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

Diff of /code/trunk/pcre_compile.c

Parent Directory | Revision Log | View Patch Patch

-revision 205 by ph10,
Fri Aug  3 13:18:33 2007 UTC
+revision 1219 by ph10,
Sun Nov 11 18:04:37 2012 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2007 University of Cambridge
+            Copyright (c) 1997-2012 University of Cambridge
  -----------------------------------------------------------------------------
  Redistribution and use in source and binary forms, with or without
 Line 43 
 supporting internal functions that are n
  #ifdef HAVE_CONFIG_H
- #include <config.h>
+ #include "config.h"
  #endif
  #define NLBLOCK cd             /* Block containing newline information */
 Line 53 
 supporting internal functions that are n
  #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(16|32)_printint() function, which
- used by pcretest. DEBUG is not defined when building a production library. */
+ is also used by pcretest. PCRE_DEBUG is not defined when building a production
+ library. We do not need to select pcre16_printint.c specially, because the
- #ifdef DEBUG
+ COMPILE_PCREx macro will already be appropriately set. */
- #include "pcre_printint.src"
+ #ifdef PCRE_DEBUG
+ /* pcre_printint.c should not include any headers */
+ #define PCRE_INCLUDED
+ #include "pcre_printint.c"
+ #undef PCRE_INCLUDED
  #endif
  /* Macro for setting individual bits in class bitmaps. */
- #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
+ #define SETBIT(a,b) a[(b)/8] |= (1 << ((b)&7))
  /* 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
-Line 72 
 to check them every time. */
+Line 77 
 to check them every time. */
  #define OFLOW_MAX (INT_MAX - 20)
+ /* Definitions to allow mutual recursion */
+ static int
+   add_list_to_class(pcre_uint8 *, pcre_uchar **, int, compile_data *,
+     const pcre_uint32 *, unsigned int);
+ static BOOL
+   compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
+     pcre_uint32 *, pcre_int32 *, pcre_uint32 *, pcre_int32 *, branch_chain *,
+     compile_data *, int *);
  /*************************************************
  *      Code parameters and static tables         *
-Line 87 
 so this number is very generous.
+Line 104 
 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. */
+ is 4 there is plenty of room for most patterns. However, the memory can get
+ filled up by repetitions of forward references, for example patterns like
+ /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
+ that the workspace is expanded using malloc() in this situation. The value
+ below is therefore a minimum, and we put a maximum on it for safety. The
+ minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
+ kicks in at the same number of forward references in all cases. */
+ #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
+ #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
- #define COMPILE_WORK_SIZE (4096)
+ /* 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_SAFETY_MARGIN (100)
+ /* Private flags added to firstchar and reqchar. */
+ #define REQ_CASELESS    (1 << 0)        /* Indicates caselessness */
+ #define REQ_VARY        (1 << 1)        /* Reqchar followed non-literal item */
+ /* Negative values for the firstchar and reqchar flags */
+ #define REQ_UNSET       (-2)
+ #define REQ_NONE        (-1)
+ /* Repeated character flags. */
+ #define UTF_LENGTH     0x10000000l      /* The char contains its length. */
  /* Table for handling escaped characters in the range '0'-'z'. Positive returns
  are simple data values; negative values are for special things like \d and so
  on. Zero means further processing is needed (for things like \x), or the escape
  is invalid. */
- #ifndef EBCDIC  /* This is the "normal" table for ASCII systems */
+ #ifndef EBCDIC
+ /* This is the "normal" table for ASCII systems or for EBCDIC systems running
+ in UTF-8 mode. */
  static const short int escapes[] = {
-,      0,      0,      0,      0,      0,      0,      0,   /* 0 - 7 */
+,                       0,
-,      0,    ':',    ';',    '<',    '=',    '>',    '?',   /* 8 - ? */
+,                       0,
-    '@', -ESC_A, -ESC_B, -ESC_C, -ESC_D, -ESC_E,      0, -ESC_G,   /* @ - G */
+,                       0,
- -ESC_H,      0,      0, -ESC_K,      0,      0,      0,      0,   /* H - O */
+,                       0,
- -ESC_P, -ESC_Q, -ESC_R, -ESC_S,      0,      0, -ESC_V, -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,
- -ESC_h,      0,      0, -ESC_k,      0,      0,  ESC_n,      0,   /* h - o */
+      CHAR_GREATER_THAN_SIGN,  CHAR_QUESTION_MARK,
- -ESC_p,      0,  ESC_r, -ESC_s,  ESC_tee,    0, -ESC_v, -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 130 
 static const short int escapes[] = {
+Line 205 
 static const short int escapes[] = {
  /*  B8 */     0,     0,      0,       0,      0,   ']',    '=',    '-',
  /*  C0 */   '{',-ESC_A, -ESC_B,  -ESC_C, -ESC_D,-ESC_E,      0, -ESC_G,
  /*  C8 */-ESC_H,     0,      0,       0,      0,     0,      0,      0,
- /*  D0 */   '}',     0, -ESC_K,       0,      0,     0,      0, -ESC_P,
+ /*  D0 */   '}',     0, -ESC_K,       0,      0,-ESC_N,      0, -ESC_P,
  /*  D8 */-ESC_Q,-ESC_R,      0,       0,      0,     0,      0,      0,
  /*  E0 */  '\\',     0, -ESC_S,       0,      0,-ESC_V, -ESC_W, -ESC_X,
  /*  E8 */     0,-ESC_Z,      0,       0,      0,     0,      0,      0,
-Line 140 
 static const short int escapes[] = {
+Line 215 
 static const short int escapes[] = {
  #endif
- /* Tables of names of POSIX character classes and their lengths. The list is
+ /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
- terminated by a zero length entry. The first three must be alpha, lower, upper,
+ searched linearly. Put all the names into a single string, in order to reduce
- as this is assumed for handling case independence. */
+ the number of relocations when a shared library is dynamically linked. The
+ string is built from string macros so that it works in UTF-8 mode on EBCDIC
- static const char *const posix_names[] = {
+ platforms. */
-   "alpha", "lower", "upper",
-   "alnum", "ascii", "blank", "cntrl", "digit", "graph",
+ typedef struct verbitem {
-   "print", "punct", "space", "word",  "xdigit" };
+   int   len;                 /* 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);
- static const uschar posix_name_lengths[] = {
+ /* 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 pcre_uint8 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. Each class is formed from a
-Line 179 
 static const int posix_class_maps[] = {
+Line 295 
 static const int posix_class_maps[] = {
    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. */
+ #ifdef SUPPORT_UCP
+ static const pcre_uchar string_PNd[]  = {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_pNd[]  = {
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_N, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_PXsp[] = {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_pXsp[] = {
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_s, CHAR_p, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_PXwd[] = {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_pXwd[] = {
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_w, CHAR_d, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar *substitutes[] = {
+   string_PNd,           /* \D */
+   string_pNd,           /* \d */
+   string_PXsp,          /* \S */       /* NOTE: Xsp is Perl space */
+   string_pXsp,          /* \s */
+   string_PXwd,          /* \W */
+   string_pXwd           /* \w */
+ };
+ static const pcre_uchar string_pL[] =   {
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_pLl[] =  {
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_pLu[] =  {
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_pXan[] = {
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_h[] =    {
+   CHAR_BACKSLASH, CHAR_h, '\0' };
+ static const pcre_uchar string_pXps[] = {
+   CHAR_BACKSLASH, CHAR_p, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_PL[] =   {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_L, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_PLl[] =  {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_L, CHAR_l, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_PLu[] =  {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_L, CHAR_u, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_PXan[] = {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_a, CHAR_n, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar string_H[] =    {
+   CHAR_BACKSLASH, CHAR_H, '\0' };
+ static const pcre_uchar string_PXps[] = {
+   CHAR_BACKSLASH, CHAR_P, CHAR_LEFT_CURLY_BRACKET,
+   CHAR_X, CHAR_p, CHAR_s, CHAR_RIGHT_CURLY_BRACKET, '\0' };
+ static const pcre_uchar *posix_substitutes[] = {
+   string_pL,            /* alpha */
+   string_pLl,           /* lower */
+   string_pLu,           /* upper */
+   string_pXan,          /* alnum */
+   NULL,                 /* ascii */
+   string_h,             /* blank */
+   NULL,                 /* cntrl */
+   string_pNd,           /* digit */
+   NULL,                 /* graph */
+   NULL,                 /* print */
+   NULL,                 /* punct */
+   string_pXps,          /* space */    /* NOTE: Xps is POSIX space */
+   string_pXwd,          /* word */
+   NULL,                 /* xdigit */
+   /* Negated cases */
+   string_PL,            /* ^alpha */
+   string_PLl,           /* ^lower */
+   string_PLu,           /* ^upper */
+   string_PXan,          /* ^alnum */
+   NULL,                 /* ^ascii */
+   string_H,             /* ^blank */
+   NULL,                 /* ^cntrl */
+   string_PNd,           /* ^digit */
+   NULL,                 /* ^graph */
+   NULL,                 /* ^print */
+   NULL,                 /* ^punct */
+   string_PXps,          /* ^space */   /* NOTE: Xps is POSIX space */
+   string_PXwd,          /* ^word */
+   NULL                  /* ^xdigit */
+ };
+ #define POSIX_SUBSIZE (sizeof(posix_substitutes) / sizeof(pcre_uchar *))
+ #endif
  #define STRING(a)  # a
  #define XSTRING(s) STRING(s)
-Line 186 
 static const int posix_class_maps[] = {
+Line 403 
 static const int posix_class_maps[] = {
  /* The texts of compile-time error messages. These are "char *" because they
  are passed to the outside world. Do not ever re-use any error number, because
  they are documented. Always add a new error instead. Messages marked DEAD below
- are no longer used. */
+ are no longer used. This used to be a table of strings, but in order to reduce
+ the number of relocations needed when a shared library is loaded dynamically,
- static const char *error_texts[] = {
+ it is now one long string. We cannot use a table of offsets, because the
-   "no error",
+ lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
-   "\\ at end of pattern",
+ simply count through to the one we want - this isn't a performance issue
-   "\\c at end of pattern",
+ because these strings are used only when there is a compilation error.
-   "unrecognized character follows \\",
-   "numbers out of order in {} quantifier",
+ 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",  /** DEAD **/
+   "operand of unlimited repeat could match the empty string\0"  /** DEAD **/
-   "internal error: unexpected repeat",
+   "internal error: unexpected repeat\0"
-   "unrecognized character after (?",
+   "unrecognized character after (? or (?-\0"
-   "POSIX named classes are supported only within a class",
+   "POSIX named classes are supported only within a class\0"
-   "missing )",
+   "missing )\0"
    /* 15 */
-   "reference to non-existent subpattern",
+   "reference to non-existent subpattern\0"
-   "erroffset passed as NULL",
+   "erroffset passed as NULL\0"
-   "unknown option bit(s) set",
+   "unknown option bit(s) set\0"
-   "missing ) after comment",
+   "missing ) after comment\0"
-   "parentheses nested too deeply",  /** DEAD **/
+   "parentheses nested too deeply\0"  /** DEAD **/
    /* 20 */
-   "regular expression is too large",
+   "regular expression is too large\0"
-   "failed to get memory",
+   "failed to get memory\0"
-   "unmatched parentheses",
+   "unmatched parentheses\0"
-   "internal error: code overflow",
+   "internal error: code overflow\0"
-   "unrecognized character after (?<",
+   "unrecognized character after (?<\0"
    /* 25 */
-   "lookbehind assertion is not fixed length",
+   "lookbehind assertion is not fixed length\0"
-   "malformed number or name after (?(",
+   "malformed number or name after (?(\0"
-   "conditional group contains more than two branches",
+   "conditional group contains more than two branches\0"
-   "assertion expected after (?(",
+   "assertion expected after (?(\0"
-   "(?R or (?[+-]digits must be followed by )",
+   "(?R or (?[+-]digits must be followed by )\0"
    /* 30 */
-   "unknown POSIX class name",
+   "unknown POSIX class name\0"
-   "POSIX collating elements are not supported",
+   "POSIX collating elements are not supported\0"
-   "this version of PCRE is not compiled with PCRE_UTF8 support",
+   "this version of PCRE is compiled without UTF support\0"
-   "spare error",  /** DEAD **/
+   "spare error\0"  /** DEAD **/
-   "character value in \\x{...} sequence is too large",
+   "character value in \\x{...} sequence is too large\0"
    /* 35 */
-   "invalid condition (?(0)",
+   "invalid condition (?(0)\0"
-   "\\C not allowed in lookbehind assertion",
+   "\\C not allowed in lookbehind assertion\0"
-   "PCRE does not support \\L, \\l, \\N, \\U, or \\u",
+   "PCRE does not support \\L, \\l, \\N{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 in subpattern name (missing terminator)",
+   "syntax error in subpattern name (missing terminator)\0"
-   "two named subpatterns have the same name",
+   "two named subpatterns have the same name\0"
-   "invalid UTF-8 string",
+   "invalid UTF-8 string\0"
    /* 45 */
-   "support for \\P, \\p, and \\X has not been compiled",
+   "support for \\P, \\p, and \\X has not been compiled\0"
-   "malformed \\P or \\p sequence",
+   "malformed \\P or \\p sequence\0"
-   "unknown property name after \\P or \\p",
+   "unknown property name after \\P or \\p\0"
-   "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)",
+   "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
-   "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")",
+   "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
    /* 50 */
-   "repeated subpattern is too long",    /** DEAD **/
+   "repeated subpattern is too long\0"    /** DEAD **/
-   "octal value is greater than \\377 (not in UTF-8 mode)",
+   "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0"
-   "internal error: overran compiling workspace",
+   "internal error: overran compiling workspace\0"
-   "internal error: previously-checked referenced subpattern not found",
+   "internal error: previously-checked referenced subpattern not found\0"
-   "DEFINE group contains more than one branch",
+   "DEFINE group contains more than one branch\0"
    /* 55 */
-   "repeating a DEFINE group is not allowed",
+   "repeating a DEFINE group is not allowed\0"  /** DEAD **/
-   "inconsistent NEWLINE options",
+   "inconsistent NEWLINE options\0"
-   "\\g is not followed by a braced name or an optionally braced non-zero number",
+   "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
-   "(?+ or (?- or (?(+ or (?(- must be followed by a non-zero number"
+   "a numbered reference must not be zero\0"
- };
+   "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 Unicode property support\0"
+   "\\c must be followed by an ASCII character\0"
+   "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
+   /* 70 */
+   "internal error: unknown opcode in find_fixedlength()\0"
+   "\\N is not supported in a class\0"
+   "too many forward references\0"
+   "disallowed Unicode code point (>= 0xd800 && <= 0xdfff)\0"
+   "invalid UTF-16 string\0"
+   /* 75 */
+   "name is too long in (*MARK), (*PRUNE), (*SKIP), or (*THEN)\0"
+   "character value in \\u.... sequence is too large\0"
+   "invalid UTF-32 string\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 278 
 For convenience, we use the same bit def
+Line 526 
 For convenience, we use the same bit def
  Then we can use ctype_digit and ctype_xdigit in the code. */
- #ifndef EBCDIC  /* This is the "normal" case, for ASCII systems */
+ /* Using a simple comparison for decimal numbers rather than a memory read
- static const unsigned char digitab[] =
+ is much faster, and the resulting code is simpler (the compiler turns it
+ into a subtraction and unsigned comparison). */
+ #define IS_DIGIT(x) ((x) >= CHAR_0 && (x) <= CHAR_9)
+ #ifndef EBCDIC
+ /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
+ UTF-8 mode. */
+ static const pcre_uint8 digitab[] =
    {
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   8- 15 */
-Line 314 
 static const unsigned char digitab[] =
+Line 572 
 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
- static const unsigned char digitab[] =
+ /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
+ static const pcre_uint8 digitab[] =
    {
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7  0 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   8- 15    */
-Line 350 
 static const unsigned char digitab[] =
+Line 611 
 static const unsigned char digitab[] =
 x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /*  0 - 7  F0 */
 x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/*  8 -255    */
- static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
+ static const pcre_uint8 ebcdic_chartab[] = { /* chartable partial dup */
 x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /*   0-  7 */
 x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /*   8- 15 */
 x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /*  16- 23 */
-Line 386 
 static const unsigned char ebcdic_charta
+Line 647 
 static const unsigned char ebcdic_charta
  #endif
- /* Definition to allow mutual recursion */
+ /*************************************************
+ *            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++ != CHAR_NULL) {};
+   if (*s == CHAR_NULL) return "Error text not found (please report)";
+   }
+ return s;
+ }
+ /*************************************************
+ *           Expand the workspace                 *
+ *************************************************/
+ /* This function is called during the second compiling phase, if the number of
+ forward references fills the existing workspace, which is originally a block on
+ the stack. A larger block is obtained from malloc() unless the ultimate limit
+ has been reached or the increase will be rather small.
+ Argument: pointer to the compile data block
+ Returns:  0 if all went well, else an error number
+ */
+ static int
+ expand_workspace(compile_data *cd)
+ {
+ pcre_uchar *newspace;
+ int newsize = cd->workspace_size * 2;
+ if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
+ if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
+     newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
+  return ERR72;
+ newspace = (PUBL(malloc))(IN_UCHARS(newsize));
+ if (newspace == NULL) return ERR21;
+ memcpy(newspace, cd->start_workspace, cd->workspace_size * sizeof(pcre_uchar));
+ cd->hwm = (pcre_uchar *)newspace + (cd->hwm - cd->start_workspace);
+ if (cd->workspace_size > COMPILE_WORK_SIZE)
+   (PUBL(free))((void *)cd->start_workspace);
+ cd->start_workspace = newspace;
+ cd->workspace_size = newsize;
+ return 0;
+ }
+ /*************************************************
+ *            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
-   compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
+ is_counted_repeat(const pcre_uchar *p)
-     int *, int *, branch_chain *, compile_data *, int *);
+ {
+ if (!IS_DIGIT(*p)) return FALSE;
+ p++;
+ while (IS_DIGIT(*p)) p++;
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
+ if (*p++ != CHAR_COMMA) return FALSE;
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
+ if (!IS_DIGIT(*p)) return FALSE;
+ p++;
+ while (IS_DIGIT(*p)) p++;
+ return (*p == CHAR_RIGHT_CURLY_BRACKET);
+ }
-Line 399 
 static BOOL
+Line 751 
 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
+ positive value for a simple escape such as \n, or 0 for a data character
- encodes one of the more complicated things such as \d. A backreference to group
+ which will be placed in chptr. A backreference to group n is returned as
- n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
+ negative n. When UTF-8 is enabled, a positive value greater than 255 may
- UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
+ be returned in chptr.
- ptr is pointing at the \. On exit, it is on the final character of the escape
+ On entry,ptr is pointing at the \. On exit, it is on the final character of the
- sequence.
+ escape sequence.
  Arguments:
    ptrptr         points to the pattern position pointer
+   chptr          points to the data character
    errorcodeptr   points to the errorcode variable
    bracount       number of previous extracting brackets
    options        the options bits
    isclass        TRUE if inside a character class
- Returns:         zero or positive => a data character
+ Returns:         zero => a data character
-                  negative => a special escape sequence
+                  positive => a special escape sequence
-                  on error, errorptr is set
+                  negative => a back reference
+                  on error, errorcodeptr is set
  */
  static int
- check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
+ check_escape(const pcre_uchar **ptrptr, pcre_uint32 *chptr, int *errorcodeptr,
-   int options, BOOL isclass)
+   int bracount, int options, BOOL isclass)
  {
- BOOL utf8 = (options & PCRE_UTF8) != 0;
+ /* PCRE_UTF16 has the same value as PCRE_UTF8. */
- const uschar *ptr = *ptrptr + 1;
+ BOOL utf = (options & PCRE_UTF8) != 0;
- int c, i;
+ const pcre_uchar *ptr = *ptrptr + 1;
+ pcre_uint32 c;
+ int escape = 0;
+ int 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. */
- if (c == 0) *errorcodeptr = ERR1;
+ if (c == CHAR_NULL) *errorcodeptr = ERR1;
- /* Non-alphamerics are literals. For digits or letters, do an initial lookup in
+ /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
- a table. A non-zero result is something that can be returned immediately.
+ in a table. A non-zero result is something that can be returned immediately.
  Otherwise further processing may be required. */
- #ifndef EBCDIC  /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
- else if (c < '0' || c > 'z') {}                           /* Not alphameric */
+ /* Not alphanumeric */
- else if ((i = escapes[c - '0']) != 0) c = i;
+ else if (c < CHAR_0 || c > CHAR_z) {}
+ else if ((i = escapes[c - CHAR_0]) != 0) { if (i > 0) c = (pcre_uint32)i; else escape = -i; }
  #else           /* EBCDIC coding */
- else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphameric */
+ /* Not alphanumeric */
- else if ((i = escapes[c - 0x48]) != 0)  c = i;
+ else if (c < CHAR_a || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
+ else if ((i = escapes[c - 0x48]) != 0)  { if (i > 0) c = (pcre_uint32)i; else escape = -i; }
  #endif
  /* Escapes that need further processing, or are illegal. */
  else
    {
-   const uschar *oldptr;
+   const pcre_uchar *oldptr;
-   BOOL braced, negated;
+   BOOL braced, negated, overflow;
+   int s;
    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 'u':
-     case 'U':
      *errorcodeptr = ERR37;
      break;
-     /* \g must be followed by a number, either plain or braced. If positive, it
+     case CHAR_u:
-     is an absolute backreference. If negative, it is a relative backreference.
+     if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
-     This is a Perl 5.10 feature. Perl 5.10 also supports \g{name} as a
+       {
-     reference to a named group. This is part of Perl's movement towards a
+       /* In JavaScript, \u must be followed by four hexadecimal numbers.
-     unified syntax for back references. As this is synonymous with \k{name}, we
+       Otherwise it is a lowercase u letter. */
-     fudge it up by pretending it really was \k. */
+       if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
+         && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
-     case 'g':
+         && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
-     if (ptr[1] == '{')
+         && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
-       {
+         {
-       const uschar *p;
+         c = 0;
-       for (p = ptr+2; *p != 0 && *p != '}'; p++)
+         for (i = 0; i < 4; ++i)
-         if (*p != '-' && (digitab[*p] & ctype_digit) == 0) break;
+           {
-       if (*p != 0 && *p != '}')
+           register pcre_uint32 cc = *(++ptr);
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
+           if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
+           c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
+ #else           /* EBCDIC coding */
+           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 defined COMPILE_PCRE8
+         if (c > (utf ? 0x10ffff : 0xff))
+ #elif defined COMPILE_PCRE16
+         if (c > (utf ? 0x10ffff : 0xffff))
+ #elif defined COMPILE_PCRE32
+         if (utf && c > 0x10ffff)
+ #endif
+           {
+           *errorcodeptr = ERR76;
+           }
+         else if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
+         }
+       }
+     else
+       *errorcodeptr = ERR37;
+     break;
+     case CHAR_U:
+     /* In JavaScript, \U is an uppercase U letter. */
+     if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *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)
+       {
+       escape = ESC_g;
+       break;
+       }
+     /* Handle the Perl-compatible cases */
+     if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
+       {
+       const pcre_uchar *p;
+       for (p = ptr+2; *p != CHAR_NULL && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
+         if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
+       if (*p != CHAR_NULL && *p != CHAR_RIGHT_CURLY_BRACKET)
          {
-         c = -ESC_k;
+         escape = ESC_k;
          break;
          }
        braced = TRUE;
-Line 489 
 else
+Line 910 
 else
        }
      else braced = FALSE;
-     if (ptr[1] == '-')
+     if (ptr[1] == CHAR_MINUS)
        {
        negated = TRUE;
        ptr++;
        }
      else negated = FALSE;
-     c = 0;
+     /* The integer range is limited by the machine's int representation. */
-     while ((digitab[ptr[1]] & ctype_digit) != 0)
+     s = 0;
-       c = c * 10 + *(++ptr) - '0';
+     overflow = FALSE;
+     while (IS_DIGIT(ptr[1]))
+       {
+       if (s > INT_MAX / 10 - 1) /* Integer overflow */
+         {
+         overflow = TRUE;
+         break;
+         }
+       s = s * 10 + (int)(*(++ptr) - CHAR_0);
+       }
+     if (overflow) /* Integer overflow */
+       {
+       while (IS_DIGIT(ptr[1]))
+         ptr++;
+       *errorcodeptr = ERR61;
+       break;
+       }
-     if (c == 0 || (braced && *(++ptr) != '}'))
+     if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
        {
        *errorcodeptr = ERR57;
-       return 0;
+       break;
+       }
+     if (s == 0)
+       {
+       *errorcodeptr = ERR58;
+       break;
        }
      if (negated)
        {
-       if (c > bracount)
+       if (s > bracount)
          {
          *errorcodeptr = ERR15;
-         return 0;
+         break;
          }
-       c = bracount - (c - 1);
+       s = bracount - (s - 1);
        }
-     c = -(ESC_REF + c);
+     escape = -s;
      break;
      /* The handling of escape sequences consisting of a string of digits
-Line 531 
 else
+Line 974 
 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';
+       /* The integer range is limited by the machine's int representation. */
-       while ((digitab[ptr[1]] & ctype_digit) != 0)
+       s = (int)(c -CHAR_0);
-         c = c * 10 + *(++ptr) - '0';
+       overflow = FALSE;
-       if (c < 10 || c <= bracount)
+       while (IS_DIGIT(ptr[1]))
+         {
+         if (s > INT_MAX / 10 - 1) /* Integer overflow */
+           {
+           overflow = TRUE;
+           break;
+           }
+         s = s * 10 + (int)(*(++ptr) - CHAR_0);
+         }
+       if (overflow) /* Integer overflow */
+         {
+         while (IS_DIGIT(ptr[1]))
+           ptr++;
+         *errorcodeptr = ERR61;
+         break;
+         }
+       if (s < 10 || s <= bracount)
          {
-         c = -(ESC_REF + c);
+         escape = -s;
          break;
          }
        ptr = oldptr;      /* Put the pointer back and fall through */
-Line 552 
 else
+Line 1011 
 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 562 
 else
+Line 1021 
 else
      /* \0 always starts an octal number, but we may drop through to here with a
      larger first octal digit. The original code used just to take the least
      significant 8 bits of octal numbers (I think this is what early Perls used
-     to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
+     to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
-     than 3 octal digits. */
+     but no more than 3 octal digits. */
-     case '0':
+     case CHAR_0:
-     c -= '0';
+     c -= CHAR_0;
-     while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7')
+     while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
-         c = c * 8 + *(++ptr) - '0';
+         c = c * 8 + *(++ptr) - CHAR_0;
-     if (!utf8 && c > 255) *errorcodeptr = ERR51;
+ #ifdef COMPILE_PCRE8
+     if (!utf && c > 0xff) *errorcodeptr = ERR51;
+ #endif
      break;
      /* \x is complicated. \x{ddd} is a character number which can be greater
-     than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
+     than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
-     treated as a data character. */
+     If not, { is treated as a data character. */
+     case CHAR_x:
+     if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
+       {
+       /* In JavaScript, \x must be followed by two hexadecimal numbers.
+       Otherwise it is a lowercase x letter. */
+       if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
+         && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0)
+         {
+         c = 0;
+         for (i = 0; i < 2; ++i)
+           {
+           register pcre_uint32 cc = *(++ptr);
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
+           if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
+           c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
+ #else           /* EBCDIC coding */
+           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
+           }
+         }
+       break;
+       }
-     case 'x':
+     if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
-     if (ptr[1] == '{')
        {
-       const uschar *pt = ptr + 2;
+       const pcre_uchar *pt = ptr + 2;
-       int count = 0;
        c = 0;
-       while ((digitab[*pt] & ctype_xdigit) != 0)
+       overflow = FALSE;
+       while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
          {
-         register int cc = *pt++;
+         register pcre_uint32 cc = *pt++;
-         if (c == 0 && cc == '0') continue;     /* Leading zeroes */
+         if (c == 0 && cc == CHAR_0) continue;     /* Leading zeroes */
-         count++;
+ #ifdef COMPILE_PCRE32
- #ifndef EBCDIC  /* ASCII coding */
+         if (c >= 0x10000000l) { overflow = TRUE; break; }
-         if (cc >= 'a') cc -= 32;               /* Convert to upper case */
+ #endif
-         c = (c << 4) + cc - ((cc < 'A')? '0' : ('A' - 10));
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
+         if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
+         c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
  #else           /* EBCDIC coding */
-         if (cc >= 'a' && cc <= 'z') cc += 64;  /* Convert to upper case */
+         if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
-         c = (c << 4) + cc - ((cc >= '0')? '0' : ('A' - 10));
+         c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
  #endif
+ #if defined COMPILE_PCRE8
+         if (c > (utf ? 0x10ffff : 0xff)) { overflow = TRUE; break; }
+ #elif defined COMPILE_PCRE16
+         if (c > (utf ? 0x10ffff : 0xffff)) { overflow = TRUE; break; }
+ #elif defined COMPILE_PCRE32
+         if (utf && c > 0x10ffff) { overflow = TRUE; break; }
+ #endif
+         }
+       if (overflow)
+         {
+         while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
+         *errorcodeptr = ERR34;
          }
-       if (*pt == '}')
+       if (*pt == CHAR_RIGHT_CURLY_BRACKET)
          {
-         if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
+         if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
          ptr = pt;
          break;
          }
-Line 612 
 else
+Line 1113 
 else
      /* Read just a single-byte hex-defined char */
      c = 0;
-     while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
+     while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
        {
-       int cc;                               /* Some compilers don't like ++ */
+       pcre_uint32 cc;                          /* Some compilers don't like */
-       cc = *(++ptr);                        /* in initializers */
+       cc = *(++ptr);                           /* ++ in initializers */
- #ifndef EBCDIC  /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
-       if (cc >= 'a') cc -= 32;              /* Convert to upper case */
+       if (cc >= CHAR_a) cc -= 32;              /* Convert to upper case */
-       c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));
+       c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
  #else           /* EBCDIC coding */
-       if (cc <= 'z') cc += 64;              /* Convert to upper case */
+       if (cc <= CHAR_z) cc += 64;              /* Convert to upper case */
-       c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));
+       c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
  #endif
        }
      break;
      /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
-     This coding is ASCII-specific, but then the whole concept of \cx is
+     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)
+     if (c == CHAR_NULL)
        {
        *errorcodeptr = ERR2;
-       return 0;
+       break;
        }
+ #ifndef EBCDIC    /* ASCII/UTF-8 coding */
- #ifndef EBCDIC  /* ASCII coding */
+     if (c > 127)  /* Excludes all non-ASCII in either mode */
-     if (c >= 'a' && c <= 'z') c -= 32;
+       {
+       *errorcodeptr = ERR68;
+       break;
+       }
+     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 664 
 else
+Line 1170 
 else
      }
    }
- *ptrptr = ptr;
+ /* Perl supports \N{name} for character names, as well as plain \N for "not
- return c;
+ newline". PCRE does not support \N{name}. However, it does support
- }
+ quantification such as \N{2,3}. */
+ if (escape == 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 && escape >= ESC_D && escape <= ESC_w)
+   escape += (ESC_DU - ESC_D);
+ /* Set the pointer to the final character before returning. */
+ *ptrptr = ptr;
+ *chptr = c;
+ return escape;
+ }
  #ifdef SUPPORT_UCP
  /*************************************************
-Line 683 
 escape sequence.
+Line 1203 
 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
+   ptypeptr       points to an unsigned int that is set to the type value
+   pdataptr       points to an unsigned int that is set to the detailed property value
    errorcodeptr   points to the error code variable
- Returns:         type value from ucp_type_table, or -1 for an invalid type
+ Returns:         TRUE if the type value was found, or FALSE for an invalid type
  */
- static int
+ static BOOL
- get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
+ get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, unsigned int *ptypeptr,
+   unsigned int *pdataptr, int *errorcodeptr)
  {
- int c, i, bot, top;
+ pcre_uchar c;
- const uschar *ptr = *ptrptr;
+ int i, bot, top;
- char name[32];
+ const pcre_uchar *ptr = *ptrptr;
+ pcre_uchar name[32];
  c = *(++ptr);
- if (c == 0) goto ERROR_RETURN;
+ if (c == CHAR_NULL) goto ERROR_RETURN;
  *negptr = FALSE;
  /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
  negation. */
- if (c == '{')
+ if (c == CHAR_LEFT_CURLY_BRACKET)
    {
-   if (ptr[1] == '^')
+   if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
      {
      *negptr = TRUE;
      ptr++;
      }
-   for (i = 0; i < (int)sizeof(name) - 1; i++)
+   for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
      {
      c = *(++ptr);
-     if (c == 0) goto ERROR_RETURN;
+     if (c == CHAR_NULL) goto ERROR_RETURN;
-     if (c == '}') break;
+     if (c == CHAR_RIGHT_CURLY_BRACKET) break;
      name[i] = c;
      }
-   if (c !='}') goto ERROR_RETURN;
+   if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
    name[i] = 0;
    }
-Line 735 
 else
+Line 1258 
 else
  /* Search for a recognized property name using binary chop */
  bot = 0;
- top = _pcre_utt_size;
+ top = PRIV(utt_size);
  while (bot < top)
    {
+   int r;
    i = (bot + top) >> 1;
-   c = strcmp(name, _pcre_utt[i].name);
+   r = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
-   if (c == 0)
+   if (r == 0)
      {
-     *dptr = _pcre_utt[i].value;
+     *ptypeptr = PRIV(utt)[i].type;
-     return _pcre_utt[i].type;
+     *pdataptr = PRIV(utt)[i].value;
+     return TRUE;
      }
-   if (c > 0) bot = i + 1; else top = i;
+   if (r > 0) bot = i + 1; else top = i;
    }
  *errorcodeptr = ERR47;
  *ptrptr = ptr;
- return -1;
+ return FALSE;
  ERROR_RETURN:
  *errorcodeptr = ERR46;
  *ptrptr = ptr;
- return -1;
+ return FALSE;
  }
  #endif
-Line 764 
 return -1;
+Line 1289 
 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 815 
 Returns:         pointer to '}' on succe
+Line 1307 
 Returns:         pointer to '}' on succe
                   current ptr on error, with errorcodeptr set non-zero
  */
- static const uschar *
+ static const pcre_uchar *
- read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
+ read_repeat_counts(const pcre_uchar *p, int *minp, int *maxp, int *errorcodeptr)
  {
  int min = 0;
  int max = -1;
-Line 824 
 int max = -1;
+Line 1316 
 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 (IS_DIGIT(*p)) min = min * 10 + (int)(*p++ - CHAR_0);
  if (min < 0 || min > 65535)
    {
    *errorcodeptr = ERR5;
-Line 834 
 if (min < 0 || min > 65535)
+Line 1326 
 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(IS_DIGIT(*p)) max = max * 10 + (int)(*p++ - CHAR_0);
      if (max < 0 || max > 65535)
        {
        *errorcodeptr = ERR5;
-Line 864 
 return p;
+Line 1356 
 return p;
  /*************************************************
- *       Find forward referenced subpattern       *
+ *  Subroutine for finding forward reference      *
  *************************************************/
- /* This function scans along a pattern's text looking for capturing
+ /* This recursive function is called only from find_parens() below. The
+ top-level call starts at the beginning of the pattern. All other calls must
+ start at a parenthesis. It scans along a pattern's text looking for capturing
  subpatterns, and counting them. If it finds a named pattern that matches the
  name it is given, it returns its number. Alternatively, if the name is NULL, it
- returns when it reaches a given numbered subpattern. This is used for forward
+ returns when it reaches a given numbered subpattern. Recursion is used to keep
- references to subpatterns. We know that if (?P< is encountered, the name will
+ track of subpatterns that reset the capturing group numbers - the (?| feature.
- be terminated by '>' because that is checked in the first pass.
+ 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:
-   ptr          current position in the pattern
+   ptrptr       address of the current character pointer (updated)
-   count        current count of capturing parens so far encountered
+   cd           compile background data
    name         name to seek, or NULL if seeking a numbered subpattern
    lorn         name length, or subpattern number if name is NULL
    xmode        TRUE if we are in /x mode
+   utf          TRUE if we are in UTF-8 / UTF-16 / UTF-32 mode
+   count        pointer to the current capturing subpattern number (updated)
  Returns:       the number of the named subpattern, or -1 if not found
  */
  static int
- find_parens(const uschar *ptr, int count, const uschar *name, int lorn,
+ find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
-   BOOL xmode)
+   BOOL xmode, BOOL utf, int *count)
  {
- const uschar *thisname;
+ pcre_uchar *ptr = *ptrptr;
+ int start_count = *count;
+ int hwm_count = start_count;
+ BOOL dup_parens = FALSE;
- for (; *ptr != 0; ptr++)
+ /* If the first character is a parenthesis, check on the type of group we are
+ dealing with. The very first call may not start with a parenthesis. */
+ if (ptr[0] == CHAR_LEFT_PARENTHESIS)
    {
-   int term;
+   /* Handle specials such as (*SKIP) or (*UTF8) etc. */
+   if (ptr[1] == CHAR_ASTERISK) ptr += 2;
+   /* Handle a normal, unnamed capturing parenthesis. */
+   else if (ptr[1] != CHAR_QUESTION_MARK)
+     {
+     *count += 1;
+     if (name == NULL && *count == lorn) return *count;
+     ptr++;
+     }
+   /* 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)
+     {
+     for (ptr += 3; *ptr != CHAR_NULL; ptr++)
+       if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
+     goto FAIL_EXIT;
+     }
+   /* Handle a condition. If it is an assertion, just carry on so that it
+   is processed as normal. If not, skip to the closing parenthesis of the
+   condition (there can't be any nested parens). */
+   else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
+     {
+     ptr += 2;
+     if (ptr[1] != CHAR_QUESTION_MARK)
+       {
+       while (*ptr != CHAR_NULL && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
+       if (*ptr != CHAR_NULL) ptr++;
+       }
+     }
+   /* Start with (? but not a condition. */
+   else
+     {
+     ptr += 2;
+     if (*ptr == CHAR_P) ptr++;                      /* Allow optional P */
+     /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
+     if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
+         ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
+       {
+       pcre_uchar term;
+       const pcre_uchar *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 == (int)(ptr - thisname) &&
+           STRNCMP_UC_UC(name, thisname, (unsigned int)lorn) == 0)
+         return *count;
+       term++;
+       }
+     }
+   }
+ /* 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++)
+   {
    /* Skip over backslashed characters and also entire \Q...\E */
-   if (*ptr == '\\')
+   if (*ptr == CHAR_BACKSLASH)
      {
-     if (*(++ptr) == 0) return -1;
+     if (*(++ptr) == CHAR_NULL) goto FAIL_EXIT;
-     if (*ptr == 'Q') for (;;)
+     if (*ptr == CHAR_Q) for (;;)
        {
-       while (*(++ptr) != 0 && *ptr != '\\');
+       while (*(++ptr) != CHAR_NULL && *ptr != CHAR_BACKSLASH) {};
-       if (*ptr == 0) return -1;
+       if (*ptr == CHAR_NULL) goto FAIL_EXIT;
-       if (*(++ptr) == 'E') break;
+       if (*(++ptr) == CHAR_E) break;
        }
      continue;
      }
-   /* Skip over character classes */
+   /* Skip over character classes; this logic must be similar to the way they
+   are handled for real. If the first character is '^', skip it. Also, if the
+   first few characters (either before or after ^) are \Q\E or \E we skip them
+   too. This makes for compatibility with Perl. Note the use of STR macros to
+   encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
-   if (*ptr == '[')
+   if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
      {
-     while (*(++ptr) != ']')
+     BOOL negate_class = FALSE;
+     for (;;)
+       {
+       if (ptr[1] == CHAR_BACKSLASH)
+         {
+         if (ptr[2] == CHAR_E)
+           ptr+= 2;
+         else if (STRNCMP_UC_C8(ptr + 2,
+                  STR_Q STR_BACKSLASH STR_E, 3) == 0)
+           ptr += 4;
+         else
+           break;
+         }
+       else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
+         {
+         negate_class = TRUE;
+         ptr++;
+         }
+       else break;
+       }
+     /* If the next character is ']', it is a data character that must be
+     skipped, except in JavaScript compatibility mode. */
+     if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
+         (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
+       ptr++;
+     while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
        {
-       if (*ptr == '\\')
+       if (*ptr == CHAR_NULL) return -1;
+       if (*ptr == CHAR_BACKSLASH)
          {
-         if (*(++ptr) == 0) return -1;
+         if (*(++ptr) == CHAR_NULL) goto FAIL_EXIT;
-         if (*ptr == 'Q') for (;;)
+         if (*ptr == CHAR_Q) for (;;)
            {
-           while (*(++ptr) != 0 && *ptr != '\\');
+           while (*(++ptr) != CHAR_NULL && *ptr != CHAR_BACKSLASH) {};
-           if (*ptr == 0) return -1;
+           if (*ptr == CHAR_NULL) goto FAIL_EXIT;
-           if (*(++ptr) == 'E') break;
+           if (*(++ptr) == CHAR_E) break;
            }
          continue;
          }
-Line 931 
 for (; *ptr != 0; ptr++)
+Line 1556 
 for (; *ptr != 0; ptr++)
    /* Skip comments in /x mode */
-   if (xmode && *ptr == '#')
+   if (xmode && *ptr == CHAR_NUMBER_SIGN)
      {
-     while (*(++ptr) != 0 && *ptr != '\n');
+     ptr++;
-     if (*ptr == 0) return -1;
+     while (*ptr != CHAR_NULL)
+       {
+       if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
+       ptr++;
+ #ifdef SUPPORT_UTF
+       if (utf) FORWARDCHAR(ptr);
+ #endif
+       }
+     if (*ptr == CHAR_NULL) goto FAIL_EXIT;
      continue;
      }
-   /* An opening parens must now be a real metacharacter */
+   /* Check for the special metacharacters */
-   if (*ptr != '(') continue;
+   if (*ptr == CHAR_LEFT_PARENTHESIS)
-   if (ptr[1] != '?')
      {
-     count++;
+     int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
-     if (name == NULL && count == lorn) return count;
+     if (rc > 0) return rc;
-     continue;
+     if (*ptr == CHAR_NULL) goto FAIL_EXIT;
      }
-   ptr += 2;
+   else if (*ptr == CHAR_RIGHT_PARENTHESIS)
-   if (*ptr == 'P') ptr++;                      /* Allow optional P */
+     {
+     if (dup_parens && *count < hwm_count) *count = hwm_count;
-   /* We have to disambiguate (?<! and (?<= from (?<name> */
+     goto FAIL_EXIT;
+     }
-   if ((*ptr != '<' || ptr[1] == '!' || ptr[1] == '=') &&
+   else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
-        *ptr != '\'')
+     {
-     continue;
+     if (*count > hwm_count) hwm_count = *count;
+     *count = start_count;
+     }
+   }
-   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
+   utf          TRUE if we are in UTF-8 / UTF-16 / UTF-32 mode
+ Returns:       the number of the found subpattern, or -1 if not found
+ */
+ static int
+ find_parens(compile_data *cd, const pcre_uchar *name, int lorn, BOOL xmode,
+   BOOL utf)
+ {
+ pcre_uchar *ptr = (pcre_uchar *)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. */
-   if (name == NULL && count == lorn) return count;
+ for (;;)
-   term = *ptr++;
+   {
-   if (term == '<') term = '>';
+   rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
-   thisname = ptr;
+   if (rc > 0 || *ptr++ == CHAR_NULL) break;
-   while (*ptr != term) ptr++;
-   if (name != NULL && lorn == ptr - thisname &&
-       strncmp((const char *)name, (const char *)thisname, lorn) == 0)
-     return count;
    }
- return -1;
+ 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, a change of option is important.
+ that do not influence this. For some calls, it makes sense to skip negative
- For some calls, it makes sense to skip negative forward and all backward
+ forward and all backward assertions, and also the \b assertion; for others it
- assertions, and also the \b assertion; for others it does not.
+ does not.
  Arguments:
    code         pointer to the start of the group
-   options      pointer to external options
-   optbit       the option bit whose changing is significant, or
-                  zero if none are
    skipassert   TRUE if certain assertions are to be skipped
  Returns:       pointer to the first significant opcode
  */
- static const uschar*
+ static const pcre_uchar*
- first_significant_code(const uschar *code, int *options, int optbit,
+ first_significant_code(const pcre_uchar *code, BOOL skipassert)
-   BOOL skipassert)
  {
  for (;;)
    {
    switch ((int)*code)
      {
-     case OP_OPT:
-     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
-       *options = (int)code[1];
-     code += 2;
-     break;
      case OP_ASSERT_NOT:
      case OP_ASSERTBACK:
      case OP_ASSERTBACK_NOT:
      if (!skipassert) return code;
      do code += GET(code, 1); while (*code == OP_ALT);
-     code += _pcre_OP_lengths[*code];
+     code += PRIV(OP_lengths)[*code];
      break;
      case OP_WORD_BOUNDARY:
-Line 1023 
 for (;;)
+Line 1690 
 for (;;)
      case OP_CALLOUT:
      case OP_CREF:
+     case OP_NCREF:
      case OP_RREF:
+     case OP_NRREF:
      case OP_DEF:
-     code += _pcre_OP_lengths[*code];
+     code += PRIV(OP_lengths)[*code];
      break;
      default:
-Line 1039 
 for (;;)
+Line 1708 
 for (;;)
  /*************************************************
- *        Find the fixed length of a pattern      *
+ *        Find the fixed length of a branch       *
  *************************************************/
- /* Scan a pattern and compute the fixed length of subject that will match it,
+ /* Scan a branch and compute the fixed length of subject that will match it,
  if the length is fixed. This is needed for dealing with backward assertions.
- In UTF8 mode, the result is in characters rather than bytes.
+ In UTF8 mode, the result is in characters rather than bytes. The branch is
+ temporarily terminated with OP_END when this function is called.
+ This function is called when a backward assertion is encountered, so that if it
+ fails, the error message can point to the correct place in the pattern.
+ However, we cannot do this when the assertion contains subroutine calls,
+ because they can be forward references. We solve this by remembering this case
+ and doing the check at the end; a flag specifies which mode we are running in.
  Arguments:
    code     points to the start of the pattern (the bracket)
-   options  the compiling options
+   utf      TRUE in UTF-8 / UTF-16 / UTF-32 mode
+   atend    TRUE if called when the pattern is complete
- Returns:   the fixed length, or -1 if there is no fixed length,
+   cd       the "compile data" structure
-              or -2 if \C was encountered
+ Returns:   the fixed length,
+              or -1 if there is no fixed length,
+              or -2 if \C was encountered (in UTF-8 mode only)
+              or -3 if an OP_RECURSE item was encountered and atend is FALSE
+              or -4 if an unknown opcode was encountered (internal error)
  */
  static int
- find_fixedlength(uschar *code, int options)
+ find_fixedlength(pcre_uchar *code, BOOL utf, BOOL atend, compile_data *cd)
  {
  int length = -1;
  register int branchlength = 0;
- register uschar *cc = code + 1 + LINK_SIZE;
+ register pcre_uchar *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. */
-Line 1068 
 branch, check the length against that of
+Line 1749 
 branch, check the length against that of
  for (;;)
    {
    int d;
-   register int op = *cc;
+   pcre_uchar *ce, *cs;
+   register pcre_uchar 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. */
      case OP_CBRA:
      case OP_BRA:
      case OP_ONCE:
+     case OP_ONCE_NC:
      case OP_COND:
-     d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options);
+     d = find_fixedlength(cc + ((op == OP_CBRA)? IMM2_SIZE : 0), utf, 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
+     /* Reached end of a branch; if it's a ket it is the end of a nested call.
-     call. If it's ALT it is an alternation in a nested call. If it is
+     If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
-     END it's the end of the outer call. All can be handled by the same code. */
+     an ALT. If it is 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:
+     case OP_ACCEPT:
+     case OP_ASSERT_ACCEPT:
      if (length < 0) length = branchlength;
        else if (length != branchlength) return -1;
      if (*cc != OP_ALT) return length;
-Line 1099 
 for (;;)
+Line 1789 
 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 = (pcre_uchar *)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 + IMM2_SIZE, utf, atend, cd);
+     if (d < 0) return d;
+     branchlength += d;
+     cc += 1 + LINK_SIZE;
+     break;
      /* Skip over assertive subpatterns */
      case OP_ASSERT:
-Line 1106 
 for (;;)
+Line 1811 
 for (;;)
      case OP_ASSERTBACK:
      case OP_ASSERTBACK_NOT:
      do cc += GET(cc, 1); while (*cc == OP_ALT);
-     /* Fall through */
+     cc += PRIV(OP_lengths)[*cc];
+     break;
      /* Skip over things that don't match chars */
-     case OP_REVERSE:
+     case OP_MARK:
+     case OP_PRUNE_ARG:
+     case OP_SKIP_ARG:
+     case OP_THEN_ARG:
+     cc += cc[1] + PRIV(OP_lengths)[*cc];
+     break;
+     case OP_CALLOUT:
+     case OP_CIRC:
+     case OP_CIRCM:
+     case OP_CLOSE:
+     case OP_COMMIT:
      case OP_CREF:
-     case OP_RREF:
      case OP_DEF:
-     case OP_OPT:
+     case OP_DOLL:
-     case OP_CALLOUT:
+     case OP_DOLLM:
-     case OP_SOD:
-     case OP_SOM:
      case OP_EOD:
      case OP_EODN:
-     case OP_CIRC:
+     case OP_FAIL:
-     case OP_DOLL:
+     case OP_NCREF:
+     case OP_NRREF:
      case OP_NOT_WORD_BOUNDARY:
+     case OP_PRUNE:
+     case OP_REVERSE:
+     case OP_RREF:
+     case OP_SET_SOM:
+     case OP_SKIP:
+     case OP_SOD:
+     case OP_SOM:
+     case OP_THEN:
      case OP_WORD_BOUNDARY:
-     cc += _pcre_OP_lengths[*cc];
+     cc += PRIV(OP_lengths)[*cc];
      break;
      /* Handle literal characters */
      case OP_CHAR:
-     case OP_CHARNC:
+     case OP_CHARI:
      case OP_NOT:
+     case OP_NOTI:
      branchlength++;
      cc += 2;
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-     if ((options & PCRE_UTF8) != 0)
+     if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
-       {
-       while ((*cc & 0xc0) == 0x80) cc++;
-       }
  #endif
      break;
-Line 1146 
 for (;;)
+Line 1867 
 for (;;)
      need to skip over a multibyte character in UTF8 mode.  */
      case OP_EXACT:
-     branchlength += GET2(cc,1);
+     case OP_EXACTI:
-     cc += 4;
+     case OP_NOTEXACT:
- #ifdef SUPPORT_UTF8
+     case OP_NOTEXACTI:
-     if ((options & PCRE_UTF8) != 0)
+     branchlength += (int)GET2(cc,1);
-       {
+     cc += 2 + IMM2_SIZE;
-       while((*cc & 0x80) == 0x80) cc++;
+ #ifdef SUPPORT_UTF
-       }
+     if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
  #endif
      break;
      case OP_TYPEEXACT:
      branchlength += GET2(cc,1);
-     cc += 4;
+     if (cc[1 + IMM2_SIZE] == OP_PROP || cc[1 + IMM2_SIZE] == OP_NOTPROP)
+       cc += 2;
+     cc += 1 + IMM2_SIZE + 1;
      break;
      /* Handle single-char matchers */
-Line 1168 
 for (;;)
+Line 1891 
 for (;;)
      cc += 2;
      /* Fall through */
+     case OP_HSPACE:
+     case OP_VSPACE:
+     case OP_NOT_HSPACE:
+     case OP_NOT_VSPACE:
      case OP_NOT_DIGIT:
      case OP_DIGIT:
      case OP_NOT_WHITESPACE:
-Line 1175 
 for (;;)
+Line 1902 
 for (;;)
      case OP_NOT_WORDCHAR:
      case OP_WORDCHAR:
      case OP_ANY:
+     case OP_ALLANY:
      branchlength++;
      cc++;
      break;
-     /* The single-byte matcher isn't allowed */
+     /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
+     otherwise \C is coded as OP_ALLANY. */
      case OP_ANYBYTE:
      return -2;
      /* Check a class for variable quantification */
- #ifdef SUPPORT_UTF8
-     case OP_XCLASS:
-     cc += GET(cc, 1) - 33;
-     /* Fall through */
- #endif
      case OP_CLASS:
      case OP_NCLASS:
-     cc += 33;
+ #if defined SUPPORT_UTF || defined COMPILE_PCRE16 || defined COMPILE_PCRE32
+     case OP_XCLASS:
+     /* The original code caused an unsigned overflow in 64 bit systems,
+     so now we use a conditional statement. */
+     if (op == OP_XCLASS)
+       cc += GET(cc, 1);
+     else
+       cc += PRIV(OP_lengths)[OP_CLASS];
+ #else
+     cc += PRIV(OP_lengths)[OP_CLASS];
+ #endif
      switch (*cc)
        {
+       case OP_CRPLUS:
+       case OP_CRMINPLUS:
        case OP_CRSTAR:
        case OP_CRMINSTAR:
        case OP_CRQUERY:
-Line 1206 
 for (;;)
+Line 1941 
 for (;;)
        case OP_CRRANGE:
        case OP_CRMINRANGE:
-       if (GET2(cc,1) != GET2(cc,3)) return -1;
+       if (GET2(cc,1) != GET2(cc,1+IMM2_SIZE)) return -1;
-       branchlength += GET2(cc,1);
+       branchlength += (int)GET2(cc,1);
-       cc += 5;
+       cc += 1 + 2 * IMM2_SIZE;
        break;
        default:
-Line 1218 
 for (;;)
+Line 1953 
 for (;;)
      /* Anything else is variable length */
-     default:
+     case OP_ANYNL:
+     case OP_BRAMINZERO:
+     case OP_BRAPOS:
+     case OP_BRAPOSZERO:
+     case OP_BRAZERO:
+     case OP_CBRAPOS:
+     case OP_EXTUNI:
+     case OP_KETRMAX:
+     case OP_KETRMIN:
+     case OP_KETRPOS:
+     case OP_MINPLUS:
+     case OP_MINPLUSI:
+     case OP_MINQUERY:
+     case OP_MINQUERYI:
+     case OP_MINSTAR:
+     case OP_MINSTARI:
+     case OP_MINUPTO:
+     case OP_MINUPTOI:
+     case OP_NOTMINPLUS:
+     case OP_NOTMINPLUSI:
+     case OP_NOTMINQUERY:
+     case OP_NOTMINQUERYI:
+     case OP_NOTMINSTAR:
+     case OP_NOTMINSTARI:
+     case OP_NOTMINUPTO:
+     case OP_NOTMINUPTOI:
+     case OP_NOTPLUS:
+     case OP_NOTPLUSI:
+     case OP_NOTPOSPLUS:
+     case OP_NOTPOSPLUSI:
+     case OP_NOTPOSQUERY:
+     case OP_NOTPOSQUERYI:
+     case OP_NOTPOSSTAR:
+     case OP_NOTPOSSTARI:
+     case OP_NOTPOSUPTO:
+     case OP_NOTPOSUPTOI:
+     case OP_NOTQUERY:
+     case OP_NOTQUERYI:
+     case OP_NOTSTAR:
+     case OP_NOTSTARI:
+     case OP_NOTUPTO:
+     case OP_NOTUPTOI:
+     case OP_PLUS:
+     case OP_PLUSI:
+     case OP_POSPLUS:
+     case OP_POSPLUSI:
+     case OP_POSQUERY:
+     case OP_POSQUERYI:
+     case OP_POSSTAR:
+     case OP_POSSTARI:
+     case OP_POSUPTO:
+     case OP_POSUPTOI:
+     case OP_QUERY:
+     case OP_QUERYI:
+     case OP_REF:
+     case OP_REFI:
+     case OP_SBRA:
+     case OP_SBRAPOS:
+     case OP_SCBRA:
+     case OP_SCBRAPOS:
+     case OP_SCOND:
+     case OP_SKIPZERO:
+     case OP_STAR:
+     case OP_STARI:
+     case OP_TYPEMINPLUS:
+     case OP_TYPEMINQUERY:
+     case OP_TYPEMINSTAR:
+     case OP_TYPEMINUPTO:
+     case OP_TYPEPLUS:
+     case OP_TYPEPOSPLUS:
+     case OP_TYPEPOSQUERY:
+     case OP_TYPEPOSSTAR:
+     case OP_TYPEPOSUPTO:
+     case OP_TYPEQUERY:
+     case OP_TYPESTAR:
+     case OP_TYPEUPTO:
+     case OP_UPTO:
+     case OP_UPTOI:
      return -1;
+     /* Catch unrecognized opcodes so that when new ones are added they
+     are not forgotten, as has happened in the past. */
+     default:
+     return -4;
      }
    }
  /* Control never gets here */
-Line 1229 
 for (;;)
+Line 2047 
 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
+   utf         TRUE in UTF-8 / UTF-16 / UTF-32 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 pcre_uchar *
- find_bracket(const uschar *code, BOOL utf8, int number)
+ PRIV(find_bracket)(const pcre_uchar *code, BOOL utf, int number)
  {
  for (;;)
    {
-   register int c = *code;
+   register pcre_uchar c = *code;
    if (c == OP_END) return NULL;
    /* XCLASS is used for classes that cannot be represented just by a bit
-Line 1257 
 for (;;)
+Line 2079 
 for (;;)
    if (c == OP_XCLASS) code += GET(code, 1);
+   /* Handle recursion */
+   else if (c == OP_REVERSE)
+     {
+     if (number < 0) return (pcre_uchar *)code;
+     code += PRIV(OP_lengths)[c];
+     }
    /* Handle capturing bracket */
-   else if (c == OP_CBRA)
+   else if (c == OP_CBRA || c == OP_SCBRA ||
+            c == OP_CBRAPOS || c == OP_SCBRAPOS)
      {
-     int n = GET2(code, 1+LINK_SIZE);
+     int n = (int)GET2(code, 1+LINK_SIZE);
-     if (n == number) return (uschar *)code;
+     if (n == number) return (pcre_uchar *)code;
-     code += _pcre_OP_lengths[c];
+     code += PRIV(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
+     {
+     switch(c)
+       {
+       case OP_TYPESTAR:
+       case OP_TYPEMINSTAR:
+       case OP_TYPEPLUS:
+       case OP_TYPEMINPLUS:
+       case OP_TYPEQUERY:
+       case OP_TYPEMINQUERY:
+       case OP_TYPEPOSSTAR:
+       case OP_TYPEPOSPLUS:
+       case OP_TYPEPOSQUERY:
+       if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+       break;
+       case OP_TYPEUPTO:
+       case OP_TYPEMINUPTO:
+       case OP_TYPEEXACT:
+       case OP_TYPEPOSUPTO:
+       if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == 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];
+       break;
+       }
+     /* Add in the fixed length from the table */
+     code += PRIV(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. */
-   else
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
-     {
+     if (utf) switch(c)
-     code += _pcre_OP_lengths[c];
- #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:
+       case OP_MINQUERYI:
        case OP_POSQUERY:
-       if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
+       case OP_POSQUERYI:
+       if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
        break;
        }
+ #else
+     (void)(utf);  /* Keep compiler happy by referencing function argument */
  #endif
      }
    }
-Line 1310 
 instance of OP_RECURSE.
+Line 2197 
 instance of OP_RECURSE.
  Arguments:
    code        points to start of expression
-   utf8        TRUE in UTF-8 mode
+   utf         TRUE in UTF-8 / UTF-16 / UTF-32 mode
  Returns:      pointer to the opcode for OP_RECURSE, or NULL if not found
  */
- static const uschar *
+ static const pcre_uchar *
- find_recurse(const uschar *code, BOOL utf8)
+ find_recurse(const pcre_uchar *code, BOOL utf)
  {
  for (;;)
    {
-   register int c = *code;
+   register pcre_uchar c = *code;
    if (c == OP_END) return NULL;
    if (c == OP_RECURSE) return code;
-Line 1330 
 for (;;)
+Line 2217 
 for (;;)
    if (c == OP_XCLASS) code += GET(code, 1);
-   /* Otherwise, we get the item's length from the table. In UTF-8 mode, opcodes
+   /* Otherwise, we can get the item's length from the table, except that for
-   that are followed by a character may be followed by a multi-byte character.
+   repeated character types, we have to test for \p and \P, which have an extra
-   The length in the table is a minimum, so we have to arrange to skip the extra
+   two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
-   bytes. */
+   must add in its length. */
    else
      {
-     code += _pcre_OP_lengths[c];
+     switch(c)
- #ifdef SUPPORT_UTF8
+       {
-     if (utf8) switch(c)
+       case OP_TYPESTAR:
+       case OP_TYPEMINSTAR:
+       case OP_TYPEPLUS:
+       case OP_TYPEMINPLUS:
+       case OP_TYPEQUERY:
+       case OP_TYPEMINQUERY:
+       case OP_TYPEPOSSTAR:
+       case OP_TYPEPOSPLUS:
+       case OP_TYPEPOSQUERY:
+       if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+       break;
+       case OP_TYPEPOSUPTO:
+       case OP_TYPEUPTO:
+       case OP_TYPEMINUPTO:
+       case OP_TYPEEXACT:
+       if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == 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];
+       break;
+       }
+     /* Add in the fixed length from the table */
+     code += PRIV(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. */
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
+     if (utf) switch(c)
        {
        case OP_CHAR:
-       case OP_CHARNC:
+       case OP_CHARI:
+       case OP_NOT:
+       case OP_NOTI:
        case OP_EXACT:
+       case OP_EXACTI:
+       case OP_NOTEXACT:
+       case OP_NOTEXACTI:
        case OP_UPTO:
+       case OP_UPTOI:
+       case OP_NOTUPTO:
+       case OP_NOTUPTOI:
        case OP_MINUPTO:
+       case OP_MINUPTOI:
+       case OP_NOTMINUPTO:
+       case OP_NOTMINUPTOI:
        case OP_POSUPTO:
+       case OP_POSUPTOI:
+       case OP_NOTPOSUPTO:
+       case OP_NOTPOSUPTOI:
        case OP_STAR:
+       case OP_STARI:
+       case OP_NOTSTAR:
+       case OP_NOTSTARI:
        case OP_MINSTAR:
+       case OP_MINSTARI:
+       case OP_NOTMINSTAR:
+       case OP_NOTMINSTARI:
        case OP_POSSTAR:
+       case OP_POSSTARI:
+       case OP_NOTPOSSTAR:
+       case OP_NOTPOSSTARI:
        case OP_PLUS:
+       case OP_PLUSI:
+       case OP_NOTPLUS:
+       case OP_NOTPLUSI:
        case OP_MINPLUS:
+       case OP_MINPLUSI:
+       case OP_NOTMINPLUS:
+       case OP_NOTMINPLUSI:
        case OP_POSPLUS:
+       case OP_POSPLUSI:
+       case OP_NOTPOSPLUS:
+       case OP_NOTPOSPLUSI:
        case OP_QUERY:
+       case OP_QUERYI:
+       case OP_NOTQUERY:
+       case OP_NOTQUERYI:
        case OP_MINQUERY:
+       case OP_MINQUERYI:
+       case OP_NOTMINQUERY:
+       case OP_NOTMINQUERYI:
        case OP_POSQUERY:
-       if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
+       case OP_POSQUERYI:
+       case OP_NOTPOSQUERY:
+       case OP_NOTPOSQUERYI:
+       if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
        break;
        }
+ #else
+     (void)(utf);  /* Keep compiler happy by referencing function argument */
  #endif
      }
    }
-Line 1374 
 for (;;)
+Line 2344 
 for (;;)
  can match the empty string or not. It is called from could_be_empty()
  below and from compile_branch() when checking for an unlimited repeat of a
  group that can match nothing. Note that first_significant_code() skips over
- assertions. If we hit an unclosed bracket, we return "empty" - this means we've
+ backward and negative forward assertions when its final argument is TRUE. If we
- struck an inner bracket whose current branch will already have been scanned.
+ hit an unclosed bracket, we return "empty" - this means we've struck an inner
+ bracket whose current branch will already have been scanned.
  Arguments:
    code        points to start of search
    endcode     points to where to stop
-   utf8        TRUE if in UTF8 mode
+   utf         TRUE if in UTF-8 / UTF-16 / UTF-32 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 pcre_uchar *code, const pcre_uchar *endcode,
+   BOOL utf, compile_data *cd)
  {
- register int c;
+ register pcre_uchar c;
- for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
+ for (code = first_significant_code(code + PRIV(OP_lengths)[*code], TRUE);
       code < endcode;
-      code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
+      code = first_significant_code(code + PRIV(OP_lengths)[c], TRUE))
    {
-   const uschar *ccode;
+   const pcre_uchar *ccode;
    c = *code;
-   /* Groups with zero repeats can of course be empty; skip them. */
+   /* Skip over forward assertions; the other assertions are skipped by
+   first_significant_code() with a TRUE final argument. */
-   if (c == OP_BRAZERO || c == OP_BRAMINZERO)
+   if (c == OP_ASSERT)
      {
-     code += _pcre_OP_lengths[c];
      do code += GET(code, 1); while (*code == OP_ALT);
      c = *code;
      continue;
      }
-   /* For other groups, scan the branches. */
+   /* 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_BRA || c == OP_CBRA || c == OP_ONCE)
+   if (c == OP_RECURSE)
      {
+     const pcre_uchar *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 ((int)GET(scode, 0) == (int)(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, utf, cd))
+         {
          empty_branch = TRUE;
+         break;
+         }
+       scode += GET(scode, 1);
+       }
+     while (*scode == OP_ALT);
+     if (!empty_branch) return FALSE;  /* All branches are non-empty */
+     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 += PRIV(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_ONCE_NC ||
+       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, utf, cd))
+           empty_branch = TRUE;
+         code += GET(code, 1);
+         }
+       while (*code == OP_ALT);
+       if (!empty_branch) return FALSE;   /* All branches are non-empty */
        }
-     while (*code == OP_ALT);
-     if (!empty_branch) return FALSE;   /* All branches are non-empty */
      c = *code;
      continue;
      }
-Line 1433 
 for (code = first_significant_code(code
+Line 2479 
 for (code = first_significant_code(code
    switch (c)
      {
-     /* Check for quantifiers after a class */
+     /* Check for quantifiers after a class. XCLASS is used for classes that
+     cannot be represented just by a bit map. This includes negated single
+     high-valued characters. The length in PRIV(OP_lengths)[] is zero; the
+     actual length is stored in the compiled code, so we must update "code"
+     here. */
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
      case OP_XCLASS:
-     ccode = code + GET(code, 1);
+     ccode = code += GET(code, 1);
      goto CHECK_CLASS_REPEAT;
  #endif
      case OP_CLASS:
      case OP_NCLASS:
-     ccode = code + 33;
+     ccode = code + PRIV(OP_lengths)[OP_CLASS];
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
      CHECK_CLASS_REPEAT:
  #endif
-Line 1481 
 for (code = first_significant_code(code
+Line 2531 
 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:
-Line 1499 
 for (code = first_significant_code(code
+Line 2551 
 for (code = first_significant_code(code
      case OP_TYPEEXACT:
      return FALSE;
+     /* These are going to continue, as they may be empty, but we have to
+     fudge the length for the \p and \P cases. */
+     case OP_TYPESTAR:
+     case OP_TYPEMINSTAR:
+     case OP_TYPEPOSSTAR:
+     case OP_TYPEQUERY:
+     case OP_TYPEMINQUERY:
+     case OP_TYPEPOSQUERY:
+     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+     break;
+     /* Same for these */
+     case OP_TYPEUPTO:
+     case OP_TYPEMINUPTO:
+     case OP_TYPEPOSUPTO:
+     if (code[1 + IMM2_SIZE] == OP_PROP || code[1 + IMM2_SIZE] == 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, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
      MINUPTO, and POSUPTO may be followed by a multibyte character */
- #ifdef SUPPORT_UTF8
+ #if defined SUPPORT_UTF && !defined COMPILE_PCRE32
      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 (utf && HAS_EXTRALEN(code[1])) code += GET_EXTRALEN(code[1]);
+     break;
      case OP_UPTO:
+     case OP_UPTOI:
      case OP_MINUPTO:
+     case OP_MINUPTOI:
      case OP_POSUPTO:
-     if (utf8) while ((code[2] & 0xc0) == 0x80) code++;
+     case OP_POSUPTOI:
+     if (utf && HAS_EXTRALEN(code[1 + IMM2_SIZE])) code += GET_EXTRALEN(code[1 + IMM2_SIZE]);
      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];
+     break;
+     /* None of the remaining opcodes are required to match a character. */
+     default:
+     break;
      }
    }
-Line 1539 
 return TRUE;
+Line 2643 
 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
+   utf         TRUE if in UTF-8 / UTF-16 / UTF-32 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,
+ could_be_empty(const pcre_uchar *code, const pcre_uchar *endcode,
-   BOOL utf8)
+   branch_chain *bcptr, BOOL utf, 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, utf, cd))
+     return FALSE;
    bcptr = bcptr->outer;
    }
  return TRUE;
-Line 1568 
 return TRUE;
+Line 2676 
 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 pcre_uchar *ptr, const pcre_uchar **endptr)
  {
- int terminator;          /* Don't combine these lines; the Solaris cc */
+ pcre_uchar terminator;          /* Don't combine these lines; the Solaris cc */
  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
- if (*(++ptr) == '^') ptr++;
+ for (++ptr; *ptr != CHAR_NULL; 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 1613 
 Returns:     a value representing the na
+Line 2758 
 Returns:     a value representing the na
  */
  static int
- check_posix_name(const uschar *ptr, int len)
+ check_posix_name(const pcre_uchar *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_UC_C8(ptr, pn, (unsigned int)len) == 0) return yield;
+   pn += posix_name_lengths[yield] + 1;
    yield++;
    }
  return -1;
-Line 1634 
 return -1;
+Line 2781 
 return -1;
  that is referenced. This means that groups can be replicated for fixed
  repetition simply by copying (because the recursion is allowed to refer to
  earlier groups that are outside the current group). However, when a group is
- optional (i.e. the minimum quantifier is zero), OP_BRAZERO is inserted before
+ optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
- it, after it has been compiled. This means that any OP_RECURSE items within it
+ inserted before it, after it has been compiled. This means that any OP_RECURSE
- that refer to the group itself or any contained groups have to have their
+ items within it that refer to the group itself or any contained groups have to
- offsets adjusted. That one of the jobs of this function. Before it is called,
+ have their offsets adjusted. That one of the jobs of this function. Before it
- the partially compiled regex must be temporarily terminated with OP_END.
+ is called, the partially compiled regex must be temporarily terminated with
+ OP_END.
  This function has been extended with the possibility of forward references for
  recursions and subroutine calls. It must also check the list of such references
-Line 1649 
 value in the reference (which is a group
+Line 2797 
 value in the reference (which is a group
  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
+   utf        TRUE in UTF-8 / UTF-16 / UTF-32 mode
    cd         contains pointers to tables etc.
    save_hwm   the hwm forward reference pointer at the start of the group
-Line 1657 
 Returns:     nothing
+Line 2805 
 Returns:     nothing
  */
  static void
- adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
+ adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
-   uschar *save_hwm)
+   pcre_uchar *save_hwm)
  {
- uschar *ptr = group;
+ pcre_uchar *ptr = group;
- while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
+ while ((ptr = (pcre_uchar *)find_recurse(ptr, utf)) != NULL)
    {
    int offset;
-   uschar *hc;
+   pcre_uchar *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);
+     offset = (int)GET(hc, 0);
      if (cd->start_code + offset == ptr + 1)
        {
        PUT(hc, 0, offset + adjust);
-Line 1684 
 while ((ptr = (uschar *)find_recurse(ptr
+Line 2833 
 while ((ptr = (uschar *)find_recurse(ptr
    if (hc >= cd->hwm)
      {
-     offset = GET(ptr, 1);
+     offset = (int)GET(ptr, 1);
      if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
      }
-Line 1709 
 Arguments:
+Line 2858 
 Arguments:
  Returns:         new code pointer
  */
- static uschar *
+ static pcre_uchar *
- auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
+ auto_callout(pcre_uchar *code, const pcre_uchar *ptr, compile_data *cd)
  {
  *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;
+ return code + 2 * LINK_SIZE;
  }
-Line 1738 
 Returns:             nothing
+Line 2887 
 Returns:             nothing
  */
  static void
- complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
+ complete_callout(pcre_uchar *previous_callout, const pcre_uchar *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 1752 
 PUT(previous_callout, 2 + LINK_SIZE, len
+Line 2901 
 PUT(previous_callout, 2 + LINK_SIZE, len
  *************************************************/
  /* This function is passed the start and end of a class range, in UTF-8 mode
- with UCP support. It searches up the characters, looking for internal ranges of
+ with UCP support. It searches up the characters, looking for ranges of
  characters in the "other" case. Each call returns the next one, updating the
- start address.
+ start address. A character with multiple other cases is returned on its own
+ with a special return value.
  Arguments:
    cptr        points to starting character value; updated
-Line 1762 
 Arguments:
+Line 2912 
 Arguments:
    ocptr       where to put start of othercase range
    odptr       where to put end of othercase range
- Yield:        TRUE when range returned; FALSE when no more
+ Yield:        -1 when no more
+when a range is returned
+               >0 the CASESET offset for char with multiple other cases
+                 in this case, ocptr contains the original
  */
- static BOOL
+ static int
- get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
+ get_othercase_range(pcre_uint32 *cptr, pcre_uint32 d, pcre_uint32 *ocptr,
-   unsigned int *odptr)
+   pcre_uint32 *odptr)
  {
- unsigned int c, othercase, next;
+ pcre_uint32 c, othercase, next;
+ unsigned int co;
+ /* Find the first character that has an other case. If it has multiple other
+ cases, return its case offset value. */
  for (c = *cptr; c <= d; c++)
-   { if ((othercase = _pcre_ucp_othercase(c)) != NOTACHAR) break; }
+   {
+   if ((co = UCD_CASESET(c)) != 0)
+     {
+     *ocptr = c++;   /* Character that has the set */
+     *cptr = c;      /* Rest of input range */
+     return (int)co;
+     }
+   if ((othercase = UCD_OTHERCASE(c)) != c) break;
+   }
- if (c > d) return FALSE;
+ if (c > d) return -1;  /* Reached end of range */
  *ocptr = othercase;
  next = othercase + 1;
  for (++c; c <= d; c++)
    {
-   if (_pcre_ucp_othercase(c) != next) break;
+   if (UCD_OTHERCASE(c) != next) break;
    next++;
    }
- *odptr = next - 1;
+ *odptr = next - 1;     /* End of othercase range */
- *cptr = c;
+ *cptr = c;             /* Rest of input range */
+ return 0;
+ }
- 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(pcre_uint32 c, unsigned int ptype, unsigned int pdata, BOOL negated)
+ {
+ #ifdef SUPPORT_UCP
+ const pcre_uint32 *p;
+ #endif
+ 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 == PRIV(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 (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
+           PRIV(ucp_gentype)[prop->chartype] == ucp_N) == negated;
+   case PT_SPACE:    /* Perl space */
+   return (PRIV(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 (PRIV(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 (PRIV(ucp_gentype)[prop->chartype] == ucp_L ||
+           PRIV(ucp_gentype)[prop->chartype] == ucp_N ||
+           c == CHAR_UNDERSCORE) == negated;
+ #ifdef SUPPORT_UCP
+   case PT_CLIST:
+   p = PRIV(ucd_caseless_sets) + prop->caseset;
+   for (;;)
+     {
+     if (c < *p) return !negated;
+     if (c == *p++) return negated;
+     }
+   break;  /* Control never reaches here */
+ #endif
+   }
+ return FALSE;
  }
  #endif  /* SUPPORT_UCP */
-Line 1803 
 whether the next thing could possibly ma
+Line 3047 
 whether the next thing could possibly ma
  sense to automatically possessify the repeated item.
  Arguments:
-   op_code       the repeated op code
+   previous      pointer to the repeated opcode
-   this          data for this item, depends on the opcode
+   utf           TRUE in UTF-8 / UTF-16 / UTF-32 mode
-   utf8          TRUE in UTF-8 mode
-   utf8_char     used for utf8 character bytes, NULL if not relevant
    ptr           next character in pattern
    options       options bits
    cd            contains pointers to tables etc.
-Line 1815 
 Returns:        TRUE if possessifying is
+Line 3057 
 Returns:        TRUE if possessifying is
  */
  static BOOL
- check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
+ check_auto_possessive(const pcre_uchar *previous, BOOL utf,
-   const uschar *ptr, int options, compile_data *cd)
+   const pcre_uchar *ptr, int options, compile_data *cd)
  {
- int next;
+ pcre_uint32 c = NOTACHAR;
+ pcre_uint32 next;
+ int escape;
+ pcre_uchar op_code = *previous++;
  /* Skip whitespace and comments in extended mode */
-Line 1826 
 if ((options & PCRE_EXTENDED) != 0)
+Line 3071 
 if ((options & PCRE_EXTENDED) != 0)
    {
    for (;;)
      {
-     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
+     while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
-     if (*ptr == '#')
+     if (*ptr == CHAR_NUMBER_SIGN)
        {
-       while (*(++ptr) != 0)
+       ptr++;
+       while (*ptr != CHAR_NULL)
+         {
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
+         ptr++;
+ #ifdef SUPPORT_UTF
+         if (utf) FORWARDCHAR(ptr);
+ #endif
+         }
        }
      else break;
      }
-Line 1839 
 if ((options & PCRE_EXTENDED) != 0)
+Line 3091 
 if ((options & PCRE_EXTENDED) != 0)
  /* If the next item is one that we can handle, get its value. A non-negative
  value is a character, a negative value is an escape value. */
- if (*ptr == '\\')
+ if (*ptr == CHAR_BACKSLASH)
    {
    int temperrorcode = 0;
-   next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
+   escape = check_escape(&ptr, &next, &temperrorcode, cd->bracount, options, FALSE);
    if (temperrorcode != 0) return FALSE;
    ptr++;    /* Point after the escape sequence */
    }
+ else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
- else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
    {
- #ifdef SUPPORT_UTF8
+   escape = 0;
-   if (utf8) { GETCHARINC(next, ptr); } else
+ #ifdef SUPPORT_UTF
+   if (utf) { GETCHARINC(next, ptr); } else
  #endif
    next = *ptr++;
    }
  else return FALSE;
  /* Skip whitespace and comments in extended mode */
-Line 1863 
 if ((options & PCRE_EXTENDED) != 0)
+Line 3114 
 if ((options & PCRE_EXTENDED) != 0)
    {
    for (;;)
      {
-     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
+     while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
-     if (*ptr == '#')
+     if (*ptr == CHAR_NUMBER_SIGN)
        {
-       while (*(++ptr) != 0)
+       ptr++;
+       while (*ptr != CHAR_NULL)
+         {
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
+         ptr++;
+ #ifdef SUPPORT_UTF
+         if (utf) FORWARDCHAR(ptr);
+ #endif
+         }
        }
      else break;
      }
-Line 1875 
 if ((options & PCRE_EXTENDED) != 0)
+Line 3133 
 if ((options & PCRE_EXTENDED) != 0)
  /* If the next thing is itself optional, we have to give up. */
- if (*ptr == '*' || *ptr == '?' || strncmp((char *)ptr, "{0,", 3) == 0)
+ if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
-   return FALSE;
+   STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
+     return FALSE;
- /* Now compare the next item with the previous opcode. If the previous is a
- positive single character match, "item" either contains the character or, if
- "item" is greater than 127 in utf8 mode, the character's bytes are in
- utf8_char. */
- /* Handle cases when the next item is a character. */
+ /* If the previous item is a character, get its value. */
- if (next >= 0) switch(op_code)
+ if (op_code == OP_CHAR || op_code == OP_CHARI ||
+     op_code == OP_NOT || op_code == OP_NOTI)
    {
-   case OP_CHAR:
+ #ifdef SUPPORT_UTF
- #ifdef SUPPORT_UTF8
+   GETCHARTEST(c, previous);
-   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+ #else
+   c = *previous;
  #endif
-   return item != next;
+   }
-   /* For CHARNC (caseless character) we must check the other case. If we have
+ /* Now compare the next item with the previous opcode. First, handle cases when
-   Unicode property support, we can use it to test the other case of
+ the next item is a character. */
-   high-valued characters. */
-   case OP_CHARNC:
+ if (escape == 0)
- #ifdef SUPPORT_UTF8
+   {
-   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+   /* For a caseless UTF match, the next character may have more than one other
+   case, which maps to the special PT_CLIST property. Check this first. */
+ #ifdef SUPPORT_UCP
+   if (utf && c != NOTACHAR && (options & PCRE_CASELESS) != 0)
+     {
+     unsigned int ocs = UCD_CASESET(next);
+     if (ocs > 0) return check_char_prop(c, PT_CLIST, ocs, op_code >= OP_NOT);
+     }
  #endif
-   if (item == next) return FALSE;
- #ifdef SUPPORT_UTF8
+   switch(op_code)
-   if (utf8)
      {
-     unsigned int othercase;
+     case OP_CHAR:
-     if (next < 128) othercase = cd->fcc[next]; else
+     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. We know that next can have only one other case,
+     because multi-other-case characters are dealt with above. */
+     case OP_CHARI:
+     if (c == next) return FALSE;
+ #ifdef SUPPORT_UTF
+     if (utf)
+       {
+       pcre_uint32 othercase;
+       if (next < 128) othercase = cd->fcc[next]; else
  #ifdef SUPPORT_UCP
-     othercase = _pcre_ucp_othercase((unsigned int)next);
+       othercase = UCD_OTHERCASE(next);
  #else
-     othercase = NOTACHAR;
+       othercase = NOTACHAR;
  #endif
-     return (unsigned int)item != othercase;
+       return c != othercase;
-     }
+       }
-   else
+     else
- #endif  /* SUPPORT_UTF8 */
+ #endif  /* SUPPORT_UTF */
-   return (item != cd->fcc[next]);  /* Non-UTF-8 mode */
+     return (c != TABLE_GET(next, cd->fcc, next));  /* Not UTF */
-   /* For OP_NOT, "item" must be a single-byte character. */
+     case OP_NOT:
+     return c == next;
-   case OP_NOT:
-   if (next < 0) return FALSE;  /* Not a character */
+     case OP_NOTI:
-   if (item == next) return TRUE;
+     if (c == next) return TRUE;
-   if ((options & PCRE_CASELESS) == 0) return FALSE;
+ #ifdef SUPPORT_UTF
- #ifdef SUPPORT_UTF8
+     if (utf)
-   if (utf8)
+       {
-     {
+       pcre_uint32 othercase;
-     unsigned int othercase;
+       if (next < 128) othercase = cd->fcc[next]; else
-     if (next < 128) othercase = cd->fcc[next]; else
  #ifdef SUPPORT_UCP
-     othercase = _pcre_ucp_othercase(next);
+       othercase = UCD_OTHERCASE(next);
  #else
-     othercase = NOTACHAR;
+       othercase = NOTACHAR;
  #endif
-     return (unsigned int)item == othercase;
+       return c == othercase;
-     }
+       }
-   else
+     else
- #endif  /* SUPPORT_UTF8 */
+ #endif  /* SUPPORT_UTF */
-   return (item == cd->fcc[next]);  /* Non-UTF-8 mode */
+     return (c == TABLE_GET(next, cd->fcc, next));  /* Not UTF */
-   case OP_DIGIT:
+     /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
-   return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
+     When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
-   case OP_NOT_DIGIT:
+     case OP_DIGIT:
-   return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
+     return next > 255 || (cd->ctypes[next] & ctype_digit) == 0;
-   case OP_WHITESPACE:
+     case OP_NOT_DIGIT:
-   return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
+     return next <= 255 && (cd->ctypes[next] & ctype_digit) != 0;
-   case OP_NOT_WHITESPACE:
+     case OP_WHITESPACE:
-   return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
+     return next > 255 || (cd->ctypes[next] & ctype_space) == 0;
-   case OP_WORDCHAR:
+     case OP_NOT_WHITESPACE:
-   return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
+     return next <= 255 && (cd->ctypes[next] & ctype_space) != 0;
-   case OP_NOT_WORDCHAR:
+     case OP_WORDCHAR:
-   return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
+     return next > 255 || (cd->ctypes[next] & ctype_word) == 0;
-   case OP_HSPACE:
+     case OP_NOT_WORDCHAR:
-   case OP_NOT_HSPACE:
+     return next <= 255 && (cd->ctypes[next] & ctype_word) != 0;
-   switch(next)
-     {
+     case OP_HSPACE:
-     case 0x09:
+     case OP_NOT_HSPACE:
-     case 0x20:
+     switch(next)
-     case 0xa0:
+       {
-     case 0x1680:
+       HSPACE_CASES:
-     case 0x180e:
+       return op_code == OP_NOT_HSPACE;
-     case 0x2000:
-     case 0x2001:
+       default:
-     case 0x2002:
+       return op_code != OP_NOT_HSPACE;
-     case 0x2003:
+       }
-     case 0x2004:
-     case 0x2005:
+     case OP_ANYNL:
-     case 0x2006:
+     case OP_VSPACE:
-     case 0x2007:
+     case OP_NOT_VSPACE:
-     case 0x2008:
+     switch(next)
-     case 0x2009:
+       {
-     case 0x200A:
+       VSPACE_CASES:
-     case 0x202f:
+       return op_code == OP_NOT_VSPACE;
-     case 0x205f:
-     case 0x3000:
+       default:
-     return op_code != OP_HSPACE;
+       return op_code != OP_NOT_VSPACE;
-     default:
+       }
-     return op_code == OP_HSPACE;
-     }
+ #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
-   case OP_VSPACE:
-   case OP_NOT_VSPACE:
-   switch(next)
-     {
-     case 0x0a:
-     case 0x0b:
-     case 0x0c:
-     case 0x0d:
-     case 0x85:
-     case 0x2028:
-     case 0x2029:
-     return op_code != OP_VSPACE;
      default:
-     return op_code == OP_VSPACE;
+     return FALSE;
      }
-   default:
-   return FALSE;
    }
+ /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
- /* Handle the case when the next item is \d, \s, etc. */
+ 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_CHARNC:
+   case OP_CHARI:
- #ifdef SUPPORT_UTF8
+   switch(escape)
-   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
- #endif
-   switch(-next)
      {
      case ESC_d:
-     return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
+     return c > 255 || (cd->ctypes[c] & ctype_digit) == 0;
      case ESC_D:
-     return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
+     return c <= 255 && (cd->ctypes[c] & ctype_digit) != 0;
      case ESC_s:
-     return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
+     return c > 255 || (cd->ctypes[c] & ctype_space) == 0;
      case ESC_S:
-     return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
+     return c <= 255 && (cd->ctypes[c] & ctype_space) != 0;
      case ESC_w:
-     return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
+     return c > 255 || (cd->ctypes[c] & ctype_word) == 0;
      case ESC_W:
-     return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
+     return c <= 255 && (cd->ctypes[c] & ctype_word) != 0;
      case ESC_h:
      case ESC_H:
-     switch(item)
+     switch(c)
        {
-       case 0x09:
+       HSPACE_CASES:
-       case 0x20:
+       return escape != ESC_h;
-       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;
+       return escape == ESC_h;
        }
      case ESC_v:
      case ESC_V:
-     switch(item)
+     switch(c)
        {
-       case 0x0a:
+       VSPACE_CASES:
-       case 0x0b:
+       return escape != ESC_v;
-       case 0x0c:
-       case 0x0d:
-       case 0x85:
-       case 0x2028:
-       case 0x2029:
-       return -next != ESC_v;
        default:
-       return -next == ESC_v;
+       return escape == 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[escape - ESC_DU];
+       escape = check_escape(&ptr, &next, &temperrorcode, 0, options, FALSE);
+       if (temperrorcode != 0) return FALSE;
+       ptr++;    /* For compatibility */
        }
+     /* Fall through */
+     case ESC_p:
+     case ESC_P:
+       {
+       unsigned int ptype = 0, pdata = 0;
+       int errorcodeptr;
+       BOOL negated;
+       ptr--;      /* Make ptr point at the p or P */
+       if (!get_ucp(&ptr, &negated, &ptype, &pdata, &errorcodeptr))
+         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_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
+           return FALSE;
+       /* Do the property check. */
+       return check_char_prop(c, ptype, pdata, (escape == 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 ||
+   return escape == ESC_D || escape == ESC_s || escape == ESC_W ||
-          next == -ESC_h || next == -ESC_v;
+          escape == ESC_h || escape == ESC_v || escape == ESC_R;
    case OP_NOT_DIGIT:
-   return next == -ESC_d;
+   return escape == ESC_d;
    case OP_WHITESPACE:
-   return next == -ESC_S || next == -ESC_d || next == -ESC_w;
+   return escape == ESC_S || escape == ESC_d || escape == ESC_w;
    case OP_NOT_WHITESPACE:
-   return next == -ESC_s || next == -ESC_h || next == -ESC_v;
+   return escape == ESC_s || escape == ESC_h || escape == ESC_v || escape == ESC_R;
    case OP_HSPACE:
-   return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
+   return escape == ESC_S || escape == ESC_H || escape == ESC_d ||
+          escape == ESC_w || escape == ESC_v || escape == ESC_R;
    case OP_NOT_HSPACE:
-   return next == -ESC_h;
+   return escape == 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;
+   return escape == ESC_V || escape == ESC_d || escape == ESC_w;
    case OP_NOT_VSPACE:
-   return next == -ESC_v;
+   return escape == ESC_v || escape == ESC_R;
    case OP_WORDCHAR:
-   return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
+   return escape == ESC_W || escape == ESC_s || escape == ESC_h ||
+          escape == ESC_v || escape == ESC_R;
    case OP_NOT_WORDCHAR:
-   return next == -ESC_w || next == -ESC_d;
+   return escape == ESC_w || escape == ESC_d;
    default:
    return FALSE;
-Line 2127 
 switch(op_code)
+Line 3424 
 switch(op_code)
  /*************************************************
+ *        Add a character or range to a class     *
+ *************************************************/
+ /* This function packages up the logic of adding a character or range of
+ characters to a class. The character values in the arguments will be within the
+ valid values for the current mode (8-bit, 16-bit, UTF, etc). This function is
+ mutually recursive with the function immediately below.
+ Arguments:
+   classbits     the bit map for characters < 256
+   uchardptr     points to the pointer for extra data
+   options       the options word
+   cd            contains pointers to tables etc.
+   start         start of range character
+   end           end of range character
+ Returns:        the number of < 256 characters added
+                 the pointer to extra data is updated
+ */
+ static int
+ add_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
+   compile_data *cd, pcre_uint32 start, pcre_uint32 end)
+ {
+ pcre_uint32 c;
+ int n8 = 0;
+ /* If caseless matching is required, scan the range and process alternate
+ cases. In Unicode, there are 8-bit characters that have alternate cases that
+ are greater than 255 and vice-versa. Sometimes we can just extend the original
+ range. */
+ if ((options & PCRE_CASELESS) != 0)
+   {
+ #ifdef SUPPORT_UCP
+   if ((options & PCRE_UTF8) != 0)
+     {
+     int rc;
+     pcre_uint32 oc, od;
+     options &= ~PCRE_CASELESS;   /* Remove for recursive calls */
+     c = start;
+     while ((rc = get_othercase_range(&c, end, &oc, &od)) >= 0)
+       {
+       /* Handle a single character that has more than one other case. */
+       if (rc > 0) n8 += add_list_to_class(classbits, uchardptr, options, cd,
+         PRIV(ucd_caseless_sets) + rc, oc);
+       /* Do nothing if the other case range is within the original range. */
+       else if (oc >= start && od <= end) continue;
+       /* Extend the original range if there is overlap, noting that if oc < c, we
+       can't have od > end because a subrange is always shorter than the basic
+       range. Otherwise, use a recursive call to add the additional range. */
+       else if (oc < start && od >= start - 1) start = oc; /* Extend downwards */
+       else if (od > end && oc <= end + 1) end = od;       /* Extend upwards */
+       else n8 += add_to_class(classbits, uchardptr, options, cd, oc, od);
+       }
+     }
+   else
+ #endif  /* SUPPORT_UCP */
+   /* Not UTF-mode, or no UCP */
+   for (c = start; c <= end && c < 256; c++)
+     {
+     SETBIT(classbits, cd->fcc[c]);
+     n8++;
+     }
+   }
+ /* Now handle the original range. Adjust the final value according to the bit
+ length - this means that the same lists of (e.g.) horizontal spaces can be used
+ in all cases. */
+ #if defined COMPILE_PCRE8
+ #ifdef SUPPORT_UTF
+   if ((options & PCRE_UTF8) == 0)
+ #endif
+   if (end > 0xff) end = 0xff;
+ #elif defined COMPILE_PCRE16
+ #ifdef SUPPORT_UTF
+   if ((options & PCRE_UTF16) == 0)
+ #endif
+   if (end > 0xffff) end = 0xffff;
+ #endif /* COMPILE_PCRE[8|16] */
+ /* If all characters are less than 256, use the bit map. Otherwise use extra
+ data. */
+ if (end < 0x100)
+   {
+   for (c = start; c <= end; c++)
+     {
+     n8++;
+     SETBIT(classbits, c);
+     }
+   }
+ else
+   {
+   pcre_uchar *uchardata = *uchardptr;
+ #ifdef SUPPORT_UTF
+   if ((options & PCRE_UTF8) != 0)  /* All UTFs use the same flag bit */
+     {
+     if (start < end)
+       {
+       *uchardata++ = XCL_RANGE;
+       uchardata += PRIV(ord2utf)(start, uchardata);
+       uchardata += PRIV(ord2utf)(end, uchardata);
+       }
+     else if (start == end)
+       {
+       *uchardata++ = XCL_SINGLE;
+       uchardata += PRIV(ord2utf)(start, uchardata);
+       }
+     }
+   else
+ #endif  /* SUPPORT_UTF */
+   /* Without UTF support, character values are constrained by the bit length,
+   and can only be > 256 for 16-bit and 32-bit libraries. */
+ #ifdef COMPILE_PCRE8
+     {}
+ #else
+   if (start < end)
+     {
+     *uchardata++ = XCL_RANGE;
+     *uchardata++ = start;
+     *uchardata++ = end;
+     }
+   else if (start == end)
+     {
+     *uchardata++ = XCL_SINGLE;
+     *uchardata++ = start;
+     }
+ #endif
+   *uchardptr = uchardata;   /* Updata extra data pointer */
+   }
+ return n8;    /* Number of 8-bit characters */
+ }
+ /*************************************************
+ *        Add a list of characters to a class     *
+ *************************************************/
+ /* This function is used for adding a list of case-equivalent characters to a
+ class, and also for adding a list of horizontal or vertical whitespace. If the
+ list is in order (which it should be), ranges of characters are detected and
+ handled appropriately. This function is mutually recursive with the function
+ above.
+ Arguments:
+   classbits     the bit map for characters < 256
+   uchardptr     points to the pointer for extra data
+   options       the options word
+   cd            contains pointers to tables etc.
+   p             points to row of 32-bit values, terminated by NOTACHAR
+   except        character to omit; this is used when adding lists of
+                   case-equivalent characters to avoid including the one we
+                   already know about
+ Returns:        the number of < 256 characters added
+                 the pointer to extra data is updated
+ */
+ static int
+ add_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr, int options,
+   compile_data *cd, const pcre_uint32 *p, unsigned int except)
+ {
+ int n8 = 0;
+ while (p[0] < NOTACHAR)
+   {
+   int n = 0;
+   if (p[0] != except)
+     {
+     while(p[n+1] == p[0] + n + 1) n++;
+     n8 += add_to_class(classbits, uchardptr, options, cd, p[0], p[n]);
+     }
+   p += n + 1;
+   }
+ return n8;
+ }
+ /*************************************************
+ *    Add characters not in a list to a class     *
+ *************************************************/
+ /* This function is used for adding the complement of a list of horizontal or
+ vertical whitespace to a class. The list must be in order.
+ Arguments:
+   classbits     the bit map for characters < 256
+   uchardptr     points to the pointer for extra data
+   options       the options word
+   cd            contains pointers to tables etc.
+   p             points to row of 32-bit values, terminated by NOTACHAR
+ Returns:        the number of < 256 characters added
+                 the pointer to extra data is updated
+ */
+ static int
+ add_not_list_to_class(pcre_uint8 *classbits, pcre_uchar **uchardptr,
+   int options, compile_data *cd, const pcre_uint32 *p)
+ {
+ BOOL utf = (options & PCRE_UTF8) != 0;
+ int n8 = 0;
+ if (p[0] > 0)
+   n8 += add_to_class(classbits, uchardptr, options, cd, 0, p[0] - 1);
+ while (p[0] < NOTACHAR)
+   {
+   while (p[1] == p[0] + 1) p++;
+   n8 += add_to_class(classbits, uchardptr, options, cd, p[0] + 1,
+     (p[1] == NOTACHAR) ? (utf ? 0x10ffffu : 0xffffffffu) : p[1] - 1);
+   p++;
+   }
+ return n8;
+ }
+ /*************************************************
  *           Compile one branch                   *
  *************************************************/
-Line 2141 
 Arguments:
+Line 3676 
 Arguments:
    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)
+   firstcharptr    place to put the first required character
-   reqbyteptr     set to the last literal character required, else < 0
+   firstcharflagsptr place to put the first character flags, or a negative number
+   reqcharptr     place to put the last required character
+   reqcharflagsptr place to put the last required character flags, or a negative number
    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
-Line 2153 
 Returns:         TRUE on success
+Line 3691 
 Returns:         TRUE on success
  */
  static BOOL
- compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
+ compile_branch(int *optionsptr, pcre_uchar **codeptr,
-   int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
+   const pcre_uchar **ptrptr, int *errorcodeptr,
+   pcre_uint32 *firstcharptr, pcre_int32 *firstcharflagsptr,
+   pcre_uint32 *reqcharptr, pcre_int32 *reqcharflagsptr,
+   branch_chain *bcptr, int cond_depth,
    compile_data *cd, int *lengthptr)
  {
  int repeat_type, op_type;
  int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */
  int bravalue = 0;
  int greedy_default, greedy_non_default;
- int firstbyte, reqbyte;
+ pcre_uint32 firstchar, reqchar;
- int zeroreqbyte, zerofirstbyte;
+ pcre_int32 firstcharflags, reqcharflags;
- int req_caseopt, reqvary, tempreqvary;
+ pcre_uint32 zeroreqchar, zerofirstchar;
- int options = *optionsptr;
+ pcre_int32 zeroreqcharflags, zerofirstcharflags;
+ pcre_int32 req_caseopt, reqvary, tempreqvary;
+ int options = *optionsptr;               /* May change dynamically */
  int after_manual_callout = 0;
  int length_prevgroup = 0;
- register int c;
+ register pcre_uint32 c;
- register uschar *code = *codeptr;
+ int escape;
- uschar *last_code = code;
+ register pcre_uchar *code = *codeptr;
- uschar *orig_code = code;
+ pcre_uchar *last_code = code;
- uschar *tempcode;
+ pcre_uchar *orig_code = code;
+ pcre_uchar *tempcode;
  BOOL inescq = FALSE;
- BOOL groupsetfirstbyte = FALSE;
+ BOOL groupsetfirstchar = FALSE;
- const uschar *ptr = *ptrptr;
+ const pcre_uchar *ptr = *ptrptr;
- const uschar *tempptr;
+ const pcre_uchar *tempptr;
- uschar *previous = NULL;
+ const pcre_uchar *nestptr = NULL;
- uschar *previous_callout = NULL;
+ pcre_uchar *previous = NULL;
- uschar *save_hwm = NULL;
+ pcre_uchar *previous_callout = NULL;
- uschar classbits[32];
+ pcre_uchar *save_hwm = NULL;
+ pcre_uint8 classbits[32];
- #ifdef SUPPORT_UTF8
- BOOL class_utf8;
+ /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
- BOOL utf8 = (options & PCRE_UTF8) != 0;
+ must not do this for other options (e.g. PCRE_EXTENDED) because they may change
- uschar *class_utf8data;
+ dynamically as we process the pattern. */
- uschar utf8_char[6];
+ #ifdef SUPPORT_UTF
+ /* PCRE_UTF[16|32] have the same value as PCRE_UTF8. */
+ BOOL utf = (options & PCRE_UTF8) != 0;
+ #ifndef COMPILE_PCRE32
+ pcre_uchar utf_chars[6];
+ #endif
  #else
- BOOL utf8 = FALSE;
+ BOOL utf = FALSE;
- uschar *utf8_char = NULL;
+ #endif
+ /* Helper variables for OP_XCLASS opcode (for characters > 255). We define
+ class_uchardata always so that it can be passed to add_to_class() always,
+ though it will not be used in non-UTF 8-bit cases. This avoids having to supply
+ alternative calls for the different cases. */
+ pcre_uchar *class_uchardata;
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
+ BOOL xclass;
+ pcre_uchar *class_uchardata_base;
  #endif
- #ifdef DEBUG
+ #ifdef PCRE_DEBUG
  if (lengthptr != NULL) DPRINTF((">> start branch\n"));
  #endif
-Line 2202 
 greedy_non_default = greedy_default ^ 1;
+Line 3762 
 greedy_non_default = greedy_default ^ 1;
  /* Initialize no first byte, no required byte. REQ_UNSET means "no char
  matching encountered yet". It gets changed to REQ_NONE if we hit something that
- matches a non-fixed char first char; reqbyte just remains unset if we never
+ matches a non-fixed char first char; reqchar just remains unset if we never
  find one.
  When we hit a repeat whose minimum is zero, we may have to adjust these values
  to take the zero repeat into account. This is implemented by setting them to
- zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
+ zerofirstbyte and zeroreqchar when such a repeat is encountered. The individual
  item types that can be repeated set these backoff variables appropriately. */
- firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
+ firstchar = reqchar = zerofirstchar = zeroreqchar = 0;
+ firstcharflags = reqcharflags = zerofirstcharflags = zeroreqcharflags = REQ_UNSET;
- /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
+ /* The variable req_caseopt contains either the REQ_CASELESS value
- according to the current setting of the caseless flag. REQ_CASELESS is a bit
+ or zero, according to the current setting of the caseless flag. The
- value > 255. It is added into the firstbyte or reqbyte variables to record the
+ REQ_CASELESS leaves the lower 28 bit empty. It is added into the
- case status of the value. This is used only for ASCII characters. */
+ firstchar or reqchar variables to record the case status of the
+ value. This is used only for ASCII characters. */
- req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
+ req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS:0;
  /* Switch on next character until the end of the branch */
  for (;; ptr++)
    {
    BOOL negate_class;
+   BOOL should_flip_negation;
    BOOL possessive_quantifier;
    BOOL is_quantifier;
    BOOL is_recurse;
    BOOL reset_bracount;
-   int class_charcount;
+   int class_has_8bitchar;
-   int class_lastchar;
+   int class_one_char;
    int newoptions;
    int recno;
    int refsign;
    int skipbytes;
-   int subreqbyte;
+   pcre_uint32 subreqchar, subfirstchar;
-   int subfirstbyte;
+   pcre_int32 subreqcharflags, subfirstcharflags;
    int terminator;
-   int mclength;
+   unsigned int mclength;
-   uschar mcbuffer[8];
+   unsigned int tempbracount;
+   pcre_uint32 ec;
+   pcre_uchar mcbuffer[8];
-   /* Get next byte in the pattern */
+   /* Get next character in the pattern */
    c = *ptr;
+   /* If we are at the end of a nested substitution, revert to the outer level
+   string. Nesting only happens one level deep. */
+   if (c == CHAR_NULL && 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 DEBUG
+ #ifdef PCRE_DEBUG
      if (code > cd->hwm) cd->hwm = code;                 /* High water info */
  #endif
-     if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
+     if (code > cd->start_workspace + cd->workspace_size -
+         WORK_SIZE_SAFETY_MARGIN)                       /* Check for overrun */
        {
        *errorcodeptr = ERR52;
        goto FAILED;
-Line 2274 
 for (;; ptr++)
+Line 3850 
 for (;; ptr++)
        goto FAILED;
        }
-     *lengthptr += code - last_code;
+     *lengthptr += (int)(code - last_code);
-     DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
+     DPRINTF(("length=%d added %d c=%c (0x%x)\n", *lengthptr,
+       (int)(code - last_code), c, 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,
-Line 2285 
 for (;; ptr++)
+Line 3862 
 for (;; ptr++)
        {
        if (previous > orig_code)
          {
-         memmove(orig_code, previous, code - previous);
+         memmove(orig_code, previous, IN_UCHARS(code - previous));
          code -= previous - orig_code;
          previous = orig_code;
          }
-Line 2301 
 for (;; ptr++)
+Line 3878 
 for (;; ptr++)
    /* In the real compile phase, just check the workspace used by the forward
    reference list. */
-   else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
+   else if (cd->hwm > cd->start_workspace + cd->workspace_size -
+            WORK_SIZE_SAFETY_MARGIN)
      {
      *errorcodeptr = ERR52;
      goto FAILED;
-Line 2309 
 for (;; ptr++)
+Line 3887 
 for (;; ptr++)
    /* If in \Q...\E, check for the end; if not, we have a literal */
-   if (inescq && c != 0)
+   if (inescq && c != CHAR_NULL)
      {
-     if (c == '\\' && ptr[1] == 'E')
+     if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
        {
        inescq = FALSE;
        ptr++;
-Line 2337 
 for (;; ptr++)
+Line 3915 
 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));