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

Diff of /code/trunk/pcre_compile.c

Parent Directory | Revision Log | View Patch Patch

-revision 300 by ph10,
Mon Jan 14 19:43:19 2008 UTC
+revision 550 by ph10,
Sun Oct 10 16:24:11 2010 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) 1997-2010 University of Cambridge
  -----------------------------------------------------------------------------
  Redistribution and use in source and binary forms, with or without
 Line 53 
 supporting internal functions that are n
  #include "pcre_internal.h"
- /* When DEBUG is defined, we need the pcre_printint() function, which is also
+ /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
- used by pcretest. DEBUG is not defined when building a production library. */
+ also used by pcretest. PCRE_DEBUG is not defined when building a production
+ library. */
- #ifdef DEBUG
+ #ifdef PCRE_DEBUG
  #include "pcre_printint.src"
  #endif
-Line 91 
 is 4 there is plenty of room. */
+Line 92 
 is 4 there is plenty of room. */
  #define COMPILE_WORK_SIZE (4096)
+ /* The overrun tests check for a slightly smaller size so that they detect the
+ overrun before it actually does run off the end of the data block. */
+ #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
  /* Table for handling escaped characters in the range '0'-'z'. Positive returns
  are simple data values; negative values are for special things like \d and so
  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,
+      -ESC_N,                  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 130 
 static const short int escapes[] = {
+Line 171 
 static const short int escapes[] = {
  /*  B8 */     0,     0,      0,       0,      0,   ']',    '=',    '-',
  /*  C0 */   '{',-ESC_A, -ESC_B,  -ESC_C, -ESC_D,-ESC_E,      0, -ESC_G,
  /*  C8 */-ESC_H,     0,      0,       0,      0,     0,      0,      0,
- /*  D0 */   '}',     0, -ESC_K,       0,      0,     0,      0, -ESC_P,
+ /*  D0 */   '}',     0, -ESC_K,       0,      0,-ESC_N,      0, -ESC_P,
  /*  D8 */-ESC_Q,-ESC_R,      0,       0,      0,     0,      0,      0,
  /*  E0 */  '\\',     0, -ESC_S,       0,      0,-ESC_V, -ESC_W, -ESC_X,
  /*  E8 */     0,-ESC_Z,      0,       0,      0,     0,      0,      0,
-Line 142 
 static const short int escapes[] = {
+Line 183 
 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;
+   int   len;                 /* Length of verb name */
-   int   op;
+   int   op;                  /* Op when no arg, or -1 if arg mandatory */
+   int   op_arg;              /* Op when arg present, or -1 if not allowed */
  } verbitem;
  static const char verbnames[] =
-   "ACCEPT\0"
+   "\0"                       /* Empty name is a shorthand for MARK */
-   "COMMIT\0"
+   STRING_MARK0
-   "F\0"
+   STRING_ACCEPT0
-   "FAIL\0"
+   STRING_COMMIT0
-   "PRUNE\0"
+   STRING_F0
-   "SKIP\0"
+   STRING_FAIL0
-   "THEN";
+   STRING_PRUNE0
+   STRING_SKIP0
- static verbitem verbs[] = {
+   STRING_THEN;
-   { 6, OP_ACCEPT },
-   { 6, OP_COMMIT },
+ static const verbitem verbs[] = {
-   { 1, OP_FAIL },
+   { 0, -1,        OP_MARK },
-   { 4, OP_FAIL },
+   { 4, -1,        OP_MARK },
-   { 5, OP_PRUNE },
+   { 6, OP_ACCEPT, -1 },
-   { 4, OP_SKIP  },
+   { 6, OP_COMMIT, -1 },
-   { 4, OP_THEN  }
+   { 1, OP_FAIL,   -1 },
+   { 4, OP_FAIL,   -1 },
+   { 5, OP_PRUNE,  OP_PRUNE_ARG },
+   { 4, OP_SKIP,   OP_SKIP_ARG  },
+   { 4, OP_THEN,   OP_THEN_ARG  }
  };
- 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 226 
 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 212 
 static const int posix_class_maps[] = {
+Line 261 
 static const int posix_class_maps[] = {
    cbit_xdigit,-1,          0              /* xdigit */
  };
+ /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
+ substitutes must be in the order of the names, defined above, and there are
+ both positive and negative cases. NULL means no substitute. */
+ #ifdef SUPPORT_UCP
+ static const uschar *substitutes[] = {
+   (uschar *)"\\P{Nd}",    /* \D */
+   (uschar *)"\\p{Nd}",    /* \d */
+   (uschar *)"\\P{Xsp}",   /* \S */       /* NOTE: Xsp is Perl space */
+   (uschar *)"\\p{Xsp}",   /* \s */
+   (uschar *)"\\P{Xwd}",   /* \W */
+   (uschar *)"\\p{Xwd}"    /* \w */
+ };
+ static const uschar *posix_substitutes[] = {
+   (uschar *)"\\p{L}",     /* alpha */
+   (uschar *)"\\p{Ll}",    /* lower */
+   (uschar *)"\\p{Lu}",    /* upper */
+   (uschar *)"\\p{Xan}",   /* alnum */
+   NULL,                   /* ascii */
+   (uschar *)"\\h",        /* blank */
+   NULL,                   /* cntrl */
+   (uschar *)"\\p{Nd}",    /* digit */
+   NULL,                   /* graph */
+   NULL,                   /* print */
+   NULL,                   /* punct */
+   (uschar *)"\\p{Xps}",   /* space */    /* NOTE: Xps is POSIX space */
+   (uschar *)"\\p{Xwd}",   /* word */
+   NULL,                   /* xdigit */
+   /* Negated cases */
+   (uschar *)"\\P{L}",     /* ^alpha */
+   (uschar *)"\\P{Ll}",    /* ^lower */
+   (uschar *)"\\P{Lu}",    /* ^upper */
+   (uschar *)"\\P{Xan}",   /* ^alnum */
+   NULL,                   /* ^ascii */
+   (uschar *)"\\H",        /* ^blank */
+   NULL,                   /* ^cntrl */
+   (uschar *)"\\P{Nd}",    /* ^digit */
+   NULL,                   /* ^graph */
+   NULL,                   /* ^print */
+   NULL,                   /* ^punct */
+   (uschar *)"\\P{Xps}",   /* ^space */   /* NOTE: Xps is POSIX space */
+   (uschar *)"\\P{Xwd}",   /* ^word */
+   NULL                    /* ^xdigit */
+ };
+ #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
+ #endif
  #define STRING(a)  # a
  #define XSTRING(s) STRING(s)
-Line 224 
 the number of relocations needed when a
+Line 320 
 the number of relocations needed when a
  it is now one long string. We cannot use a table of offsets, because the
  lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
  simply count through to the one we want - this isn't a performance issue
- because these strings are used only when there is a compilation error. */
+ because these strings are used only when there is a compilation error.
+ Each substring ends with \0 to insert a null character. This includes the final
+ substring, so that the whole string ends with \0\0, which can be detected when
+ counting through. */
  static const char error_texts[] =
    "no error\0"
-Line 271 
 static const char error_texts[] =
+Line 371 
 static const char error_texts[] =
    /* 35 */
    "invalid condition (?(0)\0"
    "\\C not allowed in lookbehind assertion\0"
-   "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
+   "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
    "number after (?C is > 255\0"
    "closing ) for (?C expected\0"
    /* 40 */
-Line 295 
 static const char error_texts[] =
+Line 395 
 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"
+   "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\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\0"
+   /* 65 */
+   "different names for subpatterns of the same number are not allowed\0"
+   "(*MARK) must have an argument\0"
+   "this version of PCRE is not compiled with PCRE_UCP support\0"
+   ;
  /* Table to identify digits and hex digits. This is used when compiling
  patterns. Note that the tables in chartables are dependent on the locale, and
-Line 321 
 For convenience, we use the same bit def
+Line 426 
 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 466 
 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 566 
 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) {};
+   if (*s == 0) return "Error text not found (please report)";
+   }
  return s;
  }
-Line 502 
 if (c == 0) *errorcodeptr = ERR1;
+Line 618 
 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 639 
 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_u:
-     case 'u':
+     case CHAR_U:
-     case '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
+     (1) A number, either plain or braced. If positive, it is an absolute
-     reference to a named group. This is part of Perl's movement towards a
+     backreference. If negative, it is a relative backreference. This is a Perl
-     unified syntax for back references. As this is synonymous with \k{name}, we
+.10 feature.
-     fudge it up by pretending it really was \k. */
+     (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 */
-     case 'g':
+     if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
-     if (ptr[1] == '{')
        {
        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 686 
 else
        }
      else braced = FALSE;
-     if (ptr[1] == '-')
+     if (ptr[1] == CHAR_MINUS)
        {
        negated = TRUE;
        ptr++;
-Line 563 
 else
+Line 695 
 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 740 
 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 766 
 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 779 
 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 790 
 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 800 
 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 828 
 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 844 
 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 852 
 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 740 
 else
+Line 878 
 else
      }
    }
+ /* Perl supports \N{name} for character names, as well as plain \N for "not
+ newline". PCRE does not support \N{name}. */
+ if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)
+   *errorcodeptr = ERR37;
+ /* If PCRE_UCP is set, we change the values for \d etc. */
+ if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
+   c -= (ESC_DU - ESC_D);
+ /* Set the pointer to the final character before returning. */
  *ptrptr = ptr;
  return c;
  }
-Line 780 
 if (c == 0) goto ERROR_RETURN;
+Line 931 
 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 942 
 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 1010 
 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 1051 
 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 1061 
 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 1091 
 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;
+   /* Handle specials such as (*SKIP) or (*UTF8) etc. */
+   if (ptr[1] == CHAR_ASTERISK) ptr += 2;
+   /* Handle a normal, unnamed capturing parenthesis. */
+   else if (ptr[1] != CHAR_QUESTION_MARK)
+     {
+     *count += 1;
+     if (name == NULL && *count == lorn) return *count;
+     ptr++;
+     }
+   /* All cases now have (? at the start. Remember when we are in a group
+   where the parenthesis numbers are duplicated. */
+   else if (ptr[2] == CHAR_VERTICAL_LINE)
+     {
+     ptr += 3;
+     dup_parens = TRUE;
+     }
+   /* Handle comments; all characters are allowed until a ket is reached. */
+   else if (ptr[2] == CHAR_NUMBER_SIGN)
+     {
+     for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
+     goto FAIL_EXIT;
+     }
+   /* 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++;
+       }
+     }
+   /* Start with (? but 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 1276 
 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;
+     goto FAIL_EXIT;
+     }
-   /* 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.
+ 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
+ */
-   count++;
+ 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 1100 
 for (;;)
+Line 1410 
 for (;;)
      case OP_CALLOUT:
      case OP_CREF:
+     case OP_NCREF:
      case OP_RREF:
+     case OP_NRREF:
      case OP_DEF:
      code += _pcre_OP_lengths[*code];
      break;
-Line 1116 
 for (;;)
+Line 1428 
 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 1468 
 branch, check the length against that of
  for (;;)
    {
    int d;
+   uschar *ce, *cs;
    register int op = *cc;
    switch (op)
      {
-Line 1152 
 for (;;)
+Line 1476 
 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 1175 
 for (;;)
+Line 1499 
 for (;;)
      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 */
      case OP_ASSERT:
-Line 1188 
 for (;;)
+Line 1527 
 for (;;)
      case OP_REVERSE:
      case OP_CREF:
+     case OP_NCREF:
      case OP_RREF:
+     case OP_NRREF:
      case OP_DEF:
      case OP_OPT:
      case OP_CALLOUT:
      case OP_SOD:
      case OP_SOM:
+     case OP_SET_SOM:
      case OP_EOD:
      case OP_EODN:
      case OP_CIRC:
-Line 1211 
 for (;;)
+Line 1553 
 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 1565 
 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 1590 
 for (;;)
      case OP_NOT_WORDCHAR:
      case OP_WORDCHAR:
      case OP_ANY:
+     case OP_ALLANY:
      branchlength++;
      cc++;
      break;
-Line 1306 
 for (;;)
+Line 1645 
 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 1334 
 for (;;)
+Line 1676 
 for (;;)
    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 */
    else if (c == OP_CBRA)
-Line 1345 
 for (;;)
+Line 1695 
 for (;;)
    /* Otherwise, we can get the item's length from the table, except that for
    repeated character types, we have to test for \p and \P, which have an extra
-   two bytes of parameters. */
+   two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
+   must add in its length. */
    else
      {
-Line 1369 
 for (;;)
+Line 1720 
 for (;;)
        case OP_TYPEPOSUPTO:
        if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
        break;
+       case OP_MARK:
+       case OP_PRUNE_ARG:
+       case OP_SKIP_ARG:
+       code += code[1];
+       break;
+       case OP_THEN_ARG:
+       code += code[1+LINK_SIZE];
+       break;
        }
      /* Add in the fixed length from the table */
-Line 1400 
 for (;;)
+Line 1761 
 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 1438 
 for (;;)
+Line 1801 
 for (;;)
    /* Otherwise, we can get the item's length from the table, except that for
    repeated character types, we have to test for \p and \P, which have an extra
-   two bytes of parameters. */
+   two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
+   must add in its length. */
    else
      {
-Line 1462 
 for (;;)
+Line 1826 
 for (;;)
        case OP_TYPEEXACT:
        if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
        break;
+       case OP_MARK:
+       case OP_PRUNE_ARG:
+       case OP_SKIP_ARG:
+       code += code[1];
+       break;
+       case OP_THEN_ARG:
+       code += code[1+LINK_SIZE];
+       break;
        }
      /* Add in the fixed length from the table */
-Line 1493 
 for (;;)
+Line 1867 
 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 1516 
 Arguments:
+Line 1892 
 Arguments:
    code        points to start of search
    endcode     points to where to stop
    utf8        TRUE if in UTF8 mode
+   cd          contains pointers to tables etc.
  Returns:      TRUE if what is matched could be empty
  */
  static BOOL
- could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
+ could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
+   compile_data *cd)
  {
  register int c;
  for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
-Line 1544 
 for (code = first_significant_code(code
+Line 1922 
 for (code = first_significant_code(code
    /* 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 1552 
 for (code = first_significant_code(code
+Line 1930 
 for (code = first_significant_code(code
      continue;
      }
+   /* For a recursion/subroutine call, if its end has been reached, which
+   implies a subroutine call, we can scan it. */
+   if (c == OP_RECURSE)
+     {
+     BOOL empty_branch = FALSE;
+     const uschar *scode = cd->start_code + GET(code, 1);
+     if (GET(scode, 1) == 0) return TRUE;    /* Unclosed */
+     do
+       {
+       if (could_be_empty_branch(scode, endcode, utf8, cd))
+         {
+         empty_branch = TRUE;
+         break;
+         }
+       scode += GET(scode, 1);
+       }
+     while (*scode == OP_ALT);
+     if (!empty_branch) return FALSE;  /* All branches are non-empty */
+     continue;
+     }
    /* For other groups, scan the branches. */
    if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
-Line 1559 
 for (code = first_significant_code(code
+Line 1959 
 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, cd))
+           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 2038 
 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 2095 
 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
+     /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
+     string. */
+     case OP_MARK:
+     case OP_PRUNE_ARG:
+     case OP_SKIP_ARG:
+     code += code[1];
+     break;
+     case OP_THEN_ARG:
+     code += code[1+LINK_SIZE];
+     break;
+     /* None of the remaining opcodes are required to match a character. */
+     default:
+     break;
      }
    }
-Line 1714 
 Arguments:
+Line 2144 
 Arguments:
    endcode     points to where to stop (current RECURSE item)
    bcptr       points to the chain of current (unclosed) branch starts
    utf8        TRUE if in UTF-8 mode
+   cd          pointers to tables etc
  Returns:      TRUE if what is matched could be empty
  */
  static BOOL
  could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
-   BOOL utf8)
+   BOOL utf8, compile_data *cd)
  {
- while (bcptr != NULL && bcptr->current >= code)
+ while (bcptr != NULL && bcptr->current_branch >= code)
    {
-   if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;
+   if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
+     return FALSE;
    bcptr = bcptr->outer;
    }
  return TRUE;
-Line 1770 
 int terminator;          /* Don't combin
+Line 2202 
 int terminator;          /* Don't combin
  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
  for (++ptr; *ptr != 0; ptr++)
    {
-   if (*ptr == '\\' && ptr[1] == ']') ptr++; else
+   if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
      {
-     if (*ptr == ']') return FALSE;
+     if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
-     if (*ptr == terminator && ptr[1] == ']')
+     if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
        {
        *endptr = ptr;
        return TRUE;
-Line 1824 
 return -1;
+Line 2256 
 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 1905 
 auto_callout(uschar *code, const uschar
+Line 2338 
 auto_callout(uschar *code, const uschar
  {
  *code++ = OP_CALLOUT;
  *code++ = 255;
- PUT(code, 0, ptr - cd->start_pattern);  /* Pattern offset */
+ PUT(code, 0, (int)(ptr - cd->start_pattern));  /* Pattern offset */
- PUT(code, LINK_SIZE, 0);                /* Default length */
+ PUT(code, LINK_SIZE, 0);                       /* Default length */
  return code + 2*LINK_SIZE;
  }
-Line 1931 
 Returns:             nothing
+Line 2364 
 Returns:             nothing
  static void
  complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
  {
- int length = ptr - cd->start_pattern - GET(previous_callout, 2);
+ int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
  PUT(previous_callout, 2 + LINK_SIZE, length);
  }
-Line 1963 
 get_othercase_range(unsigned int *cptr,
+Line 2396 
 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 1972 
 next = othercase + 1;
+Line 2405 
 next = othercase + 1;
  for (++c; c <= d; c++)
    {
-   if (_pcre_ucp_othercase(c) != next) break;
+   if (UCD_OTHERCASE(c) != next) break;
    next++;
    }
-Line 1981 
 for (++c; c <= d; c++)
+Line 2414 
 for (++c; c <= d; c++)
  return TRUE;
  }
+ /*************************************************
+ *        Check a character and a property        *
+ *************************************************/
+ /* This function is called by check_auto_possessive() when a property item
+ is adjacent to a fixed character.
+ Arguments:
+   c            the character
+   ptype        the property type
+   pdata        the data for the type
+   negated      TRUE if it's a negated property (\P or \p{^)
+ Returns:       TRUE if auto-possessifying is OK
+ */
+ static BOOL
+ check_char_prop(int c, int ptype, int pdata, BOOL negated)
+ {
+ const ucd_record *prop = GET_UCD(c);
+ switch(ptype)
+   {
+   case PT_LAMP:
+   return (prop->chartype == ucp_Lu ||
+           prop->chartype == ucp_Ll ||
+           prop->chartype == ucp_Lt) == negated;
+   case PT_GC:
+   return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
+   case PT_PC:
+   return (pdata == prop->chartype) == negated;
+   case PT_SC:
+   return (pdata == prop->script) == negated;
+   /* These are specials */
+   case PT_ALNUM:
+   return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
+           _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
+   case PT_SPACE:    /* Perl space */
+   return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
+           c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
+           == negated;
+   case PT_PXSPACE:  /* POSIX space */
+   return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
+           c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
+           c == CHAR_FF || c == CHAR_CR)
+           == negated;
+   case PT_WORD:
+   return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
+           _pcre_ucp_gentype[prop->chartype] == ucp_N ||
+           c == CHAR_UNDERSCORE) == negated;
+   }
+ return FALSE;
+ }
  #endif  /* SUPPORT_UCP */
-Line 1994 
 whether the next thing could possibly ma
+Line 2490 
 whether the next thing could possibly ma
  sense to automatically possessify the repeated item.
  Arguments:
-   op_code       the repeated op code
+   previous      pointer to the repeated opcode
-   this          data for this item, depends on the opcode
    utf8          TRUE in UTF-8 mode
-   utf8_char     used for utf8 character bytes, NULL if not relevant
    ptr           next character in pattern
    options       options bits
    cd            contains pointers to tables etc.
-Line 2006 
 Returns:        TRUE if possessifying is
+Line 2500 
 Returns:        TRUE if possessifying is
  */
  static BOOL
- check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
+ check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
-   const uschar *ptr, int options, compile_data *cd)
+   int options, compile_data *cd)
  {
- int next;
+ int c, next;
+ int op_code = *previous++;
  /* Skip whitespace and comments in extended mode */
-Line 2018 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2513 
 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 2030 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2525 
 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 2055 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2550 
 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 2066 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2561 
 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
- "item" is greater than 127 in utf8 mode, the character's bytes are in
- utf8_char. */
- /* Handle cases when the next item is a character. */
+ /* Now compare the next item with the previous opcode. First, handle cases when
+ the next item is a character. */
  if (next >= 0) switch(op_code)
    {
    case OP_CHAR:
  #ifdef SUPPORT_UTF8
-   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+   GETCHARTEST(c, previous);
+ #else
+   c = *previous;
  #endif
-   return item != next;
+   return c != next;
    /* For CHARNC (caseless character) we must check the other case. If we have
    Unicode property support, we can use it to test the other case of
-Line 2091 
 if (next >= 0) switch(op_code)
+Line 2584 
 if (next >= 0) switch(op_code)
    case OP_CHARNC:
  #ifdef SUPPORT_UTF8
-   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+   GETCHARTEST(c, previous);
+ #else
+   c = *previous;
  #endif
-   if (item == next) return FALSE;
+   if (c == next) return FALSE;
  #ifdef SUPPORT_UTF8
    if (utf8)
      {
      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
-     return (unsigned int)item != othercase;
+     return (unsigned int)c != othercase;
      }
    else
  #endif  /* SUPPORT_UTF8 */
-   return (item != cd->fcc[next]);  /* Non-UTF-8 mode */
+   return (c != cd->fcc[next]);  /* Non-UTF-8 mode */
-   /* For OP_NOT, "item" must be a single-byte character. */
+   /* For OP_NOT, its data is always a single-byte character. */
    case OP_NOT:
-   if (next < 0) return FALSE;  /* Not a character */
+   if ((c = *previous) == next) return TRUE;
-   if (item == next) return TRUE;
    if ((options & PCRE_CASELESS) == 0) return FALSE;
  #ifdef SUPPORT_UTF8
    if (utf8)
-Line 2122 
 if (next >= 0) switch(op_code)
+Line 2616 
 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
-     return (unsigned int)item == othercase;
+     return (unsigned int)c == othercase;
      }
    else
  #endif  /* SUPPORT_UTF8 */
-   return (item == cd->fcc[next]);  /* Non-UTF-8 mode */
+   return (c == cd->fcc[next]);  /* Non-UTF-8 mode */
+   /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
+   When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
    case OP_DIGIT:
    return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
-Line 2173 
 if (next >= 0) switch(op_code)
+Line 2670 
 if (next >= 0) switch(op_code)
      case 0x202f:
      case 0x205f:
      case 0x3000:
-     return op_code != OP_HSPACE;
+     return op_code == OP_NOT_HSPACE;
      default:
-     return op_code == OP_HSPACE;
+     return op_code != OP_NOT_HSPACE;
      }
+   case OP_ANYNL:
    case OP_VSPACE:
    case OP_NOT_VSPACE:
    switch(next)
-Line 2189 
 if (next >= 0) switch(op_code)
+Line 2687 
 if (next >= 0) switch(op_code)
      case 0x85:
      case 0x2028:
      case 0x2029:
-     return op_code != OP_VSPACE;
+     return op_code == OP_NOT_VSPACE;
      default:
-     return op_code == OP_VSPACE;
+     return op_code != OP_NOT_VSPACE;
      }
+ #ifdef SUPPORT_UCP
+   case OP_PROP:
+   return check_char_prop(next, previous[0], previous[1], FALSE);
+   case OP_NOTPROP:
+   return check_char_prop(next, previous[0], previous[1], TRUE);
+ #endif
    default:
    return FALSE;
    }
- /* Handle the case when the next item is \d, \s, etc. */
+ /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
+ is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
+ generated only when PCRE_UCP is *not* set, that is, when only ASCII
+ characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
+ replaced by OP_PROP codes when PCRE_UCP is set. */
  switch(op_code)
    {
    case OP_CHAR:
    case OP_CHARNC:
  #ifdef SUPPORT_UTF8
-   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+   GETCHARTEST(c, previous);
+ #else
+   c = *previous;
  #endif
    switch(-next)
      {
      case ESC_d:
-     return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
+     return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
      case ESC_D:
-     return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
+     return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
      case ESC_s:
-     return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
+     return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
      case ESC_S:
-     return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
+     return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
      case ESC_w:
-     return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
+     return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
      case ESC_W:
-     return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
+     return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
      case ESC_h:
      case ESC_H:
-     switch(item)
+     switch(c)
        {
        case 0x09:
        case 0x20:
-Line 2258 
 switch(op_code)
+Line 2770 
 switch(op_code)
      case ESC_v:
      case ESC_V:
-     switch(item)
+     switch(c)
        {
        case 0x0a:
        case 0x0b:
-Line 2272 
 switch(op_code)
+Line 2784 
 switch(op_code)
        return -next == ESC_v;
        }
+     /* When PCRE_UCP is set, these values get generated for \d etc. Find
+     their substitutions and process them. The result will always be either
+     -ESC_p or -ESC_P. Then fall through to process those values. */
+ #ifdef SUPPORT_UCP
+     case ESC_du:
+     case ESC_DU:
+     case ESC_wu:
+     case ESC_WU:
+     case ESC_su:
+     case ESC_SU:
+       {
+       int temperrorcode = 0;
+       ptr = substitutes[-next - ESC_DU];
+       next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
+       if (temperrorcode != 0) return FALSE;
+       ptr++;    /* For compatibility */
+       }
+     /* Fall through */
+     case ESC_p:
+     case ESC_P:
+       {
+       int ptype, pdata, errorcodeptr;
+       BOOL negated;
+       ptr--;      /* Make ptr point at the p or P */
+       ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
+       if (ptype < 0) return FALSE;
+       ptr++;      /* Point past the final curly ket */
+       /* If the property item is optional, we have to give up. (When generated
+       from \d etc by PCRE_UCP, this test will have been applied much earlier,
+       to the original \d etc. At this point, ptr will point to a zero byte. */
+       if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
+         strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
+           return FALSE;
+       /* Do the property check. */
+       return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
+       }
+ #endif
      default:
      return FALSE;
      }
+   /* In principle, support for Unicode properties should be integrated here as
+   well. It means re-organizing the above code so as to get hold of the property
+   values before switching on the op-code. However, I wonder how many patterns
+   combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
+   these op-codes are never generated.) */
    case OP_DIGIT:
    return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
-          next == -ESC_h || next == -ESC_v;
+          next == -ESC_h || next == -ESC_v || next == -ESC_R;
    case OP_NOT_DIGIT:
    return next == -ESC_d;
    case OP_WHITESPACE:
-   return next == -ESC_S || next == -ESC_d || next == -ESC_w;
+   return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
    case OP_NOT_WHITESPACE:
    return next == -ESC_s || next == -ESC_h || next == -ESC_v;
    case OP_HSPACE:
-   return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
+   return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
+          next == -ESC_w || next == -ESC_v || next == -ESC_R;
    case OP_NOT_HSPACE:
    return next == -ESC_h;
    /* Can't have \S in here because VT matches \S (Perl anomaly) */
+   case OP_ANYNL:
    case OP_VSPACE:
    return next == -ESC_V || next == -ESC_d || next == -ESC_w;
    case OP_NOT_VSPACE:
-   return next == -ESC_v;
+   return next == -ESC_v || next == -ESC_R;
    case OP_WORDCHAR:
-   return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
+   return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
+          next == -ESC_v || next == -ESC_R;
    case OP_NOT_WORDCHAR:
    return next == -ESC_w || next == -ESC_d;
-Line 2367 
 BOOL inescq = FALSE;
+Line 2933 
 BOOL inescq = FALSE;
  BOOL groupsetfirstbyte = FALSE;
  const uschar *ptr = *ptrptr;
  const uschar *tempptr;
+ const uschar *nestptr = NULL;
  uschar *previous = NULL;
  uschar *previous_callout = NULL;
  uschar *save_hwm = NULL;
-Line 2383 
 BOOL utf8 = FALSE;
+Line 2950 
 BOOL utf8 = FALSE;
  uschar *utf8_char = NULL;
  #endif
- #ifdef DEBUG
+ #ifdef PCRE_DEBUG
  if (lengthptr != NULL) DPRINTF((">> start branch\n"));
  #endif
-Line 2437 
 for (;; ptr++)
+Line 3004 
 for (;; ptr++)
    c = *ptr;
+   /* If we are at the end of a nested substitution, revert to the outer level
+   string. Nesting only happens one level deep. */
+   if (c == 0 && nestptr != NULL)
+     {
+     ptr = nestptr;
+     nestptr = NULL;
+     c = *ptr;
+     }
    /* If we are in the pre-compile phase, accumulate the length used for the
    previous cycle of this loop. */
    if (lengthptr != NULL)
      {
- #ifdef DEBUG
+ #ifdef PCRE_DEBUG
      if (code > cd->hwm) cd->hwm = code;                 /* High water info */
  #endif
-     if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
+     if (code > cd->start_workspace + WORK_SIZE_CHECK)   /* Check for overrun */
        {
        *errorcodeptr = ERR52;
        goto FAILED;
-Line 2467 
 for (;; ptr++)
+Line 3044 
 for (;; ptr++)
        goto FAILED;
        }
-     *lengthptr += code - last_code;
+     *lengthptr += (int)(code - last_code);
      DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
      /* If "previous" is set and it is not at the start of the work space, move
-Line 2494 
 for (;; ptr++)
+Line 3071 
 for (;; ptr++)
    /* In the real compile phase, just check the workspace used by the forward
    reference list. */
-   else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
+   else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
      {
      *errorcodeptr = ERR52;
      goto FAILED;
-Line 2504 
 for (;; ptr++)
+Line 3081 
 for (;; ptr++)
    if (inescq && c != 0)
      {
-     if (c == '\\' && ptr[1] == 'E')
+     if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
        {
        inescq = FALSE;
        ptr++;
-Line 2530 
 for (;; ptr++)
+Line 3107 
 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 2546 
 for (;; ptr++)
+Line 3124 
 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 2571 
 for (;; ptr++)
+Line 3149 
 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 2584 
 for (;; ptr++)
+Line 3162 
 for (;; ptr++)
          *errorcodeptr = ERR20;
          goto FAILED;
          }
-       *lengthptr += code - last_code;   /* To include callout length */
+       *lengthptr += (int)(code - last_code);   /* To include callout length */
        DPRINTF((">> end branch\n"));
        }
      return TRUE;
-Line 2594 
 for (;; ptr++)
+Line 3172 
 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 2603 
 for (;; ptr++)
+Line 3181 
 for (;; ptr++)
      *code++ = OP_CIRC;
      break;
-     case '$':
+     case CHAR_DOLLAR_SIGN:
      previous = NULL;
      *code++ = OP_DOLL;
      break;
-Line 2611 
 for (;; ptr++)
+Line 3189 
 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 2631 
 for (;; ptr++)
+Line 3209 
 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 ||
+          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 2654 
 for (;; ptr++)
+Line 3243 
 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;
        }
+     /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
+     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). */
-Line 2688 
 for (;; ptr++)
+Line 3295 
 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 2704 
 for (;; ptr++)
+Line 3311 
 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 2737 
 for (;; ptr++)
+Line 3344 
 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))
+            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++;
            }
-         posix_class = check_posix_name(ptr, tempptr - ptr);
+         posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
          if (posix_class < 0)
            {
            *errorcodeptr = ERR30;
-Line 2774 
 for (;; ptr++)
+Line 3381 
 for (;; ptr++)
          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
            posix_class = 0;
-         /* We build the bit map for the POSIX class in a chunk of local store
+         /* When PCRE_UCP is set, some of the POSIX classes are converted to
-         because we may be adding and subtracting from it, and we don't want to
+         different escape sequences that use Unicode properties. */
-         subtract bits that may be in the main map already. At the end we or the
-         result into the bit map that is being built. */
+ #ifdef SUPPORT_UCP
+         if ((options & PCRE_UCP) != 0)
+           {
+           int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
+           if (posix_substitutes[pc] != NULL)
+             {
+             nestptr = tempptr + 1;
+             ptr = posix_substitutes[pc] - 1;
+             continue;
+             }
+           }
+ #endif
+         /* In the non-UCP case, we build the bit map for the POSIX class in a
+         chunk of local store because we may be adding and subtracting from it,
+         and we don't want to subtract bits that may be in the main map already.
+         At the end we or the result into the bit map that is being built. */
          posix_class *= 3;
-Line 2821 
 for (;; ptr++)
+Line 3443 
 for (;; ptr++)
        /* Backslash may introduce a single character, or it may introduce one
        of the specials, which just set a flag. The sequence \b is a special
-       case. Inside a class (and only there) it is treated as backspace.
+       case. Inside a class (and only there) it is treated as backspace. We
-       Elsewhere it marks a word boundary. Other escapes have preset maps ready
+       assume that other escapes have more than one character in them, so set
-       to 'or' into the one we are building. We assume they have more than one
+       class_charcount bigger than one. Unrecognized escapes fall through and
-       character in them, so set class_charcount bigger than one. */
+       are either treated as literal characters (by default), or are faulted if
+       PCRE_EXTRA is set. */
-       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_R) c = '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 2850 
 for (;; ptr++)
+Line 3471 
 for (;; ptr++)
            register const uschar *cbits = cd->cbits;
            class_charcount += 2;     /* Greater than 1 is what matters */
-           /* Save time by not doing this in the pre-compile phase. */
+           switch (-c)
-           if (lengthptr == NULL) switch (-c)
              {
+ #ifdef SUPPORT_UCP
+             case ESC_du:     /* These are the values given for \d etc */
+             case ESC_DU:     /* when PCRE_UCP is set. We replace the */
+             case ESC_wu:     /* escape sequence with an appropriate \p */
+             case ESC_WU:     /* or \P to test Unicode properties instead */
+             case ESC_su:     /* of the default ASCII testing. */
+             case ESC_SU:
+             nestptr = ptr;
+             ptr = substitutes[-c - ESC_DU] - 1;  /* Just before substitute */
+             class_charcount -= 2;                /* Undo! */
+             continue;
+ #endif
              case ESC_d:
              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
              continue;
-Line 2883 
 for (;; ptr++)
+Line 3514 
 for (;; ptr++)
              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */
              continue;
-             default:    /* Not recognized; fall through */
+             case ESC_h:
-             break;      /* Need "default" setting to stop compiler warning. */
-             }
-           /* In the pre-compile phase, just do the recognition. */
-           else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
-                    c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
-           /* We need to deal with \H, \h, \V, and \v in both phases because
-           they use extra memory. */
-           if (-c == ESC_h)
-             {
              SETBIT(classbits, 0x09); /* VT */
              SETBIT(classbits, 0x20); /* SPACE */
              SETBIT(classbits, 0xa0); /* NSBP */
-Line 2920 
 for (;; ptr++)
+Line 3538 
 for (;; ptr++)
                }
  #endif
              continue;
-             }
-           if (-c == ESC_H)
+             case ESC_H:
-             {
              for (c = 0; c < 32; c++)
                {
                int x = 0xff;
-Line 2965 
 for (;; ptr++)
+Line 3581 
 for (;; ptr++)
                }
  #endif
              continue;
-             }
-           if (-c == ESC_v)
+             case ESC_v:
-             {
              SETBIT(classbits, 0x0a); /* LF */
              SETBIT(classbits, 0x0b); /* VT */
              SETBIT(classbits, 0x0c); /* FF */
-Line 2984 
 for (;; ptr++)
+Line 3598 
 for (;; ptr++)
                }
  #endif
              continue;
-             }
-           if (-c == ESC_V)
+             case ESC_V:
-             {
              for (c = 0; c < 32; c++)
                {
                int x = 0xff;
-Line 3017 
 for (;; ptr++)
+Line 3629 
 for (;; ptr++)
                }
  #endif
              continue;
-             }
-           /* We need to deal with \P and \p in both phases. */
  #ifdef SUPPORT_UCP
-           if (-c == ESC_p || -c == ESC_P)
+             case ESC_p:
-             {
+             case ESC_P:
-             BOOL negated;
+               {
-             int pdata;
+               BOOL negated;
-             int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
+               int pdata;
-             if (ptype < 0) goto FAILED;
+               int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
-             class_utf8 = TRUE;
+               if (ptype < 0) goto FAILED;
-             *class_utf8data++ = ((-c == ESC_p) != negated)?
+               class_utf8 = TRUE;
-               XCL_PROP : XCL_NOTPROP;
+               *class_utf8data++ = ((-c == ESC_p) != negated)?
-             *class_utf8data++ = ptype;
+                 XCL_PROP : XCL_NOTPROP;
-             *class_utf8data++ = pdata;
+               *class_utf8data++ = ptype;
-             class_charcount -= 2;   /* Not a < 256 character */
+               *class_utf8data++ = pdata;
-             continue;
+               class_charcount -= 2;   /* Not a < 256 character */
-             }
+               continue;
+               }
  #endif
-           /* Unrecognized escapes are faulted if PCRE is running in its
+             /* Unrecognized escapes are faulted if PCRE is running in its
-           strict mode. By default, for compatibility with Perl, they are
+             strict mode. By default, for compatibility with Perl, they are
-           treated as literals. */
+             treated as literals. */
-           if ((options & PCRE_EXTRA) != 0)
+             default:
-             {
+             if ((options & PCRE_EXTRA) != 0)
-             *errorcodeptr = ERR7;
+               {
-             goto FAILED;
+               *errorcodeptr = ERR7;
+               goto FAILED;
+               }
+             class_charcount -= 2;  /* Undo the default count from above */
+             c = *ptr;              /* Get the final character and fall through */
+             break;
              }
-           class_charcount -= 2;  /* Undo the default count from above */
-           c = *ptr;              /* Get the final character and fall through */
            }
          /* Fall through if we have a single character (c >= 0). This may be
-Line 3062 
 for (;; ptr++)
+Line 3674 
 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 3072 
 for (;; ptr++)
+Line 3684 
 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 3112 
 for (;; ptr++)
+Line 3725 
 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;
-           /* \b is backspace; \X is literal X; \R is literal R; any other
+           /* \b is backspace; any other special means the '-' was literal */
-           special means the '-' was literal */
            if (d < 0)
              {
-             if (d == -ESC_b) d = '\b';
+             if (d == -ESC_b) d = CHAR_BS; else
-             else if (d == -ESC_X) d = 'X';
-             else if (d == -ESC_R) d = 'R'; else
                {
                ptr = oldptr;
                goto LONE_SINGLE_CHARACTER;  /* A few lines below */
-Line 3145 
 for (;; ptr++)
+Line 3755 
 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 3265 
 for (;; ptr++)
+Line 3875 
 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 3290 
 for (;; ptr++)
+Line 3900 
 for (;; ptr++)
          }
        }
-     /* Loop until ']' reached. This "while" is the end of the "do" above. */
+     /* Loop until ']' reached. This "while" is the end of the "do" far above.
+     If we are at the end of an internal nested string, revert to the outer
+     string. */
+     while (((c = *(++ptr)) != 0 ||
+            (nestptr != NULL &&
+              (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
+            (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
-     while ((c = *(++ptr)) != 0 && (c != ']' || inescq));
+     /* Check for missing terminating ']' */
-     if (c == 0)                          /* Missing terminating ']' */
+     if (c == 0)
        {
        *errorcodeptr = ERR6;
        goto FAILED;
        }
- /* This code has been disabled because it would mean that \s counts as
- an explicit \r or \n reference, and that's not really what is wanted. Now
- we set the flag only if there is a literal "\r" or "\n" in the class. */
- #if 0
-     /* Remember whether \r or \n are in this class */
-     if (negate_class)
-       {
-       if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
-       }
-     else
-       {
-       if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
-       }
- #endif
      /* If class_charcount is 1, we saw precisely one character whose value is
      less than 256. As long as there were no characters >= 128 and there was no
      use of \p or \P, in other words, no use of any XCLASS features, we can
-Line 3382 
 we set the flag only if there is a liter
+Line 3980 
 we set the flag only if there is a liter
      /* 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
+     such as \S in the class, and PCRE_UCP is not set, because in that case all
-     the class, so any that were explicitly given as well can be ignored. If
+     characters > 255 are in the class, so any that were explicitly given as
-     (when there are explicit characters > 255 that must be listed) there are no
+     well can be ignored. If (when there are explicit characters > 255 that must
-     characters < 256, we can omit the bitmap in the actual compiled code. */
+     be listed) there are no characters < 256, we can omit the bitmap in the
+     actual compiled code. */
  #ifdef SUPPORT_UTF8
-     if (class_utf8 && !should_flip_negation)
+     if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
        {
        *class_utf8data++ = XCL_END;    /* Marks the end of extra data */
        *code++ = OP_XCLASS;
-Line 3414 
 we set the flag only if there is a liter
+Line 4013 
 we set the flag only if there is a liter
        }
  #endif
-     /* If there are no characters > 255, set the opcode to OP_CLASS or
+     /* If there are no characters > 255, or they are all to be included or
-     OP_NCLASS, depending on whether the whole class was negated and whether
+     excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
-     there were negative specials such as \S in the class. Then copy the 32-byte
+     whole class was negated and whether there were negative specials such as \S
-     map into the code vector, negating it if necessary. */
+     (non-UCP) 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 3437 
 we set the flag only if there is a liter
+Line 4037 
 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 3488 
 we set the flag only if there is a liter
+Line 4088 
 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 3541 
 we set the flag only if there is a liter
+Line 4141 
 we set the flag only if there is a liter
        if (!possessive_quantifier &&
            repeat_max < 0 &&
-           check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
+           check_auto_possessive(previous, utf8, ptr + 1, options, cd))
-             options, cd))
          {
          repeat_type = 0;    /* Force greedy */
          possessive_quantifier = TRUE;
-Line 3563 
 we set the flag only if there is a liter
+Line 4162 
 we set the flag only if there is a liter
        c = previous[1];
        if (!possessive_quantifier &&
            repeat_max < 0 &&
-           check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
+           check_auto_possessive(previous, utf8, ptr + 1, options, cd))
          {
          repeat_type = 0;    /* Force greedy */
          possessive_quantifier = TRUE;
-Line 3587 
 we set the flag only if there is a liter
+Line 4186 
 we set the flag only if there is a liter
        if (!possessive_quantifier &&
            repeat_max < 0 &&
-           check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
+           check_auto_possessive(previous, utf8, ptr + 1, options, cd))
          {
          repeat_type = 0;    /* Force greedy */
          possessive_quantifier = TRUE;
-Line 3609 
 we set the flag only if there is a liter
+Line 4208 
 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 3759 
 we set the flag only if there is a liter
+Line 4363 
 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 4396 
 we set the flag only if there is a liter
        {
        register int i;
        int ketoffset = 0;
-       int len = code - previous;
+       int len = (int)(code - previous);
        uschar *bralink = NULL;
        /* Repeating a DEFINE group is pointless */
-Line 3808 
 we set the flag only if there is a liter
+Line 4417 
 we set the flag only if there is a liter
          {
          register uschar *ket = previous;
          do ket += GET(ket, 1); while (*ket != OP_KET);
-         ketoffset = code - ket;
+         ketoffset = (int)(code - ket);
          }
        /* The case of a zero minimum is special because of the need to stick
-Line 3820 
 we set the flag only if there is a liter
+Line 4429 
 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 3866 
 we set the flag only if there is a liter
+Line 4485 
 we set the flag only if there is a liter
            /* We chain together the bracket offset fields that have to be
            filled in later when the ends of the brackets are reached. */
-           offset = (bralink == NULL)? 0 : previous - bralink;
+           offset = (bralink == NULL)? 0 : (int)(previous - bralink);
            bralink = previous;
            PUTINC(previous, 0, offset);
            }
-Line 3887 
 we set the flag only if there is a liter
+Line 4506 
 we set the flag only if there is a liter
            {
            /* In the pre-compile phase, we don't actually do the replication. We
            just adjust the length as if we had. Do some paranoid checks for
-           potential integer overflow. */
+           potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
+           integer type when available, otherwise double. */
            if (lengthptr != NULL)
              {
              int delta = (repeat_min - 1)*length_prevgroup;
-             if ((double)(repeat_min - 1)*(double)length_prevgroup >
+             if ((INT64_OR_DOUBLE)(repeat_min - 1)*
-                                                             (double)INT_MAX ||
+                   (INT64_OR_DOUBLE)length_prevgroup >
+                     (INT64_OR_DOUBLE)INT_MAX ||
                  OFLOW_MAX - *lengthptr < delta)
                {
                *errorcodeptr = ERR20;
-Line 3939 
 we set the flag only if there is a liter
+Line 4560 
 we set the flag only if there is a liter
          just adjust the length as if we had. For each repetition we must add 1
          to the length for BRAZERO and for all but the last repetition we must
          add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
-         paranoid checks to avoid integer overflow. */
+         paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
+         a 64-bit integer type when available, otherwise double. */
          if (lengthptr != NULL && repeat_max > 0)
            {
            int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
 - 2*LINK_SIZE;   /* Last one doesn't nest */
-           if ((double)repeat_max *
+           if ((INT64_OR_DOUBLE)repeat_max *
-                 (double)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
+                 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
-                   > (double)INT_MAX ||
+                   > (INT64_OR_DOUBLE)INT_MAX ||
                OFLOW_MAX - *lengthptr < delta)
              {
              *errorcodeptr = ERR20;
-Line 3972 
 we set the flag only if there is a liter
+Line 4594 
 we set the flag only if there is a liter
              {
              int offset;
              *code++ = OP_BRA;
-             offset = (bralink == NULL)? 0 : code - bralink;
+             offset = (bralink == NULL)? 0 : (int)(code - bralink);
              bralink = code;
              PUTINC(code, 0, offset);
              }
-Line 3993 
 we set the flag only if there is a liter
+Line 4615 
 we set the flag only if there is a liter
          while (bralink != NULL)
            {
            int oldlinkoffset;
-           int offset = code - bralink + 1;
+           int offset = (int)(code - bralink + 1);
            uschar *bra = code - offset;
            oldlinkoffset = GET(bra, 1);
            bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
-Line 4024 
 we set the flag only if there is a liter
+Line 4646 
 we set the flag only if there is a liter
            uschar *scode = bracode;
            do
              {
-             if (could_be_empty_branch(scode, ketcode, utf8))
+             if (could_be_empty_branch(scode, ketcode, utf8, cd))
                {
                *bracode += OP_SBRA - OP_BRA;
                break;
-Line 4036 
 we set the flag only if there is a liter
+Line 4658 
 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 4060 
 we set the flag only if there is a liter
+Line 4689 
 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);
-       len = code - tempcode;
+       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 = (int)(code - tempcode);
        if (len > 0) switch (*tempcode)
          {
          case OP_STAR:  *tempcode = OP_POSSTAR; break;
-Line 4083 
 we set the flag only if there is a liter
+Line 4721 
 we set the flag only if there is a liter
          case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
          case OP_NOTUPTO:  *tempcode = OP_NOTPOSUPTO; break;
+         /* Because we are moving code along, we must ensure that any
+         pending recursive references are updated. */
          default:
+         *code = OP_END;
+         adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
          memmove(tempcode + 1+LINK_SIZE, tempcode, len);
          code += 1 + LINK_SIZE;
          len += 1 + LINK_SIZE;
-Line 4110 
 we set the flag only if there is a liter
+Line 4753 
 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 4119 
 we set the flag only if there is a liter
+Line 4762 
 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 || ptr[1] == ':'))
        {
        int i, namelen;
+       int arglen = 0;
        const char *vn = verbnames;
-       const uschar *name = ++ptr;
+       const uschar *name = ptr + 1;
+       const uschar *arg = NULL;
        previous = NULL;
-       while ((cd->ctypes[*++ptr] & ctype_letter) != 0);
+       while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
-       if (*ptr == ':')
+       namelen = (int)(ptr - name);
+       if (*ptr == CHAR_COLON)
          {
-         *errorcodeptr = ERR59;   /* Not supported */
+         arg = ++ptr;
-         goto FAILED;
+         while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
+           || *ptr == '_') ptr++;
+         arglen = (int)(ptr - arg);
          }
-       if (*ptr != ')')
+       if (*ptr != CHAR_RIGHT_PARENTHESIS)
          {
          *errorcodeptr = ERR60;
          goto FAILED;
          }
-       namelen = ptr - name;
+       /* Scan the table of verb names */
        for (i = 0; i < verbcount; i++)
          {
          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;
-           break;
+           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);
+               }
+             }
+           /* Handle the cases with/without an argument */
+           if (arglen == 0)
+             {
+             if (verbs[i].op < 0)   /* Argument is mandatory */
+               {
+               *errorcodeptr = ERR66;
+               goto FAILED;
+               }
+             *code = verbs[i].op;
+             if (*code++ == OP_THEN)
+               {
+               PUT(code, 0, code - bcptr->current_branch - 1);
+               code += LINK_SIZE;
+               }
+             }
+           else
+             {
+             if (verbs[i].op_arg < 0)   /* Argument is forbidden */
+               {
+               *errorcodeptr = ERR59;
+               goto FAILED;
+               }
+             *code = verbs[i].op_arg;
+             if (*code++ == OP_THEN_ARG)
+               {
+               PUT(code, 0, code - bcptr->current_branch - 1);
+               code += LINK_SIZE;
+               }
+             *code++ = arglen;
+             memcpy(code, arg, arglen);
+             code += arglen;
+             *code++ = 0;
+             }
+           break;  /* Found verb, exit loop */
            }
          vn += verbs[i].len + 1;
          }
-       if (i < verbcount) continue;
-       *errorcodeptr = ERR60;
+       if (i < verbcount) continue;    /* Successfully handled a verb */
+       *errorcodeptr = ERR60;          /* Verb not recognized */
        goto FAILED;
        }
      /* 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 4165 
 we set the flag only if there is a liter
+Line 4867 
 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 4177 
 we set the flag only if there is a liter
+Line 4879 
 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 4209 
 we set the flag only if there is a liter
+Line 4911 
 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 4221 
 we set the flag only if there is a liter
+Line 4924 
 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 4231 
 we set the flag only if there is a liter
+Line 4934 
 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 4264 
 we set the flag only if there is a liter
+Line 4967 
 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;
+         namelen = (int)(ptr - name);
-         if ((terminator > 0 && *ptr++ != terminator) || *ptr++ != ')')
+         if ((terminator > 0 && *ptr++ != terminator) ||
+             *ptr++ != CHAR_RIGHT_PARENTHESIS)
            {
            ptr--;      /* Error offset */
            *errorcodeptr = ERR26;
-Line 4291 
 we set the flag only if there is a liter
+Line 4995 
 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 4303 
 we set the flag only if there is a liter
+Line 5007 
 we set the flag only if there is a liter
            }
          /* Otherwise (did not start with "+" or "-"), start by looking for the
-         name. */
+         name. If we find a name, add one to the opcode to change OP_CREF or
+         OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
+         except they record that the reference was originally to a name. The
+         information is used to check duplicate names. */
          slot = cd->name_table;
          for (i = 0; i < cd->names_found; i++)
-Line 4318 
 we set the flag only if there is a liter
+Line 5025 
 we set the flag only if there is a liter
            {
            recno = GET2(slot, 0);
            PUT2(code, 2+LINK_SIZE, recno);
+           code[1+LINK_SIZE]++;
            }
          /* 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);
+           code[1+LINK_SIZE]++;
            }
          /* If terminator == 0 it means that the name followed directly after
-Line 4343 
 we set the flag only if there is a liter
+Line 5052 
 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 4353 
 we set the flag only if there is a liter
+Line 5062 
 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 4363 
 we set the flag only if there is a liter
+Line 5072 
 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;
-Line 4388 
 we set the flag only if there is a liter
+Line 5097 
 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 4408 
 we set the flag only if there is a liter
+Line 5117 
 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 4431 
 we set the flag only if there is a liter
+Line 5140 
 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 4457 
 we set the flag only if there is a liter
+Line 5166 
 we set the flag only if there is a liter
              goto FAILED;
              }
            *code++ = n;
-           PUT(code, 0, ptr - cd->start_pattern + 1);  /* Pattern offset */
+           PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
-           PUT(code, LINK_SIZE, 0);                    /* Default length */
+           PUT(code, LINK_SIZE, 0);                          /* Default length */
            code += 2 * LINK_SIZE;
            }
          previous = NULL;
-Line 4466 
 we set the flag only if there is a liter
+Line 5175 
 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 4483 
 we set the flag only if there is a liter
+Line 5193 
 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++;
-           namelen = ptr - name;
+           namelen = (int)(ptr - name);
            /* In the pre-compile phase, just do a syntax check. */
-Line 4516 
 we set the flag only if there is a liter
+Line 5227 
 we set the flag only if there is a liter
                }
              }
-           /* In the real compile, create the entry in the table */
+           /* In the real compile, create the entry in the table, maintaining
+           alphabetical order. Duplicate names for different numbers are
+           permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
+           number are always OK. (An existing number can be re-used if (?|
+           appears in the pattern.) In either event, a duplicate name results in
+           a duplicate entry in the table, even if the number is the same. This
+           is because the number of names, and hence the table size, is computed
+           in the pre-compile, and it affects various numbers and pointers which
+           would all have to be modified, and the compiled code moved down, if
+           duplicates with the same number were omitted from the table. This
+           doesn't seem worth the hassle. However, *different* names for the
+           same number are not permitted. */
            else
              {
+             BOOL dupname = FALSE;
              slot = cd->name_table;
              for (i = 0; i < cd->names_found; i++)
                {
                int crc = memcmp(name, slot+2, namelen);
-Line 4528 
 we set the flag only if there is a liter
+Line 5252 
 we set the flag only if there is a liter
                  {
                  if (slot[2+namelen] == 0)
                    {
-                   if ((options & PCRE_DUPNAMES) == 0)
+                   if (GET2(slot, 0) != cd->bracount + 1 &&
+                       (options & PCRE_DUPNAMES) == 0)
                      {
                      *errorcodeptr = ERR43;
                      goto FAILED;
                      }
+                   else dupname = TRUE;
                    }
-                 else crc = -1;      /* Current name is substring */
+                 else crc = -1;      /* Current name is a substring */
                  }
+               /* Make space in the table and break the loop for an earlier
+               name. For a duplicate or later name, carry on. We do this for
+               duplicates so that in the simple case (when ?(| is not used) they
+               are in order of their numbers. */
                if (crc < 0)
                  {
                  memmove(slot + cd->name_entry_size, slot,
                    (cd->names_found - i) * cd->name_entry_size);
                  break;
                  }
+               /* Continue the loop for a later or duplicate name */
                slot += cd->name_entry_size;
                }
+             /* For non-duplicate names, check for a duplicate number before
+             adding the new name. */
+             if (!dupname)
+               {
+               uschar *cslot = cd->name_table;
+               for (i = 0; i < cd->names_found; i++)
+                 {
+                 if (cslot != slot)
+                   {
+                   if (GET2(cslot, 0) == cd->bracount + 1)
+                     {
+                     *errorcodeptr = ERR65;
+                     goto FAILED;
+                     }
+                   }
+                 else i--;
+                 cslot += cd->name_entry_size;
+                 }
+               }
              PUT2(slot, 0, cd->bracount + 1);
              memcpy(slot + 2, name, namelen);
              slot[2+namelen] = 0;
              }
            }
-         /* In both cases, count the number of names we've encountered. */
+         /* In both pre-compile and compile, count the number of names we've
+         encountered. */
-         ptr++;                    /* Move past > or ' */
          cd->names_found++;
+         ptr++;                    /* Move past > or ' */
          goto NUMBERED_GROUP;
          /* ------------------------------------------------------------ */
-         case '&':                 /* Perl recursion/subroutine syntax */
+         case CHAR_AMPERSAND:            /* Perl recursion/subroutine syntax */
-         terminator = ')';
+         terminator = CHAR_RIGHT_PARENTHESIS;
          is_recurse = TRUE;
          /* Fall through */
-Line 4568 
 we set the flag only if there is a liter
+Line 5325 
 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;
          while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
-         namelen = ptr - name;
+         namelen = (int)(ptr - name);
          /* In the pre-compile phase, do a syntax check and set a dummy
          reference number. */
-Line 4619 
 we set the flag only if there is a liter
+Line 5376 
 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 4635 
 we set the flag only if there is a liter
+Line 5392 
 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)
-Line 4656 
 we set the flag only if there is a liter
+Line 5422 
 we set the flag only if there is a liter
                goto FAILED;
                }
              }
-           else if (refsign == '-')
+           else if (refsign == CHAR_MINUS)
              {
              if ((digitab[ptr[1]] & ctype_digit) == 0)
                goto OTHER_CHAR_AFTER_QUERY;
-Line 4665 
 we set the flag only if there is a liter
+Line 5431 
 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 4687 
 we set the flag only if there is a liter
+Line 5453 
 we set the flag only if there is a liter
                goto FAILED;
                }
              }
-           else if (refsign == '+')
+           else if (refsign == CHAR_PLUS)
              {
              if (recno == 0)
                {
-Line 4714 
 we set the flag only if there is a liter
+Line 5480 
 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;
                  }
+               /* Fudge the value of "called" so that when it is inserted as an
+               offset below, what it actually inserted is the reference number
+               of the group. */
                called = cd->start_code + recno;
-               PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);
+               PUTINC(cd->hwm, 0, (int)(code + 2 + LINK_SIZE - cd->start_code));
                }
              /* If not a forward reference, and the subpattern is still open,
-Line 4735 
 we set the flag only if there is a liter
+Line 5507 
 we set the flag only if there is a liter
              recursion that could loop for ever, and diagnose that case. */
              else if (GET(called, 1) == 0 &&
-                      could_be_empty(called, code, bcptr, utf8))
+                      could_be_empty(called, code, bcptr, utf8, cd))
                {
                *errorcodeptr = ERR40;
                goto FAILED;
-Line 4751 
 we set the flag only if there is a liter
+Line 5523 
 we set the flag only if there is a liter
            code += 1 + LINK_SIZE;
            *code = OP_RECURSE;
-           PUT(code, 1, called - cd->start_code);
+           PUT(code, 1, (int)(called - cd->start_code));
            code += 1 + LINK_SIZE;
            *code = OP_KET;
-Line 4773 
 we set the flag only if there is a liter
+Line 5545 
 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 4820 
 we set the flag only if there is a liter
+Line 5592 
 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
+           else
              {
              if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
                {
                *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). */
 &