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

Diff of /code/trunk/pcre_compile.c

Parent Directory | Revision Log | View Patch Patch

-revision 505 by ph10,
Tue Mar  9 16:50:47 2010 UTC
+revision 842 by ph10,
Sat Dec 31 15:19:04 2011 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2010 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)_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
-Line 88 
 so this number is very generous.
+Line 92 
 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 */
+ /* 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 124 
 static const short int escapes[] = {
+Line 143 
 static const short int escapes[] = {
       -ESC_H,                  0,
 ,                       -ESC_K,
 ,                       0,
-,                       0,
+      -ESC_N,                  0,
       -ESC_P,                  -ESC_Q,
       -ESC_R,                  -ESC_S,
 ,                       0,
-Line 171 
 static const short int escapes[] = {
+Line 190 
 static const short int escapes[] = {
  /*  B8 */     0,     0,      0,       0,      0,   ']',    '=',    '-',
  /*  C0 */   '{',-ESC_A, -ESC_B,  -ESC_C, -ESC_D,-ESC_E,      0, -ESC_G,
  /*  C8 */-ESC_H,     0,      0,       0,      0,     0,      0,      0,
- /*  D0 */   '}',     0, -ESC_K,       0,      0,     0,      0, -ESC_P,
+ /*  D0 */   '}',     0, -ESC_K,       0,      0,-ESC_N,      0, -ESC_P,
  /*  D8 */-ESC_Q,-ESC_R,      0,       0,      0,     0,      0,      0,
  /*  E0 */  '\\',     0, -ESC_S,       0,      0,-ESC_V, -ESC_W, -ESC_X,
  /*  E8 */     0,-ESC_Z,      0,       0,      0,     0,      0,      0,
-Line 188 
 string is built from string macros so th
+Line 207 
 string is built from string macros so th
  platforms. */
  typedef struct verbitem {
-   int   len;
+   int   len;                 /* Length of verb name */
-   int   op;
+   int   op;                  /* Op when no arg, or -1 if arg mandatory */
+   int   op_arg;              /* Op when arg present, or -1 if not allowed */
  } verbitem;
  static const char verbnames[] =
+   "\0"                       /* Empty name is a shorthand for MARK */
+   STRING_MARK0
    STRING_ACCEPT0
    STRING_COMMIT0
    STRING_F0
-Line 202 
 static const char verbnames[] =
+Line 224 
 static const char verbnames[] =
    STRING_THEN;
  static const verbitem verbs[] = {
-   { 6, OP_ACCEPT },
+   { 0, -1,        OP_MARK },
-   { 6, OP_COMMIT },
+   { 4, -1,        OP_MARK },
-   { 1, OP_FAIL },
+   { 6, OP_ACCEPT, -1 },
-   { 4, OP_FAIL },
+   { 6, OP_COMMIT, -1 },
-   { 5, OP_PRUNE },
+   { 1, OP_FAIL,   -1 },
-   { 4, OP_SKIP  },
+   { 4, OP_FAIL,   -1 },
-   { 4, OP_THEN  }
+   { 5, OP_PRUNE,  OP_PRUNE_ARG },
+   { 4, OP_SKIP,   OP_SKIP_ARG  },
+   { 4, OP_THEN,   OP_THEN_ARG  }
  };
  static const int verbcount = sizeof(verbs)/sizeof(verbitem);
-Line 226 
 static const char posix_names[] =
+Line 250 
 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 256 
 static const int posix_class_maps[] = {
+Line 280 
 static const int posix_class_maps[] = {
    cbit_xdigit,-1,          0              /* xdigit */
  };
+ /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
+ substitutes must be in the order of the names, defined above, and there are
+ both positive and negative cases. NULL means no substitute. */
+ #ifdef SUPPORT_UCP
+ static const pcre_uchar string_PNd[]  = {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_pNd[]  = {
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_N, CHAR_d, 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_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 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 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 */
+   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 */
+ };
+ #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
+ #endif
  #define STRING(a)  # a
  #define XSTRING(s) STRING(s)
-Line 268 
 the number of relocations needed when a
+Line 393 
 the number of relocations needed when a
  it is now one long string. We cannot use a table of offsets, because the
  lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
  simply count through to the one we want - this isn't a performance issue
  because these strings are used only when there is a compilation error.
  Each substring ends with \0 to insert a null character. This includes the final
  substring, so that the whole string ends with \0\0, which can be detected when
  counting through. */
  static const char error_texts[] =
-Line 319 
 static const char error_texts[] =
+Line 444 
 static const char error_texts[] =
    /* 35 */
    "invalid condition (?(0)\0"
    "\\C not allowed in lookbehind assertion\0"
-   "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
+   "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
    "number after (?C is > 255\0"
    "closing ) for (?C expected\0"
    /* 40 */
-Line 341 
 static const char error_texts[] =
+Line 466 
 static const char error_texts[] =
    "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"
-   "(*VERB) with an argument is not supported\0"
+   "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
    /* 60 */
    "(*VERB) not recognized\0"
    "number is too big\0"
-Line 353 
 static const char error_texts[] =
+Line 478 
 static const char error_texts[] =
    "digit expected after (?+\0"
    "] is an invalid data character in JavaScript compatibility mode\0"
    /* 65 */
-   "different names for subpatterns of the same number are not allowed\0";
+   "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"
+   "\\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 UTF-8/16 code point (>= 0xd800 && <= 0xdfff)\0"
+   ;
  /* Table to identify digits and hex digits. This is used when compiling
  patterns. Note that the tables in chartables are dependent on the locale, and
-Line 371 
 For convenience, we use the same bit def
+Line 506 
 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 415 
 static const unsigned char digitab[] =
+Line 556 
 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 450 
 static const unsigned char digitab[] =
+Line 591 
 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 489 
 static const unsigned char ebcdic_charta
+Line 630 
 static const unsigned char ebcdic_charta
  /* Definition to allow mutual recursion */
  static BOOL
-   compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
+   compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
      int *, int *, branch_chain *, compile_data *, int *);
-Line 511 
 static const char *
+Line 652 
 static const char *
  find_error_text(int n)
  {
  const char *s = error_texts;
  for (; n > 0; n--)
    {
    while (*s++ != 0) {};
    if (*s == 0) return "Error text not found (please report)";
    }
  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 545 
 Returns:         zero or positive => a d
+Line 758 
 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 564 
 in a table. A non-zero result is somethi
+Line 779 
 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 < 'a' || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
  else if ((i = escapes[c - 0x48]) != 0)  c = i;
  #endif
-Line 576 
 else if ((i = escapes[c - 0x48]) != 0)
+Line 793 
 else if ((i = escapes[c - 0x48]) != 0)
  else
    {
-   const uschar *oldptr;
+   const pcre_uchar *oldptr;
    BOOL braced, negated;
    switch (c)
-Line 586 
 else
+Line 803 
 else
      case CHAR_l:
      case CHAR_L:
-     case CHAR_N:
+     *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
+           }
+         }
+       }
+     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 609 
 else
+Line 857 
 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 619 
 else
+Line 868 
 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 639 
 else
+Line 888 
 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 692 
 else
+Line 950 
 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 729 
 else
+Line 997 
 else
      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;
+     if (!utf && c > 0xff) *errorcodeptr = ERR51;
      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 756 
 else
+Line 1045 
 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
+ #ifdef COMPILE_PCRE8
+         if (c > (utf ? 0x10ffff : 0xff)) { c = -1; break; }
+ #else
+ #ifdef COMPILE_PCRE16
+         if (c > (utf ? 0x10ffff : 0xffff)) { c = -1; break; }
+ #endif
+ #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 772 
 else
+Line 1075 
 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 787 
 else
+Line 1090 
 else
      break;
      /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
-     This coding is ASCII-specific, but then the whole concept of \cx is
+     An error is given if the byte following \c is not an ASCII character. This
+     coding is ASCII-specific, but then the whole concept of \cx is
      ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
      case CHAR_c:
-Line 797 
 else
+Line 1101 
 else
        *errorcodeptr = ERR2;
        break;
        }
+ #ifndef EBCDIC    /* ASCII/UTF-8 coding */
- #ifndef EBCDIC  /* ASCII/UTF-8 coding */
+     if (c > 127)  /* Excludes all non-ASCII in either mode */
+       {
+       *errorcodeptr = ERR68;
+       break;
+       }
      if (c >= CHAR_a && c <= CHAR_z) c -= 32;
      c ^= 0x40;
- #else           /* EBCDIC coding */
+ #else             /* EBCDIC coding */
      if (c >= CHAR_a && c <= CHAR_z) c += 64;
      c ^= 0xC0;
  #endif
-Line 824 
 else
+Line 1132 
 else
      }
    }
+ /* Perl supports \N{name} for character names, as well as plain \N for "not
+ 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 &&
+      !is_counted_repeat(ptr+2))
+   *errorcodeptr = ERR37;
+ /* If PCRE_UCP is set, we change the values for \d etc. */
+ if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
+   c -= (ESC_DU - ESC_D);
+ /* Set the pointer to the final character before returning. */
  *ptrptr = ptr;
  return c;
  }
-Line 850 
 Returns:         type value from ucp_typ
+Line 1173 
 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 871 
 if (c == CHAR_LEFT_CURLY_BRACKET)
+Line 1194 
 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 895 
 else
+Line 1218 
 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 924 
 return -1;
+Line 1247 
 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 975 
 Returns:         pointer to '}' on succe
+Line 1265 
 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 984 
 int max = -1;
+Line 1274 
 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 999 
 if (*p == CHAR_RIGHT_CURLY_BRACKET) max
+Line 1289 
 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 1032 
 top-level call starts at the beginning o
+Line 1322 
 top-level call starts at the beginning o
  start at a parenthesis. It scans along a pattern's text looking for capturing
  subpatterns, and counting them. If it finds a named pattern that matches the
  name it is given, it returns its number. Alternatively, if the name is NULL, it
- returns when it reaches a given numbered subpattern. We know that if (?P< is
+ returns when it reaches a given numbered subpattern. Recursion is used to keep
- encountered, the name will be terminated by '>' because that is checked in the
+ track of subpatterns that reset the capturing group numbers - the (?| feature.
- first pass. Recursion is used to keep track of subpatterns that reset the
- capturing group numbers - the (?| feature.
+ This function was originally called only from the second pass, in which we know
+ that if (?< or (?' or (?P< is encountered, the name will be correctly
+ terminated because that is checked in the first pass. There is now one call to
+ this function in the first pass, to check for a recursive back reference by
+ name (so that we can make the whole group atomic). In this case, we need check
+ only up to the current position in the pattern, and that is still OK because
+ and previous occurrences will have been checked. To make this work, the test
+ for "end of pattern" is a check against cd->end_pattern in the main loop,
+ instead of looking for a binary zero. This means that the special first-pass
+ call can adjust cd->end_pattern temporarily. (Checks for binary zero while
+ processing items within the loop are OK, because afterwards the main loop will
+ terminate.)
  Arguments:
    ptrptr       address of the current character pointer (updated)
-Line 1043 
 Arguments:
+Line 1344 
 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
+   utf          TRUE if we are in UTF-8 / UTF-16 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, 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 1062 
 dealing with. The very first call may no
+Line 1364 
 dealing with. The very first call may no
  if (ptr[0] == CHAR_LEFT_PARENTHESIS)
    {
-   if (ptr[1] == CHAR_QUESTION_MARK &&
+   /* Handle specials such as (*SKIP) or (*UTF8) etc. */
-       ptr[2] == CHAR_VERTICAL_LINE)
+   if (ptr[1] == CHAR_ASTERISK) ptr += 2;
+   /* Handle a normal, unnamed capturing parenthesis. */
+   else if (ptr[1] != CHAR_QUESTION_MARK)
+     {
+     *count += 1;
+     if (name == NULL && *count == lorn) return *count;
+     ptr++;
+     }
+   /* All cases now have (? at the start. Remember when we are in a group
+   where the parenthesis numbers are duplicated. */
+   else if (ptr[2] == CHAR_VERTICAL_LINE)
      {
      ptr += 3;
      dup_parens = TRUE;
      }
-   /* Handle a normal, unnamed capturing parenthesis */
+   /* Handle comments; all characters are allowed until a ket is reached. */
-   else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
+   else if (ptr[2] == CHAR_NUMBER_SIGN)
      {
-     *count += 1;
+     for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
-     if (name == NULL && *count == lorn) return *count;
+     goto FAIL_EXIT;
-     ptr++;
      }
    /* Handle a condition. If it is an assertion, just carry on so that it
    is processed as normal. If not, skip to the closing parenthesis of the
-   condition (there can't be any nested parens. */
+   condition (there can't be any nested parens). */
    else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
      {
-Line 1092 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
+Line 1408 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
        }
      }
-   /* We have either (? or (* and not a condition */
+   /* Start with (? but not a condition. */
    else
      {
-Line 1105 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
+Line 1421 
 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 1113 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
+Line 1429 
 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 1121 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
+Line 1437 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
    }
  /* Past any initial parenthesis handling, scan for parentheses or vertical
- bars. */
+ bars. Stop if we get to cd->end_pattern. Note that this is important for the
+ first-pass call when this value is temporarily adjusted to stop at the current
+ position. So DO NOT change this to a test for binary zero. */
- for (; *ptr != 0; ptr++)
+ for (; ptr < cd->end_pattern; ptr++)
    {
    /* Skip over backslashed characters and also entire \Q...\E */
-Line 1154 
 for (; *ptr != 0; ptr++)
+Line 1472 
 for (; *ptr != 0; 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 1197 
 for (; *ptr != 0; ptr++)
+Line 1515 
 for (; *ptr != 0; ptr++)
    if (xmode && *ptr == CHAR_NUMBER_SIGN)
      {
-     while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
+     ptr++;
+     while (*ptr != 0)
+       {
+       if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
+       ptr++;
+ #ifdef SUPPORT_UTF
+       if (utf) FORWARDCHAR(ptr);
+ #endif
+       }
      if (*ptr == 0) goto FAIL_EXIT;
      continue;
      }
-Line 1206 
 for (; *ptr != 0; ptr++)
+Line 1532 
 for (; *ptr != 0; ptr++)
    if (*ptr == CHAR_LEFT_PARENTHESIS)
      {
-     int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, 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 1214 
 for (; *ptr != 0; ptr++)
+Line 1540 
 for (; *ptr != 0; ptr++)
    else if (*ptr == CHAR_RIGHT_PARENTHESIS)
      {
      if (dup_parens && *count < hwm_count) *count = hwm_count;
-     *ptrptr = ptr;
+     goto FAIL_EXIT;
-     return -1;
      }
    else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
-Line 1253 
 Arguments:
+Line 1578 
 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
+   utf          TRUE if we are in UTF-8 / UTF-16 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 utf)
  {
- uschar *ptr = (uschar *)cd->start_pattern;
+ pcre_uchar *ptr = (pcre_uchar *)cd->start_pattern;
  int count = 0;
  int rc;
-Line 1271 
 matching closing parens. That is why we
+Line 1598 
 matching closing parens. That is why we
  for (;;)
    {
-   rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
+   rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
    if (rc > 0 || *ptr++ == 0) break;
    }
-Line 1287 
 return rc;
+Line 1614 
 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 1334 
 for (;;)
+Line 1651 
 for (;;)
      case OP_RREF:
      case OP_NRREF:
      case OP_DEF:
-     code += _pcre_OP_lengths[*code];
+     code += PRIV(OP_lengths)[*code];
      break;
      default:
-Line 1364 
 and doing the check at the end; a flag s
+Line 1681 
 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 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 1388 
 branch, check the length against that of
+Line 1706 
 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 1425 
 for (;;)
+Line 1752 
 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 1441 
 for (;;)
+Line 1768 
 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
+ #ifdef SUPPORT_UTF
-     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 1482 
 for (;;)
+Line 1824 
 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
+ #ifdef SUPPORT_UTF
-     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 += 2;
-     cc += 4;
+     cc += 1 + IMM2_SIZE + 1;
      break;
      /* Handle single-char matchers */
-Line 1503 
 for (;;)
+Line 1847 
 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 1515 
 for (;;)
+Line 1863 
 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
      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 1542 
 for (;;)
+Line 1893 
 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 1554 
 for (;;)
+Line 1905 
 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 1576 
 length.
+Line 2010 
 length.
  Arguments:
    code        points to start of expression
-   utf8        TRUE in UTF-8 mode
+   utf         TRUE in UTF-8 / UTF-16 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 1600 
 for (;;)
+Line 2035 
 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
    repeated character types, we have to test for \p and \P, which have an extra
-   two bytes of parameters. */
+   two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
+   must add in its length. */
    else
      {
-Line 1637 
 for (;;)
+Line 2074 
 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:
+       case OP_PRUNE_ARG:
+       case OP_SKIP_ARG:
+       code += code[1];
+       break;
+       case OP_THEN_ARG:
+       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
+ #ifdef SUPPORT_UTF
-     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 1688 
 instance of OP_RECURSE.
+Line 2149 
 instance of OP_RECURSE.
  Arguments:
    code        points to start of expression
-   utf8        TRUE in UTF-8 mode
+   utf         TRUE in UTF-8 / UTF-16 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 1710 
 for (;;)
+Line 2171 
 for (;;)
    /* Otherwise, we can get the item's length from the table, except that for
    repeated character types, we have to test for \p and \P, which have an extra
-   two bytes of parameters. */
+   two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
+   must add in its length. */
    else
      {
-Line 1732 
 for (;;)
+Line 2194 
 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:
+       case OP_PRUNE_ARG:
+       case OP_SKIP_ARG:
+       code += code[1];
+       break;
+       case OP_THEN_ARG:
+       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
+ #ifdef SUPPORT_UTF
-     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 1789 
 bracket whose current branch will alread
+Line 2275 
 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 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;
    /* Skip over forward assertions; the other assertions are skipped by
    first_significant_code() with a TRUE final argument. */
-Line 1818 
 for (code = first_significant_code(code
+Line 2304 
 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;
          }
        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 1868 
 for (code = first_significant_code(code
+Line 2387 
 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 1886 
 for (code = first_significant_code(code
+Line 2405 
 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 1898 
 for (code = first_significant_code(code
+Line 2417 
 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 1939 
 for (code = first_significant_code(code
+Line 2458 
 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 1972 
 for (code = first_significant_code(code
+Line 2492 
 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 1980 
 for (code = first_significant_code(code
+Line 2501 
 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
+ #ifdef SUPPORT_UTF
      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
+     /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
+     string. */
+     case OP_MARK:
+     case OP_PRUNE_ARG:
+     case OP_SKIP_ARG:
+     code += code[1];
+     break;
+     case OP_THEN_ARG:
+     code += code[1];
+     break;
      /* None of the remaining opcodes are required to match a character. */
      default:
      break;
      }
    }
-Line 2023 
 return TRUE;
+Line 2567 
 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 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 2073 
 where Perl recognizes it as the POSIX cl
+Line 2619 
 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 2081 
 Returns:   TRUE or FALSE
+Line 2638 
 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 2118 
 Returns:     a value representing the na
+Line 2682 
 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 2157 
 value in the reference (which is a group
+Line 2721 
 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 mode
    cd         contains pointers to tables etc.
    save_hwm   the hwm forward reference pointer at the start of the group
-Line 2165 
 Returns:     nothing
+Line 2729 
 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 2218 
 Arguments:
+Line 2782 
 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, ptr - cd->start_pattern);  /* Pattern offset */
+ PUT(code, 0, (int)(ptr - cd->start_pattern));  /* Pattern offset */
- PUT(code, LINK_SIZE, 0);                /* Default length */
+ PUT(code, LINK_SIZE, 0);                       /* Default length */
- return code + 2*LINK_SIZE;
+ return code + 2 * LINK_SIZE;
  }
-Line 2247 
 Returns:             nothing
+Line 2811 
 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 = ptr - cd->start_pattern - GET(previous_callout, 2);
+ int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
  PUT(previous_callout, 2 + LINK_SIZE, length);
  }
-Line 2299 
 for (++c; c <= d; c++)
+Line 2863 
 for (++c; c <= d; c++)
  return TRUE;
  }
+ /*************************************************
+ *        Check a character and a property        *
+ *************************************************/
+ /* This function is called by check_auto_possessive() when a property item
+ is adjacent to a fixed character.
+ Arguments:
+   c            the character
+   ptype        the property type
+   pdata        the data for the type
+   negated      TRUE if it's a negated property (\P or \p{^)
+ Returns:       TRUE if auto-possessifying is OK
+ */
+ static BOOL
+ check_char_prop(int c, int ptype, int pdata, BOOL negated)
+ {
+ const ucd_record *prop = GET_UCD(c);
+ switch(ptype)
+   {
+   case PT_LAMP:
+   return (prop->chartype == ucp_Lu ||
+           prop->chartype == ucp_Ll ||
+           prop->chartype == ucp_Lt) == negated;
+   case PT_GC:
+   return (pdata == PRIV(ucp_gentype)[prop->chartype]) == negated;
+   case PT_PC:
+   return (pdata == prop->chartype) == negated;
+   case PT_SC:
+   return (pdata == prop->script) == negated;
+   /* These are specials */
+   case PT_ALNUM:
+   return (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
+           PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
+   case PT_SPACE:    /* Perl space */
+   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 (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 (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
+           PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
+           c == CHAR_UNDERSCORE) == negated;
+   }
+ return FALSE;
+ }
  #endif  /* SUPPORT_UCP */
-Line 2312 
 whether the next thing could possibly ma
+Line 2939 
 whether the next thing could possibly ma
  sense to automatically possessify the repeated item.
  Arguments:
-   op_code       the repeated op code
+   previous      pointer to the repeated opcode
-   this          data for this item, depends on the opcode
+   utf           TRUE in UTF-8 / UTF-16 mode
-   utf8          TRUE in UTF-8 mode
-   utf8_char     used for utf8 character bytes, NULL if not relevant
    ptr           next character in pattern
    options       options bits
    cd            contains pointers to tables etc.
-Line 2324 
 Returns:        TRUE if possessifying is
+Line 2949 
 Returns:        TRUE if possessifying is
  */
  static BOOL
- check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
+ check_auto_possessive(const pcre_uchar *previous, BOOL utf,
-   const uschar *ptr, int options, compile_data *cd)
+   const pcre_uchar *ptr, int options, compile_data *cd)
  {
- int next;
+ pcre_int32 c, next;
+ int op_code = *previous++;
  /* Skip whitespace and comments in extended mode */
-Line 2335 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2961 
 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)
        {
-       while (*(++ptr) != 0)
+       ptr++;
+       while (*ptr != 0)
+         {
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
+         ptr++;
+ #ifdef SUPPORT_UTF
+         if (utf) FORWARDCHAR(ptr);
+ #endif
+         }
        }
      else break;
      }
-Line 2355 
 if (*ptr == CHAR_BACKSLASH)
+Line 2988 
 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 2372 
 if ((options & PCRE_EXTENDED) != 0)
+Line 3003 
 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)
        {
-       while (*(++ptr) != 0)
+       ptr++;
+       while (*ptr != 0)
+         {
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
+         ptr++;
+ #ifdef SUPPORT_UTF
+         if (utf) FORWARDCHAR(ptr);
+ #endif
+         }
        }
      else break;
      }
-Line 2385 
 if ((options & PCRE_EXTENDED) != 0)
+Line 3023 
 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. If the previous is a
+ /* Now compare the next item with the previous opcode. First, handle cases when
- positive single character match, "item" either contains the character or, if
+ the next item is a character. */
- "item" is greater than 127 in utf8 mode, the character's bytes are in
- utf8_char. */
- /* Handle cases when the next item is a character. */
  if (next >= 0) switch(op_code)
    {
    case OP_CHAR:
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+   GETCHARTEST(c, previous);
  #else
-   (void)(utf8_char);  /* Keep compiler happy by referencing function argument */
+   c = *previous;
  #endif
-   return item != next;
+   return c != next;
-   /* For CHARNC (caseless character) we must check the other case. If we have
+   /* 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. */
-   case OP_CHARNC:
+   case OP_CHARI:
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+   GETCHARTEST(c, previous);
- #endif
+ #else
-   if (item == next) return FALSE;
+   c = *previous;
- #ifdef SUPPORT_UTF8
+ #endif
-   if (utf8)
+   if (c == next) return FALSE;
+ #ifdef SUPPORT_UTF
+   if (utf)
      {
      unsigned int othercase;
      if (next < 128) othercase = cd->fcc[next]; else
-Line 2425 
 if (next >= 0) switch(op_code)
+Line 3060 
 if (next >= 0) switch(op_code)
  #else
      othercase = NOTACHAR;
  #endif
-     return (unsigned int)item != othercase;
+     return (unsigned int)c != othercase;
      }
    else
- #endif  /* SUPPORT_UTF8 */
+ #endif  /* SUPPORT_UTF */
-   return (item != cd->fcc[next]);  /* Non-UTF-8 mode */
+   return (c != TABLE_GET(next, cd->fcc, next));  /* Non-UTF-8 mode */
-   /* For OP_NOT, "item" must be a single-byte character. */
+   /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
+   opcodes are not used for multi-byte characters, because they are coded using
+   an XCLASS instead. */
    case OP_NOT:
-   if (item == next) return TRUE;
+   return (c = *previous) == next;
-   if ((options & PCRE_CASELESS) == 0) return FALSE;
- #ifdef SUPPORT_UTF8
+   case OP_NOTI:
-   if (utf8)
+   if ((c = *previous) == next) return TRUE;
+ #ifdef SUPPORT_UTF
+   if (utf)
      {
      unsigned int othercase;
      if (next < 128) othercase = cd->fcc[next]; else
-Line 2446 
 if (next >= 0) switch(op_code)
+Line 3085 
 if (next >= 0) switch(op_code)
  #else
      othercase = NOTACHAR;
  #endif
-     return (unsigned int)item == othercase;
+     return (unsigned int)c == othercase;
      }
    else
- #endif  /* SUPPORT_UTF8 */
+ #endif  /* SUPPORT_UTF */
-   return (item == cd->fcc[next]);  /* Non-UTF-8 mode */
+   return (c == TABLE_GET(next, cd->fcc, next));  /* Non-UTF-8 mode */
+   /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
+   When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
    case OP_DIGIT:
    return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
-Line 2493 
 if (next >= 0) switch(op_code)
+Line 3135 
 if (next >= 0) switch(op_code)
      case 0x202f:
      case 0x205f:
      case 0x3000:
-     return op_code != OP_HSPACE;
+     return op_code == OP_NOT_HSPACE;
      default:
-     return op_code == OP_HSPACE;
+     return op_code != OP_NOT_HSPACE;
      }
+   case OP_ANYNL:
    case OP_VSPACE:
    case OP_NOT_VSPACE:
    switch(next)
-Line 2509 
 if (next >= 0) switch(op_code)
+Line 3152 
 if (next >= 0) switch(op_code)
      case 0x85:
      case 0x2028:
      case 0x2029:
-     return op_code != OP_VSPACE;
+     return op_code == OP_NOT_VSPACE;
      default:
-     return op_code == OP_VSPACE;
+     return op_code != OP_NOT_VSPACE;
      }
+ #ifdef SUPPORT_UCP
+   case OP_PROP:
+   return check_char_prop(next, previous[0], previous[1], FALSE);
+   case OP_NOTPROP:
+   return check_char_prop(next, previous[0], previous[1], TRUE);
+ #endif
    default:
    return FALSE;
    }
- /* Handle the case when the next item is \d, \s, etc. */
+ /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
+ is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
+ generated only when PCRE_UCP is *not* set, that is, when only ASCII
+ characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
+ replaced by OP_PROP codes when PCRE_UCP is set. */
  switch(op_code)
    {
    case OP_CHAR:
-   case OP_CHARNC:
+   case OP_CHARI:
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+   GETCHARTEST(c, previous);
+ #else
+   c = *previous;
  #endif
    switch(-next)
      {
      case ESC_d:
-     return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
+     return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
      case ESC_D:
-     return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
+     return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
      case ESC_s:
-     return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
+     return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
      case ESC_S:
-     return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
+     return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
      case ESC_w:
-     return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
+     return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
      case ESC_W:
-     return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
+     return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
      case ESC_h:
      case ESC_H:
-     switch(item)
+     switch(c)
        {
        case 0x09:
        case 0x20:
-Line 2578 
 switch(op_code)
+Line 3235 
 switch(op_code)
      case ESC_v:
      case ESC_V:
-     switch(item)
+     switch(c)
        {
        case 0x0a:
        case 0x0b:
-Line 2592 
 switch(op_code)
+Line 3249 
 switch(op_code)
        return -next == ESC_v;
        }
+     /* When PCRE_UCP is set, these values get generated for \d etc. Find
+     their substitutions and process them. The result will always be either
+     -ESC_p or -ESC_P. Then fall through to process those values. */
+ #ifdef SUPPORT_UCP
+     case ESC_du:
+     case ESC_DU:
+     case ESC_wu:
+     case ESC_WU:
+     case ESC_su:
+     case ESC_SU:
+       {
+       int temperrorcode = 0;
+       ptr = substitutes[-next - ESC_DU];
+       next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
+       if (temperrorcode != 0) return FALSE;
+       ptr++;    /* For compatibility */
+       }
+     /* Fall through */
+     case ESC_p:
+     case ESC_P:
+       {
+       int ptype, pdata, errorcodeptr;
+       BOOL negated;
+       ptr--;      /* Make ptr point at the p or P */
+       ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
+       if (ptype < 0) return FALSE;
+       ptr++;      /* Point past the final curly ket */
+       /* If the property item is optional, we have to give up. (When generated
+       from \d etc by PCRE_UCP, this test will have been applied much earlier,
+       to the original \d etc. At this point, ptr will point to a zero byte. */
+       if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
+         STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
+           return FALSE;
+       /* Do the property check. */
+       return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
+       }
+ #endif
      default:
      return FALSE;
      }
+   /* In principle, support for Unicode properties should be integrated here as
+   well. It means re-organizing the above code so as to get hold of the property
+   values before switching on the op-code. However, I wonder how many patterns
+   combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
+   these op-codes are never generated.) */
    case OP_DIGIT:
    return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
-          next == -ESC_h || next == -ESC_v;
+          next == -ESC_h || next == -ESC_v || next == -ESC_R;
    case OP_NOT_DIGIT:
    return next == -ESC_d;
    case OP_WHITESPACE:
-   return next == -ESC_S || next == -ESC_d || next == -ESC_w;
+   return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
    case OP_NOT_WHITESPACE:
    return next == -ESC_s || next == -ESC_h || next == -ESC_v;
    case OP_HSPACE:
-   return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
+   return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
+          next == -ESC_w || next == -ESC_v || next == -ESC_R;
    case OP_NOT_HSPACE:
    return next == -ESC_h;
    /* Can't have \S in here because VT matches \S (Perl anomaly) */
+   case OP_ANYNL:
    case OP_VSPACE:
    return next == -ESC_V || next == -ESC_d || next == -ESC_w;
    case OP_NOT_VSPACE:
-   return next == -ESC_v;
+   return next == -ESC_v || next == -ESC_R;
    case OP_WORDCHAR:
-   return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
+   return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
+          next == -ESC_v || next == -ESC_R;
    case OP_NOT_WORDCHAR:
    return next == -ESC_w || next == -ESC_d;
-Line 2652 
 Arguments:
+Line 3363 
 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 2664 
 Returns:         TRUE on success
+Line 3376 
 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;
- uschar *previous = NULL;
+ const pcre_uchar *nestptr = NULL;
- uschar *previous_callout = NULL;
+ pcre_uchar *previous = NULL;
- uschar *save_hwm = NULL;
+ pcre_uchar *previous_callout = NULL;
- uschar classbits[32];
+ pcre_uchar *save_hwm = NULL;
+ pcre_uint8 classbits[32];
- #ifdef SUPPORT_UTF8
- BOOL class_utf8;
+ /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
- BOOL utf8 = (options & PCRE_UTF8) != 0;
+ must not do this for other options (e.g. PCRE_EXTENDED) because they may change
- uschar *class_utf8data;
+ dynamically as we process the pattern. */
- uschar *class_utf8data_base;
- uschar utf8_char[6];
+ #ifdef SUPPORT_UTF
+ /* PCRE_UTF16 has 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). */
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
+ BOOL xclass;
+ pcre_uchar *class_uchardata;
+ pcre_uchar *class_uchardata_base;
  #endif
  #ifdef PCRE_DEBUG
-Line 2714 
 greedy_non_default = greedy_default ^ 1;
+Line 3437 
 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 2741 
 for (;; ptr++)
+Line 3465 
 for (;; ptr++)
    BOOL is_quantifier;
    BOOL is_recurse;
    BOOL reset_bracount;
-   int class_charcount;
+   int class_has_8bitchar;
-   int class_lastchar;
+   int class_single_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;
+   /* If we are at the end of a nested substitution, revert to the outer level
+   string. Nesting only happens one level deep. */
+   if (c == 0 && nestptr != NULL)
+     {
+     ptr = nestptr;
+     nestptr = NULL;
+     c = *ptr;
+     }
    /* If we are in the pre-compile phase, accumulate the length used for the
    previous cycle of this loop. */
-Line 2765 
 for (;; ptr++)
+Line 3500 
 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 2787 
 for (;; ptr++)
+Line 3523 
 for (;; ptr++)
        goto FAILED;
        }
-     *lengthptr += code - last_code;
+     *lengthptr += (int)(code - last_code);
-     DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
+     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 2798 
 for (;; ptr++)
+Line 3535 
 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 2814 
 for (;; ptr++)
+Line 3551 
 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 2862 
 for (;; ptr++)
+Line 3600 
 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)
        {
-       while (*(++ptr) != 0)
+       ptr++;
+       while (*ptr != 0)
          {
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
+         ptr++;
+ #ifdef SUPPORT_UTF
+         if (utf) FORWARDCHAR(ptr);
+ #endif
          }
        if (*ptr != 0) continue;
-Line 2894 
 for (;; ptr++)
+Line 3637 
 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 2905 
 for (;; ptr++)
+Line 3648 
 for (;; ptr++)
          *errorcodeptr = ERR20;
          goto FAILED;
          }
-       *lengthptr += code - last_code;   /* To include callout length */
+       *lengthptr += (int)(code - last_code);   /* To include callout length */
        DPRINTF((">> end branch\n"));
        }
      return TRUE;
-Line 2916 
 for (;; ptr++)
+Line 3659 
 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 2990 
 for (;; ptr++)
+Line 3734 
 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 3010 
 for (;; ptr++)
+Line 3753 
 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 3021 
 for (;; ptr++)
+Line 3764 
 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. */
+     < 256 character. class_single_char will be 1 if the class contains only
+     a single character. */
-     class_charcount = 0;
+     class_has_8bitchar = 0;
-     class_lastchar = -1;
+     class_single_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
      than 256), 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;                           /* No chars >= 256 */
-     class_utf8data = code + LINK_SIZE + 2;    /* For UTF-8 items */
+     class_uchardata = code + LINK_SIZE + 2;   /* For UTF-8 items */
-     class_utf8data_base = class_utf8data;     /* For resetting in pass 1 */
+     class_uchardata_base = class_uchardata;   /* For resetting in pass 1 */
  #endif
      /* Process characters until ] is reached. By writing this as a "do" it
-Line 3047 
 for (;; ptr++)
+Line 3791 
 for (;; ptr++)
      if (c != 0) do
        {
-       const uschar *oldptr;
+       const pcre_uchar *oldptr;
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-       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). */
        if (lengthptr != NULL)
          {
-         *lengthptr += class_utf8data - class_utf8data_base;
+         *lengthptr += class_uchardata - class_uchardata_base;
-         class_utf8data = class_utf8data_base;
+         class_uchardata = class_uchardata_base;
          }
  #endif
        /* Inside \Q...\E everything is literal except \E */
-Line 3093 
 for (;; ptr++)
+Line 3838 
 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 3110 
 for (;; ptr++)
+Line 3855 
 for (;; ptr++)
            ptr++;
            }
-         posix_class = check_posix_name(ptr, tempptr - ptr);
+         posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
          if (posix_class < 0)
            {
            *errorcodeptr = ERR30;
-Line 3124 
 for (;; ptr++)
+Line 3869 
 for (;; ptr++)
          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
            posix_class = 0;
-         /* We build the bit map for the POSIX class in a chunk of local store
+         /* When PCRE_UCP is set, some of the POSIX classes are converted to
-         because we may be adding and subtracting from it, and we don't want to
+         different escape sequences that use Unicode properties. */
-         subtract bits that may be in the main map already. At the end we or the
-         result into the bit map that is being built. */
+ #ifdef SUPPORT_UCP
+         if ((options & PCRE_UCP) != 0)
+           {
+           int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
+           if (posix_substitutes[pc] != NULL)
+             {
+             nestptr = tempptr + 1;
+             ptr = posix_substitutes[pc] - 1;
+             continue;
+             }
+           }
+ #endif
+         /* In the non-UCP case, we build the bit map for the POSIX class in a
+         chunk of local store because we may be adding and subtracting from it,
+         and we don't want to subtract bits that may be in the main map already.
+         At the end we or the result into the bit map that is being built. */
          posix_class *= 3;
          /* 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 3165 
 for (;; ptr++)
+Line 3925 
 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 characters. */
+         class_has_8bitchar = 1;
+         /* Every class contains at least two characters. */
+         class_single_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.
+       case. Inside a class (and only there) it is treated as backspace. We
-       Elsewhere it marks a word boundary. Other escapes have preset maps ready
+       assume that other escapes have more than one character in them, so
-       to 'or' into the one we are building. We assume they have more than one
+       speculatively set both class_has_8bitchar and class_single_char bigger
-       character in them, so set class_charcount bigger than one. */
+       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)
          {
          c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
          if (*errorcodeptr != 0) goto FAILED;
-         if (-c == ESC_b) c = CHAR_BS;       /* \b is backspace in a class */
+         if (-c == ESC_b) c = CHAR_BS;    /* \b is backspace in a class */
-         else if (-c == ESC_X) c = CHAR_X;   /* \X is literal X in a class */
+         else if (-c == ESC_N)            /* \N is not supported in a class */
-         else if (-c == ESC_R) c = CHAR_R;   /* \R is literal R 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 3197 
 for (;; ptr++)
+Line 3965 
 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_single_char += 2;
-           /* Save time by not doing this in the pre-compile phase. */
+           switch (-c)
-           if (lengthptr == NULL) switch (-c)
              {
+ #ifdef SUPPORT_UCP
+             case ESC_du:     /* These are the values given for \d etc */
+             case ESC_DU:     /* when PCRE_UCP is set. We replace the */
+             case ESC_wu:     /* escape sequence with an appropriate \p */
+             case ESC_WU:     /* or \P to test Unicode properties instead */
+             case ESC_su:     /* of the default ASCII testing. */
+             case ESC_SU:
+             nestptr = ptr;
+             ptr = substitutes[-c - ESC_DU] - 1;  /* Just before substitute */
+             class_has_8bitchar--;                /* Undo! */
+             continue;
+ #endif
              case ESC_d:
              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
              continue;
-Line 3222 
 for (;; ptr++)
+Line 4003 
 for (;; ptr++)
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
              continue;
+             /* 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. */
              case ESC_s:
-             for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
+             classbits[0] |= cbits[cbit_space];
-             classbits[1] &= ~0x08;   /* Perl 5.004 onwards omits VT from \s */
+             classbits[1] |= cbits[cbit_space+1] & ~0x08;
+             for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
              continue;
              case ESC_S:
-Line 3233 
 for (;; ptr++)
+Line 4019 
 for (;; ptr++)
              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */
              continue;
-             default:    /* Not recognized; fall through */
+             case ESC_h:
-             break;      /* Need "default" setting to stop compiler warning. */
-             }
-           /* In the pre-compile phase, just do the recognition. */
-           else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
-                    c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
-           /* We need to deal with \H, \h, \V, and \v in both phases because
-           they use extra memory. */
-           if (-c == ESC_h)
-             {
              SETBIT(classbits, 0x09); /* VT */
              SETBIT(classbits, 0x20); /* SPACE */
              SETBIT(classbits, 0xa0); /* NSBP */
- #ifdef SUPPORT_UTF8
+ #ifndef COMPILE_PCRE8
-             if (utf8)
+             xclass = TRUE;
+             *class_uchardata++ = XCL_SINGLE;
+             *class_uchardata++ = 0x1680;
+             *class_uchardata++ = XCL_SINGLE;
+             *class_uchardata++ = 0x180e;
+             *class_uchardata++ = XCL_RANGE;
+             *class_uchardata++ = 0x2000;
+             *class_uchardata++ = 0x200a;
+             *class_uchardata++ = XCL_SINGLE;
+             *class_uchardata++ = 0x202f;
+             *class_uchardata++ = XCL_SINGLE;
+             *class_uchardata++ = 0x205f;
+             *class_uchardata++ = XCL_SINGLE;
+             *class_uchardata++ = 0x3000;
+ #elif defined SUPPORT_UTF
+             if (utf)
                {
-               class_utf8 = TRUE;
+               xclass = TRUE;
-               *class_utf8data++ = XCL_SINGLE;
+               *class_uchardata++ = XCL_SINGLE;
-               class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x1680, class_uchardata);
-               *class_utf8data++ = XCL_SINGLE;
+               *class_uchardata++ = XCL_SINGLE;
-               class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x180e, class_uchardata);
-               *class_utf8data++ = XCL_RANGE;
+               *class_uchardata++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x2000, class_uchardata);
-               class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x200a, class_uchardata);
-               *class_utf8data++ = XCL_SINGLE;
+               *class_uchardata++ = XCL_SINGLE;
-               class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x202f, class_uchardata);
-               *class_utf8data++ = XCL_SINGLE;
+               *class_uchardata++ = XCL_SINGLE;
-               class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x205f, class_uchardata);
-               *class_utf8data++ = XCL_SINGLE;
+               *class_uchardata++ = XCL_SINGLE;
-               class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x3000, class_uchardata);
                }
  #endif
              continue;
-             }
-           if (-c == ESC_H)
+             case ESC_H:
-             {
              for (c = 0; c < 32; c++)
                {
                int x = 0xff;
-Line 3286 
 for (;; ptr++)
+Line 4072 
 for (;; ptr++)
                  }
                classbits[c] |= x;
                }
+ #ifndef COMPILE_PCRE8
- #ifdef SUPPORT_UTF8
+             xclass = TRUE;
-             if (utf8)
+             *class_uchardata++ = XCL_RANGE;
+             *class_uchardata++ = 0x0100;
+             *class_uchardata++ = 0x167f;
+             *class_uchardata++ = XCL_RANGE;
+             *class_uchardata++ = 0x1681;
+             *class_uchardata++ = 0x180d;
+             *class_uchardata++ = XCL_RANGE;
+             *class_uchardata++ = 0x180f;
+             *class_uchardata++ = 0x1fff;
+             *class_uchardata++ = XCL_RANGE;
+             *class_uchardata++ = 0x200b;
+             *class_uchardata++ = 0x202e;
+             *class_uchardata++ = XCL_RANGE;
+             *class_uchardata++ = 0x2030;
+             *class_uchardata++ = 0x205e;
+             *class_uchardata++ = XCL_RANGE;
+             *class_uchardata++ = 0x2060;
+             *class_uchardata++ = 0x2fff;
+             *class_uchardata++ = XCL_RANGE;
+             *class_uchardata++ = 0x3001;
+ #ifdef SUPPORT_UTF
+             if (utf)
+               class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
+             else
+ #endif
+               *class_uchardata++ = 0xffff;
+ #elif defined SUPPORT_UTF
+             if (utf)
                {
-               class_utf8 = TRUE;
+               xclass = TRUE;
-               *class_utf8data++ = XCL_RANGE;
+               *class_uchardata++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
-               class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x167f, class_uchardata);
-               *class_utf8data++ = XCL_RANGE;
+               *class_uchardata++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x1681, class_uchardata);
-               class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x180d, class_uchardata);
-               *class_utf8data++ = XCL_RANGE;
+               *class_uchardata++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x180f, class_uchardata);
-               class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x1fff, class_uchardata);
-               *class_utf8data++ = XCL_RANGE;
+               *class_uchardata++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x200b, class_uchardata);
-               class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x202e, class_uchardata);
-               *class_utf8data++ = XCL_RANGE;
+               *class_uchardata++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x2030, class_uchardata);
-               class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x205e, class_uchardata);
-               *class_utf8data++ = XCL_RANGE;
+               *class_uchardata++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x2060, class_uchardata);
-               class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x2fff, class_uchardata);
-               *class_utf8data++ = XCL_RANGE;
+               *class_uchardata++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x3001, class_uchardata);
-               class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
                }
  #endif
              continue;
-             }
-           if (-c == ESC_v)
+             case ESC_v:
-             {
              SETBIT(classbits, 0x0a); /* LF */
              SETBIT(classbits, 0x0b); /* VT */
              SETBIT(classbits, 0x0c); /* FF */
              SETBIT(classbits, 0x0d); /* CR */
              SETBIT(classbits, 0x85); /* NEL */
- #ifdef SUPPORT_UTF8
+ #ifndef COMPILE_PCRE8
-             if (utf8)
+             xclass = TRUE;
+             *class_uchardata++ = XCL_RANGE;
+             *class_uchardata++ = 0x2028;
+             *class_uchardata++ = 0x2029;
+ #elif defined SUPPORT_UTF
+             if (utf)
                {
-               class_utf8 = TRUE;
+               xclass = TRUE;
-               *class_utf8data++ = XCL_RANGE;
+               *class_uchardata++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x2028, class_uchardata);
-               class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x2029, class_uchardata);
                }
  #endif
              continue;
-             }
-           if (-c == ESC_V)
+             case ESC_V:
-             {
              for (c = 0; c < 32; c++)
                {
                int x = 0xff;
-Line 3354 
 for (;; ptr++)
+Line 4168 
 for (;; ptr++)
                classbits[c] |= x;
                }
- #ifdef SUPPORT_UTF8
+ #ifndef COMPILE_PCRE8
-             if (utf8)
+             xclass = TRUE;
+             *class_uchardata++ = XCL_RANGE;
+             *class_uchardata++ = 0x0100;
+             *class_uchardata++ = 0x2027;
+             *class_uchardata++ = XCL_RANGE;
+             *class_uchardata++ = 0x202a;
+ #ifdef SUPPORT_UTF
+             if (utf)
+               class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
+             else
+ #endif
+               *class_uchardata++ = 0xffff;
+ #elif defined SUPPORT_UTF
+             if (utf)
                {
-               class_utf8 = TRUE;
+               xclass = TRUE;
-               *class_utf8data++ = XCL_RANGE;
+               *class_uchardata++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x0100, class_uchardata);
-               class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x2027, class_uchardata);
-               *class_utf8data++ = XCL_RANGE;
+               *class_uchardata++ = XCL_RANGE;
-               class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x202a, class_uchardata);
-               class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(0x10ffff, class_uchardata);
                }
  #endif
              continue;
-             }
-           /* We need to deal with \P and \p in both phases. */
  #ifdef SUPPORT_UCP
-           if (-c == ESC_p || -c == ESC_P)
+             case ESC_p:
-             {
+             case ESC_P:
-             BOOL negated;
+               {
-             int pdata;
+               BOOL negated;
-             int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
+               int pdata;
-             if (ptype < 0) goto FAILED;
+               int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
-             class_utf8 = TRUE;
+               if (ptype < 0) goto FAILED;
-             *class_utf8data++ = ((-c == ESC_p) != negated)?
+               xclass = TRUE;
-               XCL_PROP : XCL_NOTPROP;
+               *class_uchardata++ = ((-c == ESC_p) != negated)?
-             *class_utf8data++ = ptype;
+                 XCL_PROP : XCL_NOTPROP;
-             *class_utf8data++ = pdata;
+               *class_uchardata++ = ptype;
-             class_charcount -= 2;   /* Not a < 256 character */
+               *class_uchardata++ = pdata;
-             continue;
+               class_has_8bitchar--;                /* Undo! */
-             }
+               continue;
+               }
  #endif
-           /* Unrecognized escapes are faulted if PCRE is running in its
+             /* Unrecognized escapes are faulted if PCRE is running in its
-           strict mode. By default, for compatibility with Perl, they are
+             strict mode. By default, for compatibility with Perl, they are
-           treated as literals. */
+             treated as literals. */
-           if ((options & PCRE_EXTRA) != 0)
+             default:
-             {
+             if ((options & PCRE_EXTRA) != 0)
-             *errorcodeptr = ERR7;
+               {
-             goto FAILED;
+               *errorcodeptr = ERR7;
+               goto FAILED;
+               }
+             class_has_8bitchar--;    /* Undo the speculative increase. */
+             class_single_char -= 2;  /* Undo the speculative increase. */
+             c = *ptr;                /* Get the final character and fall through */
+             break;
              }
-           class_charcount -= 2;  /* Undo the default count from above */
-           c = *ptr;              /* Get the final character and fall through */
            }
          /* Fall through if we have a single character (c >= 0). This may be
-         greater than 256 in UTF-8 mode. */
+         greater than 256. */
          }   /* End of backslash handling */
-Line 3450 
 for (;; ptr++)
+Line 4278 
 for (;; ptr++)
            goto LONE_SINGLE_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 3468 
 for (;; ptr++)
+Line 4296 
 for (;; ptr++)
            d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
            if (*errorcodeptr != 0) goto FAILED;
-           /* \b is backspace; \X is literal X; \R is literal R; any other
+           /* \b is backspace; any other special means the '-' was literal */
-           special means the '-' was literal */
            if (d < 0)
              {
-             if (d == -ESC_b) d = CHAR_BS;
+             if (d == -ESC_b) d = CHAR_BS; else
-             else if (d == -ESC_X) d = CHAR_X;
-             else if (d == -ESC_R) d = CHAR_R; else
                {
                ptr = oldptr;
                goto LONE_SINGLE_CHARACTER;  /* A few lines below */
-Line 3498 
 for (;; ptr++)
+Line 4323 
 for (;; ptr++)
          if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
+         /* Since we found a character range, single character optimizations
+         cannot be done anymore. */
+         class_single_char = 2;
          /* 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 defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
-         if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
+         if ((d > 255) || (utf && ((options & PCRE_CASELESS) != 0 && d > 127)))
+ #elif defined  SUPPORT_UTF
+         if (utf && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
+ #elif !(defined COMPILE_PCRE8)
+         if (d > 255)
+ #endif
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
            {
-           class_utf8 = TRUE;
+           xclass = 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
+ #ifndef COMPILE_PCRE8
+           if (utf && (options & PCRE_CASELESS) != 0)
+ #else
            if ((options & PCRE_CASELESS) != 0)
+ #endif
              {
              unsigned int occ, ocd;
              unsigned int cc = c;
-Line 3539 
 for (;; ptr++)
+Line 4378 
 for (;; ptr++)
                if (occ == ocd)
                  {
-                 *class_utf8data++ = XCL_SINGLE;
+                 *class_uchardata++ = XCL_SINGLE;
                  }
                else
                  {
-                 *class_utf8data++ = XCL_RANGE;
+                 *class_uchardata++ = XCL_RANGE;
-                 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
+                 class_uchardata += PRIV(ord2utf)(occ, class_uchardata);
                  }
-               class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
+               class_uchardata += PRIV(ord2utf)(ocd, class_uchardata);
                }
              }
  #endif  /* SUPPORT_UCP */
-Line 3554 
 for (;; ptr++)
+Line 4393 
 for (;; ptr++)
            /* Now record the original range, possibly modified for UCP caseless
            overlapping ranges. */
-           *class_utf8data++ = XCL_RANGE;
+           *class_uchardata++ = XCL_RANGE;
-           class_utf8data += _pcre_ord2utf8(c, class_utf8data);
+ #ifdef SUPPORT_UTF
-           class_utf8data += _pcre_ord2utf8(d, class_utf8data);
+ #ifndef COMPILE_PCRE8
+           if (utf)
+             {
+             class_uchardata += PRIV(ord2utf)(c, class_uchardata);
+             class_uchardata += PRIV(ord2utf)(d, class_uchardata);
+             }
+           else
+             {
+             *class_uchardata++ = c;
+             *class_uchardata++ = d;
+             }
+ #else
+           class_uchardata += PRIV(ord2utf)(c, class_uchardata);
+           class_uchardata += PRIV(ord2utf)(d, class_uchardata);
+ #endif
+ #else /* SUPPORT_UTF */
+           *class_uchardata++ = c;
+           *class_uchardata++ = d;
+ #endif /* SUPPORT_UTF */
            /* 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
+           caseless matching for UTF characters > 127; we can use the bit map
-           for the smaller ones. */
+           for the smaller ones. As for 16 bit characters without UTF, we
+           can still use  */
  #ifdef SUPPORT_UCP
-           continue;    /* With next character in the class */
+ #ifndef COMPILE_PCRE8
- #else
+           if (utf)
-           if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
+ #endif
+             continue;    /* With next character in the class */
+ #endif  /* SUPPORT_UCP */
+ #if defined SUPPORT_UTF && !defined(SUPPORT_UCP) && !(defined COMPILE_PCRE8)
+           if (utf)
+             {
+             if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
+             /* Adjust upper limit and fall through to set up the map */
+             d = 127;
+             }
+           else
+             {
+             if (c > 255) continue;
+             /* Adjust upper limit and fall through to set up the map */
+             d = 255;
+             }
+ #elif defined SUPPORT_UTF && !defined(SUPPORT_UCP)
+           if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
            /* Adjust upper limit and fall through to set up the map */
            d = 127;
+ #else
- #endif  /* SUPPORT_UCP */
+           if (c > 255) continue;
+           /* Adjust upper limit and fall through to set up the map */
+           d = 255;
+ #endif  /* SUPPORT_UTF && !SUPPORT_UCP && !COMPILE_PCRE8 */
            }
- #endif  /* SUPPORT_UTF8 */
+ #endif  /* SUPPORT_UTF || !COMPILE_PCRE8 */
-         /* We use the bit map for all cases when not in UTF-8 mode; else
+         /* We use the bit map for 8 bit mode, or when the characters fall
-         ranges that lie entirely within 0-127 when there is UCP support; else
+         partially or entirely to [0-255] ([0-127] for UCP) ranges. */
-         for partial ranges without UCP support. */
-         class_charcount += d - c + 1;
+         class_has_8bitchar = 1;
-         class_lastchar = d;
          /* We can save a bit of time by skipping this in the pre-compile. */
-Line 3589 
 for (;; ptr++)
+Line 4464 
 for (;; ptr++)
            classbits[c/8] |= (1 << (c&7));
            if ((options & PCRE_CASELESS) != 0)
              {
-             int uc = cd->fcc[c];           /* flip case */
+             int uc = cd->fcc[c]; /* flip case */
              classbits[uc/8] |= (1 << (uc&7));
              }
            }
-Line 3601 
 for (;; ptr++)
+Line 4476 
 for (;; ptr++)
        non-escape char, or after \ that introduces a single character or for an
        apparent range that isn't. */
        LONE_SINGLE_CHARACTER:
+       /* Only the value of 1 matters for class_single_char. */
+       if (class_single_char < 2) class_single_char++;
+       /* If class_charcount is 1, we saw precisely one character. As long as
+       there were no negated characters >= 128 and there was no use of \p or \P,
+       in other words, no use of any XCLASS features, we can optimize.
+       In UTF-8 mode, we can optimize the negative case only if there were no
+       characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
+       operate on single-bytes characters only. This is an historical hangover.
+       Maybe one day we can tidy these opcodes to handle multi-byte characters.
+       The optimization throws away the bit map. We turn the item into a
+-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
+       Note that OP_NOT[I] does not support multibyte characters. 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_UTF
+       if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET
+         && (!utf || !negate_class || c < (MAX_VALUE_FOR_SINGLE_CHAR + 1)))
+ #else
+       if (class_single_char == 1 && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
+ #endif
+         {
+         ptr++;
+         zeroreqchar = reqchar;
+         /* The OP_NOT[I] opcodes work on single characters only. */
+         if (negate_class)
+           {
+           if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
+           zerofirstchar = firstchar;
+           *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
+           *code++ = c;
+           goto NOT_CHAR;
+           }
+         /* For a single, positive character, get the value into mcbuffer, and
+         then we can handle this with the normal one-character code. */
+ #ifdef SUPPORT_UTF
+         if (utf && c > MAX_VALUE_FOR_SINGLE_CHAR)
+           mclength = PRIV(ord2utf)(c, mcbuffer);
+         else
+ #endif
+           {
+           mcbuffer[0] = c;
+           mclength = 1;
+           }
+         goto ONE_CHAR;
+         }       /* End of 1-char optimization */
-       /* Handle a character that cannot go in the bit map */
+       /* Handle a character that cannot go in the bit map. */
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF && !(defined COMPILE_PCRE8)
-       if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
+       if ((c > 255) || (utf && ((options & PCRE_CASELESS) != 0 && c > 127)))
-         {
+ #elif defined SUPPORT_UTF
-         class_utf8 = TRUE;
+       if (utf && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
-         *class_utf8data++ = XCL_SINGLE;
+ #elif !(defined COMPILE_PCRE8)
-         class_utf8data += _pcre_ord2utf8(c, class_utf8data);
+       if (c > 255)
+ #endif
+ #if defined SUPPORT_UTF || !(defined COMPILE_PCRE8)
+         {
+         xclass = TRUE;
+         *class_uchardata++ = XCL_SINGLE;
+ #ifdef SUPPORT_UTF
+ #ifndef COMPILE_PCRE8
+         /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
+         if (!utf)
+           *class_uchardata++ = c;
+         else
+ #endif
+           class_uchardata += PRIV(ord2utf)(c, class_uchardata);
+ #else /* SUPPORT_UTF */
+         *class_uchardata++ = c;
+ #endif /* SUPPORT_UTF */
  #ifdef SUPPORT_UCP
+ #ifdef COMPILE_PCRE8
          if ((options & PCRE_CASELESS) != 0)
+ #else
+         /* In non 8 bit mode, we can get here even if we are not in UTF mode. */
+         if (utf && (options & PCRE_CASELESS) != 0)
+ #endif
            {
            unsigned int othercase;
            if ((othercase = UCD_OTHERCASE(c)) != c)
              {
-             *class_utf8data++ = XCL_SINGLE;
+             *class_uchardata++ = XCL_SINGLE;
-             class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
+             class_uchardata += PRIV(ord2utf)(othercase, class_uchardata);
              }
            }
  #endif  /* SUPPORT_UCP */
          }
        else
- #endif  /* SUPPORT_UTF8 */
+ #endif  /* SUPPORT_UTF || COMPILE_PCRE16 */
        /* Handle a single-byte character */
          {
+         class_has_8bitchar = 1;
          classbits[c/8] |= (1 << (c&7));
          if ((options & PCRE_CASELESS) != 0)
            {
-           c = cd->fcc[c];   /* flip case */
+           c = cd->fcc[c]; /* flip case */
            classbits[c/8] |= (1 << (c&7));
            }
-         class_charcount++;
-         class_lastchar = c;
          }
        }
-     /* Loop until ']' reached. This "while" is the end of the "do" above. */
+     /* Loop until ']' reached. This "while" is the end of the "do" far above.
+     If we are at the end of an internal nested string, revert to the outer
+     string. */
+     while (((c = *(++ptr)) != 0 ||
+            (nestptr != NULL &&
+              (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
+            (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
-     while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
+     /* Check for missing terminating ']' */
-     if (c == 0)                          /* Missing terminating ']' */
+     if (c == 0)
        {
        *errorcodeptr = ERR6;
        goto FAILED;
        }
+     /* If this is the first thing in the branch, there can be no first char
- /* This code has been disabled because it would mean that \s counts as
+     setting, whatever the repeat count. Any reqchar setting must remain
- an explicit \r or \n reference, and that's not really what is wanted. Now
+     unchanged after any kind of repeat. */
- we set the flag only if there is a literal "\r" or "\n" in the class. */
+     if (firstchar == REQ_UNSET) firstchar = REQ_NONE;
- #if 0
+     zerofirstchar = firstchar;
-     /* Remember whether \r or \n are in this class */
+     zeroreqchar = reqchar;
-     if (negate_class)
-       {
-       if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
-       }
-     else
-       {
-       if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
-       }
- #endif
-     /* If class_charcount is 1, we saw precisely one character whose value is
-     less than 256. As long as there were no characters >= 128 and there was no
-     use of \p or \P, in other words, no use of any XCLASS features, we can
-     optimize.
-     In UTF-8 mode, we can optimize the negative case only if there were no
-     characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
-     operate on single-bytes only. This is an historical hangover. Maybe one day
-     we can tidy these opcodes to handle multi-byte characters.
-     The optimization throws away the bit map. We turn the item into a
--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
-     can cause firstbyte to be set. Otherwise, there can be no first char if
-     this item is first, whatever repeat count may follow. In the case of
-     reqbyte, save the previous value for reinstating. */
- #ifdef SUPPORT_UTF8
-     if (class_charcount == 1 && !class_utf8 &&
-       (!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
-     such as \S in the class, because in that case all characters > 255 are in
+     such as \S in the class, and PCRE_UCP is not set, because in that case all
-     the class, so any that were explicitly given as well can be ignored. If
+     characters > 255 are in the class, so any that were explicitly given as
-     (when there are explicit characters > 255 that must be listed) there are no
+     well can be ignored. If (when there are explicit characters > 255 that must
-     characters < 256, we can omit the bitmap in the actual compiled code. */
+     be listed) there are no characters < 256, we can omit the bitmap in the
+     actual compiled code. */
- #ifdef SUPPORT_UTF8
-     if (class_utf8 && !should_flip_negation)
+ #ifdef SUPPORT_UTF
+     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
-     /* If there are no characters > 255, set the opcode to OP_CLASS or
+     /* If there are no characters > 255, or they are all to be included or
-     OP_NCLASS, depending on whether the whole class was negated and whether
+     excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
-     there were negative specials such as \S in the class. Then copy the 32-byte
+     whole class was negated and whether there were negative specials such as \S
-     map into the code vector, negating it if necessary. */
+     (non-UCP) in the class. Then copy the 32-byte map into the code vector,
+     negating it if necessary. */
      *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
-     if (negate_class)
+     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);
+     NOT_CHAR:
      break;
-Line 3817 
 we set the flag only if there is a liter
+Line 4708 
 we set the flag only if there is a liter
      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 3828 
 we set the flag only if there is a liter
+Line 4719 
 we set the flag only if there is a liter
      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 3852 
 we set the flag only if there is a liter
+Line 4743 
 we set the flag only if there is a liter
        }
      else repeat_type = greedy_default;
+     /* If previous was a recursion call, wrap it in atomic brackets so that
+     previous becomes the atomic group. All recursions were so wrapped in the
+     past, but it no longer happens for non-repeated recursions. In fact, the
+     repeated ones could be re-implemented independently so as not to need this,
+     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 was a character match, abolish the item and generate a
      repeat item instead. If a char item has a minumum of more than one, ensure
-     that it is set in reqbyte - it might not be if a sequence such as x{3} is
+     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 firstbyte
+     the first thing in a branch because the x will have gone into firstchar
      instead.  */
-     if (*previous == OP_CHAR || *previous == OP_CHARNC)
+     if (*previous == OP_CHAR || *previous == OP_CHARI)
        {
-       /* Deal with UTF-8 characters that take up more than one byte. It's
+       op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
+       /* 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
+ #ifdef SUPPORT_UTF
-       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 (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 3892 
 we set the flag only if there is a liter
+Line 4813 
 we set the flag only if there is a liter
        if (!possessive_quantifier &&
            repeat_max < 0 &&
-           check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
+           check_auto_possessive(previous, utf, ptr + 1, options, cd))
-             options, cd))
          {
          repeat_type = 0;    /* Force greedy */
          possessive_quantifier = TRUE;
-Line 3905 
 we set the flag only if there is a liter
+Line 4825 
 we set the flag only if there is a liter
      /* 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
+     repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
-     currently used only for single-byte chars. */
+     are currently used only for single-byte chars. */
-     else if (*previous == OP_NOT)
+     else if (*previous == OP_NOT || *previous == OP_NOTI)
        {
-       op_type = OP_NOTSTAR - OP_STAR;  /* Use "not" opcodes */
+       op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
        c = previous[1];
        if (!possessive_quantifier &&
            repeat_max < 0 &&
-           check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
+           check_auto_possessive(previous, utf, ptr + 1, options, cd))
          {
          repeat_type = 0;    /* Force greedy */
          possessive_quantifier = TRUE;
-Line 3931 
 we set the flag only if there is a liter
+Line 4851 
 we set the flag only if there is a liter
      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;