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

Diff of /code/trunk/pcre_compile.c

Parent Directory | Revision Log | View Patch Patch

-revision 85 by nigel,
Sat Feb 24 21:41:13 2007 UTC
+revision 708 by ph10,
Fri Sep 23 11:03:03 2011 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2005 University of Cambridge
+            Copyright (c) 1997-2011 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"
- /* When DEBUG is defined, we need the pcre_printint() function, which is also
+ /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
- used by pcretest. DEBUG is not defined when building a production library. */
+ also used by pcretest. PCRE_DEBUG is not defined when building a production
+ library. */
- #ifdef DEBUG
+ #ifdef PCRE_DEBUG
  #include "pcre_printint.src"
  #endif
+ /* Macro for setting individual bits in class bitmaps. */
+ #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
+ /* Maximum length value to check against when making sure that the integer that
+ holds the compiled pattern length does not overflow. We make it a bit less than
+ INT_MAX to allow for adding in group terminating bytes, so that we don't have
+ to check them every time. */
+ #define OFLOW_MAX (INT_MAX - 20)
  /*************************************************
  *      Code parameters and static tables         *
  *************************************************/
- /* Maximum number of items on the nested bracket stacks at compile time. This
+ /* This value specifies the size of stack workspace that is used during the
- applies to the nesting of all kinds of parentheses. It does not limit
+ first pre-compile phase that determines how much memory is required. The regex
- un-nested, non-capturing parentheses. This number can be made bigger if
+ is partly compiled into this space, but the compiled parts are discarded as
- necessary - it is used to dimension one int and one unsigned char vector at
+ soon as they can be, so that hopefully there will never be an overrun. The code
- compile time. */
+ does, however, check for an overrun. The largest amount I've seen used is 218,
+ so this number is very generous.
+ The same workspace is used during the second, actual compile phase for
+ remembering forward references to groups so that they can be filled in at the
+ end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
+ is 4 there is plenty of room. */
+ #define COMPILE_WORK_SIZE (4096)
- #define BRASTACK_SIZE 200
+ /* The overrun tests check for a slightly smaller size so that they detect the
+ overrun before it actually does run off the end of the data block. */
+ #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
  /* Table for handling escaped characters in the range '0'-'z'. Positive returns
-Line 72 
 are simple data values; negative values
+Line 103 
 are simple data values; negative values
  on. Zero means further processing is needed (for things like \x), or the escape
  is invalid. */
- #if !EBCDIC   /* This is the "normal" table for ASCII systems */
+ #ifndef EBCDIC
+ /* This is the "normal" table for ASCII systems 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,      0,      0,      0,      0,      0,   /* H - O */
+,                       0,
- -ESC_P, -ESC_Q,      0, -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,      0,      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,
+      -ESC_N,                  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 96 
 static const short int escapes[] = {
+Line 162 
 static const short int escapes[] = {
  /*  70 */     0,     0,      0,       0,      0,     0,      0,      0,
  /*  78 */     0,   '`',    ':',     '#',    '@',  '\'',    '=',    '"',
  /*  80 */     0,     7, -ESC_b,       0, -ESC_d, ESC_e,  ESC_f,      0,
- /*  88 */     0,     0,      0,     '{',      0,     0,      0,      0,
+ /*  88 */-ESC_h,     0,      0,     '{',      0,     0,      0,      0,
- /*  90 */     0,     0,      0,     'l',      0, ESC_n,      0, -ESC_p,
+ /*  90 */     0,     0, -ESC_k,     'l',      0, ESC_n,      0, -ESC_p,
  /*  98 */     0, ESC_r,      0,     '}',      0,     0,      0,      0,
- /*  A0 */     0,   '~', -ESC_s, ESC_tee,      0,     0, -ESC_w,      0,
+ /*  A0 */     0,   '~', -ESC_s, ESC_tee,      0,-ESC_v, -ESC_w,      0,
  /*  A8 */     0,-ESC_z,      0,       0,      0,   '[',      0,      0,
  /*  B0 */     0,     0,      0,       0,      0,     0,      0,      0,
  /*  B8 */     0,     0,      0,       0,      0,   ']',    '=',    '-',
  /*  C0 */   '{',-ESC_A, -ESC_B,  -ESC_C, -ESC_D,-ESC_E,      0, -ESC_G,
- /*  C8 */     0,     0,      0,       0,      0,     0,      0,      0,
+ /*  C8 */-ESC_H,     0,      0,       0,      0,     0,      0,      0,
- /*  D0 */   '}',     0,      0,       0,      0,     0,      0, -ESC_P,
+ /*  D0 */   '}',     0, -ESC_K,       0,      0,-ESC_N,      0, -ESC_P,
- /*  D8 */-ESC_Q,     0,      0,       0,      0,     0,      0,      0,
+ /*  D8 */-ESC_Q,-ESC_R,      0,       0,      0,     0,      0,      0,
- /*  E0 */  '\\',     0, -ESC_S,       0,      0,     0, -ESC_W, -ESC_X,
+ /*  E0 */  '\\',     0, -ESC_S,       0,      0,-ESC_V, -ESC_W, -ESC_X,
  /*  E8 */     0,-ESC_Z,      0,       0,      0,     0,      0,      0,
  /*  F0 */     0,     0,      0,       0,      0,     0,      0,      0,
  /*  F8 */     0,     0,      0,       0,      0,     0,      0,      0
-Line 115 
 static const short int escapes[] = {
+Line 181 
 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, upper, lower,
+ 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;                 /* Length of verb name */
+   int   op;                  /* Op when no arg, or -1 if arg mandatory */
+   int   op_arg;              /* Op when arg present, or -1 if not allowed */
+ } verbitem;
+ static const char verbnames[] =
+   "\0"                       /* Empty name is a shorthand for MARK */
+   STRING_MARK0
+   STRING_ACCEPT0
+   STRING_COMMIT0
+   STRING_F0
+   STRING_FAIL0
+   STRING_PRUNE0
+   STRING_SKIP0
+   STRING_THEN;
+ static const verbitem verbs[] = {
+   { 0, -1,        OP_MARK },
+   { 4, -1,        OP_MARK },
+   { 6, OP_ACCEPT, -1 },
+   { 6, OP_COMMIT, -1 },
+   { 1, OP_FAIL,   -1 },
+   { 4, OP_FAIL,   -1 },
+   { 5, OP_PRUNE,  OP_PRUNE_ARG },
+   { 4, OP_SKIP,   OP_SKIP_ARG  },
+   { 4, OP_THEN,   OP_THEN_ARG  }
+ };
+ static const int verbcount = sizeof(verbs)/sizeof(verbitem);
+ /* 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 };
- /* Table of class bit maps for each POSIX class; up to three may be combined
+ /* Table of class bit maps for each POSIX class. Each class is formed from a
- to form the class. The table for [:blank:] is dynamically modified to remove
+ base map, with an optional addition or removal of another map. Then, for some
- the vertical space characters. */
+ classes, there is some additional tweaking: for [:blank:] the vertical space
+ characters are removed, and for [:alpha:] and [:alnum:] the underscore
+ character is removed. The triples in the table consist of the base map offset,
+ second map offset or -1 if no second map, and a non-negative value for map
+ addition or a negative value for map subtraction (if there are two maps). The
+ absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
+ remove vertical space characters, 2 => remove underscore. */
  static const int posix_class_maps[] = {
-   cbit_lower, cbit_upper, -1,             /* alpha */
+   cbit_word,  cbit_digit, -2,             /* alpha */
-   cbit_lower, -1,         -1,             /* lower */
+   cbit_lower, -1,          0,             /* lower */
-   cbit_upper, -1,         -1,             /* upper */
+   cbit_upper, -1,          0,             /* upper */
-   cbit_digit, cbit_lower, cbit_upper,     /* alnum */
+   cbit_word,  -1,          2,             /* alnum - word without underscore */
-   cbit_print, cbit_cntrl, -1,             /* ascii */
+   cbit_print, cbit_cntrl,  0,             /* ascii */
-   cbit_space, -1,         -1,             /* blank - a GNU extension */
+   cbit_space, -1,          1,             /* blank - a GNU extension */
-   cbit_cntrl, -1,         -1,             /* cntrl */
+   cbit_cntrl, -1,          0,             /* cntrl */
-   cbit_digit, -1,         -1,             /* digit */
+   cbit_digit, -1,          0,             /* digit */
-   cbit_graph, -1,         -1,             /* graph */
+   cbit_graph, -1,          0,             /* graph */
-   cbit_print, -1,         -1,             /* print */
+   cbit_print, -1,          0,             /* print */
-   cbit_punct, -1,         -1,             /* punct */
+   cbit_punct, -1,          0,             /* punct */
-   cbit_space, -1,         -1,             /* space */
+   cbit_space, -1,          0,             /* space */
-   cbit_word,  -1,         -1,             /* word - a Perl extension */
+   cbit_word,  -1,          0,             /* word - a Perl extension */
-   cbit_xdigit,-1,         -1              /* xdigit */
+   cbit_xdigit,-1,          0              /* xdigit */
  };
+ /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
+ substitutes must be in the order of the names, defined above, and there are
+ both positive and negative cases. NULL means no substitute. */
- /* The texts of compile-time error messages. These are "char *" because they
+ #ifdef SUPPORT_UCP
- are passed to the outside world. */
+ static const uschar *substitutes[] = {
+   (uschar *)"\\P{Nd}",    /* \D */
+   (uschar *)"\\p{Nd}",    /* \d */
+   (uschar *)"\\P{Xsp}",   /* \S */       /* NOTE: Xsp is Perl space */
+   (uschar *)"\\p{Xsp}",   /* \s */
+   (uschar *)"\\P{Xwd}",   /* \W */
+   (uschar *)"\\p{Xwd}"    /* \w */
+ };
+ static const uschar *posix_substitutes[] = {
+   (uschar *)"\\p{L}",     /* alpha */
+   (uschar *)"\\p{Ll}",    /* lower */
+   (uschar *)"\\p{Lu}",    /* upper */
+   (uschar *)"\\p{Xan}",   /* alnum */
+   NULL,                   /* ascii */
+   (uschar *)"\\h",        /* blank */
+   NULL,                   /* cntrl */
+   (uschar *)"\\p{Nd}",    /* digit */
+   NULL,                   /* graph */
+   NULL,                   /* print */
+   NULL,                   /* punct */
+   (uschar *)"\\p{Xps}",   /* space */    /* NOTE: Xps is POSIX space */
+   (uschar *)"\\p{Xwd}",   /* word */
+   NULL,                   /* xdigit */
+   /* Negated cases */
+   (uschar *)"\\P{L}",     /* ^alpha */
+   (uschar *)"\\P{Ll}",    /* ^lower */
+   (uschar *)"\\P{Lu}",    /* ^upper */
+   (uschar *)"\\P{Xan}",   /* ^alnum */
+   NULL,                   /* ^ascii */
+   (uschar *)"\\H",        /* ^blank */
+   NULL,                   /* ^cntrl */
+   (uschar *)"\\P{Nd}",    /* ^digit */
+   NULL,                   /* ^graph */
+   NULL,                   /* ^print */
+   NULL,                   /* ^punct */
+   (uschar *)"\\P{Xps}",   /* ^space */   /* NOTE: Xps is POSIX space */
+   (uschar *)"\\P{Xwd}",   /* ^word */
+   NULL                    /* ^xdigit */
+ };
+ #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
+ #endif
- static const char *error_texts[] = {
+ #define STRING(a)  # a
-   "no error",
+ #define XSTRING(s) STRING(s)
-   "\\ at end of pattern",
-   "\\c at end of pattern",
+ /* The texts of compile-time error messages. These are "char *" because they
-   "unrecognized character follows \\",
+ are passed to the outside world. Do not ever re-use any error number, because
-   "numbers out of order in {} quantifier",
+ they are documented. Always add a new error instead. Messages marked DEAD below
+ 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,
+ it is now one long string. We cannot use a table of offsets, because the
+ lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
+ simply count through to the one we want - this isn't a performance issue
+ because these strings are used only when there is a compilation error.
+ Each substring ends with \0 to insert a null character. This includes the final
+ substring, so that the whole string ends with \0\0, which can be detected when
+ counting through. */
+ static const char error_texts[] =
+   "no error\0"
+   "\\ at end of pattern\0"
+   "\\c at end of pattern\0"
+   "unrecognized character follows \\\0"
+   "numbers out of order in {} quantifier\0"
    /* 5 */
-   "number too big in {} quantifier",
+   "number too big in {} quantifier\0"
-   "missing terminating ] for character class",
+   "missing terminating ] for character class\0"
-   "invalid escape sequence in character class",
+   "invalid escape sequence in character class\0"
-   "range out of order in character class",
+   "range out of order in character class\0"
-   "nothing to repeat",
+   "nothing to repeat\0"
    /* 10 */
-   "operand of unlimited repeat could match the empty string",
+   "operand of unlimited repeat could match the empty string\0"  /** DEAD **/
-   "internal error: unexpected repeat",
+   "internal error: unexpected repeat\0"
-   "unrecognized character after (?",
+   "unrecognized character after (? or (?-\0"
-   "POSIX named classes are supported only within a class",
+   "POSIX named classes are supported only within a class\0"
-   "missing )",
+   "missing )\0"
    /* 15 */
-   "reference to non-existent subpattern",
+   "reference to non-existent subpattern\0"
-   "erroffset passed as NULL",
+   "erroffset passed as NULL\0"
-   "unknown option bit(s) set",
+   "unknown option bit(s) set\0"
-   "missing ) after comment",
+   "missing ) after comment\0"
-   "parentheses nested too deeply",
+   "parentheses nested too deeply\0"  /** DEAD **/
    /* 20 */
-   "regular expression too large",
+   "regular expression is too large\0"
-   "failed to get memory",
+   "failed to get memory\0"
-   "unmatched parentheses",
+   "unmatched parentheses\0"
-   "internal error: code overflow",
+   "internal error: code overflow\0"
-   "unrecognized character after (?<",
+   "unrecognized character after (?<\0"
    /* 25 */
-   "lookbehind assertion is not fixed length",
+   "lookbehind assertion is not fixed length\0"
-   "malformed number after (?(",
+   "malformed number or name after (?(\0"
-   "conditional group contains more than two branches",
+   "conditional group contains more than two branches\0"
-   "assertion expected after (?(",
+   "assertion expected after (?(\0"
-   "(?R or (?digits must be followed by )",
+   "(?R or (?[+-]digits must be followed by )\0"
    /* 30 */
-   "unknown POSIX class name",
+   "unknown POSIX class name\0"
-   "POSIX collating elements are not supported",
+   "POSIX collating elements are not supported\0"
-   "this version of PCRE is not compiled with PCRE_UTF8 support",
+   "this version of PCRE is not compiled with PCRE_UTF8 support\0"
-   "spare error",
+   "spare error\0"  /** DEAD **/
-   "character value in \\x{...} sequence is too large",
+   "character value in \\x{...} sequence is too large\0"
    /* 35 */
-   "invalid condition (?(0)",
+   "invalid condition (?(0)\0"
-   "\\C not allowed in lookbehind assertion",
+   "\\C not allowed in lookbehind assertion\0"
-   "PCRE does not support \\L, \\l, \\N, \\U, or \\u",
+   "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
-   "number after (?C is > 255",
+   "number after (?C is > 255\0"
-   "closing ) for (?C expected",
+   "closing ) for (?C expected\0"
    /* 40 */
-   "recursive call could loop indefinitely",
+   "recursive call could loop indefinitely\0"
-   "unrecognized character after (?P",
+   "unrecognized character after (?P\0"
-   "syntax error after (?P",
+   "syntax error in subpattern name (missing terminator)\0"
-   "two named groups have the same name",
+   "two named subpatterns have the same name\0"
-   "invalid UTF-8 string",
+   "invalid UTF-8 string\0"
    /* 45 */
-   "support for \\P, \\p, and \\X has not been compiled",
+   "support for \\P, \\p, and \\X has not been compiled\0"
-   "malformed \\P or \\p sequence",
+   "malformed \\P or \\p sequence\0"
-   "unknown property name after \\P or \\p"
+   "unknown property name after \\P or \\p\0"
- };
+   "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
+   "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
+   /* 50 */
+   "repeated subpattern is too long\0"    /** DEAD **/
+   "octal value is greater than \\377 (not in UTF-8 mode)\0"
+   "internal error: overran compiling workspace\0"
+   "internal error: previously-checked referenced subpattern not found\0"
+   "DEFINE group contains more than one branch\0"
+   /* 55 */
+   "repeating a DEFINE group is not allowed\0"  /** DEAD **/
+   "inconsistent NEWLINE options\0"
+   "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
+   "a numbered reference must not be zero\0"
+   "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\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\0"
+   /* 65 */
+   "different names for subpatterns of the same number are not allowed\0"
+   "(*MARK) must have an argument\0"
+   "this version of PCRE is not compiled with PCRE_UCP support\0"
+   "\\c must be followed by an ASCII character\0"
+   "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
+   ;
  /* Table to identify digits and hex digits. This is used when compiling
  patterns. Note that the tables in chartables are dependent on the locale, and
-Line 229 
 For convenience, we use the same bit def
+Line 428 
 For convenience, we use the same bit def
  Then we can use ctype_digit and ctype_xdigit in the code. */
- #if !EBCDIC    /* This is the "normal" case, for ASCII systems */
+ #ifndef EBCDIC
+ /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
+ UTF-8 mode. */
  static const unsigned char digitab[] =
    {
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7 */
-Line 265 
 static const unsigned char digitab[] =
+Line 468 
 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 279 
 static const unsigned char digitab[] =
+Line 485 
 static const unsigned char digitab[] =
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 40 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  72- |     */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 50 */
-x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  88- �     */
+x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  88- 95    */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 60 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ?     */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
-Line 313 
 static const unsigned char ebcdic_charta
+Line 519 
 static const unsigned char ebcdic_charta
 x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 */
 x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /*  72- |  */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 */
-x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /*  88- �  */
+x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /*  88- 95 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 */
 x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ?  */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
-Line 340 
 static const unsigned char ebcdic_charta
+Line 546 
 static const unsigned char ebcdic_charta
  /* Definition to allow mutual recursion */
  static BOOL
-   compile_regex(int, int, int *, uschar **, const uschar **, int *, BOOL, int,
+   compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int,
-     int *, int *, branch_chain *, compile_data *);
+     int *, int *, branch_chain *, compile_data *, int *);
+ /*************************************************
+ *            Find an error text                  *
+ *************************************************/
+ /* The error texts are now all in one long string, to save on relocations. As
+ some of the text is of unknown length, we can't use a table of offsets.
+ Instead, just count through the strings. This is not a performance issue
+ because it happens only when there has been a compilation error.
+ Argument:   the error number
+ Returns:    pointer to the error string
+ */
+ static const char *
+ find_error_text(int n)
+ {
+ const char *s = error_texts;
+ for (; n > 0; n--)
+   {
+   while (*s++ != 0) {};
+   if (*s == 0) return "Error text not found (please report)";
+   }
+ return s;
+ }
+ /*************************************************
+ *            Check for counted repeat            *
+ *************************************************/
+ /* This function is called when a '{' is encountered in a place where it might
+ start a quantifier. It looks ahead to see if it really is a quantifier or not.
+ It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
+ where the ddds are digits.
+ Arguments:
+   p         pointer to the first char after '{'
+ Returns:    TRUE or FALSE
+ */
+ static BOOL
+ is_counted_repeat(const uschar *p)
+ {
+ if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
+ while ((digitab[*p] & ctype_digit) != 0) p++;
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
+ if (*p++ != CHAR_COMMA) return FALSE;
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
+ if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
+ while ((digitab[*p] & ctype_digit) != 0) p++;
+ return (*p == CHAR_RIGHT_CURLY_BRACKET);
+ }
-Line 351 
 static BOOL
+Line 616 
 static BOOL
  /* This function is called when a \ has been encountered. It either returns a
  positive value for a simple escape such as \n, or a negative value which
- encodes one of the more complicated things such as \d. When UTF-8 is enabled,
+ encodes one of the more complicated things such as \d. A backreference to group
- a positive value greater than 255 may be returned. On entry, ptr is pointing at
+ n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
- the \. On exit, it is on the final character of the escape sequence.
+ UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
+ ptr is pointing at the \. On exit, it is on the final character of the escape
+ sequence.
  Arguments:
    ptrptr         points to the pattern position pointer
-Line 364 
 Arguments:
+Line 631 
 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
  check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
    int options, BOOL isclass)
  {
- const uschar *ptr = *ptrptr;
+ BOOL utf8 = (options & PCRE_UTF8) != 0;
+ const uschar *ptr = *ptrptr + 1;
  int c, i;
+ GETCHARINCTEST(c, ptr);           /* Get character value, increment pointer */
+ ptr--;                            /* Set pointer back to the last byte */
  /* If backslash is at the end of the pattern, it's an error. */
- c = *(++ptr);
  if (c == 0) *errorcodeptr = ERR1;
- /* Non-alphamerics are literals. For digits or letters, do an initial lookup in
+ /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
- a table. A non-zero result is something that can be returned immediately.
+ in a table. A non-zero result is something that can be returned immediately.
  Otherwise further processing may be required. */
- #if !EBCDIC    /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII/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           /* 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 397 
 else if ((i = escapes[c - 0x48]) != 0)
+Line 667 
 else if ((i = escapes[c - 0x48]) != 0)
  else
    {
    const uschar *oldptr;
+   BOOL braced, negated;
    switch (c)
      {
      /* A number of Perl escapes are not handled by PCRE. We give an explicit
      error. */
-     case 'l':
+     case CHAR_l:
-     case 'L':
+     case CHAR_L:
-     case 'N':
+     case CHAR_u:
-     case 'u':
+     case CHAR_U:
-     case 'U':
      *errorcodeptr = ERR37;
      break;
+     /* In a character class, \g is just a literal "g". Outside a character
+     class, \g must be followed by one of a number of specific things:
+     (1) A number, either plain or braced. If positive, it is an absolute
+     backreference. If negative, it is a relative backreference. This is a Perl
+.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 CHAR_g:
+     if (isclass) break;
+     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] == CHAR_MINUS)
+       {
+       negated = TRUE;
+       ptr++;
+       }
+     else negated = FALSE;
+     c = 0;
+     while ((digitab[ptr[1]] & ctype_digit) != 0)
+       c = c * 10 + *(++ptr) - CHAR_0;
+     if (c < 0)   /* Integer overflow */
+       {
+       *errorcodeptr = ERR61;
+       break;
+       }
+     if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
+       {
+       *errorcodeptr = ERR57;
+       break;
+       }
+     if (c == 0)
+       {
+       *errorcodeptr = ERR58;
+       break;
+       }
+     if (negated)
+       {
+       if (c > bracount)
+         {
+         *errorcodeptr = ERR15;
+         break;
+         }
+       c = bracount - (c - 1);
+       }
+     c = -(ESC_REF + c);
+     break;
      /* The handling of escape sequences consisting of a string of digits
      starting with one that is not zero is not straightforward. By experiment,
      the way Perl works seems to be as follows:
-Line 422 
 else
+Line 777 
 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 443 
 else
+Line 803 
 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 451 
 else
+Line 811 
 else
        }
      /* \0 always starts an octal number, but we may drop through to here with a
-     larger first octal digit. */
+     larger first octal digit. The original code used just to take the least
+     significant 8 bits of octal numbers (I think this is what early Perls used
-     case '0':
+     to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
-     c -= '0';
+     than 3 octal digits. */
-     while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7')
-         c = c * 8 + *(++ptr) - '0';
+     case CHAR_0:
-     c &= 255;     /* Take least significant 8 bits */
+     c -= CHAR_0;
+     while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
+         c = c * 8 + *(++ptr) - CHAR_0;
+     if (!utf8 && c > 255) *errorcodeptr = ERR51;
      break;
-     /* \x is complicated when UTF-8 is enabled. \x{ddd} is a character number
+     /* \x is complicated. \x{ddd} is a character number which can be greater
-     which can be greater than 0xff, but only if the ddd are hex digits. */
+     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:
- #ifdef SUPPORT_UTF8
+     if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
-     if (ptr[1] == '{' && (options & PCRE_UTF8) != 0)
        {
        const uschar *pt = ptr + 2;
-       register int count = 0;
+       int count = 0;
        c = 0;
        while ((digitab[*pt] & ctype_xdigit) != 0)
          {
-         int cc = *pt++;
+         register int cc = *pt++;
+         if (c == 0 && cc == CHAR_0) continue;     /* Leading zeroes */
          count++;
- #if !EBCDIC    /* ASCII coding */
-         if (cc >= 'a') cc -= 32;               /* Convert to upper case */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
-         c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));
+         if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
- #else          /* EBCDIC coding */
+         c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
-         if (cc >= 'a' && cc <= 'z') cc += 64;  /* Convert to upper case */
+ #else           /* EBCDIC coding */
-         c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));
+         if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
+         c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
  #endif
          }
-       if (*pt == '}')
+       if (*pt == CHAR_RIGHT_CURLY_BRACKET)
          {
-         if (c < 0 || count > 8) *errorcodeptr = ERR34;
+         if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
          ptr = pt;
          break;
          }
        /* If the sequence of hex digits does not end with '}', then we don't
        recognize this construct; fall through to the normal \x handling. */
        }
- #endif
-     /* Read just a single hex char */
+     /* Read just a single-byte hex-defined char */
      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 */
- #if !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 */
+ #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;
-     /* Other special escapes not starting with a digit are straightforward */
+     /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
+     An error is given if the byte following \c is not an ASCII character. This
+     coding is ASCII-specific, but then the whole concept of \cx is
+     ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
-     case 'c':
+     case CHAR_c:
      c = *(++ptr);
      if (c == 0)
        {
        *errorcodeptr = ERR2;
-       return 0;
+       break;
        }
+ #ifndef EBCDIC    /* ASCII/UTF-8 coding */
-     /* A letter is upper-cased; then the 0x40 bit is flipped. This coding
+     if (c > 127)  /* Excludes all non-ASCII in either mode */
-     is ASCII-specific, but then the whole concept of \cx is ASCII-specific.
+       {
-     (However, an EBCDIC equivalent has now been added.) */
+       *errorcodeptr = ERR68;
+       break;
- #if !EBCDIC    /* ASCII coding */
+       }
-     if (c >= 'a' && c <= 'z') c -= 32;
+     if (c >= CHAR_a && c <= CHAR_z) c -= 32;
      c ^= 0x40;
- #else          /* EBCDIC coding */
+ #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 550 
 else
+Line 920 
 else
      }
    }
+ /* Perl supports \N{name} for character names, as well as plain \N for "not
+ newline". PCRE does not support \N{name}. However, it does support
+ quantification such as \N{2,3}. */
+ if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
+      !is_counted_repeat(ptr+2))
+   *errorcodeptr = ERR37;
+ /* If PCRE_UCP is set, we change the values for \d etc. */
+ if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
+   c -= (ESC_DU - ESC_D);
+ /* Set the pointer to the final character before returning. */
  *ptrptr = ptr;
  return c;
  }
-Line 569 
 escape sequence.
+Line 954 
 escape sequence.
  Argument:
    ptrptr         points to the pattern position pointer
    negptr         points to a boolean that is set TRUE for negation else FALSE
+   dptr           points to an int that is set to the detailed property value
    errorcodeptr   points to the error code variable
- Returns:     value from ucp_type_table, or -1 for an invalid type
+ Returns:         type value from ucp_type_table, or -1 for an invalid type
  */
  static int
- get_ucp(const uschar **ptrptr, BOOL *negptr, int *errorcodeptr)
+ get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
  {
  int c, i, bot, top;
  const uschar *ptr = *ptrptr;
- char name[4];
+ char name[32];
  c = *(++ptr);
  if (c == 0) goto ERROR_RETURN;
  *negptr = FALSE;
- /* \P or \p can be followed by a one- or two-character name in {}, optionally
+ /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
- preceded by ^ for negation. */
+ 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 <= 2; 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 !='}')   /* Try to distinguish error cases */
+   if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
-     {
-     while (*(++ptr) != 0 && *ptr != '}');
-     if (*ptr == '}') goto UNKNOWN_RETURN; else goto ERROR_RETURN;
-     }
    name[i] = 0;
    }
-Line 628 
 top = _pcre_utt_size;
+Line 1010 
 top = _pcre_utt_size;
  while (bot < top)
    {
-   i = (bot + top)/2;
+   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) return _pcre_utt[i].value;
+   if (c == 0)
+     {
+     *dptr = _pcre_utt[i].value;
+     return _pcre_utt[i].type;
+     }
    if (c > 0) bot = i + 1; else top = i;
    }
- UNKNOWN_RETURN:
  *errorcodeptr = ERR47;
  *ptrptr = ptr;
  return -1;
-Line 650 
 return -1;
+Line 1035 
 return -1;
  /*************************************************
- *            Check for counted repeat            *
- *************************************************/
- /* This function is called when a '{' is encountered in a place where it might
- start a quantifier. It looks ahead to see if it really is a quantifier or not.
- It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
- where the ddds are digits.
- Arguments:
-   p         pointer to the first char after '{'
- Returns:    TRUE or FALSE
- */
- static BOOL
- is_counted_repeat(const uschar *p)
- {
- if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
- while ((digitab[*p] & ctype_digit) != 0) p++;
- if (*p == '}') return TRUE;
- if (*p++ != ',') return FALSE;
- if (*p == '}') return TRUE;
- if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
- while ((digitab[*p] & ctype_digit) != 0) p++;
- return (*p == '}');
- }
- /*************************************************
  *         Read repeat counts                     *
  *************************************************/
-Line 710 
 int max = -1;
+Line 1062 
 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 720 
 if (min < 0 || min > 65535)
+Line 1072 
 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 750 
 return p;
+Line 1102 
 return p;
  /*************************************************
- *      Find first significant op code            *
+ *  Subroutine for finding forward reference      *
  *************************************************/
- /* This is called by several functions that scan a compiled expression looking
+ /* This recursive function is called only from find_parens() below. The
- for a fixed first character, or an anchoring op code etc. It skips over things
+ top-level call starts at the beginning of the pattern. All other calls must
- that do not influence this. For some calls, a change of option is important.
+ start at a parenthesis. It scans along a pattern's text looking for capturing
- For some calls, it makes sense to skip negative forward and all backward
+ subpatterns, and counting them. If it finds a named pattern that matches the
- assertions, and also the \b assertion; for others it does not.
+ name it is given, it returns its number. Alternatively, if the name is NULL, it
+ returns when it reaches a given numbered subpattern. Recursion is used to keep
+ track of subpatterns that reset the capturing group numbers - the (?| feature.
+ This function was originally called only from the second pass, in which we know
+ that if (?< or (?' or (?P< is encountered, the name will be correctly
+ terminated because that is checked in the first pass. There is now one call to
+ this function in the first pass, to check for a recursive back reference by
+ name (so that we can make the whole group atomic). In this case, we need check
+ only up to the current position in the pattern, and that is still OK because
+ and previous occurrences will have been checked. To make this work, the test
+ for "end of pattern" is a check against cd->end_pattern in the main loop,
+ instead of looking for a binary zero. This means that the special first-pass
+ call can adjust cd->end_pattern temporarily. (Checks for binary zero while
+ processing items within the loop are OK, because afterwards the main loop will
+ terminate.)
  Arguments:
-   code         pointer to the start of the group
+   ptrptr       address of the current character pointer (updated)
-   options      pointer to external options
+   cd           compile background data
-   optbit       the option bit whose changing is significant, or
+   name         name to seek, or NULL if seeking a numbered subpattern
-                  zero if none are
+   lorn         name length, or subpattern number if name is NULL
-   skipassert   TRUE if certain assertions are to be skipped
+   xmode        TRUE if we are in /x mode
+   utf8         TRUE if we are in UTF-8 mode
+   count        pointer to the current capturing subpattern number (updated)
- Returns:       pointer to the first significant opcode
+ Returns:       the number of the named subpattern, or -1 if not found
  */
- static const uschar*
+ static int
- first_significant_code(const uschar *code, int *options, int optbit,
+ find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
-   BOOL skipassert)
+   BOOL xmode, BOOL utf8, int *count)
  {
- for (;;)
+ uschar *ptr = *ptrptr;
-   {
+ int start_count = *count;
-   switch ((int)*code)
+ int hwm_count = start_count;
-     {
+ BOOL dup_parens = FALSE;
-     case OP_OPT:
-     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
-       *options = (int)code[1];
-     code += 2;
-     break;
-     case OP_ASSERT_NOT:
+ /* If the first character is a parenthesis, check on the type of group we are
-     case OP_ASSERTBACK:
+ dealing with. The very first call may not start with a parenthesis. */
-     case OP_ASSERTBACK_NOT:
-     if (!skipassert) return code;
-     do code += GET(code, 1); while (*code == OP_ALT);
-     code += _pcre_OP_lengths[*code];
-     break;
-     case OP_WORD_BOUNDARY:
+ if (ptr[0] == CHAR_LEFT_PARENTHESIS)
-     case OP_NOT_WORD_BOUNDARY:
+   {
-     if (!skipassert) return code;
+   /* Handle specials such as (*SKIP) or (*UTF8) etc. */
-     /* Fall through */
-     case OP_CALLOUT:
+   if (ptr[1] == CHAR_ASTERISK) ptr += 2;
-     case OP_CREF:
-     case OP_BRANUMBER:
-     code += _pcre_OP_lengths[*code];
-     break;
-     default:
+   /* Handle a normal, unnamed capturing parenthesis. */
-     return code;
+   else if (ptr[1] != CHAR_QUESTION_MARK)
+     {
+     *count += 1;
+     if (name == NULL && *count == lorn) return *count;
+     ptr++;
      }
-   }
- /* Control never reaches here */
- }
+   /* All cases now have (? at the start. Remember when we are in a group
+   where the parenthesis numbers are duplicated. */
+   else if (ptr[2] == CHAR_VERTICAL_LINE)
+     {
+     ptr += 3;
+     dup_parens = TRUE;
+     }
+   /* Handle comments; all characters are allowed until a ket is reached. */
- /*************************************************
+   else if (ptr[2] == CHAR_NUMBER_SIGN)
- *        Find the fixed length of a pattern      *
+     {
- *************************************************/
+     for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
+     goto FAIL_EXIT;
+     }
- /* Scan a pattern and compute the fixed length of subject that will match it,
+   /* Handle a condition. If it is an assertion, just carry on so that it
- if the length is fixed. This is needed for dealing with backward assertions.
+   is processed as normal. If not, skip to the closing parenthesis of the
- In UTF8 mode, the result is in characters rather than bytes.
+   condition (there can't be any nested parens). */
- Arguments:
+   else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
-   code     points to the start of the pattern (the bracket)
+     {
-   options  the compiling options
+     ptr += 2;
+     if (ptr[1] != CHAR_QUESTION_MARK)
+       {
+       while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
+       if (*ptr != 0) ptr++;
+       }
+     }
- Returns:   the fixed length, or -1 if there is no fixed length,
+   /* Start with (? but not a condition. */
-              or -2 if \C was encountered
- */
- static int
+   else
- find_fixedlength(uschar *code, int options)
+     {
- {
+     ptr += 2;
- int length = -1;
+     if (*ptr == CHAR_P) ptr++;                      /* Allow optional P */
- register int branchlength = 0;
+     /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
- register uschar *cc = code + 1 + LINK_SIZE;
- /* Scan along the opcodes for this branch. If we get to the end of the
+     if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
- branch, check the length against that of the other branches. */
+         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++;
+       }
+     }
+   }
- for (;;)
+ /* Past any initial parenthesis handling, scan for parentheses or vertical
+ bars. Stop if we get to cd->end_pattern. Note that this is important for the
+ first-pass call when this value is temporarily adjusted to stop at the current
+ position. So DO NOT change this to a test for binary zero. */
+ for (; ptr < cd->end_pattern; ptr++)
    {
-   int d;
+   /* Skip over backslashed characters and also entire \Q...\E */
-   register int op = *cc;
-   if (op >= OP_BRA) op = OP_BRA;
-   switch (op)
+   if (*ptr == CHAR_BACKSLASH)
      {
-     case OP_BRA:
+     if (*(++ptr) == 0) goto FAIL_EXIT;
-     case OP_ONCE:
+     if (*ptr == CHAR_Q) for (;;)
-     case OP_COND:
+       {
-     d = find_fixedlength(cc, options);
+       while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
-     if (d < 0) return d;
+       if (*ptr == 0) goto FAIL_EXIT;
-     branchlength += d;
+       if (*(++ptr) == CHAR_E) break;
-     do cc += GET(cc, 1); while (*cc == OP_ALT);
+       }
-     cc += 1 + LINK_SIZE;
+     continue;
-     break;
+     }
-     /* Reached end of a branch; if it's a ket it is the end of a nested
+   /* 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 == CHAR_LEFT_SQUARE_BRACKET)
+     {
+     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 == 0) return -1;
+       if (*ptr == CHAR_BACKSLASH)
+         {
+         if (*(++ptr) == 0) goto FAIL_EXIT;
+         if (*ptr == CHAR_Q) for (;;)
+           {
+           while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
+           if (*ptr == 0) goto FAIL_EXIT;
+           if (*(++ptr) == CHAR_E) break;
+           }
+         continue;
+         }
+       }
+     continue;
+     }
+   /* Skip comments in /x mode */
+   if (xmode && *ptr == CHAR_NUMBER_SIGN)
+     {
+     ptr++;
+     while (*ptr != 0)
+       {
+       if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
+       ptr++;
+ #ifdef SUPPORT_UTF8
+       if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
+ #endif
+       }
+     if (*ptr == 0) goto FAIL_EXIT;
+     continue;
+     }
+   /* Check for the special metacharacters */
+   if (*ptr == CHAR_LEFT_PARENTHESIS)
+     {
+     int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
+     if (rc > 0) return rc;
+     if (*ptr == 0) goto FAIL_EXIT;
+     }
+   else if (*ptr == CHAR_RIGHT_PARENTHESIS)
+     {
+     if (dup_parens && *count < hwm_count) *count = hwm_count;
+     goto FAIL_EXIT;
+     }
+   else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
+     {
+     if (*count > hwm_count) hwm_count = *count;
+     *count = start_count;
+     }
+   }
+ FAIL_EXIT:
+ *ptrptr = ptr;
+ return -1;
+ }
+ /*************************************************
+ *       Find forward referenced subpattern       *
+ *************************************************/
+ /* This function scans along a pattern's text looking for capturing
+ subpatterns, and counting them. If it finds a named pattern that matches the
+ name it is given, it returns its number. Alternatively, if the name is NULL, it
+ returns when it reaches a given numbered subpattern. This is used for forward
+ references to subpatterns. We 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
+   utf8         TRUE if we are in UTF-8 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,
+   BOOL utf8)
+ {
+ 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, utf8, &count);
+   if (rc > 0 || *ptr++ == 0) break;
+   }
+ return rc;
+ }
+ /*************************************************
+ *      Find first significant op code            *
+ *************************************************/
+ /* This is called by several functions that scan a compiled expression looking
+ for a fixed first character, or an anchoring op code etc. It skips over things
+ that do not influence this. For some calls, it makes sense to skip negative
+ forward and all backward assertions, and also the \b assertion; for others it
+ does not.
+ Arguments:
+   code         pointer to the start of the group
+   skipassert   TRUE if certain assertions are to be skipped
+ Returns:       pointer to the first significant opcode
+ */
+ static const uschar*
+ first_significant_code(const uschar *code, BOOL skipassert)
+ {
+ for (;;)
+   {
+   switch ((int)*code)
+     {
+     case OP_ASSERT_NOT:
+     case OP_ASSERTBACK:
+     case OP_ASSERTBACK_NOT:
+     if (!skipassert) return code;
+     do code += GET(code, 1); while (*code == OP_ALT);
+     code += _pcre_OP_lengths[*code];
+     break;
+     case OP_WORD_BOUNDARY:
+     case OP_NOT_WORD_BOUNDARY:
+     if (!skipassert) return code;
+     /* Fall through */
+     case OP_CALLOUT:
+     case OP_CREF:
+     case OP_NCREF:
+     case OP_RREF:
+     case OP_NRREF:
+     case OP_DEF:
+     code += _pcre_OP_lengths[*code];
+     break;
+     default:
+     return code;
+     }
+   }
+ /* Control never reaches here */
+ }
+ /*************************************************
+ *        Find the fixed length of a branch       *
+ *************************************************/
+ /* 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. 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)
+   utf8     TRUE in UTF-8 mode
+   atend    TRUE if called when the pattern is complete
+   cd       the "compile data" structure
+ Returns:   the fixed length,
+              or -1 if there is no fixed length,
+              or -2 if \C was encountered
+              or -3 if an OP_RECURSE item was encountered and atend is FALSE
+ */
+ static int
+ find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
+ {
+ int length = -1;
+ register int branchlength = 0;
+ register uschar *cc = code + 1 + LINK_SIZE;
+ /* Scan along the opcodes for this branch. If we get to the end of the
+ branch, check the length against that of the other branches. */
+ for (;;)
+   {
+   int d;
+   uschar *ce, *cs;
+   register int op = *cc;
+   switch (op)
+     {
+     /* We only need to continue for OP_CBRA (normal capturing bracket) and
+     OP_BRA (normal non-capturing bracket) because the other variants of these
+     opcodes are all concerned with unlimited repeated groups, which of course
+     are not of fixed length. They will cause a -1 response from the default
+     case of this switch. */
+     case OP_CBRA:
+     case OP_BRA:
+     case OP_ONCE:
+     case OP_COND:
+     d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
+     if (d < 0) return d;
+     branchlength += d;
+     do cc += GET(cc, 1); while (*cc == OP_ALT);
+     cc += 1 + LINK_SIZE;
+     break;
+     /* Reached end of a branch; if it's a ket it is the end of a nested
      call. If it's ALT it is an alternation in a nested call. If it is
-     END it's the end of the outer call. All can be handled by the same code. */
+     END it's the end of the outer call. All can be handled by the same code.
+     Note that we must not include the OP_KETRxxx opcodes here, because they
+     all imply an unlimited repeat. */
      case OP_ALT:
      case OP_KET:
-     case OP_KETRMAX:
-     case OP_KETRMIN:
      case OP_END:
      if (length < 0) length = branchlength;
        else if (length != branchlength) return -1;
-Line 873 
 for (;;)
+Line 1530 
 for (;;)
      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, utf8, atend, cd);
+     if (d < 0) return d;
+     branchlength += d;
+     cc += 1 + LINK_SIZE;
+     break;
      /* Skip over assertive subpatterns */
      case OP_ASSERT:
-Line 885 
 for (;;)
+Line 1557 
 for (;;)
      /* Skip over things that don't match chars */
      case OP_REVERSE:
-     case OP_BRANUMBER:
      case OP_CREF:
-     case OP_OPT:
+     case OP_NCREF:
+     case OP_RREF:
+     case OP_NRREF:
+     case OP_DEF:
      case OP_CALLOUT:
      case OP_SOD:
      case OP_SOM:
+     case OP_SET_SOM:
      case OP_EOD:
      case OP_EODN:
      case OP_CIRC:
+     case OP_CIRCM:
      case OP_DOLL:
+     case OP_DOLLM:
      case OP_NOT_WORD_BOUNDARY:
      case OP_WORD_BOUNDARY:
      cc += _pcre_OP_lengths[*cc];
-Line 903 
 for (;;)
+Line 1580 
 for (;;)
      /* Handle literal characters */
      case OP_CHAR:
-     case OP_CHARNC:
+     case OP_CHARI:
+     case OP_NOT:
+     case OP_NOTI:
      branchlength++;
      cc += 2;
  #ifdef SUPPORT_UTF8
-     if ((options & PCRE_UTF8) != 0)
+     if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
-       {
-       while ((*cc & 0xc0) == 0x80) cc++;
-       }
  #endif
      break;
-Line 921 
 for (;;)
+Line 1597 
 for (;;)
      branchlength += GET2(cc,1);
      cc += 4;
  #ifdef SUPPORT_UTF8
-     if ((options & PCRE_UTF8) != 0)
+     if (utf8 && 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 937 
 for (;;)
+Line 1611 
 for (;;)
      case OP_PROP:
      case OP_NOTPROP:
-     cc++;
+     cc += 2;
      /* Fall through */
      case OP_NOT_DIGIT:
-Line 947 
 for (;;)
+Line 1621 
 for (;;)
      case OP_NOT_WORDCHAR:
      case OP_WORDCHAR:
      case OP_ANY:
+     case OP_ALLANY:
      branchlength++;
      cc++;
      break;
-Line 1001 
 for (;;)
+Line 1676 
 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. The function is global in the C sense
+ so that it can be called from pcre_study() when finding the minimum matching
+ length.
  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
  */
- static const uschar *
+ const uschar *
- find_bracket(const uschar *code, BOOL utf8, int number)
+ _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
  {
- #ifndef SUPPORT_UTF8
- utf8 = utf8;               /* Stop pedantic compilers complaining */
- #endif
  for (;;)
    {
    register int c = *code;
    if (c == OP_END) return NULL;
-   else if (c > OP_BRA)
+   /* XCLASS is used for classes that cannot be represented just by a bit
+   map. This includes negated single high-valued characters. The length in
+   the table is zero; the actual length is stored in the compiled code. */
+   if (c == OP_XCLASS) code += GET(code, 1);
+   /* Handle recursion */
+   else if (c == OP_REVERSE)
+     {
+     if (number < 0) return (uschar *)code;
+     code += _pcre_OP_lengths[c];
+     }
+   /* Handle capturing bracket */
+   else if (c == OP_CBRA || c == OP_SCBRA ||
+            c == OP_CBRAPOS || c == OP_SCBRAPOS)
      {
-     int n = c - OP_BRA;
+     int n = GET2(code, 1+LINK_SIZE);
-     if (n > EXTRACT_BASIC_MAX) n = GET2(code, 2+LINK_SIZE);
      if (n == number) return (uschar *)code;
-     code += _pcre_OP_lengths[OP_BRA];
+     code += _pcre_OP_lengths[c];
      }
+   /* Otherwise, we can get the item's length from the table, except that for
+   repeated character types, we have to test for \p and \P, which have an extra
+   two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
+   must add in its length. */
    else
      {
-     code += _pcre_OP_lengths[c];
+     switch(c)
+       {
+       case OP_TYPESTAR:
+       case OP_TYPEMINSTAR:
+       case OP_TYPEPLUS:
+       case OP_TYPEMINPLUS:
+       case OP_TYPEQUERY:
+       case OP_TYPEMINQUERY:
+       case OP_TYPEPOSSTAR:
+       case OP_TYPEPOSPLUS:
+       case OP_TYPEPOSQUERY:
+       if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+       break;
- #ifdef SUPPORT_UTF8
+       case OP_TYPEUPTO:
+       case OP_TYPEMINUPTO:
+       case OP_TYPEEXACT:
+       case OP_TYPEPOSUPTO:
+       if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
+       break;
-     /* In UTF-8 mode, opcodes that are followed by a character may be followed
+       case OP_MARK:
-     by a multi-byte character. The length in the table is a minimum, so we have
+       case OP_PRUNE_ARG:
-     to scan along to skip the extra bytes. All opcodes are less than 128, so we
+       case OP_SKIP_ARG:
-     can use relatively efficient code. */
+       code += code[1];
+       break;
+       case OP_THEN_ARG:
+       code += code[1+LINK_SIZE];
+       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)
        {
        case OP_CHAR:
-       case OP_CHARNC:
+       case OP_CHARI:
        case OP_EXACT:
+       case OP_EXACTI:
        case OP_UPTO:
+       case OP_UPTOI:
        case OP_MINUPTO:
+       case OP_MINUPTOI:
+       case OP_POSUPTO:
+       case OP_POSUPTOI:
        case OP_STAR:
+       case OP_STARI:
        case OP_MINSTAR:
+       case OP_MINSTARI:
+       case OP_POSSTAR:
+       case OP_POSSTARI:
        case OP_PLUS:
+       case OP_PLUSI:
        case OP_MINPLUS:
+       case OP_MINPLUSI:
+       case OP_POSPLUS:
+       case OP_POSPLUSI:
        case OP_QUERY:
+       case OP_QUERYI:
        case OP_MINQUERY:
-       while ((*code & 0xc0) == 0x80) code++;
+       case OP_MINQUERYI:
-       break;
+       case OP_POSQUERY:
+       case OP_POSQUERYI:
-       /* XCLASS is used for classes that cannot be represented just by a bit
+       if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
-       map. This includes negated single high-valued characters. The length in
-       the table is zero; the actual length is stored in the compiled code. */
-       case OP_XCLASS:
-       code += GET(code, 1) + 1;
        break;
        }
+ #else
+     (void)(utf8);  /* Keep compiler happy by referencing function argument */
  #endif
      }
    }
-Line 1092 
 Returns:      pointer to the opcode for
+Line 1833 
 Returns:      pointer to the opcode for
  static const uschar *
  find_recurse(const uschar *code, BOOL utf8)
  {
- #ifndef SUPPORT_UTF8
- utf8 = utf8;               /* Stop pedantic compilers complaining */
- #endif
  for (;;)
    {
    register int c = *code;
    if (c == OP_END) return NULL;
-   else if (c == OP_RECURSE) return code;
+   if (c == OP_RECURSE) return code;
-   else if (c > OP_BRA)
-     {
+   /* XCLASS is used for classes that cannot be represented just by a bit
-     code += _pcre_OP_lengths[OP_BRA];
+   map. This includes negated single high-valued characters. The length in
-     }
+   the table is zero; the actual length is stored in the compiled code. */
+   if (c == OP_XCLASS) code += GET(code, 1);
+   /* Otherwise, we can get the item's length from the table, except that for
+   repeated character types, we have to test for \p and \P, which have an extra
+   two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
+   must add in its length. */
    else
      {
-     code += _pcre_OP_lengths[c];
+     switch(c)
+       {
+       case OP_TYPESTAR:
+       case OP_TYPEMINSTAR:
+       case OP_TYPEPLUS:
+       case OP_TYPEMINPLUS:
+       case OP_TYPEQUERY:
+       case OP_TYPEMINQUERY:
+       case OP_TYPEPOSSTAR:
+       case OP_TYPEPOSPLUS:
+       case OP_TYPEPOSQUERY:
+       if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+       break;
- #ifdef SUPPORT_UTF8
+       case OP_TYPEPOSUPTO:
+       case OP_TYPEUPTO:
+       case OP_TYPEMINUPTO:
+       case OP_TYPEEXACT:
+       if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
+       break;
+       case OP_MARK:
+       case OP_PRUNE_ARG:
+       case OP_SKIP_ARG:
+       code += code[1];
+       break;
+       case OP_THEN_ARG:
+       code += code[1+LINK_SIZE];
+       break;
+       }
+     /* Add in the fixed length from the table */
+     code += _pcre_OP_lengths[c];
      /* In UTF-8 mode, opcodes that are followed by a character may be followed
      by a multi-byte character. The length in the table is a minimum, so we have
-     to scan along to skip the extra bytes. All opcodes are less than 128, so we
+     to arrange to skip the extra bytes. */
-     can use relatively efficient code. */
+ #ifdef SUPPORT_UTF8
      if (utf8) switch(c)
        {
        case OP_CHAR:
-       case OP_CHARNC:
+       case OP_CHARI:
        case OP_EXACT:
+       case OP_EXACTI:
        case OP_UPTO:
+       case OP_UPTOI:
        case OP_MINUPTO:
+       case OP_MINUPTOI:
+       case OP_POSUPTO:
+       case OP_POSUPTOI:
        case OP_STAR:
+       case OP_STARI:
        case OP_MINSTAR:
+       case OP_MINSTARI:
+       case OP_POSSTAR:
+       case OP_POSSTARI:
        case OP_PLUS:
+       case OP_PLUSI:
        case OP_MINPLUS:
+       case OP_MINPLUSI:
+       case OP_POSPLUS:
+       case OP_POSPLUSI:
        case OP_QUERY:
+       case OP_QUERYI:
        case OP_MINQUERY:
-       while ((*code & 0xc0) == 0x80) code++;
+       case OP_MINQUERYI:
-       break;
+       case OP_POSQUERY:
+       case OP_POSQUERYI:
-       /* XCLASS is used for classes that cannot be represented just by a bit
+       if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
-       map. This includes negated single high-valued characters. The length in
-       the table is zero; the actual length is stored in the compiled code. */
-       case OP_XCLASS:
-       code += GET(code, 1) + 1;
        break;
        }
+ #else
+     (void)(utf8);  /* Keep compiler happy by referencing function argument */
  #endif
      }
    }
-Line 1152 
 for (;;)
+Line 1940 
 for (;;)
  *************************************************/
  /* This function scans through a branch of a compiled pattern to see whether it
- can match the empty string or not. It is called only from could_be_empty()
+ can match the empty string or not. It is called from could_be_empty()
- below. Note that first_significant_code() skips over assertions. If we hit an
+ below and from compile_branch() when checking for an unlimited repeat of a
- unclosed bracket, we return "empty" - this means we've struck an inner bracket
+ group that can match nothing. Note that first_significant_code() skips over
- whose current branch will already have been scanned.
+ backward and negative forward assertions when its final argument is TRUE. If we
+ hit an unclosed bracket, we return "empty" - this means we've struck an inner
+ bracket whose current branch will already have been scanned.
  Arguments:
    code        points to start of search
    endcode     points to where to stop
    utf8        TRUE if in UTF8 mode
+   cd          contains pointers to tables etc.
  Returns:      TRUE if what is matched could be empty
  */
  static BOOL
- could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
+ could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
+   compile_data *cd)
  {
  register int c;
- for (code = first_significant_code(code + 1 + LINK_SIZE, NULL, 0, TRUE);
+ for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
       code < endcode;
-      code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
+      code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
    {
    const uschar *ccode;
    c = *code;
-   if (c >= OP_BRA)
+   /* Skip over forward assertions; the other assertions are skipped by
+   first_significant_code() with a TRUE final argument. */
+   if (c == OP_ASSERT)
+     {
+     do code += GET(code, 1); while (*code == OP_ALT);
+     c = *code;
+     continue;
+     }
+   /* For a recursion/subroutine call, if its end has been reached, which
+   implies a backward reference subroutine call, we can scan it. If it's a
+   forward reference subroutine call, we can't. To detect forward reference
+   we have to scan up the list that is kept in the workspace. This function is
+   called only when doing the real compile, not during the pre-compile that
+   measures the size of the compiled pattern. */
+   if (c == OP_RECURSE)
      {
+     const uschar *scode;
      BOOL empty_branch;
-     if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */
-     /* Scan a closed bracket */
+     /* Test for forward reference */
+     for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
+       if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
+     /* Not a forward reference, test for completed backward reference */
      empty_branch = FALSE;
+     scode = cd->start_code + GET(code, 1);
+     if (GET(scode, 1) == 0) return TRUE;    /* Unclosed */
+     /* Completed backwards reference */
      do
        {
-       if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
+       if (could_be_empty_branch(scode, endcode, utf8, cd))
+         {
          empty_branch = TRUE;
-       code += GET(code, 1);
+         break;
+         }
+       scode += GET(scode, 1);
        }
-     while (*code == OP_ALT);
+     while (*scode == OP_ALT);
-     if (!empty_branch) return FALSE;   /* All branches are non-empty */
-     code += 1 + LINK_SIZE;
+     if (!empty_branch) return FALSE;  /* All branches are non-empty */
-     c = *code;
+     continue;
+     }
+   /* Groups with zero repeats can of course be empty; skip them. */
+   if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
+       c == OP_BRAPOSZERO)
+     {
+     code += _pcre_OP_lengths[c];
+     do code += GET(code, 1); while (*code == OP_ALT);
+     c = *code;
+     continue;
+     }
+   /* A nested group that is already marked as "could be empty" can just be
+   skipped. */
+   if (c == OP_SBRA  || c == OP_SBRAPOS ||
+       c == OP_SCBRA || c == OP_SCBRAPOS)
+     {
+     do code += GET(code, 1); while (*code == OP_ALT);
+     c = *code;
+     continue;
+     }
+   /* For other groups, scan the branches. */
+   if (c == OP_BRA  || c == OP_BRAPOS ||
+       c == OP_CBRA || c == OP_CBRAPOS ||
+       c == OP_ONCE || c == OP_COND)
+     {
+     BOOL empty_branch;
+     if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed 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. */
+     if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
+       code += GET(code, 1);
+     else
+       {
+       empty_branch = FALSE;
+       do
+         {
+         if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
+           empty_branch = TRUE;
+         code += GET(code, 1);
+         }
+       while (*code == OP_ALT);
+       if (!empty_branch) return FALSE;   /* All branches are non-empty */
+       }
+     c = *code;
+     continue;
      }
-   else switch (c)
+   /* Handle the other opcodes */
+   switch (c)
      {
-     /* Check for quantifiers after a class */
+     /* Check for quantifiers after a class. XCLASS is used for classes that
+     cannot be represented just by a bit map. This includes negated single
+     high-valued characters. The length in _pcre_OP_lengths[] is zero; the
+     actual length is stored in the compiled code, so we must update "code"
+     here. */
  #ifdef SUPPORT_UTF8
      case OP_XCLASS:
-     ccode = code + GET(code, 1);
+     ccode = code += GET(code, 1);
      goto CHECK_CLASS_REPEAT;
  #endif
-Line 1247 
 for (code = first_significant_code(code
+Line 2129 
 for (code = first_significant_code(code
      case OP_NOT_WORDCHAR:
      case OP_WORDCHAR:
      case OP_ANY:
+     case OP_ALLANY:
      case OP_ANYBYTE:
      case OP_CHAR:
-     case OP_CHARNC:
+     case OP_CHARI:
      case OP_NOT:
+     case OP_NOTI:
      case OP_PLUS:
      case OP_MINPLUS:
+     case OP_POSPLUS:
      case OP_EXACT:
      case OP_NOTPLUS:
      case OP_NOTMINPLUS:
+     case OP_NOTPOSPLUS:
      case OP_NOTEXACT:
      case OP_TYPEPLUS:
      case OP_TYPEMINPLUS:
+     case OP_TYPEPOSPLUS:
      case OP_TYPEEXACT:
      return FALSE;
+     /* These are going to continue, as they may be empty, but we have to
+     fudge the length for the \p and \P cases. */
+     case OP_TYPESTAR:
+     case OP_TYPEMINSTAR:
+     case OP_TYPEPOSSTAR:
+     case OP_TYPEQUERY:
+     case OP_TYPEMINQUERY:
+     case OP_TYPEPOSQUERY:
+     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+     break;
+     /* Same for these */
+     case OP_TYPEUPTO:
+     case OP_TYPEMINUPTO:
+     case OP_TYPEPOSUPTO:
+     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
+     break;
      /* End of branch */
      case OP_KET:
      case OP_KETRMAX:
      case OP_KETRMIN:
+     case OP_KETRPOS:
      case OP_ALT:
      return TRUE;
-     /* In UTF-8 mode, STAR, MINSTAR, QUERY, MINQUERY, UPTO, and MINUPTO  may be
+     /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
-     followed by a multibyte character */
+     MINUPTO, and POSUPTO may be followed by a multibyte character */
  #ifdef SUPPORT_UTF8
      case OP_STAR:
+     case OP_STARI:
      case OP_MINSTAR:
+     case OP_MINSTARI:
+     case OP_POSSTAR:
+     case OP_POSSTARI:
      case OP_QUERY:
+     case OP_QUERYI:
      case OP_MINQUERY:
+     case OP_MINQUERYI:
+     case OP_POSQUERY:
+     case OP_POSQUERYI:
+     if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
+     break;
      case OP_UPTO:
+     case OP_UPTOI:
      case OP_MINUPTO:
-     if (utf8) while ((code[2] & 0xc0) == 0x80) code++;
+     case OP_MINUPTOI:
+     case OP_POSUPTO:
+     case OP_POSUPTOI:
+     if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
      break;
  #endif
+     /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
+     string. */
+     case OP_MARK:
+     case OP_PRUNE_ARG:
+     case OP_SKIP_ARG:
+     code += code[1];
+     break;
+     case OP_THEN_ARG:
+     code += code[1+LINK_SIZE];
+     break;
+     /* None of the remaining opcodes are required to match a character. */
+     default:
+     break;
      }
    }
-Line 1299 
 return TRUE;
+Line 2240 
 return TRUE;
  the current branch of the current pattern to see if it could match the empty
  string. If it could, we must look outwards for branches at other levels,
  stopping when we pass beyond the bracket which is the subject of the recursion.
+ This function is called only during the real compile, not during the
+ pre-compile.
  Arguments:
    code        points to start of the recursion
    endcode     points to where to stop (current RECURSE item)
    bcptr       points to the chain of current (unclosed) branch starts
    utf8        TRUE if in UTF-8 mode
+   cd          pointers to tables etc
  Returns:      TRUE if what is matched could be empty
  */
  static BOOL
  could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
-   BOOL utf8)
+   BOOL utf8, compile_data *cd)
  {
- while (bcptr != NULL && bcptr->current >= code)
+ while (bcptr != NULL && bcptr->current_branch >= code)
    {
-   if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;
+   if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
+     return FALSE;
    bcptr = bcptr->outer;
    }
  return TRUE;
-Line 1328 
 return TRUE;
+Line 2273 
 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.
+ A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
+ It seems that the appearance of a nested POSIX class supersedes an apparent
+ external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
+ a digit.
+ In Perl, unescaped square brackets may also appear as part of class names. For
+ example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
+ [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
+ seem right at all. PCRE does not allow closing square brackets in POSIX class
+ names.
- 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)
-   return TRUE;
+     ptr++;
+   else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
+   else
+     {
+     if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
+       {
+       *endptr = ptr;
+       return TRUE;
+       }
+     if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
+          (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
+           ptr[1] == CHAR_EQUALS_SIGN) &&
+         check_posix_syntax(ptr, endptr))
+       return FALSE;
+     }
    }
  return FALSE;
  }
-Line 1375 
 Returns:     a value representing the na
+Line 2357 
 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 1394 
 return -1;
+Line 2378 
 return -1;
  that is referenced. This means that groups can be replicated for fixed
  repetition simply by copying (because the recursion is allowed to refer to
  earlier groups that are outside the current group). However, when a group is
- optional (i.e. the minimum quantifier is zero), OP_BRAZERO is inserted before
+ optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
- it, after it has been compiled. This means that any OP_RECURSE items within it
+ inserted before it, after it has been compiled. This means that any OP_RECURSE
- that refer to the group itself or any contained groups have to have their
+ items within it that refer to the group itself or any contained groups have to
- offsets adjusted. That is the job of this function. Before it is called, the
+ have their offsets adjusted. That one of the jobs of this function. Before it
- partially compiled regex must be temporarily terminated with OP_END.
+ is called, the partially compiled regex must be temporarily terminated with
+ OP_END.
+ This function has been extended with the possibility of forward references for
+ recursions and subroutine calls. It must also check the list of such references
+ for the group we are dealing with. If it finds that one of the recursions in
+ the current group is on this list, it adjusts the offset in the list, not the
+ value in the reference (which is a group number).
  Arguments:
    group      points to the start of the group
    adjust     the amount by which the group is to be moved
    utf8       TRUE in UTF-8 mode
    cd         contains pointers to tables etc.
+   save_hwm   the hwm forward reference pointer at the start of the group
  Returns:     nothing
  */
  static void
- adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd)
+ adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
+   uschar *save_hwm)
  {
  uschar *ptr = group;
  while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
    {
-   int offset = GET(ptr, 1);
+   int offset;
-   if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
+   uschar *hc;
+   /* See if this recursion is on the forward reference list. If so, adjust the
+   reference. */
+   for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
+     {
+     offset = GET(hc, 0);
+     if (cd->start_code + offset == ptr + 1)
+       {
+       PUT(hc, 0, offset + adjust);
+       break;
+       }
+     }
+   /* Otherwise, adjust the recursion offset if it's after the start of this
+   group. */
+   if (hc >= cd->hwm)
+     {
+     offset = GET(ptr, 1);
+     if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
+     }
    ptr += 1 + LINK_SIZE;
    }
  }
-Line 1443 
 auto_callout(uschar *code, const uschar
+Line 2460 
 auto_callout(uschar *code, const uschar
  {
  *code++ = OP_CALLOUT;
  *code++ = 255;
- PUT(code, 0, ptr - cd->start_pattern);  /* Pattern offset */
+ PUT(code, 0, (int)(ptr - cd->start_pattern));  /* Pattern offset */
- PUT(code, LINK_SIZE, 0);                /* Default length */
+ PUT(code, LINK_SIZE, 0);                       /* Default length */
  return code + 2*LINK_SIZE;
  }
-Line 1469 
 Returns:             nothing
+Line 2486 
 Returns:             nothing
  static void
  complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
  {
- int length = ptr - cd->start_pattern - GET(previous_callout, 2);
+ int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
  PUT(previous_callout, 2 + LINK_SIZE, length);
  }
-Line 1495 
 Yield:        TRUE when range returned;
+Line 2512 
 Yield:        TRUE when range returned;
  */
  static BOOL
- get_othercase_range(int *cptr, int d, int *ocptr, int *odptr)
+ get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
+   unsigned int *odptr)
  {
- int c, chartype, othercase, next;
+ unsigned int c, othercase, next;
  for (c = *cptr; c <= d; c++)
-   {
+   { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
-   if (_pcre_ucp_findchar(c, &chartype, &othercase) == ucp_L && othercase != 0)
-     break;
-   }
  if (c > d) return FALSE;
-Line 1512 
 next = othercase + 1;
+Line 2527 
 next = othercase + 1;
  for (++c; c <= d; c++)
    {
-   if (_pcre_ucp_findchar(c, &chartype, &othercase) != ucp_L ||
+   if (UCD_OTHERCASE(c) != next) break;
-         othercase != next)
-     break;
    next++;
    }
-Line 1523 
 for (++c; c <= d; c++)
+Line 2536 
 for (++c; c <= d; c++)
  return TRUE;
  }
+ /*************************************************
+ *        Check a character and a property        *
+ *************************************************/
+ /* This function is called by check_auto_possessive() when a property item
+ is adjacent to a fixed character.
+ Arguments:
+   c            the character
+   ptype        the property type
+   pdata        the data for the type
+   negated      TRUE if it's a negated property (\P or \p{^)
+ Returns:       TRUE if auto-possessifying is OK
+ */
+ static BOOL
+ check_char_prop(int c, int ptype, int pdata, BOOL negated)
+ {
+ const ucd_record *prop = GET_UCD(c);
+ switch(ptype)
+   {
+   case PT_LAMP:
+   return (prop->chartype == ucp_Lu ||
+           prop->chartype == ucp_Ll ||
+           prop->chartype == ucp_Lt) == negated;
+   case PT_GC:
+   return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
+   case PT_PC:
+   return (pdata == prop->chartype) == negated;
+   case PT_SC:
+   return (pdata == prop->script) == negated;
+   /* These are specials */
+   case PT_ALNUM:
+   return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
+           _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
+   case PT_SPACE:    /* Perl space */
+   return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
+           c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
+           == negated;
+   case PT_PXSPACE:  /* POSIX space */
+   return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
+           c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
+           c == CHAR_FF || c == CHAR_CR)
+           == negated;
+   case PT_WORD:
+   return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
+           _pcre_ucp_gentype[prop->chartype] == ucp_N ||
+           c == CHAR_UNDERSCORE) == negated;
+   }
+ return FALSE;
+ }
  #endif  /* SUPPORT_UCP */
+ /*************************************************
+ *     Check if auto-possessifying is possible    *
+ *************************************************/
+ /* This function is called for unlimited repeats of certain items, to see
+ whether the next thing could possibly match the repeated item. If not, it makes
+ sense to automatically possessify the repeated item.
+ Arguments:
+   previous      pointer to the repeated opcode
+   utf8          TRUE in UTF-8 mode
+   ptr           next character in pattern
+   options       options bits
+   cd            contains pointers to tables etc.
+ Returns:        TRUE if possessifying is wanted
+ */
+ static BOOL
+ check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
+   int options, compile_data *cd)
+ {
+ int c, next;
+ int op_code = *previous++;
+ /* Skip whitespace and comments in extended mode */
+ if ((options & PCRE_EXTENDED) != 0)
+   {
+   for (;;)
+     {
+     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
+     if (*ptr == CHAR_NUMBER_SIGN)
+       {
+       ptr++;
+       while (*ptr != 0)
+         {
+         if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
+         ptr++;
+ #ifdef SUPPORT_UTF8
+         if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
+ #endif
+         }
+       }
+     else break;
+     }
+   }
+ /* 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 == CHAR_BACKSLASH)
+   {
+   int temperrorcode = 0;
+   next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
+   if (temperrorcode != 0) return FALSE;
+   ptr++;    /* Point after the escape sequence */
+   }
+ else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
+   {
+ #ifdef SUPPORT_UTF8
+   if (utf8) { GETCHARINC(next, ptr); } else
+ #endif
+   next = *ptr++;
+   }
+ else return FALSE;
+ /* Skip whitespace and comments in extended mode */
+ if ((options & PCRE_EXTENDED) != 0)
+   {
+   for (;;)
+     {
+     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
+     if (*ptr == CHAR_NUMBER_SIGN)
+       {
+       ptr++;
+       while (*ptr != 0)
+         {
+         if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
+         ptr++;
+ #ifdef SUPPORT_UTF8
+         if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
+ #endif
+         }
+       }
+     else break;
+     }
+   }
+ /* If the next thing is itself optional, we have to give up. */
+ if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
+   strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
+     return FALSE;
+ /* Now compare the next item with the previous opcode. First, handle cases when
+ the next item is a character. */
+ if (next >= 0) switch(op_code)
+   {
+   case OP_CHAR:
+ #ifdef SUPPORT_UTF8
+   GETCHARTEST(c, previous);
+ #else
+   c = *previous;
+ #endif
+   return c != next;
+   /* For CHARI (caseless character) we must check the other case. If we have
+   Unicode property support, we can use it to test the other case of
+   high-valued characters. */
+   case OP_CHARI:
+ #ifdef SUPPORT_UTF8
+   GETCHARTEST(c, previous);
+ #else
+   c = *previous;
+ #endif
+   if (c == next) return FALSE;
+ #ifdef SUPPORT_UTF8
+   if (utf8)
+     {
+     unsigned int othercase;
+     if (next < 128) othercase = cd->fcc[next]; else
+ #ifdef SUPPORT_UCP
+     othercase = UCD_OTHERCASE((unsigned int)next);
+ #else
+     othercase = NOTACHAR;
+ #endif
+     return (unsigned int)c != othercase;
+     }
+   else
+ #endif  /* SUPPORT_UTF8 */
+   return (c != cd->fcc[next]);  /* Non-UTF-8 mode */
+   /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
+   opcodes are not used for multi-byte characters, because they are coded using
+   an XCLASS instead. */
+   case OP_NOT:
+   return (c = *previous) == next;
+   case OP_NOTI:
+   if ((c = *previous) == next) return TRUE;
+ #ifdef SUPPORT_UTF8
+   if (utf8)
+     {
+     unsigned int othercase;
+     if (next < 128) othercase = cd->fcc[next]; else
+ #ifdef SUPPORT_UCP
+     othercase = UCD_OTHERCASE(next);
+ #else
+     othercase = NOTACHAR;
+ #endif
+     return (unsigned int)c == othercase;
+     }
+   else
+ #endif  /* SUPPORT_UTF8 */
+   return (c == cd->fcc[next]);  /* Non-UTF-8 mode */
+   /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
+   When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
+   case OP_DIGIT:
+   return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
+   case OP_NOT_DIGIT:
+   return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
+   case OP_WHITESPACE:
+   return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
+   case OP_NOT_WHITESPACE:
+   return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
+   case OP_WORDCHAR:
+   return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
+   case OP_NOT_WORDCHAR:
+   return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
+   case OP_HSPACE:
+   case OP_NOT_HSPACE:
+   switch(next)
+     {
+     case 0x09:
+     case 0x20:
+     case 0xa0:
+     case 0x1680:
+     case 0x180e:
+     case 0x2000:
+     case 0x2001:
+     case 0x2002:
+     case 0x2003:
+     case 0x2004:
+     case 0x2005:
+     case 0x2006:
+     case 0x2007:
+     case 0x2008:
+     case 0x2009:
+     case 0x200A:
+     case 0x202f:
+     case 0x205f:
+     case 0x3000:
+     return op_code == OP_NOT_HSPACE;
+     default:
+     return op_code != OP_NOT_HSPACE;
+     }
+   case OP_ANYNL:
+   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_NOT_VSPACE;
+     default:
+     return op_code != OP_NOT_VSPACE;
+     }
+ #ifdef SUPPORT_UCP
+   case OP_PROP:
+   return check_char_prop(next, previous[0], previous[1], FALSE);
+   case OP_NOTPROP:
+   return check_char_prop(next, previous[0], previous[1], TRUE);
+ #endif
+   default:
+   return FALSE;
+   }
+ /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
+ is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
+ generated only when PCRE_UCP is *not* set, that is, when only ASCII
+ characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
+ replaced by OP_PROP codes when PCRE_UCP is set. */
+ switch(op_code)
+   {
+   case OP_CHAR:
+   case OP_CHARI:
+ #ifdef SUPPORT_UTF8
+   GETCHARTEST(c, previous);
+ #else
+   c = *previous;
+ #endif
+   switch(-next)
+     {
+     case ESC_d:
+     return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
+     case ESC_D:
+     return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
+     case ESC_s:
+     return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
+     case ESC_S:
+     return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
+     case ESC_w:
+     return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
+     case ESC_W:
+     return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
+     case ESC_h:
+     case ESC_H:
+     switch(c)
+       {
+       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(c)
+       {
+       case 0x0a:
+       case 0x0b:
+       case 0x0c:
+       case 0x0d:
+       case 0x85:
+       case 0x2028:
+       case 0x2029:
+       return -next != ESC_v;
+       default:
+       return -next == ESC_v;
+       }
+     /* When PCRE_UCP is set, these values get generated for \d etc. Find
+     their substitutions and process them. The result will always be either
+     -ESC_p or -ESC_P. Then fall through to process those values. */
+ #ifdef SUPPORT_UCP
+     case ESC_du:
+     case ESC_DU:
+     case ESC_wu:
+     case ESC_WU:
+     case ESC_su:
+     case ESC_SU:
+       {
+       int temperrorcode = 0;
+       ptr = substitutes[-next - ESC_DU];
+       next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
+       if (temperrorcode != 0) return FALSE;
+       ptr++;    /* For compatibility */
+       }
+     /* Fall through */
+     case ESC_p:
+     case ESC_P:
+       {
+       int ptype, pdata, errorcodeptr;
+       BOOL negated;
+       ptr--;      /* Make ptr point at the p or P */
+       ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
+       if (ptype < 0) return FALSE;
+       ptr++;      /* Point past the final curly ket */
+       /* If the property item is optional, we have to give up. (When generated
+       from \d etc by PCRE_UCP, this test will have been applied much earlier,
+       to the original \d etc. At this point, ptr will point to a zero byte. */
+       if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
+         strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
+           return FALSE;
+       /* Do the property check. */
+       return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
+       }
+ #endif
+     default:
+     return FALSE;
+     }
+   /* In principle, support for Unicode properties should be integrated here as
+   well. It means re-organizing the above code so as to get hold of the property
+   values before switching on the op-code. However, I wonder how many patterns
+   combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
+   these op-codes are never generated.) */
+   case OP_DIGIT:
+   return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
+          next == -ESC_h || next == -ESC_v || next == -ESC_R;
+   case OP_NOT_DIGIT:
+   return next == -ESC_d;
+   case OP_WHITESPACE:
+   return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
+   case OP_NOT_WHITESPACE:
+   return next == -ESC_s || next == -ESC_h || next == -ESC_v;
+   case OP_HSPACE:
+   return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
+          next == -ESC_w || next == -ESC_v || next == -ESC_R;
+   case OP_NOT_HSPACE:
+   return next == -ESC_h;
+   /* Can't have \S in here because VT matches \S (Perl anomaly) */
+   case OP_ANYNL:
+   case OP_VSPACE:
+   return next == -ESC_V || next == -ESC_d || next == -ESC_w;
+   case OP_NOT_VSPACE:
+   return next == -ESC_v || next == -ESC_R;
+   case OP_WORDCHAR:
+   return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
+          next == -ESC_v || next == -ESC_R;
+   case OP_NOT_WORDCHAR:
+   return next == -ESC_w || next == -ESC_d;
+   default:
+   return FALSE;
+   }
+ /* Control does not reach here */
+ }
  /*************************************************
  *           Compile one branch                   *
  *************************************************/
- /* Scan the pattern, compiling it into the code vector. If the options are
+ /* Scan the pattern, compiling it into the a vector. If the options are
  changed during the branch, the pointer is used to change the external options
- bits.
+ bits. This function is used during the pre-compile phase when we are trying
+ to find out the amount of memory needed, as well as during the real compile
+ phase. The value of lengthptr distinguishes the two phases.
  Arguments:
    optionsptr     pointer to the option bits
-   brackets       points to number of extracting brackets used
    codeptr        points to the pointer to the current code point
    ptrptr         points to the current pattern pointer
    errorcodeptr   points to error code variable
    firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
    reqbyteptr     set to the last literal character required, else < 0
    bcptr          points to current branch chain
+   cond_depth     conditional nesting depth
    cd             contains pointers to tables etc.
+   lengthptr      NULL during the real compile phase
+                  points to length accumulator during pre-compile phase
  Returns:         TRUE on success
                   FALSE, with *errorcodeptr set non-zero on error
  */
  static BOOL
- compile_branch(int *optionsptr, int *brackets, uschar **codeptr,
+ compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
-   const uschar **ptrptr, int *errorcodeptr, int *firstbyteptr,
+   int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
-   int *reqbyteptr, branch_chain *bcptr, compile_data *cd)
+   int cond_depth, compile_data *cd, int *lengthptr)
  {
  int repeat_type, op_type;
  int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */
-Line 1561 
 int greedy_default, greedy_non_default;
+Line 3062 
 int greedy_default, greedy_non_default;
  int firstbyte, reqbyte;
  int zeroreqbyte, zerofirstbyte;
  int req_caseopt, reqvary, tempreqvary;
- int condcount = 0;
+ int options = *optionsptr;               /* May change dynamically */
- int options = *optionsptr;
  int after_manual_callout = 0;
+ int length_prevgroup = 0;
  register int c;
  register uschar *code = *codeptr;
+ uschar *last_code = code;
+ uschar *orig_code = code;
  uschar *tempcode;
  BOOL inescq = FALSE;
  BOOL groupsetfirstbyte = FALSE;
  const uschar *ptr = *ptrptr;
  const uschar *tempptr;
+ const uschar *nestptr = NULL;
  uschar *previous = NULL;
  uschar *previous_callout = NULL;
+ uschar *save_hwm = NULL;
  uschar classbits[32];
+ /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
+ must not do this for other options (e.g. PCRE_EXTENDED) because they may change
+ dynamically as we process the pattern. */
  #ifdef SUPPORT_UTF8
  BOOL class_utf8;
  BOOL utf8 = (options & PCRE_UTF8) != 0;
  uschar *class_utf8data;
+ uschar *class_utf8data_base;
  uschar utf8_char[6];
  #else
  BOOL utf8 = FALSE;
  #endif
+ #ifdef PCRE_DEBUG
+ if (lengthptr != NULL) DPRINTF((">> start branch\n"));
+ #endif
  /* Set up the default and non-default settings for greediness */
  greedy_default = ((options & PCRE_UNGREEDY) != 0);
-Line 1613 
 req_caseopt = ((options & PCRE_CASELESS)
+Line 3127 
 req_caseopt = ((options & PCRE_CASELESS)
  for (;; ptr++)
    {
    BOOL negate_class;
+   BOOL should_flip_negation;
    BOOL possessive_quantifier;
    BOOL is_quantifier;
+   BOOL is_recurse;
+   BOOL reset_bracount;
    int class_charcount;
    int class_lastchar;
    int newoptions;
    int recno;
+   int refsign;
    int skipbytes;
    int subreqbyte;
    int subfirstbyte;
+   int terminator;
    int mclength;
    uschar mcbuffer[8];
-   /* Next byte in the pattern */
+   /* Get next byte in the pattern */
    c = *ptr;
-   /* If in \Q...\E, check for the end; if not, we have a literal */
+   /* If we are at the end of a nested substitution, revert to the outer level
+   string. Nesting only happens one level deep. */
+   if (c == 0 && nestptr != NULL)
+     {
+     ptr = nestptr;
+     nestptr = NULL;
+     c = *ptr;
+     }
+   /* If we are in the pre-compile phase, accumulate the length used for the
+   previous cycle of this loop. */
+   if (lengthptr != NULL)
+     {
+ #ifdef PCRE_DEBUG
+     if (code > cd->hwm) cd->hwm = code;                 /* High water info */
+ #endif
+     if (code > cd->start_workspace + WORK_SIZE_CHECK)   /* Check for overrun */
+       {
+       *errorcodeptr = ERR52;
+       goto FAILED;
+       }
+     /* There is at least one situation where code goes backwards: this is the
+     case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
+     the class is simply eliminated. However, it is created first, so we have to
+     allow memory for it. Therefore, don't ever reduce the length at this point.
+     */
+     if (code < last_code) code = last_code;
+     /* Paranoid check for integer overflow */
+     if (OFLOW_MAX - *lengthptr < code - last_code)
+       {
+       *errorcodeptr = ERR20;
+       goto FAILED;
+       }
+     *lengthptr += (int)(code - last_code);
+     DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
+     /* If "previous" is set and it is not at the start of the work space, move
+     it back to there, in order to avoid filling up the work space. Otherwise,
+     if "previous" is NULL, reset the current code pointer to the start. */
+     if (previous != NULL)
+       {
+       if (previous > orig_code)
+         {
+         memmove(orig_code, previous, code - previous);
+         code -= previous - orig_code;
+         previous = orig_code;
+         }
+       }
+     else code = orig_code;
+     /* Remember where this code item starts so we can pick up the length
+     next time round. */
+     last_code = code;
+     }
+   /* In the real compile phase, just check the workspace used by the forward
+   reference list. */
+   else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
+     {
+     *errorcodeptr = ERR52;
+     goto FAILED;
+     }
+   /* If in \Q...\E, check for the end; if not, we have a literal */
    if (inescq && c != 0)
      {
-     if (c == '\\' && ptr[1] == 'E')
+     if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
        {
        inescq = FALSE;
        ptr++;
-Line 1643 
 for (;; ptr++)
+Line 3235 
 for (;; ptr++)
        {
        if (previous_callout != NULL)
          {
-         complete_callout(previous_callout, ptr, cd);
+         if (lengthptr == NULL)  /* Don't attempt in pre-compile phase */
+           complete_callout(previous_callout, ptr, cd);
          previous_callout = NULL;
          }
        if ((options & PCRE_AUTO_CALLOUT) != 0)
-Line 1658 
 for (;; ptr++)
+Line 3251 
 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)
      {
-     complete_callout(previous_callout, ptr, cd);
+     if (lengthptr == NULL)      /* Don't attempt in pre-compile phase */
+       complete_callout(previous_callout, ptr, cd);
      previous_callout = NULL;
      }
-   /* In extended mode, skip white space and comments */
+   /* In extended mode, skip white space and comments. */
    if ((options & PCRE_EXTENDED) != 0)
      {
      if ((cd->ctypes[c] & ctype_space) != 0) continue;
-     if (c == '#')
+     if (c == CHAR_NUMBER_SIGN)
        {
-       /* The space before the ; is to avoid a warning on a silly compiler
+       ptr++;
-       on the Macintosh. */
+       while (*ptr != 0)
-       while ((c = *(++ptr)) != 0 && c != NEWLINE) ;
+         {
-       if (c != 0) continue;   /* Else fall through to handle end of string */
+         if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
+         ptr++;
+ #ifdef SUPPORT_UTF8
+         if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
+ #endif
+         }
+       if (*ptr != 0) continue;
+       /* Else fall through to handle end of string */
+       c = 0;
        }
      }
-Line 1692 
 for (;; ptr++)
+Line 3296 
 for (;; ptr++)
    switch(c)
      {
-     /* The branch terminates at end of string, |, or ). */
+     /* ===================================================================*/
+     case 0:                        /* The branch terminates at string end */
-     case 0:
+     case CHAR_VERTICAL_LINE:       /* or | or ) */
-     case '|':
+     case CHAR_RIGHT_PARENTHESIS:
-     case ')':
      *firstbyteptr = firstbyte;
      *reqbyteptr = reqbyte;
      *codeptr = code;
      *ptrptr = ptr;
+     if (lengthptr != NULL)
+       {
+       if (OFLOW_MAX - *lengthptr < code - last_code)
+         {
+         *errorcodeptr = ERR20;
+         goto FAILED;
+         }
+       *lengthptr += (int)(code - last_code);   /* To include callout length */
+       DPRINTF((">> end branch\n"));
+       }
      return TRUE;
+     /* ===================================================================*/
      /* Handle single-character metacharacters. In multiline mode, ^ disables
      the setting of any following char as a first character. */
-     case '^':
+     case CHAR_CIRCUMFLEX_ACCENT:
+     previous = NULL;
      if ((options & PCRE_MULTILINE) != 0)
        {
        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
+       *code++ = OP_CIRCM;
        }
-     previous = NULL;
+     else *code++ = OP_CIRC;
-     *code++ = OP_CIRC;
      break;
-     case '$':
+     case CHAR_DOLLAR_SIGN:
      previous = NULL;
-     *code++ = OP_DOLL;
+     *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
      break;
      /* There can never be a first char if '.' is first, whatever happens about
      repeats. The value of reqbyte doesn't change either. */
-     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;
-     /* Character classes. If the included characters are all < 255 in value, we
-     build a 32-byte bitmap of the permitted characters, except in the special
+     /* ===================================================================*/
-     case where there is only one such character. For negated classes, we build
+     /* Character classes. If the included characters are all < 256, we build a
-     the map as usual, then invert it at the end. However, we use a different
+-byte bitmap of the permitted characters, except in the special case
-     opcode so that data characters > 255 can be handled correctly.
+     where there is only one such character. For negated classes, we build the
+     map as usual, then invert it at the end. However, we use a different opcode
+     so that data characters > 255 can be handled correctly.
      If the class contains characters outside the 0-255 range, a different
      opcode is compiled. It may optionally have a bit map for characters < 256,
      but those above are are explicitly listed afterwards. A flag byte tells
      whether the bitmap is present, and whether this is a negated class or not.
-     */
-     case '[':
+     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). For higher valued UTF-8
+     of just a single character (as long as it's < 256). However, For higher
-     characters, we don't yet do any optimization. */
+     valued UTF-8 characters, we don't yet do any optimization. */
      class_charcount = 0;
      class_lastchar = -1;
+     /* Initialize the 32-char bit map to all zeros. We build the map in a
+     temporary bit of memory, in case the class contains only 1 character (less
+     than 256), because in that case the compiled code doesn't use the bit map.
+     */
+     memset(classbits, 0, 32 * sizeof(uschar));
  #ifdef SUPPORT_UTF8
      class_utf8 = FALSE;                       /* No chars >= 256 */
-     class_utf8data = code + LINK_SIZE + 34;   /* For UTF-8 items */
+     class_utf8data = code + LINK_SIZE + 2;    /* For UTF-8 items */
+     class_utf8data_base = class_utf8data;     /* For resetting in pass 1 */
  #endif
-     /* Initialize the 32-char bit map to all zeros. We have to build the
-     map in a temporary bit of store, in case the class contains only 1
-     character (< 256), because in that case the compiled code doesn't use the
-     bit map. */
-     memset(classbits, 0, 32 * sizeof(uschar));
      /* Process characters until ] is reached. By writing this as a "do" it
-     means that an initial ] is taken as a data character. The first pass
+     means that an initial ] is taken as a data character. At the start of the
-     through the regex checked the overall syntax, so we don't need to be very
+     loop, c contains the first byte of the character. */
-     strict here. At the start of the loop, c contains the first byte of the
-     character. */
-     do
+     if (c != 0) do
        {
+       const uschar *oldptr;
  #ifdef SUPPORT_UTF8
        if (utf8 && c > 127)
          {                           /* Braces are required because the */
          GETCHARLEN(c, ptr, ptr);    /* macro generates multiple statements */
          }
+       /* In the pre-compile phase, accumulate the length of any UTF-8 extra
+       data and reset the pointer. This is so that very large classes that
+       contain a zillion UTF-8 characters no longer overwrite the work space
+       (which is on the stack). */
+       if (lengthptr != NULL)
+         {
+         *lengthptr += class_utf8data - class_utf8data_base;
+         class_utf8data = class_utf8data_base;
+         }
  #endif
        /* Inside \Q...\E everything is literal except \E */
        if (inescq)
          {
-         if (c == '\\' && ptr[1] == 'E')
+         if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)  /* If we are at \E */
            {
-           inescq = FALSE;
+           inescq = FALSE;                   /* Reset literal state */
-           ptr++;
+           ptr++;                            /* Skip the 'E' */
-           continue;
+           continue;                         /* Carry on with next */
            }
-         else goto LONE_SINGLE_CHARACTER;
+         goto CHECK_RANGE;                   /* Could be range if \E follows */
          }
        /* Handle POSIX class names. Perl allows a negation extension of the
-Line 1821 
 for (;; ptr++)
+Line 3494 
 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, i;
+         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++;
            }
-         posix_class = check_posix_name(ptr, tempptr - ptr);
+         posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
          if (posix_class < 0)
            {
            *errorcodeptr = ERR30;
-Line 1856 
 for (;; ptr++)
+Line 3531 
 for (;; ptr++)
          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
            posix_class = 0;
-         /* Or into the map we are building up to 3 of the static class
+         /* When PCRE_UCP is set, some of the POSIX classes are converted to
-         tables, or their negations. The [:blank:] class sets up the same
+         different escape sequences that use Unicode properties. */
-         chars as the [:space:] class (all white space). We remove the vertical
-         white space chars afterwards. */
-         posix_class *= 3;
+ #ifdef SUPPORT_UCP
-         for (i = 0; i < 3; i++)
+         if ((options & PCRE_UCP) != 0)
            {
-           BOOL blankclass = strncmp((char *)ptr, "blank", 5) == 0;
+           int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
-           int taboffset = posix_class_maps[posix_class + i];
+           if (posix_substitutes[pc] != NULL)
-           if (taboffset < 0) break;
-           if (local_negate)
-             {
-             if (i == 0)
-               for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+taboffset];
-             else
-               for (c = 0; c < 32; c++) classbits[c] &= ~cbits[c+taboffset];
-             if (blankclass) classbits[1] |= 0x3c;
-             }
-           else
              {
-             for (c = 0; c < 32; c++) classbits[c] |= cbits[c+taboffset];
+             nestptr = tempptr + 1;
-             if (blankclass) classbits[1] &= ~0x3c;
+             ptr = posix_substitutes[pc] - 1;
+             continue;
              }
            }
+ #endif
+         /* In the non-UCP case, we build the bit map for the POSIX class in a
+         chunk of local store because we may be adding and subtracting from it,
+         and we don't want to subtract bits that may be in the main map already.
+         At the end we or the result into the bit map that is being built. */
+         posix_class *= 3;
+         /* Copy in the first table (always present) */
+         memcpy(pbits, cbits + posix_class_maps[posix_class],
+* sizeof(uschar));
+         /* If there is a second table, add or remove it as required. */
+         taboffset = posix_class_maps[posix_class + 1];
+         tabopt = posix_class_maps[posix_class + 2];
+         if (taboffset >= 0)
+           {
+           if (tabopt >= 0)
+             for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
+           else
+             for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
+           }
+         /* Not see if we need to remove any special characters. An option
+         value of 1 removes vertical space and 2 removes underscore. */
+         if (tabopt < 0) tabopt = -tabopt;
+         if (tabopt == 1) pbits[1] &= ~0x3c;
+           else if (tabopt == 2) pbits[11] &= 0x7f;
+         /* Add the POSIX table or its complement into the main table that is
+         being built and we are done. */
+         if (local_negate)
+           for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
+         else
+           for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
          ptr = tempptr + 1;
          class_charcount = 10;  /* Set > 1; assumes more than 1 per class */
-Line 1888 
 for (;; ptr++)
+Line 3592 
 for (;; ptr++)
          }
        /* Backslash may introduce a single character, or it may introduce one
-       of the specials, which just set a flag. Escaped items are checked for
+       of the specials, which just set a flag. The sequence \b is a special
-       validity in the pre-compiling pass. The sequence \b is a special case.
+       case. Inside a class (and only there) it is treated as backspace. We
-       Inside a class (and only there) it is treated as backspace. Elsewhere
+       assume that other escapes have more than one character in them, so set
-       it marks a word boundary. Other escapes have preset maps ready to
+       class_charcount bigger than one. Unrecognized escapes fall through and
-       or into the one we are building. We assume they have more than one
+       are either treated as literal characters (by default), or are faulted if
-       character in them, so set class_charcount bigger than one. */
+       PCRE_EXTRA is set. */
-       if (c == '\\')
+       if (c == CHAR_BACKSLASH)
          {
-         c = check_escape(&ptr, errorcodeptr, *brackets, options, TRUE);
+         c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
+         if (*errorcodeptr != 0) goto FAILED;
-         if (-c == ESC_b) c = '\b';       /* \b is backslash in a class */
+         if (-c == ESC_b) c = 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_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)
            {
            register const uschar *cbits = cd->cbits;
            class_charcount += 2;     /* Greater than 1 is what matters */
            switch (-c)
              {
+ #ifdef SUPPORT_UCP
+             case ESC_du:     /* These are the values given for \d etc */
+             case ESC_DU:     /* when PCRE_UCP is set. We replace the */
+             case ESC_wu:     /* escape sequence with an appropriate \p */
+             case ESC_WU:     /* or \P to test Unicode properties instead */
+             case ESC_su:     /* of the default ASCII testing. */
+             case ESC_SU:
+             nestptr = ptr;
+             ptr = substitutes[-c - ESC_DU] - 1;  /* Just before substitute */
+             class_charcount -= 2;                /* Undo! */
+             continue;
+ #endif
              case ESC_d:
              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
              continue;
              case ESC_D:
+             should_flip_negation = TRUE;
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
              continue;
-Line 1930 
 for (;; ptr++)
+Line 3649 
 for (;; ptr++)
              continue;
              case ESC_W:
+             should_flip_negation = TRUE;
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
              continue;
+             /* Perl 5.004 onwards omits VT from \s, but we must preserve it
+             if it was previously set by something earlier in the character
+             class. */
              case ESC_s:
-             for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
+             classbits[0] |= cbits[cbit_space];
-             classbits[1] &= ~0x08;   /* Perl 5.004 onwards omits VT from \s */
+             classbits[1] |= cbits[cbit_space+1] & ~0x08;
+             for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
              continue;
              case ESC_S:
+             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_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;
+             case 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;
+             case ESC_v:
+             SETBIT(classbits, 0x0a); /* LF */
+             SETBIT(classbits, 0x0b); /* VT */
+             SETBIT(classbits, 0x0c); /* FF */
+             SETBIT(classbits, 0x0d); /* CR */
+             SETBIT(classbits, 0x85); /* NEL */
+ #ifdef SUPPORT_UTF8
+             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;
+             case ESC_V:
+             for (c = 0; c < 32; c++)
+               {
+               int x = 0xff;
+               switch (c)
+                 {
+                 case 0x0a/8: x ^= 1 << (0x0a%8);
+                              x ^= 1 << (0x0b%8);
+                              x ^= 1 << (0x0c%8);
+                              x ^= 1 << (0x0d%8);
+                              break;
+                 case 0x85/8: x ^= 1 << (0x85%8); break;
+                 default: break;
+                 }
+               classbits[c] |= x;
+               }
+ #ifdef SUPPORT_UTF8
+             if (utf8)
+               {
+               class_utf8 = TRUE;
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
+               *class_utf8data++ = XCL_RANGE;
+               class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
+               class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
+               }
+ #endif
+             continue;
  #ifdef SUPPORT_UCP
              case ESC_p:
              case ESC_P:
                {
                BOOL negated;
-               int property = get_ucp(&ptr, &negated, errorcodeptr);
+               int pdata;
-               if (property < 0) goto FAILED;
+               int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
+               if (ptype < 0) goto FAILED;
                class_utf8 = TRUE;
                *class_utf8data++ = ((-c == ESC_p) != negated)?
                  XCL_PROP : XCL_NOTPROP;
-               *class_utf8data++ = property;
+               *class_utf8data++ = ptype;
+               *class_utf8data++ = pdata;
                class_charcount -= 2;   /* Not a < 256 character */
+               continue;
                }
-             continue;
  #endif
              /* Unrecognized escapes are faulted if PCRE is running in its
              strict mode. By default, for compatibility with Perl, they are
              treated as literals. */
-Line 1969 
 for (;; ptr++)
+Line 3812 
 for (;; ptr++)
                *errorcodeptr = ERR7;
                goto FAILED;
                }
-             c = *ptr;              /* The final character */
              class_charcount -= 2;  /* Undo the default count from above */
+             c = *ptr;              /* Get the final character and fall through */
+             break;
              }
            }
          /* Fall through if we have a single character (c >= 0). This may be
-         > 256 in UTF-8 mode. */
+         greater than 256 in UTF-8 mode. */
          }   /* End of backslash handling */
        /* A single character may be followed by '-' to form a range. However,
        Perl does not permit ']' to be the end of the range. A '-' character
-       here is treated as a literal. */
+       at the end is treated as a literal. Perl ignores orphaned \E sequences
+       entirely. The code for handling \Q and \E is messy. */
+       CHECK_RANGE:
+       while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
+         {
+         inescq = FALSE;
+         ptr += 2;
+         }
+       oldptr = ptr;
+       /* Remember \r or \n */
+       if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
-       if (ptr[1] == '-' && ptr[2] != ']')
+       /* Check for range */
+       if (!inescq && ptr[1] == CHAR_MINUS)
          {
          int d;
     &n