/[pcre]/code/tags/pcre-2.08a/pcre.c

Diff of /code/tags/pcre-2.08a/pcre.c

Parent Directory | Revision Log | View Patch Patch

-revision 19 by nigel,
Sat Feb 24 21:38:33 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 33 
 restrictions:
+Line 37 
 restrictions:
  /* #define DEBUG */
- /* Use a macro for debugging printing, 'cause that eliminates the the use
+ /* Use a macro for debugging printing, 'cause that eliminates the use of #ifdef
- of #ifdef inline, and there are *still* stupid compilers about that don't like
+ inline, and there are *still* stupid compilers about that don't like indented
- indented pre-processor statements. I suppose it's only been 10 years... */
+ pre-processor statements. I suppose it's only been 10 years... */
  #ifdef DEBUG
  #define DPRINTF(p) printf p
-Line 56 
 the external pcre header. */
+Line 60 
 the external pcre header. */
  #endif
+ /* Number of items on the nested bracket stacks at compile time. This should
+ not be set greater than 200. */
+ #define BRASTACK_SIZE 200
  /* Min and max values for the common repeats; for the maxima, 0 => infinity */
  static const char rep_min[] = { 0, 0, 1, 1, 0, 0 };
-Line 66 
 static const char rep_max[] = { 0, 0, 0,
+Line 76 
 static const char rep_max[] = { 0, 0, 0,
  #ifdef DEBUG
  static const char *OP_names[] = {
    "End", "\\A", "\\B", "\\b", "\\D", "\\d",
-   "\\S", "\\s", "\\W", "\\w", "Cut", "\\Z", "^", "$", "Any", "chars",
+   "\\S", "\\s", "\\W", "\\w", "\\Z", "\\z",
-   "not",
+   "Opt", "^", "$", "Any", "chars", "not",
    "*", "*?", "+", "+?", "?", "??", "{", "{", "{",
    "*", "*?", "+", "+?", "?", "??", "{", "{", "{",
    "*", "*?", "+", "+?", "?", "??", "{", "{", "{",
    "*", "*?", "+", "+?", "?", "??", "{", "{",
-   "class", "negclass", "Ref",
+   "class", "Ref",
-   "Alt", "Ket", "KetRmax", "KetRmin", "Assert", "Assert not", "Once",
+   "Alt", "Ket", "KetRmax", "KetRmin", "Assert", "Assert not",
+   "AssertB", "AssertB not", "Reverse", "Once", "Cond", "Cref",
    "Brazero", "Braminzero", "Bra"
  };
  #endif
-Line 93 
 static const short int escapes[] = {
+Line 104 
 static const short int escapes[] = {
    '`',      7, -ESC_b,      0, -ESC_d,     27,   '\f',      0,   /* ` - g */
 ,      0,      0,      0,      0,      0,   '\n',      0,   /* h - o */
 ,      0,   '\r', -ESC_s,   '\t',      0,      0, -ESC_w,   /* p - w */
-,      0,      0                                            /* x - z */
+,      0, -ESC_z                                            /* x - z */
  };
  /* Definition to allow mutual recursion */
  static BOOL
-   compile_regex(int, int *, uschar **, const uschar **, const char **);
+   compile_regex(int, int, int *, uschar **, const uschar **, const char **,
+     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 */
-   BOOL   offset_overflow;       /* Set if too many extractions */
-   BOOL   caseless;              /* Case-independent flag */
-   BOOL   runtime_caseless;      /* Caseless forced at run time */
-   BOOL   multiline;             /* Multiline flag */
-   BOOL   notbol;                /* NOTBOL flag */
-   BOOL   noteol;                /* NOTEOL flag */
-   BOOL   dotall;                /* Dot matches any char */
-   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 */
-   jmp_buf fail_env;             /* Environment for longjump() break out */
-   const uschar *end_match_ptr;  /* Subject position at end match */
-   int     end_offset_top;       /* Highwater mark at end of match */
- } match_data;
-Line 141 
 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 158 
 return PCRE_VERSION;
+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 177 
 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 218 
 while (length-- > 0)
+Line 225 
 while (length-- > 0)
  /*************************************************
- *         Check subpattern for empty operand     *
- *************************************************/
- /* This function checks a bracketed subpattern to see if any of the paths
- through it could match an empty string. This is used to diagnose an error if
- such a subpattern is followed by a quantifier with an unlimited upper bound.
- Argument:
-   code      points to the opening bracket
- Returns:    TRUE or FALSE
- */
- static BOOL
- could_be_empty(uschar *code)
- {
- do {
-   uschar *cc = code + 3;
-   /* Scan along the opcodes for this branch; as soon as we find something
-   that matches a non-empty string, break out and advance to test the next
-   branch. If we get to the end of the branch, return TRUE for the whole
-   sub-expression. */
-   for (;;)
-     {
-     /* Test an embedded subpattern; if it could not be empty, break the
-     loop. Otherwise carry on in the branch. */
-     if ((int)(*cc) >= OP_BRA || (int)(*cc) == OP_ONCE)
-       {
-       if (!could_be_empty(cc)) break;
-       do cc += (cc[1] << 8) + cc[2]; while (*cc == OP_ALT);
-       cc += 3;
-       }
-     else switch (*cc)
-       {
-       /* Reached end of a branch: the subpattern may match the empty string */
-       case OP_ALT:
-       case OP_KET:
-       case OP_KETRMAX:
-       case OP_KETRMIN:
-       return TRUE;
-       /* Skip over entire bracket groups with zero lower bound */
-       case OP_BRAZERO:
-       case OP_BRAMINZERO:
-       cc++;
-       /* Fall through */
-       /* Skip over assertive subpatterns */
-       case OP_ASSERT:
-       case OP_ASSERT_NOT:
-       do cc += (cc[1] << 8) + cc[2]; while (*cc == OP_ALT);
-       cc += 3;
-       break;
-       /* Skip over things that don't match chars */
-       case OP_SOD:
-       case OP_EOD:
-       case OP_CIRC:
-       case OP_DOLL:
-       case OP_NOT_WORD_BOUNDARY:
-       case OP_WORD_BOUNDARY:
-       cc++;
-       break;
-       /* Skip over simple repeats with zero lower bound */
-       case OP_STAR:
-       case OP_MINSTAR:
-       case OP_QUERY:
-       case OP_MINQUERY:
-       case OP_NOTSTAR:
-       case OP_NOTMINSTAR:
-       case OP_NOTQUERY:
-       case OP_NOTMINQUERY:
-       case OP_TYPESTAR:
-       case OP_TYPEMINSTAR:
-       case OP_TYPEQUERY:
-       case OP_TYPEMINQUERY:
-       cc += 2;
-       break;
-       /* Skip over UPTOs (lower bound is zero) */
-       case OP_UPTO:
-       case OP_MINUPTO:
-       case OP_TYPEUPTO:
-       case OP_TYPEMINUPTO:
-       cc += 4;
-       break;
-       /* Check a class or a back reference for a zero minimum */
-       case OP_CLASS:
-       case OP_NEGCLASS:
-       case OP_REF:
-       cc += (*cc == OP_REF)? 2 : 33;
-       switch (*cc)
-         {
-         case OP_CRSTAR:
-         case OP_CRMINSTAR:
-         case OP_CRQUERY:
-         case OP_CRMINQUERY:
-         cc++;
-         break;
-         case OP_CRRANGE:
-         case OP_CRMINRANGE:
-         if ((cc[1] << 8) + cc[2] != 0) goto NEXT_BRANCH;
-         cc += 3;
-         break;
-         default:
-         goto NEXT_BRANCH;
-         }
-       break;
-       /* Anything else matches at least one character */
-       default:
-       goto NEXT_BRANCH;
-       }
-     }
-   NEXT_BRANCH:
-   code += (code[1] << 8) + code[2];
-   }
- while (*code == OP_ALT);
- /* No branches match the empty string */
- return FALSE;
- }
- /*************************************************
  *            Handle escapes                      *
  *************************************************/
-Line 378 
 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 386 
 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 429 
 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 455 
 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 464 
 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 482 
 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)
        {
-       case 'X':
-       c = -ESC_X;      /* This could be a lookup if it ever got into Perl */
-       break;
        default:
        *errorptr = ERR3;
        break;
-Line 522 
 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 557 
 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 600 
 return p;
+Line 464 
 return p;
  /*************************************************
+ *        Find the fixed length of a pattern      *
+ *************************************************/
+ /* Scan a pattern and compute the fixed length of subject that will match it,
+ if the length is fixed. This is needed for dealing with backward assertions.
+ Arguments:
+   code     points to the start of the pattern (the bracket)
+ Returns:   the fixed length, or -1 if there is no fixed length
+ */
+ static int
+ find_fixedlength(uschar *code)
+ {
+ int length = -1;
+ register int branchlength = 0;
+ register uschar *cc = code + 3;
+ /* Scan along the opcodes for this branch. If we get to the end of the
+ branch, check the length against that of the other branches. */
+ for (;;)
+   {
+   int d;
+   register int op = *cc;
+   if (op >= OP_BRA) op = OP_BRA;
+   switch (op)
+     {
+     case OP_BRA:
+     case OP_ONCE:
+     case OP_COND:
+     d = find_fixedlength(cc);
+     if (d < 0) return -1;
+     branchlength += d;
+     do cc += (cc[1] << 8) + cc[2]; while (*cc == OP_ALT);
+     cc += 3;
+     break;
+     /* Reached end of a branch; if it's a ket it is the end of a nested
+     call. If it's ALT it is an alternation in a nested call. If it is
+     END it's the end of the outer call. All can be handled by the same code. */
+     case OP_ALT:
+     case OP_KET:
+     case OP_KETRMAX:
+     case OP_KETRMIN:
+     case OP_END:
+     if (length < 0) length = branchlength;
+       else if (length != branchlength) return -1;
+     if (*cc != OP_ALT) return length;
+     cc += 3;
+     branchlength = 0;
+     break;
+     /* Skip over assertive subpatterns */
+     case OP_ASSERT:
+     case OP_ASSERT_NOT:
+     case OP_ASSERTBACK:
+     case OP_ASSERTBACK_NOT:
+     do cc += (cc[1] << 8) + cc[2]; while (*cc == OP_ALT);
+     cc += 3;
+     break;
+     /* Skip over things that don't match chars */
+     case OP_REVERSE:
+     cc++;
+     /* Fall through */
+     case OP_CREF:
+     case OP_OPT:
+     cc++;
+     /* Fall through */
+     case OP_SOD:
+     case OP_EOD:
+     case OP_EODN:
+     case OP_CIRC:
+     case OP_DOLL:
+     case OP_NOT_WORD_BOUNDARY:
+     case OP_WORD_BOUNDARY:
+     cc++;
+     break;
+     /* Handle char strings */
+     case OP_CHARS:
+     branchlength += *(++cc);
+     cc += *cc + 1;
+     break;
+     /* Handle exact repetitions */
+     case OP_EXACT:
+     case OP_TYPEEXACT:
+     branchlength += (cc[1] << 8) + cc[2];
+     cc += 4;
+     break;
+     /* Handle single-char matchers */
+     case OP_NOT_DIGIT:
+     case OP_DIGIT:
+     case OP_NOT_WHITESPACE:
+     case OP_WHITESPACE:
+     case OP_NOT_WORDCHAR:
+     case OP_WORDCHAR:
+     case OP_ANY:
+     branchlength++;
+     cc++;
+     break;
+     /* Check a class for variable quantification */
+     case OP_CLASS:
+     cc += (*cc == OP_REF)? 2 : 33;
+     switch (*cc)
+       {
+       case OP_CRSTAR:
+       case OP_CRMINSTAR:
+       case OP_CRQUERY:
+       case OP_CRMINQUERY:
+       return -1;
+       case OP_CRRANGE:
+       case OP_CRMINRANGE:
+       if ((cc[1] << 8) + cc[2] != (cc[3] << 8) + cc[4]) return -1;
+       branchlength += (cc[1] << 8) + cc[2];
+       cc += 5;
+       break;
+       default:
+       branchlength++;
+       }
+     break;
+     /* Anything else is variable length */
+     default:
+     return -1;
+     }
+   }
+ /* Control never gets here */
+ }
+ /*************************************************
  *           Compile one branch                   *
  *************************************************/
  /* Scan the pattern, compiling it into the code vector.
  Arguments:
-   options    the option bits
+   options      the option bits
-   bracket    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
+   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)
+   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;
  const uschar *ptr = *ptrptr;
- const uschar *oldptr;
+ const uschar *tempptr;
  uschar *previous = NULL;
  uschar class[32];
-Line 636 
 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++)
    {
    BOOL negate_class;
-   int  class_charcount;
+   int class_charcount;
-   int  class_lastchar;
+   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 690 
 for (;; ptr++)
+Line 726 
 for (;; ptr++)
      case '[':
      previous = code;
+     *code++ = OP_CLASS;
-     /* If the first character is '^', set the negation flag, and use a
+     /* If the first character is '^', set the negation flag and skip it. */
-     different opcode. This only matters if caseless matching is specified at
-     runtime. */
      if ((c = *(++ptr)) == '^')
        {
        negate_class = TRUE;
-       *code++ = OP_NEGCLASS;
        c = *(++ptr);
        }
-     else
+     else negate_class = FALSE;
-       {
-       negate_class = FALSE;
-       *code++ = OP_CLASS;
-       }
      /* Keep a count of chars so that we can optimize the case of just a single
      character. */
-Line 741 
 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 803 
 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 825 
 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 840 
 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 887 
 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 921 
 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 964 
 for (;; ptr++)
+Line 996 
 for (;; ptr++)
      create a suitable repeat item. The code is shared with single-character
      repeats by adding a suitable offset into repeat_type. */
-     else if ((int)*previous < OP_EOD || *previous == OP_ANY)
+     else if ((int)*previous < OP_EODN || *previous == OP_ANY)
        {
        op_type = OP_TYPESTAR - OP_STAR;  /* Use type opcodes */
        c = *previous;
        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 1008 
 for (;; ptr++)
+Line 1048 
 for (;; ptr++)
          /* If the mininum is 1 and the previous item was a character string,
          we either have to put back the item that got cancelled if the string
          length was 1, or add the character back onto the end of a longer
-         string. For a character type nothing need be done; it will just get put
+         string. For a character type nothing need be done; it will just get
-         back naturally. */
+         put back naturally. Note that the final character is always going to
+         get added below. */
          else if (*previous == OP_CHARS)
            {
            if (code == previous) code += 2; else previous[1]++;
            }
+         /*  For a single negated character we also have to put back the
+         item that got cancelled. */
+         else if (*previous == OP_NOT) code++;
          /* If the maximum is unlimited, insert an OP_STAR. */
          if (repeat_max < 0)
-Line 1042 
 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_NEGCLASS ||
+     else if (*previous == OP_CLASS || *previous == OP_REF)
-              *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 1065 
 for (;; ptr++)
+Line 1115 
 for (;; ptr++)
        }
      /* If previous was a bracket group, we may have to replicate it in certain
-     cases. If the maximum repeat count is unlimited, check that the bracket
+     cases. */
-     group cannot match the empty string, and diagnose an error if it can. */
-     else if ((int)*previous >= OP_BRA)
+     else if ((int)*previous >= OP_BRA || (int)*previous == OP_ONCE ||
+              (int)*previous == OP_COND)
        {
-       int i;
+       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
+       from the current code pointer. There may be an OP_OPT setting following
+       the final KET, so we can't find the end just by going back from the code
+       pointer. */
+       if (repeat_max == -1)
+         {
+         register uschar *ket = previous;
+         do ket += (ket[1] << 8) + ket[2]; while (*ket != OP_KET);
+         ketoffset = code - ket;
+         }
+       /* The case of a zero minimum is special because of the need to stick
+       OP_BRAZERO in front of it, and because the group appears once in the
+       data, whereas in other cases it appears the minimum number of times. For
+       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_max == -1 && could_be_empty(previous))
+       if (repeat_min == 0)
          {
-         *errorptr = ERR10;
+         /* If we set up a required char from the bracket, we must back off
-         goto FAILED;
+         to the previous value and reset the countlits value too. */
-         }
+         if (subcountlits > 0)
+           {
+           *reqchar = prevreqchar;
+           *countlits -= subcountlits;
+           }
-       /* If the minimum is greater than zero, and the maximum is unlimited or
+         /* If the maximum is also zero, we just omit the group from the output
-       equal to the minimum, the first copy remains where it is, and is
+         altogether. */
-       replicated up to the minimum number of times. This case includes the +
-       repeat, but of course no replication is needed in that case. */
-       if (repeat_min > 0 && (repeat_max == -1 || repeat_max == repeat_min))
+         if (repeat_max == 0)
-         {
-         for (i = 1; i < repeat_min; i++)
            {
-           memcpy(code, previous, len);
+           code = previous;
-           code += len;
+           goto END_REPEAT;
            }
-         }
-       /* If the minimum is zero, stick BRAZERO in front of the first copy.
+         /* If the maximum is 1 or unlimited, we just have to stick in the
-       Then, if there is a fixed upper limit, replicated up to that many times,
+         BRAZERO and do no more at this point. */
-       sticking BRAZERO in front of all the optional ones. */
-       else
+         if (repeat_max <= 1)
-         {
-         if (repeat_min == 0)
            {
            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. */
+       /* If the maximum is unlimited, set a repeater in the final copy. We
+       can't just offset backwards from the current code point, because we
+       don't know if there's been an options resetting after the ket. The
+       correct offset was computed above. */
-       if (repeat_max == -1) code[-3] = OP_KETRMAX + repeat_type;
+       else code[-ketoffset] = OP_KETRMAX + repeat_type;
        }
      /* Else there's some kind of shambles */
-Line 1135 
 for (;; ptr++)
+Line 1279 
 for (;; ptr++)
      /* In all case we no longer have a previous item. */
+     END_REPEAT:
      previous = NULL;
      break;
-     /* Start of nested bracket sub-expression, or comment or lookahead.
+     /* Start of nested bracket sub-expression, or comment or lookahead or
-     First deal with special things that can come after a bracket; all are
+     lookbehind or option setting or condition. First deal with special things
-     introduced by ?, and the appearance of any of them means that this is not a
+     that can come after a bracket; all are introduced by ?, and the appearance
-     referencing group. They were checked for validity in the first pass over
+     of any of them means that this is not a referencing group. They were
-     the string, so we don't have to check for syntax errors here.  */
+     checked for validity in the first pass over the string, so we don't have to
+     check for syntax errors here.  */
      case '(':
-     previous = code;              /* Only real brackets can be repeated */
+     newoptions = options;
+     condref = -1;
      if (*(++ptr) == '?')
        {
-       bravalue = OP_BRA;
+       int set, unset;
+       int *optset;
        switch (*(++ptr))
          {
-         case '#':
+         case '#':                 /* Comment; skip to ket */
-         case 'i':
-         case 'm':
-         case 's':
-         case 'x':
-         case 'U':
-         case 'X':
          ptr++;
          while (*ptr != ')') ptr++;
-         previous = NULL;
          continue;
          case ':':                 /* Non-extracting bracket */
+         bravalue = OP_BRA;
          ptr++;
          break;
-         case '=':                 /* Assertions can't be repeated */
+         case '(':
+         bravalue = OP_COND;       /* Conditional group */
+         if ((cd->ctypes[*(++ptr)] & ctype_digit) != 0)
+           {
+           condref = *ptr - '0';
+           while (*(++ptr) != ')') condref = condref*10 + *ptr - '0';
+           ptr++;
+           }
+         else ptr--;
+         break;
+         case '=':                 /* Positive lookahead */
          bravalue = OP_ASSERT;
          ptr++;
-         previous = NULL;
          break;
-         case '!':
+         case '!':                 /* Negative lookahead */
          bravalue = OP_ASSERT_NOT;
          ptr++;
-         previous = NULL;
          break;
-         case '>':                         /* "Match once" brackets */
+         case '<':                 /* Lookbehinds */
-         if ((options & PCRE_EXTRA) != 0)  /* Not yet standard */
+         switch (*(++ptr))
            {
-           bravalue = OP_ONCE;
+           case '=':               /* Positive lookbehind */
+           bravalue = OP_ASSERTBACK;
            ptr++;
-           previous = NULL;
            break;
+           case '!':               /* Negative lookbehind */
+           bravalue = OP_ASSERTBACK_NOT;
+           ptr++;
+           break;
+           default:                /* Syntax error */
+           *errorptr = ERR24;
+           goto FAILED;
            }
-         /* Else fall through */
+         break;
-         default:
+         case '>':                 /* One-time brackets */
-         *errorptr = ERR12;
+         bravalue = OP_ONCE;
-         goto FAILED;
+         ptr++;
+         break;
+         default:                  /* Option setting */
+         set = unset = 0;
+         optset = &set;
+         while (*ptr != ')' && *ptr != ':')
+           {
+           switch (*ptr++)
+             {
+             case '-': optset = &unset; break;
+             case 'i': *optset |= PCRE_CASELESS; break;
+             case 'm': *optset |= PCRE_MULTILINE; break;
+             case 's': *optset |= PCRE_DOTALL; break;
+             case 'x': *optset |= PCRE_EXTENDED; break;
+             case 'U': *optset |= PCRE_UNGREEDY; break;
+             case 'X': *optset |= PCRE_EXTRA; break;
+             default:
+             *errorptr = ERR12;
+             goto FAILED;
+             }
+           }
+         /* Set up the changed option bits, but don't change anything yet. */
+         newoptions = (options | set) & (~unset);
+         /* If the options ended with ')' this is not the start of a nested
+         group with option changes, so the options change at this level. At top
+         level there is nothing else to be done (the options will in fact have
+         been set from the start of compiling as a result of the first pass) but
+         at an inner level we must compile code to change the ims options if
+         necessary, and pass the new setting back so that it can be put at the
+         start of any following branches, and when this group ends, a resetting
+         item can be compiled. */
+         if (*ptr == ')')
+           {
+           if ((options & PCRE_INGROUP) != 0 &&
+               (options & PCRE_IMS) != (newoptions & PCRE_IMS))
+             {
+             *code++ = OP_OPT;
+             *code++ = *optchanged = newoptions & PCRE_IMS;
+             }
+           options = newoptions;  /* Change options at this level */
+           previous = NULL;       /* This item can't be repeated */
+           continue;              /* It is complete */
+           }
+         /* If the options ended with ':' we are heading into a nested group
+         with possible change of options. Such groups are non-capturing and are
+         not assertions of any kind. All we need to do is skip over the ':';
+         the newoptions value is handled below. */
+         bravalue = OP_BRA;
+         ptr++;
          }
        }
-     /* Else we have a referencing group */
+     /* Else we have a referencing group; adjust the opcode. */
      else
        {
-Line 1209 
 for (;; ptr++)
+Line 1428 
 for (;; ptr++)
        bravalue = OP_BRA + *brackets;
        }
-     /* Process nested bracketed re; at end pointer is on the bracket. We copy
+     /* Process nested bracketed re. Assertions may not be repeated, but other
-     code into a non-register variable in order to be able to pass its address
+     kinds can be. We copy code into a non-register variable in order to be able
-     because some compilers complain otherwise. */
+     to pass its address because some compilers complain otherwise. Pass in a
+     new setting for the ims options if they have changed. */
+     previous = (bravalue >= OP_ONCE)? code : NULL;
      *code = bravalue;
+     tempcode = code;
+     if (!compile_regex(
+          options | PCRE_INGROUP,       /* Set for all nested groups */
+          ((options & PCRE_IMS) != (newoptions & PCRE_IMS))?
+            newoptions & PCRE_IMS : -1, /* Pass ims options if changed */
+          brackets,                     /* Bracket level */
+          &tempcode,                    /* Where to put code (updated) */
+          &ptr,                         /* Input pointer (updated) */
+          errorptr,                     /* Where to put an error message */
+          (bravalue == OP_ASSERTBACK ||
+           bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
+          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
+     group, while tempcode has been updated to point past the end of the group
+     and any option resetting that may follow it. The pattern pointer (ptr)
+     is on the bracket. */
+     /* If this is a conditional bracket, check that there are no more than
+     two branches in the group. */
+     if (bravalue == OP_COND)
        {
-       uschar *mcode = code;
+       uschar *tc = code;
-       if (!compile_regex(options, brackets, &mcode, &ptr, errorptr))
+       condcount = 0;
+       do {
+          condcount++;
+          tc += (tc[1] << 8) | tc[2];
+          }
+       while (*tc != OP_KET);
+       if (condcount > 2)
+         {
+         *errorptr = ERR27;
          goto FAILED;
-       code = mcode;
+         }
        }
+     /* 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;
+     /* Error if hit end of pattern */
      if (*ptr != ')')
        {
        *errorptr = ERR14;
-Line 1233 
 for (;; ptr++)
+Line 1514 
 for (;; ptr++)
      for validity in the pre-compiling pass. */
      case '\\':
-     oldptr = ptr;
+     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 1247 
 for (;; ptr++)
+Line 1528 
 for (;; ptr++)
        {
        if (-c >= ESC_REF)
          {
-         int refnum = -c - ESC_REF;
-         if (*brackets < refnum)
-           {
-           *errorptr = ERR15;
-           goto FAILED;
-           }
          previous = code;
          *code++ = OP_REF;
-         *code++ = refnum;
+         *code++ = -c - ESC_REF;
          }
        else
          {
-         previous = (-c > ESC_b && -c < ESC_X)? code : NULL;
+         previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
          *code++ = -c;
          }
        continue;
-Line 1267 
 for (;; ptr++)
+Line 1542 
 for (;; ptr++)
      /* Data character: reset and fall through */
-     ptr = oldptr;
+     ptr = tempptr;
      c = '\\';
      /* Handle a run of data characters until a metacharacter is encountered.
-Line 1285 
 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 1300 
 for (;; ptr++)
+Line 1575 
 for (;; ptr++)
        if (c == '\\')
          {
-         oldptr = ptr;
+         tempptr = ptr;
-         c = check_escape(&ptr, errorptr, *brackets, options, FALSE);
+         c = check_escape(&ptr, errorptr, *brackets, options, FALSE, cd);
-         if (c < 0) { ptr = oldptr; break; }
+         if (c < 0) { ptr = tempptr; break; }
          }
        /* Ordinary character or single-char escape */
-Line 1313 
 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 1343 
 return FALSE;
+Line 1624 
 return FALSE;
  /* On entry, ptr is pointing past the bracket character, but on return
  it points to the closing bracket, or vertical bar, or end of string.
  The code variable is pointing at the byte into which the BRA operator has been
- stored.
+ stored. If the ims options are changed at the start (for a (?ims: group) or
+ during any branch, we need to insert an OP_OPT item at the start of every
+ following branch to ensure they get set correctly at run time, and also pass
+ the new options into every subsequent branch compile.
  Argument:
-   options   the option bits
+   options     the option bits
-   brackets  -> int containing the number of extracting brackets used
+   optchanged  new ims options to set as if (?ims) were at the start, or -1
-   codeptr   -> the address of the current code pointer
+                for no change
-   ptrptr    -> the address of the current pattern pointer
+   brackets    -> int containing the number of extracting brackets used
-   errorptr  -> pointer to error message
+   codeptr     -> the address of the current code pointer
+   ptrptr      -> the address of the current pattern pointer
+   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
+ Returns:      TRUE on success
  */
  static BOOL
- compile_regex(int options, int *brackets, uschar **codeptr,
+ compile_regex(int options, int optchanged, int *brackets, uschar **codeptr,
-   const uschar **ptrptr, const char **errorptr)
+   const uschar **ptrptr, const char **errorptr, BOOL lookbehind, int condref,
+   int *reqchar, int *countlits, compile_data *cd)
  {
  const uschar *ptr = *ptrptr;
  uschar *code = *codeptr;
+ 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
+ number as an OP_CREF item. */
+ if (condref > 0)
+   {
+   *code++ = OP_CREF;
+   *code++ = condref;
+   }
+ /* Loop for each alternative branch */
  for (;;)
    {
    int length;
-   uschar *last_branch = code;
-   code += 3;
+   /* Handle change of options */
-   if (!compile_branch(options, brackets, &code, &ptr, errorptr))
+   if (optchanged >= 0)
+     {
+     *code++ = OP_OPT;
+     *code++ = optchanged;
+     options = (options & ~PCRE_IMS) | optchanged;
+     }
+   /* Set up dummy OP_REVERSE if lookbehind assertion */
+   if (lookbehind)
+     {
+     *code++ = OP_REVERSE;
+     reverse_count = code;
+     *code++ = 0;
+     *code++ = 0;
+     }
+   /* Now compile the branch */
+   if (!compile_branch(options, brackets, &code, &ptr, errorptr, &optchanged,
+       &branchreqchar, &branchcountlits, cd))
      {
      *ptrptr = ptr;
      return FALSE;
-Line 1381 
 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. */
+   if (lookbehind)
+     {
+     *code = OP_END;
+     length = find_fixedlength(last_branch);
+     DPRINTF(("fixed length = %d\n", length));
+     if (length < 0)
+       {
+       *errorptr = ERR25;
+       *ptrptr = ptr;
+       return FALSE;
+       }
+     reverse_count[0] = (length >> 8);
+     reverse_count[1] = length & 255;
+     }
    /* Reached end of expression, either ')' or end of pattern. Insert a
    terminating ket and the length of the whole bracketed item, and return,
-   leaving the pointer at the terminating char. */
+   leaving the pointer at the terminating char. If any of the ims options
+   were changed inside the group, compile a resetting op-code following. */
    if (*ptr != '|')
      {
-Line 1391 
 for (;;)
+Line 1761 
 for (;;)
      *code++ = OP_KET;
      *code++ = length >> 8;
      *code++ = length & 255;
+     if (optchanged >= 0)
+       {
+       *code++ = OP_OPT;
+       *code++ = oldoptions;
+       }
      *codeptr = code;
      *ptrptr = ptr;
      return TRUE;
-Line 1399 
 for (;;)
+Line 1774 
 for (;;)
    /* Another branch follows; insert an "or" node and advance the pointer. */
    *code = OP_ALT;
+   last_branch = code;
+   code += 3;
    ptr++;
    }
  /* Control never reaches here */
-Line 1406 
 for (;;)
+Line 1783 
 for (;;)
+ /*************************************************
+ *      Find first significant op code            *
+ *************************************************/
+ /* This is called by several functions that scan a compiled expression looking
+ for a fixed first character, or an anchoring op code etc. It skips over things
+ that do not influence this. For one application, a change of caseless option is
+ important.
+ Arguments:
+   code       pointer to the start of the group
+   options    pointer to external options
+   optbit     the option bit whose changing is significant, or
+              zero if none are
+   optstop    TRUE to return on option change, otherwise change the options
+                value and continue
+ Returns:     pointer to the first significant opcode
+ */
+ static const uschar*
+ first_significant_code(const uschar *code, int *options, int optbit,
+   BOOL optstop)
+ {
+ for (;;)
+   {
+   switch ((int)*code)
+     {
+     case OP_OPT:
+     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
+       {
+       if (optstop) return code;
+       *options = (int)code[1];
+       }
+     code += 2;
+     break;
+     case OP_CREF:
+     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:
+     do code += (code[1] << 8) + code[2]; while (*code == OP_ALT);
+     code += 3;
+     break;
+     default:
+     return code;
+     }
+   }
+ /* Control never reaches here */
+ }
  /*************************************************
  *          Check for anchored expression         *
  *************************************************/
-Line 1416 
 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.
- Argument:  points to start of expression (the bracket)
+ Arguments:
- Returns:   TRUE or FALSE
+   code       points to start of expression (the bracket)
+   options    points to the options setting
+ Returns:     TRUE or FALSE
  */
  static BOOL
- is_anchored(register const uschar *code, BOOL multiline)
+ is_anchored(register const uschar *code, int *options)
  {
  do {
-    int op = (int)code[3];
+    const uschar *scode = first_significant_code(code + 3, options,
-    if (op >= OP_BRA || op == OP_ASSERT || op == OP_ONCE)
+      PCRE_MULTILINE, FALSE);
-      { if (!is_anchored(code+3, multiline)) return FALSE; }
+    register int op = *scode;
-    else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR)
+    if (op >= OP_BRA || op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
-      { if (code[4] != OP_ANY) return FALSE; }
+      { if (!is_anchored(scode, options)) return FALSE; }
-    else if (op != OP_SOD && (multiline || op != OP_CIRC)) return FALSE;
+    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))
+      return FALSE;
     code += (code[1] << 8) + code[2];
     }
  while (*code == OP_ALT);
-Line 1443 
 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 1457 
 static BOOL
+Line 1908 
 static BOOL
  is_startline(const uschar *code)
  {
  do {
-    if ((int)code[3] >= OP_BRA || code[3] == OP_ASSERT)
+    const uschar *scode = first_significant_code(code + 3, NULL, 0, FALSE);
-      { if (!is_startline(code+3)) return FALSE; }
+    register int op = *scode;
-    else if (code[3] != OP_CIRC) return FALSE;
+    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];
     }
  while (*code == OP_ALT);
-Line 1478 
 Consider each alternative branch. If the
+Line 1933 
 Consider each alternative branch. If the
  a bracket all of whose alternatives start with the same char (recurse ad lib),
  then we return that char, otherwise -1.
- Argument:  points to start of expression (the bracket)
+ Arguments:
- Returns:   -1 or the fixed first char
+   code       points to start of expression (the bracket)
+   options    pointer to the options (used to check casing changes)
+ Returns:     -1 or the fixed first char
  */
  static int
- find_firstchar(uschar *code)
+ find_firstchar(const uschar *code, int *options)
  {
  register int c = -1;
- do
+ do {
-   {
+    int d;
-   register int charoffset = 4;
+    const uschar *scode = first_significant_code(code + 3, options,
+      PCRE_CASELESS, TRUE);
-   if ((int)code[3] >= OP_BRA || code[3] == OP_ASSERT)
+    register int op = *scode;
-     {
-     register int d;
+    if (op >= OP_BRA) op = OP_BRA;
-     if ((d = find_firstchar(code+3)) < 0) return -1;
-     if (c < 0) c = d; else if (c != d) return -1;
+    switch(op)
-     }
+      {
+      default:
-   else switch(code[3])
+      return -1;
-     {
-     default:
+      case OP_BRA:
-     return -1;
+      case OP_ASSERT:
+      case OP_ONCE:
-     case OP_EXACT:       /* Fall through */
+      case OP_COND:
-     charoffset++;
+      if ((d = find_firstchar(scode, options)) < 0) return -1;
+      if (c < 0) c = d; else if (c != d) return -1;
-     case OP_CHARS:       /* Fall through */
+      break;
-     charoffset++;
+      case OP_EXACT:       /* Fall through */
+      scode++;
+      case OP_CHARS:       /* Fall through */
+      scode++;
+      case OP_PLUS:
+      case OP_MINPLUS:
+      if (c < 0) c = scode[1]; else if (c != scode[1]) return -1;
+      break;
+      }
-     case OP_PLUS:
+    code += (code[1] << 8) + code[2];
-     case OP_MINPLUS:
+    }
-     if (c < 0) c = code[charoffset]; else if (c != code[charoffset]) return -1;
-     break;
-     }
-   code += (code[1] << 8) + code[2];
-   }
  while (*code == OP_ALT);
  return c;
  }
  /*************************************************
  *        Compile a Regular Expression            *
  *************************************************/
-Line 1533 
 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 1540 
 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 spaces = 0;
  int length = 3;      /* For initial BRA plus length */
  int runlength;
- int c, size;
+ int c, size, reqchar, countlits;
  int bracount = 0;
- int brastack[200];
  int top_backref = 0;
+ int branch_extra = 0;
+ int branch_newextra;
  unsigned int brastackptr = 0;
  uschar *code;
  const uschar *ptr;
+ compile_data compile_block;
+ int brastack[BRASTACK_SIZE];
+ uschar bralenstack[BRASTACK_SIZE];
  #ifdef DEBUG
  uschar *code_base, *code_end;
-Line 1579 
 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 1595 
 while ((c = *(++ptr)) != 0)
+Line 2075 
 while ((c = *(++ptr)) != 0)
    int min, max;
    int class_charcount;
-   if ((pcre_ctypes[c] & ctype_space) != 0)
+   if ((options & PCRE_EXTENDED) != 0)
-     {
-     if ((options & PCRE_EXTENDED) != 0) continue;
-     spaces++;
-     }
-   if (c == '#' && (options & PCRE_EXTENDED) != 0)
      {
-     while ((c = *(++ptr)) != 0 && c != '\n');
+     if ((compile_block.ctypes[c] & ctype_space) != 0) continue;
-     continue;
+     if (c == '#')
+       {
+       while ((c = *(++ptr)) != 0 && c != '\n');
+       continue;
+       }
      }
    switch(c)
-Line 1617 
 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 1637 
 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 1663 
 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 1678 
 while ((c = *(++ptr)) != 0)
+Line 2156 
 while ((c = *(++ptr)) != 0)
      if (ptr[1] == '?') ptr++;
      continue;
-     /* An alternation contains an offset to the next branch or ket. */
+     /* An alternation contains an offset to the next branch or ket. If any ims
+     options changed in the previous branch(es), and/or if we are in a
+     lookbehind assertion, extra space will be needed at the start of the
+     branch. This is handled by branch_extra. */
      case '|':
-     length += 3;
+     length += 3 + branch_extra;
      continue;
      /* A character class uses 33 characters. Don't worry about character types
-Line 1695 
 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 1712 
 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 1728 
 while ((c = *(++ptr)) != 0)
+Line 2211 
 while ((c = *(++ptr)) != 0)
      /* Brackets may be genuine groups or special things */
      case '(':
+     branch_newextra = 0;
      /* Handle special forms of bracket, which all start (? */
-     if (ptr[1] == '?') switch (c = ptr[2])
+     if (ptr[1] == '?')
        {
-       /* Skip over comments entirely */
+       int set, unset;
-       case '#':
+       int *optset;
-       ptr += 3;
-       while (*ptr != 0 && *ptr != ')') ptr++;
-       if (*ptr == 0)
-         {
-         *errorptr = ERR18;
-         goto PCRE_ERROR_RETURN;
-         }
-       continue;
-       /* Non-referencing groups and lookaheads just move the pointer on, and
+       switch (c = ptr[2])
-       then behave like a non-special bracket, except that they don't increment
+         {
-       the count of extracting brackets. */
+         /* Skip over comments entirely */
+         case '#':
-       case ':':
+         ptr += 3;
-       case '=':
+         while (*ptr != 0 && *ptr != ')') ptr++;
-       case '!':
+         if (*ptr == 0)
-       ptr += 2;
+           {
-       break;
+           *errorptr = ERR18;
+           goto PCRE_ERROR_RETURN;
+           }
+         continue;
-       /* Ditto for the "once only" bracket, allowed only if the extra bit
+         /* Non-referencing groups and lookaheads just move the pointer on, and
-       is set. */
+         then behave like a non-special bracket, except that they don't increment
+         the count of extracting brackets. Ditto for the "once only" bracket,
+         which is in Perl from version 5.005. */
-       case '>':
+         case ':':
-       if ((options & PCRE_EXTRA) != 0)
+         case '=':
-         {
+         case '!':
+         case '>':
          ptr += 2;
          break;
-         }
-       /* Else fall through */
-       /* Else loop setting valid options until ) is met. Anything else is an
+         /* Lookbehinds are in Perl from version 5.005 */
-       error. */
-       default:
+         case '<':
-       ptr += 2;
+         if (ptr[3] == '=' || ptr[3] == '!')
-       for (;; ptr++)
-         {
-         if ((c = *ptr) == 'i')
            {
-           options |= PCRE_CASELESS;
+           ptr += 3;
-           continue;
+           branch_newextra = 3;
-           }
+           length += 3;         /* For the first branch */
-         else if ((c = *ptr) == 'm')
+           break;
-           {
-           options |= PCRE_MULTILINE;
-           continue;
            }
-         else if (c == 's')
+         *errorptr = ERR24;
-           {
+         goto PCRE_ERROR_RETURN;
-           options |= PCRE_DOTALL;
-           continue;
+         /* Conditionals are in Perl from version 5.005. The bracket must either
+         be followed by a number (for bracket reference) or by an assertion
+         group. */
+         case '(':
+         if ((compile_block.ctypes[ptr[3]] & ctype_digit) != 0)
+           {
+           ptr += 4;
+           length += 2;
+           while ((compile_block.ctypes[*ptr] & ctype_digit) != 0) ptr++;
+           if (*ptr != ')')
+             {
+             *errorptr = ERR26;
+             goto PCRE_ERROR_RETURN;
+             }
            }
-         else if (c == 'x')
+         else   /* An assertion must follow */
            {
-           options |= PCRE_EXTENDED;
+           ptr++;   /* Can treat like ':' as far as spacing is concerned */
-           length -= spaces;          /* Already counted spaces */
-           continue;
+           if (ptr[2] != '?' || strchr("=!<", ptr[3]) == NULL)
+             {
+             ptr += 2;    /* To get right offset in message */
+             *errorptr = ERR28;
+             goto PCRE_ERROR_RETURN;
+             }
            }
-         else if (c == 'X')
+         break;
+         /* Else loop checking valid options until ) is met. Anything else is an
+         error. If we are without any brackets, i.e. at top level, the settings
+         act as if specified in the options, so massage the options immediately.
+         This is for backward compatibility with Perl 5.004. */
+         default:
+         set = unset = 0;
+         optset = &set;
+         ptr += 2;
+         for (;; ptr++)
            {
-           options |= PCRE_EXTRA;
+           c = *ptr;
-           continue;
+           switch (c)
+             {
+             case 'i':
+             *optset |= PCRE_CASELESS;
+             continue;
+             case 'm':
+             *optset |= PCRE_MULTILINE;
+             continue;
+             case 's':
+             *optset |= PCRE_DOTALL;
+             continue;
+             case 'x':
+             *optset |= PCRE_EXTENDED;
+             continue;
+             case 'X':
+             *optset |= PCRE_EXTRA;
+             continue;
+             case 'U':
+             *optset |= PCRE_UNGREEDY;
+             continue;
+             case '-':
+             optset = &unset;
+             continue;
+             /* A termination by ')' indicates an options-setting-only item;
+             this is global at top level; otherwise nothing is done here and
+             it is handled during the compiling process on a per-bracket-group
+             basis. */
+             case ')':
+             if (brastackptr == 0)
+               {
+               options = (options | set) & (~unset);
+               set = unset = 0;     /* To save length */
+               }
+             /* Fall through */
+             /* A termination by ':' indicates the start of a nested group with
+             the given options set. This is again handled at compile time, but
+             we must allow for compiled space if any of the ims options are
+             set. We also have to allow for resetting space at the end of
+             the group, which is why 4 is added to the length and not just 2.
+             If there are several changes of options within the same group, this
+             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. 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;
+             /* Unrecognized option character */
+             default:
+             *errorptr = ERR12;
+             goto PCRE_ERROR_RETURN;
+             }
            }
-         else if (c == 'U')
+         /* If we hit a closing bracket, that's it - this is a freestanding
+         option-setting. We need to ensure that branch_extra is updated if
+         necessary. The only values branch_newextra can have here are 0 or 2.
+         If the value is 2, then branch_extra must either be 2 or 5, depending
+         on whether this is a lookbehind group or not. */
+         END_OPTIONS:
+         if (c == ')')
            {
-           options |= PCRE_UNGREEDY;
+           if (branch_newextra == 2 && (branch_extra == 0 || branch_extra == 3))
+             branch_extra += branch_newextra;
            continue;
            }
-         else if (c == ')') break;
-         *errorptr = ERR12;
+         /* If options were terminated by ':' control comes here. Fall through
-         goto PCRE_ERROR_RETURN;
+         to handle the group below. */
          }
-       continue;                      /* End of this bracket handling */
        }
      /* Extracting brackets must be counted so we can process escapes in a
-Line 1817 
 while ((c = *(++ptr)) != 0)
+Line 2398 
 while ((c = *(++ptr)) != 0)
      else bracount++;
      /* Non-special forms of bracket. Save length for computing whole length
-     at end if there's a repeat that requires duplication of the group. */
+     at end if there's a repeat that requires duplication of the group. Also
+     save the current value of branch_extra, and start the new group with
+     the new value. If non-zero, this will either be 2 for a (?imsx: group, or 3
+     for a lookbehind assertion. */
      if (brastackptr >= sizeof(brastack)/sizeof(int))
        {
-Line 1825 
 while ((c = *(++ptr)) != 0)
+Line 2409 
 while ((c = *(++ptr)) != 0)
        goto PCRE_ERROR_RETURN;
        }
+     bralenstack[brastackptr] = branch_extra;
+     branch_extra = branch_newextra;
      brastack[brastackptr++] = length;
      length += 3;
      continue;
-Line 1832 
 while ((c = *(++ptr)) != 0)
+Line 2419 
 while ((c = *(++ptr)) != 0)
      /* Handle ket. Look for subsequent max/min; for certain sets of values we
      have to replicate this bracket up to that many times. If brastackptr is
 this is an unmatched bracket which will generate an error, but take care
-     not to try to access brastack[-1]. */
+     not to try to access brastack[-1] when computing the length and restoring
+     the branch_extra value. */
      case ')':
      length += 3;
        {
        int minval = 1;
        int maxval = 1;
-       int duplength = (brastackptr > 0)? length - brastack[--brastackptr] : 0;
+       int duplength;
+       if (brastackptr > 0)
+         {
+         duplength = length - brastack[--brastackptr];
+         branch_extra = bralenstack[brastackptr];
+         }
+       else duplength = 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 1875 
 while ((c = *(++ptr)) != 0)
+Line 2486 
 while ((c = *(++ptr)) != 0)
      runlength = 0;
      do
        {
-       if ((pcre_ctypes[c] & ctype_space) != 0)
+       if ((options & PCRE_EXTENDED) != 0)
-         {
-         if ((options & PCRE_EXTENDED) != 0) continue;
-         spaces++;
-         }
-       if (c == '#' && (options & PCRE_EXTENDED) != 0)
          {
-         while ((c = *(++ptr)) != 0 && c != '\n');
+         if ((compile_block.ctypes[c] & ctype_space) != 0) continue;
-         continue;
+         if (c == '#')
+           {
+           while ((c = *(++ptr)) != 0 && c != '\n');
+           continue;
+           }
          }
        /* Backslash may introduce a data char or a metacharacter; stop the
-Line 1893 
 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 1905 
 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 1940 
 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 1949 
 ptr = (const uschar *)pattern;
+Line 2561 
 ptr = (const uschar *)pattern;
  code = re->code;
  *code = OP_BRA;
  bracount = 0;
- (void)compile_regex(options, &bracount, &code, &ptr, errorptr);
+ (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 1966 
 if debugging, leave the test till after
+Line 2579 
 if debugging, leave the test till after
  if (code - re->code > length) *errorptr = ERR23;
  #endif
+ /* Give an error if there's back reference to a non-existent capturing
+ subpattern. */
+ if (top_backref > re->top_bracket) *errorptr = ERR15;
  /* Failed to compile */
  if (*errorptr != NULL)
-Line 1976 
 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)
    {
-   if (is_anchored(re->code, (options & PCRE_MULTILINE) != 0))
+   int temp_options = options;
+   if (is_anchored(re->code, &temp_options))
      re->options |= PCRE_ANCHORED;
    else
      {
-     int ch = find_firstchar(re->code);
+     int ch = find_firstchar(re->code, &temp_options);
      if (ch >= 0)
        {
        re->first_char = ch;
-Line 1999 
 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
- printf("Length = %d top_bracket = %d top_backref=%d\n",
+ printf("Length = %d top_bracket = %d top_backref = %d\n",
    length, re->top_bracket, re->top_backref);
  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 2025 
 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 2042 
 while (code < code_end)
+Line 2681 
 while (code < code_end)
    else switch(*code)
      {
+     case OP_OPT:
+     printf(" %.2x %s", code[1], OP_names[*code]);
+     code++;
+     break;
+     case OP_COND:
+     printf("%3d Cond", (code[1] << 8) + code[2]);
+     code += 2;
+     break;
+     case OP_CREF:
+     printf(" %.2d %s", code[1], OP_names[*code]);
+     code++;
+     break;
      case OP_CHARS:
      charlength = *(++code);
      printf("%3d ", charlength);
-Line 2055 
 while (code < code_end)
+Line 2709 
 while (code < code_end)
      case OP_KET:
      case OP_ASSERT:
      case OP_ASSERT_NOT:
+     case OP_ASSERTBACK:
+     case OP_ASSERTBACK_NOT:
      case OP_ONCE:
      printf("%3d %s", (code[1] << 8) + code[2], OP_names[*code]);
      code += 2;
      break;
+     case OP_REVERSE:
+     printf("%3d %s", (code[1] << 8) + code[2], OP_names[*code]);
+     code += 2;
+     break;
      case OP_STAR:
      case OP_MINSTAR:
      case OP_PLUS:
-Line 2133 
 while (code < code_end)
+Line 2794 
 while (code < code_end)
      goto CLASS_REF_REPEAT;
      case OP_CLASS:
-     case OP_NEGCLASS:
        {
        int i, min, max;
+       code++;
-       if (*code++ == OP_CLASS) printf("    [");
+       printf("    [");
-         else printf("   ^[");
        for (i = 0; i < 256; i++)
          {
-Line 2220 
 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                *
  *************************************************/
- /* If a back reference hasn't been set, the match fails.
+ /* If a back reference hasn't been set, the length that is passed is greater
+ than the number of characters left in the string, so the match fails.
  Arguments:
-   number      reference number
+   offset      index into the offset vector
    eptr        points into the subject
    length      length to be matched
    md          points to match data block
+   ims         the ims flags
  Returns:      TRUE if matched
  */
  static BOOL
- match_ref(int number, register const uschar *eptr, int length, match_data *md)
+ match_ref(int offset, register const uschar *eptr, int length, match_data *md,
+   unsigned long int ims)
  {
- const uschar *p = md->start_subject + md->offset_vector[number];
+ const uschar *p = md->start_subject + md->offset_vector[offset];
  #ifdef DEBUG
  if (eptr >= md->end_subject)
-Line 2294 
 printf("\n");
+Line 2916 
 printf("\n");
  /* Always fail if not enough characters left */
- if (length > md->end_subject - p) return FALSE;
+ if (length > md->end_subject - eptr) return FALSE;
  /* Separate the caselesss case for speed */
- if (md->caseless)
+ 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 2312 
 return TRUE;
+Line 2937 
 return TRUE;
  *         Match from current position            *
  *************************************************/
- /* On entry ecode points to the first opcode, and eptr to the first character.
+ /* On entry ecode points to the first opcode, and eptr to the first character
+ in the subject string, while eptrb holds the value of eptr at the start of the
+ last bracketed group - used for breaking infinite loops matching zero-length
+ strings.
  Arguments:
     eptr        pointer in subject
     ecode       position in code
     offset_top  current top pointer
     md          pointer to "static" info for the match
+    ims         current /i, /m, and /s options
+    condassert  TRUE if called to check a condition assertion
+    eptrb       eptr at start of last bracket
  Returns:       TRUE if matched
  */
  static BOOL
- match(register const uschar *eptr, register const uschar *ecode, int offset_top,
+ match(register const uschar *eptr, register const uschar *ecode,
-   match_data *md)
+   int offset_top, match_data *md, unsigned long int ims, BOOL condassert,
+   const uschar *eptrb)
  {
+ unsigned long int original_ims = ims;   /* Save for resetting on ')' */
  for (;;)
    {
+   int op = (int)*ecode;
    int min, max, ctype;
    register int i;
    register int c;
    BOOL minimize = FALSE;
-   /* Opening bracket. Check the alternative branches in turn, failing if none
+   /* Opening capturing bracket. If there is space in the offset vector, save
-   match. We have to set the start offset if required and there is space
+   the current subject position in the working slot at the top of the vector. We
-   in the offset vector so that it is available for subsequent back references
+   mustn't change the current values of the data slot, because they may be set
-   if the bracket matches. However, if the bracket fails, we must put back the
+   from a previous iteration of this group, and be referred to by a reference
-   previous value of both offsets in case they were set by a previous copy of
+   inside the group.
-   the same bracket. Don't worry about setting the flag for the error case here;
-   that is handled in the code for KET. */
+   If the bracket fails to match, we need to restore this value and also the
+   values of the final offsets, in case they were set by a previous iteration of
+   the same bracket.
+   If there isn't enough space in the offset vector, treat this as if it were a
+   non-capturing bracket. Don't worry about setting the flag for the error case
+   here; that is handled in the code for KET. */
-   if ((int)*ecode >= OP_BRA)
+   if (op > OP_BRA)
      {
-     int number = (*ecode - OP_BRA) << 1;
+     int number = op - OP_BRA;
-     int save_offset1 = 0, save_offset2 = 0;
+     int offset = number << 1;
-     DPRINTF(("start bracket %d\n", number/2));
+ #ifdef DEBUG
+     printf("start bracket %d subject=", number);
+     pchars(eptr, 16, TRUE, md);
+     printf("\n");
+ #endif
-     if (number > 0 && number < md->offset_end)
+     if (offset < md->offset_max)
        {
-       save_offset1 = md->offset_vector[number];
+       int save_offset1 = md->offset_vector[offset];
-       save_offset2 = md->offset_vector[number+1];
+       int save_offset2 = md->offset_vector[offset+1];
-       md->offset_vector[number] = eptr - md->start_subject;
+       int save_offset3 = md->offset_vector[md->offset_end - number];
+       DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
+       md->offset_vector[md->offset_end - number] = eptr - md->start_subject;
+       do
+         {
+         if (match(eptr, ecode+3, offset_top, md, ims, FALSE, eptr)) return TRUE;
+         ecode += (ecode[1] << 8) + ecode[2];
+         }
+       while (*ecode == OP_ALT);
-       DPRINTF(("saving %d %d\n", save_offset1, save_offset2));
+       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;
+       return FALSE;
        }
-     /* Recurse for all the alternatives. */
+     /* Insufficient room for saving captured contents */
+     else op = OP_BRA;
+     }
+   /* Other types of node can be handled by a switch */
+   switch(op)
+     {
+     case OP_BRA:     /* Non-capturing bracket: optimized */
+     DPRINTF(("start bracket 0\n"));
      do
        {
-       if (match(eptr, ecode+3, offset_top, md)) return TRUE;
+       if (match(eptr, ecode+3, offset_top, md, ims, FALSE, eptr)) return TRUE;
        ecode += (ecode[1] << 8) + ecode[2];
        }
      while (*ecode == OP_ALT);
+     DPRINTF(("bracket 0 failed\n"));
+     return FALSE;
-     DPRINTF(("bracket %d failed\n", number/2));
+     /* 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. */
+     case OP_COND:
+     if (ecode[3] == OP_CREF)         /* Condition is extraction test */
+       {
+       int offset = ecode[4] << 1;    /* Doubled reference number */
+       return match(eptr,
+         ecode + ((offset < offset_top && md->offset_vector[offset] >= 0)?
+: 3 + (ecode[1] << 8) + ecode[2]),
+         offset_top, md, ims, FALSE, eptr);
+       }
+     /* The condition is an assertion. Call match() to evaluate it - setting
+     the final argument TRUE causes it to stop at the end of an assertion. */
-     if (number > 0 && number < md->offset_end)
+     else
        {
-       md->offset_vector[number] = save_offset1;
+       if (match(eptr, ecode+3, offset_top, md, ims, TRUE, NULL))
-       md->offset_vector[number+1] = save_offset2;
+         {
+         ecode += 3 + (ecode[4] << 8) + ecode[5];
+         while (*ecode == OP_ALT) ecode += (ecode[1] << 8) + ecode[2];
+         }
+       else ecode += (ecode[1] << 8) + ecode[2];
+       return match(eptr, ecode+3, offset_top, md, ims, FALSE, eptr);
        }
+     /* Control never reaches here */
-     return FALSE;
+     /* Skip over conditional reference data if encountered (should not be) */
-     }
-   /* Other types of node can be handled by a switch */
+     case OP_CREF:
+     ecode += 2;
+     break;
+     /* 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. */
-   switch(*ecode)
-     {
      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;
-     /* The equivalent of Prolog's "cut" - if the rest doesn't match, the
+     /* Change option settings */
-     whole thing doesn't match, so we have to get out via a longjmp(). */
-     case OP_CUT:
+     case OP_OPT:
-     if (match(eptr, ecode+1, offset_top, md)) return TRUE;
+     ims = ecode[1];
-     longjmp(md->fail_env, 1);
+     ecode += 2;
+     DPRINTF(("ims set to %02x\n", ims));
+     break;
      /* Assertion brackets. Check the alternative branches in turn - the
      matching won't pass the KET for an assertion. If any one branch matches,
-     the assertion is true. */
+     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. */
      case OP_ASSERT:
+     case OP_ASSERTBACK:
      do
        {
-       if (match(eptr, ecode+3, offset_top, md)) break;
+       if (match(eptr, ecode+3, offset_top, md, ims, FALSE, NULL)) break;
        ecode += (ecode[1] << 8) + ecode[2];
        }
      while (*ecode == OP_ALT);
      if (*ecode == OP_KET) return FALSE;
+     /* If checking an assertion for a condition, return TRUE. */
+     if (condassert) return TRUE;
      /* Continue from after the assertion, updating the offsets high water
      mark, since extracts may have been taken during the assertion. */
-Line 2418 
 for (;;)
+Line 3122 
 for (;;)
      /* Negative assertion: all branches must fail to match */
      case OP_ASSERT_NOT:
+     case OP_ASSERTBACK_NOT:
      do
        {
-       if (match(eptr, ecode+3, offset_top, md)) return FALSE;
+       if (match(eptr, ecode+3, offset_top, md, ims, FALSE, NULL)) return FALSE;
        ecode += (ecode[1] << 8) + ecode[2];
        }
      while (*ecode == OP_ALT);
+     if (condassert) return TRUE;
      ecode += 3;
      continue;
+     /* Move the subject pointer back. This occurs only at the start of
+     each branch of a lookbehind assertion. If we are too close to the start to
+     move back, this match function fails. */
+     case OP_REVERSE:
+     eptr -= (ecode[1] << 8) + ecode[2];
+     if (eptr < md->start_subject) return FALSE;
+     ecode += 3;
+     break;
      /* "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. 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, leaving the subject pointer. */
+     matches, we carry on as at the end of a normal bracket, leaving the subject
+     pointer. */
      case OP_ONCE:
-     do
        {
-       if (match(eptr, ecode+3, offset_top, md)) break;
+       const uschar *prev = ecode;
-       ecode += (ecode[1] << 8) + ecode[2];
-       }
-     while (*ecode == OP_ALT);
-     if (*ecode == OP_KET) return FALSE;
-     /* Continue as from after the assertion, updating the offsets high water
+       do
-     mark, since extracts may have been taken. */
+         {
+         if (match(eptr, ecode+3, offset_top, md, ims, FALSE, eptr)) break;
+         ecode += (ecode[1] << 8) + ecode[2];
+         }
+       while (*ecode == OP_ALT);
-     do ecode += (ecode[1] << 8) + ecode[2]; while (*ecode == OP_ALT);
+       /* If hit the end of the group (which could be repeated), fail */
-     ecode += 3;
-     offset_top = md->end_offset_top;
+       if (*ecode != OP_ONCE && *ecode != OP_ALT) return FALSE;
-     eptr = md->end_match_ptr;
-     continue;
+       /* Continue as from after the assertion, updating the offsets high water
+       mark, since extracts may have been taken. */
+       do ecode += (ecode[1] << 8) + ecode[2]; 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 == eptrb)
+         {
+         ecode += 3;
+         break;
+         }
+       /* The repeating kets try the rest of the pattern or restart from the
+       preceding bracket, in the appropriate order. We need to reset any options
+       that changed within the bracket before re-running it, so check the next
+       opcode. */
+       if (ecode[3] == OP_OPT)
+         {
+         ims = (ims & ~PCRE_IMS) | ecode[4];
+         DPRINTF(("ims set to %02x at group repeat\n", ims));
+         }
+       if (*ecode == OP_KETRMIN)
+         {
+         if (match(eptr, ecode+3, offset_top, md, ims, FALSE, eptr) ||
+             match(eptr, prev, offset_top, md, ims, FALSE, eptr)) return TRUE;
+         }
+       else  /* OP_KETRMAX */
+         {
+         if (match(eptr, prev, offset_top, md, ims, FALSE, eptr) ||
+             match(eptr, ecode+3, offset_top, md, ims, FALSE, eptr)) return TRUE;
+         }
+       }
+     return FALSE;
      /* An alternation is the end of a branch; scan along to find the end of the
      bracketed group and go to there. */
-Line 2467 
 for (;;)
+Line 3227 
 for (;;)
      case OP_BRAZERO:
        {
        const uschar *next = ecode+1;
-       if (match(eptr, next, offset_top, md)) return TRUE;
+       if (match(eptr, next, offset_top, md, ims, FALSE, eptr)) return TRUE;
        do next += (next[1] << 8) + next[2]; while (*next == OP_ALT);
        ecode = next + 3;
        }
-Line 2477 
 for (;;)
+Line 3237 
 for (;;)
        {
        const uschar *next = ecode+1;
        do next += (next[1] << 8) + next[2]; while (*next == OP_ALT);
-       if (match(eptr, next+3, offset_top, md)) return TRUE;
+       if (match(eptr, next+3, offset_top, md, ims, FALSE, eptr)) return TRUE;
        ecode++;
        }
-     break;;
+     break;
      /* End of a group, repeated or non-repeating. If we are at the end of
      an assertion "group", stop matching and return TRUE, but record the
-     current high water mark for use by positive assertions. */
+     current high water mark for use by positive assertions. Do this also
+     for the "once" (not-backup up) groups. */
      case OP_KET:
      case OP_KETRMIN:
      case OP_KETRMAX:
        {
-       int number;
        const uschar *prev = ecode - (ecode[1] << 8) - ecode[2];
-       if (*prev == OP_ASSERT || *prev == OP_ASSERT_NOT || *prev == OP_ONCE)
+       if (*prev == OP_ASSERT || *prev == OP_ASSERT_NOT ||
+           *prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT ||
+           *prev == OP_ONCE)
          {
          md->end_match_ptr = eptr;      /* For ONCE */
          md->end_offset_top = offset_top;
          return TRUE;
          }
-       /* In all other cases we have to check the group number back at the
+       /* In all other cases except a conditional group we have to check the
-       start and if necessary complete handling an extraction by setting the
+       group number back at the start and if necessary complete handling an
-       final offset and bumping the high water mark. */
+       extraction by setting the offsets and bumping the high water mark. */
-       number = (*prev - OP_BRA) << 1;
+       if (*prev != OP_COND)
+         {
+         int number = *prev - OP_BRA;
+         int offset = number << 1;
-       DPRINTF(("end bracket %d\n", number/2));
+         DPRINTF(("end bracket %d\n", number));
-       if (number > 0)
+         if (number > 0)
-         {
-         if (number >= md->offset_end) md->offset_overflow = TRUE; else
            {
-           md->offset_vector[number+1] = eptr - md->start_subject;
+           if (offset >= md->offset_max) md->offset_overflow = TRUE; else
-           if (offset_top <= number) offset_top = number + 2;
+             {
+             md->offset_vector[offset] =
+               md->offset_vector[md->offset_end - number];
+             md->offset_vector[offset+1] = eptr - md->start_subject;
+             if (offset_top <= offset) offset_top = offset + 2;
+             }
            }
          }
-       /* For a non-repeating ket, just advance to the next node and continue at
+       /* Reset the value of the ims flags, in case they got changed during
-       this level. */
+       the group. */
+       ims = original_ims;
+       DPRINTF(("ims reset to %02x\n", ims));
-       if (*ecode == OP_KET)
+       /* 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 == eptrb)
          {
          ecode += 3;
          break;
-Line 2531 
 for (;;)
+Line 3308 
 for (;;)
        if (*ecode == OP_KETRMIN)
          {
-         if (match(eptr, ecode+3, offset_top, md) ||
+         if (match(eptr, ecode+3, offset_top, md, ims, FALSE, eptr) ||
-             match(eptr, prev, offset_top, md)) return TRUE;
+             match(eptr, prev, offset_top, md, ims, FALSE, eptr)) return TRUE;
          }
        else  /* OP_KETRMAX */
          {
-         if (match(eptr, prev, offset_top, md) ||
+         if (match(eptr, prev, offset_top, md, ims, FALSE, eptr) ||
-             match(eptr, ecode+3, offset_top, md)) return TRUE;
+             match(eptr, ecode+3, offset_top, md, ims, FALSE, eptr)) return TRUE;
          }
        }
      return FALSE;
-Line 2546 
 for (;;)
+Line 3323 
 for (;;)
      case OP_CIRC:
      if (md->notbol && eptr == md->start_subject) return FALSE;
-     if (md->multiline)
+     if ((ims & PCRE_MULTILINE) != 0)
        {
        if (eptr != md->start_subject && eptr[-1] != '\n') return FALSE;
        ecode++;
-Line 2561 
 for (;;)
+Line 3338 
 for (;;)
      ecode++;
      break;
-     /* Assert before internal newline if multiline, or before
+     /* Assert before internal newline if multiline, or before a terminating
-     a terminating newline unless endonly is set, else end of subject unless
+     newline unless endonly is set, else end of subject unless noteol is set. */
-     noteol is set. */
      case OP_DOLL:
-     if (md->noteol && eptr >= md->end_subject) return FALSE;
+     if ((ims & PCRE_MULTILINE) != 0)
-     if (md->multiline)
        {
-       if (eptr < md->end_subject && *eptr != '\n') return FALSE;
+       if (eptr < md->end_subject) { if (*eptr != '\n') return FALSE; }
+         else { if (md->noteol) return FALSE; }
        ecode++;
        break;
        }
-     else if (!md->endonly)
+     else
        {
-       if (eptr < md->end_subject - 1 ||
+       if (md->noteol) return FALSE;
-          (eptr == md->end_subject - 1 && *eptr != '\n')) return FALSE;
+       if (!md->endonly)
-       ecode++;
+         {
-       break;
+         if (eptr < md->end_subject - 1 ||
+            (eptr == md->end_subject - 1 && *eptr != '\n')) return FALSE;
+         ecode++;
+         break;
+         }
        }
      /* ... else fall through */
-     /* End of subject assertion */
+     /* End of subject assertion (\z) */
      case OP_EOD:
      if (eptr < md->end_subject) return FALSE;
      ecode++;
      break;
+     /* End of subject or ending \n assertion (\Z) */
+     case OP_EODN:
+     if (eptr < md->end_subject - 1 ||
+        (eptr == md->end_subject - 1 && *eptr != '\n')) return FALSE;
+     ecode++;
+     break;
      /* Word boundary assertions */
      case OP_NOT_WORD_BOUNDARY:
      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 2607 
 for (;;)
+Line 3396 
 for (;;)
      /* Match a single character type; inline for speed */
      case OP_ANY:
-     if (!md->dotall && eptr < md->end_subject && *eptr == '\n') return FALSE;
+     if ((ims & PCRE_DOTALL) == 0 && eptr < md->end_subject && *eptr == '\n')
+       return FALSE;
      if (eptr++ >= md->end_subject) return FALSE;
      ecode++;
      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 2659 
 for (;;)
+Line 3455 
 for (;;)
      case OP_REF:
        {
        int length;
-       int number = ecode[1] << 1;                /* Doubled reference number */
+       int offset = ecode[1] << 1;                /* Doubled reference number */
        ecode += 2;                                /* Advance past the item */
-       if (number >= offset_top || md->offset_vector[number] < 0)
+       /* If the reference is unset, set the length to be longer than the amount
-         {
+       of subject left; this ensures that every attempt at a match fails. We
-         md->errorcode = PCRE_ERROR_BADREF;
+       can't just fail here, because of the possibility of quantifiers with zero
-         return FALSE;
+       minima. */
-         }
+       length = (offset >= offset_top || md->offset_vector[offset] < 0)?
+         md->end_subject - eptr + 1 :
+         md->offset_vector[offset+1] - md->offset_vector[offset];
-       length = md->offset_vector[number+1] - md->offset_vector[number];
+       /* Set up for repetition, or handle the non-repeated case */
        switch (*ecode)
          {
-Line 2695 
 for (;;)
+Line 3494 
 for (;;)
          break;
          default:               /* No repeat follows */
-         if (!match_ref(number, eptr, length, md)) return FALSE;
+         if (!match_ref(offset, eptr, length, md, ims)) return FALSE;
          eptr += length;
          continue;              /* With the main loop */
          }
-Line 2711 
 for (;;)
+Line 3510 
 for (;;)
        for (i = 1; i <= min; i++)
          {
-         if (!match_ref(number, eptr, length, md)) return FALSE;
+         if (!match_ref(offset, eptr, length, md, ims)) return FALSE;
          eptr += length;
          }
-Line 2726 
 for (;;)
+Line 3525 
 for (;;)
          {
          for (i = min;; i++)
            {
-           if (match(eptr, ecode, offset_top, md)) return TRUE;
+           if (match(eptr, ecode, offset_top, md, ims, FALSE, eptrb))
-           if (i >= max || !match_ref(number, eptr, length, md))
+             return TRUE;
+           if (i >= max || !match_ref(offset, eptr, length, md, ims))
              return FALSE;
            eptr += length;
            }
-Line 2741 
 for (;;)
+Line 3541 
 for (;;)
          const uschar *pp = eptr;
          for (i = min; i < max; i++)
            {
-           if (!match_ref(number, eptr, length, md)) break;
+           if (!match_ref(offset, eptr, length, md, ims)) break;
            eptr += length;
            }
          while (eptr >= pp)
            {
-           if (match(eptr, ecode, offset_top, md)) return TRUE;
+           if (match(eptr, ecode, offset_top, md, ims, FALSE, eptrb))
+             return TRUE;
            eptr -= length;
            }
          return FALSE;
-Line 2754 
 for (;;)
+Line 3555 
 for (;;)
        }
      /* Control never gets here */
      /* Match a character class, possibly repeatedly. Look past the end of the
      item to see if there is repeat information following. Then obey similar
-     code to character type repeats - written out again for speed. If caseless
+     code to character type repeats - written out again for speed. */
-     matching was set at runtime but not at compile time, we have to check both
-     versions of a character, and we have to behave differently for positive and
-     negative classes. This is the only time where OP_CLASS and OP_NEGCLASS are
-     treated differently. */
      case OP_CLASS:
-     case OP_NEGCLASS:
        {
-       BOOL nasty_case = *ecode == OP_NEGCLASS && md->runtime_caseless;
        const uschar *data = ecode + 1;  /* Save for matching */
        ecode += 33;                     /* Advance past the item */
-Line 2804 
 for (;;)
+Line 3601 
 for (;;)
          {
          if (eptr >= md->end_subject) return FALSE;
          c = *eptr++;
+         if ((data[c/8] & (1 << (c&7))) != 0) continue;
-         /* Either not runtime caseless, or it was a positive class. For
-         runtime caseless, continue if either case is in the map. */
-         if (!nasty_case)
-           {
-           if ((data[c/8] & (1 << (c&7))) != 0) continue;
-           if (md->runtime_caseless)
-             {
-             c = pcre_fcc[c];
-             if ((data[c/8] & (1 << (c&7))) != 0) continue;
-             }
-           }
-         /* Runtime caseless and it was a negative class. Continue only if
-         both cases are in the map. */
-         else
-           {
-           if ((data[c/8] & (1 << (c&7))) == 0) return FALSE;
-           c = pcre_fcc[c];
-           if ((data[c/8] & (1 << (c&7))) != 0) continue;
-           }
          return FALSE;
          }
-Line 2843 
 for (;;)
+Line 3617 
 for (;;)
          {
          for (i = min;; i++)
            {
-           if (match(eptr, ecode, offset_top, md)) return TRUE;
+           if (match(eptr, ecode, offset_top, md, ims, FALSE, eptrb))
+             return TRUE;
            if (i >= max || eptr >= md->end_subject) return FALSE;
            c = *eptr++;
+           if ((data[c/8] & (1 << (c&7))) != 0) continue;
-           /* Either not runtime caseless, or it was a positive class. For
-           runtime caseless, continue if either case is in the map. */
-           if (!nasty_case)
-             {
-             if ((data[c/8] & (1 << (c&7))) != 0) continue;
-             if (md->runtime_caseless)
-               {
-               c = pcre_fcc[c];
-               if ((data[c/8] & (1 << (c&7))) != 0) continue;
-               }
-             }
-           /* Runtime caseless and it was a negative class. Continue only if
-           both cases are in the map. */
-           else
-             {
-             if ((data[c/8] & (1 << (c&7))) == 0) return FALSE;
-             c = pcre_fcc[c];
-             if ((data[c/8] & (1 << (c&7))) != 0) continue;
-             }
            return FALSE;
            }
          /* Control never gets here */
-Line 2884 
 for (;;)
+Line 3636 
 for (;;)
            {
            if (eptr >= md->end_subject) break;
            c = *eptr;
+           if ((data[c/8] & (1 << (c&7))) != 0) continue;
-           /* Either not runtime caseless, or it was a positive class. For
-           runtime caseless, continue if either case is in the map. */
-           if (!nasty_case)
-             {
-             if ((data[c/8] & (1 << (c&7))) != 0) continue;
-             if (md->runtime_caseless)
-               {
-               c = pcre_fcc[c];
-               if ((data[c/8] & (1 << (c&7))) != 0) continue;
-               }
-             }
-           /* Runtime caseless and it was a negative class. Continue only if
-           both cases are in the map. */
-           else
-             {
-             if ((data[c/8] & (1 << (c&7))) == 0) break;
-             c = pcre_fcc[c];
-             if ((data[c/8] & (1 << (c&7))) != 0) continue;
-             }
            break;
            }
          while (eptr >= pp)
-           if (match(eptr--, ecode, offset_top, md)) return TRUE;
+           if (match(eptr--, ecode, offset_top, md, ims, FALSE, eptrb))
+             return TRUE;
          return FALSE;
          }
        }
-Line 2939 
 for (;;)
+Line 3669 
 for (;;)
  #endif
        if (length > md->end_subject - eptr) return FALSE;
-       if (md->caseless)
+       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 2996 
 for (;;)
+Line 3728 
 for (;;)
      DPRINTF(("matching %c{%d,%d} against subject %.*s\n", c, min, max,
        max, eptr));
-     if (md->caseless)
+     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)
          {
          for (i = min;; i++)
            {
-           if (match(eptr, ecode, offset_top, md)) return TRUE;
+           if (match(eptr, ecode, offset_top, md, ims, FALSE, eptrb))
-           if (i >= max || eptr >= md->end_subject || c != pcre_lcc[*eptr++])
+             return TRUE;
+           if (i >= max || eptr >= md->end_subject ||
+               c != md->lcc[*eptr++])
              return FALSE;
            }
          /* Control never gets here */
-Line 3016 
 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)
-           if (match(eptr--, ecode, offset_top, md)) return TRUE;
+           if (match(eptr--, ecode, offset_top, md, ims, FALSE, eptrb))
+             return TRUE;
          return FALSE;
          }
        /* Control never gets here */
-Line 3036 
 for (;;)
+Line 3772 
 for (;;)
          {
          for (i = min;; i++)
            {
-           if (match(eptr, ecode, offset_top, md)) return TRUE;
+           if (match(eptr, ecode, offset_top, md, ims, FALSE, eptrb))
+             return TRUE;
            if (i >= max || eptr >= md->end_subject || c != *eptr++) return FALSE;
            }
          /* Control never gets here */
-Line 3050 
 for (;;)
+Line 3787 
 for (;;)
            eptr++;
            }
          while (eptr >= pp)
-          if (match(eptr--, ecode, offset_top, md)) return TRUE;
+          if (match(eptr--, ecode, offset_top, md, ims, FALSE, eptrb))
+            return TRUE;
          return FALSE;
          }
        }
-Line 3061 
 for (;;)
+Line 3799 
 for (;;)
      case OP_NOT:
      if (eptr >= md->end_subject) return FALSE;
      ecode++;
-     if (md->caseless)
+     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 3121 
 for (;;)
+Line 3859 
 for (;;)
      DPRINTF(("negative matching %c{%d,%d} against subject %.*s\n", c, min, max,
        max, eptr));
-     if (md->caseless)
+     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)
          {
          for (i = min;; i++)
            {
-           if (match(eptr, ecode, offset_top, md)) return TRUE;
+           if (match(eptr, ecode, offset_top, md, ims, FALSE, eptrb))
-           if (i >= max || eptr >= md->end_subject || c == pcre_lcc[*eptr++])
+             return TRUE;
+           if (i >= max || eptr >= md->end_subject ||
+               c == md->lcc[*eptr++])
              return FALSE;
            }
          /* Control never gets here */
-Line 3141 
 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)
-           if (match(eptr--, ecode, offset_top, md)) return TRUE;
+           if (match(eptr--, ecode, offset_top, md, ims, FALSE, eptrb))
+             return TRUE;
          return FALSE;
          }
        /* Control never gets here */
-Line 3161 
 for (;;)
+Line 3903 
 for (;;)
          {
          for (i = min;; i++)
            {
-           if (match(eptr, ecode, offset_top, md)) return TRUE;
+           if (match(eptr, ecode, offset_top, md, ims, FALSE, eptrb))
+             return TRUE;
            if (i >= max || eptr >= md->end_subject || c == *eptr++) return FALSE;
            }
          /* Control never gets here */
-Line 3175 
 for (;;)
+Line 3918 
 for (;;)
            eptr++;
            }
          while (eptr >= pp)
-          if (match(eptr--, ecode, offset_top, md)) return TRUE;
+          if (match(eptr--, ecode, offset_top, md, ims, FALSE, eptrb))
+            return TRUE;
          return FALSE;
          }
        }
-Line 3225 
 for (;;)
+Line 3969 
 for (;;)
      if (min > 0) switch(ctype)
        {
        case OP_ANY:
-       if (!md->dotall)
+       if ((ims & PCRE_DOTALL) == 0)
          { for (i = 1; i <= min; i++) if (*eptr++ == '\n') return FALSE; }
        else eptr += min;
        break;
        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 3266 
 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)) return TRUE;
+         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++, md->dotall))
-             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 3289 
 for (;;)
+Line 4065 
 for (;;)
        switch(ctype)
          {
          case OP_ANY:
-         if (!md->dotall)
+         if ((ims & PCRE_DOTALL) == 0)
            {
            for (i = min; i < max; i++)
              {
-Line 3308 
 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 3317 
 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 3326 
 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 3335 
 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 3344 
 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 3353 
 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 3361 
 for (;;)
+Line 4137 
 for (;;)
          }
        while (eptr >= pp)
-         if (match(eptr--, ecode, offset_top, md)) return TRUE;
+         if (match(eptr--, ecode, offset_top, md, ims, FALSE, eptrb))
+           return TRUE;
        return FALSE;
        }
      /* Control never gets here */
-Line 3384 
 for (;;)
+Line 4161 
 for (;;)
- /*************************************************
- *         Segregate setjmp()                     *
- *************************************************/
- /* The -Wall option of gcc gives warnings for all local variables when setjmp()
- is used, even if the coding conforms to the rules of ANSI C. To avoid this, we
- hide it in a separate function. This is called only when PCRE_EXTRA is set,
- since it's needed only for the extension \X option, and with any luck, a good
- compiler will spot the tail recursion and compile it efficiently.
- Arguments:
-    eptr        pointer in subject
-    ecode       position in code
-    offset_top  current top pointer
-    md          pointer to "static" info for the match
- Returns:       TRUE if matched
- */
- static BOOL
- match_with_setjmp(const uschar *eptr, const uschar *ecode, int offset_top,
-   match_data *match_block)
- {
- return setjmp(match_block->fail_env) == 0 &&
-       match(eptr, ecode, offset_top, match_block);
- }
  /*************************************************
  *         Execute a Regular Expression           *
-Line 3426 
 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 3438 
 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 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 3462 
 match_block.start_subject = (const uscha
+Line 4217 
 match_block.start_subject = (const uscha
  match_block.end_subject = match_block.start_subject + length;
  end_subject = match_block.end_subject;
- match_block.caseless  = ((re->options | options) & PCRE_CASELESS) != 0;
+ match_block.endonly = (re->options & PCRE_DOLLAR_ENDONLY) != 0;
- match_block.runtime_caseless = match_block.caseless &&
-   (re->options & PCRE_CASELESS) == 0;
- match_block.multiline = ((re->options | options) & PCRE_MULTILINE) != 0;
- match_block.dotall    = ((re->options | options) & PCRE_DOTALL) != 0;
- match_block.endonly   = ((re->options | options) & PCRE_DOLLAR_ENDONLY) != 0;
  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. */
+ ims = re->options & (PCRE_CASELESS|PCRE_MULTILINE|PCRE_DOTALL);
  /* If the expression has got more back references than the offsets supplied can
  hold, we get a temporary bit of working store to use during the matching.
  Otherwise, we can use the vector supplied, rounding down its size to a multiple
- of 2. */
+ of 3. */
- ocount = offsetcount & (-2);
+ ocount = offsetcount - (offsetcount % 3);
- if (re->top_backref > 0 && re->top_backref >= ocount/2)
+ if (re->top_backref > 0 && re->top_backref >= ocount/3)
    {
-   ocount = re->top_backref * 2 + 2;
+   ocount = re->top_backref * 3 + 3;
    match_block.offset_vector = (int *)(pcre_malloc)(ocount * sizeof(int));
    if (match_block.offset_vector == NULL) return PCRE_ERROR_NOMEMORY;
    using_temporary_offsets = TRUE;
-Line 3492 
 if (re->top_backref > 0 && re->top_backr
+Line 4252 
 if (re->top_backref > 0 && re->top_backr
  else match_block.offset_vector = offsets;
  match_block.offset_end = ocount;
+ match_block.offset_max = (2*ocount)/3;
  match_block.offset_overflow = FALSE;
  /* Compute the minimum number of offsets that we need to reset each time. Doing
-Line 3501 
 in the pattern. */
+Line 4262 
 in the pattern. */
  resetcount = 2 + re->top_bracket * 2;
  if (resetcount > offsetcount) resetcount = ocount;
- /* If MULTILINE is set at exec time but was not set at compile time, and the
+ /* Reset the working variable associated with each extraction. These should
- anchored flag is set, we must re-check because a setting provoked by ^ in the
+ never be used unless previously set, but they get saved and restored, and so we
- pattern is not right in multi-line mode. Calling is_anchored() again here does
+ initialize them to avoid reading uninitialized locations. */
- the right check, because multiline is now set. If it now yields FALSE, the
- expression must have had ^ starting some of its branches. Check to see if
- that is true for *all* branches, and if so, set the startline flag. */
- if (match_block. multiline && anchored && (re->options & PCRE_MULTILINE) == 0 &&
+ if (match_block.offset_vector != NULL)
-     !is_anchored(re->code, match_block.multiline))
    {
-   anchored = FALSE;
+   register int *iptr = match_block.offset_vector + ocount;
-   if (is_startline(re->code)) startline = TRUE;
+   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
  an unanchored pattern, of course. If there's no first char and the pattern was
- studied, the may be a bitmap of possible first characters. However, we can
+ studied, there may be a bitmap of possible first characters. */
- use this only if the caseless state of the studying was correct. */
  if (!anchored)
    {
    if ((re->options & PCRE_FIRSTSET) != 0)
      {
      first_char = re->first_char;
-     if (match_block.caseless) first_char = pcre_lcc[first_char];
+     if ((ims & PCRE_CASELESS) != 0) first_char = match_block.lcc[first_char];
      }
    else
      if (!startline && extra != NULL &&
-       (extra->options & PCRE_STUDY_MAPPED) != 0 &&
+       (extra->options & PCRE_STUDY_MAPPED) != 0)
-       ((extra->options & PCRE_STUDY_CASELESS) != 0) == match_block.caseless)
          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 3552 
 do
+Line 4324 
 do
    if (first_char >= 0)
      {
-     if (match_block.caseless)
+     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 3571 
 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 3588 
 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
    where we had to get some local store to hold offsets for backreferences, copy
    those back references that we can. In this case there need not be overflow
-   if certain parts of the pattern were not used.
+   if certain parts of the pattern were not used. */
-   Before starting the match, we have to set up a longjmp() target to enable
-   the "cut" operation to fail a match completely without backtracking. This
-   is done in a separate function to avoid compiler warnings. We need not do
-   it unless PCRE_EXTRA is set, since only in that case is the "cut" operation
-   enabled. */
-   if ((re->options & PCRE_EXTRA) != 0)
+   match_block.start_match = start_match;
-     {
+   if (!match(start_match, re->code, 2, &match_block, ims, FALSE, start_match))
-     if (!match_with_setjmp(start_match, re->code, 2, &match_block))
+     continue;
-       continue;
-     }
-   else if (!match(start_match, re->code, 2, &match_block)) continue;
    /* Copy the offset information from temporary store if necessary */
-Line 3636 
 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.19
 


changed lines


 
Added in v.37
 Legend:



Removed from v.19
 


changed lines


 
Added in v.37
-Removed from v.19
+Added in v.37

	ViewVC Help
Powered by ViewVC 1.1.5