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

Diff of /code/trunk/pcre_exec.c

Parent Directory | Revision Log | View Patch Patch

-revision 604 by ph10,
Thu Jun  2 19:04:54 2011 UTC
+revision 850 by zherczeg,
Wed Jan  4 17:29:11 2012 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2011 University of Cambridge
+            Copyright (c) 1997-2012 University of Cambridge
  -----------------------------------------------------------------------------
  Redistribution and use in source and binary forms, with or without
 Line 57 
 possible. There are also some static sup
  #undef min
  #undef max
  /* 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     1  /* Called to check a condition assertion */
 Line 76 
 negative to avoid the external error cod
  #define MATCH_ACCEPT       (-999)
  #define MATCH_COMMIT       (-998)
  #define MATCH_KETRPOS      (-997)
- #define MATCH_PRUNE        (-996)
+ #define MATCH_ONCE         (-996)
- #define MATCH_SKIP         (-995)
+ #define MATCH_PRUNE        (-995)
- #define MATCH_SKIP_ARG     (-994)
+ #define MATCH_SKIP         (-994)
- #define MATCH_THEN         (-993)
+ #define MATCH_SKIP_ARG     (-993)
+ #define MATCH_THEN         (-992)
- /* 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,
-Line 120 
 Returns:     nothing
+Line 113 
 Returns:     nothing
  */
  static void
- pchars(const uschar *p, int length, BOOL is_subject, match_data *md)
+ pchars(const pcre_uchar *p, int length, BOOL is_subject, match_data *md)
  {
  unsigned int c;
  if (is_subject && length > md->end_subject - p) length = md->end_subject - p;
-Line 137 
 while (length-- > 0)
+Line 130 
 while (length-- > 0)
  /* Normally, if a back reference hasn't been set, the length that is passed is
  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:
-Line 151 
 Returns:      < 0 if not matched, otherw
+Line 144 
 Returns:      < 0 if not matched, otherw
  */
  static int
- match_ref(int offset, register USPTR eptr, int length, match_data *md,
+ match_ref(int offset, register PCRE_PUCHAR eptr, int length, match_data *md,
    BOOL caseless)
  {
- USPTR eptr_start = eptr;
+ PCRE_PUCHAR eptr_start = eptr;
- register USPTR p = md->start_subject + md->offset_vector[offset];
+ register PCRE_PUCHAR p = md->start_subject + md->offset_vector[offset];
  #ifdef PCRE_DEBUG
  if (eptr >= md->end_subject)
-Line 180 
 ASCII characters. */
+Line 173 
 ASCII characters. */
  if (caseless)
    {
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
  #ifdef SUPPORT_UCP
-   if (md->utf8)
+   if (md->utf)
      {
      /* Match characters up to the end of the reference. NOTE: the number of
      bytes matched may differ, because there are some characters whose upper and
      lower case versions code as different numbers of bytes. For example, U+023A
      (2 bytes in UTF-8) is the upper case version of U+2C65 (3 bytes in UTF-8);
      a sequence of 3 of the former uses 6 bytes, as does a sequence of two of
      the latter. It is important, therefore, to check the length along the
      reference, not along the subject (earlier code did this wrong). */
-     USPTR endptr = p + length;
+     PCRE_PUCHAR endptr = p + length;
      while (p < endptr)
        {
        int c, d;
-Line 209 
 if (caseless)
+Line 202 
 if (caseless)
    /* 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; }
+       {
-     }
+       if (TABLE_GET(*p, md->lcc, *p) != TABLE_GET(*eptr, md->lcc, *eptr)) return -1;
+       p++;
+       eptr++;
+       }
+     }
    }
  /* In the caseful case, we can just compare the bytes, whether or not we
  are in UTF-8 mode. */
  else
    {
    if (eptr + length > md->end_subject) return -1;
    while (length-- > 0) if (*p++ != *eptr++) return -1;
    }
- return eptr - eptr_start;
+ return (int)(eptr - eptr_start);
  }
-Line 276 
 enum { RM1=1, RM2,  RM3,  RM4,  RM5,  RM
+Line 273 
 enum { RM1=1, RM2,  RM3,  RM4,  RM5,  RM
         RM31,  RM32, RM33, RM34, RM35, RM36, RM37, RM38, RM39, RM40,
         RM41,  RM42, RM43, RM44, RM45, RM46, RM47, RM48, RM49, RM50,
         RM51,  RM52, RM53, RM54, RM55, RM56, RM57, RM58, RM59, RM60,
-        RM61,  RM62, RM63, RM64 };
+        RM61,  RM62, RM63, RM64, RM65, RM66 };
  /* These versions of the macros use the stack, as normal. There are debugging
  versions and production versions. Note that the "rw" argument of RMATCH isn't
-Line 289 
 actually used in this definition. */
+Line 286 
 actually used in this definition. */
  #define RMATCH(ra,rb,rc,rd,re,rw) \
    { \
    printf("match() called in line %d\n", __LINE__); \
-   rrc = match(ra,rb,mstart,markptr,rc,rd,re,rdepth+1); \
+   rrc = match(ra,rb,mstart,rc,rd,re,rdepth+1); \
    printf("to line %d\n", __LINE__); \
    }
  #define RRETURN(ra) \
-Line 299 
 actually used in this definition. */
+Line 296 
 actually used in this definition. */
    }
  #else
  #define RMATCH(ra,rb,rc,rd,re,rw) \
-   rrc = match(ra,rb,mstart,markptr,rc,rd,re,rdepth+1)
+   rrc = match(ra,rb,mstart,rc,rd,re,rdepth+1)
  #define RRETURN(ra) return ra
  #endif
-Line 314 
 argument of match(), which never changes
+Line 311 
 argument of match(), which never changes
  #define RMATCH(ra,rb,rc,rd,re,rw)\
    {\
-   heapframe *newframe = (heapframe *)(pcre_stack_malloc)(sizeof(heapframe));\
+   heapframe *newframe = (heapframe *)(PUBL(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->Xeptrb = re;\
    newframe->Xrdepth = frame->Xrdepth + 1;\
-Line 336 
 argument of match(), which never changes
+Line 332 
 argument of match(), which never changes
    {\
    heapframe *oldframe = frame;\
    frame = oldframe->Xprevframe;\
-   (pcre_stack_free)(oldframe);\
+   (PUBL(stack_free))(oldframe);\
    if (frame != NULL)\
      {\
      rrc = ra;\
-Line 353 
 typedef struct heapframe {
+Line 349 
 typedef struct heapframe {
    /* Function arguments that may change */
-   USPTR Xeptr;
+   PCRE_PUCHAR Xeptr;
-   const uschar *Xecode;
+   const pcre_uchar *Xecode;
-   USPTR Xmstart;
+   PCRE_PUCHAR Xmstart;
-   USPTR Xmarkptr;
    int Xoffset_top;
    eptrblock *Xeptrb;
    unsigned int Xrdepth;
    /* Function local variables */
-   USPTR Xcallpat;
+   PCRE_PUCHAR Xcallpat;
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-   USPTR Xcharptr;
+   PCRE_PUCHAR Xcharptr;
  #endif
-   USPTR Xdata;
+   PCRE_PUCHAR Xdata;
-   USPTR Xnext;
+   PCRE_PUCHAR Xnext;
-   USPTR Xpp;
+   PCRE_PUCHAR Xpp;
-   USPTR Xprev;
+   PCRE_PUCHAR Xprev;
-   USPTR Xsaved_eptr;
+   PCRE_PUCHAR Xsaved_eptr;
    recursion_info Xnew_recursive;
-Line 383 
 typedef struct heapframe {
+Line 378 
 typedef struct heapframe {
    int Xprop_type;
    int Xprop_value;
    int Xprop_fail_result;
-   int Xprop_category;
-   int Xprop_chartype;
-   int Xprop_script;
    int Xoclength;
-   uschar Xocchars[8];
+   pcre_uchar Xocchars[6];
  #endif
    int Xcodelink;
-Line 429 
 returns a negative (error) response, the
+Line 421 
 returns a negative (error) response, the
  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
+ appear several times in the code. We set the "hit end" flag if the pointer is
  at the end of the subject and also past the start of the subject (i.e.
  something has been matched). For hard partial matching, we then return
  immediately. The second one is used when we already know we are past the end of
-Line 440 
 the subject. */
+Line 432 
 the subject. */
        eptr > md->start_used_ptr) \
      { \
      md->hitend = TRUE; \
-     if (md->partial > 1) MRRETURN(PCRE_ERROR_PARTIAL); \
+     if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL); \
      }
  #define SCHECK_PARTIAL()\
    if (md->partial != 0 && eptr > md->start_used_ptr) \
      { \
      md->hitend = TRUE; \
-     if (md->partial > 1) MRRETURN(PCRE_ERROR_PARTIAL); \
+     if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL); \
      }
  /* Performance note: It might be tempting to extract commonly used fields from
- the md structure (e.g. utf8, end_subject) into individual variables to improve
+ the md structure (e.g. utf, end_subject) into individual variables to improve
  performance. Tests using gcc on a SPARC disproved this; in the first case, it
  made performance worse.
-Line 461 
 Arguments:
+Line 453 
 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
     eptrb       pointer to chain of blocks containing eptr at start of
-Line 476 
 Returns:       MATCH_MATCH if matched
+Line 467 
 Returns:       MATCH_MATCH if matched
  */
  static int
- match(REGISTER USPTR eptr, REGISTER const uschar *ecode, USPTR mstart,
+ match(REGISTER PCRE_PUCHAR eptr, REGISTER const pcre_uchar *ecode,
-   const uschar *markptr, int offset_top, match_data *md, eptrblock *eptrb,
+   PCRE_PUCHAR mstart, int offset_top, match_data *md, eptrblock *eptrb,
    unsigned int rdepth)
  {
  /* These variables do not need to be preserved over recursion in this function,
-Line 487 
 so they can be ordinary variables in all
+Line 478 
 so they can be ordinary variables in all
  register int  rrc;         /* Returns from recursive calls */
  register int  i;           /* Used for loops not involving calls to RMATCH() */
  register unsigned int c;   /* Character values not kept over RMATCH() calls */
- register BOOL utf8;        /* Local copy of UTF-8 flag for speed */
+ register BOOL utf;         /* Local copy of UTF flag for speed */
  BOOL minimize, possessive; /* Quantifier options */
  BOOL caseless;
-Line 499 
 heap storage. Set up the top-level frame
+Line 490 
 heap storage. Set up the top-level frame
  heap whenever RMATCH() does a "recursion". See the macro definitions above. */
  #ifdef NO_RECURSE
- heapframe *frame = (heapframe *)(pcre_stack_malloc)(sizeof(heapframe));
+ heapframe *frame = (heapframe *)(PUBL(stack_malloc))(sizeof(heapframe));
  if (frame == NULL) RRETURN(PCRE_ERROR_NOMEMORY);
  frame->Xprevframe = NULL;            /* Marks the top level */
-Line 508 
 frame->Xprevframe = NULL;            /*
+Line 499 
 frame->Xprevframe = NULL;            /*
  frame->Xeptr = eptr;
  frame->Xecode = ecode;
  frame->Xmstart = mstart;
- frame->Xmarkptr = markptr;
  frame->Xoffset_top = offset_top;
  frame->Xeptrb = eptrb;
  frame->Xrdepth = rdepth;
-Line 522 
 HEAP_RECURSE:
+Line 512 
 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 eptrb              frame->Xeptrb
  #define rdepth             frame->Xrdepth
  /* Ditto for the local variables */
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
  #define charptr            frame->Xcharptr
  #endif
  #define callpat            frame->Xcallpat
-Line 550 
 HEAP_RECURSE:
+Line 539 
 HEAP_RECURSE:
  #define prop_type          frame->Xprop_type
  #define prop_value         frame->Xprop_value
  #define prop_fail_result   frame->Xprop_fail_result
- #define prop_category      frame->Xprop_category
- #define prop_chartype      frame->Xprop_chartype
- #define prop_script        frame->Xprop_script
  #define oclength           frame->Xoclength
  #define occhars            frame->Xocchars
  #endif
-Line 590 
 declarations can be cut out in a block.
+Line 576 
 declarations can be cut out in a block.
  below are for variables that do not have to be preserved over a recursive call
  to RMATCH(). */
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
- const uschar *charptr;
+ const pcre_uchar *charptr;
  #endif
- const uschar *callpat;
+ const pcre_uchar *callpat;
- const uschar *data;
+ const pcre_uchar *data;
- const uschar *next;
+ const pcre_uchar *next;
- USPTR         pp;
+ PCRE_PUCHAR       pp;
- const uschar *prev;
+ const pcre_uchar *prev;
- USPTR         saved_eptr;
+ PCRE_PUCHAR       saved_eptr;
  recursion_info new_recursive;
  BOOL cur_is_word;
  BOOL condition;
  BOOL prev_is_word;
-Line 610 
 BOOL prev_is_word;
+Line 596 
 BOOL prev_is_word;
  int prop_type;
  int prop_value;
  int prop_fail_result;
- int prop_category;
- int prop_chartype;
- int prop_script;
  int oclength;
- uschar occhars[8];
+ pcre_uchar occhars[6];
  #endif
  int codelink;
-Line 632 
 int stacksave[REC_STACK_SAVE_MAX];
+Line 615 
 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
-Line 642 
 the alternative names that are used. */
+Line 625 
 the alternative names that are used. */
  #define code_offset   codelink
  #define condassert    condition
  #define matched_once  prev_is_word
+ #define foc           number
  /* These statements are here to stop the compiler complaining about unitialized
  variables. */
-Line 667 
 defined). However, RMATCH isn't like a f
+Line 651 
 defined). However, RMATCH isn't like a f
  complicated macro. It has to be used in one particular way. This shouldn't,
  however, impact performance when true recursion is being used. */
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
- utf8 = md->utf8;       /* Local copy of the flag */
+ utf = md->utf;       /* Local copy of the flag */
  #else
- utf8 = FALSE;
+ utf = FALSE;
  #endif
  /* First check that we haven't called match() too many times, or that we
-Line 680 
 if (md->match_call_count++ >= md->match_
+Line 664 
 if (md->match_call_count++ >= md->match_
  if (rdepth >= md->match_limit_recursion) RRETURN(PCRE_ERROR_RECURSIONLIMIT);
  /* At the start of a group with an unlimited repeat that may match an empty
  string, the variable md->match_function_type is set to MATCH_CBEGROUP. It is
  done this way to save having to use another function argument, which would take
  up space on the stack. See also MATCH_CONDASSERT below.
  When MATCH_CBEGROUP is set, add the current subject pointer to the chain of
-Line 705 
 for (;;)
+Line 689 
 for (;;)
    {
    minimize = possessive = FALSE;
    op = *ecode;
    switch(op)
      {
      case OP_MARK:
-     markptr = ecode + 2;
+     md->nomatch_mark = ecode + 2;
-     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode] + ecode[1], offset_top, md,
+     md->mark = NULL;    /* In case previously set by assertion */
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
        eptrb, RM55);
+     if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
+          md->mark == NULL) md->mark = ecode + 2;
      /* A return of MATCH_SKIP_ARG means that matching failed at SKIP with an
      argument, and we must check whether that argument matches this MARK's
-Line 720 
 for (;;)
+Line 707 
 for (;;)
      position and return MATCH_SKIP. Otherwise, pass back the return code
      unaltered. */
-     if (rrc == MATCH_SKIP_ARG &&
+     else if (rrc == MATCH_SKIP_ARG &&
-         strcmp((char *)markptr, (char *)(md->start_match_ptr)) == 0)
+         STRCMP_UC_UC(ecode + 2, 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);
+     RRETURN(MATCH_NOMATCH);
      /* COMMIT overrides PRUNE, SKIP, and THEN */
      case OP_COMMIT:
-     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
        eptrb, RM52);
      if (rrc != MATCH_NOMATCH && rrc != MATCH_PRUNE &&
          rrc != MATCH_SKIP && rrc != MATCH_SKIP_ARG &&
          rrc != MATCH_THEN)
        RRETURN(rrc);
-     MRRETURN(MATCH_COMMIT);
+     RRETURN(MATCH_COMMIT);
      /* PRUNE overrides THEN */
      case OP_PRUNE:
-     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
        eptrb, RM51);
      if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
-     MRRETURN(MATCH_PRUNE);
+     RRETURN(MATCH_PRUNE);
      case OP_PRUNE_ARG:
-     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode] + ecode[1], offset_top, md,
+     md->nomatch_mark = ecode + 2;
+     md->mark = NULL;    /* In case previously set by assertion */
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
        eptrb, RM56);
+     if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
+          md->mark == NULL) md->mark = ecode + 2;
      if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
-     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,
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
        eptrb, RM53);
      if (rrc != MATCH_NOMATCH && rrc != MATCH_PRUNE && rrc != MATCH_THEN)
        RRETURN(rrc);
      md->start_match_ptr = eptr;   /* Pass back current position */
-     MRRETURN(MATCH_SKIP);
+     RRETURN(MATCH_SKIP);
+     /* Note that, for Perl compatibility, SKIP with an argument does NOT set
+     nomatch_mark. There is a flag that disables this opcode when re-matching a
+     pattern that ended with a SKIP for which there was not a matching MARK. */
      case OP_SKIP_ARG:
-     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode] + ecode[1], offset_top, md,
+     if (md->ignore_skip_arg)
+       {
+       ecode += PRIV(OP_lengths)[*ecode] + ecode[1];
+       break;
+       }
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
        eptrb, RM57);
      if (rrc != MATCH_NOMATCH && rrc != MATCH_PRUNE && rrc != MATCH_THEN)
        RRETURN(rrc);
      /* Pass back the current skip name by overloading md->start_match_ptr and
      returning the special MATCH_SKIP_ARG return code. This will either be
-     caught by a matching MARK, or get to the top, where it is treated the same
+     caught by a matching MARK, or get to the top, where it causes a rematch
-     as PRUNE. */
+     with the md->ignore_skip_arg flag set. */
      md->start_match_ptr = ecode + 2;
      RRETURN(MATCH_SKIP_ARG);
-     /* For THEN (and THEN_ARG) we pass back the address of the bracket or
+     /* For THEN (and THEN_ARG) we pass back the address of the opcode, so that
-     the alt that is at the start of the current branch. This makes it possible
+     the branch in which it occurs can be determined. Overload the start of
-     to skip back past alternatives that precede the THEN within the current
+     match pointer to do this. */
-     branch. */
      case OP_THEN:
-     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
        eptrb, RM54);
      if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-     md->start_match_ptr = ecode - GET(ecode, 1);
+     md->start_match_ptr = ecode;
-     MRRETURN(MATCH_THEN);
+     RRETURN(MATCH_THEN);
      case OP_THEN_ARG:
-     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode] + ecode[1+LINK_SIZE],
+     md->nomatch_mark = ecode + 2;
-       offset_top, md, eptrb, RM58);
+     md->mark = NULL;    /* In case previously set by assertion */
+     RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top,
+       md, eptrb, RM58);
+     if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
+          md->mark == NULL) md->mark = ecode + 2;
      if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-     md->start_match_ptr = ecode - GET(ecode, 1);
+     md->start_match_ptr = ecode;
-     md->mark = ecode + LINK_SIZE + 2;
      RRETURN(MATCH_THEN);
+     /* Handle an atomic group that does not contain any capturing parentheses.
+     This can be handled like an assertion. Prior to 8.13, all atomic groups
+     were handled this way. In 8.13, the code was changed as below for ONCE, so
+     that backups pass through the group and thereby reset captured values.
+     However, this uses a lot more stack, so in 8.20, atomic groups that do not
+     contain any captures generate OP_ONCE_NC, which can be handled in the old,
+     less stack intensive way.
+     Check the alternative branches in turn - the matching won't pass the KET
+     for this kind of subpattern. If any one branch matches, we carry on as at
+     the end of a normal bracket, leaving the subject pointer, but resetting
+     the start-of-match value in case it was changed by \K. */
+     case OP_ONCE_NC:
+     prev = ecode;
+     saved_eptr = eptr;
+     do
+       {
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM64);
+       if (rrc == MATCH_MATCH)  /* Note: _not_ MATCH_ACCEPT */
+         {
+         mstart = md->start_match_ptr;
+         break;
+         }
+       if (rrc == MATCH_THEN)
+         {
+         next = ecode + GET(ecode,1);
+         if (md->start_match_ptr < next &&
+             (*ecode == OP_ALT || *next == OP_ALT))
+           rrc = MATCH_NOMATCH;
+         }
+       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       ecode += GET(ecode,1);
+       }
+     while (*ecode == OP_ALT);
+     /* If hit the end of the group (which could be repeated), fail */
+     if (*ecode != OP_ONCE_NC && *ecode != OP_ALT) RRETURN(MATCH_NOMATCH);
+     /* Continue as from after the group, updating the offsets high water
+     mark, since extracts may have been taken. */
+     do ecode += GET(ecode, 1); while (*ecode == OP_ALT);
+     offset_top = md->end_offset_top;
+     eptr = md->end_match_ptr;
+     /* For a non-repeating ket, just continue at this level. This also
+     happens for a repeating ket if no characters were matched in the group.
+     This is the forcible breaking of infinite loops as implemented in Perl
+.005. */
+     if (*ecode == OP_KET || eptr == saved_eptr)
+       {
+       ecode += 1+LINK_SIZE;
+       break;
+       }
+     /* The repeating kets try the rest of the pattern or restart from the
+     preceding bracket, in the appropriate order. The second "call" of match()
+     uses tail recursion, to avoid using another stack frame. */
+     if (*ecode == OP_KETRMIN)
+       {
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM65);
+       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       ecode = prev;
+       goto TAIL_RECURSE;
+       }
+     else  /* OP_KETRMAX */
+       {
+       md->match_function_type = MATCH_CBEGROUP;
+       RMATCH(eptr, prev, offset_top, md, eptrb, RM66);
+       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       ecode += 1 + LINK_SIZE;
+       goto TAIL_RECURSE;
+       }
+     /* Control never gets here */
      /* 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
+     the group. A failure to match might occur after the group has succeeded,
-     values of the final offsets, in case they were set by a previous iteration
+     if something later on doesn't match. For this reason, we need to restore
-     of the same bracket.
+     the working value and also the values of the final offsets, in case they
+     were set by a previous iteration of the same bracket.
      If there isn't enough space in the offset vector, treat this as if it were
      a non-capturing bracket. Don't worry about setting the flag for the error
-Line 820 
 for (;;)
+Line 901 
 for (;;)
      case OP_SCBRA:
      number = GET2(ecode, 1+LINK_SIZE);
      offset = number << 1;
  #ifdef PCRE_DEBUG
      printf("start bracket %d\n", number);
      printf("subject=");
-Line 841 
 for (;;)
+Line 922 
 for (;;)
        for (;;)
          {
          if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
-         RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
+         RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
            eptrb, RM1);
-         if (rrc != MATCH_NOMATCH &&
+         if (rrc == MATCH_ONCE) break;  /* Backing up through an atomic group */
-             (rrc != MATCH_THEN || md->start_match_ptr != ecode))
-           RRETURN(rrc);
+         /* If we backed up to a THEN, check whether it is within the current
+         branch by comparing the address of the THEN that is passed back with
+         the end of the branch. If it is within the current branch, and the
+         branch is one of two or more alternatives (it either starts or ends
+         with OP_ALT), we have reached the limit of THEN's action, so convert
+         the return code to NOMATCH, which will cause normal backtracking to
+         happen from now on. Otherwise, THEN is passed back to an outer
+         alternative. This implements Perl's treatment of parenthesized groups,
+         where a group not containing | does not affect the current alternative,
+         that is, (X) is NOT the same as (X|(*F)). */
+         if (rrc == MATCH_THEN)
+           {
+           next = ecode + GET(ecode,1);
+           if (md->start_match_ptr < next &&
+               (*ecode == OP_ALT || *next == OP_ALT))
+             rrc = MATCH_NOMATCH;
+           }
+         /* Anything other than NOMATCH is passed back. */
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
          md->capture_last = save_capture_last;
          ecode += GET(ecode, 1);
          if (*ecode != OP_ALT) break;
          }
        DPRINTF(("bracket %d failed\n", number));
        md->offset_vector[offset] = save_offset1;
        md->offset_vector[offset+1] = save_offset2;
        md->offset_vector[md->offset_end - number] = save_offset3;
-       if (rrc != MATCH_THEN) md->mark = markptr;
+       /* At this point, rrc will be one of MATCH_ONCE or MATCH_NOMATCH. */
-       RRETURN(MATCH_NOMATCH);
+       RRETURN(rrc);
        }
      /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
-Line 873 
 for (;;)
+Line 975 
 for (;;)
      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-     /* Non-capturing bracket, except for possessive with unlimited repeat. Loop
+     /* Non-capturing or atomic group, except for possessive with unlimited
-     for all the alternatives. When we get to the final alternative within the
+     repeat and ONCE group with no captures. Loop for all the alternatives.
-     brackets, we would return the result of a recursive call to match()
-     whatever happened. We can reduce stack usage by turning this into a tail
+     When we get to the final alternative within the brackets, we used to return
-     recursion, except in the case of a possibly empty group.*/
+     the result of a recursive call to match() whatever happened so it was
+     possible to reduce stack usage by turning this into a tail recursion,
+     except in the case of a possibly empty group. However, now that there is
+     the possiblity of (*THEN) occurring in the final alternative, this
+     optimization is no longer always possible.
+     We can optimize if we know there are no (*THEN)s in the pattern; at present
+     this is the best that can be done.
+     MATCH_ONCE is returned when the end of an atomic group is successfully
+     reached, but subsequent matching fails. It passes back up the tree (causing
+     captured values to be reset) until the original atomic group level is
+     reached. This is tested by comparing md->once_target with the start of the
+     group. At this point, the return is converted into MATCH_NOMATCH so that
+     previous backup points can be taken. */
+     case OP_ONCE:
      case OP_BRA:
      case OP_SBRA:
      DPRINTF(("start non-capturing bracket\n"));
      for (;;)
        {
-       if (ecode[GET(ecode, 1)] != OP_ALT)   /* Final alternative */
+       if (op >= OP_SBRA || op == OP_ONCE) md->match_function_type = MATCH_CBEGROUP;
+       /* If this is not a possibly empty group, and there are no (*THEN)s in
+       the pattern, and this is the final alternative, optimize as described
+       above. */
+       else if (!md->hasthen && ecode[GET(ecode, 1)] != OP_ALT)
          {
-         if (op >= OP_SBRA)   /* Possibly empty group */
+         ecode += PRIV(OP_lengths)[*ecode];
-           {
-           md->match_function_type = MATCH_CBEGROUP;
-           RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md, eptrb,
-             RM48);
-           if (rrc == MATCH_NOMATCH) md->mark = markptr;
-           RRETURN(rrc);
-           }
-         /* Not a possibly empty group; use tail recursion */
-         ecode += _pcre_OP_lengths[*ecode];
-         DPRINTF(("bracket 0 tail recursion\n"));
          goto TAIL_RECURSE;
          }
-       /* For non-final alternatives, continue the loop for a NOMATCH result;
+       /* In all other cases, we have to make another call to match(). */
-       otherwise return. */
-       if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
+       RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md, eptrb,
-       RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md, eptrb,
          RM2);
-       if (rrc != MATCH_NOMATCH &&
-           (rrc != MATCH_THEN || md->start_match_ptr != ecode))
+       /* See comment in the code for capturing groups above about handling
+       THEN. */
+       if (rrc == MATCH_THEN)
+         {
+         next = ecode + GET(ecode,1);
+         if (md->start_match_ptr < next &&
+             (*ecode == OP_ALT || *next == OP_ALT))
+           rrc = MATCH_NOMATCH;
+         }
+       if (rrc != MATCH_NOMATCH)
+         {
+         if (rrc == MATCH_ONCE)
+           {
+           const pcre_uchar *scode = ecode;
+           if (*scode != OP_ONCE)           /* If not at start, find it */
+             {
+             while (*scode == OP_ALT) scode += GET(scode, 1);
+             scode -= GET(scode, 1);
+             }
+           if (md->once_target == scode) rrc = MATCH_NOMATCH;
+           }
          RRETURN(rrc);
+         }
        ecode += GET(ecode, 1);
+       if (*ecode != OP_ALT) break;
        }
-     /* Control never reaches here. */
-     /* Handle possessive capturing brackets with an unlimited repeat. We come
+     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;
-Line 939 
 for (;;)
+Line 1076 
 for (;;)
      if (offset < md->offset_max)
        {
        matched_once = FALSE;
-       code_offset = ecode - md->start_code;
+       code_offset = (int)(ecode - md->start_code);
        save_offset1 = md->offset_vector[offset];
        save_offset2 = md->offset_vector[offset+1];
-Line 947 
 for (;;)
+Line 1084 
 for (;;)
        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,
+         RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
            eptrb, RM63);
          if (rrc == MATCH_KETRPOS)
            {
            offset_top = md->end_offset_top;
            eptr = md->end_match_ptr;
            ecode = md->start_code + code_offset;
            save_capture_last = md->capture_last;
            matched_once = TRUE;
            continue;
            }
-         if (rrc != MATCH_NOMATCH &&
-             (rrc != MATCH_THEN || md->start_match_ptr != ecode))
+         /* See comment in the code for capturing groups above about handling
-           RRETURN(rrc);
+         THEN. */
+         if (rrc == MATCH_THEN)
+           {
+           next = ecode + GET(ecode,1);
+           if (md->start_match_ptr < next &&
+               (*ecode == OP_ALT || *next == OP_ALT))
+             rrc = MATCH_NOMATCH;
+           }
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
          md->capture_last = save_capture_last;
          ecode += GET(ecode, 1);
          if (*ecode != OP_ALT) break;
          }
        if (!matched_once)
          {
          md->offset_vector[offset] = save_offset1;
          md->offset_vector[offset+1] = save_offset2;
          md->offset_vector[md->offset_end - number] = save_offset3;
          }
-       if (rrc != MATCH_THEN) md->mark = markptr;
        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. */
-Line 1009 
 for (;;)
+Line 1155 
 for (;;)
      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
      /* Non-capturing possessive bracket with unlimited repeat. We come here
      from BRAZERO with allow_zero = TRUE. The code is similar to the above,
      without the capturing complication. It is written out separately for speed
      and cleanliness. */
      case OP_BRAPOS:
      case OP_SBRAPOS:
      allow_zero = FALSE;
      POSSESSIVE_NON_CAPTURE:
      matched_once = FALSE;
-     code_offset = ecode - md->start_code;
+     code_offset = (int)(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,
+       RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
-         eptrb, RM64);
+         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))
+       /* See comment in the code for capturing groups above about handling
-         RRETURN(rrc);
+       THEN. */
+       if (rrc == MATCH_THEN)
+         {
+         next = ecode + GET(ecode,1);
+         if (md->start_match_ptr < next &&
+             (*ecode == OP_ALT || *next == OP_ALT))
+           rrc = MATCH_NOMATCH;
+         }
+       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
        ecode += GET(ecode, 1);
        if (*ecode != OP_ALT) break;
        }
      if (matched_once || allow_zero)
        {
        ecode += 1 + LINK_SIZE;
        break;
        }
      RRETURN(MATCH_NOMATCH);
      /* Control never reaches here. */
-Line 1053 
 for (;;)
+Line 1210 
 for (;;)
      /* 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:
-Line 1065 
 for (;;)
+Line 1221 
 for (;;)
      if (ecode[LINK_SIZE+1] == OP_CALLOUT)
        {
-       if (pcre_callout != NULL)
+       if (PUBL(callout) != NULL)
          {
-         pcre_callout_block cb;
+         PUBL(callout_block) cb;
-         cb.version          = 1;   /* Version 1 of the callout block */
+         cb.version          = 2;   /* Version 1 of the callout block */
          cb.callout_number   = ecode[LINK_SIZE+2];
          cb.offset_vector    = md->offset_vector;
          cb.subject          = (PCRE_SPTR)md->start_subject;
-Line 1080 
 for (;;)
+Line 1236 
 for (;;)
          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);
+         cb.mark             = md->nomatch_mark;
+         if ((rrc = (*PUBL(callout))(&cb)) > 0) RRETURN(MATCH_NOMATCH);
          if (rrc < 0) RRETURN(rrc);
          }
-       ecode += _pcre_OP_lengths[OP_CALLOUT];
+       ecode += PRIV(OP_lengths)[OP_CALLOUT];
        }
      condcode = ecode[LINK_SIZE+1];
-Line 1100 
 for (;;)
+Line 1257 
 for (;;)
        else
          {
          int recno = GET2(ecode, LINK_SIZE + 2);   /* Recursion group number*/
-         condition =  (recno == RREF_ANY || recno == md->recursive->group_num);
+         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)
+         if (!condition && condcode == OP_NRREF)
            {
-           uschar *slotA = md->name_table;
+           pcre_uchar *slotA = md->name_table;
            for (i = 0; i < md->name_count; i++)
              {
              if (GET2(slotA, 0) == recno) break;
-Line 1122 
 for (;;)
+Line 1279 
 for (;;)
            if (i < md->name_count)
              {
-             uschar *slotB = slotA;
+             pcre_uchar *slotB = slotA;
              while (slotB > md->name_table)
                {
                slotB -= md->name_entry_size;
-               if (strcmp((char *)slotA + 2, (char *)slotB + 2) == 0)
+               if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
                  {
                  condition = GET2(slotB, 0) == md->recursive->group_num;
                  if (condition) break;
-Line 1142 
 for (;;)
+Line 1299 
 for (;;)
                for (i++; i < md->name_count; i++)
                  {
                  slotB += md->name_entry_size;
-                 if (strcmp((char *)slotA + 2, (char *)slotB + 2) == 0)
+                 if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
                    {
                    condition = GET2(slotB, 0) == md->recursive->group_num;
                    if (condition) break;
-Line 1155 
 for (;;)
+Line 1312 
 for (;;)
          /* Chose branch according to the condition */
-         ecode += condition? 3 : GET(ecode, 1);
+         ecode += condition? 1 + IMM2_SIZE : GET(ecode, 1);
          }
        }
-Line 1172 
 for (;;)
+Line 1329 
 for (;;)
        if (!condition && condcode == OP_NCREF)
          {
          int refno = offset >> 1;
-         uschar *slotA = md->name_table;
+         pcre_uchar *slotA = md->name_table;
          for (i = 0; i < md->name_count; i++)
            {
-Line 1186 
 for (;;)
+Line 1343 
 for (;;)
          if (i < md->name_count)
            {
-           uschar *slotB = slotA;
+           pcre_uchar *slotB = slotA;
            while (slotB > md->name_table)
              {
              slotB -= md->name_entry_size;
-             if (strcmp((char *)slotA + 2, (char *)slotB + 2) == 0)
+             if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
                {
                offset = GET2(slotB, 0) << 1;
                condition = offset < offset_top &&
-Line 1208 
 for (;;)
+Line 1365 
 for (;;)
              for (i++; i < md->name_count; i++)
                {
                slotB += md->name_entry_size;
-               if (strcmp((char *)slotA + 2, (char *)slotB + 2) == 0)
+               if (STRCMP_UC_UC(slotA + IMM2_SIZE, slotB + IMM2_SIZE) == 0)
                  {
                  offset = GET2(slotB, 0) << 1;
                  condition = offset < offset_top &&
-Line 1223 
 for (;;)
+Line 1380 
 for (;;)
        /* Chose branch according to the condition */
-       ecode += condition? 3 : GET(ecode, 1);
+       ecode += condition? 1 + IMM2_SIZE : GET(ecode, 1);
        }
      else if (condcode == OP_DEF)     /* DEFINE - always false */
-Line 1238 
 for (;;)
+Line 1395 
 for (;;)
      else
        {
        md->match_function_type = MATCH_CONDASSERT;
        RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM3);
        if (rrc == MATCH_MATCH)
          {
+         if (md->end_offset_top > offset_top)
+           offset_top = md->end_offset_top;  /* Captures may have happened */
          condition = TRUE;
          ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE + 2);
          while (*ecode == OP_ALT) ecode += GET(ecode, 1);
          }
-       else if (rrc != MATCH_NOMATCH &&
-               (rrc != MATCH_THEN || md->start_match_ptr != ecode))
+       /* PCRE doesn't allow the effect of (*THEN) to escape beyond an
+       assertion; it is therefore treated as NOMATCH. */
+       else if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN)
          {
          RRETURN(rrc);         /* Need braces because of following else */
          }
-Line 1258 
 for (;;)
+Line 1420 
 for (;;)
          }
        }
-     /* We are now at the branch that is to be obeyed. As there is only one,
+     /* We are now at the branch that is to be obeyed. As there is only one, can
-     we can use tail recursion to avoid using another stack frame, except when
+     use tail recursion to avoid using another stack frame, except when there is
-     we have an unlimited repeat of a possibly empty group. If the second
+     unlimited repeat of a possibly empty group. In the latter case, a recursive
-     alternative doesn't exist, we can just plough on. */
+     call to match() is always required, unless the second alternative doesn't
+     exist, in which case we can just plough on. Note that, for compatibility
+     with Perl, the | in a conditional group is NOT treated as creating two
+     alternatives. If a THEN is encountered in the branch, it propagates out to
+     the enclosing alternative (unless nested in a deeper set of alternatives,
+     of course). */
      if (condition || *ecode == OP_ALT)
        {
-       ecode += 1 + LINK_SIZE;
+       if (op != OP_SCOND)
-       if (op == OP_SCOND)        /* Possibly empty group */
          {
-         md->match_function_type = MATCH_CBEGROUP;
+         ecode += 1 + LINK_SIZE;
-         RMATCH(eptr, ecode, offset_top, md, eptrb, RM49);
+         goto TAIL_RECURSE;
-         RRETURN(rrc);
          }
-       else goto TAIL_RECURSE;
+       md->match_function_type = MATCH_CBEGROUP;
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM49);
+       RRETURN(rrc);
        }
-     else                         /* Condition false & no alternative */
+      /* Condition false & no alternative; continue after the group. */
+     else
        {
        ecode += 1 + LINK_SIZE;
        }
-Line 1301 
 for (;;)
+Line 1472 
 for (;;)
        md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
        if (offset_top <= offset) offset_top = offset + 2;
        }
-     ecode += 3;
+     ecode += 1 + IMM2_SIZE;
      break;
-     /* End of the pattern, either real or forced. If we are in a top-level
+     /* End of the pattern, either real or forced. */
-     recursion, we should restore the offsets appropriately and continue from
-     after the call. */
-     case OP_ACCEPT:
      case OP_END:
-     if (md->recursive != NULL && md->recursive->group_num == 0)
+     case OP_ACCEPT:
-       {
+     case OP_ASSERT_ACCEPT:
-       recursion_info *rec = md->recursive;
-       DPRINTF(("End of pattern in a (?0) recursion\n"));
-       md->recursive = rec->prevrec;
-       memmove(md->offset_vector, rec->offset_save,
-         rec->saved_max * sizeof(int));
-       offset_top = rec->save_offset_top;
-       ecode = rec->after_call;
-       break;
-       }
-     /* Otherwise, if we have matched an empty string, fail if PCRE_NOTEMPTY is
+     /* If we have matched an empty string, fail if not in an assertion and not
-     set, or if PCRE_NOTEMPTY_ATSTART is set and we have matched at the start of
+     in a recursion if either PCRE_NOTEMPTY is set, or if PCRE_NOTEMPTY_ATSTART
-     the subject. In both cases, backtracking will then try other alternatives,
+     is set and we have matched at the start of the subject. In both cases,
-     if any. */
+     backtracking will then try other alternatives, if any. */
-     if (eptr == mstart &&
+     if (eptr == mstart && op != OP_ASSERT_ACCEPT &&
-         (md->notempty ||
+          md->recursive == NULL &&
-           (md->notempty_atstart &&
+          (md->notempty ||
-             mstart == md->start_subject + md->start_offset)))
+            (md->notempty_atstart &&
-       MRRETURN(MATCH_NOMATCH);
+              mstart == md->start_subject + md->start_offset)))
+       RRETURN(MATCH_NOMATCH);
      /* Otherwise, we have a match. */
-Line 1341 
 for (;;)
+Line 1501 
 for (;;)
      md->start_match_ptr = mstart;       /* and the start (\K can modify) */
      /* For some reason, the macros don't work properly if an expression is
-     given as the argument to MRRETURN when the heap is in use. */
+     given as the argument to RRETURN when the heap is in use. */
      rrc = (op == OP_END)? MATCH_MATCH : MATCH_ACCEPT;
-     MRRETURN(rrc);
+     RRETURN(rrc);
      /* Assertion brackets. Check the alternative branches in turn - the
      matching won't pass the KET for an assertion. If any one branch matches,
      the assertion is true. Lookbehind assertions have an OP_REVERSE item at the
      start of each branch to move the current point backwards, so the code at
      this level is identical to the lookahead case. When the assertion is part
      of a condition, we want to return immediately afterwards. The caller of
      this incarnation of the match() function will have set MATCH_CONDASSERT in
      md->match_function type, and one of these opcodes will be the first opcode
      that is processed. We use a local variable that is preserved over calls to
      match() to remember this case. */
      case OP_ASSERT:
-Line 1364 
 for (;;)
+Line 1524 
 for (;;)
        condassert = TRUE;
        md->match_function_type = 0;
        }
      else condassert = FALSE;
      do
        {
        RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM4);
-Line 1374 
 for (;;)
+Line 1534 
 for (;;)
          mstart = md->start_match_ptr;   /* In case \K reset it */
          break;
          }
-       if (rrc != MATCH_NOMATCH &&
-           (rrc != MATCH_THEN || md->start_match_ptr != ecode))
+       /* PCRE does not allow THEN to escape beyond an assertion; it is treated
-         RRETURN(rrc);
+       as NOMATCH. */
+       if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
        ecode += GET(ecode, 1);
        }
      while (*ecode == OP_ALT);
-     if (*ecode == OP_KET) MRRETURN(MATCH_NOMATCH);
+     if (*ecode == OP_KET) RRETURN(MATCH_NOMATCH);
      /* If checking an assertion for a condition, return MATCH_MATCH. */
-Line 1406 
 for (;;)
+Line 1568 
 for (;;)
        condassert = TRUE;
        md->match_function_type = 0;
        }
      else condassert = FALSE;
      do
        {
        RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM5);
-       if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) MRRETURN(MATCH_NOMATCH);
+       if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) RRETURN(MATCH_NOMATCH);
        if (rrc == MATCH_SKIP || rrc == MATCH_PRUNE || rrc == MATCH_COMMIT)
          {
          do ecode += GET(ecode,1); while (*ecode == OP_ALT);
          break;
          }
-       if (rrc != MATCH_NOMATCH &&
-           (rrc != MATCH_THEN || md->start_match_ptr != ecode))
+       /* PCRE does not allow THEN to escape beyond an assertion; it is treated
-         RRETURN(rrc);
+       as NOMATCH. */
+       if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
        ecode += GET(ecode,1);
        }
      while (*ecode == OP_ALT);
      if (condassert) RRETURN(MATCH_MATCH);  /* Condition assertion */
      ecode += 1 + LINK_SIZE;
      continue;
-Line 1435 
 for (;;)
+Line 1599 
 for (;;)
      back a number of characters, not bytes. */
      case OP_REVERSE:
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-     if (utf8)
+     if (utf)
        {
        i = GET(ecode, 1);
        while (i-- > 0)
          {
          eptr--;
-         if (eptr < md->start_subject) MRRETURN(MATCH_NOMATCH);
+         if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
          BACKCHAR(eptr);
          }
        }
-Line 1453 
 for (;;)
+Line 1617 
 for (;;)
        {
        eptr -= GET(ecode, 1);
-       if (eptr < md->start_subject) MRRETURN(MATCH_NOMATCH);
+       if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
        }
      /* Save the earliest consulted character, then skip to next op code */
-Line 1467 
 for (;;)
+Line 1631 
 for (;;)
      function is able to force a failure. */
      case OP_CALLOUT:
-     if (pcre_callout != NULL)
+     if (PUBL(callout) != NULL)
        {
-       pcre_callout_block cb;
+       PUBL(callout_block) cb;
-       cb.version          = 1;   /* Version 1 of the callout block */
+       cb.version          = 2;   /* Version 1 of the callout block */
        cb.callout_number   = ecode[1];
        cb.offset_vector    = md->offset_vector;
        cb.subject          = (PCRE_SPTR)md->start_subject;
-Line 1482 
 for (;;)
+Line 1646 
 for (;;)
        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);
+       cb.mark             = md->nomatch_mark;
+       if ((rrc = (*PUBL(callout))(&cb)) > 0) RRETURN(MATCH_NOMATCH);
        if (rrc < 0) RRETURN(rrc);
        }
      ecode += 2 + 2*LINK_SIZE;
-Line 1492 
 for (;;)
+Line 1657 
 for (;;)
      offset data is the offset to the starting bracket from the start of the
      whole pattern. (This is so that it works from duplicated subpatterns.)
-     If there are any capturing brackets started but not finished, we have to
+     The state of the capturing groups is preserved over recursion, and
-     save their starting points and reinstate them after the recursion. However,
+     re-instated afterwards. We don't know how many are started and not yet
-     we don't know how many such there are (offset_top records the completed
+     finished (offset_top records the completed total) so we just have to save
-     total) so we just have to save all the potential data. There may be up to
+     all the potential data. There may be up to 65535 such values, which is too
-such values, which is too large to put on the stack, but using malloc
+     large to put on the stack, but using malloc for small numbers seems
-     for small numbers seems expensive. As a compromise, the stack is used when
+     expensive. As a compromise, the stack is used when there are no more than
-     there are no more than REC_STACK_SAVE_MAX values to store; otherwise malloc
+     REC_STACK_SAVE_MAX values to store; otherwise malloc is used.
-     is used. A problem is what to do if the malloc fails ... there is no way of
-     returning to the top level with an error. Save the top REC_STACK_SAVE_MAX
-     values on the stack, and accept that the rest may be wrong.
      There are also other values that have to be saved. We use a chained
      sequence of blocks that actually live on the stack. Thanks to Robin Houston
-     for the original version of this logic. */
+     for the original version of this logic. It has, however, been hacked around
+     a lot, so he is not to blame for the current way it works. */
      case OP_RECURSE:
        {
+       recursion_info *ri;
+       int recno;
        callpat = md->start_code + GET(ecode, 1);
-       new_recursive.group_num = (callpat == md->start_code)? 0 :
+       recno = (callpat == md->start_code)? 0 :
          GET2(callpat, 1 + LINK_SIZE);
+       /* Check for repeating a recursion without advancing the subject pointer.
+       This should catch convoluted mutual recursions. (Some simple cases are
+       caught at compile time.) */
+       for (ri = md->recursive; ri != NULL; ri = ri->prevrec)
+         if (recno == ri->group_num && eptr == ri->subject_position)
+           RRETURN(PCRE_ERROR_RECURSELOOP);
        /* Add to "recursing stack" */
+       new_recursive.group_num = recno;
+       new_recursive.subject_position = eptr;
        new_recursive.prevrec = md->recursive;
        md->recursive = &new_recursive;
-       /* Find where to continue from afterwards */
+       /* Where to continue from afterwards */
        ecode += 1 + LINK_SIZE;
-       new_recursive.after_call = ecode;
-       /* Now save the offset data. */
+       /* Now save the offset data */
        new_recursive.saved_max = md->offset_end;
        if (new_recursive.saved_max <= REC_STACK_SAVE_MAX)
-Line 1531 
 for (;;)
+Line 1706 
 for (;;)
        else
          {
          new_recursive.offset_save =
-           (int *)(pcre_malloc)(new_recursive.saved_max * sizeof(int));
+           (int *)(PUBL(malloc))(new_recursive.saved_max * sizeof(int));
          if (new_recursive.offset_save == NULL) RRETURN(PCRE_ERROR_NOMEMORY);
          }
        memcpy(new_recursive.offset_save, md->offset_vector,
              new_recursive.saved_max * sizeof(int));
-       new_recursive.save_offset_top = offset_top;
-       /* OK, now we can do the recursion. For each top-level alternative we
+       /* OK, now we can do the recursion. After processing each alternative,
-       restore the offset and recursion data. */
+       restore the offset data. If there were nested recursions, md->recursive
+       might be changed, so reset it before looping. */
        DPRINTF(("Recursing into group %d\n", new_recursive.group_num));
        cbegroup = (*callpat >= OP_SBRA);
        do
          {
          if (cbegroup) md->match_function_type = MATCH_CBEGROUP;
-         RMATCH(eptr, callpat + _pcre_OP_lengths[*callpat], offset_top,
+         RMATCH(eptr, callpat + PRIV(OP_lengths)[*callpat], offset_top,
            md, eptrb, RM6);
+         memcpy(md->offset_vector, new_recursive.offset_save,
+             new_recursive.saved_max * sizeof(int));
+         md->recursive = new_recursive.prevrec;
          if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT)
            {
            DPRINTF(("Recursion matched\n"));
-           md->recursive = new_recursive.prevrec;
            if (new_recursive.offset_save != stacksave)
-             (pcre_free)(new_recursive.offset_save);
+             (PUBL(free))(new_recursive.offset_save);
-           MRRETURN(MATCH_MATCH);
+           /* Set where we got to in the subject, and reset the start in case
+           it was changed by \K. This *is* propagated back out of a recursion,
+           for Perl compatibility. */
+           eptr = md->end_match_ptr;
+           mstart = md->start_match_ptr;
+           goto RECURSION_MATCHED;        /* Exit loop; end processing */
            }
-         else if (rrc != MATCH_NOMATCH &&
-                 (rrc != MATCH_THEN || md->start_match_ptr != ecode))
+         /* PCRE does not allow THEN to escape beyond a recursion; it is treated
+         as NOMATCH. */
+         else if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN)
            {
            DPRINTF(("Recursion gave error %d\n", rrc));
            if (new_recursive.offset_save != stacksave)
-             (pcre_free)(new_recursive.offset_save);
+             (PUBL(free))(new_recursive.offset_save);
            RRETURN(rrc);
            }
          md->recursive = &new_recursive;
-         memcpy(md->offset_vector, new_recursive.offset_save,
-             new_recursive.saved_max * sizeof(int));
          callpat += GET(callpat, 1);
          }
        while (*callpat == OP_ALT);
-Line 1576 
 for (;;)
+Line 1760 
 for (;;)
        DPRINTF(("Recursion didn't match\n"));
        md->recursive = new_recursive.prevrec;
        if (new_recursive.offset_save != stacksave)
-         (pcre_free)(new_recursive.offset_save);
+         (PUBL(free))(new_recursive.offset_save);
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
-       }
-     /* Control never reaches here */
-     /* "Once" brackets are like assertion brackets except that after a match,
-     the point in the subject string is not moved back. Thus there can never be
-     a move back into the brackets. Friedl calls these "atomic" subpatterns.
-     Check the alternative branches in turn - the matching won't pass the KET
-     for this kind of subpattern. If any one branch matches, we carry on as at
-     the end of a normal bracket, leaving the subject pointer, but resetting
-     the start-of-match value in case it was changed by \K. */
-     case OP_ONCE:
-     prev = ecode;
-     saved_eptr = eptr;
-     do
-       {
-       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM7);
-       if (rrc == MATCH_MATCH)  /* Note: _not_ MATCH_ACCEPT */
-         {
-         mstart = md->start_match_ptr;
-         break;
-         }
-       if (rrc != MATCH_NOMATCH &&
-           (rrc != MATCH_THEN || md->start_match_ptr != ecode))
-         RRETURN(rrc);
-       ecode += GET(ecode,1);
-       }
-     while (*ecode == OP_ALT);
-     /* If hit the end of the group (which could be repeated), fail */
-     if (*ecode != OP_ONCE && *ecode != OP_ALT) RRETURN(MATCH_NOMATCH);
-     /* Continue as from after the assertion, updating the offsets high water
-     mark, since extracts may have been taken. */
-     do ecode += GET(ecode, 1); while (*ecode == OP_ALT);
-     offset_top = md->end_offset_top;
-     eptr = md->end_match_ptr;
-     /* For a non-repeating ket, just continue at this level. This also
-     happens for a repeating ket if no characters were matched in the group.
-     This is the forcible breaking of infinite loops as implemented in Perl
-.005. If there is an options reset, it will get obeyed in the normal
-     course of events. */
-     if (*ecode == OP_KET || eptr == saved_eptr)
-       {
-       ecode += 1+LINK_SIZE;
-       break;
        }
-     /* The repeating kets try the rest of the pattern or restart from the
+     RECURSION_MATCHED:
-     preceding bracket, in the appropriate order. The second "call" of match()
+     break;
-     uses tail recursion, to avoid using another stack frame. */
-     if (*ecode == OP_KETRMIN)
-       {
-       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM8);
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-       ecode = prev;
-       goto TAIL_RECURSE;
-       }
-     else  /* OP_KETRMAX */
-       {
-       md->match_function_type = MATCH_CBEGROUP;
-       RMATCH(eptr, prev, offset_top, md, eptrb, RM9);
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-       ecode += 1 + LINK_SIZE;
-       goto TAIL_RECURSE;
-       }
-     /* Control never gets here */
      /* An alternation is the end of a branch; scan along to find the end of the
      bracketed group and go to there. */
-Line 1665 
 for (;;)
+Line 1779 
 for (;;)
      at all - i.e. it could be ()* or ()? or even (){0} in the pattern. Brackets
      with fixed upper repeat limits are compiled as a number of copies, with the
      optional ones preceded by BRAZERO or BRAMINZERO. */
      case OP_BRAZERO:
      next = ecode + 1;
      RMATCH(eptr, next, offset_top, md, eptrb, RM10);
-Line 1673 
 for (;;)
+Line 1787 
 for (;;)
      do next += GET(next, 1); while (*next == OP_ALT);
      ecode = next + 1 + LINK_SIZE;
      break;
      case OP_BRAMINZERO:
      next = ecode + 1;
      do next += GET(next, 1); while (*next == OP_ALT);
-Line 1687 
 for (;;)
+Line 1801 
 for (;;)
      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;
-Line 1702 
 for (;;)
+Line 1816 
 for (;;)
      case OP_KET:
      case OP_KETRMIN:
      case OP_KETRMAX:
      case OP_KETRPOS:
      prev = ecode - GET(ecode, 1);
      /* If this was a group that remembered the subject start, in order to break
      infinite repeats of empty string matches, retrieve the subject start from
      the chain. Otherwise, set it NULL. */
-     if (*prev >= OP_SBRA)
+     if (*prev >= OP_SBRA || *prev == OP_ONCE)
        {
        saved_eptr = eptrb->epb_saved_eptr;   /* Value at start of group */
        eptrb = eptrb->epb_prev;              /* Backup to previous group */
        }
      else saved_eptr = NULL;
-     /* If we are at the end of an assertion group or an atomic group, stop
+     /* If we are at the end of an assertion group or a non-capturing atomic
-     matching and return MATCH_MATCH, but record the current high water mark for
+     group, stop matching and return MATCH_MATCH, but record the current high
-     use by positive assertions. We also need to record the match start in case
+     water mark for use by positive assertions. We also need to record the match
-     it was changed by \K. */
+     start in case it was changed by \K. */
-     if (*prev == OP_ASSERT || *prev == OP_ASSERT_NOT ||
+     if ((*prev >= OP_ASSERT && *prev <= OP_ASSERTBACK_NOT) ||
-         *prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT ||
+          *prev == OP_ONCE_NC)
-         *prev == OP_ONCE)
        {
-       md->end_match_ptr = eptr;      /* For ONCE */
+       md->end_match_ptr = eptr;      /* For ONCE_NC */
        md->end_offset_top = offset_top;
        md->start_match_ptr = mstart;
-       MRRETURN(MATCH_MATCH);
+       RRETURN(MATCH_MATCH);         /* Sets md->mark */
        }
      /* For capturing groups we have to check the group number back at the start
      and if necessary complete handling an extraction by setting the offsets and
-     bumping the high water mark. Note that whole-pattern recursion is coded as
+     bumping the high water mark. Whole-pattern recursion is coded as a recurse
-     a recurse into group 0, so it won't be picked up here. Instead, we catch it
+     into group 0, so it won't be picked up here. Instead, we catch it when the
-     when the OP_END is reached. Other recursion is handled here. */
+     OP_END is reached. Other recursion is handled here. We just have to record
+     the current subject position and start match pointer and give a MATCH
+     return. */
      if (*prev == OP_CBRA || *prev == OP_SCBRA ||
          *prev == OP_CBRAPOS || *prev == OP_SCBRAPOS)
-Line 1748 
 for (;;)
+Line 1863 
 for (;;)
        printf("\n");
  #endif
+       /* Handle a recursively called group. */
+       if (md->recursive != NULL && md->recursive->group_num == number)
+         {
+         md->end_match_ptr = eptr;
+         md->start_match_ptr = mstart;
+         RRETURN(MATCH_MATCH);
+         }
+       /* Deal with capturing */
        md->capture_last = number;
        if (offset >= md->offset_max) md->offset_overflow = TRUE; else
          {
+         /* If offset is greater than offset_top, it means that we are
+         "skipping" a capturing group, and that group's offsets must be marked
+         unset. In earlier versions of PCRE, all the offsets were unset at the
+         start of matching, but this doesn't work because atomic groups and
+         assertions can cause a value to be set that should later be unset.
+         Example: matching /(?>(a))b|(a)c/ against "ac". This sets group 1 as
+         part of the atomic group, but this is not on the final matching path,
+         so must be unset when 2 is set. (If there is no group 2, there is no
+         problem, because offset_top will then be 2, indicating no capture.) */
+         if (offset > offset_top)
+           {
+           register int *iptr = md->offset_vector + offset_top;
+           register int *iend = md->offset_vector + offset;
+           while (iptr < iend) *iptr++ = -1;
+           }
+         /* Now make the extraction */
          md->offset_vector[offset] =
            md->offset_vector[md->offset_end - number];
          md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
          if (offset_top <= offset) offset_top = offset + 2;
          }
-       /* Handle a recursively called group. Restore the offsets
-       appropriately and continue from after the call. */
-       if (md->recursive != NULL && md->recursive->group_num == number)
-         {
-         recursion_info *rec = md->recursive;
-         DPRINTF(("Recursion (%d) succeeded - continuing\n", number));
-         md->recursive = rec->prevrec;
-         memcpy(md->offset_vector, rec->offset_save,
-           rec->saved_max * sizeof(int));
-         offset_top = rec->save_offset_top;
-         ecode = rec->after_call;
-         break;
-         }
        }
-     /* For a non-repeating ket, just continue at this level. This also
+     /* For an ordinary non-repeating ket, just continue at this level. This
-     happens for a repeating ket if no characters were matched in the group.
+     also happens for a repeating ket if no characters were matched in the
-     This is the forcible breaking of infinite loops as implemented in Perl
+     group. This is the forcible breaking of infinite loops as implemented in
-.005. If there is an options reset, it will get obeyed in the normal
+     Perl 5.005. For a non-repeating atomic group that includes captures,
-     course of events. */
+     establish a backup point by processing the rest of the pattern at a lower
+     level. If this results in a NOMATCH return, pass MATCH_ONCE back to the
+     original OP_ONCE level, thereby bypassing intermediate backup points, but
+     resetting any captures that happened along the way. */
      if (*ecode == OP_KET || eptr == saved_eptr)
        {
-       ecode += 1 + LINK_SIZE;
+       if (*prev == OP_ONCE)
+         {
+         RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM12);
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
+         RRETURN(MATCH_ONCE);
+         }
+       ecode += 1 + LINK_SIZE;    /* Carry on at this level */
        break;
        }
      /* OP_KETRPOS is a possessive repeating ket. Remember the current position,
      and return the MATCH_KETRPOS. This makes it possible to do the repeats one
      at a time from the outer level, thus saving stack. */
      if (*ecode == OP_KETRPOS)
        {
        md->end_match_ptr = eptr;
        md->end_offset_top = offset_top;
        RRETURN(MATCH_KETRPOS);
        }
      /* The 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. */
+     an atomic group or an unlimited repeat of a group that can match an empty
+     string. */
      if (*ecode == OP_KETRMIN)
        {
-       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM12);
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM7);
        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       if (*prev == OP_ONCE)
+         {
+         RMATCH(eptr, prev, offset_top, md, eptrb, RM8);
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
+         RRETURN(MATCH_ONCE);
+         }
        if (*prev >= OP_SBRA)    /* Could match an empty string */
          {
          md->match_function_type = MATCH_CBEGROUP;
          RMATCH(eptr, prev, offset_top, md, eptrb, RM50);
          RRETURN(rrc);
          }
-Line 1816 
 for (;;)
+Line 1964 
 for (;;)
        }
      else  /* OP_KETRMAX */
        {
        if (*prev >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
        RMATCH(eptr, prev, offset_top, md, eptrb, RM13);
+       if (rrc == MATCH_ONCE && md->once_target == prev) rrc = MATCH_NOMATCH;
        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       if (*prev == OP_ONCE)
+         {
+         RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM9);
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         md->once_target = prev;
+         RRETURN(MATCH_ONCE);
+         }
        ecode += 1 + LINK_SIZE;
        goto TAIL_RECURSE;
        }
-Line 1827 
 for (;;)
+Line 1983 
 for (;;)
      /* Not multiline mode: start of subject assertion, unless notbol. */
      case OP_CIRC:
-     if (md->notbol && eptr == md->start_subject) MRRETURN(MATCH_NOMATCH);
+     if (md->notbol && eptr == md->start_subject) RRETURN(MATCH_NOMATCH);
      /* Start of subject assertion */
      case OP_SOD:
-     if (eptr != md->start_subject) MRRETURN(MATCH_NOMATCH);
+     if (eptr != md->start_subject) RRETURN(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 (md->notbol && eptr == md->start_subject) RRETURN(MATCH_NOMATCH);
      if (eptr != md->start_subject &&
          (eptr == md->end_subject || !WAS_NEWLINE(eptr)))
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
      ecode++;
      break;
      /* Start of match assertion */
      case OP_SOM:
-     if (eptr != md->start_subject + md->start_offset) MRRETURN(MATCH_NOMATCH);
+     if (eptr != md->start_subject + md->start_offset) RRETURN(MATCH_NOMATCH);
      ecode++;
      break;
-Line 1865 
 for (;;)
+Line 2021 
 for (;;)
      case OP_DOLLM:
      if (eptr < md->end_subject)
-       { if (!IS_NEWLINE(eptr)) MRRETURN(MATCH_NOMATCH); }
+       { if (!IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH); }
      else
        {
-       if (md->noteol) MRRETURN(MATCH_NOMATCH);
+       if (md->noteol) RRETURN(MATCH_NOMATCH);
        SCHECK_PARTIAL();
        }
      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->noteol) RRETURN(MATCH_NOMATCH);
      if (!md->endonly) goto ASSERT_NL_OR_EOS;
      /* ... else fall through for endonly */
-Line 1886 
 for (;;)
+Line 2042 
 for (;;)
      /* End of subject assertion (\z) */
      case OP_EOD:
-     if (eptr < md->end_subject) MRRETURN(MATCH_NOMATCH);
+     if (eptr < md->end_subject) RRETURN(MATCH_NOMATCH);
      SCHECK_PARTIAL();
      ecode++;
      break;
-Line 1897 
 for (;;)
+Line 2053 
 for (;;)
      ASSERT_NL_OR_EOS:
      if (eptr < md->end_subject &&
          (!IS_NEWLINE(eptr) || eptr != md->end_subject - md->nllen))
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
      /* Either at end of string or \n before end. */
-Line 1916 
 for (;;)
+Line 2072 
 for (;;)
        be "non-word" characters. Remember the earliest consulted character for
        partial matching. */
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-       if (utf8)
+       if (utf)
          {
          /* Get status of previous character */
          if (eptr == md->start_subject) prev_is_word = FALSE; else
            {
-           USPTR lastptr = eptr - 1;
+           PCRE_PUCHAR lastptr = eptr - 1;
-           while((*lastptr & 0xc0) == 0x80) lastptr--;
+           BACKCHAR(lastptr);
            if (lastptr < md->start_used_ptr) md->start_used_ptr = lastptr;
            GETCHAR(c, lastptr);
  #ifdef SUPPORT_UCP
-Line 1989 
 for (;;)
+Line 2145 
 for (;;)
              }
            else
  #endif
-           prev_is_word = ((md->ctypes[eptr[-1]] & ctype_word) != 0);
+           prev_is_word = MAX_255(eptr[-1])
+             && ((md->ctypes[eptr[-1]] & ctype_word) != 0);
            }
          /* Get status of next character */
-Line 2012 
 for (;;)
+Line 2169 
 for (;;)
            }
          else
  #endif
-         cur_is_word = ((md->ctypes[*eptr] & ctype_word) != 0);
+         cur_is_word = MAX_255(*eptr)
+           && ((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)
-         MRRETURN(MATCH_NOMATCH);
+         RRETURN(MATCH_NOMATCH);
        }
      break;
      /* Match a single character type; inline for speed */
      case OP_ANY:
-     if (IS_NEWLINE(eptr)) MRRETURN(MATCH_NOMATCH);
+     if (IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH);
      /* Fall through */
      case OP_ALLANY:
-     if (eptr++ >= md->end_subject)
+     if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
-       {
+       {                            /* not be updated before SCHECK_PARTIAL. */
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
-     if (utf8) while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
+     eptr++;
+ #ifdef SUPPORT_UTF
+     if (utf) ACROSSCHAR(eptr < md->end_subject, *eptr, eptr++);
+ #endif
      ecode++;
      break;
-Line 2043 
 for (;;)
+Line 2204 
 for (;;)
      any byte, even newline, independent of the setting of PCRE_DOTALL. */
      case OP_ANYBYTE:
-     if (eptr++ >= md->end_subject)
+     if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
-       {
+       {                            /* not be updated before SCHECK_PARTIAL. */
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
+     eptr++;
      ecode++;
      break;
-Line 2055 
 for (;;)
+Line 2217 
 for (;;)
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
      if (
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
         c < 256 &&
  #endif
         (md->ctypes[c] & ctype_digit) != 0
         )
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
      ecode++;
      break;
-Line 2072 
 for (;;)
+Line 2234 
 for (;;)
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
      if (
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
-        c >= 256 ||
+        c > 255 ||
  #endif
         (md->ctypes[c] & ctype_digit) == 0
         )
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
      ecode++;
      break;
-Line 2089 
 for (;;)
+Line 2251 
 for (;;)
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
      if (
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
         c < 256 &&
  #endif
         (md->ctypes[c] & ctype_space) != 0
         )
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
      ecode++;
      break;
-Line 2106 
 for (;;)
+Line 2268 
 for (;;)
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
      if (
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
-        c >= 256 ||
+        c > 255 ||
  #endif
         (md->ctypes[c] & ctype_space) == 0
         )
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
      ecode++;
      break;
-Line 2123 
 for (;;)
+Line 2285 
 for (;;)
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
      if (
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
         c < 256 &&
  #endif
         (md->ctypes[c] & ctype_word) != 0
         )
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
      ecode++;
      break;
-Line 2140 
 for (;;)
+Line 2302 
 for (;;)
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
      if (
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
-        c >= 256 ||
+        c > 255 ||
  #endif
         (md->ctypes[c] & ctype_word) == 0
         )
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
      ecode++;
      break;
-Line 2157 
 for (;;)
+Line 2319 
 for (;;)
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
      switch(c)
        {
-       default: MRRETURN(MATCH_NOMATCH);
+       default: RRETURN(MATCH_NOMATCH);
        case 0x000d:
        if (eptr < md->end_subject && *eptr == 0x0a) eptr++;
        break;
-Line 2176 
 for (;;)
+Line 2338 
 for (;;)
        case 0x0085:
        case 0x2028:
        case 0x2029:
-       if (md->bsr_anycrlf) MRRETURN(MATCH_NOMATCH);
+       if (md->bsr_anycrlf) RRETURN(MATCH_NOMATCH);
        break;
        }
      ecode++;
-Line 2186 
 for (;;)
+Line 2348 
 for (;;)
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
      switch(c)
-Line 2211 
 for (;;)
+Line 2373 
 for (;;)
        case 0x202f:    /* NARROW NO-BREAK SPACE */
        case 0x205f:    /* MEDIUM MATHEMATICAL SPACE */
        case 0x3000:    /* IDEOGRAPHIC SPACE */
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      ecode++;
      break;
-Line 2220 
 for (;;)
+Line 2382 
 for (;;)
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
      switch(c)
        {
-       default: MRRETURN(MATCH_NOMATCH);
+       default: RRETURN(MATCH_NOMATCH);
        case 0x09:      /* HT */
        case 0x20:      /* SPACE */
        case 0xa0:      /* NBSP */
-Line 2254 
 for (;;)
+Line 2416 
 for (;;)
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
      switch(c)
-Line 2267 
 for (;;)
+Line 2429 
 for (;;)
        case 0x85:      /* NEL */
        case 0x2028:    /* LINE SEPARATOR */
        case 0x2029:    /* PARAGRAPH SEPARATOR */
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      ecode++;
      break;
-Line 2276 
 for (;;)
+Line 2438 
 for (;;)
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
      switch(c)
        {
-       default: MRRETURN(MATCH_NOMATCH);
+       default: RRETURN(MATCH_NOMATCH);
        case 0x0a:      /* LF */
        case 0x0b:      /* VT */
        case 0x0c:      /* FF */
-Line 2303 
 for (;;)
+Line 2465 
 for (;;)
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
        {
-Line 2312 
 for (;;)
+Line 2474 
 for (;;)
        switch(ecode[1])
          {
          case PT_ANY:
-         if (op == OP_NOTPROP) MRRETURN(MATCH_NOMATCH);
+         if (op == OP_NOTPROP) RRETURN(MATCH_NOMATCH);
          break;
          case PT_LAMP:
          if ((prop->chartype == ucp_Lu ||
               prop->chartype == ucp_Ll ||
               prop->chartype == ucp_Lt) == (op == OP_NOTPROP))
-           MRRETURN(MATCH_NOMATCH);
+           RRETURN(MATCH_NOMATCH);
          break;
          case PT_GC:
-         if ((ecode[2] != _pcre_ucp_gentype[prop->chartype]) == (op == OP_PROP))
+         if ((ecode[2] != PRIV(ucp_gentype)[prop->chartype]) == (op == OP_PROP))
-           MRRETURN(MATCH_NOMATCH);
+           RRETURN(MATCH_NOMATCH);
          break;
          case PT_PC:
          if ((ecode[2] != prop->chartype) == (op == OP_PROP))
-           MRRETURN(MATCH_NOMATCH);
+           RRETURN(MATCH_NOMATCH);
          break;
          case PT_SC:
          if ((ecode[2] != prop->script) == (op == OP_PROP))
-           MRRETURN(MATCH_NOMATCH);
+           RRETURN(MATCH_NOMATCH);
          break;
          /* These are specials */
          case PT_ALNUM:
-         if ((_pcre_ucp_gentype[prop->chartype] == ucp_L ||
+         if ((PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
-              _pcre_ucp_gentype[prop->chartype] == ucp_N) == (op == OP_NOTPROP))
+              PRIV(ucp_gentype)[prop->chartype] == ucp_N) == (op == OP_NOTPROP))
-           MRRETURN(MATCH_NOMATCH);
+           RRETURN(MATCH_NOMATCH);
          break;
          case PT_SPACE:    /* Perl space */
-         if ((_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
+         if ((PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
               c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
                 == (op == OP_NOTPROP))
-           MRRETURN(MATCH_NOMATCH);
+           RRETURN(MATCH_NOMATCH);
          break;
          case PT_PXSPACE:  /* POSIX space */
-         if ((_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
+         if ((PRIV(ucp_gentype)[prop->chartype] == ucp_Z ||
               c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
               c == CHAR_FF || c == CHAR_CR)
                 == (op == OP_NOTPROP))
-           MRRETURN(MATCH_NOMATCH);
+           RRETURN(MATCH_NOMATCH);
          break;
          case PT_WORD:
-         if ((_pcre_ucp_gentype[prop->chartype] == ucp_L ||
+         if ((PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
-              _pcre_ucp_gentype[prop->chartype] == ucp_N ||
+              PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
               c == CHAR_UNDERSCORE) == (op == OP_NOTPROP))
-           MRRETURN(MATCH_NOMATCH);
+           RRETURN(MATCH_NOMATCH);
          break;
          /* This should never occur */
-Line 2384 
 for (;;)
+Line 2546 
 for (;;)
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
+     if (UCD_CATEGORY(c) == ucp_M) RRETURN(MATCH_NOMATCH);
+     while (eptr < md->end_subject)
        {
-       int category = UCD_CATEGORY(c);
+       int len = 1;
-       if (category == ucp_M) MRRETURN(MATCH_NOMATCH);
+       if (!utf) c = *eptr; else { GETCHARLEN(c, eptr, len); }
-       while (eptr < md->end_subject)
+       if (UCD_CATEGORY(c) != ucp_M) break;
-         {
+       eptr += len;
-         int len = 1;
-         if (!utf8) c = *eptr; else
-           {
-           GETCHARLEN(c, eptr, len);
-           }
-         category = UCD_CATEGORY(c);
-         if (category != ucp_M) break;
-         eptr += len;
-         }
        }
      ecode++;
      break;
-Line 2416 
 for (;;)
+Line 2571 
 for (;;)
      loops). */
      case OP_REF:
      case OP_REFI:
      caseless = op == OP_REFI;
      offset = GET2(ecode, 1) << 1;               /* Doubled ref number */
-     ecode += 3;
+     ecode += 1 + IMM2_SIZE;
      /* If the reference is unset, there are two possibilities:
-Line 2459 
 for (;;)
+Line 2614 
 for (;;)
        case OP_CRMINRANGE:
        minimize = (*ecode == OP_CRMINRANGE);
        min = GET2(ecode, 1);
-       max = GET2(ecode, 3);
+       max = GET2(ecode, 1 + IMM2_SIZE);
        if (max == 0) max = INT_MAX;
-       ecode += 5;
+       ecode += 1 + 2 * IMM2_SIZE;
        break;
        default:               /* No repeat follows */
        if ((length = match_ref(offset, eptr, length, md, caseless)) < 0)
          {
          CHECK_PARTIAL();
-         MRRETURN(MATCH_NOMATCH);
+         RRETURN(MATCH_NOMATCH);
          }
        eptr += length;
        continue;              /* With the main loop */
        }
      /* Handle repeated back references. If the length of the reference is
-     zero, just continue with the main loop. */
+     zero, just continue with the main loop. If the length is negative, it
+     means the reference is unset in non-Java-compatible mode. If the minimum is
+     zero, we can continue at the same level without recursion. For any other
+     minimum, carrying on will result in NOMATCH. */
      if (length == 0) continue;
+     if (length < 0 && min == 0) continue;
      /* First, ensure the minimum number of matches are present. We get back
      the length of the reference string explicitly rather than passing the
-Line 2485 
 for (;;)
+Line 2644 
 for (;;)
      for (i = 1; i <= min; i++)
        {
        int slength;
        if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
          {
          CHECK_PARTIAL();
-         MRRETURN(MATCH_NOMATCH);
+         RRETURN(MATCH_NOMATCH);
          }
        eptr += slength;
        }
-Line 2505 
 for (;;)
+Line 2664 
 for (;;)
        {
        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 (fi >= max) RRETURN(MATCH_NOMATCH);
          if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
            {
            CHECK_PARTIAL();
-           MRRETURN(MATCH_NOMATCH);
+           RRETURN(MATCH_NOMATCH);
            }
          eptr += slength;
          }
-Line 2526 
 for (;;)
+Line 2685 
 for (;;)
        pp = eptr;
        for (i = min; i < max; i++)
          {
          int slength;
          if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
            {
            CHECK_PARTIAL();
-Line 2540 
 for (;;)
+Line 2699 
 for (;;)
          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
          eptr -= length;
          }
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      /* Control never gets here */
-Line 2558 
 for (;;)
+Line 2717 
 for (;;)
      case OP_NCLASS:
      case OP_CLASS:
        {
+       /* The data variable is saved across frames, so the byte map needs to
+       be stored there. */
+ #define BYTE_MAP ((pcre_uint8 *)data)
        data = ecode + 1;                /* Save for matching */
-       ecode += 33;                     /* Advance past the item */
+       ecode += 1 + (32 / sizeof(pcre_uchar)); /* Advance past the item */
        switch (*ecode)
          {
-Line 2580 
 for (;;)
+Line 2742 
 for (;;)
          case OP_CRMINRANGE:
          minimize = (*ecode == OP_CRMINRANGE);
          min = GET2(ecode, 1);
-         max = GET2(ecode, 3);
+         max = GET2(ecode, 1 + IMM2_SIZE);
          if (max == 0) max = INT_MAX;
-         ecode += 5;
+         ecode += 1 + 2 * IMM2_SIZE;
          break;
          default:               /* No repeat follows */
-Line 2592 
 for (;;)
+Line 2754 
 for (;;)
        /* First, ensure the minimum number of matches are present. */
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-       /* UTF-8 mode */
+       if (utf)
-       if (utf8)
          {
          for (i = 1; i <= min; i++)
            {
            if (eptr >= md->end_subject)
              {
              SCHECK_PARTIAL();
-             MRRETURN(MATCH_NOMATCH);
+             RRETURN(MATCH_NOMATCH);
              }
            GETCHARINC(c, eptr);
            if (c > 255)
              {
-             if (op == OP_CLASS) MRRETURN(MATCH_NOMATCH);
+             if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
              }
            else
-             {
+             if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
-             if ((data[c/8] & (1 << (c&7))) == 0) MRRETURN(MATCH_NOMATCH);
-             }
            }
          }
        else
  #endif
-       /* Not UTF-8 mode */
+       /* Not UTF mode */
          {
          for (i = 1; i <= min; i++)
            {
            if (eptr >= md->end_subject)
              {
              SCHECK_PARTIAL();
-             MRRETURN(MATCH_NOMATCH);
+             RRETURN(MATCH_NOMATCH);
              }
            c = *eptr++;
-           if ((data[c/8] & (1 << (c&7))) == 0) MRRETURN(MATCH_NOMATCH);
+ #ifndef COMPILE_PCRE8
+           if (c > 255)
+             {
+             if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
+             }
+           else
+ #endif
+             if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
            }
          }
-Line 2640 
 for (;;)
+Line 2806 
 for (;;)
        if (minimize)
          {
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-         /* UTF-8 mode */
+         if (utf)
-         if (utf8)
            {
            for (fi = min;; fi++)
              {
              RMATCH(eptr, ecode, offset_top, md, eptrb, RM16);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max) MRRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              GETCHARINC(c, eptr);
              if (c > 255)
                {
-               if (op == OP_CLASS) MRRETURN(MATCH_NOMATCH);
+               if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
                }
              else
-               {
+               if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
-               if ((data[c/8] & (1 << (c&7))) == 0) MRRETURN(MATCH_NOMATCH);
-               }
              }
            }
          else
  #endif
-         /* Not UTF-8 mode */
+         /* Not UTF mode */
            {
            for (fi = min;; fi++)
              {
              RMATCH(eptr, ecode, offset_top, md, eptrb, RM17);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max) MRRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              c = *eptr++;
-             if ((data[c/8] & (1 << (c&7))) == 0) MRRETURN(MATCH_NOMATCH);
+ #ifndef COMPILE_PCRE8
+             if (c > 255)
+               {
+               if (op == OP_CLASS) RRETURN(MATCH_NOMATCH);
+               }
+             else
+ #endif
+               if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
              }
            }
          /* Control never gets here */
-Line 2692 
 for (;;)
+Line 2862 
 for (;;)
          {
          pp = eptr;
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-         /* UTF-8 mode */
+         if (utf)
-         if (utf8)
            {
            for (i = min; i < max; i++)
              {
-Line 2710 
 for (;;)
+Line 2879 
 for (;;)
                if (op == OP_CLASS) break;
                }
              else
-               {
+               if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) break;
-               if ((data[c/8] & (1 << (c&7))) == 0) break;
-               }
              eptr += len;
              }
            for (;;)
-Line 2725 
 for (;;)
+Line 2892 
 for (;;)
            }
          else
  #endif
-           /* Not UTF-8 mode */
+           /* Not UTF mode */
            {
            for (i = min; i < max; i++)
              {
-Line 2735 
 for (;;)
+Line 2902 
 for (;;)
                break;
                }
              c = *eptr;
-             if ((data[c/8] & (1 << (c&7))) == 0) break;
+ #ifndef COMPILE_PCRE8
+             if (c > 255)
+               {
+               if (op == OP_CLASS) break;
+               }
+             else
+ #endif
+               if ((BYTE_MAP[c/8] & (1 << (c&7))) == 0) break;
              eptr++;
              }
            while (eptr >= pp)
-Line 2746 
 for (;;)
+Line 2920 
 for (;;)
              }
            }
-         MRRETURN(MATCH_NOMATCH);
+         RRETURN(MATCH_NOMATCH);
          }
+ #undef BYTE_MAP
        }
      /* Control never gets here */
-Line 2756 
 for (;;)
+Line 2931 
 for (;;)
      when UTF-8 mode mode is supported. Nevertheless, we may not be in UTF-8
      mode, because Unicode properties are supported in non-UTF-8 mode. */
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
      case OP_XCLASS:
        {
        data = ecode + 1 + LINK_SIZE;                /* Save for matching */
-Line 2781 
 for (;;)
+Line 2956 
 for (;;)
          case OP_CRMINRANGE:
          minimize = (*ecode == OP_CRMINRANGE);
          min = GET2(ecode, 1);
-         max = GET2(ecode, 3);
+         max = GET2(ecode, 1 + IMM2_SIZE);
          if (max == 0) max = INT_MAX;
-         ecode += 5;
+         ecode += 1 + 2 * IMM2_SIZE;
          break;
          default:               /* No repeat follows */
-Line 2798 
 for (;;)
+Line 2973 
 for (;;)
          if (eptr >= md->end_subject)
            {
            SCHECK_PARTIAL();
-           MRRETURN(MATCH_NOMATCH);
+           RRETURN(MATCH_NOMATCH);
            }
          GETCHARINCTEST(c, eptr);
-         if (!_pcre_xclass(c, data)) MRRETURN(MATCH_NOMATCH);
+         if (!PRIV(xclass)(c, data, utf)) RRETURN(MATCH_NOMATCH);
          }
        /* If max == min we can continue with the main loop without the
-Line 2818 
 for (;;)
+Line 2993 
 for (;;)
            {
            RMATCH(eptr, ecode, offset_top, md, eptrb, RM20);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-           if (fi >= max) MRRETURN(MATCH_NOMATCH);
+           if (fi >= max) RRETURN(MATCH_NOMATCH);
            if (eptr >= md->end_subject)
              {
              SCHECK_PARTIAL();
-             MRRETURN(MATCH_NOMATCH);
+             RRETURN(MATCH_NOMATCH);
              }
            GETCHARINCTEST(c, eptr);
-           if (!_pcre_xclass(c, data)) MRRETURN(MATCH_NOMATCH);
+           if (!PRIV(xclass)(c, data, utf)) RRETURN(MATCH_NOMATCH);
            }
          /* Control never gets here */
          }
-Line 2843 
 for (;;)
+Line 3018 
 for (;;)
              SCHECK_PARTIAL();
              break;
              }
+ #ifdef SUPPORT_UTF
            GETCHARLENTEST(c, eptr, len);
-           if (!_pcre_xclass(c, data)) break;
+ #else
+           c = *eptr;
+ #endif
+           if (!PRIV(xclass)(c, data, utf)) break;
            eptr += len;
            }
          for(;;)
-Line 2852 
 for (;;)
+Line 3031 
 for (;;)
            RMATCH(eptr, ecode, offset_top, md, eptrb, RM21);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (eptr-- == pp) break;        /* Stop if tried at original pos */
-           if (utf8) BACKCHAR(eptr);
+ #ifdef SUPPORT_UTF
+           if (utf) BACKCHAR(eptr);
+ #endif
            }
-         MRRETURN(MATCH_NOMATCH);
+         RRETURN(MATCH_NOMATCH);
          }
        /* Control never gets here */
-Line 2864 
 for (;;)
+Line 3045 
 for (;;)
      /* Match a single character, casefully */
      case OP_CHAR:
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-     if (utf8)
+     if (utf)
        {
        length = 1;
        ecode++;
-Line 2873 
 for (;;)
+Line 3054 
 for (;;)
        if (length > md->end_subject - eptr)
          {
          CHECK_PARTIAL();             /* Not SCHECK_PARTIAL() */
-         MRRETURN(MATCH_NOMATCH);
+         RRETURN(MATCH_NOMATCH);
          }
-       while (length-- > 0) if (*ecode++ != *eptr++) MRRETURN(MATCH_NOMATCH);
+       while (length-- > 0) if (*ecode++ != *eptr++) RRETURN(MATCH_NOMATCH);
        }
      else
  #endif
+     /* Not UTF mode */
-     /* Non-UTF-8 mode */
        {
        if (md->end_subject - eptr < 1)
          {
          SCHECK_PARTIAL();            /* This one can use SCHECK_PARTIAL() */
-         MRRETURN(MATCH_NOMATCH);
+         RRETURN(MATCH_NOMATCH);
          }
-       if (ecode[1] != *eptr++) MRRETURN(MATCH_NOMATCH);
+       if (ecode[1] != *eptr++) RRETURN(MATCH_NOMATCH);
        ecode += 2;
        }
      break;
-     /* Match a single character, caselessly */
+     /* Match a single character, caselessly. If we are at the end of the
+     subject, give up immediately. */
      case OP_CHARI:
- #ifdef SUPPORT_UTF8
+     if (eptr >= md->end_subject)
-     if (utf8)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
+ #ifdef SUPPORT_UTF
+     if (utf)
        {
        length = 1;
        ecode++;
        GETCHARLEN(fc, ecode, length);
-       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. */
+       we know that its other case must also be one byte long, so we can use the
+       fast lookup table. We know that there is at least one byte left in the
+       subject. */
        if (fc < 128)
          {
-         if (md->lcc[*ecode++] != md->lcc[*eptr++]) MRRETURN(MATCH_NOMATCH);
+         if (md->lcc[fc]
+             != TABLE_GET(*eptr, md->lcc, *eptr)) RRETURN(MATCH_NOMATCH);
+         ecode++;
+         eptr++;
          }
-       /* Otherwise we must pick up the subject character */
+       /* Otherwise we must pick up the subject character. Note that we cannot
+       use the value of "length" to check for sufficient bytes left, because the
+       other case of the character may have more or fewer bytes.  */
        else
          {
-Line 2932 
 for (;;)
+Line 3120 
 for (;;)
  #ifdef SUPPORT_UCP
            if (dc != UCD_OTHERCASE(fc))
  #endif
-             MRRETURN(MATCH_NOMATCH);
+             RRETURN(MATCH_NOMATCH);
            }
          }
        }
      else
- #endif   /* SUPPORT_UTF8 */
+ #endif   /* SUPPORT_UTF */
-     /* Non-UTF-8 mode */
+     /* Not UTF mode */
        {
-       if (md->end_subject - eptr < 1)
+       if (TABLE_GET(ecode[1], md->lcc, ecode[1])
-         {
+           != TABLE_GET(*eptr, md->lcc, *eptr)) RRETURN(MATCH_NOMATCH);
-         SCHECK_PARTIAL();            /* This one can use SCHECK_PARTIAL() */
+       eptr++;
-         MRRETURN(MATCH_NOMATCH);
-         }
-       if (md->lcc[ecode[1]] != md->lcc[*eptr++]) MRRETURN(MATCH_NOMATCH);
        ecode += 2;
        }
      break;
-Line 2956 
 for (;;)
+Line 3141 
 for (;;)
      case OP_EXACT:
      case OP_EXACTI:
      min = max = GET2(ecode, 1);
-     ecode += 3;
+     ecode += 1 + IMM2_SIZE;
      goto REPEATCHAR;
      case OP_POSUPTO:
-Line 2971 
 for (;;)
+Line 3156 
 for (;;)
      min = 0;
      max = GET2(ecode, 1);
      minimize = *ecode == OP_MINUPTO || *ecode == OP_MINUPTOI;
-     ecode += 3;
+     ecode += 1 + IMM2_SIZE;
      goto REPEATCHAR;
      case OP_POSSTAR:
-Line 3019 
 for (;;)
+Line 3204 
 for (;;)
      /* Common code for all repeated single-character matches. */
      REPEATCHAR:
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-     if (utf8)
+     if (utf)
        {
        length = 1;
        charptr = ecode;
-Line 3036 
 for (;;)
+Line 3221 
 for (;;)
          unsigned int othercase;
          if (op >= OP_STARI &&     /* Caseless */
              (othercase = UCD_OTHERCASE(fc)) != fc)
-           oclength = _pcre_ord2utf8(othercase, occhars);
+           oclength = PRIV(ord2utf)(othercase, occhars);
          else oclength = 0;
  #endif  /* SUPPORT_UCP */
          for (i = 1; i <= min; i++)
            {
            if (eptr <= md->end_subject - length &&
-             memcmp(eptr, charptr, length) == 0) eptr += length;
+             memcmp(eptr, charptr, IN_UCHARS(length)) == 0) eptr += length;
  #ifdef SUPPORT_UCP
            else if (oclength > 0 &&
                     eptr <= md->end_subject - oclength &&
-                    memcmp(eptr, occhars, oclength) == 0) eptr += oclength;
+                    memcmp(eptr, occhars, IN_UCHARS(oclength)) == 0) eptr += oclength;
  #endif  /* SUPPORT_UCP */
            else
              {
              CHECK_PARTIAL();
-             MRRETURN(MATCH_NOMATCH);
+             RRETURN(MATCH_NOMATCH);
              }
            }
-Line 3064 
 for (;;)
+Line 3249 
 for (;;)
              {
              RMATCH(eptr, ecode, offset_top, md, eptrb, RM22);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max) MRRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
              if (eptr <= md->end_subject - length &&
-               memcmp(eptr, charptr, length) == 0) eptr += length;
+               memcmp(eptr, charptr, IN_UCHARS(length)) == 0) eptr += length;
  #ifdef SUPPORT_UCP
              else if (oclength > 0 &&
                       eptr <= md->end_subject - oclength &&
-                      memcmp(eptr, occhars, oclength) == 0) eptr += oclength;
+                      memcmp(eptr, occhars, IN_UCHARS(oclength)) == 0) eptr += oclength;
  #endif  /* SUPPORT_UCP */
              else
                {
                CHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              }
            /* Control never gets here */
-Line 3087 
 for (;;)
+Line 3272 
 for (;;)
            for (i = min; i < max; i++)
              {
              if (eptr <= md->end_subject - length &&
-                 memcmp(eptr, charptr, length) == 0) eptr += length;
+                 memcmp(eptr, charptr, IN_UCHARS(length)) == 0) eptr += length;
  #ifdef SUPPORT_UCP
              else if (oclength > 0 &&
                       eptr <= md->end_subject - oclength &&
-                      memcmp(eptr, occhars, oclength) == 0) eptr += oclength;
+                      memcmp(eptr, occhars, IN_UCHARS(oclength)) == 0) eptr += oclength;
  #endif  /* SUPPORT_UCP */
              else
                {
-Line 3106 
 for (;;)
+Line 3291 
 for (;;)
              {
              RMATCH(eptr, ecode, offset_top, md, eptrb, RM23);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (eptr == pp) { MRRETURN(MATCH_NOMATCH); }
+             if (eptr == pp) { RRETURN(MATCH_NOMATCH); }
  #ifdef SUPPORT_UCP
              eptr--;
              BACKCHAR(eptr);
-Line 3123 
 for (;;)
+Line 3308 
 for (;;)
        value of fc will always be < 128. */
        }
      else
- #endif  /* SUPPORT_UTF8 */
+ #endif  /* SUPPORT_UTF */
+       /* When not in UTF-8 mode, load a single-byte character. */
+       fc = *ecode++;
-     /* When not in UTF-8 mode, load a single-byte character. */
+     /* The value of fc at this point is always one character, though we may
+     or may not be in UTF mode. The code is duplicated for the caseless and
-     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
      caseful cases, for speed, since matching characters is likely to be quite
      common. First, ensure the minimum number of matches are present. If min =
      max, continue at the same level without recursing. Otherwise, if
-Line 3143 
 for (;;)
+Line 3326 
 for (;;)
      if (op >= OP_STARI)  /* Caseless */
        {
-       fc = md->lcc[fc];
+ #ifdef COMPILE_PCRE8
+       /* fc must be < 128 if UTF is enabled. */
+       foc = md->fcc[fc];
+ #else
+ #ifdef SUPPORT_UTF
+ #ifdef SUPPORT_UCP
+       if (utf && fc > 127)
+         foc = UCD_OTHERCASE(fc);
+ #else
+       if (utf && fc > 127)
+         foc = fc;
+ #endif /* SUPPORT_UCP */
+       else
+ #endif /* SUPPORT_UTF */
+         foc = TABLE_GET(fc, md->fcc, fc);
+ #endif /* COMPILE_PCRE8 */
        for (i = 1; i <= min; i++)
          {
          if (eptr >= md->end_subject)
            {
            SCHECK_PARTIAL();
-           MRRETURN(MATCH_NOMATCH);
+           RRETURN(MATCH_NOMATCH);
            }
-         if (fc != md->lcc[*eptr++]) MRRETURN(MATCH_NOMATCH);
+         if (fc != *eptr && foc != *eptr) RRETURN(MATCH_NOMATCH);
+         eptr++;
          }
        if (min == max) continue;
        if (minimize)
-Line 3160 
 for (;;)
+Line 3360 
 for (;;)
            {
            RMATCH(eptr, ecode, offset_top, md, eptrb, RM24);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-           if (fi >= max) MRRETURN(MATCH_NOMATCH);
+           if (fi >= max) RRETURN(MATCH_NOMATCH);
            if (eptr >= md->end_subject)
              {
              SCHECK_PARTIAL();
-             MRRETURN(MATCH_NOMATCH);
+             RRETURN(MATCH_NOMATCH);
              }
-           if (fc != md->lcc[*eptr++]) MRRETURN(MATCH_NOMATCH);
+           if (fc != *eptr && foc != *eptr) RRETURN(MATCH_NOMATCH);
+           eptr++;
            }
          /* Control never gets here */
          }
-Line 3180 
 for (;;)
+Line 3381 
 for (;;)
              SCHECK_PARTIAL();
              break;
              }
-           if (fc != md->lcc[*eptr]) break;
+           if (fc != *eptr && foc != *eptr) break;
            eptr++;
            }
-Line 3192 
 for (;;)
+Line 3393 
 for (;;)
            eptr--;
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            }
-         MRRETURN(MATCH_NOMATCH);
+         RRETURN(MATCH_NOMATCH);
          }
        /* Control never gets here */
        }
-Line 3206 
 for (;;)
+Line 3407 
 for (;;)
          if (eptr >= md->end_subject)
            {
            SCHECK_PARTIAL();
-           MRRETURN(MATCH_NOMATCH);
+           RRETURN(MATCH_NOMATCH);
            }
-         if (fc != *eptr++) MRRETURN(MATCH_NOMATCH);
+         if (fc != *eptr++) RRETURN(MATCH_NOMATCH);
          }
        if (min == max) continue;
-Line 3219 
 for (;;)
+Line 3420 
 for (;;)
            {
            RMATCH(eptr, ecode, offset_top, md, eptrb, RM26);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-           if (fi >= max) MRRETURN(MATCH_NOMATCH);
+           if (fi >= max) RRETURN(MATCH_NOMATCH);
            if (eptr >= md->end_subject)
              {
              SCHECK_PARTIAL();
-             MRRETURN(MATCH_NOMATCH);
+             RRETURN(MATCH_NOMATCH);
              }
-           if (fc != *eptr++) MRRETURN(MATCH_NOMATCH);
+           if (fc != *eptr++) RRETURN(MATCH_NOMATCH);
            }
          /* Control never gets here */
          }
-Line 3250 
 for (;;)
+Line 3451 
 for (;;)
            eptr--;
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            }
-         MRRETURN(MATCH_NOMATCH);
+         RRETURN(MATCH_NOMATCH);
          }
        }
      /* Control never gets here */
-Line 3259 
 for (;;)
+Line 3460 
 for (;;)
      checking can be multibyte. */
      case OP_NOT:
      case OP_NOTI:
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-       MRRETURN(MATCH_NOMATCH);
+       RRETURN(MATCH_NOMATCH);
        }
      ecode++;
      GETCHARINCTEST(c, eptr);
      if (op == OP_NOTI)         /* The caseless case */
        {
- #ifdef SUPPORT_UTF8
+       register int ch, och;
-       if (c < 256)
+       ch = *ecode++;
- #endif
+ #ifdef COMPILE_PCRE8
-       c = md->lcc[c];
+       /* ch must be < 128 if UTF is enabled. */
-       if (md->lcc[*ecode++] == c) MRRETURN(MATCH_NOMATCH);
+       och = md->fcc[ch];
+ #else
+ #ifdef SUPPORT_UTF
+ #ifdef SUPPORT_UCP
+       if (utf && ch > 127)
+         och = UCD_OTHERCASE(ch);
+ #else
+       if (utf && ch > 127)
+         och = ch;
+ #endif /* SUPPORT_UCP */
+       else
+ #endif /* SUPPORT_UTF */
+         och = TABLE_GET(ch, md->fcc, ch);
+ #endif /* COMPILE_PCRE8 */
+       if (ch == c || och == c) RRETURN(MATCH_NOMATCH);
        }
      else    /* Caseful */
        {
-       if (*ecode++ == c) MRRETURN(MATCH_NOMATCH);
+       if (*ecode++ == c) RRETURN(MATCH_NOMATCH);
        }
      break;
-Line 3291 
 for (;;)
+Line 3506 
 for (;;)
      case OP_NOTEXACT:
      case OP_NOTEXACTI:
      min = max = GET2(ecode, 1);
-     ecode += 3;
+     ecode += 1 + IMM2_SIZE;
      goto REPEATNOTCHAR;
      case OP_NOTUPTO:
-Line 3301 
 for (;;)
+Line 3516 
 for (;;)
      min = 0;
      max = GET2(ecode, 1);
      minimize = *ecode == OP_NOTMINUPTO || *ecode == OP_NOTMINUPTOI;
-     ecode += 3;
+     ecode += 1 + IMM2_SIZE;
      goto REPEATNOTCHAR;
      case OP_NOTPOSSTAR:
-Line 3333 
 for (;;)
+Line 3548 
 for (;;)
      possessive = TRUE;
      min = 0;
      max = GET2(ecode, 1);
-     ecode += 3;
+     ecode += 1 + IMM2_SIZE;
      goto REPEATNOTCHAR;
      case OP_NOTSTAR:
-Line 3372 
 for (;;)
+Line 3587 
 for (;;)
      if (op >= OP_NOTSTARI)     /* Caseless */
        {
-       fc = md->lcc[fc];
+ #ifdef COMPILE_PCRE8
+       /* fc must be < 128 if UTF is enabled. */
+       foc = md->fcc[fc];
+ #else
+ #ifdef SUPPORT_UTF
+ #ifdef SUPPORT_UCP
+       if (utf && fc > 127)
+         foc = UCD_OTHERCASE(fc);
+ #else
+       if (utf && fc > 127)
+         foc = fc;
+ #endif /* SUPPORT_UCP */
+       else
+ #endif /* SUPPORT_UTF */
+         foc = TABLE_GET(fc, md->fcc, fc);
+ #endif /* COMPILE_PCRE8 */
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-       /* UTF-8 mode */
+       if (utf)
-       if (utf8)
          {
          register unsigned int d;
          for (i = 1; i <= min; i++)
-Line 3384 
 for (;;)
+Line 3613 
 for (;;)
            if (eptr >= md->end_subject)
              {
              SCHECK_PARTIAL();
-             MRRETURN(MATCH_NOMATCH);
+             RRETURN(MATCH_NOMATCH);
              }
            GETCHARINC(d, eptr);
-           if (d < 256) d = md->lcc[d];
+           if (fc == d || foc == d) RRETURN(MATCH_NOMATCH);
-           if (fc == d) MRRETURN(MATCH_NOMATCH);
            }
          }
        else
  #endif
+       /* Not UTF mode */
-       /* Not UTF-8 mode */
          {
          for (i = 1; i <= min; i++)
            {
            if (eptr >= md->end_subject)
              {
              SCHECK_PARTIAL();
-             MRRETURN(MATCH_NOMATCH);
+             RRETURN(MATCH_NOMATCH);
              }
-           if (fc == md->lcc[*eptr++]) MRRETURN(MATCH_NOMATCH);
+           if (fc == *eptr || foc == *eptr) RRETURN(MATCH_NOMATCH);
+           eptr++;
            }
          }
-Line 3411 
 for (;;)
+Line 3639 
 for (;;)
        if (minimize)
          {
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-         /* UTF-8 mode */
+         if (utf)
-         if (utf8)
            {
            register unsigned int d;
            for (fi = min;; fi++)
              {
              RMATCH(eptr, ecode, offset_top, md, eptrb, RM28);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max) MRRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              GETCHARINC(d, eptr);
-             if (d < 256) d = md->lcc[d];
+             if (fc == d || foc == d) RRETURN(MATCH_NOMATCH);
-             if (fc == d) MRRETURN(MATCH_NOMATCH);
              }
            }
          else
  #endif
-         /* Not UTF-8 mode */
+         /* Not UTF mode */
            {
            for (fi = min;; fi++)
              {
              RMATCH(eptr, ecode, offset_top, md, eptrb, RM29);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max) MRRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
-             if (fc == md->lcc[*eptr++]) MRRETURN(MATCH_NOMATCH);
+             if (fc == *eptr || foc == *eptr) RRETURN(MATCH_NOMATCH);
+             eptr++;
              }
            }
          /* Control never gets here */
-Line 3457 
 for (;;)
+Line 3684 
 for (;;)
          {
          pp = eptr;
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-         /* UTF-8 mode */
+         if (utf)
-         if (utf8)
            {
            register unsigned int d;
            for (i = min; i < max; i++)
-Line 3471 
 for (;;)
+Line 3697 
 for (;;)
                break;
                }
              GETCHARLEN(d, eptr, len);
-             if (d < 256) d = md->lcc[d];
+             if (fc == d || foc == d) break;
-             if (fc == d) break;
              eptr += len;
              }
          if (possessive) continue;
-Line 3486 
 for (;;)
+Line 3711 
 for (;;)
            }
          else
  #endif
-         /* Not UTF-8 mode */
+         /* Not UTF mode */
            {
            for (i = min; i < max; i++)
              {
-Line 3495 
 for (;;)
+Line 3720 
 for (;;)
                SCHECK_PARTIAL();
                break;
                }
-             if (fc == md->lcc[*eptr]) break;
+             if (fc == *eptr || foc == *eptr) break;
              eptr++;
              }
            if (possessive) continue;
-Line 3507 
 for (;;)
+Line 3732 
 for (;;)
              }
            }
-         MRRETURN(MATCH_NOMATCH);
+         RRETURN(MATCH_NOMATCH);
          }
        /* Control never gets here */
        }
-Line 3516 
 for (;;)
+Line 3741 
 for (;;)
      else
        {
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-       /* UTF-8 mode */
+       if (utf)
-       if (utf8)
          {
          register unsigned int d;
          for (i = 1; i <= min; i++)
-Line 3526 
 for (;;)
+Line 3750 
 for (;;)
            if (eptr >= md->end_subject)
              {
              SCHECK_PARTIAL();
-             MRRETURN(MATCH_NOMATCH);
+             RRETURN(MATCH_NOMATCH);
              }
            GETCHARINC(d, eptr);
-           if (fc == d) MRRETURN(MATCH_NOMATCH);
+           if (fc == d) RRETURN(MATCH_NOMATCH);
            }
          }
        else
  #endif
-       /* Not UTF-8 mode */
+       /* Not UTF mode */
          {
          for (i = 1; i <= min; i++)
            {
            if (eptr >= md->end_subject)
              {
              SCHECK_PARTIAL();
-             MRRETURN(MATCH_NOMATCH);
+             RRETURN(MATCH_NOMATCH);
              }
-           if (fc == *eptr++) MRRETURN(MATCH_NOMATCH);
+           if (fc == *eptr++) RRETURN(MATCH_NOMATCH);
            }
          }
-Line 3551 
 for (;;)
+Line 3775 
 for (;;)
        if (minimize)
          {
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-         /* UTF-8 mode */
+         if (utf)
-         if (utf8)
            {
            register unsigned int d;
            for (fi = min;; fi++)
              {
              RMATCH(eptr, ecode, offset_top, md, eptrb, RM32);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max) MRRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              GETCHARINC(d, eptr);
-             if (fc == d) MRRETURN(MATCH_NOMATCH);
+             if (fc == d) RRETURN(MATCH_NOMATCH);
              }
            }
          else
  #endif
-         /* Not UTF-8 mode */
+         /* Not UTF mode */
            {
            for (fi = min;; fi++)
              {
              RMATCH(eptr, ecode, offset_top, md, eptrb, RM33);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max) MRRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
-             if (fc == *eptr++) MRRETURN(MATCH_NOMATCH);
+             if (fc == *eptr++) RRETURN(MATCH_NOMATCH);
              }
            }
          /* Control never gets here */
-Line 3596 
 for (;;)
+Line 3819 
 for (;;)
          {
          pp = eptr;
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-         /* UTF-8 mode */
+         if (utf)
-         if (utf8)
            {
            register unsigned int d;
            for (i = min; i < max; i++)
-Line 3624 
 for (;;)
+Line 3846 
 for (;;)
            }
          else
  #endif
-         /* Not UTF-8 mode */
+         /* Not UTF mode */
            {
            for (i = min; i < max; i++)
              {
-Line 3645 
 for (;;)
+Line 3867 
 for (;;)
              }
            }
-         MRRETURN(MATCH_NOMATCH);
+         RRETURN(MATCH_NOMATCH);
          }
        }
      /* Control never gets here */
-Line 3657 
 for (;;)
+Line 3879 
 for (;;)
      case OP_TYPEEXACT:
      min = max = GET2(ecode, 1);
      minimize = TRUE;
-     ecode += 3;
+     ecode += 1 + IMM2_SIZE;
      goto REPEATTYPE;
      case OP_TYPEUPTO:
-Line 3665 
 for (;;)
+Line 3887 
 for (;;)
      min = 0;
      max = GET2(ecode, 1);
      minimize = *ecode == OP_TYPEMINUPTO;
-     ecode += 3;
+     ecode += 1 + IMM2_SIZE;
      goto REPEATTYPE;
      case OP_TYPEPOSSTAR:
-Line 3693 
 for (;;)
+Line 3915 
 for (;;)
      possessive = TRUE;
      min = 0;
      max = GET2(ecode, 1);
-     ecode += 3;
+     ecode += 1 + IMM2_SIZE;
      goto REPEATTYPE;
      case OP_TYPESTAR:
-Line 3739 
 for (;;)
+Line 3961 
 for (;;)
          switch(prop_type)
            {
            case PT_ANY:
-           if (prop_fail_result) MRRETURN(MATCH_NOMATCH);
+           if (prop_fail_result) RRETURN(MATCH_NOMATCH);
            for (i = 1; i <= min; i++)
              {
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
              }
-Line 3754 
 for (;;)
+Line 3976 
 for (;;)
            case PT_LAMP:
            for (i = 1; i <= min; i++)
              {
+             int chartype;
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_chartype = UCD_CHARTYPE(c);
+             chartype = UCD_CHARTYPE(c);
-             if ((prop_chartype == ucp_Lu ||
+             if ((chartype == ucp_Lu ||
-                  prop_chartype == ucp_Ll ||
+                  chartype == ucp_Ll ||
-                  prop_chartype == ucp_Lt) == prop_fail_result)
+                  chartype == ucp_Lt) == prop_fail_result)
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
              }
            break;
-Line 3774 
 for (;;)
+Line 3997 
 for (;;)
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == prop_value) == prop_fail_result)
-             if ((prop_category == prop_value) == prop_fail_result)
+               RRETURN(MATCH_NOMATCH);
-               MRRETURN(MATCH_NOMATCH);
              }
            break;
-Line 3789 
 for (;;)
+Line 4011 
 for (;;)
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_chartype = UCD_CHARTYPE(c);
+             if ((UCD_CHARTYPE(c) == prop_value) == prop_fail_result)
-             if ((prop_chartype == prop_value) == prop_fail_result)
+               RRETURN(MATCH_NOMATCH);
-               MRRETURN(MATCH_NOMATCH);
              }
            break;
-Line 3804 
 for (;;)
+Line 4025 
 for (;;)
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_script = UCD_SCRIPT(c);
+             if ((UCD_SCRIPT(c) == prop_value) == prop_fail_result)
-             if ((prop_script == prop_value) == prop_fail_result)
+               RRETURN(MATCH_NOMATCH);
-               MRRETURN(MATCH_NOMATCH);
              }
            break;
            case PT_ALNUM:
            for (i = 1; i <= min; i++)
              {
+             int category;
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             category = UCD_CATEGORY(c);
-             if ((prop_category == ucp_L || prop_category == ucp_N)
+             if ((category == ucp_L || category == ucp_N) == prop_fail_result)
-                    == prop_fail_result)
+               RRETURN(MATCH_NOMATCH);
-               MRRETURN(MATCH_NOMATCH);
              }
            break;
-Line 3835 
 for (;;)
+Line 4055 
 for (;;)
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
-             if ((prop_category == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
                   c == CHAR_FF || c == CHAR_CR)
                     == prop_fail_result)
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
              }
            break;
-Line 3852 
 for (;;)
+Line 4071 
 for (;;)
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
-             if ((prop_category == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
                   c == CHAR_VT || c == CHAR_FF || c == CHAR_CR)
                     == prop_fail_result)
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
              }
            break;
            case PT_WORD:
            for (i = 1; i <= min; i++)
              {
+             int category;
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             category = UCD_CATEGORY(c);
-             if ((prop_category == ucp_L || prop_category == ucp_N ||
+             if ((category == ucp_L || category == ucp_N || c == CHAR_UNDERSCORE)
-                  c == CHAR_UNDERSCORE)
                     == prop_fail_result)
-               MRRETURN(MATCH_NOMATCH);
+               RRETURN(MATCH_NOMATCH);
              }
            break;
-Line 3897 
 for (;;)
+Line 4115 
 for (;;)
            if (eptr >= md->end_subject)
              {
              SCHECK_PARTIAL();
-             MRRETURN(MATCH_NOMATCH);
+             RRETURN(MATCH_NOMATCH);
              }
            GETCHARINCTEST(c, eptr);
-           prop_category = UCD_CATEGORY(c);
+           if (UCD_CATEGORY(c) == ucp_M) RRETURN(MATCH_NOMATCH);
-           if (prop_category == ucp_M) MRRETURN(MATCH_NOMATCH);
            while (eptr < md->end_subject)
              {
              int len = 1;
-             if (!utf8) c = *eptr;
+             if (!utf) c = *eptr; else { GETCHARLEN(c, eptr, len); }
-               else { GETCHARLEN(c, eptr, len); }
+             if (UCD_CATEGORY(c) != ucp_M) break;
-             prop_category = UCD_CATEGORY(c);
-             if (prop_category != ucp_M) break;
              eptr += len;
              }
            }
-Line 3919 
 for (;;)
+Line 4134 
 for (;;)
  /* Handle all other cases when the coding is UTF-8 */
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-       if (utf8) switch(ctype)
+       if (utf) switch(ctype)
          {
          case OP_ANY:
          for (i = 1; i <= min; i++)
-Line 3928 
 for (;;)
+Line 4143 
 for (;;)
            if (eptr >= md->end_subject)
              {
              SCHECK_PARTIAL();
-             MRRETURN(MATCH_NOMATCH);
+             RRETURN(MATCH_NOMATCH);
              }
-           if (IS_NEWLINE(eptr)) MRRETURN(MATCH_NOMATCH);
+           if (IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH);
            eptr++;
-           while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
+           ACROSSCHAR(eptr < md->end_subject, *eptr, eptr++);
            }
          break;
-Line 3942 
 for (;;)
+Line 4157 
 for (;;)
            if (eptr >= md->end_subject)