/[pcre]/code/tags/pcre-3.2/pcre.c

Diff of /code/tags/pcre-3.2/pcre.c

Parent Directory | Revision Log | View Patch Patch

-revision 23 by nigel,
Sat Feb 24 21:38:41 2007 UTC
+revision 37 by nigel,
Sat Feb 24 21:39:09 2007 UTC
 Line 9 
 the file Tech.Notes for some information
  Written by: Philip Hazel <ph10@cam.ac.uk>
-            Copyright (c) 1998 University of Cambridge
+            Copyright (c) 1997-1999 University of Cambridge
  -----------------------------------------------------------------------------
  Permission is granted to anyone to use this software for any purpose on any
 Line 25 
 restrictions:
 . Altered versions must be plainly marked as such, and must not be
     misrepresented as being the original software.
+. If PCRE is embedded in any software that is released under the GNU
+    General Purpose Licence (GPL), then the terms of that licence shall
+    supersede any condition above with which it is incompatible.
  -----------------------------------------------------------------------------
  */
-Line 107 
 static const short int escapes[] = {
+Line 111 
 static const short int escapes[] = {
  static BOOL
    compile_regex(int, int, int *, uschar **, const uschar **, const char **,
-     BOOL, int);
+     BOOL, int, int *, int *, compile_data *);
- /* Structure for passing "static" information around between the functions
- doing the matching, so that they are thread-safe. */
- typedef struct match_data {
-   int    errorcode;             /* As it says */
-   int   *offset_vector;         /* Offset vector */
-   int    offset_end;            /* One past the end */
-   int    offset_max;            /* The maximum usable for return data */
-   BOOL   offset_overflow;       /* Set if too many extractions */
-   BOOL   notbol;                /* NOTBOL flag */
-   BOOL   noteol;                /* NOTEOL flag */
-   BOOL   endonly;               /* Dollar not before final \n */
-   const uschar *start_subject;  /* Start of the subject string */
-   const uschar *end_subject;    /* End of the subject string */
-   const uschar *end_match_ptr;  /* Subject position at end match */
-   int     end_offset_top;       /* Highwater mark at end of match */
- } match_data;
-Line 145 
 void  (*pcre_free)(void *) = free;
+Line 131 
 void  (*pcre_free)(void *) = free;
  /*************************************************
+ *             Default character tables           *
+ *************************************************/
+ /* A default set of character tables is included in the PCRE binary. Its source
+ is built by the maketables auxiliary program, which uses the default C ctypes
+ functions, and put in the file chartables.c. These tables are used by PCRE
+ whenever the caller of pcre_compile() does not provide an alternate set of
+ tables. */
+ #include "chartables.c"
+ /*************************************************
  *          Return version string                 *
  *************************************************/
 Line 162 
 return PCRE_VERSION;
  *************************************************/
  /* This function picks potentially useful data out of the private
- structure.
+ structure. The public options are passed back in an int - though the
+ re->options field has been expanded to a long int, all the public options
+ at the low end of it, and so even on 16-bit systems this will still be OK.
+ Therefore, I haven't changed the API for pcre_info().
  Arguments:
    external_re   points to compiled code
-Line 181 
 pcre_info(const pcre *external_re, int *
+Line 184 
 pcre_info(const pcre *external_re, int *
  const real_pcre *re = (const real_pcre *)external_re;
  if (re == NULL) return PCRE_ERROR_NULL;
  if (re->magic_number != MAGIC_NUMBER) return PCRE_ERROR_BADMAGIC;
- if (optptr != NULL) *optptr = (re->options & PUBLIC_OPTIONS);
+ if (optptr != NULL) *optptr = (int)(re->options & PUBLIC_OPTIONS);
  if (first_char != NULL)
    *first_char = ((re->options & PCRE_FIRSTSET) != 0)? re->first_char :
       ((re->options & PCRE_STARTLINE) != 0)? -1 : -2;
-Line 237 
 Arguments:
+Line 240 
 Arguments:
    bracount   number of previous extracting brackets
    options    the options bits
    isclass    TRUE if inside a character class
+   cd         pointer to char tables block
  Returns:     zero or positive => a data character
               negative => a special escape sequence
-Line 245 
 Returns:     zero or positive => a data
+Line 249 
 Returns:     zero or positive => a data
  static int
  check_escape(const uschar **ptrptr, const char **errorptr, int bracount,
-   int options, BOOL isclass)
+   int options, BOOL isclass, compile_data *cd)
  {
  const uschar *ptr = *ptrptr;
  int c = *(++ptr) & 255;   /* Ensure > 0 on signed-char systems */
-Line 288 
 else
+Line 292 
 else
        {
        oldptr = ptr;
        c -= '0';
-       while ((pcre_ctypes[ptr[1]] & ctype_digit) != 0)
+       while ((cd->ctypes[ptr[1]] & ctype_digit) != 0)
          c = c * 10 + *(++ptr) - '0';
        if (c < 10 || c <= bracount)
          {
-Line 314 
 else
+Line 318 
 else
      case '0':
      c -= '0';
-     while(i++ < 2 && (pcre_ctypes[ptr[1]] & ctype_digit) != 0 &&
+     while(i++ < 2 && (cd->ctypes[ptr[1]] & ctype_digit) != 0 &&
        ptr[1] != '8' && ptr[1] != '9')
          c = c * 8 + *(++ptr) - '0';
      break;
-Line 323 
 else
+Line 327 
 else
      case 'x':
      c = 0;
-     while (i++ < 2 && (pcre_ctypes[ptr[1]] & ctype_xdigit) != 0)
+     while (i++ < 2 && (cd->ctypes[ptr[1]] & ctype_xdigit) != 0)
        {
        ptr++;
-       c = c * 16 + pcre_lcc[*ptr] -
+       c = c * 16 + cd->lcc[*ptr] -
-         (((pcre_ctypes[*ptr] & ctype_digit) != 0)? '0' : 'W');
+         (((cd->ctypes[*ptr] & ctype_digit) != 0)? '0' : 'W');
        }
      break;
-Line 341 
 else
+Line 345 
 else
      /* A letter is upper-cased; then the 0x40 bit is flipped */
-     if (c >= 'a' && c <= 'z') c = pcre_fcc[c];
+     if (c >= 'a' && c <= 'z') c = cd->fcc[c];
      c ^= 0x40;
      break;
      /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
      other alphameric following \ is an error if PCRE_EXTRA was set; otherwise,
-     for Perl compatibility, it is a literal. */
+     for Perl compatibility, it is a literal. This code looks a bit odd, but
+     there used to be some cases other than the default, and there may be again
+     in future, so I haven't "optimized" it. */
      default:
      if ((options & PCRE_EXTRA) != 0) switch(c)
-Line 377 
 where the ddds are digits.
+Line 383 
 where the ddds are digits.
  Arguments:
    p         pointer to the first char after '{'
+   cd        pointer to char tables block
  Returns:    TRUE or FALSE
  */
  static BOOL
- is_counted_repeat(const uschar *p)
+ is_counted_repeat(const uschar *p, compile_data *cd)
  {
- if ((pcre_ctypes[*p++] & ctype_digit) == 0) return FALSE;
+ if ((cd->ctypes[*p++] & ctype_digit) == 0) return FALSE;
- while ((pcre_ctypes[*p] & ctype_digit) != 0) p++;
+ while ((cd->ctypes[*p] & ctype_digit) != 0) p++;
  if (*p == '}') return TRUE;
  if (*p++ != ',') return FALSE;
  if (*p == '}') return TRUE;
- if ((pcre_ctypes[*p++] & ctype_digit) == 0) return FALSE;
+ if ((cd->ctypes[*p++] & ctype_digit) == 0) return FALSE;
- while ((pcre_ctypes[*p] & ctype_digit) != 0) p++;
+ while ((cd->ctypes[*p] & ctype_digit) != 0) p++;
  return (*p == '}');
  }
-Line 412 
 Arguments:
+Line 419 
 Arguments:
    maxp       pointer to int for max
               returned as -1 if no max
    errorptr   points to pointer to error message
+   cd         pointer to character tables clock
  Returns:     pointer to '}' on success;
               current ptr on error, with errorptr set
  */
  static const uschar *
- read_repeat_counts(const uschar *p, int *minp, int *maxp, const char **errorptr)
+ read_repeat_counts(const uschar *p, int *minp, int *maxp,
+   const char **errorptr, compile_data *cd)
  {
  int min = 0;
  int max = -1;
- while ((pcre_ctypes[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0';
+ while ((cd->ctypes[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0';
  if (*p == '}') max = min; else
    {
    if (*(++p) != '}')
      {
      max = 0;
-     while((pcre_ctypes[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0';
+     while((cd->ctypes[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0';
      if (max < min)
        {
        *errorptr = ERR4;
-Line 526 
 for (;;)
+Line 535 
 for (;;)
      case OP_REVERSE:
      cc++;
+     /* Fall through */
      case OP_CREF:
      case OP_OPT:
-Line 615 
 for (;;)
+Line 625 
 for (;;)
  /* Scan the pattern, compiling it into the code vector.
  Arguments:
-   options     the option bits
+   options      the option bits
-   brackets    points to number of brackets used
+   brackets     points to number of brackets used
-   code        points to the pointer to the current code point
+   code         points to the pointer to the current code point
-   ptrptr      points to the current pattern pointer
+   ptrptr       points to the current pattern pointer
-   errorptr    points to pointer to error message
+   errorptr     points to pointer to error message
-   optchanged  set to the value of the last OP_OPT item compiled
+   optchanged   set to the value of the last OP_OPT item compiled
+   reqchar      set to the last literal character required, else -1
+   countlits    set to count of mandatory literal characters
+   cd           contains pointers to tables
- Returns:      TRUE on success
+ Returns:       TRUE on success
-               FALSE, with *errorptr set on error
+                FALSE, with *errorptr set on error
  */
  static BOOL
  compile_branch(int options, int *brackets, uschar **codeptr,
-   const uschar **ptrptr, const char **errorptr, int *optchanged)
+   const uschar **ptrptr, const char **errorptr, int *optchanged,
+   int *reqchar, int *countlits, compile_data *cd)
  {
  int repeat_type, op_type;
  int repeat_min, repeat_max;
  int bravalue, length;
  int greedy_default, greedy_non_default;
+ int prevreqchar;
+ int condcount = 0;
+ int subcountlits = 0;
  register int c;
  register uschar *code = *codeptr;
  uschar *tempcode;
-Line 647 
 uschar class[32];
+Line 664 
 uschar class[32];
  greedy_default = ((options & PCRE_UNGREEDY) != 0);
  greedy_non_default = greedy_default ^ 1;
+ /* Initialize no required char, and count of literals */
+ *reqchar = prevreqchar = -1;
+ *countlits = 0;
  /* Switch on next character until the end of the branch */
  for (;; ptr++)
-Line 656 
 for (;; ptr++)
+Line 678 
 for (;; ptr++)
    int class_lastchar;
    int newoptions;
    int condref;
+   int subreqchar;
    c = *ptr;
    if ((options & PCRE_EXTENDED) != 0)
      {
-     if ((pcre_ctypes[c] & ctype_space) != 0) continue;
+     if ((cd->ctypes[c] & ctype_space) != 0) continue;
      if (c == '#')
        {
        while ((c = *(++ptr)) != 0 && c != '\n');
-Line 748 
 for (;; ptr++)
+Line 771 
 for (;; ptr++)
        if (c == '\\')
          {
-         c = check_escape(&ptr, errorptr, *brackets, options, TRUE);
+         c = check_escape(&ptr, errorptr, *brackets, options, TRUE, cd);
          if (-c == ESC_b) c = '\b';
          else if (c < 0)
            {
+           register const uschar *cbits = cd->cbits;
            class_charcount = 10;
            switch (-c)
              {
              case ESC_d:
-             for (c = 0; c < 32; c++) class[c] |= pcre_cbits[c+cbit_digit];
+             for (c = 0; c < 32; c++) class[c] |= cbits[c+cbit_digit];
              continue;
              case ESC_D:
-             for (c = 0; c < 32; c++) class[c] |= ~pcre_cbits[c+cbit_digit];
+             for (c = 0; c < 32; c++) class[c] |= ~cbits[c+cbit_digit];
              continue;
              case ESC_w:
              for (c = 0; c < 32; c++)
-               class[c] |= (pcre_cbits[c] | pcre_cbits[c+cbit_word]);
+               class[c] |= (cbits[c+cbit_digit] | cbits[c+cbit_word]);
              continue;
              case ESC_W:
              for (c = 0; c < 32; c++)
-               class[c] |= ~(pcre_cbits[c] | pcre_cbits[c+cbit_word]);
+               class[c] |= ~(cbits[c+cbit_digit] | cbits[c+cbit_word]);
              continue;
              case ESC_s:
-             for (c = 0; c < 32; c++) class[c] |= pcre_cbits[c+cbit_space];
+             for (c = 0; c < 32; c++) class[c] |= cbits[c+cbit_space];
              continue;
              case ESC_S:
-             for (c = 0; c < 32; c++) class[c] |= ~pcre_cbits[c+cbit_space];
+             for (c = 0; c < 32; c++) class[c] |= ~cbits[c+cbit_space];
              continue;
              default:
-Line 810 
 for (;; ptr++)
+Line 834 
 for (;; ptr++)
          if (d == '\\')
            {
-           d = check_escape(&ptr, errorptr, *brackets, options, TRUE);
+           d = check_escape(&ptr, errorptr, *brackets, options, TRUE, cd);
            if (d < 0)
              {
              if (d == -ESC_b) d = '\b'; else
-Line 832 
 for (;; ptr++)
+Line 856 
 for (;; ptr++)
            class[c/8] |= (1 << (c&7));
            if ((options & PCRE_CASELESS) != 0)
              {
-             int uc = pcre_fcc[c];           /* flip case */
+             int uc = cd->fcc[c];           /* flip case */
              class[uc/8] |= (1 << (uc&7));
              }
            class_charcount++;                /* in case a one-char range */
-Line 847 
 for (;; ptr++)
+Line 871 
 for (;; ptr++)
        class [c/8] |= (1 << (c&7));
        if ((options & PCRE_CASELESS) != 0)
          {
-         c = pcre_fcc[c];   /* flip case */
+         c = cd->fcc[c];   /* flip case */
          class[c/8] |= (1 << (c&7));
          }
        class_charcount++;
-Line 894 
 for (;; ptr++)
+Line 918 
 for (;; ptr++)
      /* Various kinds of repeat */
      case '{':
-     if (!is_counted_repeat(ptr+1)) goto NORMAL_CHAR;
+     if (!is_counted_repeat(ptr+1, cd)) goto NORMAL_CHAR;
-     ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorptr);
+     ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorptr, cd);
      if (*errorptr != NULL) goto FAILED;
      goto REPEAT;
-Line 928 
 for (;; ptr++)
+Line 952 
 for (;; ptr++)
        { repeat_type = greedy_non_default; ptr++; }
      else repeat_type = greedy_default;
-     /* If the maximum is zero then the minimum must also be zero; Perl allows
-     this case, so we do too - by simply omitting the item altogether. */
-     if (repeat_max == 0) code = previous;
      /* If previous was a string of characters, chop off the last one and use it
      as the subject of the repeat. If there was only one character, we can
-     abolish the previous item altogether. */
+     abolish the previous item altogether. A repeat with a zero minimum wipes
+     out any reqchar setting, backing up to the previous value. We must also
+     adjust the countlits value. */
-     else if (*previous == OP_CHARS)
+     if (*previous == OP_CHARS)
        {
        int len = previous[1];
+       if (repeat_min == 0) *reqchar = prevreqchar;
+       *countlits += repeat_min - 1;
        if (len == 1)
          {
          c = previous[2];
-Line 978 
 for (;; ptr++)
+Line 1003 
 for (;; ptr++)
        code = previous;
        OUTPUT_SINGLE_REPEAT:
-       repeat_type += op_type;      /* Combine both values for many cases */
+       /* If the maximum is zero then the minimum must also be zero; Perl allows
+       this case, so we do too - by simply omitting the item altogether. */
+       if (repeat_max == 0) goto END_REPEAT;
+       /* Combine the op_type with the repeat_type */
+       repeat_type += op_type;
        /* A minimum of zero is handled either as the special case * or ?, or as
        an UPTO, with the maximum given. */
-Line 1055 
 for (;; ptr++)
+Line 1088 
 for (;; ptr++)
        }
      /* If previous was a character class or a back reference, we put the repeat
-     stuff after it. */
+     stuff after it, but just skip the item if the repeat was {0,0}. */
      else if (*previous == OP_CLASS || *previous == OP_REF)
        {
+       if (repeat_max == 0)
+         {
+         code = previous;
+         goto END_REPEAT;
+         }
        if (repeat_min == 0 && repeat_max == -1)
          *code++ = OP_CRSTAR + repeat_type;
        else if (repeat_min == 1 && repeat_max == -1)
-Line 1082 
 for (;; ptr++)
+Line 1120 
 for (;; ptr++)
      else if ((int)*previous >= OP_BRA || (int)*previous == OP_ONCE ||
               (int)*previous == OP_COND)
        {
-       int i, ketoffset = 0;
+       register int i;
+       int ketoffset = 0;
        int len = code - previous;
+       uschar *bralink = NULL;
        /* If the maximum repeat count is unlimited, find the end of the bracket
        by scanning through from the start, and compute the offset back to it
-Line 1098 
 for (;; ptr++)
+Line 1138 
 for (;; ptr++)
          ketoffset = code - ket;
          }
-       /* If the minimum is greater than zero, and the maximum is unlimited or
+       /* The case of a zero minimum is special because of the need to stick
-       equal to the minimum, the first copy remains where it is, and is
+       OP_BRAZERO in front of it, and because the group appears once in the
-       replicated up to the minimum number of times. This case includes the +
+       data, whereas in other cases it appears the minimum number of times. For
-       repeat, but of course no replication is needed in that case. */
+       this reason, it is simplest to treat this case separately, as otherwise
+       the code gets far too mess. There are several special subcases when the
+       minimum is zero. */
-       if (repeat_min > 0 && (repeat_max == -1 || repeat_max == repeat_min))
+       if (repeat_min == 0)
          {
-         for (i = 1; i < repeat_min; i++)
+         /* If we set up a required char from the bracket, we must back off
+         to the previous value and reset the countlits value too. */
+         if (subcountlits > 0)
            {
-           memcpy(code, previous, len);
+           *reqchar = prevreqchar;
-           code += len;
+           *countlits -= subcountlits;
            }
-         }
-       /* If the minimum is zero, stick BRAZERO in front of the first copy.
+         /* If the maximum is also zero, we just omit the group from the output
-       Then, if there is a fixed upper limit, replicated up to that many times,
+         altogether. */
-       sticking BRAZERO in front of all the optional ones. */
-       else
+         if (repeat_max == 0)
-         {
+           {
-         if (repeat_min == 0)
+           code = previous;
+           goto END_REPEAT;
+           }
+         /* If the maximum is 1 or unlimited, we just have to stick in the
+         BRAZERO and do no more at this point. */
+         if (repeat_max <= 1)
            {
            memmove(previous+1, previous, len);
            code++;
            *previous++ = OP_BRAZERO + repeat_type;
            }
+         /* If the maximum is greater than 1 and limited, we have to replicate
+         in a nested fashion, sticking OP_BRAZERO before each set of brackets.
+         The first one has to be handled carefully because it's the original
+         copy, which has to be moved up. The remainder can be handled by code
+         that is common with the non-zero minimum case below. We just have to
+         adjust the value or repeat_max, since one less copy is required. */
+         else
+           {
+           int offset;
+           memmove(previous+4, previous, len);
+           code += 4;
+           *previous++ = OP_BRAZERO + repeat_type;
+           *previous++ = OP_BRA;
+           /* We chain together the bracket offset fields that have to be
+           filled in later when the ends of the brackets are reached. */
+           offset = (bralink == NULL)? 0 : previous - bralink;
+           bralink = previous;
+           *previous++ = offset >> 8;
+           *previous++ = offset & 255;
+           }
+         repeat_max--;
+         }
+       /* If the minimum is greater than zero, replicate the group as many
+       times as necessary, and adjust the maximum to the number of subsequent
+       copies that we need. */
+       else
+         {
          for (i = 1; i < repeat_min; i++)
            {
            memcpy(code, previous, len);
            code += len;
            }
+         if (repeat_max > 0) repeat_max -= repeat_min;
+         }
+       /* This code is common to both the zero and non-zero minimum cases. If
+       the maximum is limited, it replicates the group in a nested fashion,
+       remembering the bracket starts on a stack. In the case of a zero minimum,
+       the first one was set up above. In all cases the repeat_max now specifies
+       the number of additional copies needed. */
-         for (i = (repeat_min > 0)? repeat_min : 1; i < repeat_max; i++)
+       if (repeat_max >= 0)
+         {
+         for (i = repeat_max - 1; i >= 0; i--)
            {
            *code++ = OP_BRAZERO + repeat_type;
+           /* All but the final copy start a new nesting, maintaining the
+           chain of brackets outstanding. */
+           if (i != 0)
+             {
+             int offset;
+             *code++ = OP_BRA;
+             offset = (bralink == NULL)? 0 : code - bralink;
+             bralink = code;
+             *code++ = offset >> 8;
+             *code++ = offset & 255;
+             }
            memcpy(code, previous, len);
            code += len;
            }
+         /* Now chain through the pending brackets, and fill in their length
+         fields (which are holding the chain links pro tem). */
+         while (bralink != NULL)
+           {
+           int oldlinkoffset;
+           int offset = code - bralink + 1;
+           uschar *bra = code - offset;
+           oldlinkoffset = (bra[1] << 8) + bra[2];
+           bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
+           *code++ = OP_KET;
+           *code++ = bra[1] = offset >> 8;
+           *code++ = bra[2] = (offset & 255);
+           }
          }
        /* If the maximum is unlimited, set a repeater in the final copy. We
-Line 1144 
 for (;; ptr++)
+Line 1266 
 for (;; ptr++)
        don't know if there's been an options resetting after the ket. The
        correct offset was computed above. */
-       if (repeat_max == -1) code[-ketoffset] = OP_KETRMAX + repeat_type;
+       else code[-ketoffset] = OP_KETRMAX + repeat_type;
        }
      /* Else there's some kind of shambles */
-Line 1157 
 for (;; ptr++)
+Line 1279 
 for (;; ptr++)
      /* In all case we no longer have a previous item. */
+     END_REPEAT:
      previous = NULL;
      break;
-Line 1191 
 for (;; ptr++)
+Line 1314 
 for (;; ptr++)
          case '(':
          bravalue = OP_COND;       /* Conditional group */
-         if ((pcre_ctypes[*(++ptr)] & ctype_digit) != 0)
+         if ((cd->ctypes[*(++ptr)] & ctype_digit) != 0)
            {
            condref = *ptr - '0';
            while (*(++ptr) != ')') condref = condref*10 + *ptr - '0';
-Line 1324 
 for (;; ptr++)
+Line 1447 
 for (;; ptr++)
           errorptr,                     /* Where to put an error message */
           (bravalue == OP_ASSERTBACK ||
            bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
-          condref))                     /* Condition reference number */
+          condref,                      /* Condition reference number */
+          &subreqchar,                  /* For possible last char */
+          &subcountlits,                /* For literal count */
+          cd))                          /* Tables block */
        goto FAILED;
      /* At the end of compiling, code is still pointing to the start of the
-Line 1337 
 for (;; ptr++)
+Line 1463 
 for (;; ptr++)
      if (bravalue == OP_COND)
        {
-       int branchcount = 0;
        uschar *tc = code;
+       condcount = 0;
        do {
-          branchcount++;
+          condcount++;
           tc += (tc[1] << 8) | tc[2];
           }
        while (*tc != OP_KET);
-       if (branchcount > 2)
+       if (condcount > 2)
          {
          *errorptr = ERR27;
          goto FAILED;
          }
        }
+     /* Handle updating of the required character. If the subpattern didn't
+     set one, leave it as it was. Otherwise, update it for normal brackets of
+     all kinds, forward assertions, and conditions with two branches. Don't
+     update the literal count for forward assertions, however. If the bracket
+     is followed by a quantifier with zero repeat, we have to back off. Hence
+     the definition of prevreqchar and subcountlits outside the main loop so
+     that they can be accessed for the back off. */
+     if (subreqchar > 0 &&
+          (bravalue >= OP_BRA || bravalue == OP_ONCE || bravalue == OP_ASSERT ||
+          (bravalue == OP_COND && condcount == 2)))
+       {
+       prevreqchar = *reqchar;
+       *reqchar = subreqchar;
+       if (bravalue != OP_ASSERT) *countlits += subcountlits;
+       }
      /* Now update the main code pointer to the end of the group. */
      code = tempcode;
-Line 1372 
 for (;; ptr++)
+Line 1515 
 for (;; ptr++)
      case '\\':
      tempptr = ptr;
-     c = check_escape(&ptr, errorptr, *brackets, options, FALSE);
+     c = check_escape(&ptr, errorptr, *brackets, options, FALSE, cd);
      /* Handle metacharacters introduced by \. For ones like \d, the ESC_ values
      are arranged to be the negation of the corresponding OP_values. For the
-Line 1417 
 for (;; ptr++)
+Line 1560 
 for (;; ptr++)
        {
        if ((options & PCRE_EXTENDED) != 0)
          {
-         if ((pcre_ctypes[c] & ctype_space) != 0) continue;
+         if ((cd->ctypes[c] & ctype_space) != 0) continue;
          if (c == '#')
            {
            while ((c = *(++ptr)) != 0 && c != '\n');
-Line 1433 
 for (;; ptr++)
+Line 1576 
 for (;; ptr++)
        if (c == '\\')
          {
          tempptr = ptr;
-         c = check_escape(&ptr, errorptr, *brackets, options, FALSE);
+         c = check_escape(&ptr, errorptr, *brackets, options, FALSE, cd);
          if (c < 0) { ptr = tempptr; break; }
          }
-Line 1445 
 for (;; ptr++)
+Line 1588 
 for (;; ptr++)
      /* This "while" is the end of the "do" above. */
-     while (length < 255 && (pcre_ctypes[c = *(++ptr)] & ctype_meta) == 0);
+     while (length < 255 && (cd->ctypes[c = *(++ptr)] & ctype_meta) == 0);
+     /* Update the last character and the count of literals */
+     prevreqchar = (length > 1)? code[-2] : *reqchar;
+     *reqchar = code[-1];
+     *countlits += length;
      /* Compute the length and set it in the data vector, and advance to
      the next state. */
-Line 1490 
 Argument:
+Line 1639 
 Argument:
    errorptr    -> pointer to error message
    lookbehind  TRUE if this is a lookbehind assertion
    condref     > 0 for OPT_CREF setting at start of conditional group
+   reqchar     -> place to put the last required character, or a negative number
+   countlits   -> place to put the shortest literal count of any branch
+   cd          points to the data block with tables pointers
  Returns:      TRUE on success
  */
  static BOOL
  compile_regex(int options, int optchanged, int *brackets, uschar **codeptr,
-   const uschar **ptrptr, const char **errorptr, BOOL lookbehind, int condref)
+   const uschar **ptrptr, const char **errorptr, BOOL lookbehind, int condref,
+   int *reqchar, int *countlits, compile_data *cd)
  {
  const uschar *ptr = *ptrptr;
  uschar *code = *codeptr;
-Line 1504 
 uschar *last_branch = code;
+Line 1657 
 uschar *last_branch = code;
  uschar *start_bracket = code;
  uschar *reverse_count = NULL;
  int oldoptions = options & PCRE_IMS;
+ int branchreqchar, branchcountlits;
+ *reqchar = -1;
+ *countlits = INT_MAX;
  code += 3;
  /* At the start of a reference-based conditional group, insert the reference
-Line 1543 
 for (;;)
+Line 1699 
 for (;;)
    /* Now compile the branch */
-   if (!compile_branch(options, brackets, &code, &ptr, errorptr, &optchanged))
+   if (!compile_branch(options, brackets, &code, &ptr, errorptr, &optchanged,
+       &branchreqchar, &branchcountlits, cd))
      {
      *ptrptr = ptr;
      return FALSE;
-Line 1555 
 for (;;)
+Line 1712 
 for (;;)
    last_branch[1] = length >> 8;
    last_branch[2] = length & 255;
+   /* Save the last required character if all branches have the same; a current
+   value of -1 means unset, while -2 means "previous branch had no last required
+   char".  */
+   if (*reqchar != -2)
+     {
+     if (branchreqchar >= 0)
+       {
+       if (*reqchar == -1) *reqchar = branchreqchar;
+       else if (*reqchar != branchreqchar) *reqchar = -2;
+       }
+     else *reqchar = -2;
+     }
+   /* Keep the shortest literal count */
+   if (branchcountlits < *countlits) *countlits = branchcountlits;
+   DPRINTF(("literal count = %d min=%d\n", branchcountlits, *countlits));
    /* If lookbehind, check that this branch matches a fixed-length string,
    and put the length into the OP_REVERSE item. Temporarily mark the end of
    the branch with OP_END. */
-Line 1649 
 for (;;)
+Line 1825 
 for (;;)
      code += 2;
      break;
+     case OP_WORD_BOUNDARY:
+     case OP_NOT_WORD_BOUNDARY:
+     code++;
+     break;
      case OP_ASSERT_NOT:
      case OP_ASSERTBACK:
      case OP_ASSERTBACK_NOT:
-Line 1676 
 all of whose alternatives start with OP_
+Line 1857 
 all of whose alternatives start with OP_
  it's anchored. However, if this is a multiline pattern, then only OP_SOD
  counts, since OP_CIRC can match in the middle.
- A branch is also implicitly anchored if it starts with .* because that will try
+ A branch is also implicitly anchored if it starts with .* and DOTALL is set,
- the rest of the pattern at all possible matching points, so there is no point
+ because that will try the rest of the pattern at all possible matching points,
- trying them again.
+ so there is no point trying them again.
  Arguments:
    code       points to start of expression (the bracket)
-Line 1696 
 do {
+Line 1877 
 do {
     register int op = *scode;
     if (op >= OP_BRA || op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
       { if (!is_anchored(scode, options)) return FALSE; }
-    else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR)
+    else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR) &&
+             (*options & PCRE_DOTALL) != 0)
       { if (scode[1] != OP_ANY) return FALSE; }
     else if (op != OP_SOD &&
             ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
-Line 1710 
 return TRUE;
+Line 1892 
 return TRUE;
  /*************************************************
- *     Check for start with \n line expression    *
+ *         Check for starting with ^ or .*        *
  *************************************************/
- /* This is called for multiline expressions to try to find out if every branch
+ /* This is called to find out if every branch starts with ^ or .* so that
- starts with ^ so that "first char" processing can be done to speed things up.
+ "first char" processing can be done to speed things up in multiline
+ matching and for non-DOTALL patterns that start with .* (which must start at
+ the beginning or after \n).
  Argument:  points to start of expression (the bracket)
  Returns:   TRUE or FALSE
-Line 1728 
 do {
+Line 1912 
 do {
     register int op = *scode;
     if (op >= OP_BRA || op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
       { if (!is_startline(scode)) return FALSE; }
+    else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR)
+      { if (scode[1] != OP_ANY) return FALSE; }
     else if (op != OP_CIRC) return FALSE;
     code += (code[1] << 8) + code[2];
     }
-Line 1813 
 Arguments:
+Line 1999 
 Arguments:
    options      various option bits
    errorptr     pointer to pointer to error text
    erroroffset  ptr offset in pattern where error was detected
+   tables       pointer to character tables or NULL
  Returns:       pointer to compiled data block, or NULL on error,
                 with errorptr and erroroffset set
-Line 1820 
 Returns:       pointer to compiled data
+Line 2007 
 Returns:       pointer to compiled data
  pcre *
  pcre_compile(const char *pattern, int options, const char **errorptr,
-   int *erroroffset)
+   int *erroroffset, const unsigned char *tables)
  {
  real_pcre *re;
  int length = 3;      /* For initial BRA plus length */
  int runlength;
- int c, size;
+ int c, size, reqchar, countlits;
  int bracount = 0;
  int top_backref = 0;
  int branch_extra = 0;
-Line 1833 
 int branch_newextra;
+Line 2020 
 int branch_newextra;
  unsigned int brastackptr = 0;
  uschar *code;
  const uschar *ptr;
+ compile_data compile_block;
  int brastack[BRASTACK_SIZE];
  uschar bralenstack[BRASTACK_SIZE];
-Line 1861 
 if ((options & ~PUBLIC_OPTIONS) != 0)
+Line 2049 
 if ((options & ~PUBLIC_OPTIONS) != 0)
    return NULL;
    }
+ /* Set up pointers to the individual character tables */
+ if (tables == NULL) tables = pcre_default_tables;
+ compile_block.lcc = tables + lcc_offset;
+ compile_block.fcc = tables + fcc_offset;
+ compile_block.cbits = tables + cbits_offset;
+ compile_block.ctypes = tables + ctypes_offset;
+ /* Reflect pattern for debugging output */
  DPRINTF(("------------------------------------------------------------------\n"));
  DPRINTF(("%s\n", pattern));
-Line 1879 
 while ((c = *(++ptr)) != 0)
+Line 2077 
 while ((c = *(++ptr)) != 0)
    if ((options & PCRE_EXTENDED) != 0)
      {
-     if ((pcre_ctypes[c] & ctype_space) != 0) continue;
+     if ((compile_block.ctypes[c] & ctype_space) != 0) continue;
      if (c == '#')
        {
        while ((c = *(++ptr)) != 0 && c != '\n');
-Line 1897 
 while ((c = *(++ptr)) != 0)
+Line 2095 
 while ((c = *(++ptr)) != 0)
      case '\\':
        {
        const uschar *save_ptr = ptr;
-       c = check_escape(&ptr, errorptr, bracount, options, FALSE);
+       c = check_escape(&ptr, errorptr, bracount, options, FALSE, &compile_block);
        if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
        if (c >= 0)
          {
-Line 1917 
 while ((c = *(++ptr)) != 0)
+Line 2115 
 while ((c = *(++ptr)) != 0)
        int refnum = -c - ESC_REF;
        if (refnum > top_backref) top_backref = refnum;
        length++;   /* For single back reference */
-       if (ptr[1] == '{' && is_counted_repeat(ptr+2))
+       if (ptr[1] == '{' && is_counted_repeat(ptr+2, &compile_block))
          {
-         ptr = read_repeat_counts(ptr+2, &min, &max, errorptr);
+         ptr = read_repeat_counts(ptr+2, &min, &max, errorptr, &compile_block);
          if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
          if ((min == 0 && (max == 1 || max == -1)) ||
            (min == 1 && max == -1))
-Line 1943 
 while ((c = *(++ptr)) != 0)
+Line 2141 
 while ((c = *(++ptr)) != 0)
      or back reference. */
      case '{':
-     if (!is_counted_repeat(ptr+1)) goto NORMAL_CHAR;
+     if (!is_counted_repeat(ptr+1, &compile_block)) goto NORMAL_CHAR;
-     ptr = read_repeat_counts(ptr+1, &min, &max, errorptr);
+     ptr = read_repeat_counts(ptr+1, &min, &max, errorptr, &compile_block);
      if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
      if ((min == 0 && (max == 1 || max == -1)) ||
        (min == 1 && max == -1))
-Line 1979 
 while ((c = *(++ptr)) != 0)
+Line 2177 
 while ((c = *(++ptr)) != 0)
        {
        if (*ptr == '\\')
          {
-         int ch = check_escape(&ptr, errorptr, bracount, options, TRUE);
+         int ch = check_escape(&ptr, errorptr, bracount, options, TRUE,
+           &compile_block);
          if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
          if (-ch == ESC_b) class_charcount++; else class_charcount = 10;
          }
-Line 1996 
 while ((c = *(++ptr)) != 0)
+Line 2195 
 while ((c = *(++ptr)) != 0)
        /* A repeat needs either 1 or 5 bytes. */
-       if (*ptr != 0 && ptr[1] == '{' && is_counted_repeat(ptr+2))
+       if (*ptr != 0 && ptr[1] == '{' && is_counted_repeat(ptr+2, &compile_block))
          {
-         ptr = read_repeat_counts(ptr+2, &min, &max, errorptr);
+         ptr = read_repeat_counts(ptr+2, &min, &max, errorptr, &compile_block);
          if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
          if ((min == 0 && (max == 1 || max == -1)) ||
            (min == 1 && max == -1))
-Line 2064 
 while ((c = *(++ptr)) != 0)
+Line 2263 
 while ((c = *(++ptr)) != 0)
          group. */
          case '(':
-         if ((pcre_ctypes[ptr[3]] & ctype_digit) != 0)
+         if ((compile_block.ctypes[ptr[3]] & ctype_digit) != 0)
            {
            ptr += 4;
            length += 2;
-           while ((pcre_ctypes[*ptr] & ctype_digit) != 0) ptr++;
+           while ((compile_block.ctypes[*ptr] & ctype_digit) != 0) ptr++;
            if (*ptr != ')')
              {
              *errorptr = ERR26;
-Line 2153 
 while ((c = *(++ptr)) != 0)
+Line 2352 
 while ((c = *(++ptr)) != 0)
              will lead to an over-estimate on the length, but this shouldn't
              matter very much. We also have to allow for resetting options at
              the start of any alternations, which we do by setting
-             branch_newextra to 2. */
+             branch_newextra to 2. Finally, we record whether the case-dependent
+             flag ever changes within the regex. This is used by the "required
+             character" code. */
              case ':':
              if (((set|unset) & PCRE_IMS) != 0)
                {
                length += 4;
                branch_newextra = 2;
+               if (((set|unset) & PCRE_CASELESS) != 0) options |= PCRE_ICHANGED;
                }
              goto END_OPTIONS;
-Line 2237 
 while ((c = *(++ptr)) != 0)
+Line 2439 
 while ((c = *(++ptr)) != 0)
        /* Leave ptr at the final char; for read_repeat_counts this happens
        automatically; for the others we need an increment. */
-       if ((c = ptr[1]) == '{' && is_counted_repeat(ptr+2))
+       if ((c = ptr[1]) == '{' && is_counted_repeat(ptr+2, &compile_block))
          {
-         ptr = read_repeat_counts(ptr+2, &minval, &maxval, errorptr);
+         ptr = read_repeat_counts(ptr+2, &minval, &maxval, errorptr,
+           &compile_block);
          if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
          }
        else if (c == '*') { minval = 0; maxval = -1; ptr++; }
        else if (c == '+') { maxval = -1; ptr++; }
        else if (c == '?') { minval = 0; ptr++; }
-       /* If there is a minimum > 1 we have to replicate up to minval-1 times;
+       /* If the minimum is zero, we have to allow for an OP_BRAZERO before the
-       if there is a limited maximum we have to replicate up to maxval-1 times
+       group, and if the maximum is greater than zero, we have to replicate
-       and allow for a BRAZERO item before each optional copy, as we also have
+       maxval-1 times; each replication acquires an OP_BRAZERO plus a nesting
-       to do before the first copy if the minimum is zero. */
+       bracket set - hence the 7. */
-       if (minval == 0) length++;
+       if (minval == 0)
-         else if (minval > 1) length += (minval - 1) * duplength;
+         {
-       if (maxval > minval) length += (maxval - minval) * (duplength + 1);
+         length++;
+         if (maxval > 0) length += (maxval - 1) * (duplength + 7);
+         }
+       /* When the minimum is greater than zero, 1 we have to replicate up to
+       minval-1 times, with no additions required in the copies. Then, if
+       there is a limited maximum we have to replicate up to maxval-1 times
+       allowing for a BRAZERO item before each optional copy and nesting
+       brackets for all but one of the optional copies. */
+       else
+         {
+         length += (minval - 1) * duplength;
+         if (maxval > minval)   /* Need this test as maxval=-1 means no limit */
+           length += (maxval - minval) * (duplength + 7) - 6;
+         }
        }
      continue;
-Line 2270 
 while ((c = *(++ptr)) != 0)
+Line 2488 
 while ((c = *(++ptr)) != 0)
        {
        if ((options & PCRE_EXTENDED) != 0)
          {
-         if ((pcre_ctypes[c] & ctype_space) != 0) continue;
+         if ((compile_block.ctypes[c] & ctype_space) != 0) continue;
          if (c == '#')
            {
            while ((c = *(++ptr)) != 0 && c != '\n');
-Line 2284 
 while ((c = *(++ptr)) != 0)
+Line 2502 
 while ((c = *(++ptr)) != 0)
        if (c == '\\')
          {
          const uschar *saveptr = ptr;
-         c = check_escape(&ptr, errorptr, bracount, options, FALSE);
+         c = check_escape(&ptr, errorptr, bracount, options, FALSE,
+           &compile_block);
          if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
          if (c < 0) { ptr = saveptr; break; }
          }
-Line 2296 
 while ((c = *(++ptr)) != 0)
+Line 2515 
 while ((c = *(++ptr)) != 0)
      /* This "while" is the end of the "do" above. */
-     while (runlength < 255 && (pcre_ctypes[c = *(++ptr)] & ctype_meta) == 0);
+     while (runlength < 255 &&
+       (compile_block.ctypes[c = *(++ptr)] & ctype_meta) == 0);
      ptr--;
      length += runlength;
-Line 2331 
 if (re == NULL)
+Line 2551 
 if (re == NULL)
  re->magic_number = MAGIC_NUMBER;
  re->options = options;
+ re->tables = tables;
  /* Set up a starting, non-extracting bracket, then compile the expression. On
  error, *errorptr will be set non-NULL, so we don't need to look at the result
-Line 2340 
 ptr = (const uschar *)pattern;
+Line 2561 
 ptr = (const uschar *)pattern;
  code = re->code;
  *code = OP_BRA;
  bracount = 0;
- (void)compile_regex(options, -1, &bracount, &code, &ptr, errorptr, FALSE, -1);
+ (void)compile_regex(options, -1, &bracount, &code, &ptr, errorptr, FALSE, -1,
+   &reqchar, &countlits, &compile_block);
  re->top_bracket = bracount;
  re->top_backref = top_backref;
-Line 2372 
 if (*errorptr != NULL)
+Line 2594 
 if (*errorptr != NULL)
    return NULL;
    }
- /* If the anchored option was not passed, set flag if we can determine that it
+ /* If the anchored option was not passed, set flag if we can determine that the
- is anchored by virtue of ^ characters or \A or anything else. Otherwise, see if
+ pattern is anchored by virtue of ^ characters or \A or anything else (such as
- we can determine what the first character has to be, because that speeds up
+ starting with .* when DOTALL is set).
- unanchored matches no end. In the case of multiline matches, an alternative is
- to set the PCRE_STARTLINE flag if all branches start with ^. */
+ Otherwise, see if we can determine what the first character has to be, because
+ that speeds up unanchored matches no end. If not, see if we can set the
+ PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
+ start with ^. and also when all branches start with .* for non-DOTALL matches.
+ */
  if ((options & PCRE_ANCHORED) == 0)
    {
-Line 2396 
 if ((options & PCRE_ANCHORED) == 0)
+Line 2622 
 if ((options & PCRE_ANCHORED) == 0)
      }
    }
+ /* Save the last required character if there are at least two literal
+ characters on all paths, or if there is no first character setting. */
+ if (reqchar >= 0 && (countlits > 1 || (re->options & PCRE_FIRSTSET) == 0))
+   {
+   re->req_char = reqchar;
+   re->options |= PCRE_REQCHSET;
+   }
  /* Print out the compiled data for debugging */
  #ifdef DEBUG
-Line 2405 
 printf("Length = %d top_bracket = %d top
+Line 2640 
 printf("Length = %d top_bracket = %d top
  if (re->options != 0)
    {
-   printf("%s%s%s%s%s%s%s%s\n",
+   printf("%s%s%s%s%s%s%s%s%s\n",
      ((re->options & PCRE_ANCHORED) != 0)? "anchored " : "",
      ((re->options & PCRE_CASELESS) != 0)? "caseless " : "",
+     ((re->options & PCRE_ICHANGED) != 0)? "case state changed " : "",
      ((re->options & PCRE_EXTENDED) != 0)? "extended " : "",
      ((re->options & PCRE_MULTILINE) != 0)? "multiline " : "",
      ((re->options & PCRE_DOTALL) != 0)? "dotall " : "",
-Line 2422 
 if ((re->options & PCRE_FIRSTSET) != 0)
+Line 2658 
 if ((re->options & PCRE_FIRSTSET) != 0)
      else printf("First char = \\x%02x\n", re->first_char);
    }
+ if ((re->options & PCRE_REQCHSET) != 0)
+   {
+   if (isprint(re->req_char)) printf("Req char = %c\n", re->req_char);
+     else printf("Req char = \\x%02x\n", re->req_char);
+   }
  code_end = code;
  code_base = code = re->code;
-Line 2637 
 return (pcre *)re;
+Line 2879 
 return (pcre *)re;
  /*************************************************
- *        Match a character type                  *
- *************************************************/
- /* Not used in all the places it might be as it's sometimes faster
- to put the code inline.
- Arguments:
-   type        the character type
-   c           the character
-   dotall      the dotall flag
- Returns:      TRUE if character is of the type
- */
- static BOOL
- match_type(int type, int c, BOOL dotall)
- {
- #ifdef DEBUG
- if (isprint(c)) printf("matching subject %c against ", c);
-   else printf("matching subject \\x%02x against ", c);
- printf("%s\n", OP_names[type]);
- #endif
- switch(type)
-   {
-   case OP_ANY:            return dotall || c != '\n';
-   case OP_NOT_DIGIT:      return (pcre_ctypes[c] & ctype_digit) == 0;
-   case OP_DIGIT:          return (pcre_ctypes[c] & ctype_digit) != 0;
-   case OP_NOT_WHITESPACE: return (pcre_ctypes[c] & ctype_space) == 0;
-   case OP_WHITESPACE:     return (pcre_ctypes[c] & ctype_space) != 0;
-   case OP_NOT_WORDCHAR:   return (pcre_ctypes[c] & ctype_word) == 0;
-   case OP_WORDCHAR:       return (pcre_ctypes[c] & ctype_word) != 0;
-   }
- return FALSE;
- }
- /*************************************************
  *          Match a back-reference                *
  *************************************************/
-Line 2695 
 Returns:      TRUE if matched
+Line 2897 
 Returns:      TRUE if matched
  static BOOL
  match_ref(int offset, register const uschar *eptr, int length, match_data *md,
-   int ims)
+   unsigned long int ims)
  {
  const uschar *p = md->start_subject + md->offset_vector[offset];
-Line 2719 
 if (length > md->end_subject - eptr) ret
+Line 2921 
 if (length > md->end_subject - eptr) ret
  /* Separate the caselesss case for speed */
  if ((ims & PCRE_CASELESS) != 0)
-   { while (length-- > 0) if (pcre_lcc[*p++] != pcre_lcc[*eptr++]) return FALSE; }
+   {
+   while (length-- > 0)
+     if (md->lcc[*p++] != md->lcc[*eptr++]) return FALSE;
+   }
  else
    { while (length-- > 0) if (*p++ != *eptr++) return FALSE; }
-Line 2751 
 Returns:       TRUE if matched
+Line 2956 
 Returns:       TRUE if matched
  static BOOL
  match(register const uschar *eptr, register const uschar *ecode,
-   int offset_top, match_data *md, int ims, BOOL condassert, const uschar *eptrb)
+   int offset_top, match_data *md, unsigned long int ims, BOOL condassert,
+   const uschar *eptrb)
  {
- int original_ims = ims;   /* Save for resetting on ')' */
+ unsigned long int original_ims = ims;   /* Save for resetting on ')' */
  for (;;)
    {
-Line 2782 
 for (;;)
+Line 2988 
 for (;;)
      int number = op - OP_BRA;
      int offset = number << 1;
-     DPRINTF(("start bracket %d\n", number));
+ #ifdef DEBUG
+     printf("start bracket %d subject=", number);
+     pchars(eptr, 16, TRUE, md);
+     printf("\n");
+ #endif
      if (offset < md->offset_max)
        {
-Line 2864 
 for (;;)
+Line 3074 
 for (;;)
      ecode += 2;
      break;
-     /* End of the pattern */
+     /* End of the pattern. If PCRE_NOTEMPTY is set, fail if we have matched
+     an empty string - recursion will then try other alternatives, if any. */
      case OP_END:
+     if (md->notempty && eptr == md->start_match) return FALSE;
      md->end_match_ptr = eptr;          /* Record where we ended */
      md->end_offset_top = offset_top;   /* and how many extracts were taken */
      return TRUE;
-Line 3172 
 for (;;)
+Line 3384 
 for (;;)
      case OP_WORD_BOUNDARY:
        {
        BOOL prev_is_word = (eptr != md->start_subject) &&
-         ((pcre_ctypes[eptr[-1]] & ctype_word) != 0);
+         ((md->ctypes[eptr[-1]] & ctype_word) != 0);
        BOOL cur_is_word = (eptr < md->end_subject) &&
-         ((pcre_ctypes[*eptr] & ctype_word) != 0);
+         ((md->ctypes[*eptr] & ctype_word) != 0);
        if ((*ecode++ == OP_WORD_BOUNDARY)?
             cur_is_word == prev_is_word : cur_is_word != prev_is_word)
          return FALSE;
-Line 3191 
 for (;;)
+Line 3403 
 for (;;)
      break;
      case OP_NOT_DIGIT:
-     if (eptr >= md->end_subject || (pcre_ctypes[*eptr++] & ctype_digit) != 0)
+     if (eptr >= md->end_subject ||
+        (md->ctypes[*eptr++] & ctype_digit) != 0)
        return FALSE;
      ecode++;
      break;
      case OP_DIGIT:
-     if (eptr >= md->end_subject || (pcre_ctypes[*eptr++] & ctype_digit) == 0)
+     if (eptr >= md->end_subject ||
+        (md->ctypes[*eptr++] & ctype_digit) == 0)
        return FALSE;
      ecode++;
      break;
      case OP_NOT_WHITESPACE:
-     if (eptr >= md->end_subject || (pcre_ctypes[*eptr++] & ctype_space) != 0)
+     if (eptr >= md->end_subject ||
+        (md->ctypes[*eptr++] & ctype_space) != 0)
        return FALSE;
      ecode++;
      break;
      case OP_WHITESPACE:
-     if (eptr >= md->end_subject || (pcre_ctypes[*eptr++] & ctype_space) == 0)
+     if (eptr >= md->end_subject ||
+        (md->ctypes[*eptr++] & ctype_space) == 0)
        return FALSE;
      ecode++;
      break;
      case OP_NOT_WORDCHAR:
-     if (eptr >= md->end_subject || (pcre_ctypes[*eptr++] & ctype_word) != 0)
+     if (eptr >= md->end_subject ||
+        (md->ctypes[*eptr++] & ctype_word) != 0)
        return FALSE;
      ecode++;
      break;
      case OP_WORDCHAR:
-     if (eptr >= md->end_subject || (pcre_ctypes[*eptr++] & ctype_word) == 0)
+     if (eptr >= md->end_subject ||
+        (md->ctypes[*eptr++] & ctype_word) == 0)
        return FALSE;
      ecode++;
      break;
-Line 3453 
 for (;;)
+Line 3671 
 for (;;)
        if (length > md->end_subject - eptr) return FALSE;
        if ((ims & PCRE_CASELESS) != 0)
          {
-         while (length-- > 0) if (pcre_lcc[*ecode++] != pcre_lcc[*eptr++]) return FALSE;
+         while (length-- > 0)
+           if (md->lcc[*ecode++] != md->lcc[*eptr++])
+             return FALSE;
          }
        else
          {
-Line 3510 
 for (;;)
+Line 3730 
 for (;;)
      if ((ims & PCRE_CASELESS) != 0)
        {
-       c = pcre_lcc[c];
+       c = md->lcc[c];
-       for (i = 1; i <= min; i++) if (c != pcre_lcc[*eptr++]) return FALSE;
+       for (i = 1; i <= min; i++)
+         if (c != md->lcc[*eptr++]) return FALSE;
        if (min == max) continue;
        if (minimize)
          {
-Line 3519 
 for (;;)
+Line 3740 
 for (;;)
            {
            if (match(eptr, ecode, offset_top, md, ims, FALSE, eptrb))
              return TRUE;
-           if (i >= max || eptr >= md->end_subject || c != pcre_lcc[*eptr++])
+           if (i >= max || eptr >= md->end_subject ||
+               c != md->lcc[*eptr++])
              return FALSE;
            }
          /* Control never gets here */
-Line 3529 
 for (;;)
+Line 3751 
 for (;;)
          const uschar *pp = eptr;
          for (i = min; i < max; i++)
            {
-           if (eptr >= md->end_subject || c != pcre_lcc[*eptr]) break;
+           if (eptr >= md->end_subject || c != md->lcc[*eptr]) break;
            eptr++;
            }
          while (eptr >= pp)
-Line 3579 
 for (;;)
+Line 3801 
 for (;;)
      ecode++;
      if ((ims & PCRE_CASELESS) != 0)
        {
-       if (pcre_lcc[*ecode++] == pcre_lcc[*eptr++]) return FALSE;
+       if (md->lcc[*ecode++] == md->lcc[*eptr++]) return FALSE;
        }
      else
        {
-Line 3639 
 for (;;)
+Line 3861 
 for (;;)
      if ((ims & PCRE_CASELESS) != 0)
        {
-       c = pcre_lcc[c];
+       c = md->lcc[c];
-       for (i = 1; i <= min; i++) if (c == pcre_lcc[*eptr++]) return FALSE;
+       for (i = 1; i <= min; i++)
+         if (c == md->lcc[*eptr++]) return FALSE;
        if (min == max) continue;
        if (minimize)
          {
-Line 3648 
 for (;;)
+Line 3871 
 for (;;)
            {
            if (match(eptr, ecode, offset_top, md, ims, FALSE, eptrb))
              return TRUE;
-           if (i >= max || eptr >= md->end_subject || c == pcre_lcc[*eptr++])
+           if (i >= max || eptr >= md->end_subject ||
+               c == md->lcc[*eptr++])
              return FALSE;
            }
          /* Control never gets here */
-Line 3658 
 for (;;)
+Line 3882 
 for (;;)
          const uschar *pp = eptr;
          for (i = min; i < max; i++)
            {
-           if (eptr >= md->end_subject || c == pcre_lcc[*eptr]) break;
+           if (eptr >= md->end_subject || c == md->lcc[*eptr]) break;
            eptr++;
            }
          while (eptr >= pp)
-Line 3752 
 for (;;)
+Line 3976 
 for (;;)
        case OP_NOT_DIGIT:
        for (i = 1; i <= min; i++)
-         if ((pcre_ctypes[*eptr++] & ctype_digit) != 0) return FALSE;
+         if ((md->ctypes[*eptr++] & ctype_digit) != 0) return FALSE;
        break;
        case OP_DIGIT:
        for (i = 1; i <= min; i++)
-         if ((pcre_ctypes[*eptr++] & ctype_digit) == 0) return FALSE;
+         if ((md->ctypes[*eptr++] & ctype_digit) == 0) return FALSE;
        break;
        case OP_NOT_WHITESPACE:
        for (i = 1; i <= min; i++)
-         if ((pcre_ctypes[*eptr++] & ctype_space) != 0) return FALSE;
+         if ((md->ctypes[*eptr++] & ctype_space) != 0) return FALSE;
        break;
        case OP_WHITESPACE:
        for (i = 1; i <= min; i++)
-         if ((pcre_ctypes[*eptr++] & ctype_space) == 0) return FALSE;
+         if ((md->ctypes[*eptr++] & ctype_space) == 0) return FALSE;
        break;
        case OP_NOT_WORDCHAR:
-       for (i = 1; i <= min; i++) if ((pcre_ctypes[*eptr++] & ctype_word) != 0)
+       for (i = 1; i <= min; i++)
-         return FALSE;
+         if ((md->ctypes[*eptr++] & ctype_word) != 0)
+           return FALSE;
        break;
        case OP_WORDCHAR:
-       for (i = 1; i <= min; i++) if ((pcre_ctypes[*eptr++] & ctype_word) == 0)
+       for (i = 1; i <= min; i++)
-         return FALSE;
+         if ((md->ctypes[*eptr++] & ctype_word) == 0)
+           return FALSE;
        break;
        }
-Line 3786 
 for (;;)
+Line 4012 
 for (;;)
      if (min == max) continue;
      /* If minimizing, we have to test the rest of the pattern before each
-     subsequent match, so inlining isn't much help; just use the function. */
+     subsequent match. */
      if (minimize)
        {
        for (i = min;; i++)
          {
          if (match(eptr, ecode, offset_top, md, ims, FALSE, eptrb)) return TRUE;
-         if (i >= max || eptr >= md->end_subject ||
+         if (i >= max || eptr >= md->end_subject) return FALSE;
-           !match_type(ctype, *eptr++, (ims & PCRE_DOTALL) != 0))
-             return FALSE;
+         c = *eptr++;
+         switch(ctype)
+           {
+           case OP_ANY:
+           if ((ims & PCRE_DOTALL) == 0 && c == '\n') return FALSE;
+           break;
+           case OP_NOT_DIGIT:
+           if ((md->ctypes[c] & ctype_digit) != 0) return FALSE;
+           break;
+           case OP_DIGIT:
+           if ((md->ctypes[c] & ctype_digit) == 0) return FALSE;
+           break;
+           case OP_NOT_WHITESPACE:
+           if ((md->ctypes[c] & ctype_space) != 0) return FALSE;
+           break;
+           case OP_WHITESPACE:
+           if  ((md->ctypes[c] & ctype_space) == 0) return FALSE;
+           break;
+           case OP_NOT_WORDCHAR:
+           if ((md->ctypes[c] & ctype_word) != 0) return FALSE;
+           break;
+           case OP_WORDCHAR:
+           if ((md->ctypes[c] & ctype_word) == 0) return FALSE;
+           break;
+           }
          }
        /* Control never gets here */
        }
-Line 3828 
 for (;;)
+Line 4084 
 for (;;)
          case OP_NOT_DIGIT:
          for (i = min; i < max; i++)
            {
-           if (eptr >= md->end_subject || (pcre_ctypes[*eptr] & ctype_digit) != 0)
+           if (eptr >= md->end_subject || (md->ctypes[*eptr] & ctype_digit) != 0)
              break;
            eptr++;
            }
-Line 3837 
 for (;;)
+Line 4093 
 for (;;)
          case OP_DIGIT:
          for (i = min; i < max; i++)
            {
-           if (eptr >= md->end_subject || (pcre_ctypes[*eptr] & ctype_digit) == 0)
+           if (eptr >= md->end_subject || (md->ctypes[*eptr] & ctype_digit) == 0)
              break;
            eptr++;
            }
-Line 3846 
 for (;;)
+Line 4102 
 for (;;)
          case OP_NOT_WHITESPACE:
          for (i = min; i < max; i++)
            {
-           if (eptr >= md->end_subject || (pcre_ctypes[*eptr] & ctype_space) != 0)
+           if (eptr >= md->end_subject || (md->ctypes[*eptr] & ctype_space) != 0)
              break;
            eptr++;
            }
-Line 3855 
 for (;;)
+Line 4111 
 for (;;)
          case OP_WHITESPACE:
          for (i = min; i < max; i++)
            {
-           if (eptr >= md->end_subject || (pcre_ctypes[*eptr] & ctype_space) == 0)
+           if (eptr >= md->end_subject || (md->ctypes[*eptr] & ctype_space) == 0)
              break;
            eptr++;
            }
-Line 3864 
 for (;;)
+Line 4120 
 for (;;)
          case OP_NOT_WORDCHAR:
          for (i = min; i < max; i++)
            {
-           if (eptr >= md->end_subject || (pcre_ctypes[*eptr] & ctype_word) != 0)
+           if (eptr >= md->end_subject || (md->ctypes[*eptr] & ctype_word) != 0)
              break;
            eptr++;
            }
-Line 3873 
 for (;;)
+Line 4129 
 for (;;)
          case OP_WORDCHAR:
          for (i = min; i < max; i++)
            {
-           if (eptr >= md->end_subject || (pcre_ctypes[*eptr] & ctype_word) == 0)
+           if (eptr >= md->end_subject || (md->ctypes[*eptr] & ctype_word) == 0)
              break;
            eptr++;
            }
-Line 3919 
 Arguments:
+Line 4175 
 Arguments:
    external_extra  points to "hints" from pcre_study() 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         points to a vector of ints to be filled in with offsets
    offsetcount     the number of elements in the vector
-Line 3931 
 Returns:          > 0 => success; value
+Line 4188 
 Returns:          > 0 => success; value
  int
  pcre_exec(const pcre *external_re, const pcre_extra *external_extra,
-   const char *subject, int length, int options, int *offsets, int offsetcount)
+   const char *subject, int length, int start_offset, int options, int *offsets,
+   int offsetcount)
  {
  int resetcount, ocount;
  int first_char = -1;
- int ims = 0;
+ int req_char = -1;
+ int req_char2 = -1;
+ unsigned long int ims = 0;
  match_data match_block;
  const uschar *start_bits = NULL;
- const uschar *start_match = (const uschar *)subject;
+ const uschar *start_match = (const uschar *)subject + start_offset;
  const uschar *end_subject;
+ const uschar *req_char_ptr = start_match - 1;
  const real_pcre *re = (const real_pcre *)external_re;
  const real_pcre_extra *extra = (const real_pcre_extra *)external_extra;
  BOOL using_temporary_offsets = FALSE;
-Line 3960 
 match_block.endonly = (re->options & PCR
+Line 4221 
 match_block.endonly = (re->options & PCR
  match_block.notbol = (options & PCRE_NOTBOL) != 0;
  match_block.noteol = (options & PCRE_NOTEOL) != 0;
+ match_block.notempty = (options & PCRE_NOTEMPTY) != 0;
  match_block.errorcode = PCRE_ERROR_NOMATCH;     /* Default error */
+ match_block.lcc = re->tables + lcc_offset;
+ match_block.ctypes = re->tables + ctypes_offset;
  /* The ims options can vary during the matching as a result of the presence
  of (?ims) items in the pattern. They are kept in a local variable so that
  restoring at the exit of a group is easy. */
-Line 3997 
 in the pattern. */
+Line 4262 
 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. */
+ if (match_block.offset_vector != NULL)
+   {
+   register int *iptr = match_block.offset_vector + ocount;
+   register int *iend = iptr - resetcount/2 + 1;
+   while (--iptr >= iend) *iptr = -1;
+   }
  /* Set up the first character to match, if available. The first_char 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
-Line 4008 
 if (!anchored)
+Line 4284 
 if (!anchored)
    if ((re->options & PCRE_FIRSTSET) != 0)
      {
      first_char = re->first_char;
-     if ((ims & PCRE_CASELESS) != 0) first_char = pcre_lcc[first_char];
+     if ((ims & PCRE_CASELESS) != 0) first_char = match_block.lcc[first_char];
      }
    else
      if (!startline && extra != NULL &&
-Line 4016 
 if (!anchored)
+Line 4292 
 if (!anchored)
          start_bits = extra->start_bits;
    }
- /* Loop for unanchored matches; for anchored regexps the loop runs just once. */
+ /* For anchored or unanchored matches, there may be a "last known required
+ character" set. If the PCRE_CASELESS is set, implying that the match starts
+ caselessly, or if there are any changes of this flag within the regex, set up
+ both cases of the character. Otherwise set the two values the same, which will
+ avoid duplicate testing (which takes significant time). This covers the vast
+ majority of cases. It will be suboptimal when the case flag changes in a regex
+ and the required character in fact is caseful. */
+ if ((re->options & PCRE_REQCHSET) != 0)
+   {
+   req_char = re->req_char;
+   req_char2 = ((re->options & (PCRE_CASELESS | PCRE_ICHANGED)) != 0)?
+     (re->tables + fcc_offset)[req_char] : req_char;
+   }
+ /* Loop for handling unanchored repeated matching attempts; for anchored regexs
+ the loop runs just once. */
  do
    {
-Line 4033 
 do
+Line 4325 
 do
    if (first_char >= 0)
      {
      if ((ims & PCRE_CASELESS) != 0)
-       while (start_match < end_subject && pcre_lcc[*start_match] != first_char)
+       while (start_match < end_subject &&
+              match_block.lcc[*start_match] != first_char)
          start_match++;
      else
        while (start_match < end_subject && *start_match != first_char)
-Line 4051 
 do
+Line 4344 
 do
        }
      }
-   /* Or to a non-unique first char */
+   /* Or to a non-unique first char after study */
    else if (start_bits != NULL)
      {
-Line 4068 
 do
+Line 4361 
 do
    printf("\n");
  #endif
+   /* If req_char is set, we know that that character must appear in the subject
+   for the match to succeed. If the first character is set, req_char must be
+   later in the subject; otherwise the test starts at the match point. This
+   optimization can save a huge amount of backtracking in patterns with nested
+   unlimited repeats that aren't going to match. We don't know what the state of
+   case matching may be when this character is hit, so test for it in both its
+   cases if necessary. However, the different cased versions will not be set up
+   unless PCRE_CASELESS was given or the casing state changes within the regex.
+   Writing separate code makes it go faster, as does using an autoincrement and
+   backing off on a match. */
+   if (req_char >= 0)
+     {
+     register const uschar *p = start_match + ((first_char >= 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_char_ptr)
+       {
+       /* Do a single test if no case difference is set up */
+       if (req_char == req_char2)
+         {
+         while (p < end_subject)
+           {
+           if (*p++ == req_char) { p--; break; }
+           }
+         }
+       /* Otherwise test for either case */
+       else
+         {
+         while (p < end_subject)
+           {
+           register int pp = *p++;
+           if (pp == req_char || pp == req_char2) { p--; break; }
+           }
+         }
+       /* If we can't find the required character, break the matching loop */
+       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_char_ptr = p;
+       }
+     }
    /* When a match occurs, substrings will be set for all internal extractions;
    we just need to set up the whole thing as substring 0 before returning. If
    there were too many extractions, set the return code to zero. In the case
-Line 4075 
 do
+Line 4421 
 do
    those back references that we can. In this case there need not be overflow
    if certain parts of the pattern were not used. */
+   match_block.start_match = start_match;
    if (!match(start_match, re->code, 2, &match_block, ims, FALSE, start_match))
      continue;
-Line 4106 
 do
+Line 4453 
 do
    DPRINTF((">>>> returning %d\n", rc));
    return rc;
    }
+ /* This "while" is the end of the "do" above */
  while (!anchored &&
         match_block.errorcode == PCRE_ERROR_NOMATCH &&
         start_match++ < end_subject);

 Legend:



Removed from v.23
 


changed lines


 
Added in v.37
 Legend:



Removed from v.23
 


changed lines


 
Added in v.37
-Removed from v.23
+Added in v.37

	ViewVC Help
Powered by ViewVC 1.1.5