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

Diff of /code/trunk/pcre_compile.c

Parent Directory | Revision Log | View Patch Patch

-revision 598 by ph10,
Sat May  7 15:37:31 2011 UTC
+revision 1055 by chpe,
Tue Oct 16 15:53:30 2012 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2011 University of Cambridge
+            Copyright (c) 1997-2012 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 PCRE_DEBUG is defined, we need the pcre_printint() function, which is
+ /* When PCRE_DEBUG is defined, we need the pcre(16|32)_printint() function, which
- also used by pcretest. PCRE_DEBUG is not defined when building a production
+ is also used by pcretest. PCRE_DEBUG is not defined when building a production
- library. */
+ library. We do not need to select pcre16_printint.c specially, because the
+ COMPILE_PCREx macro will already be appropriately set. */
  #ifdef PCRE_DEBUG
- #include "pcre_printint.src"
+ /* pcre_printint.c should not include any headers */
+ #define PCRE_INCLUDED
+ #include "pcre_printint.c"
+ #undef PCRE_INCLUDED
  #endif
  /* Macro for setting individual bits in class bitmaps. */
- #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
+ #define SETBIT(a,b) a[(b)/8] |= (1 << ((b)&7))
  /* Maximum length value to check against when making sure that the integer that
  holds the compiled pattern length does not overflow. We make it a bit less than
-Line 73 
 to check them every time. */
+Line 77 
 to check them every time. */
  #define OFLOW_MAX (INT_MAX - 20)
+ /* Definitions to allow mutual recursion */
+ static int
+   add_list_to_class(pcre_uint8 *, pcre_uchar **, int, compile_data *,
+     const pcre_uint32 *, unsigned int);
+ static BOOL
+   compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL,
+     int, int, int *, int *, branch_chain *, compile_data *, int *);
  /*************************************************
  *      Code parameters and static tables         *
-Line 88 
 so this number is very generous.
+Line 103 
 so this number is very generous.
  The same workspace is used during the second, actual compile phase for
  remembering forward references to groups so that they can be filled in at the
  end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
- is 4 there is plenty of room. */
+ is 4 there is plenty of room for most patterns. However, the memory can get
+ filled up by repetitions of forward references, for example patterns like
+ /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
+ that the workspace is expanded using malloc() in this situation. The value
+ below is therefore a minimum, and we put a maximum on it for safety. The
+ minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
+ kicks in at the same number of forward references in all cases. */
- #define COMPILE_WORK_SIZE (4096)
+ #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
+ #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
  /* 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)
+ #define WORK_SIZE_SAFETY_MARGIN (100)
+ /* Private flags added to firstchar and reqchar. */
+ #define REQ_CASELESS   0x10000000l      /* Indicates caselessness */
+ #define REQ_VARY       0x20000000l      /* Reqchar followed non-literal item */
+ #define REQ_MASK       (REQ_CASELESS | REQ_VARY)
+ /* Repeated character flags. */
+ #define UTF_LENGTH     0x10000000l      /* The char contains its length. */
  /* 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
-Line 231 
 static const char posix_names[] =
+Line 262 
 static const char posix_names[] =
    STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
    STRING_word0  STRING_xdigit;
- static const uschar posix_name_lengths[] = {
+ static const pcre_uint8 posix_name_lengths[] = {
 , 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
  /* Table of class bit maps for each POSIX class. Each class is formed from a
-Line 266 
 substitutes must be in the order of the
+Line 297 
 substitutes must be in the order of the
  both positive and negative cases. NULL means no substitute. */
  #ifdef SUPPORT_UCP
- static const uschar *substitutes[] = {
+ static const pcre_uchar string_PNd[]  = {
-   (uschar *)"\\P{Nd}",    /* \D */
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
-   (uschar *)"\\p{Nd}",    /* \d */
+   CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
-   (uschar *)"\\P{Xsp}",   /* \S */       /* NOTE: Xsp is Perl space */
+ static const pcre_uchar string_pNd[]  = {
-   (uschar *)"\\p{Xsp}",   /* \s */
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
-   (uschar *)"\\P{Xwd}",   /* \W */
+   CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
-   (uschar *)"\\p{Xwd}"    /* \w */
+ static const pcre_uchar string_PXsp[] = {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_pXsp[] = {
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_PXwd[] = {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_pXwd[] = {
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar *substitutes[] = {
+   string_PNd,           /* \D */
+   string_pNd,           /* \d */
+   string_PXsp,          /* \S */       /* NOTE: Xsp is Perl space */
+   string_pXsp,          /* \s */
+   string_PXwd,          /* \W */
+   string_pXwd           /* \w */
  };
- static const uschar *posix_substitutes[] = {
+ static const pcre_uchar string_pL[] =   {
-   (uschar *)"\\p{L}",     /* alpha */
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
-   (uschar *)"\\p{Ll}",    /* lower */
+   CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
-   (uschar *)"\\p{Lu}",    /* upper */
+ static const pcre_uchar string_pLl[] =  {
-   (uschar *)"\\p{Xan}",   /* alnum */
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
-   NULL,                   /* ascii */
+   CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
-   (uschar *)"\\h",        /* blank */
+ static const pcre_uchar string_pLu[] =  {
-   NULL,                   /* cntrl */
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
-   (uschar *)"\\p{Nd}",    /* digit */
+   CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
-   NULL,                   /* graph */
+ static const pcre_uchar string_pXan[] = {
-   NULL,                   /* print */
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
-   NULL,                   /* punct */
+   CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
-   (uschar *)"\\p{Xps}",   /* space */    /* NOTE: Xps is POSIX space */
+ static const pcre_uchar string_h[] =    {
-   (uschar *)"\\p{Xwd}",   /* word */
+   CHAR_BACKSLASH, CHAR_h, '\0' };
-   NULL,                   /* xdigit */
+ static const pcre_uchar string_pXps[] = {
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_PL[] =   {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_PLl[] =  {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_PLu[] =  {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_PXan[] = {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_H[] =    {
+   CHAR_BACKSLASH, CHAR_H, '\0' };
+ static const pcre_uchar string_PXps[] = {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar *posix_substitutes[] = {
+   string_pL,            /* alpha */
+   string_pLl,           /* lower */
+   string_pLu,           /* upper */
+   string_pXan,          /* alnum */
+   NULL,                 /* ascii */
+   string_h,             /* blank */
+   NULL,                 /* cntrl */
+   string_pNd,           /* digit */
+   NULL,                 /* graph */
+   NULL,                 /* print */
+   NULL,                 /* punct */
+   string_pXps,          /* space */    /* NOTE: Xps is POSIX space */
+   string_pXwd,          /* word */
+   NULL,                 /* xdigit */
    /* Negated cases */
-   (uschar *)"\\P{L}",     /* ^alpha */
+   string_PL,            /* ^alpha */
-   (uschar *)"\\P{Ll}",    /* ^lower */
+   string_PLl,           /* ^lower */
-   (uschar *)"\\P{Lu}",    /* ^upper */
+   string_PLu,           /* ^upper */
-   (uschar *)"\\P{Xan}",   /* ^alnum */
+   string_PXan,          /* ^alnum */
-   NULL,                   /* ^ascii */
+   NULL,                 /* ^ascii */
-   (uschar *)"\\H",        /* ^blank */
+   string_H,             /* ^blank */
-   NULL,                   /* ^cntrl */
+   NULL,                 /* ^cntrl */
-   (uschar *)"\\P{Nd}",    /* ^digit */
+   string_PNd,           /* ^digit */
-   NULL,                   /* ^graph */
+   NULL,                 /* ^graph */
-   NULL,                   /* ^print */
+   NULL,                 /* ^print */
-   NULL,                   /* ^punct */
+   NULL,                 /* ^punct */
-   (uschar *)"\\P{Xps}",   /* ^space */   /* NOTE: Xps is POSIX space */
+   string_PXps,          /* ^space */   /* NOTE: Xps is POSIX space */
-   (uschar *)"\\P{Xwd}",   /* ^word */
+   string_PXwd,          /* ^word */
-   NULL                    /* ^xdigit */
+   NULL                  /* ^xdigit */
  };
- #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
+ #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
  #endif
  #define STRING(a)  # a
-Line 365 
 static const char error_texts[] =
+Line 450 
 static const char error_texts[] =
    /* 30 */
    "unknown POSIX class name\0"
    "POSIX collating elements are not supported\0"
-   "this version of PCRE is not compiled with PCRE_UTF8 support\0"
+   "this version of PCRE is compiled without UTF support\0"
    "spare error\0"  /** DEAD **/
    "character value in \\x{...} sequence is too large\0"
    /* 35 */
-Line 388 
 static const char error_texts[] =
+Line 473 
 static const char error_texts[] =
    "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
    /* 50 */
    "repeated subpattern is too long\0"    /** DEAD **/
-   "octal value is greater than \\377 (not in UTF-8 mode)\0"
+   "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0"
    "internal error: overran compiling workspace\0"
    "internal error: previously-checked referenced subpattern not found\0"
    "DEFINE group contains more than one branch\0"
    /* 55 */
-   "repeating a DEFINE group is not allowed\0"
+   "repeating a DEFINE group is not allowed\0"  /** DEAD **/
    "inconsistent NEWLINE options\0"
    "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
    "a numbered reference must not be zero\0"
-Line 407 
 static const char error_texts[] =
+Line 492 
 static const char error_texts[] =
    /* 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"
+   "this version of PCRE is not compiled with Unicode property support\0"
    "\\c must be followed by an ASCII character\0"
+   "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
+   /* 70 */
+   "internal error: unknown opcode in find_fixedlength()\0"
+   "\\N is not supported in a class\0"
+   "too many forward references\0"
+   "disallowed Unicode code point (>= 0xd800 && <= 0xdfff)\0"
+   "invalid UTF-16 string\0"
+   /* 75 */
+   "name is too long in (*MARK), (*PRUNE), (*SKIP), or (*THEN)\0"
+   "character value in \\u.... sequence is too large\0"
+   "invalid UTF-32 string\0"
    ;
  /* Table to identify digits and hex digits. This is used when compiling
-Line 427 
 For convenience, we use the same bit def
+Line 523 
 For convenience, we use the same bit def
  Then we can use ctype_digit and ctype_xdigit in the code. */
+ /* Using a simple comparison for decimal numbers rather than a memory read
+ is much faster, and the resulting code is simpler (the compiler turns it
+ into a subtraction and unsigned comparison). */
+ #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
  #ifndef EBCDIC
  /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
  UTF-8 mode. */
- static const unsigned char digitab[] =
+ static const pcre_uint8 digitab[] =
    {
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   8- 15 */
-Line 471 
 static const unsigned char digitab[] =
+Line 573 
 static const unsigned char digitab[] =
  /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
- static const unsigned char digitab[] =
+ static const pcre_uint8 digitab[] =
    {
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7  0 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   8- 15    */
-Line 506 
 static const unsigned char digitab[] =
+Line 608 
 static const unsigned char digitab[] =
 x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /*  0 - 7  F0 */
 x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/*  8 -255    */
- static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
+ static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
 x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /*   0-  7 */
 x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /*   8- 15 */
 x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /*  16- 23 */
-Line 542 
 static const unsigned char ebcdic_charta
+Line 644 
 static const unsigned char ebcdic_charta
  #endif
- /* Definition to allow mutual recursion */
- static BOOL
-   compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
-     int *, int *, branch_chain *, compile_data *, int *);
  /*************************************************
-Line 577 
 return s;
+Line 673 
 return s;
  /*************************************************
+ *           Expand the workspace                 *
+ *************************************************/
+ /* This function is called during the second compiling phase, if the number of
+ forward references fills the existing workspace, which is originally a block on
+ the stack. A larger block is obtained from malloc() unless the ultimate limit
+ has been reached or the increase will be rather small.
+ Argument: pointer to the compile data block
+ Returns:  0 if all went well, else an error number
+ */
+ static int
+ expand_workspace(compile_data *cd)
+ {
+ pcre_uchar *newspace;
+ int newsize = cd->workspace_size * 2;
+ if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
+ if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
+     newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
+  return ERR72;
+ newspace = (PUBL(malloc))(IN_UCHARS(newsize));
+ if (newspace == NULL) return ERR21;
+ memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
+ cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
+ if (cd->workspace_size > COMPILE_WORK_SIZE)
+   (PUBL(free))((void *)cd->start_workspace);
+ cd->start_workspace = newspace;
+ cd->workspace_size = newsize;
+ return 0;
+ }
+ /*************************************************
+ *            Check for counted repeat            *
+ *************************************************/
+ /* This function is called when a '{' is encountered in a place where it might
+ start a quantifier. It looks ahead to see if it really is a quantifier or not.
+ It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
+ where the ddds are digits.
+ Arguments:
+   p         pointer to the first char after '{'
+ Returns:    TRUE or FALSE
+ */
+ static BOOL
+ is_counted_repeat(const pcre_uchar *p)
+ {
+ if (!IS_DIGIT(*p)) return FALSE;
+ p++;
+ while (IS_DIGIT(*p)) p++;
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
+ if (*p++ != CHAR_COMMA) return FALSE;
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
+ if (!IS_DIGIT(*p)) return FALSE;
+ p++;
+ while (IS_DIGIT(*p)) p++;
+ return (*p == CHAR_RIGHT_CURLY_BRACKET);
+ }
+ /*************************************************
  *            Handle escapes                      *
  *************************************************/
-Line 601 
 Returns:         zero or positive => a d
+Line 769 
 Returns:         zero or positive => a d
  */
  static int
- check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
+ check_escape(const pcre_uchar **ptrptr, int *errorcodeptr, int bracount,
    int options, BOOL isclass)
  {
- BOOL utf8 = (options & PCRE_UTF8) != 0;
+ /* PCRE_UTF16 has the same value as PCRE_UTF8. */
- const uschar *ptr = *ptrptr + 1;
+ BOOL utf = (options & PCRE_UTF8) != 0;
- int c, i;
+ const pcre_uchar *ptr = *ptrptr + 1;
+ pcre_int32 c;
+ int i;
  GETCHARINCTEST(c, ptr);           /* Get character value, increment pointer */
  ptr--;                            /* Set pointer back to the last byte */
-Line 620 
 in a table. A non-zero result is somethi
+Line 790 
 in a table. A non-zero result is somethi
  Otherwise further processing may be required. */
  #ifndef EBCDIC  /* ASCII/UTF-8 coding */
- else if (c < CHAR_0 || c > CHAR_z) {}                     /* Not alphanumeric */
+ /* Not alphanumeric */
+ else if (c < CHAR_0 || c > CHAR_z) {}
  else if ((i = escapes[c - CHAR_0]) != 0) c = i;
  #else           /* EBCDIC coding */
- else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphanumeric */
+ /* Not alphanumeric */
+ else if (c < CHAR_a || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
  else if ((i = escapes[c - 0x48]) != 0)  c = i;
  #endif
-Line 632 
 else if ((i = escapes[c - 0x48]) != 0)
+Line 804 
 else if ((i = escapes[c - 0x48]) != 0)
  else
    {
-   const uschar *oldptr;
+   const pcre_uchar *oldptr;
    BOOL braced, negated;
    switch (c)
-Line 642 
 else
+Line 814 
 else
      case CHAR_l:
      case CHAR_L:
+     *errorcodeptr = ERR37;
+     break;
      case CHAR_u:
+     if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
+       {
+       /* In JavaScript, \u must be followed by four hexadecimal numbers.
+       Otherwise it is a lowercase u letter. */
+       if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
+         && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
+         && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
+         && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
+         {
+         c = 0;
+         for (i = 0; i < 4; ++i)
+           {
+           register int cc = *(++ptr);
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
+           if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
+           c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
+ #else           /* EBCDIC coding */
+           if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
+           c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
+ #endif
+           }
+ #if defined COMPILE_PCRE8
+         if (c > (utf ? 0x10ffff : 0xff))
+ #elif defined COMPILE_PCRE16
+         if (c > (utf ? 0x10ffff : 0xffff))
+ #elif defined COMPILE_PCRE32
+         if (utf && c > 0x10ffff)
+ #endif
+           {
+           *errorcodeptr = ERR76;
+           }
+         else if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
+         }
+       }
+     else
+       *errorcodeptr = ERR37;
+     break;
      case CHAR_U:
-     *errorcodeptr = ERR37;
+     /* In JavaScript, \U is an uppercase U letter. */
+     if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
      break;
-     /* \g must be followed by one of a number of specific things:
+     /* In a character class, \g is just a literal "g". Outside a character
+     class, \g must be followed by one of a number of specific things:
      (1) A number, either plain or braced. If positive, it is an absolute
      backreference. If negative, it is a relative backreference. This is a Perl
-Line 664 
 else
+Line 880 
 else
      the -ESC_g code (cf \k). */
      case CHAR_g:
+     if (isclass) break;
      if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
        {
        c = -ESC_g;
-Line 674 
 else
+Line 891 
 else
      if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
        {
-       const uschar *p;
+       const pcre_uchar *p;
        for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
-         if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
+         if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
        if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
          {
          c = -ESC_k;
-Line 694 
 else
+Line 911 
 else
        }
      else negated = FALSE;
+     /* The integer range is limited by the machine's int representation. */
      c = 0;
-     while ((digitab[ptr[1]] & ctype_digit) != 0)
+     while (IS_DIGIT(ptr[1]))
+       {
+       if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
+         {
+         c = -1;
+         break;
+         }
        c = c * 10 + *(++ptr) - CHAR_0;
+       }
-     if (c < 0)   /* Integer overflow */
+     if (((unsigned int)c) > INT_MAX) /* Integer overflow */
        {
+       while (IS_DIGIT(ptr[1]))
+         ptr++;
        *errorcodeptr = ERR61;
        break;
        }
-Line 747 
 else
+Line 973 
 else
      if (!isclass)
        {
        oldptr = ptr;
+       /* The integer range is limited by the machine's int representation. */
        c -= CHAR_0;
-       while ((digitab[ptr[1]] & ctype_digit) != 0)
+       while (IS_DIGIT(ptr[1]))
+         {
+         if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
+           {
+           c = -1;
+           break;
+           }
          c = c * 10 + *(++ptr) - CHAR_0;
-       if (c < 0)    /* Integer overflow */
+         }
+       if (((unsigned int)c) > INT_MAX) /* Integer overflow */
          {
+         while (IS_DIGIT(ptr[1]))
+           ptr++;
          *errorcodeptr = ERR61;
          break;
          }
-Line 777 
 else
+Line 1013 
 else
      /* \0 always starts an octal number, but we may drop through to here with a
      larger first octal digit. The original code used just to take the least
      significant 8 bits of octal numbers (I think this is what early Perls used
-     to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
+     to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
-     than 3 octal digits. */
+     but no more than 3 octal digits. */
      case CHAR_0:
      c -= CHAR_0;
      while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
          c = c * 8 + *(++ptr) - CHAR_0;
-     if (!utf8 && c > 255) *errorcodeptr = ERR51;
+ #ifdef COMPILE_PCRE8
+     if (!utf && c > 0xff) *errorcodeptr = ERR51;
+ #endif
      break;
      /* \x is complicated. \x{ddd} is a character number which can be greater
-     than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
+     than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
-     treated as a data character. */
+     If not, { is treated as a data character. */
      case CHAR_x:
+     if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
+       {
+       /* In JavaScript, \x must be followed by two hexadecimal numbers.
+       Otherwise it is a lowercase x letter. */
+       if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
+         && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
+         {
+         c = 0;
+         for (i = 0; i < 2; ++i)
+           {
+           register int cc = *(++ptr);
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
+           if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
+           c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
+ #else           /* EBCDIC coding */
+           if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
+           c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
+ #endif
+           }
+         }
+       break;
+       }
      if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
        {
-       const uschar *pt = ptr + 2;
+       const pcre_uchar *pt = ptr + 2;
-       int count = 0;
        c = 0;
-       while ((digitab[*pt] & ctype_xdigit) != 0)
+       while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
          {
          register int cc = *pt++;
          if (c == 0 && cc == CHAR_0) continue;     /* Leading zeroes */
-         count++;
  #ifndef EBCDIC  /* ASCII/UTF-8 coding */
          if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
-Line 811 
 else
+Line 1070 
 else
          if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
          c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
  #endif
+ #if defined COMPILE_PCRE8
+         if (c > (utf ? 0x10ffff : 0xff)) { c = -1; break; }
+ #elif defined COMPILE_PCRE16
+         if (c > (utf ? 0x10ffff : 0xffff)) { c = -1; break; }
+ #elif defined COMPILE_PCRE32
+         if (utf && c > 0x10ffff) { c = -1; break; }
+ #endif
+         }
+       if (c < 0)
+         {
+         while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
+         *errorcodeptr = ERR34;
          }
        if (*pt == CHAR_RIGHT_CURLY_BRACKET)
          {
-         if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
+         if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
          ptr = pt;
          break;
          }
-Line 827 
 else
+Line 1100 
 else
      /* Read just a single-byte hex-defined char */
      c = 0;
-     while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
+     while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
        {
        int cc;                                  /* Some compilers don't like */
        cc = *(++ptr);                           /* ++ in initializers */
-Line 885 
 else
+Line 1158 
 else
    }
  /* Perl supports \N{name} for character names, as well as plain \N for "not
- newline". PCRE does not support \N{name}. */
+ newline". PCRE does not support \N{name}. However, it does support
+ quantification such as \N{2,3}. */
- if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)
+ if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
+      !is_counted_repeat(ptr+2))
    *errorcodeptr = ERR37;
  /* If PCRE_UCP is set, we change the values for \d etc. */
-Line 923 
 Returns:         type value from ucp_typ
+Line 1198 
 Returns:         type value from ucp_typ
  */
  static int
- get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
+ get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
  {
  int c, i, bot, top;
- const uschar *ptr = *ptrptr;
+ const pcre_uchar *ptr = *ptrptr;
- char name[32];
+ pcre_uchar name[32];
  c = *(++ptr);
  if (c == 0) goto ERROR_RETURN;
-Line 944 
 if (c == CHAR_LEFT_CURLY_BRACKET)
+Line 1219 
 if (c == CHAR_LEFT_CURLY_BRACKET)
      *negptr = TRUE;
      ptr++;
      }
-   for (i = 0; i < (int)sizeof(name) - 1; i++)
+   for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
      {
      c = *(++ptr);
      if (c == 0) goto ERROR_RETURN;
-Line 968 
 else
+Line 1243 
 else
  /* Search for a recognized property name using binary chop */
  bot = 0;
- top = _pcre_utt_size;
+ top = PRIV(utt_size);
  while (bot < top)
    {
    i = (bot + top) >> 1;
-   c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
+   c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
    if (c == 0)
      {
-     *dptr = _pcre_utt[i].value;
+     *dptr = PRIV(utt)[i].value;
-     return _pcre_utt[i].type;
+     return PRIV(utt)[i].type;
      }
    if (c > 0) bot = i + 1; else top = i;
    }
-Line 997 
 return -1;
+Line 1272 
 return -1;
  /*************************************************
- *            Check for counted repeat            *
- *************************************************/
- /* This function is called when a '{' is encountered in a place where it might
- start a quantifier. It looks ahead to see if it really is a quantifier or not.
- It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
- where the ddds are digits.
- Arguments:
-   p         pointer to the first char after '{'
- Returns:    TRUE or FALSE
- */
- static BOOL
- is_counted_repeat(const uschar *p)
- {
- if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
- while ((digitab[*p] & ctype_digit) != 0) p++;
- if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
- if (*p++ != CHAR_COMMA) return FALSE;
- 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 == CHAR_RIGHT_CURLY_BRACKET);
- }
- /*************************************************
  *         Read repeat counts                     *
  *************************************************/
-Line 1048 
 Returns:         pointer to '}' on succe
+Line 1290 
 Returns:         pointer to '}' on succe
                   current ptr on error, with errorcodeptr set non-zero
  */
- static const uschar *
+ static const pcre_uchar *
- read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
+ read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
  {
  int min = 0;
  int max = -1;
-Line 1057 
 int max = -1;
+Line 1299 
 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++ - CHAR_0;
+ while (IS_DIGIT(*p)) min = min * 10 + *p++ - CHAR_0;
  if (min < 0 || min > 65535)
    {
    *errorcodeptr = ERR5;
-Line 1072 
 if (*p == CHAR_RIGHT_CURLY_BRACKET) max
+Line 1314 
 if (*p == CHAR_RIGHT_CURLY_BRACKET) max
    if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
      {
      max = 0;
-     while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
+     while(IS_DIGIT(*p)) max = max * 10 + *p++ - CHAR_0;
      if (max < 0 || max > 65535)
        {
        *errorcodeptr = ERR5;
-Line 1127 
 Arguments:
+Line 1369 
 Arguments:
    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
-   utf8         TRUE if we are in UTF-8 mode
+   utf          TRUE if we are in UTF-8 / UTF-16 / UTF-32 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_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
+ find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
-   BOOL xmode, BOOL utf8, int *count)
+   BOOL xmode, BOOL utf, int *count)
  {
- uschar *ptr = *ptrptr;
+ pcre_uchar *ptr = *ptrptr;
  int start_count = *count;
  int hwm_count = start_count;
  BOOL dup_parens = FALSE;
-Line 1204 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
+Line 1446 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
          ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
        {
        int term;
-       const uschar *thisname;
+       const pcre_uchar *thisname;
        *count += 1;
        if (name == NULL && *count == lorn) return *count;
        term = *ptr++;
-Line 1212 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
+Line 1454 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
        thisname = ptr;
        while (*ptr != term) ptr++;
        if (name != NULL && lorn == ptr - thisname &&
-           strncmp((const char *)name, (const char *)thisname, lorn) == 0)
+           STRNCMP_UC_UC(name, thisname, lorn) == 0)
          return *count;
        term++;
        }
-Line 1255 
 for (; ptr < cd->end_pattern; ptr++)
+Line 1497 
 for (; ptr < cd->end_pattern; ptr++)
          {
          if (ptr[2] == CHAR_E)
            ptr+= 2;
-         else if (strncmp((const char *)ptr+2,
+         else if (STRNCMP_UC_C8(ptr + 2,
                   STR_Q STR_BACKSLASH STR_E, 3) == 0)
            ptr += 4;
          else
-Line 1303 
 for (; ptr < cd->end_pattern; ptr++)
+Line 1545 
 for (; ptr < cd->end_pattern; ptr++)
        {
        if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
        ptr++;
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-       if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
+       if (utf) FORWARDCHAR(ptr);
  #endif
        }
      if (*ptr == 0) goto FAIL_EXIT;
-Line 1315 
 for (; ptr < cd->end_pattern; ptr++)
+Line 1557 
 for (; ptr < cd->end_pattern; ptr++)
    if (*ptr == CHAR_LEFT_PARENTHESIS)
      {
-     int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
+     int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
      if (rc > 0) return rc;
      if (*ptr == 0) goto FAIL_EXIT;
      }
-Line 1361 
 Arguments:
+Line 1603 
 Arguments:
    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
-   utf8         TRUE if we are in UTF-8 mode
+   utf          TRUE if we are in UTF-8 / UTF-16 / UTF-32 mode
  Returns:       the number of the found subpattern, or -1 if not found
  */
  static int
- find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
+ find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
-   BOOL utf8)
+   BOOL utf)
  {
- uschar *ptr = (uschar *)cd->start_pattern;
+ pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
  int count = 0;
  int rc;
-Line 1381 
 matching closing parens. That is why we
+Line 1623 
 matching closing parens. That is why we
  for (;;)
    {
-   rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
+   rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
    if (rc > 0 || *ptr++ == 0) break;
    }
-Line 1397 
 return rc;
+Line 1639 
 return rc;
  /* This is called by several functions that scan a compiled expression looking
  for a fixed first character, or an anchoring op code etc. It skips over things
- that do not influence this. For some calls, a change of option is important.
+ that do not influence this. For some calls, it makes sense to skip negative
- For some calls, it makes sense to skip negative forward and all backward
+ forward and all backward assertions, and also the \b assertion; for others it
- assertions, and also the \b assertion; for others it does not.
+ does not.
  Arguments:
    code         pointer to the start of the group
-   options      pointer to external options
-   optbit       the option bit whose changing is significant, or
-                  zero if none are
    skipassert   TRUE if certain assertions are to be skipped
  Returns:       pointer to the first significant opcode
  */
- static const uschar*
+ static const pcre_uchar*
- first_significant_code(const uschar *code, int *options, int optbit,
+ first_significant_code(const pcre_uchar *code, BOOL skipassert)
-   BOOL skipassert)
  {
  for (;;)
    {
    switch ((int)*code)
      {
-     case OP_OPT:
-     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
-       *options = (int)code[1];
-     code += 2;
-     break;
      case OP_ASSERT_NOT:
      case OP_ASSERTBACK:
      case OP_ASSERTBACK_NOT:
      if (!skipassert) return code;
      do code += GET(code, 1); while (*code == OP_ALT);
-     code += _pcre_OP_lengths[*code];
+     code += PRIV(OP_lengths)[*code];
      break;
      case OP_WORD_BOUNDARY:
-Line 1444 
 for (;;)
+Line 1676 
 for (;;)
      case OP_RREF:
      case OP_NRREF:
      case OP_DEF:
-     code += _pcre_OP_lengths[*code];
+     code += PRIV(OP_lengths)[*code];
      break;
      default:
-Line 1474 
 and doing the check at the end; a flag s
+Line 1706 
 and doing the check at the end; a flag s
  Arguments:
    code     points to the start of the pattern (the bracket)
-   options  the compiling options
+   utf      TRUE in UTF-8 / UTF-16 / UTF-32 mode
    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,
-              or -2 if \C was encountered
+              or -2 if \C was encountered (in UTF-8 mode only)
               or -3 if an OP_RECURSE item was encountered and atend is FALSE
+              or -4 if an unknown opcode was encountered (internal error)
  */
  static int
- find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
+ find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
  {
  int length = -1;
  register int branchlength = 0;
- register uschar *cc = code + 1 + LINK_SIZE;
+ register pcre_uchar *cc = code + 1 + LINK_SIZE;
  /* Scan along the opcodes for this branch. If we get to the end of the
  branch, check the length against that of the other branches. */
-Line 1498 
 branch, check the length against that of
+Line 1731 
 branch, check the length against that of
  for (;;)
    {
    int d;
-   uschar *ce, *cs;
+   pcre_uchar *ce, *cs;
    register int op = *cc;
    switch (op)
      {
+     /* We only need to continue for OP_CBRA (normal capturing bracket) and
+     OP_BRA (normal non-capturing bracket) because the other variants of these
+     opcodes are all concerned with unlimited repeated groups, which of course
+     are not of fixed length. */
      case OP_CBRA:
      case OP_BRA:
      case OP_ONCE:
+     case OP_ONCE_NC:
      case OP_COND:
-     d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
+     d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, atend, cd);
      if (d < 0) return d;
      branchlength += d;
      do cc += GET(cc, 1); while (*cc == OP_ALT);
      cc += 1 + LINK_SIZE;
      break;
-     /* Reached end of a branch; if it's a ket it is the end of a nested
+     /* Reached end of a branch; if it's a ket it is the end of a nested call.
-     call. If it's ALT it is an alternation in a nested call. If it is
+     If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
-     END it's the end of the outer call. All can be handled by the same code. */
+     an ALT. If it is END it's the end of the outer call. All can be handled by
+     the same code. Note that we must not include the OP_KETRxxx opcodes here,
+     because they all imply an unlimited repeat. */
      case OP_ALT:
      case OP_KET:
-     case OP_KETRMAX:
-     case OP_KETRMIN:
      case OP_END:
+     case OP_ACCEPT:
+     case OP_ASSERT_ACCEPT:
      if (length < 0) length = branchlength;
        else if (length != branchlength) return -1;
      if (*cc != OP_ALT) return length;
-Line 1535 
 for (;;)
+Line 1777 
 for (;;)
      case OP_RECURSE:
      if (!atend) return -3;
-     cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */
+     cs = ce = (pcre_uchar *)cd->start_code + GET(cc, 1);  /* Start subpattern */
-     do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */
+     do ce += GET(ce, 1); while (*ce == OP_ALT);           /* End subpattern */
-     if (cc > cs && cc < ce) return -1;                /* Recursion */
+     if (cc > cs && cc < ce) return -1;                    /* Recursion */
-     d = find_fixedlength(cs + 2, options, atend, cd);
+     d = find_fixedlength(cs + IMM2_SIZE, utf, atend, cd);
      if (d < 0) return d;
      branchlength += d;
      cc += 1 + LINK_SIZE;
-Line 1551 
 for (;;)
+Line 1793 
 for (;;)
      case OP_ASSERTBACK:
      case OP_ASSERTBACK_NOT:
      do cc += GET(cc, 1); while (*cc == OP_ALT);
-     /* Fall through */
+     cc += PRIV(OP_lengths)[*cc];
+     break;
      /* Skip over things that don't match chars */
-     case OP_REVERSE:
+     case OP_MARK:
+     case OP_PRUNE_ARG:
+     case OP_SKIP_ARG:
+     case OP_THEN_ARG:
+     cc += cc[1] + PRIV(OP_lengths)[*cc];
+     break;
+     case OP_CALLOUT:
+     case OP_CIRC:
+     case OP_CIRCM:
+     case OP_CLOSE:
+     case OP_COMMIT:
      case OP_CREF:
-     case OP_NCREF:
-     case OP_RREF:
-     case OP_NRREF:
      case OP_DEF:
-     case OP_OPT:
+     case OP_DOLL:
-     case OP_CALLOUT:
+     case OP_DOLLM:
-     case OP_SOD:
-     case OP_SOM:
-     case OP_SET_SOM:
      case OP_EOD:
      case OP_EODN:
-     case OP_CIRC:
+     case OP_FAIL:
-     case OP_DOLL:
+     case OP_NCREF:
+     case OP_NRREF:
      case OP_NOT_WORD_BOUNDARY:
+     case OP_PRUNE:
+     case OP_REVERSE:
+     case OP_RREF:
+     case OP_SET_SOM:
+     case OP_SKIP:
+     case OP_SOD:
+     case OP_SOM:
+     case OP_THEN:
      case OP_WORD_BOUNDARY:
-     cc += _pcre_OP_lengths[*cc];
+     cc += PRIV(OP_lengths)[*cc];
      break;
      /* Handle literal characters */
      case OP_CHAR:
-     case OP_CHARNC:
+     case OP_CHARI:
      case OP_NOT:
+     case OP_NOTI:
      branchlength++;
      cc += 2;
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
-     if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
+     if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
-       cc += _pcre_utf8_table4[cc[-1] & 0x3f];
  #endif
      break;
-Line 1592 
 for (;;)
+Line 1849 
 for (;;)
      need to skip over a multibyte character in UTF8 mode.  */
      case OP_EXACT:
+     case OP_EXACTI:
+     case OP_NOTEXACT:
+     case OP_NOTEXACTI:
      branchlength += GET2(cc,1);
-     cc += 4;
+     cc += 2 + IMM2_SIZE;
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
-     if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
+     if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
-       cc += _pcre_utf8_table4[cc[-1] & 0x3f];
  #endif
      break;
      case OP_TYPEEXACT:
      branchlength += GET2(cc,1);
-     if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
+     if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP)
-     cc += 4;
+       cc += 2;
+     cc += 1 + IMM2_SIZE + 1;
      break;
      /* Handle single-char matchers */
-Line 1613 
 for (;;)
+Line 1873 
 for (;;)
      cc += 2;
      /* Fall through */
+     case OP_HSPACE:
+     case OP_VSPACE:
+     case OP_NOT_HSPACE:
+     case OP_NOT_VSPACE:
      case OP_NOT_DIGIT:
      case OP_DIGIT:
      case OP_NOT_WHITESPACE:
-Line 1625 
 for (;;)
+Line 1889 
 for (;;)
      cc++;
      break;
-     /* The single-byte matcher isn't allowed */
+     /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
+     otherwise \C is coded as OP_ALLANY. */
      case OP_ANYBYTE:
      return -2;
      /* Check a class for variable quantification */
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32
      case OP_XCLASS:
-     cc += GET(cc, 1) - 33;
+     cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
      /* Fall through */
  #endif
      case OP_CLASS:
      case OP_NCLASS:
-     cc += 33;
+     cc += PRIV(OP_lengths)[OP_CLASS];
      switch (*cc)
        {
+       case OP_CRPLUS:
+       case OP_CRMINPLUS:
        case OP_CRSTAR:
        case OP_CRMINSTAR:
        case OP_CRQUERY:
-Line 1652 
 for (;;)
+Line 1919 
 for (;;)
        case OP_CRRANGE:
        case OP_CRMINRANGE:
-       if (GET2(cc,1) != GET2(cc,3)) return -1;
+       if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
        branchlength += GET2(cc,1);
-       cc += 5;
+       cc += 1 + 2 * IMM2_SIZE;
        break;
        default:
-Line 1664 
 for (;;)
+Line 1931 
 for (;;)
      /* Anything else is variable length */
-     default:
+     case OP_ANYNL:
+     case OP_BRAMINZERO:
+     case OP_BRAPOS:
+     case OP_BRAPOSZERO:
+     case OP_BRAZERO:
+     case OP_CBRAPOS:
+     case OP_EXTUNI:
+     case OP_KETRMAX:
+     case OP_KETRMIN:
+     case OP_KETRPOS:
+     case OP_MINPLUS:
+     case OP_MINPLUSI:
+     case OP_MINQUERY:
+     case OP_MINQUERYI:
+     case OP_MINSTAR:
+     case OP_MINSTARI:
+     case OP_MINUPTO:
+     case OP_MINUPTOI:
+     case OP_NOTMINPLUS:
+     case OP_NOTMINPLUSI:
+     case OP_NOTMINQUERY:
+     case OP_NOTMINQUERYI:
+     case OP_NOTMINSTAR:
+     case OP_NOTMINSTARI:
+     case OP_NOTMINUPTO:
+     case OP_NOTMINUPTOI:
+     case OP_NOTPLUS:
+     case OP_NOTPLUSI:
+     case OP_NOTPOSPLUS:
+     case OP_NOTPOSPLUSI:
+     case OP_NOTPOSQUERY:
+     case OP_NOTPOSQUERYI:
+     case OP_NOTPOSSTAR:
+     case OP_NOTPOSSTARI:
+     case OP_NOTPOSUPTO:
+     case OP_NOTPOSUPTOI:
+     case OP_NOTQUERY:
+     case OP_NOTQUERYI:
+     case OP_NOTSTAR:
+     case OP_NOTSTARI:
+     case OP_NOTUPTO:
+     case OP_NOTUPTOI:
+     case OP_PLUS:
+     case OP_PLUSI:
+     case OP_POSPLUS:
+     case OP_POSPLUSI:
+     case OP_POSQUERY:
+     case OP_POSQUERYI:
+     case OP_POSSTAR:
+     case OP_POSSTARI:
+     case OP_POSUPTO:
+     case OP_POSUPTOI:
+     case OP_QUERY:
+     case OP_QUERYI:
+     case OP_REF:
+     case OP_REFI:
+     case OP_SBRA:
+     case OP_SBRAPOS:
+     case OP_SCBRA:
+     case OP_SCBRAPOS:
+     case OP_SCOND:
+     case OP_SKIPZERO:
+     case OP_STAR:
+     case OP_STARI:
+     case OP_TYPEMINPLUS:
+     case OP_TYPEMINQUERY:
+     case OP_TYPEMINSTAR:
+     case OP_TYPEMINUPTO:
+     case OP_TYPEPLUS:
+     case OP_TYPEPOSPLUS:
+     case OP_TYPEPOSQUERY:
+     case OP_TYPEPOSSTAR:
+     case OP_TYPEPOSUPTO:
+     case OP_TYPEQUERY:
+     case OP_TYPESTAR:
+     case OP_TYPEUPTO:
+     case OP_UPTO:
+     case OP_UPTOI:
      return -1;
+     /* Catch unrecognized opcodes so that when new ones are added they
+     are not forgotten, as has happened in the past. */
+     default:
+     return -4;
      }
    }
  /* Control never gets here */
-Line 1686 
 length.
+Line 2036 
 length.
  Arguments:
    code        points to start of expression
-   utf8        TRUE in UTF-8 mode
+   utf         TRUE in UTF-8 / UTF-16 / UTF-32 mode
    number      the required bracket number or negative to find a lookbehind
  Returns:      pointer to the opcode for the bracket, or NULL if not found
  */
- const uschar *
+ const pcre_uchar *
- _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
+ PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
  {
  for (;;)
    {
    register int c = *code;
    if (c == OP_END) return NULL;
    /* XCLASS is used for classes that cannot be represented just by a bit
-Line 1710 
 for (;;)
+Line 2061 
 for (;;)
    else if (c == OP_REVERSE)
      {
-     if (number < 0) return (uschar *)code;
+     if (number < 0) return (pcre_uchar *)code;
-     code += _pcre_OP_lengths[c];
+     code += PRIV(OP_lengths)[c];
      }
    /* Handle capturing bracket */
-   else if (c == OP_CBRA)
+   else if (c == OP_CBRA || c == OP_SCBRA ||
+            c == OP_CBRAPOS || c == OP_SCBRAPOS)
      {
      int n = GET2(code, 1+LINK_SIZE);
-     if (n == number) return (uschar *)code;
+     if (n == number) return (pcre_uchar *)code;
-     code += _pcre_OP_lengths[c];
+     code += PRIV(OP_lengths)[c];
      }
    /* Otherwise, we can get the item's length from the table, except that for
-Line 1748 
 for (;;)
+Line 2100 
 for (;;)
        case OP_TYPEMINUPTO:
        case OP_TYPEEXACT:
        case OP_TYPEPOSUPTO:
-       if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
+       if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
+         code += 2;
        break;
        case OP_MARK:
-Line 1758 
 for (;;)
+Line 2111 
 for (;;)
        break;
        case OP_THEN_ARG:
-       code += code[1+LINK_SIZE];
+       code += code[1];
        break;
        }
      /* Add in the fixed length from the table */
-     code += _pcre_OP_lengths[c];
+     code += PRIV(OP_lengths)[c];
    /* In UTF-8 mode, opcodes that are followed by a character may be followed by
    a multi-byte character. The length in the table is a minimum, so we have to
    arrange to skip the extra bytes. */
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
-     if (utf8) switch(c)
+     if (utf) switch(c)
        {
        case OP_CHAR:
-       case OP_CHARNC:
+       case OP_CHARI:
        case OP_EXACT:
+       case OP_EXACTI:
        case OP_UPTO:
+       case OP_UPTOI:
        case OP_MINUPTO:
+       case OP_MINUPTOI:
        case OP_POSUPTO:
+       case OP_POSUPTOI:
        case OP_STAR:
+       case OP_STARI:
        case OP_MINSTAR:
+       case OP_MINSTARI:
        case OP_POSSTAR:
+       case OP_POSSTARI:
        case OP_PLUS:
+       case OP_PLUSI:
        case OP_MINPLUS:
+       case OP_MINPLUSI:
        case OP_POSPLUS:
+       case OP_POSPLUSI:
        case OP_QUERY:
+       case OP_QUERYI:
        case OP_MINQUERY:
+       case OP_MINQUERYI:
        case OP_POSQUERY:
-       if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
+       case OP_POSQUERYI:
+       if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
        break;
        }
  #else
-     (void)(utf8);  /* Keep compiler happy by referencing function argument */
+     (void)(utf);  /* Keep compiler happy by referencing function argument */
  #endif
      }
    }
-Line 1809 
 instance of OP_RECURSE.
+Line 2175 
 instance of OP_RECURSE.
  Arguments:
    code        points to start of expression
-   utf8        TRUE in UTF-8 mode
+   utf         TRUE in UTF-8 / UTF-16 / UTF-32 mode
  Returns:      pointer to the opcode for OP_RECURSE, or NULL if not found
  */
- static const uschar *
+ static const pcre_uchar *
- find_recurse(const uschar *code, BOOL utf8)
+ find_recurse(const pcre_uchar *code, BOOL utf)
  {
  for (;;)
    {
-Line 1854 
 for (;;)
+Line 2220 
 for (;;)
        case OP_TYPEUPTO:
        case OP_TYPEMINUPTO:
        case OP_TYPEEXACT:
-       if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
+       if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
+         code += 2;
        break;
        case OP_MARK:
-Line 1864 
 for (;;)
+Line 2231 
 for (;;)
        break;
        case OP_THEN_ARG:
-       code += code[1+LINK_SIZE];
+       code += code[1];
        break;
        }
      /* Add in the fixed length from the table */
-     code += _pcre_OP_lengths[c];
+     code += PRIV(OP_lengths)[c];
      /* In UTF-8 mode, opcodes that are followed by a character may be followed
      by a multi-byte character. The length in the table is a minimum, so we have
      to arrange to skip the extra bytes. */
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
-     if (utf8) switch(c)
+     if (utf) switch(c)
        {
        case OP_CHAR:
-       case OP_CHARNC:
+       case OP_CHARI:
+       case OP_NOT:
+       case OP_NOTI:
        case OP_EXACT:
+       case OP_EXACTI:
+       case OP_NOTEXACT:
+       case OP_NOTEXACTI:
        case OP_UPTO:
+       case OP_UPTOI:
+       case OP_NOTUPTO:
+       case OP_NOTUPTOI:
        case OP_MINUPTO:
+       case OP_MINUPTOI:
+       case OP_NOTMINUPTO:
+       case OP_NOTMINUPTOI:
        case OP_POSUPTO:
+       case OP_POSUPTOI:
+       case OP_NOTPOSUPTO:
+       case OP_NOTPOSUPTOI:
        case OP_STAR:
+       case OP_STARI:
+       case OP_NOTSTAR:
+       case OP_NOTSTARI:
        case OP_MINSTAR:
+       case OP_MINSTARI:
+       case OP_NOTMINSTAR:
+       case OP_NOTMINSTARI:
        case OP_POSSTAR:
+       case OP_POSSTARI:
+       case OP_NOTPOSSTAR:
+       case OP_NOTPOSSTARI:
        case OP_PLUS:
+       case OP_PLUSI:
+       case OP_NOTPLUS:
+       case OP_NOTPLUSI:
        case OP_MINPLUS:
+       case OP_MINPLUSI:
+       case OP_NOTMINPLUS:
+       case OP_NOTMINPLUSI:
        case OP_POSPLUS:
+       case OP_POSPLUSI:
+       case OP_NOTPOSPLUS:
+       case OP_NOTPOSPLUSI:
        case OP_QUERY:
+       case OP_QUERYI:
+       case OP_NOTQUERY:
+       case OP_NOTQUERYI:
        case OP_MINQUERY:
+       case OP_MINQUERYI:
+       case OP_NOTMINQUERY:
+       case OP_NOTMINQUERYI:
        case OP_POSQUERY:
-       if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
+       case OP_POSQUERYI:
+       case OP_NOTPOSQUERY:
+       case OP_NOTPOSQUERYI:
+       if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
        break;
        }
  #else
-     (void)(utf8);  /* Keep compiler happy by referencing function argument */
+     (void)(utf);  /* Keep compiler happy by referencing function argument */
  #endif
      }
    }
-Line 1921 
 bracket whose current branch will alread
+Line 2329 
 bracket whose current branch will alread
  Arguments:
    code        points to start of search
    endcode     points to where to stop
-   utf8        TRUE if in UTF8 mode
+   utf         TRUE if in UTF-8 / UTF-16 / UTF-32 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 pcre_uchar *code, const pcre_uchar *endcode,
-   compile_data *cd)
+   BOOL utf, compile_data *cd)
  {
  register int c;
- for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
+ for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
       code < endcode;
-      code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
+      code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
    {
-   const uschar *ccode;
+   const pcre_uchar *ccode;
    c = *code;
-Line 1950 
 for (code = first_significant_code(code
+Line 2358 
 for (code = first_significant_code(code
      continue;
      }
-   /* Groups with zero repeats can of course be empty; skip them. */
-   if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
-     {
-     code += _pcre_OP_lengths[c];
-     do code += GET(code, 1); while (*code == OP_ALT);
-     c = *code;
-     continue;
-     }
    /* For a recursion/subroutine call, if its end has been reached, which
-   implies a subroutine call, we can scan it. */
+   implies a backward reference subroutine call, we can scan it. If it's a
+   forward reference subroutine call, we can't. To detect forward reference
+   we have to scan up the list that is kept in the workspace. This function is
+   called only when doing the real compile, not during the pre-compile that
+   measures the size of the compiled pattern. */
    if (c == OP_RECURSE)
      {
-     BOOL empty_branch = FALSE;
+     const pcre_uchar *scode;
-     const uschar *scode = cd->start_code + GET(code, 1);
+     BOOL empty_branch;
+     /* Test for forward reference */
+     for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
+       if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
+     /* Not a forward reference, test for completed backward reference */
+     empty_branch = FALSE;
+     scode = cd->start_code + GET(code, 1);
      if (GET(scode, 1) == 0) return TRUE;    /* Unclosed */
+     /* Completed backwards reference */
      do
        {
-       if (could_be_empty_branch(scode, endcode, utf8, cd))
+       if (could_be_empty_branch(scode, endcode, utf, cd))
          {
          empty_branch = TRUE;
          break;
-Line 1978 
 for (code = first_significant_code(code
+Line 2393 
 for (code = first_significant_code(code
        scode += GET(scode, 1);
        }
      while (*scode == OP_ALT);
      if (!empty_branch) return FALSE;  /* All branches are non-empty */
      continue;
      }
+   /* Groups with zero repeats can of course be empty; skip them. */
+   if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
+       c == OP_BRAPOSZERO)
+     {
+     code += PRIV(OP_lengths)[c];
+     do code += GET(code, 1); while (*code == OP_ALT);
+     c = *code;
+     continue;
+     }
+   /* A nested group that is already marked as "could be empty" can just be
+   skipped. */
+   if (c == OP_SBRA  || c == OP_SBRAPOS ||
+       c == OP_SCBRA || c == OP_SCBRAPOS)
+     {
+     do code += GET(code, 1); while (*code == OP_ALT);
+     c = *code;
+     continue;
+     }
    /* For other groups, scan the branches. */
-   if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
+   if (c == OP_BRA  || c == OP_BRAPOS ||
+       c == OP_CBRA || c == OP_CBRAPOS ||
+       c == OP_ONCE || c == OP_ONCE_NC ||
+       c == OP_COND)
      {
      BOOL empty_branch;
      if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */
-Line 2000 
 for (code = first_significant_code(code
+Line 2441 
 for (code = first_significant_code(code
        empty_branch = FALSE;
        do
          {
-         if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
+         if (!empty_branch && could_be_empty_branch(code, endcode, utf, cd))
            empty_branch = TRUE;
          code += GET(code, 1);
          }
-Line 2018 
 for (code = first_significant_code(code
+Line 2459 
 for (code = first_significant_code(code
      {
      /* Check for quantifiers after a class. XCLASS is used for classes that
      cannot be represented just by a bit map. This includes negated single
-     high-valued characters. The length in _pcre_OP_lengths[] is zero; the
+     high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
      actual length is stored in the compiled code, so we must update "code"
      here. */
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
      case OP_XCLASS:
      ccode = code += GET(code, 1);
      goto CHECK_CLASS_REPEAT;
-Line 2030 
 for (code = first_significant_code(code
+Line 2471 
 for (code = first_significant_code(code
      case OP_CLASS:
      case OP_NCLASS:
-     ccode = code + 33;
+     ccode = code + PRIV(OP_lengths)[OP_CLASS];
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
      CHECK_CLASS_REPEAT:
  #endif
-Line 2071 
 for (code = first_significant_code(code
+Line 2512 
 for (code = first_significant_code(code
      case OP_ALLANY:
      case OP_ANYBYTE:
      case OP_CHAR:
-     case OP_CHARNC:
+     case OP_CHARI:
      case OP_NOT:
+     case OP_NOTI:
      case OP_PLUS:
      case OP_MINPLUS:
      case OP_POSPLUS:
-Line 2104 
 for (code = first_significant_code(code
+Line 2546 
 for (code = first_significant_code(code
      case OP_TYPEUPTO:
      case OP_TYPEMINUPTO:
      case OP_TYPEPOSUPTO:
-     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
+     if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == OP_NOTPROP)
+       code += 2;
      break;
      /* End of branch */
-Line 2112 
 for (code = first_significant_code(code
+Line 2555 
 for (code = first_significant_code(code
      case OP_KET:
      case OP_KETRMAX:
      case OP_KETRMIN:
+     case OP_KETRPOS:
      case OP_ALT:
      return TRUE;
      /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
      MINUPTO, and POSUPTO may be followed by a multibyte character */
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
      case OP_STAR:
+     case OP_STARI:
      case OP_MINSTAR:
+     case OP_MINSTARI:
      case OP_POSSTAR:
+     case OP_POSSTARI:
      case OP_QUERY:
+     case OP_QUERYI:
      case OP_MINQUERY:
+     case OP_MINQUERYI:
      case OP_POSQUERY:
-     if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
+     case OP_POSQUERYI:
+     if (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
      break;
      case OP_UPTO:
+     case OP_UPTOI:
      case OP_MINUPTO:
+     case OP_MINUPTOI:
      case OP_POSUPTO:
-     if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
+     case OP_POSUPTOI:
+     if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
      break;
  #endif
-Line 2145 
 for (code = first_significant_code(code
+Line 2598 
 for (code = first_significant_code(code
      break;
      case OP_THEN_ARG:
-     code += code[1+LINK_SIZE];
+     code += code[1];
      break;
      /* None of the remaining opcodes are required to match a character. */
-Line 2168 
 return TRUE;
+Line 2621 
 return TRUE;
  the current branch of the current pattern to see if it could match the empty
  string. If it could, we must look outwards for branches at other levels,
  stopping when we pass beyond the bracket which is the subject of the recursion.
+ This function is called only during the real compile, not during the
+ pre-compile.
  Arguments:
    code        points to start of the recursion
    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
+   utf         TRUE if in UTF-8 / UTF-16 / UTF-32 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,
+ could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
-   BOOL utf8, compile_data *cd)
+   branch_chain *bcptr, BOOL utf, compile_data *cd)
  {
  while (bcptr != NULL && bcptr->current_branch >= code)
    {
-   if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
+   if (!could_be_empty_branch(bcptr->current_branch, endcode, utf, cd))
      return FALSE;
    bcptr = bcptr->outer;
    }
-Line 2218 
 where Perl recognizes it as the POSIX cl
+Line 2673 
 where Perl recognizes it as the POSIX cl
  "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
  I think.
+ A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
+ It seems that the appearance of a nested POSIX class supersedes an apparent
+ external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
+ a digit.
+ In Perl, unescaped square brackets may also appear as part of class names. For
+ example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
+ [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
+ seem right at all. PCRE does not allow closing square brackets in POSIX class
+ names.
  Arguments:
    ptr      pointer to the initial [
    endptr   where to return the end pointer
-Line 2226 
 Returns:   TRUE or FALSE
+Line 2692 
 Returns:   TRUE or FALSE
  */
  static BOOL
- check_posix_syntax(const uschar *ptr, const uschar **endptr)
+ check_posix_syntax(const pcre_uchar *ptr, const pcre_uchar **endptr)
  {
  int terminator;          /* Don't combine these lines; the Solaris cc */
  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
  for (++ptr; *ptr != 0; ptr++)
    {
-   if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
+   if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
+     ptr++;
+   else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
+   else
      {
-     if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
      if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
        {
        *endptr = ptr;
        return TRUE;
        }
+     if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
+          (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
+           ptr[1] == CHAR_EQUALS_SIGN) &&
+         check_posix_syntax(ptr, endptr))
+       return FALSE;
      }
    }
  return FALSE;
-Line 2263 
 Returns:     a value representing the na
+Line 2736 
 Returns:     a value representing the na
  */
  static int
- check_posix_name(const uschar *ptr, int len)
+ check_posix_name(const pcre_uchar *ptr, int len)
  {
  const char *pn = posix_names;
  register int yield = 0;
  while (posix_name_lengths[yield] != 0)
    {
    if (len == posix_name_lengths[yield] &&
-     strncmp((const char *)ptr, pn, len) == 0) return yield;
+     STRNCMP_UC_C8(ptr, pn, len) == 0) return yield;
    pn += posix_name_lengths[yield] + 1;
    yield++;
    }
-Line 2302 
 value in the reference (which is a group
+Line 2775 
 value in the reference (which is a group
  Arguments:
    group      points to the start of the group
    adjust     the amount by which the group is to be moved
-   utf8       TRUE in UTF-8 mode
+   utf        TRUE in UTF-8 / UTF-16 / UTF-32 mode
    cd         contains pointers to tables etc.
    save_hwm   the hwm forward reference pointer at the start of the group
-Line 2310 
 Returns:     nothing
+Line 2783 
 Returns:     nothing
  */
  static void
- adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
+ adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
-   uschar *save_hwm)
+   pcre_uchar *save_hwm)
  {
- uschar *ptr = group;
+ pcre_uchar *ptr = group;
- while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
+ while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
    {
    int offset;
-   uschar *hc;
+   pcre_uchar *hc;
    /* See if this recursion is on the forward reference list. If so, adjust the
    reference. */
-Line 2363 
 Arguments:
+Line 2836 
 Arguments:
  Returns:         new code pointer
  */
- static uschar *
+ static pcre_uchar *
- auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
+ auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
  {
  *code++ = OP_CALLOUT;
  *code++ = 255;
  PUT(code, 0, (int)(ptr - cd->start_pattern));  /* Pattern offset */
  PUT(code, LINK_SIZE, 0);                       /* Default length */
- return code + 2*LINK_SIZE;
+ return code + 2 * LINK_SIZE;
  }
-Line 2392 
 Returns:             nothing
+Line 2865 
 Returns:             nothing
  */
  static void
- complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
+ complete_callout(pcre_uchar *previous_callout, const pcre_uchar *ptr, compile_data *cd)
  {
  int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
  PUT(previous_callout, 2 + LINK_SIZE, length);
-Line 2406 
 PUT(previous_callout, 2 + LINK_SIZE, len
+Line 2879 
 PUT(previous_callout, 2 + LINK_SIZE, len
  *************************************************/
  /* This function is passed the start and end of a class range, in UTF-8 mode
- with UCP support. It searches up the characters, looking for internal ranges of
+ with UCP support. It searches up the characters, looking for ranges of
  characters in the "other" case. Each call returns the next one, updating the
- start address.
+ start address. A character with multiple other cases is returned on its own
+ with a special return value.
  Arguments:
    cptr        points to starting character value; updated
-Line 2416 
 Arguments:
+Line 2890 
 Arguments:
    ocptr       where to put start of othercase range
    odptr       where to put end of othercase range
- Yield:        TRUE when range returned; FALSE when no more
+ Yield:        -1 when no more
+when a range is returned
+               >0 the CASESET offset for char with multiple other cases
+                 in this case, ocptr contains the original
  */
- static BOOL
+ static int
  get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
    unsigned int *odptr)
  {
  unsigned int c, othercase, next;
+ int co;
+ /* Find the first character that has an other case. If it has multiple other
+ cases, return its case offset value. */
  for (c = *cptr; c <= d; c++)
-   { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
+   {
+   if ((co = UCD_CASESET(c)) != 0)
+     {
+     *ocptr = c++;   /* Character that has the set */
+     *cptr = c;      /* Rest of input range */
+     return co;
+     }
+   if ((othercase = UCD_OTHERCASE(c)) != c) break;
+   }
- if (c > d) return FALSE;
+ if (c > d) return -1;  /* Reached end of range */
  *ocptr = othercase;
  next = othercase + 1;
-Line 2439 
 for (++c; c <= d; c++)
+Line 2928 
 for (++c; c <= d; c++)
    next++;
    }
- *odptr = next - 1;
+ *odptr = next - 1;     /* End of othercase range */
- *cptr = c;
+ *cptr = c;             /* Rest of input range */
+ return 0;
- return TRUE;
  }
-Line 2466 
 Returns:       TRUE if auto-possessifyin
+Line 2954 
 Returns:       TRUE if auto-possessifyin
  static BOOL
  check_char_prop(int c, int ptype, int pdata, BOOL negated)
  {
+ #ifdef SUPPORT_UCP
+ const pcre_uint32 *p;
+ #endif
  const ucd_record *prop = GET_UCD(c);
  switch(ptype)
    {
    case PT_LAMP:
-Line 2475 
 switch(ptype)
+Line 2968 
 switch(ptype)
            prop->chartype == ucp_Lt) == negated;
    case PT_GC:
-   return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
+   return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
    case PT_PC:
    return (pdata == prop->chartype) == negated;
-Line 2486 
 switch(ptype)
+Line 2979 
 switch(ptype)
    /* These are specials */
    case PT_ALNUM:
-   return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
+   return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
-           _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
+           PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
    case PT_SPACE:    /* Perl space */
-   return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
+   return (PRIV(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 ||
+   return (PRIV(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 ||
+   return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
-           _pcre_ucp_gentype[prop->chartype] == ucp_N ||
+           PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
            c == CHAR_UNDERSCORE) == negated;
+ #ifdef SUPPORT_UCP
+   case PT_CLIST:
+   p = PRIV(ucd_caseless_sets) + prop->caseset;
+   for (;;)
+     {
+     if ((unsigned int)c < *p) return !negated;
+     if ((unsigned int)c == *p++) return negated;
+     }
+   break;  /* Control never reaches here */
+ #endif
    }
  return FALSE;
  }
  #endif  /* SUPPORT_UCP */
-Line 2521 
 sense to automatically possessify the re
+Line 3026 
 sense to automatically possessify the re
  Arguments:
    previous      pointer to the repeated opcode
-   utf8          TRUE in UTF-8 mode
+   utf           TRUE in UTF-8 / UTF-16 / UTF-32 mode
    ptr           next character in pattern
    options       options bits
    cd            contains pointers to tables etc.
-Line 2530 
 Returns:        TRUE if possessifying is
+Line 3035 
 Returns:        TRUE if possessifying is
  */
  static BOOL
- check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
+ check_auto_possessive(const pcre_uchar *previous, BOOL utf,
-   int options, compile_data *cd)
+   const pcre_uchar *ptr, int options, compile_data *cd)
  {
- int c, next;
+ pcre_int32 c = NOTACHAR;
+ pcre_int32 next;
  int op_code = *previous++;
  /* Skip whitespace and comments in extended mode */
-Line 2542 
 if ((options & PCRE_EXTENDED) != 0)
+Line 3048 
 if ((options & PCRE_EXTENDED) != 0)
    {
    for (;;)
      {
-     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
+     while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
      if (*ptr == CHAR_NUMBER_SIGN)
        {
        ptr++;
-Line 2550 
 if ((options & PCRE_EXTENDED) != 0)
+Line 3056 
 if ((options & PCRE_EXTENDED) != 0)
          {
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
          ptr++;
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-         if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
+         if (utf) FORWARDCHAR(ptr);
  #endif
          }
        }
-Line 2569 
 if (*ptr == CHAR_BACKSLASH)
+Line 3075 
 if (*ptr == CHAR_BACKSLASH)
    if (temperrorcode != 0) return FALSE;
    ptr++;    /* Point after the escape sequence */
    }
+ else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
- else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
    {
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-   if (utf8) { GETCHARINC(next, ptr); } else
+   if (utf) { GETCHARINC(next, ptr); } else
  #endif
    next = *ptr++;
    }
  else return FALSE;
  /* Skip whitespace and comments in extended mode */
-Line 2586 
 if ((options & PCRE_EXTENDED) != 0)
+Line 3090 
 if ((options & PCRE_EXTENDED) != 0)
    {
    for (;;)
      {
-     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
+     while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
      if (*ptr == CHAR_NUMBER_SIGN)
        {
        ptr++;
-Line 2594 
 if ((options & PCRE_EXTENDED) != 0)
+Line 3098 
 if ((options & PCRE_EXTENDED) != 0)
          {
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
          ptr++;
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-         if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
+         if (utf) FORWARDCHAR(ptr);
  #endif
          }
        }
-Line 2606 
 if ((options & PCRE_EXTENDED) != 0)
+Line 3110 
 if ((options & PCRE_EXTENDED) != 0)
  /* If the next thing is itself optional, we have to give up. */
  if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
-   strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
+   STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
      return FALSE;
- /* Now compare the next item with the previous opcode. First, handle cases when
+ /* If the previous item is a character, get its value. */
- the next item is a character. */
- if (next >= 0) switch(op_code)
+ if (op_code == OP_CHAR || op_code == OP_CHARI ||
+     op_code == OP_NOT || op_code == OP_NOTI)
    {
-   case OP_CHAR:
+ #ifdef SUPPORT_UTF
- #ifdef SUPPORT_UTF8
    GETCHARTEST(c, previous);
  #else
    c = *previous;
  #endif
-   return c != next;
+   }
-   /* For CHARNC (caseless character) we must check the other case. If we have
+ /* Now compare the next item with the previous opcode. First, handle cases when
-   Unicode property support, we can use it to test the other case of
+ the next item is a character. */
-   high-valued characters. */
-   case OP_CHARNC:
+ if (next >= 0)
- #ifdef SUPPORT_UTF8
+   {
-   GETCHARTEST(c, previous);
+   /* For a caseless UTF match, the next character may have more than one other
- #else
+   case, which maps to the special PT_CLIST property. Check this first. */
-   c = *previous;
+ #ifdef SUPPORT_UCP
+   if (utf && (unsigned int)c != NOTACHAR && (options & PCRE_CASELESS) != 0)
+     {
+     int ocs = UCD_CASESET(next);
+     if (ocs > 0) return check_char_prop(c, PT_CLIST, ocs, op_code >= OP_NOT);
+     }
  #endif
-   if (c == next) return FALSE;
- #ifdef SUPPORT_UTF8
+   switch(op_code)
-   if (utf8)
      {
-     unsigned int othercase;
+     case OP_CHAR:
-     if (next < 128) othercase = cd->fcc[next]; else
+     return c != next;
+     /* For CHARI (caseless character) we must check the other case. If we have
+     Unicode property support, we can use it to test the other case of
+     high-valued characters. We know that next can have only one other case,
+     because multi-other-case characters are dealt with above. */
+     case OP_CHARI:
+     if (c == next) return FALSE;
+ #ifdef SUPPORT_UTF
+     if (utf)
+       {
+       unsigned int othercase;
+       if (next < 128) othercase = cd->fcc[next]; else
  #ifdef SUPPORT_UCP
-     othercase = UCD_OTHERCASE((unsigned int)next);
+       othercase = UCD_OTHERCASE((unsigned int)next);
  #else
-     othercase = NOTACHAR;
+       othercase = NOTACHAR;
  #endif
-     return (unsigned int)c != othercase;
+       return (unsigned int)c != othercase;
-     }
+       }
-   else
+     else
- #endif  /* SUPPORT_UTF8 */
+ #endif  /* SUPPORT_UTF */
-   return (c != cd->fcc[next]);  /* Non-UTF-8 mode */
+     return (c != TABLE_GET((unsigned int)next, cd->fcc, next));  /* Not UTF */
-   /* For OP_NOT, its data is always a single-byte character. */
+     case OP_NOT:
+     return c == next;
-   case OP_NOT:
-   if ((c = *previous) == next) return TRUE;
+     case OP_NOTI:
-   if ((options & PCRE_CASELESS) == 0) return FALSE;
+     if (c == next) return TRUE;
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-   if (utf8)
+     if (utf)
-     {
+       {
-     unsigned int othercase;
+       unsigned int othercase;
-     if (next < 128) othercase = cd->fcc[next]; else
+       if (next < 128) othercase = cd->fcc[next]; else
  #ifdef SUPPORT_UCP
-     othercase = UCD_OTHERCASE(next);
+       othercase = UCD_OTHERCASE((unsigned int)next);
  #else
-     othercase = NOTACHAR;
+       othercase = NOTACHAR;
  #endif
-     return (unsigned int)c == othercase;
+       return (unsigned int)c == othercase;
-     }
+       }
-   else
+     else
- #endif  /* SUPPORT_UTF8 */
+ #endif  /* SUPPORT_UTF */
-   return (c == cd->fcc[next]);  /* Non-UTF-8 mode */
+     return (c == TABLE_GET((unsigned int)next, cd->fcc, next));  /* Not UTF */
-   /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
+     /* 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. */
+     When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
-   case OP_DIGIT:
+     case OP_DIGIT:
-   return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
+     return next > 255 || (cd->ctypes[next] & ctype_digit) == 0;
-   case OP_NOT_DIGIT:
+     case OP_NOT_DIGIT:
-   return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
+     return next <= 255 && (cd->ctypes[next] & ctype_digit) != 0;
-   case OP_WHITESPACE:
+     case OP_WHITESPACE:
-   return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
+     return next > 255 || (cd->ctypes[next] & ctype_space) == 0;
-   case OP_NOT_WHITESPACE:
+     case OP_NOT_WHITESPACE:
-   return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
+     return next <= 255 && (cd->ctypes[next] & ctype_space) != 0;
-   case OP_WORDCHAR:
+     case OP_WORDCHAR:
-   return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
+     return next > 255 || (cd->ctypes[next] & ctype_word) == 0;
-   case OP_NOT_WORDCHAR:
+     case OP_NOT_WORDCHAR:
-   return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
+     return next <= 255 && (cd->ctypes[next] & ctype_word) != 0;
-   case OP_HSPACE:
+     case OP_HSPACE:
-   case OP_NOT_HSPACE:
+     case OP_NOT_HSPACE:
-   switch(next)
+     switch(next)
-     {
+       {
-     case 0x09:
+       HSPACE_CASES:
-     case 0x20:
+       return op_code == OP_NOT_HSPACE;
-     case 0xa0:
-     case 0x1680:
-     case 0x180e:
-     case 0x2000:
-     case 0x2001:
-     case 0x2002:
-     case 0x2003:
-     case 0x2004:
-     case 0x2005:
-     case 0x2006:
-     case 0x2007:
-     case 0x2008:
-     case 0x2009:
-     case 0x200A:
-     case 0x202f:
-     case 0x205f:
-     case 0x3000:
-     return op_code == OP_NOT_HSPACE;
-     default:
-     return op_code != OP_NOT_HSPACE;
-     }
-   case OP_ANYNL:
+       default:
-   case OP_VSPACE:
+       return op_code != OP_NOT_HSPACE;
-   case OP_NOT_VSPACE:
+       }
-   switch(next)
-     {
+     case OP_ANYNL:
-     case 0x0a:
+     case OP_VSPACE:
-     case 0x0b:
+     case OP_NOT_VSPACE:
-     case 0x0c:
+     switch(next)
-     case 0x0d:
+       {
-     case 0x85:
+       VSPACE_CASES:
-     case 0x2028:
+       return op_code == OP_NOT_VSPACE;
-     case 0x2029:
-     return op_code == OP_NOT_VSPACE;
+       default:
-     default:
+       return op_code != OP_NOT_VSPACE;
-     return op_code != OP_NOT_VSPACE;
+       }
-     }
  #ifdef SUPPORT_UCP
-   case OP_PROP:
+     case OP_PROP:
-   return check_char_prop(next, previous[0], previous[1], FALSE);
+     return check_char_prop(next, previous[0], previous[1], FALSE);
-   case OP_NOTPROP:
+     case OP_NOTPROP:
-   return check_char_prop(next, previous[0], previous[1], TRUE);
+     return check_char_prop(next, previous[0], previous[1], TRUE);
  #endif
-   default:
+     default:
-   return FALSE;
+     return FALSE;
+     }
    }
  /* 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
-Line 2758 
 replaced by OP_PROP codes when PCRE_UCP
+Line 3256 
 replaced by OP_PROP codes when PCRE_UCP
  switch(op_code)
    {
    case OP_CHAR:
-   case OP_CHARNC:
+   case OP_CHARI:
- #ifdef SUPPORT_UTF8
-   GETCHARTEST(c, previous);
- #else
-   c = *previous;
- #endif
    switch(-next)
      {
      case ESC_d:
-     return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
+     return c > 255 || (cd->ctypes[c] & ctype_digit) == 0;
      case ESC_D:
-     return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
+     return c <= 255 && (cd->ctypes[c] & ctype_digit) != 0;
      case ESC_s:
-     return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
+     return c > 255 || (cd->ctypes[c] & ctype_space) == 0;
      case ESC_S:
-     return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
+     return c <= 255 && (cd->ctypes[c] & ctype_space) != 0;
      case ESC_w:
-     return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
+     return c > 255 || (cd->ctypes[c] & ctype_word) == 0;
      case ESC_W:
-     return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
+     return c <= 255 && (cd->ctypes[c] & ctype_word) != 0;
      case ESC_h:
      case ESC_H:
      switch(c)
        {
-       case 0x09:
+       HSPACE_CASES:
-       case 0x20:
-       case 0xa0:
-       case 0x1680:
-       case 0x180e:
-       case 0x2000:
-       case 0x2001:
-       case 0x2002:
-       case 0x2003:
-       case 0x2004:
-       case 0x2005:
-       case 0x2006:
-       case 0x2007:
-       case 0x2008:
-       case 0x2009:
-       case 0x200A:
-       case 0x202f:
-       case 0x205f:
-       case 0x3000:
        return -next != ESC_h;
        default:
        return -next == ESC_h;
        }
-Line 2816 
 switch(op_code)
+Line 3292 
 switch(op_code)
      case ESC_V:
      switch(c)
        {
-       case 0x0a:
+       VSPACE_CASES:
-       case 0x0b:
-       case 0x0c:
-       case 0x0d:
-       case 0x85:
-       case 0x2028:
-       case 0x2029:
        return -next != ESC_v;
        default:
        return -next == ESC_v;
        }
-Line 2864 
 switch(op_code)
+Line 3335 
 switch(op_code)
        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)
+         STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
            return FALSE;
        /* Do the property check. */
-Line 2891 
 switch(op_code)
+Line 3362 
 switch(op_code)
    return next == -ESC_d;
    case OP_WHITESPACE:
-   return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
+   return next == -ESC_S || next == -ESC_d || next == -ESC_w;
    case OP_NOT_WHITESPACE:
-   return next == -ESC_s || next == -ESC_h || next == -ESC_v;
+   return next == -ESC_s || next == -ESC_h || next == -ESC_v || next == -ESC_R;
    case OP_HSPACE:
    return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
-Line 2928 
 switch(op_code)
+Line 3399 
 switch(op_code)
  /*************************************************
+ *        Add a character or range to a class     *
+ *************************************************/
+ /* This function packages up the logic of adding a character or range of
+ characters to a class. The character values in the arguments will be within the
+ valid values for the current mode (8-bit, 16-bit, UTF, etc). This function is
+ mutually recursive with the function immediately below.
+ Arguments:
+   classbits     the bit map for characters < 256
+   uchardptr     points to the pointer for extra data
+   options       the options word
+   cd            contains pointers to tables etc.
+   start         start of range character
+   end           end of range character
+ Returns:        the number of < 256 characters added
+                 the pointer to extra data is updated
+ */
+ static int
+ add_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
+   compile_data *cd, unsigned int start, unsigned int end)
+ {
+ unsigned int c;
+ int n8 = 0;
+ /* If caseless matching is required, scan the range and process alternate
+ cases. In Unicode, there are 8-bit characters that have alternate cases that
+ are greater than 255 and vice-versa. Sometimes we can just extend the original
+ range. */
+ if ((options & PCRE_CASELESS) != 0)
+   {
+ #ifdef SUPPORT_UCP
+   if ((options & PCRE_UTF8) != 0)
+     {
+     int rc;
+     unsigned int oc, od;
+     options &= ~PCRE_CASELESS;   /* Remove for recursive calls */
+     c = start;
+     while ((rc = get_othercase_range(&c, end, &oc, &od)) >= 0)
+       {
+       /* Handle a single character that has more than one other case. */
+       if (rc > 0) n8 += add_list_to_class(classbits, uchardptr, options, cd,
+         PRIV(ucd_caseless_sets) + rc, oc);
+       /* Do nothing if the other case range is within the original range. */
+       else if (oc >= start && od <= end) continue;
+       /* Extend the original range if there is overlap, noting that if oc < c, we
+       can't have od > end because a subrange is always shorter than the basic
+       range. Otherwise, use a recursive call to add the additional range. */
+       else if (oc < start && od >= start - 1) start = oc; /* Extend downwards */
+       else if (od > end && oc <= end + 1) end = od;       /* Extend upwards */
+       else n8 += add_to_class(classbits, uchardptr, options, cd, oc, od);
+       }
+     }
+   else
+ #endif  /* SUPPORT_UCP */
+   /* Not UTF-mode, or no UCP */
+   for (c = start; c <= end && c < 256; c++)
+     {
+     SETBIT(classbits, cd->fcc[c]);
+     n8++;
+     }
+   }
+ /* Now handle the original range. Adjust the final value according to the bit
+ length - this means that the same lists of (e.g.) horizontal spaces can be used
+ in all cases. */
+ #if defined COMPILE_PCRE8
+ #ifdef SUPPORT_UTF
+   if ((options & PCRE_UTF8) == 0)
+ #endif
+   if (end > 0xff) end = 0xff;
+ #elif defined COMPILE_PCRE16
+ #ifdef SUPPORT_UTF
+   if ((options & PCRE_UTF16) == 0)
+ #endif
+   if (end > 0xffff) end = 0xffff;
+ #elif defined COMPILE_PCRE32
+ #ifdef SUPPORT_UTF
+   if ((options & PCRE_UTF32) == 0)
+     if (end > 0xffffu) end = 0xffffu; // FIXMEchpe rebase fix this
+ #endif
+ #endif /* COMPILE_PCRE[8|16|32] */
+ /* If all characters are less than 256, use the bit map. Otherwise use extra
+ data. */
+ if (end < 0x100)
+   {
+   for (c = start; c <= end; c++)
+     {
+     n8++;
+     SETBIT(classbits, c);
+     }
+   }
+ else
+   {
+   pcre_uchar *uchardata = *uchardptr;
+ #ifdef SUPPORT_UTF
+   if ((options & PCRE_UTF8) != 0)  /* All UTFs use the same flag bit */
+     {
+     if (start < end)
+       {
+       *uchardata++ = XCL_RANGE;
+       uchardata += PRIV(ord2utf)(start, uchardata);
+       uchardata += PRIV(ord2utf)(end, uchardata);
+       }
+     else if (start == end)
+       {
+       *uchardata++ = XCL_SINGLE;
+       uchardata += PRIV(ord2utf)(start, uchardata);
+       }
+     }
+   else
+ #endif  /* SUPPORT_UTF */
+   /* Without UTF support, character values are constrained by the bit length,
+   and can only be > 256 for 16-bit and 32-bit libraries. */
+ #ifdef COMPILE_PCRE8
+     {}
+ #else
+   if (start < end)
+     {
+     *uchardata++ = XCL_RANGE;
+     *uchardata++ = start;
+     *uchardata++ = end;
+     }
+   else if (start == end)
+     {
+     *uchardata++ = XCL_SINGLE;
+     *uchardata++ = start;
+     }
+ #endif
+   *uchardptr = uchardata;   /* Updata extra data pointer */
+   }
+ return n8;    /* Number of 8-bit characters */
+ }
+ /*************************************************
+ *        Add a list of characters to a class     *
+ *************************************************/
+ /* This function is used for adding a list of case-equivalent characters to a
+ class, and also for adding a list of horizontal or vertical whitespace. If the
+ list is in order (which it should be), ranges of characters are detected and
+ handled appropriately. This function is mutually recursive with the function
+ above.
+ Arguments:
+   classbits     the bit map for characters < 256
+   uchardptr     points to the pointer for extra data
+   options       the options word
+   cd            contains pointers to tables etc.
+   p             points to row of 32-bit values, terminated by NOTACHAR
+   except        character to omit; this is used when adding lists of
+                   case-equivalent characters to avoid including the one we
+                   already know about
+ Returns:        the number of < 256 characters added
+                 the pointer to extra data is updated
+ */
+ static int
+ add_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
+   compile_data *cd, const pcre_uint32 *p, unsigned int except)
+ {
+ int n8 = 0;
+ while (p[0] < NOTACHAR)
+   {
+   int n = 0;
+   if (p[0] != except)
+     {
+     while(p[n+1] == p[0] + n + 1) n++;
+     n8 += add_to_class(classbits, uchardptr, options, cd, p[0], p[n]);
+     }
+   p += n + 1;
+   }
+ return n8;
+ }
+ /*************************************************
+ *    Add characters not in a list to a class     *
+ *************************************************/
+ /* This function is used for adding the complement of a list of horizontal or
+ vertical whitespace to a class. The list must be in order.
+ Arguments:
+   classbits     the bit map for characters < 256
+   uchardptr     points to the pointer for extra data
+   options       the options word
+   cd            contains pointers to tables etc.
+   p             points to row of 32-bit values, terminated by NOTACHAR
+ Returns:        the number of < 256 characters added
+                 the pointer to extra data is updated
+ */
+ static int
+ add_not_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr,
+   int options, compile_data *cd, const pcre_uint32 *p)
+ {
+ int n8 = 0;
+ if (p[0] > 0)
+   n8 += add_to_class(classbits, uchardptr, options, cd, 0, p[0] - 1);
+ while (p[0] < NOTACHAR)
+   {
+   while (p[1] == p[0] + 1) p++;
+   n8 += add_to_class(classbits, uchardptr, options, cd, p[0] + 1,
+     (p[1] == NOTACHAR)? 0x10ffff : p[1] - 1);
+   p++;
+   }
+ return n8;
+ }
+ /*************************************************
  *           Compile one branch                   *
  *************************************************/
-Line 2942 
 Arguments:
+Line 3655 
 Arguments:
    codeptr        points to the pointer to the current code point
    ptrptr         points to the current pattern pointer
    errorcodeptr   points to error code variable
-   firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
+   firstcharptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
-   reqbyteptr     set to the last literal character required, else < 0
+   reqcharptr     set to the last literal character required, else < 0
    bcptr          points to current branch chain
+   cond_depth     conditional nesting depth
    cd             contains pointers to tables etc.
    lengthptr      NULL during the real compile phase
                   points to length accumulator during pre-compile phase
-Line 2954 
 Returns:         TRUE on success
+Line 3668 
 Returns:         TRUE on success
  */
  static BOOL
- compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
+ compile_branch(int *optionsptr, pcre_uchar **codeptr,
-   int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
+   const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
+   pcre_int32 *reqcharptr, branch_chain *bcptr, int cond_depth,
    compile_data *cd, int *lengthptr)
  {
  int repeat_type, op_type;
  int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */
  int bravalue = 0;
  int greedy_default, greedy_non_default;
- int firstbyte, reqbyte;
+ pcre_int32 firstchar, reqchar;
- int zeroreqbyte, zerofirstbyte;
+ pcre_int32 zeroreqchar, zerofirstchar;
- int req_caseopt, reqvary, tempreqvary;
+ pcre_int32 req_caseopt, reqvary, tempreqvary;
- int options = *optionsptr;
+ int options = *optionsptr;               /* May change dynamically */
  int after_manual_callout = 0;
  int length_prevgroup = 0;
  register int c;
- register uschar *code = *codeptr;
+ register pcre_uchar *code = *codeptr;
- uschar *last_code = code;
+ pcre_uchar *last_code = code;
- uschar *orig_code = code;
+ pcre_uchar *orig_code = code;
- uschar *tempcode;
+ pcre_uchar *tempcode;
  BOOL inescq = FALSE;
- BOOL groupsetfirstbyte = FALSE;
+ BOOL groupsetfirstchar = FALSE;
- const uschar *ptr = *ptrptr;
+ const pcre_uchar *ptr = *ptrptr;
- const uschar *tempptr;
+ const pcre_uchar *tempptr;
- const uschar *nestptr = NULL;
+ const pcre_uchar *nestptr = NULL;
- uschar *previous = NULL;
+ pcre_uchar *previous = NULL;
- uschar *previous_callout = NULL;
+ pcre_uchar *previous_callout = NULL;
- uschar *save_hwm = NULL;
+ pcre_uchar *save_hwm = NULL;
- uschar classbits[32];
+ pcre_uint8 classbits[32];
- #ifdef SUPPORT_UTF8
+ /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
- BOOL class_utf8;
+ must not do this for other options (e.g. PCRE_EXTENDED) because they may change
- BOOL utf8 = (options & PCRE_UTF8) != 0;
+ dynamically as we process the pattern. */
- uschar *class_utf8data;
- uschar *class_utf8data_base;
+ #ifdef SUPPORT_UTF
- uschar utf8_char[6];
+ /* PCRE_UTF(16|32) have the same value as PCRE_UTF8. */
+ BOOL utf = (options & PCRE_UTF8) != 0;
+ pcre_uchar utf_chars[6];
  #else
- BOOL utf8 = FALSE;
+ BOOL utf = FALSE;
- uschar *utf8_char = NULL;
+ #endif
+ /* Helper variables for OP_XCLASS opcode (for characters > 255). We define
+ class_uchardata always so that it can be passed to add_to_class() always,
+ though it will not be used in non-UTF 8-bit cases. This avoids having to supply
+ alternative calls for the different cases. */
+ pcre_uchar *class_uchardata;
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
+ BOOL xclass;
+ pcre_uchar *class_uchardata_base;
  #endif
  #ifdef PCRE_DEBUG
-Line 3005 
 greedy_non_default = greedy_default ^ 1;
+Line 3732 
 greedy_non_default = greedy_default ^ 1;
  /* Initialize no first byte, no required byte. REQ_UNSET means "no char
  matching encountered yet". It gets changed to REQ_NONE if we hit something that
- matches a non-fixed char first char; reqbyte just remains unset if we never
+ matches a non-fixed char first char; reqchar just remains unset if we never
  find one.
  When we hit a repeat whose minimum is zero, we may have to adjust these values
  to take the zero repeat into account. This is implemented by setting them to
- zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
+ zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
  item types that can be repeated set these backoff variables appropriately. */
- firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
+ firstchar = reqchar = zerofirstchar = zeroreqchar = REQ_UNSET;
- /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
+ /* The variable req_caseopt contains either the REQ_CASELESS value
- according to the current setting of the caseless flag. REQ_CASELESS is a bit
+ or zero, according to the current setting of the caseless flag. The
- value > 255. It is added into the firstbyte or reqbyte variables to record the
+ REQ_CASELESS leaves the lower 28 bit empty. It is added into the
- case status of the value. This is used only for ASCII characters. */
+ firstchar or reqchar variables to record the case status of the
+ value. This is used only for ASCII characters. */
- req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
+ req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
  /* Switch on next character until the end of the branch */
-Line 3032 
 for (;; ptr++)
+Line 3760 
 for (;; ptr++)
    BOOL is_quantifier;
    BOOL is_recurse;
    BOOL reset_bracount;
-   int class_charcount;
+   int class_has_8bitchar;
-   int class_lastchar;
+   int class_one_char;
    int newoptions;
    int recno;
    int refsign;
    int skipbytes;
-   int subreqbyte;
+   int subreqchar;
-   int subfirstbyte;
+   int subfirstchar;
    int terminator;
    int mclength;
-   uschar mcbuffer[8];
+   int tempbracount;
+   pcre_uchar mcbuffer[8];
-   /* Get next byte in the pattern */
+   /* Get next character in the pattern */
    c = *ptr;
-Line 3066 
 for (;; ptr++)
+Line 3795 
 for (;; ptr++)
  #ifdef PCRE_DEBUG
      if (code > cd->hwm) cd->hwm = code;                 /* High water info */
  #endif
-     if (code > cd->start_workspace + WORK_SIZE_CHECK)   /* Check for overrun */
+     if (code > cd->start_workspace + cd->workspace_size -
+         WORK_SIZE_SAFETY_MARGIN)                       /* Check for overrun */
        {
        *errorcodeptr = ERR52;
        goto FAILED;
-Line 3089 
 for (;; ptr++)
+Line 3819 
 for (;; ptr++)
        }
      *lengthptr += (int)(code - last_code);
-     DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
+     DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
+       (int)(code - last_code), c, c));
      /* If "previous" is set and it is not at the start of the work space, move
      it back to there, in order to avoid filling up the work space. Otherwise,
-Line 3099 
 for (;; ptr++)
+Line 3830 
 for (;; ptr++)
        {
        if (previous > orig_code)
          {
-         memmove(orig_code, previous, code - previous);
+         memmove(orig_code, previous, IN_UCHARS(code - previous));
          code -= previous - orig_code;
          previous = orig_code;
          }
-Line 3115 
 for (;; ptr++)
+Line 3846 
 for (;; ptr++)
    /* In the real compile phase, just check the workspace used by the forward
    reference list. */
-   else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
+   else if (cd->hwm > cd->start_workspace + cd->workspace_size -
+            WORK_SIZE_SAFETY_MARGIN)
      {
      *errorcodeptr = ERR52;
      goto FAILED;
-Line 3163 
 for (;; ptr++)
+Line 3895 
 for (;; ptr++)
      previous_callout = NULL;
      }
-   /* In extended mode, skip white space and comments */
+   /* In extended mode, skip white space and comments. */
    if ((options & PCRE_EXTENDED) != 0)
      {
-     if ((cd->ctypes[c] & ctype_space) != 0) continue;
+     if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
      if (c == CHAR_NUMBER_SIGN)
        {
        ptr++;
-Line 3175 
 for (;; ptr++)
+Line 3907 
 for (;; ptr++)
          {
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
          ptr++;
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-         if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
+         if (utf) FORWARDCHAR(ptr);
  #endif
          }
        if (*ptr != 0) continue;
-Line 3200 
 for (;; ptr++)
+Line 3932 
 for (;; ptr++)
      case 0:                        /* The branch terminates at string end */
      case CHAR_VERTICAL_LINE:       /* or | or ) */
      case CHAR_RIGHT_PARENTHESIS:
-     *firstbyteptr = firstbyte;
+     *firstcharptr = firstchar;
-     *reqbyteptr = reqbyte;
+     *reqcharptr = reqchar;
      *codeptr = code;
      *ptrptr = ptr;
      if (lengthptr != NULL)
-Line 3222 
 for (;; ptr++)
+Line 3954 
 for (;; ptr++)
      the setting of any following char as a first character. */
      case CHAR_CIRCUMFLEX_ACCENT:
+     previous = NULL;
      if ((options & PCRE_MULTILINE) != 0)
        {
-       if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
+       if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
+       *code++ = OP_CIRCM;
        }
-     previous = NULL;
+     else *code++ = OP_CIRC;
-     *code++ = OP_CIRC;
      break;
      case CHAR_DOLLAR_SIGN:
      previous = NULL;
-     *code++ = OP_DOLL;
+     *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
      break;
      /* There can never be a first char if '.' is first, whatever happens about
-     repeats. The value of reqbyte doesn't change either. */
+     repeats. The value of reqchar doesn't change either. */
      case CHAR_DOT:
-     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
+     if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
-     zerofirstbyte = firstbyte;
+     zerofirstchar = firstchar;
-     zeroreqbyte = reqbyte;
+     zeroreqchar = reqchar;
      previous = code;
      *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
      break;
-Line 3296 
 for (;; ptr++)
+Line 4029 
 for (;; ptr++)
          {
          if (ptr[1] == CHAR_E)
            ptr++;
-         else if (strncmp((const char *)ptr+1,
+         else if (STRNCMP_UC_C8(ptr + 1, STR_Q STR_BACKSLASH STR_E, 3) == 0)
-                           STR_Q STR_BACKSLASH STR_E, 3) == 0)
            ptr += 3;
          else
            break;
-Line 3316 
 for (;; ptr++)
+Line 4048 
 for (;; ptr++)
          (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
        {
        *code++ = negate_class? OP_ALLANY : OP_FAIL;
-       if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
+       if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
-       zerofirstbyte = firstbyte;
+       zerofirstchar = firstchar;
        break;
        }
-Line 3327 
 for (;; ptr++)
+Line 4059 
 for (;; ptr++)
      should_flip_negation = FALSE;
-     /* Keep a count of chars with values < 256 so that we can optimize the case
+     /* For optimization purposes, we track some properties of the class:
-     of just a single character (as long as it's < 256). However, For higher
+     class_has_8bitchar will be non-zero if the class contains at least one <
-     valued UTF-8 characters, we don't yet do any optimization. */
+character; class_one_char will be 1 if the class contains just one
+     character. */
-     class_charcount = 0;
+     class_has_8bitchar = 0;
-     class_lastchar = -1;
+     class_one_char = 0;
      /* Initialize the 32-char bit map to all zeros. We build the map in a
-     temporary bit of memory, in case the class contains only 1 character (less
+     temporary bit of memory, in case the class contains fewer than two
-     than 256), because in that case the compiled code doesn't use the bit map.
+-bit characters because in that case the compiled code doesn't use the bit
-     */
+     map. */
-     memset(classbits, 0, 32 * sizeof(uschar));
+     memset(classbits, 0, 32 * sizeof(pcre_uint8));
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
-     class_utf8 = FALSE;                       /* No chars >= 256 */
+     xclass = FALSE;
-     class_utf8data = code + LINK_SIZE + 2;    /* For UTF-8 items */
+     class_uchardata = code + LINK_SIZE + 2;   /* For XCLASS items */
-     class_utf8data_base = class_utf8data;     /* For resetting in pass 1 */
+     class_uchardata_base = class_uchardata;   /* Save the start */
  #endif
      /* Process characters until ] is reached. By writing this as a "do" it
-Line 3353 
 for (;; ptr++)
+Line 4086 
 for (;; ptr++)
      if (c != 0) do
        {
-       const uschar *oldptr;
+       const pcre_uchar *oldptr;
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
-       if (utf8 && c > 127)
+       if (utf && HAS_EXTRALEN(c))
          {                           /* Braces are required because the */
          GETCHARLEN(c, ptr, ptr);    /* macro generates multiple statements */
          }
+ #endif
-       /* In the pre-compile phase, accumulate the length of any UTF-8 extra
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
+       /* In the pre-compile phase, accumulate the length of any 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
+       contain a zillion > 255 characters no longer overwrite the work space
-       (which is on the stack). */
+       (which is on the stack). We have to remember that there was XCLASS data,
+       however. */
-       if (lengthptr != NULL)
-         {
+       if (lengthptr != NULL && class_uchardata > class_uchardata_base)
-         *lengthptr += class_utf8data - class_utf8data_base;
+         {
-         class_utf8data = class_utf8data_base;
+         xclass = TRUE;
+         *lengthptr += class_uchardata - class_uchardata_base;
+         class_uchardata = class_uchardata_base;
          }
  #endif
        /* Inside \Q...\E everything is literal except \E */
-Line 3399 
 for (;; ptr++)
+Line 4135 
 for (;; ptr++)
          {
          BOOL local_negate = FALSE;
          int posix_class, taboffset, tabopt;
-         register const uschar *cbits = cd->cbits;
+         register const pcre_uint8 *cbits = cd->cbits;
-         uschar pbits[32];
+         pcre_uint8 pbits[32];
          if (ptr[1] != CHAR_COLON)
            {
-Line 3427 
 for (;; ptr++)
+Line 4163 
 for (;; ptr++)
          alpha. This relies on the fact that the class table starts with
          alpha, lower, upper as the first 3 entries. */
          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
            posix_class = 0;
          /* When PCRE_UCP is set, some of the POSIX classes are converted to
-Line 3455 
 for (;; ptr++)
+Line 4191 
 for (;; ptr++)
          /* Copy in the first table (always present) */
          memcpy(pbits, cbits + posix_class_maps[posix_class],
-* sizeof(uschar));
+* sizeof(pcre_uint8));
          /* If there is a second table, add or remove it as required. */
-Line 3470 
 for (;; ptr++)
+Line 4206 
 for (;; ptr++)
              for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
            }
-         /* Not see if we need to remove any special characters. An option
+         /* Now see if we need to remove any special characters. An option
          value of 1 removes vertical space and 2 removes underscore. */
          if (tabopt < 0) tabopt = -tabopt;
-Line 3486 
 for (;; ptr++)
+Line 4222 
 for (;; ptr++)
            for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
          ptr = tempptr + 1;
-         class_charcount = 10;  /* Set > 1; assumes more than 1 per class */
+         /* Every class contains at least one < 256 character. */
+         class_has_8bitchar = 1;
+         /* Every class contains at least two characters. */
+         class_one_char = 2;
          continue;    /* End of POSIX syntax handling */
          }
        /* 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. We
-       assume that other escapes have more than one character in them, so set
+       assume that other escapes have more than one character in them, so
-       class_charcount bigger than one. Unrecognized escapes fall through and
+       speculatively set both class_has_8bitchar and class_one_char bigger
-       are either treated as literal characters (by default), or are faulted if
+       than one. Unrecognized escapes fall through and are either treated
+       as literal characters (by default), or are faulted if
        PCRE_EXTRA is set. */
        if (c == CHAR_BACKSLASH)
-Line 3504 
 for (;; ptr++)
+Line 4244 
 for (;; ptr++)
          if (*errorcodeptr != 0) goto FAILED;
          if (-c == ESC_b) c = CHAR_BS;    /* \b is backspace in a class */
+         else if (-c == ESC_N)            /* \N is not supported in a class */
+           {
+           *errorcodeptr = ERR71;
+           goto FAILED;
+           }
          else if (-c == ESC_Q)            /* Handle start of quoted string */
            {
            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
-Line 3517 
 for (;; ptr++)
+Line 4262 
 for (;; ptr++)
          if (c < 0)
            {
-           register const uschar *cbits = cd->cbits;
+           register const pcre_uint8 *cbits = cd->cbits;
-           class_charcount += 2;     /* Greater than 1 is what matters */
+           /* Every class contains at least two < 256 characters. */
+           class_has_8bitchar++;
+           /* Every class contains at least two characters. */
+           class_one_char += 2;
            switch (-c)
              {
-Line 3531 
 for (;; ptr++)
+Line 4279 
 for (;; ptr++)
              case ESC_SU:
              nestptr = ptr;
              ptr = substitutes[-c - ESC_DU] - 1;  /* Just before substitute */
-             class_charcount -= 2;                /* Undo! */
+             class_has_8bitchar--;                /* Undo! */
              continue;
  #endif
              case ESC_d:
-Line 3554 
 for (;; ptr++)
+Line 4302 
 for (;; ptr++)
              /* Perl 5.004 onwards omits VT from \s, but we must preserve it
              if it was previously set by something earlier in the character
-             class. */
+             class. Luckily, the value of CHAR_VT is 0x0b in both ASCII and
+             EBCDIC, so we lazily just adjust the appropriate bit. */
              case ESC_s:
              classbits[0] |= cbits[cbit_space];
-Line 3567 
 for (;; ptr++)
+Line 4316 
 for (;; ptr++)
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */
              continue;
+             /* The rest apply in both UCP and non-UCP cases. */
              case ESC_h:
-             SETBIT(classbits, 0x09); /* VT */
+             (void)add_list_to_class(classbits, &class_uchardata, options, cd,
-             SETBIT(classbits, 0x20); /* SPACE */
+               PRIV(hspace_list), NOTACHAR);
-             SETBIT(classbits, 0xa0); /* NSBP */
- #ifdef SUPPORT_UTF8
-             if (utf8)
-               {
-               class_utf8 = TRUE;
-               *class_utf8data++ = XCL_SINGLE;
-               class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
-               *class_utf8data++ = XCL_SINGLE;
-               class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
-               *class_utf8data++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
-               class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
-               *class_utf8data++ = XCL_SINGLE;
-               class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
-               *class_utf8data++ = XCL_SINGLE;
-               class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
-               *class_utf8data++ = XCL_SINGLE;
-               class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
-               }
- #endif
              continue;
              case ESC_H:
-             for (c = 0; c < 32; c++)
+             (void)add_not_list_to_class(classbits, &class_uchardata, options,
-               {
+               cd, PRIV(hspace_list));
-               int x = 0xff;
-               switch (c)
-                 {
-                 case 0x09/8: x ^= 1 << (0x09%8); break;
-                 case 0x20/8: x ^= 1 << (0x20%8); break;
-                 case 0xa0/8: x ^= 1 << (0xa0%8); break;
-                 default: break;
-                 }
-               classbits[c] |= x;
-               }
- #ifdef SUPPORT_UTF8
-             if (utf8)
-               {
-               class_utf8 = TRUE;
-               *class_utf8data++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
-               class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
-               *class_utf8data++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
-               class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
-               *class_utf8data++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
-               class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
-               *class_utf8data++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
-               class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
-               *class_utf8data++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
-               class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
-               *class_utf8data++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
-               class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
-               *class_utf8data++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
-               class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
-               }
- #endif
              continue;
              case ESC_v:
-             SETBIT(classbits, 0x0a); /* LF */
+             (void)add_list_to_class(classbits, &class_uchardata, options, cd,
-             SETBIT(classbits, 0x0b); /* VT */
+               PRIV(vspace_list), NOTACHAR);
-             SETBIT(classbits, 0x0c); /* FF */
-             SETBIT(classbits, 0x0d); /* CR */
-             SETBIT(classbits, 0x85); /* NEL */
- #ifdef SUPPORT_UTF8
-             if (utf8)
-               {
-               class_utf8 = TRUE;
-               *class_utf8data++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
-               class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
-               }
- #endif
              continue;
              case ESC_V:
-             for (c = 0; c < 32; c++)
+             (void)add_not_list_to_class(classbits, &class_uchardata, options,
-               {
+               cd, PRIV(vspace_list));
-               int x = 0xff;
-               switch (c)
-                 {
-                 case 0x0a/8: x ^= 1 << (0x0a%8);
-                              x ^= 1 << (0x0b%8);
-                              x ^= 1 << (0x0c%8);
-                              x ^= 1 << (0x0d%8);
-                              break;
-                 case 0x85/8: x ^= 1 << (0x85%8); break;
-                 default: break;
-                 }
-               classbits[c] |= x;
-               }
- #ifdef SUPPORT_UTF8
-             if (utf8)
-               {
-               class_utf8 = TRUE;
-               *class_utf8data++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
-               class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
-               *class_utf8data++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
-               class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
-               }
- #endif
              continue;
  #ifdef SUPPORT_UCP
-Line 3692 
 for (;; ptr++)
+Line 4347 
 for (;; ptr++)
                int pdata;
                int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
                if (ptype < 0) goto FAILED;
-               class_utf8 = TRUE;
+               *class_uchardata++ = ((-c == ESC_p) != negated)?
-               *class_utf8data++ = ((-c == ESC_p) != negated)?
                  XCL_PROP : XCL_NOTPROP;
-               *class_utf8data++ = ptype;
+               *class_uchardata++ = ptype;
-               *class_utf8data++ = pdata;
+               *class_uchardata++ = pdata;
-               class_charcount -= 2;   /* Not a < 256 character */
+               class_has_8bitchar--;                /* Undo! */
                continue;
                }
  #endif
-Line 3711 
 for (;; ptr++)
+Line 4365 
 for (;; ptr++)
                *errorcodeptr = ERR7;
                goto FAILED;
                }
-             class_charcount -= 2;  /* Undo the default count from above */
+             class_has_8bitchar--;    /* Undo the speculative increase. */
-             c = *ptr;              /* Get the final character and fall through */
+             class_one_char -= 2;     /* Undo the speculative increase. */
+             c = *ptr;                /* Get the final character and fall through */
              break;
              }
            }
-         /* Fall through if we have a single character (c >= 0). This may be
+         /* Fall through if the escape just defined a single character (c >= 0).
-         greater than 256 in UTF-8 mode. */
+         This may be greater than 256. */
          }   /* End of backslash handling */
-       /* A single character may be followed by '-' to form a range. However,
+       /* A character may be followed by '-' to form a range. However, Perl does
-       Perl does not permit ']' to be the end of the range. A '-' character
+       not permit ']' to be the end of the range. A '-' character at the end is
-       at the end is treated as a literal. Perl ignores orphaned \E sequences
+       treated as a literal. Perl ignores orphaned \E sequences entirely. The
-       entirely. The code for handling \Q and \E is messy. */
+       code for handling \Q and \E is messy. */
        CHECK_RANGE:
        while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
-Line 3733 
 for (;; ptr++)
+Line 4388 
 for (;; ptr++)
          inescq = FALSE;
          ptr += 2;
          }
        oldptr = ptr;
-       /* Remember \r or \n */
+       /* Remember if \r or \n were explicitly used */
        if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
-Line 3759 
 for (;; ptr++)
+Line 4413 
 for (;; ptr++)
            inescq = TRUE;
            break;
            }
+         /* Minus (hyphen) at the end of a class is treated as a literal, so put
+         back the pointer and jump to handle the character that preceded it. */
          if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
            {
            ptr = oldptr;
-           goto LONE_SINGLE_CHARACTER;
+           goto CLASS_SINGLE_CHARACTER;
            }
+         /* Otherwise, we have a potential range; pick up the next character */
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-         if (utf8)
+         if (utf)
            {                           /* Braces are required because the */
            GETCHARLEN(d, ptr, ptr);    /* macro generates multiple statements */
            }
-Line 3784 
 for (;; ptr++)
+Line 4443 
 for (;; ptr++)
            d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
            if (*errorcodeptr != 0) goto FAILED;
-           /* \b is backspace; any other special means the '-' was literal */
+           /* \b is backspace; any other special means the '-' was literal. */
            if (d < 0)
              {
              if (d == -ESC_b) d = CHAR_BS; else
                {
                ptr = oldptr;
-               goto LONE_SINGLE_CHARACTER;  /* A few lines below */
+               goto CLASS_SINGLE_CHARACTER;  /* A few lines below */
                }
              }
-           }
-         /* Check that the two values are in the correct order. Optimize
-         one-character ranges */
-         if (d < c)
-           {
-           *errorcodeptr = ERR8;
-           goto FAILED;
-           }
-         if (d == c) goto LONE_SINGLE_CHARACTER;  /* A few lines below */
-         /* Remember \r or \n */
-         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
-         matching for characters > 127 is available only if UCP support is
-         available. */
- #ifdef SUPPORT_UTF8
-         if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
-           {
-           class_utf8 = TRUE;
-           /* With UCP support, we can find the other case equivalents of
-           the relevant characters. There may be several ranges. Optimize how
-           they fit with the basic range. */
- #ifdef SUPPORT_UCP
-           if ((options & PCRE_CASELESS) != 0)
-             {
-             unsigned int occ, ocd;
-             unsigned int cc = c;
-             unsigned int origd = d;
-             while (get_othercase_range(&cc, origd, &occ, &ocd))
-               {
-               if (occ >= (unsigned int)c &&
-                   ocd <= (unsigned int)d)
-                 continue;                          /* Skip embedded ranges */
-               if (occ < (unsigned int)c  &&
-                   ocd >= (unsigned int)c - 1)      /* Extend the basic range */
-                 {                                  /* if there is overlap,   */
-                 c = occ;                           /* noting that if occ < c */
-                 continue;                          /* we can't have ocd > d  */
-                 }                                  /* because a subrange is  */
-               if (ocd > (unsigned int)d &&
-                   occ <= (unsigned int)d + 1)      /* always shorter than    */
-                 {                                  /* the basic range.       */
-                 d = ocd;
-                 continue;
-                 }
-               if (occ == ocd)
-                 {
-                 *class_utf8data++ = XCL_SINGLE;
-                 }
-               else
-                 {
-                 *class_utf8data++ = XCL_RANGE;
-                 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
-                 }
-               class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
-               }
-             }
- #endif  /* SUPPORT_UCP */
-           /* Now record the original range, possibly modified for UCP caseless
-           overlapping ranges. */
-           *class_utf8data++ = XCL_RANGE;
-           class_utf8data += _pcre_ord2utf8(c, class_utf8data);
-           class_utf8data += _pcre_ord2utf8(d, class_utf8data);
-           /* With UCP support, we are done. Without UCP support, there is no
-           caseless matching for UTF-8 characters > 127; we can use the bit map
-           for the smaller ones. */
- #ifdef SUPPORT_UCP
-           continue;    /* With next character in the class */
- #else
-           if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
-           /* Adjust upper limit and fall through to set up the map */
-           d = 127;
- #endif  /* SUPPORT_UCP */
            }
- #endif  /* SUPPORT_UTF8 */
-         /* We use the bit map for all cases when not in UTF-8 mode; else
-         ranges that lie entirely within 0-127 when there is UCP support; else
-         for partial ranges without UCP support. */
-         class_charcount += d - c + 1;
+         /* Check that the two values are in the correct order. Optimize
-         class_lastchar = d;
+         one-character ranges. */
-         /* We can save a bit of time by skipping this in the pre-compile. */
-         if (lengthptr == NULL) for (; c <= d; c++)
+         if (d < c)
            {
-           classbits[c/8] |= (1 << (c&7));
+           *errorcodeptr = ERR8;
-           if ((options & PCRE_CASELESS) != 0)
+           goto FAILED;
-             {
-             int uc = cd->fcc[c];           /* flip case */
-             classbits[uc/8] |= (1 << (uc&7));
-             }
            }
+         if (d == c) goto CLASS_SINGLE_CHARACTER;  /* A few lines below */
-         continue;   /* Go get the next char in the class */
+         /* We have found a character range, so single character optimizations
-         }
+         cannot be done anymore. Any value greater than 1 indicates that there
+         is more than one character. */
+         class_one_char = 2;
-       /* Handle a lone single character - we can get here for a normal
+         /* Remember an explicit \r or \n, and add the range to the class. */
-       non-escape char, or after \ that introduces a single character or for an
-       apparent range that isn't. */
-       LONE_SINGLE_CHARACTER:
+         if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
+         class_has_8bitchar +=
+           add_to_class(classbits, &class_uchardata, options, cd, c, d);
+         continue;   /* Go get the next char in the class */
+         }
-       /* Handle a character that cannot go in the bit map */
+       /* Handle a single character - we can get here for a normal non-escape
+       char, or after \ that introduces a single character or for an apparent
+       range that isn't. Only the value 1 matters for class_one_char, so don't
+       increase it if it is already 2 or more ... just in case there's a class
+       with a zillion characters in it. */
+       CLASS_SINGLE_CHARACTER:
+       if (class_one_char < 2) class_one_char++;
+       /* If class_one_char is 1, we have the first single character in the
+       class, and there have been no prior ranges, or XCLASS items generated by
+       escapes. If this is the final character in the class, we can optimize by
+       turning the item into a 1-character OP_CHAR[I] if it's positive, or
+       OP_NOT[I] if it's negative. In the positive case, it can cause firstchar
+       to be set. Otherwise, there can be no first char if this item is first,
+       whatever repeat count may follow. In the case of reqchar, save the
+       previous value for reinstating. */
- #ifdef SUPPORT_UTF8
+       if (class_one_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
-       if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
          {
-         class_utf8 = TRUE;
+         ptr++;
-         *class_utf8data++ = XCL_SINGLE;
+         zeroreqchar = reqchar;
-         class_utf8data += _pcre_ord2utf8(c, class_utf8data);
- #ifdef SUPPORT_UCP
+         if (negate_class)
-         if ((options & PCRE_CASELESS) != 0)
            {
-           unsigned int othercase;
+ #ifdef SUPPORT_UCP
-           if ((othercase = UCD_OTHERCASE(c)) != c)
+             // FIXMEchpe pcreuint32?
+           int d;
+ #endif
+           if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
+           zerofirstchar = firstchar;
+           /* For caseless UTF-8 mode when UCP support is available, check
+           whether this character has more than one other case. If so, generate
+           a special OP_NOTPROP item instead of OP_NOTI. */
+ #ifdef SUPPORT_UCP
+           if (utf && (options & PCRE_CASELESS) != 0 &&
+               (d = UCD_CASESET(c)) != 0)
+             {
+             *code++ = OP_NOTPROP;
+             *code++ = PT_CLIST;
+             *code++ = d;
+             }
+           else
+ #endif
+           /* Char has only one other case, or UCP not available */
              {
-             *class_utf8data++ = XCL_SINGLE;
+             *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
-             class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
+             if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
+               code += PRIV(ord2utf)(c, code);
+             else
+ #endif
+               *code++ = c;
              }
+           /* We are finished with this character class */
+           goto END_CLASS;
            }
- #endif  /* SUPPORT_UCP */
-         }
+         /* For a single, positive character, get the value into mcbuffer, and
-       else
+         then we can handle this with the normal one-character code. */
- #endif  /* SUPPORT_UTF8 */
-       /* Handle a single-byte character */
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
-         {
+         if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
-         classbits[c/8] |= (1 << (c&7));
+           mclength = PRIV(ord2utf)(c, mcbuffer);
-         if ((options & PCRE_CASELESS) != 0)
+         else
+ #endif
            {
-           c = cd->fcc[c];   /* flip case */
+           mcbuffer[0] = c;
-           classbits[c/8] |= (1 << (c&7));
+           mclength = 1;
            }
-         class_charcount++;
+         goto ONE_CHAR;
-         class_lastchar = c;
+         }       /* End of 1-char optimization */
-         }
+       /* There is more than one character in the class, or an XCLASS item
+       has been generated. Add this character to the class. */
+       class_has_8bitchar +=
+         add_to_class(classbits, &class_uchardata, options, cd, c, c);
        }
      /* Loop until ']' reached. This "while" is the end of the "do" far above.
-Line 3971 
 for (;; ptr++)
+Line 4583 
 for (;; ptr++)
        goto FAILED;
        }
-     /* If class_charcount is 1, we saw precisely one character whose value is
+     /* We will need an XCLASS if data has been placed in class_uchardata. In
-     less than 256. As long as there were no characters >= 128 and there was no
+     the second phase this is a sufficient test. However, in the pre-compile
-     use of \p or \P, in other words, no use of any XCLASS features, we can
+     phase, class_uchardata gets emptied to prevent workspace overflow, so it
-     optimize.
+     only if the very last character in the class needs XCLASS will it contain
+     anything at this point. For this reason, xclass gets set TRUE above when
-     In UTF-8 mode, we can optimize the negative case only if there were no
+     uchar_classdata is emptied, and that's why this code is the way it is here
-     characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
+     instead of just doing a test on class_uchardata below. */
-     operate on single-bytes only. This is an historical hangover. Maybe one day
-     we can tidy these opcodes to handle multi-byte characters.
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
+     if (class_uchardata > class_uchardata_base) xclass = TRUE;
-     The optimization throws away the bit map. We turn the item into a
+ #endif
--character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
-     that OP_NOT does not support multibyte characters. In the positive case, it
+     /* If this is the first thing in the branch, there can be no first char
-     can cause firstbyte to be set. Otherwise, there can be no first char if
+     setting, whatever the repeat count. Any reqchar setting must remain
-     this item is first, whatever repeat count may follow. In the case of
+     unchanged after any kind of repeat. */
-     reqbyte, save the previous value for reinstating. */
+     if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
- #ifdef SUPPORT_UTF8
+     zerofirstchar = firstchar;
-     if (class_charcount == 1 && !class_utf8 &&
+     zeroreqchar = reqchar;
-       (!utf8 || !negate_class || class_lastchar < 128))
- #else
-     if (class_charcount == 1)
- #endif
-       {
-       zeroreqbyte = reqbyte;
-       /* The OP_NOT opcode works on one-byte characters only. */
-       if (negate_class)
-         {
-         if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
-         zerofirstbyte = firstbyte;
-         *code++ = OP_NOT;
-         *code++ = class_lastchar;
-         break;
-         }
-       /* For a single, positive character, get the value into mcbuffer, and
-       then we can handle this with the normal one-character code. */
- #ifdef SUPPORT_UTF8
-       if (utf8 && class_lastchar > 127)
-         mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
-       else
- #endif
-         {
-         mcbuffer[0] = class_lastchar;
-         mclength = 1;
-         }
-       goto ONE_CHAR;
-       }       /* End of 1-char optimization */
-     /* The general case - not the one-char optimization. If this is the first
-     thing in the branch, there can be no first char setting, whatever the
-     repeat count. Any reqbyte setting must remain unchanged after any kind of
-     repeat. */
-     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
-     zerofirstbyte = firstbyte;
-     zeroreqbyte = reqbyte;
      /* If there are characters with values > 255, we have to compile an
      extended class, with its own opcode, unless there was a negated special
-Line 4040 
 for (;; ptr++)
+Line 4611 
 for (;; ptr++)
      be listed) there are no characters < 256, we can omit the bitmap in the
      actual compiled code. */
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-     if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
+     if (xclass && (!should_flip_negation || (options & PCRE_UCP) != 0))
+ #elif !defined COMPILE_PCRE8
+     if (xclass && !should_flip_negation)
+ #endif
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
        {
-       *class_utf8data++ = XCL_END;    /* Marks the end of extra data */
+       *class_uchardata++ = XCL_END;    /* Marks the end of extra data */
        *code++ = OP_XCLASS;
        code += LINK_SIZE;
-       *code = negate_class? XCL_NOT : 0;
+       *code = negate_class? XCL_NOT:0;
        /* If the map is required, move up the extra data to make room for it;
        otherwise just move the code pointer to the end of the extra data. */
-       if (class_charcount > 0)
+       if (class_has_8bitchar > 0)
          {
          *code++ |= XCL_MAP;
-         memmove(code + 32, code, class_utf8data - code);
+         memmove(code + (32 / sizeof(pcre_uchar)), code,
+           IN_UCHARS(class_uchardata - code));
          memcpy(code, classbits, 32);
-         code = class_utf8data + 32;
+         code = class_uchardata + (32 / sizeof(pcre_uchar));
          }
-       else code = class_utf8data;
+       else code = class_uchardata;
        /* Now fill in the complete length of the item */
-       PUT(previous, 1, code - previous);
+       PUT(previous, 1, (int)(code - previous));
        break;   /* End of class handling */
        }
  #endif
-Line 4074 
 for (;; ptr++)
+Line 4650 
 for (;; ptr++)
      negating it if necessary. */
      *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
-     if (negate_class)
+     if (lengthptr == NULL)    /* Save time in the pre-compile phase */
-       {
-       if (lengthptr == NULL)    /* Save time in the pre-compile phase */
-         for (c = 0; c < 32; c++) code[c] = ~classbits[c];
-       }
-     else
        {
+       if (negate_class)
+         for (c = 0; c < 32; c++) classbits[c] = ~classbits[c];
        memcpy(code, classbits, 32);
        }
-     code += 32;
+     code += 32 / sizeof(pcre_uchar);
+     END_CLASS:
      break;
-Line 4120 
 for (;; ptr++)
+Line 4695 
 for (;; ptr++)
      if (repeat_min == 0)
        {
-       firstbyte = zerofirstbyte;    /* Adjust for zero repeat */
+       firstchar = zerofirstchar;    /* Adjust for zero repeat */
-       reqbyte = zeroreqbyte;        /* Ditto */
+       reqchar = zeroreqchar;        /* Ditto */
        }
      /* Remember whether this is a variable length repeat */
-Line 4131 
 for (;; ptr++)
+Line 4706 
 for (;; ptr++)
      op_type = 0;                    /* Default single-char op codes */
      possessive_quantifier = FALSE;  /* Default not possessive quantifier */
-     /* Save start of previous item, in case we have to move it up to make space
+     /* Save start of previous item, in case we have to move it up in order to
-     for an inserted OP_ONCE for the additional '+' extension. */
+     insert something before it. */
      tempcode = previous;
-Line 4155 
 for (;; ptr++)
+Line 4730 
 for (;; ptr++)
        }
      else repeat_type = greedy_default;
-     /* If previous was a character match, abolish the item and generate a
+     /* If previous was a recursion call, wrap it in atomic brackets so that
-     repeat item instead. If a char item has a minumum of more than one, ensure
+     previous becomes the atomic group. All recursions were so wrapped in the
-     that it is set in reqbyte - it might not be if a sequence such as x{3} is
+     past, but it no longer happens for non-repeated recursions. In fact, the
-     the first thing in a branch because the x will have gone into firstbyte
+     repeated ones could be re-implemented independently so as not to need this,
-     instead.  */
+     but for the moment we rely on the code for repeating groups. */
+     if (*previous == OP_RECURSE)
+       {
+       memmove(previous + 1 + LINK_SIZE, previous, IN_UCHARS(1 + LINK_SIZE));
+       *previous = OP_ONCE;
+       PUT(previous, 1, 2 + 2*LINK_SIZE);
+       previous[2 + 2*LINK_SIZE] = OP_KET;
+       PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
+       code += 2 + 2 * LINK_SIZE;
+       length_prevgroup = 3 + 3*LINK_SIZE;
+       /* When actually compiling, we need to check whether this was a forward
+       reference, and if so, adjust the offset. */
+       if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
+         {
+         int offset = GET(cd->hwm, -LINK_SIZE);
+         if (offset == previous + 1 - cd->start_code)
+           PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
+         }
+       }
+     /* Now handle repetition for the different types of item. */
-     if (*previous == OP_CHAR || *previous == OP_CHARNC)
+     /* If previous was a character or negated character match, abolish the item
+     and generate a repeat item instead. If a char item has a minimum of more
+     than one, ensure that it is set in reqchar - it might not be if a sequence
+     such as x{3} is the first thing in a branch because the x will have gone
+     into firstchar instead.  */
+     if (*previous == OP_CHAR || *previous == OP_CHARI
+         || *previous == OP_NOT || *previous == OP_NOTI)
        {
-       /* Deal with UTF-8 characters that take up more than one byte. It's
+       switch (*previous)
+         {
+         default: /* Make compiler happy. */
+         case OP_CHAR:  op_type = OP_STAR - OP_STAR; break;
+         case OP_CHARI: op_type = OP_STARI - OP_STAR; break;
+         case OP_NOT:   op_type = OP_NOTSTAR - OP_STAR; break;
+         case OP_NOTI:  op_type = OP_NOTSTARI - OP_STAR; break;
+         }
+       /* Deal with UTF characters that take up more than one character. It's
        easier to write this out separately than try to macrify it. Use c to
-       hold the length of the character in bytes, plus 0x80 to flag that it's a
+       hold the length of the character in bytes, plus UTF_LENGTH to flag that
-       length rather than a small character. */
+       it's a length rather than a small character. */
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
-       if (utf8 && (code[-1] & 0x80) != 0)
+       if (utf && NOT_FIRSTCHAR(code[-1]))
          {
-         uschar *lastchar = code - 1;
+         pcre_uchar *lastchar = code - 1;
-         while((*lastchar & 0xc0) == 0x80) lastchar--;
+         BACKCHAR(lastchar);
-         c = code - lastchar;            /* Length of UTF-8 character */
+         c = (int)(code - lastchar);     /* Length of UTF-8 character */
-         memcpy(utf8_char, lastchar, c); /* Save the char */
+         memcpy(utf_chars, lastchar, IN_UCHARS(c)); /* Save the char */
-         c |= 0x80;                      /* Flag c as a length */
+         c |= UTF_LENGTH;                /* Flag c as a length */
          }
        else
- #endif
+ #endif /* SUPPORT_UTF */
-       /* Handle the case of a single byte - either with no UTF8 support, or
-       with UTF-8 disabled, or for a UTF-8 character < 128. */
+       /* Handle the case of a single charater - either with no UTF support, or
+       with UTF disabled, or for a single character UTF character. */
          {
          c = code[-1];
-         if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
+         if (*previous <= OP_CHARI && repeat_min > 1)
+           reqchar = c | req_caseopt | cd->req_varyopt;
          }
        /* If the repetition is unlimited, it pays to see if the next thing on
-Line 4195 
 for (;; ptr++)
+Line 4809 
 for (;; ptr++)
        if (!possessive_quantifier &&
            repeat_max < 0 &&
-           check_auto_possessive(previous, utf8, ptr + 1, options, cd))
+           check_auto_possessive(previous, utf, ptr + 1, options, cd))
          {
          repeat_type = 0;    /* Force greedy */
          possessive_quantifier = TRUE;
-Line 4204 
 for (;; ptr++)
+Line 4818 
 for (;; ptr++)
        goto OUTPUT_SINGLE_REPEAT;   /* Code shared with single character types */
        }
-     /* If previous was a single negated character ([^a] or similar), we use
-     one of the special opcodes, replacing it. The code is shared with single-
-     character repeats by setting opt_type to add a suitable offset into
-     repeat_type. We can also test for auto-possessification. OP_NOT is
-     currently used only for single-byte chars. */
-     else if (*previous == OP_NOT)
-       {
-       op_type = OP_NOTSTAR - OP_STAR;  /* Use "not" opcodes */
-       c = previous[1];
-       if (!possessive_quantifier &&
-           repeat_max < 0 &&
-           check_auto_possessive(previous, utf8, ptr + 1, options, cd))
-         {
-         repeat_type = 0;    /* Force greedy */
-         possessive_quantifier = TRUE;
-         }
-       goto OUTPUT_SINGLE_REPEAT;
-       }
      /* If previous was a character type match (\d or similar), abolish it and
      create a suitable repeat item. The code is shared with single-character
      repeats by setting op_type to add a suitable offset into repeat_type. Note
-Line 4233 
 for (;; ptr++)
+Line 4827 
 for (;; ptr++)
      else if (*previous < OP_EODN)
        {
-       uschar *oldcode;
+       pcre_uchar *oldcode;
        int prop_type, prop_value;
        op_type = OP_TYPESTAR - OP_STAR;  /* Use type opcodes */
        c = *previous;
        if (!possessive_quantifier &&
            repeat_max < 0 &&
-           check_auto_possessive(previous, utf8, ptr + 1, options, cd))
+           check_auto_possessive(previous, utf, ptr + 1, options, cd))
          {
          repeat_type = 0;    /* Force greedy */
          possessive_quantifier = TRUE;
-Line 4320 
 for (;; ptr++)
+Line 4914 
 for (;; ptr++)
          we have to insert the character for the previous code. For a repeated
          Unicode property match, there are two extra bytes that define the
          required property. In UTF-8 mode, long characters have their length in
-         c, with the 0x80 bit as a flag. */
+         c, with the UTF_LENGTH bit as a flag. */
          if (repeat_max < 0)
            {
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
-           if (utf8 && c >= 128)
+           if (utf && (c & UTF_LENGTH) != 0)
              {
-             memcpy(code, utf8_char, c & 7);
+             memcpy(code, utf_chars, IN_UCHARS(c & 7));
              code += c & 7;
              }
            else
-Line 4349 
 for (;; ptr++)
+Line 4943 
 for (;; ptr++)
          else if (repeat_max != repeat_min)
            {
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
-           if (utf8 && c >= 128)
+           if (utf && (c & UTF_LENGTH) != 0)
              {
-             memcpy(code, utf8_char, c & 7);
+             memcpy(code, utf_chars, IN_UCHARS(c & 7));
              code += c & 7;
              }
            else
-Line 4379 
 for (;; ptr++)
+Line 4973 
 for (;; ptr++)
        /* The character or character type itself comes last in all cases. */
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
-       if (utf8 && c >= 128)
+       if (utf && (c & UTF_LENGTH) != 0)
          {
-         memcpy(code, utf8_char, c & 7);
+         memcpy(code, utf_chars, IN_UCHARS(c & 7));
          code += c & 7;
          }
        else
-Line 4406 
 for (;; ptr++)
+Line 5000 
 for (;; ptr++)
      else if (*previous == OP_CLASS ||
               *previous == OP_NCLASS ||
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
               *previous == OP_XCLASS ||
  #endif
-              *previous == OP_REF)
+              *previous == OP_REF ||
+              *previous == OP_REFI)
        {
        if (repeat_max == 0)
          {
-Line 4443 
 for (;; ptr++)
+Line 5038 
 for (;; ptr++)
        }
      /* If previous was a bracket group, we may have to replicate it in certain
-     cases. */
+     cases. Note that at this point we can encounter only the "basic" bracket
+     opcodes such as BRA and CBRA, as this is the place where they get converted
+     into the more special varieties such as BRAPOS and SBRA. A test for >=
+     OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
+     ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
+     repetition of assertions, but now it does, for Perl compatibility. */
-     else if (*previous == OP_BRA  || *previous == OP_CBRA ||
+     else if (*previous >= OP_ASSERT && *previous <= OP_COND)
-              *previous == OP_ONCE || *previous == OP_COND)
        {
        register int i;
-       int ketoffset = 0;
        int len = (int)(code - previous);
-       uschar *bralink = NULL;
+       pcre_uchar *bralink = NULL;
+       pcre_uchar *brazeroptr = NULL;
-       /* Repeating a DEFINE group is pointless */
+       /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
+       we just ignore the repeat. */
        if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
-         {
+         goto END_REPEAT;
-         *errorcodeptr = ERR55;
-         goto FAILED;
-         }
-       /* If the maximum repeat count is unlimited, find the end of the bracket
+       /* There is no sense in actually repeating assertions. The only potential
-       by scanning through from the start, and compute the offset back to it
+       use of repetition is in cases when the assertion is optional. Therefore,
-       from the current code pointer. There may be an OP_OPT setting following
+       if the minimum is greater than zero, just ignore the repeat. If the
-       the final KET, so we can't find the end just by going back from the code
+       maximum is not not zero or one, set it to 1. */
-       pointer. */
+       if (*previous < OP_ONCE)    /* Assertion */
-       if (repeat_max == -1)
+         {
-         {
+         if (repeat_min > 0) goto END_REPEAT;
-         register uschar *ket = previous;
+         if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
-         do ket += GET(ket, 1); while (*ket != OP_KET);
-         ketoffset = (int)(code - ket);
          }
        /* The case of a zero minimum is special because of the need to stick
-Line 4492 
 for (;; ptr++)
+Line 5087 
 for (;; ptr++)
          **   goto END_REPEAT;
          **   }
-         However, that fails when a group is referenced as a subroutine from
+         However, that fails when a group or a subgroup within it is referenced
-         elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
+         as a subroutine from elsewhere in the pattern, so now we stick in
-         so that it is skipped on execution. As we don't have a list of which
+         OP_SKIPZERO in front of it so that it is skipped on execution. As we
-         groups are referenced, we cannot do this selectively.
+         don't have a list of which groups are referenced, we cannot do this
+         selectively.