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

Diff of /code/trunk/pcre_compile.c

Parent Directory | Revision Log | View Patch Patch

-revision 170 by ph10,
Mon Jun  4 11:21:13 2007 UTC
+revision 454 by ph10,
Tue Sep 22 09:42:11 2009 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2007 University of Cambridge
+            Copyright (c) 1997-2009 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 58 
 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         *
  *************************************************/
-Line 82 
 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. */
- #ifndef EBCDIC  /* This is the "normal" table for ASCII systems */
+ #ifndef EBCDIC
+ /* This is the "normal" table for ASCII systems or for EBCDIC systems running
+ in UTF-8 mode. */
  static const short int escapes[] = {
-,      0,      0,      0,      0,      0,      0,      0,   /* 0 - 7 */
+,                       0,
-,      0,    ':',    ';',    '<',    '=',    '>',    '?',   /* 8 - ? */
+,                       0,
-    '@', -ESC_A, -ESC_B, -ESC_C, -ESC_D, -ESC_E,      0, -ESC_G,   /* @ - G */
+,                       0,
-,      0,      0, -ESC_K,      0,      0,      0,      0,   /* H - O */
+,                       0,
- -ESC_P, -ESC_Q, -ESC_R, -ESC_S,      0,      0,      0, -ESC_W,   /* P - W */
+,                       0,
- -ESC_X,      0, -ESC_Z,    '[',   '\\',    ']',    '^',    '_',   /* X - _ */
+      CHAR_COLON,              CHAR_SEMICOLON,
-    '`',      7, -ESC_b,      0, -ESC_d,  ESC_e,  ESC_f,      0,   /* ` - g */
+      CHAR_LESS_THAN_SIGN,     CHAR_EQUALS_SIGN,
-,      0,      0, -ESC_k,      0,      0,  ESC_n,      0,   /* h - o */
+      CHAR_GREATER_THAN_SIGN,  CHAR_QUESTION_MARK,
- -ESC_p,      0,  ESC_r, -ESC_s,  ESC_tee,    0,      0, -ESC_w,   /* p - w */
+      CHAR_COMMERCIAL_AT,      -ESC_A,
-,      0, -ESC_z                                            /* x - z */
+      -ESC_B,                  -ESC_C,
+      -ESC_D,                  -ESC_E,
+,                       -ESC_G,
+      -ESC_H,                  0,
+,                       -ESC_K,
+,                       0,
+,                       0,
+      -ESC_P,                  -ESC_Q,
+      -ESC_R,                  -ESC_S,
+,                       0,
+      -ESC_V,                  -ESC_W,
+      -ESC_X,                  0,
+      -ESC_Z,                  CHAR_LEFT_SQUARE_BRACKET,
+      CHAR_BACKSLASH,          CHAR_RIGHT_SQUARE_BRACKET,
+      CHAR_CIRCUMFLEX_ACCENT,  CHAR_UNDERSCORE,
+      CHAR_GRAVE_ACCENT,       7,
+      -ESC_b,                  0,
+      -ESC_d,                  ESC_e,
+      ESC_f,                   0,
+      -ESC_h,                  0,
+,                       -ESC_k,
+,                       0,
+      ESC_n,                   0,
+      -ESC_p,                  0,
+      ESC_r,                   -ESC_s,
+      ESC_tee,                 0,
+      -ESC_v,                  -ESC_w,
+,                       0,
+      -ESC_z
  };
- #else           /* This is the "abnormal" table for EBCDIC systems */
+ #else
+ /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
  static const short int escapes[] = {
  /*  48 */     0,     0,      0,     '.',    '<',   '(',    '+',    '|',
  /*  50 */   '&',     0,      0,       0,      0,     0,      0,      0,
-Line 106 
 static const short int escapes[] = {
+Line 156 
 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, -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,-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 125 
 static const short int escapes[] = {
+Line 175 
 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. The
+ string is built from string macros so that it works in UTF-8 mode on EBCDIC
- static const char *const posix_names[] = {
+ platforms. */
-   "alpha", "lower", "upper",
-   "alnum", "ascii", "blank", "cntrl", "digit", "graph",
+ typedef struct verbitem {
-   "print", "punct", "space", "word",  "xdigit" };
+   int   len;
+   int   op;
+ } verbitem;
+ static const char verbnames[] =
+   STRING_ACCEPT0
+   STRING_COMMIT0
+   STRING_F0
+   STRING_FAIL0
+   STRING_PRUNE0
+   STRING_SKIP0
+   STRING_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[] =
+   STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
+   STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
+   STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
+   STRING_word0  STRING_xdigit;
  static const uschar posix_name_lengths[] = {
 , 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
-Line 171 
 static const int posix_class_maps[] = {
+Line 257 
 static const int posix_class_maps[] = {
  /* The texts of compile-time error messages. These are "char *" because they
  are passed to the outside world. Do not ever re-use any error number, because
  they are documented. Always add a new error instead. Messages marked DEAD below
- are no longer used. */
+ are no longer used. This used to be a table of strings, but in order to reduce
+ the number of relocations needed when a shared library is loaded dynamically,
- static const char *error_texts[] = {
+ it is now one long string. We cannot use a table of offsets, because the
-   "no error",
+ lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
-   "\\ at end of pattern",
+ simply count through to the one we want - this isn't a performance issue
-   "\\c at end of pattern",
+ because these strings are used only when there is a compilation error. */
-   "unrecognized character follows \\",
-   "numbers out of order in {} quantifier",
+ 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",  /** DEAD **/
+   "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",  /** DEAD **/
+   "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 or name 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",  /** DEAD **/
+   "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 in subpattern name (missing terminator)",
+   "syntax error in subpattern name (missing terminator)\0"
-   "two named subpatterns 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)",
+   "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
-   "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")",
+   "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
    /* 50 */
-   "repeated subpattern is too long",
+   "repeated subpattern is too long\0"    /** DEAD **/
-   "octal value is greater than \\377 (not in UTF-8 mode)",
+   "octal value is greater than \\377 (not in UTF-8 mode)\0"
-   "internal error: overran compiling workspace",
+   "internal error: overran compiling workspace\0"
-   "internal error: previously-checked referenced subpattern not found",
+   "internal error: previously-checked referenced subpattern not found\0"
-   "DEFINE group contains more than one branch",
+   "DEFINE group contains more than one branch\0"
    /* 55 */
-   "repeating a DEFINE group is not allowed",
+   "repeating a DEFINE group is not allowed\0"
-   "inconsistent NEWLINE options",
+   "inconsistent NEWLINE options\0"
-   "\\g is not followed by an (optionally braced) non-zero number",
+   "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
-   "(?+ or (?- or (?(+ or (?(- must be followed by a non-zero number"
+   "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 263 
 For convenience, we use the same bit def
+Line 360 
 For convenience, we use the same bit def
  Then we can use ctype_digit and ctype_xdigit in the code. */
- #ifndef EBCDIC  /* This is the "normal" case, for ASCII systems */
+ #ifndef EBCDIC
+ /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
+ UTF-8 mode. */
  static const unsigned char digitab[] =
    {
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7 */
-Line 299 
 static const unsigned char digitab[] =
+Line 400 
 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 not running in UTF-8 mode. */
  static const unsigned char digitab[] =
    {
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7  0 */
-Line 374 
 static const unsigned char ebcdic_charta
+Line 478 
 static const unsigned char ebcdic_charta
  /* Definition to allow mutual recursion */
  static BOOL
-   compile_regex(int, int, uschar **, const uschar **, int *, BOOL, int, int *,
+   compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
-     int *, branch_chain *, compile_data *, int *);
+     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 400 
 Arguments:
+Line 526 
 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 418 
 ptr--;                            /* Set
+Line 544 
 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. */
- #ifndef EBCDIC  /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
- else if (c < '0' || c > 'z') {}                           /* Not alphameric */
+ else if (c < CHAR_0 || c > CHAR_z) {}                     /* Not alphanumeric */
- else if ((i = escapes[c - '0']) != 0) c = i;
+ else if ((i = escapes[c - CHAR_0]) != 0) c = i;
  #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 443 
 else
+Line 569 
 else
      /* A number of Perl escapes are not handled by PCRE. We give an explicit
      error. */
-     case 'l':
+     case CHAR_l:
-     case 'L':
+     case CHAR_L:
-     case 'N':
+     case CHAR_N:
-     case 'u':
+     case CHAR_u:
-     case 'U':
+     case CHAR_U:
      *errorcodeptr = ERR37;
      break;
-     /* \g must be followed by a number, either plain or braced. If positive, it
+     /* \g must be followed by one of a number of specific things:
-     is an absolute backreference. If negative, it is a relative backreference.
-     This is a Perl 5.10 feature. */
+     (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':
+     case CHAR_g:
-     if (ptr[1] == '{')
+     if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
        {
+       c = -ESC_g;
+       break;
+       }
+     /* Handle the Perl-compatible cases */
+     if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
+       {
+       const uschar *p;
+       for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
+         if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
+       if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
+         {
+         c = -ESC_k;
+         break;
+         }
        braced = TRUE;
        ptr++;
        }
      else braced = FALSE;
-     if (ptr[1] == '-')
+     if (ptr[1] == CHAR_MINUS)
        {
        negated = TRUE;
        ptr++;
-Line 472 
 else
+Line 626 
 else
      c = 0;
      while ((digitab[ptr[1]] & ctype_digit) != 0)
-       c = c * 10 + *(++ptr) - '0';
+       c = c * 10 + *(++ptr) - CHAR_0;
+     if (c < 0)   /* Integer overflow */
+       {
+       *errorcodeptr = ERR61;
+       break;
+       }
-     if (c == 0 || (braced && *(++ptr) != '}'))
+     if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
        {
        *errorcodeptr = ERR57;
-       return 0;
+       break;
+       }
+     if (c == 0)
+       {
+       *errorcodeptr = ERR58;
+       break;
        }
      if (negated)
-Line 485 
 else
+Line 651 
 else
        if (c > bracount)
          {
          *errorcodeptr = ERR15;
-         return 0;
+         break;
          }
        c = bracount - (c - 1);
        }
-Line 505 
 else
+Line 671 
 else
      value is greater than 377, the least significant 8 bits are taken. Inside a
      character class, \ followed by a digit is always an octal number. */
-     case '1': case '2': case '3': case '4': case '5':
+     case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
-     case '6': case '7': case '8': case '9':
+     case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
      if (!isclass)
        {
        oldptr = ptr;
-       c -= '0';
+       c -= CHAR_0;
        while ((digitab[ptr[1]] & ctype_digit) != 0)
-         c = c * 10 + *(++ptr) - '0';
+         c = c * 10 + *(++ptr) - CHAR_0;
+       if (c < 0)    /* Integer overflow */
+         {
+         *errorcodeptr = ERR61;
+         break;
+         }
        if (c < 10 || c <= bracount)
          {
          c = -(ESC_REF + c);
-Line 526 
 else
+Line 697 
 else
      generates a binary zero byte and treats the digit as a following literal.
      Thus we have to pull back the pointer by one. */
-     if ((c = *ptr) >= '8')
+     if ((c = *ptr) >= CHAR_8)
        {
        ptr--;
        c = 0;
-Line 539 
 else
+Line 710 
 else
      to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
      than 3 octal digits. */
-     case '0':
+     case CHAR_0:
-     c -= '0';
+     c -= CHAR_0;
-     while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7')
+     while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
-         c = c * 8 + *(++ptr) - '0';
+         c = c * 8 + *(++ptr) - CHAR_0;
      if (!utf8 && c > 255) *errorcodeptr = ERR51;
      break;
-Line 550 
 else
+Line 721 
 else
      than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
      treated as a data character. */
-     case 'x':
+     case CHAR_x:
-     if (ptr[1] == '{')
+     if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
        {
        const uschar *pt = ptr + 2;
        int count = 0;
-Line 560 
 else
+Line 731 
 else
        while ((digitab[*pt] & ctype_xdigit) != 0)
          {
          register int cc = *pt++;
-         if (c == 0 && cc == '0') continue;     /* Leading zeroes */
+         if (c == 0 && cc == CHAR_0) continue;     /* Leading zeroes */
          count++;
- #ifndef EBCDIC  /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
-         if (cc >= 'a') cc -= 32;               /* Convert to upper case */
+         if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
-         c = (c << 4) + cc - ((cc < 'A')? '0' : ('A' - 10));
+         c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
  #else           /* EBCDIC coding */
-         if (cc >= 'a' && cc <= 'z') cc += 64;  /* Convert to upper case */
+         if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
-         c = (c << 4) + cc - ((cc >= '0')? '0' : ('A' - 10));
+         c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
  #endif
          }
-       if (*pt == '}')
+       if (*pt == CHAR_RIGHT_CURLY_BRACKET)
          {
          if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
          ptr = pt;
-Line 588 
 else
+Line 759 
 else
      c = 0;
      while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
        {
-       int cc;                               /* Some compilers don't like ++ */
+       int cc;                                  /* Some compilers don't like */
-       cc = *(++ptr);                        /* in initializers */
+       cc = *(++ptr);                           /* ++ in initializers */
- #ifndef EBCDIC  /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
-       if (cc >= 'a') cc -= 32;              /* Convert to upper case */
+       if (cc >= CHAR_a) cc -= 32;              /* Convert to upper case */
-       c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));
+       c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
  #else           /* EBCDIC coding */
-       if (cc <= 'z') cc += 64;              /* Convert to upper case */
+       if (cc <= CHAR_z) cc += 64;              /* Convert to upper case */
-       c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));
+       c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
  #endif
        }
      break;
-Line 604 
 else
+Line 775 
 else
      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':
+     case CHAR_c:
      c = *(++ptr);
      if (c == 0)
        {
        *errorcodeptr = ERR2;
-       return 0;
+       break;
        }
- #ifndef EBCDIC  /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
-     if (c >= 'a' && c <= 'z') c -= 32;
+     if (c >= CHAR_a && c <= CHAR_z) c -= 32;
      c ^= 0x40;
  #else           /* EBCDIC coding */
-     if (c >= 'a' && c <= 'z') c += 64;
+     if (c >= CHAR_a && c <= CHAR_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 678 
 if (c == 0) goto ERROR_RETURN;
+Line 849 
 if (c == 0) goto ERROR_RETURN;
  /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
  negation. */
- if (c == '{')
+ if (c == CHAR_LEFT_CURLY_BRACKET)
    {
-   if (ptr[1] == '^')
+   if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
      {
      *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;
-     if (c == '}') break;
+     if (c == CHAR_RIGHT_CURLY_BRACKET) break;
      name[i] = c;
      }
-   if (c !='}') goto ERROR_RETURN;
+   if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
    name[i] = 0;
    }
-Line 714 
 top = _pcre_utt_size;
+Line 885 
 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 757 
 is_counted_repeat(const uschar *p)
+Line 928 
 is_counted_repeat(const uschar *p)
  {
  if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
  while ((digitab[*p] & ctype_digit) != 0) p++;
- if (*p == '}') return TRUE;
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
- if (*p++ != ',') return FALSE;
+ if (*p++ != CHAR_COMMA) return FALSE;
- if (*p == '}') return TRUE;
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
  if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
  while ((digitab[*p] & ctype_digit) != 0) p++;
- return (*p == '}');
+ return (*p == CHAR_RIGHT_CURLY_BRACKET);
  }
-Line 798 
 int max = -1;
+Line 969 
 int max = -1;
  /* Read the minimum value and do a paranoid check: a negative value indicates
  an integer overflow. */
- while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0';
+ while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
  if (min < 0 || min > 65535)
    {
    *errorcodeptr = ERR5;
-Line 808 
 if (min < 0 || min > 65535)
+Line 979 
 if (min < 0 || min > 65535)
  /* Read the maximum value if there is one, and again do a paranoid on its size.
  Also, max must not be less than min. */
- if (*p == '}') max = min; else
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
    {
-   if (*(++p) != '}')
+   if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
      {
      max = 0;
-     while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0';
+     while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
      if (max < 0 || max > 65535)
        {
        *errorcodeptr = ERR5;
-Line 838 
 return p;
+Line 1009 
 return p;
  /*************************************************
- *       Find forward referenced subpattern       *
+ *  Subroutine for finding forward reference      *
  *************************************************/
- /* This function scans along a pattern's text looking for capturing
+ /* This recursive function is called only from find_parens() below. The
+ top-level call starts at the beginning of the pattern. All other calls must
+ start at a parenthesis. It scans along a pattern's text looking for capturing
  subpatterns, and counting them. If it finds a named pattern that matches the
  name it is given, it returns its number. Alternatively, if the name is NULL, it
- returns when it reaches a given numbered subpattern. This is used for forward
+ returns when it reaches a given numbered subpattern. We know that if (?P< is
- references to subpatterns. We know that if (?P< is encountered, the name will
+ encountered, the name will be terminated by '>' because that is checked in the
- be terminated by '>' because that is checked in the first pass.
+ first pass. Recursion is used to keep track of subpatterns that reset the
+ capturing group numbers - the (?| feature.
  Arguments:
-   ptr          current position in the pattern
+   ptrptr       address of the current character pointer (updated)
-   count        current count of capturing parens so far encountered
+   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
+   count        pointer to the current capturing subpattern number (updated)
  Returns:       the number of the named subpattern, or -1 if not found
  */
  static int
- find_parens(const uschar *ptr, int count, const uschar *name, int lorn,
+ find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
-   BOOL xmode)
+   BOOL xmode, int *count)
  {
- const uschar *thisname;
+ uschar *ptr = *ptrptr;
+ int start_count = *count;
+ int hwm_count = start_count;
+ BOOL dup_parens = FALSE;
- for (; *ptr != 0; ptr++)
+ /* If the first character is a parenthesis, check on the type of group we are
+ dealing with. The very first call may not start with a parenthesis. */
+ if (ptr[0] == CHAR_LEFT_PARENTHESIS)
    {
-   int term;
+   if (ptr[1] == CHAR_QUESTION_MARK &&
+       ptr[2] == CHAR_VERTICAL_LINE)
+     {
+     ptr += 3;
+     dup_parens = TRUE;
+     }
+   /* Handle a normal, unnamed capturing parenthesis */
+   else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
+     {
+     *count += 1;
+     if (name == NULL && *count == lorn) return *count;
+     ptr++;
+     }
+   /* Handle a condition. If it is an assertion, just carry on so that it
+   is processed as normal. If not, skip to the closing parenthesis of the
+   condition (there can't be any nested parens. */
+   else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
+     {
+     ptr += 2;
+     if (ptr[1] != CHAR_QUESTION_MARK)
+       {
+       while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
+       if (*ptr != 0) ptr++;
+       }
+     }
+   /* We have either (? or (* and not a condition */
+   else
+     {
+     ptr += 2;
+     if (*ptr == CHAR_P) ptr++;                      /* Allow optional P */
+     /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
+     if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
+         ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
+       {
+       int term;
+       const uschar *thisname;
+       *count += 1;
+       if (name == NULL && *count == lorn) return *count;
+       term = *ptr++;
+       if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
+       thisname = ptr;
+       while (*ptr != term) ptr++;
+       if (name != NULL && lorn == ptr - thisname &&
+           strncmp((const char *)name, (const char *)thisname, lorn) == 0)
+         return *count;
+       term++;
+       }
+     }
+   }
+ /* Past any initial parenthesis handling, scan for parentheses or vertical
+ bars. */
+ for (; *ptr != 0; ptr++)
+   {
    /* Skip over backslashed characters and also entire \Q...\E */
-   if (*ptr == '\\')
+   if (*ptr == CHAR_BACKSLASH)
      {
-     if (*(++ptr) == 0) return -1;
+     if (*(++ptr) == 0) goto FAIL_EXIT;
-     if (*ptr == 'Q') for (;;)
+     if (*ptr == CHAR_Q) for (;;)
        {
-       while (*(++ptr) != 0 && *ptr != '\\');
+       while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
-       if (*ptr == 0) return -1;
+       if (*ptr == 0) goto FAIL_EXIT;
-       if (*(++ptr) == 'E') break;
+       if (*(++ptr) == CHAR_E) break;
        }
      continue;
      }
-   /* Skip over character classes */
+   /* 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. Note the use of STR macros to
+   encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
-   if (*ptr == '[')
+   if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
      {
-     while (*(++ptr) != ']')
+     BOOL negate_class = FALSE;
+     for (;;)
+       {
+       if (ptr[1] == CHAR_BACKSLASH)
+         {
+         if (ptr[2] == CHAR_E)
+           ptr+= 2;
+         else if (strncmp((const char *)ptr+2,
+                  STR_Q STR_BACKSLASH STR_E, 3) == 0)
+           ptr += 4;
+         else
+           break;
+         }
+       else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
+         {
+         negate_class = TRUE;
+         ptr++;
+         }
+       else break;
+       }
+     /* If the next character is ']', it is a data character that must be
+     skipped, except in JavaScript compatibility mode. */
+     if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
+         (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
+       ptr++;
+     while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
        {
-       if (*ptr == '\\')
+       if (*ptr == 0) return -1;
+       if (*ptr == CHAR_BACKSLASH)
          {
-         if (*(++ptr) == 0) return -1;
+         if (*(++ptr) == 0) goto FAIL_EXIT;
-         if (*ptr == 'Q') for (;;)
+         if (*ptr == CHAR_Q) for (;;)
            {
-           while (*(++ptr) != 0 && *ptr != '\\');
+           while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
-           if (*ptr == 0) return -1;
+           if (*ptr == 0) goto FAIL_EXIT;
-           if (*(++ptr) == 'E') break;
+           if (*(++ptr) == CHAR_E) break;
            }
          continue;
          }
-Line 905 
 for (; *ptr != 0; ptr++)
+Line 1180 
 for (; *ptr != 0; ptr++)
    /* Skip comments in /x mode */
-   if (xmode && *ptr == '#')
+   if (xmode && *ptr == CHAR_NUMBER_SIGN)
      {
-     while (*(++ptr) != 0 && *ptr != '\n');
+     while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
-     if (*ptr == 0) return -1;
+     if (*ptr == 0) goto FAIL_EXIT;
      continue;
      }
-   /* An opening parens must now be a real metacharacter */
+   /* Check for the special metacharacters */
-   if (*ptr != '(') continue;
+   if (*ptr == CHAR_LEFT_PARENTHESIS)
-   if (ptr[1] != '?')
      {
-     count++;
+     int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
-     if (name == NULL && count == lorn) return count;
+     if (rc > 0) return rc;
-     continue;
+     if (*ptr == 0) goto FAIL_EXIT;
+     }
+   else if (*ptr == CHAR_RIGHT_PARENTHESIS)
+     {
+     if (dup_parens && *count < hwm_count) *count = hwm_count;
+     *ptrptr = ptr;
+     return -1;
+     }
+   else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
+     {
+     if (*count > hwm_count) hwm_count = *count;
+     *count = start_count;
      }
+   }
-   ptr += 2;
+ FAIL_EXIT:
-   if (*ptr == 'P') ptr++;                      /* Allow optional P */
+ *ptrptr = ptr;
+ return -1;
+ }
-   /* 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++;
+ *       Find forward referenced subpattern       *
-   if (term == '<') term = '>';
+ *************************************************/
-   thisname = ptr;
-   while (*ptr != term) ptr++;
+ /* This function scans along a pattern's text looking for capturing
-   if (name != NULL && lorn == ptr - thisname &&
+ subpatterns, and counting them. If it finds a named pattern that matches the
-       strncmp((const char *)name, (const char *)thisname, lorn) == 0)
+ name it is given, it returns its number. Alternatively, if the name is NULL, it
-     return count;
+ returns when it reaches a given numbered subpattern. This is used for forward
+ references to subpatterns. We used to be able to start this scan from the
+ current compiling point, using the current count value from cd->bracount, and
+ do it all in a single loop, but the addition of the possibility of duplicate
+ subpattern numbers means that we have to scan from the very start, in order to
+ take account of such duplicates, and to use a recursive function to keep track
+ of the different types of group.
+ Arguments:
+   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 found subpattern, or -1 if not found
+ */
+ static int
+ find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
+ {
+ uschar *ptr = (uschar *)cd->start_pattern;
+ int count = 0;
+ int rc;
+ /* If the pattern does not start with an opening parenthesis, the first call
+ to find_parens_sub() will scan right to the end (if necessary). However, if it
+ does start with a parenthesis, find_parens_sub() will return when it hits the
+ matching closing parens. That is why we have to have a loop. */
+ for (;;)
+   {
+   rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
+   if (rc > 0 || *ptr++ == 0) break;
    }
- return -1;
+ return rc;
  }
  /*************************************************
  *      Find first significant op code            *
  *************************************************/
-Line 1013 
 for (;;)
+Line 1331 
 for (;;)
  /*************************************************
- *        Find the fixed length of a pattern      *
+ *        Find the fixed length of a branch       *
  *************************************************/
- /* Scan a pattern and compute the fixed length of subject that will match it,
+ /* Scan a branch and compute the fixed length of subject that will match it,
  if the length is fixed. This is needed for dealing with backward assertions.
- In UTF8 mode, the result is in characters rather than bytes.
+ In UTF8 mode, the result is in characters rather than bytes. The branch is
+ temporarily terminated with OP_END when this function is called.
+ This function is called when a backward assertion is encountered, so that if it
+ fails, the error message can point to the correct place in the pattern.
+ However, we cannot do this when the assertion contains subroutine calls,
+ because they can be forward references. We solve this by remembering this case
+ and doing the check at the end; a flag specifies which mode we are running in.
  Arguments:
    code     points to the start of the pattern (the bracket)
    options  the compiling options
+   atend    TRUE if called when the pattern is complete
+   cd       the "compile data" structure
- Returns:   the fixed length, or -1 if there is no fixed length,
+ Returns:   the fixed length,
+              or -1 if there is no fixed length,
               or -2 if \C was encountered
+              or -3 if an OP_RECURSE item was encountered and atend is FALSE
  */
  static int
- find_fixedlength(uschar *code, int options)
+ find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
  {
  int length = -1;
-Line 1042 
 branch, check the length against that of
+Line 1371 
 branch, check the length against that of
  for (;;)
    {
    int d;
+   uschar *ce, *cs;
    register int op = *cc;
    switch (op)
      {
      case OP_CBRA:
      case OP_BRA:
      case OP_ONCE:
      case OP_COND:
-     d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options);
+     d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
      if (d < 0) return d;
      branchlength += d;
      do cc += GET(cc, 1); while (*cc == OP_ALT);
-Line 1072 
 for (;;)
+Line 1401 
 for (;;)
      cc += 1 + LINK_SIZE;
      branchlength = 0;
      break;
+     /* A true recursion implies not fixed length, but a subroutine call may
+     be OK. If the subroutine is a forward reference, we can't deal with
+     it until the end of the pattern, so return -3. */
+     case OP_RECURSE:
+     if (!atend) return -3;
+     cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */
+     do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */
+     if (cc > cs && cc < ce) return -1;                /* Recursion */
+     d = find_fixedlength(cs + 2, options, atend, cd);
+     if (d < 0) return d;
+     branchlength += d;
+     cc += 1 + LINK_SIZE;
+     break;
      /* Skip over assertive subpatterns */
-Line 1109 
 for (;;)
+Line 1453 
 for (;;)
      branchlength++;
      cc += 2;
  #ifdef SUPPORT_UTF8
-     if ((options & PCRE_UTF8) != 0)
+     if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
-       {
+       cc += _pcre_utf8_table4[cc[-1] & 0x3f];
-       while ((*cc & 0xc0) == 0x80) cc++;
-       }
  #endif
      break;
-Line 1123 
 for (;;)
+Line 1465 
 for (;;)
      branchlength += GET2(cc,1);
      cc += 4;
  #ifdef SUPPORT_UTF8
-     if ((options & PCRE_UTF8) != 0)
+     if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
-       {
+       cc += _pcre_utf8_table4[cc[-1] & 0x3f];
-       while((*cc & 0x80) == 0x80) cc++;
-       }
  #endif
      break;
      case OP_TYPEEXACT:
      branchlength += GET2(cc,1);
+     if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
      cc += 4;
      break;
-Line 1149 
 for (;;)
+Line 1490 
 for (;;)
      case OP_NOT_WORDCHAR:
      case OP_WORDCHAR:
      case OP_ANY:
+     case OP_ALLANY:
      branchlength++;
      cc++;
      break;
-Line 1203 
 for (;;)
+Line 1545 
 for (;;)
  /*************************************************
- *    Scan compiled regex for numbered bracket    *
+ *    Scan compiled regex for specific bracket    *
  *************************************************/
  /* This little function scans through a compiled pattern until it finds a
- capturing bracket with the given number.
+ capturing bracket with the given number, or, if the number is negative, an
+ instance of OP_REVERSE for a lookbehind.
  Arguments:
    code        points to start of expression
    utf8        TRUE in UTF-8 mode
-   number      the required bracket number
+   number      the required bracket number or negative to find a lookbehind
  Returns:      pointer to the opcode for the bracket, or NULL if not found
  */
-Line 1230 
 for (;;)
+Line 1573 
 for (;;)
    the table is zero; the actual length is stored in the compiled code. */
    if (c == OP_XCLASS) code += GET(code, 1);
+   /* Handle recursion */
+   else if (c == OP_REVERSE)
+     {
+     if (number < 0) return (uschar *)code;
+     code += _pcre_OP_lengths[c];
+     }
    /* Handle capturing bracket */
-Line 1240 
 for (;;)
+Line 1591 
 for (;;)
      code += _pcre_OP_lengths[c];
      }
-   /* In UTF-8 mode, opcodes that are followed by a character may be followed by
+   /* Otherwise, we can get the item's length from the table, except that for
-   a multi-byte character. The length in the table is a minimum, so we have to
+   repeated character types, we have to test for \p and \P, which have an extra
-   arrange to skip the extra bytes. */
+   two bytes of parameters. */
    else
      {
+     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;
+       case OP_TYPEUPTO:
+       case OP_TYPEMINUPTO:
+       case OP_TYPEEXACT:
+       case OP_TYPEPOSUPTO:
+       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
+   arrange to skip the extra bytes. */
  #ifdef SUPPORT_UTF8
      if (utf8) switch(c)
        {
-Line 1268 
 for (;;)
+Line 1648 
 for (;;)
        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
        break;
        }
+ #else
+     (void)(utf8);  /* Keep compiler happy by referencing function argument */
  #endif
      }
    }
-Line 1304 
 for (;;)
+Line 1686 
 for (;;)
    if (c == OP_XCLASS) code += GET(code, 1);
-   /* Otherwise, we get the item's length from the table. In UTF-8 mode, opcodes
+   /* Otherwise, we can get the item's length from the table, except that for
-   that are followed by a character may be followed by a multi-byte character.
+   repeated character types, we have to test for \p and \P, which have an extra
-   The length in the table is a minimum, so we have to arrange to skip the extra
+   two bytes of parameters. */
-   bytes. */
    else
      {
+     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;
+       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 arrange to skip the extra bytes. */
  #ifdef SUPPORT_UTF8
      if (utf8) switch(c)
        {
-Line 1333 
 for (;;)
+Line 1743 
 for (;;)
        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
        break;
        }
+ #else
+     (void)(utf8);  /* Keep compiler happy by referencing function argument */
  #endif
      }
    }
-Line 1348 
 for (;;)
+Line 1760 
 for (;;)
  can match the empty string or not. It is called from could_be_empty()
  below and from compile_branch() when checking for an unlimited repeat of a
  group that can match nothing. Note that first_significant_code() skips over
- assertions. If we hit an unclosed bracket, we return "empty" - this means we've
+ backward and negative forward assertions when its final argument is TRUE. If we
- struck an inner bracket whose current branch will already have been scanned.
+ 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 1370 
 for (code = first_significant_code(code
+Line 1783 
 for (code = first_significant_code(code
    const uschar *ccode;
    c = *code;
+   /* 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)
+   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)
+   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 */
-     /* Scan a closed bracket */
+     /* If a conditional group has only one branch, there is a second, implied,
+     empty branch, so just skip over the conditional, because it could be empty.
+     Otherwise, scan the individual branches of the group. */
-     empty_branch = FALSE;
+     if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
-     do
-       {
-       if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
-         empty_branch = TRUE;
        code += GET(code, 1);
+     else
+       {
+       empty_branch = FALSE;
+       do
+         {
+         if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
+           empty_branch = TRUE;
+         code += GET(code, 1);
+         }
+       while (*code == OP_ALT);
+       if (!empty_branch) return FALSE;   /* All branches are non-empty */
        }
-     while (*code == OP_ALT);
-     if (!empty_branch) return FALSE;   /* All branches are non-empty */
+     c = *code;
-     c = *code;
      continue;
      }
-Line 1406 
 for (code = first_significant_code(code
+Line 1838 
 for (code = first_significant_code(code
    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 1454 
 for (code = first_significant_code(code
+Line 1890 
 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:
-Line 1472 
 for (code = first_significant_code(code
+Line 1909 
 for (code = first_significant_code(code
      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 1490 
 for (code = first_significant_code(code
+Line 1947 
 for (code = first_significant_code(code
      case OP_QUERY:
      case OP_MINQUERY:
      case OP_POSQUERY:
+     if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
+     break;
      case OP_UPTO:
      case OP_MINUPTO:
      case OP_POSUPTO:
-     if (utf8) while ((code[2] & 0xc0) == 0x80) code++;
+     if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
      break;
  #endif
      }
-Line 1541 
 return TRUE;
+Line 2001 
 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 == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
-   return TRUE;
+     {
+     if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
+     if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
+       {
+       *endptr = ptr;
+       return TRUE;
+       }
+     }
    }
  return FALSE;
  }
-Line 1588 
 Returns:     a value representing the na
+Line 2067 
 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 1607 
 return -1;
+Line 2088 
 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 one of the jobs of this function. Before it is called,
+ have their offsets adjusted. That one of the jobs of this function. Before it
- the 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
-Line 1634 
 adjust_recurse(uschar *group, int adjust
+Line 2116 
 adjust_recurse(uschar *group, int adjust
    uschar *save_hwm)
  {
  uschar *ptr = group;
  while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
    {
    int offset;
-Line 1745 
 get_othercase_range(unsigned int *cptr,
+Line 2228 
 get_othercase_range(unsigned int *cptr,
  unsigned int c, othercase, next;
  for (c = *cptr; c <= d; c++)
-   { if ((othercase = _pcre_ucp_othercase(c)) != NOTACHAR) break; }
+   { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
  if (c > d) return FALSE;
-Line 1754 
 next = othercase + 1;
+Line 2237 
 next = othercase + 1;
  for (++c; c <= d; c++)
    {
-   if (_pcre_ucp_othercase(c) != next) break;
+   if (UCD_OTHERCASE(c) != next) break;
    next++;
    }
-Line 1800 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2283 
 if ((options & PCRE_EXTENDED) != 0)
    for (;;)
      {
      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
-     if (*ptr == '#')
+     if (*ptr == CHAR_NUMBER_SIGN)
        {
        while (*(++ptr) != 0)
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
-Line 1812 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2295 
 if ((options & PCRE_EXTENDED) != 0)
  /* 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 == '\\')
+ if (*ptr == CHAR_BACKSLASH)
    {
    int temperrorcode = 0;
    next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
-Line 1837 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2320 
 if ((options & PCRE_EXTENDED) != 0)
    for (;;)
      {
      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
-     if (*ptr == '#')
+     if (*ptr == CHAR_NUMBER_SIGN)
        {
        while (*(++ptr) != 0)
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
-Line 1848 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2331 
 if ((options & PCRE_EXTENDED) != 0)
  /* If the next thing is itself optional, we have to give up. */
- if (*ptr == '*' || *ptr == '?' || strncmp((char *)ptr, "{0,", 3) == 0)
+ if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
-   return FALSE;
+   strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
+     return FALSE;
  /* Now compare the next item with the previous opcode. If the previous is a
  positive single character match, "item" either contains the character or, if
-Line 1864 
 if (next >= 0) switch(op_code)
+Line 2348 
 if (next >= 0) switch(op_code)
    case OP_CHAR:
  #ifdef SUPPORT_UTF8
    if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+ #else
+   (void)(utf8_char);  /* Keep compiler happy by referencing function argument */
  #endif
    return item != next;
-Line 1882 
 if (next >= 0) switch(op_code)
+Line 2368 
 if (next >= 0) switch(op_code)
      unsigned int othercase;
      if (next < 128) othercase = cd->fcc[next]; else
  #ifdef SUPPORT_UCP
-     othercase = _pcre_ucp_othercase((unsigned int)next);
+     othercase = UCD_OTHERCASE((unsigned int)next);
  #else
      othercase = NOTACHAR;
  #endif
-Line 1895 
 if (next >= 0) switch(op_code)
+Line 2381 
 if (next >= 0) switch(op_code)
    /* For OP_NOT, "item" must be a single-byte character. */
    case OP_NOT:
-   if (next < 0) return FALSE;  /* Not a character */
    if (item == next) return TRUE;
    if ((options & PCRE_CASELESS) == 0) return FALSE;
  #ifdef SUPPORT_UTF8
-Line 1904 
 if (next >= 0) switch(op_code)
+Line 2389 
 if (next >= 0) switch(op_code)
      unsigned int othercase;
      if (next < 128) othercase = cd->fcc[next]; else
  #ifdef SUPPORT_UCP
-     othercase = _pcre_ucp_othercase(next);
+     othercase = UCD_OTHERCASE(next);
  #else
      othercase = NOTACHAR;
  #endif
-Line 1932 
 if (next >= 0) switch(op_code)
+Line 2417 
 if (next >= 0) switch(op_code)
    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;
    }
-Line 1966 
 switch(op_code)
+Line 2495 
 switch(op_code)
      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;
+   return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
+          next == -ESC_h || next == -ESC_v;
    case OP_NOT_DIGIT:
    return next == -ESC_d;
-Line 1980 
 switch(op_code)
+Line 2554 
 switch(op_code)
    return next == -ESC_S || next == -ESC_d || next == -ESC_w;
    case OP_NOT_WHITESPACE:
-   return next == -ESC_s;
+   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;
+   return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
    case OP_NOT_WORDCHAR:
    return next == -ESC_w || next == -ESC_d;
-Line 2056 
 uschar classbits[32];
+Line 2643 
 uschar classbits[32];
  BOOL class_utf8;
  BOOL utf8 = (options & PCRE_UTF8) != 0;
  uschar *class_utf8data;
+ uschar *class_utf8data_base;
  uschar utf8_char[6];
  #else
  BOOL utf8 = FALSE;
-Line 2095 
 req_caseopt = ((options & PCRE_CASELESS)
+Line 2683 
 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;
-Line 2135 
 for (;; ptr++)
+Line 2725 
 for (;; ptr++)
      */
      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));
-Line 2172 
 for (;; ptr++)
+Line 2771 
 for (;; ptr++)
    if (inescq && c != 0)
      {
-     if (c == '\\' && ptr[1] == 'E')
+     if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
        {
        inescq = FALSE;
        ptr++;
-Line 2198 
 for (;; ptr++)
+Line 2797 
 for (;; ptr++)
    /* Fill in length of a previous callout, except when the next thing is
    a quantifier. */
-   is_quantifier = c == '*' || c == '+' || c == '?' ||
+   is_quantifier =
-     (c == '{' && is_counted_repeat(ptr+1));
+     c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
+     (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
    if (!is_quantifier && previous_callout != NULL &&
         after_manual_callout-- <= 0)
-Line 2214 
 for (;; ptr++)
+Line 2814 
 for (;; ptr++)
    if ((options & PCRE_EXTENDED) != 0)
      {
      if ((cd->ctypes[c] & ctype_space) != 0) continue;
-     if (c == '#')
+     if (c == CHAR_NUMBER_SIGN)
        {
        while (*(++ptr) != 0)
          {
-Line 2239 
 for (;; ptr++)
+Line 2839 
 for (;; ptr++)
      {
      /* ===================================================================*/
      case 0:                        /* The branch terminates at string end */
-     case '|':                      /* or | or ) */
+     case CHAR_VERTICAL_LINE:       /* or | or ) */
-     case ')':
+     case CHAR_RIGHT_PARENTHESIS:
      *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"));
        }
-Line 2257 
 for (;; ptr++)
+Line 2862 
 for (;; ptr++)
      /* Handle single-character metacharacters. In multiline mode, ^ disables
      the setting of any following char as a first character. */
-     case '^':
+     case CHAR_CIRCUMFLEX_ACCENT:
      if ((options & PCRE_MULTILINE) != 0)
        {
        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
-Line 2266 
 for (;; ptr++)
+Line 2871 
 for (;; ptr++)
      *code++ = OP_CIRC;
      break;
-     case '$':
+     case CHAR_DOLLAR_SIGN:
      previous = NULL;
      *code++ = OP_DOLL;
      break;
-Line 2274 
 for (;; ptr++)
+Line 2879 
 for (;; ptr++)
      /* There can never be a first char if '.' is first, whatever happens about
      repeats. The value of reqbyte doesn't change either. */
-     case '.':
+     case CHAR_DOT:
      if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
      zerofirstbyte = firstbyte;
      zeroreqbyte = reqbyte;
      previous = code;
-     *code++ = OP_ANY;
+     *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
      break;
-Line 2294 
 for (;; ptr++)
+Line 2899 
 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.
-     */
-     case '[':
+     In JavaScript compatibility mode, an isolated ']' causes an error. In
+     default (Perl) mode, it is treated as a data character. */
+     case CHAR_RIGHT_SQUARE_BRACKET:
+     if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
+       {
+       *errorcodeptr = ERR64;
+       goto FAILED;
+       }
+     goto NORMAL_CHAR;
+     case CHAR_LEFT_SQUARE_BRACKET:
      previous = code;
      /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
      they are encountered at the top level, so we'll do that too. */
-     if ((ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
+     if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
-         check_posix_syntax(ptr, &tempptr, cd))
+          ptr[1] == CHAR_EQUALS_SIGN) &&
+         check_posix_syntax(ptr, &tempptr))
        {
-       *errorcodeptr = (ptr[1] == ':')? ERR13 : ERR31;
+       *errorcodeptr = (ptr[1] == CHAR_COLON)? 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 == CHAR_BACKSLASH)
+         {
+         if (ptr[1] == CHAR_E)
+           ptr++;
+         else if (strncmp((const char *)ptr+1,
+                           STR_Q STR_BACKSLASH STR_E, 3) == 0)
+           ptr += 3;
+         else
+           break;
+         }
+       else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
+         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 == CHAR_RIGHT_SQUARE_BRACKET &&
+         (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). However, For higher
      valued UTF-8 characters, we don't yet do any optimization. */
-Line 2338 
 for (;; ptr++)
+Line 2985 
 for (;; ptr++)
  #ifdef SUPPORT_UTF8
      class_utf8 = FALSE;                       /* No chars >= 256 */
      class_utf8data = code + LINK_SIZE + 2;    /* For UTF-8 items */
+     class_utf8data_base = class_utf8data;     /* For resetting in pass 1 */
  #endif
      /* Process characters until ] is reached. By writing this as a "do" it
-Line 2353 
 for (;; ptr++)
+Line 3001 
 for (;; ptr++)
          {                           /* 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 we are at \E */
+         if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)  /* If we are at \E */
            {
            inescq = FALSE;                   /* Reset literal state */
            ptr++;                            /* Skip the 'E' */
-Line 2374 
 for (;; ptr++)
+Line 3034 
 for (;; ptr++)
        [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
 .6 and 5.8 do. */
-       if (c == '[' &&
+       if (c == CHAR_LEFT_SQUARE_BRACKET &&
-           (ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
+           (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
-           check_posix_syntax(ptr, &tempptr, cd))
+            ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
          {
          BOOL local_negate = FALSE;
          int posix_class, taboffset, tabopt;
          register const uschar *cbits = cd->cbits;
          uschar pbits[32];
-         if (ptr[1] != ':')
+         if (ptr[1] != CHAR_COLON)
            {
            *errorcodeptr = ERR31;
            goto FAILED;
            }
          ptr += 2;
-         if (*ptr == '^')
+         if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
            {
            local_negate = TRUE;
+           should_flip_negation = TRUE;  /* Note negative special */
            ptr++;
            }
-Line 2459 
 for (;; ptr++)
+Line 3120 
 for (;; ptr++)
        of the specials, which just set a flag. The sequence \b is a special
        case. Inside a class (and only there) it is treated as backspace.
        Elsewhere it marks a word boundary. Other escapes have preset maps ready
-       to or into the one we are building. We assume they have more than one
+       to 'or' into the one we are building. We assume they have more than one
        character in them, so set class_charcount bigger than one. */
-       if (c == '\\')
+       if (c == CHAR_BACKSLASH)
          {
          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 = CHAR_BS;       /* \b is backspace in a class */
-         else if (-c == ESC_X) c = 'X';   /* \X is literal X in a class */
+         else if (-c == ESC_X) c = CHAR_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_R) c = CHAR_R;   /* \R is literal R in a class */
          else if (-c == ESC_Q)            /* Handle start of quoted string */
            {
-           if (ptr[1] == '\\' && ptr[2] == 'E')
+           if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
              {
              ptr += 2; /* avoid empty string */
              }
            else inescq = TRUE;
            continue;
            }
+         else if (-c == ESC_E) continue;  /* Ignore orphan \E */
          if (c < 0)
            {
-Line 2494 
 for (;; ptr++)
+Line 3156 
 for (;; ptr++)
              continue;
              case ESC_D:
+             should_flip_negation = TRUE;
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
              continue;
-Line 2502 
 for (;; ptr++)
+Line 3165 
 for (;; ptr++)
              continue;
              case ESC_W:
+             should_flip_negation = TRUE;
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
              continue;
-Line 2511 
 for (;; ptr++)
+Line 3175 
 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;
-             case ESC_E: /* Perl ignores an orphan \E */
+             default:    /* Not recognized; fall through */
+             break;      /* Need "default" setting to stop compiler warning. */
+             }
+           /* In the pre-compile phase, just do the recognition. */
+           else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
+                    c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
+           /* We need to deal with \H, \h, \V, and \v in both phases because
+           they use extra memory. */
+           if (-c == ESC_h)
+             {
+             SETBIT(classbits, 0x09); /* VT */
+             SETBIT(classbits, 0x20); /* SPACE */
+             SETBIT(classbits, 0xa0); /* NSBP */
+ #ifdef SUPPORT_UTF8
+             if (utf8)
+               {
+               class_utf8 = TRUE;
+               *class_utf8data++ = XCL_SINGLE;
+               class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
+               *class_utf8data++ = XCL_SINGLE;
+               class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
+               *class_utf8data++ = XCL_SINGLE;
+               class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
+               *class_utf8data++ = XCL_SINGLE;
+               class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
+               *class_utf8data++ = XCL_SINGLE;
+               class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
+               }
+ #endif
+             continue;
+             }
+           if (-c == ESC_H)
+             {
+             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;
+               }
+ #ifdef SUPPORT_UTF8
+             if (utf8)
+               {
+               class_utf8 = TRUE;
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
+               }
+ #endif
              continue;
+             }
-             default:    /* Not recognized; fall through */
+           if (-c == ESC_v)
-             break;      /* Need "default" setting to stop compiler warning. */
+             {
+             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;
              }
-           /* In the pre-compile phase, just do the recognition. */
+           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;
+               }
-           else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
+ #ifdef SUPPORT_UTF8
-                    c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
+             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. */
-Line 2570 
 for (;; ptr++)
+Line 3359 
 for (;; ptr++)
        entirely. The code for handling \Q and \E is messy. */
        CHECK_RANGE:
-       while (ptr[1] == '\\' && ptr[2] == 'E')
+       while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
          {
          inescq = FALSE;
          ptr += 2;
-Line 2578 
 for (;; ptr++)
+Line 3367 
 for (;; ptr++)
        oldptr = ptr;
-       if (!inescq && ptr[1] == '-')
+       /* Remember \r or \n */
+       if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
+       /* Check for range */
+       if (!inescq && ptr[1] == CHAR_MINUS)
          {
          int d;
          ptr += 2;
-         while (*ptr == '\\' && ptr[1] == 'E') ptr += 2;
+         while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
          /* If we hit \Q (not followed by \E) at this point, go into escaped
          mode. */
-         while (*ptr == '\\' && ptr[1] == 'Q')
+         while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
            {
            ptr += 2;
-           if (*ptr == '\\' && ptr[1] == 'E') { ptr += 2; continue; }
+           if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
+             { ptr += 2; continue; }
            inescq = TRUE;
            break;
            }
-         if (*ptr == 0 || (!inescq && *ptr == ']'))
+         if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
            {
            ptr = oldptr;
            goto LONE_SINGLE_CHARACTER;
-Line 2614 
 for (;; ptr++)
+Line 3410 
 for (;; ptr++)
          not any of the other escapes. Perl 5.6 treats a hyphen as a literal
          in such circumstances. */
-         if (!inescq && d == '\\')
+         if (!inescq && d == CHAR_BACKSLASH)
            {
            d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
            if (*errorcodeptr != 0) goto FAILED;
-           /* \b is backslash; \X is literal X; \R is literal R; any other
+           /* \b is backspace; \X is literal X; \R is literal R; any other
            special means the '-' was literal */
            if (d < 0)
              {
-             if (d == -ESC_b) d = '\b';
+             if (d == -ESC_b) d = CHAR_BS;
-             else if (d == -ESC_X) d = 'X';
+             else if (d == -ESC_X) d = CHAR_X;
-             else if (d == -ESC_R) d = 'R'; else
+             else if (d == -ESC_R) d = CHAR_R; else
                {
                ptr = oldptr;
                goto LONE_SINGLE_CHARACTER;  /* A few lines below */
-Line 2645 
 for (;; ptr++)
+Line 3441 
 for (;; ptr++)
          if (d == c) goto LONE_SINGLE_CHARACTER;  /* A few lines below */
+         /* Remember \r or \n */
+         if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
          /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
          matching, we have to use an XCLASS with extra data items. Caseless
          matching for characters > 127 is available only if UCP support is
-Line 2667 
 for (;; ptr++)
+Line 3467 
 for (;; ptr++)
              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 2759 
 for (;; ptr++)
+Line 3563 
 for (;; ptr++)
          if ((options & PCRE_CASELESS) != 0)
            {
            unsigned int othercase;
-           if ((othercase = _pcre_ucp_othercase(c)) != NOTACHAR)
+           if ((othercase = UCD_OTHERCASE(c)) != c)
              {
              *class_utf8data++ = XCL_SINGLE;
              class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
-Line 2786 
 for (;; ptr++)
+Line 3590 
 for (;; ptr++)
      /* Loop until ']' reached. This "while" is the end of the "do" above. */
-     while ((c = *(++ptr)) != 0 && (c != ']' || inescq));
+     while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
      if (c == 0)                          /* Missing terminating ']' */
        {
-Line 2794 
 for (;; ptr++)
+Line 3598 
 for (;; ptr++)
        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
      /* 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 2809 
 for (;; ptr++)
+Line 3635 
 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 2855 
 for (;; ptr++)
+Line 3679 
 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 in the actual compiled code. */
+     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;
-Line 2885 
 for (;; ptr++)
+Line 3712 
 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];
        }
      else
        {
-       *code++ = OP_CLASS;
        memcpy(code, classbits, 32);
        }
      code += 32;
-Line 2909 
 for (;; ptr++)
+Line 3735 
 for (;; ptr++)
      /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
      has been tested above. */
-     case '{':
+     case CHAR_LEFT_CURLY_BRACKET:
      if (!is_quantifier) goto NORMAL_CHAR;
      ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
      if (*errorcodeptr != 0) goto FAILED;
      goto REPEAT;
-     case '*':
+     case CHAR_ASTERISK:
      repeat_min = 0;
      repeat_max = -1;
      goto REPEAT;
-     case '+':
+     case CHAR_PLUS:
      repeat_min = 1;
      repeat_max = -1;
      goto REPEAT;
-     case '?':
+     case CHAR_QUESTION_MARK:
      repeat_min = 0;
      repeat_max = 1;
-Line 2960 
 for (;; ptr++)
+Line 3786 
 for (;; ptr++)
      but if PCRE_UNGREEDY is set, it works the other way round. We change the
      repeat type to the non-default. */
-     if (ptr[1] == '+')
+     if (ptr[1] == CHAR_PLUS)
        {
        repeat_type = 0;                  /* Force greedy */
        possessive_quantifier = TRUE;
        ptr++;
        }
-     else if (ptr[1] == '?')
+     else if (ptr[1] == CHAR_QUESTION_MARK)
        {
        repeat_type = greedy_non_default;
        ptr++;
-Line 3081 
 for (;; ptr++)
+Line 3907 
 for (;; ptr++)
        if (repeat_max == 0) goto END_REPEAT;
+       /*--------------------------------------------------------------------*/
+       /* This code is obsolete from release 8.00; the restriction was finally
+       removed: */
        /* 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 3231 
 for (;; ptr++)
+Line 4062 
 for (;; ptr++)
          goto END_REPEAT;
          }
+       /*--------------------------------------------------------------------*/
+       /* This code is obsolete from release 8.00; the restriction was finally
+       removed: */
        /* 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 3270 
 for (;; ptr++)
+Line 4106 
 for (;; ptr++)
          goto FAILED;
          }
-       /* This is a paranoid check to stop integer overflow later on */
-       if (len > MAX_DUPLENGTH)
-         {
-         *errorcodeptr = ERR50;
-         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 3300 
 for (;; ptr++)
+Line 4128 
 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)
-           {
-           code = previous;
-           goto END_REPEAT;
-           }
-         /* If the maximum is 1 or unlimited, we just have to stick in the
+         ** if (repeat_max == 0)
-         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
+         **   code = previous;
-         any internal or forward referenced group, because the offset is from
+         **   goto END_REPEAT;
-         the start of the whole regex. Temporarily terminate the pattern while
+         **   }
-         doing this. */
+         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 (repeat_max <= 1)
+         if (repeat_max <= 1)    /* Covers 0, 1, and unlimited */
            {
            *code = OP_END;
            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 3366 
 for (;; ptr++)
+Line 4204 
 for (;; ptr++)
          if (repeat_min > 1)
            {
            /* In the pre-compile phase, we don't actually do the replication. We
-           just adjust the length as if we had. */
+           just adjust the length as if we had. Do some paranoid checks for
+           potential integer overflow. */
            if (lengthptr != NULL)
-             *lengthptr += (repeat_min - 1)*length_prevgroup;
+             {
+             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 */
-Line 3407 
 for (;; ptr++)
+Line 4256 
 for (;; ptr++)
          /* 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. */
+         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)
-           *lengthptr += repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
+           {
-- 2*LINK_SIZE;  /* Last one doesn't nest */
+           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 */
-Line 3493 
 for (;; ptr++)
+Line 4354 
 for (;; ptr++)
          }
        }
+     /* 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 3517 
 for (;; ptr++)
+Line 4385 
 for (;; ptr++)
      if (possessive_quantifier)
        {
        int len;
-       if (*tempcode == OP_EXACT || *tempcode == OP_TYPEEXACT ||
-           *tempcode == OP_NOTEXACT)
+       if (*tempcode == OP_TYPEEXACT)
+         tempcode += _pcre_OP_lengths[*tempcode] +
+           ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
+       else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
+         {
          tempcode += _pcre_OP_lengths[*tempcode];
+ #ifdef SUPPORT_UTF8
+         if (utf8 && tempcode[-1] >= 0xc0)
+           tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
+ #endif
+         }
        len = code - tempcode;
        if (len > 0) switch (*tempcode)
          {
-Line 3563 
 for (;; ptr++)
+Line 4442 
 for (;; ptr++)
      /* ===================================================================*/
      /* Start of nested parenthesized sub-expression, or comment or lookahead or
      lookbehind or option setting or condition or all the other extended
-     parenthesis forms. First deal with the specials; all are introduced by ?,
+     parenthesis forms.  */
-     and the appearance of any of them means that this is not a capturing
-     group. */
-     case '(':
+     case CHAR_LEFT_PARENTHESIS:
      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) == CHAR_ASTERISK && (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 == CHAR_COLON)
+         {
+         *errorcodeptr = ERR59;   /* Not supported */
+         goto FAILED;
+         }
+       if (*ptr != CHAR_RIGHT_PARENTHESIS)
+         {
+         *errorcodeptr = ERR60;
+         goto FAILED;
+         }
+       namelen = ptr - name;
+       for (i = 0; i < verbcount; i++)
+         {
+         if (namelen == verbs[i].len &&
+             strncmp((char *)name, vn, namelen) == 0)
+           {
+           /* Check for open captures before ACCEPT */
+           if (verbs[i].op == OP_ACCEPT)
+             {
+             open_capitem *oc;
+             cd->had_accept = TRUE;
+             for (oc = cd->open_caps; oc != NULL; oc = oc->next)
+               {
+               *code++ = OP_CLOSE;
+               PUT2INC(code, 0, oc->number);
+               }
+             }
+           *code++ = verbs[i].op;
+           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 == CHAR_QUESTION_MARK)
        {
        int i, set, unset, namelen;
        int *optset;
-Line 3582 
 for (;; ptr++)
+Line 4510 
 for (;; ptr++)
        switch (*(++ptr))
          {
-         case '#':                 /* Comment; skip to ket */
+         case CHAR_NUMBER_SIGN:                 /* Comment; skip to ket */
          ptr++;
-         while (*ptr != 0 && *ptr != ')') ptr++;
+         while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
          if (*ptr == 0)
            {
            *errorcodeptr = ERR18;
-Line 3594 
 for (;; ptr++)
+Line 4522 
 for (;; ptr++)
          /* ------------------------------------------------------------ */
-         case ':':                 /* Non-capturing bracket */
+         case CHAR_VERTICAL_LINE:  /* Reset capture count for each branch */
+         reset_bracount = TRUE;
+         /* Fall through */
+         /* ------------------------------------------------------------ */
+         case CHAR_COLON:          /* Non-capturing bracket */
          bravalue = OP_BRA;
          ptr++;
          break;
          /* ------------------------------------------------------------ */
-         case '(':
+         case CHAR_LEFT_PARENTHESIS:
          bravalue = OP_COND;       /* Conditional group */
          /* A condition can be an assertion, a number (referring to a numbered
-Line 3621 
 for (;; ptr++)
+Line 4554 
 for (;; ptr++)
          the switch. This will take control down to where bracketed groups,
          including assertions, are processed. */
-         if (ptr[1] == '?' && (ptr[2] == '=' || ptr[2] == '!' || ptr[2] == '<'))
+         if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
+             ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
            break;
          /* Most other conditions use OP_CREF (a couple change to OP_RREF
-Line 3629 
 for (;; ptr++)
+Line 4563 
 for (;; ptr++)
          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] == '&')
+         if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
            {
            terminator = -1;
            ptr += 2;
-Line 3643 
 for (;; ptr++)
+Line 4577 
 for (;; ptr++)
          /* Check for a test for a named group's having been set, using the Perl
          syntax (?(<name>) or (?('name') */
-         else if (ptr[1] == '<')
+         else if (ptr[1] == CHAR_LESS_THAN_SIGN)
            {
-           terminator = '>';
+           terminator = CHAR_GREATER_THAN_SIGN;
            ptr++;
            }
-         else if (ptr[1] == '\'')
+         else if (ptr[1] == CHAR_APOSTROPHE)
            {
-           terminator = '\'';
+           terminator = CHAR_APOSTROPHE;
            ptr++;
            }
          else
            {
            terminator = 0;
-           if (ptr[1] == '-' || ptr[1] == '+') refsign = *(++ptr);
+           if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
            }
          /* We now expect to read a name; any thing else is an error */
-Line 3676 
 for (;; ptr++)
+Line 4610 
 for (;; ptr++)
            {
            if (recno >= 0)
              recno = ((digitab[*ptr] & ctype_digit) != 0)?
-               recno * 10 + *ptr - '0' : -1;
+               recno * 10 + *ptr - CHAR_0 : -1;
            ptr++;
            }
          namelen = ptr - name;
-         if ((terminator > 0 && *ptr++ != terminator) || *ptr++ != ')')
+         if ((terminator > 0 && *ptr++ != terminator) ||
+             *ptr++ != CHAR_RIGHT_PARENTHESIS)
            {
            ptr--;      /* Error offset */
            *errorcodeptr = ERR26;
-Line 3695 
 for (;; ptr++)
+Line 4630 
 for (;; ptr++)
          /* In the real compile we do the work of looking for the actual
          reference. If the string started with "+" or "-" we require the rest to
          be digits, in which case recno will be set. */
          if (refsign > 0)
            {
            if (recno <= 0)
              {
              *errorcodeptr = ERR58;
              goto FAILED;
              }
-           if (refsign == '-')
+           recno = (refsign == CHAR_MINUS)?
+             cd->bracount - recno + 1 : recno +cd->bracount;
+           if (recno <= 0 || recno > cd->final_bracount)
              {
-             recno = cd->bracount - recno + 1;
+             *errorcodeptr = ERR15;
-             if (recno <= 0)
+             goto FAILED;
-               {
-               *errorcodeptr = ERR15;
-               goto FAILED;
-               }
              }
-           else recno += cd->bracount;
            PUT2(code, 2+LINK_SIZE, recno);
            break;
            }
          /* Otherwise (did not start with "+" or "-"), start by looking for the
          name. */
          slot = cd->name_table;
          for (i = 0; i < cd->names_found; i++)
            {
-Line 3737 
 for (;; ptr++)
+Line 4669 
 for (;; ptr++)
          /* Search the pattern for a forward reference */
-         else if ((i = find_parens(ptr, cd->bracount, name, namelen,
+         else if ((i = find_parens(cd, name, namelen,
                          (options & PCRE_EXTENDED) != 0)) > 0)
            {
            PUT2(code, 2+LINK_SIZE, i);
-Line 3758 
 for (;; ptr++)
+Line 4690 
 for (;; ptr++)
          /* Check for (?(R) for recursion. Allow digits after R to specify a
          specific group number. */
-         else if (*name == 'R')
+         else if (*name == CHAR_R)
            {
            recno = 0;
            for (i = 1; i < namelen; i++)
-Line 3768 
 for (;; ptr++)
+Line 4700 
 for (;; ptr++)
                *errorcodeptr = ERR15;
                goto FAILED;
                }
-             recno = recno * 10 + name[i] - '0';
+             recno = recno * 10 + name[i] - CHAR_0;
              }
            if (recno == 0) recno = RREF_ANY;
            code[1+LINK_SIZE] = OP_RREF;      /* Change test type */
-Line 3778 
 for (;; ptr++)
+Line 4710 
 for (;; ptr++)
          /* Similarly, check for the (?(DEFINE) "condition", which is always
          false. */
-         else if (namelen == 6 && strncmp((char *)name, "DEFINE", 6) == 0)
+         else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
            {
            code[1+LINK_SIZE] = OP_DEF;
            skipbytes = 1;
            }
-         /* Check for the "name" actually being a subpattern number. */
+         /* 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)
+         else if (recno > 0 && recno <= cd->final_bracount)
            {
            PUT2(code, 2+LINK_SIZE, recno);
            }
-Line 3802 
 for (;; ptr++)
+Line 4735 
 for (;; ptr++)
          /* ------------------------------------------------------------ */
-         case '=':                 /* Positive lookahead */
+         case CHAR_EQUALS_SIGN:                 /* Positive lookahead */
          bravalue = OP_ASSERT;
          ptr++;
          break;
          /* ------------------------------------------------------------ */
-         case '!':                 /* Negative lookahead */
+         case CHAR_EXCLAMATION_MARK:            /* Negative lookahead */
-         bravalue = OP_ASSERT_NOT;
          ptr++;
+         if (*ptr == CHAR_RIGHT_PARENTHESIS)    /* Optimize (?!) */
+           {
+           *code++ = OP_FAIL;
+           previous = NULL;
+           continue;
+           }
+         bravalue = OP_ASSERT_NOT;
          break;
          /* ------------------------------------------------------------ */
-         case '<':                 /* Lookbehind or named define */
+         case CHAR_LESS_THAN_SIGN:              /* Lookbehind or named define */
          switch (ptr[1])
            {
-           case '=':               /* Positive lookbehind */
+           case CHAR_EQUALS_SIGN:               /* Positive lookbehind */
            bravalue = OP_ASSERTBACK;
            ptr += 2;
            break;
-           case '!':               /* Negative lookbehind */
+           case CHAR_EXCLAMATION_MARK:          /* Negative lookbehind */
            bravalue = OP_ASSERTBACK_NOT;
            ptr += 2;
            break;
-Line 3839 
 for (;; ptr++)
+Line 4778 
 for (;; ptr++)
          /* ------------------------------------------------------------ */
-         case '>':                 /* One-time brackets */
+         case CHAR_GREATER_THAN_SIGN:           /* One-time brackets */
          bravalue = OP_ONCE;
          ptr++;
          break;
          /* ------------------------------------------------------------ */
-         case 'C':                 /* Callout - may be followed by digits; */
+         case CHAR_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';
+             n = n * 10 + *ptr - CHAR_0;
-           if (*ptr != ')')
+           if (*ptr != CHAR_RIGHT_PARENTHESIS)
              {
              *errorcodeptr = ERR39;
              goto FAILED;
-Line 3874 
 for (;; ptr++)
+Line 4813 
 for (;; ptr++)
          /* ------------------------------------------------------------ */
-         case 'P':                 /* Python-style named subpattern handling */
+         case CHAR_P:              /* Python-style named subpattern handling */
-         if (*(++ptr) == '=' || *ptr == '>')  /* Reference or recursion */
+         if (*(++ptr) == CHAR_EQUALS_SIGN ||
+             *ptr == CHAR_GREATER_THAN_SIGN)  /* Reference or recursion */
            {
-           is_recurse = *ptr == '>';
+           is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
-           terminator = ')';
+           terminator = CHAR_RIGHT_PARENTHESIS;
            goto NAMED_REF_OR_RECURSE;
            }
-         else if (*ptr != '<')    /* Test for Python-style definition */
+         else if (*ptr != CHAR_LESS_THAN_SIGN)  /* Test for Python-style defn */
            {
            *errorcodeptr = ERR41;
            goto FAILED;
-Line 3891 
 for (;; ptr++)
+Line 4831 
 for (;; ptr++)
          /* ------------------------------------------------------------ */
          DEFINE_NAME:    /* Come here from (?< handling */
-         case '\'':
+         case CHAR_APOSTROPHE:
            {
-           terminator = (*ptr == '<')? '>' : '\'';
+           terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
+             CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
            name = ++ptr;
            while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
-Line 3967 
 for (;; ptr++)
+Line 4908 
 for (;; ptr++)
          /* ------------------------------------------------------------ */
-         case '&':                 /* Perl recursion/subroutine syntax */
+         case CHAR_AMPERSAND:            /* Perl recursion/subroutine syntax */
-         terminator = ')';
+         terminator = CHAR_RIGHT_PARENTHESIS;
          is_recurse = TRUE;
          /* Fall through */
          /* We come here from the Python syntax above that handles both
          references (?P=name) and recursion (?P>name), as well as falling
-         through from the Perl recursion syntax (?&name). */
+         through from the Perl recursion syntax (?&name). We also come here from
+         the Perl \k<name> or \k'name' back reference syntax and the \k{name}
+         .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
          NAMED_REF_OR_RECURSE:
          name = ++ptr;
-Line 3986 
 for (;; ptr++)
+Line 4929 
 for (;; ptr++)
          if (lengthptr != NULL)
            {
+           if (namelen == 0)
+             {
+             *errorcodeptr = ERR62;
+             goto FAILED;
+             }
            if (*ptr != terminator)
              {
              *errorcodeptr = ERR42;
-Line 3999 
 for (;; ptr++)
+Line 4947 
 for (;; ptr++)
            recno = 0;
            }
-         /* In the real compile, seek the name in the table */
+         /* In the real compile, seek the name in the table. We check the name
+         first, and then check that we have reached the end of the name in the
+         table. That way, if the name that is longer than any in the table,
+         the comparison will fail without reading beyond the table entry. */
          else
            {
            slot = cd->name_table;
            for (i = 0; i < cd->names_found; i++)
              {
-             if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
+             if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
+                 slot[2+namelen] == 0)
+               break;
              slot += cd->name_entry_size;
              }
-Line 4015 
 for (;; ptr++)
+Line 4968 
 for (;; ptr++)
              recno = GET2(slot, 0);
              }
            else if ((recno =                /* Forward back reference */
-                     find_parens(ptr, cd->bracount, name, namelen,
+                     find_parens(cd, name, namelen,
                        (options & PCRE_EXTENDED) != 0)) <= 0)
              {
              *errorcodeptr = ERR15;
-Line 4031 
 for (;; ptr++)
+Line 4984 
 for (;; ptr++)
          /* ------------------------------------------------------------ */
-         case 'R':                 /* Recursion */
+         case CHAR_R:              /* Recursion */
          ptr++;                    /* Same as (?0)      */
          /* Fall through */
          /* ------------------------------------------------------------ */
-         case '-': case '+':
+         case CHAR_MINUS: case CHAR_PLUS:  /* Recursion or subroutine */
-         case '0': case '1': case '2': case '3': case '4':   /* Recursion or */
+         case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
-         case '5': case '6': case '7': case '8': case '9':   /* subroutine */
+         case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
            {
            const uschar *called;
+           terminator = CHAR_RIGHT_PARENTHESIS;
+           /* Come here from the \g<...> and \g'...' code (Oniguruma
+           compatibility). However, the syntax has been checked to ensure that
+           the ... are a (signed) number, so that neither ERR63 nor ERR29 will
+           be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
+           ever be taken. */
+           HANDLE_NUMERICAL_RECURSION:
-           if ((refsign = *ptr) == '+') ptr++;
+           if ((refsign = *ptr) == CHAR_PLUS)
-           else if (refsign == '-')
+             {
+             ptr++;
+             if ((digitab[*ptr] & ctype_digit) == 0)
+               {
+               *errorcodeptr = ERR63;
+               goto FAILED;
+               }
+             }
+           else if (refsign == CHAR_MINUS)
              {
              if ((digitab[ptr[1]] & ctype_digit) == 0)
                goto OTHER_CHAR_AFTER_QUERY;
              ptr++;
              }
            recno = 0;
            while((digitab[*ptr] & ctype_digit) != 0)
-             recno = recno * 10 + *ptr++ - '0';
+             recno = recno * 10 + *ptr++ - CHAR_0;
-           if (*ptr != ')')
+           if (*ptr != terminator)
              {
              *errorcodeptr = ERR29;
              goto FAILED;
              }
-           if (refsign == '-')
+           if (refsign == CHAR_MINUS)
              {
              if (recno == 0)
                {
                *errorcodeptr = ERR58;
                goto FAILED;
                }
              recno = cd->bracount - recno + 1;
              if (recno <= 0)
                {
                *errorcodeptr = ERR15;
                goto FAILED;
                }
              }
-           else if (refsign == '+')
+           else if (refsign == CHAR_PLUS)
              {
              if (recno == 0)
                {
                *errorcodeptr = ERR58;
                goto FAILED;
                }
              recno += cd->bracount;
              }
            /* Come here from code above that handles a named recursion */
-Line 4108 
 for (;; ptr++)
+Line 5078 
 for (;; ptr++)
              if (called == NULL)
                {
-               if (find_parens(ptr, cd->bracount, NULL, recno,
+               if (find_parens(cd, NULL, recno,
-                    (options & PCRE_EXTENDED) != 0) < 0)
+                     (options & PCRE_EXTENDED) != 0) < 0)
                  {
                  *errorcodeptr = ERR15;
                  goto FAILED;
-Line 4157 
 for (;; ptr++)
+Line 5127 
 for (;; ptr++)
          /* ------------------------------------------------------------ */
          default:              /* Other characters: check option setting */
          OTHER_CHAR_AFTER_QUERY:
          set = unset = 0;
          optset = &set;
-         while (*ptr != ')' && *ptr != ':')
+         while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
            {
            switch (*ptr++)
              {
-             case '-': optset = &unset; break;
+             case CHAR_MINUS: optset = &unset; break;
-             case 'J':    /* Record that it changed in the external options */
+             case CHAR_J:    /* Record that it changed in the external options */
              *optset |= PCRE_DUPNAMES;
-             cd->external_options |= PCRE_JCHANGED;
+             cd->external_flags |= PCRE_JCHANGED;
              break;
-             case 'i': *optset |= PCRE_CASELESS; break;
+             case CHAR_i: *optset |= PCRE_CASELESS; break;
-             case 'm': *optset |= PCRE_MULTILINE; break;
+             case CHAR_m: *optset |= PCRE_MULTILINE; break;
-             case 's': *optset |= PCRE_DOTALL; break;
+             case CHAR_s: *optset |= PCRE_DOTALL; break;
-             case 'x': *optset |= PCRE_EXTENDED; break;
+             case CHAR_x: *optset |= PCRE_EXTENDED; break;
-             case 'U': *optset |= PCRE_UNGREEDY; break;
+             case CHAR_U: *optset |= PCRE_UNGREEDY; break;
-             case 'X': *optset |= PCRE_EXTRA; break;
+             case CHAR_X: *optset |= PCRE_EXTRA; break;
              default:  *errorcodeptr = ERR12;
                        ptr--;    /* Correct the offset */
-Line 4208 
 for (;; ptr++)
+Line 5178 
 for (;; ptr++)
          both phases.
          If we are not at the pattern start, compile code to change the ims
-         options if this setting actually changes any of them. We also pass the
+         options if this setting actually changes any of them, and reset the
-         new setting back so that it can be put at the start of any following
+         greedy defaults and the case value for firstbyte and reqbyte. */
-         branches, and when this group ends (if we are in a group), a resetting
-         item can be compiled. */
-         if (*ptr == ')')
+         if (*ptr == CHAR_RIGHT_PARENTHESIS)
            {
            if (code == cd->start_code + 1 + LINK_SIZE &&
                 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
              {
              cd->external_options = newoptions;
-             options = newoptions;
              }
           else
              {
-Line 4228 
 for (;; ptr++)
+Line 5195 
 for (;; ptr++)
                *code++ = OP_OPT;
                *code++ = newoptions & PCRE_IMS;
                }
-             /* Change options at this level, and pass them back for use
-             in subsequent branches. Reset the greedy defaults and the case
-             value for firstbyte and reqbyte. */
-             *optionsptr = options = newoptions;
              greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
              greedy_non_default = greedy_default ^ 1;
-             req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
+             req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
              }
+           /* Change options at this level, and pass them back for use
+           in subsequent branches. When not at the start of the pattern, this
+           information is also necessary so that a resetting item can be
+           compiled at the end of a group (if we are in a group). */
+           *optionsptr = options = newoptions;
            previous = NULL;       /* This item can't be repeated */
            continue;              /* It is complete */
            }
-Line 4292 
 for (;; ptr++)
+Line 5259 
 for (;; ptr++)
           errorcodeptr,                 /* Where to put an error message */
           (bravalue == OP_ASSERTBACK ||
            bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
+          reset_bracount,               /* True if (?| group */
           skipbytes,                    /* Skip over bracket number */
           &subfirstbyte,                /* For possible first char */
           &subreqbyte,                  /* For possible last char */
-Line 4308 
 for (;; ptr++)
+Line 5276 
 for (;; ptr++)
      is on the bracket. */
      /* If this is a conditional bracket, check that there are no more than
-     two branches in the group, or just one if it's a DEFINE group. */
+     two branches in the group, or just one if it's a DEFINE group. We do this
+     in the real compile phase, not in the pre-pass, where the whole group may
+     not be available. */
-     if (bravalue == OP_COND)
+     if (bravalue == OP_COND && lengthptr == NULL)
        {
        uschar *tc = code;
        int condcount = 0;
-Line 4351 
 for (;; ptr++)
+Line 5321 
 for (;; ptr++)
      /* Error if hit end of pattern */
-     if (*ptr != ')')
+     if (*ptr != CHAR_RIGHT_PARENTHESIS)
        {
        *errorcodeptr = ERR14;
        goto FAILED;
        }
-     /* In the pre-compile phase, update the length by the length of the nested
+     /* In the pre-compile phase, update the length by the length of the group,
-     group, less the brackets at either end. Then reduce the compiled code to
+     less the brackets at either end. Then reduce the compiled code to just a
-     just the brackets so that it doesn't use much memory if it is duplicated by
+     set of non-capturing brackets so that it doesn't use much memory if it is
-     a quantifier. */
+     duplicated by a quantifier.*/
      if (lengthptr != NULL)
        {
+       if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
+         {
+         *errorcodeptr = ERR20;
+         goto FAILED;
+         }
        *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
-       code++;
+       *code++ = OP_BRA;
        PUTINC(code, 0, 1 + LINK_SIZE);
        *code++ = OP_KET;
        PUTINC(code, 0, 1 + LINK_SIZE);
+       break;    /* No need to waste time with special character handling */
        }
      /* Otherwise update the main code pointer to the end of the group. */
-     else code = tempcode;
+     code = tempcode;
      /* For a DEFINE group, required and first character settings are not
      relevant. */
-Line 4443 
 for (;; ptr++)
+Line 5419 
 for (;; ptr++)
      We can test for values between ESC_b and ESC_Z for the latter; this may
      have to change if any new ones are ever created. */
-     case '\\':
+     case CHAR_BACKSLASH:
      tempptr = ptr;
      c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
      if (*errorcodeptr != 0) goto FAILED;
-Line 4452 
 for (;; ptr++)
+Line 5428 
 for (;; ptr++)
        {
        if (-c == ESC_Q)            /* Handle start of quoted string */
          {
-         if (ptr[1] == '\\' && ptr[2] == 'E') ptr += 2; /* avoid empty string */
+         if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
-           else inescq = TRUE;
+           ptr += 2;               /* avoid empty string */
+             else inescq = TRUE;
          continue;
          }
-Line 4470 
 for (;; ptr++)
+Line 5447 
 for (;; ptr++)
        zerofirstbyte = firstbyte;
        zeroreqbyte = reqbyte;
-       /* \k<name> or \k'name' is a back reference by name (Perl syntax) */
+       /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
+       is a subroutine call by number (Oniguruma syntax). In fact, the value
+       -ESC_g is returned only for these cases. So we don't need to check for <
+       or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
+       -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
+       that is a synonym for a named back reference). */
+       if (-c == ESC_g)
+         {
+         const uschar *p;
+         save_hwm = cd->hwm;   /* Normally this is set when '(' is read */
+         terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
+           CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
+         /* These two statements stop the compiler for warning about possibly
+         unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
+         fact, because we actually check for a number below, the paths that
+         would actually be in error are never taken. */
+         skipbytes = 0;
+         reset_bracount = FALSE;
+         /* Test for a name */
+         if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
+           {
+           BOOL isnumber = TRUE;
+           for (p = ptr + 1; *p != 0 && *p != terminator; p++)
+             {
+             if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
+             if ((cd->ctypes[*p] & ctype_word) == 0) break;
+             }
+           if (*p != terminator)
+             {
+             *errorcodeptr = ERR57;
+             break;
+             }
+           if (isnumber)
+             {
+             ptr++;
+             goto HANDLE_NUMERICAL_RECURSION;
+             }
+           is_recurse = TRUE;
+           goto NAMED_REF_OR_RECURSE;
+           }
+         /* Test a signed number in angle brackets or quotes. */
+         p = ptr + 2;
+         while ((digitab[*p] & ctype_digit) != 0) p++;
+         if (*p != terminator)
+           {
+           *errorcodeptr = ERR57;
+           break;
+           }
+         ptr++;
+         goto HANDLE_NUMERICAL_RECURSION;
+         }
+       /* \k<name> or \k'name' is a back reference by name (Perl syntax).
+       We also support \k{name} (.NET syntax) */
-       if (-c == ESC_k && (ptr[1] == '<' || ptr[1] == '\''))
+       if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
+           ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
          {
          is_recurse = FALSE;
-         terminator = (*(++ptr) == '<')? '>' : '\'';
+         terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
+           CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
+           CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
          goto NAMED_REF_OR_RECURSE;
          }
-Line 4576 
 for (;; ptr++)
+Line 5616 
 for (;; ptr++)
      *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
      for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
+     /* Remember if \r or \n were seen */
+     if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
+       cd->external_flags |= PCRE_HASCRORLF;
      /* Set the first and required bytes appropriately. If no previous first
      byte, set it from this character, but revert to none on a zero repeat.
      Otherwise, leave the firstbyte value alone, and don't change it on a zero
-Line 4641 
 This function is used during the pre-com
+Line 5686 
 This function is used during the pre-com
  out the amount of memory needed, as well as during the real compile phase. The
  value of lengthptr distinguishes the two phases.
- Argument:
+ Arguments:
    options        option bits, including any changes for this subpattern
    oldims         previous settings of ims option bits
    codeptr        -> the address of the current code pointer
    ptrptr         -> the address of the current pattern pointer
    errorcodeptr   -> pointer to error code variable
    lookbehind     TRUE if this is a lookbehind assertion
+   reset_bracount TRUE to reset the count for each branch
    skipbytes      skip this many bytes at start (for brackets and OP_COND)
    firstbyteptr   place to put the first required character, or a negative number
    reqbyteptr     place to put the last required character, or a negative number
-Line 4661 
 Returns:         TRUE on success
+Line 5707 
 Returns:         TRUE on success
  static BOOL
  compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
-   int *errorcodeptr, BOOL lookbehind, int skipbytes, int *firstbyteptr,
+   int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
-   int *reqbyteptr, branch_chain *bcptr, compile_data *cd, int *lengthptr)
+   int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
+   int *lengthptr)
  {
  const uschar *ptr = *ptrptr;
  uschar *code = *codeptr;
  uschar *last_branch = code;
  uschar *start_bracket = code;
  uschar *reverse_count = NULL;
+ open_capitem capitem;
+ int capnumber = 0;
  int firstbyte, reqbyte;
  int branchfirstbyte, branchreqbyte;
  int length;
+ int orig_bracount;
+ int max_bracount;
  branch_chain bc;
  bc.outer = bcptr;
-Line 4693 
 the code that abstracts option settings
+Line 5744 
 the code that abstracts option settings
  them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
  pre-compile phase to find out whether anything has yet been compiled or not. */
+ /* If this is a capturing subpattern, add to the chain of open capturing items
+ so that we can detect them if (*ACCEPT) is encountered. */
+ if (*code == OP_CBRA)
+   {
+   capnumber = GET2(code, 1 + LINK_SIZE);
+   capitem.number = capnumber;
+   capitem.next = cd->open_caps;
+   cd->open_caps = &capitem;
+   }
  /* Offset is set zero to mark that this bracket is still open */
  PUT(code, 1, 0);
-Line 4700 
 code += 1 + LINK_SIZE + skipbytes;
+Line 5762 
 code += 1 + LINK_SIZE + skipbytes;
  /* Loop for each alternative branch */
+ orig_bracount = max_bracount = cd->bracount;
  for (;;)
    {
+   /* For a (?| group, reset the capturing bracket count so that each branch
+   uses the same numbers. */
+   if (reset_bracount) cd->bracount = orig_bracount;
    /* Handle a change of ims options at the start of the branch */
    if ((options & PCRE_IMS) != oldims)
-Line 4731 
 for (;;)
+Line 5799 
 for (;;)
      return FALSE;
      }
+   /* Keep the highest bracket count in case (?| was used and some branch
+   has fewer than the rest. */
+   if (cd->bracount > max_bracount) max_bracount = cd->bracount;
    /* In the real compile phase, there is some post-processing to be done. */
    if (lengthptr == NULL)
 #