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

Diff of /code/trunk/pcre_compile.c

Parent Directory | Revision Log | View Patch Patch

-revision 87 by nigel,
Sat Feb 24 21:41:21 2007 UTC
+revision 349 by ph10,
Wed Jul  2 18:42:11 2008 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2006 University of Cambridge
+            Copyright (c) 1997-2008 University of Cambridge
  -----------------------------------------------------------------------------
  Redistribution and use in source and binary forms, with or without
 Line 42 
 POSSIBILITY OF SUCH DAMAGE.
  supporting internal functions that are not used by other modules. */
+ #ifdef HAVE_CONFIG_H
+ #include "config.h"
+ #endif
+ #define NLBLOCK cd             /* Block containing newline information */
+ #define PSSTART start_pattern  /* Field containing processed string start */
+ #define PSEND   end_pattern    /* Field containing processed string end */
  #include "pcre_internal.h"
-Line 53 
 used by pcretest. DEBUG is not defined w
+Line 61 
 used by pcretest. DEBUG is not defined w
  #endif
+ /* Macro for setting individual bits in class bitmaps. */
+ #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
+ /* Maximum length value to check against when making sure that the integer that
+ holds the compiled pattern length does not overflow. We make it a bit less than
+ INT_MAX to allow for adding in group terminating bytes, so that we don't have
+ to check them every time. */
+ #define OFLOW_MAX (INT_MAX - 20)
  /*************************************************
  *      Code parameters and static tables         *
  *************************************************/
- /* Maximum number of items on the nested bracket stacks at compile time. This
+ /* This value specifies the size of stack workspace that is used during the
- applies to the nesting of all kinds of parentheses. It does not limit
+ first pre-compile phase that determines how much memory is required. The regex
- un-nested, non-capturing parentheses. This number can be made bigger if
+ is partly compiled into this space, but the compiled parts are discarded as
- necessary - it is used to dimension one int and one unsigned char vector at
+ soon as they can be, so that hopefully there will never be an overrun. The code
- compile time. */
+ does, however, check for an overrun. The largest amount I've seen used is 218,
+ so this number is very generous.
+ The same workspace is used during the second, actual compile phase for
+ remembering forward references to groups so that they can be filled in at the
+ end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
+ is 4 there is plenty of room. */
- #define BRASTACK_SIZE 200
+ #define COMPILE_WORK_SIZE (4096)
  /* Table for handling escaped characters in the range '0'-'z'. Positive returns
-Line 72 
 are simple data values; negative values
+Line 97 
 are simple data values; negative values
  on. Zero means further processing is needed (for things like \x), or the escape
  is invalid. */
- #if !EBCDIC   /* This is the "normal" table for ASCII systems */
+ #ifndef EBCDIC  /* This is the "normal" table for ASCII systems */
  static const short int escapes[] = {
 ,      0,      0,      0,      0,      0,      0,      0,   /* 0 - 7 */
 ,      0,    ':',    ';',    '<',    '=',    '>',    '?',   /* 8 - ? */
     '@', -ESC_A, -ESC_B, -ESC_C, -ESC_D, -ESC_E,      0, -ESC_G,   /* @ - G */
-,      0,      0,      0,      0,      0,      0,      0,   /* H - O */
+ -ESC_H,      0,      0, -ESC_K,      0,      0,      0,      0,   /* H - O */
- -ESC_P, -ESC_Q,      0, -ESC_S,      0,      0,      0, -ESC_W,   /* P - W */
+ -ESC_P, -ESC_Q, -ESC_R, -ESC_S,      0,      0, -ESC_V, -ESC_W,   /* P - W */
  -ESC_X,      0, -ESC_Z,    '[',   '\\',    ']',    '^',    '_',   /* X - _ */
     '`',      7, -ESC_b,      0, -ESC_d,  ESC_e,  ESC_f,      0,   /* ` - g */
-,      0,      0,      0,      0,      0,  ESC_n,      0,   /* h - o */
+ -ESC_h,      0,      0, -ESC_k,      0,      0,  ESC_n,      0,   /* h - o */
- -ESC_p,      0,  ESC_r, -ESC_s,  ESC_tee,    0,      0, -ESC_w,   /* p - w */
+ -ESC_p,      0,  ESC_r, -ESC_s,  ESC_tee,    0, -ESC_v, -ESC_w,   /* p - w */
 ,      0, -ESC_z                                            /* x - z */
  };
- #else         /* This is the "abnormal" table for EBCDIC systems */
+ #else           /* This is the "abnormal" table for EBCDIC systems */
  static const short int escapes[] = {
  /*  48 */     0,     0,      0,     '.',    '<',   '(',    '+',    '|',
  /*  50 */   '&',     0,      0,       0,      0,     0,      0,      0,
-Line 96 
 static const short int escapes[] = {
+Line 121 
 static const short int escapes[] = {
  /*  70 */     0,     0,      0,       0,      0,     0,      0,      0,
  /*  78 */     0,   '`',    ':',     '#',    '@',  '\'',    '=',    '"',
  /*  80 */     0,     7, -ESC_b,       0, -ESC_d, ESC_e,  ESC_f,      0,
- /*  88 */     0,     0,      0,     '{',      0,     0,      0,      0,
+ /*  88 */-ESC_h,     0,      0,     '{',      0,     0,      0,      0,
- /*  90 */     0,     0,      0,     'l',      0, ESC_n,      0, -ESC_p,
+ /*  90 */     0,     0, -ESC_k,     'l',      0, ESC_n,      0, -ESC_p,
  /*  98 */     0, ESC_r,      0,     '}',      0,     0,      0,      0,
- /*  A0 */     0,   '~', -ESC_s, ESC_tee,      0,     0, -ESC_w,      0,
+ /*  A0 */     0,   '~', -ESC_s, ESC_tee,      0,-ESC_v, -ESC_w,      0,
  /*  A8 */     0,-ESC_z,      0,       0,      0,   '[',      0,      0,
  /*  B0 */     0,     0,      0,       0,      0,     0,      0,      0,
  /*  B8 */     0,     0,      0,       0,      0,   ']',    '=',    '-',
  /*  C0 */   '{',-ESC_A, -ESC_B,  -ESC_C, -ESC_D,-ESC_E,      0, -ESC_G,
- /*  C8 */     0,     0,      0,       0,      0,     0,      0,      0,
+ /*  C8 */-ESC_H,     0,      0,       0,      0,     0,      0,      0,
- /*  D0 */   '}',     0,      0,       0,      0,     0,      0, -ESC_P,
+ /*  D0 */   '}',     0, -ESC_K,       0,      0,     0,      0, -ESC_P,
- /*  D8 */-ESC_Q,     0,      0,       0,      0,     0,      0,      0,
+ /*  D8 */-ESC_Q,-ESC_R,      0,       0,      0,     0,      0,      0,
- /*  E0 */  '\\',     0, -ESC_S,       0,      0,     0, -ESC_W, -ESC_X,
+ /*  E0 */  '\\',     0, -ESC_S,       0,      0,-ESC_V, -ESC_W, -ESC_X,
  /*  E8 */     0,-ESC_Z,      0,       0,      0,     0,      0,      0,
  /*  F0 */     0,     0,      0,       0,      0,     0,      0,      0,
  /*  F8 */     0,     0,      0,       0,      0,     0,      0,      0
-Line 115 
 static const short int escapes[] = {
+Line 140 
 static const short int escapes[] = {
  #endif
- /* Tables of names of POSIX character classes and their lengths. The list is
+ /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
- terminated by a zero length entry. The first three must be alpha, lower, upper,
+ searched linearly. Put all the names into a single string, in order to reduce
- as this is assumed for handling case independence. */
+ the number of relocations when a shared library is dynamically linked. */
- static const char *const posix_names[] = {
+ typedef struct verbitem {
-   "alpha", "lower", "upper",
+   int   len;
-   "alnum", "ascii", "blank", "cntrl", "digit", "graph",
+   int   op;
-   "print", "punct", "space", "word",  "xdigit" };
+ } verbitem;
+ static const char verbnames[] =
+   "ACCEPT\0"
+   "COMMIT\0"
+   "F\0"
+   "FAIL\0"
+   "PRUNE\0"
+   "SKIP\0"
+   "THEN";
+ static const verbitem verbs[] = {
+   { 6, OP_ACCEPT },
+   { 6, OP_COMMIT },
+   { 1, OP_FAIL },
+   { 4, OP_FAIL },
+   { 5, OP_PRUNE },
+   { 4, OP_SKIP  },
+   { 4, OP_THEN  }
+ };
+ static const int verbcount = sizeof(verbs)/sizeof(verbitem);
+ /* Tables of names of POSIX character classes and their lengths. The names are
+ now all in a single string, to reduce the number of relocations when a shared
+ library is dynamically loaded. The list of lengths is terminated by a zero
+ length entry. The first three must be alpha, lower, upper, as this is assumed
+ for handling case independence. */
+ static const char posix_names[] =
+   "alpha\0"  "lower\0"  "upper\0"  "alnum\0"  "ascii\0"  "blank\0"
+   "cntrl\0"  "digit\0"  "graph\0"  "print\0"  "punct\0"  "space\0"
+   "word\0"   "xdigit";
  static const uschar posix_name_lengths[] = {
 , 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
-Line 155 
 static const int posix_class_maps[] = {
+Line 213 
 static const int posix_class_maps[] = {
  };
- /* The texts of compile-time error messages. These are "char *" because they
+ #define STRING(a)  # a
- are passed to the outside world. */
+ #define XSTRING(s) STRING(s)
- static const char *error_texts[] = {
+ /* The texts of compile-time error messages. These are "char *" because they
-   "no error",
+ are passed to the outside world. Do not ever re-use any error number, because
-   "\\ at end of pattern",
+ they are documented. Always add a new error instead. Messages marked DEAD below
-   "\\c at end of pattern",
+ are no longer used. This used to be a table of strings, but in order to reduce
-   "unrecognized character follows \\",
+ the number of relocations needed when a shared library is loaded dynamically,
-   "numbers out of order in {} quantifier",
+ it is now one long string. We cannot use a table of offsets, because the
+ lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
+ simply count through to the one we want - this isn't a performance issue
+ because these strings are used only when there is a compilation error. */
+ static const char error_texts[] =
+   "no error\0"
+   "\\ at end of pattern\0"
+   "\\c at end of pattern\0"
+   "unrecognized character follows \\\0"
+   "numbers out of order in {} quantifier\0"
    /* 5 */
-   "number too big in {} quantifier",
+   "number too big in {} quantifier\0"
-   "missing terminating ] for character class",
+   "missing terminating ] for character class\0"
-   "invalid escape sequence in character class",
+   "invalid escape sequence in character class\0"
-   "range out of order in character class",
+   "range out of order in character class\0"
-   "nothing to repeat",
+   "nothing to repeat\0"
    /* 10 */
-   "operand of unlimited repeat could match the empty string",
+   "operand of unlimited repeat could match the empty string\0"  /** DEAD **/
-   "internal error: unexpected repeat",
+   "internal error: unexpected repeat\0"
-   "unrecognized character after (?",
+   "unrecognized character after (? or (?-\0"
-   "POSIX named classes are supported only within a class",
+   "POSIX named classes are supported only within a class\0"
-   "missing )",
+   "missing )\0"
    /* 15 */
-   "reference to non-existent subpattern",
+   "reference to non-existent subpattern\0"
-   "erroffset passed as NULL",
+   "erroffset passed as NULL\0"
-   "unknown option bit(s) set",
+   "unknown option bit(s) set\0"
-   "missing ) after comment",
+   "missing ) after comment\0"
-   "parentheses nested too deeply",
+   "parentheses nested too deeply\0"  /** DEAD **/
    /* 20 */
-   "regular expression too large",
+   "regular expression is too large\0"
-   "failed to get memory",
+   "failed to get memory\0"
-   "unmatched parentheses",
+   "unmatched parentheses\0"
-   "internal error: code overflow",
+   "internal error: code overflow\0"
-   "unrecognized character after (?<",
+   "unrecognized character after (?<\0"
    /* 25 */
-   "lookbehind assertion is not fixed length",
+   "lookbehind assertion is not fixed length\0"
-   "malformed number after (?(",
+   "malformed number or name after (?(\0"
-   "conditional group contains more than two branches",
+   "conditional group contains more than two branches\0"
-   "assertion expected after (?(",
+   "assertion expected after (?(\0"
-   "(?R or (?digits must be followed by )",
+   "(?R or (?[+-]digits must be followed by )\0"
    /* 30 */
-   "unknown POSIX class name",
+   "unknown POSIX class name\0"
-   "POSIX collating elements are not supported",
+   "POSIX collating elements are not supported\0"
-   "this version of PCRE is not compiled with PCRE_UTF8 support",
+   "this version of PCRE is not compiled with PCRE_UTF8 support\0"
-   "spare error",
+   "spare error\0"  /** DEAD **/
-   "character value in \\x{...} sequence is too large",
+   "character value in \\x{...} sequence is too large\0"
    /* 35 */
-   "invalid condition (?(0)",
+   "invalid condition (?(0)\0"
-   "\\C not allowed in lookbehind assertion",
+   "\\C not allowed in lookbehind assertion\0"
-   "PCRE does not support \\L, \\l, \\N, \\U, or \\u",
+   "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
-   "number after (?C is > 255",
+   "number after (?C is > 255\0"
-   "closing ) for (?C expected",
+   "closing ) for (?C expected\0"
    /* 40 */
-   "recursive call could loop indefinitely",
+   "recursive call could loop indefinitely\0"
-   "unrecognized character after (?P",
+   "unrecognized character after (?P\0"
-   "syntax error after (?P",
+   "syntax error in subpattern name (missing terminator)\0"
-   "two named groups have the same name",
+   "two named subpatterns have the same name\0"
-   "invalid UTF-8 string",
+   "invalid UTF-8 string\0"
    /* 45 */
-   "support for \\P, \\p, and \\X has not been compiled",
+   "support for \\P, \\p, and \\X has not been compiled\0"
-   "malformed \\P or \\p sequence",
+   "malformed \\P or \\p sequence\0"
-   "unknown property name after \\P or \\p"
+   "unknown property name after \\P or \\p\0"
- };
+   "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
+   "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
+   /* 50 */
+   "repeated subpattern is too long\0"    /** DEAD **/
+   "octal value is greater than \\377 (not in UTF-8 mode)\0"
+   "internal error: overran compiling workspace\0"
+   "internal error: previously-checked referenced subpattern not found\0"
+   "DEFINE group contains more than one branch\0"
+   /* 55 */
+   "repeating a DEFINE group is not allowed\0"
+   "inconsistent NEWLINE options\0"
+   "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
+   "a numbered reference must not be zero\0"
+   "(*VERB) with an argument is not supported\0"
+   /* 60 */
+   "(*VERB) not recognized\0"
+   "number is too big\0"
+   "subpattern name expected\0"
+   "digit expected after (?+\0"
+   "] is an invalid data character in JavaScript compatibility mode";
  /* Table to identify digits and hex digits. This is used when compiling
-Line 235 
 For convenience, we use the same bit def
+Line 322 
 For convenience, we use the same bit def
  Then we can use ctype_digit and ctype_xdigit in the code. */
- #if !EBCDIC    /* This is the "normal" case, for ASCII systems */
+ #ifndef EBCDIC  /* This is the "normal" case, for ASCII systems */
  static const unsigned char digitab[] =
    {
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7 */
-Line 271 
 static const unsigned char digitab[] =
+Line 358 
 static const unsigned char digitab[] =
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
- #else          /* This is the "abnormal" case, for EBCDIC systems */
+ #else           /* This is the "abnormal" case, for EBCDIC systems */
  static const unsigned char digitab[] =
    {
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7  0 */
-Line 285 
 static const unsigned char digitab[] =
+Line 372 
 static const unsigned char digitab[] =
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 40 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  72- |     */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 50 */
-x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  88- �     */
+x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  88- 95    */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 60 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ?     */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
-Line 319 
 static const unsigned char ebcdic_charta
+Line 406 
 static const unsigned char ebcdic_charta
 x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 */
 x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /*  72- |  */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 */
-x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /*  88- �  */
+x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /*  88- 95 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 */
 x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ?  */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
-Line 346 
 static const unsigned char ebcdic_charta
+Line 433 
 static const unsigned char ebcdic_charta
  /* Definition to allow mutual recursion */
  static BOOL
-   compile_regex(int, int, int *, uschar **, const uschar **, int *, BOOL, int,
+   compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
-     int *, int *, branch_chain *, compile_data *);
+     int *, int *, branch_chain *, compile_data *, int *);
+ /*************************************************
+ *            Find an error text                  *
+ *************************************************/
+ /* The error texts are now all in one long string, to save on relocations. As
+ some of the text is of unknown length, we can't use a table of offsets.
+ Instead, just count through the strings. This is not a performance issue
+ because it happens only when there has been a compilation error.
+ Argument:   the error number
+ Returns:    pointer to the error string
+ */
+ static const char *
+ find_error_text(int n)
+ {
+ const char *s = error_texts;
+ for (; n > 0; n--) while (*s++ != 0);
+ return s;
+ }
  /*************************************************
-Line 357 
 static BOOL
+Line 466 
 static BOOL
  /* 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,
+ encodes one of the more complicated things such as \d. A backreference to group
- a positive value greater than 255 may be returned. On entry, ptr is pointing at
+ n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
- the \. On exit, it is on the final character of the escape sequence.
+ 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
-Line 370 
 Arguments:
+Line 481 
 Arguments:
  Returns:         zero or positive => a data character
                   negative => a special escape sequence
-                  on error, errorptr is set
+                  on error, errorcodeptr is set
  */
  static int
-Line 388 
 ptr--;                            /* Set
+Line 499 
 ptr--;                            /* Set
  if (c == 0) *errorcodeptr = ERR1;
- /* Non-alphamerics are literals. For digits or letters, do an initial lookup in
+ /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
- a table. A non-zero result is something that can be returned immediately.
+ in a table. A non-zero result is something that can be returned immediately.
  Otherwise further processing may be required. */
- #if !EBCDIC    /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII coding */
- else if (c < '0' || c > 'z') {}                           /* Not alphameric */
+ else if (c < '0' || c > 'z') {}                           /* Not alphanumeric */
  else if ((i = escapes[c - '0']) != 0) c = i;
- #else          /* EBCDIC coding */
+ #else           /* EBCDIC coding */
- else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphameric */
+ else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphanumeric */
  else if ((i = escapes[c - 0x48]) != 0)  c = i;
  #endif
-Line 406 
 else if ((i = escapes[c - 0x48]) != 0)
+Line 517 
 else if ((i = escapes[c - 0x48]) != 0)
  else
    {
    const uschar *oldptr;
+   BOOL braced, negated;
    switch (c)
      {
      /* A number of Perl escapes are not handled by PCRE. We give an explicit
-Line 419 
 else
+Line 532 
 else
      *errorcodeptr = ERR37;
      break;
+     /* \g must be followed by one of a number of specific things:
+     (1) A number, either plain or braced. If positive, it is an absolute
+     backreference. If negative, it is a relative backreference. This is a Perl
+.10 feature.
+     (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
+     is part of Perl's movement towards a unified syntax for back references. As
+     this is synonymous with \k{name}, we fudge it up by pretending it really
+     was \k.
+     (3) For Oniguruma compatibility we also support \g followed by a name or a
+     number either in angle brackets or in single quotes. However, these are
+     (possibly recursive) subroutine calls, _not_ backreferences. Just return
+     the -ESC_g code (cf \k). */
+     case 'g':
+     if (ptr[1] == '<' || ptr[1] == '\'')
+       {
+       c = -ESC_g;
+       break;
+       }
+     /* Handle the Perl-compatible cases */
+     if (ptr[1] == '{')
+       {
+       const uschar *p;
+       for (p = ptr+2; *p != 0 && *p != '}'; p++)
+         if (*p != '-' && (digitab[*p] & ctype_digit) == 0) break;
+       if (*p != 0 && *p != '}')
+         {
+         c = -ESC_k;
+         break;
+         }
+       braced = TRUE;
+       ptr++;
+       }
+     else braced = FALSE;
+     if (ptr[1] == '-')
+       {
+       negated = TRUE;
+       ptr++;
+       }
+     else negated = FALSE;
+     c = 0;
+     while ((digitab[ptr[1]] & ctype_digit) != 0)
+       c = c * 10 + *(++ptr) - '0';
+     if (c < 0)   /* Integer overflow */
+       {
+       *errorcodeptr = ERR61;
+       break;
+       }
+     if (braced && *(++ptr) != '}')
+       {
+       *errorcodeptr = ERR57;
+       break;
+       }
+     if (c == 0)
+       {
+       *errorcodeptr = ERR58;
+       break;
+       }
+     if (negated)
+       {
+       if (c > bracount)
+         {
+         *errorcodeptr = ERR15;
+         break;
+         }
+       c = bracount - (c - 1);
+       }
+     c = -(ESC_REF + c);
+     break;
      /* The handling of escape sequences consisting of a string of digits
      starting with one that is not zero is not straightforward. By experiment,
      the way Perl works seems to be as follows:
-Line 440 
 else
+Line 635 
 else
        c -= '0';
        while ((digitab[ptr[1]] & ctype_digit) != 0)
          c = c * 10 + *(++ptr) - '0';
+       if (c < 0)    /* Integer overflow */
+         {
+         *errorcodeptr = ERR61;
+         break;
+         }
        if (c < 10 || c <= bracount)
          {
          c = -(ESC_REF + c);
-Line 460 
 else
+Line 660 
 else
        }
      /* \0 always starts an octal number, but we may drop through to here with a
-     larger first octal digit. */
+     larger first octal digit. The original code used just to take the least
+     significant 8 bits of octal numbers (I think this is what early Perls used
+     to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
+     than 3 octal digits. */
      case '0':
      c -= '0';
      while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7')
          c = c * 8 + *(++ptr) - '0';
-     c &= 255;     /* Take least significant 8 bits */
+     if (!utf8 && c > 255) *errorcodeptr = ERR51;
      break;
      /* \x is complicated. \x{ddd} is a character number which can be greater
-Line 486 
 else
+Line 689 
 else
          if (c == 0 && cc == '0') continue;     /* Leading zeroes */
          count++;
- #if !EBCDIC    /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII coding */
          if (cc >= 'a') cc -= 32;               /* Convert to upper case */
          c = (c << 4) + cc - ((cc < 'A')? '0' : ('A' - 10));
- #else          /* EBCDIC coding */
+ #else           /* EBCDIC coding */
          if (cc >= 'a' && cc <= 'z') cc += 64;  /* Convert to upper case */
          c = (c << 4) + cc - ((cc >= '0')? '0' : ('A' - 10));
  #endif
-Line 513 
 else
+Line 716 
 else
        {
        int cc;                               /* Some compilers don't like ++ */
        cc = *(++ptr);                        /* in initializers */
- #if !EBCDIC    /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII coding */
        if (cc >= 'a') cc -= 32;              /* Convert to upper case */
        c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));
- #else          /* EBCDIC coding */
+ #else           /* EBCDIC coding */
        if (cc <= 'z') cc += 64;              /* Convert to upper case */
        c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));
  #endif
        }
      break;
-     /* Other special escapes not starting with a digit are straightforward */
+     /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
+     This coding is ASCII-specific, but then the whole concept of \cx is
+     ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
      case 'c':
      c = *(++ptr);
      if (c == 0)
        {
        *errorcodeptr = ERR2;
-       return 0;
+       break;
        }
-     /* A letter is upper-cased; then the 0x40 bit is flipped. This coding
+ #ifndef EBCDIC  /* ASCII coding */
-     is ASCII-specific, but then the whole concept of \cx is ASCII-specific.
-     (However, an EBCDIC equivalent has now been added.) */
- #if !EBCDIC    /* ASCII coding */
      if (c >= 'a' && c <= 'z') c -= 32;
      c ^= 0x40;
- #else          /* EBCDIC coding */
+ #else           /* EBCDIC coding */
      if (c >= 'a' && c <= 'z') c += 64;
      c ^= 0xC0;
  #endif
      break;
      /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
-     other alphameric following \ is an error if PCRE_EXTRA was set; otherwise,
+     other alphanumeric following \ is an error if PCRE_EXTRA was set;
-     for Perl compatibility, it is a literal. This code looks a bit odd, but
+     otherwise, for Perl compatibility, it is a literal. This code looks a bit
-     there used to be some cases other than the default, and there may be again
+     odd, but there used to be some cases other than the default, and there may
-     in future, so I haven't "optimized" it. */
+     be again in future, so I haven't "optimized" it. */
      default:
      if ((options & PCRE_EXTRA) != 0) switch(c)
-Line 610 
 if (c == '{')
+Line 811 
 if (c == '{')
      *negptr = TRUE;
      ptr++;
      }
-   for (i = 0; i < sizeof(name) - 1; i++)
+   for (i = 0; i < (int)sizeof(name) - 1; i++)
      {
      c = *(++ptr);
      if (c == 0) goto ERROR_RETURN;
-Line 639 
 top = _pcre_utt_size;
+Line 840 
 top = _pcre_utt_size;
  while (bot < top)
    {
    i = (bot + top) >> 1;
-   c = strcmp(name, _pcre_utt[i].name);
+   c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
    if (c == 0)
      {
      *dptr = _pcre_utt[i].value;
-Line 763 
 return p;
+Line 964 
 return p;
  /*************************************************
+ *       Find forward referenced subpattern       *
+ *************************************************/
+ /* This function scans along a pattern's text looking for capturing
+ subpatterns, and counting them. If it finds a named pattern that matches the
+ name it is given, it returns its number. Alternatively, if the name is NULL, it
+ returns when it reaches a given numbered subpattern. This is used for forward
+ references to subpatterns. We know that if (?P< is encountered, the name will
+ be terminated by '>' because that is checked in the first pass.
+ Arguments:
+   ptr          current position in the pattern
+   cd           compile background data
+   name         name to seek, or NULL if seeking a numbered subpattern
+   lorn         name length, or subpattern number if name is NULL
+   xmode        TRUE if we are in /x mode
+ Returns:       the number of the named subpattern, or -1 if not found
+ */
+ static int
+ find_parens(const uschar *ptr, compile_data *cd, const uschar *name, int lorn,
+   BOOL xmode)
+ {
+ const uschar *thisname;
+ int count = cd->bracount;
+ for (; *ptr != 0; ptr++)
+   {
+   int term;
+   /* Skip over backslashed characters and also entire \Q...\E */
+   if (*ptr == '\\')
+     {
+     if (*(++ptr) == 0) return -1;
+     if (*ptr == 'Q') for (;;)
+       {
+       while (*(++ptr) != 0 && *ptr != '\\');
+       if (*ptr == 0) return -1;
+       if (*(++ptr) == 'E') break;
+       }
+     continue;
+     }
+   /* Skip over character classes; this logic must be similar to the way they
+   are handled for real. If the first character is '^', skip it. Also, if the
+   first few characters (either before or after ^) are \Q\E or \E we skip them
+   too. This makes for compatibility with Perl. */
+   if (*ptr == '[')
+     {
+     BOOL negate_class = FALSE;
+     for (;;)
+       {
+       int c = *(++ptr);
+       if (c == '\\')
+         {
+         if (ptr[1] == 'E') ptr++;
+           else if (strncmp((const char *)ptr+1, "Q\\E", 3) == 0) ptr += 3;
+             else break;
+         }
+       else if (!negate_class && c == '^')
+         negate_class = TRUE;
+       else break;
+       }
+     /* If the next character is ']', it is a data character that must be
+     skipped, except in JavaScript compatibility mode. */
+     if (ptr[1] == ']' && (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
+       ptr++;
+     while (*(++ptr) != ']')
+       {
+       if (*ptr == 0) return -1;
+       if (*ptr == '\\')
+         {
+         if (*(++ptr) == 0) return -1;
+         if (*ptr == 'Q') for (;;)
+           {
+           while (*(++ptr) != 0 && *ptr != '\\');
+           if (*ptr == 0) return -1;
+           if (*(++ptr) == 'E') break;
+           }
+         continue;
+         }
+       }
+     continue;
+     }
+   /* Skip comments in /x mode */
+   if (xmode && *ptr == '#')
+     {
+     while (*(++ptr) != 0 && *ptr != '\n');
+     if (*ptr == 0) return -1;
+     continue;
+     }
+   /* An opening parens must now be a real metacharacter */
+   if (*ptr != '(') continue;
+   if (ptr[1] != '?' && ptr[1] != '*')
+     {
+     count++;
+     if (name == NULL && count == lorn) return count;
+     continue;
+     }
+   ptr += 2;
+   if (*ptr == 'P') ptr++;                      /* Allow optional P */
+   /* We have to disambiguate (?<! and (?<= from (?<name> */
+   if ((*ptr != '<' || ptr[1] == '!' || ptr[1] == '=') &&
+        *ptr != '\'')
+     continue;
+   count++;
+   if (name == NULL && count == lorn) return count;
+   term = *ptr++;
+   if (term == '<') term = '>';
+   thisname = ptr;
+   while (*ptr != term) ptr++;
+   if (name != NULL && lorn == ptr - thisname &&
+       strncmp((const char *)name, (const char *)thisname, lorn) == 0)
+     return count;
+   }
+ return -1;
+ }
+ /*************************************************
  *      Find first significant op code            *
  *************************************************/
-Line 811 
 for (;;)
+Line 1149 
 for (;;)
      case OP_CALLOUT:
      case OP_CREF:
-     case OP_BRANUMBER:
+     case OP_RREF:
+     case OP_DEF:
      code += _pcre_OP_lengths[*code];
      break;
-Line 856 
 for (;;)
+Line 1195 
 for (;;)
    {
    int d;
    register int op = *cc;
-   if (op >= OP_BRA) op = OP_BRA;
    switch (op)
      {
+     case OP_CBRA:
      case OP_BRA:
      case OP_ONCE:
      case OP_COND:
-     d = find_fixedlength(cc, options);
+     d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options);
      if (d < 0) return d;
      branchlength += d;
      do cc += GET(cc, 1); while (*cc == OP_ALT);
-Line 898 
 for (;;)
+Line 1236 
 for (;;)
      /* Skip over things that don't match chars */
      case OP_REVERSE:
-     case OP_BRANUMBER:
      case OP_CREF:
+     case OP_RREF:
+     case OP_DEF:
      case OP_OPT:
      case OP_CALLOUT:
      case OP_SOD:
-Line 917 
 for (;;)
+Line 1256 
 for (;;)
      case OP_CHAR:
      case OP_CHARNC:
+     case OP_NOT:
      branchlength++;
      cc += 2;
  #ifdef SUPPORT_UTF8
-Line 943 
 for (;;)
+Line 1283 
 for (;;)
      case OP_TYPEEXACT:
      branchlength += GET2(cc,1);
+     if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
      cc += 4;
      break;
-Line 960 
 for (;;)
+Line 1301 
 for (;;)
      case OP_NOT_WORDCHAR:
      case OP_WORDCHAR:
      case OP_ANY:
+     case OP_ALLANY:
      branchlength++;
      cc++;
      break;
-Line 1031 
 Returns:      pointer to the opcode for
+Line 1373 
 Returns:      pointer to the opcode for
  static const uschar *
  find_bracket(const uschar *code, BOOL utf8, int number)
  {
- #ifndef SUPPORT_UTF8
- utf8 = utf8;               /* Stop pedantic compilers complaining */
- #endif
  for (;;)
    {
    register int c = *code;
    if (c == OP_END) return NULL;
-   else if (c > OP_BRA)
+   /* XCLASS is used for classes that cannot be represented just by a bit
+   map. This includes negated single high-valued characters. The length in
+   the table is zero; the actual length is stored in the compiled code. */
+   if (c == OP_XCLASS) code += GET(code, 1);
+   /* Handle capturing bracket */
+   else if (c == OP_CBRA)
      {
-     int n = c - OP_BRA;
+     int n = GET2(code, 1+LINK_SIZE);
-     if (n > EXTRACT_BASIC_MAX) n = GET2(code, 2+LINK_SIZE);
      if (n == number) return (uschar *)code;
-     code += _pcre_OP_lengths[OP_BRA];
+     code += _pcre_OP_lengths[c];
      }
+   /* Otherwise, we can get the item's length from the table, except that for
+   repeated character types, we have to test for \p and \P, which have an extra
+   two bytes of parameters. */
    else
      {
-     code += _pcre_OP_lengths[c];
+     switch(c)
+       {
+       case OP_TYPESTAR:
+       case OP_TYPEMINSTAR:
+       case OP_TYPEPLUS:
+       case OP_TYPEMINPLUS:
+       case OP_TYPEQUERY:
+       case OP_TYPEMINQUERY:
+       case OP_TYPEPOSSTAR:
+       case OP_TYPEPOSPLUS:
+       case OP_TYPEPOSQUERY:
+       if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+       break;
- #ifdef SUPPORT_UTF8
+       case OP_TYPEUPTO:
+       case OP_TYPEMINUPTO:
+       case OP_TYPEEXACT:
+       case OP_TYPEPOSUPTO:
+       if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
+       break;
+       }
-     /* In UTF-8 mode, opcodes that are followed by a character may be followed
+     /* Add in the fixed length from the table */
-     by a multi-byte character. The length in the table is a minimum, so we have
-     to scan along to skip the extra bytes. All opcodes are less than 128, so we
+     code += _pcre_OP_lengths[c];
-     can use relatively efficient code. */
+   /* In UTF-8 mode, opcodes that are followed by a character may be followed by
+   a multi-byte character. The length in the table is a minimum, so we have to
+   arrange to skip the extra bytes. */
+ #ifdef SUPPORT_UTF8
      if (utf8) switch(c)
        {
        case OP_CHAR:
-Line 1064 
 for (;;)
+Line 1437 
 for (;;)
        case OP_EXACT:
        case OP_UPTO:
        case OP_MINUPTO:
+       case OP_POSUPTO:
        case OP_STAR:
        case OP_MINSTAR:
+       case OP_POSSTAR:
        case OP_PLUS:
        case OP_MINPLUS:
+       case OP_POSPLUS:
        case OP_QUERY:
        case OP_MINQUERY:
-       while ((*code & 0xc0) == 0x80) code++;
+       case OP_POSQUERY:
-       break;
+       if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
-       /* XCLASS is used for classes that cannot be represented just by a bit
-       map. This includes negated single high-valued characters. The length in
-       the table is zero; the actual length is stored in the compiled code. */
-       case OP_XCLASS:
-       code += GET(code, 1) + 1;
        break;
        }
  #endif
-Line 1105 
 Returns:      pointer to the opcode for
+Line 1474 
 Returns:      pointer to the opcode for
  static const uschar *
  find_recurse(const uschar *code, BOOL utf8)
  {
- #ifndef SUPPORT_UTF8
- utf8 = utf8;               /* Stop pedantic compilers complaining */
- #endif
  for (;;)
    {
    register int c = *code;
    if (c == OP_END) return NULL;
-   else if (c == OP_RECURSE) return code;
+   if (c == OP_RECURSE) return code;
-   else if (c > OP_BRA)
-     {
+   /* XCLASS is used for classes that cannot be represented just by a bit
-     code += _pcre_OP_lengths[OP_BRA];
+   map. This includes negated single high-valued characters. The length in
-     }
+   the table is zero; the actual length is stored in the compiled code. */
+   if (c == OP_XCLASS) code += GET(code, 1);
+   /* Otherwise, we can get the item's length from the table, except that for
+   repeated character types, we have to test for \p and \P, which have an extra
+   two bytes of parameters. */
    else
      {
-     code += _pcre_OP_lengths[c];
+     switch(c)
+       {
+       case OP_TYPESTAR:
+       case OP_TYPEMINSTAR:
+       case OP_TYPEPLUS:
+       case OP_TYPEMINPLUS:
+       case OP_TYPEQUERY:
+       case OP_TYPEMINQUERY:
+       case OP_TYPEPOSSTAR:
+       case OP_TYPEPOSPLUS:
+       case OP_TYPEPOSQUERY:
+       if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+       break;
- #ifdef SUPPORT_UTF8
+       case OP_TYPEPOSUPTO:
+       case OP_TYPEUPTO:
+       case OP_TYPEMINUPTO:
+       case OP_TYPEEXACT:
+       if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
+       break;
+       }
+     /* Add in the fixed length from the table */
+     code += _pcre_OP_lengths[c];
      /* In UTF-8 mode, opcodes that are followed by a character may be followed
      by a multi-byte character. The length in the table is a minimum, so we have
-     to scan along to skip the extra bytes. All opcodes are less than 128, so we
+     to arrange to skip the extra bytes. */
-     can use relatively efficient code. */
+ #ifdef SUPPORT_UTF8
      if (utf8) switch(c)
        {
        case OP_CHAR:
-Line 1136 
 for (;;)
+Line 1530 
 for (;;)
        case OP_EXACT:
        case OP_UPTO:
        case OP_MINUPTO:
+       case OP_POSUPTO:
        case OP_STAR:
        case OP_MINSTAR:
+       case OP_POSSTAR:
        case OP_PLUS:
        case OP_MINPLUS:
+       case OP_POSPLUS:
        case OP_QUERY:
        case OP_MINQUERY:
-       while ((*code & 0xc0) == 0x80) code++;
+       case OP_POSQUERY:
-       break;
+       if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
-       /* XCLASS is used for classes that cannot be represented just by a bit
-       map. This includes negated single high-valued characters. The length in
-       the table is zero; the actual length is stored in the compiled code. */
-       case OP_XCLASS:
-       code += GET(code, 1) + 1;
        break;
        }
  #endif
-Line 1165 
 for (;;)
+Line 1555 
 for (;;)
  *************************************************/
  /* This function scans through a branch of a compiled pattern to see whether it
- can match the empty string or not. It is called only from could_be_empty()
+ can match the empty string or not. It is called from could_be_empty()
- below. Note that first_significant_code() skips over assertions. If we hit an
+ below and from compile_branch() when checking for an unlimited repeat of a
- unclosed bracket, we return "empty" - this means we've struck an inner bracket
+ group that can match nothing. Note that first_significant_code() skips over
- whose current branch will already have been scanned.
+ backward and negative forward assertions when its final argument is TRUE. If we
+ hit an unclosed bracket, we return "empty" - this means we've struck an inner
+ bracket whose current branch will already have been scanned.
  Arguments:
    code        points to start of search
-Line 1182 
 static BOOL
+Line 1574 
 static BOOL
  could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
  {
  register int c;
- for (code = first_significant_code(code + 1 + LINK_SIZE, NULL, 0, TRUE);
+ for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
       code < endcode;
       code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
    {
-Line 1190 
 for (code = first_significant_code(code
+Line 1582 
 for (code = first_significant_code(code
    c = *code;
-   if (c >= OP_BRA)
+   /* Skip over forward assertions; the other assertions are skipped by
+   first_significant_code() with a TRUE final argument. */
+   if (c == OP_ASSERT)
+     {
+     do code += GET(code, 1); while (*code == OP_ALT);
+     c = *code;
+     continue;
+     }
+   /* Groups with zero repeats can of course be empty; skip them. */
+   if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
+     {
+     code += _pcre_OP_lengths[c];
+     do code += GET(code, 1); while (*code == OP_ALT);
+     c = *code;
+     continue;
+     }
+   /* For other groups, scan the branches. */
+   if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
      {
      BOOL empty_branch;
      if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */
-Line 1206 
 for (code = first_significant_code(code
+Line 1620 
 for (code = first_significant_code(code
        }
      while (*code == OP_ALT);
      if (!empty_branch) return FALSE;   /* All branches are non-empty */
-     code += 1 + LINK_SIZE;
      c = *code;
+     continue;
      }
-   else switch (c)
+   /* Handle the other opcodes */
+   switch (c)
      {
-     /* Check for quantifiers after a class */
+     /* Check for quantifiers after a class. XCLASS is used for classes that
+     cannot be represented just by a bit map. This includes negated single
+     high-valued characters. The length in _pcre_OP_lengths[] is zero; the
+     actual length is stored in the compiled code, so we must update "code"
+     here. */
  #ifdef SUPPORT_UTF8
      case OP_XCLASS:
-     ccode = code + GET(code, 1);
+     ccode = code += GET(code, 1);
      goto CHECK_CLASS_REPEAT;
  #endif
-Line 1260 
 for (code = first_significant_code(code
+Line 1680 
 for (code = first_significant_code(code
      case OP_NOT_WORDCHAR:
      case OP_WORDCHAR:
      case OP_ANY:
+     case OP_ALLANY:
      case OP_ANYBYTE:
      case OP_CHAR:
      case OP_CHARNC:
      case OP_NOT:
      case OP_PLUS:
      case OP_MINPLUS:
+     case OP_POSPLUS:
      case OP_EXACT:
      case OP_NOTPLUS:
      case OP_NOTMINPLUS:
+     case OP_NOTPOSPLUS:
      case OP_NOTEXACT:
      case OP_TYPEPLUS:
      case OP_TYPEMINPLUS:
+     case OP_TYPEPOSPLUS:
      case OP_TYPEEXACT:
      return FALSE;
+     /* These are going to continue, as they may be empty, but we have to
+     fudge the length for the \p and \P cases. */
+     case OP_TYPESTAR:
+     case OP_TYPEMINSTAR:
+     case OP_TYPEPOSSTAR:
+     case OP_TYPEQUERY:
+     case OP_TYPEMINQUERY:
+     case OP_TYPEPOSQUERY:
+     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+     break;
+     /* Same for these */
+     case OP_TYPEUPTO:
+     case OP_TYPEMINUPTO:
+     case OP_TYPEPOSUPTO:
+     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
+     break;
      /* End of branch */
      case OP_KET:
-Line 1283 
 for (code = first_significant_code(code
+Line 1727 
 for (code = first_significant_code(code
      case OP_ALT:
      return TRUE;
-     /* In UTF-8 mode, STAR, MINSTAR, QUERY, MINQUERY, UPTO, and MINUPTO  may be
+     /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
-     followed by a multibyte character */
+     MINUPTO, and POSUPTO may be followed by a multibyte character */
  #ifdef SUPPORT_UTF8
      case OP_STAR:
      case OP_MINSTAR:
+     case OP_POSSTAR:
      case OP_QUERY:
      case OP_MINQUERY:
+     case OP_POSQUERY:
      case OP_UPTO:
      case OP_MINUPTO:
+     case OP_POSUPTO:
      if (utf8) while ((code[2] & 0xc0) == 0x80) code++;
      break;
  #endif
-Line 1341 
 return TRUE;
+Line 1788 
 return TRUE;
  *************************************************/
  /* This function is called when the sequence "[:" or "[." or "[=" is
- encountered in a character class. It checks whether this is followed by an
+ encountered in a character class. It checks whether this is followed by a
- optional ^ and then a sequence of letters, terminated by a matching ":]" or
+ sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
- ".]" or "=]".
+ reach an unescaped ']' without the special preceding character, return FALSE.
+ Originally, this function only recognized a sequence of letters between the
+ terminators, but it seems that Perl recognizes any sequence of characters,
+ though of course unknown POSIX names are subsequently rejected. Perl gives an
+ "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
+ didn't consider this to be a POSIX class. Likewise for [:1234:].
+ The problem in trying to be exactly like Perl is in the handling of escapes. We
+ have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
+ class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
+ below handles the special case of \], but does not try to do any other escape
+ processing. This makes it different from Perl for cases such as [:l\ower:]
+ where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
+ "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
+ I think.
- Argument:
+ Arguments:
    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)
+ check_posix_syntax(const uschar *ptr, const uschar **endptr)
  {
  int terminator;          /* Don't combine these lines; the Solaris cc */
  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
- if (*(++ptr) == '^') ptr++;
+ for (++ptr; *ptr != 0; ptr++)
- while ((cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
- if (*ptr == terminator && ptr[1] == ']')
    {
-   *endptr = ptr;
+   if (*ptr == '\\' && ptr[1] == ']') ptr++; else
-   return TRUE;
+     {
+     if (*ptr == ']') return FALSE;
+     if (*ptr == terminator && ptr[1] == ']')
+       {
+       *endptr = ptr;
+       return TRUE;
+       }
+     }
    }
  return FALSE;
  }
-Line 1388 
 Returns:     a value representing the na
+Line 1854 
 Returns:     a value representing the na
  static int
  check_posix_name(const uschar *ptr, int len)
  {
+ const char *pn = posix_names;
  register int yield = 0;
  while (posix_name_lengths[yield] != 0)
    {
    if (len == posix_name_lengths[yield] &&
-     strncmp((const char *)ptr, posix_names[yield], len) == 0) return yield;
+     strncmp((const char *)ptr, pn, len) == 0) return yield;
+   pn += posix_name_lengths[yield] + 1;
    yield++;
    }
  return -1;
-Line 1407 
 return -1;
+Line 1875 
 return -1;
  that is referenced. This means that groups can be replicated for fixed
  repetition simply by copying (because the recursion is allowed to refer to
  earlier groups that are outside the current group). However, when a group is
- optional (i.e. the minimum quantifier is zero), OP_BRAZERO is inserted before
+ optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
- it, after it has been compiled. This means that any OP_RECURSE items within it
+ inserted before it, after it has been compiled. This means that any OP_RECURSE
- that refer to the group itself or any contained groups have to have their
+ items within it that refer to the group itself or any contained groups have to
- offsets adjusted. That is the job of this function. Before it is called, the
+ have their offsets adjusted. That one of the jobs of this function. Before it
- partially compiled regex must be temporarily terminated with OP_END.
+ is called, the partially compiled regex must be temporarily terminated with
+ OP_END.
+ This function has been extended with the possibility of forward references for
+ recursions and subroutine calls. It must also check the list of such references
+ for the group we are dealing with. If it finds that one of the recursions in
+ the current group is on this list, it adjusts the offset in the list, not the
+ value in the reference (which is a group number).
  Arguments:
    group      points to the start of the group
    adjust     the amount by which the group is to be moved
    utf8       TRUE in UTF-8 mode
    cd         contains pointers to tables etc.
+   save_hwm   the hwm forward reference pointer at the start of the group
  Returns:     nothing
  */
  static void
- adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd)
+ adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
+   uschar *save_hwm)
  {
  uschar *ptr = group;
  while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
    {
-   int offset = GET(ptr, 1);
+   int offset;
-   if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
+   uschar *hc;
+   /* See if this recursion is on the forward reference list. If so, adjust the
+   reference. */
+   for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
+     {
+     offset = GET(hc, 0);
+     if (cd->start_code + offset == ptr + 1)
+       {
+       PUT(hc, 0, offset + adjust);
+       break;
+       }
+     }
+   /* Otherwise, adjust the recursion offset if it's after the start of this
+   group. */
+   if (hc >= cd->hwm)
+     {
+     offset = GET(ptr, 1);
+     if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
+     }
    ptr += 1 + LINK_SIZE;
    }
  }
-Line 1508 
 Yield:        TRUE when range returned;
+Line 2009 
 Yield:        TRUE when range returned;
  */
  static BOOL
- get_othercase_range(int *cptr, int d, int *ocptr, int *odptr)
+ get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
+   unsigned int *odptr)
  {
- int c, othercase, next;
+ unsigned int c, othercase, next;
  for (c = *cptr; c <= d; c++)
-   { if ((othercase = _pcre_ucp_othercase(c)) >= 0) break; }
+   { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
  if (c > d) return FALSE;
-Line 1522 
 next = othercase + 1;
+Line 2024 
 next = othercase + 1;
  for (++c; c <= d; c++)
    {
-   if (_pcre_ucp_othercase(c) != next) break;
+   if (UCD_OTHERCASE(c) != next) break;
    next++;
    }
-Line 1534 
 return TRUE;
+Line 2036 
 return TRUE;
  #endif  /* SUPPORT_UCP */
  /*************************************************
- *           Compile one branch                   *
+ *     Check if auto-possessifying is possible    *
  *************************************************/
- /* Scan the pattern, compiling it into the code vector. If the options are
+ /* This function is called for unlimited repeats of certain items, to see
- changed during the branch, the pointer is used to change the external options
+ whether the next thing could possibly match the repeated item. If not, it makes
- bits.
+ sense to automatically possessify the repeated item.
  Arguments:
-   optionsptr     pointer to the option bits
+   op_code       the repeated op code
-   brackets       points to number of extracting brackets used
+   this          data for this item, depends on the opcode
-   codeptr        points to the pointer to the current code point
+   utf8          TRUE in UTF-8 mode
-   ptrptr         points to the current pattern pointer
+   utf8_char     used for utf8 character bytes, NULL if not relevant
-   errorcodeptr   points to error code variable
+   ptr           next character in pattern
-   firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
+   options       options bits
-   reqbyteptr     set to the last literal character required, else < 0
+   cd            contains pointers to tables etc.
-   bcptr          points to current branch chain
-   cd             contains pointers to tables etc.
- Returns:         TRUE on success
+ Returns:        TRUE if possessifying is wanted
-                  FALSE, with *errorcodeptr set non-zero on error
  */
  static BOOL
- compile_branch(int *optionsptr, int *brackets, uschar **codeptr,
+ check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
-   const uschar **ptrptr, int *errorcodeptr, int *firstbyteptr,
+   const uschar *ptr, int options, compile_data *cd)
-   int *reqbyteptr, branch_chain *bcptr, compile_data *cd)
  {
- int repeat_type, op_type;
+ int next;
- int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */
- int bravalue = 0;
+ /* Skip whitespace and comments in extended mode */
- int greedy_default, greedy_non_default;
- int firstbyte, reqbyte;
+ if ((options & PCRE_EXTENDED) != 0)
- int zeroreqbyte, zerofirstbyte;
+   {
- int req_caseopt, reqvary, tempreqvary;
+   for (;;)
- int condcount = 0;
+     {
- int options = *optionsptr;
+     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
- int after_manual_callout = 0;
+     if (*ptr == '#')
- register int c;
+       {
- register uschar *code = *codeptr;
+       while (*(++ptr) != 0)
- uschar *tempcode;
+         if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
- BOOL inescq = FALSE;
+       }
- BOOL groupsetfirstbyte = FALSE;
+     else break;
- const uschar *ptr = *ptrptr;
+     }
+   }
+ /* If the next item is one that we can handle, get its value. A non-negative
+ value is a character, a negative value is an escape value. */
+ if (*ptr == '\\')
+   {
+   int temperrorcode = 0;
+   next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
+   if (temperrorcode != 0) return FALSE;
+   ptr++;    /* Point after the escape sequence */
+   }
+ else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
+   {
+ #ifdef SUPPORT_UTF8
+   if (utf8) { GETCHARINC(next, ptr); } else
+ #endif
+   next = *ptr++;
+   }
+ else return FALSE;
+ /* Skip whitespace and comments in extended mode */
+ if ((options & PCRE_EXTENDED) != 0)
+   {
+   for (;;)
+     {
+     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
+     if (*ptr == '#')
+       {
+       while (*(++ptr) != 0)
+         if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
+       }
+     else break;
+     }
+   }
+ /* If the next thing is itself optional, we have to give up. */
+ if (*ptr == '*' || *ptr == '?' || strncmp((char *)ptr, "{0,", 3) == 0)
+   return FALSE;
+ /* Now compare the next item with the previous opcode. If the previous is a
+ positive single character match, "item" either contains the character or, if
+ "item" is greater than 127 in utf8 mode, the character's bytes are in
+ utf8_char. */
+ /* Handle cases when the next item is a character. */
+ if (next >= 0) switch(op_code)
+   {
+   case OP_CHAR:
+ #ifdef SUPPORT_UTF8
+   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+ #endif
+   return item != next;
+   /* For CHARNC (caseless character) we must check the other case. If we have
+   Unicode property support, we can use it to test the other case of
+   high-valued characters. */
+   case OP_CHARNC:
+ #ifdef SUPPORT_UTF8
+   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+ #endif
+   if (item == next) return FALSE;
+ #ifdef SUPPORT_UTF8
+   if (utf8)
+     {
+     unsigned int othercase;
+     if (next < 128) othercase = cd->fcc[next]; else
+ #ifdef SUPPORT_UCP
+     othercase = UCD_OTHERCASE((unsigned int)next);
+ #else
+     othercase = NOTACHAR;
+ #endif
+     return (unsigned int)item != othercase;
+     }
+   else
+ #endif  /* SUPPORT_UTF8 */
+   return (item != cd->fcc[next]);  /* Non-UTF-8 mode */
+   /* For OP_NOT, "item" must be a single-byte character. */
+   case OP_NOT:
+   if (item == next) return TRUE;
+   if ((options & PCRE_CASELESS) == 0) return FALSE;
+ #ifdef SUPPORT_UTF8
+   if (utf8)
+     {
+     unsigned int othercase;
+     if (next < 128) othercase = cd->fcc[next]; else
+ #ifdef SUPPORT_UCP
+     othercase = UCD_OTHERCASE(next);
+ #else
+     othercase = NOTACHAR;
+ #endif
+     return (unsigned int)item == othercase;
+     }
+   else
+ #endif  /* SUPPORT_UTF8 */
+   return (item == cd->fcc[next]);  /* Non-UTF-8 mode */
+   case OP_DIGIT:
+   return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
+   case OP_NOT_DIGIT:
+   return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
+   case OP_WHITESPACE:
+   return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
+   case OP_NOT_WHITESPACE:
+   return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
+   case OP_WORDCHAR:
+   return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
+   case OP_NOT_WORDCHAR:
+   return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
+   case OP_HSPACE:
+   case OP_NOT_HSPACE:
+   switch(next)
+     {
+     case 0x09:
+     case 0x20:
+     case 0xa0:
+     case 0x1680:
+     case 0x180e:
+     case 0x2000:
+     case 0x2001:
+     case 0x2002:
+     case 0x2003:
+     case 0x2004:
+     case 0x2005:
+     case 0x2006:
+     case 0x2007:
+     case 0x2008:
+     case 0x2009:
+     case 0x200A:
+     case 0x202f:
+     case 0x205f:
+     case 0x3000:
+     return op_code != OP_HSPACE;
+     default:
+     return op_code == OP_HSPACE;
+     }
+   case OP_VSPACE:
+   case OP_NOT_VSPACE:
+   switch(next)
+     {
+     case 0x0a:
+     case 0x0b:
+     case 0x0c:
+     case 0x0d:
+     case 0x85:
+     case 0x2028:
+     case 0x2029:
+     return op_code != OP_VSPACE;
+     default:
+     return op_code == OP_VSPACE;
+     }
+   default:
+   return FALSE;
+   }
+ /* Handle the case when the next item is \d, \s, etc. */
+ switch(op_code)
+   {
+   case OP_CHAR:
+   case OP_CHARNC:
+ #ifdef SUPPORT_UTF8
+   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+ #endif
+   switch(-next)
+     {
+     case ESC_d:
+     return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
+     case ESC_D:
+     return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
+     case ESC_s:
+     return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
+     case ESC_S:
+     return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
+     case ESC_w:
+     return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
+     case ESC_W:
+     return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
+     case ESC_h:
+     case ESC_H:
+     switch(item)
+       {
+       case 0x09:
+       case 0x20:
+       case 0xa0:
+       case 0x1680:
+       case 0x180e:
+       case 0x2000:
+       case 0x2001:
+       case 0x2002:
+       case 0x2003:
+       case 0x2004:
+       case 0x2005:
+       case 0x2006:
+       case 0x2007:
+       case 0x2008:
+       case 0x2009:
+       case 0x200A:
+       case 0x202f:
+       case 0x205f:
+       case 0x3000:
+       return -next != ESC_h;
+       default:
+       return -next == ESC_h;
+       }
+     case ESC_v:
+     case ESC_V:
+     switch(item)
+       {
+       case 0x0a:
+       case 0x0b:
+       case 0x0c:
+       case 0x0d:
+       case 0x85:
+       case 0x2028:
+       case 0x2029:
+       return -next != ESC_v;
+       default:
+       return -next == ESC_v;
+       }
+     default:
+     return FALSE;
+     }
+   case OP_DIGIT:
+   return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
+          next == -ESC_h || next == -ESC_v;
+   case OP_NOT_DIGIT:
+   return next == -ESC_d;
+   case OP_WHITESPACE:
+   return next == -ESC_S || next == -ESC_d || next == -ESC_w;
+   case OP_NOT_WHITESPACE:
+   return next == -ESC_s || next == -ESC_h || next == -ESC_v;
+   case OP_HSPACE:
+   return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
+   case OP_NOT_HSPACE:
+   return next == -ESC_h;
+   /* Can't have \S in here because VT matches \S (Perl anomaly) */
+   case OP_VSPACE:
+   return next == -ESC_V || next == -ESC_d || next == -ESC_w;
+   case OP_NOT_VSPACE:
+   return next == -ESC_v;
+   case OP_WORDCHAR:
+   return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
+   case OP_NOT_WORDCHAR:
+   return next == -ESC_w || next == -ESC_d;
+   default:
+   return FALSE;
+   }
+ /* Control does not reach here */
+ }
+ /*************************************************
+ *           Compile one branch                   *
+ *************************************************/
+ /* Scan the pattern, compiling it into the a vector. If the options are
+ changed during the branch, the pointer is used to change the external options
+ bits. This function is used during the pre-compile phase when we are trying
+ to find out the amount of memory needed, as well as during the real compile
+ phase. The value of lengthptr distinguishes the two phases.
+ Arguments:
+   optionsptr     pointer to the option bits
+   codeptr        points to the pointer to the current code point
+   ptrptr         points to the current pattern pointer
+   errorcodeptr   points to error code variable
+   firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
+   reqbyteptr     set to the last literal character required, else < 0
+   bcptr          points to current branch chain
+   cd             contains pointers to tables etc.
+   lengthptr      NULL during the real compile phase
+                  points to length accumulator during pre-compile phase
+ Returns:         TRUE on success
+                  FALSE, with *errorcodeptr set non-zero on error
+ */
+ static BOOL
+ compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
+   int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
+   compile_data *cd, int *lengthptr)
+ {
+ int repeat_type, op_type;
+ int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */
+ int bravalue = 0;
+ int greedy_default, greedy_non_default;
+ int firstbyte, reqbyte;
+ int zeroreqbyte, zerofirstbyte;
+ int req_caseopt, reqvary, tempreqvary;
+ int options = *optionsptr;
+ int after_manual_callout = 0;
+ int length_prevgroup = 0;
+ register int c;
+ register uschar *code = *codeptr;
+ uschar *last_code = code;
+ uschar *orig_code = code;
+ uschar *tempcode;
+ BOOL inescq = FALSE;
+ BOOL groupsetfirstbyte = FALSE;
+ const uschar *ptr = *ptrptr;
  const uschar *tempptr;
  uschar *previous = NULL;
  uschar *previous_callout = NULL;
+ uschar *save_hwm = NULL;
  uschar classbits[32];
  #ifdef SUPPORT_UTF8
  BOOL class_utf8;
  BOOL utf8 = (options & PCRE_UTF8) != 0;
  uschar *class_utf8data;
+ uschar *class_utf8data_base;
  uschar utf8_char[6];
  #else
  BOOL utf8 = FALSE;
+ uschar *utf8_char = NULL;
+ #endif
+ #ifdef DEBUG
+ if (lengthptr != NULL) DPRINTF((">> start branch\n"));
  #endif
  /* Set up the default and non-default settings for greediness */
-Line 1621 
 req_caseopt = ((options & PCRE_CASELESS)
+Line 2467 
 req_caseopt = ((options & PCRE_CASELESS)
  for (;; ptr++)
    {
    BOOL negate_class;
+   BOOL should_flip_negation;
    BOOL possessive_quantifier;
    BOOL is_quantifier;
+   BOOL is_recurse;
+   BOOL reset_bracount;
    int class_charcount;
    int class_lastchar;
    int newoptions;
    int recno;
+   int refsign;
    int skipbytes;
    int subreqbyte;
    int subfirstbyte;
+   int terminator;
    int mclength;
    uschar mcbuffer[8];
-   /* Next byte in the pattern */
+   /* Get next byte in the pattern */
    c = *ptr;
+   /* If we are in the pre-compile phase, accumulate the length used for the
+   previous cycle of this loop. */
+   if (lengthptr != NULL)
+     {
+ #ifdef DEBUG
+     if (code > cd->hwm) cd->hwm = code;                 /* High water info */
+ #endif
+     if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
+       {
+       *errorcodeptr = ERR52;
+       goto FAILED;
+       }
+     /* There is at least one situation where code goes backwards: this is the
+     case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
+     the class is simply eliminated. However, it is created first, so we have to
+     allow memory for it. Therefore, don't ever reduce the length at this point.
+     */
+     if (code < last_code) code = last_code;
+     /* Paranoid check for integer overflow */
+     if (OFLOW_MAX - *lengthptr < code - last_code)
+       {
+       *errorcodeptr = ERR20;
+       goto FAILED;
+       }
+     *lengthptr += code - last_code;
+     DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
+     /* If "previous" is set and it is not at the start of the work space, move
+     it back to there, in order to avoid filling up the work space. Otherwise,
+     if "previous" is NULL, reset the current code pointer to the start. */
+     if (previous != NULL)
+       {
+       if (previous > orig_code)
+         {
+         memmove(orig_code, previous, code - previous);
+         code -= previous - orig_code;
+         previous = orig_code;
+         }
+       }
+     else code = orig_code;
+     /* Remember where this code item starts so we can pick up the length
+     next time round. */
+     last_code = code;
+     }
+   /* In the real compile phase, just check the workspace used by the forward
+   reference list. */
+   else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
+     {
+     *errorcodeptr = ERR52;
+     goto FAILED;
+     }
    /* If in \Q...\E, check for the end; if not, we have a literal */
    if (inescq && c != 0)
-Line 1651 
 for (;; ptr++)
+Line 2565 
 for (;; ptr++)
        {
        if (previous_callout != NULL)
          {
-         complete_callout(previous_callout, ptr, cd);
+         if (lengthptr == NULL)  /* Don't attempt in pre-compile phase */
+           complete_callout(previous_callout, ptr, cd);
          previous_callout = NULL;
          }
        if ((options & PCRE_AUTO_CALLOUT) != 0)
-Line 1672 
 for (;; ptr++)
+Line 2587 
 for (;; ptr++)
    if (!is_quantifier && previous_callout != NULL &&
         after_manual_callout-- <= 0)
      {
-     complete_callout(previous_callout, ptr, cd);
+     if (lengthptr == NULL)      /* Don't attempt in pre-compile phase */
+       complete_callout(previous_callout, ptr, cd);
      previous_callout = NULL;
      }
-Line 1683 
 for (;; ptr++)
+Line 2599 
 for (;; ptr++)
      if ((cd->ctypes[c] & ctype_space) != 0) continue;
      if (c == '#')
        {
-       /* The space before the ; is to avoid a warning on a silly compiler
+       while (*(++ptr) != 0)
-       on the Macintosh. */
+         {
-       while ((c = *(++ptr)) != 0 && c != NEWLINE) ;
+         if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
-       if (c != 0) continue;   /* Else fall through to handle end of string */
+         }
+       if (*ptr != 0) continue;
+       /* Else fall through to handle end of string */
+       c = 0;
        }
      }
-Line 1700 
 for (;; ptr++)
+Line 2620 
 for (;; ptr++)
    switch(c)
      {
-     /* The branch terminates at end of string, |, or ). */
+     /* ===================================================================*/
+     case 0:                        /* The branch terminates at string end */
-     case 0:
+     case '|':                      /* or | or ) */
-     case '|':
      case ')':
      *firstbyteptr = firstbyte;
      *reqbyteptr = reqbyte;
      *codeptr = code;
      *ptrptr = ptr;
+     if (lengthptr != NULL)
+       {
+       if (OFLOW_MAX - *lengthptr < code - last_code)
+         {
+         *errorcodeptr = ERR20;
+         goto FAILED;
+         }
+       *lengthptr += code - last_code;   /* To include callout length */
+       DPRINTF((">> end branch\n"));
+       }
      return TRUE;
+     /* ===================================================================*/
      /* Handle single-character metacharacters. In multiline mode, ^ disables
      the setting of any following char as a first character. */
-Line 1736 
 for (;; ptr++)
+Line 2667 
 for (;; ptr++)
      zerofirstbyte = firstbyte;
      zeroreqbyte = reqbyte;
      previous = code;
-     *code++ = OP_ANY;
+     *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
      break;
+     /* ===================================================================*/
      /* Character classes. If the included characters are all < 256, we build a
 -byte bitmap of the permitted characters, except in the special case
      where there is only one such character. For negated classes, we build the
-Line 1749 
 for (;; ptr++)
+Line 2682 
 for (;; ptr++)
      opcode is compiled. It may optionally have a bit map for characters < 256,
      but those above are are explicitly listed afterwards. A flag byte tells
      whether the bitmap is present, and whether this is a negated class or not.
-     */
+     In JavaScript compatibility mode, an isolated ']' causes an error. In
+     default (Perl) mode, it is treated as a data character. */
+     case ']':
+     if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
+       {
+       *errorcodeptr = ERR64;
+       goto FAILED;
+       }
+     goto NORMAL_CHAR;
      case '[':
      previous = code;
-Line 1758 
 for (;; ptr++)
+Line 2701 
 for (;; ptr++)
      they are encountered at the top level, so we'll do that too. */
      if ((ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
-         check_posix_syntax(ptr, &tempptr, cd))
+         check_posix_syntax(ptr, &tempptr))
        {
        *errorcodeptr = (ptr[1] == ':')? ERR13 : ERR31;
        goto FAILED;
        }
-     /* If the first character is '^', set the negation flag and skip it. */
+     /* If the first character is '^', set the negation flag and skip it. Also,
+     if the first few characters (either before or after ^) are \Q\E or \E we
+     skip them too. This makes for compatibility with Perl. */
-     if ((c = *(++ptr)) == '^')
+     negate_class = FALSE;
+     for (;;)
        {
-       negate_class = TRUE;
        c = *(++ptr);
+       if (c == '\\')
+         {
+         if (ptr[1] == 'E') ptr++;
+           else if (strncmp((const char *)ptr+1, "Q\\E", 3) == 0) ptr += 3;
+             else break;
+         }
+       else if (!negate_class && c == '^')
+         negate_class = TRUE;
+       else break;
        }
-     else
+     /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
+     an initial ']' is taken as a data character -- the code below handles
+     that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
+     [^] must match any character, so generate OP_ALLANY. */
+     if (c ==']' && (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
        {
-       negate_class = FALSE;
+       *code++ = negate_class? OP_ALLANY : OP_FAIL;
+       if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
+       zerofirstbyte = firstbyte;
+       break;
        }
+     /* If a class contains a negative special such as \S, we need to flip the
+     negation flag at the end, so that support for characters > 255 works
+     correctly (they are all included in the class). */
+     should_flip_negation = FALSE;
      /* Keep a count of chars with values < 256 so that we can optimize the case
-     of just a single character (as long as it's < 256). For higher valued UTF-8
+     of just a single character (as long as it's < 256). However, For higher
-     characters, we don't yet do any optimization. */
+     valued UTF-8 characters, we don't yet do any optimization. */
      class_charcount = 0;
      class_lastchar = -1;
+     /* Initialize the 32-char bit map to all zeros. We build the map in a
+     temporary bit of memory, in case the class contains only 1 character (less
+     than 256), because in that case the compiled code doesn't use the bit map.
+     */
+     memset(classbits, 0, 32 * sizeof(uschar));
  #ifdef SUPPORT_UTF8
      class_utf8 = FALSE;                       /* No chars >= 256 */
-     class_utf8data = code + LINK_SIZE + 34;   /* For UTF-8 items */
+     class_utf8data = code + LINK_SIZE + 2;    /* For UTF-8 items */
+     class_utf8data_base = class_utf8data;     /* For resetting in pass 1 */
  #endif
-     /* Initialize the 32-char bit map to all zeros. We have to build the
-     map in a temporary bit of store, in case the class contains only 1
-     character (< 256), because in that case the compiled code doesn't use the
-     bit map. */
-     memset(classbits, 0, 32 * sizeof(uschar));
      /* Process characters until ] is reached. By writing this as a "do" it
-     means that an initial ] is taken as a data character. The first pass
+     means that an initial ] is taken as a data character. At the start of the
-     through the regex checked the overall syntax, so we don't need to be very
+     loop, c contains the first byte of the character. */
-     strict here. At the start of the loop, c contains the first byte of the
-     character. */
-     do
+     if (c != 0) do
        {
+       const uschar *oldptr;
  #ifdef SUPPORT_UTF8
        if (utf8 && c > 127)
          {                           /* Braces are required because the */
          GETCHARLEN(c, ptr, ptr);    /* macro generates multiple statements */
          }
+       /* In the pre-compile phase, accumulate the length of any UTF-8 extra
+       data and reset the pointer. This is so that very large classes that
+       contain a zillion UTF-8 characters no longer overwrite the work space
+       (which is on the stack). */
+       if (lengthptr != NULL)
+         {
+         *lengthptr += class_utf8data - class_utf8data_base;
+         class_utf8data = class_utf8data_base;
+         }
  #endif
        /* Inside \Q...\E everything is literal except \E */
        if (inescq)
          {
-         if (c == '\\' && ptr[1] == 'E')
+         if (c == '\\' && ptr[1] == 'E')     /* If we are at \E */
            {
-           inescq = FALSE;
+           inescq = FALSE;                   /* Reset literal state */
-           ptr++;
+           ptr++;                            /* Skip the 'E' */
-           continue;
+           continue;                         /* Carry on with next */
            }
-         else goto LONE_SINGLE_CHARACTER;
+         goto CHECK_RANGE;                   /* Could be range if \E follows */
          }
        /* Handle POSIX class names. Perl allows a negation extension of the
-Line 1831 
 for (;; ptr++)
+Line 2813 
 for (;; ptr++)
        if (c == '[' &&
            (ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
-           check_posix_syntax(ptr, &tempptr, cd))
+           check_posix_syntax(ptr, &tempptr))
          {
          BOOL local_negate = FALSE;
          int posix_class, taboffset, tabopt;
-Line 1848 
 for (;; ptr++)
+Line 2830 
 for (;; ptr++)
          if (*ptr == '^')
            {
            local_negate = TRUE;
+           should_flip_negation = TRUE;  /* Note negative special */
            ptr++;
            }
-Line 1911 
 for (;; ptr++)
+Line 2894 
 for (;; ptr++)
          }
        /* Backslash may introduce a single character, or it may introduce one
-       of the specials, which just set a flag. Escaped items are checked for
+       of the specials, which just set a flag. The sequence \b is a special
-       validity in the pre-compiling pass. The sequence \b is a special case.
+       case. Inside a class (and only there) it is treated as backspace.
-       Inside a class (and only there) it is treated as backspace. Elsewhere
+       Elsewhere it marks a word boundary. Other escapes have preset maps ready
-       it marks a word boundary. Other escapes have preset maps ready to
+       to 'or' into the one we are building. We assume they have more than one
-       or into the one we are building. We assume they have more than one
        character in them, so set class_charcount bigger than one. */
        if (c == '\\')
          {
-         c = check_escape(&ptr, errorcodeptr, *brackets, options, TRUE);
+         c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
+         if (*errorcodeptr != 0) goto FAILED;
-         if (-c == ESC_b) c = '\b';       /* \b is backslash in a class */
+         if (-c == ESC_b) c = '\b';       /* \b is backspace in a class */
          else if (-c == ESC_X) c = 'X';   /* \X is literal X in a class */
+         else if (-c == ESC_R) c = 'R';   /* \R is literal R in a class */
          else if (-c == ESC_Q)            /* Handle start of quoted string */
            {
            if (ptr[1] == '\\' && ptr[2] == 'E')
-Line 1933 
 for (;; ptr++)
+Line 2917 
 for (;; ptr++)
            else inescq = TRUE;
            continue;
            }
+         else if (-c == ESC_E) continue;  /* Ignore orphan \E */
          if (c < 0)
            {
            register const uschar *cbits = cd->cbits;
            class_charcount += 2;     /* Greater than 1 is what matters */
-           switch (-c)
+           /* Save time by not doing this in the pre-compile phase. */
+           if (lengthptr == NULL) switch (-c)
              {
              case ESC_d:
              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
              continue;
              case ESC_D:
+             should_flip_negation = TRUE;
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
              continue;
-Line 1953 
 for (;; ptr++)
+Line 2942 
 for (;; ptr++)
              continue;
              case ESC_W:
+             should_flip_negation = TRUE;
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
              continue;
-Line 1962 
 for (;; ptr++)
+Line 2952 
 for (;; ptr++)
              continue;
              case ESC_S:
+             should_flip_negation = TRUE;
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */
              continue;
- #ifdef SUPPORT_UCP
+             default:    /* Not recognized; fall through */
-             case ESC_p:
+             break;      /* Need "default" setting to stop compiler warning. */
-             case ESC_P:
+             }
+           /* In the pre-compile phase, just do the recognition. */
+           else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
+                    c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
+           /* We need to deal with \H, \h, \V, and \v in both phases because
+           they use extra memory. */
+           if (-c == ESC_h)
+             {
+             SETBIT(classbits, 0x09); /* VT */
+             SETBIT(classbits, 0x20); /* SPACE */
+             SETBIT(classbits, 0xa0); /* NSBP */
+ #ifdef SUPPORT_UTF8
+             if (utf8)
                {
-               BOOL negated;
-               int pdata;
-               int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
-               if (ptype < 0) goto FAILED;
                class_utf8 = TRUE;
-               *class_utf8data++ = ((-c == ESC_p) != negated)?
+               *class_utf8data++ = XCL_SINGLE;
-                 XCL_PROP : XCL_NOTPROP;
+               class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
-               *class_utf8data++ = ptype;
+               *class_utf8data++ = XCL_SINGLE;
-               *class_utf8data++ = pdata;
+               class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
-               class_charcount -= 2;   /* Not a < 256 character */
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
+               *class_utf8data++ = XCL_SINGLE;
+               class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
+               *class_utf8data++ = XCL_SINGLE;
+               class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
+               *class_utf8data++ = XCL_SINGLE;
+               class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
                }
-             continue;
  #endif
+             continue;
+             }
-             /* Unrecognized escapes are faulted if PCRE is running in its
+           if (-c == ESC_H)
-             strict mode. By default, for compatibility with Perl, they are
+             {
-             treated as literals. */
+             for (c = 0; c < 32; c++)
+               {
+               int x = 0xff;
+               switch (c)
+                 {
+                 case 0x09/8: x ^= 1 << (0x09%8); break;
+                 case 0x20/8: x ^= 1 << (0x20%8); break;
+                 case 0xa0/8: x ^= 1 << (0xa0%8); break;
+                 default: break;
+                 }
+               classbits[c] |= x;
+               }
-             default:
+ #ifdef SUPPORT_UTF8
-             if ((options & PCRE_EXTRA) != 0)
+             if (utf8)
                {
-               *errorcodeptr = ERR7;
+               class_utf8 = TRUE;
-               goto FAILED;
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
                }
-             c = *ptr;              /* The final character */
+ #endif
-             class_charcount -= 2;  /* Undo the default count from above */
+             continue;
              }
-           }
-         /* Fall through if we have a single character (c >= 0). This may be
-         > 256 in UTF-8 mode. */
-         }   /* End of backslash handling */
-       /* A single character may be followed by '-' to form a range. However,
+           if (-c == ESC_v)
+             {
+             SETBIT(classbits, 0x0a); /* LF */
+             SETBIT(classbits, 0x0b); /* VT */
+             SETBIT(classbits, 0x0c); /* FF */
+             SETBIT(classbits, 0x0d); /* CR */
+             SETBIT(classbits, 0x85); /* NEL */
+ #ifdef SUPPORT_UTF8
+             if (utf8)
+               {
+               class_utf8 = TRUE;
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
+               }
+ #endif
+             continue;
+             }
+           if (-c == ESC_V)
+             {
+             for (c = 0; c < 32; c++)
+               {
+               int x = 0xff;
+               switch (c)
+                 {
+                 case 0x0a/8: x ^= 1 << (0x0a%8);
+                              x ^= 1 << (0x0b%8);
+                              x ^= 1 << (0x0c%8);
+                              x ^= 1 << (0x0d%8);
+                              break;
+                 case 0x85/8: x ^= 1 << (0x85%8); break;
+                 default: break;
+                 }
+               classbits[c] |= x;
+               }
+ #ifdef SUPPORT_UTF8
+             if (utf8)
+               {
+               class_utf8 = TRUE;
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
+               }
+ #endif
+             continue;
+             }
+           /* We need to deal with \P and \p in both phases. */
+ #ifdef SUPPORT_UCP
+           if (-c == ESC_p || -c == ESC_P)
+             {
+             BOOL negated;
+             int pdata;
+             int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
+             if (ptype < 0) goto FAILED;
+             class_utf8 = TRUE;
+             *class_utf8data++ = ((-c == ESC_p) != negated)?
+               XCL_PROP : XCL_NOTPROP;
+             *class_utf8data++ = ptype;
+             *class_utf8data++ = pdata;
+             class_charcount -= 2;   /* Not a < 256 character */
+             continue;
+             }
+ #endif
+           /* Unrecognized escapes are faulted if PCRE is running in its
+           strict mode. By default, for compatibility with Perl, they are
+           treated as literals. */
+           if ((options & PCRE_EXTRA) != 0)
+             {
+             *errorcodeptr = ERR7;
+             goto FAILED;
+             }
+           class_charcount -= 2;  /* Undo the default count from above */
+           c = *ptr;              /* Get the final character and fall through */
+           }
+         /* Fall through if we have a single character (c >= 0). This may be
+         greater than 256 in UTF-8 mode. */
+         }   /* End of backslash handling */
+       /* A single character may be followed by '-' to form a range. However,
        Perl does not permit ']' to be the end of the range. A '-' character
-       here is treated as a literal. */
+       at the end is treated as a literal. Perl ignores orphaned \E sequences
+       entirely. The code for handling \Q and \E is messy. */
+       CHECK_RANGE:
+       while (ptr[1] == '\\' && ptr[2] == 'E')
+         {
+         inescq = FALSE;
+         ptr += 2;
+         }
+       oldptr = ptr;
+       /* Remember \r or \n */
-       if (ptr[1] == '-' && ptr[2] != ']')
+       if (c == '\r' || c == '\n') cd->external_flags |= PCRE_HASCRORLF;
+       /* Check for range */
+       if (!inescq && ptr[1] == '-')
          {
          int d;
          ptr += 2;
+         while (*ptr == '\\' && ptr[1] == 'E') ptr += 2;
+         /* If we hit \Q (not followed by \E) at this point, go into escaped
+         mode. */
+         while (*ptr == '\\' && ptr[1] == 'Q')
+           {
+           ptr += 2;
+           if (*ptr == '\\' && ptr[1] == 'E') { ptr += 2; continue; }
+           inescq = TRUE;
+           break;
+           }
+         if (*ptr == 0 || (!inescq && *ptr == ']'))
+           {
+           ptr = oldptr;
+           goto LONE_SINGLE_CHARACTER;
+           }
  #ifdef SUPPORT_UTF8
          if (utf8)
-Line 2026 
 for (;; ptr++)
+Line 3186 
 for (;; ptr++)
          not any of the other escapes. Perl 5.6 treats a hyphen as a literal
          in such circumstances. */
-         if (d == '\\')
+         if (!inescq && d == '\\')
            {
-           const uschar *oldptr = ptr;
+           d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
-           d = check_escape(&ptr, errorcodeptr, *brackets, options, TRUE);
+           if (*errorcodeptr != 0) goto FAILED;
-           /* \b is backslash; \X is literal X; any other special means the '-'
+           /* \b is backspace; \X is literal X; \R is literal R; any other
-           was literal */
+           special means the '-' was literal */
            if (d < 0)
              {
              if (d == -ESC_b) d = '\b';
-             else if (d == -ESC_X) d = 'X'; else
+             else if (d == -ESC_X) d = 'X';
+             else if (d == -ESC_R) d = 'R'; else
                {
-               ptr = oldptr - 2;
+               ptr = oldptr;
                goto LONE_SINGLE_CHARACTER;  /* A few lines below */
                }
              }
            }
-         /* The check that the two values are in the correct order happens in
+         /* Check that the two values are in the correct order. Optimize
-         the pre-pass. Optimize one-character ranges */
+         one-character ranges */
+         if (d < c)
+           {
+           *errorcodeptr = ERR8;
+           goto FAILED;
+           }
          if (d == c) goto LONE_SINGLE_CHARACTER;  /* A few lines below */
+         /* Remember \r or \n */
+         if (d == '\r' || d == '\n') cd->external_flags |= PCRE_HASCRORLF;
          /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
          matching, we have to use an XCLASS with extra data items. Caseless
          matching for characters > 127 is available only if UCP support is
-Line 2067 
 for (;; ptr++)
+Line 3238 
 for (;; ptr++)
  #ifdef SUPPORT_UCP
            if ((options & PCRE_CASELESS) != 0)
              {
-             int occ, ocd;
+             unsigned int occ, ocd;
-             int cc = c;
+             unsigned int cc = c;
-             int origd = d;
+             unsigned int origd = d;
              while (get_othercase_range(&cc, origd, &occ, &ocd))
                {
-               if (occ >= c && ocd <= d) continue;  /* Skip embedded ranges */
+               if (occ >= (unsigned int)c &&
+                   ocd <= (unsigned int)d)
+                 continue;                          /* Skip embedded ranges */
-               if (occ < c  && ocd >= c - 1)        /* Extend the basic range */
+               if (occ < (unsigned int)c  &&
+                   ocd >= (unsigned int)c - 1)      /* Extend the basic range */
                  {                                  /* if there is overlap,   */
                  c = occ;                           /* noting that if occ < c */
                  continue;                          /* we can't have ocd > d  */
                  }                                  /* because a subrange is  */
-               if (ocd > d && occ <= d + 1)         /* always shorter than    */
+               if (ocd > (unsigned int)d &&
+                   occ <= (unsigned int)d + 1)      /* always shorter than    */
                  {                                  /* the basic range.       */
                  d = ocd;
                  continue;
-Line 2127 
 for (;; ptr++)
+Line 3302 
 for (;; ptr++)
          ranges that lie entirely within 0-127 when there is UCP support; else
          for partial ranges without UCP support. */
-         for (; c <= d; c++)
+         class_charcount += d - c + 1;
+         class_lastchar = d;
+         /* We can save a bit of time by skipping this in the pre-compile. */
+         if (lengthptr == NULL) for (; c <= d; c++)
            {
            classbits[c/8] |= (1 << (c&7));
            if ((options & PCRE_CASELESS) != 0)
-Line 2135 
 for (;; ptr++)
+Line 3315 
 for (;; ptr++)
              int uc = cd->fcc[c];           /* flip case */
              classbits[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 */
-Line 2160 
 for (;; ptr++)
+Line 3338 
 for (;; ptr++)
  #ifdef SUPPORT_UCP
          if ((options & PCRE_CASELESS) != 0)
            {
-           int othercase;
+           unsigned int othercase;
-           if ((othercase = _pcre_ucp_othercase(c)) >= 0)
+           if ((othercase = UCD_OTHERCASE(c)) != c)
              {
              *class_utf8data++ = XCL_SINGLE;
              class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
-Line 2186 
 for (;; ptr++)
+Line 3364 
 for (;; ptr++)
          }
        }
-     /* Loop until ']' reached; the check for end of string happens inside the
+     /* Loop until ']' reached. This "while" is the end of the "do" above. */
-     loop. This "while" is the end of the "do" above. */
+     while ((c = *(++ptr)) != 0 && (c != ']' || inescq));
+     if (c == 0)                          /* Missing terminating ']' */
+       {
+       *errorcodeptr = ERR6;
+       goto FAILED;
+       }
+ /* This code has been disabled because it would mean that \s counts as
+ an explicit \r or \n reference, and that's not really what is wanted. Now
+ we set the flag only if there is a literal "\r" or "\n" in the class. */
+ #if 0
+     /* Remember whether \r or \n are in this class */
+     if (negate_class)
+       {
+       if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
+       }
+     else
+       {
+       if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
+       }
+ #endif
-     while ((c = *(++ptr)) != ']' || inescq);
      /* If class_charcount is 1, we saw precisely one character whose value is
-     less than 256. In non-UTF-8 mode we can always optimize. In UTF-8 mode, we
+     less than 256. As long as there were no characters >= 128 and there was no
-     can optimize the negative case only if there were no characters >= 128
+     use of \p or \P, in other words, no use of any XCLASS features, we can
-     because OP_NOT and the related opcodes like OP_NOTSTAR operate on
+     optimize.
-     single-bytes only. This is an historical hangover. Maybe one day we can
-     tidy these opcodes to handle multi-byte characters.
+     In UTF-8 mode, we can optimize the negative case only if there were no
+     characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
+     operate on single-bytes only. This is an historical hangover. Maybe one day
+     we can tidy these opcodes to handle multi-byte characters.
      The optimization throws away the bit map. We turn the item into a
 -character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
-Line 2206 
 for (;; ptr++)
+Line 3411 
 for (;; ptr++)
      reqbyte, save the previous value for reinstating. */
  #ifdef SUPPORT_UTF8
-     if (class_charcount == 1 &&
+     if (class_charcount == 1 && !class_utf8 &&
-           (!utf8 ||
+       (!utf8 || !negate_class || class_lastchar < 128))
-           (!class_utf8 && (!negate_class || class_lastchar < 128))))
  #else
      if (class_charcount == 1)
  #endif
-Line 2252 
 for (;; ptr++)
+Line 3455 
 for (;; ptr++)
      zeroreqbyte = reqbyte;
      /* If there are characters with values > 255, we have to compile an
-     extended class, with its own opcode. If there are no characters < 256,
+     extended class, with its own opcode, unless there was a negated special
-     we can omit the bitmap. */
+     such as \S in the class, because in that case all characters > 255 are in
+     the class, so any that were explicitly given as well can be ignored. If
+     (when there are explicit characters > 255 that must be listed) there are no
+     characters < 256, we can omit the bitmap in the actual compiled code. */
  #ifdef SUPPORT_UTF8
-     if (class_utf8)
+     if (class_utf8 && !should_flip_negation)
        {
        *class_utf8data++ = XCL_END;    /* Marks the end of extra data */
        *code++ = OP_XCLASS;
        code += LINK_SIZE;
        *code = negate_class? XCL_NOT : 0;
-       /* If the map is required, install it, and move on to the end of
+       /* If the map is required, move up the extra data to make room for it;
-       the extra data */
+       otherwise just move the code pointer to the end of the extra data. */
        if (class_charcount > 0)
          {
          *code++ |= XCL_MAP;
+         memmove(code + 32, code, class_utf8data - code);
          memcpy(code, classbits, 32);
-         code = class_utf8data;
+         code = class_utf8data + 32;
-         }
-       /* If the map is not required, slide down the extra data. */
-       else
-         {
-         int len = class_utf8data - (code + 33);
-         memmove(code + 1, code + 33, len);
-         code += len + 1;
          }
+       else code = class_utf8data;
        /* Now fill in the complete length of the item */
-Line 2289 
 for (;; ptr++)
+Line 3488 
 for (;; ptr++)
        }
  #endif
-     /* If there are no characters > 255, negate the 32-byte map if necessary,
+     /* If there are no characters > 255, set the opcode to OP_CLASS or
-     and copy it into the code vector. If this is the first thing in the branch,
+     OP_NCLASS, depending on whether the whole class was negated and whether
-     there can be no first char setting, whatever the repeat count. Any reqbyte
+     there were negative specials such as \S in the class. Then copy the 32-byte
-     setting must remain unchanged after any kind of repeat. */
+     map into the code vector, negating it if necessary. */
+     *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
      if (negate_class)
        {
-       *code++ = OP_NCLASS;
+       if (lengthptr == NULL)    /* Save time in the pre-compile phase */
-       for (c = 0; c < 32; c++) code[c] = ~classbits[c];
+         for (c = 0; c < 32; c++) code[c] = ~classbits[c];
        }
      else
        {
-       *code++ = OP_CLASS;
        memcpy(code, classbits, 32);
        }
      code += 32;
      break;
+     /* ===================================================================*/
      /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
      has been tested above. */
-Line 2374 
 for (;; ptr++)
+Line 3575 
 for (;; ptr++)
        }
      else repeat_type = greedy_default;
-     /* If previous was a recursion, we need to wrap it inside brackets so that
-     it can be replicated if necessary. */
-     if (*previous == OP_RECURSE)
-       {
-       memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
-       code += 1 + LINK_SIZE;
-       *previous = OP_BRA;
-       PUT(previous, 1, code - previous);
-       *code = OP_KET;
-       PUT(code, 1, code - previous);
-       code += 1 + LINK_SIZE;
-       }
      /* If previous was a character match, abolish the item and generate a
      repeat item instead. If a char item has a minumum of more than one, ensure
      that it is set in reqbyte - it might not be if a sequence such as x{3} is
-Line 2421 
 for (;; ptr++)
+Line 3608 
 for (;; ptr++)
          if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
          }
+       /* If the repetition is unlimited, it pays to see if the next thing on
+       the line is something that cannot possibly match this character. If so,
+       automatically possessifying this item gains some performance in the case
+       where the match fails. */
+       if (!possessive_quantifier &&
+           repeat_max < 0 &&
+           check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
+             options, cd))
+         {
+         repeat_type = 0;    /* Force greedy */
+         possessive_quantifier = TRUE;
+         }
        goto OUTPUT_SINGLE_REPEAT;   /* Code shared with single character types */
        }
      /* If previous was a single negated character ([^a] or similar), we use
      one of the special opcodes, replacing it. The code is shared with single-
      character repeats by setting opt_type to add a suitable offset into
-     repeat_type. OP_NOT is currently used only for single-byte chars. */
+     repeat_type. We can also test for auto-possessification. OP_NOT is
+     currently used only for single-byte chars. */
      else if (*previous == OP_NOT)
        {
        op_type = OP_NOTSTAR - OP_STAR;  /* Use "not" opcodes */
        c = previous[1];
+       if (!possessive_quantifier &&
+           repeat_max < 0 &&
+           check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
+         {
+         repeat_type = 0;    /* Force greedy */
+         possessive_quantifier = TRUE;
+         }
        goto OUTPUT_SINGLE_REPEAT;
        }
-Line 2450 
 for (;; ptr++)
+Line 3659 
 for (;; ptr++)
        op_type = OP_TYPESTAR - OP_STAR;  /* Use type opcodes */
        c = *previous;
+       if (!possessive_quantifier &&
+           repeat_max < 0 &&
+           check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
+         {
+         repeat_type = 0;    /* Force greedy */
+         possessive_quantifier = TRUE;
+         }
        OUTPUT_SINGLE_REPEAT:
        if (*previous == OP_PROP || *previous == OP_NOTPROP)
          {
-Line 2469 
 for (;; ptr++)
+Line 3686 
 for (;; ptr++)
        /* All real repeats make it impossible to handle partial matching (maybe
        one day we will be able to remove this restriction). */
-       if (repeat_max != 1) cd->nopartial = TRUE;
+       if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
        /* Combine the op_type with the repeat_type */
-Line 2490 
 for (;; ptr++)
+Line 3707 
 for (;; ptr++)
          }
        /* A repeat minimum of 1 is optimized into some special cases. If the
-       maximum is unlimited, we use OP_PLUS. Otherwise, the original item it
+       maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
        left in place and, if the maximum is greater than 1, we use OP_UPTO with
        one less than the maximum. */
-Line 2543 
 for (;; ptr++)
+Line 3760 
 for (;; ptr++)
            }
          /* Else insert an UPTO if the max is greater than the min, again
-         preceded by the character, for the previously inserted code. */
+         preceded by the character, for the previously inserted code. If the
+         UPTO is just for 1 instance, we can use QUERY instead. */
          else if (repeat_max != repeat_min)
            {
-Line 2562 
 for (;; ptr++)
+Line 3780 
 for (;; ptr++)
              *code++ = prop_value;
              }
            repeat_max -= repeat_min;
-           *code++ = OP_UPTO + repeat_type;
-           PUT2INC(code, 0, repeat_max);
+           if (repeat_max == 1)
+             {
+             *code++ = OP_QUERY + repeat_type;
+             }
+           else
+             {
+             *code++ = OP_UPTO + repeat_type;
+             PUT2INC(code, 0, repeat_max);
+             }
            }
          }
-Line 2610 
 for (;; ptr++)
+Line 3836 
 for (;; ptr++)
        /* All real repeats make it impossible to handle partial matching (maybe
        one day we will be able to remove this restriction). */
-       if (repeat_max != 1) cd->nopartial = TRUE;
+       if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
        if (repeat_min == 0 && repeat_max == -1)
          *code++ = OP_CRSTAR + repeat_type;
-Line 2630 
 for (;; ptr++)
+Line 3856 
 for (;; ptr++)
      /* If previous was a bracket group, we may have to replicate it in certain
      cases. */
-     else if (*previous >= OP_BRA || *previous == OP_ONCE ||
+     else if (*previous == OP_BRA  || *previous == OP_CBRA ||
-              *previous == OP_COND)
+              *previous == OP_ONCE || *previous == OP_COND)
        {
        register int i;
        int ketoffset = 0;
        int len = code - previous;
        uschar *bralink = NULL;
+       /* Repeating a DEFINE group is pointless */
+       if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
+         {
+         *errorcodeptr = ERR55;
+         goto FAILED;
+         }
        /* 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
-Line 2660 
 for (;; ptr++)
+Line 3894 
 for (;; ptr++)
        if (repeat_min == 0)
          {
-         /* If the maximum is also zero, we just omit the group from the output
+         /* If the maximum is also zero, we used to just omit the group from the
-         altogether. */
+         output altogether, like this:
-         if (repeat_max == 0)
+         ** if (repeat_max == 0)
-           {
+         **   {
-           code = previous;
+         **   code = previous;
-           goto END_REPEAT;
+         **   goto END_REPEAT;
-           }
+         **   }
+         However, that fails when a group is referenced as a subroutine from
+         elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
+         so that it is skipped on execution. As we don't have a list of which
+         groups are referenced, we cannot do this selectively.
+         If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
+         and do no more at this point. However, we do need to adjust any
+         OP_RECURSE calls inside the group that refer to the group itself or any
+         internal or forward referenced group, because the offset is from the
+         start of the whole regex. Temporarily terminate the pattern while doing
+         this. */
-         /* If the maximum is 1 or unlimited, we just have to stick in the
+         if (repeat_max <= 1)    /* Covers 0, 1, and unlimited */
-         BRAZERO and do no more at this point. However, we do need to adjust
-         any OP_RECURSE calls inside the group that refer to the group itself or
-         any internal group, because the offset is from the start of the whole
-         regex. Temporarily terminate the pattern while doing this. */
-         if (repeat_max <= 1)
            {
            *code = OP_END;
-           adjust_recurse(previous, 1, utf8, cd);
+           adjust_recurse(previous, 1, utf8, cd, save_hwm);
            memmove(previous+1, previous, len);
            code++;
+           if (repeat_max == 0)
+             {
+             *previous++ = OP_SKIPZERO;
+             goto END_REPEAT;
+             }
            *previous++ = OP_BRAZERO + repeat_type;
            }
-Line 2696 
 for (;; ptr++)
+Line 3941 
 for (;; ptr++)
            {
            int offset;
            *code = OP_END;
-           adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd);
+           adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
            memmove(previous + 2 + LINK_SIZE, previous, len);
            code += 2 + LINK_SIZE;
            *previous++ = OP_BRAZERO + repeat_type;
-Line 2716 
 for (;; ptr++)
+Line 3961 
 for (;; ptr++)
        /* 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. If we set a first char from the group, and didn't
-       set a required char, copy the latter from the former. */
+       set a required char, copy the latter from the former. If there are any
+       forward reference subroutine calls in the group, there will be entries on
+       the workspace list; replicate these with an appropriate increment. */
        else
          {
          if (repeat_min > 1)
            {
-           if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
+           /* In the pre-compile phase, we don't actually do the replication. We
-           for (i = 1; i < repeat_min; i++)
+           just adjust the length as if we had. Do some paranoid checks for
+           potential integer overflow. */
+           if (lengthptr != NULL)
+             {
+             int delta = (repeat_min - 1)*length_prevgroup;
+             if ((double)(repeat_min - 1)*(double)length_prevgroup >
+                                                             (double)INT_MAX ||
+                 OFLOW_MAX - *lengthptr < delta)
+               {
+               *errorcodeptr = ERR20;
+               goto FAILED;
+               }
+             *lengthptr += delta;
+             }
+           /* This is compiling for real */
+           else
              {
-             memcpy(code, previous, len);
+             if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
-             code += len;
+             for (i = 1; i < repeat_min; i++)
+               {
+               uschar *hc;
+               uschar *this_hwm = cd->hwm;
+               memcpy(code, previous, len);
+               for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
+                 {
+                 PUT(cd->hwm, 0, GET(hc, 0) + len);
+                 cd->hwm += LINK_SIZE;
+                 }
+               save_hwm = this_hwm;
+               code += len;
+               }
              }
            }
          if (repeat_max > 0) repeat_max -= repeat_min;
          }
-Line 2736 
 for (;; ptr++)
+Line 4014 
 for (;; ptr++)
        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. */
+       the number of additional copies needed. Again, we must remember to
+       replicate entries on the forward reference list. */
        if (repeat_max >= 0)
          {
-         for (i = repeat_max - 1; i >= 0; i--)
+         /* In the pre-compile phase, we don't actually do the replication. We
+         just adjust the length as if we had. For each repetition we must add 1
+         to the length for BRAZERO and for all but the last repetition we must
+         add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
+         paranoid checks to avoid integer overflow. */
+         if (lengthptr != NULL && repeat_max > 0)
+           {
+           int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
+- 2*LINK_SIZE;   /* Last one doesn't nest */
+           if ((double)repeat_max *
+                 (double)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
+                   > (double)INT_MAX ||
+               OFLOW_MAX - *lengthptr < delta)
+             {
+             *errorcodeptr = ERR20;
+             goto FAILED;
+             }
+           *lengthptr += delta;
+           }
+         /* This is compiling for real */
+         else for (i = repeat_max - 1; i >= 0; i--)
            {
+           uschar *hc;
+           uschar *this_hwm = cd->hwm;
            *code++ = OP_BRAZERO + repeat_type;
            /* All but the final copy start a new nesting, maintaining the
-Line 2757 
 for (;; ptr++)
+Line 4062 
 for (;; ptr++)
              }
            memcpy(code, previous, len);
+           for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
+             {
+             PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
+             cd->hwm += LINK_SIZE;
+             }
+           save_hwm = this_hwm;
            code += len;
            }
-Line 2779 
 for (;; ptr++)
+Line 4090 
 for (;; ptr++)
        /* 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. */
+       correct offset was computed above.
-       else code[-ketoffset] = OP_KETRMAX + repeat_type;
+       Then, when we are doing the actual compile phase, check to see whether
+       this group is a non-atomic one that could match an empty string. If so,
+       convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
+       that runtime checking can be done. [This check is also applied to
+       atomic groups at runtime, but in a different way.] */
+       else
+         {
+         uschar *ketcode = code - ketoffset;
+         uschar *bracode = ketcode - GET(ketcode, 1);
+         *ketcode = OP_KETRMAX + repeat_type;
+         if (lengthptr == NULL && *bracode != OP_ONCE)
+           {
+           uschar *scode = bracode;
+           do
+             {
+             if (could_be_empty_branch(scode, ketcode, utf8))
+               {
+               *bracode += OP_SBRA - OP_BRA;
+               break;
+               }
+             scode += GET(scode, 1);
+             }
+           while (*scode == OP_ALT);
+           }
+         }
        }
+     /* If previous is OP_FAIL, it was generated by an empty class [] in
+     JavaScript mode. The other ways in which OP_FAIL can be generated, that is
+     by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
+     error above. We can just ignore the repeat in JS case. */
+     else if (*previous == OP_FAIL) goto END_REPEAT;
      /* Else there's some kind of shambles */
      else
-Line 2792 
 for (;; ptr++)
+Line 4135 
 for (;; ptr++)
        goto FAILED;
        }
-     /* If the character following a repeat is '+', we wrap the entire repeated
+     /* If the character following a repeat is '+', or if certain optimization
-     item inside OP_ONCE brackets. This is just syntactic sugar, taken from
+     tests above succeeded, possessive_quantifier is TRUE. For some of the
-     Sun's Java package. The repeated item starts at tempcode, not at previous,
+     simpler opcodes, there is an special alternative opcode for this. For
-     which might be the first part of a string whose (former) last char we
+     anything else, we wrap the entire repeated item inside OP_ONCE brackets.
-     repeated. However, we don't support '+' after a greediness '?'. */
+     The '+' notation is just syntactic sugar, taken from Sun's Java package,
+     but the special opcodes can optimize it a bit. The repeated item starts at
+     tempcode, not at previous, which might be the first part of a string whose
+     (former) last char we repeated.
+     Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
+     an 'upto' may follow. We skip over an 'exact' item, and then test the
+     length of what remains before proceeding. */
      if (possessive_quantifier)
        {
-       int len = code - tempcode;
+       int len;
-       memmove(tempcode + 1+LINK_SIZE, tempcode, len);
+       if (*tempcode == OP_EXACT || *tempcode == OP_TYPEEXACT ||
-       code += 1 + LINK_SIZE;
+           *tempcode == OP_NOTEXACT)
-       len += 1 + LINK_SIZE;
+         tempcode += _pcre_OP_lengths[*tempcode] +
-       tempcode[0] = OP_ONCE;
+           ((*tempcode == OP_TYPEEXACT &&
-       *code++ = OP_KET;
+              (tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP))? 2:0);
-       PUTINC(code, 0, len);
+       len = code - tempcode;
-       PUT(tempcode, 1, len);
+       if (len > 0) switch (*tempcode)
+         {
+         case OP_STAR:  *tempcode = OP_POSSTAR; break;
+         case OP_PLUS:  *tempcode = OP_POSPLUS; break;
+         case OP_QUERY: *tempcode = OP_POSQUERY; break;
+         case OP_UPTO:  *tempcode = OP_POSUPTO; break;
+         case OP_TYPESTAR:  *tempcode = OP_TYPEPOSSTAR; break;
+         case OP_TYPEPLUS:  *tempcode = OP_TYPEPOSPLUS; break;
+         case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
+         case OP_TYPEUPTO:  *tempcode = OP_TYPEPOSUPTO; break;
+         case OP_NOTSTAR:  *tempcode = OP_NOTPOSSTAR; break;
+         case OP_NOTPLUS:  *tempcode = OP_NOTPOSPLUS; break;
+         case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
+         case OP_NOTUPTO:  *tempcode = OP_NOTPOSUPTO; break;
+         default:
+         memmove(tempcode + 1+LINK_SIZE, tempcode, len);
+         code += 1 + LINK_SIZE;
+         len += 1 + LINK_SIZE;
+         tempcode[0] = OP_ONCE;
+         *code++ = OP_KET;
+         PUTINC(code, 0, len);
+         PUT(tempcode, 1, len);
+         break;
+         }
        }
      /* In all case we no longer have a previous item. We also set the
-Line 2820 
 for (;; ptr++)
+Line 4196 
 for (;; ptr++)
      break;
-     /* Start of nested bracket sub-expression, or comment or lookahead or
+     /* ===================================================================*/
-     lookbehind or option setting or condition. First deal with special things
+     /* Start of nested parenthesized sub-expression, or comment or lookahead or
-     that can come after a bracket; all are introduced by ?, and the appearance
+     lookbehind or option setting or condition or all the other extended
-     of any of them means that this is not a referencing group. They were
+     parenthesis forms.  */
-     checked for validity in the first pass over the string, so we don't have to
-     check for syntax errors here.  */
      case '(':
      newoptions = options;
      skipbytes = 0;
+     bravalue = OP_CBRA;
+     save_hwm = cd->hwm;
+     reset_bracount = FALSE;
+     /* First deal with various "verbs" that can be introduced by '*'. */
+     if (*(++ptr) == '*' && (cd->ctypes[ptr[1]] & ctype_letter) != 0)
+       {
+       int i, namelen;
+       const char *vn = verbnames;
+       const uschar *name = ++ptr;
+       previous = NULL;
+       while ((cd->ctypes[*++ptr] & ctype_letter) != 0);
+       if (*ptr == ':')
+         {
+         *errorcodeptr = ERR59;   /* Not supported */
+         goto FAILED;
+         }
+       if (*ptr != ')')
+         {
+         *errorcodeptr = ERR60;
+         goto FAILED;
+         }
+       namelen = ptr - name;
+       for (i = 0; i < verbcount; i++)
+         {
+         if (namelen == verbs[i].len &&
+             strncmp((char *)name, vn, namelen) == 0)
+           {
+           *code = verbs[i].op;
+           if (*code++ == OP_ACCEPT) cd->had_accept = TRUE;
+           break;
+           }
+         vn += verbs[i].len + 1;
+         }
+       if (i < verbcount) continue;
+       *errorcodeptr = ERR60;
+       goto FAILED;
+       }
+     /* Deal with the extended parentheses; all are introduced by '?', and the
+     appearance of any of them means that this is not a capturing group. */
-     if (*(++ptr) == '?')
+     else if (*ptr == '?')
        {
-       int set, unset;
+       int i, set, unset, namelen;
        int *optset;
+       const uschar *name;
+       uschar *slot;
        switch (*(++ptr))
          {
          case '#':                 /* Comment; skip to ket */
          ptr++;
-         while (*ptr != ')') ptr++;
+         while (*ptr != 0 && *ptr != ')') ptr++;
+         if (*ptr == 0)
+           {
+           *errorcodeptr = ERR18;
+           goto FAILED;
+           }
          continue;
-         case ':':                 /* Non-extracting bracket */
+         /* ------------------------------------------------------------ */
+         case '|':                 /* Reset capture count for each branch */
+         reset_bracount = TRUE;
+         /* Fall through */
+         /* ------------------------------------------------------------ */
+         case ':':                 /* Non-capturing bracket */
          bravalue = OP_BRA;
          ptr++;
          break;
+         /* ------------------------------------------------------------ */
          case '(':
          bravalue = OP_COND;       /* Conditional group */
-         /* Condition to test for recursion */
+         /* A condition can be an assertion, a number (referring to a numbered
+         group), a name (referring to a named group), or 'R', referring to
+         recursion. R<digits> and R&name are also permitted for recursion tests.
+         There are several syntaxes for testing a named group: (?(name)) is used
+         by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
+         There are two unfortunate ambiguities, caused by history. (a) 'R' can
+         be the recursive thing or the name 'R' (and similarly for 'R' followed
+         by digits), and (b) a number could be a name that consists of digits.
+         In both cases, we look for a name first; if not found, we try the other
+         cases. */
+         /* For conditions that are assertions, check the syntax, and then exit
+         the switch. This will take control down to where bracketed groups,
+         including assertions, are processed. */
+         if (ptr[1] == '?' && (ptr[2] == '=' || ptr[2] == '!' || ptr[2] == '<'))
+           break;
+         /* Most other conditions use OP_CREF (a couple change to OP_RREF
+         below), and all need to skip 3 bytes at the start of the group. */
-         if (ptr[1] == 'R')
+         code[1+LINK_SIZE] = OP_CREF;
+         skipbytes = 3;
+         refsign = -1;
+         /* Check for a test for recursion in a named group. */
+         if (ptr[1] == 'R' && ptr[2] == '&')
            {
-           code[1+LINK_SIZE] = OP_CREF;
+           terminator = -1;
-           PUT2(code, 2+LINK_SIZE, CREF_RECURSE);
+           ptr += 2;
-           skipbytes = 3;
+           code[1+LINK_SIZE] = OP_RREF;    /* Change the type of test */
-           ptr += 3;
            }
-         /* Condition to test for a numbered subpattern match. We know that
+         /* Check for a test for a named group's having been set, using the Perl
-         if a digit follows ( then there will just be digits until ) because
+         syntax (?(<name>) or (?('name') */
-         the syntax was checked in the first pass. */
-         else if ((digitab[ptr[1]] && ctype_digit) != 0)
+         else if (ptr[1] == '<')
            {
-           int condref;                 /* Don't amalgamate; some compilers */
+           terminator = '>';
-           condref = *(++ptr) - '0';    /* grumble at autoincrement in declaration */
-           while (*(++ptr) != ')') condref = condref*10 + *ptr - '0';
-           if (condref == 0)
-             {
-             *errorcodeptr = ERR35;
-             goto FAILED;
-             }
            ptr++;
-           code[1+LINK_SIZE] = OP_CREF;
-           PUT2(code, 2+LINK_SIZE, condref);
-           skipbytes = 3;
            }
-         /* For conditions that are assertions, we just fall through, having
+         else if (ptr[1] == '\'')
-         set bravalue above. */
+           {
-         break;
+           terminator = '\'';
+           ptr++;
-         case '=':                 /* Positive lookahead */
+           }
-         bravalue = OP_ASSERT;
+         else
-         ptr++;
+           {
-         break;
+           terminator = 0;
+           if (ptr[1] == '-' || ptr[1] == '+') refsign = *(++ptr);
+           }
-         case '!':                 /* Negative lookahead */
+         /* We now expect to read a name; any thing else is an error */
-         bravalue = OP_ASSERT_NOT;
-         ptr++;
-         break;
-         case '<':                 /* Lookbehinds */
+         if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
-         switch (*(++ptr))
            {
-           case '=':               /* Positive lookbehind */
+           ptr += 1;  /* To get the right offset */
-           bravalue = OP_ASSERTBACK;
+           *errorcodeptr = ERR28;
-           ptr++;
+           goto FAILED;
-           break;
+           }
-           case '!':               /* Negative lookbehind */
+         /* Read the name, but also get it as a number if it's all digits */
-           bravalue = OP_ASSERTBACK_NOT;
+         recno = 0;
+         name = ++ptr;
+         while ((cd->ctypes[*ptr] & ctype_word) != 0)
+           {
+           if (recno >= 0)
+             recno = ((digitab[*ptr] & ctype_digit) != 0)?
+               recno * 10 + *ptr - '0' : -1;
            ptr++;
-           break;
            }
-         break;
+         namelen = ptr - name;
-         case '>':                 /* One-time brackets */
+         if ((terminator > 0 && *ptr++ != terminator) || *ptr++ != ')')
-         bravalue = OP_ONCE;
+           {
-         ptr++;
+           ptr--;      /* Error offset */
-         break;
+           *errorcodeptr = ERR26;
+           goto FAILED;
+           }
-         case 'C':                 /* Callout - may be followed by digits; */
+         /* Do no further checking in the pre-compile phase. */
-         previous_callout = code;  /* Save for later completion */
-         after_manual_callout = 1; /* Skip one item before completing */
+         if (lengthptr != NULL) break;
-         *code++ = OP_CALLOUT;     /* Already checked that the terminating */
-           {                       /* closing parenthesis is present. */
+         /* In the real compile we do the work of looking for the actual
-           int n = 0;
+         reference. If the string started with "+" or "-" we require the rest to
-           while ((digitab[*(++ptr)] & ctype_digit) != 0)
+         be digits, in which case recno will be set. */
-             n = n * 10 + *ptr - '0';
-           if (n > 255)
+         if (refsign > 0)
+           {
+           if (recno <= 0)
              {
-             *errorcodeptr = ERR38;
+             *errorcodeptr = ERR58;
              goto FAILED;
              }
-           *code++ = n;
+           recno = (refsign == '-')?
-           PUT(code, 0, ptr - cd->start_pattern + 1);  /* Pattern offset */
+             cd->bracount - recno + 1 : recno +cd->bracount;
-           PUT(code, LINK_SIZE, 0);                    /* Default length */
+           if (recno <= 0 || recno > cd->final_bracount)
-           code += 2 * LINK_SIZE;
+             {
+             *errorcodeptr = ERR15;
+             goto FAILED;
+             }
+           PUT2(code, 2+LINK_SIZE, recno);
+           break;
            }
-         previous = NULL;
-         continue;
-         case 'P':                 /* Named subpattern handling */
+         /* Otherwise (did not start with "+" or "-"), start by looking for the
-         if (*(++ptr) == '<')      /* Definition */
+         name. */
-           {
-           int i, namelen;
-           uschar *slot = cd->name_table;
-           const uschar *name;     /* Don't amalgamate; some compilers */
-           name = ++ptr;           /* grumble at autoincrement in declaration */
-           while (*ptr++ != '>');
+         slot = cd->name_table;
-           namelen = ptr - name - 1;
+         for (i = 0; i < cd->names_found; i++)
+           {
+           if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
+           slot += cd->name_entry_size;
+           }
-           for (i = 0; i < cd->names_found; i++)
+         /* Found a previous named subpattern */
+         if (i < cd->names_found)
+           {
+           recno = GET2(slot, 0);
+           PUT2(code, 2+LINK_SIZE, recno);
+           }
+         /* Search the pattern for a forward reference */
+         else if ((i = find_parens(ptr, cd, name, namelen,
+                         (options & PCRE_EXTENDED) != 0)) > 0)
+           {
+           PUT2(code, 2+LINK_SIZE, i);
+           }
+         /* If terminator == 0 it means that the name followed directly after
+         the opening parenthesis [e.g. (?(abc)...] and in this case there are
+         some further alternatives to try. For the cases where terminator != 0
+         [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
+         now checked all the possibilities, so give an error. */
+         else if (terminator != 0)
+           {
+           *errorcodeptr = ERR15;
+           goto FAILED;
+           }
+         /* Check for (?(R) for recursion. Allow digits after R to specify a
+         specific group number. */
+         else if (*name == 'R')
+           {
+           recno = 0;
+           for (i = 1; i < namelen; i++)
+             {
+             if ((digitab[name[i]] & ctype_digit) == 0)
+               {
+               *errorcodeptr = ERR15;
+               goto FAILED;
+               }
+             recno = recno * 10 + name[i] - '0';
+             }
+           if (recno == 0) recno = RREF_ANY;
+           code[1+LINK_SIZE] = OP_RREF;      /* Change test type */
+           PUT2(code, 2+LINK_SIZE, recno);
+           }
+         /* Similarly, check for the (?(DEFINE) "condition", which is always
+         false. */
+         else if (namelen == 6 && strncmp((char *)name, "DEFINE", 6) == 0)
+           {
+           code[1+LINK_SIZE] = OP_DEF;
+           skipbytes = 1;
+           }
+         /* Check for the "name" actually being a subpattern number. We are
+         in the second pass here, so final_bracount is set. */
+         else if (recno > 0 && recno <= cd->final_bracount)
+           {
+           PUT2(code, 2+LINK_SIZE, recno);
+           }
+         /* Either an unidentified subpattern, or a reference to (?(0) */
+         else
+           {
+           *errorcodeptr = (recno == 0)? ERR35: ERR15;
+           goto FAILED;
+           }
+         break;
+         /* ------------------------------------------------------------ */
+         case '=':                 /* Positive lookahead */
+         bravalue = OP_ASSERT;
+         ptr++;
+         break;
+         /* ------------------------------------------------------------ */
+         case '!':                 /* Negative lookahead */
+         ptr++;
+         if (*ptr == ')')          /* Optimize (?!) */
+           {
+           *code++ = OP_FAIL;
+           previous = NULL;
+           continue;
+           }
+         bravalue = OP_ASSERT_NOT;
+         break;
+         /* ------------------------------------------------------------ */
+         case '<':                 /* Lookbehind or named define */
+         switch (ptr[1])
+           {
+           case '=':               /* Positive lookbehind */
+           bravalue = OP_ASSERTBACK;
+           ptr += 2;
+           break;
+           case '!':               /* Negative lookbehind */
+           bravalue = OP_ASSERTBACK_NOT;
+           ptr += 2;
+           break;
+           default:                /* Could be name define, else bad */
+           if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
+           ptr++;                  /* Correct offset for error */
+           *errorcodeptr = ERR24;
+           goto FAILED;
+           }
+         break;
+         /* ------------------------------------------------------------ */
+         case '>':                 /* One-time brackets */
+         bravalue = OP_ONCE;
+         ptr++;
+         break;
+         /* ------------------------------------------------------------ */
+         case 'C':                 /* Callout - may be followed by digits; */
+         previous_callout = code;  /* Save for later completion */
+         after_manual_callout = 1; /* Skip one item before completing */
+         *code++ = OP_CALLOUT;
+           {
+           int n = 0;
+           while ((digitab[*(++ptr)] & ctype_digit) != 0)
+             n = n * 10 + *ptr - '0';
+           if (*ptr != ')')
+             {
+             *errorcodeptr = ERR39;
+             goto FAILED;
+             }
+           if (n > 255)
+             {
+             *errorcodeptr = ERR38;
+             goto FAILED;
+             }
+           *code++ = n;
+           PUT(code, 0, ptr - cd->start_pattern + 1);  /* Pattern offset */
+           PUT(code, LINK_SIZE, 0);                    /* Default length */
+           code += 2 * LINK_SIZE;
+           }
+         previous