/[pcre]/code/branches/pcre16/pcre_compile.c

Diff of /code/branches/pcre16/pcre_compile.c

Parent Directory | Revision Log | View Patch Patch

-revision 285 by ph10,
Wed Dec 12 17:03:50 2007 UTC
+revision 455 by ph10,
Sat Sep 26 19:12:32 2009 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2007 University of Cambridge
+            Copyright (c) 1997-2009 University of Cambridge
  -----------------------------------------------------------------------------
  Redistribution and use in source and binary forms, with or without
 Line 97 
 are simple data values; negative values
  on. Zero means further processing is needed (for things like \x), or the escape
  is invalid. */
- #ifndef EBCDIC  /* This is the "normal" table for ASCII systems */
+ #ifndef EBCDIC
+ /* This is the "normal" table for ASCII systems or for EBCDIC systems running
+ in UTF-8 mode. */
  static const short int escapes[] = {
-,      0,      0,      0,      0,      0,      0,      0,   /* 0 - 7 */
+,                       0,
-,      0,    ':',    ';',    '<',    '=',    '>',    '?',   /* 8 - ? */
+,                       0,
-    '@', -ESC_A, -ESC_B, -ESC_C, -ESC_D, -ESC_E,      0, -ESC_G,   /* @ - G */
+,                       0,
- -ESC_H,      0,      0, -ESC_K,      0,      0,      0,      0,   /* H - O */
+,                       0,
- -ESC_P, -ESC_Q, -ESC_R, -ESC_S,      0,      0, -ESC_V, -ESC_W,   /* P - W */
+,                       0,
- -ESC_X,      0, -ESC_Z,    '[',   '\\',    ']',    '^',    '_',   /* X - _ */
+      CHAR_COLON,              CHAR_SEMICOLON,
-    '`',      7, -ESC_b,      0, -ESC_d,  ESC_e,  ESC_f,      0,   /* ` - g */
+      CHAR_LESS_THAN_SIGN,     CHAR_EQUALS_SIGN,
- -ESC_h,      0,      0, -ESC_k,      0,      0,  ESC_n,      0,   /* h - o */
+      CHAR_GREATER_THAN_SIGN,  CHAR_QUESTION_MARK,
- -ESC_p,      0,  ESC_r, -ESC_s,  ESC_tee,    0, -ESC_v, -ESC_w,   /* p - w */
+      CHAR_COMMERCIAL_AT,      -ESC_A,
-,      0, -ESC_z                                            /* x - z */
+      -ESC_B,                  -ESC_C,
+      -ESC_D,                  -ESC_E,
+,                       -ESC_G,
+      -ESC_H,                  0,
+,                       -ESC_K,
+,                       0,
+,                       0,
+      -ESC_P,                  -ESC_Q,
+      -ESC_R,                  -ESC_S,
+,                       0,
+      -ESC_V,                  -ESC_W,
+      -ESC_X,                  0,
+      -ESC_Z,                  CHAR_LEFT_SQUARE_BRACKET,
+      CHAR_BACKSLASH,          CHAR_RIGHT_SQUARE_BRACKET,
+      CHAR_CIRCUMFLEX_ACCENT,  CHAR_UNDERSCORE,
+      CHAR_GRAVE_ACCENT,       7,
+      -ESC_b,                  0,
+      -ESC_d,                  ESC_e,
+      ESC_f,                   0,
+      -ESC_h,                  0,
+,                       -ESC_k,
+,                       0,
+      ESC_n,                   0,
+      -ESC_p,                  0,
+      ESC_r,                   -ESC_s,
+      ESC_tee,                 0,
+      -ESC_v,                  -ESC_w,
+,                       0,
+      -ESC_z
  };
- #else           /* This is the "abnormal" table for EBCDIC systems */
+ #else
+ /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
  static const short int escapes[] = {
  /*  48 */     0,     0,      0,     '.',    '<',   '(',    '+',    '|',
  /*  50 */   '&',     0,      0,       0,      0,     0,      0,      0,
-Line 142 
 static const short int escapes[] = {
+Line 177 
 static const short int escapes[] = {
  /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
  searched linearly. Put all the names into a single string, in order to reduce
- the number of relocations when a shared library is dynamically linked. */
+ the number of relocations when a shared library is dynamically linked. The
+ string is built from string macros so that it works in UTF-8 mode on EBCDIC
+ platforms. */
  typedef struct verbitem {
    int   len;
-Line 150 
 typedef struct verbitem {
+Line 187 
 typedef struct verbitem {
  } verbitem;
  static const char verbnames[] =
-   "ACCEPT\0"
+   STRING_ACCEPT0
-   "COMMIT\0"
+   STRING_COMMIT0
-   "F\0"
+   STRING_F0
-   "FAIL\0"
+   STRING_FAIL0
-   "PRUNE\0"
+   STRING_PRUNE0
-   "SKIP\0"
+   STRING_SKIP0
-   "THEN";
+   STRING_THEN;
- static verbitem verbs[] = {
+ static const verbitem verbs[] = {
    { 6, OP_ACCEPT },
    { 6, OP_COMMIT },
    { 1, OP_FAIL },
-Line 168 
 static verbitem verbs[] = {
+Line 205 
 static verbitem verbs[] = {
    { 4, OP_THEN  }
  };
- static int verbcount = sizeof(verbs)/sizeof(verbitem);
+ static const int verbcount = sizeof(verbs)/sizeof(verbitem);
  /* Tables of names of POSIX character classes and their lengths. The names are
-Line 178 
 length entry. The first three must be al
+Line 215 
 length entry. The first three must be al
  for handling case independence. */
  static const char posix_names[] =
-   "alpha\0"  "lower\0"  "upper\0"  "alnum\0"  "ascii\0"  "blank\0"
+   STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
-   "cntrl\0"  "digit\0"  "graph\0"  "print\0"  "punct\0"  "space\0"
+   STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
-   "word\0"   "xdigit";
+   STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
+   STRING_word0  STRING_xdigit;
  static const uschar posix_name_lengths[] = {
 , 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
-Line 295 
 static const char error_texts[] =
+Line 333 
 static const char error_texts[] =
    /* 55 */
    "repeating a DEFINE group is not allowed\0"
    "inconsistent NEWLINE options\0"
-   "\\g is not followed by a braced name or an optionally braced non-zero number\0"
+   "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
-   "(?+ or (?- or (?(+ or (?(- must be followed by a non-zero number\0"
+   "a numbered reference must not be zero\0"
    "(*VERB) with an argument is not supported\0"
    /* 60 */
    "(*VERB) not recognized\0"
    "number is too big\0"
    "subpattern name expected\0"
-   "digit expected after (?+";
+   "digit expected after (?+\0"
+   "] is an invalid data character in JavaScript compatibility mode";
  /* Table to identify digits and hex digits. This is used when compiling
-Line 321 
 For convenience, we use the same bit def
+Line 360 
 For convenience, we use the same bit def
  Then we can use ctype_digit and ctype_xdigit in the code. */
- #ifndef EBCDIC  /* This is the "normal" case, for ASCII systems */
+ #ifndef EBCDIC
+ /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
+ UTF-8 mode. */
  static const unsigned char digitab[] =
    {
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7 */
-Line 357 
 static const unsigned char digitab[] =
+Line 400 
 static const unsigned char digitab[] =
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
- #else           /* This is the "abnormal" case, for EBCDIC systems */
+ #else
+ /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
  static const unsigned char digitab[] =
    {
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7  0 */
-Line 454 
 static const char *
+Line 500 
 static const char *
  find_error_text(int n)
  {
  const char *s = error_texts;
- for (; n > 0; n--) while (*s++ != 0);
+ for (; n > 0; n--) while (*s++ != 0) {};
  return s;
  }
-Line 502 
 if (c == 0) *errorcodeptr = ERR1;
+Line 548 
 if (c == 0) *errorcodeptr = ERR1;
  in a table. A non-zero result is something that can be returned immediately.
  Otherwise further processing may be required. */
- #ifndef EBCDIC  /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
- else if (c < '0' || c > 'z') {}                           /* Not alphanumeric */
+ else if (c < CHAR_0 || c > CHAR_z) {}                     /* Not alphanumeric */
- else if ((i = escapes[c - '0']) != 0) c = i;
+ else if ((i = escapes[c - CHAR_0]) != 0) c = i;
  #else           /* EBCDIC coding */
  else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphanumeric */
-Line 523 
 else
+Line 569 
 else
      /* A number of Perl escapes are not handled by PCRE. We give an explicit
      error. */
-     case 'l':
+     case CHAR_l:
-     case 'L':
+     case CHAR_L:
-     case 'N':
+     case CHAR_N:
-     case 'u':
+     case CHAR_u:
-     case 'U':
+     case CHAR_U:
      *errorcodeptr = ERR37;
      break;
-     /* \g must be followed by a number, either plain or braced. If positive, it
+     /* \g must be followed by one of a number of specific things:
-     is an absolute backreference. If negative, it is a relative backreference.
-     This is a Perl 5.10 feature. Perl 5.10 also supports \g{name} as a
-     reference to a named group. This is part of Perl's movement towards a
-     unified syntax for back references. As this is synonymous with \k{name}, we
-     fudge it up by pretending it really was \k. */
-     case 'g':
+     (1) A number, either plain or braced. If positive, it is an absolute
-     if (ptr[1] == '{')
+     backreference. If negative, it is a relative backreference. This is a Perl
+.10 feature.
+     (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
+     is part of Perl's movement towards a unified syntax for back references. As
+     this is synonymous with \k{name}, we fudge it up by pretending it really
+     was \k.
+     (3) For Oniguruma compatibility we also support \g followed by a name or a
+     number either in angle brackets or in single quotes. However, these are
+     (possibly recursive) subroutine calls, _not_ backreferences. Just return
+     the -ESC_g code (cf \k). */
+     case CHAR_g:
+     if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
+       {
+       c = -ESC_g;
+       break;
+       }
+     /* Handle the Perl-compatible cases */
+     if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
        {
        const uschar *p;
-       for (p = ptr+2; *p != 0 && *p != '}'; p++)
+       for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
-         if (*p != '-' && (digitab[*p] & ctype_digit) == 0) break;
+         if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
-       if (*p != 0 && *p != '}')
+       if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
          {
          c = -ESC_k;
          break;
-Line 554 
 else
+Line 617 
 else
        }
      else braced = FALSE;
-     if (ptr[1] == '-')
+     if (ptr[1] == CHAR_MINUS)
        {
        negated = TRUE;
        ptr++;
-Line 563 
 else
+Line 626 
 else
      c = 0;
      while ((digitab[ptr[1]] & ctype_digit) != 0)
-       c = c * 10 + *(++ptr) - '0';
+       c = c * 10 + *(++ptr) - CHAR_0;
-     if (c < 0)
+     if (c < 0)   /* Integer overflow */
        {
        *errorcodeptr = ERR61;
        break;
        }
-     if (c == 0 || (braced && *(++ptr) != '}'))
+     if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
        {
        *errorcodeptr = ERR57;
        break;
        }
+     if (c == 0)
+       {
+       *errorcodeptr = ERR58;
+       break;
+       }
      if (negated)
        {
        if (c > bracount)
-Line 602 
 else
+Line 671 
 else
      value is greater than 377, the least significant 8 bits are taken. Inside a
      character class, \ followed by a digit is always an octal number. */
-     case '1': case '2': case '3': case '4': case '5':
+     case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
-     case '6': case '7': case '8': case '9':
+     case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
      if (!isclass)
        {
        oldptr = ptr;
-       c -= '0';
+       c -= CHAR_0;
        while ((digitab[ptr[1]] & ctype_digit) != 0)
-         c = c * 10 + *(++ptr) - '0';
+         c = c * 10 + *(++ptr) - CHAR_0;
-       if (c < 0)
+       if (c < 0)    /* Integer overflow */
          {
          *errorcodeptr = ERR61;
          break;
-Line 628 
 else
+Line 697 
 else
      generates a binary zero byte and treats the digit as a following literal.
      Thus we have to pull back the pointer by one. */
-     if ((c = *ptr) >= '8')
+     if ((c = *ptr) >= CHAR_8)
        {
        ptr--;
        c = 0;
-Line 641 
 else
+Line 710 
 else
      to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
      than 3 octal digits. */
-     case '0':
+     case CHAR_0:
-     c -= '0';
+     c -= CHAR_0;
-     while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7')
+     while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
-         c = c * 8 + *(++ptr) - '0';
+         c = c * 8 + *(++ptr) - CHAR_0;
      if (!utf8 && c > 255) *errorcodeptr = ERR51;
      break;
-Line 652 
 else
+Line 721 
 else
      than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
      treated as a data character. */
-     case 'x':
+     case CHAR_x:
-     if (ptr[1] == '{')
+     if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
        {
        const uschar *pt = ptr + 2;
        int count = 0;
-Line 662 
 else
+Line 731 
 else
        while ((digitab[*pt] & ctype_xdigit) != 0)
          {
          register int cc = *pt++;
-         if (c == 0 && cc == '0') continue;     /* Leading zeroes */
+         if (c == 0 && cc == CHAR_0) continue;     /* Leading zeroes */
          count++;
- #ifndef EBCDIC  /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
-         if (cc >= 'a') cc -= 32;               /* Convert to upper case */
+         if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
-         c = (c << 4) + cc - ((cc < 'A')? '0' : ('A' - 10));
+         c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
  #else           /* EBCDIC coding */
-         if (cc >= 'a' && cc <= 'z') cc += 64;  /* Convert to upper case */
+         if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
-         c = (c << 4) + cc - ((cc >= '0')? '0' : ('A' - 10));
+         c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
  #endif
          }
-       if (*pt == '}')
+       if (*pt == CHAR_RIGHT_CURLY_BRACKET)
          {
          if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
          ptr = pt;
-Line 690 
 else
+Line 759 
 else
      c = 0;
      while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
        {
-       int cc;                               /* Some compilers don't like ++ */
+       int cc;                                  /* Some compilers don't like */
-       cc = *(++ptr);                        /* in initializers */
+       cc = *(++ptr);                           /* ++ in initializers */
- #ifndef EBCDIC  /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
-       if (cc >= 'a') cc -= 32;              /* Convert to upper case */
+       if (cc >= CHAR_a) cc -= 32;              /* Convert to upper case */
-       c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));
+       c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
  #else           /* EBCDIC coding */
-       if (cc <= 'z') cc += 64;              /* Convert to upper case */
+       if (cc <= CHAR_z) cc += 64;              /* Convert to upper case */
-       c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));
+       c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
  #endif
        }
      break;
-Line 706 
 else
+Line 775 
 else
      This coding is ASCII-specific, but then the whole concept of \cx is
      ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
-     case 'c':
+     case CHAR_c:
      c = *(++ptr);
      if (c == 0)
        {
-Line 714 
 else
+Line 783 
 else
        break;
        }
- #ifndef EBCDIC  /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
-     if (c >= 'a' && c <= 'z') c -= 32;
+     if (c >= CHAR_a && c <= CHAR_z) c -= 32;
      c ^= 0x40;
  #else           /* EBCDIC coding */
-     if (c >= 'a' && c <= 'z') c += 64;
+     if (c >= CHAR_a && c <= CHAR_z) c += 64;
      c ^= 0xC0;
  #endif
      break;
-Line 780 
 if (c == 0) goto ERROR_RETURN;
+Line 849 
 if (c == 0) goto ERROR_RETURN;
  /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
  negation. */
- if (c == '{')
+ if (c == CHAR_LEFT_CURLY_BRACKET)
    {
-   if (ptr[1] == '^')
+   if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
      {
      *negptr = TRUE;
      ptr++;
-Line 791 
 if (c == '{')
+Line 860 
 if (c == '{')
      {
      c = *(++ptr);
      if (c == 0) goto ERROR_RETURN;
-     if (c == '}') break;
+     if (c == CHAR_RIGHT_CURLY_BRACKET) break;
      name[i] = c;
      }
-   if (c !='}') goto ERROR_RETURN;
+   if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
    name[i] = 0;
    }
-Line 859 
 is_counted_repeat(const uschar *p)
+Line 928 
 is_counted_repeat(const uschar *p)
  {
  if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
  while ((digitab[*p] & ctype_digit) != 0) p++;
- if (*p == '}') return TRUE;
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
- if (*p++ != ',') return FALSE;
+ if (*p++ != CHAR_COMMA) return FALSE;
- if (*p == '}') return TRUE;
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
  if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
  while ((digitab[*p] & ctype_digit) != 0) p++;
- return (*p == '}');
+ return (*p == CHAR_RIGHT_CURLY_BRACKET);
  }
-Line 900 
 int max = -1;
+Line 969 
 int max = -1;
  /* Read the minimum value and do a paranoid check: a negative value indicates
  an integer overflow. */
- while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0';
+ while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
  if (min < 0 || min > 65535)
    {
    *errorcodeptr = ERR5;
-Line 910 
 if (min < 0 || min > 65535)
+Line 979 
 if (min < 0 || min > 65535)
  /* Read the maximum value if there is one, and again do a paranoid on its size.
  Also, max must not be less than min. */
- if (*p == '}') max = min; else
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
    {
-   if (*(++p) != '}')
+   if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
      {
      max = 0;
-     while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0';
+     while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
      if (max < 0 || max > 65535)
        {
        *errorcodeptr = ERR5;
-Line 940 
 return p;
+Line 1009 
 return p;
  /*************************************************
- *       Find forward referenced subpattern       *
+ *  Subroutine for finding forward reference      *
  *************************************************/
- /* This function scans along a pattern's text looking for capturing
+ /* This recursive function is called only from find_parens() below. The
+ top-level call starts at the beginning of the pattern. All other calls must
+ start at a parenthesis. It scans along a pattern's text looking for capturing
  subpatterns, and counting them. If it finds a named pattern that matches the
  name it is given, it returns its number. Alternatively, if the name is NULL, it
- returns when it reaches a given numbered subpattern. This is used for forward
+ returns when it reaches a given numbered subpattern. We know that if (?P< is
- references to subpatterns. We know that if (?P< is encountered, the name will
+ encountered, the name will be terminated by '>' because that is checked in the
- be terminated by '>' because that is checked in the first pass.
+ first pass. Recursion is used to keep track of subpatterns that reset the
+ capturing group numbers - the (?| feature.
  Arguments:
-   ptr          current position in the pattern
+   ptrptr       address of the current character pointer (updated)
-   count        current count of capturing parens so far encountered
+   cd           compile background data
    name         name to seek, or NULL if seeking a numbered subpattern
    lorn         name length, or subpattern number if name is NULL
    xmode        TRUE if we are in /x mode
+   count        pointer to the current capturing subpattern number (updated)
  Returns:       the number of the named subpattern, or -1 if not found
  */
  static int
- find_parens(const uschar *ptr, int count, const uschar *name, int lorn,
+ find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
-   BOOL xmode)
+   BOOL xmode, int *count)
  {
- const uschar *thisname;
+ uschar *ptr = *ptrptr;
+ int start_count = *count;
+ int hwm_count = start_count;
+ BOOL dup_parens = FALSE;
- for (; *ptr != 0; ptr++)
+ /* If the first character is a parenthesis, check on the type of group we are
+ dealing with. The very first call may not start with a parenthesis. */
+ if (ptr[0] == CHAR_LEFT_PARENTHESIS)
    {
-   int term;
+   if (ptr[1] == CHAR_QUESTION_MARK &&
+       ptr[2] == CHAR_VERTICAL_LINE)
+     {
+     ptr += 3;
+     dup_parens = TRUE;
+     }
+   /* Handle a normal, unnamed capturing parenthesis */
+   else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
+     {
+     *count += 1;
+     if (name == NULL && *count == lorn) return *count;
+     ptr++;
+     }
+   /* Handle a condition. If it is an assertion, just carry on so that it
+   is processed as normal. If not, skip to the closing parenthesis of the
+   condition (there can't be any nested parens. */
+   else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
+     {
+     ptr += 2;
+     if (ptr[1] != CHAR_QUESTION_MARK)
+       {
+       while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
+       if (*ptr != 0) ptr++;
+       }
+     }
+   /* We have either (? or (* and not a condition */
+   else
+     {
+     ptr += 2;
+     if (*ptr == CHAR_P) ptr++;                      /* Allow optional P */
+     /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
+     if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
+         ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
+       {
+       int term;
+       const uschar *thisname;
+       *count += 1;
+       if (name == NULL && *count == lorn) return *count;
+       term = *ptr++;
+       if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
+       thisname = ptr;
+       while (*ptr != term) ptr++;
+       if (name != NULL && lorn == ptr - thisname &&
+           strncmp((const char *)name, (const char *)thisname, lorn) == 0)
+         return *count;
+       term++;
+       }
+     }
+   }
+ /* Past any initial parenthesis handling, scan for parentheses or vertical
+ bars. */
+ for (; *ptr != 0; ptr++)
+   {
    /* Skip over backslashed characters and also entire \Q...\E */
-   if (*ptr == '\\')
+   if (*ptr == CHAR_BACKSLASH)
      {
-     if (*(++ptr) == 0) return -1;
+     if (*(++ptr) == 0) goto FAIL_EXIT;
-     if (*ptr == 'Q') for (;;)
+     if (*ptr == CHAR_Q) for (;;)
        {
-       while (*(++ptr) != 0 && *ptr != '\\');
+       while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
-       if (*ptr == 0) return -1;
+       if (*ptr == 0) goto FAIL_EXIT;
-       if (*(++ptr) == 'E') break;
+       if (*(++ptr) == CHAR_E) break;
        }
      continue;
      }
-   /* Skip over character classes */
+   /* Skip over character classes; this logic must be similar to the way they
+   are handled for real. If the first character is '^', skip it. Also, if the
+   first few characters (either before or after ^) are \Q\E or \E we skip them
+   too. This makes for compatibility with Perl. Note the use of STR macros to
+   encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
-   if (*ptr == '[')
+   if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
      {
-     while (*(++ptr) != ']')
+     BOOL negate_class = FALSE;
+     for (;;)
+       {
+       if (ptr[1] == CHAR_BACKSLASH)
+         {
+         if (ptr[2] == CHAR_E)
+           ptr+= 2;
+         else if (strncmp((const char *)ptr+2,
+                  STR_Q STR_BACKSLASH STR_E, 3) == 0)
+           ptr += 4;
+         else
+           break;
+         }
+       else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
+         {
+         negate_class = TRUE;
+         ptr++;
+         }
+       else break;
+       }
+     /* If the next character is ']', it is a data character that must be
+     skipped, except in JavaScript compatibility mode. */
+     if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
+         (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
+       ptr++;
+     while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
        {
        if (*ptr == 0) return -1;
-       if (*ptr == '\\')
+       if (*ptr == CHAR_BACKSLASH)
          {
-         if (*(++ptr) == 0) return -1;
+         if (*(++ptr) == 0) goto FAIL_EXIT;
-         if (*ptr == 'Q') for (;;)
+         if (*ptr == CHAR_Q) for (;;)
            {
-           while (*(++ptr) != 0 && *ptr != '\\');
+           while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
-           if (*ptr == 0) return -1;
+           if (*ptr == 0) goto FAIL_EXIT;
-           if (*(++ptr) == 'E') break;
+           if (*(++ptr) == CHAR_E) break;
            }
          continue;
          }
-Line 1008 
 for (; *ptr != 0; ptr++)
+Line 1180 
 for (; *ptr != 0; ptr++)
    /* Skip comments in /x mode */
-   if (xmode && *ptr == '#')
+   if (xmode && *ptr == CHAR_NUMBER_SIGN)
      {
-     while (*(++ptr) != 0 && *ptr != '\n');
+     while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
-     if (*ptr == 0) return -1;
+     if (*ptr == 0) goto FAIL_EXIT;
      continue;
      }
-   /* An opening parens must now be a real metacharacter */
+   /* Check for the special metacharacters */
-   if (*ptr != '(') continue;
+   if (*ptr == CHAR_LEFT_PARENTHESIS)
-   if (ptr[1] != '?' && ptr[1] != '*')
      {
-     count++;
+     int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
-     if (name == NULL && count == lorn) return count;
+     if (rc > 0) return rc;
-     continue;
+     if (*ptr == 0) goto FAIL_EXIT;
      }
-   ptr += 2;
+   else if (*ptr == CHAR_RIGHT_PARENTHESIS)
-   if (*ptr == 'P') ptr++;                      /* Allow optional P */
+     {
+     if (dup_parens && *count < hwm_count) *count = hwm_count;
+     *ptrptr = ptr;
+     return -1;
+     }
-   /* We have to disambiguate (?<! and (?<= from (?<name> */
+   else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
+     {
+     if (*count > hwm_count) hwm_count = *count;
+     *count = start_count;
+     }
+   }
-   if ((*ptr != '<' || ptr[1] == '!' || ptr[1] == '=') &&
+ FAIL_EXIT:
-        *ptr != '\'')
+ *ptrptr = ptr;
-     continue;
+ return -1;
+ }
+ /*************************************************
+ *       Find forward referenced subpattern       *
+ *************************************************/
+ /* This function scans along a pattern's text looking for capturing
+ subpatterns, and counting them. If it finds a named pattern that matches the
+ name it is given, it returns its number. Alternatively, if the name is NULL, it
+ returns when it reaches a given numbered subpattern. This is used for forward
+ references to subpatterns. We used to be able to start this scan from the
+ current compiling point, using the current count value from cd->bracount, and
+ do it all in a single loop, but the addition of the possibility of duplicate
+ subpattern numbers means that we have to scan from the very start, in order to
+ take account of such duplicates, and to use a recursive function to keep track
+ of the different types of group.
-   count++;
+ Arguments:
+   cd           compile background data
+   name         name to seek, or NULL if seeking a numbered subpattern
+   lorn         name length, or subpattern number if name is NULL
+   xmode        TRUE if we are in /x mode
+ Returns:       the number of the found subpattern, or -1 if not found
+ */
+ static int
+ find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
+ {
+ uschar *ptr = (uschar *)cd->start_pattern;
+ int count = 0;
+ int rc;
+ /* If the pattern does not start with an opening parenthesis, the first call
+ to find_parens_sub() will scan right to the end (if necessary). However, if it
+ does start with a parenthesis, find_parens_sub() will return when it hits the
+ matching closing parens. That is why we have to have a loop. */
-   if (name == NULL && count == lorn) return count;
+ for (;;)
-   term = *ptr++;
+   {
-   if (term == '<') term = '>';
+   rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
-   thisname = ptr;
+   if (rc > 0 || *ptr++ == 0) break;
-   while (*ptr != term) ptr++;
-   if (name != NULL && lorn == ptr - thisname &&
-       strncmp((const char *)name, (const char *)thisname, lorn) == 0)
-     return count;
    }
- return -1;
+ return rc;
  }
  /*************************************************
  *      Find first significant op code            *
  *************************************************/
-Line 1116 
 for (;;)
+Line 1331 
 for (;;)
  /*************************************************
- *        Find the fixed length of a pattern      *
+ *        Find the fixed length of a branch       *
  *************************************************/
- /* Scan a pattern and compute the fixed length of subject that will match it,
+ /* Scan a branch and compute the fixed length of subject that will match it,
  if the length is fixed. This is needed for dealing with backward assertions.
- In UTF8 mode, the result is in characters rather than bytes.
+ In UTF8 mode, the result is in characters rather than bytes. The branch is
+ temporarily terminated with OP_END when this function is called.
+ This function is called when a backward assertion is encountered, so that if it
+ fails, the error message can point to the correct place in the pattern.
+ However, we cannot do this when the assertion contains subroutine calls,
+ because they can be forward references. We solve this by remembering this case
+ and doing the check at the end; a flag specifies which mode we are running in.
  Arguments:
    code     points to the start of the pattern (the bracket)
    options  the compiling options
+   atend    TRUE if called when the pattern is complete
+   cd       the "compile data" structure
- Returns:   the fixed length, or -1 if there is no fixed length,
+ Returns:   the fixed length,
+              or -1 if there is no fixed length,
               or -2 if \C was encountered
+              or -3 if an OP_RECURSE item was encountered and atend is FALSE
  */
  static int
- find_fixedlength(uschar *code, int options)
+ find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
  {
  int length = -1;
-Line 1145 
 branch, check the length against that of
+Line 1371 
 branch, check the length against that of
  for (;;)
    {
    int d;
+   uschar *ce, *cs;
    register int op = *cc;
    switch (op)
      {
-Line 1152 
 for (;;)
+Line 1379 
 for (;;)
      case OP_BRA:
      case OP_ONCE:
      case OP_COND:
-     d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options);
+     d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
      if (d < 0) return d;
      branchlength += d;
      do cc += GET(cc, 1); while (*cc == OP_ALT);
-Line 1174 
 for (;;)
+Line 1401 
 for (;;)
      cc += 1 + LINK_SIZE;
      branchlength = 0;
      break;
+     /* A true recursion implies not fixed length, but a subroutine call may
+     be OK. If the subroutine is a forward reference, we can't deal with
+     it until the end of the pattern, so return -3. */
+     case OP_RECURSE:
+     if (!atend) return -3;
+     cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */
+     do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */
+     if (cc > cs && cc < ce) return -1;                /* Recursion */
+     d = find_fixedlength(cs + 2, options, atend, cd);
+     if (d < 0) return d;
+     branchlength += d;
+     cc += 1 + LINK_SIZE;
+     break;
      /* Skip over assertive subpatterns */
-Line 1211 
 for (;;)
+Line 1453 
 for (;;)
      branchlength++;
      cc += 2;
  #ifdef SUPPORT_UTF8
-     if ((options & PCRE_UTF8) != 0)
+     if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
-       {
+       cc += _pcre_utf8_table4[cc[-1] & 0x3f];
-       while ((*cc & 0xc0) == 0x80) cc++;
-       }
  #endif
      break;
-Line 1225 
 for (;;)
+Line 1465 
 for (;;)
      branchlength += GET2(cc,1);
      cc += 4;
  #ifdef SUPPORT_UTF8
-     if ((options & PCRE_UTF8) != 0)
+     if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
-       {
+       cc += _pcre_utf8_table4[cc[-1] & 0x3f];
-       while((*cc & 0x80) == 0x80) cc++;
-       }
  #endif
      break;
-Line 1252 
 for (;;)
+Line 1490 
 for (;;)
      case OP_NOT_WORDCHAR:
      case OP_WORDCHAR:
      case OP_ANY:
+     case OP_ALLANY:
      branchlength++;
      cc++;
      break;
-Line 1306 
 for (;;)
+Line 1545 
 for (;;)
  /*************************************************
- *    Scan compiled regex for numbered bracket    *
+ *    Scan compiled regex for specific bracket    *
  *************************************************/
  /* This little function scans through a compiled pattern until it finds a
- capturing bracket with the given number.
+ capturing bracket with the given number, or, if the number is negative, an
+ instance of OP_REVERSE for a lookbehind. The function is global in the C sense
+ so that it can be called from pcre_study() when finding the minimum matching
+ length.
  Arguments:
    code        points to start of expression
    utf8        TRUE in UTF-8 mode
-   number      the required bracket number
+   number      the required bracket number or negative to find a lookbehind
  Returns:      pointer to the opcode for the bracket, or NULL if not found
  */
- static const uschar *
+ const uschar *
- find_bracket(const uschar *code, BOOL utf8, int number)
+ _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
  {
  for (;;)
    {
-Line 1333 
 for (;;)
+Line 1575 
 for (;;)
    the table is zero; the actual length is stored in the compiled code. */
    if (c == OP_XCLASS) code += GET(code, 1);
+   /* Handle recursion */
+   else if (c == OP_REVERSE)
+     {
+     if (number < 0) return (uschar *)code;
+     code += _pcre_OP_lengths[c];
+     }
    /* Handle capturing bracket */
-Line 1400 
 for (;;)
+Line 1650 
 for (;;)
        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
        break;
        }
+ #else
+     (void)(utf8);  /* Keep compiler happy by referencing function argument */
  #endif
      }
    }
-Line 1493 
 for (;;)
+Line 1745 
 for (;;)
        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
        break;
        }
+ #else
+     (void)(utf8);  /* Keep compiler happy by referencing function argument */
  #endif
      }
    }
-Line 1531 
 for (code = first_significant_code(code
+Line 1785 
 for (code = first_significant_code(code
    const uschar *ccode;
    c = *code;
    /* Skip over forward assertions; the other assertions are skipped by
    first_significant_code() with a TRUE final argument. */
    if (c == OP_ASSERT)
      {
      do code += GET(code, 1); while (*code == OP_ALT);
      c = *code;
      continue;
      }
    /* Groups with zero repeats can of course be empty; skip them. */
-   if (c == OP_BRAZERO || c == OP_BRAMINZERO)
+   if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
      {
      code += _pcre_OP_lengths[c];
      do code += GET(code, 1); while (*code == OP_ALT);
-Line 1559 
 for (code = first_significant_code(code
+Line 1813 
 for (code = first_significant_code(code
      BOOL empty_branch;
      if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */
-     /* Scan a closed bracket */
+     /* If a conditional group has only one branch, there is a second, implied,
+     empty branch, so just skip over the conditional, because it could be empty.
+     Otherwise, scan the individual branches of the group. */
-     empty_branch = FALSE;
+     if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
-     do
-       {
-       if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
-         empty_branch = TRUE;
        code += GET(code, 1);
+     else
+       {
+       empty_branch = FALSE;
+       do
+         {
+         if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
+           empty_branch = TRUE;
+         code += GET(code, 1);
+         }
+       while (*code == OP_ALT);
+       if (!empty_branch) return FALSE;   /* All branches are non-empty */
        }
-     while (*code == OP_ALT);
-     if (!empty_branch) return FALSE;   /* All branches are non-empty */
      c = *code;
      continue;
      }
-Line 1630 
 for (code = first_significant_code(code
+Line 1892 
 for (code = first_significant_code(code
      case OP_NOT_WORDCHAR:
      case OP_WORDCHAR:
      case OP_ANY:
+     case OP_ALLANY:
      case OP_ANYBYTE:
      case OP_CHAR:
      case OP_CHARNC:
-Line 1686 
 for (code = first_significant_code(code
+Line 1949 
 for (code = first_significant_code(code
      case OP_QUERY:
      case OP_MINQUERY:
      case OP_POSQUERY:
+     if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
+     break;
      case OP_UPTO:
      case OP_MINUPTO:
      case OP_POSUPTO:
-     if (utf8) while ((code[2] & 0xc0) == 0x80) code++;
+     if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
      break;
  #endif
      }
-Line 1737 
 return TRUE;
+Line 2003 
 return TRUE;
  *************************************************/
  /* This function is called when the sequence "[:" or "[." or "[=" is
- encountered in a character class. It checks whether this is followed by an
+ encountered in a character class. It checks whether this is followed by a
- optional ^ and then a sequence of letters, terminated by a matching ":]" or
+ sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
- ".]" or "=]".
+ reach an unescaped ']' without the special preceding character, return FALSE.
+ Originally, this function only recognized a sequence of letters between the
+ terminators, but it seems that Perl recognizes any sequence of characters,
+ though of course unknown POSIX names are subsequently rejected. Perl gives an
+ "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
+ didn't consider this to be a POSIX class. Likewise for [:1234:].
+ The problem in trying to be exactly like Perl is in the handling of escapes. We
+ have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
+ class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
+ below handles the special case of \], but does not try to do any other escape
+ processing. This makes it different from Perl for cases such as [:l\ower:]
+ where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
+ "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
+ I think.
- Argument:
+ Arguments:
    ptr      pointer to the initial [
    endptr   where to return the end pointer
-   cd       pointer to compile data
  Returns:   TRUE or FALSE
  */
  static BOOL
- check_posix_syntax(const uschar *ptr, const uschar **endptr, compile_data *cd)
+ check_posix_syntax(const uschar *ptr, const uschar **endptr)
  {
  int terminator;          /* Don't combine these lines; the Solaris cc */
  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
- if (*(++ptr) == '^') ptr++;
+ for (++ptr; *ptr != 0; ptr++)
- while ((cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
- if (*ptr == terminator && ptr[1] == ']')
    {
-   *endptr = ptr;
+   if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
-   return TRUE;
+     {
+     if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
+     if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
+       {
+       *endptr = ptr;
+       return TRUE;
+       }
+     }
    }
  return FALSE;
  }
-Line 1805 
 return -1;
+Line 2090 
 return -1;
  that is referenced. This means that groups can be replicated for fixed
  repetition simply by copying (because the recursion is allowed to refer to
  earlier groups that are outside the current group). However, when a group is
- optional (i.e. the minimum quantifier is zero), OP_BRAZERO is inserted before
+ optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
- it, after it has been compiled. This means that any OP_RECURSE items within it
+ inserted before it, after it has been compiled. This means that any OP_RECURSE
- that refer to the group itself or any contained groups have to have their
+ items within it that refer to the group itself or any contained groups have to
- offsets adjusted. That one of the jobs of this function. Before it is called,
+ have their offsets adjusted. That one of the jobs of this function. Before it
- the partially compiled regex must be temporarily terminated with OP_END.
+ is called, the partially compiled regex must be temporarily terminated with
+ OP_END.
  This function has been extended with the possibility of forward references for
  recursions and subroutine calls. It must also check the list of such references
-Line 1944 
 get_othercase_range(unsigned int *cptr,
+Line 2230 
 get_othercase_range(unsigned int *cptr,
  unsigned int c, othercase, next;
  for (c = *cptr; c <= d; c++)
-   { if ((othercase = _pcre_ucp_othercase(c)) != NOTACHAR) break; }
+   { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
  if (c > d) return FALSE;
-Line 1953 
 next = othercase + 1;
+Line 2239 
 next = othercase + 1;
  for (++c; c <= d; c++)
    {
-   if (_pcre_ucp_othercase(c) != next) break;
+   if (UCD_OTHERCASE(c) != next) break;
    next++;
    }
-Line 1999 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2285 
 if ((options & PCRE_EXTENDED) != 0)
    for (;;)
      {
      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
-     if (*ptr == '#')
+     if (*ptr == CHAR_NUMBER_SIGN)
        {
        while (*(++ptr) != 0)
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
-Line 2011 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2297 
 if ((options & PCRE_EXTENDED) != 0)
  /* If the next item is one that we can handle, get its value. A non-negative
  value is a character, a negative value is an escape value. */
- if (*ptr == '\\')
+ if (*ptr == CHAR_BACKSLASH)
    {
    int temperrorcode = 0;
    next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
-Line 2036 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2322 
 if ((options & PCRE_EXTENDED) != 0)
    for (;;)
      {
      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
-     if (*ptr == '#')
+     if (*ptr == CHAR_NUMBER_SIGN)
        {
        while (*(++ptr) != 0)
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
-Line 2047 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2333 
 if ((options & PCRE_EXTENDED) != 0)
  /* If the next thing is itself optional, we have to give up. */
- if (*ptr == '*' || *ptr == '?' || strncmp((char *)ptr, "{0,", 3) == 0)
+ if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
-   return FALSE;
+   strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
+     return FALSE;
  /* Now compare the next item with the previous opcode. If the previous is a
  positive single character match, "item" either contains the character or, if
-Line 2063 
 if (next >= 0) switch(op_code)
+Line 2350 
 if (next >= 0) switch(op_code)
    case OP_CHAR:
  #ifdef SUPPORT_UTF8
    if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+ #else
+   (void)(utf8_char);  /* Keep compiler happy by referencing function argument */
  #endif
    return item != next;
-Line 2081 
 if (next >= 0) switch(op_code)
+Line 2370 
 if (next >= 0) switch(op_code)
      unsigned int othercase;
      if (next < 128) othercase = cd->fcc[next]; else
  #ifdef SUPPORT_UCP
-     othercase = _pcre_ucp_othercase((unsigned int)next);
+     othercase = UCD_OTHERCASE((unsigned int)next);
  #else
      othercase = NOTACHAR;
  #endif
-Line 2094 
 if (next >= 0) switch(op_code)
+Line 2383 
 if (next >= 0) switch(op_code)
    /* For OP_NOT, "item" must be a single-byte character. */
    case OP_NOT:
-   if (next < 0) return FALSE;  /* Not a character */
    if (item == next) return TRUE;
    if ((options & PCRE_CASELESS) == 0) return FALSE;
  #ifdef SUPPORT_UTF8
-Line 2103 
 if (next >= 0) switch(op_code)
+Line 2391 
 if (next >= 0) switch(op_code)
      unsigned int othercase;
      if (next < 128) othercase = cd->fcc[next]; else
  #ifdef SUPPORT_UCP
-     othercase = _pcre_ucp_othercase(next);
+     othercase = UCD_OTHERCASE(next);
  #else
      othercase = NOTACHAR;
  #endif
-Line 2357 
 uschar classbits[32];
+Line 2645 
 uschar classbits[32];
  BOOL class_utf8;
  BOOL utf8 = (options & PCRE_UTF8) != 0;
  uschar *class_utf8data;
+ uschar *class_utf8data_base;
  uschar utf8_char[6];
  #else
  BOOL utf8 = FALSE;
-Line 2396 
 req_caseopt = ((options & PCRE_CASELESS)
+Line 2685 
 req_caseopt = ((options & PCRE_CASELESS)
  for (;; ptr++)
    {
    BOOL negate_class;
    BOOL should_flip_negation;
    BOOL possessive_quantifier;
    BOOL is_quantifier;
    BOOL is_recurse;
-Line 2484 
 for (;; ptr++)
+Line 2773 
 for (;; ptr++)
    if (inescq && c != 0)
      {
-     if (c == '\\' && ptr[1] == 'E')
+     if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
        {
        inescq = FALSE;
        ptr++;
-Line 2510 
 for (;; ptr++)
+Line 2799 
 for (;; ptr++)
    /* Fill in length of a previous callout, except when the next thing is
    a quantifier. */
-   is_quantifier = c == '*' || c == '+' || c == '?' ||
+   is_quantifier =
-     (c == '{' && is_counted_repeat(ptr+1));
+     c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
+     (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
    if (!is_quantifier && previous_callout != NULL &&
         after_manual_callout-- <= 0)
-Line 2526 
 for (;; ptr++)
+Line 2816 
 for (;; ptr++)
    if ((options & PCRE_EXTENDED) != 0)
      {
      if ((cd->ctypes[c] & ctype_space) != 0) continue;
-     if (c == '#')
+     if (c == CHAR_NUMBER_SIGN)
        {
        while (*(++ptr) != 0)
          {
-Line 2551 
 for (;; ptr++)
+Line 2841 
 for (;; ptr++)
      {
      /* ===================================================================*/
      case 0:                        /* The branch terminates at string end */
-     case '|':                      /* or | or ) */
+     case CHAR_VERTICAL_LINE:       /* or | or ) */
-     case ')':
+     case CHAR_RIGHT_PARENTHESIS:
      *firstbyteptr = firstbyte;
      *reqbyteptr = reqbyte;
      *codeptr = code;
-Line 2574 
 for (;; ptr++)
+Line 2864 
 for (;; ptr++)
      /* Handle single-character metacharacters. In multiline mode, ^ disables
      the setting of any following char as a first character. */
-     case '^':
+     case CHAR_CIRCUMFLEX_ACCENT:
      if ((options & PCRE_MULTILINE) != 0)
        {
        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
-Line 2583 
 for (;; ptr++)
+Line 2873 
 for (;; ptr++)
      *code++ = OP_CIRC;
      break;
-     case '$':
+     case CHAR_DOLLAR_SIGN:
      previous = NULL;
      *code++ = OP_DOLL;
      break;
-Line 2591 
 for (;; ptr++)
+Line 2881 
 for (;; ptr++)
      /* There can never be a first char if '.' is first, whatever happens about
      repeats. The value of reqbyte doesn't change either. */
-     case '.':
+     case CHAR_DOT:
      if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
      zerofirstbyte = firstbyte;
      zeroreqbyte = reqbyte;
      previous = code;
-     *code++ = OP_ANY;
+     *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
      break;
-Line 2611 
 for (;; ptr++)
+Line 2901 
 for (;; ptr++)
      opcode is compiled. It may optionally have a bit map for characters < 256,
      but those above are are explicitly listed afterwards. A flag byte tells
      whether the bitmap is present, and whether this is a negated class or not.
-     */
-     case '[':
+     In JavaScript compatibility mode, an isolated ']' causes an error. In
+     default (Perl) mode, it is treated as a data character. */
+     case CHAR_RIGHT_SQUARE_BRACKET:
+     if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
+       {
+       *errorcodeptr = ERR64;
+       goto FAILED;
+       }
+     goto NORMAL_CHAR;
+     case CHAR_LEFT_SQUARE_BRACKET:
      previous = code;
      /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
      they are encountered at the top level, so we'll do that too. */
-     if ((ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
+     if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
-         check_posix_syntax(ptr, &tempptr, cd))
+          ptr[1] == CHAR_EQUALS_SIGN) &&
+         check_posix_syntax(ptr, &tempptr))
        {
-       *errorcodeptr = (ptr[1] == ':')? ERR13 : ERR31;
+       *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
        goto FAILED;
        }
-Line 2634 
 for (;; ptr++)
+Line 2935 
 for (;; ptr++)
      for (;;)
        {
        c = *(++ptr);
-       if (c == '\\')
+       if (c == CHAR_BACKSLASH)
          {
-         if (ptr[1] == 'E') ptr++;
+         if (ptr[1] == CHAR_E)
-           else if (strncmp((const char *)ptr+1, "Q\\E", 3) == 0) ptr += 3;
+           ptr++;
-             else break;
+         else if (strncmp((const char *)ptr+1,
+                           STR_Q STR_BACKSLASH STR_E, 3) == 0)
+           ptr += 3;
+         else
+           break;
          }
-       else if (!negate_class && c == '^')
+       else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
          negate_class = TRUE;
        else break;
        }
-     /* If a class contains a negative special such as \S, we need to flip the
+     /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
-     negation flag at the end, so that support for characters > 255 works
+     an initial ']' is taken as a data character -- the code below handles
+     that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
+     [^] must match any character, so generate OP_ALLANY. */
+     if (c == CHAR_RIGHT_SQUARE_BRACKET &&
+         (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
+       {
+       *code++ = negate_class? OP_ALLANY : OP_FAIL;
+       if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
+       zerofirstbyte = firstbyte;
+       break;
+       }
+     /* If a class contains a negative special such as \S, we need to flip the
+     negation flag at the end, so that support for characters > 255 works
      correctly (they are all included in the class). */
      should_flip_negation = FALSE;
-Line 2668 
 for (;; ptr++)
+Line 2987 
 for (;; ptr++)
  #ifdef SUPPORT_UTF8
      class_utf8 = FALSE;                       /* No chars >= 256 */
      class_utf8data = code + LINK_SIZE + 2;    /* For UTF-8 items */
+     class_utf8data_base = class_utf8data;     /* For resetting in pass 1 */
  #endif
      /* Process characters until ] is reached. By writing this as a "do" it
-Line 2683 
 for (;; ptr++)
+Line 3003 
 for (;; ptr++)
          {                           /* Braces are required because the */
          GETCHARLEN(c, ptr, ptr);    /* macro generates multiple statements */
          }
+       /* In the pre-compile phase, accumulate the length of any UTF-8 extra
+       data and reset the pointer. This is so that very large classes that
+       contain a zillion UTF-8 characters no longer overwrite the work space
+       (which is on the stack). */
+       if (lengthptr != NULL)
+         {
+         *lengthptr += class_utf8data - class_utf8data_base;
+         class_utf8data = class_utf8data_base;
+         }
  #endif
        /* Inside \Q...\E everything is literal except \E */
        if (inescq)
          {
-         if (c == '\\' && ptr[1] == 'E')     /* If we are at \E */
+         if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)  /* If we are at \E */
            {
            inescq = FALSE;                   /* Reset literal state */
            ptr++;                            /* Skip the 'E' */
-Line 2704 
 for (;; ptr++)
+Line 3036 
 for (;; ptr++)
        [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
 .6 and 5.8 do. */
-       if (c == '[' &&
+       if (c == CHAR_LEFT_SQUARE_BRACKET &&
-           (ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
+           (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
-           check_posix_syntax(ptr, &tempptr, cd))
+            ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
          {
          BOOL local_negate = FALSE;
          int posix_class, taboffset, tabopt;
          register const uschar *cbits = cd->cbits;
          uschar pbits[32];
-         if (ptr[1] != ':')
+         if (ptr[1] != CHAR_COLON)
            {
            *errorcodeptr = ERR31;
            goto FAILED;
            }
          ptr += 2;
-         if (*ptr == '^')
+         if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
            {
            local_negate = TRUE;
            should_flip_negation = TRUE;  /* Note negative special */
            ptr++;
            }
-Line 2793 
 for (;; ptr++)
+Line 3125 
 for (;; ptr++)
        to 'or' into the one we are building. We assume they have more than one
        character in them, so set class_charcount bigger than one. */
-       if (c == '\\')
+       if (c == CHAR_BACKSLASH)
          {
          c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
          if (*errorcodeptr != 0) goto FAILED;
-         if (-c == ESC_b) c = '\b';       /* \b is backspace in a class */
+         if (-c == ESC_b) c = CHAR_BS;       /* \b is backspace in a class */
-         else if (-c == ESC_X) c = 'X';   /* \X is literal X in a class */
+         else if (-c == ESC_X) c = CHAR_X;   /* \X is literal X in a class */
-         else if (-c == ESC_R) c = 'R';   /* \R is literal R in a class */
+         else if (-c == ESC_R) c = CHAR_R;   /* \R is literal R in a class */
          else if (-c == ESC_Q)            /* Handle start of quoted string */
            {
-           if (ptr[1] == '\\' && ptr[2] == 'E')
+           if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
              {
              ptr += 2; /* avoid empty string */
              }
-Line 2826 
 for (;; ptr++)
+Line 3158 
 for (;; ptr++)
              continue;
              case ESC_D:
              should_flip_negation = TRUE;
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
              continue;
-Line 2835 
 for (;; ptr++)
+Line 3167 
 for (;; ptr++)
              continue;
              case ESC_W:
              should_flip_negation = TRUE;
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
              continue;
-Line 2845 
 for (;; ptr++)
+Line 3177 
 for (;; ptr++)
              continue;
              case ESC_S:
              should_flip_negation = TRUE;
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */
              continue;
-Line 3029 
 for (;; ptr++)
+Line 3361 
 for (;; ptr++)
        entirely. The code for handling \Q and \E is messy. */
        CHECK_RANGE:
-       while (ptr[1] == '\\' && ptr[2] == 'E')
+       while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
          {
          inescq = FALSE;
          ptr += 2;
-Line 3039 
 for (;; ptr++)
+Line 3371 
 for (;; ptr++)
        /* Remember \r or \n */
-       if (c == '\r' || c == '\n') cd->external_flags |= PCRE_HASCRORLF;
+       if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
        /* Check for range */
-       if (!inescq && ptr[1] == '-')
+       if (!inescq && ptr[1] == CHAR_MINUS)
          {
          int d;
          ptr += 2;
-         while (*ptr == '\\' && ptr[1] == 'E') ptr += 2;
+         while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
          /* If we hit \Q (not followed by \E) at this point, go into escaped
          mode. */
-         while (*ptr == '\\' && ptr[1] == 'Q')
+         while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
            {
            ptr += 2;
-           if (*ptr == '\\' && ptr[1] == 'E') { ptr += 2; continue; }
+           if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
+             { ptr += 2; continue; }
            inescq = TRUE;
            break;
            }
-         if (*ptr == 0 || (!inescq && *ptr == ']'))
+         if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
            {
            ptr = oldptr;
            goto LONE_SINGLE_CHARACTER;
-Line 3079 
 for (;; ptr++)
+Line 3412 
 for (;; ptr++)
          not any of the other escapes. Perl 5.6 treats a hyphen as a literal
          in such circumstances. */
-         if (!inescq && d == '\\')
+         if (!inescq && d == CHAR_BACKSLASH)
            {
            d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
            if (*errorcodeptr != 0) goto FAILED;
-Line 3089 
 for (;; ptr++)
+Line 3422 
 for (;; ptr++)
            if (d < 0)
              {
-             if (d == -ESC_b) d = '\b';
+             if (d == -ESC_b) d = CHAR_BS;
-             else if (d == -ESC_X) d = 'X';
+             else if (d == -ESC_X) d = CHAR_X;
-             else if (d == -ESC_R) d = 'R'; else
+             else if (d == -ESC_R) d = CHAR_R; else
                {
                ptr = oldptr;
                goto LONE_SINGLE_CHARACTER;  /* A few lines below */
-Line 3112 
 for (;; ptr++)
+Line 3445 
 for (;; ptr++)
          /* Remember \r or \n */
-         if (d == '\r' || d == '\n') cd->external_flags |= PCRE_HASCRORLF;
+         if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
          /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
          matching, we have to use an XCLASS with extra data items. Caseless
-Line 3232 
 for (;; ptr++)
+Line 3565 
 for (;; ptr++)
          if ((options & PCRE_CASELESS) != 0)
            {
            unsigned int othercase;
-           if ((othercase = _pcre_ucp_othercase(c)) != NOTACHAR)
+           if ((othercase = UCD_OTHERCASE(c)) != c)
              {
              *class_utf8data++ = XCL_SINGLE;
              class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
-Line 3259 
 for (;; ptr++)
+Line 3592 
 for (;; ptr++)
      /* Loop until ']' reached. This "while" is the end of the "do" above. */
-     while ((c = *(++ptr)) != 0 && (c != ']' || inescq));
+     while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
      if (c == 0)                          /* Missing terminating ']' */
        {
-Line 3348 
 we set the flag only if there is a liter
+Line 3681 
 we set the flag only if there is a liter
      zeroreqbyte = reqbyte;
      /* If there are characters with values > 255, we have to compile an
      extended class, with its own opcode, unless there was a negated special
      such as \S in the class, because in that case all characters > 255 are in
      the class, so any that were explicitly given as well can be ignored. If
      (when there are explicit characters > 255 that must be listed) there are no
      characters < 256, we can omit the bitmap in the actual compiled code. */
-Line 3381 
 we set the flag only if there is a liter
+Line 3714 
 we set the flag only if there is a liter
        }
  #endif
      /* If there are no characters > 255, set the opcode to OP_CLASS or
      OP_NCLASS, depending on whether the whole class was negated and whether
      there were negative specials such as \S in the class. Then copy the 32-byte
      map into the code vector, negating it if necessary. */
      *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
      if (negate_class)
        {
-Line 3404 
 we set the flag only if there is a liter
+Line 3737 
 we set the flag only if there is a liter
      /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
      has been tested above. */
-     case '{':
+     case CHAR_LEFT_CURLY_BRACKET:
      if (!is_quantifier) goto NORMAL_CHAR;
      ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
      if (*errorcodeptr != 0) goto FAILED;
      goto REPEAT;
-     case '*':
+     case CHAR_ASTERISK:
      repeat_min = 0;
      repeat_max = -1;
      goto REPEAT;
-     case '+':
+     case CHAR_PLUS:
      repeat_min = 1;
      repeat_max = -1;
      goto REPEAT;
-     case '?':
+     case CHAR_QUESTION_MARK:
      repeat_min = 0;
      repeat_max = 1;
-Line 3455 
 we set the flag only if there is a liter
+Line 3788 
 we set the flag only if there is a liter
      but if PCRE_UNGREEDY is set, it works the other way round. We change the
      repeat type to the non-default. */
-     if (ptr[1] == '+')
+     if (ptr[1] == CHAR_PLUS)
        {
        repeat_type = 0;                  /* Force greedy */
        possessive_quantifier = TRUE;
        ptr++;
        }
-     else if (ptr[1] == '?')
+     else if (ptr[1] == CHAR_QUESTION_MARK)
        {
        repeat_type = greedy_non_default;
        ptr++;
-Line 3576 
 we set the flag only if there is a liter
+Line 3909 
 we set the flag only if there is a liter
        if (repeat_max == 0) goto END_REPEAT;
+       /*--------------------------------------------------------------------*/
+       /* This code is obsolete from release 8.00; the restriction was finally
+       removed: */
        /* All real repeats make it impossible to handle partial matching (maybe
        one day we will be able to remove this restriction). */
-       if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
+       /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
+       /*--------------------------------------------------------------------*/
        /* Combine the op_type with the repeat_type */
-Line 3726 
 we set the flag only if there is a liter
+Line 4064 
 we set the flag only if there is a liter
          goto END_REPEAT;
          }
+       /*--------------------------------------------------------------------*/
+       /* This code is obsolete from release 8.00; the restriction was finally
+       removed: */
        /* All real repeats make it impossible to handle partial matching (maybe
        one day we will be able to remove this restriction). */
-       if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
+       /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
+       /*--------------------------------------------------------------------*/
        if (repeat_min == 0 && repeat_max == -1)
          *code++ = OP_CRSTAR + repeat_type;
-Line 3787 
 we set the flag only if there is a liter
+Line 4130 
 we set the flag only if there is a liter
        if (repeat_min == 0)
          {
-         /* If the maximum is also zero, we just omit the group from the output
+         /* If the maximum is also zero, we used to just omit the group from the
-         altogether. */
+         output altogether, like this:
-         if (repeat_max == 0)
-           {
-           code = previous;
-           goto END_REPEAT;
-           }
-         /* If the maximum is 1 or unlimited, we just have to stick in the
+         ** if (repeat_max == 0)
-         BRAZERO and do no more at this point. However, we do need to adjust
+         **   {
-         any OP_RECURSE calls inside the group that refer to the group itself or
+         **   code = previous;
-         any internal or forward referenced group, because the offset is from
+         **   goto END_REPEAT;
-         the start of the whole regex. Temporarily terminate the pattern while
+         **   }
-         doing this. */
+         However, that fails when a group is referenced as a subroutine from
+         elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
+         so that it is skipped on execution. As we don't have a list of which
+         groups are referenced, we cannot do this selectively.
+         If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
+         and do no more at this point. However, we do need to adjust any
+         OP_RECURSE calls inside the group that refer to the group itself or any
+         internal or forward referenced group, because the offset is from the
+         start of the whole regex. Temporarily terminate the pattern while doing
+         this. */
-         if (repeat_max <= 1)
+         if (repeat_max <= 1)    /* Covers 0, 1, and unlimited */
            {
            *code = OP_END;
            adjust_recurse(previous, 1, utf8, cd, save_hwm);
            memmove(previous+1, previous, len);
            code++;
+           if (repeat_max == 0)
+             {
+             *previous++ = OP_SKIPZERO;
+             goto END_REPEAT;
+             }
            *previous++ = OP_BRAZERO + repeat_type;
            }
-Line 4003 
 we set the flag only if there is a liter
+Line 4356 
 we set the flag only if there is a liter
          }
        }
+     /* If previous is OP_FAIL, it was generated by an empty class [] in
+     JavaScript mode. The other ways in which OP_FAIL can be generated, that is
+     by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
+     error above. We can just ignore the repeat in JS case. */
+     else if (*previous == OP_FAIL) goto END_REPEAT;
      /* Else there's some kind of shambles */
      else
-Line 4027 
 we set the flag only if there is a liter
+Line 4387 
 we set the flag only if there is a liter
      if (possessive_quantifier)
        {
        int len;
-       if (*tempcode == OP_EXACT || *tempcode == OP_TYPEEXACT ||
-           *tempcode == OP_NOTEXACT)
+       if (*tempcode == OP_TYPEEXACT)
          tempcode += _pcre_OP_lengths[*tempcode] +
-           ((*tempcode == OP_TYPEEXACT &&
+           ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
-              (tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP))? 2:0);
+       else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
+         {
+         tempcode += _pcre_OP_lengths[*tempcode];
+ #ifdef SUPPORT_UTF8
+         if (utf8 && tempcode[-1] >= 0xc0)
+           tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
+ #endif
+         }
        len = code - tempcode;
        if (len > 0) switch (*tempcode)
          {
-Line 4077 
 we set the flag only if there is a liter
+Line 4446 
 we set the flag only if there is a liter
      lookbehind or option setting or condition or all the other extended
      parenthesis forms.  */
-     case '(':
+     case CHAR_LEFT_PARENTHESIS:
      newoptions = options;
      skipbytes = 0;
      bravalue = OP_CBRA;
-Line 4086 
 we set the flag only if there is a liter
+Line 4455 
 we set the flag only if there is a liter
      /* First deal with various "verbs" that can be introduced by '*'. */
-     if (*(++ptr) == '*' && (cd->ctypes[ptr[1]] & ctype_letter) != 0)
+     if (*(++ptr) == CHAR_ASTERISK && (cd->ctypes[ptr[1]] & ctype_letter) != 0)
        {
        int i, namelen;
        const char *vn = verbnames;
        const uschar *name = ++ptr;
        previous = NULL;
-       while ((cd->ctypes[*++ptr] & ctype_letter) != 0);
+       while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
-       if (*ptr == ':')
+       if (*ptr == CHAR_COLON)
          {
          *errorcodeptr = ERR59;   /* Not supported */
          goto FAILED;
          }
-       if (*ptr != ')')
+       if (*ptr != CHAR_RIGHT_PARENTHESIS)
          {
          *errorcodeptr = ERR60;
          goto FAILED;
-Line 4109 
 we set the flag only if there is a liter
+Line 4478 
 we set the flag only if there is a liter
          if (namelen == verbs[i].len &&
              strncmp((char *)name, vn, namelen) == 0)
            {
-           *code = verbs[i].op;
+           /* Check for open captures before ACCEPT */
-           if (*code++ == OP_ACCEPT) cd->had_accept = TRUE;
+           if (verbs[i].op == OP_ACCEPT)
+             {
+             open_capitem *oc;
+             cd->had_accept = TRUE;
+             for (oc = cd->open_caps; oc != NULL; oc = oc->next)
+               {
+               *code++ = OP_CLOSE;
+               PUT2INC(code, 0, oc->number);
+               }
+             }
+           *code++ = verbs[i].op;
            break;
            }
          vn += verbs[i].len + 1;
-Line 4123 
 we set the flag only if there is a liter
+Line 4503 
 we set the flag only if there is a liter
      /* Deal with the extended parentheses; all are introduced by '?', and the
      appearance of any of them means that this is not a capturing group. */
-     else if (*ptr == '?')
+     else if (*ptr == CHAR_QUESTION_MARK)
        {
        int i, set, unset, namelen;
        int *optset;
-Line 4132 
 we set the flag only if there is a liter
+Line 4512 
 we set the flag only if there is a liter
        switch (*(++ptr))
          {
-         case '#':                 /* Comment; skip to ket */
+         case CHAR_NUMBER_SIGN:                 /* Comment; skip to ket */
          ptr++;
-         while (*ptr != 0 && *ptr != ')') ptr++;
+         while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
          if (*ptr == 0)
            {
            *errorcodeptr = ERR18;
-Line 4144 
 we set the flag only if there is a liter
+Line 4524 
 we set the flag only if there is a liter
          /* ------------------------------------------------------------ */
-         case '|':                 /* Reset capture count for each branch */
+         case CHAR_VERTICAL_LINE:  /* Reset capture count for each branch */
          reset_bracount = TRUE;
          /* Fall through */
          /* ------------------------------------------------------------ */
-         case ':':                 /* Non-capturing bracket */
+         case CHAR_COLON:          /* Non-capturing bracket */
          bravalue = OP_BRA;
          ptr++;
          break;
          /* ------------------------------------------------------------ */
-         case '(':
+         case CHAR_LEFT_PARENTHESIS:
          bravalue = OP_COND;       /* Conditional group */
          /* A condition can be an assertion, a number (referring to a numbered
-Line 4176 
 we set the flag only if there is a liter
+Line 4556 
 we set the flag only if there is a liter
          the switch. This will take control down to where bracketed groups,
          including assertions, are processed. */
-         if (ptr[1] == '?' && (ptr[2] == '=' || ptr[2] == '!' || ptr[2] == '<'))
+         if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
+             ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
            break;
          /* Most other conditions use OP_CREF (a couple change to OP_RREF
-Line 4188 
 we set the flag only if there is a liter
+Line 4569 
 we set the flag only if there is a liter
          /* Check for a test for recursion in a named group. */
-         if (ptr[1] == 'R' && ptr[2] == '&')
+         if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
            {
            terminator = -1;
            ptr += 2;
-Line 4198 
 we set the flag only if there is a liter
+Line 4579 
 we set the flag only if there is a liter
          /* Check for a test for a named group's having been set, using the Perl
          syntax (?(<name>) or (?('name') */
-         else if (ptr[1] == '<')
+         else if (ptr[1] == CHAR_LESS_THAN_SIGN)
            {
-           terminator = '>';
+           terminator = CHAR_GREATER_THAN_SIGN;
            ptr++;
            }
-         else if (ptr[1] == '\'')
+         else if (ptr[1] == CHAR_APOSTROPHE)
            {
-           terminator = '\'';
+           terminator = CHAR_APOSTROPHE;
            ptr++;
            }
          else
            {
            terminator = 0;
-           if (ptr[1] == '-' || ptr[1] == '+') refsign = *(++ptr);
+           if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
            }
          /* We now expect to read a name; any thing else is an error */
-Line 4231 
 we set the flag only if there is a liter
+Line 4612 
 we set the flag only if there is a liter
            {
            if (recno >= 0)
              recno = ((digitab[*ptr] & ctype_digit) != 0)?
-               recno * 10 + *ptr - '0' : -1;
+               recno * 10 + *ptr - CHAR_0 : -1;
            ptr++;
            }
          namelen = ptr - name;
-         if ((terminator > 0 && *ptr++ != terminator) || *ptr++ != ')')
+         if ((terminator > 0 && *ptr++ != terminator) ||
+             *ptr++ != CHAR_RIGHT_PARENTHESIS)
            {
            ptr--;      /* Error offset */
            *errorcodeptr = ERR26;
-Line 4258 
 we set the flag only if there is a liter
+Line 4640 
 we set the flag only if there is a liter
              *errorcodeptr = ERR58;
              goto FAILED;
              }
-           recno = (refsign == '-')?
+           recno = (refsign == CHAR_MINUS)?
              cd->bracount - recno + 1 : recno +cd->bracount;
            if (recno <= 0 || recno > cd->final_bracount)
              {
-Line 4289 
 we set the flag only if there is a liter
+Line 4671 
 we set the flag only if there is a liter
          /* Search the pattern for a forward reference */
-         else if ((i = find_parens(ptr, cd->bracount, name, namelen,
+         else if ((i = find_parens(cd, name, namelen,
                          (options & PCRE_EXTENDED) != 0)) > 0)
            {
            PUT2(code, 2+LINK_SIZE, i);
-Line 4310 
 we set the flag only if there is a liter
+Line 4692 
 we set the flag only if there is a liter
          /* Check for (?(R) for recursion. Allow digits after R to specify a
          specific group number. */
-         else if (*name == 'R')
+         else if (*name == CHAR_R)
            {
            recno = 0;
            for (i = 1; i < namelen; i++)
-Line 4320 
 we set the flag only if there is a liter
+Line 4702 
 we set the flag only if there is a liter
                *errorcodeptr = ERR15;
                goto FAILED;
                }
-             recno = recno * 10 + name[i] - '0';
+             recno = recno * 10 + name[i] - CHAR_0;
              }
            if (recno == 0) recno = RREF_ANY;
            code[1+LINK_SIZE] = OP_RREF;      /* Change test type */
-Line 4330 
 we set the flag only if there is a liter
+Line 4712 
 we set the flag only if there is a liter
          /* Similarly, check for the (?(DEFINE) "condition", which is always
          false. */
-         else if (namelen == 6 && strncmp((char *)name, "DEFINE", 6) == 0)
+         else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
            {
            code[1+LINK_SIZE] = OP_DEF;
            skipbytes = 1;
            }
          /* Check for the "name" actually being a subpattern number. We are
          in the second pass here, so final_bracount is set. */
          else if (recno > 0 && recno <= cd->final_bracount)
            {
-Line 4355 
 we set the flag only if there is a liter
+Line 4737 
 we set the flag only if there is a liter
          /* ------------------------------------------------------------ */
-         case '=':                 /* Positive lookahead */
+         case CHAR_EQUALS_SIGN:                 /* Positive lookahead */
          bravalue = OP_ASSERT;
          ptr++;
          break;
          /* ------------------------------------------------------------ */
-         case '!':                 /* Negative lookahead */
+         case CHAR_EXCLAMATION_MARK:            /* Negative lookahead */
          ptr++;
-         if (*ptr == ')')          /* Optimize (?!) */
+         if (*ptr == CHAR_RIGHT_PARENTHESIS)    /* Optimize (?!) */
            {
            *code++ = OP_FAIL;
            previous = NULL;
-Line 4375 
 we set the flag only if there is a liter
+Line 4757 
 we set the flag only if there is a liter
          /* ------------------------------------------------------------ */
-         case '<':                 /* Lookbehind or named define */
+         case CHAR_LESS_THAN_SIGN:              /* Lookbehind or named define */
          switch (ptr[1])
            {
-           case '=':               /* Positive lookbehind */
+           case CHAR_EQUALS_SIGN:               /* Positive lookbehind */
            bravalue = OP_ASSERTBACK;
            ptr += 2;
            break;
-           case '!':               /* Negative lookbehind */
+           case CHAR_EXCLAMATION_MARK:          /* Negative lookbehind */
            bravalue = OP_ASSERTBACK_NOT;
            ptr += 2;
            break;
-Line 4398 
 we set the flag only if there is a liter
+Line 4780 
 we set the flag only if there is a liter
          /* ------------------------------------------------------------ */
-         case '>':                 /* One-time brackets */
+         case CHAR_GREATER_THAN_SIGN:           /* One-time brackets */
          bravalue = OP_ONCE;
          ptr++;
          break;
          /* ------------------------------------------------------------ */
-         case 'C':                 /* Callout - may be followed by digits; */
+         case CHAR_C:                 /* Callout - may be followed by digits; */
          previous_callout = code;  /* Save for later completion */
          after_manual_callout = 1; /* Skip one item before completing */
          *code++ = OP_CALLOUT;
            {
            int n = 0;
            while ((digitab[*(++ptr)] & ctype_digit) != 0)
-             n = n * 10 + *ptr - '0';
+             n = n * 10 + *ptr - CHAR_0;
-           if (*ptr != ')')
+           if (*ptr != CHAR_RIGHT_PARENTHESIS)
              {
              *errorcodeptr = ERR39;
              goto FAILED;
-Line 4433 
 we set the flag only if there is a liter
+Line 4815 
 we set the flag only if there is a liter
          /* ------------------------------------------------------------ */
-         case 'P':                 /* Python-style named subpattern handling */
+         case CHAR_P:              /* Python-style named subpattern handling */
-         if (*(++ptr) == '=' || *ptr == '>')  /* Reference or recursion */
+         if (*(++ptr) == CHAR_EQUALS_SIGN ||
+             *ptr == CHAR_GREATER_THAN_SIGN)  /* Reference or recursion */
            {
-           is_recurse = *ptr == '>';
+           is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
-           terminator = ')';
+           terminator = CHAR_RIGHT_PARENTHESIS;
            goto NAMED_REF_OR_RECURSE;
            }
-         else if (*ptr != '<')    /* Test for Python-style definition */
+         else if (*ptr != CHAR_LESS_THAN_SIGN)  /* Test for Python-style defn */
            {
            *errorcodeptr = ERR41;
            goto FAILED;
-Line 4450 
 we set the flag only if there is a liter
+Line 4833 
 we set the flag only if there is a liter
          /* ------------------------------------------------------------ */
          DEFINE_NAME:    /* Come here from (?< handling */
-         case '\'':
+         case CHAR_APOSTROPHE:
            {
-           terminator = (*ptr == '<')? '>' : '\'';
+           terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
+             CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
            name = ++ptr;
            while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
-Line 4526 
 we set the flag only if there is a liter
+Line 4910 
 we set the flag only if there is a liter
          /* ------------------------------------------------------------ */
-         case '&':                 /* Perl recursion/subroutine syntax */
+         case CHAR_AMPERSAND:            /* Perl recursion/subroutine syntax */
-         terminator = ')';
+         terminator = CHAR_RIGHT_PARENTHESIS;
          is_recurse = TRUE;
          /* Fall through */
-Line 4535 
 we set the flag only if there is a liter
+Line 4919 
 we set the flag only if there is a liter
          references (?P=name) and recursion (?P>name), as well as falling
          through from the Perl recursion syntax (?&name). We also come here from
          the Perl \k<name> or \k'name' back reference syntax and the \k{name}
-         .NET syntax. */
+         .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
          NAMED_REF_OR_RECURSE:
          name = ++ptr;
-Line 4551 
 we set the flag only if there is a liter
+Line 4935 
 we set the flag only if there is a liter
              {
              *errorcodeptr = ERR62;
              goto FAILED;
              }
            if (*ptr != terminator)
              {
              *errorcodeptr = ERR42;
-Line 4565 
 we set the flag only if there is a liter
+Line 4949 
 we set the flag only if there is a liter
            recno = 0;
            }
          /* In the real compile, seek the name in the table. We check the name
          first, and then check that we have reached the end of the name in the
          table. That way, if the name that is longer than any in the table,
          the comparison will fail without reading beyond the table entry. */
-Line 4576 
 we set the flag only if there is a liter
+Line 4960 
 we set the flag only if there is a liter
            for (i = 0; i < cd->names_found; i++)
              {
              if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
                  slot[2+namelen] == 0)
                break;
              slot += cd->name_entry_size;
              }
-Line 4586 
 we set the flag only if there is a liter
+Line 4970 
 we set the flag only if there is a liter
              recno = GET2(slot, 0);
              }
            else if ((recno =                /* Forward back reference */
-                     find_parens(ptr, cd->bracount, name, namelen,
+                     find_parens(cd, name, namelen,
                        (options & PCRE_EXTENDED) != 0)) <= 0)
              {
              *errorcodeptr = ERR15;
-Line 4602 
 we set the flag only if there is a liter
+Line 4986 
 we set the flag only if there is a liter
          /* ------------------------------------------------------------ */
-         case 'R':                 /* Recursion */
+         case CHAR_R:              /* Recursion */
          ptr++;                    /* Same as (?0)      */
          /* Fall through */
          /* ------------------------------------------------------------ */
-         case '-': case '+':
+         case CHAR_MINUS: case CHAR_PLUS:  /* Recursion or subroutine */
-         case '0': case '1': case '2': case '3': case '4':   /* Recursion or */
+         case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
-         case '5': case '6': case '7': case '8': case '9':   /* subroutine */
+         case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
            {
            const uschar *called;
+           terminator = CHAR_RIGHT_PARENTHESIS;
+           /* Come here from the \g<...> and \g'...' code (Oniguruma
+           compatibility). However, the syntax has been checked to ensure that
+           the ... are a (signed) number, so that neither ERR63 nor ERR29 will
+           be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
+           ever be taken. */
+           HANDLE_NUMERICAL_RECURSION:
-           if ((refsign = *ptr) == '+')
+           if ((refsign = *ptr) == CHAR_PLUS)
              {
              ptr++;
              if ((digitab[*ptr] & ctype_digit) == 0)
                {
                *errorcodeptr = ERR63;
                goto FAILED;
                }
              }
-           else if (refsign == '-')
+           else if (refsign == CHAR_MINUS)
              {
              if ((digitab[ptr[1]] & ctype_digit) == 0)
                goto OTHER_CHAR_AFTER_QUERY;
-Line 4632 
 we set the flag only if there is a liter
+Line 5025 
 we set the flag only if there is a liter
            recno = 0;
            while((digitab[*ptr] & ctype_digit) != 0)
-             recno = recno * 10 + *ptr++ - '0';
+             recno = recno * 10 + *ptr++ - CHAR_0;
-           if (*ptr != ')')
+           if (*ptr != terminator)
              {
              *errorcodeptr = ERR29;
              goto FAILED;
              }
-           if (refsign == '-')
+           if (refsign == CHAR_MINUS)
              {
              if (recno == 0)
                {
-Line 4654 
 we set the flag only if there is a liter
+Line 5047 
 we set the flag only if there is a liter
                goto FAILED;
                }
              }
-           else if (refsign == '+')
+           else if (refsign == CHAR_PLUS)
              {
              if (recno == 0)
                {
-Line 4681 
 we set the flag only if there is a liter
+Line 5074 
 we set the flag only if there is a liter
            if (lengthptr == NULL)
              {
              *code = OP_END;
-             if (recno != 0) called = find_bracket(cd->start_code, utf8, recno);
+             if (recno != 0)
+               called = _pcre_find_bracket(cd->start_code, utf8, recno);
              /* Forward reference */
              if (called == NULL)
                {
-               if (find_parens(ptr, cd->bracount, NULL, recno,
+               if (find_parens(cd, NULL, recno,
-                    (options & PCRE_EXTENDED) != 0) < 0)
+                     (options & PCRE_EXTENDED) != 0) < 0)
                  {
                  *errorcodeptr = ERR15;
                  goto FAILED;
-Line 4740 
 we set the flag only if there is a liter
+Line 5134 
 we set the flag only if there is a liter
          set = unset = 0;
          optset = &set;
-         while (*ptr != ')' && *ptr != ':')
+         while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
            {
            switch (*ptr++)
              {
-             case '-': optset = &unset; break;
+             case CHAR_MINUS: optset = &unset; break;
-             case 'J':    /* Record that it changed in the external options */
+             case CHAR_J:    /* Record that it changed in the external options */
              *optset |= PCRE_DUPNAMES;
              cd->external_flags |= PCRE_JCHANGED;
              break;
-             case 'i': *optset |= PCRE_CASELESS; break;
+             case CHAR_i: *optset |= PCRE_CASELESS; break;
-             case 'm': *optset |= PCRE_MULTILINE; break;
+             case CHAR_m: *optset |= PCRE_MULTILINE; break;
-             case 's': *optset |= PCRE_DOTALL; break;
+             case CHAR_s: *optset |= PCRE_DOTALL; break;
-             case 'x': *optset |= PCRE_EXTENDED; break;
+             case CHAR_x: *optset |= PCRE_EXTENDED; break;
-             case 'U': *optset |= PCRE_UNGREEDY; break;
+             case CHAR_U: *optset |= PCRE_UNGREEDY; break;
-             case 'X': *optset |= PCRE_EXTRA; break;
+             case CHAR_X: *optset |= PCRE_EXTRA; break;
              default:  *errorcodeptr = ERR12;
                        ptr--;    /* Correct the offset */
-Line 4787 
 we set the flag only if there is a liter
+Line 5181 
 we set the flag only if there is a liter
          both phases.
          If we are not at the pattern start, compile code to change the ims
-         options if this setting actually changes any of them. We also pass the
+         options if this setting actually changes any of them, and reset the
-         new setting back so that it can be put at the start of any following
+         greedy defaults and the case value for firstbyte and reqbyte. */
-         branches, and when this group ends (if we are in a group), a resetting
-         item can be compiled. */
-         if (*ptr == ')')
+         if (*ptr == CHAR_RIGHT_PARENTHESIS)
            {
            if (code == cd->start_code + 1 + LINK_SIZE &&
                 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
              {
              cd->external_options = newoptions;
-             options = newoptions;
              }
           else
              {
-Line 4807 
 we set the flag only if there is a liter
+Line 5198 
 we set the flag only if there is a liter
                *code++ = OP_OPT;
                *code++ = newoptions & PCRE_IMS;
                }
-             /* Change options at this level, and pass them back for use
-             in subsequent branches. Reset the greedy defaults and the case
-             value for firstbyte and reqbyte. */
-             *optionsptr = options = newoptions;
              greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
              greedy_non_default = greedy_default ^ 1;
-             req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
+             req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
              }
+           /* Change options at this level, and pass them back for use
+           in subsequent branches. When not at the start of the pattern, this
+           information is also necessary so that a resetting item can be
+           compiled at the end of a group (if we are in a group). */
+           *optionsptr = options = newoptions;
            previous = NULL;       /* This item can't be repeated */
            continue;              /* It is complete */
            }
-Line 4933 
 we set the flag only if there is a liter
+Line 5324 
 we set the flag only if there is a liter
      /* Error if hit end of pattern */
-     if (*ptr != ')')
+     if (*ptr != CHAR_RIGHT_PARENTHESIS)
        {
        *errorcodeptr = ERR14;
        goto FAILED;
-Line 5031 
 we set the flag only if there is a liter
+Line 5422 
 we set the flag only if there is a liter
      We can test for values between ESC_b and ESC_Z for the latter; this may
      have to change if any new ones are ever created. */
-     case '\\':
+     case CHAR_BACKSLASH:
      tempptr = ptr;
      c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
      if (*errorcodeptr != 0) goto FAILED;
-Line 5040 
 we set the flag only if there is a liter
+Line 5431 
 we set the flag only if there is a liter
        {
        if (-c == ESC_Q)            /* Handle start of quoted string */
          {
-         if (ptr[1] == '\\' && ptr[2] == 'E') ptr += 2; /* avoid empty string */
+         if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
-           else inescq = TRUE;
+           ptr += 2;               /* avoid empty string */
+             else inescq = TRUE;
          continue;
          }
-Line 5058 
 we set the flag only if there is a liter
+Line 5450 
 we set the flag only if there is a liter
        zerofirstbyte = firstbyte;
        zeroreqbyte = reqbyte;
+       /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
+       is a subroutine call by number (Oniguruma syntax). In fact, the value
+       -ESC_g is returned only for these cases. So we don't need to check for <
+       or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
+       -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
+       that is a synonym for a named back reference). */
+       if (-c == ESC_g)
+         {
+         const uschar *p;
+         save_hwm = cd->hwm;   /* Normally this is set when '(' is read */
+         terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
+           CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
+         /* These two statements stop the compiler for warning about possibly
+         unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
+         fact, because we actually check for a number below, the paths that
+         would actually be in error are never taken. */
+         skipbytes = 0;
+         reset_bracount = FALSE;
+         /* Test for a name */
+         if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
+           {
+           BOOL isnumber = TRUE;
+           for (p = ptr + 1; *p != 0 && *p != terminator; p++)
+             {
+             if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
+             if ((cd->ctypes[*p] & ctype_word) == 0) break;
+             }
+           if (*p != terminator)
+             {
+             *errorcodeptr = ERR57;
+             break;
+             }
+           if (isnumber)
+             {
+             ptr++;
+             goto HANDLE_NUMERICAL_RECURSION;
+             }
+           is_recurse = TRUE;
+           goto NAMED_REF_OR_RECURSE;
+           }
+         /* Test a signed number in angle brackets or quotes. */
+         p = ptr + 2;
+         while ((digitab[*p] & ctype_digit) != 0) p++;
+         if (*p != terminator)
+           {
+           *errorcodeptr = ERR57;
+           break;
+           }
+         ptr++;
+         goto HANDLE_NUMERICAL_RECURSION;
+         }
        /* \k<name> or \k'name' is a back reference by name (Perl syntax).
        We also support \k{name} (.NET syntax) */
-       if (-c == ESC_k && (ptr[1] == '<' || ptr[1] == '\'' || ptr[1] == '{'))
+       if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
+           ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
          {
          is_recurse = FALSE;
-         terminator = (*(++ptr) == '<')? '>' : (*ptr == '\'')? '\'' : '}';
+         terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
+           CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
+           CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
          goto NAMED_REF_OR_RECURSE;
          }
-Line 5167 
 we set the flag only if there is a liter
+Line 5621 
 we set the flag only if there is a liter
      /* Remember if \r or \n were seen */
-     if (mcbuffer[0] == '\r' || mcbuffer[0] == '\n')
+     if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
        cd->external_flags |= PCRE_HASCRORLF;
      /* Set the first and required bytes appropriately. If no previous first
-Line 5265 
 uschar *code = *codeptr;
+Line 5719 
 uschar *code = *codeptr;
  uschar *last_branch = code;
  uschar *start_bracket = code;
  uschar *reverse_count = NULL;
+ open_capitem capitem;
+ int capnumber = 0;
  int firstbyte, reqbyte;
  int branchfirstbyte, branchreqbyte;
  int length;
-Line 5291 
 the code that abstracts option settings
+Line 5747 
 the code that abstracts option settings
  them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
  pre-compile phase to find out whether anything has yet been compiled or not. */
+ /* If this is a capturing subpattern, add to the chain of open capturing items
+ so that we can detect them if (*ACCEPT) is encountered. */
+ if (*code == OP_CBRA)
+   {
+   capnumber = GET2(code, 1 + LINK_SIZE);
+   capitem.number = capnumber;
+   capitem.next = cd->open_caps;
+   cd->open_caps = &capitem;
+   }
  /* Offset is set zero to mark that this bracket is still open */
  PUT(code, 1, 0);
-Line 5385 
 for (;;)
+Line 5852 
 for (;;)
      /* 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. */
+     branch with OP_END. If the branch contains OP_RECURSE, the result is -3
+     because there may be forward references that we can't check here. Set a
+     flag to cause another lookbehind check at the end. Why not do it all at the
+     end? Because common, erroneous checks are picked up here and the offset of
+     the problem can be shown. */
      if (lookbehind)
        {
        int fixed_length;
        *code = OP_END;
-       fixed_length = find_fixedlength(last_branch, options);
+       fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
        DPRINTF(("fixed length = %d\n", fixed_length));
-       if (fixed_length < 0)
+       if (fixed_length == -3)
+         {
+         cd->check_lookbehind = TRUE;
+         }
+       else if (fixed_length < 0)
          {
          *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
          *ptrptr = ptr;
          return FALSE;
          }
-       PUT(reverse_count, 0, fixed_length);
+       else { PUT(reverse_count, 0, fixed_length); }
        }
      }
-Line 5412 
 for (;;)
+Line 5887 
 for (;;)
    compile a resetting op-code following, except at the very end of the pattern.
    Return leaving the pointer at the terminating char. */
-   if (*ptr != '|')
+   if (*ptr != CHAR_VERTICAL_LINE)
      {
      if (lengthptr == NULL)
        {
-Line 5426 
 for (;;)
+Line 5901 
 for (;;)
          }
        while (branch_length > 0);
        }
+     /* If it was a capturing subpattern, remove it from the chain. */
+     if (capnumber > 0) cd->open_caps = cd->open_caps->next;
      /* Fill in the ket */
-Line 5435 
 for (;;)
+Line 5914 
 for (;;)
      /* Resetting option if needed */
-     if ((options & PCRE_IMS) != oldims && *ptr == ')')
+     if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
        {
        *code++ = OP_OPT;
        *code++ = oldims;
-Line 5564 
 do {
+Line 6043 
 do {
       if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
       }
-    /* .* is not anchored unless DOTALL is set and it isn't in brackets that
+    /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
-    are or may be referenced. */
+    it isn't in brackets that are or may be referenced. */
     else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
-              op == OP_TYPEPOSSTAR) &&
+              op == OP_TYPEPOSSTAR))
-             (*options & PCRE_DOTALL) != 0)
       {
-      if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
+      if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
+        return FALSE;
       }
     /* Check for explicit anchoring */
-Line 5617 
 do {
+Line 6096 
 do {
       NULL, 0, FALSE);
     register int op = *scode;
+    /* If we are at the start of a conditional assertion group, *both* the
+    conditional assertion *and* what follows the condition must satisfy the test
+    for start of line. Other kinds of condition fail. Note that there may be an
+    auto-callout at the start of a condition. */
+    if (op == OP_COND)
+      {
+      scode += 1 + LINK_SIZE;
+      if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
+      switch (*scode)
+        {
+        case OP_CREF:
+        case OP_RREF:
+        case OP_DEF:
+        return FALSE;
+        default:     /* Assertion */
+        if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
+        do scode += GET(scode, 1); while (*scode == OP_ALT);
+        scode += 1 + LINK_SIZE;
+        break;
+        }
+      scode = first_significant_code(scode, NULL, 0, FALSE);
+      op = *scode;
+      }
     /* Non-capturing brackets */
     if (op == OP_BRA)
-Line 5635 
 do {
+Line 6140 
 do {
     /* Other brackets */
-    else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
+    else if (op == OP_ASSERT || op == OP_ONCE)
-      { if (!is_startline(scode, bracket_map, backref_map)) return FALSE; }
+      {
+      if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
+      }
     /* .* means "start at start or after \n" if it isn't in brackets that
     may be referenced. */
-Line 5753 
 Returns:        pointer to compiled data
+Line 6260 
 Returns:        pointer to compiled data
                  with errorptr and erroroffset set
  */
- PCRE_EXP_DEFN pcre *
+ PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
  pcre_compile(const char *pattern, int options, const char **errorptr,
    int *erroroffset, const unsigned char *tables)
  {
-Line 5761 
 return pcre_compile2(pattern, options, N
+Line 6268 
 return pcre_compile2(pattern, options, N
  }
- PCRE_EXP_DEFN pcre *
+ PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
  pcre_compile2(const char *pattern, int options, int *errorcodeptr,
    const char **errorptr, int *erroroffset, const unsigned char *tables)
  {
-Line 5770 
 int length = 1;  /* For final END opcode
+Line 6277 
 int length = 1;  /* For final END opcode
  int firstbyte, reqbyte, newline;
  int errorcode = 0;
  int skipatstart = 0;
- #ifdef SUPPORT_UTF8
+ BOOL utf8 = (options & PCRE_UTF8) != 0;
- BOOL utf8;
- #endif
  size_t size;
  uschar *code;
  const uschar *codestart;
-Line 5788 
 to fill in forward references to subpatt
+Line 6293 
 to fill in forward references to subpatt
  uschar cworkspace[COMPILE_WORK_SIZE];
  /* Set this early so that early errors get offset 0. */
  ptr = (const uschar *)pattern;
-Line 5816 
 if (erroroffset == NULL)
+Line 6320 
 if (erroroffset == NULL)
  *erroroffset = 0;
- /* Can't support UTF8 unless PCRE has been compiled to include the code. */
- #ifdef SUPPORT_UTF8
- utf8 = (options & PCRE_UTF8) != 0;
- if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
-      (*erroroffset = _pcre_valid_utf8((uschar *)pattern, -1)) >= 0)
-   {
-   errorcode = ERR44;
-   goto PCRE_EARLY_ERROR_RETURN2;
-   }
- #else
- if ((options & PCRE_UTF8) != 0)
-   {
-   errorcode = ERR32;
-   goto PCRE_EARLY_ERROR_RETURN;
-   }
- #endif
- if ((options & ~PUBLIC_OPTIONS) != 0)
-   {
-   errorcode = ERR17;
-   goto PCRE_EARLY_ERROR_RETURN;
-   }
  /* Set up pointers to the individual character tables */
  if (tables == NULL) tables = _pcre_default_tables;
-Line 5848 
 cd->fcc = tables + fcc_offset;
+Line 6328 
 cd->fcc = tables + fcc_offset;
  cd->cbits = tables + cbits_offset;
  cd->ctypes = tables + ctypes_offset;
+ /* Check that all undefined public option bits are zero */
+ if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
+   {
+   errorcode = ERR17;
+   goto PCRE_EARLY_ERROR_RETURN;
+   }
  /* Check for global one-time settings at the start of the pattern, and remember
  the offset for later. */
- while (ptr[skipatstart] == '(' && ptr[skipatstart+1] == '*')
+ while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
+        ptr[skipatstart+1] == CHAR_ASTERISK)
    {
    int newnl = 0;
    int newbsr = 0;
-   if (strncmp((char *)(ptr+skipatstart+2), "CR)", 3) == 0)
+   if (strncmp((char *)(ptr+skipatstart+2), STRING_UTF8_RIGHTPAR, 5) == 0)
+     { skipatstart += 7; options |= PCRE_UTF8; continue; }
+   if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
      { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
-   else if (strncmp((char *)(ptr+skipatstart+2), "LF)", 3)  == 0)
+   else if (strncmp((char *)(ptr+skipatstart+2), STRING_LF_RIGHTPAR, 3)  == 0)
      { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
-   else if (strncmp((char *)(ptr+skipatstart+2), "CRLF)", 5)  == 0)
+   else if (strncmp((char *)(ptr+skipatstart+2), STRING_CRLF_RIGHTPAR, 5)  == 0)
      { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
-   else if (strncmp((char *)(ptr+skipatstart+2), "ANY)", 4) == 0)
+   else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANY_RIGHTPAR, 4) == 0)
      { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
-   else if (strncmp((char *)(ptr+skipatstart+2), "ANYCRLF)", 8)  == 0)
+   else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANYCRLF_RIGHTPAR, 8) == 0)
      { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
-   else if (strncmp((char *)(ptr+skipatstart+2), "BSR_ANYCRLF)", 12) == 0)
+   else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
      { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
-   else if (strncmp((char *)(ptr+skipatstart+2), "BSR_UNICODE)", 12) == 0)
+   else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
      { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
    if (newnl != 0)
-Line 5879 
 while (ptr[skipatstart] == '(' && ptr[sk
+Line 6371 
 while (ptr[skipatstart] == '(' && ptr[sk
    else break;
    }
+ /* Can't support UTF8 unless PCRE has been compiled to include the code. */
+ #ifdef SUPPORT_UTF8
+ if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
+      (*erroroffset = _pcre_valid_utf8((uschar *)pattern, -1)) >= 0)
+   {
+   errorcode = ERR44;
+   goto PCRE_EARLY_ERROR_RETURN2;
+   }
+ #else
+ if (utf8)
+   {
+   errorcode = ERR32;
+   goto PCRE_EARLY_ERROR_RETURN;
+   }
+ #endif
  /* Check validity of \R options. */
  switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
-Line 5897 
 current code allows for fixed one- or tw
+Line 6406 
 current code allows for fixed one- or tw
  switch (options & PCRE_NEWLINE_BITS)
    {
    case 0: newline = NEWLINE; break;   /* Build-time default */
-   case PCRE_NEWLINE_CR: newline = '\r'; break;
+   case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
-   case PCRE_NEWLINE_LF: newline = '\n'; break;
+   case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
    case PCRE_NEWLINE_CR+
-        PCRE_NEWLINE_LF: newline = ('\r' << 8) | '\n'; break;
+        PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
    case PCRE_NEWLINE_ANY: newline = -1; break;
    case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
    default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
-Line 5961 
 cd->end_pattern = (const uschar *)(patte
+Line 6470 
 cd->end_pattern = (const uschar *)(patte
  cd->req_varyopt = 0;
  cd->external_options = options;
  cd->external_flags = 0;
+ cd->open_caps = NULL;
  /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
  don't need to look at the result of the function here. The initial options have
-Line 6035 
 cd->start_code = codestart;
+Line 6545 
 cd->start_code = codestart;
  cd->hwm = cworkspace;
  cd->req_varyopt = 0;
  cd->had_accept = FALSE;
+ cd->check_lookbehind = FALSE;
+ cd->open_caps = NULL;
  /* Set up a starting, non-extracting bracket, then compile the expression. On
  error, errorcode will be set non-zero, so we don't need to look at the result
-Line 6073 
 while (errorcode == 0 && cd->hwm > cwork
+Line 6585 
 while (errorcode == 0 && cd->hwm > cwork
    cd->hwm -= LINK_SIZE;
    offset = GET(cd->hwm, 0);
    recno = GET(codestart, offset);
-   groupptr = find_bracket(codestart, (re->options & PCRE_UTF8) != 0, recno);
+   groupptr = _pcre_find_bracket(codestart, utf8, recno);
    if (groupptr == NULL) errorcode = ERR53;
      else PUT(((uschar *)codestart), offset, groupptr - codestart);
    }
-Line 6083 
 subpattern. */
+Line 6595 
 subpattern. */
  if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
+ /* If there were any lookbehind assertions that contained OP_RECURSE
+ (recursions or subroutine calls), a flag is set for them to be checked here,
+ because they may contain forward references. Actual recursions can't be fixed
+ length, but subroutine calls can. It is done like this so that those without
+ OP_RECURSE that are not fixed length get a diagnosic with a useful offset. The
+ exceptional ones forgo this. We scan the pattern to check that they are fixed
+ length, and set their lengths. */
+ if (cd->check_lookbehind)
+   {
+   uschar *cc = (uschar *)codestart;
+   /* Loop, searching for OP_REVERSE items, and process those that do not have
+   their length set. (Actually, it will also re-process any that have a length
+   of zero, but that is a pathological case, and it does no harm.) When we find
+   one, we temporarily terminate the branch it is in while we scan it. */
+   for (cc = (uschar *)_pcre_find_bracket(codestart, utf8, -1);
+        cc != NULL;
+        cc = (uschar *)_pcre_find_bracket(cc, utf8, -1))
+     {
+     if (GET(cc, 1) == 0)
+       {
+       int fixed_length;
+       uschar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);
+       int end_op = *be;
+       *be = OP_END;
+       fixed_length = find_fixedlength(cc, re->options, TRUE, cd);
+       *be = end_op;
+       DPRINTF(("fixed length = %d\n", fixed_length));
+       if (fixed_length < 0)
+         {
+         errorcode = (fixed_length == -2)? ERR36 : ERR25;
+         break;
+         }
+       PUT(cc, 1, fixed_length);
+       }
+     cc += 1 + LINK_SIZE;
+     }
+   }
  /* Failed to compile, or error while post-processing */
  if (errorcode != 0)

 Legend:



Removed from v.285
 


changed lines


 
Added in v.455
 Legend:



Removed from v.285
 


changed lines


 
Added in v.455
-Removed from v.285
+Added in v.455

webmaster@exim.org	ViewVC Help
Powered by ViewVC 1.1.12