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

Diff of /code/trunk/pcre_exec.c

Parent Directory | Revision Log | View Patch Patch

-revision 87 by nigel,
Sat Feb 24 21:41:21 2007 UTC
+revision 443 by ph10,
Sun Sep 13 16:00:08 2009 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2006 University of Cambridge
+            Copyright (c) 1997-2009 University of Cambridge
  -----------------------------------------------------------------------------
  Redistribution and use in source and binary forms, with or without
 Line 42 
 POSSIBILITY OF SUCH DAMAGE.
  pattern matching using an NFA algorithm, trying to mimic Perl as closely as
  possible. There are also some static supporting functions. */
+ #ifdef HAVE_CONFIG_H
+ #include "config.h"
+ #endif
+ #define NLBLOCK md             /* Block containing newline information */
+ #define PSSTART start_subject  /* Field containing processed string start */
+ #define PSEND   end_subject    /* Field containing processed string end */
  #include "pcre_internal.h"
+ /* Undefine some potentially clashing cpp symbols */
- /* Structure for building a chain of data that actually lives on the
+ #undef min
- stack, for holding the values of the subject pointer at the start of each
+ #undef max
- subpattern, so as to detect when an empty string has been matched by a
- subpattern - to break infinite loops. When NO_RECURSE is set, these blocks
- are on the heap, not on the stack. */
- typedef struct eptrblock {
-   struct eptrblock *epb_prev;
-   USPTR epb_saved_eptr;
- } eptrblock;
  /* Flag bits for the match() function */
- #define match_condassert   0x01    /* Called to check a condition assertion */
+ #define match_condassert     0x01  /* Called to check a condition assertion */
- #define match_isgroup      0x02    /* Set if start of bracketed group */
+ #define match_cbegroup       0x02  /* Could-be-empty unlimited repeat group */
  /* Non-error returns from the match() function. Error returns are externally
  defined PCRE_ERROR_xxx codes, which are all negative. */
 Line 68 
 defined PCRE_ERROR_xxx codes, which are
  #define MATCH_MATCH        1
  #define MATCH_NOMATCH      0
+ /* Special internal returns from the match() function. Make them sufficiently
+ negative to avoid the external error codes. */
+ #define MATCH_COMMIT       (-999)
+ #define MATCH_PRUNE        (-998)
+ #define MATCH_SKIP         (-997)
+ #define MATCH_THEN         (-996)
  /* Maximum number of ints of offset to save on the stack for recursive calls.
  If the offset vector is bigger, malloc is used. This should be a multiple of 3,
  because the offset vector is always a multiple of 3 long. */
-Line 101 
 Returns:     nothing
+Line 109 
 Returns:     nothing
  static void
  pchars(const uschar *p, int length, BOOL is_subject, match_data *md)
  {
- int c;
+ unsigned int c;
  if (is_subject && length > md->end_subject - p) length = md->end_subject - p;
  while (length-- > 0)
    if (isprint(c = *(p++))) printf("%c", c); else printf("\\x%02x", c);
-Line 150 
 printf("\n");
+Line 158 
 printf("\n");
  if (length > md->end_subject - eptr) return FALSE;
- /* Separate the caselesss case for speed */
+ /* Separate the caseless case for speed. In UTF-8 mode we can only do this
+ properly if Unicode properties are supported. Otherwise, we can check only
+ ASCII characters. */
  if ((ims & PCRE_CASELESS) != 0)
    {
+ #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UCP
+   if (md->utf8)
+     {
+     USPTR endptr = eptr + length;
+     while (eptr < endptr)
+       {
+       int c, d;
+       GETCHARINC(c, eptr);
+       GETCHARINC(d, p);
+       if (c != d && c != UCD_OTHERCASE(d)) return FALSE;
+       }
+     }
+   else
+ #endif
+ #endif
+   /* The same code works when not in UTF-8 mode and in UTF-8 mode when there
+   is no UCP support. */
    while (length-- > 0)
-     if (md->lcc[*p++] != md->lcc[*eptr++]) return FALSE;
+     { if (md->lcc[*p++] != md->lcc[*eptr++]) return FALSE; }
    }
+ /* In the caseful case, we can just compare the bytes, whether or not we
+ are in UTF-8 mode. */
  else
    { while (length-- > 0) if (*p++ != *eptr++) return FALSE; }
-Line 186 
 calls by keeping local variables that ne
+Line 220 
 calls by keeping local variables that ne
  obtained from malloc() instead instead of on the stack. Macros are used to
  achieve this so that the actual code doesn't look very different to what it
  always used to.
+ The original heap-recursive code used longjmp(). However, it seems that this
+ can be very slow on some operating systems. Following a suggestion from Stan
+ Switzer, the use of longjmp() has been abolished, at the cost of having to
+ provide a unique number for each call to RMATCH. There is no way of generating
+ a sequence of numbers at compile time in C. I have given them names, to make
+ them stand out more clearly.
+ Crude tests on x86 Linux show a small speedup of around 5-8%. However, on
+ FreeBSD, avoiding longjmp() more than halves the time taken to run the standard
+ tests. Furthermore, not using longjmp() means that local dynamic variables
+ don't have indeterminate values; this has meant that the frame size can be
+ reduced because the result can be "passed back" by straight setting of the
+ variable instead of being passed in the frame.
  ****************************************************************************
  ***************************************************************************/
+ /* Numbers for RMATCH calls. When this list is changed, the code at HEAP_RETURN
+ below must be updated in sync.  */
+ enum { RM1=1, RM2,  RM3,  RM4,  RM5,  RM6,  RM7,  RM8,  RM9,  RM10,
+        RM11,  RM12, RM13, RM14, RM15, RM16, RM17, RM18, RM19, RM20,
+        RM21,  RM22, RM23, RM24, RM25, RM26, RM27, RM28, RM29, RM30,
+        RM31,  RM32, RM33, RM34, RM35, RM36, RM37, RM38, RM39, RM40,
+        RM41,  RM42, RM43, RM44, RM45, RM46, RM47, RM48, RM49, RM50,
+        RM51,  RM52, RM53, RM54 };
  /* These versions of the macros use the stack, as normal. There are debugging
- versions and production versions. */
+ versions and production versions. Note that the "rw" argument of RMATCH isn't
+ actuall used in this definition. */
  #ifndef NO_RECURSE
  #define REGISTER register
  #ifdef DEBUG
- #define RMATCH(rx,ra,rb,rc,rd,re,rf,rg) \
+ #define RMATCH(ra,rb,rc,rd,re,rf,rg,rw) \
    { \
    printf("match() called in line %d\n", __LINE__); \
-   rx = match(ra,rb,rc,rd,re,rf,rg,rdepth+1); \
+   rrc = match(ra,rb,mstart,rc,rd,re,rf,rg,rdepth+1); \
    printf("to line %d\n", __LINE__); \
    }
  #define RRETURN(ra) \
-Line 208 
 versions and production versions. */
+Line 267 
 versions and production versions. */
    return ra; \
    }
  #else
- #define RMATCH(rx,ra,rb,rc,rd,re,rf,rg) \
+ #define RMATCH(ra,rb,rc,rd,re,rf,rg,rw) \
-   rx = match(ra,rb,rc,rd,re,rf,rg,rdepth+1)
+   rrc = match(ra,rb,mstart,rc,rd,re,rf,rg,rdepth+1)
  #define RRETURN(ra) return ra
  #endif
  #else
- /* These versions of the macros manage a private stack on the heap. Note
+ /* These versions of the macros manage a private stack on the heap. Note that
- that the rd argument of RMATCH isn't actually used. It's the md argument of
+ the "rd" argument of RMATCH isn't actually used in this definition. It's the md
- match(), which never changes. */
+ argument of match(), which never changes. */
  #define REGISTER
- #define RMATCH(rx,ra,rb,rc,rd,re,rf,rg)\
+ #define RMATCH(ra,rb,rc,rd,re,rf,rg,rw)\
    {\
    heapframe *newframe = (pcre_stack_malloc)(sizeof(heapframe));\
-   if (setjmp(frame->Xwhere) == 0)\
+   frame->Xwhere = rw; \
-     {\
+   newframe->Xeptr = ra;\
-     newframe->Xeptr = ra;\
+   newframe->Xecode = rb;\
-     newframe->Xecode = rb;\
+   newframe->Xmstart = mstart;\
-     newframe->Xoffset_top = rc;\
+   newframe->Xoffset_top = rc;\
-     newframe->Xims = re;\
+   newframe->Xims = re;\
-     newframe->Xeptrb = rf;\
+   newframe->Xeptrb = rf;\
-     newframe->Xflags = rg;\
+   newframe->Xflags = rg;\
-     newframe->Xrdepth = frame->Xrdepth + 1;\
+   newframe->Xrdepth = frame->Xrdepth + 1;\
-     newframe->Xprevframe = frame;\
+   newframe->Xprevframe = frame;\
-     frame = newframe;\
+   frame = newframe;\
-     DPRINTF(("restarting from line %d\n", __LINE__));\
+   DPRINTF(("restarting from line %d\n", __LINE__));\
-     goto HEAP_RECURSE;\
+   goto HEAP_RECURSE;\
-     }\
+   L_##rw:\
-   else\
+   DPRINTF(("jumped back to line %d\n", __LINE__));\
-     {\
-     DPRINTF(("longjumped back to line %d\n", __LINE__));\
-     frame = md->thisframe;\
-     rx = frame->Xresult;\
-     }\
    }
  #define RRETURN(ra)\
-Line 254 
 match(), which never changes. */
+Line 308 
 match(), which never changes. */
    (pcre_stack_free)(newframe);\
    if (frame != NULL)\
      {\
-     frame->Xresult = ra;\
+     rrc = ra;\
-     md->thisframe = frame;\
+     goto HEAP_RETURN;\
-     longjmp(frame->Xwhere, 1);\
      }\
    return ra;\
    }
-Line 269 
 typedef struct heapframe {
+Line 322 
 typedef struct heapframe {
    /* Function arguments that may change */
-   const uschar *Xeptr;
+   USPTR Xeptr;
    const uschar *Xecode;
+   USPTR Xmstart;
    int Xoffset_top;
    long int Xims;
    eptrblock *Xeptrb;
    int Xflags;
-   int Xrdepth;
+   unsigned int Xrdepth;
    /* Function local variables */
-   const uschar *Xcallpat;
+   USPTR Xcallpat;
-   const uschar *Xcharptr;
+ #ifdef SUPPORT_UTF8
-   const uschar *Xdata;
+   USPTR Xcharptr;
-   const uschar *Xnext;
+ #endif
-   const uschar *Xpp;
+   USPTR Xdata;
-   const uschar *Xprev;
+   USPTR Xnext;
-   const uschar *Xsaved_eptr;
+   USPTR Xpp;
+   USPTR Xprev;
+   USPTR Xsaved_eptr;
    recursion_info Xnew_recursive;
    BOOL Xcur_is_word;
    BOOL Xcondition;
-   BOOL Xminimize;
    BOOL Xprev_is_word;
    unsigned long int Xoriginal_ims;
-Line 303 
 typedef struct heapframe {
+Line 358 
 typedef struct heapframe {
    int Xprop_category;
    int Xprop_chartype;
    int Xprop_script;
-   int *Xprop_test_variable;
+   int Xoclength;
+   uschar Xocchars[8];
  #endif
+   int Xcodelink;
    int Xctype;
-   int Xfc;
+   unsigned int Xfc;
    int Xfi;
    int Xlength;
    int Xmax;
-Line 321 
 typedef struct heapframe {
+Line 378 
 typedef struct heapframe {
    eptrblock Xnewptrb;
-   /* Place to pass back result, and where to jump back to */
+   /* Where to jump back to */
-   int  Xresult;
+   int Xwhere;
-   jmp_buf Xwhere;
  } heapframe;
-Line 340 
 typedef struct heapframe {
+Line 396 
 typedef struct heapframe {
  *         Match from current position            *
  *************************************************/
- /* On entry ecode points to the first opcode, and eptr to the first character
+ /* This function is called recursively in many circumstances. Whenever it
- in the subject string, while eptrb holds the value of eptr at the start of the
- last bracketed group - used for breaking infinite loops matching zero-length
- strings. This function is called recursively in many circumstances. Whenever it
  returns a negative (error) response, the outer incarnation must also return the
- same response.
+ same response. */
+ /* These macros pack up tests that are used for partial matching, and which
+ appears several times in the code. We set the "hit end" flag if the pointer is
+ at the end of the subject and also past the start of the subject (i.e.
+ something has been matched). For hard partial matching, we then return
+ immediately. The second one is used when we already know we are past the end of
+ the subject. */
+ #define CHECK_PARTIAL()\
+   if (md->partial != 0 && eptr >= md->end_subject && eptr > mstart)\
+     {\
+     md->hitend = TRUE;\
+     if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);\
+     }
- Performance note: It might be tempting to extract commonly used fields from the
+ #define SCHECK_PARTIAL()\
- md structure (e.g. utf8, end_subject) into individual variables to improve
+   if (md->partial && eptr > mstart)\
+     {\
+     md->hitend = TRUE;\
+     if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL);\
+     }
+ /* Performance note: It might be tempting to extract commonly used fields from
+ the md structure (e.g. utf8, end_subject) into individual variables to improve
  performance. Tests using gcc on a SPARC disproved this; in the first case, it
  made performance worse.
  Arguments:
-    eptr        pointer in subject
+    eptr        pointer to current character in subject
-    ecode       position in code
+    ecode       pointer to current position in compiled code
+    mstart      pointer to the current match start position (can be modified
+                  by encountering \K)
     offset_top  current top pointer
     md          pointer to "static" info for the match
     ims         current /i, /m, and /s options
-Line 362 
 Arguments:
+Line 439 
 Arguments:
                   brackets - for testing for empty matches
     flags       can contain
                   match_condassert - this is an assertion condition
-                  match_isgroup - this is the start of a bracketed group
+                  match_cbegroup - this is the start of an unlimited repeat
+                    group that can match an empty string
     rdepth      the recursion depth
  Returns:       MATCH_MATCH if matched            )  these values are >= 0
-Line 372 
 Returns:       MATCH_MATCH if matched
+Line 450 
 Returns:       MATCH_MATCH if matched
  */
  static int
- match(REGISTER USPTR eptr, REGISTER const uschar *ecode,
+ match(REGISTER USPTR eptr, REGISTER const uschar *ecode, USPTR mstart,
    int offset_top, match_data *md, unsigned long int ims, eptrblock *eptrb,
-   int flags, int rdepth)
+   int flags, unsigned int rdepth)
  {
  /* These variables do not need to be preserved over recursion in this function,
- so they can be ordinary variables in all cases. Mark them with "register"
+ so they can be ordinary variables in all cases. Mark some of them with
- because they are used a lot in loops. */
+ "register" because they are used a lot in loops. */
- register int  rrc;    /* Returns from recursive calls */
+ register int  rrc;         /* Returns from recursive calls */
- register int  i;      /* Used for loops not involving calls to RMATCH() */
+ register int  i;           /* Used for loops not involving calls to RMATCH() */
- register int  c;      /* Character values not kept over RMATCH() calls */
+ register unsigned int c;   /* Character values not kept over RMATCH() calls */
- register BOOL utf8;   /* Local copy of UTF-8 flag for speed */
+ register BOOL utf8;        /* Local copy of UTF-8 flag for speed */
+ BOOL minimize, possessive; /* Quantifier options */
+ int condcode;
  /* When recursion is not being used, all "local" variables that have to be
  preserved over calls to RMATCH() are part of a "frame" which is obtained from
-Line 398 
 frame->Xprevframe = NULL;            /*
+Line 479 
 frame->Xprevframe = NULL;            /*
  frame->Xeptr = eptr;
  frame->Xecode = ecode;
+ frame->Xmstart = mstart;
  frame->Xoffset_top = offset_top;
  frame->Xims = ims;
  frame->Xeptrb = eptrb;
-Line 412 
 HEAP_RECURSE:
+Line 494 
 HEAP_RECURSE:
  #define eptr               frame->Xeptr
  #define ecode              frame->Xecode
+ #define mstart             frame->Xmstart
  #define offset_top         frame->Xoffset_top
  #define ims                frame->Xims
  #define eptrb              frame->Xeptrb
-Line 424 
 HEAP_RECURSE:
+Line 507 
 HEAP_RECURSE:
  #define charptr            frame->Xcharptr
  #endif
  #define callpat            frame->Xcallpat
+ #define codelink           frame->Xcodelink
  #define data               frame->Xdata
  #define next               frame->Xnext
  #define pp                 frame->Xpp
-Line 434 
 HEAP_RECURSE:
+Line 518 
 HEAP_RECURSE:
  #define cur_is_word        frame->Xcur_is_word
  #define condition          frame->Xcondition
- #define minimize           frame->Xminimize
  #define prev_is_word       frame->Xprev_is_word
  #define original_ims       frame->Xoriginal_ims
-Line 446 
 HEAP_RECURSE:
+Line 529 
 HEAP_RECURSE:
  #define prop_category      frame->Xprop_category
  #define prop_chartype      frame->Xprop_chartype
  #define prop_script        frame->Xprop_script
- #define prop_test_variable frame->Xprop_test_variable
+ #define oclength           frame->Xoclength
+ #define occhars            frame->Xocchars
  #endif
  #define ctype              frame->Xctype
-Line 470 
 HEAP_RECURSE:
+Line 554 
 HEAP_RECURSE:
  get preserved during recursion in the normal way. In this environment, fi and
  i, and fc and c, can be the same variables. */
- #else
+ #else         /* NO_RECURSE not defined */
  #define fi i
  #define fc c
-Line 489 
 recursion_info new_recursive;      /* wi
+Line 573 
 recursion_info new_recursive;      /* wi
                                     /* that do not have to be preserved over  */
  BOOL cur_is_word;                  /* a recursive call to RMATCH().          */
  BOOL condition;
- BOOL minimize;
  BOOL prev_is_word;
  unsigned long int original_ims;
-Line 501 
 int prop_fail_result;
+Line 584 
 int prop_fail_result;
  int prop_category;
  int prop_chartype;
  int prop_script;
- int *prop_test_variable;
+ int oclength;
+ uschar occhars[8];
  #endif
+ int codelink;
  int ctype;
  int length;
  int max;
-Line 516 
 int save_offset1, save_offset2, save_off
+Line 601 
 int save_offset1, save_offset2, save_off
  int stacksave[REC_STACK_SAVE_MAX];
  eptrblock newptrb;
- #endif
+ #endif     /* NO_RECURSE */
  /* These statements are here to stop the compiler complaining about unitialized
  variables. */
-Line 524 
 variables. */
+Line 609 
 variables. */
  #ifdef SUPPORT_UCP
  prop_value = 0;
  prop_fail_result = 0;
- prop_test_variable = NULL;
  #endif
+ /* This label is used for tail recursion, which is used in a few cases even
+ when NO_RECURSE is not defined, in order to reduce the amount of stack that is
+ used. Thanks to Ian Taylor for noticing this possibility and sending the
+ original patch. */
+ TAIL_RECURSE:
  /* OK, now we can get on with the real code of the function. Recursive calls
  are specified by the macro RMATCH and RRETURN is used to return. When
  NO_RECURSE is *not* defined, these just turn into a recursive call to match()
-Line 535 
 defined). However, RMATCH isn't like a f
+Line 627 
 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
+ utf8 = md->utf8;       /* Local copy of the flag */
+ #else
+ utf8 = FALSE;
+ #endif
  /* First check that we haven't called match() too many times, or that we
  haven't exceeded the recursive call limit. */
-Line 542 
 if (md->match_call_count++ >= md->match_
+Line 640 
 if (md->match_call_count++ >= md->match_
  if (rdepth >= md->match_limit_recursion) RRETURN(PCRE_ERROR_RECURSIONLIMIT);
  original_ims = ims;    /* Save for resetting on ')' */
- utf8 = md->utf8;       /* Local copy of the flag */
- /* At the start of a bracketed group, add the current subject pointer to the
+ /* At the start of a group with an unlimited repeat that may match an empty
- stack of such pointers, to be re-instated at the end of the group when we hit
+ string, the match_cbegroup flag is set. When this is the case, add the current
- the closing ket. When match() is called in other circumstances, we don't add to
+ subject pointer to the chain of such remembered pointers, to be checked when we
- this stack. */
+ hit the closing ket, in order to break infinite loops that match no characters.
+ When match() is called in other circumstances, don't add to the chain. The
+ match_cbegroup flag must NOT be used with tail recursion, because the memory
+ block that is used is on the stack, so a new one may be required for each
+ match(). */
- if ((flags & match_isgroup) != 0)
+ if ((flags & match_cbegroup) != 0)
    {
-   newptrb.epb_prev = eptrb;
    newptrb.epb_saved_eptr = eptr;
+   newptrb.epb_prev = eptrb;
    eptrb = &newptrb;
    }
- /* Now start processing the operations. */
+ /* Now start processing the opcodes. */
  for (;;)
    {
+   minimize = possessive = FALSE;
    op = *ecode;
-   minimize = FALSE;
-   /* For partial matching, remember if we ever hit the end of the subject after
-   matching at least one subject character. */
-   if (md->partial &&
-       eptr >= md->end_subject &&
-       eptr > md->start_match)
-     md->hitend = TRUE;
-   /* Opening capturing bracket. If there is space in the offset vector, save
-   the current subject position in the working slot at the top of the vector. We
-   mustn't change the current values of the data slot, because they may be set
-   from a previous iteration of this group, and be referred to by a reference
-   inside the group.
-   If the bracket fails to match, we need to restore this value and also the
-   values of the final offsets, in case they were set by a previous iteration of
-   the same bracket.
-   If there isn't enough space in the offset vector, treat this as if it were a
-   non-capturing bracket. Don't worry about setting the flag for the error case
-   here; that is handled in the code for KET. */
-   if (op > OP_BRA)
+   switch(op)
      {
-     number = op - OP_BRA;
+     case OP_FAIL:
+     RRETURN(MATCH_NOMATCH);
-     /* For extended extraction brackets (large number), we have to fish out the
-     number from a dummy opcode at the start. */
-     if (number > EXTRACT_BASIC_MAX)
+     case OP_PRUNE:
-       number = GET2(ecode, 2+LINK_SIZE);
+     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
+       ims, eptrb, flags, RM51);
+     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     RRETURN(MATCH_PRUNE);
+     case OP_COMMIT:
+     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
+       ims, eptrb, flags, RM52);
+     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     RRETURN(MATCH_COMMIT);
+     case OP_SKIP:
+     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
+       ims, eptrb, flags, RM53);
+     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     md->start_match_ptr = eptr;   /* Pass back current position */
+     RRETURN(MATCH_SKIP);
+     case OP_THEN:
+     RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
+       ims, eptrb, flags, RM54);
+     if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+     RRETURN(MATCH_THEN);
+     /* Handle a capturing bracket. If there is space in the offset vector, save
+     the current subject position in the working slot at the top of the vector.
+     We mustn't change the current values of the data slot, because they may be
+     set from a previous iteration of this group, and be referred to by a
+     reference inside the group.
+     If the bracket fails to match, we need to restore this value and also the
+     values of the final offsets, in case they were set by a previous iteration
+     of the same bracket.
+     If there isn't enough space in the offset vector, treat this as if it were
+     a non-capturing bracket. Don't worry about setting the flag for the error
+     case here; that is handled in the code for KET. */
+     case OP_CBRA:
+     case OP_SCBRA:
+     number = GET2(ecode, 1+LINK_SIZE);
      offset = number << 1;
  #ifdef DEBUG
-     printf("start bracket %d subject=", number);
+     printf("start bracket %d\n", number);
+     printf("subject=");
      pchars(eptr, 16, TRUE, md);
      printf("\n");
  #endif
-Line 612 
 for (;;)
+Line 730 
 for (;;)
        DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
        md->offset_vector[md->offset_end - number] = eptr - md->start_subject;
+       flags = (op == OP_SCBRA)? match_cbegroup : 0;
        do
          {
-         RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb,
+         RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
-           match_isgroup);
+           ims, eptrb, flags, RM1);
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
          md->capture_last = save_capture_last;
          ecode += GET(ecode, 1);
          }
-Line 631 
 for (;;)
+Line 750 
 for (;;)
        RRETURN(MATCH_NOMATCH);
        }
-     /* Insufficient room for saving captured contents */
+     /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
+     as a non-capturing bracket. */
-     else op = OP_BRA;
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-     }
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-   /* Other types of node can be handled by a switch */
+     DPRINTF(("insufficient capture room: treat as non-capturing\n"));
-   switch(op)
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-     {
+     /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-     case OP_BRA:     /* Non-capturing bracket: optimized */
-     DPRINTF(("start bracket 0\n"));
+     /* Non-capturing bracket. Loop for all the alternatives. When we get to the
-     do
+     final alternative within the 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 recursion, except in the case when match_cbegroup
+     is set.*/
+     case OP_BRA:
+     case OP_SBRA:
+     DPRINTF(("start non-capturing bracket\n"));
+     flags = (op >= OP_SBRA)? match_cbegroup : 0;
+     for (;;)
        {
-       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb,
+       if (ecode[GET(ecode, 1)] != OP_ALT)   /* Final alternative */
-         match_isgroup);
+         {
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         if (flags == 0)    /* Not a possibly empty group */
+           {
+           ecode += _pcre_OP_lengths[*ecode];
+           DPRINTF(("bracket 0 tail recursion\n"));
+           goto TAIL_RECURSE;
+           }
+         /* Possibly empty group; can't use tail recursion. */
+         RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md, ims,
+           eptrb, flags, RM48);
+         RRETURN(rrc);
+         }
+       /* For non-final alternatives, continue the loop for a NOMATCH result;
+       otherwise return. */
+       RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md, ims,
+         eptrb, flags, RM2);
+       if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
        ecode += GET(ecode, 1);
        }
-     while (*ecode == OP_ALT);
+     /* Control never reaches here. */
-     DPRINTF(("bracket 0 failed\n"));
-     RRETURN(MATCH_NOMATCH);
      /* 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. */
+     exactly what going to the ket would do. As there is only one branch to be
+     obeyed, we can use tail recursion to avoid using another stack frame. */
      case OP_COND:
-     if (ecode[LINK_SIZE+1] == OP_CREF) /* Condition extract or recurse test */
+     case OP_SCOND:
+     codelink= GET(ecode, 1);
+     /* Because of the way auto-callout works during compile, a callout item is
+     inserted between OP_COND and an assertion condition. */
+     if (ecode[LINK_SIZE+1] == OP_CALLOUT)
+       {
+       if (pcre_callout != NULL)
+         {
+         pcre_callout_block cb;
+         cb.version          = 1;   /* Version 1 of the callout block */
+         cb.callout_number   = ecode[LINK_SIZE+2];
+         cb.offset_vector    = md->offset_vector;
+         cb.subject          = (PCRE_SPTR)md->start_subject;
+         cb.subject_length   = md->end_subject - md->start_subject;
+         cb.start_match      = mstart - md->start_subject;
+         cb.current_position = eptr - md->start_subject;
+         cb.pattern_position = GET(ecode, LINK_SIZE + 3);
+         cb.next_item_length = GET(ecode, 3 + 2*LINK_SIZE);
+         cb.capture_top      = offset_top/2;
+         cb.capture_last     = md->capture_last;
+         cb.callout_data     = md->callout_data;
+         if ((rrc = (*pcre_callout)(&cb)) > 0) RRETURN(MATCH_NOMATCH);
+         if (rrc < 0) RRETURN(rrc);
+         }
+       ecode += _pcre_OP_lengths[OP_CALLOUT];
+       }
+     condcode = ecode[LINK_SIZE+1];
+     /* Now see what the actual condition is */
+     if (condcode == OP_RREF)         /* Recursion test */
+       {
+       offset = GET2(ecode, LINK_SIZE + 2);     /* Recursion group number*/
+       condition = md->recursive != NULL &&
+         (offset == RREF_ANY || offset == md->recursive->group_num);
+       ecode += condition? 3 : GET(ecode, 1);
+       }
+     else if (condcode == OP_CREF)    /* Group used test */
        {
        offset = GET2(ecode, LINK_SIZE+2) << 1;  /* Doubled ref number */
-       condition = (offset == CREF_RECURSE * 2)?
+       condition = offset < offset_top && md->offset_vector[offset] >= 0;
-         (md->recursive != NULL) :
+       ecode += condition? 3 : GET(ecode, 1);
-         (offset < offset_top && md->offset_vector[offset] >= 0);
+       }
-       RMATCH(rrc, eptr, ecode + (condition?
-         (LINK_SIZE + 4) : (LINK_SIZE + 1 + GET(ecode, 1))),
+     else if (condcode == OP_DEF)     /* DEFINE - always false */
-         offset_top, md, ims, eptrb, match_isgroup);
+       {
-       RRETURN(rrc);
+       condition = FALSE;
+       ecode += GET(ecode, 1);
        }
      /* The condition is an assertion. Call match() to evaluate it - setting
-     the final argument TRUE causes it to stop at the end of an assertion. */
+     the final argument match_condassert causes it to stop at the end of an
+     assertion. */
      else
        {
-       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,
-           match_condassert | match_isgroup);
+           match_condassert, RM3);
        if (rrc == MATCH_MATCH)
          {
-         ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE+2);
+         condition = TRUE;
+         ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE + 2);
          while (*ecode == OP_ALT) ecode += GET(ecode, 1);
          }
-       else if (rrc != MATCH_NOMATCH)
+       else if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN)
          {
          RRETURN(rrc);         /* Need braces because of following else */
          }
-       else ecode += GET(ecode, 1);
+       else
-       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb,
+         {
-         match_isgroup);
+         condition = FALSE;
-       RRETURN(rrc);
+         ecode += codelink;
+         }
        }
-     /* Control never reaches here */
-     /* Skip over conditional reference or large extraction number data if
+     /* We are now at the branch that is to be obeyed. As there is only one,
-     encountered. */
+     we can use tail recursion to avoid using another stack frame, except when
+     match_cbegroup is required for an unlimited repeat of a possibly empty
+     group. If the second alternative doesn't exist, we can just plough on. */
-     case OP_CREF:
+     if (condition || *ecode == OP_ALT)
-     case OP_BRANUMBER:
+       {
-     ecode += 3;
+       ecode += 1 + LINK_SIZE;
+       if (op == OP_SCOND)        /* Possibly empty group */
+         {
+         RMATCH(eptr, ecode, offset_top, md, ims, eptrb, match_cbegroup, RM49);
+         RRETURN(rrc);
+         }
+       else                       /* Group must match something */
+         {
+         flags = 0;
+         goto TAIL_RECURSE;
+         }
+       }
+     else                         /* Condition false & no alternative */
+       {
+       ecode += 1 + LINK_SIZE;
+       }
      break;
-     /* End of the pattern. If we are in a recursion, we should restore the
-     offsets appropriately and continue from after the call. */
+     /* End of the pattern, either real or forced. If we are in a top-level
+     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)
        {
-Line 713 
 for (;;)
+Line 924 
 for (;;)
        md->recursive = rec->prevrec;
        memmove(md->offset_vector, rec->offset_save,
          rec->saved_max * sizeof(int));
-       md->start_match = rec->save_start;
+       mstart = rec->save_start;
        ims = original_ims;
        ecode = rec->after_call;
        break;
        }
-     /* Otherwise, if PCRE_NOTEMPTY is set, fail if we have matched an empty
+     /* Otherwise, if we have matched an empty string, fail if PCRE_NOTEMPTY is
-     string - backtracking will then try other alternatives, if any. */
+     set, or if PCRE_NOTEMPTY_ATSTART is set and we have matched at the start of
+     the subject. In both cases, backtracking will then try other alternatives,
+     if any. */
+     if (eptr == mstart &&
+         (md->notempty ||
+           (md->notempty_atstart &&
+             mstart == md->start_subject + md->start_offset)))
+       RRETURN(MATCH_NOMATCH);
+     /* Otherwise, we have a match. */
-     if (md->notempty && eptr == md->start_match) RRETURN(MATCH_NOMATCH);
+     md->end_match_ptr = eptr;           /* Record where we ended */
-     md->end_match_ptr = eptr;          /* Record where we ended */
+     md->end_offset_top = offset_top;    /* and how many extracts were taken */
-     md->end_offset_top = offset_top;   /* and how many extracts were taken */
+     md->start_match_ptr = mstart;       /* and the start (\K can modify) */
      RRETURN(MATCH_MATCH);
      /* Change option settings */
-Line 745 
 for (;;)
+Line 966 
 for (;;)
      case OP_ASSERTBACK:
      do
        {
-       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL, 0,
-         match_isgroup);
+         RM4);
        if (rrc == MATCH_MATCH) break;
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
        ecode += GET(ecode, 1);
        }
      while (*ecode == OP_ALT);
-Line 772 
 for (;;)
+Line 993 
 for (;;)
      case OP_ASSERTBACK_NOT:
      do
        {
-       RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL,
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, NULL, 0,
-         match_isgroup);
+         RM5);
        if (rrc == MATCH_MATCH) RRETURN(MATCH_NOMATCH);
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
        ecode += GET(ecode,1);
        }
      while (*ecode == OP_ALT);
-Line 794 
 for (;;)
+Line 1015 
 for (;;)
  #ifdef SUPPORT_UTF8
      if (utf8)
        {
-       c = GET(ecode,1);
+       i = GET(ecode, 1);
-       for (i = 0; i < c; i++)
+       while (i-- > 0)
          {
          eptr--;
          if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
-         BACKCHAR(eptr)
+         BACKCHAR(eptr);
          }
        }
      else
-Line 808 
 for (;;)
+Line 1029 
 for (;;)
      /* No UTF-8 support, or not in UTF-8 mode: count is byte count */
        {
-       eptr -= GET(ecode,1);
+       eptr -= GET(ecode, 1);
        if (eptr < md->start_subject) RRETURN(MATCH_NOMATCH);
        }
-     /* Skip to next op code */
+     /* Save the earliest consulted character, then skip to next op code */
+     if (eptr < md->start_used_ptr) md->start_used_ptr = eptr;
      ecode += 1 + LINK_SIZE;
      break;
-Line 830 
 for (;;)
+Line 1052 
 for (;;)
        cb.offset_vector    = md->offset_vector;
        cb.subject          = (PCRE_SPTR)md->start_subject;
        cb.subject_length   = md->end_subject - md->start_subject;
-       cb.start_match      = md->start_match - md->start_subject;
+       cb.start_match      = mstart - md->start_subject;
        cb.current_position = eptr - md->start_subject;
        cb.pattern_position = GET(ecode, 2);
        cb.next_item_length = GET(ecode, 2 + LINK_SIZE);
-Line 865 
 for (;;)
+Line 1087 
 for (;;)
      case OP_RECURSE:
        {
        callpat = md->start_code + GET(ecode, 1);
-       new_recursive.group_num = *callpat - OP_BRA;
+       new_recursive.group_num = (callpat == md->start_code)? 0 :
+         GET2(callpat, 1 + LINK_SIZE);
-       /* For extended extraction brackets (large number), we have to fish out
-       the number from a dummy opcode at the start. */
-       if (new_recursive.group_num > EXTRACT_BASIC_MAX)
-         new_recursive.group_num = GET2(callpat, 2+LINK_SIZE);
        /* Add to "recursing stack" */
-Line 897 
 for (;;)
+Line 1114 
 for (;;)
        memcpy(new_recursive.offset_save, md->offset_vector,
              new_recursive.saved_max * sizeof(int));
-       new_recursive.save_start = md->start_match;
+       new_recursive.save_start = mstart;
-       md->start_match = eptr;
+       mstart = eptr;
        /* OK, now we can do the recursion. For each top-level alternative we
        restore the offset and recursion data. */
        DPRINTF(("Recursing into group %d\n", new_recursive.group_num));
+       flags = (*callpat >= OP_SBRA)? match_cbegroup : 0;
        do
          {
-         RMATCH(rrc, eptr, callpat + 1 + LINK_SIZE, offset_top, md, ims,
+         RMATCH(eptr, callpat + _pcre_OP_lengths[*callpat], offset_top,
-             eptrb, match_isgroup);
+           md, ims, eptrb, flags, RM6);
          if (rrc == MATCH_MATCH)
            {
            DPRINTF(("Recursion matched\n"));
-Line 916 
 for (;;)
+Line 1134 
 for (;;)
              (pcre_free)(new_recursive.offset_save);
            RRETURN(MATCH_MATCH);
            }
-         else if (rrc != MATCH_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);
            RRETURN(rrc);
            }
-Line 945 
 for (;;)
+Line 1165 
 for (;;)
      the end of a normal bracket, leaving the subject pointer. */
      case OP_ONCE:
-       {
+     prev = ecode;
-       prev = ecode;
+     saved_eptr = eptr;
-       saved_eptr = eptr;
-       do
+     do
-         {
+       {
-         RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims,
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb, 0, RM7);
-           eptrb, match_isgroup);
+       if (rrc == MATCH_MATCH) break;
-         if (rrc == MATCH_MATCH) break;
+       if (rrc != MATCH_NOMATCH && rrc != MATCH_THEN) RRETURN(rrc);
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       ecode += GET(ecode,1);
-         ecode += GET(ecode,1);
+       }
-         }
+     while (*ecode == OP_ALT);
-       while (*ecode == OP_ALT);
-       /* If hit the end of the group (which could be repeated), fail */
+     /* If hit the end of the group (which could be repeated), fail */
-       if (*ecode != OP_ONCE && *ecode != OP_ALT) RRETURN(MATCH_NOMATCH);
+     if (*ecode != OP_ONCE && *ecode != OP_ALT) RRETURN(MATCH_NOMATCH);
-       /* Continue as from after the assertion, updating the offsets high water
+     /* Continue as from after the assertion, updating the offsets high water
-       mark, since extracts may have been taken. */
+     mark, since extracts may have been taken. */
-       do ecode += GET(ecode,1); while (*ecode == OP_ALT);
+     do ecode += GET(ecode, 1); while (*ecode == OP_ALT);
-       offset_top = md->end_offset_top;
+     offset_top = md->end_offset_top;
-       eptr = md->end_match_ptr;
+     eptr = md->end_match_ptr;
-       /* For a non-repeating ket, just continue at this level. This also
+     /* 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.
+     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
+     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. */
+     course of events. */
-       if (*ecode == OP_KET || eptr == saved_eptr)
+     if (*ecode == OP_KET || eptr == saved_eptr)
-         {
+       {
-         ecode += 1+LINK_SIZE;
+       ecode += 1+LINK_SIZE;
-         break;
+       break;
-         }
+       }
-       /* The repeating kets try the rest of the pattern or restart from the
+     /* The repeating kets try the rest of the pattern or restart from the
-       preceding bracket, in the appropriate order. We need to reset any options
+     preceding bracket, in the appropriate order. The second "call" of match()
-       that changed within the bracket before re-running it, so check the next
+     uses tail recursion, to avoid using another stack frame. We need to reset
-       opcode. */
+     any options that changed within the bracket before re-running it, so
+     check the next opcode. */
-       if (ecode[1+LINK_SIZE] == OP_OPT)
+     if (ecode[1+LINK_SIZE] == OP_OPT)
-         {
+       {
-         ims = (ims & ~PCRE_IMS) | ecode[4];
+       ims = (ims & ~PCRE_IMS) | ecode[4];
-         DPRINTF(("ims set to %02lx at group repeat\n", ims));
+       DPRINTF(("ims set to %02lx at group repeat\n", ims));
-         }
+       }
-       if (*ecode == OP_KETRMIN)
+     if (*ecode == OP_KETRMIN)
-         {
+       {
-         RMATCH(rrc, eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb, 0);
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb, 0, RM8);
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-         RMATCH(rrc, eptr, prev, offset_top, md, ims, eptrb, match_isgroup);
+       ecode = prev;
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       flags = 0;
-         }
+       goto TAIL_RECURSE;
-       else  /* OP_KETRMAX */
-         {
-         RMATCH(rrc, eptr, prev, offset_top, md, ims, eptrb, match_isgroup);
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-         RMATCH(rrc, eptr, ecode + 1+LINK_SIZE, offset_top, md, ims, eptrb, 0);
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-         }
        }
-     RRETURN(MATCH_NOMATCH);
+     else  /* OP_KETRMAX */
+       {
+       RMATCH(eptr, prev, offset_top, md, ims, eptrb, match_cbegroup, RM9);
+       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       ecode += 1 + LINK_SIZE;
+       flags = 0;
+       goto TAIL_RECURSE;
+       }
+     /* Control never gets here */
      /* An alternation is the end of a branch; scan along to find the end of the
      bracketed group and go to there. */
-Line 1018 
 for (;;)
+Line 1238 
 for (;;)
      do ecode += GET(ecode,1); while (*ecode == OP_ALT);
      break;
-     /* BRAZERO and BRAMINZERO occur just before a bracket group, indicating
+     /* BRAZERO, BRAMINZERO and SKIPZERO occur just before a bracket group,
-     that it may occur zero times. It may repeat infinitely, or not at all -
+     indicating that it may occur zero times. It may repeat infinitely, or not
-     i.e. it could be ()* or ()? in the pattern. Brackets with fixed upper
+     at all - i.e. it could be ()* or ()? or even (){0} in the pattern. Brackets
-     repeat limits are compiled as a number of copies, with the optional ones
+     with fixed upper repeat limits are compiled as a number of copies, with the
-     preceded by BRAZERO or BRAMINZERO. */
+     optional ones preceded by BRAZERO or BRAMINZERO. */
      case OP_BRAZERO:
        {
        next = ecode+1;
-       RMATCH(rrc, eptr, next, offset_top, md, ims, eptrb, match_isgroup);
+       RMATCH(eptr, next, offset_top, md, ims, eptrb, 0, RM10);
        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
        do next += GET(next,1); while (*next == OP_ALT);
-       ecode = next + 1+LINK_SIZE;
+       ecode = next + 1 + LINK_SIZE;
        }
      break;
      case OP_BRAMINZERO:
        {
        next = ecode+1;
-       do next += GET(next,1); while (*next == OP_ALT);
+       do next += GET(next, 1); while (*next == OP_ALT);
-       RMATCH(rrc, eptr, next + 1+LINK_SIZE, offset_top, md, ims, eptrb,
+       RMATCH(eptr, next + 1+LINK_SIZE, offset_top, md, ims, eptrb, 0, RM11);
-         match_isgroup);
        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
        ecode++;
        }
      break;
-     /* End of a group, repeated or non-repeating. If we are at the end of
+     case OP_SKIPZERO:
-     an assertion "group", stop matching and return MATCH_MATCH, but record the
+       {
-     current high water mark for use by positive assertions. Do this also
+       next = ecode+1;
-     for the "once" (not-backup up) groups. */
+       do next += GET(next,1); while (*next == OP_ALT);
+       ecode = next + 1 + LINK_SIZE;
+       }
+     break;
+     /* End of a group, repeated or non-repeating. */
      case OP_KET:
      case OP_KETRMIN:
      case OP_KETRMAX:
-       {
+     prev = ecode - GET(ecode, 1);
-       prev = ecode - GET(ecode, 1);
-       saved_eptr = eptrb->epb_saved_eptr;
-       /* Back up the stack of bracket start pointers. */
+     /* If this was a group that remembered the subject start, in order to break
+     infinite repeats of empty string matches, retrieve the subject start from
+     the chain. Otherwise, set it NULL. */
-       eptrb = eptrb->epb_prev;
+     if (*prev >= OP_SBRA)
+       {
-       if (*prev == OP_ASSERT || *prev == OP_ASSERT_NOT ||
+       saved_eptr = eptrb->epb_saved_eptr;   /* Value at start of group */
-           *prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT ||
+       eptrb = eptrb->epb_prev;              /* Backup to previous group */
-           *prev == OP_ONCE)
+       }
-         {
+     else saved_eptr = NULL;
-         md->end_match_ptr = eptr;      /* For ONCE */
-         md->end_offset_top = offset_top;
-         RRETURN(MATCH_MATCH);
-         }
-       /* In all other cases except a conditional group we have to check the
+     /* If we are at the end of an assertion group, stop matching and return
-       group number back at the start and if necessary complete handling an
+     MATCH_MATCH, but record the current high water mark for use by positive
-       extraction by setting the offsets and bumping the high water mark. */
+     assertions. Do this also for the "once" (atomic) groups. */
-       if (*prev != OP_COND)
+     if (*prev == OP_ASSERT || *prev == OP_ASSERT_NOT ||
-         {
+         *prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT ||
-         number = *prev - OP_BRA;
+         *prev == OP_ONCE)
+       {
+       md->end_match_ptr = eptr;      /* For ONCE */
+       md->end_offset_top = offset_top;
+       RRETURN(MATCH_MATCH);
+       }
-         /* For extended extraction brackets (large number), we have to fish out
+     /* For capturing groups we have to check the group number back at the start
-         the number from a dummy opcode at the start. */
+     and if necessary complete handling an extraction by setting the offsets and
+     bumping the high water mark. Note that whole-pattern recursion is coded as
+     a recurse into group 0, so it won't be picked up here. Instead, we catch it
+     when the OP_END is reached. Other recursion is handled here. */
-         if (number > EXTRACT_BASIC_MAX) number = GET2(prev, 2+LINK_SIZE);
+     if (*prev == OP_CBRA || *prev == OP_SCBRA)
-         offset = number << 1;
+       {
+       number = GET2(prev, 1+LINK_SIZE);
+       offset = number << 1;
  #ifdef DEBUG
-         printf("end bracket %d", number);
+       printf("end bracket %d", number);
-         printf("\n");
+       printf("\n");
  #endif
-         /* Test for a numbered group. This includes groups called as a result
+       md->capture_last = number;
-         of recursion. Note that whole-pattern recursion is coded as a recurse
+       if (offset >= md->offset_max) md->offset_overflow = TRUE; else
-         into group 0, so it won't be picked up here. Instead, we catch it when
+         {
-         the OP_END is reached. */
+         md->offset_vector[offset] =
+           md->offset_vector[md->offset_end - number];
-         if (number > 0)
+         md->offset_vector[offset+1] = eptr - md->start_subject;
-           {
+         if (offset_top <= offset) offset_top = offset + 2;
-           md->capture_last = number;
+         }
-           if (offset >= md->offset_max) md->offset_overflow = TRUE; else
-             {
-             md->offset_vector[offset] =
-               md->offset_vector[md->offset_end - number];
-             md->offset_vector[offset+1] = eptr - md->start_subject;
-             if (offset_top <= offset) offset_top = offset + 2;
-             }
-           /* Handle a recursively called group. Restore the offsets
+       /* Handle a recursively called group. Restore the offsets
-           appropriately and continue from after the call. */
+       appropriately and continue from after the call. */
-           if (md->recursive != NULL && md->recursive->group_num == number)
+       if (md->recursive != NULL && md->recursive->group_num == number)
-             {
+         {
-             recursion_info *rec = md->recursive;
+         recursion_info *rec = md->recursive;
-             DPRINTF(("Recursion (%d) succeeded - continuing\n", number));
+         DPRINTF(("Recursion (%d) succeeded - continuing\n", number));
-             md->recursive = rec->prevrec;
+         md->recursive = rec->prevrec;
-             md->start_match = rec->save_start;
+         mstart = rec->save_start;
-             memcpy(md->offset_vector, rec->offset_save,
+         memcpy(md->offset_vector, rec->offset_save,
-               rec->saved_max * sizeof(int));
+           rec->saved_max * sizeof(int));
-             ecode = rec->after_call;
+         ecode = rec->after_call;
-             ims = original_ims;
+         ims = original_ims;
-             break;
+         break;
-             }
-           }
          }
+       }
-       /* Reset the value of the ims flags, in case they got changed during
+     /* For both capturing and non-capturing groups, reset the value of the ims
-       the group. */
+     flags, in case they got changed during the group. */
-       ims = original_ims;
+     ims = original_ims;
-       DPRINTF(("ims reset to %02lx\n", ims));
+     DPRINTF(("ims reset to %02lx\n", ims));
-       /* For a non-repeating ket, just continue at this level. This also
+     /* 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.
+     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
+     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. */
+     course of events. */
-       if (*ecode == OP_KET || eptr == saved_eptr)
+     if (*ecode == OP_KET || eptr == saved_eptr)
-         {
+       {
-         ecode += 1 + LINK_SIZE;
+       ecode += 1 + LINK_SIZE;
-         break;
+       break;
-         }
+       }
-       /* The repeating kets try the rest of the pattern or restart from the
+     /* The repeating kets try the rest of the pattern or restart from the
-       preceding bracket, in the appropriate order. */
+     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. */
-       if (*ecode == OP_KETRMIN)
+     flags = (*prev >= OP_SBRA)? match_cbegroup : 0;
-         {
-         RMATCH(rrc, eptr, ecode + 1+LINK_SIZE, offset_top, md, ims, eptrb, 0);
+     if (*ecode == OP_KETRMIN)
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       {
-         RMATCH(rrc, eptr, prev, offset_top, md, ims, eptrb, match_isgroup);
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, ims, eptrb, 0, RM12);
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-         }
+       if (flags != 0)    /* Could match an empty string */
-       else  /* OP_KETRMAX */
          {
-         RMATCH(rrc, eptr, prev, offset_top, md, ims, eptrb, match_isgroup);
+         RMATCH(eptr, prev, offset_top, md, ims, eptrb, flags, RM50);
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         RRETURN(rrc);
-         RMATCH(rrc, eptr, ecode + 1+LINK_SIZE, offset_top, md, ims, eptrb, 0);
-         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
          }
+       ecode = prev;
+       goto TAIL_RECURSE;
        }
+     else  /* OP_KETRMAX */
-     RRETURN(MATCH_NOMATCH);
+       {
+       RMATCH(eptr, prev, offset_top, md, ims, eptrb, flags, RM13);
+       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       ecode += 1 + LINK_SIZE;
+       flags = 0;
+       goto TAIL_RECURSE;
+       }
+     /* Control never gets here */
      /* Start of subject unless notbol, or after internal newline if multiline */
-Line 1168 
 for (;;)
+Line 1398 
 for (;;)
      if (md->notbol && eptr == md->start_subject) RRETURN(MATCH_NOMATCH);
      if ((ims & PCRE_MULTILINE) != 0)
        {
-       if (eptr != md->start_subject && eptr[-1] != NEWLINE)
+       if (eptr != md->start_subject &&
+           (eptr == md->end_subject || !WAS_NEWLINE(eptr)))
          RRETURN(MATCH_NOMATCH);
        ecode++;
        break;
-Line 1189 
 for (;;)
+Line 1420 
 for (;;)
      ecode++;
      break;
+     /* Reset the start of match point */
+     case OP_SET_SOM:
+     mstart = eptr;
+     ecode++;
+     break;
      /* Assert before internal newline if multiline, or before a terminating
      newline unless endonly is set, else end of subject unless noteol is set. */
-Line 1196 
 for (;;)
+Line 1434 
 for (;;)
      if ((ims & PCRE_MULTILINE) != 0)
        {
        if (eptr < md->end_subject)
-         { if (*eptr != NEWLINE) RRETURN(MATCH_NOMATCH); }
+         { if (!IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH); }
        else
          { if (md->noteol) RRETURN(MATCH_NOMATCH); }
        ecode++;
-Line 1207 
 for (;;)
+Line 1445 
 for (;;)
        if (md->noteol) RRETURN(MATCH_NOMATCH);
        if (!md->endonly)
          {
-         if (eptr < md->end_subject - 1 ||
+         if (eptr != md->end_subject &&
-            (eptr == md->end_subject - 1 && *eptr != NEWLINE))
+             (!IS_NEWLINE(eptr) || eptr != md->end_subject - md->nllen))
            RRETURN(MATCH_NOMATCH);
          ecode++;
          break;
          }
        }
-     /* ... else fall through */
+     /* ... else fall through for endonly */
      /* End of subject assertion (\z) */
-Line 1226 
 for (;;)
+Line 1464 
 for (;;)
      /* End of subject or ending \n assertion (\Z) */
      case OP_EODN:
-     if (eptr < md->end_subject - 1 ||
+     if (eptr != md->end_subject &&
-        (eptr == md->end_subject - 1 && *eptr != NEWLINE)) RRETURN(MATCH_NOMATCH);
+         (!IS_NEWLINE(eptr) || eptr != md->end_subject - md->nllen))
+       RRETURN(MATCH_NOMATCH);
      ecode++;
      break;
-Line 1239 
 for (;;)
+Line 1478 
 for (;;)
        /* Find out if the previous and current characters are "word" characters.
        It takes a bit more work in UTF-8 mode. Characters > 255 are assumed to
-       be "non-word" characters. */
+       be "non-word" characters. Remember the earliest consulted character for
+       partial matching. */
  #ifdef SUPPORT_UTF8
        if (utf8)
          {
          if (eptr == md->start_subject) prev_is_word = FALSE; else
            {
-           const uschar *lastptr = eptr - 1;
+           USPTR lastptr = eptr - 1;
            while((*lastptr & 0xc0) == 0x80) lastptr--;
+           if (lastptr < md->start_used_ptr) md->start_used_ptr = lastptr;
            GETCHAR(c, lastptr);
            prev_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;
            }
-         if (eptr >= md->end_subject) cur_is_word = FALSE; else
+         if (eptr >= md->end_subject)
+           {
+           SCHECK_PARTIAL();
+           cur_is_word = FALSE;
+           }
+         else
            {
            GETCHAR(c, eptr);
            cur_is_word = c < 256 && (md->ctypes[c] & ctype_word) != 0;
-Line 1260 
 for (;;)
+Line 1506 
 for (;;)
        else
  #endif
-       /* More streamlined when not in UTF-8 mode */
+       /* Not in UTF-8 mode */
          {
-         prev_is_word = (eptr != md->start_subject) &&
+         if (eptr == md->start_subject) prev_is_word = FALSE; else
-           ((md->ctypes[eptr[-1]] & ctype_word) != 0);
+           {
-         cur_is_word = (eptr < md->end_subject) &&
+           if (eptr <= md->start_used_ptr) md->start_used_ptr = eptr - 1;
-           ((md->ctypes[*eptr] & ctype_word) != 0);
+           prev_is_word = ((md->ctypes[eptr[-1]] & ctype_word) != 0);
+           }
+         if (eptr >= md->end_subject)
+           {
+           SCHECK_PARTIAL();
+           cur_is_word = FALSE;
+           }
+         else cur_is_word = ((md->ctypes[*eptr] & ctype_word) != 0);
          }
        /* Now see if the situation is what we want */
-Line 1280 
 for (;;)
+Line 1533 
 for (;;)
      /* Match a single character type; inline for speed */
      case OP_ANY:
-     if ((ims & PCRE_DOTALL) == 0 && eptr < md->end_subject && *eptr == NEWLINE)
+     if (IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH);
+     /* Fall through */
+     case OP_ALLANY:
+     if (eptr++ >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
        RRETURN(MATCH_NOMATCH);
-     if (eptr++ >= md->end_subject) RRETURN(MATCH_NOMATCH);
+       }
- #ifdef SUPPORT_UTF8
+     if (utf8) while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
-     if (utf8)
-       while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
- #endif
      ecode++;
      break;
-Line 1294 
 for (;;)
+Line 1550 
 for (;;)
      any byte, even newline, independent of the setting of PCRE_DOTALL. */
      case OP_ANYBYTE:
-     if (eptr++ >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr++ >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      ecode++;
      break;
      case OP_NOT_DIGIT:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
  #ifdef SUPPORT_UTF8
-Line 1312 
 for (;;)
+Line 1576 
 for (;;)
      break;
      case OP_DIGIT:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
  #ifdef SUPPORT_UTF8
-Line 1325 
 for (;;)
+Line 1593 
 for (;;)
      break;
      case OP_NOT_WHITESPACE:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
  #ifdef SUPPORT_UTF8
-Line 1338 
 for (;;)
+Line 1610 
 for (;;)
      break;
      case OP_WHITESPACE:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
  #ifdef SUPPORT_UTF8
-Line 1351 
 for (;;)
+Line 1627 
 for (;;)
      break;
      case OP_NOT_WORDCHAR:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
  #ifdef SUPPORT_UTF8
-Line 1364 
 for (;;)
+Line 1644 
 for (;;)
      break;
      case OP_WORDCHAR:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
      if (
  #ifdef SUPPORT_UTF8
-Line 1376 
 for (;;)
+Line 1660 
 for (;;)
      ecode++;
      break;
- #ifdef SUPPORT_UCP
+     case OP_ANYNL:
-     /* Check the next character by Unicode property. We will get here only
+     if (eptr >= md->end_subject)
-     if the support is in the binary; otherwise a compile-time error occurs. */
+       {
+       SCHECK_PARTIAL();
-     case OP_PROP:
+       RRETURN(MATCH_NOMATCH);
-     case OP_NOTPROP:
+       }
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
      GETCHARINCTEST(c, eptr);
+     switch(c)
        {
-       int chartype, script;
+       default: RRETURN(MATCH_NOMATCH);
-       int category = _pcre_ucp_findprop(c, &chartype, &script);
+       case 0x000d:
+       if (eptr < md->end_subject && *eptr == 0x0a) eptr++;
+       break;
-       switch(ecode[1])
+       case 0x000a:
-         {
+       break;
-         case PT_ANY:
-         if (op == OP_NOTPROP) RRETURN(MATCH_NOMATCH);
-         break;
-         case PT_LAMP:
+       case 0x000b:
-         if ((chartype == ucp_Lu ||
+       case 0x000c:
-              chartype == ucp_Ll ||
+       case 0x0085:
-              chartype == ucp_Lt) == (op == OP_NOTPROP))
+       case 0x2028:
-           RRETURN(MATCH_NOMATCH);
+       case 0x2029:
-          break;
+       if (md->bsr_anycrlf) RRETURN(MATCH_NOMATCH);
+       break;
+       }
+     ecode++;
+     break;
+     case OP_NOT_HSPACE:
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
+     GETCHARINCTEST(c, eptr);
+     switch(c)
+       {
+       default: break;
+       case 0x09:      /* HT */
+       case 0x20:      /* SPACE */
+       case 0xa0:      /* NBSP */
+       case 0x1680:    /* OGHAM SPACE MARK */
+       case 0x180e:    /* MONGOLIAN VOWEL SEPARATOR */
+       case 0x2000:    /* EN QUAD */
+       case 0x2001:    /* EM QUAD */
+       case 0x2002:    /* EN SPACE */
+       case 0x2003:    /* EM SPACE */
+       case 0x2004:    /* THREE-PER-EM SPACE */
+       case 0x2005:    /* FOUR-PER-EM SPACE */
+       case 0x2006:    /* SIX-PER-EM SPACE */
+       case 0x2007:    /* FIGURE SPACE */
+       case 0x2008:    /* PUNCTUATION SPACE */
+       case 0x2009:    /* THIN SPACE */
+       case 0x200A:    /* HAIR SPACE */
+       case 0x202f:    /* NARROW NO-BREAK SPACE */
+       case 0x205f:    /* MEDIUM MATHEMATICAL SPACE */
+       case 0x3000:    /* IDEOGRAPHIC SPACE */
+       RRETURN(MATCH_NOMATCH);
+       }
+     ecode++;
+     break;
+     case OP_HSPACE:
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
+     GETCHARINCTEST(c, eptr);
+     switch(c)
+       {
+       default: RRETURN(MATCH_NOMATCH);
+       case 0x09:      /* HT */
+       case 0x20:      /* SPACE */
+       case 0xa0:      /* NBSP */
+       case 0x1680:    /* OGHAM SPACE MARK */
+       case 0x180e:    /* MONGOLIAN VOWEL SEPARATOR */
+       case 0x2000:    /* EN QUAD */
+       case 0x2001:    /* EM QUAD */
+       case 0x2002:    /* EN SPACE */
+       case 0x2003:    /* EM SPACE */
+       case 0x2004:    /* THREE-PER-EM SPACE */
+       case 0x2005:    /* FOUR-PER-EM SPACE */
+       case 0x2006:    /* SIX-PER-EM SPACE */
+       case 0x2007:    /* FIGURE SPACE */
+       case 0x2008:    /* PUNCTUATION SPACE */
+       case 0x2009:    /* THIN SPACE */
+       case 0x200A:    /* HAIR SPACE */
+       case 0x202f:    /* NARROW NO-BREAK SPACE */
+       case 0x205f:    /* MEDIUM MATHEMATICAL SPACE */
+       case 0x3000:    /* IDEOGRAPHIC SPACE */
+       break;
+       }
+     ecode++;
+     break;
+     case OP_NOT_VSPACE:
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
+     GETCHARINCTEST(c, eptr);
+     switch(c)
+       {
+       default: break;
+       case 0x0a:      /* LF */
+       case 0x0b:      /* VT */
+       case 0x0c:      /* FF */
+       case 0x0d:      /* CR */
+       case 0x85:      /* NEL */
+       case 0x2028:    /* LINE SEPARATOR */
+       case 0x2029:    /* PARAGRAPH SEPARATOR */
+       RRETURN(MATCH_NOMATCH);
+       }
+     ecode++;
+     break;
+     case OP_VSPACE:
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
+     GETCHARINCTEST(c, eptr);
+     switch(c)
+       {
+       default: RRETURN(MATCH_NOMATCH);
+       case 0x0a:      /* LF */
+       case 0x0b:      /* VT */
+       case 0x0c:      /* FF */
+       case 0x0d:      /* CR */
+       case 0x85:      /* NEL */
+       case 0x2028:    /* LINE SEPARATOR */
+       case 0x2029:    /* PARAGRAPH SEPARATOR */
+       break;
+       }
+     ecode++;
+     break;
+ #ifdef SUPPORT_UCP
+     /* Check the next character by Unicode property. We will get here only
+     if the support is in the binary; otherwise a compile-time error occurs. */
+     case OP_PROP:
+     case OP_NOTPROP:
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
+     GETCHARINCTEST(c, eptr);
+       {
+       const ucd_record *prop = GET_UCD(c);
+       switch(ecode[1])
+         {
+         case PT_ANY:
+         if (op == OP_NOTPROP) RRETURN(MATCH_NOMATCH);
+         break;
+         case PT_LAMP:
+         if ((prop->chartype == ucp_Lu ||
+              prop->chartype == ucp_Ll ||
+              prop->chartype == ucp_Lt) == (op == OP_NOTPROP))
+           RRETURN(MATCH_NOMATCH);
+          break;
          case PT_GC:
-         if ((ecode[2] != category) == (op == OP_PROP))
+         if ((ecode[2] != _pcre_ucp_gentype[prop->chartype]) == (op == OP_PROP))
            RRETURN(MATCH_NOMATCH);
          break;
          case PT_PC:
-         if ((ecode[2] != chartype) == (op == OP_PROP))
+         if ((ecode[2] != prop->chartype) == (op == OP_PROP))
            RRETURN(MATCH_NOMATCH);
          break;
          case PT_SC:
-         if ((ecode[2] != script) == (op == OP_PROP))
+         if ((ecode[2] != prop->script) == (op == OP_PROP))
            RRETURN(MATCH_NOMATCH);
          break;
          default:
          RRETURN(PCRE_ERROR_INTERNAL);
-         break;
          }
        ecode += 3;
-Line 1429 
 for (;;)
+Line 1855 
 for (;;)
      is in the binary; otherwise a compile-time error occurs. */
      case OP_EXTUNI:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      GETCHARINCTEST(c, eptr);
        {
-       int chartype, script;
+       int category = UCD_CATEGORY(c);
-       int category = _pcre_ucp_findprop(c, &chartype, &script);
        if (category == ucp_M) RRETURN(MATCH_NOMATCH);
        while (eptr < md->end_subject)
          {
-Line 1442 
 for (;;)
+Line 1871 
 for (;;)
            {
            GETCHARLEN(c, eptr, len);
            }
-         category = _pcre_ucp_findprop(c, &chartype, &script);
+         category = UCD_CATEGORY(c);
          if (category != ucp_M) break;
          eptr += len;
          }
-Line 1463 
 for (;;)
+Line 1892 
 for (;;)
      case OP_REF:
        {
        offset = GET2(ecode, 1) << 1;               /* Doubled ref number */
-       ecode += 3;                                 /* Advance past item */
+       ecode += 3;
+       /* If the reference is unset, there are two possibilities:
-       /* If the reference is unset, set the length to be longer than the amount
+       (a) In the default, Perl-compatible state, set the length to be longer
-       of subject left; this ensures that every attempt at a match fails. We
+       than the amount of subject left; this ensures that every attempt at a
-       can't just fail here, because of the possibility of quantifiers with zero
+       match fails. We can't just fail here, because of the possibility of
-       minima. */
+       quantifiers with zero minima.
-       length = (offset >= offset_top || md->offset_vector[offset] < 0)?
+       (b) If the JavaScript compatibility flag is set, set the length to zero
-         md->end_subject - eptr + 1 :
+       so that the back reference matches an empty string.
-         md->offset_vector[offset+1] - md->offset_vector[offset];
+       Otherwise, set the length to the length of what was matched by the
+       referenced subpattern. */
+       if (offset >= offset_top || md->offset_vector[offset] < 0)
+         length = (md->jscript_compat)? 0 : md->end_subject - eptr + 1;
+       else
+         length = md->offset_vector[offset+1] - md->offset_vector[offset];
        /* Set up for repetition, or handle the non-repeated case */
-Line 1501 
 for (;;)
+Line 1939 
 for (;;)
          break;
          default:               /* No repeat follows */
-         if (!match_ref(offset, eptr, length, md, ims)) RRETURN(MATCH_NOMATCH);
+         if (!match_ref(offset, eptr, length, md, ims))
+           {
+           CHECK_PARTIAL();
+           RRETURN(MATCH_NOMATCH);
+           }
          eptr += length;
          continue;              /* With the main loop */
          }
-Line 1517 
 for (;;)
+Line 1959 
 for (;;)
        for (i = 1; i <= min; i++)
          {
-         if (!match_ref(offset, eptr, length, md, ims)) RRETURN(MATCH_NOMATCH);
+         if (!match_ref(offset, eptr, length, md, ims))
+           {
+           CHECK_PARTIAL();
+           RRETURN(MATCH_NOMATCH);
+           }
          eptr += length;
          }
-Line 1532 
 for (;;)
+Line 1978 
 for (;;)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM14);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-           if (fi >= max || !match_ref(offset, eptr, length, md, ims))
+           if (fi >= max) RRETURN(MATCH_NOMATCH);
+           if (!match_ref(offset, eptr, length, md, ims))
+             {
+             CHECK_PARTIAL();
              RRETURN(MATCH_NOMATCH);
+             }
            eptr += length;
            }
          /* Control never gets here */
-Line 1553 
 for (;;)
+Line 2003 
 for (;;)
            }
          while (eptr >= pp)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM15);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            eptr -= length;
            }
-Line 1562 
 for (;;)
+Line 2012 
 for (;;)
        }
      /* Control never gets here */
      /* Match a bit-mapped character class, possibly repeatedly. This op code is
      used when all the characters in the class have values in the range 0-255,
      and either the matching is caseful, or the characters are in the range
-Line 1618 
 for (;;)
+Line 2066 
 for (;;)
          {
          for (i = 1; i <= min; i++)
            {
-           if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            GETCHARINC(c, eptr);
            if (c > 255)
              {
-Line 1636 
 for (;;)
+Line 2088 
 for (;;)
          {
          for (i = 1; i <= min; i++)
            {
-           if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            c = *eptr++;
            if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
            }
-Line 1658 
 for (;;)
+Line 2114 
 for (;;)
            {
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM16);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
              GETCHARINC(c, eptr);
              if (c > 255)
                {
-Line 1678 
 for (;;)
+Line 2139 
 for (;;)
            {
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM17);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
              c = *eptr++;
              if ((data[c/8] & (1 << (c&7))) == 0) RRETURN(MATCH_NOMATCH);
              }
-Line 1715 
 for (;;)
+Line 2181 
 for (;;)
              }
            for (;;)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM18);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (eptr-- == pp) break;        /* Stop if tried at original pos */
              BACKCHAR(eptr);
-Line 1734 
 for (;;)
+Line 2200 
 for (;;)
              }
            while (eptr >= pp)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM19);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              eptr--;
              }
-Line 1747 
 for (;;)
+Line 2213 
 for (;;)
      /* Match an extended character class. This opcode is encountered only
-     in UTF-8 mode, because that's the only time it is compiled. */
+     when UTF-8 mode mode is supported. Nevertheless, we may not be in UTF-8
+     mode, because Unicode properties are supported in non-UTF-8 mode. */
  #ifdef SUPPORT_UTF8
      case OP_XCLASS:
-Line 1788 
 for (;;)
+Line 2255 
 for (;;)
        for (i = 1; i <= min; i++)
          {
-         if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+         if (eptr >= md->end_subject)
-         GETCHARINC(c, eptr);
+           {
+           SCHECK_PARTIAL();
+           RRETURN(MATCH_NOMATCH);
+           }
+         GETCHARINCTEST(c, eptr);
          if (!_pcre_xclass(c, data)) RRETURN(MATCH_NOMATCH);
          }
-Line 1805 
 for (;;)
+Line 2276 
 for (;;)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM20);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-           if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+           if (fi >= max) RRETURN(MATCH_NOMATCH);
-           GETCHARINC(c, eptr);
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           GETCHARINCTEST(c, eptr);
            if (!_pcre_xclass(c, data)) RRETURN(MATCH_NOMATCH);
            }
          /* Control never gets here */
-Line 1823 
 for (;;)
+Line 2299 
 for (;;)
            {
            int len = 1;
            if (eptr >= md->end_subject) break;
-           GETCHARLEN(c, eptr, len);
+           GETCHARLENTEST(c, eptr, len);
            if (!_pcre_xclass(c, data)) break;
            eptr += len;
            }
          for(;;)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM21);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (eptr-- == pp) break;        /* Stop if tried at original pos */
-           BACKCHAR(eptr)
+           if (utf8) BACKCHAR(eptr);
            }
          RRETURN(MATCH_NOMATCH);
          }
-Line 1850 
 for (;;)
+Line 2326 
 for (;;)
        length = 1;
        ecode++;
        GETCHARLEN(fc, ecode, length);
-       if (length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
+       if (length > md->end_subject - eptr)
+         {
+         CHECK_PARTIAL();             /* Not SCHECK_PARTIAL() */
+         RRETURN(MATCH_NOMATCH);
+         }
        while (length-- > 0) if (*ecode++ != *eptr++) RRETURN(MATCH_NOMATCH);
        }
      else
-Line 1858 
 for (;;)
+Line 2338 
 for (;;)
      /* Non-UTF-8 mode */
        {
-       if (md->end_subject - eptr < 1) RRETURN(MATCH_NOMATCH);
+       if (md->end_subject - eptr < 1)
+         {
+         SCHECK_PARTIAL();            /* This one can use SCHECK_PARTIAL() */
+         RRETURN(MATCH_NOMATCH);
+         }
        if (ecode[1] != *eptr++) RRETURN(MATCH_NOMATCH);
        ecode += 2;
        }
-Line 1874 
 for (;;)
+Line 2358 
 for (;;)
        ecode++;
        GETCHARLEN(fc, ecode, length);
-       if (length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
+       if (length > md->end_subject - eptr)
+         {
+         CHECK_PARTIAL();             /* Not SCHECK_PARTIAL() */
+         RRETURN(MATCH_NOMATCH);
+         }
        /* If the pattern character's value is < 128, we have only one byte, and
        can use the fast lookup table. */
-Line 1888 
 for (;;)
+Line 2376 
 for (;;)
        else
          {
-         int dc;
+         unsigned int dc;
          GETCHARINC(dc, eptr);
          ecode += length;
-Line 1898 
 for (;;)
+Line 2386 
 for (;;)
          if (fc != dc)
            {
  #ifdef SUPPORT_UCP
-           if (dc != _pcre_ucp_othercase(fc))
+           if (dc != UCD_OTHERCASE(fc))
  #endif
              RRETURN(MATCH_NOMATCH);
            }
-Line 1909 
 for (;;)
+Line 2397 
 for (;;)
      /* Non-UTF-8 mode */
        {
-       if (md->end_subject - eptr < 1) RRETURN(MATCH_NOMATCH);
+       if (md->end_subject - eptr < 1)
+         {
+         SCHECK_PARTIAL();            /* This one can use SCHECK_PARTIAL() */
+         RRETURN(MATCH_NOMATCH);
+         }
        if (md->lcc[ecode[1]] != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);
        ecode += 2;
        }
      break;
-     /* Match a single character repeatedly; different opcodes share code. */
+     /* Match a single character repeatedly. */
      case OP_EXACT:
      min = max = GET2(ecode, 1);
      ecode += 3;
      goto REPEATCHAR;
+     case OP_POSUPTO:
+     possessive = TRUE;
+     /* Fall through */
      case OP_UPTO:
      case OP_MINUPTO:
      min = 0;
-Line 1930 
 for (;;)
+Line 2426 
 for (;;)
      ecode += 3;
      goto REPEATCHAR;
+     case OP_POSSTAR:
+     possessive = TRUE;
+     min = 0;
+     max = INT_MAX;
+     ecode++;
+     goto REPEATCHAR;
+     case OP_POSPLUS:
+     possessive = TRUE;
+     min = 1;
+     max = INT_MAX;
+     ecode++;
+     goto REPEATCHAR;
+     case OP_POSQUERY:
+     possessive = TRUE;
+     min = 0;
+     max = 1;
+     ecode++;
+     goto REPEATCHAR;
      case OP_STAR:
      case OP_MINSTAR:
      case OP_PLUS:
-Line 1938 
 for (;;)
+Line 2455 
 for (;;)
      case OP_MINQUERY:
      c = *ecode++ - OP_STAR;
      minimize = (c & 1) != 0;
      min = rep_min[c];                 /* Pick up values from tables; */
      max = rep_max[c];                 /* zero for max => infinity */
      if (max == 0) max = INT_MAX;
-     /* Common code for all repeated single-character matches. We can give
+     /* Common code for all repeated single-character matches. */
-     up quickly if there are fewer than the minimum number of characters left in
-     the subject. */
      REPEATCHAR:
  #ifdef SUPPORT_UTF8
-Line 1953 
 for (;;)
+Line 2469 
 for (;;)
        length = 1;
        charptr = ecode;
        GETCHARLEN(fc, ecode, length);
-       if (min * length > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
        ecode += length;
        /* Handle multibyte character matching specially here. There is
-Line 1961 
 for (;;)
+Line 2476 
 for (;;)
        if (length > 1)
          {
-         int oclength = 0;
-         uschar occhars[8];
  #ifdef SUPPORT_UCP
-         int othercase;
+         unsigned int othercase;
          if ((ims & PCRE_CASELESS) != 0 &&
-             (othercase = _pcre_ucp_othercase(fc)) >= 0 &&
+             (othercase = UCD_OTHERCASE(fc)) != fc)
-              othercase >= 0)
            oclength = _pcre_ord2utf8(othercase, occhars);
+         else oclength = 0;
  #endif  /* SUPPORT_UCP */
          for (i = 1; i <= min; i++)
            {
-           if (memcmp(eptr, charptr, length) == 0) eptr += length;
+           if (eptr <= md->end_subject - length &&
-           /* Need braces because of following else */
+             memcmp(eptr, charptr, length) == 0) eptr += length;
-           else if (oclength == 0) { RRETURN(MATCH_NOMATCH); }
+ #ifdef SUPPORT_UCP
+           else if (oclength > 0 &&
+                    eptr <= md->end_subject - oclength &&
+                    memcmp(eptr, occhars, oclength) == 0) eptr += oclength;
+ #endif  /* SUPPORT_UCP */
            else
              {
-             if (memcmp(eptr, occhars, oclength) != 0) RRETURN(MATCH_NOMATCH);
+             CHECK_PARTIAL();
-             eptr += oclength;
+             RRETURN(MATCH_NOMATCH);
              }
            }
-Line 1990 
 for (;;)
+Line 2506 
 for (;;)
            {
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM22);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
-             if (memcmp(eptr, charptr, length) == 0) eptr += length;
+             if (eptr <= md->end_subject - length &&
-             /* Need braces because of following else */
+               memcmp(eptr, charptr, length) == 0) eptr += length;
-             else if (oclength == 0) { RRETURN(MATCH_NOMATCH); }
+ #ifdef SUPPORT_UCP
+             else if (oclength > 0 &&
+                      eptr <= md->end_subject - oclength &&
+                      memcmp(eptr, occhars, oclength) == 0) eptr += oclength;
+ #endif  /* SUPPORT_UCP */
              else
                {
-               if (memcmp(eptr, occhars, oclength) != 0) RRETURN(MATCH_NOMATCH);
+               CHECK_PARTIAL();
-               eptr += oclength;
+               RRETURN(MATCH_NOMATCH);
                }
              }
            /* Control never gets here */
            }
-         else
+         else  /* Maximize */
            {
            pp = eptr;
            for (i = min; i < max; i++)
              {
-             if (eptr > md->end_subject - length) break;
+             if (eptr <= md->end_subject - length &&
-             if (memcmp(eptr, charptr, length) == 0) eptr += length;
+                 memcmp(eptr, charptr, length) == 0) eptr += length;
-             else if (oclength == 0) break;
+ #ifdef SUPPORT_UCP
-             else
+             else if (oclength > 0 &&
-               {
+                      eptr <= md->end_subject - oclength &&
-               if (memcmp(eptr, occhars, oclength) != 0) break;
+                      memcmp(eptr, occhars, oclength) == 0) eptr += oclength;
-               eptr += oclength;
+ #endif  /* SUPPORT_UCP */
-               }
+             else break;
+             }
+           if (possessive) continue;
+           for(;;)
+             {
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM23);
+             if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+             if (eptr == pp) { RRETURN(MATCH_NOMATCH); }
+ #ifdef SUPPORT_UCP
+             eptr--;
+             BACKCHAR(eptr);
+ #else   /* without SUPPORT_UCP */
+             eptr -= length;
+ #endif  /* SUPPORT_UCP */
              }
-           while (eptr >= pp)
-            {
-            RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
-            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-            eptr -= length;
-            }
-           RRETURN(MATCH_NOMATCH);
            }
          /* Control never gets here */
          }
-Line 2037 
 for (;;)
+Line 2566 
 for (;;)
  #endif  /* SUPPORT_UTF8 */
      /* When not in UTF-8 mode, load a single-byte character. */
-       {
-       if (min > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
+     fc = *ecode++;
-       fc = *ecode++;
-       }
      /* The value of fc at this point is always less than 256, though we may or
      may not be in UTF-8 mode. The code is duplicated for the caseless and
-Line 2058 
 for (;;)
+Line 2585 
 for (;;)
        {
        fc = md->lcc[fc];
        for (i = 1; i <= min; i++)
+         {
+         if (eptr >= md->end_subject)
+           {
+           SCHECK_PARTIAL();
+           RRETURN(MATCH_NOMATCH);
+           }
          if (fc != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);
+         }
        if (min == max) continue;
        if (minimize)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM24);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-           if (fi >= max || eptr >= md->end_subject ||
+           if (fi >= max) RRETURN(MATCH_NOMATCH);
-               fc != md->lcc[*eptr++])
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
              RRETURN(MATCH_NOMATCH);
+             }
+           if (fc != md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);
            }
          /* Control never gets here */
          }
-       else
+       else  /* Maximize */
          {
          pp = eptr;
          for (i = min; i < max; i++)
-Line 2080 
 for (;;)
+Line 2618 
 for (;;)
            if (eptr >= md->end_subject || fc != md->lcc[*eptr]) break;
            eptr++;
            }
+         if (possessive) continue;
          while (eptr >= pp)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM25);
            eptr--;
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            }
-Line 2095 
 for (;;)
+Line 2636 
 for (;;)
      else
        {
-       for (i = 1; i <= min; i++) if (fc != *eptr++) RRETURN(MATCH_NOMATCH);
+       for (i = 1; i <= min; i++)
+         {
+         if (eptr >= md->end_subject)
+           {
+           SCHECK_PARTIAL();
+           RRETURN(MATCH_NOMATCH);
+           }
+         if (fc != *eptr++) RRETURN(MATCH_NOMATCH);
+         }
        if (min == max) continue;
        if (minimize)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM26);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-           if (fi >= max || eptr >= md->end_subject || fc != *eptr++)
+           if (fi >= max) RRETURN(MATCH_NOMATCH);
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
              RRETURN(MATCH_NOMATCH);
+             }
+           if (fc != *eptr++) RRETURN(MATCH_NOMATCH);
            }
          /* Control never gets here */
          }
-       else
+       else  /* Maximize */
          {
          pp = eptr;
          for (i = min; i < max; i++)
-Line 2116 
 for (;;)
+Line 2672 
 for (;;)
            if (eptr >= md->end_subject || fc != *eptr) break;
            eptr++;
            }
+         if (possessive) continue;
          while (eptr >= pp)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM27);
            eptr--;
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            }
-Line 2131 
 for (;;)
+Line 2689 
 for (;;)
      checking can be multibyte. */
      case OP_NOT:
-     if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+     if (eptr >= md->end_subject)
+       {
+       SCHECK_PARTIAL();
+       RRETURN(MATCH_NOMATCH);
+       }
      ecode++;
      GETCHARINCTEST(c, eptr);
      if ((ims & PCRE_CASELESS) != 0)
-Line 2168 
 for (;;)
+Line 2730 
 for (;;)
      ecode += 3;
      goto REPEATNOTCHAR;
+     case OP_NOTPOSSTAR:
+     possessive = TRUE;
+     min = 0;
+     max = INT_MAX;
+     ecode++;
+     goto REPEATNOTCHAR;
+     case OP_NOTPOSPLUS:
+     possessive = TRUE;
+     min = 1;
+     max = INT_MAX;
+     ecode++;
+     goto REPEATNOTCHAR;
+     case OP_NOTPOSQUERY:
+     possessive = TRUE;
+     min = 0;
+     max = 1;
+     ecode++;
+     goto REPEATNOTCHAR;
+     case OP_NOTPOSUPTO:
+     possessive = TRUE;
+     min = 0;
+     max = GET2(ecode, 1);
+     ecode += 3;
+     goto REPEATNOTCHAR;
      case OP_NOTSTAR:
      case OP_NOTMINSTAR:
      case OP_NOTPLUS:
-Line 2180 
 for (;;)
+Line 2770 
 for (;;)
      max = rep_max[c];                 /* zero for max => infinity */
      if (max == 0) max = INT_MAX;
-     /* Common code for all repeated single-byte matches. We can give up quickly
+     /* Common code for all repeated single-byte matches. */
-     if there are fewer than the minimum number of bytes left in the
-     subject. */
      REPEATNOTCHAR:
-     if (min > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
      fc = *ecode++;
      /* The code is duplicated for the caseless and caseful cases, for speed,
-Line 2207 
 for (;;)
+Line 2794 
 for (;;)
        /* UTF-8 mode */
        if (utf8)
          {
-         register int d;
+         register unsigned int d;
          for (i = 1; i <= min; i++)
            {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            GETCHARINC(d, eptr);
            if (d < 256) d = md->lcc[d];
            if (fc == d) RRETURN(MATCH_NOMATCH);
-Line 2221 
 for (;;)
+Line 2813 
 for (;;)
        /* Not UTF-8 mode */
          {
          for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            if (fc == md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);
+           }
          }
        if (min == max) continue;
-Line 2232 
 for (;;)
+Line 2831 
 for (;;)
          /* UTF-8 mode */
          if (utf8)
            {
-           register int d;
+           register unsigned int d;
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM28);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
              GETCHARINC(d, eptr);
              if (d < 256) d = md->lcc[d];
-             if (fi >= max || eptr >= md->end_subject || fc == d)
+             if (fc == d) RRETURN(MATCH_NOMATCH);
-               RRETURN(MATCH_NOMATCH);
              }
            }
          else
-Line 2249 
 for (;;)
+Line 2853 
 for (;;)
            {
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM29);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max || eptr >= md->end_subject || fc == md->lcc[*eptr++])
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
                RRETURN(MATCH_NOMATCH);
+               }
+             if (fc == md->lcc[*eptr++]) RRETURN(MATCH_NOMATCH);
              }
            }
          /* Control never gets here */
-Line 2268 
 for (;;)
+Line 2877 
 for (;;)
          /* UTF-8 mode */
          if (utf8)
            {
-           register int d;
+           register unsigned int d;
            for (i = min; i < max; i++)
              {
              int len = 1;
-Line 2278 
 for (;;)
+Line 2887 
 for (;;)
              if (fc == d) break;
              eptr += len;
              }
-           for(;;)
+         if (possessive) continue;
+         for(;;)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM30);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (eptr-- == pp) break;        /* Stop if tried at original pos */
              BACKCHAR(eptr);
-Line 2295 
 for (;;)
+Line 2905 
 for (;;)
              if (eptr >= md->end_subject || fc == md->lcc[*eptr]) break;
              eptr++;
              }
+           if (possessive) continue;
            while (eptr >= pp)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM31);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              eptr--;
              }
-Line 2316 
 for (;;)
+Line 2927 
 for (;;)
        /* UTF-8 mode */
        if (utf8)
          {
-         register int d;
+         register unsigned int d;
          for (i = 1; i <= min; i++)
            {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            GETCHARINC(d, eptr);
            if (fc == d) RRETURN(MATCH_NOMATCH);
            }
-Line 2328 
 for (;;)
+Line 2944 
 for (;;)
        /* Not UTF-8 mode */
          {
          for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            if (fc == *eptr++) RRETURN(MATCH_NOMATCH);
+           }
          }
        if (min == max) continue;
-Line 2339 
 for (;;)
+Line 2962 
 for (;;)
          /* UTF-8 mode */
          if (utf8)
            {
-           register int d;
+           register unsigned int d;
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM32);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             GETCHARINC(d, eptr);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
-             if (fi >= max || eptr >= md->end_subject || fc == d)
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
                RRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINC(d, eptr);
+             if (fc == d) RRETURN(MATCH_NOMATCH);
              }
            }
          else
-Line 2355 
 for (;;)
+Line 2983 
 for (;;)
            {
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM33);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max || eptr >= md->end_subject || fc == *eptr++)
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
                RRETURN(MATCH_NOMATCH);
+               }
+             if (fc == *eptr++) RRETURN(MATCH_NOMATCH);
              }
            }
          /* Control never gets here */
-Line 2374 
 for (;;)
+Line 3007 
 for (;;)
          /* UTF-8 mode */
          if (utf8)
            {
-           register int d;
+           register unsigned int d;
            for (i = min; i < max; i++)
              {
              int len = 1;
-Line 2383 
 for (;;)
+Line 3016 
 for (;;)
              if (fc == d) break;
              eptr += len;
              }
+           if (possessive) continue;
            for(;;)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM34);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (eptr-- == pp) break;        /* Stop if tried at original pos */
              BACKCHAR(eptr);
-Line 2400 
 for (;;)
+Line 3034 
 for (;;)
              if (eptr >= md->end_subject || fc == *eptr) break;
              eptr++;
              }
+           if (possessive) continue;
            while (eptr >= pp)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM35);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              eptr--;
              }
-Line 2431 
 for (;;)
+Line 3066 
 for (;;)
      ecode += 3;
      goto REPEATTYPE;
+     case OP_TYPEPOSSTAR:
+     possessive = TRUE;
+     min = 0;
+     max = INT_MAX;
+     ecode++;
+     goto REPEATTYPE;
+     case OP_TYPEPOSPLUS:
+     possessive = TRUE;
+     min = 1;
+     max = INT_MAX;
+     ecode++;
+     goto REPEATTYPE;
+     case OP_TYPEPOSQUERY:
+     possessive = TRUE;
+     min = 0;
+     max = 1;
+     ecode++;
+     goto REPEATTYPE;
+     case OP_TYPEPOSUPTO:
+     possessive = TRUE;
+     min = 0;
+     max = GET2(ecode, 1);
+     ecode += 3;
+     goto REPEATTYPE;
      case OP_TYPESTAR:
      case OP_TYPEMINSTAR:
      case OP_TYPEPLUS:
-Line 2462 
 for (;;)
+Line 3125 
 for (;;)
      /* First, ensure the minimum number of matches are present. Use inline
      code for maximizing the speed, and do the type test once at the start
-     (i.e. keep it out of the loop). Also we can test that there are at least
+     (i.e. keep it out of the loop). Separate the UTF-8 code completely as that
-     the minimum number of bytes before we start. This isn't as effective in
-     UTF-8 mode, but it does no harm. Separate the UTF-8 code completely as that
      is tidier. Also separate the UCP code, which can be the same for both UTF-8
      and single-bytes. */
-     if (min > md->end_subject - eptr) RRETURN(MATCH_NOMATCH);
      if (min > 0)
        {
  #ifdef SUPPORT_UCP
-Line 2480 
 for (;;)
+Line 3140 
 for (;;)
            if (prop_fail_result) RRETURN(MATCH_NOMATCH);
            for (i = 1; i <= min; i++)
              {
-             if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
-             GETCHARINC(c, eptr);
+               {
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
              }
            break;
            case PT_LAMP:
            for (i = 1; i <= min; i++)
              {
-             if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
-             GETCHARINC(c, eptr);
+               {
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+               SCHECK_PARTIAL();
-             if ((prop_chartype == ucp_Lu ||
+               RRETURN(MATCH_NOMATCH);
-                  prop_chartype == ucp_Ll ||
+               }
+             GETCHARINCTEST(c, eptr);
+             prop_chartype = UCD_CHARTYPE(c);
+             if ((prop_chartype == ucp_Lu ||
+                  prop_chartype == ucp_Ll ||
                   prop_chartype == ucp_Lt) == prop_fail_result)
                RRETURN(MATCH_NOMATCH);
              }
-Line 2501 
 for (;;)
+Line 3169 
 for (;;)
            case PT_GC:
            for (i = 1; i <= min; i++)
              {
-             if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
-             GETCHARINC(c, eptr);
+               {
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             prop_category = UCD_CATEGORY(c);
              if ((prop_category == prop_value) == prop_fail_result)
                RRETURN(MATCH_NOMATCH);
              }
-Line 2512 
 for (;;)
+Line 3184 
 for (;;)
            case PT_PC:
            for (i = 1; i <= min; i++)
              {
-             if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
-             GETCHARINC(c, eptr);
+               {
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             prop_chartype = UCD_CHARTYPE(c);
              if ((prop_chartype == prop_value) == prop_fail_result)
                RRETURN(MATCH_NOMATCH);
              }
-Line 2523 
 for (;;)
+Line 3199 
 for (;;)
            case PT_SC:
            for (i = 1; i <= min; i++)
              {
-             if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
-             GETCHARINC(c, eptr);
+               {
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
+             GETCHARINCTEST(c, eptr);
+             prop_script = UCD_SCRIPT(c);
              if ((prop_script == prop_value) == prop_fail_result)
                RRETURN(MATCH_NOMATCH);
              }
-Line 2533 
 for (;;)
+Line 3213 
 for (;;)
            default:
            RRETURN(PCRE_ERROR_INTERNAL);
-           break;
            }
          }
-Line 2544 
 for (;;)
+Line 3223 
 for (;;)
          {
          for (i = 1; i <= min; i++)
            {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            GETCHARINCTEST(c, eptr);
-           prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+           prop_category = UCD_CATEGORY(c);
            if (prop_category == ucp_M) RRETURN(MATCH_NOMATCH);
            while (eptr < md->end_subject)
              {
              int len = 1;
-             if (!utf8) c = *eptr; else
+             if (!utf8) c = *eptr;
-               {
+               else { GETCHARLEN(c, eptr, len); }
-               GETCHARLEN(c, eptr, len);
+             prop_category = UCD_CATEGORY(c);
-               }
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
              if (prop_category != ucp_M) break;
              eptr += len;
              }
-Line 2572 
 for (;;)
+Line 3254 
 for (;;)
          case OP_ANY:
          for (i = 1; i <= min; i++)
            {
-           if (eptr >= md->end_subject ||
+           if (eptr >= md->end_subject)
-              (*eptr++ == NEWLINE && (ims & PCRE_DOTALL) == 0))
+             {
+             SCHECK_PARTIAL();
              RRETURN(MATCH_NOMATCH);
+             }
+           if (IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH);
+           eptr++;
+           while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
+           }
+         break;
+         case OP_ALLANY:
+         for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           eptr++;
            while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
            }
          break;
          case OP_ANYBYTE:
+         if (eptr > md->end_subject - min) RRETURN(MATCH_NOMATCH);
          eptr += min;
          break;
+         case OP_ANYNL:
+         for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           GETCHARINC(c, eptr);
+           switch(c)
+             {
+             default: RRETURN(MATCH_NOMATCH);
+             case 0x000d:
+             if (eptr < md->end_subject && *eptr == 0x0a) eptr++;
+             break;
+             case 0x000a:
+             break;
+             case 0x000b:
+             case 0x000c:
+             case 0x0085:
+             case 0x2028:
+             case 0x2029:
+             if (md->bsr_anycrlf) RRETURN(MATCH_NOMATCH);
+             break;
+             }
+           }
+         break;
+         case OP_NOT_HSPACE:
+         for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           GETCHARINC(c, eptr);
+           switch(c)
+             {
+             default: break;
+             case 0x09:      /* HT */
+             case 0x20:      /* SPACE */
+             case 0xa0:      /* NBSP */
+             case 0x1680:    /* OGHAM SPACE MARK */
+             case 0x180e:    /* MONGOLIAN VOWEL SEPARATOR */
+             case 0x2000:    /* EN QUAD */
+             case 0x2001:    /* EM QUAD */
+             case 0x2002:    /* EN SPACE */
+             case 0x2003:    /* EM SPACE */
+             case 0x2004:    /* THREE-PER-EM SPACE */
+             case 0x2005:    /* FOUR-PER-EM SPACE */
+             case 0x2006:    /* SIX-PER-EM SPACE */
+             case 0x2007:    /* FIGURE SPACE */
+             case 0x2008:    /* PUNCTUATION SPACE */
+             case 0x2009:    /* THIN SPACE */
+             case 0x200A:    /* HAIR SPACE */
+             case 0x202f:    /* NARROW NO-BREAK SPACE */
+             case 0x205f:    /* MEDIUM MATHEMATICAL SPACE */
+             case 0x3000:    /* IDEOGRAPHIC SPACE */
+             RRETURN(MATCH_NOMATCH);
+             }
+           }
+         break;
+         case OP_HSPACE:
+         for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           GETCHARINC(c, eptr);
+           switch(c)
+             {
+             default: RRETURN(MATCH_NOMATCH);
+             case 0x09:      /* HT */
+             case 0x20:      /* SPACE */
+             case 0xa0:      /* NBSP */
+             case 0x1680:    /* OGHAM SPACE MARK */
+             case 0x180e:    /* MONGOLIAN VOWEL SEPARATOR */
+             case 0x2000:    /* EN QUAD */
+             case 0x2001:    /* EM QUAD */
+             case 0x2002:    /* EN SPACE */
+             case 0x2003:    /* EM SPACE */
+             case 0x2004:    /* THREE-PER-EM SPACE */
+             case 0x2005:    /* FOUR-PER-EM SPACE */
+             case 0x2006:    /* SIX-PER-EM SPACE */
+             case 0x2007:    /* FIGURE SPACE */
+             case 0x2008:    /* PUNCTUATION SPACE */
+             case 0x2009:    /* THIN SPACE */
+             case 0x200A:    /* HAIR SPACE */
+             case 0x202f:    /* NARROW NO-BREAK SPACE */
+             case 0x205f:    /* MEDIUM MATHEMATICAL SPACE */
+             case 0x3000:    /* IDEOGRAPHIC SPACE */
+             break;
+             }
+           }
+         break;
+         case OP_NOT_VSPACE:
+         for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           GETCHARINC(c, eptr);
+           switch(c)
+             {
+             default: break;
+             case 0x0a:      /* LF */
+             case 0x0b:      /* VT */
+             case 0x0c:      /* FF */
+             case 0x0d:      /* CR */
+             case 0x85:      /* NEL */
+             case 0x2028:    /* LINE SEPARATOR */
+             case 0x2029:    /* PARAGRAPH SEPARATOR */
+             RRETURN(MATCH_NOMATCH);
+             }
+           }
+         break;
+         case OP_VSPACE:
+         for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           GETCHARINC(c, eptr);
+           switch(c)
+             {
+             default: RRETURN(MATCH_NOMATCH);
+             case 0x0a:      /* LF */
+             case 0x0b:      /* VT */
+             case 0x0c:      /* FF */
+             case 0x0d:      /* CR */
+             case 0x85:      /* NEL */
+             case 0x2028:    /* LINE SEPARATOR */
+             case 0x2029:    /* PARAGRAPH SEPARATOR */
+             break;
+             }
+           }
+         break;
          case OP_NOT_DIGIT:
          for (i = 1; i <= min; i++)
            {
-           if (eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            GETCHARINC(c, eptr);
            if (c < 128 && (md->ctypes[c] & ctype_digit) != 0)
              RRETURN(MATCH_NOMATCH);
-Line 2596 
 for (;;)
+Line 3450 
 for (;;)
          case OP_DIGIT:
          for (i = 1; i <= min; i++)
            {
-           if (eptr >= md->end_subject ||
+           if (eptr >= md->end_subject)
-              *eptr >= 128 || (md->ctypes[*eptr++] & ctype_digit) == 0)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           if (*eptr >= 128 || (md->ctypes[*eptr++] & ctype_digit) == 0)
              RRETURN(MATCH_NOMATCH);
            /* No need to skip more bytes - we know it's a 1-byte character */
            }
-Line 2606 
 for (;;)
+Line 3464 
 for (;;)
          case OP_NOT_WHITESPACE:
          for (i = 1; i <= min; i++)
            {
-           if (eptr >= md->end_subject ||
+           if (eptr >= md->end_subject)
-              (*eptr < 128 && (md->ctypes[*eptr++] & ctype_space) != 0))
+             {
+             SCHECK_PARTIAL();
              RRETURN(MATCH_NOMATCH);
-           while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
+             }
+           if (*eptr < 128 && (md->ctypes[*eptr] & ctype_space) != 0)
+             RRETURN(MATCH_NOMATCH);
+           while (++eptr < md->end_subject && (*eptr & 0xc0) == 0x80);
            }
          break;
          case OP_WHITESPACE:
          for (i = 1; i <= min; i++)
            {
-           if (eptr >= md->end_subject ||
+           if (eptr >= md->end_subject)
-              *eptr >= 128 || (md->ctypes[*eptr++] & ctype_space) == 0)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           if (*eptr >= 128 || (md->ctypes[*eptr++] & ctype_space) == 0)
              RRETURN(MATCH_NOMATCH);
            /* No need to skip more bytes - we know it's a 1-byte character */
            }
-Line 2627 
 for (;;)
+Line 3493 
 for (;;)
          for (i = 1; i <= min; i++)
            {
            if (eptr >= md->end_subject ||
-              (*eptr < 128 && (md->ctypes[*eptr++] & ctype_word) != 0))
+              (*eptr < 128 && (md->ctypes[*eptr] & ctype_word) != 0))
              RRETURN(MATCH_NOMATCH);
-           while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
+           while (++eptr < md->end_subject && (*eptr & 0xc0) == 0x80);
            }
          break;
          case OP_WORDCHAR:
          for (i = 1; i <= min; i++)
            {
-           if (eptr >= md->end_subject ||
+           if (eptr >= md->end_subject)
-              *eptr >= 128 || (md->ctypes[*eptr++] & ctype_word) == 0)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           if (*eptr >= 128 || (md->ctypes[*eptr++] & ctype_word) == 0)
              RRETURN(MATCH_NOMATCH);
            /* No need to skip more bytes - we know it's a 1-byte character */
            }
-Line 2656 
 for (;;)
+Line 3526 
 for (;;)
        switch(ctype)
          {
          case OP_ANY:
-         if ((ims & PCRE_DOTALL) == 0)
+         for (i = 1; i <= min; i++)
            {
-           for (i = 1; i <= min; i++)
+           if (eptr >= md->end_subject)
-             if (*eptr++ == NEWLINE) RRETURN(MATCH_NOMATCH);
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           if (IS_NEWLINE(eptr)) RRETURN(MATCH_NOMATCH);
+           eptr++;
            }
-         else eptr += min;
+         break;
+         case OP_ALLANY:
+         if (eptr > md->end_subject - min)
+           {
+           SCHECK_PARTIAL();
+           RRETURN(MATCH_NOMATCH);
+           }
+         eptr += min;
          break;
          case OP_ANYBYTE:
+         if (eptr > md->end_subject - min)
+           {
+           SCHECK_PARTIAL();
+           RRETURN(MATCH_NOMATCH);
+           }
          eptr += min;
          break;
+         case OP_ANYNL:
+         for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           switch(*eptr++)
+             {
+             default: RRETURN(MATCH_NOMATCH);
+             case 0x000d:
+             if (eptr < md->end_subject && *eptr == 0x0a) eptr++;
+             break;
+             case 0x000a:
+             break;
+             case 0x000b:
+             case 0x000c:
+             case 0x0085:
+             if (md->bsr_anycrlf) RRETURN(MATCH_NOMATCH);
+             break;
+             }
+           }
+         break;
+         case OP_NOT_HSPACE:
+         for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           switch(*eptr++)
+             {
+             default: break;
+             case 0x09:      /* HT */
+             case 0x20:      /* SPACE */
+             case 0xa0:      /* NBSP */
+             RRETURN(MATCH_NOMATCH);
+             }
+           }
+         break;
+         case OP_HSPACE:
+         for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           switch(*eptr++)
+             {
+             default: RRETURN(MATCH_NOMATCH);
+             case 0x09:      /* HT */
+             case 0x20:      /* SPACE */
+             case 0xa0:      /* NBSP */
+             break;
+             }
+           }
+         break;
+         case OP_NOT_VSPACE:
+         for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           switch(*eptr++)
+             {
+             default: break;
+             case 0x0a:      /* LF */
+             case 0x0b:      /* VT */
+             case 0x0c:      /* FF */
+             case 0x0d:      /* CR */
+             case 0x85:      /* NEL */
+             RRETURN(MATCH_NOMATCH);
+             }
+           }
+         break;
+         case OP_VSPACE:
+         for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           switch(*eptr++)
+             {
+             default: RRETURN(MATCH_NOMATCH);
+             case 0x0a:      /* LF */
+             case 0x0b:      /* VT */
+             case 0x0c:      /* FF */
+             case 0x0d:      /* CR */
+             case 0x85:      /* NEL */
+             break;
+             }
+           }
+         break;
          case OP_NOT_DIGIT:
          for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            if ((md->ctypes[*eptr++] & ctype_digit) != 0) RRETURN(MATCH_NOMATCH);
+           }
          break;
          case OP_DIGIT:
          for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            if ((md->ctypes[*eptr++] & ctype_digit) == 0) RRETURN(MATCH_NOMATCH);
+           }
          break;
          case OP_NOT_WHITESPACE:
          for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            if ((md->ctypes[*eptr++] & ctype_space) != 0) RRETURN(MATCH_NOMATCH);
+           }
          break;
          case OP_WHITESPACE:
          for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            if ((md->ctypes[*eptr++] & ctype_space) == 0) RRETURN(MATCH_NOMATCH);
+           }
          break;
          case OP_NOT_WORDCHAR:
          for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            if ((md->ctypes[*eptr++] & ctype_word) != 0)
              RRETURN(MATCH_NOMATCH);
+           }
          break;
          case OP_WORDCHAR:
          for (i = 1; i <= min; i++)
+           {
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            if ((md->ctypes[*eptr++] & ctype_word) == 0)
              RRETURN(MATCH_NOMATCH);
+           }
          break;
          default:
-Line 2723 
 for (;;)
+Line 3759 
 for (;;)
            case PT_ANY:
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM36);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
              GETCHARINC(c, eptr);
              if (prop_fail_result) RRETURN(MATCH_NOMATCH);
              }
-           break;
+           /* Control never gets here */
            case PT_LAMP:
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM37);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
              GETCHARINC(c, eptr);
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+             prop_chartype = UCD_CHARTYPE(c);
              if ((prop_chartype == ucp_Lu ||
                   prop_chartype == ucp_Ll ||
                   prop_chartype == ucp_Lt) == prop_fail_result)
                RRETURN(MATCH_NOMATCH);
              }
-           break;
+           /* Control never gets here */
            case PT_GC:
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM38);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
              GETCHARINC(c, eptr);
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+             prop_category = UCD_CATEGORY(c);
              if ((prop_category == prop_value) == prop_fail_result)
                RRETURN(MATCH_NOMATCH);
              }
-           break;
+           /* Control never gets here */
            case PT_PC:
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM39);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
              GETCHARINC(c, eptr);
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+             prop_chartype = UCD_CHARTYPE(c);
              if ((prop_chartype == prop_value) == prop_fail_result)
                RRETURN(MATCH_NOMATCH);
              }
-           break;
+           /* Control never gets here */
            case PT_SC:
            for (fi = min;; fi++)
              {
-             RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+             RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM40);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-             if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+             if (fi >= max) RRETURN(MATCH_NOMATCH);
+             if (eptr >= md->end_subject)
+               {
+               SCHECK_PARTIAL();
+               RRETURN(MATCH_NOMATCH);
+               }
              GETCHARINC(c, eptr);
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+             prop_script = UCD_SCRIPT(c);
              if ((prop_script == prop_value) == prop_fail_result)
                RRETURN(MATCH_NOMATCH);
              }
-           break;
+           /* Control never gets here */
            default:
            RRETURN(PCRE_ERROR_INTERNAL);
-           break;
            }
          }
-Line 2798 
 for (;;)
+Line 3858 
 for (;;)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM41);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-           if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+           if (fi >= max) RRETURN(MATCH_NOMATCH);
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
            GETCHARINCTEST(c, eptr);
-           prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
+           prop_category = UCD_CATEGORY(c);
            if (prop_category == ucp_M) RRETURN(MATCH_NOMATCH);
            while (eptr < md->end_subject)
              {
              int len = 1;
-             if (!utf8) c = *eptr; else
+             if (!utf8) c = *eptr;
-               {
+               else { GETCHARLEN(c, eptr, len); }
-               GETCHARLEN(c, eptr, len);
+             prop_category = UCD_CATEGORY(c);
-               }
-             prop_category = _pcre_ucp_findprop(c, &prop_chartype, &prop_script);
              if (prop_category != ucp_M) break;
              eptr += len;
              }
-Line 2827 
 for (;;)
+Line 3890 
 for (;;)
          {
          for (fi = min;; fi++)
            {
-           RMATCH(rrc, eptr, ecode, offset_top, md, ims, eptrb, 0);
+           RMATCH(eptr, ecode, offset_top, md, ims, eptrb, 0, RM42);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-           if (fi >= max || eptr >= md->end_subject) RRETURN(MATCH_NOMATCH);
+           if (fi >= max) RRETURN(MATCH_NOMATCH);
+           if (eptr >= md->end_subject)
+             {
+             SCHECK_PARTIAL();
+             RRETURN(MATCH_NOMATCH);
+             }
+           if (ctype == OP_ANY && IS_NEWLINE(eptr))
+             RRETURN(MATCH_NOMATCH);
            GETCHARINC(c, eptr);
            switch(ctype)
              {
-             case OP_ANY:
+             case OP_ANY:        /* This is the non-NL case */
-             if ((ims & PCRE_DOTALL) == 0 && c == NEWLINE) RRETURN(MATCH_NOMATCH);
+             case OP_ALLANY:
+             case OP_ANYBYTE:
              break;
-             case OP_ANYBYTE:
+             case OP_ANYNL:
+             switch(c)
+               {
+               default: RRETURN(MATCH_NOMATCH);
+               case 0x000d:
+               if (eptr < md->end_subject && *eptr == 0x0a) eptr++;
+               break;
+               case 0x000a:
+               break;
+               case 0x000b:
+               case 0x000c:
+               case 0x0085:
+               case 0x2028:
+               case 0x2029:
+               if (md->bsr_anycrlf) RRETURN(MATCH_NOMATCH);
+               break;
+               }
+             break;
+             case OP_NOT_HSPACE:
+             switch(c)
+               {
+               default: break;
+               case 0x09:      /* HT */
+               case 0x20:      /* SPACE */
+               case 0xa0:      /* NBSP */
+               case 0x1680:    /* OGHAM SPACE MARK */
+               case 0x180e:    /* MONGOLIAN VOWEL SEPARATOR */
+               case 0x2000:    /* EN QUAD */
+               case 0x2001:    /* EM QUAD */
+               case 0x2002:    /* EN SPACE */
+               case 0x2003:    /* EM SPACE */
+               case 0x2004:    /* THREE-PER-EM SPACE */
+               case 0x2005:    /* FOUR-PER-EM SPACE */
+               case 0x2006:    /* SIX-PER-EM SPACE */
+               case 0x2007:    /* FIGURE SPACE */
+               case 0x2008:    /* PUNCTUATION SPACE */
+               case 0x2009:    /* THIN SPACE */
+               case 0x200A:    /* HAIR SPACE */
+               case 0x202f:    /* NARROW NO-BREAK SPACE */
+               case 0x205f:    /* MEDIUM MATHEMATICAL SPACE */
+               case 0x3000:    /* IDEOGRAPHIC SPACE */
+               RRETURN(MATCH_NOMATCH);
+               }
+             break;
+             case OP_HSPACE:
+             switch(c)
+               {
+               default: RRETURN(MATCH_NOMATCH);
+               case 0x09:      /* HT */
+               case 0x20:      /* SPACE */
+               case 0xa0:      /* NBSP */
+               case 0x1680:    /* OGHAM SPACE MARK */
+               case 0x180e:    /* MONGOLIAN VOWEL SEPARATOR */
+               case 0x2000:    /* EN QUAD */
+               case 0x2001:    /* EM QUAD */
+               case 0x2002:    /* EN SPACE */
+               case 0x2003:    /* EM SPACE */
+               case 0x2004:    /* THREE-PER-EM SPACE */
+               case 0x2005:    /* FOUR-PER-EM SPACE */
+               case 0x2006:    /* SIX-PER-EM SPACE */
+               case 0x2007:    /* FIGURE SPACE */
+               case 0x2008:    /* PUNCTUATION SPACE */
+               case 0x2009:    /* THIN SPACE */
+               case 0x200A:    /* HAIR SPACE */
+               case 0x202f:    /* NARROW NO-BREAK SPACE */
+               case 0x205f:    /* MEDIUM MATHEMATICAL SPACE */
+               case 0x3000:    /* IDEOGRAPHIC SPACE */
+               break;
+               }
+             break;
+             case OP_NOT_VSPACE:
+             switch(c)
+               {
+               default: break;
+               case 0x0a:      /* LF */
+               case 0x0b:      /* VT */
+               case 0x0c:      /* FF */
+               case 0x0d:      /* CR */
+               case 0x85:      /* NEL */
+               case 0x2028:    /* LINE SEPARATOR */
+               case 0x2029:    /* PARAGRAPH SEPARATOR */
+               RRETURN(MATCH_NOMATCH);
+               }
+             break;
+             case OP_VSPACE:
+             switch(c)
+               {
+               default: RRETURN(MATCH_NOMATCH);
+               case 0x0a:      /* LF */