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

Diff of /code/trunk/pcre_exec.c

Parent Directory | Revision Log | View Patch Patch

-revision 604 by ph10,
Thu Jun  2 19:04:54 2011 UTC
+revision 677 by ph10,
Sun Aug 28 10:50:07 2011 UTC
 Line 57 
 possible. There are also some static sup
  #undef min
  #undef max
  /* Values for setting in md->match_function_type to indicate two special types
  of call to match(). We do it this way to save on using another stack variable,
  as stack usage is to be discouraged. */
  #define MATCH_CONDASSERT     1  /* Called to check a condition assertion */
 Line 76 
 negative to avoid the external error cod
  #define MATCH_ACCEPT       (-999)
  #define MATCH_COMMIT       (-998)
  #define MATCH_KETRPOS      (-997)
- #define MATCH_PRUNE        (-996)
+ #define MATCH_ONCE         (-996)
- #define MATCH_SKIP         (-995)
+ #define MATCH_PRUNE        (-995)
- #define MATCH_SKIP_ARG     (-994)
+ #define MATCH_SKIP         (-994)
- #define MATCH_THEN         (-993)
+ #define MATCH_SKIP_ARG     (-993)
+ #define MATCH_THEN         (-992)
  /* This is a convenience macro for code that occurs many times. */
-Line 137 
 while (length-- > 0)
+Line 138 
 while (length-- > 0)
  /* Normally, if a back reference hasn't been set, the length that is passed is
  negative, so the match always fails. However, in JavaScript compatibility mode,
  the length passed is zero. Note that in caseless UTF-8 mode, the number of
  subject bytes matched may be different to the number of reference bytes.
  Arguments:
-Line 184 
 if (caseless)
+Line 185 
 if (caseless)
  #ifdef SUPPORT_UCP
    if (md->utf8)
      {
      /* Match characters up to the end of the reference. NOTE: the number of
      bytes matched may differ, because there are some characters whose upper and
      lower case versions code as different numbers of bytes. For example, U+023A
      (2 bytes in UTF-8) is the upper case version of U+2C65 (3 bytes in UTF-8);
      a sequence of 3 of the former uses 6 bytes, as does a sequence of two of
      the latter. It is important, therefore, to check the length along the
      reference, not along the subject (earlier code did this wrong). */
      USPTR endptr = p + length;
      while (p < endptr)
        {
-Line 209 
 if (caseless)
+Line 210 
 if (caseless)
    /* The same code works when not in UTF-8 mode and in UTF-8 mode when there
    is no UCP support. */
      {
      if (eptr + length > md->end_subject) return -1;
      while (length-- > 0)
        { if (md->lcc[*p++] != md->lcc[*eptr++]) return -1; }
      }
    }
  /* In the caseful case, we can just compare the bytes, whether or not we
  are in UTF-8 mode. */
  else
    {
    if (eptr + length > md->end_subject) return -1;
    while (length-- > 0) if (*p++ != *eptr++) return -1;
    }
  return eptr - eptr_start;
-Line 276 
 enum { RM1=1, RM2,  RM3,  RM4,  RM5,  RM
+Line 277 
 enum { RM1=1, RM2,  RM3,  RM4,  RM5,  RM
         RM31,  RM32, RM33, RM34, RM35, RM36, RM37, RM38, RM39, RM40,
         RM41,  RM42, RM43, RM44, RM45, RM46, RM47, RM48, RM49, RM50,
         RM51,  RM52, RM53, RM54, RM55, RM56, RM57, RM58, RM59, RM60,
-        RM61,  RM62, RM63, RM64 };
+        RM61,  RM62, RM63 };
  /* These versions of the macros use the stack, as normal. There are debugging
  versions and production versions. Note that the "rw" argument of RMATCH isn't
-Line 383 
 typedef struct heapframe {
+Line 384 
 typedef struct heapframe {
    int Xprop_type;
    int Xprop_value;
    int Xprop_fail_result;
-   int Xprop_category;
-   int Xprop_chartype;
-   int Xprop_script;
    int Xoclength;
    uschar Xocchars[8];
  #endif
-Line 477 
 Returns:       MATCH_MATCH if matched
+Line 475 
 Returns:       MATCH_MATCH if matched
  static int
  match(REGISTER USPTR eptr, REGISTER const uschar *ecode, USPTR mstart,
    const uschar *markptr, int offset_top, match_data *md, eptrblock *eptrb,
    unsigned int rdepth)
  {
  /* These variables do not need to be preserved over recursion in this function,
-Line 550 
 HEAP_RECURSE:
+Line 548 
 HEAP_RECURSE:
  #define prop_type          frame->Xprop_type
  #define prop_value         frame->Xprop_value
  #define prop_fail_result   frame->Xprop_fail_result
- #define prop_category      frame->Xprop_category
- #define prop_chartype      frame->Xprop_chartype
- #define prop_script        frame->Xprop_script
  #define oclength           frame->Xoclength
  #define occhars            frame->Xocchars
  #endif
-Line 590 
 declarations can be cut out in a block.
+Line 585 
 declarations can be cut out in a block.
  below are for variables that do not have to be preserved over a recursive call
  to RMATCH(). */
  #ifdef SUPPORT_UTF8
  const uschar *charptr;
  #endif
  const uschar *callpat;
  const uschar *data;
  const uschar *next;
  USPTR         pp;
  const uschar *prev;
  USPTR         saved_eptr;
  recursion_info new_recursive;
  BOOL cur_is_word;
  BOOL condition;
  BOOL prev_is_word;
-Line 610 
 BOOL prev_is_word;
+Line 605 
 BOOL prev_is_word;
  int prop_type;
  int prop_value;
  int prop_fail_result;
- int prop_category;
- int prop_chartype;
- int prop_script;
  int oclength;
  uschar occhars[8];
  #endif
-Line 632 
 int stacksave[REC_STACK_SAVE_MAX];
+Line 624 
 int stacksave[REC_STACK_SAVE_MAX];
  eptrblock newptrb;
  #endif     /* NO_RECURSE */
  /* To save space on the stack and in the heap frame, I have doubled up on some
  of the local variables that are used only in localised parts of the code, but
  still need to be preserved over recursive calls of match(). These macros define
  the alternative names that are used. */
  #define allow_zero    cur_is_word
-Line 680 
 if (md->match_call_count++ >= md->match_
+Line 672 
 if (md->match_call_count++ >= md->match_
  if (rdepth >= md->match_limit_recursion) RRETURN(PCRE_ERROR_RECURSIONLIMIT);
  /* At the start of a group with an unlimited repeat that may match an empty
  string, the variable md->match_function_type is set to MATCH_CBEGROUP. It is
  done this way to save having to use another function argument, which would take
  up space on the stack. See also MATCH_CONDASSERT below.
  When MATCH_CBEGROUP is set, add the current subject pointer to the chain of
-Line 705 
 for (;;)
+Line 697 
 for (;;)
    {
    minimize = possessive = FALSE;
    op = *ecode;
    switch(op)
      {
      case OP_MARK:
-Line 808 
 for (;;)
+Line 800 
 for (;;)
      subject position in the working slot at the top of the vector. We mustn't
      change the current values of the data slot, because they may be set from a
      previous iteration of this group, and be referred to by a reference inside
-     the group. If we fail to match, we need to restore this value and also the
+     the group. A failure to match might occur after the group has succeeded,
-     values of the final offsets, in case they were set by a previous iteration
+     if something later on doesn't match. For this reason, we need to restore
-     of the same bracket.
+     the working value and also the values of the final offsets, in case they
+     were set by a previous iteration of the same bracket.
      If there isn't enough space in the offset vector, treat this as if it were
      a non-capturing bracket. Don't worry about setting the flag for the error
-Line 820 
 for (;;)
+Line 813 
 for (;;)
      case OP_SCBRA:
      number = GET2(ecode, 1+LINK_SIZE);
      offset = number << 1;
  #ifdef PCRE_DEBUG
      printf("start bracket %d\n", number);
      printf("subject=");
-Line 841 
 for (;;)
+Line 834 
 for (;;)
        for (;;)
          {
          if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
          RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
            eptrb, RM1);
+         if (rrc == MATCH_ONCE) break;  /* Backing up through an atomic group */
          if (rrc != MATCH_NOMATCH &&
              (rrc != MATCH_THEN || md->start_match_ptr != ecode))
            RRETURN(rrc);
          md->capture_last = save_capture_last;
          ecode += GET(ecode, 1);
          if (*ecode != OP_ALT) break;
          }
        DPRINTF(("bracket %d failed\n", number));
        md->offset_vector[offset] = save_offset1;
        md->offset_vector[offset+1] = save_offset2;
        md->offset_vector[md->offset_end - number] = save_offset3;
-       if (rrc != MATCH_THEN) md->mark = markptr;
+       /* At this point, rrc will be one of MATCH_ONCE, MATCH_NOMATCH, or
-       RRETURN(MATCH_NOMATCH);
+       MATCH_THEN. */
+       if (rrc != MATCH_THEN && md->mark == NULL) md->mark = markptr;
+       RRETURN(((rrc == MATCH_ONCE)? MATCH_ONCE:MATCH_NOMATCH));
        }
      /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
-Line 873 
 for (;;)
+Line 869 
 for (;;)
      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
-     /* Non-capturing bracket, except for possessive with unlimited repeat. Loop
+     /* Non-capturing or atomic group, except for possessive with unlimited
-     for all the alternatives. When we get to the final alternative within the
+     repeat. Loop for all the alternatives. When we get to the final alternative
-     brackets, we would return the result of a recursive call to match()
+     within the brackets, we used to return the result of a recursive call to
-     whatever happened. We can reduce stack usage by turning this into a tail
+     match() whatever happened so it was possible to reduce stack usage by
-     recursion, except in the case of a possibly empty group.*/
+     turning this into a tail recursion, except in the case of a possibly empty
+     group. However, now that there is the possiblity of (*THEN) occurring in
+     the final alternative, this optimization is no longer possible.
+     MATCH_ONCE is returned when the end of an atomic group is successfully
+     reached, but subsequent matching fails. It passes back up the tree (causing
+     captured values to be reset) until the original atomic group level is
+     reached. This is tested by comparing md->once_target with the start of the
+     group. At this point, the return is converted into MATCH_NOMATCH so that
+     previous backup points can be taken. */
+     case OP_ONCE:
      case OP_BRA:
      case OP_SBRA:
      DPRINTF(("start non-capturing bracket\n"));
      for (;;)
        {
-       if (ecode[GET(ecode, 1)] != OP_ALT)   /* Final alternative */
+       if (op >= OP_SBRA || op == OP_ONCE) md->match_function_type = MATCH_CBEGROUP;
-         {
+       RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md, eptrb,
-         if (op >= OP_SBRA)   /* Possibly empty group */
-           {
-           md->match_function_type = MATCH_CBEGROUP;
-           RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md, eptrb,
-             RM48);
-           if (rrc == MATCH_NOMATCH) md->mark = markptr;
-           RRETURN(rrc);
-           }
-         /* Not a possibly empty group; use tail recursion */
-         ecode += _pcre_OP_lengths[*ecode];
-         DPRINTF(("bracket 0 tail recursion\n"));
-         goto TAIL_RECURSE;
-         }
-       /* For non-final alternatives, continue the loop for a NOMATCH result;
-       otherwise return. */
-       if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
-       RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md, eptrb,
          RM2);
        if (rrc != MATCH_NOMATCH &&
            (rrc != MATCH_THEN || md->start_match_ptr != ecode))
+         {
+         if (rrc == MATCH_ONCE)
+           {
+           const uschar *scode = ecode;
+           if (*scode != OP_ONCE)           /* If not at start, find it */
+             {
+             while (*scode == OP_ALT) scode += GET(scode, 1);
+             scode -= GET(scode, 1);
+             }
+           if (md->once_target == scode) rrc = MATCH_NOMATCH;
+           }
          RRETURN(rrc);
+         }
        ecode += GET(ecode, 1);
+       if (*ecode != OP_ALT) break;
        }
-     /* Control never reaches here. */
+     if (rrc != MATCH_THEN && md->mark == NULL) md->mark = markptr;
+     RRETURN(MATCH_NOMATCH);
      /* Handle possessive capturing brackets with an unlimited repeat. We come
      here from BRAZERO with allow_zero set TRUE. The offset_vector values are
      handled similarly to the normal case above. However, the matching is
      different. The end of these brackets will always be OP_KETRPOS, which
      returns MATCH_KETRPOS without going further in the pattern. By this means
      we can handle the group by iteration rather than recursion, thereby
      reducing the amount of stack needed. */
      case OP_CBRAPOS:
      case OP_SCBRAPOS:
      allow_zero = FALSE;
      POSSESSIVE_CAPTURE:
      number = GET2(ecode, 1+LINK_SIZE);
      offset = number << 1;
-Line 939 
 for (;;)
+Line 941 
 for (;;)
      if (offset < md->offset_max)
        {
        matched_once = FALSE;
        code_offset = ecode - md->start_code;
        save_offset1 = md->offset_vector[offset];
        save_offset2 = md->offset_vector[offset+1];
-Line 947 
 for (;;)
+Line 949 
 for (;;)
        save_capture_last = md->capture_last;
        DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
        /* Each time round the loop, save the current subject position for use
        when the group matches. For MATCH_MATCH, the group has matched, so we
        restart it with a new subject starting position, remembering that we had
        at least one match. For MATCH_NOMATCH, carry on with the alternatives, as
        usual. If we haven't matched any alternatives in any iteration, check to
        see if a previous iteration matched. If so, the group has matched;
        continue from afterwards. Otherwise it has failed; restore the previous
        capture values before returning NOMATCH. */
        for (;;)
          {
          md->offset_vector[md->offset_end - number] =
            (int)(eptr - md->start_subject);
          if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
          RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
            eptrb, RM63);
          if (rrc == MATCH_KETRPOS)
            {
            offset_top = md->end_offset_top;
            eptr = md->end_match_ptr;
            ecode = md->start_code + code_offset;
            save_capture_last = md->capture_last;
            matched_once = TRUE;
            continue;
            }
          if (rrc != MATCH_NOMATCH &&
              (rrc != MATCH_THEN || md->start_match_ptr != ecode))
            RRETURN(rrc);
          md->capture_last = save_capture_last;
          ecode += GET(ecode, 1);
          if (*ecode != OP_ALT) break;
          }
        if (!matched_once)
          {
          md->offset_vector[offset] = save_offset1;
          md->offset_vector[offset+1] = save_offset2;
          md->offset_vector[md->offset_end - number] = save_offset3;
          }
-       if (rrc != MATCH_THEN) md->mark = markptr;
+       if (rrc != MATCH_THEN && md->mark == NULL) md->mark = markptr;
        if (allow_zero || matched_once)
          {
          ecode += 1 + LINK_SIZE;
          break;
          }
        RRETURN(MATCH_NOMATCH);
        }
      /* FALL THROUGH ... Insufficient room for saving captured contents. Treat
      as a non-capturing bracket. */
-Line 1009 
 for (;;)
+Line 1011 
 for (;;)
      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
      /* VVVVVVVVVVVVVVVVVVVVVVVVV */
      /* Non-capturing possessive bracket with unlimited repeat. We come here
      from BRAZERO with allow_zero = TRUE. The code is similar to the above,
      without the capturing complication. It is written out separately for speed
      and cleanliness. */
      case OP_BRAPOS:
      case OP_SBRAPOS:
      allow_zero = FALSE;
      POSSESSIVE_NON_CAPTURE:
      matched_once = FALSE;
      code_offset = ecode - md->start_code;
      for (;;)
        {
        if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
        RMATCH(eptr, ecode + _pcre_OP_lengths[*ecode], offset_top, md,
-         eptrb, RM64);
+         eptrb, RM48);
        if (rrc == MATCH_KETRPOS)
          {
+         offset_top = md->end_offset_top;
          eptr = md->end_match_ptr;
          ecode = md->start_code + code_offset;
          matched_once = TRUE;
          continue;
          }
        if (rrc != MATCH_NOMATCH &&
            (rrc != MATCH_THEN || md->start_match_ptr != ecode))
          RRETURN(rrc);
        ecode += GET(ecode, 1);
        if (*ecode != OP_ALT) break;
        }
      if (matched_once || allow_zero)
        {
        ecode += 1 + LINK_SIZE;
        break;
        }
      RRETURN(MATCH_NOMATCH);
      /* Control never reaches here. */
-Line 1053 
 for (;;)
+Line 1056 
 for (;;)
      /* Conditional group: compilation checked that there are no more than
      two branches. If the condition is false, skipping the first branch takes us
      past the end if there is only one branch, but that's OK because that is
-     exactly what going to the ket would do. As there is only one branch to be
+     exactly what going to the ket would do. */
-     obeyed, we can use tail recursion to avoid using another stack frame. */
      case OP_COND:
      case OP_SCOND:
-Line 1068 
 for (;;)
+Line 1070 
 for (;;)
        if (pcre_callout != NULL)
          {
          pcre_callout_block cb;
-         cb.version          = 1;   /* Version 1 of the callout block */
+         cb.version          = 2;   /* Version 1 of the callout block */
          cb.callout_number   = ecode[LINK_SIZE+2];
          cb.offset_vector    = md->offset_vector;
          cb.subject          = (PCRE_SPTR)md->start_subject;
-Line 1080 
 for (;;)
+Line 1082 
 for (;;)
          cb.capture_top      = offset_top/2;
          cb.capture_last     = md->capture_last;
          cb.callout_data     = md->callout_data;
+         cb.mark             = markptr;
          if ((rrc = (*pcre_callout)(&cb)) > 0) MRRETURN(MATCH_NOMATCH);
          if (rrc < 0) RRETURN(rrc);
          }
-Line 1238 
 for (;;)
+Line 1241 
 for (;;)
      else
        {
        md->match_function_type = MATCH_CONDASSERT;
        RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM3);
        if (rrc == MATCH_MATCH)
          {
+         if (md->end_offset_top > offset_top)
+           offset_top = md->end_offset_top;  /* Captures may have happened */
          condition = TRUE;
          ecode += 1 + LINK_SIZE + GET(ecode, LINK_SIZE + 2);
          while (*ecode == OP_ALT) ecode += GET(ecode, 1);
-Line 1259 
 for (;;)
+Line 1264 
 for (;;)
        }
      /* We are now at the branch that is to be obeyed. As there is only one,
-     we can use tail recursion to avoid using another stack frame, except when
+     we used to use tail recursion to avoid using another stack frame, except
-     we have an unlimited repeat of a possibly empty group. If the second
+     when there was unlimited repeat of a possibly empty group. However, that
-     alternative doesn't exist, we can just plough on. */
+     strategy no longer works because of the possibilty of (*THEN) being
+     encountered in the branch. A recursive call to match() is always required,
+     unless the second alternative doesn't exist, in which case we can just
+     plough on. */
      if (condition || *ecode == OP_ALT)
        {
-       ecode += 1 + LINK_SIZE;
+       if (op == OP_SCOND) md->match_function_type = MATCH_CBEGROUP;
-       if (op == OP_SCOND)        /* Possibly empty group */
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM49);
-         {
+       if (rrc == MATCH_THEN && md->start_match_ptr == ecode)
-         md->match_function_type = MATCH_CBEGROUP;
+         rrc = MATCH_NOMATCH;
-         RMATCH(eptr, ecode, offset_top, md, eptrb, RM49);
+       RRETURN(rrc);
-         RRETURN(rrc);
-         }
-       else goto TAIL_RECURSE;
        }
      else                         /* Condition false & no alternative */
        {
-Line 1305 
 for (;;)
+Line 1310 
 for (;;)
      break;
-     /* End of the pattern, either real or forced. If we are in a top-level
+     /* End of the pattern, either real or forced. */
-     recursion, we should restore the offsets appropriately and continue from
-     after the call. */
-     case OP_ACCEPT:
      case OP_END:
-     if (md->recursive != NULL && md->recursive->group_num == 0)
+     case OP_ACCEPT:
-       {
+     case OP_ASSERT_ACCEPT:
-       recursion_info *rec = md->recursive;
-       DPRINTF(("End of pattern in a (?0) recursion\n"));
-       md->recursive = rec->prevrec;
-       memmove(md->offset_vector, rec->offset_save,
-         rec->saved_max * sizeof(int));
-       offset_top = rec->save_offset_top;
-       ecode = rec->after_call;
-       break;
-       }
-     /* Otherwise, if we have matched an empty string, fail if PCRE_NOTEMPTY is
+     /* If we have matched an empty string, fail if not in an assertion and not
-     set, or if PCRE_NOTEMPTY_ATSTART is set and we have matched at the start of
+     in a recursion if either PCRE_NOTEMPTY is set, or if PCRE_NOTEMPTY_ATSTART
-     the subject. In both cases, backtracking will then try other alternatives,
+     is set and we have matched at the start of the subject. In both cases,
-     if any. */
+     backtracking will then try other alternatives, if any. */
-     if (eptr == mstart &&
+     if (eptr == mstart && op != OP_ASSERT_ACCEPT &&
-         (md->notempty ||
+          md->recursive == NULL &&
-           (md->notempty_atstart &&
+          (md->notempty ||
-             mstart == md->start_subject + md->start_offset)))
+            (md->notempty_atstart &&
+              mstart == md->start_subject + md->start_offset)))
        MRRETURN(MATCH_NOMATCH);
      /* Otherwise, we have a match. */
-Line 1350 
 for (;;)
+Line 1344 
 for (;;)
      matching won't pass the KET for an assertion. If any one branch matches,
      the assertion is true. Lookbehind assertions have an OP_REVERSE item at the
      start of each branch to move the current point backwards, so the code at
      this level is identical to the lookahead case. When the assertion is part
      of a condition, we want to return immediately afterwards. The caller of
      this incarnation of the match() function will have set MATCH_CONDASSERT in
      md->match_function type, and one of these opcodes will be the first opcode
      that is processed. We use a local variable that is preserved over calls to
      match() to remember this case. */
      case OP_ASSERT:
-Line 1364 
 for (;;)
+Line 1358 
 for (;;)
        condassert = TRUE;
        md->match_function_type = 0;
        }
      else condassert = FALSE;
      do
        {
        RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, NULL, RM4);
        if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT)
          {
          mstart = md->start_match_ptr;   /* In case \K reset it */
+         markptr = md->mark;
          break;
          }
        if (rrc != MATCH_NOMATCH &&
-Line 1380 
 for (;;)
+Line 1375 
 for (;;)
        ecode += GET(ecode, 1);
        }
      while (*ecode == OP_ALT);
      if (*ecode == OP_KET) MRRETURN(MATCH_NOMATCH);
      /* If checking an assertion for a condition, return MATCH_MATCH. */
-Line 1406 
 for (;;)
+Line 1401 
 for (;;)
        condassert = TRUE;
        md->match_function_type = 0;
        }
      else condassert = FALSE;
      do
        {
-Line 1425 
 for (;;)
+Line 1420 
 for (;;)
      while (*ecode == OP_ALT);
      if (condassert) RRETURN(MATCH_MATCH);  /* Condition assertion */
      ecode += 1 + LINK_SIZE;
      continue;
-Line 1470 
 for (;;)
+Line 1465 
 for (;;)
      if (pcre_callout != NULL)
        {
        pcre_callout_block cb;
-       cb.version          = 1;   /* Version 1 of the callout block */
+       cb.version          = 2;   /* Version 1 of the callout block */
        cb.callout_number   = ecode[1];
        cb.offset_vector    = md->offset_vector;
        cb.subject          = (PCRE_SPTR)md->start_subject;
-Line 1482 
 for (;;)
+Line 1477 
 for (;;)
        cb.capture_top      = offset_top/2;
        cb.capture_last     = md->capture_last;
        cb.callout_data     = md->callout_data;
+       cb.mark             = markptr;
        if ((rrc = (*pcre_callout)(&cb)) > 0) MRRETURN(MATCH_NOMATCH);
        if (rrc < 0) RRETURN(rrc);
        }
-Line 1492 
 for (;;)
+Line 1488 
 for (;;)
      offset data is the offset to the starting bracket from the start of the
      whole pattern. (This is so that it works from duplicated subpatterns.)
-     If there are any capturing brackets started but not finished, we have to
+     The state of the capturing groups is preserved over recursion, and
-     save their starting points and reinstate them after the recursion. However,
+     re-instated afterwards. We don't know how many are started and not yet
-     we don't know how many such there are (offset_top records the completed
+     finished (offset_top records the completed total) so we just have to save
-     total) so we just have to save all the potential data. There may be up to
+     all the potential data. There may be up to 65535 such values, which is too
-such values, which is too large to put on the stack, but using malloc
+     large to put on the stack, but using malloc for small numbers seems
-     for small numbers seems expensive. As a compromise, the stack is used when
+     expensive. As a compromise, the stack is used when there are no more than
-     there are no more than REC_STACK_SAVE_MAX values to store; otherwise malloc
+     REC_STACK_SAVE_MAX values to store; otherwise malloc is used.
-     is used. A problem is what to do if the malloc fails ... there is no way of
-     returning to the top level with an error. Save the top REC_STACK_SAVE_MAX
-     values on the stack, and accept that the rest may be wrong.
      There are also other values that have to be saved. We use a chained
      sequence of blocks that actually live on the stack. Thanks to Robin Houston
-     for the original version of this logic. */
+     for the original version of this logic. It has, however, been hacked around
+     a lot, so he is not to blame for the current way it works. */
      case OP_RECURSE:
        {
+       recursion_info *ri;
+       int recno;
        callpat = md->start_code + GET(ecode, 1);
-       new_recursive.group_num = (callpat == md->start_code)? 0 :
+       recno = (callpat == md->start_code)? 0 :
          GET2(callpat, 1 + LINK_SIZE);
+       /* Check for repeating a recursion without advancing the subject pointer.
+       This should catch convoluted mutual recursions. (Some simple cases are
+       caught at compile time.) */
+       for (ri = md->recursive; ri != NULL; ri = ri->prevrec)
+         if (recno == ri->group_num && eptr == ri->subject_position)
+           RRETURN(PCRE_ERROR_RECURSELOOP);
        /* Add to "recursing stack" */
+       new_recursive.group_num = recno;
+       new_recursive.subject_position = eptr;
        new_recursive.prevrec = md->recursive;
        md->recursive = &new_recursive;
-       /* Find where to continue from afterwards */
+       /* Where to continue from afterwards */
        ecode += 1 + LINK_SIZE;
-       new_recursive.after_call = ecode;
-       /* Now save the offset data. */
+       /* Now save the offset data */
        new_recursive.saved_max = md->offset_end;
        if (new_recursive.saved_max <= REC_STACK_SAVE_MAX)
-Line 1534 
 for (;;)
+Line 1540 
 for (;;)
            (int *)(pcre_malloc)(new_recursive.saved_max * sizeof(int));
          if (new_recursive.offset_save == NULL) RRETURN(PCRE_ERROR_NOMEMORY);
          }
        memcpy(new_recursive.offset_save, md->offset_vector,
              new_recursive.saved_max * sizeof(int));
-       new_recursive.save_offset_top = offset_top;
-       /* OK, now we can do the recursion. For each top-level alternative we
+       /* OK, now we can do the recursion. After processing each alternative,
-       restore the offset and recursion data. */
+       restore the offset data. If there were nested recursions, md->recursive
+       might be changed, so reset it before looping. */
        DPRINTF(("Recursing into group %d\n", new_recursive.group_num));
        cbegroup = (*callpat >= OP_SBRA);
-Line 1549 
 for (;;)
+Line 1554 
 for (;;)
          if (cbegroup) md->match_function_type = MATCH_CBEGROUP;
          RMATCH(eptr, callpat + _pcre_OP_lengths[*callpat], offset_top,
            md, eptrb, RM6);
+         memcpy(md->offset_vector, new_recursive.offset_save,
+             new_recursive.saved_max * sizeof(int));
          if (rrc == MATCH_MATCH || rrc == MATCH_ACCEPT)
            {
            DPRINTF(("Recursion matched\n"));
            md->recursive = new_recursive.prevrec;
            if (new_recursive.offset_save != stacksave)
              (pcre_free)(new_recursive.offset_save);
-           MRRETURN(MATCH_MATCH);
+           /* Set where we got to in the subject, and reset the start in case
+           it was changed by \K. This *is* propagated back out of a recursion,
+           for Perl compatibility. */
+           eptr = md->end_match_ptr;
+           mstart = md->start_match_ptr;
+           goto RECURSION_MATCHED;        /* Exit loop; end processing */
            }
          else if (rrc != MATCH_NOMATCH &&
                  (rrc != MATCH_THEN || md->start_match_ptr != ecode))
-Line 1567 
 for (;;)
+Line 1581 
 for (;;)
            }
          md->recursive = &new_recursive;
-         memcpy(md->offset_vector, new_recursive.offset_save,
-             new_recursive.saved_max * sizeof(int));
          callpat += GET(callpat, 1);
          }
        while (*callpat == OP_ALT);
-Line 1579 
 for (;;)
+Line 1591 
 for (;;)
          (pcre_free)(new_recursive.offset_save);
        MRRETURN(MATCH_NOMATCH);
        }
-     /* Control never reaches here */
-     /* "Once" brackets are like assertion brackets except that after a match,
-     the point in the subject string is not moved back. Thus there can never be
-     a move back into the brackets. Friedl calls these "atomic" subpatterns.
-     Check the alternative branches in turn - the matching won't pass the KET
-     for this kind of subpattern. If any one branch matches, we carry on as at
-     the end of a normal bracket, leaving the subject pointer, but resetting
-     the start-of-match value in case it was changed by \K. */
-     case OP_ONCE:
-     prev = ecode;
-     saved_eptr = eptr;
-     do
-       {
-       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM7);
-       if (rrc == MATCH_MATCH)  /* Note: _not_ MATCH_ACCEPT */
-         {
-         mstart = md->start_match_ptr;
-         break;
-         }
-       if (rrc != MATCH_NOMATCH &&
-           (rrc != MATCH_THEN || md->start_match_ptr != ecode))
-         RRETURN(rrc);
-       ecode += GET(ecode,1);
-       }
-     while (*ecode == OP_ALT);
-     /* If hit the end of the group (which could be repeated), fail */
-     if (*ecode != OP_ONCE && *ecode != OP_ALT) RRETURN(MATCH_NOMATCH);
-     /* Continue as from after the assertion, updating the offsets high water
+     RECURSION_MATCHED:
-     mark, since extracts may have been taken. */
+     break;
-     do ecode += GET(ecode, 1); while (*ecode == OP_ALT);
-     offset_top = md->end_offset_top;
-     eptr = md->end_match_ptr;
-     /* For a non-repeating ket, just continue at this level. This also
-     happens for a repeating ket if no characters were matched in the group.
-     This is the forcible breaking of infinite loops as implemented in Perl
-.005. If there is an options reset, it will get obeyed in the normal
-     course of events. */
-     if (*ecode == OP_KET || eptr == saved_eptr)
-       {
-       ecode += 1+LINK_SIZE;
-       break;
-       }
-     /* The repeating kets try the rest of the pattern or restart from the
-     preceding bracket, in the appropriate order. The second "call" of match()
-     uses tail recursion, to avoid using another stack frame. */
-     if (*ecode == OP_KETRMIN)
-       {
-       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM8);
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-       ecode = prev;
-       goto TAIL_RECURSE;
-       }
-     else  /* OP_KETRMAX */
-       {
-       md->match_function_type = MATCH_CBEGROUP;
-       RMATCH(eptr, prev, offset_top, md, eptrb, RM9);
-       if (rrc != MATCH_NOMATCH) RRETURN(rrc);
-       ecode += 1 + LINK_SIZE;
-       goto TAIL_RECURSE;
-       }
-     /* Control never gets here */
      /* An alternation is the end of a branch; scan along to find the end of the
      bracketed group and go to there. */
-Line 1665 
 for (;;)
+Line 1607 
 for (;;)
      at all - i.e. it could be ()* or ()? or even (){0} in the pattern. Brackets
      with fixed upper repeat limits are compiled as a number of copies, with the
      optional ones preceded by BRAZERO or BRAMINZERO. */
      case OP_BRAZERO:
      next = ecode + 1;
      RMATCH(eptr, next, offset_top, md, eptrb, RM10);
-Line 1673 
 for (;;)
+Line 1615 
 for (;;)
      do next += GET(next, 1); while (*next == OP_ALT);
      ecode = next + 1 + LINK_SIZE;
      break;
      case OP_BRAMINZERO:
      next = ecode + 1;
      do next += GET(next, 1); while (*next == OP_ALT);
-Line 1687 
 for (;;)
+Line 1629 
 for (;;)
      do next += GET(next,1); while (*next == OP_ALT);
      ecode = next + 1 + LINK_SIZE;
      break;
      /* BRAPOSZERO occurs before a possessive bracket group. Don't do anything
      here; just jump to the group, with allow_zero set TRUE. */
      case OP_BRAPOSZERO:
      op = *(++ecode);
      allow_zero = TRUE;
      if (op == OP_CBRAPOS || op == OP_SCBRAPOS) goto POSSESSIVE_CAPTURE;
        goto POSSESSIVE_NON_CAPTURE;
-Line 1702 
 for (;;)
+Line 1644 
 for (;;)
      case OP_KET:
      case OP_KETRMIN:
      case OP_KETRMAX:
      case OP_KETRPOS:
      prev = ecode - GET(ecode, 1);
      /* If this was a group that remembered the subject start, in order to break
      infinite repeats of empty string matches, retrieve the subject start from
      the chain. Otherwise, set it NULL. */
-     if (*prev >= OP_SBRA)
+     if (*prev >= OP_SBRA || *prev == OP_ONCE)
        {
        saved_eptr = eptrb->epb_saved_eptr;   /* Value at start of group */
        eptrb = eptrb->epb_prev;              /* Backup to previous group */
        }
      else saved_eptr = NULL;
-     /* If we are at the end of an assertion group or an atomic group, stop
+     /* If we are at the end of an assertion group, stop matching and return
-     matching and return MATCH_MATCH, but record the current high water mark for
+     MATCH_MATCH, but record the current high water mark for use by positive
-     use by positive assertions. We also need to record the match start in case
+     assertions. We also need to record the match start in case it was changed
-     it was changed by \K. */
+     by \K. */
      if (*prev == OP_ASSERT || *prev == OP_ASSERT_NOT ||
-         *prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT ||
+         *prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT)
-         *prev == OP_ONCE)
        {
        md->end_match_ptr = eptr;      /* For ONCE */
        md->end_offset_top = offset_top;
        md->start_match_ptr = mstart;
-       MRRETURN(MATCH_MATCH);
+       MRRETURN(MATCH_MATCH);         /* Sets md->mark */
        }
      /* For capturing groups we have to check the group number back at the start
      and if necessary complete handling an extraction by setting the offsets and
-     bumping the high water mark. Note that whole-pattern recursion is coded as
+     bumping the high water mark. Whole-pattern recursion is coded as a recurse
-     a recurse into group 0, so it won't be picked up here. Instead, we catch it
+     into group 0, so it won't be picked up here. Instead, we catch it when the
-     when the OP_END is reached. Other recursion is handled here. */
+     OP_END is reached. Other recursion is handled here. We just have to record
+     the current subject position and start match pointer and give a MATCH
+     return. */
      if (*prev == OP_CBRA || *prev == OP_SCBRA ||
          *prev == OP_CBRAPOS || *prev == OP_SCBRAPOS)
-Line 1748 
 for (;;)
+Line 1691 
 for (;;)
        printf("\n");
  #endif
+       /* Handle a recursively called group. */
+       if (md->recursive != NULL && md->recursive->group_num == number)
+         {
+         md->end_match_ptr = eptr;
+         md->start_match_ptr = mstart;
+         RRETURN(MATCH_MATCH);
+         }
+       /* Deal with capturing */
        md->capture_last = number;
        if (offset >= md->offset_max) md->offset_overflow = TRUE; else
          {
+         /* If offset is greater than offset_top, it means that we are
+         "skipping" a capturing group, and that group's offsets must be marked
+         unset. In earlier versions of PCRE, all the offsets were unset at the
+         start of matching, but this doesn't work because atomic groups and
+         assertions can cause a value to be set that should later be unset.
+         Example: matching /(?>(a))b|(a)c/ against "ac". This sets group 1 as
+         part of the atomic group, but this is not on the final matching path,
+         so must be unset when 2 is set. (If there is no group 2, there is no
+         problem, because offset_top will then be 2, indicating no capture.) */
+         if (offset > offset_top)
+           {
+           register int *iptr = md->offset_vector + offset_top;
+           register int *iend = md->offset_vector + offset;
+           while (iptr < iend) *iptr++ = -1;
+           }
+         /* Now make the extraction */
          md->offset_vector[offset] =
            md->offset_vector[md->offset_end - number];
          md->offset_vector[offset+1] = (int)(eptr - md->start_subject);
          if (offset_top <= offset) offset_top = offset + 2;
          }
-       /* Handle a recursively called group. Restore the offsets
-       appropriately and continue from after the call. */
-       if (md->recursive != NULL && md->recursive->group_num == number)
-         {
-         recursion_info *rec = md->recursive;
-         DPRINTF(("Recursion (%d) succeeded - continuing\n", number));
-         md->recursive = rec->prevrec;
-         memcpy(md->offset_vector, rec->offset_save,
-           rec->saved_max * sizeof(int));
-         offset_top = rec->save_offset_top;
-         ecode = rec->after_call;
-         break;
-         }
        }
-     /* For a non-repeating ket, just continue at this level. This also
+     /* For an ordinary non-repeating ket, just continue at this level. This
-     happens for a repeating ket if no characters were matched in the group.
+     also happens for a repeating ket if no characters were matched in the
-     This is the forcible breaking of infinite loops as implemented in Perl
+     group. This is the forcible breaking of infinite loops as implemented in
-.005. If there is an options reset, it will get obeyed in the normal
+     Perl 5.005. For a non-repeating atomic group, establish a backup point by
-     course of events. */
+     processing the rest of the pattern at a lower level. If this results in a
+     NOMATCH return, pass MATCH_ONCE back to the original OP_ONCE level, thereby
+     bypassing intermediate backup points, but resetting any captures that
+     happened along the way. */
      if (*ecode == OP_KET || eptr == saved_eptr)
        {
-       ecode += 1 + LINK_SIZE;
+       if (*prev == OP_ONCE)
+         {
+         RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM12);
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
+         RRETURN(MATCH_ONCE);
+         }
+       ecode += 1 + LINK_SIZE;    /* Carry on at this level */
        break;
        }
      /* OP_KETRPOS is a possessive repeating ket. Remember the current position,
      and return the MATCH_KETRPOS. This makes it possible to do the repeats one
      at a time from the outer level, thus saving stack. */
      if (*ecode == OP_KETRPOS)
        {
        md->end_match_ptr = eptr;
        md->end_offset_top = offset_top;
        RRETURN(MATCH_KETRPOS);
        }
      /* The normal repeating kets try the rest of the pattern or restart from
      the preceding bracket, in the appropriate order. In the second case, we can
      use tail recursion to avoid using another stack frame, unless we have an
-     unlimited repeat of a group that can match an empty string. */
+     an atomic group or an unlimited repeat of a group that can match an empty
+     string. */
      if (*ecode == OP_KETRMIN)
        {
-       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM12);
+       RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM7);
        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       if (*prev == OP_ONCE)
+         {
+         RMATCH(eptr, prev, offset_top, md, eptrb, RM8);
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         md->once_target = prev;  /* Level at which to change to MATCH_NOMATCH */
+         RRETURN(MATCH_ONCE);
+         }
        if (*prev >= OP_SBRA)    /* Could match an empty string */
          {
          md->match_function_type = MATCH_CBEGROUP;
          RMATCH(eptr, prev, offset_top, md, eptrb, RM50);
          RRETURN(rrc);
          }
-Line 1816 
 for (;;)
+Line 1792 
 for (;;)
        }
      else  /* OP_KETRMAX */
        {
        if (*prev >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
        RMATCH(eptr, prev, offset_top, md, eptrb, RM13);
+       if (rrc == MATCH_ONCE && md->once_target == prev) rrc = MATCH_NOMATCH;
        if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+       if (*prev == OP_ONCE)
+         {
+         RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM9);
+         if (rrc != MATCH_NOMATCH) RRETURN(rrc);
+         md->once_target = prev;
+         RRETURN(MATCH_ONCE);
+         }
        ecode += 1 + LINK_SIZE;
        goto TAIL_RECURSE;
        }
-Line 1828 
 for (;;)
+Line 1812 
 for (;;)
      case OP_CIRC:
      if (md->notbol && eptr == md->start_subject) MRRETURN(MATCH_NOMATCH);
      /* Start of subject assertion */
      case OP_SOD:
      if (eptr != md->start_subject) MRRETURN(MATCH_NOMATCH);
      ecode++;
      break;
      /* Multiline mode: start of subject unless notbol, or after any newline. */
      case OP_CIRCM:
-Line 1874 
 for (;;)
+Line 1858 
 for (;;)
      ecode++;
      break;
      /* Not multiline mode: assert before a terminating newline or before end of
      subject unless noteol is set. */
      case OP_DOLL:
-Line 2030 
 for (;;)
+Line 2014 
 for (;;)
      /* Fall through */
      case OP_ALLANY:
-     if (eptr++ >= md->end_subject)
+     if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
-       {
+       {                            /* not be updated before SCHECK_PARTIAL. */
        SCHECK_PARTIAL();
        MRRETURN(MATCH_NOMATCH);
        }
+     eptr++;
      if (utf8) while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
      ecode++;
      break;
-Line 2043 
 for (;;)
+Line 2028 
 for (;;)
      any byte, even newline, independent of the setting of PCRE_DOTALL. */
      case OP_ANYBYTE:
-     if (eptr++ >= md->end_subject)
+     if (eptr >= md->end_subject)   /* DO NOT merge the eptr++ here; it must */
-       {
+       {                            /* not be updated before SCHECK_PARTIAL. */
        SCHECK_PARTIAL();
        MRRETURN(MATCH_NOMATCH);
        }
+     eptr++;
      ecode++;
      break;
-Line 2163 
 for (;;)
+Line 2149 
 for (;;)
      switch(c)
        {
        default: MRRETURN(MATCH_NOMATCH);
        case 0x000d:
        if (eptr < md->end_subject && *eptr == 0x0a) eptr++;
        break;
-Line 2387 
 for (;;)
+Line 2373 
 for (;;)
        MRRETURN(MATCH_NOMATCH);
        }
      GETCHARINCTEST(c, eptr);
+     if (UCD_CATEGORY(c) == ucp_M) MRRETURN(MATCH_NOMATCH);
+     while (eptr < md->end_subject)
        {
-       int category = UCD_CATEGORY(c);
+       int len = 1;
-       if (category == ucp_M) MRRETURN(MATCH_NOMATCH);
+       if (!utf8) c = *eptr; else { GETCHARLEN(c, eptr, len); }
-       while (eptr < md->end_subject)
+       if (UCD_CATEGORY(c) != ucp_M) break;
-         {
+       eptr += len;
-         int len = 1;
-         if (!utf8) c = *eptr; else
-           {
-           GETCHARLEN(c, eptr, len);
-           }
-         category = UCD_CATEGORY(c);
-         if (category != ucp_M) break;
-         eptr += len;
-         }
        }
      ecode++;
      break;
-Line 2416 
 for (;;)
+Line 2395 
 for (;;)
      loops). */
      case OP_REF:
      case OP_REFI:
      caseless = op == OP_REFI;
      offset = GET2(ecode, 1) << 1;               /* Doubled ref number */
      ecode += 3;
-Line 2485 
 for (;;)
+Line 2464 
 for (;;)
      for (i = 1; i <= min; i++)
        {
        int slength;
        if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
          {
          CHECK_PARTIAL();
-Line 2505 
 for (;;)
+Line 2484 
 for (;;)
        {
        for (fi = min;; fi++)
          {
          int slength;
          RMATCH(eptr, ecode, offset_top, md, eptrb, RM14);
          if (rrc != MATCH_NOMATCH) RRETURN(rrc);
          if (fi >= max) MRRETURN(MATCH_NOMATCH);
-Line 2526 
 for (;;)
+Line 2505 
 for (;;)
        pp = eptr;
        for (i = min; i < max; i++)
          {
          int slength;
          if ((slength = match_ref(offset, eptr, length, md, caseless)) < 0)
            {
            CHECK_PARTIAL();
-Line 3259 
 for (;;)
+Line 3238 
 for (;;)
      checking can be multibyte. */
      case OP_NOT:
      case OP_NOTI:
      if (eptr >= md->end_subject)
        {
        SCHECK_PARTIAL();
-Line 3754 
 for (;;)
+Line 3733 
 for (;;)
            case PT_LAMP:
            for (i = 1; i <= min; i++)
              {
+             int chartype;
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_chartype = UCD_CHARTYPE(c);
+             chartype = UCD_CHARTYPE(c);
-             if ((prop_chartype == ucp_Lu ||
+             if ((chartype == ucp_Lu ||
-                  prop_chartype == ucp_Ll ||
+                  chartype == ucp_Ll ||
-                  prop_chartype == ucp_Lt) == prop_fail_result)
+                  chartype == ucp_Lt) == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            break;
-Line 3777 
 for (;;)
+Line 3757 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == prop_value) == prop_fail_result)
-             if ((prop_category == prop_value) == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            break;
-Line 3792 
 for (;;)
+Line 3771 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_chartype = UCD_CHARTYPE(c);
+             if ((UCD_CHARTYPE(c) == prop_value) == prop_fail_result)
-             if ((prop_chartype == prop_value) == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            break;
-Line 3807 
 for (;;)
+Line 3785 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_script = UCD_SCRIPT(c);
+             if ((UCD_SCRIPT(c) == prop_value) == prop_fail_result)
-             if ((prop_script == prop_value) == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            break;
-Line 3816 
 for (;;)
+Line 3793 
 for (;;)
            case PT_ALNUM:
            for (i = 1; i <= min; i++)
              {
+             int category;
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             category = UCD_CATEGORY(c);
-             if ((prop_category == ucp_L || prop_category == ucp_N)
+             if ((category == ucp_L || category == ucp_N) == prop_fail_result)
-                    == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            break;
-Line 3838 
 for (;;)
+Line 3815 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
-             if ((prop_category == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
                   c == CHAR_FF || c == CHAR_CR)
                     == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
-Line 3855 
 for (;;)
+Line 3831 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
-             if ((prop_category == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
                   c == CHAR_VT || c == CHAR_FF || c == CHAR_CR)
                     == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
-Line 3866 
 for (;;)
+Line 3841 
 for (;;)
            case PT_WORD:
            for (i = 1; i <= min; i++)
              {
+             int category;
              if (eptr >= md->end_subject)
                {
                SCHECK_PARTIAL();
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             category = UCD_CATEGORY(c);
-             if ((prop_category == ucp_L || prop_category == ucp_N ||
+             if ((category == ucp_L || category == ucp_N || c == CHAR_UNDERSCORE)
-                  c == CHAR_UNDERSCORE)
                     == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
-Line 3900 
 for (;;)
+Line 3875 
 for (;;)
              MRRETURN(MATCH_NOMATCH);
              }
            GETCHARINCTEST(c, eptr);
-           prop_category = UCD_CATEGORY(c);
+           if (UCD_CATEGORY(c) == ucp_M) MRRETURN(MATCH_NOMATCH);
-           if (prop_category == ucp_M) MRRETURN(MATCH_NOMATCH);
            while (eptr < md->end_subject)
              {
              int len = 1;
-             if (!utf8) c = *eptr;
+             if (!utf8) c = *eptr; else { GETCHARLEN(c, eptr, len); }
-               else { GETCHARLEN(c, eptr, len); }
+             if (UCD_CATEGORY(c) != ucp_M) break;
-             prop_category = UCD_CATEGORY(c);
-             if (prop_category != ucp_M) break;
              eptr += len;
              }
            }
-Line 3966 
 for (;;)
+Line 3938 
 for (;;)
            switch(c)
              {
              default: MRRETURN(MATCH_NOMATCH);
              case 0x000d:
              if (eptr < md->end_subject && *eptr == 0x0a) eptr++;
              break;
-Line 4243 
 for (;;)
+Line 4215 
 for (;;)
            switch(*eptr++)
              {
              default: MRRETURN(MATCH_NOMATCH);
              case 0x000d:
              if (eptr < md->end_subject && *eptr == 0x0a) eptr++;
              break;
              case 0x000a:
              break;
-Line 4453 
 for (;;)
+Line 4425 
 for (;;)
            case PT_LAMP:
            for (fi = min;; fi++)
              {
+             int chartype;
              RMATCH(eptr, ecode, offset_top, md, eptrb, RM37);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
-Line 4462 
 for (;;)
+Line 4435 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_chartype = UCD_CHARTYPE(c);
+             chartype = UCD_CHARTYPE(c);
-             if ((prop_chartype == ucp_Lu ||
+             if ((chartype == ucp_Lu ||
-                  prop_chartype == ucp_Ll ||
+                  chartype == ucp_Ll ||
-                  prop_chartype == ucp_Lt) == prop_fail_result)
+                  chartype == ucp_Lt) == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            /* Control never gets here */
-Line 4482 
 for (;;)
+Line 4455 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == prop_value) == prop_fail_result)
-             if ((prop_category == prop_value) == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            /* Control never gets here */
-Line 4500 
 for (;;)
+Line 4472 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_chartype = UCD_CHARTYPE(c);
+             if ((UCD_CHARTYPE(c) == prop_value) == prop_fail_result)
-             if ((prop_chartype == prop_value) == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            /* Control never gets here */
-Line 4518 
 for (;;)
+Line 4489 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_script = UCD_SCRIPT(c);
+             if ((UCD_SCRIPT(c) == prop_value) == prop_fail_result)
-             if ((prop_script == prop_value) == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            /* Control never gets here */
-Line 4527 
 for (;;)
+Line 4497 
 for (;;)
            case PT_ALNUM:
            for (fi = min;; fi++)
              {
+             int category;
              RMATCH(eptr, ecode, offset_top, md, eptrb, RM59);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
-Line 4536 
 for (;;)
+Line 4507 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             category = UCD_CATEGORY(c);
-             if ((prop_category == ucp_L || prop_category == ucp_N)
+             if ((category == ucp_L || category == ucp_N) == prop_fail_result)
-                    == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
              }
            /* Control never gets here */
-Line 4555 
 for (;;)
+Line 4525 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
-             if ((prop_category == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
                   c == CHAR_FF || c == CHAR_CR)
                     == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
-Line 4575 
 for (;;)
+Line 4544 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
-             if ((prop_category == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
                   c == CHAR_VT || c == CHAR_FF || c == CHAR_CR)
                     == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
-Line 4586 
 for (;;)
+Line 4554 
 for (;;)
            case PT_WORD:
            for (fi = min;; fi++)
              {
+             int category;
              RMATCH(eptr, ecode, offset_top, md, eptrb, RM62);
              if (rrc != MATCH_NOMATCH) RRETURN(rrc);
              if (fi >= max) MRRETURN(MATCH_NOMATCH);
-Line 4595 
 for (;;)
+Line 4564 
 for (;;)
                MRRETURN(MATCH_NOMATCH);
                }
              GETCHARINCTEST(c, eptr);
-             prop_category = UCD_CATEGORY(c);
+             category = UCD_CATEGORY(c);
-             if ((prop_category == ucp_L ||
+             if ((category == ucp_L ||
-                  prop_category == ucp_N ||
+                  category == ucp_N ||
                   c == CHAR_UNDERSCORE)
                     == prop_fail_result)
                MRRETURN(MATCH_NOMATCH);
-Line 4627 
 for (;;)
+Line 4596 
 for (;;)
              MRRETURN(MATCH_NOMATCH);
              }
            GETCHARINCTEST(c, eptr);
-           prop_category = UCD_CATEGORY(c);
+           if (UCD_CATEGORY(c) == ucp_M) MRRETURN(MATCH_NOMATCH);
-           if (prop_category == ucp_M) MRRETURN(MATCH_NOMATCH);
            while (eptr < md->end_subject)
              {
              int len = 1;
-             if (!utf8) c = *eptr;
+             if (!utf8) c = *eptr; else { GETCHARLEN(c, eptr, len); }
-               else { GETCHARLEN(c, eptr, len); }
+             if (UCD_CATEGORY(c) != ucp_M) break;
-             prop_category = UCD_CATEGORY(c);
-             if (prop_category != ucp_M) break;
              eptr += len;
              }
            }
          }
        else
  #endif     /* SUPPORT_UCP */
-Line 4961 
 for (;;)
+Line 4926 
 for (;;)
            case PT_LAMP:
            for (i = min; i < max; i++)
              {
+             int chartype;
              int len = 1;
              if (eptr >= md->end_subject)
                {
-Line 4968 
 for (;;)
+Line 4934 
 for (;;)
                break;
                }
              GETCHARLENTEST(c, eptr, len);
-             prop_chartype = UCD_CHARTYPE(c);
+             chartype = UCD_CHARTYPE(c);
-             if ((prop_chartype == ucp_Lu ||
+             if ((chartype == ucp_Lu ||
-                  prop_chartype == ucp_Ll ||
+                  chartype == ucp_Ll ||
-                  prop_chartype == ucp_Lt) == prop_fail_result)
+                  chartype == ucp_Lt) == prop_fail_result)
                break;
              eptr+= len;
              }
-Line 4987 
 for (;;)
+Line 4953 
 for (;;)
                break;
                }
              GETCHARLENTEST(c, eptr, len);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == prop_value) == prop_fail_result) break;
-             if ((prop_category == prop_value) == prop_fail_result)
-               break;
              eptr+= len;
              }
            break;
-Line 5004 
 for (;;)
+Line 4968 
 for (;;)
                break;
                }
              GETCHARLENTEST(c, eptr, len);
-             prop_chartype = UCD_CHARTYPE(c);
+             if ((UCD_CHARTYPE(c) == prop_value) == prop_fail_result) break;
-             if ((prop_chartype == prop_value) == prop_fail_result)
-               break;
              eptr+= len;
              }
            break;
-Line 5021 
 for (;;)
+Line 4983 
 for (;;)
                break;
                }
              GETCHARLENTEST(c, eptr, len);
-             prop_script = UCD_SCRIPT(c);
+             if ((UCD_SCRIPT(c) == prop_value) == prop_fail_result) break;
-             if ((prop_script == prop_value) == prop_fail_result)
-               break;
              eptr+= len;
              }
            break;
-Line 5031 
 for (;;)
+Line 4991 
 for (;;)
            case PT_ALNUM:
            for (i = min; i < max; i++)
              {
+             int category;
              int len = 1;
              if (eptr >= md->end_subject)
                {
-Line 5038 
 for (;;)
+Line 4999 
 for (;;)
                break;
                }
              GETCHARLENTEST(c, eptr, len);
-             prop_category = UCD_CATEGORY(c);
+             category = UCD_CATEGORY(c);
-             if ((prop_category == ucp_L || prop_category == ucp_N)
+             if ((category == ucp_L || category == ucp_N) == prop_fail_result)
-                  == prop_fail_result)
                break;
              eptr+= len;
              }
-Line 5056 
 for (;;)
+Line 5016 
 for (;;)
                break;
                }
              GETCHARLENTEST(c, eptr, len);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
-             if ((prop_category == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
                   c == CHAR_FF || c == CHAR_CR)
                   == prop_fail_result)
                break;
-Line 5075 
 for (;;)
+Line 5034 
 for (;;)
                break;
                }
              GETCHARLENTEST(c, eptr, len);
-             prop_category = UCD_CATEGORY(c);
+             if ((UCD_CATEGORY(c) == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
-             if ((prop_category == ucp_Z || c == CHAR_HT || c == CHAR_NL ||
                   c == CHAR_VT || c == CHAR_FF || c == CHAR_CR)
                   == prop_fail_result)
                break;
-Line 5087 
 for (;;)
+Line 5045 
 for (;;)
            case PT_WORD:
            for (i = min; i < max; i++)
              {
+             int category;
              int len = 1;
              if (eptr >= md->end_subject)
                {
-Line 5094 
 for (;;)
+Line 5053 
 for (;;)
                break;
                }
              GETCHARLENTEST(c, eptr, len);
-             prop_category = UCD_CATEGORY(c);
+             category = UCD_CATEGORY(c);
-             if ((prop_category == ucp_L || prop_category == ucp_N ||
+             if ((category == ucp_L || category == ucp_N ||
                   c == CHAR_UNDERSCORE) == prop_fail_result)
                break;
              eptr+= len;
-Line 5125 
 for (;;)
+Line 5084 
 for (;;)
          {
          for (i = min; i < max; i++)
            {
+           int len = 1;
            if (eptr >= md->end_subject)
              {
              SCHECK_PARTIAL();
              break;
              }
-           GETCHARINCTEST(c, eptr);
+           if (!utf8) c = *eptr; else { GETCHARLEN(c, eptr, len); }
-           prop_category = UCD_CATEGORY(c);
+           if (UCD_CATEGORY(c) == ucp_M) break;
-           if (prop_category == ucp_M) break;
+           eptr += len;
            while (eptr < md->end_subject)
              {
-             int len = 1;
+             len = 1;
-             if (!utf8) c = *eptr; else
+             if (!utf8) c = *eptr; else { GETCHARLEN(c, eptr, len); }
-               {
+             if (UCD_CATEGORY(c) != ucp_M) break;
-               GETCHARLEN(c, eptr, len);
-               }
-             prop_category = UCD_CATEGORY(c);
-             if (prop_category != ucp_M) break;
              eptr += len;
              }
            }
-Line 5157 
 for (;;)
+Line 5113 
 for (;;)
            if (eptr-- == pp) break;        /* Stop if tried at original pos */
            for (;;)                        /* Move back over one extended */
              {
-             int len = 1;
              if (!utf8) c = *eptr; else
                {
                BACKCHAR(eptr);
-               GETCHARLEN(c, eptr, len);
+               GETCHAR(c, eptr);
                }
-             prop_category = UCD_CATEGORY(c);
+             if (UCD_CATEGORY(c) != ucp_M) break;
-             if (prop_category != ucp_M) break;
              eptr--;
              }
            }
-Line 5228 
 for (;;)
+Line 5182 
 for (;;)
                while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
                }
              }
-           else eptr = md->end_subject;   /* Unlimited UTF-8 repeat */
+           else
+             {
+             eptr = md->end_subject;   /* Unlimited UTF-8 repeat */
+             SCHECK_PARTIAL();
+             }
            break;
            /* The byte case is the same as non-UTF8 */
-Line 5449 
 for (;;)
+Line 5407 
 for (;;)
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            if (eptr-- == pp) break;        /* Stop if tried at original pos */
            BACKCHAR(eptr);
            if (ctype == OP_ANYNL && eptr > pp  && *eptr == '\n' &&
                eptr[-1] == '\r') eptr--;
            }
          }
-Line 5661 
 for (;;)
+Line 5619 
 for (;;)
            RMATCH(eptr, ecode, offset_top, md, eptrb, RM47);
            if (rrc != MATCH_NOMATCH) RRETURN(rrc);
            eptr--;
            if (ctype == OP_ANYNL && eptr > pp  && *eptr == '\n' &&
                eptr[-1] == '\r') eptr--;
            }
          }
-Line 5701 
 switch (frame->Xwhere)
+Line 5659 
 switch (frame->Xwhere)
    LBL( 9) LBL(10) LBL(11) LBL(12) LBL(13) LBL(14) LBL(15) LBL(17)
    LBL(19) LBL(24) LBL(25) LBL(26) LBL(27) LBL(29) LBL(31) LBL(33)
    LBL(35) LBL(43) LBL(47) LBL(48) LBL(49) LBL(50) LBL(51) LBL(52)
-   LBL(53) LBL(54) LBL(55) LBL(56) LBL(57) LBL(58) LBL(63) LBL(64)
+   LBL(53) LBL(54) LBL(55) LBL(56) LBL(57) LBL(58) LBL(63)
  #ifdef SUPPORT_UTF8
    LBL(16) LBL(18) LBL(20) LBL(21) LBL(22) LBL(23) LBL(28) LBL(30)
    LBL(32) LBL(34) LBL(42) LBL(46)
-Line 5803 
 pcre_exec(const pcre *argument_re, const
+Line 5761 
 pcre_exec(const pcre *argument_re, const
    PCRE_SPTR subject, int length, int start_offset, int options, int *offsets,
    int offsetcount)
  {
- int rc, resetcount, ocount;
+ int rc, ocount, arg_offset_max;
  int first_byte = -1;
  int req_byte = -1;
  int req_byte2 = -1;
-Line 5839 
 if (re == NULL || subject == NULL ||
+Line 5797 
 if (re == NULL || subject == NULL ||
  if (offsetcount < 0) return PCRE_ERROR_BADCOUNT;
  if (start_offset < 0 || start_offset > length) return PCRE_ERROR_BADOFFSET;
- /* This information is for finding all the numbers associated with a given
+ /* These two settings are used in the code for checking a UTF-8 string that
- name, for condition testing. */
+ follows immediately afterwards. Other values in the md block are used only
+ during "normal" pcre_exec() processing, not when the JIT support is in use,
+ so they are set up later. */
+ utf8 = md->utf8 = (re->options & PCRE_UTF8) != 0;
+ md->partial = ((options & PCRE_PARTIAL_HARD) != 0)? 2 :
+               ((options & PCRE_PARTIAL_SOFT) != 0)? 1 : 0;
+ /* Check a UTF-8 string if required. Pass back the character offset and error
+ code for an invalid string if a results vector is available. */
+ #ifdef SUPPORT_UTF8
+ if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0)
+   {
+   int erroroffset;
+   int errorcode = _pcre_valid_utf8((USPTR)subject, length, &erroroffset);
+   if (errorcode != 0)
+     {
+     if (offsetcount >= 2)
+       {
+       offsets[0] = erroroffset;
+       offsets[1] = errorcode;
+       }
+     return (errorcode <= PCRE_UTF8_ERR5 && md->partial > 1)?
+       PCRE_ERROR_SHORTUTF8 : PCRE_ERROR_BADUTF8;
+     }
+   /* Check that a start_offset points to the start of a UTF-8 character. */
+   if (start_offset > 0 && start_offset < length &&
+       (((USPTR)subject)[start_offset] & 0xc0) == 0x80)
+     return PCRE_ERROR_BADUTF8_OFFSET;
+   }
+ #endif
+ /* If the pattern was successfully studied with JIT support, run the JIT
+ executable instead of the rest of this function. Most options must be set at
+ compile time for the JIT code to be usable. Fallback to the normal code path if
+ an unsupported flag is set. In particular, JIT does not support partial
+ matching. */
+ #ifdef SUPPORT_JIT
+ if (extra_data != NULL
+     && (extra_data->flags & PCRE_EXTRA_EXECUTABLE_JIT) != 0
+     && extra_data->executable_jit != NULL
+     && (options & ~(PCRE_NO_UTF8_CHECK | PCRE_NOTBOL | PCRE_NOTEOL |
+                     PCRE_NOTEMPTY | PCRE_NOTEMPTY_ATSTART)) == 0)
+   return _pcre_jit_exec(re, extra_data->executable_jit, subject, length,
+     start_offset, options, ((extra_data->flags & PCRE_EXTRA_MATCH_LIMIT) == 0)
+     ? MATCH_LIMIT : extra_data->match_limit, offsets, offsetcount);
+ #endif
+ /* Carry on with non-JIT matching. This information is for finding all the
+ numbers associated with a given name, for condition testing. */
  md->name_table = (uschar *)re + re->name_table_offset;
  md->name_count = re->name_count;
-Line 5907 
 md->end_subject = md->start_subject + le
+Line 5917 
 md->end_subject = md->start_subject + le
  end_subject = md->end_subject;
  md->endonly = (re->options & PCRE_DOLLAR_ENDONLY) != 0;
- utf8 = md->utf8 = (re->options & PCRE_UTF8) != 0;
  md->use_ucp = (re->options & PCRE_UCP) != 0;
  md->jscript_compat = (re->options & PCRE_JAVASCRIPT_COMPAT) != 0;
+ /* Some options are unpacked into BOOL variables in the hope that testing
+ them will be faster than individual option bits. */
  md->notbol = (options & PCRE_NOTBOL) != 0;
  md->noteol = (options & PCRE_NOTEOL) != 0;
  md->notempty = (options & PCRE_NOTEMPTY) != 0;
  md->notempty_atstart = (options & PCRE_NOTEMPTY_ATSTART) != 0;
- md->partial = ((options & PCRE_PARTIAL_HARD) != 0)? 2 :
-               ((options & PCRE_PARTIAL_SOFT) != 0)? 1 : 0;
  md->hitend = FALSE;
  md->mark = NULL;                        /* In case never set */
-Line 5998 
 defined (though never set). So there's n
+Line 6009 
 defined (though never set). So there's n
  if (md->partial && (re->flags & PCRE_NOPARTIAL) != 0)
    return PCRE_ERROR_BADPARTIAL;
- /* Check a UTF-8 string if required. Pass back the character offset and error
- code if a results vector is available. */
- #ifdef SUPPORT_UTF8
- if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0)
-   {
-   int errorcode;
-   int tb = _pcre_valid_utf8((USPTR)subject, length, &errorcode);
-   if (tb >= 0)
-     {
-     if (offsetcount >= 2)
-       {
-       offsets[0] = tb;
-       offsets[1] = errorcode;
-       }
-     return (errorcode <= PCRE_UTF8_ERR5 && md->partial > 1)?
-       PCRE_ERROR_SHORTUTF8 : PCRE_ERROR_BADUTF8;
-     }
-   if (start_offset > 0 && start_offset < length)
-     {
-     tb = ((USPTR)subject)[start_offset] & 0xc0;
-     if (tb == 0x80) return PCRE_ERROR_BADUTF8_OFFSET;
-     }
-   }
- #endif
  /* If the expression has got more back references than the offsets supplied can
  hold, we get a temporary chunk of working store to use during the matching.
  Otherwise, we can use the vector supplied, rounding down its size to a multiple
  of 3. */
  ocount = offsetcount - (offsetcount % 3);
+ arg_offset_max = (2*ocount)/3;
  if (re->top_backref > 0 && re->top_backref >= ocount/3)
    {
-Line 6046 
 md->offset_max = (2*ocount)/3;
+Line 6032 
 md->offset_max = (2*ocount)/3;
  md->offset_overflow = FALSE;
  md->capture_last = -1;
- /* Compute the minimum number of offsets that we need to reset each time. Doing
- this makes a huge difference to execution time when there aren't many brackets
- in the pattern. */
- resetcount = 2 + re->top_bracket * 2;
- if (resetcount > offsetcount) resetcount = ocount;
  /* Reset the working variable associated with each extraction. These should
  never be used unless previously set, but they get saved and restored, and so we
- initialize them to avoid reading uninitialized locations. */
+ initialize them to avoid reading uninitialized locations. Also, unset the
+ offsets for the matched string. This is really just for tidiness with callouts,
+ in case they inspect these fields. */
  if (md->offset_vector != NULL)
    {
    register int *iptr = md->offset_vector + ocount;
-   register int *iend = iptr - resetcount/2 + 1;
+   register int *iend = iptr - re->top_bracket;
+   if (iend < md->offset_vector + 2) iend = md->offset_vector + 2;
    while (--iptr >= iend) *iptr = -1;
+   md->offset_vector[0] = md->offset_vector[1] = -1;
    }
  /* Set up the first character to match, if available. The first_byte value is
-Line 6095 
 if ((re->flags & PCRE_REQCHSET) != 0)
+Line 6078 
 if ((re->flags & PCRE_REQCHSET) != 0)
    }
  /* ==========================================================================*/
  /* Loop for handling unanchored repeated matching attempts; for anchored regexs
-Line 6105 
 for(;;)
+Line 6090 
 for(;;)
    USPTR save_end_subject = end_subject;
    USPTR new_start_match;
-   /* Reset the maximum number of extractions we might see. */
-   if (md->offset_vector != NULL)
-     {
-     register int *iptr = md->offset_vector;
-     register int *iend = iptr + resetcount;
-     while (iptr < iend) *iptr++ = -1;
-     }
    /* If firstline is TRUE, the start of the match is constrained to the first
    line of a multiline string. That is, the match must be before or at the first
    newline. Implement this by temporarily adjusting end_subject so that we stop
-Line 6302 
 for(;;)
+Line 6278 
 for(;;)
    md->start_match_ptr = start_match;
    md->start_used_ptr = start_match;
    md->match_call_count = 0;
    md->match_function_type = 0;
+   md->end_offset_top = 0;
    rc = match(start_match, md->start_code, start_match, NULL, 2, md, NULL, 0);
    if (md->hitend && start_partial == NULL) start_partial = md->start_used_ptr;
-Line 6413 
 if (rc == MATCH_MATCH || rc == MATCH_ACC
+Line 6390 
 if (rc == MATCH_MATCH || rc == MATCH_ACC
    {
    if (using_temporary_offsets)
      {
-     if (offsetcount >= 4)
+     if (arg_offset_max >= 4)
        {
        memcpy(offsets + 2, md->offset_vector + 2,
-         (offsetcount - 2) * sizeof(int));
+         (arg_offset_max - 2) * sizeof(int));
        DPRINTF(("Copied offsets from temporary memory\n"));
        }
-     if (md->end_offset_top > offsetcount) md->offset_overflow = TRUE;
+     if (md->end_offset_top > arg_offset_max) md->offset_overflow = TRUE;
      DPRINTF(("Freeing temporary memory\n"));
      (pcre_free)(md->offset_vector);
      }
-   /* Set the return code to the number of captured strings, or 0 if there are
+   /* Set the return code to the number of captured strings, or 0 if there were
    too many to fit into the vector. */
-   rc = md->offset_overflow? 0 : md->end_offset_top/2;
+   rc = (md->offset_overflow && md->end_offset_top >= arg_offset_max)?
+: md->end_offset_top/2;
+   /* If there is space in the offset vector, set any unused pairs at the end of
+   the pattern to -1 for backwards compatibility. It is documented that this
+   happens. In earlier versions, the whole set of potential capturing offsets
+   was set to -1 each time round the loop, but this is handled differently now.
+   "Gaps" are set to -1 dynamically instead (this fixes a bug). Thus, it is only
+   those at the end that need unsetting here. We can't just unset them all at
+   the start of the whole thing because they may get set in one branch that is
+   not the final matching branch. */
+   if (md->end_offset_top/2 <= re->top_bracket && offsets != NULL)
+     {
+     register int *iptr, *iend;
+     int resetcount = 2 + re->top_bracket * 2;
+     if (resetcount > offsetcount) resetcount = ocount;
+     iptr = offsets + md->end_offset_top;
+     iend = offsets + resetcount;
+     while (iptr < iend) *iptr++ = -1;
+     }
    /* If there is space, set up the whole thing as substring 0. The value of
    md->start_match_ptr might be modified if \K was encountered on the success

 Legend:



Removed from v.604
 


changed lines


 
Added in v.677
 Legend:



Removed from v.604
 


changed lines


 
Added in v.677
-Removed from v.604
+Added in v.677

webmaster@exim.org	ViewVC Help
Powered by ViewVC 1.1.12