[pcre-dev] pcre_dfa_exec optimization

Kezdőlap
Üzenet törlése
Szerző: Ronen Hod
Dátum:  
Címzett: pcre-dev@exim.org
Tárgy: [pcre-dev] pcre_dfa_exec optimization
Hi,

I encountered serious performance issues so I had to do a quick fix for my needs. It is not thread-safe, it has assumptions regarding the number of states (which are sufficient for me), and I am not 100% sure that I understand all the implications (maybe it can be optimized further).
My application as a whole runs 4x faster now, and I assume that pcre_dfa_exec() is ~10x faster on my data.
Attached is the code with the changes. Enable/Disable them using: "#define CRESCENDO_CHECK_FOR_DUPLICATES".

Thanks, Ronen.

/*************************************************
*      Perl-Compatible Regular Expressions       *
*************************************************/


/* PCRE is a library of functions to support regular expressions whose syntax
and semantics are as close as possible to those of the Perl 5 language (but see
below for why this module is different).

                       Written by Philip Hazel
           Copyright (c) 1997-2009 University of Cambridge


-----------------------------------------------------------------------------
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

    * Redistributions of source code must retain the above copyright notice,
      this list of conditions and the following disclaimer.


    * Redistributions in binary form must reproduce the above copyright
      notice, this list of conditions and the following disclaimer in the
      documentation and/or other materials provided with the distribution.


    * Neither the name of the University of Cambridge nor the names of its
      contributors may be used to endorse or promote products derived from
      this software without specific prior written permission.


THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
POSSIBILITY OF SUCH DAMAGE.
-----------------------------------------------------------------------------
*/


/* This module contains the external function pcre_dfa_exec(), which is an
alternative matching function that uses a sort of DFA algorithm (not a true
FSM). This is NOT Perl- compatible, but it has advantages in certain
applications. */


#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"


/* For use to indent debugging output */

#define SP "                   "



/*************************************************
*      Code parameters and static tables         *
*************************************************/


/* These are offsets that are used to turn the OP_TYPESTAR and friends opcodes
into others, under special conditions. A gap of 20 between the blocks should be
enough. The resulting opcodes don't have to be less than 256 because they are
never stored, so we push them well clear of the normal opcodes. */

#define OP_PROP_EXTRA       300
#define OP_EXTUNI_EXTRA     320
#define OP_ANYNL_EXTRA      340
#define OP_HSPACE_EXTRA     360
#define OP_VSPACE_EXTRA     380



/* This table identifies those opcodes that are followed immediately by a
character that is to be tested in some way. This makes is possible to
centralize the loading of these characters. In the case of Type * etc, the
"character" is the opcode for \D, \d, \S, \s, \W, or \w, which will always be a
small value. ***NOTE*** If the start of this table is modified, the two tables
that follow must also be modified. */

static const uschar coptable[] = {
  0,                             /* End                                    */
  0, 0, 0, 0, 0,                 /* \A, \G, \K, \B, \b                     */
  0, 0, 0, 0, 0, 0,              /* \D, \d, \S, \s, \W, \w                 */
  0, 0, 0,                       /* Any, AllAny, Anybyte                   */
  0, 0, 0,                       /* NOTPROP, PROP, EXTUNI                  */
  0, 0, 0, 0, 0,                 /* \R, \H, \h, \V, \v                     */
  0, 0, 0, 0, 0,                 /* \Z, \z, Opt, ^, $                      */
  1,                             /* Char                                   */
  1,                             /* Charnc                                 */
  1,                             /* not                                    */
  /* Positive single-char repeats                                          */
  1, 1, 1, 1, 1, 1,              /* *, *?, +, +?, ?, ??                    */
  3, 3, 3,                       /* upto, minupto, exact                   */
  1, 1, 1, 3,                    /* *+, ++, ?+, upto+                      */
  /* Negative single-char repeats - only for chars < 256                   */
  1, 1, 1, 1, 1, 1,              /* NOT *, *?, +, +?, ?, ??                */
  3, 3, 3,                       /* NOT upto, minupto, exact               */
  1, 1, 1, 3,                    /* NOT *+, ++, ?+, updo+                  */
  /* Positive type repeats                                                 */
  1, 1, 1, 1, 1, 1,              /* Type *, *?, +, +?, ?, ??               */
  3, 3, 3,                       /* Type upto, minupto, exact              */
  1, 1, 1, 3,                    /* Type *+, ++, ?+, upto+                 */
  /* Character class & ref repeats                                         */
  0, 0, 0, 0, 0, 0,              /* *, *?, +, +?, ?, ??                    */
  0, 0,                          /* CRRANGE, CRMINRANGE                    */
  0,                             /* CLASS                                  */
  0,                             /* NCLASS                                 */
  0,                             /* XCLASS - variable length               */
  0,                             /* REF                                    */
  0,                             /* RECURSE                                */
  0,                             /* CALLOUT                                */
  0,                             /* Alt                                    */
  0,                             /* Ket                                    */
  0,                             /* KetRmax                                */
  0,                             /* KetRmin                                */
  0,                             /* Assert                                 */
  0,                             /* Assert not                             */
  0,                             /* Assert behind                          */
  0,                             /* Assert behind not                      */
  0,                             /* Reverse                                */
  0, 0, 0, 0,                    /* ONCE, BRA, CBRA, COND                  */
  0, 0, 0,                       /* SBRA, SCBRA, SCOND                     */
  0,                             /* CREF                                   */
  0,                             /* RREF                                   */
  0,                             /* DEF                                    */
  0, 0,                          /* BRAZERO, BRAMINZERO                    */
  0, 0, 0, 0,                    /* PRUNE, SKIP, THEN, COMMIT              */
  0, 0, 0                        /* FAIL, ACCEPT, SKIPZERO                 */
};


/* These 2 tables allow for compact code for testing for \D, \d, \S, \s, \W,
and \w */

static const uschar toptable1[] = {
  0, 0, 0, 0, 0, 0,
  ctype_digit, ctype_digit,
  ctype_space, ctype_space,
  ctype_word,  ctype_word,
  0, 0                            /* OP_ANY, OP_ALLANY */
};


static const uschar toptable2[] = {
  0, 0, 0, 0, 0, 0,
  ctype_digit, 0,
  ctype_space, 0,
  ctype_word,  0,
  1, 1                            /* OP_ANY, OP_ALLANY */
};



/* Structure for holding data about a particular state, which is in effect the
current data for an active path through the match tree. It must consist
entirely of ints because the working vector we are passed, and which we put
these structures in, is a vector of ints. */

typedef struct stateblock {
  int offset;                     /* Offset to opcode */
  int count;                      /* Count for repeats */
  int ims;                        /* ims flag bits */
  int data;                       /* Some use extra data */
} stateblock;


#define INTS_PER_STATEBLOCK (sizeof(stateblock)/sizeof(int))


#ifdef DEBUG
/*************************************************
*             Print character string             *
*************************************************/


/* Character string printing function for debugging.

Arguments:
  p            points to string
  length       number of bytes
  f            where to print


Returns:       nothing
*/


static void
pchars(unsigned char *p, int length, FILE *f)
{
int c;
while (length-- > 0)
  {
  if (isprint(c = *(p++)))
    fprintf(f, "%c", c);
  else
    fprintf(f, "\\x%02x", c);
  }
}
#endif




/*************************************************
*    Execute a Regular Expression - DFA engine   *
*************************************************/


/* This internal function applies a compiled pattern to a subject string,
starting at a given point, using a DFA engine. This function is called from the
external one, possibly multiple times if the pattern is not anchored. The
function calls itself recursively for some kinds of subpattern.

Arguments:
  md                the match_data block with fixed information
  this_start_code   the opening bracket of this subexpression's code
  current_subject   where we currently are in the subject string
  start_offset      start offset in the subject string
  offsets           vector to contain the matching string offsets
  offsetcount       size of same
  workspace         vector of workspace
  wscount           size of same
  ims               the current ims flags
  rlevel            function call recursion level
  recursing         regex recursive call level


Returns:            > 0 => number of match offset pairs placed in offsets
                    = 0 => offsets overflowed; longest matches are present
                     -1 => failed to match
                   < -1 => some kind of unexpected problem


The following macros are used for adding states to the two state vectors (one
for the current character, one for the following character). */

#define CRESCENDO_CHECK_FOR_DUPLICATES
#ifndef CRESCENDO_CHECK_FOR_DUPLICATES

#define ADD_ACTIVE(x,y) \
  if (active_count++ < wscount) \
    { \
    next_active_state->offset = (x); \
    next_active_state->count  = (y); \
    next_active_state->ims    = ims; \
    next_active_state++; \
    DPRINTF(("%.*sADD_ACTIVE(%d,%d)\n", rlevel*2-2, SP, (x), (y))); \
    } \
  else return PCRE_ERROR_DFA_WSSIZE


#define ADD_ACTIVE_DATA(x,y,z) \
  if (active_count++ < wscount) \
    { \
    next_active_state->offset = (x); \
    next_active_state->count  = (y); \
    next_active_state->ims    = ims; \
    next_active_state->data   = (z); \
    next_active_state++; \
    DPRINTF(("%.*sADD_ACTIVE_DATA(%d,%d,%d)\n", rlevel*2-2, SP, (x), (y), (z))); \
    } \
  else return PCRE_ERROR_DFA_WSSIZE


#define ADD_NEW(x,y) \
  if (new_count++ < wscount) \
    { \
    next_new_state->offset = (x); \
    next_new_state->count  = (y); \
    next_new_state->ims    = ims; \
    next_new_state++; \
    DPRINTF(("%.*sADD_NEW(%d,%d)\n", rlevel*2-2, SP, (x), (y))); \
    } \
  else return PCRE_ERROR_DFA_WSSIZE


#define ADD_NEW_DATA(x,y,z) \
  if (new_count++ < wscount) \
    { \
    next_new_state->offset = (x); \
    next_new_state->count  = (y); \
    next_new_state->ims    = ims; \
    next_new_state->data   = (z); \
    next_new_state++; \
    DPRINTF(("%.*sADD_NEW_DATA(%d,%d,%d)\n", rlevel*2-2, SP, (x), (y), (z))); \
    } \
  else return PCRE_ERROR_DFA_WSSIZE


#else    // #ifndef CRESCENDO_CHECK_FOR_DUPLICATES


#define ADD_ACTIVE(x,y) \
  if (active_count < wscount) \
    { \
        if (nextActiveStateOffsetArray[x] == nextActiveStateIterationNo && \
                    doesOffsetCountExist(active_states, active_count, (x), (y))) { \
            DPRINTF(("%.*sDuplicate state: skipped\n", rlevel*2-2, SP)); \
        } \
        else { \
            active_count++; \
            next_active_state->offset = (x); \
            next_active_state->count  = (y); \
            next_active_state->ims    = ims; \
            next_active_state++; \
            nextActiveStateOffsetArray[x] = nextActiveStateIterationNo; \
            DPRINTF(("%.*sADD_ACTIVE(%d,%d)\n", rlevel*2-2, SP, (x), (y))); \
        } \
    } \
  else return PCRE_ERROR_DFA_WSSIZE


#define ADD_ACTIVE_DATA(x,y,z) \
  if (active_count++ < wscount) \
    { \
        if (nextActiveStateOffsetArray[x] == nextActiveStateIterationNo && \
                    doesOffsetCountExist(active_states, active_count, (x), (y))) { \
            DPRINTF(("%.*sDuplicate state: skipped\n", rlevel*2-2, SP)); \
        } \
        else { \
            active_count++; \
            next_active_state->offset = (x); \
            next_active_state->count  = (y); \
            next_active_state->ims    = ims; \
            next_active_state->data   = (z); \
            next_active_state++; \
            nextActiveStateOffsetArray[x] = nextActiveStateIterationNo; \
            DPRINTF(("%.*sADD_ACTIVE_DATA(%d,%d,%d)\n", rlevel*2-2, SP, (x), (y), (z))); \
        } \
    } \
  else return PCRE_ERROR_DFA_WSSIZE


#define ADD_NEW(x,y) \
  if (new_count < wscount) \
    { \
        if (nextNewStateOffsetArray[x] == nextActiveStateIterationNo + 1 && \
                    doesOffsetCountExist(new_states, new_count, (x), (y))) { \
            DPRINTF(("%.*sDuplicate state: skipped\n", rlevel*2-2, SP)); \
        } \
        else { \
            new_count++; \
            next_new_state->offset = (x); \
            next_new_state->count  = (y); \
            next_new_state->ims    = ims; \
            next_new_state++; \
            nextNewStateOffsetArray[x] = nextActiveStateIterationNo + 1; \
            DPRINTF(("%.*sADD_NEW(%d,%d)\n", rlevel*2-2, SP, (x), (y))); \
        } \
    } \
  else return PCRE_ERROR_DFA_WSSIZE


#define ADD_NEW_DATA(x,y,z) \
  if (new_count < wscount) \
    { \
        if (nextNewStateOffsetArray[x] == nextActiveStateIterationNo + 1 && \
                    doesOffsetCountExist(new_states, new_count, (x), (y))) { \
            DPRINTF(("%.*sDuplicate state: skipped\n", rlevel*2-2, SP)); \
        } \
        else { \
            new_count++; \
            next_new_state->offset = (x); \
            next_new_state->count  = (y); \
            next_new_state->ims    = ims; \
            next_new_state->data   = (z); \
            next_new_state++; \
            nextNewStateOffsetArray[x] = nextActiveStateIterationNo + 1; \
            DPRINTF(("%.*sADD_NEW_DATA(%d,%d,%d)\n", rlevel*2-2, SP, (x), (y), (z))); \
        } \
    } \
  else return PCRE_ERROR_DFA_WSSIZE


/******************************************************************************/

int doesOffsetCountExist(stateblock *states, int size, int offset, int count)
{
    int i;
    for (i = 0; i < size; i++) {
        if (states[i].offset == offset && states[i].count == count) {
            return 1;
        }
    }
    return 0;
}


/******************************************************************************/
#endif    // #ifndef CRESCENDO_CHECK_FOR_DUPLICATES


/* And now, here is the code */

static int
internal_dfa_exec(
dfa_match_data *md,
const uschar *this_start_code,
const uschar *current_subject,
int start_offset,
int *offsets,
int offsetcount,
int *workspace,
int wscount,
int ims,
int rlevel,
int recursing)
{

#ifdef CRESCENDO_CHECK_FOR_DUPLICATES
static int nextActiveStateIterationNo;
#define OFFSET_ARRAY_N_ELEM 30000
static int *nextActiveStateOffsetArray;
static int *nextNewStateOffsetArray;
#endif    // #ifdef CRESCENDO_CHECK_FOR_DUPLICATES


stateblock *active_states, *new_states, *temp_states;
stateblock *next_active_state, *next_new_state;

const uschar *ctypes, *lcc, *fcc;
const uschar *ptr;
const uschar *end_code, *first_op;

int active_count, new_count, match_count;

/* Some fields in the md block are frequently referenced, so we load them into
independent variables in the hope that this will perform better. */

const uschar *start_subject = md->start_subject;
const uschar *end_subject = md->end_subject;
const uschar *start_code = md->start_code;

#ifdef SUPPORT_UTF8
BOOL utf8 = (md->poptions & PCRE_UTF8) != 0;
#else
BOOL utf8 = FALSE;
#endif

#ifdef CRESCENDO_CHECK_FOR_DUPLICATES
if (nextActiveStateOffsetArray == NULL) {
    nextActiveStateOffsetArray = (int *)malloc(sizeof(*nextActiveStateOffsetArray) * OFFSET_ARRAY_N_ELEM);
    nextNewStateOffsetArray = (int *)malloc(sizeof(*nextNewStateOffsetArray) * OFFSET_ARRAY_N_ELEM);
}
if (nextActiveStateIterationNo == 0 || nextActiveStateIterationNo > 1000000000) {
    memset(nextActiveStateOffsetArray, 0, sizeof(*nextActiveStateOffsetArray) * OFFSET_ARRAY_N_ELEM);
    memset(nextNewStateOffsetArray, 0, sizeof(*nextNewStateOffsetArray) * OFFSET_ARRAY_N_ELEM);
    nextActiveStateIterationNo = 0;
}
#endif    // #ifdef CRESCENDO_CHECK_FOR_DUPLICATES


rlevel++;
offsetcount &= (-2);

wscount -= 2;
wscount = (wscount - (wscount % (INTS_PER_STATEBLOCK * 2))) /
          (2 * INTS_PER_STATEBLOCK);


DPRINTF(("\n%.*s---------------------\n"
"%.*sCall to internal_dfa_exec f=%d r=%d\n",
rlevel*2-2, SP, rlevel*2-2, SP, rlevel, recursing));

ctypes = md->tables + ctypes_offset;
lcc = md->tables + lcc_offset;
fcc = md->tables + fcc_offset;

match_count = PCRE_ERROR_NOMATCH; /* A negative number */

active_states = (stateblock *)(workspace + 2);
next_new_state = new_states = active_states + wscount;
new_count = 0;

first_op = this_start_code + 1 + LINK_SIZE +
((*this_start_code == OP_CBRA || *this_start_code == OP_SCBRA)? 2:0);

/* The first thing in any (sub) pattern is a bracket of some sort. Push all
the alternative states onto the list, and find out where the end is. This
makes is possible to use this function recursively, when we want to stop at a
matching internal ket rather than at the end.

If the first opcode in the first alternative is OP_REVERSE, we are dealing with
a backward assertion. In that case, we have to find out the maximum amount to
move back, and set up each alternative appropriately. */

if (*first_op == OP_REVERSE)
{
int max_back = 0;
int gone_back;

  end_code = this_start_code;
  do
    {
    int back = GET(end_code, 2+LINK_SIZE);
    if (back > max_back) max_back = back;
    end_code += GET(end_code, 1);
    }
  while (*end_code == OP_ALT);


/* If we can't go back the amount required for the longest lookbehind
pattern, go back as far as we can; some alternatives may still be viable. */

#ifdef SUPPORT_UTF8
/* In character mode we have to step back character by character */

  if (utf8)
    {
    for (gone_back = 0; gone_back < max_back; gone_back++)
      {
      if (current_subject <= start_subject) break;
      current_subject--;
      while (current_subject > start_subject &&
             (*current_subject & 0xc0) == 0x80)
        current_subject--;
      }
    }
  else
#endif


/* In byte-mode we can do this quickly. */

    {
    gone_back = (current_subject - max_back < start_subject)?
      current_subject - start_subject : max_back;
    current_subject -= gone_back;
    }


/* Now we can process the individual branches. */

  end_code = this_start_code;
  do
    {
    int back = GET(end_code, 2+LINK_SIZE);
    if (back <= gone_back)
      {
      int bstate = end_code - start_code + 2 + 2*LINK_SIZE;
      ADD_NEW_DATA(-bstate, 0, gone_back - back);
      }
    end_code += GET(end_code, 1);
    }
  while (*end_code == OP_ALT);
 }


/* This is the code for a "normal" subpattern (not a backward assertion). The
start of a whole pattern is always one of these. If we are at the top level,
we may be asked to restart matching from the same point that we reached for a
previous partial match. We still have to scan through the top-level branches to
find the end state. */

else
{
end_code = this_start_code;

/* Restarting */

  if (rlevel == 1 && (md->moptions & PCRE_DFA_RESTART) != 0)
    {
    do { end_code += GET(end_code, 1); } while (*end_code == OP_ALT);
    new_count = workspace[1];
    if (!workspace[0])
      memcpy(new_states, active_states, new_count * sizeof(stateblock));
    }


/* Not restarting */

  else
    {
    int length = 1 + LINK_SIZE +
      ((*this_start_code == OP_CBRA || *this_start_code == OP_SCBRA)? 2:0);
    do
      {
      ADD_NEW(end_code - start_code + length, 0);
      end_code += GET(end_code, 1);
      length = 1 + LINK_SIZE;
      }
    while (*end_code == OP_ALT);
    }
  }


workspace[0] = 0;    /* Bit indicating which vector is current */


DPRINTF(("%.*sEnd state = %d\n", rlevel*2-2, SP, end_code - start_code));

/* Loop for scanning the subject */

ptr = current_subject;
for (;;)
{
int i, j;
int clen, dlen;
unsigned int c, d;

/* Make the new state list into the active state list and empty the
new state list. */

#ifdef CRESCENDO_CHECK_FOR_DUPLICATES
  int *tmpStateOffsetArray = nextActiveStateOffsetArray;
  nextActiveStateOffsetArray = nextNewStateOffsetArray;
  nextNewStateOffsetArray = tmpStateOffsetArray;
  nextActiveStateIterationNo++;
#endif    // #ifdef CRESCENDO_CHECK_FOR_DUPLICATES


temp_states = active_states;
active_states = new_states;
new_states = temp_states;
active_count = new_count;
new_count = 0;

  workspace[0] ^= 1;              /* Remember for the restarting feature */
  workspace[1] = active_count;


#ifdef DEBUG
printf("%.*sNext character: rest of subject = \"", rlevel*2-2, SP);
pchars((uschar *)ptr, strlen((char *)ptr), stdout);
printf("\"\n");

  printf("%.*sActive states: ", rlevel*2-2, SP);
  for (i = 0; i < active_count; i++)
    printf("%d/%d ", active_states[i].offset, active_states[i].count);
  printf("\n");
#endif


/* Set the pointers for adding new states */

next_active_state = active_states + active_count;
next_new_state = new_states;

/* Load the current character from the subject outside the loop, as many
different states may want to look at it, and we assume that at least one
will. */

  if (ptr < end_subject)
    {
    clen = 1;        /* Number of bytes in the character */
#ifdef SUPPORT_UTF8
    if (utf8) { GETCHARLEN(c, ptr, clen); } else
#endif  /* SUPPORT_UTF8 */
    c = *ptr;
    }
  else
    {
    clen = 0;        /* This indicates the end of the subject */
    c = NOTACHAR;    /* This value should never actually be used */
    }


/* Scan up the active states and act on each one. The result of an action
may be to add more states to the currently active list (e.g. on hitting a
parenthesis) or it may be to put states on the new list, for considering
when we move the character pointer on. */

  for (i = 0; i < active_count; i++)
    {
    stateblock *current_state = active_states + i;
    const uschar *code;
    int state_offset = current_state->offset;
    int count, codevalue, rrc;


#ifdef DEBUG
    printf ("%.*sProcessing state %d c=", rlevel*2-2, SP, state_offset);
    if (clen == 0) printf("EOL\n");
      else if (c > 32 && c < 127) printf("'%c'\n", c);
        else printf("0x%02x\n", c);
#endif


    /* This variable is referred to implicity in the ADD_xxx macros. */


    ims = current_state->ims;


    /* A negative offset is a special case meaning "hold off going to this
    (negated) state until the number of characters in the data field have
    been skipped". */


    if (state_offset < 0)
      {
      if (current_state->data > 0)
        {
        DPRINTF(("%.*sSkipping this character\n", rlevel*2-2, SP));
        ADD_NEW_DATA(state_offset, current_state->count,
          current_state->data - 1);
        continue;
        }
      else
        {
        current_state->offset = state_offset = -state_offset;
        }
      }


#ifndef CRESCENDO_CHECK_FOR_DUPLICATES
    /* Check for a duplicate state with the same count, and skip if found. */
    for (j = 0; j < i; j++)
      {


      if (active_states[j].offset == state_offset &&
          active_states[j].count == current_state->count)
        {
        DPRINTF(("%.*sDuplicate state: skipped\n", rlevel*2-2, SP));
        goto NEXT_ACTIVE_STATE;
        }
      }
#endif    // #ifndef CRESCENDO_CHECK_FOR_DUPLICATES


    /* The state offset is the offset to the opcode */


    code = start_code + state_offset;
    codevalue = *code;


    /* If this opcode is followed by an inline character, load it. It is
    tempting to test for the presence of a subject character here, but that
    is wrong, because sometimes zero repetitions of the subject are
    permitted.


    We also use this mechanism for opcodes such as OP_TYPEPLUS that take an
    argument that is not a data character - but is always one byte long. We
    have to take special action to deal with  \P, \p, \H, \h, \V, \v and \X in
    this case. To keep the other cases fast, convert these ones to new opcodes.
    */


    if (coptable[codevalue] > 0)
      {
      dlen = 1;
#ifdef SUPPORT_UTF8
      if (utf8) { GETCHARLEN(d, (code + coptable[codevalue]), dlen); } else
#endif  /* SUPPORT_UTF8 */
      d = code[coptable[codevalue]];
      if (codevalue >= OP_TYPESTAR)
        {
        switch(d)
          {
          case OP_ANYBYTE: return PCRE_ERROR_DFA_UITEM;
          case OP_NOTPROP:
          case OP_PROP: codevalue += OP_PROP_EXTRA; break;
          case OP_ANYNL: codevalue += OP_ANYNL_EXTRA; break;
          case OP_EXTUNI: codevalue += OP_EXTUNI_EXTRA; break;
          case OP_NOT_HSPACE:
          case OP_HSPACE: codevalue += OP_HSPACE_EXTRA; break;
          case OP_NOT_VSPACE:
          case OP_VSPACE: codevalue += OP_VSPACE_EXTRA; break;
          default: break;
          }
        }
      }
    else
      {
      dlen = 0;         /* Not strictly necessary, but compilers moan */
      d = NOTACHAR;     /* if these variables are not set. */
      }



    /* Now process the individual opcodes */


    switch (codevalue)
      {


/* ========================================================================== */
      /* Reached a closing bracket. If not at the end of the pattern, carry
      on with the next opcode. Otherwise, unless we have an empty string and
      PCRE_NOTEMPTY is set, save the match data, shifting up all previous
      matches so we always have the longest first. */


      case OP_KET:
      case OP_KETRMIN:
      case OP_KETRMAX:
      if (code != end_code)
        {
        ADD_ACTIVE(state_offset + 1 + LINK_SIZE, 0);
        if (codevalue != OP_KET)
          {
          ADD_ACTIVE(state_offset - GET(code, 1), 0);
          }
        }
      else if (ptr > current_subject || (md->moptions & PCRE_NOTEMPTY) == 0)
        {
        if (match_count < 0) match_count = (offsetcount >= 2)? 1 : 0;
          else if (match_count > 0 && ++match_count * 2 >= offsetcount)
            match_count = 0;
        count = ((match_count == 0)? offsetcount : match_count * 2) - 2;
        if (count > 0) memmove(offsets + 2, offsets, count * sizeof(int));
        if (offsetcount >= 2)
          {
          offsets[0] = current_subject - start_subject;
          offsets[1] = ptr - start_subject;
          DPRINTF(("%.*sSet matched string = \"%.*s\"\n", rlevel*2-2, SP,
            offsets[1] - offsets[0], current_subject));
          }
        if ((md->moptions & PCRE_DFA_SHORTEST) != 0)
          {
          DPRINTF(("%.*sEnd of internal_dfa_exec %d: returning %d\n"
            "%.*s---------------------\n\n", rlevel*2-2, SP, rlevel,
            match_count, rlevel*2-2, SP));
          return match_count;
          }
        }
      break;


/* ========================================================================== */
      /* These opcodes add to the current list of states without looking
      at the current character. */


      /*-----------------------------------------------------------------*/
      case OP_ALT:
      do { code += GET(code, 1); } while (*code == OP_ALT);
      ADD_ACTIVE(code - start_code, 0);
      break;


      /*-----------------------------------------------------------------*/
      case OP_BRA:
      case OP_SBRA:
      do
        {
        ADD_ACTIVE(code - start_code + 1 + LINK_SIZE, 0);
        code += GET(code, 1);
        }
      while (*code == OP_ALT);
      break;


      /*-----------------------------------------------------------------*/
      case OP_CBRA:
      case OP_SCBRA:
      ADD_ACTIVE(code - start_code + 3 + LINK_SIZE,  0);
      code += GET(code, 1);
      while (*code == OP_ALT)
        {
        ADD_ACTIVE(code - start_code + 1 + LINK_SIZE,  0);
        code += GET(code, 1);
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_BRAZERO:
      case OP_BRAMINZERO:
      ADD_ACTIVE(state_offset + 1, 0);
      code += 1 + GET(code, 2);
      while (*code == OP_ALT) code += GET(code, 1);
      ADD_ACTIVE(code - start_code + 1 + LINK_SIZE, 0);
      break;


      /*-----------------------------------------------------------------*/
      case OP_SKIPZERO:
      code += 1 + GET(code, 2);
      while (*code == OP_ALT) code += GET(code, 1);
      ADD_ACTIVE(code - start_code + 1 + LINK_SIZE, 0);
      break;


      /*-----------------------------------------------------------------*/
      case OP_CIRC:
      if ((ptr == start_subject && (md->moptions & PCRE_NOTBOL) == 0) ||
          ((ims & PCRE_MULTILINE) != 0 &&
            ptr != end_subject &&
            WAS_NEWLINE(ptr)))
        { ADD_ACTIVE(state_offset + 1, 0); }
      break;


      /*-----------------------------------------------------------------*/
      case OP_EOD:
      if (ptr >= end_subject) { ADD_ACTIVE(state_offset + 1, 0); }
      break;


      /*-----------------------------------------------------------------*/
      case OP_OPT:
      ims = code[1];
      ADD_ACTIVE(state_offset + 2, 0);
      break;


      /*-----------------------------------------------------------------*/
      case OP_SOD:
      if (ptr == start_subject) { ADD_ACTIVE(state_offset + 1, 0); }
      break;


      /*-----------------------------------------------------------------*/
      case OP_SOM:
      if (ptr == start_subject + start_offset) { ADD_ACTIVE(state_offset + 1, 0); }
      break;



/* ========================================================================== */
      /* These opcodes inspect the next subject character, and sometimes
      the previous one as well, but do not have an argument. The variable
      clen contains the length of the current character and is zero if we are
      at the end of the subject. */


      /*-----------------------------------------------------------------*/
      case OP_ANY:
      if (clen > 0 && !IS_NEWLINE(ptr))
        { ADD_NEW(state_offset + 1, 0); }
      break;


      /*-----------------------------------------------------------------*/
      case OP_ALLANY:
      if (clen > 0)
        { ADD_NEW(state_offset + 1, 0); }
      break;


      /*-----------------------------------------------------------------*/
      case OP_EODN:
      if (clen == 0 || (IS_NEWLINE(ptr) && ptr == end_subject - md->nllen))
        { ADD_ACTIVE(state_offset + 1, 0); }
      break;


      /*-----------------------------------------------------------------*/
      case OP_DOLL:
      if ((md->moptions & PCRE_NOTEOL) == 0)
        {
        if (clen == 0 ||
            ((md->poptions & PCRE_DOLLAR_ENDONLY) == 0 && IS_NEWLINE(ptr) &&
               ((ims & PCRE_MULTILINE) != 0 || ptr == end_subject - md->nllen)
            ))
          { ADD_ACTIVE(state_offset + 1, 0); }
        }
      else if ((ims & PCRE_MULTILINE) != 0 && IS_NEWLINE(ptr))
        { ADD_ACTIVE(state_offset + 1, 0); }
      break;


      /*-----------------------------------------------------------------*/


      case OP_DIGIT:
      case OP_WHITESPACE:
      case OP_WORDCHAR:
      if (clen > 0 && c < 256 &&
            ((ctypes[c] & toptable1[codevalue]) ^ toptable2[codevalue]) != 0)
        { ADD_NEW(state_offset + 1, 0); }
      break;


      /*-----------------------------------------------------------------*/
      case OP_NOT_DIGIT:
      case OP_NOT_WHITESPACE:
      case OP_NOT_WORDCHAR:
      if (clen > 0 && (c >= 256 ||
            ((ctypes[c] & toptable1[codevalue]) ^ toptable2[codevalue]) != 0))
        { ADD_NEW(state_offset + 1, 0); }
      break;


      /*-----------------------------------------------------------------*/
      case OP_WORD_BOUNDARY:
      case OP_NOT_WORD_BOUNDARY:
        {
        int left_word, right_word;


        if (ptr > start_subject)
          {
          const uschar *temp = ptr - 1;
#ifdef SUPPORT_UTF8
          if (utf8) BACKCHAR(temp);
#endif
          GETCHARTEST(d, temp);
          left_word = d < 256 && (ctypes[d] & ctype_word) != 0;
          }
        else left_word = 0;


        if (clen > 0) right_word = c < 256 && (ctypes[c] & ctype_word) != 0;
          else right_word = 0;


        if ((left_word == right_word) == (codevalue == OP_NOT_WORD_BOUNDARY))
          { ADD_ACTIVE(state_offset + 1, 0); }
        }
      break;



      /*-----------------------------------------------------------------*/
      /* 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.
      */


#ifdef SUPPORT_UCP
      case OP_PROP:
      case OP_NOTPROP:
      if (clen > 0)
        {
        BOOL OK;
        const ucd_record * prop = GET_UCD(c);
        switch(code[1])
          {
          case PT_ANY:
          OK = TRUE;
          break;


          case PT_LAMP:
          OK = prop->chartype == ucp_Lu || prop->chartype == ucp_Ll || prop->chartype == ucp_Lt;
          break;


          case PT_GC:
          OK = _pcre_ucp_gentype[prop->chartype] == code[2];
          break;


          case PT_PC:
          OK = prop->chartype == code[2];
          break;


          case PT_SC:
          OK = prop->script == code[2];
          break;


          /* Should never occur, but keep compilers from grumbling. */


          default:
          OK = codevalue != OP_PROP;
          break;
          }


        if (OK == (codevalue == OP_PROP)) { ADD_NEW(state_offset + 3, 0); }
        }
      break;
#endif




/* ========================================================================== */
      /* These opcodes likewise inspect the subject character, but have an
      argument that is not a data character. It is one of these opcodes:
      OP_ANY, OP_ALLANY, OP_DIGIT, OP_NOT_DIGIT, OP_WHITESPACE, OP_NOT_SPACE,
      OP_WORDCHAR, OP_NOT_WORDCHAR. The value is loaded into d. */


      case OP_TYPEPLUS:
      case OP_TYPEMINPLUS:
      case OP_TYPEPOSPLUS:
      count = current_state->count;  /* Already matched */
      if (count > 0) { ADD_ACTIVE(state_offset + 2, 0); }
      if (clen > 0)
        {
        if ((c >= 256 && d != OP_DIGIT && d != OP_WHITESPACE && d != OP_WORDCHAR) ||
            (c < 256 &&
              (d != OP_ANY || !IS_NEWLINE(ptr)) &&
              ((ctypes[c] & toptable1[d]) ^ toptable2[d]) != 0))
          {
          if (count > 0 && codevalue == OP_TYPEPOSPLUS)
            {
            active_count--;            /* Remove non-match possibility */
            next_active_state--;
            }
          count++;
          ADD_NEW(state_offset, count);
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_TYPEQUERY:
      case OP_TYPEMINQUERY:
      case OP_TYPEPOSQUERY:
      ADD_ACTIVE(state_offset + 2, 0);
      if (clen > 0)
        {
        if ((c >= 256 && d != OP_DIGIT && d != OP_WHITESPACE && d != OP_WORDCHAR) ||
            (c < 256 &&
              (d != OP_ANY || !IS_NEWLINE(ptr)) &&
              ((ctypes[c] & toptable1[d]) ^ toptable2[d]) != 0))
          {
          if (codevalue == OP_TYPEPOSQUERY)
            {
            active_count--;            /* Remove non-match possibility */
            next_active_state--;
            }
          ADD_NEW(state_offset + 2, 0);
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_TYPESTAR:
      case OP_TYPEMINSTAR:
      case OP_TYPEPOSSTAR:
      ADD_ACTIVE(state_offset + 2, 0);
      if (clen > 0)
        {
        if ((c >= 256 && d != OP_DIGIT && d != OP_WHITESPACE && d != OP_WORDCHAR) ||
            (c < 256 &&
              (d != OP_ANY || !IS_NEWLINE(ptr)) &&
              ((ctypes[c] & toptable1[d]) ^ toptable2[d]) != 0))
          {
          if (codevalue == OP_TYPEPOSSTAR)
            {
            active_count--;            /* Remove non-match possibility */
            next_active_state--;
            }
          ADD_NEW(state_offset, 0);
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_TYPEEXACT:
      count = current_state->count;  /* Number already matched */
      if (clen > 0)
        {
        if ((c >= 256 && d != OP_DIGIT && d != OP_WHITESPACE && d != OP_WORDCHAR) ||
            (c < 256 &&
              (d != OP_ANY || !IS_NEWLINE(ptr)) &&
              ((ctypes[c] & toptable1[d]) ^ toptable2[d]) != 0))
          {
          if (++count >= GET2(code, 1))
            { ADD_NEW(state_offset + 4, 0); }
          else
            { ADD_NEW(state_offset, count); }
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_TYPEUPTO:
      case OP_TYPEMINUPTO:
      case OP_TYPEPOSUPTO:
      ADD_ACTIVE(state_offset + 4, 0);
      count = current_state->count;  /* Number already matched */
      if (clen > 0)
        {
        if ((c >= 256 && d != OP_DIGIT && d != OP_WHITESPACE && d != OP_WORDCHAR) ||
            (c < 256 &&
              (d != OP_ANY || !IS_NEWLINE(ptr)) &&
              ((ctypes[c] & toptable1[d]) ^ toptable2[d]) != 0))
          {
          if (codevalue == OP_TYPEPOSUPTO)
            {
            active_count--;           /* Remove non-match possibility */
            next_active_state--;
            }
          if (++count >= GET2(code, 1))
            { ADD_NEW(state_offset + 4, 0); }
          else
            { ADD_NEW(state_offset, count); }
          }
        }
      break;


/* ========================================================================== */
      /* These are virtual opcodes that are used when something like
      OP_TYPEPLUS has OP_PROP, OP_NOTPROP, OP_ANYNL, or OP_EXTUNI as its
      argument. It keeps the code above fast for the other cases. The argument
      is in the d variable. */


#ifdef SUPPORT_UCP
      case OP_PROP_EXTRA + OP_TYPEPLUS:
      case OP_PROP_EXTRA + OP_TYPEMINPLUS:
      case OP_PROP_EXTRA + OP_TYPEPOSPLUS:
      count = current_state->count;           /* Already matched */
      if (count > 0) { ADD_ACTIVE(state_offset + 4, 0); }
      if (clen > 0)
        {
        BOOL OK;
        const ucd_record * prop = GET_UCD(c);
        switch(code[2])
          {
          case PT_ANY:
          OK = TRUE;
          break;


          case PT_LAMP:
          OK = prop->chartype == ucp_Lu || prop->chartype == ucp_Ll || prop->chartype == ucp_Lt;
          break;


          case PT_GC:
          OK = _pcre_ucp_gentype[prop->chartype] == code[3];
          break;


          case PT_PC:
          OK = prop->chartype == code[3];
          break;


          case PT_SC:
          OK = prop->script == code[3];
          break;


          /* Should never occur, but keep compilers from grumbling. */


          default:
          OK = codevalue != OP_PROP;
          break;
          }


        if (OK == (d == OP_PROP))
          {
          if (count > 0 && codevalue == OP_PROP_EXTRA + OP_TYPEPOSPLUS)
            {
            active_count--;           /* Remove non-match possibility */
            next_active_state--;
            }
          count++;
          ADD_NEW(state_offset, count);
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_EXTUNI_EXTRA + OP_TYPEPLUS:
      case OP_EXTUNI_EXTRA + OP_TYPEMINPLUS:
      case OP_EXTUNI_EXTRA + OP_TYPEPOSPLUS:
      count = current_state->count;  /* Already matched */
      if (count > 0) { ADD_ACTIVE(state_offset + 2, 0); }
      if (clen > 0 && UCD_CATEGORY(c) != ucp_M)
        {
        const uschar *nptr = ptr + clen;
        int ncount = 0;
        if (count > 0 && codevalue == OP_EXTUNI_EXTRA + OP_TYPEPOSPLUS)
          {
          active_count--;           /* Remove non-match possibility */
          next_active_state--;
          }
        while (nptr < end_subject)
          {
          int nd;
          int ndlen = 1;
          GETCHARLEN(nd, nptr, ndlen);
          if (UCD_CATEGORY(nd) != ucp_M) break;
          ncount++;
          nptr += ndlen;
          }
        count++;
        ADD_NEW_DATA(-state_offset, count, ncount);
        }
      break;
#endif


      /*-----------------------------------------------------------------*/
      case OP_ANYNL_EXTRA + OP_TYPEPLUS:
      case OP_ANYNL_EXTRA + OP_TYPEMINPLUS:
      case OP_ANYNL_EXTRA + OP_TYPEPOSPLUS:
      count = current_state->count;  /* Already matched */
      if (count > 0) { ADD_ACTIVE(state_offset + 2, 0); }
      if (clen > 0)
        {
        int ncount = 0;
        switch (c)
          {
          case 0x000b:
          case 0x000c:
          case 0x0085:
          case 0x2028:
          case 0x2029:
          if ((md->moptions & PCRE_BSR_ANYCRLF) != 0) break;
          goto ANYNL01;


          case 0x000d:
          if (ptr + 1 < end_subject && ptr[1] == 0x0a) ncount = 1;
          /* Fall through */


          ANYNL01:
          case 0x000a:
          if (count > 0 && codevalue == OP_ANYNL_EXTRA + OP_TYPEPOSPLUS)
            {
            active_count--;           /* Remove non-match possibility */
            next_active_state--;
            }
          count++;
          ADD_NEW_DATA(-state_offset, count, ncount);
          break;


          default:
          break;
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_VSPACE_EXTRA + OP_TYPEPLUS:
      case OP_VSPACE_EXTRA + OP_TYPEMINPLUS:
      case OP_VSPACE_EXTRA + OP_TYPEPOSPLUS:
      count = current_state->count;  /* Already matched */
      if (count > 0) { ADD_ACTIVE(state_offset + 2, 0); }
      if (clen > 0)
        {
        BOOL OK;
        switch (c)
          {
          case 0x000a:
          case 0x000b:
          case 0x000c:
          case 0x000d:
          case 0x0085:
          case 0x2028:
          case 0x2029:
          OK = TRUE;
          break;


          default:
          OK = FALSE;
          break;
          }


        if (OK == (d == OP_VSPACE))
          {
          if (count > 0 && codevalue == OP_VSPACE_EXTRA + OP_TYPEPOSPLUS)
            {
            active_count--;           /* Remove non-match possibility */
            next_active_state--;
            }
          count++;
          ADD_NEW_DATA(-state_offset, count, 0);
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_HSPACE_EXTRA + OP_TYPEPLUS:
      case OP_HSPACE_EXTRA + OP_TYPEMINPLUS:
      case OP_HSPACE_EXTRA + OP_TYPEPOSPLUS:
      count = current_state->count;  /* Already matched */
      if (count > 0) { ADD_ACTIVE(state_offset + 2, 0); }
      if (clen > 0)
        {
        BOOL OK;
        switch (c)
          {
          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 */
          OK = TRUE;
          break;


          default:
          OK = FALSE;
          break;
          }


        if (OK == (d == OP_HSPACE))
          {
          if (count > 0 && codevalue == OP_HSPACE_EXTRA + OP_TYPEPOSPLUS)
            {
            active_count--;           /* Remove non-match possibility */
            next_active_state--;
            }
          count++;
          ADD_NEW_DATA(-state_offset, count, 0);
          }
        }
      break;


      /*-----------------------------------------------------------------*/
#ifdef SUPPORT_UCP
      case OP_PROP_EXTRA + OP_TYPEQUERY:
      case OP_PROP_EXTRA + OP_TYPEMINQUERY:
      case OP_PROP_EXTRA + OP_TYPEPOSQUERY:
      count = 4;
      goto QS1;


      case OP_PROP_EXTRA + OP_TYPESTAR:
      case OP_PROP_EXTRA + OP_TYPEMINSTAR:
      case OP_PROP_EXTRA + OP_TYPEPOSSTAR:
      count = 0;


      QS1:


      ADD_ACTIVE(state_offset + 4, 0);
      if (clen > 0)
        {
        BOOL OK;
        const ucd_record * prop = GET_UCD(c);
        switch(code[2])
          {
          case PT_ANY:
          OK = TRUE;
          break;


          case PT_LAMP:
          OK = prop->chartype == ucp_Lu || prop->chartype == ucp_Ll || prop->chartype == ucp_Lt;
          break;


          case PT_GC:
          OK = _pcre_ucp_gentype[prop->chartype] == code[3];
          break;


          case PT_PC:
          OK = prop->chartype == code[3];
          break;


          case PT_SC:
          OK = prop->script == code[3];
          break;


          /* Should never occur, but keep compilers from grumbling. */


          default:
          OK = codevalue != OP_PROP;
          break;
          }


        if (OK == (d == OP_PROP))
          {
          if (codevalue == OP_PROP_EXTRA + OP_TYPEPOSSTAR ||
              codevalue == OP_PROP_EXTRA + OP_TYPEPOSQUERY)
            {
            active_count--;           /* Remove non-match possibility */
            next_active_state--;
            }
          ADD_NEW(state_offset + count, 0);
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_EXTUNI_EXTRA + OP_TYPEQUERY:
      case OP_EXTUNI_EXTRA + OP_TYPEMINQUERY:
      case OP_EXTUNI_EXTRA + OP_TYPEPOSQUERY:
      count = 2;
      goto QS2;


      case OP_EXTUNI_EXTRA + OP_TYPESTAR:
      case OP_EXTUNI_EXTRA + OP_TYPEMINSTAR:
      case OP_EXTUNI_EXTRA + OP_TYPEPOSSTAR:
      count = 0;


      QS2:


      ADD_ACTIVE(state_offset + 2, 0);
      if (clen > 0 && UCD_CATEGORY(c) != ucp_M)
        {
        const uschar *nptr = ptr + clen;
        int ncount = 0;
        if (codevalue == OP_EXTUNI_EXTRA + OP_TYPEPOSSTAR ||
            codevalue == OP_EXTUNI_EXTRA + OP_TYPEPOSQUERY)
          {
          active_count--;           /* Remove non-match possibility */
          next_active_state--;
          }
        while (nptr < end_subject)
          {
          int nd;
          int ndlen = 1;
          GETCHARLEN(nd, nptr, ndlen);
          if (UCD_CATEGORY(nd) != ucp_M) break;
          ncount++;
          nptr += ndlen;
          }
        ADD_NEW_DATA(-(state_offset + count), 0, ncount);
        }
      break;
#endif


      /*-----------------------------------------------------------------*/
      case OP_ANYNL_EXTRA + OP_TYPEQUERY:
      case OP_ANYNL_EXTRA + OP_TYPEMINQUERY:
      case OP_ANYNL_EXTRA + OP_TYPEPOSQUERY:
      count = 2;
      goto QS3;


      case OP_ANYNL_EXTRA + OP_TYPESTAR:
      case OP_ANYNL_EXTRA + OP_TYPEMINSTAR:
      case OP_ANYNL_EXTRA + OP_TYPEPOSSTAR:
      count = 0;


      QS3:
      ADD_ACTIVE(state_offset + 2, 0);
      if (clen > 0)
        {
        int ncount = 0;
        switch (c)
          {
          case 0x000b:
          case 0x000c:
          case 0x0085:
          case 0x2028:
          case 0x2029:
          if ((md->moptions & PCRE_BSR_ANYCRLF) != 0) break;
          goto ANYNL02;


          case 0x000d:
          if (ptr + 1 < end_subject && ptr[1] == 0x0a) ncount = 1;
          /* Fall through */


          ANYNL02:
          case 0x000a:
          if (codevalue == OP_ANYNL_EXTRA + OP_TYPEPOSSTAR ||
              codevalue == OP_ANYNL_EXTRA + OP_TYPEPOSQUERY)
            {
            active_count--;           /* Remove non-match possibility */
            next_active_state--;
            }
          ADD_NEW_DATA(-(state_offset + count), 0, ncount);
          break;


          default:
          break;
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_VSPACE_EXTRA + OP_TYPEQUERY:
      case OP_VSPACE_EXTRA + OP_TYPEMINQUERY:
      case OP_VSPACE_EXTRA + OP_TYPEPOSQUERY:
      count = 2;
      goto QS4;


      case OP_VSPACE_EXTRA + OP_TYPESTAR:
      case OP_VSPACE_EXTRA + OP_TYPEMINSTAR:
      case OP_VSPACE_EXTRA + OP_TYPEPOSSTAR:
      count = 0;


      QS4:
      ADD_ACTIVE(state_offset + 2, 0);
      if (clen > 0)
        {
        BOOL OK;
        switch (c)
          {
          case 0x000a:
          case 0x000b:
          case 0x000c:
          case 0x000d:
          case 0x0085:
          case 0x2028:
          case 0x2029:
          OK = TRUE;
          break;


          default:
          OK = FALSE;
          break;
          }
        if (OK == (d == OP_VSPACE))
          {
          if (codevalue == OP_VSPACE_EXTRA + OP_TYPEPOSSTAR ||
              codevalue == OP_VSPACE_EXTRA + OP_TYPEPOSQUERY)
            {
            active_count--;           /* Remove non-match possibility */
            next_active_state--;
            }
          ADD_NEW_DATA(-(state_offset + count), 0, 0);
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_HSPACE_EXTRA + OP_TYPEQUERY:
      case OP_HSPACE_EXTRA + OP_TYPEMINQUERY:
      case OP_HSPACE_EXTRA + OP_TYPEPOSQUERY:
      count = 2;
      goto QS5;


      case OP_HSPACE_EXTRA + OP_TYPESTAR:
      case OP_HSPACE_EXTRA + OP_TYPEMINSTAR:
      case OP_HSPACE_EXTRA + OP_TYPEPOSSTAR:
      count = 0;


      QS5:
      ADD_ACTIVE(state_offset + 2, 0);
      if (clen > 0)
        {
        BOOL OK;
        switch (c)
          {
          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 */
          OK = TRUE;
          break;


          default:
          OK = FALSE;
          break;
          }


        if (OK == (d == OP_HSPACE))
          {
          if (codevalue == OP_HSPACE_EXTRA + OP_TYPEPOSSTAR ||
              codevalue == OP_HSPACE_EXTRA + OP_TYPEPOSQUERY)
            {
            active_count--;           /* Remove non-match possibility */
            next_active_state--;
            }
          ADD_NEW_DATA(-(state_offset + count), 0, 0);
          }
        }
      break;


      /*-----------------------------------------------------------------*/
#ifdef SUPPORT_UCP
      case OP_PROP_EXTRA + OP_TYPEEXACT:
      case OP_PROP_EXTRA + OP_TYPEUPTO:
      case OP_PROP_EXTRA + OP_TYPEMINUPTO:
      case OP_PROP_EXTRA + OP_TYPEPOSUPTO:
      if (codevalue != OP_PROP_EXTRA + OP_TYPEEXACT)
        { ADD_ACTIVE(state_offset + 6, 0); }
      count = current_state->count;  /* Number already matched */
      if (clen > 0)
        {
        BOOL OK;
        const ucd_record * prop = GET_UCD(c);
        switch(code[4])
          {
          case PT_ANY:
          OK = TRUE;
          break;


          case PT_LAMP:
          OK = prop->chartype == ucp_Lu || prop->chartype == ucp_Ll || prop->chartype == ucp_Lt;
          break;


          case PT_GC:
          OK = _pcre_ucp_gentype[prop->chartype] == code[5];
          break;


          case PT_PC:
          OK = prop->chartype == code[5];
          break;


          case PT_SC:
          OK = prop->script == code[5];
          break;


          /* Should never occur, but keep compilers from grumbling. */


          default:
          OK = codevalue != OP_PROP;
          break;
          }


        if (OK == (d == OP_PROP))
          {
          if (codevalue == OP_PROP_EXTRA + OP_TYPEPOSUPTO)
            {
            active_count--;           /* Remove non-match possibility */
            next_active_state--;
            }
          if (++count >= GET2(code, 1))
            { ADD_NEW(state_offset + 6, 0); }
          else
            { ADD_NEW(state_offset, count); }
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_EXTUNI_EXTRA + OP_TYPEEXACT:
      case OP_EXTUNI_EXTRA + OP_TYPEUPTO:
      case OP_EXTUNI_EXTRA + OP_TYPEMINUPTO:
      case OP_EXTUNI_EXTRA + OP_TYPEPOSUPTO:
      if (codevalue != OP_EXTUNI_EXTRA + OP_TYPEEXACT)
        { ADD_ACTIVE(state_offset + 4, 0); }
      count = current_state->count;  /* Number already matched */
      if (clen > 0 && UCD_CATEGORY(c) != ucp_M)
        {
        const uschar *nptr = ptr + clen;
        int ncount = 0;
        if (codevalue == OP_EXTUNI_EXTRA + OP_TYPEPOSUPTO)
          {
          active_count--;           /* Remove non-match possibility */
          next_active_state--;
          }
        while (nptr < end_subject)
          {
          int nd;
          int ndlen = 1;
          GETCHARLEN(nd, nptr, ndlen);
          if (UCD_CATEGORY(nd) != ucp_M) break;
          ncount++;
          nptr += ndlen;
          }
        if (++count >= GET2(code, 1))
          { ADD_NEW_DATA(-(state_offset + 4), 0, ncount); }
        else
          { ADD_NEW_DATA(-state_offset, count, ncount); }
        }
      break;
#endif


      /*-----------------------------------------------------------------*/
      case OP_ANYNL_EXTRA + OP_TYPEEXACT:
      case OP_ANYNL_EXTRA + OP_TYPEUPTO:
      case OP_ANYNL_EXTRA + OP_TYPEMINUPTO:
      case OP_ANYNL_EXTRA + OP_TYPEPOSUPTO:
      if (codevalue != OP_ANYNL_EXTRA + OP_TYPEEXACT)
        { ADD_ACTIVE(state_offset + 4, 0); }
      count = current_state->count;  /* Number already matched */
      if (clen > 0)
        {
        int ncount = 0;
        switch (c)
          {
          case 0x000b:
          case 0x000c:
          case 0x0085:
          case 0x2028:
          case 0x2029:
          if ((md->moptions & PCRE_BSR_ANYCRLF) != 0) break;
          goto ANYNL03;


          case 0x000d:
          if (ptr + 1 < end_subject && ptr[1] == 0x0a) ncount = 1;
          /* Fall through */


          ANYNL03:
          case 0x000a:
          if (codevalue == OP_ANYNL_EXTRA + OP_TYPEPOSUPTO)
            {
            active_count--;           /* Remove non-match possibility */
            next_active_state--;
            }
          if (++count >= GET2(code, 1))
            { ADD_NEW_DATA(-(state_offset + 4), 0, ncount); }
          else
            { ADD_NEW_DATA(-state_offset, count, ncount); }
          break;


          default:
          break;
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_VSPACE_EXTRA + OP_TYPEEXACT:
      case OP_VSPACE_EXTRA + OP_TYPEUPTO:
      case OP_VSPACE_EXTRA + OP_TYPEMINUPTO:
      case OP_VSPACE_EXTRA + OP_TYPEPOSUPTO:
      if (codevalue != OP_VSPACE_EXTRA + OP_TYPEEXACT)
        { ADD_ACTIVE(state_offset + 4, 0); }
      count = current_state->count;  /* Number already matched */
      if (clen > 0)
        {
        BOOL OK;
        switch (c)
          {
          case 0x000a:
          case 0x000b:
          case 0x000c:
          case 0x000d:
          case 0x0085:
          case 0x2028:
          case 0x2029:
          OK = TRUE;
          break;


          default:
          OK = FALSE;
          }


        if (OK == (d == OP_VSPACE))
          {
          if (codevalue == OP_VSPACE_EXTRA + OP_TYPEPOSUPTO)
            {
            active_count--;           /* Remove non-match possibility */
            next_active_state--;
            }
          if (++count >= GET2(code, 1))
            { ADD_NEW_DATA(-(state_offset + 4), 0, 0); }
          else
            { ADD_NEW_DATA(-state_offset, count, 0); }
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_HSPACE_EXTRA + OP_TYPEEXACT:
      case OP_HSPACE_EXTRA + OP_TYPEUPTO:
      case OP_HSPACE_EXTRA + OP_TYPEMINUPTO:
      case OP_HSPACE_EXTRA + OP_TYPEPOSUPTO:
      if (codevalue != OP_HSPACE_EXTRA + OP_TYPEEXACT)
        { ADD_ACTIVE(state_offset + 4, 0); }
      count = current_state->count;  /* Number already matched */
      if (clen > 0)
        {
        BOOL OK;
        switch (c)
          {
          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 */
          OK = TRUE;
          break;


          default:
          OK = FALSE;
          break;
          }


        if (OK == (d == OP_HSPACE))
          {
          if (codevalue == OP_HSPACE_EXTRA + OP_TYPEPOSUPTO)
            {
            active_count--;           /* Remove non-match possibility */
            next_active_state--;
            }
          if (++count >= GET2(code, 1))
            { ADD_NEW_DATA(-(state_offset + 4), 0, 0); }
          else
            { ADD_NEW_DATA(-state_offset, count, 0); }
          }
        }
      break;


/* ========================================================================== */
      /* These opcodes are followed by a character that is usually compared
      to the current subject character; it is loaded into d. We still get
      here even if there is no subject character, because in some cases zero
      repetitions are permitted. */


      /*-----------------------------------------------------------------*/
      case OP_CHAR:
      if (clen > 0 && c == d) { ADD_NEW(state_offset + dlen + 1, 0); }
      break;


      /*-----------------------------------------------------------------*/
      case OP_CHARNC:
      if (clen == 0) break;


#ifdef SUPPORT_UTF8
      if (utf8)
        {
        if (c == d) { ADD_NEW(state_offset + dlen + 1, 0); } else
          {
          unsigned int othercase;
          if (c < 128) othercase = fcc[c]; else


          /* If we have Unicode property support, we can use it to test the
          other case of the character. */


#ifdef SUPPORT_UCP
          othercase = UCD_OTHERCASE(c);
#else
          othercase = NOTACHAR;
#endif


          if (d == othercase) { ADD_NEW(state_offset + dlen + 1, 0); }
          }
        }
      else
#endif  /* SUPPORT_UTF8 */


      /* Non-UTF-8 mode */
        {
        if (lcc[c] == lcc[d]) { ADD_NEW(state_offset + 2, 0); }
        }
      break;



#ifdef SUPPORT_UCP
      /*-----------------------------------------------------------------*/
      /* This is a tricky one because it can match more than one character.
      Find out how many characters to skip, and then set up a negative state
      to wait for them to pass before continuing. */


      case OP_EXTUNI:
      if (clen > 0 && UCD_CATEGORY(c) != ucp_M)
        {
        const uschar *nptr = ptr + clen;
        int ncount = 0;
        while (nptr < end_subject)
          {
          int nclen = 1;
          GETCHARLEN(c, nptr, nclen);
          if (UCD_CATEGORY(c) != ucp_M) break;
          ncount++;
          nptr += nclen;
          }
        ADD_NEW_DATA(-(state_offset + 1), 0, ncount);
        }
      break;
#endif


      /*-----------------------------------------------------------------*/
      /* This is a tricky like EXTUNI because it too can match more than one
      character (when CR is followed by LF). In this case, set up a negative
      state to wait for one character to pass before continuing. */


      case OP_ANYNL:
      if (clen > 0) switch(c)
        {
        case 0x000b:
        case 0x000c:
        case 0x0085:
        case 0x2028:
        case 0x2029:
        if ((md->moptions & PCRE_BSR_ANYCRLF) != 0) break;


        case 0x000a:
        ADD_NEW(state_offset + 1, 0);
        break;


        case 0x000d:
        if (ptr + 1 < end_subject && ptr[1] == 0x0a)
          {
          ADD_NEW_DATA(-(state_offset + 1), 0, 1);
          }
        else
          {
          ADD_NEW(state_offset + 1, 0);
          }
        break;
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_NOT_VSPACE:
      if (clen > 0) switch(c)
        {
        case 0x000a:
        case 0x000b:
        case 0x000c:
        case 0x000d:
        case 0x0085:
        case 0x2028:
        case 0x2029:
        break;


        default:
        ADD_NEW(state_offset + 1, 0);
        break;
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_VSPACE:
      if (clen > 0) switch(c)
        {
        case 0x000a:
        case 0x000b:
        case 0x000c:
        case 0x000d:
        case 0x0085:
        case 0x2028:
        case 0x2029:
        ADD_NEW(state_offset + 1, 0);
        break;


        default: break;
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_NOT_HSPACE:
      if (clen > 0) switch(c)
        {
        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;


        default:
        ADD_NEW(state_offset + 1, 0);
        break;
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_HSPACE:
      if (clen > 0) switch(c)
        {
        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 */
        ADD_NEW(state_offset + 1, 0);
        break;
        }
      break;


      /*-----------------------------------------------------------------*/
      /* Match a negated single character. This is only used for one-byte
      characters, that is, we know that d < 256. The character we are
      checking (c) can be multibyte. */


      case OP_NOT:
      if (clen > 0)
        {
        unsigned int otherd = ((ims & PCRE_CASELESS) != 0)? fcc[d] : d;
        if (c != d && c != otherd) { ADD_NEW(state_offset + dlen + 1, 0); }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_PLUS:
      case OP_MINPLUS:
      case OP_POSPLUS:
      case OP_NOTPLUS:
      case OP_NOTMINPLUS:
      case OP_NOTPOSPLUS:
      count = current_state->count;  /* Already matched */
      if (count > 0) { ADD_ACTIVE(state_offset + dlen + 1, 0); }
      if (clen > 0)
        {
        unsigned int otherd = NOTACHAR;
        if ((ims & PCRE_CASELESS) != 0)
          {
#ifdef SUPPORT_UTF8
          if (utf8 && d >= 128)
            {
#ifdef SUPPORT_UCP
            otherd = UCD_OTHERCASE(d);
#endif  /* SUPPORT_UCP */
            }
          else
#endif  /* SUPPORT_UTF8 */
          otherd = fcc[d];
          }
        if ((c == d || c == otherd) == (codevalue < OP_NOTSTAR))
          {
          if (count > 0 &&
              (codevalue == OP_POSPLUS || codevalue == OP_NOTPOSPLUS))
            {
            active_count--;             /* Remove non-match possibility */
            next_active_state--;
            }
          count++;
          ADD_NEW(state_offset, count);
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_QUERY:
      case OP_MINQUERY:
      case OP_POSQUERY:
      case OP_NOTQUERY:
      case OP_NOTMINQUERY:
      case OP_NOTPOSQUERY:
      ADD_ACTIVE(state_offset + dlen + 1, 0);
      if (clen > 0)
        {
        unsigned int otherd = NOTACHAR;
        if ((ims & PCRE_CASELESS) != 0)
          {
#ifdef SUPPORT_UTF8
          if (utf8 && d >= 128)
            {
#ifdef SUPPORT_UCP
            otherd = UCD_OTHERCASE(d);
#endif  /* SUPPORT_UCP */
            }
          else
#endif  /* SUPPORT_UTF8 */
          otherd = fcc[d];
          }
        if ((c == d || c == otherd) == (codevalue < OP_NOTSTAR))
          {
          if (codevalue == OP_POSQUERY || codevalue == OP_NOTPOSQUERY)
            {
            active_count--;            /* Remove non-match possibility */
            next_active_state--;
            }
          ADD_NEW(state_offset + dlen + 1, 0);
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_STAR:
      case OP_MINSTAR:
      case OP_POSSTAR:
      case OP_NOTSTAR:
      case OP_NOTMINSTAR:
      case OP_NOTPOSSTAR:
      ADD_ACTIVE(state_offset + dlen + 1, 0);
      if (clen > 0)
        {
        unsigned int otherd = NOTACHAR;
        if ((ims & PCRE_CASELESS) != 0)
          {
#ifdef SUPPORT_UTF8
          if (utf8 && d >= 128)
            {
#ifdef SUPPORT_UCP
            otherd = UCD_OTHERCASE(d);
#endif  /* SUPPORT_UCP */
            }
          else
#endif  /* SUPPORT_UTF8 */
          otherd = fcc[d];
          }
        if ((c == d || c == otherd) == (codevalue < OP_NOTSTAR))
          {
          if (codevalue == OP_POSSTAR || codevalue == OP_NOTPOSSTAR)
            {
            active_count--;            /* Remove non-match possibility */
            next_active_state--;
            }
          ADD_NEW(state_offset, 0);
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_EXACT:
      case OP_NOTEXACT:
      count = current_state->count;  /* Number already matched */
      if (clen > 0)
        {
        unsigned int otherd = NOTACHAR;
        if ((ims & PCRE_CASELESS) != 0)
          {
#ifdef SUPPORT_UTF8
          if (utf8 && d >= 128)
            {
#ifdef SUPPORT_UCP
            otherd = UCD_OTHERCASE(d);
#endif  /* SUPPORT_UCP */
            }
          else
#endif  /* SUPPORT_UTF8 */
          otherd = fcc[d];
          }
        if ((c == d || c == otherd) == (codevalue < OP_NOTSTAR))
          {
          if (++count >= GET2(code, 1))
            { ADD_NEW(state_offset + dlen + 3, 0); }
          else
            { ADD_NEW(state_offset, count); }
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_UPTO:
      case OP_MINUPTO:
      case OP_POSUPTO:
      case OP_NOTUPTO:
      case OP_NOTMINUPTO:
      case OP_NOTPOSUPTO:
      ADD_ACTIVE(state_offset + dlen + 3, 0);
      count = current_state->count;  /* Number already matched */
      if (clen > 0)
        {
        unsigned int otherd = NOTACHAR;
        if ((ims & PCRE_CASELESS) != 0)
          {
#ifdef SUPPORT_UTF8
          if (utf8 && d >= 128)
            {
#ifdef SUPPORT_UCP
            otherd = UCD_OTHERCASE(d);
#endif  /* SUPPORT_UCP */
            }
          else
#endif  /* SUPPORT_UTF8 */
          otherd = fcc[d];
          }
        if ((c == d || c == otherd) == (codevalue < OP_NOTSTAR))
          {
          if (codevalue == OP_POSUPTO || codevalue == OP_NOTPOSUPTO)
            {
            active_count--;             /* Remove non-match possibility */
            next_active_state--;
            }
          if (++count >= GET2(code, 1))
            { ADD_NEW(state_offset + dlen + 3, 0); }
          else
            { ADD_NEW(state_offset, count); }
          }
        }
      break;



/* ========================================================================== */
      /* These are the class-handling opcodes */


      case OP_CLASS:
      case OP_NCLASS:
      case OP_XCLASS:
        {
        BOOL isinclass = FALSE;
        int next_state_offset;
        const uschar *ecode;


        /* For a simple class, there is always just a 32-byte table, and we
        can set isinclass from it. */


        if (codevalue != OP_XCLASS)
          {
          ecode = code + 33;
          if (clen > 0)
            {
            isinclass = (c > 255)? (codevalue == OP_NCLASS) :
              ((code[1 + c/8] & (1 << (c&7))) != 0);
            }
          }


        /* An extended class may have a table or a list of single characters,
        ranges, or both, and it may be positive or negative. There's a
        function that sorts all this out. */


        else
         {
         ecode = code + GET(code, 1);
         if (clen > 0) isinclass = _pcre_xclass(c, code + 1 + LINK_SIZE);
         }


        /* At this point, isinclass is set for all kinds of class, and ecode
        points to the byte after the end of the class. If there is a
        quantifier, this is where it will be. */


        next_state_offset = ecode - start_code;


        switch (*ecode)
          {
          case OP_CRSTAR:
          case OP_CRMINSTAR:
          ADD_ACTIVE(next_state_offset + 1, 0);
          if (isinclass) { ADD_NEW(state_offset, 0); }
          break;


          case OP_CRPLUS:
          case OP_CRMINPLUS:
          count = current_state->count;  /* Already matched */
          if (count > 0) { ADD_ACTIVE(next_state_offset + 1, 0); }
          if (isinclass) { count++; ADD_NEW(state_offset, count); }
          break;


          case OP_CRQUERY:
          case OP_CRMINQUERY:
          ADD_ACTIVE(next_state_offset + 1, 0);
          if (isinclass) { ADD_NEW(next_state_offset + 1, 0); }
          break;


          case OP_CRRANGE:
          case OP_CRMINRANGE:
          count = current_state->count;  /* Already matched */
          if (count >= GET2(ecode, 1))
            { ADD_ACTIVE(next_state_offset + 5, 0); }
          if (isinclass)
            {
            int max = GET2(ecode, 3);
            if (++count >= max && max != 0)   /* Max 0 => no limit */
              { ADD_NEW(next_state_offset + 5, 0); }
            else
              { ADD_NEW(state_offset, count); }
            }
          break;


          default:
          if (isinclass) { ADD_NEW(next_state_offset, 0); }
          break;
          }
        }
      break;


/* ========================================================================== */
      /* These are the opcodes for fancy brackets of various kinds. We have
      to use recursion in order to handle them. The "always failing" assersion
      (?!) is optimised when compiling to OP_FAIL, so we have to support that,
      though the other "backtracking verbs" are not supported. */


      case OP_FAIL:
      break;


      case OP_ASSERT:
      case OP_ASSERT_NOT:
      case OP_ASSERTBACK:
      case OP_ASSERTBACK_NOT:
        {
        int rc;
        int local_offsets[2];
        int local_workspace[1000];
        const uschar *endasscode = code + GET(code, 1);


        while (*endasscode == OP_ALT) endasscode += GET(endasscode, 1);


        rc = internal_dfa_exec(
          md,                                   /* static match data */
          code,                                 /* this subexpression's code */
          ptr,                                  /* where we currently are */
          ptr - start_subject,                  /* start offset */
          local_offsets,                        /* offset vector */
          sizeof(local_offsets)/sizeof(int),    /* size of same */
          local_workspace,                      /* workspace vector */
          sizeof(local_workspace)/sizeof(int),  /* size of same */
          ims,                                  /* the current ims flags */
          rlevel,                               /* function recursion level */
          recursing);                           /* pass on regex recursion */


        if ((rc >= 0) == (codevalue == OP_ASSERT || codevalue == OP_ASSERTBACK))
            { ADD_ACTIVE(endasscode + LINK_SIZE + 1 - start_code, 0); }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_COND:
      case OP_SCOND:
        {
        int local_offsets[1000];
        int local_workspace[1000];
        int codelink = GET(code, 1);
        int condcode;


        /* Because of the way auto-callout works during compile, a callout item
        is inserted between OP_COND and an assertion condition. This does not
        happen for the other conditions. */


        if (code[LINK_SIZE+1] == OP_CALLOUT)
          {
          rrc = 0;
          if (pcre_callout != NULL)
            {
            pcre_callout_block cb;
            cb.version          = 1;   /* Version 1 of the callout block */
            cb.callout_number   = code[LINK_SIZE+2];
            cb.offset_vector    = offsets;
            cb.subject          = (PCRE_SPTR)start_subject;
            cb.subject_length   = end_subject - start_subject;
            cb.start_match      = current_subject - start_subject;
            cb.current_position = ptr - start_subject;
            cb.pattern_position = GET(code, LINK_SIZE + 3);
            cb.next_item_length = GET(code, 3 + 2*LINK_SIZE);
            cb.capture_top      = 1;
            cb.capture_last     = -1;
            cb.callout_data     = md->callout_data;
            if ((rrc = (*pcre_callout)(&cb)) < 0) return rrc;   /* Abandon */
            }
          if (rrc > 0) break;                      /* Fail this thread */
          code += _pcre_OP_lengths[OP_CALLOUT];    /* Skip callout data */
          }


        condcode = code[LINK_SIZE+1];


        /* Back reference conditions are not supported */


        if (condcode == OP_CREF) return PCRE_ERROR_DFA_UCOND;


        /* The DEFINE condition is always false */


        if (condcode == OP_DEF)
          { ADD_ACTIVE(state_offset + codelink + LINK_SIZE + 1, 0); }


        /* The only supported version of OP_RREF is for the value RREF_ANY,
        which means "test if in any recursion". We can't test for specifically
        recursed groups. */


        else if (condcode == OP_RREF)
          {
          int value = GET2(code, LINK_SIZE+2);
          if (value != RREF_ANY) return PCRE_ERROR_DFA_UCOND;
          if (recursing > 0)
            { ADD_ACTIVE(state_offset + LINK_SIZE + 4, 0); }
          else { ADD_ACTIVE(state_offset + codelink + LINK_SIZE + 1, 0); }
          }


        /* Otherwise, the condition is an assertion */


        else
          {
          int rc;
          const uschar *asscode = code + LINK_SIZE + 1;
          const uschar *endasscode = asscode + GET(asscode, 1);


          while (*endasscode == OP_ALT) endasscode += GET(endasscode, 1);


          rc = internal_dfa_exec(
            md,                                   /* fixed match data */
            asscode,                              /* this subexpression's code */
            ptr,                                  /* where we currently are */
            ptr - start_subject,                  /* start offset */
            local_offsets,                        /* offset vector */
            sizeof(local_offsets)/sizeof(int),    /* size of same */
            local_workspace,                      /* workspace vector */
            sizeof(local_workspace)/sizeof(int),  /* size of same */
            ims,                                  /* the current ims flags */
            rlevel,                               /* function recursion level */
            recursing);                           /* pass on regex recursion */


          if ((rc >= 0) ==
                (condcode == OP_ASSERT || condcode == OP_ASSERTBACK))
            { ADD_ACTIVE(endasscode + LINK_SIZE + 1 - start_code, 0); }
          else
            { ADD_ACTIVE(state_offset + codelink + LINK_SIZE + 1, 0); }
          }
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_RECURSE:
        {
        int local_offsets[1000];
        int local_workspace[1000];
        int rc;


        DPRINTF(("%.*sStarting regex recursion %d\n", rlevel*2-2, SP,
          recursing + 1));


        rc = internal_dfa_exec(
          md,                                   /* fixed match data */
          start_code + GET(code, 1),            /* this subexpression's code */
          ptr,                                  /* where we currently are */
          ptr - start_subject,                  /* start offset */
          local_offsets,                        /* offset vector */
          sizeof(local_offsets)/sizeof(int),    /* size of same */
          local_workspace,                      /* workspace vector */
          sizeof(local_workspace)/sizeof(int),  /* size of same */
          ims,                                  /* the current ims flags */
          rlevel,                               /* function recursion level */
          recursing + 1);                       /* regex recurse level */


        DPRINTF(("%.*sReturn from regex recursion %d: rc=%d\n", rlevel*2-2, SP,
          recursing + 1, rc));


        /* Ran out of internal offsets */


        if (rc == 0) return PCRE_ERROR_DFA_RECURSE;


        /* For each successful matched substring, set up the next state with a
        count of characters to skip before trying it. Note that the count is in
        characters, not bytes. */


        if (rc > 0)
          {
          for (rc = rc*2 - 2; rc >= 0; rc -= 2)
            {
            const uschar *p = start_subject + local_offsets[rc];
            const uschar *pp = start_subject + local_offsets[rc+1];
            int charcount = local_offsets[rc+1] - local_offsets[rc];
            while (p < pp) if ((*p++ & 0xc0) == 0x80) charcount--;
            if (charcount > 0)
              {
              ADD_NEW_DATA(-(state_offset + LINK_SIZE + 1), 0, (charcount - 1));
              }
            else
              {
              ADD_ACTIVE(state_offset + LINK_SIZE + 1, 0);
              }
            }
          }
        else if (rc != PCRE_ERROR_NOMATCH) return rc;
        }
      break;


      /*-----------------------------------------------------------------*/
      case OP_ONCE:
        {
        int local_offsets[2];
        int local_workspace[1000];


        int rc = internal_dfa_exec(
          md,                                   /* fixed match data */
          code,                                 /* this subexpression's code */
          ptr,                                  /* where we currently are */
          ptr - start_subject,                  /* start offset */
          local_offsets,                        /* offset vector */
          sizeof(local_offsets)/sizeof(int),    /* size of same */
          local_workspace,                      /* workspace vector */
          sizeof(local_workspace)/sizeof(int),  /* size of same */
          ims,                                  /* the current ims flags */
          rlevel,                               /* function recursion level */
          recursing);                           /* pass on regex recursion */


        if (rc >= 0)
          {
          const uschar *end_subpattern = code;
          int charcount = local_offsets[1] - local_offsets[0];
          int next_state_offset, repeat_state_offset;


          do { end_subpattern += GET(end_subpattern, 1); }
            while (*end_subpattern == OP_ALT);
          next_state_offset = end_subpattern - start_code + LINK_SIZE + 1;


          /* If the end of this subpattern is KETRMAX or KETRMIN, we must
          arrange for the repeat state also to be added to the relevant list.
          Calculate the offset, or set -1 for no repeat. */


          repeat_state_offset = (*end_subpattern == OP_KETRMAX ||
                                 *end_subpattern == OP_KETRMIN)?
            end_subpattern - start_code - GET(end_subpattern, 1) : -1;


          /* If we have matched an empty string, add the next state at the
          current character pointer. This is important so that the duplicate
          checking kicks in, which is what breaks infinite loops that match an
          empty string. */


          if (charcount == 0)
            {
            ADD_ACTIVE(next_state_offset, 0);
            }


          /* Optimization: if there are no more active states, and there
          are no new states yet set up, then skip over the subject string
          right here, to save looping. Otherwise, set up the new state to swing
          into action when the end of the substring is reached. */


          else if (i + 1 >= active_count && new_count == 0)
            {
            ptr += charcount;
            clen = 0;
            ADD_NEW(next_state_offset, 0);


            /* If we are adding a repeat state at the new character position,
            we must fudge things so that it is the only current state.
            Otherwise, it might be a duplicate of one we processed before, and
            that would cause it to be skipped. */


            if (repeat_state_offset >= 0)
              {
              next_active_state = active_states;
              active_count = 0;
              i = -1;
              ADD_ACTIVE(repeat_state_offset, 0);
              }
            }
          else
            {
            const uschar *p = start_subject + local_offsets[0];
            const uschar *pp = start_subject + local_offsets[1];
            while (p < pp) if ((*p++ & 0xc0) == 0x80) charcount--;
            ADD_NEW_DATA(-next_state_offset, 0, (charcount - 1));
            if (repeat_state_offset >= 0)
              { ADD_NEW_DATA(-repeat_state_offset, 0, (charcount - 1)); }
            }


          }
        else if (rc != PCRE_ERROR_NOMATCH) return rc;
        }
      break;



/* ========================================================================== */
      /* Handle callouts */


      case OP_CALLOUT:
      rrc = 0;
      if (pcre_callout != NULL)
        {
        pcre_callout_block cb;
        cb.version          = 1;   /* Version 1 of the callout block */
        cb.callout_number   = code[1];
        cb.offset_vector    = offsets;
        cb.subject          = (PCRE_SPTR)start_subject;
        cb.subject_length   = end_subject - start_subject;
        cb.start_match      = current_subject - start_subject;
        cb.current_position = ptr - start_subject;
        cb.pattern_position = GET(code, 2);
        cb.next_item_length = GET(code, 2 + LINK_SIZE);
        cb.capture_top      = 1;
        cb.capture_last     = -1;
        cb.callout_data     = md->callout_data;
        if ((rrc = (*pcre_callout)(&cb)) < 0) return rrc;   /* Abandon */
        }
      if (rrc == 0)
        { ADD_ACTIVE(state_offset + _pcre_OP_lengths[OP_CALLOUT], 0); }
      break;



/* ========================================================================== */
      default:        /* Unsupported opcode */
      return PCRE_ERROR_DFA_UITEM;
      }


    NEXT_ACTIVE_STATE: continue;


    }      /* End of loop scanning active states */


/* We have finished the processing at the current subject character. If no
new states have been set for the next character, we have found all the
matches that we are going to find. If we are at the top level and partial
matching has been requested, check for appropriate conditions. */

  if (new_count <= 0)
    {
    if (match_count < 0 &&                     /* No matches found */
        rlevel == 1 &&                         /* Top level match function */
        (md->moptions & PCRE_PARTIAL) != 0 &&  /* Want partial matching */
        ptr >= end_subject &&                  /* Reached end of subject */
        ptr > current_subject)                 /* Matched non-empty string */
      {
      if (offsetcount >= 2)
        {
        offsets[0] = current_subject - start_subject;
        offsets[1] = end_subject - start_subject;
        }
      match_count = PCRE_ERROR_PARTIAL;
      }


    DPRINTF(("%.*sEnd of internal_dfa_exec %d: returning %d\n"
      "%.*s---------------------\n\n", rlevel*2-2, SP, rlevel, match_count,
      rlevel*2-2, SP));
    break;        /* In effect, "return", but see the comment below */
    }


/* One or more states are active for the next character. */

  ptr += clen;    /* Advance to next subject character */
  }               /* Loop to move along the subject string */


/* Control gets here from "break" a few lines above. We do it this way because
if we use "return" above, we have compiler trouble. Some compilers warn if
there's nothing here because they think the function doesn't return a value. On
the other hand, if we put a dummy statement here, some more clever compilers
complain that it can't be reached. Sigh. */

return match_count;
}




/*************************************************
*    Execute a Regular Expression - DFA engine   *
*************************************************/


/* This external function applies a compiled re to a subject string using a DFA
engine. This function calls the internal function multiple times if the pattern
is not anchored.

Arguments:
  argument_re     points to the compiled expression
  extra_data      points to extra data or is NULL
  subject         points to the subject string
  length          length of subject string (may contain binary zeros)
  start_offset    where to start in the subject string
  options         option bits
  offsets         vector of match offsets
  offsetcount     size of same
  workspace       workspace vector
  wscount         size of same


Returns:          > 0 => number of match offset pairs placed in offsets
                  = 0 => offsets overflowed; longest matches are present
                   -1 => failed to match
                 < -1 => some kind of unexpected problem
*/


PCRE_EXP_DEFN int PCRE_CALL_CONVENTION
pcre_dfa_exec(const pcre *argument_re, const pcre_extra *extra_data,
const char *subject, int length, int start_offset, int options, int *offsets,
int offsetcount, int *workspace, int wscount)
{
real_pcre *re = (real_pcre *)argument_re;
dfa_match_data match_block;
dfa_match_data *md = &match_block;
BOOL utf8, anchored, startline, firstline;
const uschar *current_subject, *end_subject, *lcc;

pcre_study_data internal_study;
const pcre_study_data *study = NULL;
real_pcre internal_re;

const uschar *req_byte_ptr;
const uschar *start_bits = NULL;
BOOL first_byte_caseless = FALSE;
BOOL req_byte_caseless = FALSE;
int first_byte = -1;
int req_byte = -1;
int req_byte2 = -1;
int newline;

/* Plausibility checks */

if ((options & ~PUBLIC_DFA_EXEC_OPTIONS) != 0) return PCRE_ERROR_BADOPTION;
if (re == NULL || subject == NULL || workspace == NULL ||
(offsets == NULL && offsetcount > 0)) return PCRE_ERROR_NULL;
if (offsetcount < 0) return PCRE_ERROR_BADCOUNT;
if (wscount < 20) return PCRE_ERROR_DFA_WSSIZE;

/* We need to find the pointer to any study data before we test for byte
flipping, so we scan the extra_data block first. This may set two fields in the
match block, so we must initialize them beforehand. However, the other fields
in the match block must not be set until after the byte flipping. */

md->tables = re->tables;
md->callout_data = NULL;

if (extra_data != NULL)
  {
  unsigned int flags = extra_data->flags;
  if ((flags & PCRE_EXTRA_STUDY_DATA) != 0)
    study = (const pcre_study_data *)extra_data->study_data;
  if ((flags & PCRE_EXTRA_MATCH_LIMIT) != 0) return PCRE_ERROR_DFA_UMLIMIT;
  if ((flags & PCRE_EXTRA_MATCH_LIMIT_RECURSION) != 0)
    return PCRE_ERROR_DFA_UMLIMIT;
  if ((flags & PCRE_EXTRA_CALLOUT_DATA) != 0)
    md->callout_data = extra_data->callout_data;
  if ((flags & PCRE_EXTRA_TABLES) != 0)
    md->tables = extra_data->tables;
  }


/* Check that the first field in the block is the magic number. If it is not,
test for a regex that was compiled on a host of opposite endianness. If this is
the case, flipped values are put in internal_re and internal_study if there was
study data too. */

if (re->magic_number != MAGIC_NUMBER)
{
re = _pcre_try_flipped(re, &internal_re, study, &internal_study);
if (re == NULL) return PCRE_ERROR_BADMAGIC;
if (study != NULL) study = &internal_study;
}

/* Set some local values */

current_subject = (const unsigned char *)subject + start_offset;
end_subject = (const unsigned char *)subject + length;
req_byte_ptr = current_subject - 1;

#ifdef SUPPORT_UTF8
utf8 = (re->options & PCRE_UTF8) != 0;
#else
utf8 = FALSE;
#endif

anchored = (options & (PCRE_ANCHORED|PCRE_DFA_RESTART)) != 0 ||
(re->options & PCRE_ANCHORED) != 0;

/* The remaining fixed data for passing around. */

md->start_code = (const uschar *)argument_re +
    re->name_table_offset + re->name_count * re->name_entry_size;
md->start_subject = (const unsigned char *)subject;
md->end_subject = end_subject;
md->moptions = options;
md->poptions = re->options;


/* If the BSR option is not set at match time, copy what was set
at compile time. */

if ((md->moptions & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) == 0)
  {
  if ((re->options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) != 0)
    md->moptions |= re->options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE);
#ifdef BSR_ANYCRLF
  else md->moptions |= PCRE_BSR_ANYCRLF;
#endif
  }


/* Handle different types of newline. The three bits give eight cases. If
nothing is set at run time, whatever was used at compile time applies. */

switch ((((options & PCRE_NEWLINE_BITS) == 0)? re->options : (pcre_uint32)options) &
         PCRE_NEWLINE_BITS)
  {
  case 0: newline = NEWLINE; break;   /* Compile-time default */
  case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
  case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
  case PCRE_NEWLINE_CR+
       PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
  case PCRE_NEWLINE_ANY: newline = -1; break;
  case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
  default: return PCRE_ERROR_BADNEWLINE;
  }


if (newline == -2)
  {
  md->nltype = NLTYPE_ANYCRLF;
  }
else if (newline < 0)
  {
  md->nltype = NLTYPE_ANY;
  }
else
  {
  md->nltype = NLTYPE_FIXED;
  if (newline > 255)
    {
    md->nllen = 2;
    md->nl[0] = (newline >> 8) & 255;
    md->nl[1] = newline & 255;
    }
  else
    {
    md->nllen = 1;
    md->nl[0] = newline;
    }
  }


/* Check a UTF-8 string if required. Unfortunately there's no way of passing
back the character offset. */

#ifdef SUPPORT_UTF8
if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0)
  {
  if (_pcre_valid_utf8((uschar *)subject, length) >= 0)
    return PCRE_ERROR_BADUTF8;
  if (start_offset > 0 && start_offset < length)
    {
    int tb = ((uschar *)subject)[start_offset];
    if (tb > 127)
      {
      tb &= 0xc0;
      if (tb != 0 && tb != 0xc0) return PCRE_ERROR_BADUTF8_OFFSET;
      }
    }
  }
#endif


/* If the exec call supplied NULL for tables, use the inbuilt ones. This
is a feature that makes it possible to save compiled regex and re-use them
in other programs later. */

if (md->tables == NULL) md->tables = _pcre_default_tables;

/* The lower casing table and the "must be at the start of a line" flag are
used in a loop when finding where to start. */

lcc = md->tables + lcc_offset;
startline = (re->flags & PCRE_STARTLINE) != 0;
firstline = (re->options & PCRE_FIRSTLINE) != 0;

/* Set up the first character to match, if available. The first_byte value is
never set for an anchored regular expression, but the anchoring may be forced
at run time, so we have to test for anchoring. The first char may be unset for
an unanchored pattern, of course. If there's no first char and the pattern was
studied, there may be a bitmap of possible first characters. */

if (!anchored)
  {
  if ((re->flags & PCRE_FIRSTSET) != 0)
    {
    first_byte = re->first_byte & 255;
    if ((first_byte_caseless = ((re->first_byte & REQ_CASELESS) != 0)) == TRUE)
      first_byte = lcc[first_byte];
    }
  else
    {
    if (startline && study != NULL &&
         (study->options & PCRE_STUDY_MAPPED) != 0)
      start_bits = study->start_bits;
    }
  }


/* For anchored or unanchored matches, there may be a "last known required
character" set. */

if ((re->flags & PCRE_REQCHSET) != 0)
{
req_byte = re->req_byte & 255;
req_byte_caseless = (re->req_byte & REQ_CASELESS) != 0;
req_byte2 = (md->tables + fcc_offset)[req_byte]; /* case flipped */
}

/* Call the main matching function, looping for a non-anchored regex after a
failed match. If not restarting, perform certain optimizations at the start of
a match. */

for (;;)
{
int rc;

  if ((options & PCRE_DFA_RESTART) == 0)
    {
    const uschar *save_end_subject = end_subject;


    /* If firstline is TRUE, the start of the match is constrained to the first
    line of a multiline string. Implement this by temporarily adjusting
    end_subject so that we stop scanning at a newline. If the match fails at
    the newline, later code breaks this loop. */


    if (firstline)
      {
      USPTR t = current_subject;
#ifdef SUPPORT_UTF8
      if (utf8)
        {
        while (t < md->end_subject && !IS_NEWLINE(t))
          {
          t++;
          while (t < end_subject && (*t & 0xc0) == 0x80) t++;
          }
        }
      else
#endif
      while (t < md->end_subject && !IS_NEWLINE(t)) t++;
      end_subject = t;
      }


    /* There are some optimizations that avoid running the match if a known
    starting point is not found, or if a known later character is not present.
    However, there is an option that disables these, for testing and for
    ensuring that all callouts do actually occur. */


    if ((options & PCRE_NO_START_OPTIMIZE) == 0)
      {


      /* Advance to a known first byte. */


      if (first_byte >= 0)
        {
        if (first_byte_caseless)
          while (current_subject < end_subject &&
                 lcc[*current_subject] != first_byte)
            current_subject++;
        else
          while (current_subject < end_subject &&
                 *current_subject != first_byte)
            current_subject++;
        }


      /* Or to just after a linebreak for a multiline match if possible */


      else if (startline)
        {
        if (current_subject > md->start_subject + start_offset)
          {
#ifdef SUPPORT_UTF8
          if (utf8)
            {
            while (current_subject < end_subject &&
                   !WAS_NEWLINE(current_subject))
              {
              current_subject++;
              while(current_subject < end_subject &&
                    (*current_subject & 0xc0) == 0x80)
                current_subject++;
              }
            }
          else
#endif
          while (current_subject < end_subject && !WAS_NEWLINE(current_subject))
            current_subject++;


          /* If we have just passed a CR and the newline option is ANY or
          ANYCRLF, and we are now at a LF, advance the match position by one
          more character. */


          if (current_subject[-1] == CHAR_CR &&
               (md->nltype == NLTYPE_ANY || md->nltype == NLTYPE_ANYCRLF) &&
               current_subject < end_subject &&
               *current_subject == CHAR_NL)
            current_subject++;
          }
        }


      /* Or to a non-unique first char after study */


      else if (start_bits != NULL)
        {
        while (current_subject < end_subject)
          {
          register unsigned int c = *current_subject;
          if ((start_bits[c/8] & (1 << (c&7))) == 0) current_subject++;
            else break;
          }
        }
      }


    /* Restore fudged end_subject */


    end_subject = save_end_subject;
    }


/* If req_byte is set, we know that that character must appear in the subject
for the match to succeed. If the first character is set, req_byte must be
later in the subject; otherwise the test starts at the match point. This
optimization can save a huge amount of work in patterns with nested unlimited
repeats that aren't going to match. Writing separate code for cased/caseless
versions makes it go faster, as does using an autoincrement and backing off
on a match.

HOWEVER: when the subject string is very, very long, searching to its end can
take a long time, and give bad performance on quite ordinary patterns. This
showed up when somebody was matching /^C/ on a 32-megabyte string... so we
don't do this when the string is sufficiently long.

ALSO: this processing is disabled when partial matching is requested, and can
also be explicitly deactivated. */

  if ((options & PCRE_NO_START_OPTIMIZE) == 0 &&
      req_byte >= 0 &&
      end_subject - current_subject < REQ_BYTE_MAX &&
      (options & PCRE_PARTIAL) == 0)
    {
    register const uschar *p = current_subject + ((first_byte >= 0)? 1 : 0);


    /* We don't need to repeat the search if we haven't yet reached the
    place we found it at last time. */


    if (p > req_byte_ptr)
      {
      if (req_byte_caseless)
        {
        while (p < end_subject)
          {
          register int pp = *p++;
          if (pp == req_byte || pp == req_byte2) { p--; break; }
          }
        }
      else
        {
        while (p < end_subject)
          {
          if (*p++ == req_byte) { p--; break; }
          }
        }


      /* If we can't find the required character, break the matching loop,
      which will cause a return or PCRE_ERROR_NOMATCH. */


      if (p >= end_subject) break;


      /* If we have found the required character, save the point where we
      found it, so that we don't search again next time round the loop if
      the start hasn't passed this character yet. */


      req_byte_ptr = p;
      }
    }


/* OK, now we can do the business */

  rc = internal_dfa_exec(
    md,                                /* fixed match data */
    md->start_code,                    /* this subexpression's code */
    current_subject,                   /* where we currently are */
    start_offset,                      /* start offset in subject */
    offsets,                           /* offset vector */
    offsetcount,                       /* size of same */
    workspace,                         /* workspace vector */
    wscount,                           /* size of same */
    re->options & (PCRE_CASELESS|PCRE_MULTILINE|PCRE_DOTALL), /* ims flags */
    0,                                 /* function recurse level */
    0);                                /* regex recurse level */


/* Anything other than "no match" means we are done, always; otherwise, carry
on only if not anchored. */

if (rc != PCRE_ERROR_NOMATCH || anchored) return rc;

/* Advance to the next subject character unless we are at the end of a line
and firstline is set. */

  if (firstline && IS_NEWLINE(current_subject)) break;
  current_subject++;
  if (utf8)
    {
    while (current_subject < end_subject && (*current_subject & 0xc0) == 0x80)
      current_subject++;
    }
  if (current_subject > end_subject) break;


/* If we have just passed a CR and we are now at a LF, and the pattern does
not contain any explicit matches for \r or \n, and the newline option is CRLF
or ANY or ANYCRLF, advance the match position by one more character. */

  if (current_subject[-1] == CHAR_CR &&
      current_subject < end_subject &&
      *current_subject == CHAR_NL &&
      (re->flags & PCRE_HASCRORLF) == 0 &&
        (md->nltype == NLTYPE_ANY ||
         md->nltype == NLTYPE_ANYCRLF ||
         md->nllen == 2))
    current_subject++;


} /* "Bumpalong" loop */

return PCRE_ERROR_NOMATCH;
}

/* End of pcre_dfa_exec.c */