     /* Recursion group number*/
+       if (pcre_callout != NULL)
-       condition = md->recursive != NULL &&
+         {
-         (offset == RREF_ANY || offset == md->recursive->group_num);
+         pcre_callout_block cb;
-       ecode += condition? 3 : GET(ecode, 1);
+         cb.version          = 1;   /* 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   = (int)(md->end_subject - md->start_subject);
+         cb.start_match      = (int)(mstart - md->start_subject);
+         cb.current_position = (int)(eptr - md->start_subject);
+         cb.pattern_position = GET(ecode, LINK_SIZE + 3);
+         cb.next_item_length = GET(ecode, 3 + 2*LINK_SIZE);
+         cb.capture_top      = offset_top/2;
+         cb.capture_last     = md->capture_last;
+         cb.callout_data     = md->callout_data;
+         if ((rrc = (*pcre_callout)(&cb)) > 0) MRRETURN(MATCH_NOMATCH);
+         if (rrc < 0) RRETURN(rrc);
+         }
+       ecode += _pcre_OP_lengths[OP_CALLOUT];
+       }
+     condcode = ecode[LINK_SIZE+1];
+     /* Now see what the actual condition is */
+     if (condcode == OP_RREF || condcode == OP_NRREF)    /* Recursion test */
+       {
+       if (md->recursive == NULL)                /* Not recursing => FALSE */
+         {
+         condition = FALSE;
+         ecode += GET(ecode, 1);
+         }
+       else
+         {
+         int recno = GET2(ecode, LINK_SIZE + 2);   /* Recursion group number*/
+         condition =  (recno == RREF_ANY || recno == md->recursive->group_num);
+         /* If the test is for recursion into a specific subpattern, and it is
+         false, but the test was set up by name, scan the table to see if the
+         name refers to any other numbers, and test them. The condition is true
+         if any one is set. */
+         if (!condition && condcode == OP_NRREF && recno != RREF_ANY)
+           {
+           uschar *slotA = md->name_table;
+           for (i = 0; i < md->name_count; i++)
+             {
+             if (GET2(slotA, 0) == recno) break;
+             slotA += md->name_entry_size;
+             }
+           /* Found a name for the number - there can be only one; duplicate
+           names for different numbers are allowed, but not vice versa. First
+           scan down for duplicates. */
+           if (i < md->name_count)
+             {
+             uschar *slotB = slotA;
+             while (slotB > md->name_table)
+               {
+               slotB -= md->name_entry_size;
+               if (strcmp((char *)slotA + 2, (char *)slotB + 2) == 0)
+                 {
+                 condition = GET2(slotB, 0) == md->recursive->group_num;
+                 if (condition) break;
+                 }
+               else break;
+               }
+             /* Scan up for duplicates */
+             if (!condition)
+               {
+               slotB = slotA;
+               for (i++; i < md->name_count; i++)
+                 {
+                 slotB += md->name_entry_size;
+                 if (strcmp((char *)slotA + 2, (char *)slotB + 2) == 0)
+                   {
+                   condition = GET2(slotB, 0) == md->recursive->group_num;
+                   if (condition) break;
+                   }
+                 else break;
+                 }
+               }
+             }
+           }
+         /* Chose branch according to the condition */
+         ecode += condition? 3 : GET(ecode, 1);
+         }
        }
-     else if (ecode[LINK_SIZE+1] == OP_CREF)    /* Group used test */
+     else if (condcode == OP_CREF || condcode == OP_NCREF)  /* Group used test */
        {
        offset = GET2(ecode, LINK_SIZE+2) << 1;  /* Doubled ref number */
        condition = offset < offset_top && md->offset_vector[offset] >= 0;
+       /* If the numbered capture is unset, but the reference was by name,
+       scan the table to see if the name refers to any other numbers, and test
+       them. The condition is true if any one is set. This is tediously similar
+       to the code above, but not close enough to try to amalgamate. */
+       if (!condition && condcode == OP_NCREF)
+         {
+         int refno = offset >> 1;
+         uschar *slotA = md->name_table;
+         for (i = 0; i < md->name_count; i++)
+           {
+           if (GET2(slotA, 0) == refno) break;
+           slotA += md->name_entry_size;
+           }
+         /* Found a name for the number - there can be only one; duplicate names
+         for different numbers are allowed, but not vice versa. First scan down
+         for duplicates. */
+         if (i < md->name_count)
+           {
+           uschar *slotB = slotA;
+           while (slotB > md->name_table)
+             {
+             slotB -= md->name_entry_size;
+             if (strcmp((char *)slotA + 2, (char *)slotB + 2) == 0)
+               {
+               offset = GET2(slotB, 0) << 1;
+               condition = offset < offset_top &&
+                 md->offset_vector[offset] >= 0;
+               if (condition) break;
+               }
+             else break;
+             }
+           /* Scan up for duplicates */
+           if (!condition)
+             {
+             slotB = slotA;
+             for (i++; i < md->name_count; i++)
+               {
+               slotB += md->name_entry_size;
+               if (strcmp((char *)slotA + 2, (char *)slotB + 2) == 0)
+                 {
+                 offset = GET2(slotB, 0) << 1;
+                 condition = offset < offset_top &&
+                   md->offset_vector[offset] >= 0;
+                 if (condition) break;
+                 }
+               else break;
+               }
+             }
+           }
+         }
+       /* Chose branch according to the condition */
        ecode += condition? 3 : GET(ecode, 1);
        }
-     else if (ecode[LINK_SIZE+1] == OP_DEF)     /* DEFINE - always false */
+     else if (condcode == OP_DEF)     /* DEFINE - always false */
        {
        condition = FALSE;
        ecode += GET(ecode, 1);
        }
      /* The condition is an assertion. Call match() to evaluate it - setting
-     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)
          {
          condition = TRUE;
          ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE + 2);
          while (*ecode == OP_ALT) ecode += GET(ecode, 1);
          }
-       else if (rrc != MATCH_NOMATCH)
+       else if (rrc != MATCH_NOMATCH &&
+               (rrc != MATCH_THEN || md->start_match_ptr != ecode))
          {
          RRETURN(rrc);         /* Need braces because of following else */
          }
        else
          {
          condition = FALSE;
-         ecode += GET(ecode, 1);
+         ecode += codelink;
          }
        }
      /* We are now at the branch that is to be obeyed. As there is only one,
-     we can use tail recursion to avoid using another stack frame, except when
+     we used to use tail recursion to avoid using another stack frame, except
-     match_cbegroup is required for an unlimited repeat of a possibly empty
+     when there was unlimited repeat of a possibly empty group. However, that
-     group. If the second alternative doesn't exist, we can just plough on. */
+     strategy no longer works because of the possibilty of (*THEN) being
+     encountered in the branch. A recursive call to match() is always required,
+     unless the second alternative doesn't exist, in which case we can just
+     plough on. */
      if (condition || *ecode == OP_ALT)
        {
-       ecode += 1 + LINK_SIZE;
+       if (op == OP_SCOND) md->match_function_type = MATCH_CBEGROUP;
-       if (op == OP_SCOND)        /* Possibly empty group */
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM49);
-         {
+       if (rrc == MATCH_THEN && md->start_match_ptr == ecode)
-         RMATCH(eptr, ecode, offset_top, md, ims, eptrb, match_cbegroup, RM49);
+         rrc = MATCH_NOMATCH;
-         RRETURN(rrc);
+       RRETURN(rrc);
-         }
-       else                       /* Group must match something */
-         {
-         flags = 0;
-         goto TAIL_RECURSE;
-         }
        }
-     else                         /* Condition false & no 2nd alternative */
+     else                         /* Condition false & no alternative */
        {
        ecode += 1 + LINK_SIZE;
        }
      break;
-     /* End of the pattern. If we are in a top-level recursion, we should
+     /* Before OP_ACCEPT there may be any number of OP_CLOSE opcodes,
-     restore the offsets appropriately and continue from after the call. */
+     to close any currently open capturing brackets. */
+     case OP_CLOSE:
+     number = GET2(ecode, 1);
+     offset = number << 1;
+ #ifdef PCRE_DEBUG
+       printf("end bracket %d at *ACCEPT", number);
+       printf("\n");
+ #endif
+     md->capture_last = number;
+     if (offset >= md->offset_max) md->offset_overflow = TRUE; else
+       {
+       md->offset_vector[offset] =
+         md->offset_vector[md->offset_end - number];
+       md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
+       if (offset_top <= offset) offset_top = offset + 2;
+       }
+     ecode += 3;
+     break;
+     /* End of the pattern, either real or forced. If we are in a recursion, we
+     should restore the offsets appropriately, and if it's a top-level
+     recursion, continue from after the call. */
+     case OP_ACCEPT:
+     case OP_ASSERT_ACCEPT:
      case OP_END:
-     if (md->recursive != NULL && md->recursive->group_num == 0)
+     if (md->recursive != NULL)
        {
        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));
-       mstart = rec->save_start;
+       offset_top = rec->save_offset_top;
-       ims = original_ims;
+       if (rec->group_num == 0)
-       ecode = rec->after_call;
+         {
-       break;
+         ecode = rec->after_call;
+         break;
+         }
        }
-     /* Otherwise, if PCRE_NOTEMPTY is set, fail if we have matched an empty
+     /* Otherwise, if we have matched an empty string, fail if not in an
-     string - backtracking will then try other alternatives, if any. */
+     assertion and if either PCRE_NOTEMPTY is set, or if PCRE_NOTEMPTY_ATSTART
+     is set and we have matched at the start of the subject. In both cases,
+     backtracking will then try other alternatives, if any. */
+     else if (eptr == mstart && op != OP_ASSERT_ACCEPT &&
+         (md->notempty ||
+           (md->notempty_atstart &&
+             mstart == md->start_subject + md->start_offset)))
+       MRRETURN(MATCH_NOMATCH);
-     if (md->notempty && eptr == mstart) RRETURN(MATCH_NOMATCH);
+     /* Otherwise, we have a match. */
      md->end_match_ptr = eptr;           /* Record where we ended */
      md->end_offset_top = offset_top;    /* and how many extracts were taken */
-     md->start_match_ptr = mstart;  /* and the start (\K can modify) */
+     md->start_match_ptr = mstart;       /* and the start (\K can modify) */
-     RRETURN(MATCH_MATCH);
-     /* 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)
-       if (rrc == MATCH_MATCH) break;
+         {
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         mstart = md->start_match_ptr;   /* In case \K reset it */
+         break;
+         }
+       if (rrc != MATCH_NOMATCH &&
+           (rrc != MATCH_THEN || md->start_match_ptr != ecode))
+         RRETURN(rrc);
+       /* See explanatory comment above under OP_CBRA. */
+       if (md->end_offset_top > offset_top)
+         {
+         register int *iptr = md->offset_vector + offset_top;
+         register int *iend = md->offset_vector + md->end_offset_top;
+         while (iptr < iend) *iptr++ = -1;
+         }
        ecode += GET(ecode, 1);
        }
      while (*ecode == OP_ALT);
-     if (*ecode == OP_KET) RRETURN(MATCH_NOMATCH);
+     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 862 
 for (;;)
+Line 1441 
 for (;;)
      offset_top = md->end_offset_top;
      continue;
-     /* Negative assertion: all branches must fail to match */
+     /* Negative assertion: all branches must fail to match. Encountering SKIP,
+     PRUNE, or COMMIT means we must assume failure without checking subsequent
+     branches. */
      case OP_ASSERT_NOT:
      case OP_ASSERTBACK_NOT:
+     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_MATCH) RRETURN(MATCH_NOMATCH);
+       if (rrc == MATCH_SKIP || rrc == MATCH_PRUNE || rrc == MATCH_COMMIT)
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         {
+         do ecode += GET(ecode,1); while (*ecode == OP_ALT);
+         break;
+         }
+       if (rrc != MATCH_NOMATCH &&
+           (rrc != MATCH_THEN || md->start_match_ptr != ecode))
+         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 894 
 for (;;)
+Line 1488 
 for (;;)
        while (i-- > 0)
          {
          eptr--;
-         if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
+         if (eptr < md->start_subject) MRRETURN(MATCH_NOMATCH);
-         BACKCHAR(eptr)
+         BACKCHAR(eptr);
          }
        }
      else
-Line 905 
 for (;;)
+Line 1499 
 for (;;)
        {
        eptr -= GET(ecode, 1);
-       if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
+       if (eptr < md->start_subject) MRRETURN(MATCH_NOMATCH);
        }
-     /* Skip to next op code */
+     /* Save the earliest consulted character, then skip to next op code */
+     if (eptr < md->start_used_ptr) md->start_used_ptr = eptr;
      ecode += 1 + LINK_SIZE;
      break;
-Line 925 
 for (;;)
+Line 1520 
 for (;;)
        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;
-       if ((rrc = (*pcre_callout)(&cb)) > 0) RRETURN(MATCH_NOMATCH);
+       if ((rrc = (*pcre_callout)(&cb)) > 0) MRRETURN(MATCH_NOMATCH);
        if (rrc < 0) RRETURN(rrc);
        }
      ecode += 2 + 2*LINK_SIZE;
-Line 988 
 for (;;)
+Line 1583 
 for (;;)
        memcpy(new_recursive.offset_save, md->offset_vector,
              new_recursive.saved_max * sizeof(int));
-       new_recursive.save_start = mstart;
+       new_recursive.save_offset_top = offset_top;
-       mstart = eptr;
        /* OK, now we can do the recursion. For each top-level alternative we
        restore the offset and recursion data. */
        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);
-         if (rrc == MATCH_MATCH)
+         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);
-           RRETURN(MATCH_MATCH);
+           MRRETURN(MATCH_MATCH);
            }
-         else if (rrc != MATCH_NOMATCH)
+         else if (rrc != MATCH_NOMATCH &&
+                 (rrc != MATCH_THEN || md->start_match_ptr != ecode))
            {
            DPRINTF(("Recursion gave error %d\n", rrc));
+           if (new_recursive.offset_save != stacksave)
+             (pcre_free)(new_recursive.offset_save);
            RRETURN(rrc);
            }
-Line 1025 
 for (;;)
+Line 1623 
 for (;;)
        md->recursive = new_recursive.prevrec;
        if (new_recursive.offset_save != stacksave)
          (pcre_free)(new_recursive.offset_save);
-       RRETURN(MATCH_NOMATCH);
+       MRRETURN(MATCH_NOMATCH);
        }
      /* Control never reaches here */
-Line 1034 
 for (;;)
+Line 1632 
 for (;;)
      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. */
+     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;
-Line 1042 
 for (;;)
+Line 1641 
 for (;;)
      do
        {
-       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb, 0, RM7);
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM7);
-       if (rrc == MATCH_MATCH) break;
+       if (rrc == MATCH_MATCH)  /* Note: _not_ MATCH_ACCEPT */
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         {
+         mstart = md->start_match_ptr;
+         break;
+         }
+       if (rrc != MATCH_NOMATCH &&
+           (rrc != MATCH_THEN || md->start_match_ptr != ecode))
+         RRETURN(rrc);
+       /* See explanatory comment above under OP_CBRA. */
+       if (md->end_offset_top > offset_top)
+         {
+         register int *iptr = md->offset_vector + offset_top;
+         register int *iend = md->offset_vector + md->end_offset_top;
+         while (iptr < iend) *iptr++ = -1;
+         }
        ecode += GET(ecode,1);
        }
      while (*ecode == OP_ALT);
-Line 1053 
 for (;;)
+Line 1668 
 for (;;)
      if (*ecode != OP_ONCE && *ecode != OP_ALT) RRETURN(MATCH_NOMATCH);
-     /* Continue as from after the assertion, updating the offsets high water
+     /* Continue after the group, updating the offsets high water mark, since
-     mark, since extracts may have been taken. */
+     extracts may have been taken. */
      do ecode += GET(ecode, 1); while (*ecode == OP_ALT);
-Line 1075 
 for (;;)
+Line 1690 
 for (;;)
      /* The repeating kets try the rest of the pattern or restart from the
      preceding bracket, in the appropriate order. The second "call" of match()
-     uses tail recursion, to avoid using another stack frame. We need to reset
+     uses tail recursion, to avoid using another stack frame. */
-     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);
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, 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);
+       md->match_function_type = MATCH_CBEGROUP;
+       RMATCH(eptr, prev, offset_top, md, eptrb, RM9);
        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
        ecode += 1 + LINK_SIZE;
-       flags = 0;
        goto TAIL_RECURSE;
        }
      /* Control never gets here */
-Line 1110 
 for (;;)
+Line 1716 
 for (;;)
      do ecode += GET(ecode,1); while (*ecode == OP_ALT);
      break;
-     /* BRAZERO and BRAMINZERO occur just before a bracket group, indicating
+     /* BRAZERO, BRAMINZERO and SKIPZERO occur just before a bracket group,
-     that it may occur zero times. It may repeat infinitely, or not at all -
+     indicating that it may occur zero times. It may repeat infinitely, or not
-     i.e. it could be ()* or ()? in the pattern. Brackets with fixed upper
+     at all - i.e. it could be ()* or ()? or even (){0} in the pattern. Brackets
-     repeat limits are compiled as a number of copies, with the optional ones
+     with fixed upper repeat limits are compiled as a number of copies, with the
-     preceded by BRAZERO or BRAMINZERO. */
+     optional ones preceded by BRAZERO or BRAMINZERO. */
      case OP_BRAZERO:
-       {
+     next = ecode + 1;
-       next = ecode+1;
+     RMATCH(eptr, next, offset_top, md, eptrb, RM10);
-       RMATCH(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;
+     do next += GET(next,1); while (*next == OP_ALT);
+     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
-Line 1154 
 for (;;)
+Line 1772 
 for (;;)
        }
      else saved_eptr = NULL;
-     /* If we are at the end of an assertion group, stop matching and return
+     /* If we are at the end of an assertion group or an atomic group, stop
-     MATCH_MATCH, but record the current high water mark for use by positive
+     matching and return MATCH_MATCH, but record the current high water mark for
-     assertions. Do this also for the "once" (atomic) groups. */
+     use by positive assertions. We also need to record the match start in case
+     it was changed by \K. */
      if (*prev == OP_ASSERT || *prev == OP_ASSERT_NOT ||
          *prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT ||
-Line 1164 
 for (;;)
+Line 1783 
 for (;;)
        {
        md->end_match_ptr = eptr;      /* For ONCE */
        md->end_offset_top = offset_top;
-       RRETURN(MATCH_MATCH);
+       md->start_match_ptr = mstart;
+       MRRETURN(MATCH_MATCH);
        }
      /* For capturing groups we have to check the group number back at the start
-Line 1173 
 for (;;)
+Line 1793 
 for (;;)
      a recurse into group 0, so it won't be picked up here. Instead, we catch it
      when the OP_END is reached. Other recursion is handled here. */
-     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;
- #ifdef DEBUG
+ #ifdef PCRE_DEBUG
        printf("end bracket %d", number);
        printf("\n");
  #endif
-Line 1188 
 for (;;)
+Line 1809 
 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;
          }
-Line 1200 
 for (;;)
+Line 1821 
 for (;;)
          recursion_info *rec = md->recursive;
          DPRINTF(("Recursion (%d) succeeded - continuing\n", number));
          md->recursive = rec->prevrec;
-         mstart = rec->save_start;
          memcpy(md->offset_vector, rec->offset_save,
            rec->saved_max * sizeof(int));
+         offset_top = rec->save_offset_top;
          ecode = rec->after_call;
-         ims = original_ims;
          break;
          }
        }
-     /* For both capturing and non-capturing groups, reset the value of the ims
-     flags, in case they got changed during the group. */
-     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
-Line 1226 
 for (;;)
+Line 1840 
 for (;;)
        ecode += 1 + LINK_SIZE;
        break;
        }
-     /* The repeating kets try the rest of the pattern or restart from the
+     /* OP_KETRPOS is a possessive repeating ket. Remember the current position,
-     preceding bracket, in the appropriate order. In the second case, we can use
+     and return the MATCH_KETRPOS. This makes it possible to do the repeats one
-     tail recursion to avoid using another stack frame, unless we have an
+     at a time from the outer level, thus saving stack. */
+     if (*ecode == OP_KETRPOS)
+       {
+       md->end_match_ptr = eptr;
+       md->end_offset_top = offset_top;
+       RRETURN(MATCH_KETRPOS);
+       }
+     /* The normal repeating kets try the rest of the pattern or restart from
+     the preceding bracket, in the appropriate order. In the second case, we can
+     use tail recursion to avoid using another stack frame, unless we have an
      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, RM12);
        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-       if (flags != 0)    /* Could match an empty string */
+       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 1248 
 for (;;)
+Line 1872 
 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_NOMATCH) RRETURN(rrc);
        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) RRETURN(MATCH_NOMATCH);
+     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)))
-         RRETURN(MATCH_NOMATCH);
-       ecode++;
-       break;
-       }
-     /* ... else fall through */
      /* Start of subject assertion */
      case OP_SOD:
-     if (eptr != md->start_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr != md->start_subject) MRRETURN(MATCH_NOMATCH);
+     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:
-     if (eptr != md->start_subject + md->start_offset) RRETURN(MATCH_NOMATCH);
+     if (eptr != md->start_subject + md->start_offset) MRRETURN(MATCH_NOMATCH);
      ecode++;
      break;
-Line 1291 
 for (;;)
+Line 1916 
 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); }
+     else
        {
-       if (eptr < md->end_subject)
+       if (md->noteol) MRRETURN(MATCH_NOMATCH);
-         { if (!IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH); }
+       SCHECK_PARTIAL();
-       else
-         { if (md->noteol) RRETURN(MATCH_NOMATCH); }
-       ecode++;
-       break;
-       }
-     else
-       {
-       if (md->noteol) RRETURN(MATCH_NOMATCH);
-       if (!md->endonly)
-         {
-         if (eptr != md->end_subject &&
-             (!IS_NEWLINE(eptr) || eptr != md->end_subject - md->nllen))
-           RRETURN(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) RRETURN(MATCH_NOMATCH);
+     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))
-       RRETURN(MATCH_NOMATCH);
+       MRRETURN(MATCH_NOMATCH);
+     /* Either at end of string or \n before end. */
+     SCHECK_PARTIAL();
      ecode++;
      break;
-Line 1342 
 for (;;)
+Line 1969 
 for (;;)
        /* Find out if the previous and current characters are "word" characters.
        It takes a bit more work in UTF-8 mode. Characters > 255 are assumed to
-       be "non-word" characters. */
+       be "non-word" characters. Remember the earliest consulted character for
+       partial matching. */
  #ifdef SUPPORT_UTF8
        if (utf8)
          {
+         /* Get status of previous character */
          if (eptr == md->start_subject) prev_is_word = FALSE; else
            {
-           const uschar *lastptr = eptr - 1;
+           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;
            }
-         if (eptr >= md->end_subject) cur_is_word = FALSE; else
+         /* Get status of next character */
+         if (eptr >= md->end_subject)
+           {
+           SCHECK_PARTIAL();
+           cur_is_word = FALSE;
+           }
+         else
            {
            GETCHAR(c, eptr);
+ #ifdef SUPPORT_UCP
+           if (md->use_ucp)
+             {
+             if (c == '_') cur_is_word = TRUE; else
+               {
+               int cat = UCD_CATEGORY(c);
+               cur_is_word = (cat == ucp_L || cat == ucp_N);
+               }
+             }
+           else
+ #endif
            cur_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;
            }
          }
        else
  #endif
-       /* More streamlined when not in UTF-8 mode */
+       /* Not in UTF-8 mode, but we may still have PCRE_UCP set, and for
+       consistency with the behaviour of \w we do use it in this case. */
          {
-         prev_is_word = (eptr != md->start_subject) &&
+         /* Get status of previous character */
-           ((md->ctypes[eptr[-1]] & ctype_word) != 0);
-         cur_is_word = (eptr < md->end_subject) &&
+         if (eptr == md->start_subject) prev_is_word = FALSE; else
-           ((md->ctypes[*eptr] & ctype_word) != 0);
+           {
+           if (eptr <= md->start_used_ptr) md->start_used_ptr = eptr - 1;
+ #ifdef SUPPORT_UCP
+           if (md->use_ucp)
+             {
+             c = eptr[-1];
+             if (c == '_') prev_is_word = TRUE; else
+               {
+               int cat = UCD_CATEGORY(c);
+               prev_is_word = (cat == ucp_L || cat == ucp_N);
+               }
+             }
+           else
+ #endif
+           prev_is_word = ((md->ctypes[eptr[-1]] & ctype_word) != 0);
+           }
+         /* Get status of next character */
+         if (eptr >= md->end_subject)
+           {
+           SCHECK_PARTIAL();
+           cur_is_word = FALSE;
+           }
+         else
+ #ifdef SUPPORT_UCP
+         if (md->use_ucp)
+           {
+           c = *eptr;
+           if (c == '_') cur_is_word = TRUE; else
+             {
+             int cat = UCD_CATEGORY(c);
+             cur_is_word = (cat == ucp_L || cat == ucp_N);
+             }
+           }
+         else
+ #endif
+         cur_is_word = ((md->ctypes[*eptr] & ctype_word) != 0);
          }
        /* Now see if the situation is what we want */
        if ((*ecode++ == OP_WORD_BOUNDARY)?
             cur_is_word == prev_is_word : cur_is_word != prev_is_word)
-         RRETURN(MATCH_NOMATCH);
+         MRRETURN(MATCH_NOMATCH);
        }
      break;
      /* Match a single character type; inline for speed */
      case OP_ANY:
-     if ((ims & PCRE_DOTALL) == 0)
+     if (IS_NEWLINE(eptr)) MRRETURN(MATCH_NOMATCH);
+     /* Fall through */
+     case OP_ALLANY:
+     if (eptr++ >= md->end_subject)
        {
-       if (IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH);
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
        }
-     if (eptr++ >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (utf8) while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
-     if (utf8)
-       while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
      ecode++;
      break;
-Line 1397 
 for (;;)
+Line 2099 
 for (;;)
      any byte, even newline, independent of the setting of PCRE_DOTALL. */
      case OP_ANYBYTE:
-     if (eptr++ >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr++ >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      ecode++;
      break;
      case OP_NOT_DIGIT:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
  #ifdef SUPPORT_UTF8
-Line 1410 
 for (;;)
+Line 2120 
 for (;;)
  #endif
         (md->ctypes[c] & ctype_digit) != 0
         )
-       RRETURN(MATCH_NOMATCH);
+       MRRETURN(MATCH_NOMATCH);
      ecode++;
      break;
      case OP_DIGIT:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
  #ifdef SUPPORT_UTF8
-Line 1423 
 for (;;)
+Line 2137 
 for (;;)
  #endif
         (md->ctypes[c] & ctype_digit) == 0
         )
-       RRETURN(MATCH_NOMATCH);
+       MRRETURN(MATCH_NOMATCH);
      ecode++;
      break;
      case OP_NOT_WHITESPACE:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
  #ifdef SUPPORT_UTF8
-Line 1436 
 for (;;)
+Line 2154 
 for (;;)
  #endif
         (md->ctypes[c] & ctype_space) != 0
         )
-       RRETURN(MATCH_NOMATCH);
+       MRRETURN(MATCH_NOMATCH);
      ecode++;
      break;
      case OP_WHITESPACE:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
  #ifdef SUPPORT_UTF8
-Line 1449 
 for (;;)
+Line 2171 
 for (;;)
  #endif
         (md->ctypes[c] & ctype_space) == 0
         )
-       RRETURN(MATCH_NOMATCH);
+       MRRETURN(MATCH_NOMATCH);
      ecode++;
      break;
      case OP_NOT_WORDCHAR:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
  #ifdef SUPPORT_UTF8
-Line 1462 
 for (;;)
+Line 2188 
 for (;;)
  #endif
         (md->ctypes[c] & ctype_word) != 0
         )
-       RRETURN(MATCH_NOMATCH);
+       MRRETURN(MATCH_NOMATCH);
      ecode++;
      break;
      case OP_WORDCHAR:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
  #ifdef SUPPORT_UTF8
-Line 1475 
 for (;;)
+Line 2205 
 for (;;)
  #endif
         (md->ctypes[c] & ctype_word) == 0
         )
-       RRETURN(MATCH_NOMATCH);
+       MRRETURN(MATCH_NOMATCH);
      ecode++;
      break;
      case OP_ANYNL:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      switch(c)
        {
-       default: RRETURN(MATCH_NOMATCH);
+       default: MRRETURN(MATCH_NOMATCH);
        case 0x000d:
        if (eptr < md->end_subject && *eptr == 0x0a) eptr++;
        break;
        case 0x000a:
+       break;
        case 0x000b:
        case 0x000c:
        case 0x0085:
        case 0x2028:
        case 0x2029:
+       if (md->bsr_anycrlf) MRRETURN(MATCH_NOMATCH);
        break;
        }
      ecode++;
      break;
      case OP_NOT_HSPACE:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      switch(c)
        {
-Line 1524 
 for (;;)
+Line 2267 
 for (;;)
        case 0x202f:    /* NARROW NO-BREAK SPACE */
        case 0x205f:    /* MEDIUM MATHEMATICAL SPACE */
        case 0x3000:    /* IDEOGRAPHIC SPACE */
-       RRETURN(MATCH_NOMATCH);
+       MRRETURN(MATCH_NOMATCH);
        }
      ecode++;
      break;
      case OP_HSPACE:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      switch(c)
        {
-       default: RRETURN(MATCH_NOMATCH);
+       default: MRRETURN(MATCH_NOMATCH);
        case 0x09:      /* HT */
        case 0x20:      /* SPACE */
        case 0xa0:      /* NBSP */
-Line 1560 
 for (;;)
+Line 2307 
 for (;;)
      break;
      case OP_NOT_VSPACE:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      switch(c)
        {
-Line 1572 
 for (;;)
+Line 2323 
 for (;;)
        case 0x85:      /* NEL */
        case 0x2028:    /* LINE SEPARATOR */
        case 0x2029:    /* PARAGRAPH SEPARATOR */
-       RRETURN(MATCH_NOMATCH);
+       MRRETURN(MATCH_NOMATCH);
        }
      ecode++;
      break;
      case OP_VSPACE:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      switch(c)
        {
-       default: RRETURN(MATCH_NOMATCH);
+       default: MRRETURN(MATCH_NOMATCH);
        case 0x0a:      /* LF */
        case 0x0b:      /* VT */
        case 0x0c:      /* FF */
-Line 1601 
 for (;;)
+Line 2356 
 for (;;)
      case OP_PROP:
      case OP_NOTPROP:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
        {
-       int chartype, script;
+       const ucd_record *prop = GET_UCD(c);
-       int category = _pcre_ucp_findprop(c, &chartype, &script);
        switch(ecode[1])
          {
          case PT_ANY:
-         if (op == OP_NOTPROP) RRETURN(MATCH_NOMATCH);
+         if (op == OP_NOTPROP) MRRETURN(MATCH_NOMATCH);
          break;
          case PT_LAMP:
-         if ((chartype == ucp_Lu ||
+         if ((prop->chartype == ucp_Lu ||
-              chartype == ucp_Ll ||
+              prop->chartype == ucp_Ll ||
-              chartype == ucp_Lt) == (op == OP_NOTPROP))
+              prop->chartype == ucp_Lt) == (op == OP_NOTPROP))
-           RRETURN(MATCH_NOMATCH);
+           MRRETURN(MATCH_NOMATCH);
-          break;
+         break;
          case PT_GC:
-         if ((ecode[2] != category) == (op == OP_PROP))
+         if ((ecode[2] != _pcre_ucp_gentype[prop->chartype]) == (op == OP_PROP))
-           RRETURN(MATCH_NOMATCH);
+           MRRETURN(MATCH_NOMATCH);
          break;
          case PT_PC:
-         if ((ecode[2] != chartype) == (op == OP_PROP))
+         if ((ecode[2] != prop->chartype) == (op == OP_PROP))
-           RRETURN(MATCH_NOMATCH);
+           MRRETURN(MATCH_NOMATCH);
          break;
          case PT_SC:
-         if ((ecode[2] != script) == (op == OP_PROP))
+         if ((ecode[2] != prop->script) == (op == OP_PROP))
-           RRETURN(MATCH_NOMATCH);
+           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 1647 
 for (;;)
+Line 2437 
 for (;;)
      is in the binary; otherwise a compile-time error occurs. */
      case OP_EXTUNI:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
        {
-       int chartype, script;
+       int category = UCD_CATEGORY(c);
-       int category = _pcre_ucp_findprop(c, &chartype, &script);
+       if (category == ucp_M) MRRETURN(MATCH_NOMATCH);
-       if (category == ucp_M) RRETURN(MATCH_NOMATCH);
        while (eptr < md->end_subject)
          {
          int len = 1;
-Line 1660 
 for (;;)
+Line 2453 
 for (;;)
            {
            GETCHARLEN(c, eptr, len);
            }
-         category = _pcre_ucp_findprop(c, &chartype, &script);
+         category = UCD_CATEGORY(c);
          if (category != ucp_M) break;
          eptr += len;
          }
-Line 1679 
 for (;;)
+Line 2472 
 for (;;)
      loops). */
      case OP_REF:
-       {
+     case OP_REFI:
-       offset = GET2(ecode, 1) << 1;               /* Doubled ref number */
+     caseless = op == OP_REFI;
-       ecode += 3;                                 /* Advance past item */
+     offset = GET2(ecode, 1) << 1;               /* Doubled ref number */
+     ecode += 3;
-       /* If the reference is unset, set the length to be longer than the amount
+     /* If the reference is unset, there are two possibilities:
-       of subject left; this ensures that every attempt at a match fails. We
-       can't just fail here, because of the possibility of quantifiers with zero
-       minima. */
-       length = (offset >= offset_top || md->offset_vector[offset] < 0)?
-         md->end_subject - eptr + 1 :
-         md->offset_vector[offset+1] - md->offset_vector[offset];
-       /* Set up for repetition, or handle the non-repeated case */
+     (a) In the default, Perl-compatible state, set the length negative;
+     this ensures that every attempt at a match fails. We can't just fail
+     here, because of the possibility of quantifiers with zero minima.
-       switch (*ecode)
+     (b) If the JavaScript compatibility flag is set, set the length to zero
-         {
+     so that the back reference matches an empty string.
-         case OP_CRSTAR:
-         case OP_CRMINSTAR:
-         case OP_CRPLUS:
-         case OP_CRMINPLUS:
-         case OP_CRQUERY:
-         case OP_CRMINQUERY:
-         c = *ecode++ - OP_CRSTAR;
-         minimize = (c & 1) != 0;
-         min = rep_min[c];                 /* Pick up values from tables; */
-         max = rep_max[c];                 /* zero for max => infinity */
-         if (max == 0) max = INT_MAX;
-         break;
-         case OP_CRRANGE:
+     Otherwise, set the length to the length of what was matched by the
-         case OP_CRMINRANGE:
+     referenced subpattern. */
-         minimize = (*ecode == OP_CRMINRANGE);
-         min = GET2(ecode, 1);
-         max = GET2(ecode, 3);
-         if (max == 0) max = INT_MAX;
-         ecode += 5;
-         break;
-         default:               /* No repeat follows */
+     if (offset >= offset_top || md->offset_vector[offset] < 0)
-         if (!match_ref(offset, eptr, length, md, ims)) RRETURN(MATCH_NOMATCH);
+       length = (md->jscript_compat)? 0 : -1;
-         eptr += length;
+     else
-         continue;              /* With the main loop */
+       length = md->offset_vector[offset+1] - md->offset_vector[offset];
+     /* Set up for repetition, or handle the non-repeated case */
+     switch (*ecode)
+       {
+       case OP_CRSTAR:
+       case OP_CRMINSTAR:
+       case OP_CRPLUS:
+       case OP_CRMINPLUS:
+       case OP_CRQUERY:
+       case OP_CRMINQUERY:
+       c = *ecode++ - OP_CRSTAR;
+       minimize = (c & 1) != 0;
+       min = rep_min[c];                 /* Pick up values from tables; */
+       max = rep_max[c];                 /* zero for max => infinity */
+       if (max == 0) max = INT_MAX;
+       break;
+       case OP_CRRANGE:
+       case OP_CRMINRANGE:
+       minimize = (*ecode == OP_CRMINRANGE);
+       min = GET2(ecode, 1);
+       max = GET2(ecode, 3);
+       if (max == 0) max = INT_MAX;
+       ecode += 5;
+       break;
+       default:               /* No repeat follows */
+       if ((length = match_ref(offset, eptr, length, md, caseless)) < 0)
+         {
+         CHECK_PARTIAL();
+         MRRETURN(MATCH_NOMATCH);
          }
+       eptr += length;
+       continue;              /* With the main loop */
+       }
-       /* If the length of the reference is zero, just continue with the
+     /* Handle repeated back references. If the length of the reference is
-       main loop. */
+     zero, just continue with the main loop. */
-       if (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)) RRETURN(MATCH_NOMATCH);
+         CHECK_PARTIAL();
-         eptr += length;
+         MRRETURN(MATCH_NOMATCH);
          }
+       eptr += slength;
+       }
-       /* If min = max, continue at the same level without recursion.
+     /* If min = max, continue at the same level without recursion.
-       They are not both allowed to be zero. */
+     They are not both allowed to be zero. */
-       if (min == max) continue;
+     if (min == max) continue;
-       /* If minimizing, keep trying and advancing the pointer */
+     /* If minimizing, keep trying and advancing the pointer */
-       if (minimize)
+     if (minimize)
+       {
+       for (fi = min;; fi++)
          {
-         for (fi = min;; fi++)
+         int slength;
+         RMATCH(eptr, ecode, offset_top, md, eptrb, RM14);
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         if (fi >= max) 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 || !match_ref(offset, eptr, length, md, ims))
-             RRETURN(MATCH_NOMATCH);
-           eptr += length;
            }
-         /* Control never gets here */
+         eptr += slength;
          }
+       /* Control never gets here */
+       }
-       /* If maximizing, find the longest string and work backwards */
+     /* If maximizing, find the longest string and work backwards */
-       else
+     else
+       {
+       pp = eptr;
+       for (i = min; i < max; i++)
          {
-         pp = eptr;
+         int slength;
-         for (i = min; i < max; i++)
+         if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
-           {
-           if (!match_ref(offset, eptr, length, md, ims)) break;
-           eptr += length;
-           }
-         while (eptr >= pp)
            {
-           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM15);
+           CHECK_PARTIAL();
-           if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+           break;
-           eptr -= length;
            }
-         RRETURN(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 */
      /* Match a bit-mapped character class, possibly repeatedly. This op code is
      used when all the characters in the class have values in the range 0-255,
      and either the matching is caseful, or the characters are in the range
-Line 1836 
 for (;;)
+Line 2654 
 for (;;)
          {
          for (i = 1; i <= min; i++)
            {
-           if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             MRRETURN(MATCH_NOMATCH);
+             }
            GETCHARINC(c, eptr);
            if (c > 255)
              {
-             if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
+             if (op == OP_CLASS) MRRETURN(MATCH_NOMATCH);
              }
            else
              {
-             if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
+             if ((data[c/8] & (1 << (c&7))) == 0) MRRETURN(MATCH_NOMATCH);
              }
            }
          }
-Line 1854 
 for (;;)
+Line 2676 
 for (;;)
          {
          for (i = 1; i <= min; i++)
            {
-           if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             MRRETURN(MATCH_NOMATCH);
+             }
            c = *eptr++;
-           if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
+           if ((data[c/8] & (1 << (c&7))) == 0) MRRETURN(MATCH_NOMATCH);
            }
          }
-Line 1876 
 for (;;)
+Line 2702 
 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 || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) MRRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
              GETCHARINC(c, eptr);
              if (c > 255)
                {
-               if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
+               if (op == OP_CLASS) MRRETURN(MATCH_NOMATCH);
                }
              else
                {
-               if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
+               if ((data[c/8] & (1 << (c&7))) == 0) MRRETURN(MATCH_NOMATCH);
                }
              }
            }
-Line 1896 
 for (;;)
+Line 2727 
 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 || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) MRRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
              c = *eptr++;
-             if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
+             if ((data[c/8] & (1 << (c&7))) == 0) MRRETURN(MATCH_NOMATCH);
              }
            }
          /* Control never gets here */
-Line 1919 
 for (;;)
+Line 2755 
 for (;;)
            for (i = min; i < max; i++)
              {
              int len = 1;
-             if (eptr >= md->end_subject) break;
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               break;
+               }
              GETCHARLEN(c, eptr, len);
              if (c > 255)
                {
-Line 1933 
 for (;;)
+Line 2773 
 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 1945 
 for (;;)
+Line 2785 
 for (;;)
            {
            for (i = min; i < max; i++)
              {
-             if (eptr >= md->end_subject) break;
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               break;
+               }
              c = *eptr;
              if ((data[c/8] & (1 << (c&7))) == 0) break;
              eptr++;
              }
            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--;
              }
            }
-         RRETURN(MATCH_NOMATCH);
+         MRRETURN(MATCH_NOMATCH);
          }
        }
      /* Control never gets here */
      /* Match an extended character class. This opcode is encountered only
-     in UTF-8 mode, because that's the only time it is compiled. */
+     when UTF-8 mode mode is supported. Nevertheless, we may not be in UTF-8
+     mode, because Unicode properties are supported in non-UTF-8 mode. */
  #ifdef SUPPORT_UTF8
      case OP_XCLASS:
-Line 2006 
 for (;;)
+Line 2851 
 for (;;)
        for (i = 1; i <= min; i++)
          {
-         if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+         if (eptr >= md->end_subject)
-         GETCHARINC(c, eptr);
+           {
-         if (!_pcre_xclass(c, data)) RRETURN(MATCH_NOMATCH);
+           SCHECK_PARTIAL();
+           MRRETURN(MATCH_NOMATCH);
+           }
+         GETCHARINCTEST(c, eptr);
+         if (!_pcre_xclass(c, data)) MRRETURN(MATCH_NOMATCH);
          }
        /* If max == min we can continue with the main loop without the
-Line 2023 
 for (;;)
+Line 2872 
 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 || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+           if (fi >= max) MRRETURN(MATCH_NOMATCH);
-           GETCHARINC(c, eptr);
+           if (eptr >= md->end_subject)
-           if (!_pcre_xclass(c, data)) RRETURN(MATCH_NOMATCH);
+             {
+             SCHECK_PARTIAL();
+             MRRETURN(MATCH_NOMATCH);
+             }
+           GETCHARINCTEST(c, eptr);
+           if (!_pcre_xclass(c, data)) MRRETURN(MATCH_NOMATCH);
            }
          /* Control never gets here */
          }
-Line 2040 
 for (;;)
+Line 2894 
 for (;;)
          for (i = min; i < max; i++)
            {
            int len = 1;
-           if (eptr >= md->end_subject) break;
+           if (eptr >= md->end_subject)
-           GETCHARLEN(c, eptr, len);
+             {
+             SCHECK_PARTIAL();
+             break;
+             }
+           GETCHARLENTEST(c, eptr, len);
            if (!_pcre_xclass(c, data)) break;
            eptr += len;
            }
          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 */
-           BACKCHAR(eptr)
+           if (utf8) BACKCHAR(eptr);
            }
-         RRETURN(MATCH_NOMATCH);
+         MRRETURN(MATCH_NOMATCH);
          }
        /* Control never gets here */
-Line 2068 
 for (;;)
+Line 2926 
 for (;;)
        length = 1;
        ecode++;
        GETCHARLEN(fc, ecode, length);
-       if (length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
+       if (length > md->end_subject - eptr)
-       while (length-- > 0) if (*ecode++ != *eptr++) RRETURN(MATCH_NOMATCH);
+         {
+         CHECK_PARTIAL();             /* Not SCHECK_PARTIAL() */
+         MRRETURN(MATCH_NOMATCH);
+         }
+       while (length-- > 0) if (*ecode++ != *eptr++) MRRETURN(MATCH_NOMATCH);
        }
      else
  #endif
      /* Non-UTF-8 mode */
        {
-       if (md->end_subject - eptr < 1) RRETURN(MATCH_NOMATCH);
+       if (md->end_subject - eptr < 1)
-       if (ecode[1] != *eptr++) RRETURN(MATCH_NOMATCH);
+         {
+         SCHECK_PARTIAL();            /* This one can use SCHECK_PARTIAL() */
+         MRRETURN(MATCH_NOMATCH);
+         }
+       if (ecode[1] != *eptr++) MRRETURN(MATCH_NOMATCH);
        ecode += 2;
        }
      break;
      /* Match a single character, caselessly */
-     case OP_CHARNC:
+     case OP_CHARI:
  #ifdef SUPPORT_UTF8
      if (utf8)
        {
-Line 2092 
 for (;;)
+Line 2958 
 for (;;)
        ecode++;
        GETCHARLEN(fc, ecode, length);
-       if (length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
+       if (length > md->end_subject - eptr)
+         {
+         CHECK_PARTIAL();             /* Not SCHECK_PARTIAL() */
+         MRRETURN(MATCH_NOMATCH);
+         }
        /* If the pattern character's value is < 128, we have only one byte, and
        can use the fast lookup table. */
        if (fc < 128)
          {
-         if (md->lcc[*ecode++] != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);
+         if (md->lcc[*ecode++] != md->lcc[*eptr++]) MRRETURN(MATCH_NOMATCH);
          }
        /* Otherwise we must pick up the subject character */
-Line 2116 
 for (;;)
+Line 2986 
 for (;;)
          if (fc != dc)
            {
  #ifdef SUPPORT_UCP
-           if (dc != _pcre_ucp_othercase(fc))
+           if (dc != UCD_OTHERCASE(fc))
  #endif
-             RRETURN(MATCH_NOMATCH);
+             MRRETURN(MATCH_NOMATCH);
            }
          }
        }
-Line 2127 
 for (;;)
+Line 2997 
 for (;;)
      /* Non-UTF-8 mode */
        {
-       if (md->end_subject - eptr < 1) RRETURN(MATCH_NOMATCH);
+       if (md->end_subject - eptr < 1)
-       if (md->lcc[ecode[1]] != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);
+         {
+         SCHECK_PARTIAL();            /* This one can use SCHECK_PARTIAL() */
+         MRRETURN(MATCH_NOMATCH);
+         }
+       if (md->lcc[ecode[1]] != md->lcc[*eptr++]) MRRETURN(MATCH_NOMATCH);
        ecode += 2;
        }
      break;
-Line 2136 
 for (;;)
+Line 3010 
 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 2160 
 for (;;)
+Line 3039 
 for (;;)
      goto REPEATCHAR;
      case OP_POSPLUS:
+     case OP_POSPLUSI:
      possessive = TRUE;
      min = 1;
      max = INT_MAX;
-Line 2167 
 for (;;)
+Line 3047 
 for (;;)
      goto REPEATCHAR;
      case OP_POSQUERY:
+     case OP_POSQUERYI:
      possessive = TRUE;
      min = 0;
      max = 1;
-Line 2174 
 for (;;)
+Line 3055 
 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;
-     /* Common code for all repeated single-character matches. We can give
+     /* Common code for all repeated single-character matches. */
-     up quickly if there are fewer than the minimum number of characters left in
-     the subject. */
      REPEATCHAR:
  #ifdef SUPPORT_UTF8
-Line 2196 
 for (;;)
+Line 3081 
 for (;;)
        length = 1;
        charptr = ecode;
        GETCHARLEN(fc, ecode, length);
-       if (min * length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
        ecode += length;
        /* Handle multibyte character matching specially here. There is
-Line 2206 
 for (;;)
+Line 3090 
 for (;;)
          {
  #ifdef SUPPORT_UCP
          unsigned int othercase;
-         if ((ims & PCRE_CASELESS) != 0 &&
+         if (op >= OP_STARI &&     /* Caseless */
-             (othercase = _pcre_ucp_othercase(fc)) != NOTACHAR)
+             (othercase = UCD_OTHERCASE(fc)) != fc)
            oclength = _pcre_ord2utf8(othercase, occhars);
          else oclength = 0;
  #endif  /* SUPPORT_UCP */
          for (i = 1; i <= min; i++)
            {
-           if (memcmp(eptr, charptr, length) == 0) eptr += length;
+           if (eptr <= md->end_subject - length &&
+             memcmp(eptr, charptr, length) == 0) eptr += length;
  #ifdef SUPPORT_UCP
-           /* Need braces because of following else */
+           else if (oclength > 0 &&
-           else if (oclength == 0) { RRETURN(MATCH_NOMATCH); }
+                    eptr <= md->end_subject - oclength &&
+                    memcmp(eptr, occhars, oclength) == 0) eptr += oclength;
+ #endif  /* SUPPORT_UCP */
            else
              {
-             if (memcmp(eptr, occhars, oclength) != 0) RRETURN(MATCH_NOMATCH);
+             CHECK_PARTIAL();
-             eptr += oclength;
+             MRRETURN(MATCH_NOMATCH);
              }
- #else   /* without SUPPORT_UCP */
-           else { RRETURN(MATCH_NOMATCH); }
- #endif  /* SUPPORT_UCP */
            }
          if (min == max) continue;
-Line 2234 
 for (;;)
+Line 3118 
 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 || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) MRRETURN(MATCH_NOMATCH);
-             if (memcmp(eptr, charptr, length) == 0) eptr += length;
+             if (eptr <= md->end_subject - length &&
+               memcmp(eptr, charptr, length) == 0) eptr += length;
  #ifdef SUPPORT_UCP
-             /* Need braces because of following else */
+             else if (oclength > 0 &&
-             else if (oclength == 0) { RRETURN(MATCH_NOMATCH); }
+                      eptr <= md->end_subject - oclength &&
+                      memcmp(eptr, occhars, oclength) == 0) eptr += oclength;
+ #endif  /* SUPPORT_UCP */
              else
                {
-               if (memcmp(eptr, occhars, oclength) != 0) RRETURN(MATCH_NOMATCH);
+               CHECK_PARTIAL();
-               eptr += oclength;
+               MRRETURN(MATCH_NOMATCH);
                }
- #else   /* without SUPPORT_UCP */
-             else { RRETURN (MATCH_NOMATCH); }
- #endif  /* SUPPORT_UCP */
              }
            /* Control never gets here */
            }
-Line 2258 
 for (;;)
+Line 3142 
 for (;;)
            pp = eptr;
            for (i = min; i < max; i++)
              {
-             if (eptr > md->end_subject - length) break;
+             if (eptr <= md->end_subject - length &&
-             if (memcmp(eptr, charptr, length) == 0) eptr += length;
+                 memcmp(eptr, charptr, length) == 0) eptr += length;
  #ifdef SUPPORT_UCP
-             else if (oclength == 0) break;
+             else if (oclength > 0 &&
+                      eptr <= md->end_subject - oclength &&
+                      memcmp(eptr, occhars, oclength) == 0) eptr += oclength;
+ #endif  /* SUPPORT_UCP */
              else
                {
-               if (memcmp(eptr, occhars, oclength) != 0) break;
+               CHECK_PARTIAL();
-               eptr += oclength;
+               break;
                }
- #else   /* without SUPPORT_UCP */
-             else break;
- #endif  /* SUPPORT_UCP */
              }
            if (possessive) continue;
            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 (rrc != MATCH_NOMATCH) RRETURN(rrc);
-            if (eptr == pp) RRETURN(MATCH_NOMATCH);
+             if (eptr == pp) { MRRETURN(MATCH_NOMATCH); }
  #ifdef SUPPORT_UCP
-            eptr--;
+             eptr--;
-            BACKCHAR(eptr);
+             BACKCHAR(eptr);
  #else   /* without SUPPORT_UCP */
-            eptr -= length;
+             eptr -= length;
  #endif  /* SUPPORT_UCP */
-            }
+             }
            }
          /* Control never gets here */
          }
-Line 2297 
 for (;;)
+Line 3182 
 for (;;)
  #endif  /* SUPPORT_UTF8 */
      /* When not in UTF-8 mode, load a single-byte character. */
-       {
-       if (min > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
+     fc = *ecode++;
-       fc = *ecode++;
-       }
      /* The value of fc at this point is always less than 256, though we may or
      may not be in UTF-8 mode. The code is duplicated for the caseless and
-Line 2314 
 for (;;)
+Line 3197 
 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++)
-         if (fc != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);
+         {
+         if (eptr >= md->end_subject)
+           {
+           SCHECK_PARTIAL();
+           MRRETURN(MATCH_NOMATCH);
+           }
+         if (fc != md->lcc[*eptr++]) MRRETURN(MATCH_NOMATCH);
+         }
        if (min == max) continue;
        if (minimize)
          {
          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 || eptr >= md->end_subject ||
+           if (fi >= max) MRRETURN(MATCH_NOMATCH);
-               fc != md->lcc[*eptr++])
+           if (eptr >= md->end_subject)
-             RRETURN(MATCH_NOMATCH);
+             {
+             SCHECK_PARTIAL();
+             MRRETURN(MATCH_NOMATCH);
+             }
+           if (fc != md->lcc[*eptr++]) MRRETURN(MATCH_NOMATCH);
            }
          /* Control never gets here */
          }
-Line 2337 
 for (;;)
+Line 3231 
 for (;;)
          pp = eptr;
          for (i = min; i < max; i++)
            {
-           if (eptr >= md->end_subject || fc != md->lcc[*eptr]) break;
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             break;
+             }
+           if (fc != md->lcc[*eptr]) break;
            eptr++;
            }
          if (possessive) continue;
          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);
            }
-         RRETURN(MATCH_NOMATCH);
+         MRRETURN(MATCH_NOMATCH);
          }
        /* Control never gets here */
        }
-Line 2356 
 for (;;)
+Line 3257 
 for (;;)
      else
        {
-       for (i = 1; i <= min; i++) if (fc != *eptr++) RRETURN(MATCH_NOMATCH);
+       for (i = 1; i <= min; i++)
+         {
+         if (eptr >= md->end_subject)
+           {
+           SCHECK_PARTIAL();
+           MRRETURN(MATCH_NOMATCH);
+           }
+         if (fc != *eptr++) MRRETURN(MATCH_NOMATCH);
+         }
        if (min == max) continue;
        if (minimize)
          {
          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 || eptr >= md->end_subject || fc != *eptr++)
+           if (fi >= max) MRRETURN(MATCH_NOMATCH);
-             RRETURN(MATCH_NOMATCH);
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             MRRETURN(MATCH_NOMATCH);
+             }
+           if (fc != *eptr++) MRRETURN(MATCH_NOMATCH);
            }
          /* Control never gets here */
          }
-Line 2374 
 for (;;)
+Line 3290 
 for (;;)
          pp = eptr;
          for (i = min; i < max; i++)
            {
-           if (eptr >= md->end_subject || fc != *eptr) break;
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             break;
+             }
+           if (fc != *eptr) break;
            eptr++;
            }
          if (possessive) continue;
          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);
            }
-         RRETURN(MATCH_NOMATCH);
+         MRRETURN(MATCH_NOMATCH);
          }
        }
      /* Control never gets here */
-Line 2393 
 for (;;)
+Line 3315 
 for (;;)
      checking can be multibyte. */
      case OP_NOT:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     case OP_NOTI:
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       MRRETURN(MATCH_NOMATCH);
+       }
      ecode++;
      GETCHARINCTEST(c, eptr);
-     if ((ims & PCRE_CASELESS) != 0)
+     if (op == OP_NOTI)         /* The caseless case */
        {
  #ifdef SUPPORT_UTF8
        if (c < 256)
  #endif
        c = md->lcc[c];
-       if (md->lcc[*ecode++] == c) RRETURN(MATCH_NOMATCH);
+       if (md->lcc[*ecode++] == c) MRRETURN(MATCH_NOMATCH);
        }
-     else
+     else    /* Caseful */
        {
-       if (*ecode++ == c) RRETURN(MATCH_NOMATCH);
+       if (*ecode++ == c) MRRETURN(MATCH_NOMATCH);
        }
      break;
-Line 2418 
 for (;;)
+Line 3345 
 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 2438 
 for (;;)
+Line 3369 
 for (;;)
      goto REPEATNOTCHAR;
      case OP_NOTPOSPLUS:
+     case OP_NOTPOSPLUSI:
      possessive = TRUE;
      min = 1;
      max = INT_MAX;
-Line 2445 
 for (;;)
+Line 3377 
 for (;;)
      goto REPEATNOTCHAR;
      case OP_NOTPOSQUERY:
+     case OP_NOTPOSQUERYI:
      possessive = TRUE;
      min = 0;
      max = 1;
-Line 2452 
 for (;;)
+Line 3385 
 for (;;)
      goto REPEATNOTCHAR;
      case OP_NOTPOSUPTO:
+     case OP_NOTPOSUPTOI:
      possessive = TRUE;
      min = 0;
      max = GET2(ecode, 1);
-Line 2459 
 for (;;)
+Line 3393 
 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 */
      if (max == 0) max = INT_MAX;
-     /* Common code for all repeated single-byte matches. We can give up quickly
+     /* Common code for all repeated single-byte matches. */
-     if there are fewer than the minimum number of bytes left in the
-     subject. */
      REPEATNOTCHAR:
-     if (min > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
      fc = *ecode++;
      /* The code is duplicated for the caseless and caseful cases, for speed,
-Line 2489 
 for (;;)
+Line 3426 
 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 2500 
 for (;;)
+Line 3437 
 for (;;)
          register unsigned int d;
          for (i = 1; i <= min; i++)
            {
-           GETCHARINC(d, eptr);
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             MRRETURN(MATCH_NOMATCH);
+             }
+           GETCHARINC(d, eptr);
            if (d < 256) d = md->lcc[d];
-           if (fc == d) RRETURN(MATCH_NOMATCH);
+           if (fc == d) MRRETURN(MATCH_NOMATCH);
            }
          }
        else
-Line 2511 
 for (;;)
+Line 3453 
 for (;;)
        /* Not UTF-8 mode */
          {
          for (i = 1; i <= min; i++)
-           if (fc == md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             MRRETURN(MATCH_NOMATCH);
+             }
+           if (fc == md->lcc[*eptr++]) MRRETURN(MATCH_NOMATCH);
+           }
          }
        if (min == max) continue;
-Line 2525 
 for (;;)
+Line 3474 
 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)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
              GETCHARINC(d, eptr);
              if (d < 256) d = md->lcc[d];
-             if (fi >= max || eptr >= md->end_subject || fc == d)
+             if (fc == d) MRRETURN(MATCH_NOMATCH);
-               RRETURN(MATCH_NOMATCH);
              }
            }
          else
-Line 2539 
 for (;;)
+Line 3493 
 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 || eptr >= md->end_subject || fc == md->lcc[*eptr++])
+             if (fi >= max) MRRETURN(MATCH_NOMATCH);
-               RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             if (fc == md->lcc[*eptr++]) MRRETURN(MATCH_NOMATCH);
              }
            }
          /* Control never gets here */
-Line 2562 
 for (;;)
+Line 3521 
 for (;;)
            for (i = min; i < max; i++)
              {
              int len = 1;
-             if (eptr >= md->end_subject) break;
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               break;
+               }
              GETCHARLEN(d, eptr, len);
              if (d < 256) d = md->lcc[d];
              if (fc == d) break;
-Line 2571 
 for (;;)
+Line 3534 
 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 2583 
 for (;;)
+Line 3546 
 for (;;)
            {
            for (i = min; i < max; i++)
              {
-             if (eptr >= md->end_subject || fc == md->lcc[*eptr]) break;
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               break;
+               }
+             if (fc == md->lcc[*eptr]) break;
              eptr++;
              }
            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--;
              }
            }
-         RRETURN(MATCH_NOMATCH);
+         MRRETURN(MATCH_NOMATCH);
          }
        /* Control never gets here */
        }
-Line 2611 
 for (;;)
+Line 3579 
 for (;;)
          register unsigned int d;
          for (i = 1; i <= min; i++)
            {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             MRRETURN(MATCH_NOMATCH);
+             }
            GETCHARINC(d, eptr);
-           if (fc == d) RRETURN(MATCH_NOMATCH);
+           if (fc == d) MRRETURN(MATCH_NOMATCH);
            }
          }
        else
-Line 2620 
 for (;;)
+Line 3593 
 for (;;)
        /* Not UTF-8 mode */
          {
          for (i = 1; i <= min; i++)
-           if (fc == *eptr++) RRETURN(MATCH_NOMATCH);
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             MRRETURN(MATCH_NOMATCH);
+             }
+           if (fc == *eptr++) MRRETURN(MATCH_NOMATCH);
+           }
          }
        if (min == max) continue;
-Line 2634 
 for (;;)
+Line 3614 
 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)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
              GETCHARINC(d, eptr);
-             if (fi >= max || eptr >= md->end_subject || fc == d)
+             if (fc == d) MRRETURN(MATCH_NOMATCH);
-               RRETURN(MATCH_NOMATCH);
              }
            }
          else
-Line 2647 
 for (;;)
+Line 3632 
 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 || eptr >= md->end_subject || fc == *eptr++)
+             if (fi >= max) MRRETURN(MATCH_NOMATCH);
-               RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             if (fc == *eptr++) MRRETURN(MATCH_NOMATCH);
              }
            }
          /* Control never gets here */
-Line 2670 
 for (;;)
+Line 3660 
 for (;;)
            for (i = min; i < max; i++)
              {
              int len = 1;
-             if (eptr >= md->end_subject) break;
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               break;
+               }
              GETCHARLEN(d, eptr, len);
              if (fc == d) break;
              eptr += len;
-Line 2678 
 for (;;)
+Line 3672 
 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 2690 
 for (;;)
+Line 3684 
 for (;;)
            {
            for (i = min; i < max; i++)
              {
-             if (eptr >= md->end_subject || fc == *eptr) break;
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               break;
+               }
+             if (fc == *eptr) break;
              eptr++;
              }
            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--;
              }
            }
-         RRETURN(MATCH_NOMATCH);
+         MRRETURN(MATCH_NOMATCH);
          }
        }
      /* Control never gets here */
-Line 2784 
 for (;;)
+Line 3783 
 for (;;)
      /* First, ensure the minimum number of matches are present. Use inline
      code for maximizing the speed, and do the type test once at the start
-     (i.e. keep it out of the loop). Also we can test that there are at least
+     (i.e. keep it out of the loop). Separate the UTF-8 code completely as that
-     the minimum number of bytes before we start. This isn't as effective in
-     UTF-8 mode, but it does no harm. Separate the UTF-8 code completely as that
      is tidier. Also separate the UCP code, which can be the same for both UTF-8
      and single-bytes. */
-     if (min > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
      if (min > 0)
        {
  #ifdef SUPPORT_UCP
-Line 2799 
 for (;;)
+Line 3795 
 for (;;)
          switch(prop_type)
            {
            case PT_ANY:
-           if (prop_fail_result) RRETURN(MATCH_NOMATCH);
+           if (prop_fail_result) MRRETURN(MATCH_NOMATCH);
            for (i = 1; i <= min; i++)
              {
-             if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
              GETCHARINCTEST(c, eptr);
              }
            break;
-Line 2810 
 for (;;)
+Line 3810 
 for (;;)
            case PT_LAMP:
            for (i = 1; i <= min; i++)
              {
-             if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
              GETCHARINCTEST(c, eptr);
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+             prop_chartype = UCD_CHARTYPE(c);
              if ((prop_chartype == ucp_Lu ||
                   prop_chartype == ucp_Ll ||
                   prop_chartype == ucp_Lt) == prop_fail_result)
-               RRETURN(MATCH_NOMATCH);
+               MRRETURN(MATCH_NOMATCH);
              }
            break;
            case PT_GC:
            for (i = 1; i <= min; i++)
              {
-             if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
              GETCHARINCTEST(c, eptr);
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+             prop_category = UCD_CATEGORY(c);
              if ((prop_category == prop_value) == prop_fail_result)
-               RRETURN(MATCH_NOMATCH);
+               MRRETURN(MATCH_NOMATCH);
              }
            break;
            case PT_PC:
            for (i = 1; i <= min; i++)
              {
-             if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
              GETCHARINCTEST(c, eptr);
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+             prop_chartype = UCD_CHARTYPE(c);
              if ((prop_chartype == prop_value) == prop_fail_result)
-               RRETURN(MATCH_NOMATCH);
+               MRRETURN(MATCH_NOMATCH);
              }
            break;
            case PT_SC:
            for (i = 1; i <= min; i++)
              {
-             if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
              GETCHARINCTEST(c, eptr);
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+             prop_script = UCD_SCRIPT(c);
              if ((prop_script == prop_value) == prop_fail_result)
-               RRETURN(MATCH_NOMATCH);
+               MRRETURN(MATCH_NOMATCH);
+             }
+           break;
+           case PT_ALNUM:
+           for (i = 1; i <= min; i++)
+             {
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             prop_category = UCD_CATEGORY(c);
+             if ((prop_category == ucp_L || prop_category == ucp_N)
+                    == prop_fail_result)
+               MRRETURN(MATCH_NOMATCH);
+             }
+           break;
+           case PT_SPACE:    /* Perl space */
+           for (i = 1; i <= min; i++)
+             {
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             prop_category = UCD_CATEGORY(c);
+             if ((prop_category == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
+                  c == CHAR_FF || c == CHAR_CR)
+                    == prop_fail_result)
+               MRRETURN(MATCH_NOMATCH);
+             }
+           break;
+           case PT_PXSPACE:  /* POSIX space */
+           for (i = 1; i <= min; i++)
+             {
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             prop_category = UCD_CATEGORY(c);
+             if ((prop_category == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
+                  c == CHAR_VT || c == CHAR_FF || c == CHAR_CR)
+                    == prop_fail_result)
+               MRRETURN(MATCH_NOMATCH);
+             }
+           break;
+           case PT_WORD:
+           for (i = 1; i <= min; i++)
+             {
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               MRRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             prop_category = UCD_CATEGORY(c);
+             if ((prop_category == ucp_L || prop_category == ucp_N ||
+                  c == CHAR_UNDERSCORE)
+                    == prop_fail_result)
+               MRRETURN(MATCH_NOMATCH);
              }
            break;
+           /* This should not occur */
            default:
            RRETURN(PCRE_ERROR_INTERNAL);
            }
-Line 2865 
 for (;;)
+Line 3950 
 for (;;)