code/trunk/pcre.c

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

/*
This 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. See
the file Tech.Notes for some information on the internals.

Written by: Philip Hazel <ph10@cam.ac.uk>

           Copyright (c) 1997-2000 University of Cambridge

-----------------------------------------------------------------------------
Permission is granted to anyone to use this software for any purpose on any
computer system, and to redistribute it freely, subject to the following
restrictions:

1. This software is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

2. The origin of this software must not be misrepresented, either by
   explicit claim or by omission.

3. Altered versions must be plainly marked as such, and must not be
   misrepresented as being the original software.

4. If PCRE is embedded in any software that is released under the GNU
   General Purpose Licence (GPL), then the terms of that licence shall
   supersede any condition above with which it is incompatible.
-----------------------------------------------------------------------------
*/


/* Define DEBUG to get debugging output on stdout. */

/* #define DEBUG */

/* Use a macro for debugging printing, 'cause that eliminates the use of #ifdef
inline, and there are *still* stupid compilers about that don't like indented
pre-processor statements. I suppose it's only been 10 years... */

#ifdef DEBUG
#define DPRINTF(p) printf p
#else
#define DPRINTF(p) /*nothing*/
#endif

/* Include the internals header, which itself includes Standard C headers plus
the external pcre header. */

#include "internal.h"


/* Allow compilation as C++ source code, should anybody want to do that. */

#ifdef __cplusplus
#define class pcre_class
#endif


/* Number of items on the nested bracket stacks at compile time. This should
not be set greater than 200. */

#define BRASTACK_SIZE 200


/* The number of bytes in a literal character string above which we can't add
any more is different when UTF-8 characters may be encountered. */

#ifdef SUPPORT_UTF8
#define MAXLIT 250
#else
#define MAXLIT 255
#endif


/* Min and max values for the common repeats; for the maxima, 0 => infinity */

static const char rep_min[] = { 0, 0, 1, 1, 0, 0 };
static const char rep_max[] = { 0, 0, 0, 0, 1, 1 };

/* Text forms of OP_ values and things, for debugging (not all used) */

#ifdef DEBUG
static const char *OP_names[] = {
  "End", "\\A", "\\B", "\\b", "\\D", "\\d",
  "\\S", "\\s", "\\W", "\\w", "\\Z", "\\z",
  "Opt", "^", "$", "Any", "chars", "not",
  "*", "*?", "+", "+?", "?", "??", "{", "{", "{",
  "*", "*?", "+", "+?", "?", "??", "{", "{", "{",
  "*", "*?", "+", "+?", "?", "??", "{", "{", "{",
  "*", "*?", "+", "+?", "?", "??", "{", "{",
  "class", "Ref", "Recurse",
  "Alt", "Ket", "KetRmax", "KetRmin", "Assert", "Assert not",
  "AssertB", "AssertB not", "Reverse", "Once", "Cond", "Cref",
  "Brazero", "Braminzero", "Bra"
};
#endif

/* 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. */

static const short int escapes[] = {
    0,      0,      0,      0,      0,      0,      0,      0,   /* 0 - 7 */
    0,      0,    ':',    ';',    '<',    '=',    '>',    '?',   /* 8 - ? */
  '@', -ESC_A, -ESC_B,      0, -ESC_D,      0,      0,      0,   /* @ - G */
    0,      0,      0,      0,      0,      0,      0,      0,   /* H - O */
    0,      0,      0, -ESC_S,      0,      0,      0, -ESC_W,   /* P - W */
    0,      0, -ESC_Z,    '[',   '\\',    ']',    '^',    '_',   /* X - _ */
  '`',      7, -ESC_b,      0, -ESC_d,     27,   '\f',      0,   /* ` - g */
    0,      0,      0,      0,      0,      0,   '\n',      0,   /* h - o */
    0,      0,   '\r', -ESC_s,   '\t',      0,      0, -ESC_w,   /* p - w */
    0,      0, -ESC_z                                            /* x - z */
};

/* 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 *posix_names[] = {
  "alpha", "lower", "upper",
  "alnum", "ascii", "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, 4, 6, 0 };

/* Table of class bit maps for each POSIX class; up to three may be combined
to form the class. */

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_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 */
  cbit_xdigit,-1,         -1              /* xdigit */
};


/* Definition to allow mutual recursion */

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

/* Structure for building a chain of data that actually lives on the
stack, for holding the values of the subject pointer at the start of each
subpattern, so as to detect when an empty string has been matched by a
subpattern - to break infinite loops. */

typedef struct eptrblock {
  struct eptrblock *prev;
  const uschar *saved_eptr;
} eptrblock;

/* Flag bits for the match() function */

#define match_condassert   0x01    /* Called to check a condition assertion */
#define match_isgroup      0x02    /* Set if start of bracketed group */


/*************************************************
*               Global variables                 *
*************************************************/

/* PCRE is thread-clean and doesn't use any global variables in the normal
sense. However, it calls memory allocation and free functions via the two
indirections below, which are can be changed by the caller, but are shared
between all threads. */

void *(*pcre_malloc)(size_t) = malloc;
void  (*pcre_free)(void *) = free;


/*************************************************
*    Macros and tables for character handling    *
*************************************************/

/* When UTF-8 encoding is being used, a character is no longer just a single
byte. The macros for character handling generate simple sequences when used in
byte-mode, and more complicated ones for UTF-8 characters. */

#ifndef SUPPORT_UTF8
#define GETCHARINC(c, eptr) c = *eptr++;
#define GETCHARLEN(c, eptr, len) c = *eptr;
#define BACKCHAR(eptr)

#else   /* SUPPORT_UTF8 */

/* Get the next UTF-8 character, advancing the pointer */

#define GETCHARINC(c, eptr) \
  c = *eptr++; \
  if (md->utf8 && (c & 0xc0) == 0xc0) \
    { \
    int a = utf8_table4[c & 0x3f];  /* Number of additional bytes */ \
    int s = 6 - a;                  /* Amount to shift next byte */  \
    c &= utf8_table3[a];            /* Low order bits from first byte */ \
    while (a-- > 0) \
      { \
      c |= (*eptr++ & 0x3f) << s; \
      s += 6; \
      } \
    }

/* Get the next UTF-8 character, not advancing the pointer, setting length */

#define GETCHARLEN(c, eptr, len) \
  c = *eptr; \
  len = 1; \
  if (md->utf8 && (c & 0xc0) == 0xc0) \
    { \
    int i; \
    int a = utf8_table4[c & 0x3f];  /* Number of additional bytes */ \
    int s = 6 - a;                  /* Amount to shift next byte */  \
    c &= utf8_table3[a];            /* Low order bits from first byte */ \
    for (i = 1; i <= a; i++) \
      { \
      c |= (eptr[i] & 0x3f) << s; \
      s += 6; \
      } \
    len += a; \
    }

/* If the pointer is not at the start of a character, move it back until
it is. */

#define BACKCHAR(eptr) while((*eptr & 0xc0) == 0x80) eptr--;

#endif


/*************************************************
*             Default character tables           *
*************************************************/

/* A default set of character tables is included in the PCRE binary. Its source
is built by the maketables auxiliary program, which uses the default C ctypes
functions, and put in the file chartables.c. These tables are used by PCRE
whenever the caller of pcre_compile() does not provide an alternate set of
tables. */

#include "chartables.c"


#ifdef SUPPORT_UTF8
/*************************************************
*           Tables for UTF-8 support             *
*************************************************/

/* These are the breakpoints for different numbers of bytes in a UTF-8
character. */

static int utf8_table1[] = { 0x7f, 0x7ff, 0xffff, 0x1fffff, 0x3ffffff, 0x7fffffff};

/* These are the indicator bits and the mask for the data bits to set in the
first byte of a character, indexed by the number of additional bytes. */

static int utf8_table2[] = { 0,    0xc0, 0xe0, 0xf0, 0xf8, 0xfc};
static int utf8_table3[] = { 0xff, 0x1f, 0x0f, 0x07, 0x03, 0x01};

/* Table of the number of extra characters, indexed by the first character
masked with 0x3f. The highest number for a valid UTF-8 character is in fact
0x3d. */

static uschar utf8_table4[] = {
  1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
  1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
  2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,
  3,3,3,3,3,3,3,3,4,4,4,4,5,5,5,5 };


/*************************************************
*       Convert character value to UTF-8         *
*************************************************/

/* This function takes an integer value in the range 0 - 0x7fffffff
and encodes it as a UTF-8 character in 0 to 6 bytes.

Arguments:
  cvalue     the character value
  buffer     pointer to buffer for result - at least 6 bytes long

Returns:     number of characters placed in the buffer
*/

static int
ord2utf8(int cvalue, uschar *buffer)
{
register int i, j;
for (i = 0; i < sizeof(utf8_table1)/sizeof(int); i++)
  if (cvalue <= utf8_table1[i]) break;
*buffer++ = utf8_table2[i] | (cvalue & utf8_table3[i]);
cvalue >>= 6 - i;
for (j = 0; j < i; j++)
  {
  *buffer++ = 0x80 | (cvalue & 0x3f);
  cvalue >>= 6;
  }
return i + 1;
}
#endif


/*************************************************
*          Return version string                 *
*************************************************/

#define STRING(a)  # a
#define XSTRING(s) STRING(s)

const char *
pcre_version(void)
{
return XSTRING(PCRE_MAJOR) "." XSTRING(PCRE_MINOR) " " XSTRING(PCRE_DATE);
}


/*************************************************
* (Obsolete) Return info about compiled pattern  *
*************************************************/

/* This is the original "info" function. It picks potentially useful data out
of the private structure, but its interface was too rigid. It remains for
backwards compatibility. The public options are passed back in an int - though
the re->options field has been expanded to a long int, all the public options
at the low end of it, and so even on 16-bit systems this will still be OK.
Therefore, I haven't changed the API for pcre_info().

Arguments:
  external_re   points to compiled code
  optptr        where to pass back the options
  first_char    where to pass back the first character,
                or -1 if multiline and all branches start ^,
                or -2 otherwise

Returns:        number of capturing subpatterns
                or negative values on error
*/

int
pcre_info(const pcre *external_re, int *optptr, int *first_char)
{
const real_pcre *re = (const real_pcre *)external_re;
if (re == NULL) return PCRE_ERROR_NULL;
if (re->magic_number != MAGIC_NUMBER) return PCRE_ERROR_BADMAGIC;
if (optptr != NULL) *optptr = (int)(re->options & PUBLIC_OPTIONS);
if (first_char != NULL)
  *first_char = ((re->options & PCRE_FIRSTSET) != 0)? re->first_char :
     ((re->options & PCRE_STARTLINE) != 0)? -1 : -2;
return re->top_bracket;
}


/*************************************************
*        Return info about compiled pattern      *
*************************************************/

/* This is a newer "info" function which has an extensible interface so
that additional items can be added compatibly.

Arguments:
  external_re      points to compiled code
  external_study   points to study data, or NULL
  what             what information is required
  where            where to put the information

Returns:           0 if data returned, negative on error
*/

int
pcre_fullinfo(const pcre *external_re, const pcre_extra *study_data, int what,
  void *where)
{
const real_pcre *re = (const real_pcre *)external_re;
const real_pcre_extra *study = (const real_pcre_extra *)study_data;

if (re == NULL || where == NULL) return PCRE_ERROR_NULL;
if (re->magic_number != MAGIC_NUMBER) return PCRE_ERROR_BADMAGIC;

switch (what)
  {
  case PCRE_INFO_OPTIONS:
  *((unsigned long int *)where) = re->options & PUBLIC_OPTIONS;
  break;

  case PCRE_INFO_SIZE:
  *((size_t *)where) = re->size;
  break;

  case PCRE_INFO_CAPTURECOUNT:
  *((int *)where) = re->top_bracket;
  break;

  case PCRE_INFO_BACKREFMAX:
  *((int *)where) = re->top_backref;
  break;

  case PCRE_INFO_FIRSTCHAR:
  *((int *)where) =
    ((re->options & PCRE_FIRSTSET) != 0)? re->first_char :
    ((re->options & PCRE_STARTLINE) != 0)? -1 : -2;
  break;

  case PCRE_INFO_FIRSTTABLE:
  *((const uschar **)where) =
    (study != NULL && (study->options & PCRE_STUDY_MAPPED) != 0)?
      study->start_bits : NULL;
  break;

  case PCRE_INFO_LASTLITERAL:
  *((int *)where) =
    ((re->options & PCRE_REQCHSET) != 0)? re->req_char : -1;
  break;

  default: return PCRE_ERROR_BADOPTION;
  }

return 0;
}


#ifdef DEBUG
/*************************************************
*        Debugging function to print chars       *
*************************************************/

/* Print a sequence of chars in printable format, stopping at the end of the
subject if the requested.

Arguments:
  p           points to characters
  length      number to print
  is_subject  TRUE if printing from within md->start_subject
  md          pointer to matching data block, if is_subject is TRUE

Returns:     nothing
*/

static void
pchars(const uschar *p, int length, BOOL is_subject, match_data *md)
{
int c;
if (is_subject && length > md->end_subject - p) length = md->end_subject - p;
while (length-- > 0)
  if (isprint(c = *(p++))) printf("%c", c); else printf("\\x%02x", c);
}
#endif


/*************************************************
*            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
  errorptr   points to the pointer to the error message
  bracount   number of previous extracting brackets
  options    the options bits
  isclass    TRUE if inside a character class
  cd         pointer to char tables block

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

static int
check_escape(const uschar **ptrptr, const char **errorptr, int bracount,
  int options, BOOL isclass, compile_data *cd)
{
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) *errorptr = ERR1;

/* Digits or letters may have special meaning; all others are literals. */

else if (c < '0' || c > 'z') {}

/* Do an initial lookup in a table. A non-zero result is something that can be
returned immediately. Otherwise further processing may be required. */

else if ((i = escapes[c - '0']) != 0) c = i;

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

else
  {
  const uschar *oldptr;
  switch (c)
    {
    /* 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 ((cd->ctypes[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 && (cd->ctypes[ptr[1]] & ctype_digit) != 0 &&
      ptr[1] != '8' && ptr[1] != '9')
        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 ((cd->ctypes[*pt] & ctype_xdigit) != 0)
        {
        count++;
        c = c * 16 + cd->lcc[*pt] -
          (((cd->ctypes[*pt] & ctype_digit) != 0)? '0' : 'W');
        pt++;
        }
      if (*pt == '}')
        {
        if (c < 0 || count > 8) *errorptr = 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 && (cd->ctypes[ptr[1]] & ctype_xdigit) != 0)
      {
      ptr++;
      c = c * 16 + cd->lcc[*ptr] -
        (((cd->ctypes[*ptr] & ctype_digit) != 0)? '0' : 'W');
      }
    break;

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

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

    /* A letter is upper-cased; then the 0x40 bit is flipped */

    if (c >= 'a' && c <= 'z') c = cd->fcc[c];
    c ^= 0x40;
    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:
      *errorptr = ERR3;
      break;
      }
    break;
    }
  }

*ptrptr = ptr;
return c;
}


/*************************************************
*            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 '{'
  cd        pointer to char tables block

Returns:    TRUE or FALSE
*/

static BOOL
is_counted_repeat(const uschar *p, compile_data *cd)
{
if ((cd->ctypes[*p++] & ctype_digit) == 0) return FALSE;
while ((cd->ctypes[*p] & ctype_digit) != 0) p++;
if (*p == '}') return TRUE;

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

if ((cd->ctypes[*p++] & ctype_digit) == 0) return FALSE;
while ((cd->ctypes[*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
  errorptr   points to pointer to error message
  cd         pointer to character tables clock

Returns:     pointer to '}' on success;
             current ptr on error, with errorptr set
*/

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

while ((cd->ctypes[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0';

if (*p == '}') max = min; else
  {
  if (*(++p) != '}')
    {
    max = 0;
    while((cd->ctypes[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0';
    if (max < min)
      {
      *errorptr = ERR4;
      return p;
      }
    }
  }

/* Do paranoid checks, then fill in the required variables, and pass back the
pointer to the terminating '}'. */

if (min > 65535 || max > 65535)
  *errorptr = ERR5;
else
  {
  *minp = min;
  *maxp = max;
  }
return p;
}


/*************************************************
*        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.

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
*/

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

register int branchlength = 0;
register uschar *cc = code + 3;

/* 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 -1;
    branchlength += d;
    do cc += (cc[1] << 8) + cc[2]; while (*cc == OP_ALT);
    cc += 3;
    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 += 3;
    branchlength = 0;
    break;

    /* Skip over assertive subpatterns */

    case OP_ASSERT:
    case OP_ASSERT_NOT:
    case OP_ASSERTBACK:
    case OP_ASSERTBACK_NOT:
    do cc += (cc[1] << 8) + cc[2]; while (*cc == OP_ALT);
    cc += 3;
    break;

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

    case OP_REVERSE:
    cc++;
    /* Fall through */

    case OP_CREF:
    case OP_OPT:
    cc++;
    /* Fall through */

    case OP_SOD:
    case OP_EOD:
    case OP_EODN:
    case OP_CIRC:
    case OP_DOLL:
    case OP_NOT_WORD_BOUNDARY:
    case OP_WORD_BOUNDARY:
    cc++;
    break;

    /* Handle char strings. In UTF-8 mode we must count characters, not bytes.
    This requires a scan of the string, unfortunately. We assume valid UTF-8
    strings, so all we do is reduce the length by one for byte whose bits are
    10xxxxxx. */

    case OP_CHARS:
    branchlength += *(++cc);
#ifdef SUPPORT_UTF8
    for (d = 1; d <= *cc; d++)
      if ((cc[d] & 0xc0) == 0x80) branchlength--;
#endif
    cc += *cc + 1;
    break;

    /* Handle exact repetitions */

    case OP_EXACT:
    case OP_TYPEEXACT:
    branchlength += (cc[1] << 8) + cc[2];
    cc += 4;
    break;

    /* Handle single-char matchers */

    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;


    /* Check a class for variable quantification */

    case OP_CLASS:
    cc += (*cc == OP_REF)? 2 : 33;

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

      case OP_CRRANGE:
      case OP_CRMINRANGE:
      if ((cc[1] << 8) + cc[2] != (cc[3] << 8) + cc[4]) return -1;
      branchlength += (cc[1] << 8) + cc[2];
      cc += 5;
      break;

      default:
      branchlength++;
      }
    break;

    /* Anything else is variable length */

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


/*************************************************
*           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;
}


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

/* Scan the pattern, compiling it into the code vector.

Arguments:
  options      the option bits
  brackets     points to number of brackets used
  code         points to the pointer to the current code point
  ptrptr       points to the current pattern pointer
  errorptr     points to pointer to error message
  optchanged   set to the value of the last OP_OPT item compiled
  reqchar      set to the last literal character required, else -1
  countlits    set to count of mandatory literal characters
  cd           contains pointers to tables

Returns:       TRUE on success
               FALSE, with *errorptr set on error
*/

static BOOL
compile_branch(int options, int *brackets, uschar **codeptr,
  const uschar **ptrptr, const char **errorptr, int *optchanged,
  int *reqchar, int *countlits, compile_data *cd)
{
int repeat_type, op_type;
int repeat_min, repeat_max;
int bravalue, length;
int greedy_default, greedy_non_default;
int prevreqchar;
int condcount = 0;
int subcountlits = 0;
register int c;
register uschar *code = *codeptr;
uschar *tempcode;
const uschar *ptr = *ptrptr;
const uschar *tempptr;
uschar *previous = NULL;
uschar class[32];

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

greedy_default = ((options & PCRE_UNGREEDY) != 0);
greedy_non_default = greedy_default ^ 1;

/* Initialize no required char, and count of literals */

*reqchar = prevreqchar = -1;
*countlits = 0;

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

for (;; ptr++)
  {
  BOOL negate_class;
  int class_charcount;
  int class_lastchar;
  int newoptions;
  int condref;
  int subreqchar;

  c = *ptr;
  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 != '\n') ;
      continue;
      }
    }

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

    case 0:
    case '|':
    case ')':
    *codeptr = code;
    *ptrptr = ptr;
    return TRUE;

    /* Handle single-character metacharacters */

    case '^':
    previous = NULL;
    *code++ = OP_CIRC;
    break;

    case '$':
    previous = NULL;
    *code++ = OP_DOLL;
    break;

    case '.':
    previous = code;
    *code++ = OP_ANY;
    break;

    /* Character classes. These always build a 32-byte bitmap of the permitted
    characters, except in the special case where there is only one character.
    For negated classes, we build the map as usual, then invert it at the end.
    */

    case '[':
    previous = code;
    *code++ = OP_CLASS;

    /* 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 so that we can optimize the case of just a single
    character. */

    class_charcount = 0;
    class_lastchar = -1;

    /* 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, because in that case the compiled code doesn't use the
    bit map. */

    memset(class, 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. */

    do
      {
      if (c == 0)
        {
        *errorptr = ERR6;
        goto FAILED;
        }

      /* 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 does. */

      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] != ':')
          {
          *errorptr = ERR31;
          goto FAILED;
          }

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

        posix_class = check_posix_name(ptr, tempptr - ptr);
        if (posix_class < 0)
          {
          *errorptr = 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. */

        posix_class *= 3;
        for (i = 0; i < 3; i++)
          {
          int taboffset = posix_class_maps[posix_class + i];
          if (taboffset < 0) break;
          if (local_negate)
            for (c = 0; c < 32; c++) class[c] |= ~cbits[c+taboffset];
          else
            for (c = 0; c < 32; c++) class[c] |= cbits[c+taboffset];
          }

        ptr = tempptr + 1;
        class_charcount = 10;  /* Set > 1; assumes more than 1 per class */
        continue;
        }

      /* 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_count bigger than one. */

      if (c == '\\')
        {
        c = check_escape(&ptr, errorptr, *brackets, options, TRUE, cd);
        if (-c == ESC_b) c = '\b';
        else if (c < 0)
          {
          register const uschar *cbits = cd->cbits;
          class_charcount = 10;
          switch (-c)
            {
            case ESC_d:
            for (c = 0; c < 32; c++) class[c] |= cbits[c+cbit_digit];
            continue;

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

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

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

            case ESC_s:
            for (c = 0; c < 32; c++) class[c] |= cbits[c+cbit_space];
            continue;

            case ESC_S:
            for (c = 0; c < 32; c++) class[c] |= ~cbits[c+cbit_space];
            continue;

            default:
            *errorptr = ERR7;
            goto FAILED;
            }
          }

        /* Fall through if single character, but don't at present allow
        chars > 255 in UTF-8 mode. */

#ifdef SUPPORT_UTF8
        if (c > 255)
          {
          *errorptr = ERR33;
          goto FAILED;
          }
#endif
        }

      /* 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;
        d = *ptr;

        if (d == 0)
          {
          *errorptr = ERR6;
          goto FAILED;
          }

        /* 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, errorptr, *brackets, options, TRUE, cd);

#ifdef SUPPORT_UTF8
          if (d > 255)
            {
            *errorptr = ERR33;
            goto FAILED;
            }
#endif
          /* \b is backslash; any other special means the '-' was literal */

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

        if (d < c)
          {
          *errorptr = ERR8;
          goto FAILED;
          }

        for (; c <= d; c++)
          {
          class[c/8] |= (1 << (c&7));
          if ((options & PCRE_CASELESS) != 0)
            {
            int uc = cd->fcc[c];           /* flip case */
            class[uc/8] |= (1 << (uc&7));
            }
          class_charcount++;                /* in case a one-char range */
          class_lastchar = c;
          }
        continue;   /* Go get the next char in the class */
        }

      /* Handle a lone single character - we can get here for a normal
      non-escape char, or after \ that introduces a single character. */

      SINGLE_CHARACTER:

      class [c/8] |= (1 << (c&7));
      if ((options & PCRE_CASELESS) != 0)
        {
        c = cd->fcc[c];   /* flip case */
        class[c/8] |= (1 << (c&7));
        }
      class_charcount++;
      class_lastchar = c;
      }

    /* Loop until ']' reached; the check for end of string happens inside the
    loop. This "while" is the end of the "do" above. */

    while ((c = *(++ptr)) != ']');

    /* If class_charcount is 1 and class_lastchar is not negative, we saw
    precisely one character. This doesn't need the whole 32-byte bit map.
    We turn it into a 1-character OP_CHAR if it's positive, or OP_NOT if
    it's negative. */

    if (class_charcount == 1 && class_lastchar >= 0)
      {
      if (negate_class)
        {
        code[-1] = OP_NOT;
        }
      else
        {
        code[-1] = OP_CHARS;
        *code++ = 1;
        }
      *code++ = class_lastchar;
      }

    /* Otherwise, negate the 32-byte map if necessary, and copy it into
    the code vector. */

    else
      {
      if (negate_class)
        for (c = 0; c < 32; c++) code[c] = ~class[c];
      else
        memcpy(code, class, 32);
      code += 32;
      }
    break;

    /* Various kinds of repeat */

    case '{':
    if (!is_counted_repeat(ptr+1, cd)) goto NORMAL_CHAR;
    ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorptr, cd);
    if (*errorptr != NULL) goto FAILED;
    goto REPEAT;

    case '*':
    repeat_min = 0;
    repeat_max = -1;
    goto REPEAT;

    case '+':
    repeat_min = 1;
    repeat_max = -1;
    goto REPEAT;

    case '?':
    repeat_min = 0;
    repeat_max = 1;

    REPEAT:
    if (previous == NULL)
      {
      *errorptr = ERR9;
      goto FAILED;
      }

    /* If the next character is '?' this is a minimizing repeat, by default,
    but if PCRE_UNGREEDY is set, it works the other way round. Advance to the
    next character. */

    if (ptr[1] == '?')
      { repeat_type = greedy_non_default; ptr++; }
    else repeat_type = greedy_default;

    /* If previous was a string of characters, chop off the last one and use it
    as the subject of the repeat. If there was only one character, we can
    abolish the previous item altogether. A repeat with a zero minimum wipes
    out any reqchar setting, backing up to the previous value. We must also
    adjust the countlits value. */

    if (*previous == OP_CHARS)
      {
      int len = previous[1];

      if (repeat_min == 0) *reqchar = prevreqchar;
      *countlits += repeat_min - 1;

      if (len == 1)
        {
        c = previous[2];
        code = previous;
        }
      else
        {
        c = previous[len+1];
        previous[1]--;
        code--;
        }
      op_type = 0;                 /* Use single-char op codes */
      goto OUTPUT_SINGLE_REPEAT;   /* Code shared with single character types */
      }

    /* If previous was a single negated character ([^a] or similar), we use
    one of the special opcodes, replacing it. The code is shared with single-
    character repeats by adding a suitable offset into repeat_type. */

    else if ((int)*previous == OP_NOT)
      {
      op_type = OP_NOTSTAR - OP_STAR;  /* Use "not" opcodes */
      c = previous[1];
      code = previous;
      goto OUTPUT_SINGLE_REPEAT;
      }

    /* If previous was a character type match (\d or similar), abolish it and
    create a suitable repeat item. The code is shared with single-character
    repeats by adding a suitable offset into repeat_type. */

    else if ((int)*previous < OP_EODN || *previous == OP_ANY)
      {
      op_type = OP_TYPESTAR - OP_STAR;  /* Use type opcodes */
      c = *previous;
      code = previous;

      OUTPUT_SINGLE_REPEAT:

      /* If the maximum is zero then the minimum must also be zero; Perl allows
      this case, so we do too - by simply omitting the item altogether. */

      if (repeat_max == 0) goto END_REPEAT;

      /* Combine the op_type with the repeat_type */

      repeat_type += op_type;

      /* A minimum of zero is handled either as the special case * or ?, or as
      an UPTO, with the maximum given. */

      if (repeat_min == 0)
        {
        if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
          else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
        else
          {
          *code++ = OP_UPTO + repeat_type;
          *code++ = repeat_max >> 8;
          *code++ = (repeat_max & 255);
          }
        }

      /* The case {1,} is handled as the special case + */

      else if (repeat_min == 1 && repeat_max == -1)
        *code++ = OP_PLUS + repeat_type;

      /* The case {n,n} is just an EXACT, while the general case {n,m} is
      handled as an EXACT followed by an UPTO. An EXACT of 1 is optimized. */

      else
        {
        if (repeat_min != 1)
          {
          *code++ = OP_EXACT + op_type;  /* NB EXACT doesn't have repeat_type */
          *code++ = repeat_min >> 8;
          *code++ = (repeat_min & 255);
          }

        /* If the mininum is 1 and the previous item was a character string,
        we either have to put back the item that got cancelled if the string
        length was 1, or add the character back onto the end of a longer
        string. For a character type nothing need be done; it will just get
        put back naturally. Note that the final character is always going to
        get added below. */

        else if (*previous == OP_CHARS)
          {
          if (code == previous) code += 2; else previous[1]++;
          }

        /*  For a single negated character we also have to put back the
        item that got cancelled. */

        else if (*previous == OP_NOT) code++;

        /* If the maximum is unlimited, insert an OP_STAR. */

        if (repeat_max < 0)
          {
          *code++ = c;
          *code++ = OP_STAR + repeat_type;
          }

        /* Else insert an UPTO if the max is greater than the min. */

        else if (repeat_max != repeat_min)
          {
          *code++ = c;
          repeat_max -= repeat_min;
          *code++ = OP_UPTO + repeat_type;
          *code++ = repeat_max >> 8;
          *code++ = (repeat_max & 255);
          }
        }

      /* The character or character type itself comes last in all cases. */

      *code++ = c;
      }

    /* If previous was a character class or a back reference, we put the repeat
    stuff after it, but just skip the item if the repeat was {0,0}. */

    else if (*previous == OP_CLASS || *previous == OP_REF)
      {
      if (repeat_max == 0)
        {
        code = previous;
        goto END_REPEAT;
        }
      if (repeat_min == 0 && repeat_max == -1)
        *code++ = OP_CRSTAR + repeat_type;
      else if (repeat_min == 1 && repeat_max == -1)
        *code++ = OP_CRPLUS + repeat_type;
      else if (repeat_min == 0 && repeat_max == 1)
        *code++ = OP_CRQUERY + repeat_type;
      else
        {
        *code++ = OP_CRRANGE + repeat_type;
        *code++ = repeat_min >> 8;
        *code++ = repeat_min & 255;
        if (repeat_max == -1) repeat_max = 0;  /* 2-byte encoding for max */
        *code++ = repeat_max >> 8;
        *code++ = repeat_max & 255;
        }
      }

    /* If previous was a bracket group, we may have to replicate it in certain
    cases. */

    else if ((int)*previous >= OP_BRA || (int)*previous == OP_ONCE ||
             (int)*previous == OP_COND)
      {
      register int i;
      int ketoffset = 0;
      int len = code - previous;
      uschar *bralink = NULL;

      /* If the maximum repeat count is unlimited, find the end of the bracket
      by scanning through from the start, and compute the offset back to it
      from the current code pointer. There may be an OP_OPT setting following
      the final KET, so we can't find the end just by going back from the code
      pointer. */

      if (repeat_max == -1)
        {
        register uschar *ket = previous;
        do ket += (ket[1] << 8) + ket[2]; while (*ket != OP_KET);
        ketoffset = code - ket;
        }

      /* The case of a zero minimum is special because of the need to stick
      OP_BRAZERO in front of it, and because the group appears once in the
      data, whereas in other cases it appears the minimum number of times. For
      this reason, it is simplest to treat this case separately, as otherwise
      the code gets far too mess. There are several special subcases when the
      minimum is zero. */

      if (repeat_min == 0)
        {
        /* If we set up a required char from the bracket, we must back off
        to the previous value and reset the countlits value too. */

        if (subcountlits > 0)
          {
          *reqchar = prevreqchar;
          *countlits -= subcountlits;
          }

        /* If the maximum is also zero, we just omit the group from the output
        altogether. */

        if (repeat_max == 0)
          {
          code = previous;
          goto END_REPEAT;
          }

        /* If the maximum is 1 or unlimited, we just have to stick in the
        BRAZERO and do no more at this point. */

        if (repeat_max <= 1)
          {
          memmove(previous+1, previous, len);
          code++;
          *previous++ = OP_BRAZERO + repeat_type;
          }

        /* If the maximum is greater than 1 and limited, we have to replicate
        in a nested fashion, sticking OP_BRAZERO before each set of brackets.
        The first one has to be handled carefully because it's the original
        copy, which has to be moved up. The remainder can be handled by code
        that is common with the non-zero minimum case below. We just have to
        adjust the value or repeat_max, since one less copy is required. */

        else
          {
          int offset;
          memmove(previous+4, previous, len);
          code += 4;
          *previous++ = OP_BRAZERO + repeat_type;
          *previous++ = OP_BRA;

          /* We chain together the bracket offset fields that have to be
          filled in later when the ends of the brackets are reached. */

          offset = (bralink == NULL)? 0 : previous - bralink;
          bralink = previous;
          *previous++ = offset >> 8;
          *previous++ = offset & 255;
          }

        repeat_max--;
        }

      /* If the minimum is greater than zero, replicate the group as many
      times as necessary, and adjust the maximum to the number of subsequent
      copies that we need. */

      else
        {
        for (i = 1; i < repeat_min; i++)
          {
          memcpy(code, previous, len);
          code += len;
          }
        if (repeat_max > 0) repeat_max -= repeat_min;
        }

      /* This code is common to both the zero and non-zero minimum cases. If
      the maximum is limited, it replicates the group in a nested fashion,
      remembering the bracket starts on a stack. In the case of a zero minimum,
      the first one was set up above. In all cases the repeat_max now specifies
      the number of additional copies needed. */

      if (repeat_max >= 0)
        {
        for (i = repeat_max - 1; i >= 0; i--)
          {
          *code++ = OP_BRAZERO + repeat_type;

          /* All but the final copy start a new nesting, maintaining the
          chain of brackets outstanding. */

          if (i != 0)
            {
            int offset;
            *code++ = OP_BRA;
            offset = (bralink == NULL)? 0 : code - bralink;
            bralink = code;
            *code++ = offset >> 8;
            *code++ = offset & 255;
            }

          memcpy(code, previous, len);
          code += len;
          }

        /* Now chain through the pending brackets, and fill in their length
        fields (which are holding the chain links pro tem). */

        while (bralink != NULL)
          {
          int oldlinkoffset;
          int offset = code - bralink + 1;
          uschar *bra = code - offset;
          oldlinkoffset = (bra[1] << 8) + bra[2];
          bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
          *code++ = OP_KET;
          *code++ = bra[1] = offset >> 8;
          *code++ = bra[2] = (offset & 255);
          }
        }

      /* If the maximum is unlimited, set a repeater in the final copy. We
      can't just offset backwards from the current code point, because we
      don't know if there's been an options resetting after the ket. The
      correct offset was computed above. */

      else code[-ketoffset] = OP_KETRMAX + repeat_type;
      }

    /* Else there's some kind of shambles */

    else
      {
      *errorptr = ERR11;
      goto FAILED;
      }

    /* In all case we no longer have a previous item. */

    END_REPEAT:
    previous = NULL;
    break;


    /* Start of nested bracket sub-expression, or comment or lookahead or
    lookbehind or option setting or condition. First deal with special things
    that can come after a bracket; all are introduced by ?, and the appearance
    of any of them means that this is not a referencing group. They were
    checked for validity in the first pass over the string, so we don't have to
    check for syntax errors here.  */

    case '(':
    newoptions = options;
    condref = -1;

    if (*(++ptr) == '?')
      {
      int set, unset;
      int *optset;

      switch (*(++ptr))
        {
        case '#':                 /* Comment; skip to ket */
        ptr++;
        while (*ptr != ')') ptr++;
        continue;

        case ':':                 /* Non-extracting bracket */
        bravalue = OP_BRA;
        ptr++;
        break;

        case '(':
        bravalue = OP_COND;       /* Conditional group */
        if ((cd->ctypes[*(++ptr)] & ctype_digit) != 0)
          {
          condref = *ptr - '0';
          while (*(++ptr) != ')') condref = condref*10 + *ptr - '0';
          if (condref == 0)
            {
            *errorptr = ERR35;
            goto FAILED;
            }
          ptr++;
          }
        else ptr--;
        break;

        case '=':                 /* Positive lookahead */
        bravalue = OP_ASSERT;
        ptr++;
        break;

        case '!':                 /* Negative lookahead */
        bravalue = OP_ASSERT_NOT;
        ptr++;
        break;

        case '<':                 /* Lookbehinds */
        switch (*(++ptr))
          {
          case '=':               /* Positive lookbehind */
          bravalue = OP_ASSERTBACK;
          ptr++;
          break;

          case '!':               /* Negative lookbehind */
          bravalue = OP_ASSERTBACK_NOT;
          ptr++;
          break;

          default:                /* Syntax error */
          *errorptr = ERR24;
          goto FAILED;
          }
        break;

        case '>':                 /* One-time brackets */
        bravalue = OP_ONCE;
        ptr++;
        break;

        case 'R':                 /* Pattern recursion */
        *code++ = OP_RECURSE;
        ptr++;
        continue;

        default:                  /* Option setting */
        set = unset = 0;
        optset = &set;

        while (*ptr != ')' && *ptr != ':')
          {
          switch (*ptr++)
            {
            case '-': optset = &unset; break;

            case 'i': *optset |= PCRE_CASELESS; break;
            case 'm': *optset |= PCRE_MULTILINE; break;
            case 's': *optset |= PCRE_DOTALL; break;
            case 'x': *optset |= PCRE_EXTENDED; break;
            case 'U': *optset |= PCRE_UNGREEDY; break;
            case 'X': *optset |= PCRE_EXTRA; break;

            default:
            *errorptr = ERR12;
            goto FAILED;
            }
          }

        /* Set up the changed option bits, but don't change anything yet. */

        newoptions = (options | set) & (~unset);

        /* If the options ended with ')' this is not the start of a nested
        group with option changes, so the options change at this level. At top
        level there is nothing else to be done (the options will in fact have
        been set from the start of compiling as a result of the first pass) but
        at an inner level we must compile code to change the ims options if
        necessary, and pass the new setting back so that it can be put at the
        start of any following branches, and when this group ends, a resetting
        item can be compiled. */

        if (*ptr == ')')
          {
          if ((options & PCRE_INGROUP) != 0 &&
              (options & PCRE_IMS) != (newoptions & PCRE_IMS))
            {
            *code++ = OP_OPT;
            *code++ = *optchanged = newoptions & PCRE_IMS;
            }
          options = newoptions;  /* Change options at this level */
          previous = NULL;       /* This item can't be repeated */
          continue;              /* It is complete */
          }

        /* If the options ended with ':' we are heading into a nested group
        with possible change of options. Such groups are non-capturing and are
        not assertions of any kind. All we need to do is skip over the ':';
        the newoptions value is handled below. */

        bravalue = OP_BRA;
        ptr++;
        }
      }

    /* Else we have a referencing group; adjust the opcode. */

    else
      {
      if (++(*brackets) > EXTRACT_MAX)
        {
        *errorptr = ERR13;
        goto FAILED;
        }
      bravalue = OP_BRA + *brackets;
      }

    /* Process nested bracketed re. Assertions may not be repeated, but other
    kinds can be. We copy code into a non-register variable in order to be able
    to pass its address because some compilers complain otherwise. Pass in a
    new setting for the ims options if they have changed. */

    previous = (bravalue >= OP_ONCE)? code : NULL;
    *code = bravalue;
    tempcode = code;

    if (!compile_regex(
         options | PCRE_INGROUP,       /* Set for all nested groups */
         ((options & PCRE_IMS) != (newoptions & PCRE_IMS))?
           newoptions & PCRE_IMS : -1, /* Pass ims options if changed */
         brackets,                     /* Bracket level */
         &tempcode,                    /* Where to put code (updated) */
         &ptr,                         /* Input pointer (updated) */
         errorptr,                     /* Where to put an error message */
         (bravalue == OP_ASSERTBACK ||
          bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
         condref,                      /* Condition reference number */
         &subreqchar,                  /* For possible last char */
         &subcountlits,                /* For literal count */
         cd))                          /* Tables block */
      goto FAILED;

    /* At the end of compiling, code is still pointing to the start of the
    group, while tempcode has been updated to point past the end of the group
    and any option resetting that may follow it. The pattern pointer (ptr)
    is on the bracket. */

    /* If this is a conditional bracket, check that there are no more than
    two branches in the group. */

    if (bravalue == OP_COND)
      {
      uschar *tc = code;
      condcount = 0;

      do {
         condcount++;
         tc += (tc[1] << 8) | tc[2];
         }
      while (*tc != OP_KET);

      if (condcount > 2)
        {
        *errorptr = ERR27;
        goto FAILED;
        }
      }

    /* Handle updating of the required character. If the subpattern didn't
    set one, leave it as it was. Otherwise, update it for normal brackets of
    all kinds, forward assertions, and conditions with two branches. Don't
    update the literal count for forward assertions, however. If the bracket
    is followed by a quantifier with zero repeat, we have to back off. Hence
    the definition of prevreqchar and subcountlits outside the main loop so
    that they can be accessed for the back off. */

    if (subreqchar > 0 &&
         (bravalue >= OP_BRA || bravalue == OP_ONCE || bravalue == OP_ASSERT ||
         (bravalue == OP_COND && condcount == 2)))
      {
      prevreqchar = *reqchar;
      *reqchar = subreqchar;
      if (bravalue != OP_ASSERT) *countlits += subcountlits;
      }

    /* Now update the main code pointer to the end of the group. */

    code = tempcode;

    /* Error if hit end of pattern */

    if (*ptr != ')')
      {
      *errorptr = ERR14;
      goto FAILED;
      }
    break;

    /* Check \ for being a real metacharacter; if not, fall through and handle
    it as a data character at the start of a string. Escape items are checked
    for validity in the pre-compiling pass. */

    case '\\':
    tempptr = ptr;
    c = check_escape(&ptr, errorptr, *brackets, options, FALSE, cd);

    /* Handle metacharacters introduced by \. For ones like \d, the ESC_ values
    are arranged to be the negation of the corresponding OP_values. For the
    back references, the values are ESC_REF plus the reference number. Only
    back references and those types that consume a character may be repeated.
    We can test for values between ESC_b and ESC_Z for the latter; this may
    have to change if any new ones are ever created. */

    if (c < 0)
      {
      if (-c >= ESC_REF)
        {
        previous = code;
        *code++ = OP_REF;
        *code++ = -c - ESC_REF;
        }
      else
        {
        previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
        *code++ = -c;
        }
      continue;
      }

    /* Data character: reset and fall through */

    ptr = tempptr;
    c = '\\';

    /* Handle a run of data characters until a metacharacter is encountered.
    The first character is guaranteed not to be whitespace or # when the
    extended flag is set. */

    NORMAL_CHAR:
    default:
    previous = code;
    *code = OP_CHARS;
    code += 2;
    length = 0;

    do
      {
      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 != '\n') ;
          if (c == 0) break;
          continue;
          }
        }

      /* Backslash may introduce a data char or a metacharacter. Escaped items
      are checked for validity in the pre-compiling pass. Stop the string
      before a metaitem. */

      if (c == '\\')
        {
        tempptr = ptr;
        c = check_escape(&ptr, errorptr, *brackets, options, FALSE, cd);
        if (c < 0) { ptr = tempptr; break; }

        /* If a character is > 127 in UTF-8 mode, we have to turn it into
        two or more characters in the UTF-8 encoding. */

#ifdef SUPPORT_UTF8
        if (c > 127 && (options & PCRE_UTF8) != 0)
          {
          uschar buffer[8];
          int len = ord2utf8(c, buffer);
          for (c = 0; c < len; c++) *code++ = buffer[c];
          length += len;
          continue;
          }
#endif
        }

      /* Ordinary character or single-char escape */

      *code++ = c;
      length++;
      }

    /* This "while" is the end of the "do" above. */

    while (length < MAXLIT && (cd->ctypes[c = *(++ptr)] & ctype_meta) == 0);

    /* Update the last character and the count of literals */

    prevreqchar = (length > 1)? code[-2] : *reqchar;
    *reqchar = code[-1];
    *countlits += length;

    /* Compute the length and set it in the data vector, and advance to
    the next state. */

    previous[1] = length;
    if (length < MAXLIT) ptr--;
    break;
    }
  }                   /* end of big loop */

/* Control never reaches here by falling through, only by a goto for all the
error states. Pass back the position in the pattern so that it can be displayed
to the user for diagnosing the error. */

FAILED:
*ptrptr = ptr;
return FALSE;
}


/*************************************************
*     Compile sequence of alternatives           *
*************************************************/

/* On entry, ptr is pointing past the bracket character, but on return
it points to the closing bracket, or vertical bar, or end of string.
The code variable is pointing at the byte into which the BRA operator has been
stored. If the ims options are changed at the start (for a (?ims: group) or
during any branch, we need to insert an OP_OPT item at the start of every
following branch to ensure they get set correctly at run time, and also pass
the new options into every subsequent branch compile.

Argument:
  options     the option bits
  optchanged  new ims options to set as if (?ims) were at the start, or -1
               for no change
  brackets    -> int containing the number of extracting brackets used
  codeptr     -> the address of the current code pointer
  ptrptr      -> the address of the current pattern pointer
  errorptr    -> pointer to error message
  lookbehind  TRUE if this is a lookbehind assertion
  condref     >= 0 for OPT_CREF setting at start of conditional group
  reqchar     -> place to put the last required character, or a negative number
  countlits   -> place to put the shortest literal count of any branch
  cd          points to the data block with tables pointers

Returns:      TRUE on success
*/

static BOOL
compile_regex(int options, int optchanged, int *brackets, uschar **codeptr,
  const uschar **ptrptr, const char **errorptr, BOOL lookbehind, int condref,
  int *reqchar, int *countlits, compile_data *cd)
{
const uschar *ptr = *ptrptr;
uschar *code = *codeptr;
uschar *last_branch = code;
uschar *start_bracket = code;
uschar *reverse_count = NULL;
int oldoptions = options & PCRE_IMS;
int branchreqchar, branchcountlits;

*reqchar = -1;
*countlits = INT_MAX;
code += 3;

/* At the start of a reference-based conditional group, insert the reference
number as an OP_CREF item. */

if (condref >= 0)
  {
  *code++ = OP_CREF;
  *code++ = condref;
  }

/* Loop for each alternative branch */

for (;;)
  {
  int length;

  /* Handle change of options */

  if (optchanged >= 0)
    {
    *code++ = OP_OPT;
    *code++ = optchanged;
    options = (options & ~PCRE_IMS) | optchanged;
    }

  /* Set up dummy OP_REVERSE if lookbehind assertion */

  if (lookbehind)
    {
    *code++ = OP_REVERSE;
    reverse_count = code;
    *code++ = 0;
    *code++ = 0;
    }

  /* Now compile the branch */

  if (!compile_branch(options, brackets, &code, &ptr, errorptr, &optchanged,
      &branchreqchar, &branchcountlits, cd))
    {
    *ptrptr = ptr;
    return FALSE;
    }

  /* Fill in the length of the last branch */

  length = code - last_branch;
  last_branch[1] = length >> 8;
  last_branch[2] = length & 255;

  /* Save the last required character if all branches have the same; a current
  value of -1 means unset, while -2 means "previous branch had no last required
  char".  */

  if (*reqchar != -2)
    {
    if (branchreqchar >= 0)
      {
      if (*reqchar == -1) *reqchar = branchreqchar;
      else if (*reqchar != branchreqchar) *reqchar = -2;
      }
    else *reqchar = -2;
    }

  /* Keep the shortest literal count */

  if (branchcountlits < *countlits) *countlits = branchcountlits;
  DPRINTF(("literal count = %d min=%d\n", branchcountlits, *countlits));

  /* If lookbehind, check that this branch matches a fixed-length string,
  and put the length into the OP_REVERSE item. Temporarily mark the end of
  the branch with OP_END. */

  if (lookbehind)
    {
    *code = OP_END;
    length = find_fixedlength(last_branch, options);
    DPRINTF(("fixed length = %d\n", length));
    if (length < 0)
      {
      *errorptr = ERR25;
      *ptrptr = ptr;
      return FALSE;
      }
    reverse_count[0] = (length >> 8);
    reverse_count[1] = length & 255;
    }

  /* Reached end of expression, either ')' or end of pattern. Insert a
  terminating ket and the length of the whole bracketed item, and return,
  leaving the pointer at the terminating char. If any of the ims options
  were changed inside the group, compile a resetting op-code following. */

  if (*ptr != '|')
    {
    length = code - start_bracket;
    *code++ = OP_KET;
    *code++ = length >> 8;
    *code++ = length & 255;
    if (optchanged >= 0)
      {
      *code++ = OP_OPT;
      *code++ = oldoptions;
      }
    *codeptr = code;
    *ptrptr = ptr;
    return TRUE;
    }

  /* Another branch follows; insert an "or" node and advance the pointer. */

  *code = OP_ALT;
  last_branch = code;
  code += 3;
  ptr++;
  }
/* Control never reaches here */
}


/*************************************************
*      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 one application, a change of caseless option is
important.

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
  optstop    TRUE to return on option change, otherwise change the options
               value and continue

Returns:     pointer to the first significant opcode
*/

static const uschar*
first_significant_code(const uschar *code, int *options, int optbit,
  BOOL optstop)
{
for (;;)
  {
  switch ((int)*code)
    {
    case OP_OPT:
    if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
      {
      if (optstop) return code;
      *options = (int)code[1];
      }
    code += 2;
    break;

    case OP_CREF:
    code += 2;
    break;

    case OP_WORD_BOUNDARY:
    case OP_NOT_WORD_BOUNDARY:
    code++;
    break;

    case OP_ASSERT_NOT:
    case OP_ASSERTBACK:
    case OP_ASSERTBACK_NOT:
    do code += (code[1] << 8) + code[2]; while (*code == OP_ALT);
    code += 3;
    break;

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


/*************************************************
*          Check for anchored expression         *
*************************************************/

/* Try to find out if this is an anchored regular expression. Consider each
alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
it's anchored. However, if this is a multiline pattern, then only OP_SOD
counts, since OP_CIRC can match in the middle.

A branch is also implicitly anchored if it starts with .* and DOTALL is set,
because that will try the rest of the pattern at all possible matching points,
so there is no point trying them again.

Arguments:
  code       points to start of expression (the bracket)
  options    points to the options setting

Returns:     TRUE or FALSE
*/

static BOOL
is_anchored(register const uschar *code, int *options)
{
do {
   const uschar *scode = first_significant_code(code + 3, options,
     PCRE_MULTILINE, FALSE);
   register int op = *scode;
   if (op >= OP_BRA || op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
     { if (!is_anchored(scode, options)) return FALSE; }
   else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR) &&
            (*options & PCRE_DOTALL) != 0)
     { if (scode[1] != OP_ANY) return FALSE; }
   else if (op != OP_SOD &&
           ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
     return FALSE;
   code += (code[1] << 8) + code[2];
   }
while (*code == OP_ALT);
return TRUE;
}


/*************************************************
*         Check for starting with ^ or .*        *
*************************************************/

/* This is called to find out if every branch starts with ^ or .* so that
"first char" processing can be done to speed things up in multiline
matching and for non-DOTALL patterns that start with .* (which must start at
the beginning or after \n).

Argument:  points to start of expression (the bracket)
Returns:   TRUE or FALSE
*/

static BOOL
is_startline(const uschar *code)
{
do {
   const uschar *scode = first_significant_code(code + 3, NULL, 0, FALSE);
   register int op = *scode;
   if (op >= OP_BRA || op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
     { if (!is_startline(scode)) return FALSE; }
   else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR)
     { if (scode[1] != OP_ANY) return FALSE; }
   else if (op != OP_CIRC) return FALSE;
   code += (code[1] << 8) + code[2];
   }
while (*code == OP_ALT);
return TRUE;
}


/*************************************************
*          Check for fixed first char            *
*************************************************/

/* Try to find out if there is a fixed first character. This is called for
unanchored expressions, as it speeds up their processing quite considerably.
Consider each alternative branch. If they all start with the same char, or with
a bracket all of whose alternatives start with the same char (recurse ad lib),
then we return that char, otherwise -1.

Arguments:
  code       points to start of expression (the bracket)
  options    pointer to the options (used to check casing changes)

Returns:     -1 or the fixed first char
*/

static int
find_firstchar(const uschar *code, int *options)
{
register int c = -1;
do {
   int d;
   const uschar *scode = first_significant_code(code + 3, options,
     PCRE_CASELESS, TRUE);
   register int op = *scode;

   if (op >= OP_BRA) op = OP_BRA;

   switch(op)
     {
     default:
     return -1;

     case OP_BRA:
     case OP_ASSERT:
     case OP_ONCE:
     case OP_COND:
     if ((d = find_firstchar(scode, options)) < 0) return -1;
     if (c < 0) c = d; else if (c != d) return -1;
     break;

     case OP_EXACT:       /* Fall through */
     scode++;

     case OP_CHARS:       /* Fall through */
     scode++;

     case OP_PLUS:
     case OP_MINPLUS:
     if (c < 0) c = scode[1]; else if (c != scode[1]) return -1;
     break;
     }

   code += (code[1] << 8) + code[2];
   }
while (*code == OP_ALT);
return c;
}


/*************************************************
*        Compile a Regular Expression            *
*************************************************/

/* This function takes a string and returns a pointer to a block of store
holding a compiled version of the expression.

Arguments:
  pattern      the regular expression
  options      various option bits
  errorptr     pointer to pointer to error text
  erroroffset  ptr offset in pattern where error was detected
  tables       pointer to character tables or NULL

Returns:       pointer to compiled data block, or NULL on error,
               with errorptr and erroroffset set
*/

pcre *
pcre_compile(const char *pattern, int options, const char **errorptr,
  int *erroroffset, const unsigned char *tables)
{
real_pcre *re;
int length = 3;      /* For initial BRA plus length */
int runlength;
int c, reqchar, countlits;
int bracount = 0;
int top_backref = 0;
int branch_extra = 0;
int branch_newextra;
unsigned int brastackptr = 0;
size_t size;
uschar *code;
const uschar *ptr;
compile_data compile_block;
int brastack[BRASTACK_SIZE];
uschar bralenstack[BRASTACK_SIZE];

#ifdef DEBUG
uschar *code_base, *code_end;
#endif

/* Can't support UTF8 unless PCRE has been compiled to include the code. */

#ifndef SUPPORT_UTF8
if ((options & PCRE_UTF8) != 0)
  {
  *errorptr = ERR32;
  return NULL;
  }
#endif

/* We can't pass back an error message if errorptr is NULL; I guess the best we
can do is just return NULL. */

if (errorptr == NULL) return NULL;
*errorptr = NULL;

/* However, we can give a message for this error */

if (erroroffset == NULL)
  {
  *errorptr = ERR16;
  return NULL;
  }
*erroroffset = 0;

if ((options & ~PUBLIC_OPTIONS) != 0)
  {
  *errorptr = ERR17;
  return NULL;
  }

/* Set up pointers to the individual character tables */

if (tables == NULL) tables = pcre_default_tables;
compile_block.lcc = tables + lcc_offset;
compile_block.fcc = tables + fcc_offset;
compile_block.cbits = tables + cbits_offset;
compile_block.ctypes = tables + ctypes_offset;

/* Reflect pattern for debugging output */

DPRINTF(("------------------------------------------------------------------\n"));
DPRINTF(("%s\n", pattern));

/* The first thing to do is to make a pass over the pattern to compute the
amount of store required to hold the compiled code. This does not have to be
perfect as long as errors are overestimates. At the same time we can detect any
internal flag settings. Make an attempt to correct for any counted white space
if an "extended" flag setting appears late in the pattern. We can't be so
clever for #-comments. */

ptr = (const uschar *)(pattern - 1);
while ((c = *(++ptr)) != 0)
  {
  int min, max;
  int class_charcount;

  if ((options & PCRE_EXTENDED) != 0)
    {
    if ((compile_block.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 != '\n') ;
      continue;
      }
    }

  switch(c)
    {
    /* A backslashed item may be an escaped "normal" character or a
    character type. For a "normal" character, put the pointers and
    character back so that tests for whitespace etc. in the input
    are done correctly. */

    case '\\':
      {
      const uschar *save_ptr = ptr;
      c = check_escape(&ptr, errorptr, bracount, options, FALSE, &compile_block);
      if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
      if (c >= 0)
        {
        ptr = save_ptr;
        c = '\\';
        goto NORMAL_CHAR;
        }
      }
    length++;

    /* A back reference needs an additional char, plus either one or 5
    bytes for a repeat. We also need to keep the value of the highest
    back reference. */

    if (c <= -ESC_REF)
      {
      int refnum = -c - ESC_REF;
      if (refnum > top_backref) top_backref = refnum;
      length++;   /* For single back reference */
      if (ptr[1] == '{' && is_counted_repeat(ptr+2, &compile_block))
        {
        ptr = read_repeat_counts(ptr+2, &min, &max, errorptr, &compile_block);
        if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
        if ((min == 0 && (max == 1 || max == -1)) ||
          (min == 1 && max == -1))
            length++;
        else length += 5;
        if (ptr[1] == '?') ptr++;
        }
      }
    continue;

    case '^':
    case '.':
    case '$':
    case '*':     /* These repeats won't be after brackets; */
    case '+':     /* those are handled separately */
    case '?':
    length++;
    continue;

    /* This covers the cases of repeats after a single char, metachar, class,
    or back reference. */

    case '{':
    if (!is_counted_repeat(ptr+1, &compile_block)) goto NORMAL_CHAR;
    ptr = read_repeat_counts(ptr+1, &min, &max, errorptr, &compile_block);
    if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
    if ((min == 0 && (max == 1 || max == -1)) ||
      (min == 1 && max == -1))
        length++;
    else
      {
      length--;   /* Uncount the original char or metachar */
      if (min == 1) length++; else if (min > 0) length += 4;
      if (max > 0) length += 4; else length += 2;
      }
    if (ptr[1] == '?') ptr++;
    continue;

    /* An alternation contains an offset to the next branch or ket. If any ims
    options changed in the previous branch(es), and/or if we are in a
    lookbehind assertion, extra space will be needed at the start of the
    branch. This is handled by branch_extra. */

    case '|':
    length += 3 + branch_extra;
    continue;

    /* A character class uses 33 characters. Don't worry about character types
    that aren't allowed in classes - they'll get picked up during the compile.
    A character class that contains only one character uses 2 or 3 bytes,
    depending on whether it is negated or not. Notice this where we can. */

    case '[':
    class_charcount = 0;
    if (*(++ptr) == '^') ptr++;
    do
      {
      if (*ptr == '\\')
        {
        int ch = check_escape(&ptr, errorptr, bracount, options, TRUE,
          &compile_block);
        if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
        if (-ch == ESC_b) class_charcount++; else class_charcount = 10;
        }
      else class_charcount++;
      ptr++;
      }
    while (*ptr != 0 && *ptr != ']');

    /* Repeats for negated single chars are handled by the general code */

    if (class_charcount == 1) length += 3; else
      {
      length += 33;

      /* A repeat needs either 1 or 5 bytes. */

      if (*ptr != 0 && ptr[1] == '{' && is_counted_repeat(ptr+2, &compile_block))
        {
        ptr = read_repeat_counts(ptr+2, &min, &max, errorptr, &compile_block);
        if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
        if ((min == 0 && (max == 1 || max == -1)) ||
          (min == 1 && max == -1))
            length++;
        else length += 5;
        if (ptr[1] == '?') ptr++;
        }
      }
    continue;

    /* Brackets may be genuine groups or special things */

    case '(':
    branch_newextra = 0;

    /* Handle special forms of bracket, which all start (? */

    if (ptr[1] == '?')
      {
      int set, unset;
      int *optset;

      switch (c = ptr[2])
        {
        /* Skip over comments entirely */
        case '#':
        ptr += 3;
        while (*ptr != 0 && *ptr != ')') ptr++;
        if (*ptr == 0)
          {
          *errorptr = ERR18;
          goto PCRE_ERROR_RETURN;
          }
        continue;

        /* Non-referencing groups and lookaheads just move the pointer on, and
        then behave like a non-special bracket, except that they don't increment
        the count of extracting brackets. Ditto for the "once only" bracket,
        which is in Perl from version 5.005. */

        case ':':
        case '=':
        case '!':
        case '>':
        ptr += 2;
        break;

        /* A recursive call to the regex is an extension, to provide the
        facility which can be obtained by $(?p{perl-code}) in Perl 5.6. */

        case 'R':
        if (ptr[3] != ')')
          {
          *errorptr = ERR29;
          goto PCRE_ERROR_RETURN;
          }
        ptr += 3;
        length += 1;
        break;

        /* Lookbehinds are in Perl from version 5.005 */

        case '<':
        if (ptr[3] == '=' || ptr[3] == '!')
          {
          ptr += 3;
          branch_newextra = 3;
          length += 3;         /* For the first branch */
          break;
          }
        *errorptr = ERR24;
        goto PCRE_ERROR_RETURN;

        /* Conditionals are in Perl from version 5.005. The bracket must either
        be followed by a number (for bracket reference) or by an assertion
        group. */

        case '(':
        if ((compile_block.ctypes[ptr[3]] & ctype_digit) != 0)
          {
          ptr += 4;
          length += 2;
          while ((compile_block.ctypes[*ptr] & ctype_digit) != 0) ptr++;
          if (*ptr != ')')
            {
            *errorptr = ERR26;
            goto PCRE_ERROR_RETURN;
            }
          }
        else   /* An assertion must follow */
          {
          ptr++;   /* Can treat like ':' as far as spacing is concerned */
          if (ptr[2] != '?' ||
             (ptr[3] != '=' && ptr[3] != '!' && ptr[3] != '<') )
            {
            ptr += 2;    /* To get right offset in message */
            *errorptr = ERR28;
            goto PCRE_ERROR_RETURN;
            }
          }
        break;

        /* Else loop checking valid options until ) is met. Anything else is an
        error. If we are without any brackets, i.e. at top level, the settings
        act as if specified in the options, so massage the options immediately.
        This is for backward compatibility with Perl 5.004. */

        default:
        set = unset = 0;
        optset = &set;
        ptr += 2;

        for (;; ptr++)
          {
          c = *ptr;
          switch (c)
            {
            case 'i':
            *optset |= PCRE_CASELESS;
            continue;

            case 'm':
            *optset |= PCRE_MULTILINE;
            continue;

            case 's':
            *optset |= PCRE_DOTALL;
            continue;

            case 'x':
            *optset |= PCRE_EXTENDED;
            continue;

            case 'X':
            *optset |= PCRE_EXTRA;
            continue;

            case 'U':
            *optset |= PCRE_UNGREEDY;
            continue;

            case '-':
            optset = &unset;
            continue;

            /* A termination by ')' indicates an options-setting-only item;
            this is global at top level; otherwise nothing is done here and
            it is handled during the compiling process on a per-bracket-group
            basis. */

            case ')':
            if (brastackptr == 0)
              {
              options = (options | set) & (~unset);
              set = unset = 0;     /* To save length */
              }
            /* Fall through */

            /* A termination by ':' indicates the start of a nested group with
            the given options set. This is again handled at compile time, but
            we must allow for compiled space if any of the ims options are
            set. We also have to allow for resetting space at the end of
            the group, which is why 4 is added to the length and not just 2.
            If there are several changes of options within the same group, this
            will lead to an over-estimate on the length, but this shouldn't
            matter very much. We also have to allow for resetting options at
            the start of any alternations, which we do by setting
            branch_newextra to 2. Finally, we record whether the case-dependent
            flag ever changes within the regex. This is used by the "required
            character" code. */

            case ':':
            if (((set|unset) & PCRE_IMS) != 0)
              {
              length += 4;
              branch_newextra = 2;
              if (((set|unset) & PCRE_CASELESS) != 0) options |= PCRE_ICHANGED;
              }
            goto END_OPTIONS;

            /* Unrecognized option character */

            default:
            *errorptr = ERR12;
            goto PCRE_ERROR_RETURN;
            }
          }

        /* If we hit a closing bracket, that's it - this is a freestanding
        option-setting. We need to ensure that branch_extra is updated if
        necessary. The only values branch_newextra can have here are 0 or 2.
        If the value is 2, then branch_extra must either be 2 or 5, depending
        on whether this is a lookbehind group or not. */

        END_OPTIONS:
        if (c == ')')
          {
          if (branch_newextra == 2 && (branch_extra == 0 || branch_extra == 3))
            branch_extra += branch_newextra;
          continue;
          }

        /* If options were terminated by ':' control comes here. Fall through
        to handle the group below. */
        }
      }

    /* Extracting brackets must be counted so we can process escapes in a
    Perlish way. */

    else bracount++;

    /* Non-special forms of bracket. Save length for computing whole length
    at end if there's a repeat that requires duplication of the group. Also
    save the current value of branch_extra, and start the new group with
    the new value. If non-zero, this will either be 2 for a (?imsx: group, or 3
    for a lookbehind assertion. */

    if (brastackptr >= sizeof(brastack)/sizeof(int))
      {
      *errorptr = ERR19;
      goto PCRE_ERROR_RETURN;
      }

    bralenstack[brastackptr] = branch_extra;
    branch_extra = branch_newextra;

    brastack[brastackptr++] = length;
    length += 3;
    continue;

    /* Handle ket. Look for subsequent max/min; for certain sets of values we
    have to replicate this bracket up to that many times. If brastackptr is
    0 this is an unmatched bracket which will generate an error, but take care
    not to try to access brastack[-1] when computing the length and restoring
    the branch_extra value. */

    case ')':
    length += 3;
      {
      int minval = 1;
      int maxval = 1;
      int duplength;

      if (brastackptr > 0)
        {
        duplength = length - brastack[--brastackptr];
        branch_extra = bralenstack[brastackptr];
        }
      else duplength = 0;

      /* Leave ptr at the final char; for read_repeat_counts this happens
      automatically; for the others we need an increment. */

      if ((c = ptr[1]) == '{' && is_counted_repeat(ptr+2, &compile_block))
        {
        ptr = read_repeat_counts(ptr+2, &minval, &maxval, errorptr,
          &compile_block);
        if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
        }
      else if (c == '*') { minval = 0; maxval = -1; ptr++; }
      else if (c == '+') { maxval = -1; ptr++; }
      else if (c == '?') { minval = 0; ptr++; }

      /* If the minimum is zero, we have to allow for an OP_BRAZERO before the
      group, and if the maximum is greater than zero, we have to replicate
      maxval-1 times; each replication acquires an OP_BRAZERO plus a nesting
      bracket set - hence the 7. */

      if (minval == 0)
        {
        length++;
        if (maxval > 0) length += (maxval - 1) * (duplength + 7);
        }

      /* When the minimum is greater than zero, 1 we have to replicate up to
      minval-1 times, with no additions required in the copies. Then, if
      there is a limited maximum we have to replicate up to maxval-1 times
      allowing for a BRAZERO item before each optional copy and nesting
      brackets for all but one of the optional copies. */

      else
        {
        length += (minval - 1) * duplength;
        if (maxval > minval)   /* Need this test as maxval=-1 means no limit */
          length += (maxval - minval) * (duplength + 7) - 6;
        }
      }
    continue;

    /* Non-special character. For a run of such characters the length required
    is the number of characters + 2, except that the maximum run length is 255.
    We won't get a skipped space or a non-data escape or the start of a #
    comment as the first character, so the length can't be zero. */

    NORMAL_CHAR:
    default:
    length += 2;
    runlength = 0;
    do
      {
      if ((options & PCRE_EXTENDED) != 0)
        {
        if ((compile_block.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 != '\n') ;
          continue;
          }
        }

      /* Backslash may introduce a data char or a metacharacter; stop the
      string before the latter. */

      if (c == '\\')
        {
        const uschar *saveptr = ptr;
        c = check_escape(&ptr, errorptr, bracount, options, FALSE,
          &compile_block);
        if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
        if (c < 0) { ptr = saveptr; break; }

#ifdef SUPPORT_UTF8
        if (c > 127 && (options & PCRE_UTF8) != 0)
          {
          int i;
          for (i = 0; i < sizeof(utf8_table1)/sizeof(int); i++)
            if (c <= utf8_table1[i]) break;
          runlength += i;
          }
#endif
        }

      /* Ordinary character or single-char escape */

      runlength++;
      }

    /* This "while" is the end of the "do" above. */

    while (runlength < MAXLIT &&
      (compile_block.ctypes[c = *(++ptr)] & ctype_meta) == 0);

    ptr--;
    length += runlength;
    continue;
    }
  }

length += 4;    /* For final KET and END */

if (length > 65539)
  {
  *errorptr = ERR20;
  return NULL;
  }

/* Compute the size of data block needed and get it, either from malloc or
externally provided function. We specify "code[0]" in the offsetof() expression
rather than just "code", because it has been reported that one broken compiler
fails on "code" because it is also an independent variable. It should make no
difference to the value of the offsetof(). */

size = length + offsetof(real_pcre, code[0]);
re = (real_pcre *)(pcre_malloc)(size);

if (re == NULL)
  {
  *errorptr = ERR21;
  return NULL;
  }

/* Put in the magic number, and save the size, options, and table pointer */

re->magic_number = MAGIC_NUMBER;
re->size = size;
re->options = options;
re->tables = tables;

/* Set up a starting, non-extracting bracket, then compile the expression. On
error, *errorptr will be set non-NULL, so we don't need to look at the result
of the function here. */

ptr = (const uschar *)pattern;
code = re->code;
*code = OP_BRA;
bracount = 0;
(void)compile_regex(options, -1, &bracount, &code, &ptr, errorptr, FALSE, -1,
  &reqchar, &countlits, &compile_block);
re->top_bracket = bracount;
re->top_backref = top_backref;

/* If not reached end of pattern on success, there's an excess bracket. */

if (*errorptr == NULL && *ptr != 0) *errorptr = ERR22;

/* Fill in the terminating state and check for disastrous overflow, but
if debugging, leave the test till after things are printed out. */

*code++ = OP_END;

#ifndef DEBUG
if (code - re->code > length) *errorptr = ERR23;
#endif

/* Give an error if there's back reference to a non-existent capturing
subpattern. */

if (top_backref > re->top_bracket) *errorptr = ERR15;

/* Failed to compile */

if (*errorptr != NULL)
  {
  (pcre_free)(re);
  PCRE_ERROR_RETURN:
  *erroroffset = ptr - (const uschar *)pattern;
  return NULL;
  }

/* If the anchored option was not passed, set flag if we can determine that the
pattern is anchored by virtue of ^ characters or \A or anything else (such as
starting with .* when DOTALL is set).

Otherwise, see if we can determine what the first character has to be, because
that speeds up unanchored matches no end. If not, see if we can set the
PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
start with ^. and also when all branches start with .* for non-DOTALL matches.
*/

if ((options & PCRE_ANCHORED) == 0)
  {
  int temp_options = options;
  if (is_anchored(re->code, &temp_options))
    re->options |= PCRE_ANCHORED;
  else
    {
    int ch = find_firstchar(re->code, &temp_options);
    if (ch >= 0)
      {
      re->first_char = ch;
      re->options |= PCRE_FIRSTSET;
      }
    else if (is_startline(re->code))
      re->options |= PCRE_STARTLINE;
    }
  }

/* Save the last required character if there are at least two literal
characters on all paths, or if there is no first character setting. */

if (reqchar >= 0 && (countlits > 1 || (re->options & PCRE_FIRSTSET) == 0))
  {
  re->req_char = reqchar;
  re->options |= PCRE_REQCHSET;
  }

/* Print out the compiled data for debugging */

#ifdef DEBUG

printf("Length = %d top_bracket = %d top_backref = %d\n",
  length, re->top_bracket, re->top_backref);

if (re->options != 0)
  {
  printf("%s%s%s%s%s%s%s%s%s\n",
    ((re->options & PCRE_ANCHORED) != 0)? "anchored " : "",
    ((re->options & PCRE_CASELESS) != 0)? "caseless " : "",
    ((re->options & PCRE_ICHANGED) != 0)? "case state changed " : "",
    ((re->options & PCRE_EXTENDED) != 0)? "extended " : "",
    ((re->options & PCRE_MULTILINE) != 0)? "multiline " : "",
    ((re->options & PCRE_DOTALL) != 0)? "dotall " : "",
    ((re->options & PCRE_DOLLAR_ENDONLY) != 0)? "endonly " : "",
    ((re->options & PCRE_EXTRA) != 0)? "extra " : "",
    ((re->options & PCRE_UNGREEDY) != 0)? "ungreedy " : "");
  }

if ((re->options & PCRE_FIRSTSET) != 0)
  {
  if (isprint(re->first_char)) printf("First char = %c\n", re->first_char);
    else printf("First char = \\x%02x\n", re->first_char);
  }

if ((re->options & PCRE_REQCHSET) != 0)
  {
  if (isprint(re->req_char)) printf("Req char = %c\n", re->req_char);
    else printf("Req char = \\x%02x\n", re->req_char);
  }

code_end = code;
code_base = code = re->code;

while (code < code_end)
  {
  int charlength;

  printf("%3d ", code - code_base);

  if (*code >= OP_BRA)
    {
    printf("%3d Bra %d", (code[1] << 8) + code[2], *code - OP_BRA);
    code += 2;
    }

  else switch(*code)
    {
    case OP_OPT:
    printf(" %.2x %s", code[1], OP_names[*code]);
    code++;
    break;

    case OP_COND:
    printf("%3d Cond", (code[1] << 8) + code[2]);
    code += 2;
    break;

    case OP_CREF:
    printf(" %.2d %s", code[1], OP_names[*code]);
    code++;
    break;

    case OP_CHARS:
    charlength = *(++code);
    printf("%3d ", charlength);
    while (charlength-- > 0)
      if (isprint(c = *(++code))) printf("%c", c); else printf("\\x%02x", c);
    break;

    case OP_KETRMAX:
    case OP_KETRMIN:
    case OP_ALT:
    case OP_KET:
    case OP_ASSERT:
    case OP_ASSERT_NOT:
    case OP_ASSERTBACK:
    case OP_ASSERTBACK_NOT:
    case OP_ONCE:
    printf("%3d %s", (code[1] << 8) + code[2], OP_names[*code]);
    code += 2;
    break;

    case OP_REVERSE:
    printf("%3d %s", (code[1] << 8) + code[2], OP_names[*code]);
    code += 2;
    break;

    case OP_STAR:
    case OP_MINSTAR:
    case OP_PLUS:
    case OP_MINPLUS:
    case OP_QUERY:
    case OP_MINQUERY:
    case OP_TYPESTAR:
    case OP_TYPEMINSTAR:
    case OP_TYPEPLUS:
    case OP_TYPEMINPLUS:
    case OP_TYPEQUERY:
    case OP_TYPEMINQUERY:
    if (*code >= OP_TYPESTAR)
      printf("    %s", OP_names[code[1]]);
    else if (isprint(c = code[1])) printf("    %c", c);
      else printf("    \\x%02x", c);
    printf("%s", OP_names[*code++]);
    break;

    case OP_EXACT:
    case OP_UPTO:
    case OP_MINUPTO:
    if (isprint(c = code[3])) printf("    %c{", c);
      else printf("    \\x%02x{", c);
    if (*code != OP_EXACT) printf("0,");
    printf("%d}", (code[1] << 8) + code[2]);
    if (*code == OP_MINUPTO) printf("?");
    code += 3;
    break;

    case OP_TYPEEXACT:
    case OP_TYPEUPTO:
    case OP_TYPEMINUPTO:
    printf("    %s{", OP_names[code[3]]);
    if (*code != OP_TYPEEXACT) printf(",");
    printf("%d}", (code[1] << 8) + code[2]);
    if (*code == OP_TYPEMINUPTO) printf("?");
    code += 3;
    break;

    case OP_NOT:
    if (isprint(c = *(++code))) printf("    [^%c]", c);
      else printf("    [^\\x%02x]", c);
    break;

    case OP_NOTSTAR:
    case OP_NOTMINSTAR:
    case OP_NOTPLUS:
    case OP_NOTMINPLUS:
    case OP_NOTQUERY:
    case OP_NOTMINQUERY:
    if (isprint(c = code[1])) printf("    [^%c]", c);
      else printf("    [^\\x%02x]", c);
    printf("%s", OP_names[*code++]);
    break;

    case OP_NOTEXACT:
    case OP_NOTUPTO:
    case OP_NOTMINUPTO:
    if (isprint(c = code[3])) printf("    [^%c]{", c);
      else printf("    [^\\x%02x]{", c);
    if (*code != OP_NOTEXACT) printf(",");
    printf("%d}", (code[1] << 8) + code[2]);
    if (*code == OP_NOTMINUPTO) printf("?");
    code += 3;
    break;

    case OP_REF:
    printf("    \\%d", *(++code));
    code ++;
    goto CLASS_REF_REPEAT;

    case OP_CLASS:
      {
      int i, min, max;
      code++;
      printf("    [");

      for (i = 0; i < 256; i++)
        {
        if ((code[i/8] & (1 << (i&7))) != 0)
          {
          int j;
          for (j = i+1; j < 256; j++)
            if ((code[j/8] & (1 << (j&7))) == 0) break;
          if (i == '-' || i == ']') printf("\\");
          if (isprint(i)) printf("%c", i); else printf("\\x%02x", i);
          if (--j > i)
            {
            printf("-");
            if (j == '-' || j == ']') printf("\\");
            if (isprint(j)) printf("%c", j); else printf("\\x%02x", j);
            }
          i = j;
          }
        }
      printf("]");
      code += 32;

      CLASS_REF_REPEAT:

      switch(*code)
        {
        case OP_CRSTAR:
        case OP_CRMINSTAR:
        case OP_CRPLUS:
        case OP_CRMINPLUS:
        case OP_CRQUERY:
        case OP_CRMINQUERY:
        printf("%s", OP_names[*code]);
        break;

        case OP_CRRANGE:
        case OP_CRMINRANGE:
        min = (code[1] << 8) + code[2];
        max = (code[3] << 8) + code[4];
        if (max == 0) printf("{%d,}", min);
        else printf("{%d,%d}", min, max);
        if (*code == OP_CRMINRANGE) printf("?");
        code += 4;
        break;

        default:
        code--;
        }
      }
    break;

    /* Anything else is just a one-node item */

    default:
    printf("    %s", OP_names[*code]);
    break;
    }

  code++;
  printf("\n");
  }
printf("------------------------------------------------------------------\n");

/* This check is done here in the debugging case so that the code that
was compiled can be seen. */

if (code - re->code > length)
  {
  *errorptr = ERR23;
  (pcre_free)(re);
  *erroroffset = ptr - (uschar *)pattern;
  return NULL;
  }
#endif

return (pcre *)re;
}


/*************************************************
*          Match a back-reference                *
*************************************************/

/* If a back reference hasn't been set, the length that is passed is greater
than the number of characters left in the string, so the match fails.

Arguments:
  offset      index into the offset vector
  eptr        points into the subject
  length      length to be matched
  md          points to match data block
  ims         the ims flags

Returns:      TRUE if matched
*/

static BOOL
match_ref(int offset, register const uschar *eptr, int length, match_data *md,
  unsigned long int ims)
{
const uschar *p = md->start_subject + md->offset_vector[offset];

#ifdef DEBUG
if (eptr >= md->end_subject)
  printf("matching subject <null>");
else
  {
  printf("matching subject ");
  pchars(eptr, length, TRUE, md);
  }
printf(" against backref ");
pchars(p, length, FALSE, md);
printf("\n");
#endif

/* Always fail if not enough characters left */

if (length > md->end_subject - eptr) return FALSE;

/* Separate the caselesss case for speed */

if ((ims & PCRE_CASELESS) != 0)
  {
  while (length-- > 0)
    if (md->lcc[*p++] != md->lcc[*eptr++]) return FALSE;
  }
else
  { while (length-- > 0) if (*p++ != *eptr++) return FALSE; }

return TRUE;
}


/*************************************************
*         Match from current position            *
*************************************************/

/* On entry ecode points to the first opcode, and eptr to the first character
in the subject string, while eptrb holds the value of eptr at the start of the
last bracketed group - used for breaking infinite loops matching zero-length
strings.

Arguments:
   eptr        pointer in subject
   ecode       position in code
   offset_top  current top pointer
   md          pointer to "static" info for the match
   ims         current /i, /m, and /s options
   eptrb       pointer to chain of blocks containing eptr at start of
                 brackets - for testing for empty matches
   flags       can contain
                 match_condassert - this is an assertion condition
                 match_isgroup - this is the start of a bracketed group

Returns:       TRUE if matched
*/

static BOOL
match(register const uschar *eptr, register const uschar *ecode,
  int offset_top, match_data *md, unsigned long int ims, eptrblock *eptrb,
  int flags)
{
unsigned long int original_ims = ims;   /* Save for resetting on ')' */
eptrblock newptrb;

/* At the start of a bracketed group, add the current subject pointer to the
stack of such pointers, to be re-instated at the end of the group when we hit
the closing ket. When match() is called in other circumstances, we don't add to
the stack. */

if ((flags & match_isgroup) != 0)
  {
  newptrb.prev = eptrb;
  newptrb.saved_eptr = eptr;
  eptrb = &newptrb;
  }

/* Now start processing the operations. */

for (;;)
  {
  int op = (int)*ecode;
  int min, max, ctype;
  register int i;
  register int c;
  BOOL minimize = FALSE;

  /* Opening capturing bracket. If there is space in the offset vector, save
  the current subject position in the working slot at the top of the vector. We
  mustn't change the current values of the data slot, because they may be set
  from a previous iteration of this group, and be referred to by a reference
  inside the group.

  If the bracket fails to match, we need to restore this value and also the
  values of the final offsets, in case they were set by a previous iteration of
  the same bracket.

  If there isn't enough space in the offset vector, treat this as if it were a
  non-capturing bracket. Don't worry about setting the flag for the error case
  here; that is handled in the code for KET. */

  if (op > OP_BRA)
    {
    int number = op - OP_BRA;
    int offset = number << 1;

#ifdef DEBUG
    printf("start bracket %d subject=", number);
    pchars(eptr, 16, TRUE, md);
    printf("\n");
#endif

    if (offset < md->offset_max)
      {
      int save_offset1 = md->offset_vector[offset];
      int save_offset2 = md->offset_vector[offset+1];
      int save_offset3 = md->offset_vector[md->offset_end - number];

      DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
      md->offset_vector[md->offset_end - number] = eptr - md->start_subject;

      do
        {
        if (match(eptr, ecode+3, offset_top, md, ims, eptrb, match_isgroup))
          return TRUE;
        ecode += (ecode[1] << 8) + ecode[2];
        }
      while (*ecode == OP_ALT);

      DPRINTF(("bracket %d failed\n", number));

      md->offset_vector[offset] = save_offset1;
      md->offset_vector[offset+1] = save_offset2;
      md->offset_vector[md->offset_end - number] = save_offset3;
      return FALSE;
      }

    /* Insufficient room for saving captured contents */

    else op = OP_BRA;
    }

  /* Other types of node can be handled by a switch */

  switch(op)
    {
    case OP_BRA:     /* Non-capturing bracket: optimized */
    DPRINTF(("start bracket 0\n"));
    do
      {
      if (match(eptr, ecode+3, offset_top, md, ims, eptrb, match_isgroup))
        return TRUE;
      ecode += (ecode[1] << 8) + ecode[2];
      }
    while (*ecode == OP_ALT);
    DPRINTF(("bracket 0 failed\n"));
    return FALSE;

    /* Conditional group: compilation checked that there are no more than
    two branches. If the condition is false, skipping the first branch takes us
    past the end if there is only one branch, but that's OK because that is
    exactly what going to the ket would do. */

    case OP_COND:
    if (ecode[3] == OP_CREF)         /* Condition is extraction test */
      {
      int offset = ecode[4] << 1;    /* Doubled reference number */
      return match(eptr,
        ecode + ((offset < offset_top && md->offset_vector[offset] >= 0)?
          5 : 3 + (ecode[1] << 8) + ecode[2]),
        offset_top, md, ims, eptrb, match_isgroup);
      }

    /* The condition is an assertion. Call match() to evaluate it - setting
    the final argument TRUE causes it to stop at the end of an assertion. */

    else
      {
      if (match(eptr, ecode+3, offset_top, md, ims, NULL,
          match_condassert | match_isgroup))
        {
        ecode += 3 + (ecode[4] << 8) + ecode[5];
        while (*ecode == OP_ALT) ecode += (ecode[1] << 8) + ecode[2];
        }
      else ecode += (ecode[1] << 8) + ecode[2];
      return match(eptr, ecode+3, offset_top, md, ims, eptrb, match_isgroup);
      }
    /* Control never reaches here */

    /* Skip over conditional reference data if encountered (should not be) */

    case OP_CREF:
    ecode += 2;
    break;

    /* End of the pattern. If PCRE_NOTEMPTY is set, fail if we have matched
    an empty string - recursion will then try other alternatives, if any. */

    case OP_END:
    if (md->notempty && eptr == md->start_match) return FALSE;
    md->end_match_ptr = eptr;          /* Record where we ended */
    md->end_offset_top = offset_top;   /* and how many extracts were taken */
    return TRUE;

    /* Change option settings */

    case OP_OPT:
    ims = ecode[1];
    ecode += 2;
    DPRINTF(("ims set to %02lx\n", ims));
    break;

    /* Assertion brackets. Check the alternative branches in turn - the
    matching won't pass the KET for an assertion. If any one branch matches,
    the assertion is true. Lookbehind assertions have an OP_REVERSE item at the
    start of each branch to move the current point backwards, so the code at
    this level is identical to the lookahead case. */

    case OP_ASSERT:
    case OP_ASSERTBACK:
    do
      {
      if (match(eptr, ecode+3, offset_top, md, ims, NULL, match_isgroup)) break;
      ecode += (ecode[1] << 8) + ecode[2];
      }
    while (*ecode == OP_ALT);
    if (*ecode == OP_KET) return FALSE;

    /* If checking an assertion for a condition, return TRUE. */

    if ((flags & match_condassert) != 0) return TRUE;

    /* Continue from after the assertion, updating the offsets high water
    mark, since extracts may have been taken during the assertion. */

    do ecode += (ecode[1] << 8) + ecode[2]; while (*ecode == OP_ALT);
    ecode += 3;
    offset_top = md->end_offset_top;
    continue;

    /* Negative assertion: all branches must fail to match */

    case OP_ASSERT_NOT:
    case OP_ASSERTBACK_NOT:
    do
      {
      if (match(eptr, ecode+3, offset_top, md, ims, NULL, match_isgroup))
        return FALSE;
      ecode += (ecode[1] << 8) + ecode[2];
      }
    while (*ecode == OP_ALT);

    if ((flags & match_condassert) != 0) return TRUE;

    ecode += 3;
    continue;

    /* Move the subject pointer back. This occurs only at the start of
    each branch of a lookbehind assertion. If we are too close to the start to
    move back, this match function fails. When working with UTF-8 we move
    back a number of characters, not bytes. */

    case OP_REVERSE:
#ifdef SUPPORT_UTF8
    c = (ecode[1] << 8) + ecode[2];
    for (i = 0; i < c; i++)
      {
      eptr--;
      BACKCHAR(eptr)
      }
#else
    eptr -= (ecode[1] << 8) + ecode[2];
#endif

    if (eptr < md->start_subject) return FALSE;
    ecode += 3;
    break;

    /* Recursion matches the current regex, nested. If there are any capturing
    brackets started but not finished, we have to save their starting points
    and reinstate them after the recursion. However, we don't know how many
    such there are (offset_top records the completed total) so we just have
    to save all the potential data. There may be up to 99 such values, which
    is a bit large to put on the stack, but using malloc for small numbers
    seems expensive. As a compromise, the stack is used when there are fewer
    than 16 values to store; otherwise malloc is used. A problem is what to do
    if the malloc fails ... there is no way of returning to the top level with
    an error. Save the top 15 values on the stack, and accept that the rest
    may be wrong. */

    case OP_RECURSE:
      {
      BOOL rc;
      int *save;
      int stacksave[15];

      c = md->offset_max;

      if (c < 16) save = stacksave; else
        {
        save = (int *)(pcre_malloc)((c+1) * sizeof(int));
        if (save == NULL)
          {
          save = stacksave;
          c = 15;
          }
        }

      for (i = 1; i <= c; i++)
        save[i] = md->offset_vector[md->offset_end - i];
      rc = match(eptr, md->start_pattern, offset_top, md, ims, eptrb,
        match_isgroup);
      for (i = 1; i <= c; i++)
        md->offset_vector[md->offset_end - i] = save[i];
      if (save != stacksave) (pcre_free)(save);
      if (!rc) return FALSE;

      /* In case the recursion has set more capturing values, save the final
      number, then move along the subject till after the recursive match,
      and advance one byte in the pattern code. */

      offset_top = md->end_offset_top;
      eptr = md->end_match_ptr;
      ecode++;
      }
    break;

    /* "Once" brackets are like assertion brackets except that after a match,
    the point in the subject string is not moved back. Thus there can never be
    a move back into the brackets. Check the alternative branches in turn - the
    matching won't pass the KET for this kind of subpattern. If any one branch
    matches, we carry on as at the end of a normal bracket, leaving the subject
    pointer. */

    case OP_ONCE:
      {
      const uschar *prev = ecode;
      const uschar *saved_eptr = eptr;

      do
        {
        if (match(eptr, ecode+3, offset_top, md, ims, eptrb, match_isgroup))
          break;
        ecode += (ecode[1] << 8) + ecode[2];
        }
      while (*ecode == OP_ALT);

      /* If hit the end of the group (which could be repeated), fail */

      if (*ecode != OP_ONCE && *ecode != OP_ALT) return FALSE;

      /* Continue as from after the assertion, updating the offsets high water
      mark, since extracts may have been taken. */

      do ecode += (ecode[1] << 8) + ecode[2]; while (*ecode == OP_ALT);

      offset_top = md->end_offset_top;
      eptr = md->end_match_ptr;

      /* For a non-repeating ket, just continue at this level. This also
      happens for a repeating ket if no characters were matched in the group.
      This is the forcible breaking of infinite loops as implemented in Perl
      5.005. If there is an options reset, it will get obeyed in the normal
      course of events. */

      if (*ecode == OP_KET || eptr == saved_eptr)
        {
        ecode += 3;
        break;
        }

      /* The repeating kets try the rest of the pattern or restart from the
      preceding bracket, in the appropriate order. We need to reset any options
      that changed within the bracket before re-running it, so check the next
      opcode. */

      if (ecode[3] == OP_OPT)
        {
        ims = (ims & ~PCRE_IMS) | ecode[4];
        DPRINTF(("ims set to %02lx at group repeat\n", ims));
        }

      if (*ecode == OP_KETRMIN)
        {
        if (match(eptr, ecode+3, offset_top, md, ims, eptrb, 0) ||
            match(eptr, prev, offset_top, md, ims, eptrb, match_isgroup))
              return TRUE;
        }
      else  /* OP_KETRMAX */
        {
        if (match(eptr, prev, offset_top, md, ims, eptrb, match_isgroup) ||
            match(eptr, ecode+3, offset_top, md, ims, eptrb, 0)) return TRUE;
        }
      }
    return FALSE;

    /* An alternation is the end of a branch; scan along to find the end of the
    bracketed group and go to there. */

    case OP_ALT:
    do ecode += (ecode[1] << 8) + ecode[2]; while (*ecode == OP_ALT);
    break;

    /* BRAZERO and BRAMINZERO occur just before a bracket group, indicating
    that it may occur zero times. It may repeat infinitely, or not at all -
    i.e. it could be ()* or ()? in the pattern. Brackets with fixed upper
    repeat limits are compiled as a number of copies, with the optional ones
    preceded by BRAZERO or BRAMINZERO. */

    case OP_BRAZERO:
      {
      const uschar *next = ecode+1;
      if (match(eptr, next, offset_top, md, ims, eptrb, match_isgroup))
        return TRUE;
      do next += (next[1] << 8) + next[2]; while (*next == OP_ALT);
      ecode = next + 3;
      }
    break;

    case OP_BRAMINZERO:
      {
      const uschar *next = ecode+1;
      do next += (next[1] << 8) + next[2]; while (*next == OP_ALT);
      if (match(eptr, next+3, offset_top, md, ims, eptrb, match_isgroup))
        return TRUE;
      ecode++;
      }
    break;

    /* End of a group, repeated or non-repeating. If we are at the end of
    an assertion "group", stop matching and return TRUE, but record the
    current high water mark for use by positive assertions. Do this also
    for the "once" (not-backup up) groups. */

    case OP_KET:
    case OP_KETRMIN:
    case OP_KETRMAX:
      {
      const uschar *prev = ecode - (ecode[1] << 8) - ecode[2];
      const uschar *saved_eptr = eptrb->saved_eptr;

      eptrb = eptrb->prev;    /* Back up the stack of bracket start pointers */

      if (*prev == OP_ASSERT || *prev == OP_ASSERT_NOT ||
          *prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT ||
          *prev == OP_ONCE)
        {
        md->end_match_ptr = eptr;      /* For ONCE */
        md->end_offset_top = offset_top;
        return TRUE;
        }

      /* In all other cases except a conditional group we have to check the
      group number back at the start and if necessary complete handling an
      extraction by setting the offsets and bumping the high water mark. */

      if (*prev != OP_COND)
        {
        int number = *prev - OP_BRA;
        int offset = number << 1;

#ifdef DEBUG
        printf("end bracket %d", number);
        printf("\n");
#endif

        if (number > 0)
          {
          if (offset >= md->offset_max) md->offset_overflow = TRUE; else
            {
            md->offset_vector[offset] =
              md->offset_vector[md->offset_end - number];
            md->offset_vector[offset+1] = eptr - md->start_subject;
            if (offset_top <= offset) offset_top = offset + 2;
            }
          }
        }

      /* Reset the value of the ims flags, in case they got changed during
      the group. */

      ims = original_ims;
      DPRINTF(("ims reset to %02lx\n", ims));

      /* For a non-repeating ket, just continue at this level. This also
      happens for a repeating ket if no characters were matched in the group.
      This is the forcible breaking of infinite loops as implemented in Perl
      5.005. If there is an options reset, it will get obeyed in the normal
      course of events. */

      if (*ecode == OP_KET || eptr == saved_eptr)
        {
        ecode += 3;
        break;
        }

      /* The repeating kets try the rest of the pattern or restart from the
      preceding bracket, in the appropriate order. */

      if (*ecode == OP_KETRMIN)
        {
        if (match(eptr, ecode+3, offset_top, md, ims, eptrb, 0) ||
            match(eptr, prev, offset_top, md, ims, eptrb, match_isgroup))
              return TRUE;
        }
      else  /* OP_KETRMAX */
        {
        if (match(eptr, prev, offset_top, md, ims, eptrb, match_isgroup) ||
            match(eptr, ecode+3, offset_top, md, ims, eptrb, 0)) return TRUE;
        }
      }
    return FALSE;

    /* Start of subject unless notbol, or after internal newline if multiline */

    case OP_CIRC:
    if (md->notbol && eptr == md->start_subject) return FALSE;
    if ((ims & PCRE_MULTILINE) != 0)
      {
      if (eptr != md->start_subject && eptr[-1] != '\n') return FALSE;
      ecode++;
      break;
      }
    /* ... else fall through */

    /* Start of subject assertion */

    case OP_SOD:
    if (eptr != md->start_subject) return FALSE;
    ecode++;
    break;

    /* Assert before internal newline if multiline, or before a terminating
    newline unless endonly is set, else end of subject unless noteol is set. */

    case OP_DOLL:
    if ((ims & PCRE_MULTILINE) != 0)
      {
      if (eptr < md->end_subject) { if (*eptr != '\n') return FALSE; }
        else { if (md->noteol) return FALSE; }
      ecode++;
      break;
      }
    else
      {
      if (md->noteol) return FALSE;
      if (!md->endonly)
        {
        if (eptr < md->end_subject - 1 ||
           (eptr == md->end_subject - 1 && *eptr != '\n')) return FALSE;

        ecode++;
        break;
        }
      }
    /* ... else fall through */

    /* End of subject assertion (\z) */

    case OP_EOD:
    if (eptr < md->end_subject) return FALSE;
    ecode++;
    break;

    /* End of subject or ending \n assertion (\Z) */

    case OP_EODN:
    if (eptr < md->end_subject - 1 ||
       (eptr == md->end_subject - 1 && *eptr != '\n')) return FALSE;
    ecode++;
    break;

    /* Word boundary assertions */

    case OP_NOT_WORD_BOUNDARY:
    case OP_WORD_BOUNDARY:
      {
      BOOL prev_is_word = (eptr != md->start_subject) &&
        ((md->ctypes[eptr[-1]] & ctype_word) != 0);
      BOOL cur_is_word = (eptr < md->end_subject) &&
        ((md->ctypes[*eptr] & ctype_word) != 0);
      if ((*ecode++ == OP_WORD_BOUNDARY)?
           cur_is_word == prev_is_word : cur_is_word != prev_is_word)
        return FALSE;
      }
    break;

    /* Match a single character type; inline for speed */

    case OP_ANY:
    if ((ims & PCRE_DOTALL) == 0 && eptr < md->end_subject && *eptr == '\n')
      return FALSE;
    if (eptr++ >= md->end_subject) return FALSE;
#ifdef SUPPORT_UTF8
    if (md->utf8)
      while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
#endif
    ecode++;
    break;

    case OP_NOT_DIGIT:
    if (eptr >= md->end_subject ||
       (md->ctypes[*eptr++] & ctype_digit) != 0)
      return FALSE;
    ecode++;
    break;

    case OP_DIGIT:
    if (eptr >= md->end_subject ||
       (md->ctypes[*eptr++] & ctype_digit) == 0)
      return FALSE;
    ecode++;
    break;

    case OP_NOT_WHITESPACE:
    if (eptr >= md->end_subject ||
       (md->ctypes[*eptr++] & ctype_space) != 0)
      return FALSE;
    ecode++;
    break;

    case OP_WHITESPACE:
    if (eptr >= md->end_subject ||
       (md->ctypes[*eptr++] & ctype_space) == 0)
      return FALSE;
    ecode++;
    break;

    case OP_NOT_WORDCHAR:
    if (eptr >= md->end_subject ||
       (md->ctypes[*eptr++] & ctype_word) != 0)
      return FALSE;
    ecode++;
    break;

    case OP_WORDCHAR:
    if (eptr >= md->end_subject ||
       (md->ctypes[*eptr++] & ctype_word) == 0)
      return FALSE;
    ecode++;
    break;

    /* Match a back reference, possibly repeatedly. Look past the end of the
    item to see if there is repeat information following. The code is similar
    to that for character classes, but repeated for efficiency. Then obey
    similar code to character type repeats - written out again for speed.
    However, if the referenced string is the empty string, always treat
    it as matched, any number of times (otherwise there could be infinite
    loops). */

    case OP_REF:
      {
      int length;
      int offset = ecode[1] << 1;                /* Doubled reference number */
      ecode += 2;                                /* Advance past the item */

      /* If the reference is unset, set the length to be longer than the amount
      of subject left; this ensures that every attempt at a match fails. We
      can't just fail here, because of the possibility of quantifiers with zero
      minima. */

      length = (offset >= offset_top || md->offset_vector[offset] < 0)?
        md->end_subject - eptr + 1 :
        md->offset_vector[offset+1] - md->offset_vector[offset];

      /* Set up for repetition, or handle the non-repeated case */

      switch (*ecode)
        {
        case OP_CRSTAR:
        case OP_CRMINSTAR:
        case OP_CRPLUS:
        case OP_CRMINPLUS:
        case OP_CRQUERY:
        case OP_CRMINQUERY:
        c = *ecode++ - OP_CRSTAR;
        minimize = (c & 1) != 0;
        min = rep_min[c];                 /* Pick up values from tables; */
        max = rep_max[c];                 /* zero for max => infinity */
        if (max == 0) max = INT_MAX;
        break;

        case OP_CRRANGE:
        case OP_CRMINRANGE:
        minimize = (*ecode == OP_CRMINRANGE);
        min = (ecode[1] << 8) + ecode[2];
        max = (ecode[3] << 8) + ecode[4];
        if (max == 0) max = INT_MAX;
        ecode += 5;
        break;

        default:               /* No repeat follows */
        if (!match_ref(offset, eptr, length, md, ims)) return FALSE;
        eptr += length;
        continue;              /* With the main loop */
        }

      /* If the length of the reference is zero, just continue with the
      main loop. */

      if (length == 0) continue;

      /* First, ensure the minimum number of matches are present. We get back
      the length of the reference string explicitly rather than passing the
      address of eptr, so that eptr can be a register variable. */

      for (i = 1; i <= min; i++)
        {
        if (!match_ref(offset, eptr, length, md, ims)) return FALSE;
        eptr += length;
        }

      /* If min = max, continue at the same level without recursion.
      They are not both allowed to be zero. */

      if (min == max) continue;

      /* If minimizing, keep trying and advancing the pointer */

      if (minimize)
        {
        for (i = min;; i++)
          {
          if (match(eptr, ecode, offset_top, md, ims, eptrb, 0))
            return TRUE;
          if (i >= max || !match_ref(offset, eptr, length, md, ims))
            return FALSE;
          eptr += length;
          }
        /* Control never gets here */
        }

      /* If maximizing, find the longest string and work backwards */

      else
        {
        const uschar *pp = eptr;
        for (i = min; i < max; i++)
          {
          if (!match_ref(offset, eptr, length, md, ims)) break;
          eptr += length;
          }
        while (eptr >= pp)
          {
          if (match(eptr, ecode, offset_top, md, ims, eptrb, 0))
            return TRUE;
          eptr -= length;
          }
        return FALSE;
        }
      }
    /* Control never gets here */


    /* Match a character class, possibly repeatedly. Look past the end of the
    item to see if there is repeat information following. Then obey similar
    code to character type repeats - written out again for speed. */

    case OP_CLASS:
      {
      const uschar *data = ecode + 1;  /* Save for matching */
      ecode += 33;                     /* Advance past the item */

      switch (*ecode)
        {
        case OP_CRSTAR:
        case OP_CRMINSTAR:
        case OP_CRPLUS:
        case OP_CRMINPLUS:
        case OP_CRQUERY:
        case OP_CRMINQUERY:
        c = *ecode++ - OP_CRSTAR;
        minimize = (c & 1) != 0;
        min = rep_min[c];                 /* Pick up values from tables; */
        max = rep_max[c];                 /* zero for max => infinity */
        if (max == 0) max = INT_MAX;
        break;

        case OP_CRRANGE:
        case OP_CRMINRANGE:
        minimize = (*ecode == OP_CRMINRANGE);
  