/tags/beta3/harbour/source/hbpcre/pcrecomp.c

/*************************************************
*      Perl-Compatible Regular Expressions       *
*************************************************/

/* PCRE is a library of functions to support regular expressions whose syntax
and semantics are as close as possible to those of the Perl 5 language.

                       Written by Philip Hazel
           Copyright (c) 1997-2005 University of Cambridge

-----------------------------------------------------------------------------
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

    * Redistributions of source code must retain the above copyright notice,
      this list of conditions and the following disclaimer.

    * Redistributions in binary form must reproduce the above copyright
      notice, this list of conditions and the following disclaimer in the
      documentation and/or other materials provided with the distribution.

    * Neither the name of the University of Cambridge nor the names of its
      contributors may be used to endorse or promote products derived from
      this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
POSSIBILITY OF SUCH DAMAGE.
-----------------------------------------------------------------------------
*/


/* This module contains the external function pcre_compile(), along with
supporting internal functions that are not used by other modules. */


#include "pcreinal.h"


/*************************************************
*      Code parameters and static tables         *
*************************************************/

/* Maximum number of items on the nested bracket stacks at compile time. This
applies to the nesting of all kinds of parentheses. It does not limit
un-nested, non-capturing parentheses. This number can be made bigger if
necessary - it is used to dimension one int and one unsigned char vector at
compile time. */

#define BRASTACK_SIZE 200


/* Table for handling escaped characters in the range '0'-'z'. Positive returns
are simple data values; negative values are for special things like \d and so
on. Zero means further processing is needed (for things like \x), or the escape
is invalid. */

#if !EBCDIC   /* This is the "normal" table for ASCII systems */
static const short int escapes[] = {
     0,      0,      0,      0,      0,      0,      0,      0,   /* 0 - 7 */
     0,      0,    ':',    ';',    '<',    '=',    '>',    '?',   /* 8 - ? */
   '@', -ESC_A, -ESC_B, -ESC_C, -ESC_D, -ESC_E,      0, -ESC_G,   /* @ - G */
     0,      0,      0,      0,      0,      0,      0,      0,   /* H - O */
-ESC_P, -ESC_Q,      0, -ESC_S,      0,      0,      0, -ESC_W,   /* P - W */
-ESC_X,      0, -ESC_Z,    '[',   '\\',    ']',    '^',    '_',   /* X - _ */
   '`',      7, -ESC_b,      0, -ESC_d,  ESC_e,  ESC_f,      0,   /* ` - g */
     0,      0,      0,      0,      0,      0,  ESC_n,      0,   /* h - o */
-ESC_p,      0,  ESC_r, -ESC_s,  ESC_tee,    0,      0, -ESC_w,   /* p - w */
     0,      0, -ESC_z                                            /* x - z */
};

#else         /* This is the "abnormal" table for EBCDIC systems */
static const short int escapes[] = {
/*  48 */     0,     0,      0,     '.',    '<',   '(',    '+',    '|',
/*  50 */   '&',     0,      0,       0,      0,     0,      0,      0,
/*  58 */     0,     0,    '!',     '$',    '*',   ')',    ';',    '~',
/*  60 */   '-',   '/',      0,       0,      0,     0,      0,      0,
/*  68 */     0,     0,    '|',     ',',    '%',   '_',    '>',    '?',
/*  70 */     0,     0,      0,       0,      0,     0,      0,      0,
/*  78 */     0,   '`',    ':',     '#',    '@',  '\'',    '=',    '"',
/*  80 */     0,     7, -ESC_b,       0, -ESC_d, ESC_e,  ESC_f,      0,
/*  88 */     0,     0,      0,     '{',      0,     0,      0,      0,
/*  90 */     0,     0,      0,     'l',      0, ESC_n,      0, -ESC_p,
/*  98 */     0, ESC_r,      0,     '}',      0,     0,      0,      0,
/*  A0 */     0,   '~', -ESC_s, ESC_tee,      0,     0, -ESC_w,      0,
/*  A8 */     0,-ESC_z,      0,       0,      0,   '[',      0,      0,
/*  B0 */     0,     0,      0,       0,      0,     0,      0,      0,
/*  B8 */     0,     0,      0,       0,      0,   ']',    '=',    '-',
/*  C0 */   '{',-ESC_A, -ESC_B,  -ESC_C, -ESC_D,-ESC_E,      0, -ESC_G,
/*  C8 */     0,     0,      0,       0,      0,     0,      0,      0,
/*  D0 */   '}',     0,      0,       0,      0,     0,      0, -ESC_P,
/*  D8 */-ESC_Q,     0,      0,       0,      0,     0,      0,      0,
/*  E0 */  '\\',     0, -ESC_S,       0,      0,     0, -ESC_W, -ESC_X,
/*  E8 */     0,-ESC_Z,      0,       0,      0,     0,      0,      0,
/*  F0 */     0,     0,      0,       0,      0,     0,      0,      0,
/*  F8 */     0,     0,      0,       0,      0,     0,      0,      0
};
#endif


/* Tables of names of POSIX character classes and their lengths. The list is
terminated by a zero length entry. The first three must be alpha, upper, lower,
as this is assumed for handling case independence. */

static const char *const posix_names[] = {
  "alpha", "lower", "upper",
  "alnum", "ascii", "blank", "cntrl", "digit", "graph",
  "print", "punct", "space", "word",  "xdigit" };

static const uschar posix_name_lengths[] = {
  5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };

/* Table of class bit maps for each POSIX class; up to three may be combined
to form the class. The table for [:blank:] is dynamically modified to remove
the vertical space characters. */

static const int posix_class_maps[] = {
  cbit_lower, cbit_upper, -1,             /* alpha */
  cbit_lower, -1,         -1,             /* lower */
  cbit_upper, -1,         -1,             /* upper */
  cbit_digit, cbit_lower, cbit_upper,     /* alnum */
  cbit_print, cbit_cntrl, -1,             /* ascii */
  cbit_space, -1,         -1,             /* blank - a GNU extension */
  cbit_cntrl, -1,         -1,             /* cntrl */
  cbit_digit, -1,         -1,             /* digit */
  cbit_graph, -1,         -1,             /* graph */
  cbit_print, -1,         -1,             /* print */
  cbit_punct, -1,         -1,             /* punct */
  cbit_space, -1,         -1,             /* space */
  cbit_word,  -1,         -1,             /* word - a Perl extension */
  cbit_xdigit,-1,         -1              /* xdigit */
};


/* The texts of compile-time error messages. These are "char *" because they
are passed to the outside world. */

static const char *error_texts[] = {
  "no error",
  "\\ at end of pattern",
  "\\c at end of pattern",
  "unrecognized character follows \\",
  "numbers out of order in {} quantifier",
  /* 5 */
  "number too big in {} quantifier",
  "missing terminating ] for character class",
  "invalid escape sequence in character class",
  "range out of order in character class",
  "nothing to repeat",
  /* 10 */
  "operand of unlimited repeat could match the empty string",
  "internal error: unexpected repeat",
  "unrecognized character after (?",
  "POSIX named classes are supported only within a class",
  "missing )",
  /* 15 */
  "reference to non-existent subpattern",
  "erroffset passed as NULL",
  "unknown option bit(s) set",
  "missing ) after comment",
  "parentheses nested too deeply",
  /* 20 */
  "regular expression too large",
  "failed to get memory",
  "unmatched parentheses",
  "internal error: code overflow",
  "unrecognized character after (?<",
  /* 25 */
  "lookbehind assertion is not fixed length",
  "malformed number after (?(",
  "conditional group contains more than two branches",
  "assertion expected after (?(",
  "(?R or (?digits must be followed by )",
  /* 30 */
  "unknown POSIX class name",
  "POSIX collating elements are not supported",
  "this version of PCRE is not compiled with PCRE_UTF8 support",
  "spare error",
  "character value in \\x{...} sequence is too large",
  /* 35 */
  "invalid condition (?(0)",
  "\\C not allowed in lookbehind assertion",
  "PCRE does not support \\L, \\l, \\N, \\U, or \\u",
  "number after (?C is > 255",
  "closing ) for (?C expected",
  /* 40 */
  "recursive call could loop indefinitely",
  "unrecognized character after (?P",
  "syntax error after (?P",
  "two named groups have the same name",
  "invalid UTF-8 string",
  /* 45 */
  "support for \\P, \\p, and \\X has not been compiled",
  "malformed \\P or \\p sequence",
  "unknown property name after \\P or \\p"
};


/* 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
may mark arbitrary characters as digits - but the PCRE compiling code expects
to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
a private table here. It costs 256 bytes, but it is a lot faster than doing
character value tests (at least in some simple cases I timed), and in some
applications one wants PCRE to compile efficiently as well as match
efficiently.

For convenience, we use the same bit definitions as in chartables:

  0x04   decimal digit
  0x08   hexadecimal digit

Then we can use ctype_digit and ctype_xdigit in the code. */

#if !EBCDIC    /* This is the "normal" case, for ASCII systems */
static const unsigned char digitab[] =
  {
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   8- 15 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  16- 23 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  24- 31 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - '  */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  ( - /  */
  0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /*  0 - 7  */
  0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /*  8 - ?  */
  0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /*  @ - G  */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  H - O  */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  P - W  */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  X - _  */
  0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /*  ` - g  */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  h - o  */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  p - w  */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  x -127 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */

#else          /* This is the "abnormal" case, for EBCDIC systems */
static const unsigned char digitab[] =
  {
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7  0 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   8- 15    */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  16- 23 10 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  24- 31    */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  32- 39 20 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  40- 47    */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  48- 55 30 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  56- 63    */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 40 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  72- |     */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 50 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  88- ?     */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 60 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ?     */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- "     */
  0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g  80 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  h -143    */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p  90 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  q -159    */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x  A0 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  y -175    */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  ^ -183 B0 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191    */
  0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /*  { - G  C0 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  H -207    */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  } - P  D0 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  Q -223    */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  \ - X  E0 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  Y -239    */
  0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /*  0 - 7  F0 */
  0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/*  8 -255    */

static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
  0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /*   0-  7 */
  0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /*   8- 15 */
  0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /*  16- 23 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  24- 31 */
  0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /*  32- 39 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  40- 47 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  48- 55 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  56- 63 */
  0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 */
  0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /*  72- |  */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 */
  0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /*  88- ?  */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 */
  0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ?  */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
  0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- "  */
  0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g  */
  0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /*  h -143 */
  0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p  */
  0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /*  q -159 */
  0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x  */
  0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /*  y -175 */
  0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  ^ -183 */
  0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
  0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /*  { - G  */
  0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /*  H -207 */
  0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /*  } - P  */
  0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /*  Q -223 */
  0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /*  \ - X  */
  0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /*  Y -239 */
  0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /*  0 - 7  */
  0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/*  8 -255 */
#endif


/* Definition to allow mutual recursion */

static BOOL
  compile_regex(int, int, int *, uschar **, const uschar **, int *, BOOL, int,
    int *, int *, branch_chain *, compile_data *);



/*************************************************
*            Handle escapes                      *
*************************************************/

/* This function is called when a \ has been encountered. It either returns a
positive value for a simple escape such as \n, or a negative value which
encodes one of the more complicated things such as \d. When UTF-8 is enabled,
a positive value greater than 255 may be returned. On entry, ptr is pointing at
the \. On exit, it is on the final character of the escape sequence.

Arguments:
  ptrptr         points to the pattern position pointer
  errorcodeptr   points to the errorcode variable
  bracount       number of previous extracting brackets
  options        the options bits
  isclass        TRUE if inside a character class

Returns:         zero or positive => a data character
                 negative => a special escape sequence
                 on error, errorptr is set
*/

static int
check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
  int options, BOOL isclass)
{
const uschar *ptr = *ptrptr;
int c, i;

/* If backslash is at the end of the pattern, it's an error. */

c = *(++ptr);
if (c == 0) *errorcodeptr = ERR1;

/* Non-alphamerics are literals. For digits or letters, do an initial lookup in
a table. A non-zero result is something that can be returned immediately.
Otherwise further processing may be required. */

#if !EBCDIC    /* ASCII coding */
else if (c < '0' || c > 'z') {}                           /* Not alphameric */
else if ((i = escapes[c - '0']) != 0) c = i;

#else          /* EBCDIC coding */
else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphameric */
else if ((i = escapes[c - 0x48]) != 0)  c = i;
#endif

/* Escapes that need further processing, or are illegal. */

else
  {
  const uschar *oldptr;
  switch (c)
    {
    /* A number of Perl escapes are not handled by PCRE. We give an explicit
    error. */

    case 'l':
    case 'L':
    case 'N':
    case 'u':
    case 'U':
    *errorcodeptr = ERR37;
    break;

    /* The handling of escape sequences consisting of a string of digits
    starting with one that is not zero is not straightforward. By experiment,
    the way Perl works seems to be as follows:

    Outside a character class, the digits are read as a decimal number. If the
    number is less than 10, or if there are that many previous extracting
    left brackets, then it is a back reference. Otherwise, up to three octal
    digits are read to form an escaped byte. Thus \123 is likely to be octal
    123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
    value is greater than 377, the least significant 8 bits are taken. Inside a
    character class, \ followed by a digit is always an octal number. */

    case '1': case '2': case '3': case '4': case '5':
    case '6': case '7': case '8': case '9':

    if (!isclass)
      {
      oldptr = ptr;
      c -= '0';
      while ((digitab[ptr[1]] & ctype_digit) != 0)
        c = c * 10 + *(++ptr) - '0';
      if (c < 10 || c <= bracount)
        {
        c = -(ESC_REF + c);
        break;
        }
      ptr = oldptr;      /* Put the pointer back and fall through */
      }

    /* Handle an octal number following \. If the first digit is 8 or 9, Perl
    generates a binary zero byte and treats the digit as a following literal.
    Thus we have to pull back the pointer by one. */

    if ((c = *ptr) >= '8')
      {
      ptr--;
      c = 0;
      break;
      }

    /* \0 always starts an octal number, but we may drop through to here with a
    larger first octal digit. */

    case '0':
    c -= '0';
    while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7')
        c = c * 8 + *(++ptr) - '0';
    c &= 255;     /* Take least significant 8 bits */
    break;

    /* \x is complicated when UTF-8 is enabled. \x{ddd} is a character number
    which can be greater than 0xff, but only if the ddd are hex digits. */

    case 'x':
#ifdef SUPPORT_UTF8
    if (ptr[1] == '{' && (options & PCRE_UTF8) != 0)
      {
      const uschar *pt = ptr + 2;
      register int count = 0;
      c = 0;
      while ((digitab[*pt] & ctype_xdigit) != 0)
        {
        int cc = *pt++;
        count++;
#if !EBCDIC    /* ASCII coding */
        if (cc >= 'a') cc -= 32;               /* Convert to upper case */
        c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));
#else          /* EBCDIC coding */
        if (cc >= 'a' && cc <= 'z') cc += 64;  /* Convert to upper case */
        c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));
#endif
        }
      if (*pt == '}')
        {
        if (c < 0 || count > 8) *errorcodeptr = ERR34;
        ptr = pt;
        break;
        }
      /* If the sequence of hex digits does not end with '}', then we don't
      recognize this construct; fall through to the normal \x handling. */
      }
#endif

    /* Read just a single hex char */

    c = 0;
    while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
      {
      int cc;                               /* Some compilers don't like ++ */
      cc = *(++ptr);                        /* in initializers */
#if !EBCDIC    /* ASCII coding */
      if (cc >= 'a') cc -= 32;              /* Convert to upper case */
      c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));
#else          /* EBCDIC coding */
      if (cc <= 'z') cc += 64;              /* Convert to upper case */
      c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));
#endif
      }
    break;

    /* Other special escapes not starting with a digit are straightforward */

    case 'c':
    c = *(++ptr);
    if (c == 0)
      {
      *errorcodeptr = ERR2;
      return 0;
      }

    /* A letter is upper-cased; then the 0x40 bit is flipped. This coding
    is ASCII-specific, but then the whole concept of \cx is ASCII-specific.
    (However, an EBCDIC equivalent has now been added.) */

#if !EBCDIC    /* ASCII coding */
    if (c >= 'a' && c <= 'z') c -= 32;
    c ^= 0x40;
#else          /* EBCDIC coding */
    if (c >= 'a' && c <= 'z') c += 64;
    c ^= 0xC0;
#endif
    break;

    /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
    other alphameric following \ is an error if PCRE_EXTRA was set; otherwise,
    for Perl compatibility, it is a literal. This code looks a bit odd, but
    there used to be some cases other than the default, and there may be again
    in future, so I haven't "optimized" it. */

    default:
    if ((options & PCRE_EXTRA) != 0) switch(c)
      {
      default:
      *errorcodeptr = ERR3;
      break;
      }
    break;
    }
  }

*ptrptr = ptr;
return c;
}



#ifdef SUPPORT_UCP
/*************************************************
*               Handle \P and \p                 *
*************************************************/

/* This function is called after \P or \p has been encountered, provided that
PCRE is compiled with support for Unicode properties. On entry, ptrptr is
pointing at the P or p. On exit, it is pointing at the final character of the
escape sequence.

Argument:
  ptrptr         points to the pattern position pointer
  negptr         points to a boolean that is set TRUE for negation else FALSE
  errorcodeptr   points to the error code variable

Returns:     value from ucp_type_table, or -1 for an invalid type
*/

static int
get_ucp(const uschar **ptrptr, BOOL *negptr, int *errorcodeptr)
{
int c, i, bot, top;
const uschar *ptr = *ptrptr;
char name[4];

*negptr = FALSE;

c = *(++ptr);
if (c == 0) goto ERROR_RETURN;

/* \P or \p can be followed by a one- or two-character name in {}, optionally
preceded by ^ for negation. */

if (c == '{')
  {
  if (ptr[1] == '^')
    {
    *negptr = TRUE;
    ptr++;
    }
  for (i = 0; i <= 2; i++)
    {
    c = *(++ptr);
    if (c == 0) goto ERROR_RETURN;
    if (c == '}') break;
    name[i] = c;
    }
  if (c !='}')   /* Try to distinguish error cases */
    {
    while (*(++ptr) != 0 && *ptr != '}');
    if (*ptr == '}') goto UNKNOWN_RETURN; else goto ERROR_RETURN;
    }
  name[i] = 0;
  }

/* Otherwise there is just one following character */

else
  {
  name[0] = c;
  name[1] = 0;
  }

*ptrptr = ptr;

/* Search for a recognized property name using binary chop */

bot = 0;
top = _pcre_utt_size;

while (bot < top)
  {
  i = (bot + top)/2;
  c = strcmp(name, _pcre_utt[i].name);
  if (c == 0) return _pcre_utt[i].value;
  if (c > 0) bot = i + 1; else top = i;
  }

UNKNOWN_RETURN:
*errorcodeptr = ERR47;
*ptrptr = ptr;
return -1;

ERROR_RETURN:
*errorcodeptr = ERR46;
*ptrptr = ptr;
return -1;
}
#endif




/*************************************************
*            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 == '}') return TRUE;

if (*p++ != ',') return FALSE;
if (*p == '}') return TRUE;

if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
while ((digitab[*p] & ctype_digit) != 0) p++;

return (*p == '}');
}



/*************************************************
*         Read repeat counts                     *
*************************************************/

/* Read an item of the form {n,m} and return the values. This is called only
after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
so the syntax is guaranteed to be correct, but we need to check the values.

Arguments:
  p              pointer to first char after '{'
  minp           pointer to int for min
  maxp           pointer to int for max
                 returned as -1 if no max
  errorcodeptr   points to error code variable

Returns:         pointer to '}' on success;
                 current ptr on error, with errorcodeptr set non-zero
*/

static const uschar *
read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
{
int min = 0;
int max = -1;

/* Read the minimum value and do a paranoid check: a negative value indicates
an integer overflow. */

while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0';
if (min < 0 || min > 65535)
  {
  *errorcodeptr = ERR5;
  return p;
  }

/* Read the maximum value if there is one, and again do a paranoid on its size.
Also, max must not be less than min. */

if (*p == '}') max = min; else
  {
  if (*(++p) != '}')
    {
    max = 0;
    while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0';
    if (max < 0 || max > 65535)
      {
      *errorcodeptr = ERR5;
      return p;
      }
    if (max < min)
      {
      *errorcodeptr = ERR4;
      return p;
      }
    }
  }

/* Fill in the required variables, and pass back the pointer to the terminating
'}'. */

*minp = min;
*maxp = max;
return p;
}



/*************************************************
*      Find first significant op code            *
*************************************************/

/* This is called by several functions that scan a compiled expression looking
for a fixed first character, or an anchoring op code etc. It skips over things
that do not influence this. For some calls, a change of option is important.
For some calls, it makes sense to skip negative forward and all backward
assertions, and also the \b assertion; for others it 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*
first_significant_code(const uschar *code, int *options, int optbit,
  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];
    break;

    case OP_WORD_BOUNDARY:
    case OP_NOT_WORD_BOUNDARY:
    if (!skipassert) return code;
    /* Fall through */

    case OP_CALLOUT:
    case OP_CREF:
    case OP_BRANUMBER:
    code += _pcre_OP_lengths[*code];
    break;

    default:
    return code;
    }
  }
/* Control never reaches here */
}




/*************************************************
*        Find the fixed length of a pattern      *
*************************************************/

/* Scan a pattern and compute the fixed length of subject that will match it,
if the length is fixed. This is needed for dealing with backward assertions.
In UTF8 mode, the result is in characters rather than bytes.

Arguments:
  code     points to the start of the pattern (the bracket)
  options  the compiling options

Returns:   the fixed length, or -1 if there is no fixed length,
             or -2 if \C was encountered
*/

static int
find_fixedlength(uschar *code, int options)
{
int length = -1;

register int branchlength = 0;
register uschar *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. */

for (;;)
  {
  int d;
  register int op = *cc;
  if (op >= OP_BRA) op = OP_BRA;

  switch (op)
    {
    case OP_BRA:
    case OP_ONCE:
    case OP_COND:
    d = find_fixedlength(cc, options);
    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
    call. If it's ALT it is an alternation in a nested call. If it is
    END it's the end of the outer call. All can be handled by the same code. */

    case OP_ALT:
    case OP_KET:
    case OP_KETRMAX:
    case OP_KETRMIN:
    case OP_END:
    if (length < 0) length = branchlength;
      else if (length != branchlength) return -1;
    if (*cc != OP_ALT) return length;
    cc += 1 + LINK_SIZE;
    branchlength = 0;
    break;

    /* Skip over assertive subpatterns */

    case OP_ASSERT:
    case OP_ASSERT_NOT:
    case OP_ASSERTBACK:
    case OP_ASSERTBACK_NOT:
    do cc += GET(cc, 1); while (*cc == OP_ALT);
    /* Fall through */

    /* Skip over things that don't match chars */

    case OP_REVERSE:
    case OP_BRANUMBER:
    case OP_CREF:
    case OP_OPT:
    case OP_CALLOUT:
    case OP_SOD:
    case OP_SOM:
    case OP_EOD:
    case OP_EODN:
    case OP_CIRC:
    case OP_DOLL:
    case OP_NOT_WORD_BOUNDARY:
    case OP_WORD_BOUNDARY:
    cc += _pcre_OP_lengths[*cc];
    break;

    /* Handle literal characters */

    case OP_CHAR:
    case OP_CHARNC:
    branchlength++;
    cc += 2;
#ifdef SUPPORT_UTF8
    if ((options & PCRE_UTF8) != 0)
      {
      while ((*cc & 0xc0) == 0x80) cc++;
      }
#endif
    break;

    /* Handle exact repetitions. The count is already in characters, but we
    need to skip over a multibyte character in UTF8 mode.  */

    case OP_EXACT:
    branchlength += GET2(cc,1);
    cc += 4;
#ifdef SUPPORT_UTF8
    if ((options & PCRE_UTF8) != 0)
      {
      while((*cc & 0x80) == 0x80) cc++;
      }
#endif
    break;

    case OP_TYPEEXACT:
    branchlength += GET2(cc,1);
    cc += 4;
    break;

    /* Handle single-char matchers */

    case OP_PROP:
    case OP_NOTPROP:
    cc++;
    /* Fall through */

    case OP_NOT_DIGIT:
    case OP_DIGIT:
    case OP_NOT_WHITESPACE:
    case OP_WHITESPACE:
    case OP_NOT_WORDCHAR:
    case OP_WORDCHAR:
    case OP_ANY:
    branchlength++;
    cc++;
    break;

    /* The single-byte matcher isn't allowed */

    case OP_ANYBYTE:
    return -2;

    /* Check a class for variable quantification */

#ifdef SUPPORT_UTF8
    case OP_XCLASS:
    cc += GET(cc, 1) - 33;
    /* Fall through */
#endif

    case OP_CLASS:
    case OP_NCLASS:
    cc += 33;

    switch (*cc)
      {
      case OP_CRSTAR:
      case OP_CRMINSTAR:
      case OP_CRQUERY:
      case OP_CRMINQUERY:
      return -1;

      case OP_CRRANGE:
      case OP_CRMINRANGE:
      if (GET2(cc,1) != GET2(cc,3)) return -1;
      branchlength += GET2(cc,1);
      cc += 5;
      break;

      default:
      branchlength++;
      }
    break;

    /* Anything else is variable length */

    default:
    return -1;
    }
  }
/* Control never gets here */
}




/*************************************************
*    Scan compiled regex for numbered bracket    *
*************************************************/

/* This little function scans through a compiled pattern until it finds a
capturing bracket with the given number.

Arguments:
  code        points to start of expression
  utf8        TRUE in UTF-8 mode
  number      the required bracket number

Returns:      pointer to the opcode for the bracket, or NULL if not found
*/

static const uschar *
find_bracket(const uschar *code, BOOL utf8, int number)
{
#ifndef SUPPORT_UTF8
utf8 = utf8;               /* Stop pedantic compilers complaining */
#endif

for (;;)
  {
  register int c = *code;
  if (c == OP_END) return NULL;
  else if (c > OP_BRA)
    {
    int n = c - OP_BRA;
    if (n > EXTRACT_BASIC_MAX) n = GET2(code, 2+LINK_SIZE);
    if (n == number) return (uschar *)code;
    code += _pcre_OP_lengths[OP_BRA];
    }
  else
    {
    code += _pcre_OP_lengths[c];

#ifdef SUPPORT_UTF8

    /* 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 scan along to skip the extra bytes. All opcodes are less than 128, so we
    can use relatively efficient code. */

    if (utf8) switch(c)
      {
      case OP_CHAR:
      case OP_CHARNC:
      case OP_EXACT:
      case OP_UPTO:
      case OP_MINUPTO:
      case OP_STAR:
      case OP_MINSTAR:
      case OP_PLUS:
      case OP_MINPLUS:
      case OP_QUERY:
      case OP_MINQUERY:
      while ((*code & 0xc0) == 0x80) code++;
      break;

      /* XCLASS is used for classes that cannot be represented just by a bit
      map. This includes negated single high-valued characters. The length in
      the table is zero; the actual length is stored in the compiled code. */

      case OP_XCLASS:
      code += GET(code, 1) + 1;
      break;
      }
#endif
    }
  }
}



/*************************************************
*   Scan compiled regex for recursion reference  *
*************************************************/

/* This little function scans through a compiled pattern until it finds an
instance of OP_RECURSE.

Arguments:
  code        points to start of expression
  utf8        TRUE in UTF-8 mode

Returns:      pointer to the opcode for OP_RECURSE, or NULL if not found
*/

static const uschar *
find_recurse(const uschar *code, BOOL utf8)
{
#ifndef SUPPORT_UTF8
utf8 = utf8;               /* Stop pedantic compilers complaining */
#endif

for (;;)
  {
  register int c = *code;
  if (c == OP_END) return NULL;
  else if (c == OP_RECURSE) return code;
  else if (c > OP_BRA)
    {
    code += _pcre_OP_lengths[OP_BRA];
    }
  else
    {
    code += _pcre_OP_lengths[c];

#ifdef SUPPORT_UTF8

    /* 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 scan along to skip the extra bytes. All opcodes are less than 128, so we
    can use relatively efficient code. */

    if (utf8) switch(c)
      {
      case OP_CHAR:
      case OP_CHARNC:
      case OP_EXACT:
      case OP_UPTO:
      case OP_MINUPTO:
      case OP_STAR:
      case OP_MINSTAR:
      case OP_PLUS:
      case OP_MINPLUS:
      case OP_QUERY:
      case OP_MINQUERY:
      while ((*code & 0xc0) == 0x80) code++;
      break;

      /* XCLASS is used for classes that cannot be represented just by a bit
      map. This includes negated single high-valued characters. The length in
      the table is zero; the actual length is stored in the compiled code. */

      case OP_XCLASS:
      code += GET(code, 1) + 1;
      break;
      }
#endif
    }
  }
}



/*************************************************
*    Scan compiled branch for non-emptiness      *
*************************************************/

/* This function scans through a branch of a compiled pattern to see whether it
can match the empty string or not. It is called only from could_be_empty()
below. Note that first_significant_code() skips over assertions. If we hit an
unclosed bracket, we return "empty" - this means we've struck an inner bracket
whose current branch will already have been scanned.

Arguments:
  code        points to start of search
  endcode     points to where to stop
  utf8        TRUE if in UTF8 mode

Returns:      TRUE if what is matched could be empty
*/

static BOOL
could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
{
register int c;
for (code = first_significant_code(code + 1 + LINK_SIZE, NULL, 0, TRUE);
     code < endcode;
     code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
  {
  const uschar *ccode;

  c = *code;

  if (c >= OP_BRA)
    {
    BOOL empty_branch;
    if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */

    /* Scan a closed bracket */

    empty_branch = FALSE;
    do
      {
      if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
        empty_branch = TRUE;
      code += GET(code, 1);
      }
    while (*code == OP_ALT);
    if (!empty_branch) return FALSE;   /* All branches are non-empty */
    code += 1 + LINK_SIZE;
    c = *code;
    }

  else switch (c)
    {
    /* Check for quantifiers after a class */

#ifdef SUPPORT_UTF8
    case OP_XCLASS:
    ccode = code + GET(code, 1);
    goto CHECK_CLASS_REPEAT;
#endif

    case OP_CLASS:
    case OP_NCLASS:
    ccode = code + 33;

#ifdef SUPPORT_UTF8
    CHECK_CLASS_REPEAT:
#endif

    switch (*ccode)
      {
      case OP_CRSTAR:            /* These could be empty; continue */
      case OP_CRMINSTAR:
      case OP_CRQUERY:
      case OP_CRMINQUERY:
      break;

      default:                   /* Non-repeat => class must match */
      case OP_CRPLUS:            /* These repeats aren't empty */
      case OP_CRMINPLUS:
      return FALSE;

      case OP_CRRANGE:
      case OP_CRMINRANGE:
      if (GET2(ccode, 1) > 0) return FALSE;  /* Minimum > 0 */
      break;
      }
    break;

    /* Opcodes that must match a character */

    case OP_PROP:
    case OP_NOTPROP:
    case OP_EXTUNI:
    case OP_NOT_DIGIT:
    case OP_DIGIT:
    case OP_NOT_WHITESPACE:
    case OP_WHITESPACE:
    case OP_NOT_WORDCHAR:
    case OP_WORDCHAR:
    case OP_ANY:
    case OP_ANYBYTE:
    case OP_CHAR:
    case OP_CHARNC:
    case OP_NOT:
    case OP_PLUS:
    case OP_MINPLUS:
    case OP_EXACT:
    case OP_NOTPLUS:
    case OP_NOTMINPLUS:
    case OP_NOTEXACT:
    case OP_TYPEPLUS:
    case OP_TYPEMINPLUS:
    case OP_TYPEEXACT:
    return FALSE;

    /* End of branch */

    case OP_KET:
    case OP_KETRMAX:
    case OP_KETRMIN:
    case OP_ALT:
    return TRUE;

    /* In UTF-8 mode, STAR, MINSTAR, QUERY, MINQUERY, UPTO, and MINUPTO  may be
    followed by a multibyte character */

#ifdef SUPPORT_UTF8
    case OP_STAR:
    case OP_MINSTAR:
    case OP_QUERY:
    case OP_MINQUERY:
    case OP_UPTO:
    case OP_MINUPTO:
    if (utf8) while ((code[2] & 0xc0) == 0x80) code++;
    break;
#endif
    }
  }

return TRUE;
}



/*************************************************
*    Scan compiled regex for non-emptiness       *
*************************************************/

/* This function is called to check for left recursive calls. We want to check
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.

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

Returns:      TRUE if what is matched could be empty
*/

static BOOL
could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
  BOOL utf8)
{
while (bcptr != NULL && bcptr->current >= code)
  {
  if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;
  bcptr = bcptr->outer;
  }
return TRUE;
}



/*************************************************
*           Check for POSIX class syntax         *
*************************************************/

/* This function is called when the sequence "[:" or "[." or "[=" is
encountered in a character class. It checks whether this is followed by an
optional ^ and then a sequence of letters, terminated by a matching ":]" or
".]" or "=]".

Argument:
  ptr      pointer to the initial [
  endptr   where to return the end pointer
  cd       pointer to compile data

Returns:   TRUE or FALSE
*/

static BOOL
check_posix_syntax(const uschar *ptr, const uschar **endptr, compile_data *cd)
{
int terminator;          /* Don't combine these lines; the Solaris cc */
terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
if (*(++ptr) == '^') ptr++;
while ((cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
if (*ptr == terminator && ptr[1] == ']')
  {
  *endptr = ptr;
  return TRUE;
  }
return FALSE;
}




/*************************************************
*          Check POSIX class name                *
*************************************************/

/* This function is called to check the name given in a POSIX-style class entry
such as [:alnum:].

Arguments:
  ptr        points to the first letter
  len        the length of the name

Returns:     a value representing the name, or -1 if unknown
*/

static int
check_posix_name(const uschar *ptr, int len)
{
register int yield = 0;
while (posix_name_lengths[yield] != 0)
  {
  if (len == posix_name_lengths[yield] &&
    strncmp((const char *)ptr, posix_names[yield], len) == 0) return yield;
  yield++;
  }
return -1;
}


/*************************************************
*    Adjust OP_RECURSE items in repeated group   *
*************************************************/

/* OP_RECURSE items contain an offset from the start of the regex to the group
that is referenced. This means that groups can be replicated for fixed
repetition simply by copying (because the recursion is allowed to refer to
earlier groups that are outside the current group). However, when a group is
optional (i.e. the minimum quantifier is zero), OP_BRAZERO is inserted before
it, after it has been compiled. This means that any OP_RECURSE items within it
that refer to the group itself or any contained groups have to have their
offsets adjusted. That is the job of this function. Before it is called, the
partially compiled regex must be temporarily terminated with OP_END.

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
  cd         contains pointers to tables etc.

Returns:     nothing
*/

static void
adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd)
{
uschar *ptr = group;
while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
  {
  int offset = GET(ptr, 1);
  if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
  ptr += 1 + LINK_SIZE;
  }
}



/*************************************************
*        Insert an automatic callout point       *
*************************************************/

/* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
callout points before each pattern item.

Arguments:
  code           current code pointer
  ptr            current pattern pointer
  cd             pointers to tables etc

Returns:         new code pointer
*/

static uschar *
auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
{
*code++ = OP_CALLOUT;
*code++ = 255;
PUT(code, 0, ptr - cd->start_pattern);  /* Pattern offset */
PUT(code, LINK_SIZE, 0);                /* Default length */
return code + 2*LINK_SIZE;
}



/*************************************************
*         Complete a callout item                *
*************************************************/

/* A callout item contains the length of the next item in the pattern, which
we can't fill in till after we have reached the relevant point. This is used
for both automatic and manual callouts.

Arguments:
  previous_callout   points to previous callout item
  ptr                current pattern pointer
  cd                 pointers to tables etc

Returns:             nothing
*/

static void
complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
{
int length = ptr - cd->start_pattern - GET(previous_callout, 2);
PUT(previous_callout, 2 + LINK_SIZE, length);
}



#ifdef SUPPORT_UCP
/*************************************************
*           Get othercase range                  *
*************************************************/

/* This function is passed the start and end of a class range, in UTF-8 mode
with UCP support. It searches up the characters, looking for internal ranges of
characters in the "other" case. Each call returns the next one, updating the
start address.

Arguments:
  cptr        points to starting character value; updated
  d           end value
  ocptr       where to put start of othercase range
  odptr       where to put end of othercase range

Yield:        TRUE when range returned; FALSE when no more
*/

static BOOL
get_othercase_range(int *cptr, int d, int *ocptr, int *odptr)
{
int c, chartype, othercase, next;

for (c = *cptr; c <= d; c++)
  {
  if (_pcre_ucp_findchar(c, &chartype, &othercase) == ucp_L && othercase != 0)
    break;
  }

if (c > d) return FALSE;

*ocptr = othercase;
next = othercase + 1;

for (++c; c <= d; c++)
  {
  if (_pcre_ucp_findchar(c, &chartype, &othercase) != ucp_L ||
        othercase != next)
    break;
  next++;
  }

*odptr = next - 1;
*cptr = c;

return TRUE;
}
#endif  /* SUPPORT_UCP */


/*************************************************
*           Compile one branch                   *
*************************************************/

/* Scan the pattern, compiling it into the code vector. If the options are
changed during the branch, the pointer is used to change the external options
bits.

Arguments:
  optionsptr     pointer to the option bits
  brackets       points to number of extracting brackets used
  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)
  reqbyteptr     set to the last literal character required, else < 0
  bcptr          points to current branch chain
  cd             contains pointers to tables etc.

Returns:         TRUE on success
                 FALSE, with *errorcodeptr set non-zero on error
*/

static BOOL
compile_branch(int *optionsptr, int *brackets, uschar **codeptr,
  const uschar **ptrptr, int *errorcodeptr, int *firstbyteptr,
  int *reqbyteptr, branch_chain *bcptr, compile_data *cd)
{
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;
int zeroreqbyte, zerofirstbyte;
int req_caseopt, reqvary, tempreqvary;
int condcount = 0;
int options = *optionsptr;
int after_manual_callout = 0;
register int c;
register uschar *code = *codeptr;
uschar *tempcode;
BOOL inescq = FALSE;
BOOL groupsetfirstbyte = FALSE;
const uschar *ptr = *ptrptr;
const uschar *tempptr;
uschar *previous = NULL;
uschar *previous_callout = NULL;
uschar classbits[32];

#ifdef SUPPORT_UTF8
BOOL class_utf8;
BOOL utf8 = (options & PCRE_UTF8) != 0;
uschar *class_utf8data;
uschar utf8_char[6];
#else
BOOL utf8 = FALSE;
#endif

/* Set up the default and non-default settings for greediness */

greedy_default = ((options & PCRE_UNGREEDY) != 0);
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
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
item types that can be repeated set these backoff variables appropriately. */

firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;

/* The variable req_caseopt contains either the REQ_CASELESS value or zero,
according to the current setting of the caseless flag. REQ_CASELESS is a bit
value > 255. It is added into the firstbyte or reqbyte 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;

/* Switch on next character until the end of the branch */

for (;; ptr++)
  {
  BOOL negate_class;
  BOOL possessive_quantifier;
  BOOL is_quantifier;
  int class_charcount;
  int class_lastchar;
  int newoptions;
  int recno;
  int skipbytes;
  int subreqbyte;
  int subfirstbyte;
  int mclength;
  uschar mcbuffer[8];

  /* Next byte in the pattern */

  c = *ptr;

  /* If in \Q...\E, check for the end; if not, we have a literal */

  if (inescq && c != 0)
    {
    if (c == '\\' && ptr[1] == 'E')
      {
      inescq = FALSE;
      ptr++;
      continue;
      }
    else
      {
      if (previous_callout != NULL)
        {
        complete_callout(previous_callout, ptr, cd);
        previous_callout = NULL;
        }
      if ((options & PCRE_AUTO_CALLOUT) != 0)
        {
        previous_callout = code;
        code = auto_callout(code, ptr, cd);
        }
      goto NORMAL_CHAR;
      }
    }

  /* Fill in length of a previous callout, except when the next thing is
  a quantifier. */

  is_quantifier = c == '*' || c == '+' || c == '?' ||
    (c == '{' && is_counted_repeat(ptr+1));

  if (!is_quantifier && previous_callout != NULL &&
       after_manual_callout-- <= 0)
    {
    complete_callout(previous_callout, ptr, cd);
    previous_callout = NULL;
    }

  /* In extended mode, skip white space and comments */

  if ((options & PCRE_EXTENDED) != 0)
    {
    if ((cd->ctypes[c] & ctype_space) != 0) continue;
    if (c == '#')
      {
      /* The space before the ; is to avoid a warning on a silly compiler
      on the Macintosh. */
      while ((c = *(++ptr)) != 0 && c != NEWLINE) ;
      if (c != 0) continue;   /* Else fall through to handle end of string */
      }
    }

  /* No auto callout for quantifiers. */

  if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
    {
    previous_callout = code;
    code = auto_callout(code, ptr, cd);
    }

  switch(c)
    {
    /* The branch terminates at end of string, |, or ). */

    case 0:
    case '|':
    case ')':
    *firstbyteptr = firstbyte;
    *reqbyteptr = reqbyte;
    *codeptr = code;
    *ptrptr = ptr;
    return TRUE;

    /* Handle single-character metacharacters. In multiline mode, ^ disables
    the setting of any following char as a first character. */

    case '^':
    if ((options & PCRE_MULTILINE) != 0)
      {
      if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
      }
    previous = NULL;
    *code++ = OP_CIRC;
    break;

    case '$':
    previous = NULL;
    *code++ = 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. */

    case '.':
    if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
    zerofirstbyte = firstbyte;
    zeroreqbyte = reqbyte;
    previous = code;
    *code++ = OP_ANY;
    break;

    /* Character classes. If the included characters are all < 255 in value, we
    build a 32-byte bitmap of the permitted characters, except in the special
    case where there is only one such character. For negated classes, we build
    the map as usual, then invert it at the end. However, we use a different
    opcode so that data characters > 255 can be handled correctly.

    If the class contains characters outside the 0-255 range, a different
    opcode is compiled. It may optionally have a bit map for characters < 256,
    but those above are are explicitly listed afterwards. A flag byte tells
    whether the bitmap is present, and whether this is a negated class or not.
    */

    case '[':
    previous = code;

    /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
    they are encountered at the top level, so we'll do that too. */

    if ((ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
        check_posix_syntax(ptr, &tempptr, cd))
      {
      *errorcodeptr = (ptr[1] == ':')? ERR13 : ERR31;
      goto FAILED;
      }

    /* If the first character is '^', set the negation flag and skip it. */

    if ((c = *(++ptr)) == '^')
      {
      negate_class = TRUE;
      c = *(++ptr);
      }
    else
      {
      negate_class = FALSE;
      }

    /* Keep a count of chars with values < 256 so that we can optimize the case
    of just a single character (as long as it's < 256). For higher valued UTF-8
    characters, we don't yet do any optimization. */

    class_charcount = 0;
    class_lastchar = -1;

#ifdef SUPPORT_UTF8
    class_utf8 = FALSE;                       /* No chars >= 256 */
    class_utf8data = code + LINK_SIZE + 34;   /* For UTF-8 items */
#endif

    /* Initialize the 32-char bit map to all zeros. We have to build the
    map in a temporary bit of store, in case the class contains only 1
    character (< 256), because in that case the compiled code doesn't use the
    bit map. */

    memset(classbits, 0, 32 * sizeof(uschar));

    /* Process characters until ] is reached. By writing this as a "do" it
    means that an initial ] is taken as a data character. The first pass
    through the regex checked the overall syntax, so we don't need to be very
    strict here. At the start of the loop, c contains the first byte of the
    character. */

    do
      {
#ifdef SUPPORT_UTF8
      if (utf8 && c > 127)
        {                           /* Braces are required because the */
        GETCHARLEN(c, ptr, ptr);    /* macro generates multiple statements */
        }
#endif

      /* Inside \Q...\E everything is literal except \E */

      if (inescq)
        {
        if (c == '\\' && ptr[1] == 'E')
          {
          inescq = FALSE;
          ptr++;
          continue;
          }
        else goto LONE_SINGLE_CHARACTER;
        }

      /* Handle POSIX class names. Perl allows a negation extension of the
      form [:^name:]. A square bracket that doesn't match the syntax is
      treated as a literal. We also recognize the POSIX constructions
      [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
      5.6 and 5.8 do. */

      if (c == '[' &&
          (ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
          check_posix_syntax(ptr, &tempptr, cd))
        {
        BOOL local_negate = FALSE;
        int posix_class, i;
        register const uschar *cbits = cd->cbits;

        if (ptr[1] != ':')
          {
          *errorcodeptr = ERR31;
          goto FAILED;
          }

        ptr += 2;
        if (*ptr == '^')
          {
          local_negate = TRUE;
          ptr++;
          }

        posix_class = check_posix_name(ptr, tempptr - ptr);
        if (posix_class < 0)
          {
          *errorcodeptr = ERR30;
          goto FAILED;
          }

        /* If matching is caseless, upper and lower are converted to
        alpha. This relies on the fact that the class table starts with
        alpha, lower, upper as the first 3 entries. */

        if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
          posix_class = 0;

        /* Or into the map we are building up to 3 of the static class
        tables, or their negations. The [:blank:] class sets up the same
        chars as the [:space:] class (all white space). We remove the vertical
        white space chars afterwards. */

        posix_class *= 3;
        for (i = 0; i < 3; i++)
          {
          BOOL blankclass = strncmp((char *)ptr, "blank", 5) == 0;
          int taboffset = posix_class_maps[posix_class + i];
          if (taboffset < 0) break;
          if (local_negate)
            {
            if (i == 0)
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+taboffset];
            else
              for (c = 0; c < 32; c++) classbits[c] &= ~cbits[c+taboffset];
            if (blankclass) classbits[1] |= 0x3c;
            }
          else
            {
            for (c = 0; c < 32; c++) classbits[c] |= cbits[c+taboffset];
            if (blankclass) classbits[1] &= ~0x3c;
            }
          }

        ptr = tempptr + 1;
        class_charcount = 10;  /* Set > 1; assumes more than 1 per class */
        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. Escaped items are checked for
      validity in the pre-compiling pass. The sequence \b is a special case.
      Inside a class (and only there) it is treated as backspace. Elsewhere
      it marks a word boundary. Other escapes have preset maps ready to
      or into the one we are building. We assume they have more than one
      character in them, so set class_charcount bigger than one. */

      if (c == '\\')
        {
        c = check_escape(&ptr, errorcodeptr, *brackets, options, TRUE);

        if (-c == ESC_b) c = '\b';       /* \b is backslash in a class */
        else if (-c == ESC_X) c = 'X';   /* \X is literal X in a class */
        else if (-c == ESC_Q)            /* Handle start of quoted string */
          {
          if (ptr[1] == '\\' && ptr[2] == 'E')
            {
            ptr += 2; /* avoid empty string */
            }
          else inescq = TRUE;
          continue;
          }

        if (c < 0)
          {
          register const uschar *cbits = cd->cbits;
          class_charcount += 2;     /* Greater than 1 is what matters */
          switch (-c)
            {
            case ESC_d:
            for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
            continue;

            case ESC_D:
            for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
            continue;

            case ESC_w:
            for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
            continue;

            case ESC_W:
            for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
            continue;

            case ESC_s:
            for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
            classbits[1] &= ~0x08;   /* Perl 5.004 onwards omits VT from \s */
            continue;

            case ESC_S:
            for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
            classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */
            continue;

#ifdef SUPPORT_UCP
            case ESC_p:
            case ESC_P:
              {
              BOOL negated;
              int property = get_ucp(&ptr, &negated, errorcodeptr);
              if (property < 0) goto FAILED;
              class_utf8 = TRUE;
              *class_utf8data++ = ((-c == ESC_p) != negated)?
                XCL_PROP : XCL_NOTPROP;
              *class_utf8data++ = property;
              class_charcount -= 2;   /* Not a < 256 character */
              }
            continue;
#endif

            /* Unrecognized escapes are faulted if PCRE is running in its
            strict mode. By default, for compatibility with Perl, they are
            treated as literals. */

            default:
            if ((options & PCRE_EXTRA) != 0)
              {
              *errorcodeptr = ERR7;
              goto FAILED;
              }
            c = *ptr;              /* The final character */
            class_charcount -= 2;  /* Undo the default count from above */
            }
          }

        /* Fall through if we have a single character (c >= 0). This may be
        > 256 in UTF-8 mode. */

        }   /* End of backslash handling */

      /* A single character may be followed by '-' to form a range. However,
      Perl does not permit ']' to be the end of the range. A '-' character
      here is treated as a literal. */

      if (ptr[1] == '-' && ptr[2] != ']')
        {
        int d;
        ptr += 2;

#ifdef SUPPORT_UTF8
        if (utf8)
          {                           /* Braces are required because the */
          GETCHARLEN(d, ptr, ptr);    /* macro generates multiple statements */
          }
        else
#endif
        d = *ptr;  /* Not UTF-8 mode */

        /* The second part of a range can be a single-character escape, but
        not any of the other escapes. Perl 5.6 treats a hyphen as a literal
        in such circumstances. */

        if (d == '\\')
          {
          const uschar *oldptr = ptr;
          d = check_escape(&ptr, errorcodeptr, *brackets, options, TRUE);

          /* \b is backslash; \X is literal X; any other special means the '-'
          was literal */

          if (d < 0)
            {
            if (d == -ESC_b) d = '\b';
            else if (d == -ESC_X) d = 'X'; else
              {
              ptr = oldptr - 2;
              goto LONE_SINGLE_CHARACTER;  /* A few lines below */
              }
            }
          }

        /* The check that the two values are in the correct order happens in
        the pre-pass. Optimize one-character ranges */

        if (d == c) goto LONE_SINGLE_CHARACTER;  /* A few lines below */

        /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
        matching, we have to use an XCLASS with extra data items. Caseless
        matching for characters > 127 is available only if UCP support is
        available. */

#ifdef SUPPORT_UTF8
        if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
          {
          class_utf8 = TRUE;

          /* With UCP support, we can find the other case equivalents of
          the relevant characters. There may be several ranges. Optimize how
          they fit with the basic range. */

#ifdef SUPPORT_UCP
          if ((options & PCRE_CASELESS) != 0)
            {
            int occ, ocd;
            int cc = c;
            int origd = d;
            while (get_othercase_range(&cc, origd, &occ, &ocd))
              {
              if (occ >= c && ocd <= d) continue;  /* Skip embedded ranges */

              if (occ < c  && ocd >= c - 1)        /* Extend the basic range */
                {                                  /* if there is overlap,   */
                c = occ;                           /* noting that if occ < c */
                continue;                          /* we can't have ocd > d  */
                }                                  /* because a subrange is  */
              if (ocd > d && occ <= d + 1)         /* always shorter than    */
                {                                  /* the basic range.       */
                d = ocd;
                continue;
                }

              if (occ == ocd)
                {
                *class_utf8data++ = XCL_SINGLE;
                }
              else
                {
                *class_utf8data++ = XCL_RANGE;
                class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
                }
              class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
              }
            }
#endif  /* SUPPORT_UCP */

          /* Now record the original range, possibly modified for UCP caseless
          overlapping ranges. */

          *class_utf8data++ = XCL_RANGE;
          class_utf8data += _pcre_ord2utf8(c, class_utf8data);
          class_utf8data += _pcre_ord2utf8(d, class_utf8data);

          /* With UCP support, we are done. Without UCP support, there is no
          caseless matching for UTF-8 characters > 127; we can use the bit map
          for the smaller ones. */

#ifdef SUPPORT_UCP
          continue;    /* With next character in the class */
#else
          if ((options & PCRE_CASELESS) == 0 || c > 127) continue;

          /* Adjust upper limit and fall through to set up the map */

          d = 127;

#endif  /* SUPPORT_UCP */
          }
#endif  /* SUPPORT_UTF8 */

        /* We use the bit map for all cases when not in UTF-8 mode; else
        ranges that lie entirely within 0-127 when there is UCP support; else
        for partial ranges without UCP support. */

        for (; c <= d; c++)
          {
          classbits[c/8] |= (1 << (c&7));
          if ((options & PCRE_CASELESS) != 0)
            {
            int